Multiple system atrophy (MSA) is defined by striatonigral and olivopontocerebellar degenerations with various combinations of autonomic, extrapyramidal, and cerebellar symptoms (1). MSA has many clinical phenotypes, such as parkinsonian-type MSA (MSA-P) and cerebellar-type MSA (MSA-C). Usually, it has difficulties in differentiating between early-stage MSA-P and idiopathic Parkinson’s disease (PD), because they share similar clinical features (2). Therefore, a effective tool to differentiate MSA-P from idiopathic PD is clinically meaningful.

The well-established semantic fluency test has been frequently used measure in research and clinical practice for assessing the ability to produce as many different words of a specific category (e.g., animal) as possible during one minute (3). Many parameters have been proposed for measuring the performance of this test, for example, the count of correct words of MSA or PD patients was lower than healthy adults (4, 5). Clustering and switching analyses are also common, which defined two key parameters based on the similarity or closeness of word meaning (6). The average count of correct words with high semantic relatedness (e.g., invertebrate animals) is defined as the mean cluster size, and the count of cluster switches. Compared to healthy adults, the PD, Alzheimer’s disease (AD), or progressive supranuclear palsy (PSP) patients exhibited fewer clusters and switches in verbal fluency tests (7, 8). However, this approach obtains limited information and depends primarily upon subjective estimation of cluster division. An automated classification approach conflicted with classic experimenter-based method in evaluating clusters (9).

An objective approach originates from graph theory, which calculates network properties to provide an assessment of network structure and changes. The graph theoretical analyses have been implemented to depict the topological dynamics of brain networks in a variety of neurodegenerative disorders (10, 11). Similarly, this approach have been used to analyze semantic fluency data. Our previous studies found that AD patients’ speech graphs exhibited fewer nodes, higher density, smaller diameter, and smaller average shortest path than PD patients (12), which was same as the comparison between PSP patients and PD patients (13), suggesting that the cognitive decline associated with denser and smaller speech graphs.

This study used graph theoretical analysis to investigate the speech topology of MSA-P patients. All participants were assessed by a well-established semantic fluency test (animal). Their verbal reactions were listed and represented as speech graphs with directionality. A node represents a correct word and an arc represents the temporal connection between sequential words. First, group differences in meta references (metalinguistic reference and metacognitive reference) and graph indices (nodes, density, diameter, and average shortest path) were examined. Second, the effects of levodopa on meta references or graph indices were explored in MSA-P and PD, respectively.

This study applied the graph theoretical analysis for investigating the topology of speech graphs in non-demented patients with MSA-P. Compared to PD patients and healthy adults, MSA-P patients produced denser and smaller speech graphs. Specifically, MSA-P patients’ speech graphs exhibited fewer nodes, higher density, smaller diameter, and smaller average shortest path in comparison to both PD patients and healthy adults. This study confirmed that graph theoretical analyses could differentiate MSA-P and PD, which is similar to a previous study. Šubert et al. (17) used the automated linguistic analysis of natural spontaneous speech and also achieved this goal. Importantly, in MSA-P patients, the positive correlations between levodopa equivalent daily dose and diameter and average shortest path were obtained. Particularly, there was no MSA-P patient who did not receive levodopa (levodopa equivalent daily dose = 0 mg/day). It suggested that higher levodopa doses were associated with producing sparser and larger speech graphs in MSA-P patients.

Additionally, the increased meta-references may be because of deficits in both lexical retrieval and executive control. Prefrontal cortex has long been considered essential for option generation (nodes) and executive functions. A fMRI study showed that the prefrontal cortex activation was related to option production in simple real-world scenarios (18). Enhanced left prefrontal activation has been observed in the word generation within a semantically similar category (19). Moreover, the left inferior frontal gyrus modulate the verbal responses production that satisfy the instruction and inhibit irrelevant words (20). It has been confirmed that executive control is realized by the prefrontal cortex that sends command signals to supervise and control the activities in other cortical and subcortical regions (21).

In addition, the basal ganglia also participated in speech initiation and programming. An electrophysiological research used microelectrode recordings and found that sensori-motor subthalamic neuronal activity was associated with generating speech programming during spoken sentence and syllable-repetition tasks (22). A MRI study investigated the contribution of basal ganglia on verbal fluency and exhibited that the caudate nucleus associated with language generation and the putamen activation improved the switching condition (23).

The speech topological abnormalities in MSA-P patients may result from aberrant connectivity of the frontal-basal ganglionic circuits compared to PD patients (24). A PET study confirmed that the frontal, temporal, parietal, and limbic hypometabolism was associated with cognitive decline in MSA patients (25). Another thin section MR study pointed out that the putaminal signal abnormalities and hyperintensities on proton density were more common in MSA-P patients compared to PD patients (26).

Neurochemical basis of speech graph is still unclear. A model suggested that the dopamine modulating fronto-striatal circuitry might relevant to the persistence and flexibility, respectively (27). To be specific, appropriate dopamine levels in prefrontal cortex promote persistent processes (words generation within a specific category or subcategory) along the mesocortical pathway, while appropriate dopamine levels in striatum accelerate flexible processes (words generation within different categories or subcategories) along the nigrostriatal pathway. Dual-state theory indicated that the prefrontal cortex with concentrated dopamine D1 receptor supports maintenance of item, while the striatum with concentrated dopamine D2 receptor favors flexibility of item (28). A previous study proposed that the levodopa, rather than dopamine agonist pramipexole, could enhance verbal fluency in PD patients (29). There also have conflicting results. For example, Subert et al. (30) found that PD patients who were treated with dopamine agonists showed improvement in sentence length, while who were treated with levodopa had no language amelioration. This study showed that levodopa could improve speech graphs in MSA-P patients. These results may be possibly by activating dopamine D1 receptors (31).

This study has limitations. First, in view of the small sample size of MSA patients with other clinical phenotypes and increasing global homogeneity, we only include patients with MSA-P, thus it’s impossible for this study to differentiate between MSA-P and other clinical phenotypes by graph theory-based analysis of semantic fluency data. Second, although we found the relationship between levodopa equivalent daily dose and graph indices, it cannot be regarded as a causal conclusion. Third, there were very few MSA-P patients took dopamine D2 receptor agonists, thus this study cannot examine the effect of activating dopamine D2 receptor on speech graphs. Future pharmacological researches should explore how antiparkinsonian medications affect speech topology in MSA patients.

This study applied graph theory to demonstrate the topological properties of speech graphs in MSA-P patients, indicating that MSA-P patients’ speech graphs were denser and smaller in comparison to PD patients and healthy adults. Moreover, higher levodopa doses were associated with producing sparser and larger speech graphs in MSA-P patients, indicating that the impaired speech topology can be partially restored by levodopa. Future studies should explore long-term successful treatments to achieve the improvement of speech topology in MSA-P patients.