Meta-analysis of structural integrity of white matter and functional connectivity in developmental stuttering

Developmental stuttering is a speech disfluency disorder characterized by repetitions, prolongations, and blocks of speech. While a number of neuroimaging studies have identified alterations in localized brain activation during speaking in persons with stuttering (PWS), it is unclear whether neuroimaging evidence converges on alterations in structural integrity of white matter and functional connectivity (FC) among multiple regions involved in supporting fluent speech. In the present study, we conducted coordinate-based meta-analyses according to the PRISMA guidelines for available publications that studied fractional anisotropy (FA) using tract-based spatial statistics (TBSS) for structural integrity and the seed-based voxel-wise FC analyses. The search retrieved 11 publications for the TBSS FA studies, 29 seed-based FC datasets from 6 publications for the resting-state, and 29 datasets from 6 publications for the task-based studies. The meta-analysis of TBSS FA revealed that PWS exhibited FA reductions in the middle and posterior segments of the left superior longitudinal fasciculus. Furthermore, the analysis of resting-state FC demonstrated that PWS had reduced FC in the right supplementary motor area and inferior parietal cortex, whereas an increase in FC was observed in the left cerebellum crus I. Conversely, we observed increased FC for task-based FC in regions implicated in speech production or sequential movements, including the anterior cingulate cortex, posterior insula, and bilateral cerebellum crus I in PWS. Functional network characterization of the altered FCs revealed that the sets of reduced resting-state and increased task-based FCs were largely distinct, but the somatomotor and striatum/thalamus networks were foci of alterations in both conditions. These observations indicate that developmental stuttering is characterized by structural and functional alterations in multiple brain networks that support speech fluency or sequential motor processes, including cortico-cortical and subcortical connections.


Introduction
Developmental stuttering is a speech disorder characterized by dysfluencies which commonly arise during early childhood.The core symptoms include repetitions and prolongations of speech sounds and blocks while speaking.However, these symptoms often result in a range of secondary problems that extend beyond speech-related domains, including psychiatric and psychological issues.For instance, social anxiety disorder is known to show a high rate of comorbidity with developmental stuttering (Iverach and Rapee, 2014).Although many cases show recovery during childhood, the prevalence rate remains as high as 1% in adulthood (Craig et al., 2002).
The neural basis of stuttering has been investigated using neuroimaging techniques for more than a quarter of a century.Using several modalities of neuroimaging, including structural magnetic resonance imaging (MRI), functional MRI (fMRI), and diffusion MRI (dMRI), evidence of brain alteration in persons with stuttering (PWS) has been reported.Particularly, altered local brain activation while speaking has been the primary focus of investigation, which resulted in several convergent findings.An early meta-analytic study utilizing positron emission tomography and fMRI has reported over-activation mainly in motor-related areas and under-activation in the auditory cortex in PWS (Brown et al., 2005).A part of the finding has been replicated in a more recent meta-analysis in which motor-related areas including the precentral gyrus showed over-activation (Zhang et al., 2022a).A further fine-grained picture of functional alteration was provided by another meta-analysis study, which stratified altered brain activations into "trait" and "state" effects of stuttering (Belyk et al., 2015(Belyk et al., , 2017)).
Such findings are largely in line with the notion that developmental stuttering is associated with functional abnormalities in the speechmotor system (Braun et al., 1997;Fox et al., 1996;Sakai et al., 2009;Toyomura et al., 2011).However, in addition to functions of the local brain regions of the motor cortex, speech motor control crucially depends on the integration of the large brain networks, including supplementary motor areas (SMA), anterior cingulate cortex (ACC), basal ganglia, and cerebellum (Chang et al., 2016;Chang and Zhu, 2013;Yang et al., 2016Yang et al., , 2019)).Therefore, possible alterations in connections between these speech-motor-related regions may be directly relevant to speech disfluency.Recent MRI studies have greatly advanced our knowledge of alterations in structural features of white matter pathways and functional connectivity (FC) in PWS.Nevertheless, the directions of alterations, including increases and decreases in structural integrity of white matter and FC measures, often exhibit complicated patterns, some even contradicting one another across different studies.Aside from age and sex-related factors, we note that methodological heterogeneities may be an additional factor contributing to the inconsistencies in results.For structural integrity, for instance, methodological differences exist at multiple levels including the MRI metrics (e.g., white matter density in the voxel-based morphometry [VBM] vs. fractional anisotropy [FA] in the dMRI) as well as analytical approaches (e.g., tract-based spatial statistics [TBSS] vs. regions-of-interest [ROI] analysis using tractography).A previous meta-analysis study (Neef et al., 2015) using the activation likelihood estimation (ALE) method combined TBSS and VBM publications up to and including 2015 and found significant white matter alterations at several locations, including the left superior longitudinal fasciculus (SLF) and posterior corpus callosum.However, given the mixing of different MRI metrics and analyses, additions of publications after 2015, and the conceptual and technical revisions and advances in the coordinate-based meta-analysis of neuroimaging data (Eickhoff et al., 2016(Eickhoff et al., , 2017;;Albajes-Eizagirre et al., 2018), the rationale for updating the previous findings is compelling.For FCs, a previous systematic review presented a summary of findings on FC alterations, published up to and including 2016 (Etchell et al., 2018).However, altered FCs were described qualitatively without formal assessment for reproducibility.
In this study, we conducted meta-analyses to synthesize previous findings of structural integrity of white matter and FC studies and to elucidate alterations of brain connections in PWS.In order to rule out the possible effects of methodological heterogeneities, studies were selected so that MRI measures and analytical approaches were comparable.For structural integrity, we focused on studies employing the whole-brain TBSS analysis of FA, as it represents the most commonly used metric for white matter microstructural integrity and provides a non-biased, hypothesis-free estimation for white matter alterations throughout the brain.To maintain methodological consistency, we did not include studies measuring white matter volume/density in VBM, as it provides a macroscopic assessment of the white matter tissue, conceptually distinct from microstructural integrity.Additionally, fiber counts from DTI tractography were omitted, given that they do not directly quantify axon numbers and are affected by methodological variables, including threshold settings for fiber tracking.
For FC, we focused on seed-based FC analysis either at rest or during specific tasks.While selecting seed regions limits the exploration of the range of FCs in the whole brain, seed-based FC allows a straightforward interpretation of obtained metrics and is the most common method for detecting abnormal FC in clinical neuroscience, including stuttering.Therefore, we chose this method for our meta-analysis over alternatives such as independent component analysis (ICA), which, while allowing for whole brain exploration, is influenced by assumptions like the number of pre-specified components and is less frequently applied for studies of developmental stuttering.Studies using other methodologies, such as graph-theoretic measures and ROI-to-ROI based estimation, were also excluded.A potential problem with the seed-based FC studies is the spatial variance of the selection of seed regions among studies.
Recent meta-analysis studies of seed-based FCs partly circumvented this problem by categorizing seed regions into functional networks, such as the somatomotor and default-mode networks (Keiser et al., 2015;Tang et al., 2022).This strategy is effective for organizing previous findings using spatially varied seed regions into the larger functional network scale.Motivated by this approach, we characterized altered FCs at the functional brain network level in addition to the regional level by classifying FC terminal regions into a functional network parcel.
Based on the previous reviews (Neef et al., 2015;Etchell et al., 2018), we hypothesized that significant FA reductions in developmental stuttering might be identified in multiple locations in the left hemisphere, particularly for the left SLF, arcuate fasciculus, and corpus callosum.For FC alterations, the previous systematic review reported reduced FCs between the basal ganglia and cortical speech and motor-related areas, such as SMA and superior temporal cortex, and suggested an increased cerebellar-cortical FC as a possible compensatory mechanism.Therefore, we hypothesized that cortico-striatum connections might show reduced FCs, whereas cerebellar-cortical connections may show increased FCs.On the other hand, although task-based FC alterations were less adequately reviewed than the resting-state FC, the review described altered connections mainly involving the anterior speech motor system, including the precentral cortex and SMA (Etchell et al., 2018).Therefore, we hypothesized that altered FCs during tasks may be found in the motor and other cortical systems.At the functional network level, the resting-state and task-based FCs alterations may be identified mainly in the speech motor-related systems in the cortical and subcortical structures, including the language and somatomotor systems, cerebellum, and striatum.

