Increased modularity of the resting‐state network in children with nonsyndromic cleft lip and palate after speech rehabilitation

Abstract Introduction Speech therapy is the primary management followed the physical management through surgery for children with nonsyndromic cleft lip and palate (NSCLP). However, the topological pattern of the resting‐state network after rehabilitation remains poorly understood. We aimed to explore the functional topological pattern of children with NSCLP after speech rehabilitation compared with healthy controls. Methods We examined 28 children with NSCLP after speech rehabilitation (age = 10.0 ± 2.3 years) and 28 healthy controls for resting‐state functional MRI. We calculated functional connections and the degree strength, betweenness centrality, network clustering coefficient (Cp), characteristic path length (Lp), global network efficiency (Eg), local network efficiency (Eloc), modularity index (Q), module number, and participation coefficient for the between‐group differences using two‐sample t tests (corrected p < .05). Additionally, we performed a correlation analysis between the Chinese language clear degree scale (CLCDS) scores and topological properties in children with NSCLP. Results We detected significant between‐group differences in the areas under the curve (AUCs) of degree strength and betweenness centrality in language‐related brain regions. There were no significant between‐group differences in module number, participation coefficient, Cp, Lp, Eg, or Eloc. However, the Q (density: 0.05–0.30) and QAUC (t = 2.46, p = .02) showed significant between‐group differences. Additionally, there was no significant correlation between topological properties of statistical between‐group differences and CLCDS scores. Conclusions Although nodal metric differences existed in the language‐related brain regions, the children with NSCLP after speech rehabilitation had similar global network properties, module numbers, and participation coefficient, but increased modularity. Our results suggested that children with NSCLP achieved speech rehabilitation through function specialization in the language‐related brain regions. The resting‐state topology pattern could be of substantive neurobiological importance and potential imaging biomarkers for speech rehabilitation.


| INTRODUC TI ON
Cleft lip and palate (CLP) is one of the most common craniofacial malformations in infants. It is estimated that the prevalence is 0.1% in live births (Centers for Disease and Prevention, 2006). CLP can be defined as two types, syndromic CLP and nonsyndromic CLP (NSCLP). Syndromic CLP is a portion of a well-known syndrome, while NSCLP is not. Even following successful palatoplasty and pharyngoplasty, the percentage of compensatory articulation errors ranged from 5% to 50% (Taib et al., 2015). Speech therapy is the primary method to correct compensatory articulation errors caused by abnormal articulation placement, including the use of the normal velopharyngeal function, the formation of correct articulation patterns, and consolidation training (Chen, 2012). Speech therapy is usually combined with principles of motor learning by visual, auditory, and touch feedback assistance (Maas et al., 2008). However, the topological pattern after speech rehabilitation in children with NSCLP is poorly understood.
Several neuroimaging studies have identified both structural and functional brain abnormalities in patients with NSCLP after speech rehabilitation. An analysis of adults with NSCLP after articulation rehabilitation found the changes in the cortical thickness, gyrification, and fractal dimensions in the regions involved in language, auditory, pronunciation planning, and execution functions (Li et al., 2020). Another study of adult speech-rehabilitated patients with CLP showed similar functional activation patterns as healthy controls, except for increased activation in the left hippocampus in a subvocalization task functional MRI study (Zhang et al., 2017). Our previous study found lower nodal shortest path length and higher nodal clustering coefficient of brain regions involved in higher-order language and social cognition, and increased small-world index of the whole brain in children with CLP after speech rehabilitation (Rao et al., 2020). Therefore, we performed a further study of the topological organization in the developing functional brain networks of children with CLP after speech rehabilitation.
Graph theory is a specific approach to investigate brain anatomical and functional networks. It has been widely applied in resting-state functional MRI studies (Medaglia, 2017), such as primary progressive aphasia (Mandelli et al., 2016(Mandelli et al., , 2018, adults who stutter (Ghaderi et al., 2018), and healthy people during infancy (Fan et al., 2011) and aging (Wu et al., 2012). Based on resting-state BOLD signals characterizing physiological information of spontaneous neural activities in the brain (Biswal et al., 1995), we assess the local, modular, and global brain networks' characterizations using graph theory. For the local nodal properties, degree strength represents information on communication ability, and betweenness centrality describes the effect on the network's information flow.
Modularity is widely accepted as one of the central organizing principles of the brain network (Bullmore & Sporns, 2009). It presents an optimal measure to balance the opposing requirements put on many changing systems: a great local specialization level, steady global integration, and the adaption of multiple or different selection criteria with time (Bullmore & Sporns, 2009;Kashtan & Alon, 2005). The detection and characterization of modular organization in the brain network can distinguish groups of anatomically and/or functionally related components that conduct specific biological functions. The participation coefficient measures intermodular connections describing a cost-effective network integration (Bertolero et al., 2015).
The global network properties, such as the local network efficiency (Eloc), global network efficiency (Eg), network clustering coefficient (Cp), and characteristic path length (Lp), represent the functional differentiation and integration of the whole-brain network. Network efficiency (Eg and Eloc) is often disrupted by changes in path length (Bassett & Bullmore, 2009), and global cognitive function might depend on long-distance connections (Lp) (Markov et al., 2013). Graph theory provides the ability to explore the local, modular, and global organization of the whole network, fundamentally different from other functional brain network analyses (Medaglia, 2017).
To our knowledge, few studies have explored the topological properties of functional brain networks in children with NSCLP. Therefore, the purpose of this study was to estimate the patterns in the topology of resting-state networks between rehabilitated children with NSCLP and healthy controls using graph theory.

