Abnormal neural response to phonological working memory demands in persistent developmental stuttering

Abstract

Persistent developmental stuttering is a neurological disorder that commonly manifests as a motor problem. Cognitive theories, however, hold that poorly developed cognitive skills are the origins of stuttering. Working memory (WM), a multicomponent cognitive system that mediates information maintenance and manipulation, is known to play an important role in speech production, leading us to postulate that the neurophysiological mechanisms underlying stuttering may be associated with a WM deficit. Using functional magnetic resonance imaging, we aimed to elucidate brain mechanisms in a phonological WM task in adults who stutter and controls. A right‐lateralized compensatory mechanism for a deficit in the rehearsal process and neural disconnections associated with the central executive dysfunction were found. Furthermore, the neural abnormalities underlying the phonological WM were independent of memory load. This study demonstrates for the first time the atypical neural responses to phonological WM in PWS, shedding new light on the underlying cause of stuttering.

Abbreviations

BA

Brodmann's area

HPWM

high‐load phonological WM

LPWM

low‐load phonological WM

PWS

people who stutter

SSI

stuttering severity instrument

WM

working memory

1. INTRODUCTION

Persistent developmental stuttering is a language disorder characterized by involuntary repetition, prolongation or block of speech sounds. It affects approximately 5% of preschool children and 1% of the adult population (Bloodstein & Ratner, 1995). Decades of neuroimaging studies have revealed that people who stutter (PWS) show abnormalities in large‐scale brain networks involving the motor/premotor cortex, the inferior frontal gyrus, the superior temporal gyrus, the basal ganglia, and the cerebellum (Braun et al., 1997; Budde, Barron, & Fox, 2014; Chang, Kenney, Loucks, & Ludlow, 2009; Chang & Zhu, 2013; Chang, Zhu, Choo, & Angstadt, 2015; Cykowski, Fox, Ingham, Ingham, & Robin, 2010; Fox et al., 1996; Lu et al., 2010; Neef, Hoang, Neef, Paulus, & Sommer, 2015; Sommer, Koch, Paulus, Weiller, & Büchel, 2002; Watkins, Smith, Davis, & Howell, 2008). Functional connectivity studies based on functional magnetic resonance imaging (fMRI) data have demonstrated altered functional connectivity in PWS within the basal ganglia‐thalamocortical loop (Chang & Zhu, 2013; Lu et al., 2010), the auditory‐motor loop (Chang & Zhu, 2013), the left inferior frontal‐premotor loop (Chang, Horwitz, Ostuni, Reynolds, & Ludlow, 2011), and the cerebello‐cerebral loop (Lu et al., 2012). These findings suggest that stuttering is associated with deficits in both the regional activation and interregional interaction necessary for speech production.

Cognitive theories postulate that stuttering originates from an ill‐developed cognitive‐linguistic system. For example, the covert repair hypothesis proposes that stuttering is derived from a phonetic planning defect (Postma & Kolk, 1993), and the EXPLAN theory attributes dysfluencies to the failure of timely projection from linguistic planning to motor execution (Howell & Au‐Yeung, 2002). These notions are supported by several lines of empirical evidence. Some studies reported that PWS have difficulties in speech planning during covert and overt speech production (Chang et al., 2009; Lu et al., 2009; Lu et al., 2010), while other studies demonstrated the phonological perception deficit in PWS (Lu et al., 2016; Pelczarski & Yaruss, 2014).

Working memory (WM) is one of the most essential components of human cognitive architecture, and it is of great importance to a range of complex cognitive activities, such as language, reasoning, and problem solving (Baddeley, 2003). According to Baddeley's influential model (Baddeley, 2003), the phonological loop is responsible for processing verbal/phonological information that comprises auditory‐based temporary storage and subvocal rehearsal processing. The phonological‐form store in WM is evidenced by the phonological similarity effect (sequences of phonologically similar sounds are remembered less well than sequences of dissimilar sounds) (Baddeley, 1992; Mueller, Seymour, Kieras, & Meyer, 2003). Subvocal rehearsal refers to the process of repeating the activated or task‐relevant phonological information to prevent rapid delay of memory trace, a finding that has been supported by the word length effect (the memory performance of words with many syllables is worse than words with fewer syllables) (Baddeley, Thomson, & Buchanan, 1975). The phonological‐specific storage is mediated by a modality‐general module, the central executive. Other models have advocated that the phonological WM comprises two distinct phonological buffers. The input buffer is engaged in speech perception, while the output buffer is involved in speech production (Jacquemot & Scott, 2006). Previous studies have demonstrated a close relationship between phonological WM and speech production (Acheson, Hamidi, Binder, & Postle, 2011; Adams & Gathercole, 1995; Jacquemot & Scott, 2006; Page, Madge, Cumming, & Norris, 2007). For example, phonological WM engages in phonological encoding during speech production in which phonetic elements are constructed (Bajaj, 2007). As a result, a relationship between phonological WM deficiency and stuttering has been hypothesized (Bajaj, 2007), with empirical evidence from behavioral studies. Behavioral performance of nonword repetition has been found to be poorer in stuttering children (Anderson & Wagovich, 2010; Hakim & Ratner, 2004; Pelczarski & Yaruss, 2016) and adults (Byrd, McGill, & Usler, 2015). The central executive component is also dysfunctional in PWS, as demonstrated in previous behavioral research showing reduced behavioral performance in dual tasks (Jones, Fox, & Jacewicz, 2012; Smits‐Bandstra, De Nil, & Rochon, 2006).