Search strategy and selection criteria
The meta-analysis was conducted in accordance with the Preferred Reporting Items for Systematic Reviews and Meta-analyses (PRISMA) guidelines (Page et al., 2021) (Fig. 1).An online search was performed on the following electronic databases: PubMed, Web of Science, and MIDLINE.One of the authors (K.M.) searched the databases using the following combination of keywords: ("stutter" OR "stuttering") AND (("magnetic resonance imaging" OR "MRI") OR ("functional magnetic resonance imaging" OR "fMRI") OR ("diffusion tensor imaging" OR "DTI") OR ("diffusion-weighted imaging" OR "DWI") OR ("connectivity")) for the combined literature of structural integrity of white matter and FC.The search was conducted on December 1st, 2022.After removing duplicates, this research yielded 470 relevant studies.The titles and abstracts of the retrieved papers were independently screened by the two authors (K.M. and R. H.), and all information was cross-checked by one investigator (R. A).In case of conflicting evaluations, they were resolved by consensus through discussion between two investigators (K.M. and R. H.) before proceeding with full-text reading.
The screening was performed separately for both the structural integrity and FC analyses.In both analyses, studies were included if they.
(1) were peer-reviewed original articles and were written in English.
(2) included participants who were diagnosed with PWS or recovered stutters (RS).(3) directly compared PWS or RS with persons who do not stutter (PWNS).(4) performed whole or nearly whole-brain analysis.
(5) reported three-dimensional peak coordinates of group difference in standard stereotaxic space of either Montreal Neurological Institute (MNI) or Talairach space.
In both analyses, studies were excluded if.
(1) participants completely overlapped with those in another publication.
(2) they performed only individual-level analysis without grouplevel analysis.
(3) they were case-reports, reviews, or conference abstracts.
For the analysis of structural integrity of white matter, studies were included if they performed the TBSS analysis of the FA value.Studies were excluded if they analyzed FA data by other methods, such as the tractography analysis.Henceforth, the studies analyzing structural integrity will be referred to as those utilizing FA analysis.
For the FC analysis, studies were included if they performed seedbased FC analysis whereas studies performing other FC analysis methods, such as ICA-based and ROI-to-ROI analysis, were excluded.When studies performed separate FC analyses by using different seeds on the same samples, they were regarded as separate datasets (Tang et al., 2022), with the exception of two studies that performed separate FC analyses by setting more than 100 seeds (Frankford et al., 2021;Shojaeilangari et al., 2021).

