Neuroplasticity mechanisms of NDBIs in autism: a review from brain connectivity to behavioral improvement

Abstract

Autism Spectrum Disorder (ASD) is a complex neurodevelopmental condition characterized by deficits in social communication and interaction. In recent years, Naturalistic Developmental Behavioral Interventions (NDBI), particularly the Early Start Denver Model (ESDM), have demonstrated significant efficacy in enhancing social cognitive functions in children with autism. This review synthesizes current neuroimaging research on ESDM and related NDBI approaches, with a focus on how these interventions promote neuroplasticity by remodeling brain connectivity within social cognitive networks. We examine the neural substrates underlying behavioral improvements and discuss the critical factors influencing intervention outcomes, including timing, intensity, and duration. By integrating findings on brain network reorganization and functional enhancement, this article aims to provide theoretical insights and practical guidance for optimizing clinical interventions and informing future research directions in ASD treatment.

Introduction

Autism spectrum disorder (ASD) is a complex neurodevelopmental condition characterized primarily by persistent deficits in social communication and interaction, alongside restricted, repetitive patterns of behaviors, interests, or activities. These core symptoms manifest early in development and significantly affect individuals’ adaptive functioning and quality of life. The neurobiological underpinnings of ASD are multifaceted and heterogeneous, reflecting a spectrum of etiologies and phenotypic presentations. Emerging evidence from neuroimaging, genetic, and neurophysiological studies has highlighted atypical brain connectivity and altered neuroplasticity as central features contributing to the core symptomatology of ASD. For instance, studies have demonstrated abnormal gray matter asymmetry in brain regions implicated in social interaction, language, and repetitive behaviors, suggesting distinct neurobiological substrates underpinning these symptom domains [1]. Functional connectivity analyses have further revealed that networks such as the default mode network (DMN), cognitive control, and sensory perceptual systems are differentially involved in social-communication deficits, underscoring the complex neural network dysfunctions in ASD [2]. Moreover, atypical hemispheric lateralization patterns, particularly in language and motor regions, have been associated with symptom severity and language delay, indicating individualized neuroanatomical deviations that may serve as stratification markers for tailored interventions [3]. These neurobiological insights are critical for understanding the heterogeneity of ASD and for guiding the development of targeted therapeutic strategies.

Naturalistic Developmental Behavioral Interventions (NDBIs) have emerged as evidence-based approaches that integrate applied behavior analysis principles with developmental science to address the core challenges of ASD in early childhood. Among these, the Early Start Denver Model (ESDM) stands out as a comprehensive, manualized intervention designed for young children with ASD, emphasizing naturalistic teaching within play and daily routines to promote social communication, cognitive, and adaptive skills. Clinical trials have demonstrated that ESDM can significantly improve language and social communication outcomes, with effects observed in both controlled efficacy studies and community-based settings [4, 5]. The model’s emphasis on parent-mediated strategies also enhances caregiver fidelity and engagement, which are pivotal for sustaining intervention gains [6]. Despite these promising outcomes, variability in individual response to ESDM highlights the need for identifying predictors of treatment efficacy. Factors such as baseline cognitive and language abilities, joint attention skills, and lower levels of restricted/repetitive behaviors have been associated with more favorable responses, informing personalized treatment planning [7]. The widespread clinical adoption of ESDM and other NDBIs necessitates ongoing research into implementation determinants, scalability, and long-term effectiveness across diverse populations.

Neuroplasticity—the brain’s capacity to reorganize structurally and functionally in response to experience—is increasingly recognized as a critical mechanism underlying the efficacy of behavioral interventions in ASD. Abnormal neuroplasticity has been implicated in the pathophysiology of ASD, affecting synaptic function, neurotransmitter systems, and neural network dynamics. For example, dysregulation of excitatory/inhibitory balance, particularly involving glutamatergic and GABAergic neurotransmission, contributes to sensory processing abnormalities and social cognition deficits characteristic of ASD [8, 9]. Advances in neuroimaging and neurophysiological techniques have revealed hyper-plasticity phenomena in motor and prefrontal cortices of autistic individuals, which may underlie motor, sensory, and executive function difficulties [10]. Interventions such as repetitive transcranial magnetic stimulation (rTMS) and targeted cognitive training leverage principles of neuroplasticity to modulate aberrant neural circuits and improve behavioral outcomes [11, 12]. Furthermore, molecular studies have identified alterations in neuroplasticity-related biomarkers, including brain-derived neurotrophic factor (BDNF) and synaptic proteins, which could serve as therapeutic targets and indicators of intervention response [13]. Understanding the neuroplastic mechanisms engaged by NDBIs like ESDM is essential for optimizing intervention strategies and enhancing their impact on brain development.

This review aims to systematically synthesize the current evidence on the neuroplasticity mechanisms underlying NDBIs, with a focus on the ESDM. By integrating findings from neuroimaging, neurophysiology, molecular biology, and clinical studies, we seek to elucidate how ESDM and related NDBIs induce changes in brain connectivity and function that translate into behavioral improvements. Additionally, we explore the implications of these mechanisms for refining intervention protocols and developing precision medicine approaches tailored to individual neurobiological profiles. Through this comprehensive analysis, we intend to advance the understanding of intervention-driven neuroplasticity in ASD and inform future research directions and clinical practices aimed at maximizing developmental outcomes for autistic children.

Neurobiological basis of autism and abnormal brain connectivity

Brain structural and functional abnormalities in ASD

ASD is characterized by a range of brain structural and functional abnormalities that underlie its core symptoms, including social communication deficits and repetitive behaviors. Neuroimaging studies have consistently revealed atypical features in gray matter volume, white matter integrity, and functional activity across various brain regions in individuals with ASD. Structural magnetic resonance imaging (MRI) investigations indicate that ASD patients often exhibit increased cortical thickness and volume, particularly in frontal and parietal lobes, as demonstrated in toddlers aged 2–4 years [14]. These alterations suggest disrupted neurodevelopmental trajectories, with specific enlargement in posterior frontal and cingulate areas distinguishing ASD from developmental delay controls. Moreover, gray matter abnormalities extend to language-related regions, where changes in cortical thickness and gyrification correlate with language impairments in school-aged children with ASD [15]. White matter abnormalities are also prevalent, with diffusion MRI studies revealing disrupted microstructural integrity and altered connectivity patterns, especially in tracts linking social brain regions [16, 17].

Functionally, ASD is associated with atypical activity and connectivity in key brain networks implicated in social cognition. The “social brain” network, encompassing the prefrontal cortex, superior temporal sulcus, amygdala, and related limbic structures, shows consistent abnormalities at multiple neuroimaging modalities [18]. For example, functional MRI (fMRI) studies report hyperconnectivity in frontal, anterior cingulate, and parahippocampal regions, alongside hypoconnectivity in precentral and orbitofrontal cortices [19]. These functional disruptions correlate with social and communication deficits measured by standardized clinical scales, such as the (ADOS), with particular emphasis on the left hippocampus and superior temporal gyrus [20]. Electroencephalography (EEG) investigations further reveal altered dynamic network interactions during face recognition tasks, including reduced N170 amplitudes and atypical connectivity across frontal, parietal, and temporal regions, reflecting greater neural effort to achieve comparable performance to typically developing peers [21]. Additionally, resting-state functional connectivity analyses identify aberrations within and between major brain networks such as the DMN, salience network, and executive control networks, which are critical for social and cognitive processing [22, 23].

At the subcortical level, volumetric increases have been reported in regions including the amygdala, hippocampus, thalamus, basal ganglia, and nucleus accumbens, structures integral to social and emotional processing [24, 25]. Notably, the amygdala exhibits volumetric and cellular changes associated with social behavior deficits, and its altered connectivity with cortical areas contributes to ASD symptomatology [26]. The cerebellum, traditionally linked to motor control, also shows structural and functional abnormalities in ASD, implicating it in cognitive and social deficits [27, 28]. These findings underscore the widespread nature of brain alterations in ASD, spanning cortical and subcortical regions involved in social cognition, language, sensory processing, and executive functions.

On the microstructural and molecular levels, disruptions in synaptic plasticity, neurotransmitter systems, and neuroinflammation have been implicated in ASD pathophysiology. For instance, dysregulation of the serotonergic system affects neurogenesis and neuronal circuit formation, contributing to the neurodevelopmental anomalies observed in ASD [29]. Altered neuron-glia interactions and microglial activation further exacerbate synaptic dysfunction and neuroinflammation [30]. Genetic studies reveal that mutations affecting synaptic proteins and cytoskeletal elements, such as SHANK3 and microtubule-associated proteins, lead to impaired synaptic connectivity and plasticity, which manifest as behavioral abnormalities characteristic of ASD [31–33]. Moreover, neuroplasticity-related signaling pathways, including PI3K-AKT/mTOR and Ras-ERK/MAPK, are frequently dysregulated, influencing neuronal growth, synaptic stabilization, and learning processes [34, 35].

In summary, ASD is marked by complex and multifaceted brain structural and functional abnormalities. These include atypical gray matter volumes and cortical thickness, disrupted white matter integrity, and altered functional connectivity within social cognition networks. Subcortical structures critical for emotional and social processing also show volumetric and connectivity changes. Underlying these macroscopic alterations are molecular and cellular dysfunctions affecting synaptic plasticity, neurotransmission, and neuroinflammation. Together, these neurobiological abnormalities provide a framework for understanding the diverse clinical manifestations of ASD and offer potential biomarkers and targets for intervention.