| Participants
The Beijing Children's Hospital and Beijing Stomatological Hospital ethical committee approved this study, and we obtained the informed consent of all children. Twenty-eight children (age = 10.0 ± 2.3 years) with NSCLP and 28 age-and sex-matched healthy controls were recruited from Beijing Children's Hospital and Beijing Stomatological Hospital. All children with CLP had already been evaluated by an experienced medical geneticist to exclude congenital syndromes. All children received a Chinese speech intelligibility test administered by three experienced speech pathologists. The children's inclusion criteria were as follows: (a) aged from 6-16 years old; (b) successful surgery of velopharyngeal insufficiency and speech therapy resting-state topology pattern could be of substantive neurobiological importance and potential imaging biomarkers for speech rehabilitation.

K E Y W O R D S
graph theory, modularity, nonsyndromic cleft lip and palate, resting-state functional MRI, speech therapy (Chinese language clear degree scale (CLCDS) scores > or =86); (c) normal hearing and vision (auditory brainstem response < 30 dB nHL); (d) average intelligence (the scores of Full-Scale Intelligence Quotient (FSIQ) using the Chinese Wechsler Intelligence Scale for Children-IV > 90); (e) Chinese as their mother tongue; and (f) righthanded. The exclusion criteria for the patients were as follows: children with clinic diagnoses of (a) velopharyngeal anatomy or structure defect; (b) speech disorder (CLCDS scores < 86); (c) dysgnosia; (d) hearing and/or vision impairments; (e) congenital disorders; (f) developmental delays; and (g) other chronic health diseases.

| Speech assessment
The CLCDS is a widely used method for the clear evaluation of speech in patients with CLP during speech therapy in China (Chen et al., 2002). One hundred Chinese words were selected that included the 21 consonants and all vowels by daily usage frequency to fill in a table, which contained all error-prone consonants and vowels of patients with CLP. Eighty-six correct phonetic words (or 86 points) were used as the cutoff point for the average level of clear Chinese speech, meeting daily oral communication (Wang et al., 1995). In children with NSCLP after speech rehabilitation, the CLCDS scores were 91.6 ± 4.0, reaching the average level of Chinese daily oral communication (see Table 1).