Data extraction and recording
For both FA and FC analyses, the peak coordinates and their t statistic of clusters of between-group differences were extracted.When only the z score or p value of the cluster was reported, we converted it to t statistic using the Seed-based d Mapping utility (https://www.sdmproject.com/utilities/?show=Statistics).When the peak t value of clusters was not either reported or unable to be converted to t value, we recorded the direction of group difference (i.e., PWS/RS > PWNS or PWS/RS < PWNS).Demographic data (e.g., age information) was also extracted.Additionally, studies reporting none of group differences were included to reduce the publication bias.For the FC analysis, studies were classified depending on whether they were resting-state or task-based FC studies.For both studies, anatomical names of seed regions were extracted.For task-based FC studies, the tasks of experimental and control conditions were extracted.
Data extractions were conducted by two authors (K.M. and R.H.) independently and all information was double-checked by the two authors.Any inconsistent results were discussed and resolved by discussion.

Meta-analysis
The coordinate-based meta-analyses for FA and FC alterations in PWS were separately performed using the Seed-based d Mapping with Permutation of Subject Images (SDM-PSI) software, version 6.21 (https://www.sdmproject.com/).The SDM approach performs conventional univariate voxel-wise testing if the effect of interest significantly deviates from null.This characteristic contrasts with other coordinatebased meta-analyses that test the spatial convergence of the effect across studies, such as the ALE technique.Beyond the application of conventional voxel-wise evaluations of the effect, the SDM-PSI offers multiple advantages.It accounts for both positive and negative alterations, ensuring that opposing findings neutralize one another.Moreover, it incorporates studies that report non-significant effects, thereby ensuring an unbiased estimation of the effect of interest.Details on the SDM-PSI can be found elsewhere (Albajes-Eizagirre et al., 2019).
We first created the data sheets that contain the MNI coordinates of cluster peaks and t statistics in each study.For studies that reported the Talairach coordinates, we used the MNI2Tal utility (https://bioimagesui teweb.github.io/webapp/mni2tal.html)for the conversion to MNI coordinates.For studies that reported only the corrected p value of the cluster, peak clusters showing PWS/RS > PWNS were coded as "P" whereas those showing PWS/RS < PWNS were coded as "N".Then we prepared the SDM table that contains the number of participants of each group and the threshold of t value in each study.For the FC analysis, conditions were additionally contained in the SDM table (the restingstate or task-based).For preprocessing, we calculated images of the lower and upper bounds of the possible effect sizes by converting the t value of each peak coordinate into effect size (Hedge's g), which was then convolved with a fully anisotropic Gaussian kernel (anisotropy = 1, isotropic full width at half maximum (FWHM) = 20 mm) using the mask images (voxel size = 2 ×2 x 2 mm) (Peters et al., 2012;Zhang et al., 2022b).The preprocessing was performed separately for TBSS and FC analyses, using FA skeleton as the mask for TBSS and gray matter mask for FC analyses.After the preprocessing, we applied maximum likelihood techniques to obtain the most likely effect size map and its standard error.The estimation of the effect sizes was based on MetaNSUE algorithms (Albajes-Eizagirre et al., 2018).Through multiple imputations of study images (number of imputations = 50) and imputations of subject images, group analyses of the subject images for each study and imputation were calculated, which was then meta-analyzed using a random-effects model for each permutation.The mean analysis was performed by calculating the weighted mean value of voxels (PWS/RS -PWNS) across different studies, with the weight of a study the inverse of the sum of the variance and the between-study heterogeneity.Then, the generated images are combined using Rubin's rules and saved the largest z value from the combined meta-analysis image.The procedure was iterated 1000 times and obtained the distribution of the z values to compare the combined meta-analysis image from unpermuted data.For the FC analysis, subgroup analyses were conducted by separating between studies of the resting-state and task-based FC.
For all the analyses, the statistical threshold was set at p = 0.005, uncorrected for multiple comparisons, with a spatial cluster threshold (k) of larger than 10 voxels as the main analysis.This threshold was recommended by previous studies in terms of the balance of the sensitivity and the control for type I error (Zhang et al., 2022b).The Threshold Free Cluster Enhancement (TFCE) correction was not possible for the TBSS analysis in the SDM-PSI.We further verified the results for FC analyses using the TFCE-corrected threshold of p = 0.05 and a spatial cluster of k > 10 for the FC analyses only.
In order to characterize altered FCs at the functional network level, altered FCs were delineated into functional network anatomy.Each FC consists of two terminal regions: the seed region and the peak of group difference reported in each source study.We identified an FC alteration when a peak coordinate of FC analysis taking a seed region in a source study is located within 20 mm from peak coordinates of significant clusters identified in our meta-analysis.We used the functional network parcellation scheme of Ji et al. (2019) for the parcellation of terminal cortical regions.That parcellation scheme is characteristic of containing "language" network (LAN) among other parcellation schemes such as Power's atlas (Power et al., 2011) and Schaefer's atlas (Schaefer et al., 2018).This parcellation scheme contained other cortical networks "somatomotor (SMN)," "cingulo-opercular (CON)," "dorsal attention (DAN)," "frontoparietal (FPN)," "auditory (AN)," "default-mode (DMN), " "posterior-multimodal (PMN)," "ventral multimodal (VMN)," "orbito-affective (OAN)," "primary visual (V1) and "secondary visual (V2)".We further included subcortical networks of "striatum and thalamus (ST/TH)," "cerebellum (CRB)," and "hippocampus and amygdala (HIPP/AMY)" to the cortical networks, which resulted in 15 functional networks in total.We classified the seed and the group difference peak into one of the functional networks based on the MNI coordinates and information provided in each source study.