Brain systems underlying WM have been extensively investigated using various WM task paradigms (Collette & Van der Linden, 2002; Owen, Mcmillan, Laird, & Bullmore, 2005). The bilateral dorsal prefrontal gyrus (Nee et al., 2012; Smith & Jonides, 1999), the left inferior frontal gyrus, the insula and the left inferior parietal lobule have consistently been reported to be activated in WM processing. Specifically, the bilateral dorsal prefrontal gyrus (Nee et al., 2012; Smith & Jonides, 1999) and the insula (Menon & Uddin, 2010) have been suggested to be involved in the central executive of WM. The left inferior frontal gyrus and the left inferior parietal lobule are thought to support the processing of subvocal rehearsal (Paulesu, Frith, & Frackowiak, 1993; Smith, Jonides, Marshuetz, & Koeppe, 1998) and the storage of phonological content (Crottazherbette, Anagnoson, & Menon, 2004; Jonides et al., 1998) and order (Kalm & Norris, 2014), respectively. In addition, the left posterior superior temporal gyrus (the planum temporale) was also found to be engaged in phonological WM during speech production (Mcgettigan et al., 2011). In particular, it has been suggested to serve the representations of auditory or audiomotor “templates” of speech information (Papoutsi et al., 2009; Warren, Wise, & Warren, 2005). Finally, subcortical regions, including the basal ganglia (Chang, Crottaz‐Herbette, & Menon, 2007; Lewis, Dove, Robbins, Barker, & Owen, 2004), the cerebellum (Kirschen, Chen, Schraedley‐Desmond, & Desmond, 2005; Marvel & Desmond, 2010b) and the midbrain (D'Ardenne et al., 2012) were also found to be associated with WM processing. Beyond functional localization, functional connectivity (Honey et al., 2002; Woodward et al., 2006) and effective connectivity (Honey et al., 2002; Ma et al., 2012) were applied to examine how the interregional interaction supports WM processing. Using psychophysiological interaction (PPI) analysis in a delayed‐encoding recognition task, an fMRI study found that the functional connectivity between the dorsal premotor area and the ipsilateral dorsolateral prefrontal cortex was required to combine two distinct operations of segmenting and binding in verbal WM, indicating the neural pathway of the central executive (Abe et al., 2007). Likewise, using dynamic causal modeling (DCM) based on fMRI data, Ma et al. (2012) found that the connection from the left posterior parietal cortex to the inferior frontal gyrus was enhanced with the increase in the digit load, while lower loads inhibited the connection from the right posterior parietal cortex to the left anterior cingulate gyrus. These results indicated that the frontal–parietal network modulated the WM load related to information processing or error monitoring (Ma et al., 2012). In addition, the roles of the connectivity between subcortical structures and the neocortex in WM have also been investigated. For instance, an fMRI study examined the functional connectivity of the basal ganglia in the encoding, maintenance and response phases using a Sternberg WM task. It was found that WM load mediated the connectivity of the left anterior caudate with the prefrontal cortex and the posterior parietal cortex in different phases of WM processing (Chang et al., 2007). Moreover, the contribution of cerebro‐cerebellar circuit to WM has been revealed in a Sternberg verbal WM task. The dorsal cerebellar dentate was found to be co‐activated with the supplementary motor area (SMA) during encoding stage, presumably engaging the process of articulatory processing. While, the ventral cerebellar dentate was co‐activated with anterior prefrontal cortex and the pre‐SMA during the retrieval stage, and this functional coupling was thought to support the manipulation of information in WM (Marvel & Desmond, 2010a). Phonological WM is obligatorily involved in WM tasks for linguistic material, and thus these findings described above suggest the neural circuits of phonological WM.

In the present fMRI study, we sought to identify the neural abnormalities specific to phonological WM deficits in PWS in the context of speech production, as stuttering is a speech‐specific disorder. An overt syllable repetition task was used, which requires a participant to memorize and manipulate phonological information simultaneously, consistent with the concept of WM. Specifically, the phonological memory was investigated by temporarily storing a string of syllables, while the manipulation was examined by requiring a reordering of the syllables. Moreover, we utilized a simple speech production task without WM demand to control the brain activation of speech production, thus allowing us to identify neural correlates specific to phonological WM. In the domain of WM, load‐sensitivity has been considered to be a critical property of the brain areas supporting storage function (Feredoes & Postle, 2007). Thus, in this study, we investigated the neural responses to phonological WM under different memory loads that was parameterized by the extent of the phonological similarity of the stimuli.

