Auditory rhythm discrimination in adults who stutter: An fMRI study

Abstract

Rhythm perception deficits have been linked to neurodevelopmental disorders affecting speech and language. Children who stutter have shown poorer rhythm discrimination and attenuated functional connectivity in rhythm-related brain areas, which may negatively impact timing control required for speech. It is unclear whether adults who stutter (AWS), who are likely to have acquired compensatory adaptations in response to rhythm processing/timing deficits, are similarly affected. We compared rhythm discrimination in AWS and controls (total n=36) during fMRI in two matched conditions: simple rhythms that consistently reinforced a periodic beat, and complex rhythms that did not (requiring greater reliance on internal timing). Consistent with an internal beat deficit hypothesis, behavioral results showed poorer complex rhythm discrimination for AWS than controls. In AWS, greater stuttering severity was associated with poorer rhythm discrimination. AWS showed increased activity within beat-based timing regions and increased functional connectivity between putamen and cerebellum (supporting interval-based timing) for simple rhythms.

1. Introduction

Developmental stuttering, characterized by an impaired rhythmic flow of speech, affects approximately 5% of preschool-age children and 1% of adults. Most commonly, the speech stream is disrupted by frequent occurrences of disfluencies such as sound or syllable repetitions, sound prolongations, or silent blocks. Theoretical models propose that stuttering results from poor auditory-motor integration (e.g., Directions Into Velocities of Articulators [DIVA], (Andrews et al., 1982; Bohland et al., 2010; Guenther, 2006, 2016; Kent, 1984). Integration between cortical motor and auditory regions is modulated by subcortical structures such as the thalamus and basal ganglia (Alexander et al., 1986) that comprise the basal ganglia thalamocortical (BGTC) network. Situated within this network are critical structures supporting temporal processing, particularly for intrinsic or internal timing, including the basal ganglia (BG) structure putamen, and the supplementary motor area (SMA), auditory, and premotor cortical regions (Grahn, 2009; Grahn & Brett, 2007; Grahn & McAuley, 2009) with the SMA and putamen forming a ‘main core timing network’ (Merchant et al., 2013).

Impaired function in multiple nodes and connections within the BGTC network has been hypothesized to be associated with stuttering pathophysiology (Alm, 2004; Etchell et al., 2014; Giraud, 2008; Kell et al., 2009; Watkins et al., 2008) for recent reviews, see (Chang et al., 2018; Chang & Guenther, 2019). Supporting this relationship, the BGTC network has been argued to facilitate speech perception and production by enabling precise prediction and timing of speech movements (Schwartze & Kotz, 2013). From this perspective, the ability to generate an internal beat (i.e., an intrinsically generated periodic timing signal) in the absence of an external rhythm is viewed as important for guiding the timing of fluent speech (Alm, 2004; Etchell et al., 2014). Thus, disruptions in internal beat generation may have a negative impact on producing fluent speech. Consistent with this internal beat deficit hypothesis, variable and/or poorer performance on speech and non-speech rhythm production and perception tasks that involve a temporal component such as tapping, clapping, and rhythm discrimination have been demonstrated by children who stutter (CWS) relative to their fluent peers ((Chang et al., 2016; Falk et al., 2014; Howell, P., Au-Yeung, J., & Rustin, L., 1997; Olander et al., 2010; Westphal, 1933; Wieland et al., 2015)but see (Hilger Allison I. et al., 2016) for reanalysis of Olander et al., 2010). Previous research into speech and nonspeech timing in adults who stutter (AWS), however, has resulted in inconsistent and mixed findings. For example, while some studies find poorer and/or more variable performance (Archibald & De Nil, 1999; Borden, 1983; Cooper & Allen, 1977; Hulstijn et al., 1992; Kleinow & Smith, 2000; Max & Gracco, 2005; Smith & Kleinow, 2000; Smits-Bandstra et al., 2006; Zelaznik et al., 1997), not surprisingly other studies fail to find differences (Max & Yudman, 2003; Neef et al., 2011; Zelaznik et al., 1994). One strong possibility for these discrepancies is the known heterogeneity of AWS in manifestation of language, motor, and psychosocial aspects of the disorder (Smith & Weber, 2017) encompassing the degree and type(s) of disfluency, situational effects, and overall experience of stuttering.