Robustness analyses
The jack-knife sensitivity analysis was performed to prevent one or a few studies from driving the results.This analysis repeated the mean statistical analysis, excluding one dataset each time.If a significant main result remains in all or most of the sensitivity analyses, it is considered replicable and robust (Radua et al., 2013;Radua and Mataix-Cols, 2009).To account for heterogeneity, τ 2 and I 2 were utilized for each identified brain region showing significant group difference, with I values of 50% or greater indicating substantial heterogeneity among the studies (Higgins and Thompson, 2002;Viechtbauer, 2005).These heterogeneity values were then reported in the SDM-PSI as standard z values with variances for each peak.Additionally, Egger's tests in SDM-PSI were conducted to examine the possibility of publication bias (Egger et al., 1997;Sterne and Egger, 2001).

Included studies and sample characteristics
This study included 11 publications for the TBSS FA analysis, with a total of 248 persons with developmental stuttering (216 PWS and 32 RS) and 243 PWNS.For the FC analysis, the study included 11 publications, with a total of 224 PWS and 234 PWNS.Within the FC publications, six publications were related to resting-state FC studies, and another six publications were related to task-based FC studies (one publication conducted both analyses).By creating a separate dataset for each seed in an individual FC study (Tang et al., 2022), we identified 58 datasets for FC (29 for resting-state and 29 for task-based FC).Fig. 1 shows the flow chart of the search strategy and study selection.Supplementary Tables and 2 summarize sample characteristics including sample size and age in each group of the individual studies for TBSS and FC, respectively.Supplementary Tables 3 and 4 summarize the findings of the retrieved TBSS and FC studies, respectively.

Analysis of TBSS FA
In comparison to PWNS, PWS exhibited decreased FA in two clusters of the left superior longitudinal fasciculus II (SLF II).The first cluster was located below the fundus of the left central sulcus and extended into part of the left arcuate fasciculus, while the second cluster was located close to the left angular and supramarginal cortex (see Table 1 and Fig. 2).

Table 1
Regions showing group difference in analysis of fractional anisotropy (FA).The significant difference between PWS and PWNS at uncorrected p < 0.005 and cluster size > 10 voxels.Abbreviations: L, Left; SLF, superior longitudinal fasciculus.

Analysis of FC
Six clusters with increased FCs for PWS were identified, including the right ACC, left anterior lateral temporal cortex (aLTC), left cerebellum crus I, left inferior frontal cortex (IFC) orbital part, right cerebellum crus I, and right middle frontal cortex.Among the six clusters, the clusters of the right ACC and the left aLTC survived the TFCE correction.No significant clusters showing reduced FCs for PWS were found (see Supplementary Figure 1 and Supplementary Table 5).

Analysis of resting-state FC
Increased resting-state FC for PWS was found in only one cluster in the left cerebellum crus I. On the other hand, reduced FCs were found in two clusters, located in the right inferior parietal cortex (IPC) and right SMA (Table 1 and Fig. 3A).The seed regions and publications that reported significant group differences around these brain regions are summarized in Table 2.These hyper-and hypo-connections were classified based on functional network anatomy (Ji et al., 2019) in Fig. 4A.The cerebellar-cortical FC showed hyper-connections, whereas the somatomotor, frontoparietal, and striatum/thalamus constituted hubs for hypo-connections.To show the distributional differences of affected functional networks, counts of terminal regions in each functional network are summarized in Fig. 4B.The striatum/thalamus and cerebellum networks contained the largest number of identified FCs for hyper-and hypo-connections, respectively.

Analysis of task-based FC
The analysis revealed increased FCs for PWS in six clusters, comprising the right ACC, left aLTC, right cerebellum crus I, left cerebellum crus I, right middle frontal cortex (MFC), and left putamen (Table 1 and Fig. 3B).The cluster in the left aLTC contained local maxima in the aLTC, the orbital part of IFC, and the insula.In addition, the tests using the TFCE-corrected threshold identified clusters in the right ACC, left aLTC, right cerebellum crus I, left orbital part of IFC, and left insula.Seed regions and publications that reported significant group differences around these brain regions are summarized in Table 3.These hyper-connections were classified at the functional network level (Fig. 4C).Between-network connections of the default mode and striatum/thalamus networks and the default mode and somatomotor networks are particularly notable.The number of terminal regions of these hyper-connections is summarized in Fig. 4D.The terminal regions were heavily distributed over the networks of somatomotor, striatum/thalamus, default mode, and cerebellum.No significant clusters showing reduced FC for PWS were observed.