2. MATERIALS AND METHODS

2.1. Participants

Thirty‐eight adults participated in the fMRI scans: 19 adults who stutter (16 males and 3 females; mean age = 26.26 years, range from 21 to 35 years) and 19 normal controls (16 males and 3 females; mean age = 24.84 years, range from 22 to 31 years). All stuttering participants started stuttering before age 12 and had not undergone treatments during the year prior to this study. We evaluated stuttering severity by collecting speech samples of spontaneous speech elicited through an interview and reading aloud. Based on the SSI‐3 (Riley, 1994), the frequency and duration of stuttering, as well as any physical concomitants related to stuttering in the speech samples, were evaluated by two independent judges. Typical symptoms were considered as stuttering events, including repetitions (multiple‐character, whole‐character), pauses between characters and prolongation of the rhyme part of the characters (Jiang, Lu, Peng, Zhu, & Howell, 2012). The stuttering severity ranged from very mild to very severe. The intraclass correlation coefficient for the two independent judges' evaluations was high (Cronbach's alpha = 0.99). All participants were physically healthy and had no history of neurological disease or psychiatric disorder. They were native Chinese speakers and were right‐handed as assessed by a handedness inventory (Snyder & Harris, 1993). Detailed clinical and demographic characteristics of the participants are presented in Table 1. The study was approved by the Institutional Review Board of Beijing MRI Center for Brain Research. The methods were carried out in accordance with the approved guidelines. Prior to the experiment, written informed consent was obtained from each participant.

Table 1.

Demographic characteristics of the two groups

	Stuttering (n = 19)	Control (n = 19)
	Mean (SD)	Mean (SD)	p value
Age (in years)	26.26 (3.43)	24.84 (2.79)	.17
Handedness	All right‐handed	All right‐handed
Education (years)	14.16 (3.17)	15.26 (2.42)	.235
SSI‐3	25.74 (7.01)	n/a
Phonological WM
Forward digit span	8.79 (1.4)	9.21 (1.31)	.345
Backward digit span	5.84 (1.7)	7 (1.61)	.041

Independent two‐sample t test was used. SD = standard deviation; s = second; n/a = not applicable.

2.2. Experimental design

We used a syllable repetition task to examine phonological WM during speech production. Participants overtly repeated three syllables (of phonological similarity or dissimilarity) in reverse order to that in which they were presented. The stimuli were Chinese monosyllables, i.e., “du” or “men,” and the sequence of syllables had no semantic interpretation. All stimuli are included in Supporting Information Table S1. The requirement of reversal places a demand on the manipulation of syllable information. The phonological similarity condition comprises three syllables with the same onset (e.g., “du,” “dan,” “da”), while phonological dissimilarity condition includes three distinct syllables (e.g., “da,” “hou,” “qu”). We adopted this construction of the conditions of phonological similarity based on the observation that the initial phoneme would be encoded during the production of Chinese syllables (Qu, Damian, & Kazanina, 2012). Phonological similarity stimuli were defined as a high‐load phonological WM (HPWM) condition, and phonological dissimilarity stimuli were considered a low‐load phonological WM (LPWM) condition based on the phonological similarity effect (Mueller et al., 2003). In the control condition, participants were required to repeat three times the same syllable that controlled the activation of general speech production in the experimental conditions. Participants were required to repeat each syllable individually with intervals, the same way in which the stimuli were presented.

A block design was employed, consisting of four blocks of HPWM, LPWM and the baseline condition, which alternatively appeared in a pseudorandom way. Each block included a 2‐s instruction and eight task trials. In each trial, syllables were exposed for 1.7 s followed by a 2.3‐s response period. Three blocks of central fixation, each with a duration of 12 s, were also interspersed among the task and control blocks as a “rest” condition. There was a 6‐s fixation at the beginning of the scan. The total scanning time of the task was 450 s.

2.3. Imaging acquisition

MRI data were continuously collected on a 3 T Siemens MRI scanner at the Beijing MRI Center for Brain Research of the Chinese Academy of Sciences. Functional images were acquired using a blood oxygen level‐dependent (BOLD)‐sensitive gradient echo‐plane‐image (EPI) sequence (TR = 2000 ms, TE = 30 ms, slice thickness = 4 mm, in‐plane resolution = 3.4 × 3.4 mm and flip angle = 90°). Thirty‐three axial slices were collected. High spatial resolution anatomical images were acquired using a T1‐weighted, magnetization‐prepared rapid acquisition gradient echo sequence (TR = 2,600 ms, TE = 3.02 ms, slice thickness = 1 mm, in‐plane resolution = 1.0 × 1.0 mm and flip angle = 8°).

2.4. Data analysis

2.4.1. Preprocessing

Image preprocessing and statistical analyses were processed using SPM8 (http://www.fil.ion.ucl.ac.uk/spm, Wellcome Department of Cognitive Neurology, University College London, London, UK). First, the EPI images were corrected for head motion, and the realigned images were coregistered to the high‐resolution T1 anatomical image, which were then normalized into Montreal Neurological Institute (MNI) stereotactic space via the unified segmentation approach. The resulting warp parameters were applied to the spatial normalization of EPI images into the MNI space with a resolution of 2 × 2 × 2 mm cubic voxels. The normalized EPI images were then smoothed with an isotropic Gaussian kernel of 8 mm full‐width at half‐maximum.