Despite this heterogeneity, one near constant finding in the literature is the well-established phenomenon that people who stutter can become temporarily fluent under conditions that include an external pacing signal, such as when speaking with a metronome or during choral speech (Adams & Ramig, 1980; Park & Logan, 2015) sometimes termed the “rhythm effect” (Azrin et al., 1968; Frankford et al., 2021). These external timing cues decrease stuttering (i.e., increase fluency) presumably because the speaker is able to rely less on a hypothesized faulty internal timing network when external pacing is provided. The use of external timing cues in stuttering is also associated with promoting ‘normalized’ brain activity patterns (i.e., similar to activity patterns found in fluent speakers) in speech-motor and auditory regions (De Nil et al., 2003; Giraud, 2008; Kell et al., 2009; Neumann et al., 2005; Toyomura et al., 2011, 2015). Further, while external timing, where the time intervals are externally specified (e.g., marked by an auditory or visual stimulus), has been argued to rely more on the cerebellum and premotor cortex, internal timing, such as the internal generation of a periodic beat, has been argued to be supported by core timing related cortical and striatal structures within the BGTC network (Alm, 2004; Coull et al., 2013).

These proposed internal and external timing networks overlap with a theoretical distinction that has been made in the literature between beat-based and interval-based timing mechanisms, respectively. Beat-based timing involves a process of stimulus-driven entrainment that establishes the persistent (internal) representation of a periodic beat and the relative encoding of time intervals within a rhythm (Jones, 1976; McAuley & Jones, 2003; Povel & Essens, 1985; Schulze, 1978). In contrast, interval-based timing (also called duration-based timing) relies on encoding the absolute time intervals between successive events in a sequence to represent a rhythm (R. B. Ivry & Hazeltine, 1995; Keele et al., 1989; Pashler, 2001). Evidence that beat- and interval-based timing engage distinct neural mechanisms comes from a number of sources (Breska & Ivry, 2018; Grahn & Brett, 2009; Grahn & McAuley, 2009; Grube, Cooper, et al., 2010; Grube, Lee, et al., 2010; Nozaradan et al., 2017; O’Boyle et al., 1996; Teki et al., 2011). The SMA and putamen within the BGTC network, in particular, have been shown to support beat-based timing, while interval-based timing is supported more robustly by the cerebellum. This distinction is further strengthened by an apparent functional dissociation revealed through research focusing on disorders that selectively impair the BG (e.g., Parkinson Disease [PD]) or cerebellum (e.g., cerebellar degeneration). Namely, findings suggest that beat-based timing is preserved, and interval-based timing is affected in patients with cerebellar dysfunction, and vice versa in patients with PD (Breska & Ivry, 2018; Grube, Cooper, et al., 2010).

Research into the neural bases of stuttering has revealed subtle abnormalities in speech motor and auditory brain regions, including both the BG and cerebellum (for a summary, see (Chang & Guenther, 2019; Etchell et al., 2014)). These findings motivated investigations into rhythm processing in children who stutter (CWS). Of particular relevance to the present study, Wieland et al. (2015) found that CWS exhibit poorer rhythm discrimination compared to controls especially for complex rhythms where the acoustic onsets defining the rhythmic pattern do not consistently mark out a periodic beat, but can be mentally ‘filled in” by the listener. These findings are consistent with the internal beat deficit hypothesis, suggesting the presence of a beat-based timing deficit in CWS.

In a follow up study to Wieland et al. (2015), Chang and colleagues (Chang et al., 2016) compared the strength of resting state functional connectivity in the BGTC network in CWS and controls in relation to the observed behavioral deficits in rhythm discrimination. Results showed that CWS had weaker BGTC network connectivity than controls. Moreover, better rhythm discrimination in controls was associated with greater functional connectivity between the putamen and other regions of the BGTC network (e.g., SMA, primary motor cortex [PMC], and superior temporal gyrus [STG]) whereas the relationship between BGTC network connectivity and rhythm discrimination performance was absent in CWS. These results are consistent with earlier work by Chang and Zhu (2013) who also reported attenuated resting state functional connectivity within the BGTC network (associated with beat-based timing) including between the putamen and SMA in CWS compared to controls.