Abnormal patterns of brain connectivity networks

ASD is characterized by pervasive abnormalities in brain connectivity networks, notably involving the DMN, social cognition networks, and executive control networks. The DMN, which typically mediates self-referential and social cognitive processes, exhibits disrupted functional connectivity in ASD, reflecting impairments in social interaction and communication. Studies employing resting-state fMRI have consistently demonstrated altered connectivity within and between the DMN, central executive network (CEN), and salience network (SN), the so-called “triple-network model,” in individuals with ASD. For instance, co-occurring attention deficit hyperactivity disorder (ADHD) in ASD patients further modulates these connectivity patterns, showing decreased within-network connectivity in the ventral DMN and increased between-network connectivity between ventral and dorsal DMN as well as between ventral DMN and executive control networks, highlighting the complexity of network dysfunction in ASD [22]. Moreover, deviations in resting-state functional connectivity (rsFC) among the DMN, ventral attention network, frontoparietal network, and somatomotor network correlate with distinct social behavioral dimensions in ASD males, where cognitive flexibility and working memory mediate these associations [36]. This underscores the critical role of these networks in the social deficits characteristic of ASD.

The executive control network, responsible for higher-order cognitive functions such as working memory and cognitive flexibility, also shows altered connectivity patterns in ASD. Functional connectivity analyses reveal that ASD individuals have decreased local connections but increased feeder connections within rich-club organization, suggesting a compensatory mechanism for disrupted local circuitry [37]. Dynamic functional network connectivity (dFNC) studies further reveal transient abnormalities in sensorimotor, visual, and DMNs, with altered dwell times and connectivity states correlating with clinical symptoms such as stereotyped behaviors [38]. These findings collectively indicate that ASD involves both static and dynamic dysregulation of large-scale brain networks.

Diffusion tensor imaging (DTI) studies complement these functional findings by revealing structural connectivity abnormalities, particularly in white matter tracts linking cortical and subcortical regions. Multimodal MRI analyses demonstrate significant alterations in cortical-cerebellar circuits, including thalamic and basal ganglia connections, which are implicated in ASD pathophysiology and can serve as diagnostic biomarkers [39]. Structural covariance networks derived from T1-weighted MRI also indicate inefficient morphological connectivity in cortico-striatal-thalamic-cortical circuits and reduced small-worldness in young children with ASD, correlating with social communication deficits [40]. Furthermore, abnormalities in language-related white matter tracts, such as the superior longitudinal fasciculus and cingulum, show reduced fractional anisotropy and increased diffusivity, predominantly in the left hemisphere, and these structural disruptions associate with language impairments and symptom severity [41].

Advanced graph neural network models leveraging high-order spatiotemporal dynamic functional connectivity have enhanced the classification accuracy of ASD, emphasizing the importance of dynamic brain network features in capturing ASD heterogeneity [42, 43]. EEG microstate analyses reveal reduced static and dynamic functional connectivity in ASD children, with interventions such as transcranial direct current stimulation (tDCS) modulating these networks [44]. Similarly, EEG studies demonstrate local hyper-connectivity alongside decreased effective connectivity across hemispheres, reflecting complex alterations in neural communication [45].

In sum, ASD is marked by aberrant connectivity patterns across multiple brain networks, including the DMN, social cognition, and executive control networks, with both functional and structural connectivity disruptions. These abnormalities manifest as altered static and dynamic connectivity states, impaired network integration and segregation, and disrupted white matter pathways. Multimodal neuroimaging combining fMRI, DTI, and EEG provides converging evidence for these network-level dysfunctions, which correlate with core ASD symptoms and offer potential biomarkers for diagnosis and targets for intervention. The developmental trajectory and comorbidity with other disorders such as ADHD further complicate the connectivity profiles, underscoring the need for individualized and age-sensitive approaches in understanding and treating ASD [17, 46, 47].

Application of neuroplasticity theory in autism

Neuroplasticity, the brain’s intrinsic ability to reorganize its structure, function, and connections in response to internal and external stimuli, is a fundamental concept in understanding brain development and adaptation. In the context of autism ASD, neuroplasticity theory provides a valuable framework to explain both the atypical neural development observed and the potential for therapeutic interventions targeting brain adaptability. Neuroplasticity encompasses various processes including synaptogenesis, neurogenesis, long-term potentiation (LTP), and long-term depression (LTD), which collectively support learning, memory, and behavioral flexibility. In ASD, these neuroplastic processes are often disrupted, manifesting as altered synaptic connectivity, excitation-inhibition imbalances, and atypical neural circuitry formation. For instance, evidence suggests an early hyperplasticity phase characterized by excessive synaptic formation and excitation-inhibition imbalance, which contrasts with other neuropsychiatric disorders like schizophrenia where synaptic pruning is excessive during adolescence [48]. Molecular studies implicate aberrant glutamate excitotoxicity, oxidative stress, and mitochondrial dysfunction as biochemical underpinnings of this atypical neuroplasticity in ASD, highlighting the multifactorial and polyetiological nature of the disorder [49]. Moreover, alterations in key neuroplasticity-related molecules such as SHANK2, NCAM1, and BDNF have been identified as biomarkers and potential therapeutic targets, underscoring the biological complexity of neuroplastic changes in autism [50]. These molecular and synaptic abnormalities contribute to the characteristic deficits in social communication, sensory processing, and repetitive behaviors seen in ASD. Importantly, neuroplasticity is not static but varies across developmental stages, suggesting critical windows during which interventions may be most effective. However, individuals with ASD often exhibit limitations in neuroplasticity, including reduced capacity for synaptic remodeling and adaptive learning, which constrains the brain’s ability to compensate for developmental disruptions [8]. This restricted plasticity may narrow the therapeutic window, emphasizing the need for early and targeted interventions that harness residual plasticity to improve cognitive and behavioral outcomes. Emerging therapeutic strategies, such as rTMS, noninvasive brain stimulation, and neurofeedback, aim to modulate neuroplasticity to ameliorate core and associated ASD symptoms by restoring excitation-inhibition balance and enhancing synaptic function [10, 11]. Additionally, interventions leveraging neuroplasticity principles, including sensory integration therapies and targeted cognitive training, have shown promise in improving sensory processing and executive functions by promoting adaptive neural reorganization [9, 51]. The dynamic interplay of genetic, environmental, and epigenetic factors further modulates neuroplasticity in ASD, suggesting that personalized approaches considering individual neurobiological profiles may optimize intervention efficacy [52]. Overall, the application of neuroplasticity theory in autism elucidates the mechanisms underlying atypical brain development and provides a scientific basis for designing interventions that capitalize on the brain’s capacity for change, thereby offering hope for improving functional outcomes in individuals with ASD.

Early Denver Model (ESDM) and neuroimaging evidence of NDBIs

ESDM’s intervention principles and implementation methods

The ESDM is a comprehensive, evidence-based early intervention approach designed specifically for young children with ASD, typically targeting toddlers and preschoolers aged 12 to 48 months. Rooted in naturalistic developmental behavioral intervention (NDBI) principles, ESDM integrates applied behavior analysis (ABA) techniques with developmental and relationship-based approaches to promote social communication, cognitive, and adaptive skills in a child-centered and play-based manner. The core philosophy of ESDM emphasizes early, intensive, and individualized intervention that is responsive to the child’s developmental level and interests, thereby fostering motivation and engagement in learning opportunities. The intervention content covers a broad range of developmental domains, including language, social interaction, imitation, play, and cognitive skills, aiming to reduce core ASD symptoms and improve overall functioning [53–55].

ESDM is applicable to a wide spectrum of children with ASD, including those with varying cognitive abilities and symptom severities. Studies have demonstrated its efficacy across different developmental profiles, with some evidence suggesting that children with higher baseline social cognitive skills, such as intention to communicate and attention to faces, tend to respond more favorably to ESDM [56]. Moreover, the intervention has been adapted successfully in diverse cultural and resource contexts, including community settings in Japan, Italy, Taiwan, and Senegal, highlighting its versatility and feasibility beyond specialized research centers [53, 57–59].

A distinctive feature of ESDM is its implementation within naturalistic environments, emphasizing the importance of embedding intervention strategies into the child’s daily routines and play contexts. This natural environment setting facilitates generalization of skills and promotes meaningful social interactions. The model encourages active parental involvement, recognizing parents as pivotal agents of intervention delivery. Parent coaching components, such as the Parent-implemented ESDM (P-ESDM), have been developed to empower caregivers with the knowledge and skills to apply ESDM strategies consistently, enhancing intervention intensity and promoting sustainability of gains. Parent involvement not only improves child outcomes in engagement, communication, and imitation but also reduces parental stress and fosters a sense of empowerment and competence [60–63].