2.4.2. Whole‐brain activation analysis

For each group, activation maps contrasting phonological WM conditions (HPWM and LPWM, respectively) to baseline were generated for each participant using the general linear model, in which the time series were convolved with the canonical hemodynamic response function. To minimize movement‐related variance, head movement parameters (the six movement parameters obtained during realignment) were incorporated as nuisance covariates. The data were high‐pass‐filtered at 128 s. Group analyses were performed using a random‐effects model. First, the whole‐brain activation maps for phonological WM were computed for PWS and controls separately by using a one‐sample t test. Then, a 2 (groups: PWS vs. controls) × 2 (conditions: HPWM vs. LPWM) repeated‐measures analysis of variance (anova) analysis was conducted. The voxel‐wise threshold was set at p < .05 corrected for multiple comparisons by using the false discovery rate (FDR) correction, with a minimum cluster extent of 20 contiguous voxels. Brain regions were estimated from the Talairach atlas (Talairach & Tournoux, 1988).

To further validate the results of group comparisons, we conducted a correlation analysis between brain activation and out‐scanner behavioral scores of WM. To do this analysis, contrast estimates (linear combination of β estimates) were extracted from functional regions of interest (ROIs) showing group differences in activation analysis for each participant. These results were then correlated with the scores of digit span.

2.4.3. ROI analysis

In addition to the whole‐brain analysis, ROI analysis was also performed. The bilateral inferior frontal gyrus (pars opercularis and triangularis), middle frontal gyrus, medial frontal cortex, insula, and inferior parietal lobule were included as ROIs that belong to the core neural networks of phonological WM (Chen & Desmond, 2005; Paulesu et al., 1993; Rottschy et al., 2012; Smith & Jonides, 1999). ROIs were anatomically defined using the AAL atlas (Tzourio‐Mazoyer et al., 2002). The mean contrast estimates for phonological WM > baseline contrasts were extracted from these ROIs for each participant. A group (PWS vs. controls) by condition (HPWM vs. LPWM) repeated‐measures anova analysis was conducted for each ROI. Statistical significance was set at p < .005 after Bonferroni correction for multiple comparisons.

2.4.4. Functional connectivity analysis

A seed‐to‐voxel functional connectivity analysis was performed using the CONN‐fMRI toolbox in SPM8 (Whitfield‐Gabrieli & Nieto‐Castanon, 2012). The seeds were selected using two criteria. First, the regions would be significantly activated across the WM conditions and participant groups that were searched by a conjunction analysis (voxelwise threshold of p < .05, FDR corrected). In addition, all the seeds were constrained within the neural networks known to be involved in phonological WM (Chen & Desmond, 2005; Paulesu et al., 1993; Rottschy et al., 2012; Smith & Jonides, 1999). Following these criteria, several regions were included: the bilateral insula (left: x = −30, y = 20, z = 6; right: x = 30, y = 24, z = 6 in Talairach coordinate), the SMA (left: x = −5, y = 14, z = 49; right: x = 2, y = 17, z = 47), the middle frontal gyrus (left: x = −32, y = 1, z = 55; right: x = 30, y = 3, z = 51), the left inferior parietal lobule (x = −38, y = −41, z = 43), and the bilateral cerebellar declive (left: x = −24, y = −61, z = −22; right: x = 28, y = −59, z = −21). These regions were created as sphere ROIs with a radius of 8 mm. Following the CompCor strategy (Behzadi, Restom, Liau, & Liu, 2007), the effects of nuisance covariates, including fluctuations in blood oxygen level‐dependent signal from cerebrospinal fluid and white matter and their derivatives were reduced. In addition, the main effects of task and the head motion noises were taken as confounds. Data were high‐pass filtered at 0.008 Hz (Yang, Jia, Siok, & Tan, 2017). Bivariate correlation coefficients that compute the correlation between the time courses of the signals in the ROIs and the signals in the rest voxels were acquired and were then transformed to Fisher's Z‐scores. As a result, for each participant, we acquired the functional connectivity maps contrasting phonological WM with baseline for each ROI and each condition. For the second level analysis, we first examined the interaction effect between group and WM load using a repeated‐measures anova analysis. Then, we examined the main effects of group (combing HPWM and LPWM conditions) using a two‐sample t test. Nonparametric permutation tests with 1,000 permutations were applied for between‐group statistical analysis using an uncorrected voxel‐wise threshold of p < .001 and an FDR‐corrected cluster‐mass threshold of p < .05.

To further elucidate the nature of group differences in functional connectivity, we conducted a post hoc analysis on the correlation coefficients for each group independently using a one‐sample t test. In addition, we computed Pearson's correlation coefficients between connectivity strength and behavioral scores as well as stuttering severity instrument (SSI) across the two groups of participants.

2.4.5. Correspondences between brain activation and functional connectivity

We also explored the correspondences between activation and functional connectivity results. Brain regions that showed group differences in both activation and functional connectivity analyses were defined as ROIs. Then, the mean contrast estimates and connectivity coefficients were extracted from these ROIs for the HPWM condition and the LPWM condition, respectively, in all participants, which were entered into a Pearson's correlation analysis.