The observation of poorer rhythm discrimination in CWS compared to age-matched controls suggests that atypical rhythm perception may contribute to the development or manifestation of childhood stuttering and its potential persistence into adulthood. From this perspective, our work investigating rhythm perception deficits in CWS (Chang et al., 2016; Wieland et al., 2015) raises significant questions about rhythm perception and its neural correlates in adults who stutter. Adults who stutter (AWS) are by definition chronic stutterers whose stuttering has persisted since childhood, who are more likely to have adopted compensatory strategies in response to attenuated functional connectivity in the BGTC network present during childhood and associated deficits in beat-based timing. One potential compensatory strategy may be a shift from engaging beat-based timing mechanisms toward interval-based timing mechanisms. Studies conducted with adults have the potential to reveal potential compensatory behavioral mechanisms or neural adaptations that may have developed over time in AWS, including whether or not they appear to be effective or maladaptive.

In the present study, AWS and age-matched controls participated in a functional magnetic resonance imaging (fMRI) study while completing a version of the rhythm discrimination task from Wieland et al. (2015) and Chang et al. (2016). Whereas the Chang et al. (2016) examined the rhythm discrimination performance outside the scanner that was examined in relation to resting state functional connectivity in children who stutter, in the current study, we examined performance on the rhythm discrimination task during fMRI; as such, our fMRI analyses involved both task-based activation and task-based connectivity analyses. Two general inter-related questions were of interest.

First, do AWS exhibit poorer rhythm discrimination performance relative to age-matched controls, as was previously observed for CWS? Based on the perspective that rhythm perception deficits are a characteristic of stuttering, regardless of age or neuroplastic changes associated with years of stuttering, we hypothesized that rhythm discrimination would be worse for AWS compared to controls. Further, in line with the internal beat deficit hypothesis, we hypothesized that reduced rhythm discrimination performance would potentially be more apparent for (or specific to) complex rhythms, which taxes internal timing ability to a greater extent than simple rhythms. We also further considered the hypothesis that rhythm discrimination performance would be correlated with stuttering severity; that is, adults who stutter with more severe stuttering would be those that have the greatest reduction in rhythm discrimination performance. An alternative possibility is that there might be no significant group differences in behavioral performance due to compensatory strategies acquired by AWS. Such compensatory strategies would have been less evident when examining children who stutter who have only been stuttering for shorter periods relative to adults who stutter.

Second, what are the neural correlates of rhythm discrimination in AWS, and in particular, are any differences in rhythm discrimination performance in AWS relative to controls supported by overlapping or different brain networks? It is plausible that AWS, similar to CWS, exhibit decreased brain activity and/or functional connectivity in the BGTC network regions supporting internal beat generation and processing beat based rhythms including putamen and SMA. We hypothesized that significant group differences in rhythm and timing network activity would be supporting evidence for a core deficit in beat-based timing in stuttering. In line with theoretical perspectives proposed by Alm (2004)that posit that a core deficit in BG-SMA system in stuttering is compensated for via cerebellar-lateral premotor connectivity, we further expected that if there is a shift from beat-based to interval-based timing for rhythm discrimination in AWS, we might expect greater involvement of the cerebellum compared to controls (See (Alm, 2004) for a theoretical account consistent with this argument). Namely, we hypothesized that AWS may rely on absolute representations of duration or interval-based timing regions/networks that are supported by the cerebellum, compensating for atypical functional connectivity of the BGTC networks that would normally support rhythm discrimination, consistent with their functional differentiation in the literature (Grube, Cooper, et al., 2010; R. Ivry, 1993; Teki et al., 2011).