The intervention is typically delivered through a combination of individual and group sessions, with staff-to-child ratios adjusted according to the child’s age, language level, and developmental quotient. For example, individual ESDM (I-ESDM) with a 1:1 ratio has shown greater effectiveness for children under 2 years with lower language age, whereas group ESDM (G-ESDM) with higher child-to-staff ratios can be effective for children aged 2 years or older with higher developmental quotients [64]. The intensity of ESDM varies across implementations, ranging from low-intensity (e.g., 2–3 h per week) to more intensive schedules (up to 25 h per week), with evidence supporting positive developmental outcomes even at lower intensities in community and inclusive preschool settings [57, 65, 66]. Table 1 provides a comprehensive summary of these core features and implementation elements of ESDM, systematically organizing the intervention principles, target population characteristics, content domains, delivery formats, intensity parameters, parent involvement strategies, and implementation settings discussed above. This structured overview facilitates comparison across different ESDM implementation models and serves as a practical reference for clinicians and researchers.

Table 1.

Core features and implementation elements of the ESDM

Component	Details
Intervention principles	Integrates ABA principles with developmental science, utilizing Naturalistic Developmental Behavioral Intervention (NDBI) methods through child-centered teaching embedded in play and daily routines
Target population	• Age: Toddlers and preschoolers aged 12–48 months • Applicable to children with ASD across varying cognitive levels and symptom severities • Children with higher baseline social cognitive skills (e.g., intention to communicate, attention to faces) typically show more favorable responses
Intervention content	• Social interaction and communication • Language development • Imitation skills • Play skills • Cognitive skills • Adaptive behaviors
Implementation formats	• I-ESDM: 1:1 ratio, more effective for children under 2 years with lower language age • G-ESDM: Higher child-to-staff ratios, effective for children aged 2 + with higher developmental quotients • P-ESDM: training parents to implement strategies at home
Intervention intensity	• Low intensity: 2–3 h per week • High intensity: up to 25 h per week • Evidence supports positive developmental outcomes even at lower intensities in community and inclusive preschool settings
Parent involvement	• Parents as pivotal intervention agents • Parent training and coaching enhance implementation fidelity • Improves child engagement, communication, and imitation abilities • Reduces parental stress and enhances sense of competence
Implementation settings	• Natural environments: home, preschool, community • Embedded within daily routines and play contexts • Successfully implemented across diverse cultural contexts (Japan, Italy, Taiwan, Senegal, etc.)
Key strengths	• Facilitates skill generalization • Enhances authenticity and meaningfulness of social interactions • Highly adaptable, implementable across varying resource contexts • Sustainable, with parent involvement ensuring intervention continuity

The naturalistic setting and parent involvement are critical for the social validity and acceptability of ESDM. Preschool staff and parents report that the integration of ESDM strategies into daily routines and collaborative networks enhances the learning environment and supports children’s developmental progress. The model’s adaptability allows it to be embedded seamlessly within existing educational and healthcare frameworks, facilitating broader access and sustainability, especially in multiethnic and socioeconomically diverse communities [67, 68]. Furthermore, fidelity of implementation by therapists and parents is a key determinant of treatment effectiveness, underscoring the importance of comprehensive training and ongoing coaching to maintain high-quality delivery of ESDM principles [69, 70].

In summary, the ESDM is a naturalistic, developmental behavioral intervention that targets young children with ASD through individualized, play-based strategies implemented in natural environments. Its core principles include early and intensive intervention, comprehensive developmental focus, and active parent participation. The flexibility in delivery formats and settings, combined with a strong emphasis on parent coaching and naturalistic contexts, underpins its effectiveness and growing acceptance in diverse clinical and community settings worldwide.

Review of neuroimaging studies

Neuroimaging techniques such as fMRI, EEG, and fNIRS have been extensively utilized to investigate the neural underpinnings of ASD and to monitor the effects of behavioral interventions. These modalities provide complementary insights into brain function, connectivity, and neuroplasticity associated with ASD and its treatment.

Resting-state fMRI (rs-fMRI) studies have revealed widespread atypical functional connectivity patterns in individuals with ASD. Large-scale analyses demonstrate both hypo- and hyperconnectivity across various brain networks, including the DMN, sensory, attentional, and subcortical systems. For example, hypo-connectivity predominantly affects sensory and higher-order attentional networks, correlating with social impairments and repetitive behaviors, while hyperconnectivity is observed mainly between the DMN and other cortical and subcortical regions, also relating to core ASD symptoms [71]. Dynamic regional brain activity analyses using metrics like dynamic regional homogeneity (dReHo) and dynamic amplitude of low-frequency fluctuations (dALFF) further highlight abnormal temporal variability in neural activity across multiple cortical areas in adults with ASD, with significant correlations to clinical severity scores [72]. These findings underscore the complex and dynamic nature of neural dysfunction in ASD.

EEG studies complement fMRI by providing high temporal resolution data, capturing neural oscillatory dynamics and connectivity alterations. EEG-based machine learning approaches have demonstrated promise in identifying ASD-related biomarkers, particularly in early infancy, with potential for early diagnosis. The integration of EEG with neuroimaging and genetic data enhances the understanding of ASD neurobiology and may facilitate the development of objective diagnostic tools [73].

fNIRS offers a non-invasive and child-friendly method to assess cortical hemodynamics, particularly useful in younger or less compliant populations. fNIRS studies have identified atypical neural responses to social and language stimuli in infants at elevated likelihood for ASD, indicating early alterations in functional brain lateralization and connectivity that precede overt behavioral symptoms [74].

Intervention studies employing neuroimaging have begun to elucidate the neural mechanisms underlying behavioral improvements in ASD. For instance, rTMS interventions in children with ASD have been associated with increased gray matter volume in cerebellar and cortical regions, alongside enhanced functional connectivity between key social brain areas such as the fusiform gyrus, temporal, frontal cortices, and precuneus. These neuroplastic changes correlate with behavioral improvements, suggesting that rTMS modulates cerebellar development and cognitive control networks [75]. Similarly, music therapy has been shown to improve social communication and functional brain connectivity in school-aged children with ASD, with ongoing trials investigating psychometric, neuroimaging, and biological outcomes to better understand the intervention’s efficacy [76].

Advanced machine learning and deep learning models applied to multimodal neuroimaging data (including fMRI, structural MRI, and diffusion tensor imaging) have improved diagnostic classification accuracy for ASD, identifying reproducible biomarkers related to altered connectivity patterns in fronto-parietal, temporal, and cerebellar regions. These models not only enhance diagnostic precision but also help delineate neural substrates associated with social deficits and cognitive impairments [17, 77]. Furthermore, domain adaptation techniques minimizing site-dependent variability in multi-site datasets have improved the robustness of neuroimaging-based ASD classification [78]. Table 2 systematically summarizes the key findings from neuroimaging research across different modalities, including structural MRI, resting-state and task-based fMRI, DTI, EEG, fNIRS, and multimodal integration approaches. For each imaging technique, the table presents the primary neurobiological abnormalities identified in ASD, the specific brain regions and networks involved, and the documented intervention effects. This comprehensive overview facilitates cross-modal comparison and highlights converging evidence from diverse neuroimaging methodologies, serving as a valuable reference for understanding the neural substrate of ASD and intervention-induced neuroplastic changes.

Table 2.

Summary of neuroimaging research and intervention effects in ASD

Neuroimaging modality	Key findings	Brain regions/networks involved	Intervention effects
Structural MRI	Increased cortical thickness and volume in toddlers aged 2–4 years; gray matter volume abnormalities; impaired white matter integrity	Frontal and parietal lobes, cingulate cortex; language-related regions; cortical-cerebellar circuits	Increased cerebellar and cortical gray matter volume following rTMS
Resting-state fMRI	Widespread patterns of hypo- and hyperconnectivity; abnormal dynamic regional brain activity; increased network temporal variability	DMN, sensory-attentional networks, cortico-striatal-thalamic circuits, executive control network	Enhanced frontal–temporal-occipital connectivity following neuromodulation; improved cerebellar-prefrontal connectivity
Task-based fMRI	Abnormal amygdala and fusiform gyrus activation during social cognition tasks; face recognition deficits	Amygdala, fusiform gyrus, superior temporal sulcus, prefrontal cortex	Functional restoration of amygdala post-intervention; normalization of social cognitive network activation
DTI	White matter microstructural abnormalities; impaired white matter tracts linking social brain regions; left hemisphere language-related white matter changes	Superior longitudinal fasciculus, cingulum, frontal–temporal-parietal connectivity pathways	Multimodal MRI studies show structural connectivity abnormalities can serve as diagnostic biomarkers
EEG	Abnormal microstate dynamics; reduced N170 amplitude during face recognition tasks; local hyperconnectivity and decreased interhemispheric effective connectivity	Frontal-parietal-temporal regions; theta and alpha frequency bands; sensorimotor, visual, and DMNs	tDCS modulates brain functional networks; neurofeedback training enhances functional connectivity and improves social communication
fNIRS	Atypical neural responses to social and language stimuli in infancy; early alterations in functional lateralization and connectivity	Frontal and temporal cortex; social cognition and language-related brain regions	Abnormal brain state transitions correlate with symptom severity and cognitive performance
Multimodal Integration	Combining structural and functional data improves classification accuracy; identifies reproducible biomarkers in frontal-parietal-temporal-cerebellar regions	Whole-brain multi-network integration; feature networks identified by machine learning	Provides biomarker foundation for personalized intervention strategies and treatment response monitoring

Note Intervention effects listed refer primarily to findings from neuromodulation studies (rTMS, tDCS) and behavioral interventions. Enhanced connectivity indicates increased functional or structural connections between brain regions post-intervention. Normalization refers to connectivity patterns approaching those observed in typically developing individuals. Specific ESDM-induced neuroplastic changes are detailed in subsequent sections and Fig. 2

Overall, neuroimaging studies employing fMRI, EEG, and fNIRS have converged on the characterization of ASD as a disorder of disrupted functional connectivity and atypical activation patterns in distributed brain networks implicated in social cognition, sensory processing, and executive function. Intervention-induced neuroplasticity is reflected in normalization or compensation of these neural patterns, supporting the utility of neuroimaging as a biomarker for treatment response. Future research integrating multimodal imaging, longitudinal designs, and advanced analytic frameworks promises to refine our understanding of the neural mechanisms of ASD and optimize individualized intervention strategies.