Sensitivity analysis, heterogeneity analysis, and publication bias
The jackknife sensitivity analyses revealed high replicability for clusters of main results, except for the left putamen in the task-based FC analysis.Supplementary Table 6 displays detailed results of the sensitivity analysis.Levels of heterogeneities (I 2 ) were acceptable in all regions in the main results (see Supplementary Table 6).Additionally, none of the clusters exhibited significant publication bias as per Egger's test (p > 0.05).

Discussion
Our coordinate-based meta-analyses, conducted using the SDM-PSI, aimed to investigate structural integrity of white matter and functional connectivity in developmental stuttering.Through a systematic literature search, we identified a total of 11 TBSS FA studies and 11 FC studies, including both resting-state (6 studies) and task-based (6 studies) analyses.Meta-analysis of the TBSS FA studies revealed clusters of reduced FA localized in the left SLF-II.The meta-analyses of FC revealed different patterns of alterations between resting-state and taskbased analyses.Specifically, analyses of resting-state FC revealed reduced FCs in the right IPC and left SMA, along with increased FC in the left cerebellum crus I. On the other hand, analyses of the task-based FC revealed predominantly increased FCs involving the ACC, left aLTC, crus I of the bilateral cerebellum, left orbital part of IFC, left posterior insula, right MFC, and left putamen.To our knowledge, our study represents the first meta-analysis examining white matter microstructure and functional connectivity in developmental stuttering as compared to nonstuttering, according to the PRISMA guidelines.A meta-analysis on white matter alteration had been previously conducted (Neef et al., 2015).However, we believe that our updated study represents an improvement over that study in that we adhered to the PRISMA guidelines and focused on TBSS FA studies, thereby avoiding the methodological heterogeneity associated with combining dMRI and VBM.

Alterations of structural integrity of white matter
Our meta-analysis identified two foci of reduced FA in the middle and posterior part of the left SLF-II.It is notable that altered FAs in the left SLF were reported in both children (middle and posterior SLF: Chang et al., 2008;Chang et al., 2015;Chow and Chang, 2017) and adults (middle SLF: Kell et al., 2009;Connally et al., 2014;Cai et al., 2014;Civier et al., 2015, posterior SLF: Watkins et al., 2008;Connally et al., 2014;Mollaei et al., 2021).The composition of the SLF is highly controversial (Janelle et al., 2022;Schurr et al., 2020).Nevertheless, it is commonly accepted that the SLF-II, which connects the angular cortex to  (Chang and Zhu, 2013) LAN L.putamen (Yang et al., 2016) SMN ST/TH L.putamen (Chang et al., 2016) ST/TH The significant difference between PWS and PWNS at uncorrected p < 0.005 and cluster size > 10 voxels.The columns labeled "Region" denote two terminal regions in altered functional connections: a region of significant group difference and seed.Classifications of the terminal regions into functional network are shown in the columns "Functional network".The classification was based on the fuctional network parcellation scheme of Ji et al. (2019) the dorsal precentral cortex and caudal MFC, constitutes the dorsal component of the SLF that is separate from the ventral component, SLF-III, which mainly connects the supramarginal cortex and inferior frontal cortex (Wang et al., 2016).It is also agreed that the SLF-II shows strong leftward lateralization (Janelle et al., 2022;Wang et al., 2016), with the parietal origin extending into the supramarginal cortex only in the left hemisphere (Wang et al., 2016).Our finding of localized FA alterations in the left SLF-II, but not in the SLF-III, is interesting considering the extensive literature implicating the more ventral pathways, such as SLF-III and arcuate fasciculus, in speech and language functions (Bernal and Altman, 2010;Makris et al., 2005;Nakajima et al., 2020).However, the left SLF-II is also considered a component of the language pathway responsible for motor planning of language function and/or syntactic processing during language production (Nakajima et al., 2020;Wang et al., 2016).Moreover, this pathway is critically involved in rapid action reprogramming that does not necessarily involve speech production per se (Hartwigsen et al., 2012).Thus, our findings suggest that developmental stuttering may involve difficulties in motor planning and execution that could be specific to speech or general rapid movements (Nakajima et al., 2020).
Our findings are partially in agreement with a previous review study that employed the ALE technique to analyze collapsed data of the voxelbased morphometry and TBSS (Neef et al., 2015).That study identified significant clusters of white matter alteration in the left SLF near the angular gyrus and below the fundus of the central sulcus.In addition, they reported significant FA reduction around the splenium part of the corpus callosum, which we failed to detect in our analysis.Nevertheless, we note several voxels showing significant FA reductions near the area, which did not reach the spatial extent threshold (k < 10).We also found another sub-extent threshold cluster of voxels in the left fronto-occipital fasciculus beneath the left inferior frontal cortex, which exhibited significant FA reductions at the voxel level.These sub-threshold clusters in our study may reach statistical significance in future meta-analyses with the inclusion of additional studies.