2. Methods

2.1. Participants

Eighteen adults who stutter (AWS), ages 18 – 53 years (M = 29.4, SD = 11.4; n = 6, female) and 18 adults who do not stutter (controls), ages 18 – 44 years (M = 25.3, SD = 6.9; n = 6, female) participated in the study. All participants were right-handed, monolingual, native speakers of English. Participants did not report any developmental, neurological, or psychiatric conditions (other than stuttering in the AWS group) and they were not taking any medication affecting the central nervous system. Participants had similar numbers of years of education (AWS, M = 15.9, SD = 2.9; Control, M = 15.5, SD = 1.8) and years of formal musical training (AWS, 0 – 13.5 years, M = 4.1, SD = 4.4; Control, 0 – 10 years, M = 4.3, SD = 4.0). Table S1 shows a comparison of the two groups on measures of expressive vocabulary, receptive vocabulary, articulation, and working memory. AWS participants had slightly higher expressive vocabulary scores than Controls (EVT-2: AWS, M = 115.1; Control, M = 108.1; t(34) = 2.00, p = 0.05), but the groups did not differ on any of the other measures. Stuttering severity for AWS participants was determined according to the Stuttering Severity Instrument 4^th edition (2009) by a certified speech-language pathologist. Measures of stuttering frequency and duration as well as any physical concomitants associated with stuttering during a speech sample were incorporated into a composite stuttering severity index (SSI-4), resulting in composite SSI-4 scores of very mild (n = 3), mild (n = 7), moderate (n = 4), and severe (n = 4) for participants in the AWS group. Control participants did not exhibit stuttering and did not report a personal or family history of stuttering. Participants were recruited from the East Lansing, MI area and were compensated for participation. All research procedures were approved by the Michigan State University Institutional Review Board.

2.2. Stimuli

Auditory rhythms were 12 simple and 12 complex rhythms selected from a larger set of rhythms (Grahn & Brett, 2009). Rhythms were 5, 6, or 7 intervals long and all intervals within a rhythm were integer multiples of a base time unit, notated in Table S2 by a ‘1’. Notated values of 2, 3, and 4 indicate that the inter-onset-intervals were 2x, 3x, or 4x the duration of the base time unit, respectively. The base time unit varied randomly from trial to trial taking on values of 220, 245, or 270 ms. The frequency of the tones marking the rhythms was fixed within a trial, but also varied randomly from trial to trial, taking on one of six values: 294, 353, 411, 470, 528, or 587 Hz.

Importantly, Standard and Different versions of a rhythm - Simple or Complex - consisted of the same set of interval durations, but in a different order (e.g., 31413 vs. 31431), providing for control over auditory sequence composition. In other words, compared with the standard sequence within each rhythm condition (Simple or Complex), the ‘different’ variant of each rhythm involved swapping the order of a pair of adjacent intervals, which were the same as the different variants in Grahn and Brett ((2009)). For simple rhythms, the intervals were organized into a sequence so that tones occurred every 4 base time units, thereby consistently reinforcing an explicit (period) beat; consistent with prior studies these simple rhythms were therefore expected to induce a strong perception of a metrical beat in listeners (Povel & Essens, 1985). In contrast, intervals comprising complex rhythms were organized into sequences in which there were not tones at every 4 base time units; that is, the temporal placement of the tones did not consistently reinforce a periodic beat. As a result, and in line with prior studies, complex rhythms thus lacked a consistently marked external beat, as compared to simple rhythms. See Figure 1 for a representative example of a simple and a complex rhythm.

Figure 1.

Diagram showing the distinction between simple and complex rhythms. The numbers represent the relative duration of each inter-onset-interval relative to the base interval (notated as ‘1’). Simple and complex rhythms were comprised of the same number of intervals but were arranged in different ways. Simple rhythms always had tone onsets on every beat (indicated by the dashed lines). Complex rhythms did not have tone onsets on every beat. The base interval (duration value for ‘1’) varied randomly from trial to trial and took on values of 220, 245, or 270 ms. A notated value of ‘2’, ‘3’, or ‘4’ corresponded to an inter-onset-interval that was 2x, 3x, or 4x as long as the duration of the inter-onset-interval for ‘1.’