Effects of intervention on social cognitive networks

Interventions targeting social cognitive deficits in ASD have demonstrated promising effects on neural connectivity within key brain networks that subserve social cognition, particularly the fronto-temporal circuits. The fronto-temporal connectivity, involving the prefrontal cortex (PFC) and temporal lobe regions such as the superior temporal sulcus and fusiform gyrus, is critical for processing facial expressions and emotional cues. Studies employing neurofeedback and noninvasive brain stimulation techniques, such as tDCS, have reported enhanced functional connectivity between these regions following intervention, which correlates with improvements in facial expression recognition and emotional processing abilities in individuals with ASD [79, 80]. For instance, tDCS applied over the frontal cortex in valproic acid-induced ASD rat models normalized disrupted functional connectivity in frontal-striato-hippocampal circuits, leading to restored sociability and cognitive flexibility, suggesting that modulation of fronto-temporal networks can remediate social cognitive dysfunctions [80]. Similarly, neurofeedback training has been associated with increased functional connectivity and reduced temporal variability in brain networks implicated in social cognition, accompanied by behavioral gains in communication and social domains [79]. These findings underscore the neuroplastic potential of fronto-temporal circuits in ASD and their responsiveness to targeted interventions.

Beyond fronto-temporal connectivity, the amygdala and its associated emotion regulation regions, including the insula and anterior cingulate cortex, play a pivotal role in processing emotional stimuli and regulating affective responses. Functional restoration of the amygdala following intervention has been documented, with increased activation and improved connectivity correlating with enhanced emotional processing and social cognition [81]. Deep brain stimulation (DBS) targeting the anterior insula in animal models of ASD also demonstrated amelioration of autism-like behaviors, including improved sociability and reduced repetitive behaviors, alongside reversal of autism-related protein expression in the insula, indicating restoration of emotion regulation circuits [82]. Moreover, task-based fMRI studies reveal that neural activity within social cognitive networks, including the amygdala, is associated with social cognitive performance across autism and schizophrenia spectrum disorders, highlighting the amygdala’s centrality in social cognition and its potential as a target for intervention [81]. Interventions such as music therapy have also been shown to enhance functional brain connectivity involving the amygdala and related regions, leading to improvements in social communication in children with autism [76].

Collectively, these studies suggest that behavioral and neuromodulatory interventions can enhance fronto-temporal connectivity and restore amygdala function, thereby improving key social cognitive processes such as facial expression recognition and emotional processing in ASD. The modulation of these neural circuits reflects underlying neuroplastic mechanisms, offering a neurobiological basis for observed behavioral improvements. Future research integrating multimodal neuroimaging and longitudinal designs is warranted to further elucidate the dynamics of these network changes and optimize targeted interventions for social cognitive enhancement in autism.

Comparison of neural mechanisms in ESDM and other NDBI approaches

The ESDM and other NDBIs, such as Pivotal Response Treatment (PRT), share a foundational emphasis on enhancing social communication and adaptive behaviors in children with ASD through naturalistic, play-based, and child-centered approaches. Neuroimaging studies investigating the neural mechanisms underlying these interventions have begun to reveal both convergent and divergent patterns of brain plasticity associated with their therapeutic effects. ESDM, which integrates applied behavior analysis with developmental and relationship-based approaches, has been shown to induce changes in neural connectivity within social brain networks, including increased functional connectivity between regions such as the prefrontal cortex, superior temporal sulcus, and amygdala, which are critical for social cognition and emotion processing. These neural alterations correspond with improvements in joint attention, language acquisition, and social reciprocity observed behaviorally. In contrast, PRT, which targets pivotal areas of motivation and self-initiation, appears to modulate neural circuits involved in reward processing and executive function, particularly within the striatum and prefrontal regions. Functional MRI studies have demonstrated that PRT can enhance activation in the ventral striatum and medial prefrontal cortex during social reward tasks, suggesting an increased salience of social stimuli and improved motivational engagement. While both ESDM and PRT promote neuroplasticity in overlapping social brain networks, the emphasis of ESDM on early developmental stages and relationship-building may preferentially strengthen connectivity in regions supporting social perception and communication, whereas PRT’s focus on motivation may more robustly engage reward-related circuitry. Furthermore, differences in intervention delivery—such as therapist-led versus parent-mediated formats—and targeted behavioral domains may contribute to distinct neural adaptations. Despite these differences, both interventions exemplify the capacity of NDBIs to harness experience-dependent plasticity, facilitating reorganization of neural networks that underpin social and communicative functions. Understanding the nuanced neural mechanisms of various NDBI approaches not only elucidates their therapeutic pathways but also informs personalized intervention strategies that optimize brain and behavioral outcomes in ASD. Future research employing longitudinal neuroimaging and multimodal assessments will be critical to delineate the specific neural signatures of each intervention and their relation to individual variability in treatment response. This comparative neurobiological perspective underscores the importance of integrating neural metrics into the evaluation and refinement of NDBI methodologies to advance precision medicine in autism care.

Quantitative findings provide measurable links between network-level alterations and clinical phenotypes relevant to NDBI targets. In a social brain network analysis, functional connectivity strength between social-related regions and sensorimotor/cingulate areas was inversely associated with ASD severity, showing significant negative correlations with ADOS communication, social interaction, communication + social interaction, and total scores (r =  − 0.38, − 0.39, − 0.40, and − 0.30; all p < 0.01), as well as with multiple SRS (Social Responsiveness Scale) subdomains and total score (r ranging from − 0.27 to − 0.40; all p < 0.01). In addition, Social Communication Questionnaire (SCQ) total score was negatively correlated with connectivity strength (r =  − 0.27, p < 0.01) [83]. Longitudinal network analyses further identified connectivity features associated with restricted and repetitive behaviors, including decreased coupling between the left superior occipital lobe and right angular gyrus (effect =  − 0.125) and between the left insula and left caudate nucleus (effect =  − 0.089) [84]. Importantly, these quantitative metrics primarily reflect network–phenotype associations relevant to NDBI targets rather than standardized pre–post intervention effect sizes derived from harmonized ESDM versus PRT trials. However, direct head-to-head quantitative comparisons of post-intervention connectivity effect sizes between ESDM and PRT remain scarce due to heterogeneous protocols and non-overlapping imaging metrics, underscoring the need for harmonized multimodal pipelines and standardized effect-size reporting in future multi-center longitudinal designs.

The relationship between intervention timing, intensity, and neural remodeling effects

Critical period theory and timing of intervention

The concept of critical periods in neurodevelopment refers to specific windows in early life during which the brain exhibits heightened plasticity and sensitivity to environmental inputs, making it especially receptive to experience-dependent modifications. These periods are characterized by dynamic synaptic formation, pruning, and circuit refinement, which are essential for establishing mature neural networks and adaptive behaviors. The timing of interventions during these critical periods profoundly influences their efficacy, as neural circuits are more malleable and capable of reorganization in response to targeted stimuli or therapies. Early intervention, particularly within the first three years of life, coincides with these critical periods and has been shown to yield more substantial improvements in developmental trajectories for children with neurodevelopmental disorders such as ASD [85, 86]. Neuroimaging studies comparing early (0–3 years) versus later interventions reveal distinct differences in brain connectivity and plasticity. As illustrated in Fig. 1, neuroplasticity peaks during the first 1–2 years of life and gradually declines thereafter through synaptic pruning. This temporal pattern creates distinct intervention windows: early interventions (0–3 years) leverage peak plasticity, active synaptogenesis, and strong circuit remodeling capacity to achieve significant improvements, whereas late interventions (6 + years) encounter narrowed plasticity windows, synaptic stabilization, and established perineuronal nets, resulting in limited therapeutic effects.

Fig. 1.