| Image acquisition
All MRI DICOM data were obtained with a 3.0T GE MRI system (with an eight-channel phased-array head coil) at the Department of Radiology (Beijing Children's Hospital). For each child, highresolution 3D T1-weighted gradient-echo anatomical and restingstate functional MRI data were obtained. Using echo planner imaging (EPI) sequences, functional images were acquired with participants keeping their eyes closed while awake. The scan parameters of EPI were set as follows: repetition time = 2000 ms, echo time = 35 ms, flip angle = 90°, field of view = 240 mm × 240 mm, and in-plane matrix = 64 × 64, which induces the spatial resolution = 3.75 mm × 3.75 mm, slice thickness = 3.0 mm, and slice gap = 1.0 mm. Each scanning session continued for 560 s, and each brain volume contained 38 axial slices. Structural images were acquired using a sagittal T1-weighted three-dimensional spoiled gradient sequence with the following parameters: 164 continuous sagittal slices, echo time = 3.516 ms, repetition time = 8.196 ms, slice thickness = 1 mm, flip angle = 13°, field of view = 256 mm × 256 mm, and matrix = 256 × 256.

| Preprocessing
The obtained DICOM data were processed using the GRETNA toolbox software (http://www.nitrc.org/proje cts/gretn a/) and Consequently, all children were included for further analysis.   (Latora & Marchiori, 2001). Cp is the mean clustering coefficient of the network's nodes, and Lp is defined as the mean shortest path through all pairs of nodes of the network (Watts & Strogatz, 1998). The weighted Eloc and Cp indicate network segregation of the brain, and Eg and Lp imply network integration of the brain.

| Statistical analysis
The chi-square test was applied for the between-group difference in gender, and the two-sample t test was used for between-group difference in age. The whole density range's topological metrics and their areas under the curve (AUCs) were calculated. Age and sex were taken as covariates. The between-group differences in topological parameters based on graph theory were performed for the statistical assessment using a series of two-sample t tests with two-tailed tests.
Furthermore, the relationships between topological parameters and CLCDS scores in patients were computed using Pearson's correlation coefficient. In addition, the Bonferroni correction (corrected p <.05) was applied for multiple comparison corrections.

| RE SULTS
In this study, whole-brain functional networks were constructed, and the local-, modular-, and global-scale properties were evaluated for both children with NSCLP and healthy controls over the density range of 0.05-0.50 (step = 0.01). Interestingly, the local nodal metrics were mostly affected in the language-related brain regions (Fujii et al., 2016). The mean nodal metric values of the two groups were projected onto the cortical surface (see Figure 2).

| Demographic characteristics
The children's demographic characteristics are presented in Table 1.
The children age of our study ranged from 6 to 16 years (10.0 ± 2.3), showing no significant between-group differences (two-sample t tests, t = −0.46, p = .96). The number of males was slightly higher than that of females in children with NSCLP and healthy controls, but the distribution showed no significant differences (chi-square test, χ 2 = 0, p = 1) (see Table 1).

| Between-group differences in nodal degree strength and betweenness centrality
Compared with healthy controls, children with NSCLP showed a higher degree strength AUC in the left middle temporal gyrus of the Wernicke area, and in the right intracalcarine cortex and occipital pole of the primary visual center (see Table 2, Figure 3a).
For nodes with increased betweenness centrality AUC in the children with NSCLP, compared with healthy controls, the central opercular cortex and parietal operculum cortex were located in the right cerebral hemisphere. The rest of the nodes with increased betweenness centrality AUC were all in the left hemisphere and mostly in the dorsal stream of the neural basis of language (Fujii et al., 2016), such as the thalamus, anterior and posterior part of the cingulate gyrus, anterior part of the middle temporal gyrus, orbitofrontal cortex, lateral occipital cortex, and supramarginal gyrus. Only the left posterior part of the middle temporal gyrus showed decreased betweenness centrality (see Table 2, Figure 3b).