2.3. Procedure

Participants completed two separate visits. The first visit consisted of speech, language, hearing, and cognitive testing, and the second visit consisted of the fMRI experiment. Prior to scanning, participants were familiarized with the experimental paradigm with four practice trials consisting of same and different variants of one simple and one complex rhythm without feedback. Two participants required a repeat of the practice session to ensure they understood the task. Practice rhythms were not used in the experiment proper.

2.3.1. Experimental task

On each trial, participants heard two successive presentations of the standard rhythm and judged whether a third comparison rhythm was the same or different than the standard. The trials were presented in six functional runs (6.75 min). Each run consisted of 24 experimental trials in which participants heard same and different variants of the 12 simple and 12 complex rhythms and an additional 10 null trials consisting of a fixation cross and lasting one or two TR (repetition time). On experimental trials, the time interval between presentations of each rhythm was 1300 ms. The order of stimuli was determined by creating three sets of randomized orders of all 12 simple and 12 complex rhythms. In each set, half of the trials were ‘same’ trials (the correct response was same) and half of the trials were ‘different’ trials (the correct response was different). Participants were given 2100 ms to respond using an MR-compatible keypad before the next trial started (index = “same”, middle = “different”). A total of 144 experimental trials were run across the session, which lasted approximately 40 minutes. E-Prime v2.0 Professional (Psychology Software Tools, Inc.) was used for stimuli presentation. Sounds were presented over MRI compatible headphones (Resonance Technology Inc, Northridge, CA). Responses were made by pressing labelled buttons on a MR-compatible keypad (Psychology Software Tools, Inc., Pittsburgh, PA).

2.3.2. Imaging protocol

The scans were acquired on a 3T GE HDx scanner with an eight-channel head coil. The functional scans consisted of 6 runs of 6 min 45 sec echo-planar imaging, starting from the most inferior regions of the cerebellum, and were acquired with the following parameters: 44 contiguous 3-mm axial slices in an interleaved order, time of echo (TE) = 27.7 ms, time of repetition (TR) = 2500 ms, parallel acceleration factor = 2, flip angle = 80 degrees, field of view (FOV) = 22 cm × 22 cm, and matrix size = 64 × 64.

After the functional data acquisition, the structural scan covering the whole brain was obtained using a volumetric inversion recovery fast spoiled gradient-recalled sequence with cerebrospinal fluid (CSF) suppression (10-minute scan time). The following parameters were used: TE = 3.8 ms, TR of acquisition = 8.6 ms, time of inversion (TI) = 831 ms, TR of inversion = 2332 ms, flip angle = 8°, FOV = 25.6 cm × 25.6 cm, matrix size = 256 × 256, slice thickness = 1 mm, 180 slices and receiver bandwidth = ± 20.8 kHz.

3. Data analysis

3.1. Behavioral data analysis

Rhythm discrimination performance was assessed by analyzing different response proportions for trials where the comparison rhythm was different from and the same as the standard rhythm and then calculating the signal detection measures d´ and c to separate perceptual sensitivity from response bias, respectively (Macmillan & Creelman, 2005). Responding ‘different’ on trials when the comparison was different from the standard corresponded to a ‘hit’ and responding ‘different’ on trials when the comparison was the same as the standard corresponded to a ‘false alarm.’ The performance measures, d’ and c, were then subjected to a separate 2-way mixed-measures ANOVA with factors of Group (AWS vs. Control) and Rhythm Type (simple vs. complex) to probe differences in rhythm discrimination for simple rhythms and complex rhythms for the AWS and Control groups, and any interactions, using an alpha level of 0.05 and operation span as a covariate. To consider individual differences in the AWS group, we additionally examined the relation between stuttering severity (as measured by SSI), working memory capacity (operation span [OSPAN]; ((Conway et al., 2005; Unsworth et al., 2005)) and simple and complex rhythm discrimination performance. The reason for including OSPAN scores as a covariate was twofold: First, subtle working memory deficits have been observed in speakers who stutter (Bajaj, 2007). Second, because the rhythm discrimination task involves listening to and maintaining a series of three rhythms, we wanted to control for any differences in working memory capacity that may contribute to rhythm discrimination performance.