Critical period framework and timing-dependent efficacy of interventions in neurodevelopmental disorders. The upper panel illustrates the developmental trajectory of neuroplasticity, showing peak synaptic density during the critical period (0–3 years) followed by gradual decline through synaptic pruning. Molecular regulatory mechanisms include CAMs (cell adhesion molecules), PNNs (perineuronal nets), microglial synaptic pruning, and BDNF (brain-derived neurotrophic factor) signaling pathway. The lower panel contrasts early intervention (0–3 years, green) with late intervention (6 + years, orange). Early interventions leverage peak plasticity, active synaptogenesis, and circuit remodeling capacity to achieve significant improvements, while late interventions face narrowed plasticity windows, synaptic stabilization, and limited remodeling capacity, resulting in reduced efficacy. This framework emphasizes the critical importance of intervention timing for optimal therapeutic outcomes

For instance, early behavioral interventions (EBI) capitalize on the heightened synaptic plasticity and critical period mechanisms, facilitating normalization or compensation of atypical neural pathways implicated in ASD. Conversely, interventions initiated beyond these critical windows often encounter reduced neural plasticity, limiting their capacity to induce lasting neurobiological and behavioral changes [86]. Animal models further elucidate this phenomenon; studies in Shank3 knockout mice, a model of ASD, demonstrate that social environment manipulations during critical periods can restore social behaviors and olfactory function, highlighting the temporal sensitivity of neural circuits to environmental modulation [87]. Moreover, interventions during critical periods can influence inhibitory-excitatory balance and synaptic plasticity, as evidenced by findings in Fragile X syndrome models where early pharmacological treatments ameliorate precocious fear-learning linked to altered amygdala excitability [88]. The closure of critical periods is regulated by molecular and cellular mechanisms involving neuronal adhesion molecules, perineuronal nets, and glial cell interactions, which collectively stabilize synaptic contacts and limit further plasticity [89–91]. These neurobiological insights underscore the importance of timely intervention to harness the brain’s intrinsic plasticity for optimal therapeutic outcomes.

Clinically, early diagnosis and screening at 18 to 24 months are vital to identify children who may benefit from early intervention, yet disparities in screening practices and access to services remain challenges that can delay intervention beyond critical periods [92, 93]. Parent-mediated and transdisciplinary intervention models initiated early have demonstrated sustained improvements in symptom severity and adaptive functioning, reinforcing the critical role of early engagement during sensitive developmental windows [94, 95]. In summary, critical period theory provides a foundational framework for understanding the timing-dependent efficacy of interventions in ASD, emphasizing that early, intensive, and developmentally informed approaches are essential to leverage neuroplasticity and improve long-term behavioral and cognitive outcomes. While individual critical period windows vary based on genetic background and symptom severity, neuroimaging biomarkers (e.g., functional connectivity patterns, EEG microstate dynamics) combined with behavioral assessments can help identify personalized intervention windows. For children missing the optimal 0–3 year period, "suboptimal but still effective" windows exist during preschool years (3–6 years), where moderate neuroplasticity remains available, though requiring higher intervention intensity and duration. Remedial strategies include combining neuromodulation techniques (rTMS/tDCS) with intensive behavioral interventions to enhance residual plasticity, and targeting neural circuits with extended developmental windows, such as cerebellar-prefrontal pathways.

Dose–response relationship between intervention intensity and neuroplasticity

The relationship between intervention intensity—encompassing frequency and duration—and neuroplasticity is a critical consideration in optimizing behavioral interventions for ASD. Evidence from diverse neurological and psychiatric conditions underscores that higher intensity interventions often yield more pronounced neuroplastic changes and functional improvements, although this relationship is not strictly linear and is influenced by multiple factors including timing, intervention type, and individual variability. For example, in stroke rehabilitation, studies have demonstrated that higher doses of constraint-induced movement therapy (CIMT) and bilateral arm training correlate with greater motor cortex plasticity and improved motor function, especially when interventions are sustained over longer periods and initiated during sensitive phases of recovery [96, 97]. Similarly, in Parkinson’s disease, a prospective observational study revealed that intensive aerobic exercise (measured as metabolic equivalents-minutes per week) induced greater increases in neurotrophic factors such as BDNF and improved brain connectivity compared to non-intensive exercise regimens, suggesting a dose-dependent enhancement of neuroplasticity [98]. These findings imply that both the frequency and cumulative duration of intervention sessions contribute substantially to the magnitude of neuroplastic adaptations.

Moreover, neurostimulation studies provide mechanistic insights into how intervention intensity modulates neuroplasticity. Transcranial alternating current stimulation (tACS), when administered repeatedly and tailored to individual oscillatory frequencies, has been shown to induce sustained after-effects on brain networks, with repeated sessions amplifying synaptic plasticity through mechanisms such as spike-timing-dependent plasticity and homeostatic plasticity [99]. Acute serotonin enhancement studies further suggest that pharmacological modulation can augment the neuroplastic effects of brain stimulation, with dosage-dependent impacts observed particularly in inhibitory stimulation paradigms [100]. These data collectively support the concept that higher intervention intensity, whether through increased session frequency or prolonged duration, can potentiate neuroplastic mechanisms, provided that interventions are appropriately timed and personalized.

However, the dose–response relationship is nuanced. For instance, in subacute stroke rehabilitation, a randomized controlled trial comparing early versus delayed and varying dosages of upper extremity virtual reality and robotic training found that higher dosage delivered in the delayed phase post-stroke yielded superior short-term motor gains, but these differences were not sustained at six months, indicating that timing interacts complexly with dose to influence long-term outcomes [101]. Similarly, aerobic exercise studies in stroke populations report heterogeneous neuroplasticity responses, with moderate to high-intensity training generally eliciting more robust effects, yet inconsistencies remain due to variability in neuroplasticity assessments and individual patient factors [102]. These findings highlight that while higher intensity interventions tend to promote greater neuroplasticity, the optimal dosing must consider individual neurodevelopmental stages, baseline neural status, and the specific neuroplasticity markers targeted.

In animal models, dose-dependent effects of interventions on neuroplasticity have also been documented. For example, long-term administration of lactate and high-intensity interval training in aged mice enhanced neuroplasticity biomarkers such as vascular endothelial growth factor (VEGF) and BDNF signaling pathways, with higher doses of lactate showing significant modulation of metabolic and neurotrophic factors in the hippocampus, which are critical for synaptic plasticity and cognitive function [103]. Similarly, supplementation with 3’-Sialyllactose during development improved cognitive outcomes and upregulated genes associated with synaptic growth and plasticity, suggesting that sustained and adequately dosed nutritional interventions can drive neuroplastic adaptations [104].

In summary, accumulating evidence across clinical and preclinical studies supports a dose–response relationship between intervention intensity and neuroplasticity, where higher frequency and longer duration of interventions generally facilitate more significant neuroplastic changes and functional improvements. Nevertheless, this relationship is modulated by factors such as timing relative to disease or developmental stage, intervention modality, and individual biological variability. Therefore, optimizing intervention intensity for ASD behavioral therapies requires a personalized approach that balances sufficient dosing to induce meaningful neuroplasticity while avoiding overstimulation or fatigue, thereby maximizing the potential for sustained behavioral improvements.

Individual differences and moderating factors of intervention effects

Individual differences significantly modulate the neuroplasticity responses and behavioral outcomes of natural developmental behavioral interventions in ASD, underscoring the necessity for personalized intervention approaches. Genetic background plays a pivotal role in shaping neurodevelopmental trajectories and intervention responsiveness. For example, genetic variants affecting mitochondrial dynamics and synaptic plasticity have been implicated in ASD pathophysiology, influencing neuronal function and plasticity mechanisms critical for behavioral adaptation [105]. Moreover, sex and gender differences contribute to heterogeneity in ASD presentation and neurobiological profiles, with females often exhibiting distinct neural activation patterns and symptom manifestations compared to males. Studies reveal that biological sex moderates resting-state EEG power and its association with behavioral phenotypes, suggesting that sex-specific neural mechanisms underlie ASD heterogeneity and may affect intervention outcomes [106, 107]. Symptom severity, particularly in core domains such as social communication and repetitive behaviors, further modulates neuroplastic responses. Clinical subgroups defined by symptom profiles demonstrate differential neuroanatomical deviations and developmental trajectories, which correspond to variable adaptive behavior outcomes [108, 109]. Comorbid conditions, affecting approximately 70% of children with ASD, represent critical yet understudied moderators of intervention outcomes. Co-occurring ADHD alters functional connectivity within the triple-network model [22, 46], motor developmental delays correlate with symptom severity [110], and sensory hypersensitivities reflect dysregulated excitatory–inhibitory balance [9, 111]. Environmental enrichment can ameliorate certain comorbid features by restoring GABAergic function [112], yet whether ESDM directly targets neural mechanisms underlying comorbid symptoms remains unclear. Anxiety may further constrain social-learning gains through hyperreactive amygdala–prefrontal circuitry and elevated stress physiology, whereas epilepsy or subclinical epileptiform activity may disrupt network synchrony and attenuate experience-dependent remodeling. These considerations motivate stratified trials across comorbidity profiles and adaptive protocols that incorporate comorbidity-specific components; in practice, comorbidity-informed tailoring (e.g., staged goals, modified intensity, and coordinated management of sleep, medication, or seizure control) may be required to optimize engagement and neuroplastic potential. Future studies should determine how comorbidities modulate intervention-induced connectivity changes to support personalized treatment algorithms.

Environmental factors, including early life experiences, caregiver interactions, and socio-cultural context, also critically shape neuroplasticity. For instance, environmental enrichment through paired housing during sensitive developmental windows has been shown to ameliorate autistic-like behaviors and restore GABAergic system function in animal models, highlighting the importance of timing and context in intervention design [112]. Furthermore, dietary patterns and selective eating behaviors characteristic of ASD can indirectly impact neural plasticity by altering gut microbiome composition and metabolic states, which in turn influence brain function [113]. These multifactorial influences underscore the complexity of ASD and the limitations of one-size-fits-all interventions.