2.5. Behavioral assessments

Following the fMRI session, behavioral assessments of phonological WM were conducted. A digit span task was adopted in which participants were asked to repeat digits both forward and backward. This task has been widely used in measuring phonological WM (Gathercole & Adams, 1993; Jeffries & Everatt, 2010; Kormos & Safar, 2008; Silver, Feldman, Bilker, & Gur, 2003). It not only taps the storage of phonological form of digits, but also engages the operations of order of phonological information, consistent with the definition of phonological WM. The length of the digit strings ranged from 3 to 12 digits in the forward direction, and from 3 to 10 digits in the backward direction. The test was terminated when participants failed in two consecutive trials of the same length, and the WM span was considered to be the number of digits in the last successful trials.

3. RESULTS

3.1. Behavioral results

Because one control participant failed to finish the backward digit span test, the result of the WM span was based on data from 19 PWS and 18 controls. The results indicated that compared to controls, PWS showed a reduced backward WM span [t(35) = −2.12, p = .041] but a comparable forward WM span [t(36) = −0.96, p = .345] (Table 1).

3.2. fMRI results

3.2.1. Whole‐brain analysis

The whole brain‐based analysis showed that when contrasted with repetition of one single syllable, the repetition of three syllables in reverse order elicited a similar and widespread activation pattern in both PWS and controls, involving the bilateral superior, middle and inferior frontal gyrus, SMA, insula, superior temporal gyrus, inferior parietal lobule, basal ganglia, thalamus, and cerebellar culmen (Supporting Information Table S2, Figure 1). These regions are in good agreement with previous findings showing the neural correlates of the phonological loop and central executive in WM (Paulesu et al., 1993; Rottschy et al., 2012; Smith et al., 1998; Smith & Jonides, 1999).

Figure 1.

Brain regions with significant activation during phonological WM in both groups of participants (voxel‐wise p < .05, FDR corrected). PWS showed significant activation in the HPWM condition (a) and the LPWM condition (c); controls showed significant activation in the HPWM condition (b) and the LPWM condition (d). L = left and R = right [Color figure can be viewed at http://wileyonlinelibrary.com]

However, no significant activations were observed for the group‐by‐condition interactions or the main effects of group. To examine the possibility that group differences might have been missed, we used a more lenient threshold (voxel‐wise threshold of p < .001, uncorrected and cluster‐level threshold of p < .05, family wise error [FWE] corrected). In this analysis, we found that PWS exhibited greater cortical activity in the right inferior frontal gyrus extending to the middle frontal gyrus than that in fluent controls (peak at x = 51, y = 36, z = 11, BA 46). No other significant activations for the between‐group comparisons were detected at this threshold (Figure 2a,b).

Figure 2.

Group differences in brain activation during phonological WM (voxel‐wise p < .001, uncorrected, and cluster‐level p < .05, FWE corrected). (a) Lateral and axial views. (b) Mean contrast estimates for the two groups extracted from the right inferior frontal gyrus showing group differences. (c) Scatter plots of the partial correlation between brain activation of the right inferior frontal gyrus and the backward digit span after controlling for the forward digit span. L = left and R = right, *p < .05, **p < .01, ***p < .001 [Color figure can be viewed at http://wileyonlinelibrary.com]

Partial correlation analysis revealed a significant correlation between brain activity in the right inferior/middle frontal gyrus and the backward WM span after controlling for the forward WM span (r = −.35, p = .038), confirming the role of this area in WM deficit in PWS (Figure 2c). No significant correlation between brain activity in this region and SSI was found (p = .416), suggesting that dysfunction of this region might be a universal mechanism in PWS.

3.2.2. ROI analysis

The ROI analysis replicated the results of the whole‐brain analysis. Specifically, we found significant main effects of group in brain activation in the right pars opercularis [F(1,36) = 9.78, p = .003] and pars triangularis of the inferior frontal gyrus [F(1,36) =8.91, p = .005]. Post hoc pairwise comparisons indicated that PWS showed higher activation in the right inferior frontal gyrus than that in controls (pars opercularis: p = .003; right pars triangularis p = .005, Bonferroni correction). However, the interaction effect between group and load was not significant [pars opercularis: F(1,36) = 1.14, p = .292; pars triangularis: F(1,36) = 1.08, p = .306)], suggesting that the group differences were not modulated by phonological memory load (Figure 3).

Figure 3.

The results of ROI analysis. (a) Axial views of the ROIs. (b) Group differences in activity level (mean contrast estimates) during phonological WM. IFG_oper = the opercular region of the inferior frontal gyrus, IFG_tri = the triangular region of the inferior frontal gyrus, L = left and R = right, *p < .05, **p < .01, ***p < .001 [Color figure can be viewed at http://wileyonlinelibrary.com]

3.2.3. Functional connectivity analysis

Similar to the activation analysis, we did not find any significant interaction effect between group and WM load in the functional connectivity analysis. Thus, we examined the main effects of group in functional connectivity by combining the two WM conditions. It was found that compared to fluent controls, PWS showed weaker functional connectivity between the left cerebellar declive and the right inferior/middle frontal gyrus (PWS: r = −.13, p < .001; controls: r = .05, p = .021) and between the right insula and the midbrain/thalamus (PWS: r = −.05, p = .035; controls: r = .13, p < .001). In contrast, PWS exhibited greater functional connectivity between the left insula and the left superior and medial frontal gyrus (PWS: r = .09, p = .001; controls: r = −.09, p < .001) than that of the controls during phonological WM. Additionally, relative to the controls, PWS exhibited increased functional connectivity between the left inferior parietal lobule and the left superior temporal gyrus (extending to the left inferior frontal gyrus and insula) (PWS: r = .05, p = .032; controls: r = −.13, p < .001) (Figure 4, Table 2).