Alterations of resting-state FC
Although alterations in connectivity directionality were sometimes found to be comparable between the resting-state and task-based FCs within individuals (Schurz et al., 2015), our observations indicate that there are predominantly disparate patterns of FC alterations between the two conditions.Hence, separate analyses are warranted for each of the conditions.
Our meta-analysis of resting-state FC revealed significant FC reduction in the SMA.This reduction was most consistently shown when the seed region was placed in the left putamen, as demonstrated by several previous studies of both children and adults who stuttering (Chang et al., 2016;Chang and Zhu, 2013;Yang et al., 2016) (see Table 2).This finding may be reasonably interpretable, given that the SMA is a crucial node for the cortico-striato-thalamo-cortical loop, which is implicated in a broad range of control processes in motor and cognition, including speech (Busan, 2020;Nachev et al., 2008).Furthermore, reduced FCs in the SMA have been reported in a prior study utilizing Granger causality analyses that showed weaker influences from SMA to putamen in PWS (Qiao et al., 2017).Associations of functional abnormalities in the SMA with stuttering and other dysfluent disorders have been indicated by several sources, including neuroimaging studies (Busan, 2020), lesion studies (Chung et al., 2004), and brain stimulation studies (Fried et al., 1991;Garnett et al., 2019).The SMA plays a critical role not only in speech motor, but also in action monitoring, motor control, and initiation of motor programs in general (Bonini et al., 2014;Nachev et al., 2008).Therefore, it is plausible that reduced functional integrations of the SMA and cortico-striato-thalamo-cortical loop contribute to speech dysfluency in developmental stuttering.In particular, a reduced FC between the putamen and SMA is relevant to neurocomputational models for stuttering.In an influential computational model of speech production, the Gradient Order DIVA model (Bohland et al., 2010), the connection constitutes an input to the motor loop, and the information flow from SMA to putamen is proposed to control the initiation of motor programs for speech production.Therefore, its aberrant function is predicted to result in problems in the initiation and fluidity of speech, as observed in developmental stuttering.The importance of functional alterations in these regions is corroborated by the functional network-level analysis, in which the reduced FCs were heavily distributed over the striatum/thalamus and somatomotor networks.
On the other hand, the interpretation of reduced FC in the right IPC is less straightforward.We are unaware of strong evidence that the right IPC is critically involved in speech motor control per se.This region contains multiple functional networks, including fronto-parietal, dorsal attention, and sensorimotor.These functional networks may suggest that this region is important for attention-demanding and controlled processes, including working memory and executive control functions (Marek and Dosenbach, 2018;Zanto and Gazzaley, 2013), particularly for sensory and motor integration.Consistently, a previous study reported that the right IPC is involved in multiple aspects of auditory working memory, such as monitoring and updating sound information, integrating auditory and motor functions (Alain et al., 2008), and maintenance of structures of pitch sequences (Royal et al., 2016).Therefore, altered functions in auditory working memory may contribute to problems in processing and monitoring of auditory sequences such as speech.On the other hand, this network also supports sequential motor control (Parlatini et al., 2017;Wise et al., 1997).As speech is a motor sequence requiring temporally precise coordination, the finding may suggest a general problem in sequential motor control in developmental stuttering.At the functional network level characterization, reduced FCs involving the somatomotor, cingulo-opercular, fronto-parietal networks may underline alterations of these functions.
We also observed increased FCs between the cerebellum and functional networks of somatomotor and language (Fig. 4).This finding may be consistent with a previous meta-analysis suggesting an increased cerebellar-cortical connection (Etchell et al., 2018).Given the existence of cortico-cerebellar loop for language and somatomotor functions (Stoodley and Schmahmann, 2009), the hyper-connections involving the cerebellum and cortical functional networks may play compensatory roles for altered motor and cognitive processes involved in speech production.

Alterations of task-based FC
In contrast to the resting-state FC, alterations of the task-based FC were characterized as a general shift toward the positive direction in several regions, including the subcortical structure.Although various task conditions are included, these contrasts are common in that experimental conditions contain more cognitive and motor components required in fluent speech production than control conditions, such as vocalization (nonspeech production vs. rest), generation of complex speech sound sequences (speech working memory task vs. repetitions of same syllables), and motor preparatory processes (cued vs. uncued trials of continuous performance task).Therefore, these contrasts are expected to share cognitive and motor processes related to fluent speech production.
Among the regions showing increased FCs, the ACC, left aLTC, left posterior insula, and left IFC are part of regions that are reliably activated during speech production (Indefrey and Levelt, 2004).The ACC is involved in motor control including vocalization and speech production (Belyk et al., 2015;Paus, 2001).The observation of increased FC during speech and non-speech production tasks (Chang et al., 2011) suggests compensatory FC for more effortful control in PWS during the tasks.The orbitalis part of the left IFC has also been reported to be involved in speech in numerous studies (Conner et al., 2019;Razorenova et al., 2020).Increased FC in PWS of this region was observed for various tasks that require overt speech/non-speech production (Chang et al., 2011;Yang et al., 2019).The left posterior insula has been recently shown that this region is involved in speech motor control by monitoring the speaker's own speech (Woolnough et al., 2019).Increased FC in this region may also reflect PWS's effortful control or monitoring of overt speech production (Chang et al., 2011;Yang et al., 2019).
It is unlikely that all of the increased FCs observed in PWS can be attributed solely to factors related to speech productions or vocalizations.In addition to speech production and vocalization factors, the ACC is implicated in multiple aspects of cognitive control, including conflict monitoring and adaptive control of behavior (Mansouri et al., 2017).The increased FCs in PWS observed in PWS for non-speech cognitive tasks (Metzger et al., 2018) may suggest effortful control of cognitive processes in general in PWS.Because the ACC cluster was identified in its dorsal segment in our analysis, this view is consistent with the classical functional subdivision model of the dorsal "cognitive" ACC and the ventral "emotional" ACC (Bush et al., 2000).However, recent evidence suggested that the dorsal ACC is also involved in social affective processes, including social cognition, reward, and motivation (Hughes and Beer, 2012;Magno et al., 2009).Given the observation that stuttering severity is influenced by situational variations in social cognition (Alm, 2014), altered social affective function in dorsal ACC may be associated with speech dysfluency.Moreover, the increased FCs in the right MFC may reflect compensatory processes for cognitive controls rather than vocalization in speech or non-speech production in PWS.Thus, it is possible that the alterations of task-based FCs in PWS can be At the functional network level, increased FCs were distributed over multiple networks.Among them, FCs between the default mode and striatum/thalamus networks and between the default mode and somatomotor networks are particularly notable.This finding is striking compared to the distribution of altered FCs during the resting state, in which no alterations involving the default mode network were identified.Functional alterations of the default mode network received less attention compared to other functional networks, such as sensorimotor and striatum/thalamus, in the literature on developmental stuttering.However, a recent study reported that, compared to controls, PWS showed an increase in FCs between the default mode network regions and a task-positive auditory region during solo reading (Garnett et al., 2022).Motivated by the default mode network interference model originally proposed for attention-deficit and hyperactivity disorder (Sonuga- Barke and Castellanos, 2007), a possibility was suggested that increased FCs between the default-mode and task-positive networks may disturb efficient network switching and possibly disrupt speech motor control and performance of other tasks (Chang et al., 2018;Garnett et al., 2022).Our observation of increased FCs of this network with sensorimotor and striatum/thalamus networks is consistent with this possibility.This finding may provide a new possible anatomical target for the intervention for improving speech fluency.

