脑连接模式对应孤独症青少年的感觉加工类型

从行为分类到脑网络

孤独症谱系的感觉差异常被描述为对声音、触觉、气味或味道过度敏感、迟钝，或主动寻求强烈刺激。此前一项大规模研究已把孤独症青少年的感觉表现归为五种表型：感觉适应型、广泛感觉差异型、以味觉和嗅觉敏感为主的一组、对环境反应偏低但主动寻求强烈感觉的一组，以及伴随运动困难和体力偏低的一组。这些分类来自行为观察，但它们在脑内是否对应不同的信息交流方式，此前并不清楚。

Western University 的 Matthew Kolisnyk 与资深作者 Ryan Stevenson 领衔的团队，试图检验这五种行为表型能否映射到具体的大脑功能连接模式。研究发表在 Biological Psychiatry: Cognitive Neuroscience and Neuroimaging。

研究如何做

团队分析了146名3至21岁孤独症儿童与青少年的数据。 caregivers 填写标准化临床问卷，记录孩子日常对各种感觉刺激的反应，据此把每名参与者归入五种表型之一。随后参与者接受静息态功能磁共振扫描，测量安静不执行任务时血氧水平的变化，通过哪些脑区同步激活与抑制来刻画功能连接。

研究者把大脑划分为200个皮层分区和32个皮层下区域，用图论计算连接强度、信息传递效率、介数中心性和聚类系数等指标，得到每人数百个脑网络数据点。这些指标被输入机器学习模型，检验模型能否仅凭脑连接数据判断孩子属于哪种感觉表型，并对五种表型的全部十组配对逐一比较。

哪些配对能被区分

模型在十组配对中成功区分了七组。表现最好的是两个行为上最极端的组：感觉适应型与广泛感觉差异型。感觉适应型儿童在前额叶皮层表现出更强的全局连接，这一区域主要参与决策与执行功能；眶额皮层连接也较高，该区域靠近双眼上方，参与把感觉信息与情绪奖赏整合起来。

广泛感觉差异型儿童则在躯体运动网络内部表现出更强的局部连接，这一网络直接处理触觉与自主身体运动。他们的初级感觉加工区域局部高度互联，但与额叶前部的通信连接较弱。研究者据此提出，这可能反映高层执行区域对来自外界的感觉信号进行调节的方式存在差异。

更专门化的表型也有各自的连接特征。以味觉和嗅觉敏感为主的一组，在后顶叶皮层表现出枢纽活动的改变，这一区域与空间推理和注意定向有关。广泛注意网络与默认模式网络的差异，也帮助算法区分了若干表型。皮层下结构对区分特定感觉组别尤其重要，这些深部结构包括丘脑和伏隔核，常在感觉信息到达皮层前充当第一道关口；它们在广泛感觉差异组和味觉嗅觉敏感组中连接模式尤为独特。

解释边界

研究有几处限制需要紧贴结论来看。机器学习模型在三组配对中未能给出统计显著的区分，例如反应偏低组与运动困难组之间。这可能意味着部分感觉表型共享重叠的神经机制，也可能需要更大样本才能检出更细微的脑差异。

感觉行为的分类主要依赖照护者填写的问卷。这类问卷是标准临床工具，但带有一定主观性，未必完全反映孤独症参与者自身的内部体验。分析也未纳入整体孤独症症状严重度、智商或共病情况，这些因素可能独立影响脑连接模式。此外，扫描时头部小幅移动会人为改变功能连接测量；研究者剔除了移动过大的数据并做了数学校正，但不同感觉组儿童对扫描环境的耐受程度若有系统差异，仍可能是潜在影响因素。

研究者指出，后续需要在独立的孤独症群体中检验这些预测模型，确认模式能否在更广泛人群中成立。

A new study reveals that the wide variety of sensory experiences reported by autistic individuals corresponds to distinct patterns of information exchange in the brain. Researchers found that specific functional connections between cortical and subcortical brain regions can predict an autistic child’s sensory profile. The findings were published in Biological Psychiatry: Cognitive Neuroscience and Neuroimaging.

Differences in sensory processing represent a defining feature for up to 87 percent of autistic individuals. These differences mean that sights, sounds, textures, smells, or tastes might be perceived with unusually high or low intensity. Some individuals actively seek out certain types of sensory input, while others go to great lengths to avoid everyday stimuli. These variations affect how people navigate their environments and can influence cognitive development and daily functioning.

While these sensory variations can seem highly individualized, previous research has shown that they tend to cluster into specific behavioral categories. In a prior large-scale study, scientists identified five distinct sensory phenotypes among autistic youth. The first phenotype is “sensory adaptive,” describing children who show minimal sensory differences from neurotypical peers. The second is “generalized sensory differences,” which involves severe sensitivities across nearly all sensory domains.

The remaining three phenotypes involve more specialized sensory responses. One group is characterized primarily by selective sensitivities to taste and smell. Another group features children who are generally underresponsive to their environment but actively seek out intense sensations. The final group is defined by movement difficulties combined with low physical energy. Until now, it was unknown exactly how these five behavioral profiles mapped onto the brain’s internal wiring.