3.2. fMRI data preprocessing

SPM12 was used for fMRI data preprocessing and statistical analysis unless specified otherwise (https://www.fil.ion.ucl.ac.uk/spm/software/spm12/). For each participant, functional images were corrected for differences in slice acquisition timings. Anatomical scans and functional volumes were co-registered to the first volume of the first scan using rigid body rotation. Framewise displacement (FD) for each volume was calculated using the estimated head movement obtained from the co-registration step. Anatomical scans were segmented and normalized to MNI space using DARTEL algorithm. The resulting spatial transformation matrices were applied functional images, which were spatially smoothed with a Gaussian kernel with FWHM of 6 mm.

3.3. fMRI data analysis for task-related activations

At the subject level general linear model (GLM), we modeled the simple and complex rhythm conditions as separate regressors. Voxels in the white matter were excluded from analysis using a dilated gray matter mask. Movement parameters obtained from the co-registration step and global fMRI signal were included as covariates. The individual beta estimates of each condition (simple, complex) were included in the group level analysis.

The group level GLM included group and condition as factors. Age, sex, SSI score, d’, years of musical training, and working memory scores were included as covariates. T contrasts were used to compare the beta estimates between conditions, groups, and their interactions. Family-wise errors were controlled using voxel-wise height threshold p < 0.005 and cluster-size threshold k > 52 voxels, which corresponded to a corrected p < 0.05. The cluster-size threshold was determined by AFNI 3dClustSim (version 18.3.05) with 10,000 Monte Carlo simulation in the gray matter mask. Noise random field for the simulation was generated using non-Gaussian auto-correlation function with fitting parameters of 0.57, 4.85 and 17.71, which were estimated using 3dFWHMx based on the residual images of the GLM (Cox et al., 2017).

3.4. Analysis of task-based functional connectivity

We were also interested in any functional connectivity differences between groups during the simple and complex rhythm conditions. Two separate seed locations, the left and right putamen, were defined using the AAL atlas (Tzourio-Mazoyer et al., 2002). The putamen was selected as the seed region due to its involvement during rhythm processing (e.g., (Grahn & McAuley, 2009)) and motor planning (Bohland et al., 2010) as well as previous findings demonstrating functional relevance to stuttering, rhythm, and speech fluency (Alm, 2004; Chang et al., 2016; Toyomura et al., 2011; Wieland et al., 2015).

Preprocessed functional data were bandpass filtered with cutoff frequency of 0.01 Hz and 0.2 Hz and nuisance variables including movement parameters and white matter signal were removed using regression (AFNI 3dTproject). BOLD responses during the rhythm discrimination task were separated and concatenated by conditions according to onsets and offsets of each trial with a 6 second shift to accommodate the delay of hemodynamic responses. Volumes with FD > 0.5 mm were discarded (Power et al., 2014). On average 9.4% and 11.6% volumes with FD > 0.5 mm for AWS and Controls, respectively, which did not differ significantly between groups (p=0.59).

The AFNI program 3dGroupInCorr was used for functional connectivity analyses. At each seed location, Pearson’s correlation coefficients were calculated between the averaged time series in the seed and each voxel’s time series in the whole brain. Fisher’s r-to-z transformation was applied to the correlation map. For group level analysis, individual z-maps were analyzed using a GLM, which included group, condition, and the group by condition interaction. Covariates included SSI score, d’, age, sex, years of musical training, and working memory scores. As in the task-based activation analyses in the previous section, family-wise errors were controlled using voxel-wise height threshold p < 0.005 and cluster-size threshold k > 52 voxels, which corresponded to a corrected p < 0.05.

Our general analysis strategy was to examine main effects of group and rhythm type, i.e., the difference between groups with simple and complex conditions collapsed, and the difference between rhythm types with the groups collapsed. In addition, we examined the interactions between group and rhythm type (i.e., the difference between simple and complex rhythm discrimination in AWS compared to controls or “double subtraction”: [AWS(Complex-Simple) – Controls(Complex-Simple)]). We also compared the difference between groups for each rhythm type and the difference between rhythm types within each group.

