Mapping the landscape of brain functional dynamics in autism spectrum disorder

自闭症谱系障碍（ASD）是一种日益普遍的神经发育疾病，包括广泛的表型和异质性疾病，如自闭症和阿斯伯格综合征。ASD患者在一生中都面临着各种各样的功能挑战，如社会认知、语言技能、执行功能和运动能力方面的困难[1]-[3]。他们的行为和大脑功能与典型发育个体（TD）的差异是由神经结构和大脑活动的非典型发育所支撑的。例如，尾状核体积的增加被发现与复杂的重复性运动行为有关。非典型脑活动在与心理理论、社会认知和执行功能[5]-[7]相关的脑区被广泛观察到。因此，了解ASD的神经生物学基础对于提高认识和提供更有针对性的干预措施至关重要。最近，人们认为ASD是一种导致信息处理改变的大脑连接障碍。在脑结构连接方面，有研究表明ASD患者白质连接的改变与功能障碍[8]，[9]有关。同时，有充分的证据表明，ASD[10]，[11]中存在广泛的功能连接障碍。例如，在形成社会功能的神经生物学基础的伏隔核（NAcc）与其他大脑区域（如丘脑和前扣带皮层[12]）之间发现了连接不足。然而，尽管大量且快速增长的FC研究，关于FC改变的结果，似乎受到ASD特征异质性的困扰，仍然不一致[10]。此外，大多数评估ASD中FC的研究都是在FC保持静态的前提下进行的。静态FC分析无法捕捉到网络或区域之间功能连通性的时间分辨转换，而以往研究的“平均”结果可能掩盖了动态的重要差异。动态FC （Dynamic FC, dFC）方法关注的是基于RS或任务的fMRI过程中功能连接的瞬时变化，近年来越来越受到关注。脑电图（EEG）和功能磁共振成像（fMRI）联合研究表明，dFC[14]和[15]有坚实的生物学起源，最近的报道表明，动态功能连接是探索ASD神经生物学基础的良好测量方法。表征dFC的指标，如居住时间和时间变异性，被认为与ASD的功能损伤显著相关。研究dFC也可以验证静态FC发现的非典型脑连接。例如，Fu等人发现下丘脑/下丘脑和感觉区域之间的dFNC增加，这扩展了先前基于静态FC分析的丘脑和感觉皮层之间超连接的发现[18]。通过将动态FC聚类成一组离散状态，研究人员试图识别以非典型动态模式为特征的脑部疾病。Hyatt和他们的同事已经确定了四种患有ASD的dFC状态，与TD相比，它们的居住时间不同。他们还发现dFC测量与ASD患者的社会认知能力得分之间存在显著关联。此外，通过dFC，机器学习已被用于诊断ASD。一项基于自闭症脑成像数据交换（Autism Brain Imaging Data Exchange，简称ABIDE）数据库的研究，使用支持向量机（SVM）分类器[20]从多层dFC中提取中心矩特征对ASD进行分类，准确率达到83%。作为一个新兴领域，dFC异常在ASD中的证据仍然有限，方法上存在异质性。进一步的研究应谨慎选择分析方法和零模型。

Autism spectrum disorder (ASD) is an increasingly prevalent neurodevelopmental condition, comprising a wide range of phenotypes and heterogeneous conditions, such as autism and Asperger's syndrome. ASD patients have to face various functioning challenges across lifespan, such as difficulties in social cognition, language skills, executive functions, and motor abilities [1]-[3]. Their deviations in behaviours and brain functions from typically developed (TD) individuals are underpinned by the atypical development of neural structures and brain activity. For instance, increased caudate volume was found to be correlated with complex repetitive motor behaviours [4]. Atypical brain activity has been widely observed in brain regions associated with theory of mind, social cognition, and executive functions [5]-[7]. Therefore, it is crucial to understand the neurobiological underpinnings of ASD both to increase awareness and inform options to give more tailored interventions. Recently, it has been posited that ASD is a disorder of brain connectivity leading to altered information processing [5]. Regarding structural brain connectivity, it has been implied that altered white-matter connectivity in ASD is associated with function impairment [8], [9]. Meanwhile, there is ample evidence of widespread functional dysconnectivity in ASD [10], [11]. As an example, underconnectivity was found between nucleus accumbens (NAcc), which forms the neurobiological basis of social functions, and other brain regions, such as thalamus, and anterior cingulate cortex [12]. However, despite the large and fast growing number of FC studies, the results regarding altered FC, plausibly plagued by heterogeneity in traits of ASD, remains inconsistent [13]. In addition, most studies assessing FC in ASD have been conducted based on the premise that FC remained static. Static FC analysis fails to capture the time-resolved transitions of functional connectivity between networks or regions and the 'on average' results from previous studies may mask important differences in dynamics. Dynamic FC (dFC) approaches, which have gained increasing attention recently, focus on the transient changes of functional connectivity during an RS- or task-based fMRI. Joint electroencephalographic (EEG) and fMRI studies have suggested a solid biological origin for dFC [14], [15], and recent reports have suggested dynamic functional connectivity as a good measurement to explore the neurobiological basis of ASD. Metrics characterising dFC, such as dwelling times and temporal variability, have been implied to be significantly associated with function impairment in ASD [16] [17]. Studies investigating dFC can also validate atypical brain connectivity discovered with static FC. For instance, Fu et al. found increased dFNC between Hypothalamus/Subthalamus and Sensory Regions which extended previous findings of hyperconnectivity between thalamus and sensory cortex based on static FC analysis [18]. By clustering dynamic FC into a discrete set of states, researchers have attempted to identify brain disorders characterised by their atypical dynamic patterns. Hyatt and their colleagues have identified four dFC states with ASD showing different dwelling times compared with TD [19]. They also found significant associations between dFC measures and social cognitive ability scores in ASD. Furthermore, with dFC, machine learning has been implemented to diagnose ASD. A study based on the Autism Brain Imaging Data Exchange (ABIDE) database reached 83% accuracy to classify ASD using central moment features extracted from multilevel dFC by support vector machines (SVM) classifiers [20]. As an emerging area, evidence for dFC abnormalities in ASD remain limited with heterogeneity in methodology. Further research should be conducted with carefully selected analytic approaches and null models.