Western University researcher Matthew Kolisnyk and senior author Ryan Stevenson led a team to investigate whether these sensory profiles arise from specific patterns of brain activity. They designed an analysis to determine if different sensory behaviors correspond to unique functional connectivity configurations in the brain. Uncovering these neural foundations could help explain the immense biological diversity seen across the autism spectrum.

To test this idea, the research team analyzed data from 146 autistic children and adolescents between the ages of three and twenty-one. The participants’ caregivers filled out a standardized clinical questionnaire detailing the children’s daily responses to various sensory stimuli. Based on the scores from these questionnaires, the researchers sorted each participant into one of the five previously established sensory phenotypes.

The participants then underwent resting-state functional magnetic resonance imaging. This brain scanning technique tracks blood oxygen levels to measure brain activity while a person rests quietly without performing a specific task. By observing which areas of the brain activate and deactivate in sync over time, researchers can map out the brain’s functional connectivity. This connectivity represents how different neural regions communicate and share information.

The researchers divided the brain scans into 200 distinct sections of the cerebral cortex and 32 subcortical regions located deep within the brain. They then applied a mathematical framework known as graph theory to quantify how information moves through this sprawling biological network. Graph theory provides specific metrics, such as how strongly regions are connected to one another and how efficiently they pass information to local neighbors.

The team also used graph theory to calculate betweenness centrality, a metric that identifies whether certain brain regions act as central hubs for long-distance communication. Another metric, called the clustering coefficient, measured how highly interconnected a region’s immediate neighbors are with one another. Together, these mathematical calculations generated hundreds of individual data points describing each participant’s unique brain network.

Finally, the team fed all these graph theory measurements into a machine learning algorithm. Machine learning involves training a computer model to recognize hidden patterns within massive datasets. The researchers tested whether the model could learn the connectivity data well enough to correctly identify which sensory phenotype a child belonged to, running comparisons between all ten possible pairs of the five groups.

The algorithm successfully differentiated between seven of the ten pairs of sensory phenotypes based purely on their brain connectivity metrics. The model performed best when comparing the two most extreme behavioral profiles. It most accurately distinguished the sensory adaptive group from the group with generalized sensory differences.

Specific brain networks emerged as strong predictors of a child’s sensory profile. Children in the sensory adaptive group showed stronger global connections in the prefrontal cortex, a region primarily involved in decision making and executive function. They also showed high connectivity in the orbitofrontal cortex, an area just above the eyes that helps integrate sensory information with emotional rewards.

Conversely, children with generalized sensory differences exhibited stronger local connections within the somatomotor network, which directly handles touch and voluntary physical movement. Their primary sensory processing regions appeared highly interconnected on a local level but lacked strong communication links to the prefrontal cortex. This pattern points to a potential difference in how higher-level executive brain regions modulate incoming sensory signals from the outside world.

The analysis also identified distinct connectivity signatures for the more specialized sensory profiles. Children characterized by isolated taste and smell sensitivities consistently showed altered hub activity in the posterior parietal cortex, a region important for spatial reasoning and directing attention. Differences in broad attentional networks and the default mode network, a system active during internal thought and daydreaming, also helped the algorithm separate several phenotypes.

Subcortical regions proved highly relevant for distinguishing specific sensory groups. These deep brain structures, which include areas like the thalamus and the nucleus accumbens, often act as the initial gatekeepers for sensory information before it reaches the cerebral cortex. The study found that subcortical connectivity patterns were particularly distinct in the generalized sensory differences group and the taste and smell sensitivity group.

The study carries several caveats regarding its interpretation. The machine learning model failed to produce statistically significant distinctions between three of the phenotype pairs, such as the underresponsive group and the movement difficulty group. This lack of differentiation suggests that some sensory profiles might share overlapping neural mechanisms, or that a larger sample size is needed to detect subtler brain differences.

The analysis relied heavily on caregiver reports to categorize the children’s sensory behaviors. While these questionnaires are standard clinical tools, they introduce a degree of subjectivity that might not perfectly capture the internal experiences of the autistic participants. The study also did not account for variations in overall autism symptom severity, intelligence quotient, or co-occurring conditions, which might independently influence brain connectivity patterns.

Small head movements during brain scans can artificially alter functional connectivity measurements. While the researchers removed data with excessive motion and applied mathematical corrections, systematic differences in how well children from different sensory groups tolerate the scanner environment remain a potential factor. Future research will need to test these predictive models on separate groups of autistic individuals to verify that the patterns hold true across broader populations.

The study, “Decoding the Neural Basis of Sensory Phenotypes in Autism,” was authored by Matthew Kolisnyk, Kathleen Lyons, Eun Jung Choi, Marlee M. Vandewouw, Bobby Stojanoski, Evdokia Anagnostou, Azadeh Kushki, Rob Nicolson, Elizabeth Kelley, Stelios Georgiades, Jason Lerch, Jennifer Crosbie, Russell Schachar, Muhammad Ayub, Jessica Jones, Paul Arnold, Xudong Liu, and Ryan Stevenson.