Figure 4.

The main effects of group in functional connectivity (voxel‐wise p < .001, uncorrected, cluster‐mass p < .05, FDR corrected). (a) Functional connectivity maps in which the controls showed greater connectivity than the PWS with the corresponding connectivity coefficients. (b) Functional connectivity maps in which the PWS showed greater connectivity than the controls with the corresponding connectivity coefficients. Cereb = cerebellum, MFG = middle frontal gyrus, Med.FG = medial frontal gyrus, IFG = inferior frontal gyrus, ins = insula, Midb = midbrain, IPL = inferior parietal lobule. L = left, R = right. *p < .05, **p < .01, ***p < .001 [Color figure can be viewed at http://wileyonlinelibrary.com]

Table 2.

Group differences in functional connectivity during phonological WM. Z‐score corresponds to the actual maximum pixel value within the brain region from the SPM (voxel‐wise, p < .001, uncorrected; cluster‐mass p < .05 FDR corrected). L = left and R = right, Cereb = cerebellum, MFG = middle frontal gyrus, IFG = inferior frontal gyrus, ins = insula, Midb = midbrain, Thala = thalamus, STG = superior temporal gyrus, med.FG = medial frontal gyrus, SFG = superior frontal gyrus

			Talairach
Seeds	Target regions	BA	x	y	z	z value	Mass
Control > stuttering
L Cereb	R MFG	10	38	38	15	4.74	1,940
		46	42	38	17	4.73
	R IFG	10	42	50	−1	4.25
		47	42	31	0	3.41
		45	48	27	−1	3.25
	R Ins	13	34	24	3	3.59
R ins	R Midb		6	−33	−8	4.79	765
	Thala		14	−25	−5	4.31
Stuttering > control
L IPL	L STG	22	−48	−8	−5	4.06	660
		38	−44	11	−9	3.73
	L IFG	47	−40	17	−9	4.05
	L Ins	13	−42	12	−1	3.53
L Ins	L Med.FG	10	−8	55	12	4.54	1,125
		9	−5	42	27	4.07
	L SFG	9	−15	48	29	3.8

However, we did not observe any significant correlations between functional connectivity and digit span and SSI.

3.2.4. Correspondences between brain activation and functional connectivity

The correlation analysis revealed that the activation of the right inferior frontal gyrus was negatively correlated with the functional connectivity between the right inferior/middle frontal gyrus and the left cerebellar declive in both WM conditions (HPWM: r = −.55, p < .001; LPWM: r = −.48, p = .003) (Figure 5).

Figure 5.

Scatter plots of the correlation between the activation in the right inferior frontal gyrus and its functional connectivity with the left cerebellum in the two WM conditions across the two groups of participants. IFG = inferior frontal gyrus, Cereb = cerebellum. *p < .05, **p < .01, ***p < .001 [Color figure can be viewed at http://wileyonlinelibrary.com]

4. DISCUSSION

This study, to the best of our knowledge, is the first to reveal the neural mechanisms underlying the phonological WM deficit in PWS, supporting the ill‐developed cognitive‐linguistic hypothesis of stuttering (Postma & Kolk, 1993). Specifically, in a phonological WM task that required simultaneous storage and manipulation of syllables, PWS exhibited greater activation in the right inferior frontal gyrus extending to the middle frontal gyrus than fluent controls, consistent with previous findings of increased brain activation in the right hemisphere in PWS (Chang et al., 2009; Fox et al., 1996; Fox et al., 2000; Neef et al., 2016; Preibisch et al., 2003). Post hoc analysis revealed significant negative correlation between the brain activity in the right inferior frontal gyrus and the backward digit span. Functional connectivity analysis illustrated altered connectivity associated with the phonological WM deficit in PWS, involving the connectivity between the right inferior/middle frontal gyrus and the left cerebellum and between the insula and midbrain and the left superior frontal gyrus. Furthermore, the neural abnormalities in PWS were identified regardless of memory load defined by the phonological similarity, suggesting that the phonological WM dysfunction is likely to be a general deficit in PWS.