| B E T WEEN -G ROUP D IFFEREN CE S IN G LOBAL ME TRI C S
There were no significant between-group differences for C p AUC , L p AUC , Eg AUC , and E loc AUC values of the network or Cp, Lp, Eg, and Eloc values of all threshold networks (see Table 3, Figure 4).

| Between-group differences in modularity
At all threshold values, functional brain networks in children with NSCLP and healthy controls showed typical modular structure properties (Q > 0.3, see Figure 4a). Furthermore, compared with healthy controls, a two-sample two-tailed t test indicated that children with NSCLP exhibited a higher modularity index (density: 0.05-0.30; see

F I G U R E 2
The projection of nodal values onto the cortical surface. a and b indicate the mean degree strength of the children with NSCLP and healthy controls, respectively. c and d indicate the mean betweenness centrality of the children with NSCLP and healthy controls, respectively. The color represents the nodal values

| Relationships between clinical characteristics and topological properties
Among the topological parameters of significant between-group differences, there was no statistical correlation between CLCDS scores (see Table 3, Table S1).

| D ISCUSS I ON
To the best of our knowledge, this resting-state functional MRI study was the first to explore brain networks' modularity in children with NSCLP. The major findings can be summarized as follows: (a) Nodal metric differences were mostly located in the language-related brain

F I G U R E 3
The distribution of the significant differences in the local nodal metrics. a, b: Between-group differences in the AUC of the degree strength and betweenness centrality, respectively. Color represents the T value. Corrected for age and sex. Among the nodes with increased degree strength, the middle temporal gyrus was involved in accessing the word/lexicon and its meaning (Saur et al., 2010;Schwartz et al., 2009). The right intracalcarine cortex and occipital pole were both located in the primary visual center. Our results implied that the communication capacity was strengthened in the middle temporal gyrus, right intracalcarine cortex, and occipital pole, which might help the circuit-level calculation and total information transmission (Betzel et al., 2016). We inferred that the left middle temporal gyrus and the primary visual center might receive specific information from more brain areas for reestablishing correct articulation patterns and placements. Li et al. (2020) found increased gyrification located in the temporal lobe in the adults with NSCLP after speech rehabilitation compared with the healthy controls, consistent with our findings. In our study, speech therapy by visual feedback might improve the degree strength of the occipital cortex and middle temporal gyrus.
For nodes with increased betweenness centrality, most brain areas were located in the left hemisphere, which is the languagedominant hemisphere. Our results showed that increased betweenness centrality was found in the left anterior part of the middle temporal gyrus, orbitofrontal cortex, lateral occipital cortex, and supramarginal gyrus. These brain areas were associated with phonological and semantic processing of language (Fujii et al., 2016). The posterior cingulate cortex provides "action" into the hippocampal memory system, and the anterior cingulate cortex (receiving from the orbitofrontal cortex) provides reward-related input into the hippocampal memory system via the posterior cingulate (Rolls, 2019).
The orbitofrontal cortex is involved in emotion and executive Corrected for age and sex. p * : p-value with the Bonferroni correction.