Personalized intervention strategies that integrate genetic, phenotypic, and environmental information are essential to optimize neuroplasticity and behavioral improvements. Tailoring interventions to individual profiles—including consideration of sex, symptom severity, sensory processing differences, and environmental context—can enhance engagement, motivation, and therapeutic efficacy [114]. Advances in technology, such as neuroimaging and EEG biomarkers, combined with behavioral assessments, facilitate the identification of subgroups and the monitoring of intervention responses, enabling precision medicine approaches in ASD [108, 109]. While our review has identified multiple predictive factors (baseline cognitive abilities, joint attention skills, symptom severity, genetic variants, sex differences, and neuroimaging markers), constructing an integrated predictive model remains a critical future direction. Multimodal approaches combining structural and functional neuroimaging with machine learning algorithms show promise for patient stratification, with features from functional connectivity patterns, white matter integrity, and cortical morphology demonstrating potential for classification and prediction. We propose that future research should develop composite predictive indices integrating behavioral assessments (ADOS scores, developmental quotients), neuroimaging biomarkers (DMN connectivity, cerebellar-prefrontal circuits, EEG microstate dynamics), and genetic profiles to identify optimal intervention candidates and personalize treatment intensity and modality selection, ultimately enabling precision medicine approaches in ASD intervention. In conclusion, recognizing and addressing individual differences and moderating factors is paramount for designing effective, personalized behavioral interventions that harness neuroplasticity mechanisms to improve outcomes in ASD.

Clinical evidence demonstrates how individual characteristics guide personalized ESDM implementation. Children with higher baseline social cognitive skills, particularly intention to communicate and attention to faces, show more favorable responses to standard protocols [56]. Age- and ability-based adjustments are critical: children under 2 years with lower language age benefit more from individual I-ESDM with 1:1 ratios, whereas those aged 2 + with higher developmental quotients respond effectively to group formats (G-ESDM) [64]. For children with comorbid motor delays and sensory hypersensitivities [110, 111], integrating sensory integration therapies alongside ESDM enhances engagement and reduces barriers to learning [9, 115]. Recognition of symptom severity subgroups with differential neuroanatomical deviations [108, 109] enables stratification for targeted intervention components, with severe repetitive behaviors benefiting from augmented environmental enrichment [112]. These examples underscore the importance of comprehensive baseline assessments encompassing cognitive, language, sensory, and neurobiological profiles to individualize ESDM parameters—intensity, format, and therapeutic targets—thereby optimizing neuroplasticity engagement and maximizing functional outcomes.

Long-term maintenance and sustainability of neuroplastic changes

While immediate neuroplastic changes following ESDM have been documented, longitudinal studies reveal variable persistence of these effects. Longitudinal transdisciplinary approaches demonstrate sustained improvements in symptom severity and adaptive functioning over 18-month follow-ups [94, 95], suggesting that intervention-induced brain connectivity changes may stabilize with continued implementation. However, the field currently lacks comprehensive 2–5 year neuroimaging follow-up data to definitively establish which neural modifications persist versus which require ongoing intervention to maintain. Preliminary evidence suggests that normalization of fronto-temporal connectivity and DMN integration correlates with sustained social communication gains, whereas executive control network enhancements may require periodic intervention reinforcement. The duration of continuous intervention appears critical, with studies indicating that at least 12–18 months of consistent ESDM delivery is necessary to consolidate neuroplastic adaptations that support long-term functional improvements [94, 95]. Future research employing longitudinal multimodal neuroimaging combined with extended behavioral assessments is essential to map the temporal dynamics of neural plasticity maintenance and identify biomarkers predicting sustained therapeutic response versus relapse trajectories.

Comprehensive analysis of neural plasticity mechanisms promoting behavioral improvement

Correlation between enhanced brain connectivity and behavioral performance

The restoration and enhancement of brain network connectivity in individuals with ASD have been increasingly recognized as crucial mechanisms underlying improvements in core behavioral domains such as social interaction, language abilities, and adaptive behaviors. Functional connectivity (FC) studies reveal that atypical connectivity patterns, particularly involving social brain networks, DMN, and cortico-striatal circuits, are closely linked to the severity of autistic symptoms and behavioral impairments. For example, reduced FC strength between social-related brain regions and motor and cingulate areas correlates negatively with communication and social interaction deficits, as measured by standardized scales like ADOS and SRS, indicating that weakened connectivity contributes directly to social dysfunction in ASD children [83]. Similarly, longitudinal analyses demonstrate that improvements in RRBIs with age are associated with specific alterations in FC, such as decreased connectivity between the left superior occipital lobe and right angular gyrus, and between the left insula and left caudate nucleus, suggesting that normalization of these connections supports behavioral amelioration [84].

Moreover, intrinsic structural connectivity within the DMN has been shown to correlate with executive function and social skills in youth with ASD, where impaired anterior–posterior DMN connectivity is linked to social regulation difficulties [116]. This relationship underscores the importance of integrated network functioning for complex social behaviors. In addition, atypical developmental trajectories of cortico-striatal connectivity, with increasing connectivity in ASD contrasting with stable or decreasing patterns in typically developing individuals, are negatively correlated with social-communication deficits and RRBIs, highlighting the role of subcortical-cortical interactions in symptom expression and potential recovery [117].

Interventional studies further support the direct link between enhanced brain connectivity and behavioral improvements. For instance, DBS targeting the basolateral amygdala in a Tbr1-deficient mouse model of autism improves social behaviors and restores amygdalar connectivity and whole-brain synchronization, indicating that modulation of specific circuits can lead to functional and behavioral recovery [118]. Non-invasive neuromodulation techniques such as tDCS applied to frontal or cerebellar regions have also demonstrated efficacy in enhancing social functioning and modulating aberrant connectivity patterns in ASD individuals [119, 120]. Similarly, exercise interventions have been shown to improve repetitive behaviors in children with ASD, with associated changes in dynamic connectivity within the triple network model encompassing default mode, salience, and executive control networks [121]. These findings collectively suggest that behavioral gains are mediated by normalization or compensation within distributed brain networks.

At the neurophysiological level, studies using EEG and fNIRS reveal that altered connectivity dynamics and brain state transitions correlate with symptom severity and cognitive performance in ASD children, with adaptive behavior mediating the relationship between neural activity and cognition [122, 123]. Machine learning approaches integrating high-order spatiotemporal features of dynamic functional connectivity have achieved high accuracy in classifying ASD and have identified critical brain networks whose connectivity patterns correlate with clinical measures such as ADOS scores, further substantiating the behavioral relevance of connectivity alterations [42, 124].

In summary, converging evidence from neuroimaging, neurophysiological, and interventional studies robustly supports a direct correlation between enhanced brain connectivity and improvements in social behavior, language abilities, and adaptive functioning in ASD. These findings emphasize the importance of targeting specific brain networks to promote neuroplasticity and behavioral recovery, and they provide a neurobiological framework for developing precision interventions aimed at modulating brain connectivity to ameliorate core symptoms of ASD.

Neural circuit remodeling mechanisms

Neural circuit remodeling is a fundamental process underlying the maturation and refinement of brain connectivity, which is critical for the development of functional neural networks and behavioral outcomes. This remodeling involves synaptic plasticity, neuronal reorganization, and large-scale neural network reconstruction, all of which contribute to the optimization of circuit performance during development and in response to experience. Synaptic plasticity, the ability of synapses to strengthen or weaken over time, is central to this process. It enables the pruning of redundant or inappropriate synaptic connections and the strengthening of relevant ones, thereby refining neural circuits to support efficient information processing. Cell adhesion molecules (CAMs), such as those from the L1 and NCAM families, play crucial roles in synaptic pruning and stabilization by mediating interactions at pre- and postsynaptic membranes and within the extracellular matrix. These molecules regulate synaptic remodeling by promoting pruning of inactive dendritic spines and reinforcing active synapses, processes that are essential during critical periods of plasticity and are often disrupted in neurodevelopmental disorders like ASD and schizophrenia [89]. Neuronal remodeling also involves the retraction and regrowth of neurites, a process tightly controlled by molecular signals such as the DnaJ-like-2 (Droj2) protein and GTP-binding proteins like Arf102F, which regulate dendrite pruning through modulation of adhesion molecules like Neuroglian. This precise spatiotemporal control ensures correct circuit wiring and is vital for sensory neuron development, as demonstrated in Drosophila models [125]. Additionally, glial cells, including microglia and astrocytes, contribute significantly to circuit remodeling by engulfing and pruning synapses during development. Microglial receptors such as GPR56 mediate synaptic refinement through recognition of phosphatidylserine on presynaptic elements, facilitating synapse elimination in a circuit- and time-dependent manner. Disruption of these glial functions impairs synaptic pruning and is implicated in neurodevelopmental pathologies [126]. Moreover, microglial remodeling is influenced by the gut microbiota, which modulates microglial function and neuronal arborization, thereby affecting social behavior and neural circuit maturation in vertebrate models [127]. The remodeling of neural circuits is also regulated by transcriptional control mechanisms and synaptic activity, which coordinate neuron-intrinsic and extrinsic molecular pathways to establish mature circuitry. Studies in invertebrate models like Caenorhabditis elegans and Drosophila melanogaster have provided mechanistic insights into how these processes are orchestrated, highlighting conserved pathways that govern circuit remodeling across species [128]. Taken together, these findings underscore the multifaceted nature of neural circuit remodeling, involving synaptic plasticity, neurite pruning, glial-mediated synapse elimination, and molecular signaling pathways that collectively shape the functional architecture of the brain during development and in response to behavioral interventions.