First, the whole‐brain analyses and anatomically based ROI analysis indicated a right‐lateralized hyperactivity in the inferior frontal gyrus in PWS relative to that of fluent controls in both HPWM and LPWM conditions. In addition, the significant brain activation‐behavior correlation further confirmed the contribution of the hyperactivity of the right inferior frontal gyrus to the WM deficit in PWS. Abnormal activity in the right inferior frontal gyrus in PWS has been repeatedly reported in previous functional studies using other paradigms (Chang et al., 2009; Kell et al., 2009; Neef et al., 2016;Neef et al., 2017 ; Preibisch et al., 2003), and it has been thought to represent a compensatory mechanism in response to the left‐side impairment (Neef et al., 2017; Preibisch et al., 2003). The left inferior frontal gyrus (Broca's area) is known to be a neural substrate of rehearsal processing in WM (Smith et al., 1998; Smith & Jonides, 1999). The hyperactivity of the right homolog of Broca's area may represent a compensation for insufficient rehearsal in PWS (Weber‐Fox, Spencer, Spruill, & Smith, 2004). Alternatively, the right inferior frontal gyrus is critical for inhibition function in both cognitive and motor processes (Aron, Fletcher, Bullmore, Sahakian, & Robbins, 2003; Hampshire, Chamberlain, Monti, Duncan, & Owen, 2010; Xue, Aron, & Poldrack, 2008), such as suppression of interfering information (Garavan et al., 1999). In addition, the right inferior frontal gyrus has also been found to support proactive and reactive control in WM (Marklund & Persson, 2012). The syllable repetition task used in the current study required participants to concurrently memorize and reorder the syllables, for which the participants may have to inhibit the inclination to repeat the syllables in the same order as they were presented. Thus, the increased activity of the right inferior frontal gyrus in PWS may result from a greater commitment of central executive function of the WM. This interpretation is consistent with previous behavioral studies showing that PWS' performance is more affected by a concurrent task than that of fluent speakers in speech (Bosshardt, 2002) and nonspeech tasks (Smits‐Bandstra, De Nil, & Rochon, 2006).

Functional connectivity analysis elucidated a weaker connectivity between the right inferior/middle frontal gyrus and the left cerebellar declive in PWS than that in controls. Furthermore, we found a significantly negative correlation between the activation of the right inferior frontal gyrus and its connectivity with the left cerebellum, suggesting that hyperactivity in this region may compensate for its reduced functional connectivity with other regions. The cerebellum is an important neural signature of stuttering (Brown, Ingham, Ingham, Laird, & Fox, 2005; Budde et al., 2014; Chang et al., 2015; Yang, Jia, Siok, & Tan, 2016). Previous studies have reported alternated resting‐state functional connectivity of the left cerebellum in PWS (Lu et al., 2012; Yang et al., 2016). In addition, the structural connectivity of the left cerebellum has been found to be negatively correlated with the severity of stuttering (Sitek et al., 2016). However, how cerebellar abnormality contributes to stuttering remains unclear. In this sense, our data provide a new clue regarding the specific contribution of cerebellar abnormalities to stuttering from the perspective of phonological WM. Beyond the well‐known motor function, unequivocal evidence has implicated the role of the cerebellum in phonological/verbal WM (Chen & Desmond, 2005; Ravizza et al., 2006). The frontocerebellar loop has been suggested to be involved in carrying out subvocal rehearsal (Chen & Desmond, 2005), whereas projections from the temporoparietal cortex to the cerebellum were thought to be related to phonological storage (Desmond, Gabrieli, Wagner, Ginier, & Glover, 1997). As a consequence, the reduced connectivity between the right inferior/middle frontal regions and the left cerebellar declive probably affects the subvocal rehearsal in PWS. Conversely, the cerebellum has also been proposed to be associated with central executive in WM (Bellebaum & Daum, 2007). In light of the possible role of the right inferior frontal gyrus in inhibition control, the reduced functional connectivity between the left cerebellum and the right inferior/middle frontal in PWS could be considered as a neural basis for executive control dysfunction in PWS.

Another main finding of this functional connectivity analysis is that PWS exhibited altered functional connectivity of the insula during phonological WM. First, we found that PWS showed weaker functional connectivity between the right insula and the right midbrain (thalamus) than that in fluent controls. Previous studies consistently revealed abnormal activation in the right insula in PWS (Brown et al., 2005; Budde et al., 2014), which has been suggested to contribute to the aberrant phonological processing (Brown et al., 2005). Overactivity in the bilateral midbrain in PWS has also been reported previously (Watkins et al., 2008), but its specific role in stuttering remains unknown. With regard to WM, a meta‐analysis proposed that the bilateral insula is a part of the cognitive control network of WM, and it has specifically been thought to be related to controlling goal‐directed behavior through maintenance of task set (Dosenbach et al., 2007). Additionally, the midbrain dopamine neurons have been thought to serve in the updating of context information in WM (D'Ardenne et al., 2012). As a result, the disconnection between the right insula and the midbrain (thalamus) in PWS may indicate an inferior ability in sustaining and updating the task‐related requirements in WM. However, due to the limited spatial resolution of the current fMRI data, we can hardly be confident about the extent to which these subcortical structures contribute to the WM deficit observed in our study. In contrast, PWS showed stronger connectivity between the left insula and the left superior frontal gyrus (BA9) than that of the controls. As demonstrated in previous studies, the left dorsal prefrontal gyrus is responsible for the central executive (Bunge, Klingberg, Jacobsen, & Gabrieli, 2000; D'Esposito et al., 1995; Nee et al., 2012). The increased functional connectivity between the left insula and the left superior frontal gyrus implies that PWS allocate more sustainable attention resources to inspecting the phonological WM processing than do fluent speakers.