Relationships between alterations in structural integrity and functional connectivity
Meta-analyses of FA, resting-state FC, and task-based FC revealed largely different patterns of alterations in PWS.Because the datasets used for these analyses are largely non-overlapping, with the exception of Yang et al. (2017) for resting-state and task-based FCs, it remains unclear whether these patterns reflect common underlying pathologies that manifest in different manners depending on MRI modality and brain state.Although it is possible that these differences simply reflect subject-level variability, it would be more informative to investigate potential relationships among these lines of findings rather than analyzing each one in isolation.
Because FCs are likely to be state-dependent, FA alterations may provide a reasonable starting point for integrating the three pieces of evidence.The observation of reduced FA in the middle and posterior parts of the SLF-II indicates the PWS have decreased structural integrity of this white matter tract in the language-dominant hemisphere.The anterior and posterior terminals of this tract cover multiple functional networks in the frontal and parietal cortex, including the language, somatomotor, dorsal attention, default-mode, and fronto-parietal networks.Therefore, its structural disruption potentially affects these functional networks.In this regard, reduced resting-state FCs involving somatomotor, fronto-parietal, language, and dorsal attention networks may be associated with reduced structural integrity in the left SLF-II.However, white matter disruption in this cortico-cortical connection can potentially explain only parts of altered resting-state FCs because the striatum/thalamus is the largest hub for the network of reduced FCs (Figs. 4A and 4B).This observation indicates the presence of functional brain disruptions without clear structural disruptions in developmental stuttering.The reduced cortico-subcortical connections involving the striatum/thalamus represent an interesting contrast with connections involving the cerebellum, which showed increased FCs with cortical functional networks, including the language network (Fig. 4A).The increased cerebellar FCs were consistently observed in the task-based FCs involving the somatomotor and striatum/thalamus networks (Fig. 4C).Given the cortico-cortical, cortical-striatal, and corticocerebellar connections jointly contribute to the production of fluent speech (Ackermann, 2008;Kotz and Schwartze, 2010;Riecker et al., 2005;Vargha-Khadem et al., 2005), these increased cerebellar FCs may reflect compensatory functions that might be shaped through adaptive processes in response to reduced structural integrity in the left SLF II and reduced FCs in the cortico-striatal connection.In support of this possibility, Sitek et al. (2016) found an increased level of resting-state FC between the left cerebellum and right putamen in PWS compared to PWNS, while stuttering severity was inversely correlated with FCs involving the bilateral cerebellum with the IFC.These findings suggest that those with mild stuttering symptoms rely on cerebellar compensation mechanisms to maintain fluency, as previous research found a link between cerebellar hyperactivity and stuttering severity in activation and intervention studies (De Nil et al., 2003;Fox et al., 2000).
When patterns of FC alterations are compared between resting state and task, the striatum/thalamus and somatomotor networks stand out, containing a relatively large number of reduced resting-state and increased task-based FCs.This observation suggests functional alterations of these networks during both resting state and performance of tasks.Functional alterations in these networks are consistent with neuroanatomical hypotheses of developmental stuttering (Bradshaw et al., 2021;Chang and Guenther, 2020).However, an interesting observation is that the directions of FC alterations are opposite between resting state and task.Although the reason for this discrepancy is unclear, a previous study suggested that greater task-based FC might compensate for the reduction in structural properties in the intra-hemispheric connection (Hermundstad et al., 2013).Assuming that reduced resting-state FCs reflect functional reductions in connections because of deficits of structural integrity in the left SLF-II and other reasons, increased task-based FCs may reflect compensatory processes in speech motor and executive processes.Given the observation that the set of reduced resting-state FCs and increased task-based FCs are largely non-overlapping except for FCs between striatum/thalamus and cingulo-opercular networks, functional reductions in the striatum/thalamus and somatomotor networks may be compensated by increased FCs with other functional networks involved in speech production and executive processes such as language, cerebellar, and cingulo-opercular networks.