TA B L E 3 Statistical analysis of modular and global metrics
F I G U R E 4 (a-d) Eg, Eloc, Cp, and Lp showed no significant differences. Shaded areas represent the standard deviation of the mean. Corrected for age and sex. Two-sample two-tailed t test. Bonferroni's correction, p < .05 function (Rudebeck & Rich, 2018). In addition, the thalamus is the essential sensory conduction relay station and associates with perceptual, cognitive, and motor processes (Moustafa et al., 2017).
The higher betweenness centrality for these brain areas suggested that the flows of language, emotion, and execution information increased (Kummer, 2011). We inferred that the children with NSCLP might be with the help of the language, motor, emotion, execution, and memory function for speech rehabilitation (measured by the CLCDS scores). Our previous study also found that the function of language-related brain areas was higher (showing lower nodal shortest path length and higher nodal clustering coefficient) for children with NSCLP after speech rehabilitation compared with controls (Rao et al., 2020). Moreover, adult speech-rehabilitated patients with CLP showed only increased activation in the left hippocampus in a subvocalization task functional MRI study (Zhang et al., 2017), which may be the pattern of speech rehabilitation in adults with NSCLP.
In addition, the only node with decreased betweenness centrality was located in the posterior part of the middle temporal gyrus (MTG). The posterior part of the MTG is a critical area in voice encoding, phonemic processes, and word selection processes during word expression (Glasser & Rilling, 2008). The betweenness centrality or information flows through the MTG in children with CLP were lower than those in healthy controls, which might indicate that after speech rehabilitation, word expression became clear, and the requirement diminished, so the flow of information was reduced through the posterior part of the MTG.
Modularity, one of the central organizing principles of complicated biological systems, has been widely used in recent years (Hartwell et al., 1999). Our findings confirmed the modular structure in resting-state brain networks. In the two groups, five intrinsically cohesive modules were identified in the resting-state functional networks, such as the default mode, sensorimotor, auditory, attention, visual, and salience networks, which were consistent with previous spontaneous brain activity studies (Mandelli et al., 2016(Mandelli et al., , 2018. habilitation. However, the higher Q in rehabilitated children with NSCLP stated more intramodules for local information transfer, which suggested that functionally related components conduct specific biological functions with more specialization because of the adaption of habilitation. We presumed that the increased modularity index was linked with the higher degree strength and betweenness centrality induced by the repair surgery and speech therapy, which improved network adaption (Guye et al., 2010). Our previous study found the increased small-world index in children with CLP after speech rehabilitation (Rao et al., 2020). We can infer that the higher modularity index caused the small-world index to increase for the function specialization. Interestingly, our result was consistent with Duncan & Small's study, which stated that the increased modularity index of resting-state brain networks was detected in patients after aphasia recovery (Duncan & Small, 2016

| LI M ITATI O N S
This study still has its limitations. First, the number of children in both groups was relatively small. Second, more work focused on the investigation of the differences in the topological organization of the functional networks in children with NSCLP before speech rehabilitation should be done in the future. Third, although the exclusion criteria of our study cannot exclude all inherent and acquired factors of impaired brain development, speech therapy may have primary effects on the brain for speech rehabilitation. Fourth, a longitudinal study of the alterations identified in language-related areas and networks in speech-rehabilitated children with NSCLP could be conducted. Fifth, the CLCDS scores in healthy controls should be estimated for the between-group difference in the future, contributing a lot to our results. Sixth, it is important to note that given the study design (i.e., no pretherapy measures and no measures of participants who did not achieve adequate speech following therapy), a causal relationship between speech therapy and topology patterns cannot yet be inferred. However, the findings of increased network modularity for participants with NSCLP following speech therapy support the need for further research in this area.

| CON CLUS I ON
There were no significant differences in global network metrics for children with NSCLP after speech therapy. However, significant differences existed in local nodal metrics for the language-related brain regions. In addition, the NSCLP group had increased network modularity (Q and Q AUC ) compared with the healthy controls. The similar global network metrics and increased network modularity provided profound insights into the neurobiological understandings of speech-rehabilitated children with NSCLP and could be potential imaging biomarkers for the estimation of speech rehabilitation.

ACK N OWLED G M ENTS
This research was supported in part by the National Natural Science

CO N FLI C T O F I NTE R E S T
The authors declare that they have no conflicts of interest.

AUTH O R CO NTR I B UTI O N S
Bo Rao designed methodology, provided software, and wrote the original draft. Hua Cheng wrote, reviewed, and edited the manuscript; administered the project; validated the data; and acquired funding. Wenjing Zhang involved in formal analysis and curated the data. Renji Chen investigated the study and provided resources. Yun Peng supervised the study and acquired funding.

PEER R E V I E W
The peer review history for this article is available at https://publo ns.com/publo n/10.1002/brb3.2094.

DATA AVA I L A B I L I T Y S TAT E M E N T
In our study, we used and analyzed the datasets, which is available from the corresponding author for a reasonable request.