The neuroplasticity mechanisms underlying ESDM intervention effects in ASD operate at multiple levels (Fig. 2). At the molecular level, ESDM activates the BDNF-TrkB signaling cascade, triggering downstream RAS-ERK and PI3K-AKT-mTOR pathways that enhance synaptic plasticity and upregulate neuroplasticity-related genes. These molecular changes drive large-scale brain network remodeling, including enhanced DMN connectivity (medial prefrontal cortex, posterior cingulate cortex, temporoparietal junction), normalization of the social cognition network (STS (Superior temporal sulcus) region, fusiform face area, amygdala), optimization of the executive control network (dorsolateral prefrontal cortex, parietal cortex), and improvement of the cortico-striatal circuit (prefrontal cortex, striatum, thalamus).

Fig. 2.

Neuroplasticity mechanisms underlying behavioral improvements in ASD following ESDM intervention. Left panel: Molecular pathway of synaptic plasticity enhancement. ESDM intervention activates BDNF-TrkB signaling, triggering downstream RAS-MEX-ERK and PI3K-AKT-mTOR pathways that modulate glutamate/GABA balance and upregulate synaptic proteins and neuroplasticity-related genes. Right panel: Brain network remodeling across four levels: (1) enhanced DMN connectivity (medial prefrontal cortex, posterior cingulate cortex, temporoparietal junction); (2) normalized social cognition network (STS region, fusiform face area, amygdala); (3) optimized executive control network (dorsolateral prefrontal cortex, parietal cortex); (4) improved cortico-striatal circuit (prefrontal cortex, striatum, thalamus). Arrows indicate the progression from molecular mechanisms to large-scale network reorganization, contributing to behavioral improvements in social communication and cognitive function

Emerging evidence establishes direct links between molecular changes and network reorganization following NDBI interventions. rTMS in children with ASD increases cerebellar gray matter volume while simultaneously enhancing functional connectivity between fusiform gyrus, temporal-frontal cortices, and precuneus, with both structural and functional changes correlating with behavioral improvements [75]. Similarly, tDCS normalizes frontal-striato-hippocampal connectivity alongside restoring sociability through modulation of glutamatergic and GABAergic neurotransmission [80]. These findings indicate that ESDM-induced activation of BDNF-TrkB signaling and downstream PI3K-AKT-mTOR and RAS-ERK pathways drives synaptic protein upregulation [8, 13], which enables large-scale network reorganization—enhanced DMN connectivity [116], normalized social cognition network function [83], and optimized executive control circuits [121]—that directly underlies behavioral gains in social communication and adaptive functioning. This molecular-to-network cascade provides mechanistic coherence linking intervention-induced neuroplasticity at multiple levels to observable clinical improvements.

Neurotransmitter systems, particularly those involving dopamine and glutamate, are integral to modulating the efficacy and plasticity of neural circuits during remodeling. Dopamine acts as a neuromodulator that influences synaptic strength and plasticity, thereby affecting learning and behavior. Glutamate, the primary excitatory neurotransmitter, mediates synaptic transmission and plasticity through its receptors, which are critical for experience-dependent remodeling during developmental critical periods. For example, in the Drosophila olfactory system, glutamatergic interneurons modulate the remodeling of olfactory sensory neuron (OSN) innervation, with glutamate signaling influencing synaptic refinement in response to odorant experience. Disruption of fragile X mental retardation protein (FMRP), a key regulator of synaptic function, alters glutamatergic and GABAergic signaling balance, impairing critical period remodeling and synaptic plasticity. This highlights the importance of neurotransmitter systems in regulating neuron-specific remodeling processes and circuit homeostasis [129]. Furthermore, dopamine signaling has been implicated in the regulation of insulin receptor activation in glial cells, which facilitates neuron-glia communication necessary for developmental clearance of transient neurons, a process essential for circuit refinement [130]. The interplay between neurotransmitter systems and molecular signaling pathways thus orchestrates synaptic remodeling and neural network reconfiguration, enabling adaptive behavioral improvements following interventions. Understanding these mechanisms provides a foundation for developing targeted therapies aimed at enhancing neuroplasticity and functional recovery in neurodevelopmental disorders such as autism.

Intervention-promoted neural functional integration

Intervention strategies in ASD have increasingly focused on enhancing neural functional integration across brain regions, aiming to improve information processing efficiency that is often disrupted in ASD. Functional integration refers to the coordinated activity and communication between distributed brain areas, which is critical for complex cognitive and social functions impaired in ASD. Evidence from neuroimaging studies demonstrates that individuals with ASD exhibit atypical functional connectivity patterns, including both hypo- and hyperconnectivity among sensory, attentional, and higher-order cognitive networks, which correlate with core symptoms such as social deficits and repetitive behaviors [71]. Behavioral interventions and neuromodulation techniques, such as rTMS and tDCS, have been shown to modulate these connectivity patterns, promoting more normalized functional integration. For instance, rTMS targeting prefrontal regions enhances connectivity between frontal, temporal, and occipital areas, concomitant with improvements in social cognition and behavior [12, 75]. Similarly, tDCS applied over the left Broca’s area combined with cognitive behavioral training improves pragmatic language skills by modulating functional networks involved in social communication [131]. These interventions appear to facilitate synchronization and coordination across brain networks, thereby enhancing the efficiency of information processing. Moreover, neurofeedback training targeting beta rhythms has demonstrated increased functional connectivity and reduced variability in brain networks, which parallels behavioral gains in communication and social domains [79]. At the neurophysiological level, alterations in neurotransmitter systems, including glutamate and GABA, contribute to atypical sensory processing and connectivity in ASD, and interventions that restore excitatory/inhibitory balance can promote functional integration [9]. Additionally, neuroplasticity-based sensory integration therapies have been shown to enhance executive function and balance by increasing activation in prefrontal regions, further supporting improved network integration [115]. Collectively, these findings underscore that interventions promoting cross-regional functional integration can remediate neural circuit dysfunctions underlying ASD symptoms, thereby improving cognitive and social processing efficiency. Future research should focus on optimizing intervention parameters and combining behavioral and neuromodulatory approaches to maximize neural integration and functional outcomes.

Neural oscillatory synchronization plays a pivotal role in facilitating functional integration by temporally coordinating neuronal activity across distributed brain regions. In ASD, disruptions in neural oscillations, particularly in the theta (4–8 Hz) and alpha frequency bands, have been implicated in deficits in social cognition and sensory processing. For example, children with ASD exhibit altered EEG microstate dynamics and reduced theta-band functional connectivity between fronto-parietal and occipito-temporal regions during spatial navigation tasks, reflecting impaired integration of spatial and cognitive information [132]. These oscillatory abnormalities contribute to inefficient information transfer and impaired multisensory integration, which are critical for social communication. Interventions targeting neural oscillations, such as neurofeedback and noninvasive brain stimulation, can enhance oscillatory synchronization and thereby improve social cognitive functions. Neurofeedback training has been shown to increase functional connectivity and reduce temporal variability in brain networks, associated with improvements in social and communication behaviors [79]. Similarly, tDCS and rTMS protocols modulate oscillatory activity and connectivity, facilitating synchronization within and between key networks involved in social cognition, such as the DMN and frontoparietal control network [12, 133]. The enhancement of neural oscillatory synchrony by these interventions likely underpins their efficacy in ameliorating social deficits in ASD. Moreover, dynamic brain state analyses using fNIRS reveal that children with ASD spend less time in specific brain states characterized by synchronized activity, and these alterations correlate with symptom severity and cognitive performance [122]. These findings highlight the critical role of neural oscillation synchronization in social cognition and suggest that interventions promoting oscillatory synchrony can restore functional integration and improve behavioral outcomes in ASD. Continued investigation into oscillatory mechanisms and their modulation will inform the development of targeted therapies to enhance social cognitive processing through improved neural synchronization.

Future research directions and technological applications

The future of ASD intervention research is poised to benefit significantly from the integration of advanced neuroimaging modalities and machine learning techniques, which together promise more precise and individualized assessments of treatment efficacy. Multimodal neuroimaging, combining structural and functional data such as resting-state fMRI, diffusion tensor imaging, and arterial spin labeling, allows for a comprehensive characterization of brain connectivity alterations underlying ASD. For instance, studies have demonstrated altered functional and structural brain networks in ASD, particularly involving the frontal cortex and amygdala-connected regions, which correlate with symptom severity and social functioning deficits [134–136]. The application of machine learning algorithms, including support vector machines and deep learning autoencoders, to these multimodal datasets enables the extraction of complex patterns and biomarkers that can distinguish ASD individuals from controls with high accuracy, facilitating early diagnosis and monitoring of intervention outcomes [136, 137]. Furthermore, artificial intelligence (AI) approaches have been employed to analyze behavioral and neuroimaging data, offering objective quantification of treatment responses and the potential for personalized intervention strategies [138, 139]. The convergence of these technologies can enhance the sensitivity and specificity of detecting neuroplastic changes induced by behavioral interventions, moving toward precision medicine in ASD.