Finally, we found that PWS exhibited greater connectivity between the left inferior parietal lobule and the left superior temporal gyrus adjunct to the inferior frontal gyrus and insula. This finding appears to be inconsistent with previous studies that reported reduced functional connectivity in the left hemisphere in PWS relative to controls (Chang et al., 2011). One plausible explanation of this hyperconnectivity is based on the relationship between functional connectivity and structural connectivity (Hermundstad et al., 2013; Honey et al., 2009). It has been proposed that greater task‐related functional connectivity may compensate for the decline in the properties of structural connectivity in memory and attention processes (Hermundstad et al., 2013). Consistently, previous anatomical studies found that PWS exhibited reduced white matter integration in the left arcuate fasciculus (Sommer et al., 2002; Watkins et al., 2008), which anatomically connects these regions. Thus, the hyperfunctional connectivity between the inferior parietal lobule and the left superior temporal gyrus in PWS may result from the impairment of this structural pathway. However, considering the structural variation between PWS (Cai et al., 2014) and the lack of measurement of the structural connectivity in the same group of subjects, further structural connectivity studies are necessary to test the hypothesis of a structural connectivity deficit. Functionally, the left inferior parietal lobule has been found to support order processing in WM (Moser, Baker, Sanchez, Rorden, & Fridriksson, 2009; Papagno et al., 2017), and PWS have demonstrated deficits in order processing of speech and nonspeech stimuli (Smits‐Bandstra, De Nil, & Saint‐Cyr, 2006). Hypothetically, the increased connectivity between the left inferior parietal lobule and other regions may be due to the greater effort required by PWS for reordering the syllables in phonological WM.

Unexpectedly, we did not observe group differences in phonological WM load in brain activation and functional connectivity analyses that were manipulated by varying the similarity of phonological stimuli. The phonological similarity effect reflects phonological storage in the phonological loop (Baddeley, 2003). Therefore, the WM deficit in PWS might not be derived from difficulty with the temporary maintenance of the sounds, but in the manipulation of the stored phonological information instead. This account is consistent with the previous findings showing no significant brain activity for the interaction between memory and manipulation conditions, suggesting that the brain activations involved in the manipulation of information were independent of memory load in WM (Collette et al., 1999). Another possibility is that the manipulation of phonological similarity of syllables in our task was not sufficient to differentiate the phonological WM load. However, because participants' speech responses in the scanner had not been monitored, the behavioral difference between high and low load condition could not be examined to testify this possibility. Further studies that systematically manipulate the loads of executive control and phonological storage are necessary to study phonological WM in PWS.

5. LIMITATIONS

One limitation of this study is that the potential contributions of dysfluency in PWS to the group differences in neural correlates of WM have not been disentangled owing to the lack of speech recording during scanning. However, such contributions, if any, would be minimal based on the fact that the syllable repetition task is almost a stutter‐free condition for PWS as suggested by previous behavioral studies using similar paradigms (Hakim & Ratner, 2004; Lu et al., 2010).

Another limitation is that only an overt speech production task was used in the present study, and thus, we can hardly determine whether the WM deficit in PWS is specific to speech‐production. Two distinct buffers have been advocated in phonological WM: the phonological input buffer is essential for speech perception, and the phonological output buffer is necessary for speech production (Jacquemot & Scott, 2006). We used a recall task with overt repetition because it would involve both phonological buffers and their connections. Further studies employing other speech and nonspeech tasks are required to address this issue.

Finally, because adults who stutter may develop compensatory strategies for coping with stuttering, the results of our study may be unable to determine the cause and effect relationship between the phonological WM deficit and stuttering. The phonological WM dysfunction might be bidirectionally related to stuttering in nature (Bajaj, 2007). However, several behavioral studies have reported similar WM defects in young children close to the onset of stuttering (i.e., age 3–5) (Anderson & Wagovich, 2010; Hakim & Ratner, 2004), implying the causal role of a deficit in WM in stuttering.

6. CONCLUSIONS

In summary, our study demonstrated atypical neural mechanisms underlying phonological WM in PWS by demonstrating regional activation deficits and problematic connections within regions that support the subvocal rehearsal and central executive functioning, respectively. These findings support the cognitive‐linguistic hypothesis of stuttering (Postma & Kolk, 1993), providing new clues to the cause of stuttering.

CONFLICT OF INTEREST

The authors declare no competing financial interests.

Supporting information

Supplementary Table 1 The stimuli of syllables used in the WM task. The numbers in the first column indicate the tone of the syllables

Supplementary Table 2 Coordinates of activation peaks associated with phonological WM. Z‐score corresponds to the actual maximum pixel value within the brain region from the SPM. HPWM = high‐load phonological working memory, LPWM = low‐load phonological working memory, L = left and R = right, BA = Brodmann's area

Yang Y, Jia F, Fox PT, Siok WT, Tan LH. Abnormal neural response to phonological working memory demands in persistent developmental stuttering. Hum Brain Mapp. 2019;40:214–225. 10.1002/hbm.24366