Limitations
This study revealed patterns of altered structural integrity and FC in PWS by conducting meta-analyses following the PRISMA guideline.However, we note several limitations in interpreting the present results.Firstly, the number of studies is small for each of the TBSS FA and FC analyses.Each analysis for resting-state and task-based FC was conducted with six studies.Therefore, we acknowledge that our analyses might fail to detect small but significant effects because of insufficient statistical power.For instance, although a reduction of FA in the splenium part of the corpus callosum failed to reach the significant cluster size, such subthreshold effects may turn out significant in future studies.Our findings need to be verified for replication in meta-analyses by including future studies.Secondly, the study populations were heterogeneous concerning age, sex, and the status of stuttering, including both persistent PWS and recovered individuals in some studies (Chang et al., 2008;Chow and Chang, 2017;Kell et al., 2009) (see Supplementary Table 1).These factors can potentially influence the results, considering the change in prevalence of stuttering along development (Howell et al., 2008;Yairi and Ambrose, 1999), sex difference in prevalence (Yairi andAmbrose, 1999, 2013), the evidence for brain functions before and after the speech fluency training (Korzeczek et al., 2021;Lu et al., 2012).In particular, the limitation on the age factor is critical for interpreting the current results.The number of studies of children younger than 12 years old was three and two for the TBSS and resting-state FC studies, respectively.None of the task-based FC studies included children.Therefore, the current results may primarily reflect adults with stuttering rather than children.As the small number of available studies did not enable us to conduct a stratified analysis, further examination is needed to examine how altered structural integrity and FC patterns change with development.Third, because the current results may primarily reflect adults with stuttering, the heterogeneity within the population may be potentially more problematic than for children.Previous studies suggested that (mal)adaptive and compensatory neural changes through years of stuttering experience may increase heterogeneity in several cognitive and psychological domains, including social anxiety, in adults with stuttering (Ibraheem and Quriba, 2014;Doneva, 2020;Iverach et al., 2018).Long-term longitudinal measurements over the development period (Chow et al., 2023) may be a promising approach for elucidating the neural mechanisms underlying such individual differences in adaptation and compensation.Lastly, we focused on seed-based FC studies for the resting-state and task-based FC meta-analyses.Because the selection of seeds depends on researchers' hypotheses, our analyses do not allow us to exhaust possible FC alterations throughout the brain.Therefore, critical FC alterations in developmental stuttering may exist that the present analysis failed to detect.Such critical yet underexamined FCs may be revealed by future studies using data-driven approaches such as ICA.However, the data-driven FC approaches are less common than seed-based analyses in both resting-state and task-based FC studies of stuttering, which makes the latter approach more appropriate for the present meta-analysis.

Conclusions
To our knowledge, this study provides the first evidence from metaanalyses following the PRISMA guideline, aimed at comparing structural integrity of white matter and FC between individuals with developmental stuttering and fluent control.Our meta-analysis revealed that the PWS had reduced FA values in the left SLF, and resting-state FC in the cortico-striatal pathway, while enhanced task-based FC in pathways that are involved in controls of speech and non-speech sequential motor and cognitive processes.Functional network characterization of the altered FCs revealed that the distributions of altered resting-state and increased task-based FCs were largely distinct.However, both conditions were common in that altered FCs were distributed over the somatomotor and striatum/thalamus networks.These observations indicate that developmental stuttering is characterized by structural and functional alterations in multiple brain networks that support speech fluency or sequential motor processes, including cortico-cortical and subcortical connections.

Declaration of Competing Interest
The authors declare that they have no competing financial interests or personal relationships that could affect the results of the research described in this manuscript.

Fig. 1 .
Fig. 1.Flowchart of the literature research and selection of eligible studies.Abbreviations: FC, functional connectivity; ICA, independent component analysis; TBSS, Tract-Based Spatial Statistics.

Fig. 2 .
Fig. 2. Meta-analysis of altered FA in PWS.Two clusters exhibiting reduced FA values in white matter pathways in the PWS compared to the PWNS are visually presented in the cold color scale (uncorrected p < 0.005, cluster size > 10).Abbreviations: SLF, superior longitudinal fasciculus.

Table 2
Regions showing group difference in analysis of resting-state functional connectivity (FC).

Table 3
Regions showing group difference in analysis of task-based FC.
The significant difference between PWS and PWNS at uncorrected p < 0.005 and cluster size > 10 voxels.Clusters in bold indicate clusters that survived the TFCE correction (p < 0.05, cluster size > 10 voxels).Numbers in parenthesis indicate the cluster size when the TFCE correction is applied.The columns labeled "Region" denote two terminal regions in altered functional connections: a region of significant group difference and seed.Classifications of the terminal regions into functional network are shown in the columns "Functional network".Abbreviations: L, Left; R, Right; ACC, anterior cingulate cortex; aLTC, anterior lateral temporal cortex; AN, Auditory network; CON, Cingulo-Opercular network; CPT, K.Matsuhashi et al.