In addition to neuroimaging and AI, emerging gene-editing and neuromodulation techniques hold promise for synergistic integration with behavioral therapies. Advances in understanding the genetic underpinnings of ASD, including the identification of brain connectivity-associated genes and gene-environment interactions such as those involving preeclampsia, suggest that targeted genetic interventions could modulate neurodevelopmental trajectories [140, 141]. Gene editing technologies like CRISPR-Cas9 may, in the future, allow correction or modulation of pathogenic variants contributing to ASD phenotypes, although ethical and safety considerations remain paramount. Concurrently, neuromodulation methods such as tDCS have been shown to modulate brain functional connectivity and network flexibility in ASD, potentially enhancing the efficacy of behavioral interventions by facilitating neuroplasticity [142]. The integration of neuromodulation with behavioral therapies could optimize intervention timing and individual responsiveness, especially when guided by neuroimaging biomarkers.

Moreover, the application of digital and assistive technologies—including virtual reality (VR), augmented reality (AR), robotics, and mobile cueing systems—offers innovative platforms to deliver and augment behavioral interventions. VR and XR technologies provide immersive, controlled environments for social and cognitive skill training, with evidence supporting improvements in social communication and engagement [143, 144]. Robotics and AI-driven social robots have shown potential in enhancing joint attention and social functioning in children with ASD [145, 146]. Mobile applications incorporating reinforcement learning algorithms can personalize motivator selection to optimize behavioral therapy outcomes [147]. These technologies enable scalable, accessible, and engaging interventions that can be tailored to individual needs and delivered beyond clinical settings, including at home and in educational environments [148, 149].

Future research should prioritize longitudinal studies employing multimodal neuroimaging combined with machine learning to elucidate the dynamic neuroplastic changes associated with various intervention modalities. Investigations into the ethical, practical, and clinical implications of integrating gene-editing and neuromodulation with behavioral therapies are essential. Additionally, the development and rigorous evaluation of technology-assisted interventions should emphasize user-centered design, inclusivity, and real-world applicability. Ultimately, the convergence of these multidisciplinary approaches promises to advance the precision, efficacy, and accessibility of ASD interventions, improving long-term outcomes and quality of life for individuals on the spectrum.

Among these multidisciplinary approaches, the most urgent clinical priority is establishing multi-center longitudinal studies with harmonized protocols to identify baseline neurobiological predictors of ESDM response, as this directly addresses the critical gap in precision medicine implementation. Technologies closest to clinical translation include multimodal neuroimaging biomarker identification [134–137] and non-invasive neuromodulation techniques (tDCS/rTMS) that have demonstrated preliminary efficacy in enhancing behavioral intervention outcomes [133, 142]. These approaches require standardized fidelity measures, site-effect minimization through domain adaptation techniques [78], and shared data elements including resting-state fMRI, EEG metrics, and molecular biomarkers such as BDNF levels to enable robust cross-cohort validation. In contrast, emerging technologies including gene-editing interventions and AI-driven social robots, while promising, necessitate extensive foundational research addressing mechanistic understanding, ethical frameworks, and safety validation before clinical implementation [140–146]. Multi-center study designs should prioritize common assessment batteries, coordinated intervention protocols across diverse cultural and socioeconomic contexts, and data-sharing infrastructures that facilitate meta-analyses and reproducibility, thereby accelerating the translation of neuroplasticity-based insights into evidence-based, personalized clinical practice.

To translate AI-driven neuroimaging biomarkers into clinical decision-making systems, a systematic implementation framework is essential. The proposed workflow comprises four sequential phases: (1) Data collection and harmonization: Multi-center consortia should adopt standardized imaging protocols (resting-state fMRI with 10-min acquisition, DTI with ≥ 30 directions, EEG with 64-channel montage) alongside common clinical batteries (ADOS-2, Vineland-3, developmental quotients), implementing domain adaptation techniques to minimize site effects [78]. (2) Model training and validation: Machine learning pipelines integrating structural, functional, and network features should be trained on discovery cohorts (n ≥ 500) with cross-validation, then externally validated on independent multi-site cohorts to establish reproducibility [77, 136, 137]. Feature importance analyses should identify key connectivity patterns (e.g., DMN anterior–posterior coupling [116], cerebellar-prefrontal connectivity [75]) as candidate biomarkers. (3) Clinical validation: Prospective longitudinal studies should assess whether baseline biomarker profiles predict ESDM response trajectories at 6-, 12-, and 18-month follow-ups, establishing sensitivity/specificity thresholds for clinical utility [134, 135]. (4) Clinical deployment: Cloud-based decision-support platforms could integrate automated preprocessing pipelines, biomarker extraction algorithms, and evidence-based recommendation engines to assist clinicians in stratifying patients into high-, moderate-, or low-likelihood ESDM responders, thereby informing personalized intensity and modality selection. Pilot implementation in 3–5 autism centers with iterative refinement based on clinician feedback, computational efficiency, and outcome tracking would establish feasibility before broader dissemination. This operationalized framework bridges the translational gap between neuroplasticity research and precision clinical practice.

Study limitations

Several methodological limitations warrant acknowledgment in interpreting the current evidence base. First, existing neuroimaging studies investigating ESDM and related NDBIs predominantly involve small sample sizes, typically ranging from 20 to 60 participants, which limits statistical power and may compromise the stability, reproducibility, and generalizability of reported connectivity patterns and neuroplastic changes [75, 79, 118]. Second, although ESDM has been implemented across diverse cultural contexts (e.g., Japan, Italy, Taiwan, and Senegal) [53, 57–59], systematic examination of intervention-effect heterogeneity across populations remains insufficient. Cultural variation in parent–child interaction styles, educational expectations, and service delivery infrastructures may moderate treatment outcomes, yet comparative cross-cultural efficacy data are scarce. Third, most studies employ short to medium-term assessment windows, typically spanning 6–18 months [54, 55, 62], precluding definitive conclusions regarding the durability and long-term maintenance of intervention-induced neuroplastic changes and behavioral gains; longitudinal follow-up extending into school age and adolescence is needed to determine whether early ESDM-driven network reorganization translates into sustained adaptive trajectories. In addition, heterogeneity in neuroimaging protocols, connectivity metrics, and clinical assessment instruments across studies hinders direct meta-analytic synthesis and quantitative comparison of effect sizes. Finally, the evidence base may be affected by publication bias (the “file drawer” problem), whereby studies reporting positive effects are more likely to be published than null or negative findings, potentially inflating perceived efficacy and obscuring circumstances under which ESDM is less effective. Consistent with this concern, ESDM responsiveness appears to vary substantially across individuals, with some children showing minimal gains despite intensive intervention [7, 56]. Potential neurobiological contributors to limited response include severe baseline connectivity disruptions that exceed the capacity for experience-dependent reorganization, reduced neuroplasticity reserves due to genetic or metabolic constraints, and co-occurring conditions (e.g., severe intellectual disability or epilepsy) that may constrain neural remodeling [105, 108]. Moreover, ASD heterogeneity suggests that certain neurobiological subtypes may require alternative or complementary strategies beyond standard NDBI approaches. Collectively, these limitations underscore the need for future large-scale, standardized, longitudinal, and cross-cultural research with transparent reporting (including null results), rigorous pre-registration, and systematic investigation of moderators and mediators of treatment non-response to robustly establish mechanisms and optimize precision intervention matching.

Conclusion

Early intervention during early developmental windows with NDBIs, particularly ESDM, offers the strongest opportunity to harness neuroplasticity and remodel brain networks supporting social cognition and executive function, thereby improving social communication and adaptive functioning in ASD. Given the pronounced heterogeneity of ASD, intervention planning should be personalized using comprehensive neurobiological and developmental profiles—including baseline functional connectivity patterns, symptom severity, and comorbidities—to optimize timing, intensity, format, and therapeutic targets. Future progress will depend on integrating multimodal neuroimaging biomarkers with machine learning–enabled prediction of treatment response within standardized, harmonized multi-center pipelines. Addressing current limitations through large-scale longitudinal studies with transparent reporting will accelerate scalable precision-medicine approaches and improve long-term developmental outcomes for individuals with autism.

Author contributions

Qian Tao: Conceptualization, Supervision, Project administration; Xianhua Shao: Conceptualization, Supervision, Investigation, Resources; Pan Yan: Methodology; Formal analysis, Data curation, Writing—original draft, Writing—review and editing, Project administration; Xiuqin Zhu: Methodology, Data curation, Writing—review and editing, Project administration. All authors provided critical feedback and helped shape the research, analysis, and manuscript. All data were generated in-house, and no paper mill was used. All authors agree to be accountable for all aspects of work ensuring integrity and accuracy.

Availability of data and materials

Not applicable.

Declarations

Ethics approval and consent to participate

Not applicable.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.