4. Results

4.1. Rhythm discrimination performance

Figure 2 displays box and whisker plots for the perceptual sensitivity measure, d´ (Panel A) and response bias measure, c (Panel B) for the AWS group and the Control group for the simple and complex rhythms. The omnibus 2 (Group: AWS vs. Controls) x 2 (Rhythm Type: Simple vs. Complex) mixed-measures ANOVA on d´ revealed a main effect of Rhythm Type, F(1, 33) = 12.90, p = 0.001, η² = 0.28, with simple rhythms (M = 2.70, 95% CI = 2.31 – 3.08) better discriminated than complex rhythms (M = 1.41, 95% CI = 1.06 – 1.76). There was no overall main effect of Group, F(1, 33) = 0.69, p = 0.42, η² = 0.02, but a marginally significant interaction between Group and Rhythm Type, F(1, 33) = 3.28, p = 0.079, η² = 0.09. Because we had the a priori hypothesis that a rhythm deficit in AWS would be specific to complex rhythms, we conducted separate between-group comparisons for simple and complex rhythms. In line with the internal beat deficit hypothesis, AWS evidenced poorer complex rhythm discrimination compared to controls, t(34) = 2.02, p = 0.05, Cohen’s d = 0.66, but there was no significant group difference for simple rhythm discrimination, t(34) = 0.66, p = 0.53, Cohen’s d = 0.23. With respect to the response bias measure, c, there was a very slight general tendency to respond ‘same’ (M = 0.13, 95% CI = 0.03 – 0.24). The ANOVA on c, however, revealed no main effect of Rhythm Type, F(1,33) = 0.07, p = 0.79, η² = 0.002, no main effect of group, F(1, 33) = 0.001, p = 0.978, η² < 0.001, and no interaction between Group and Rhythm Type, F(1,33) = 0.65, p = 0.43, η² = 0.02.

Figure 2.

Behavioral results. Top panel: Perceptual sensitivity, d’ (Panel A) and response bias, c (Panel B) comparing adults who stutter (AWS) to age-matched adults who do not stutter (controls) for simple and complex rhythms. Bottom panel: Relation between Stuttering Severity Instrument (SSI-4) and d’ after controlling for working memory for simple rhythms (Panel C) and complex rhythms (Panel D). Greater stuttering severity is associated with worse rhythm discrimination for both simple and complex rhythms.

Overall, there were large individual differences in rhythm discrimination performance for both groups; moreover, individuals in the AWS group showed a wide range of stuttering severity, ranging from very mild to severe. One factor that contributed to individual differences was operation span (a measure of working memory capacity). Operation span was positively correlated with both simple rhythm discrimination (r = 0.39, p = 0.019) and complex rhythm discrimination (r = 0.40, p = 0.15). To further examine rhythm discrimination performance in AWS, Figure 3 shows the relation between stuttering severity (as measured by SSI) and simple rhythm discrimination (Panel C) and complex rhythm discrimination (Panel D) after partialling out the contribution of working memory (as measured by the operation span score). In general, rhythm discrimination performance decreased with increased stuttering severity for both simple rhythms, r(15) = −0.57, p = 0.018 and complex rhythms, r(15) = −0.65, p < 0.005.

Figure 3.

Functional MRI results. All statistical maps shown at corrected p<.05.

Panel A. The right putamen and left supramarginal gyrus (SMG) showed greater activation during simple rhythm discrimination relative to complex rhythm discrimination in all participants (i.e., collapsing across both groups).

Panel B. Brain regions where activation differed significantly between AWS and control groups. The AWS group showed overall increased activation relative to controls in the SMA when rhythm conditions were collapsed (Top). This group contrast appears to reflect mainly group differences in the complex rhythm condition (Middle). Group differences during simple rhythm condition revealed significantly increased activation in AWS relative to controls in the bilateral STG (Bottom).

Panel C. Brain regions where activation was significantly modulated by performance on the rhythm discrimination task (d’). Relative to controls, AWS showed significantly greater modulatory effects of overall rhythm processing (i.e., when simple and complex rhythms were collapsed; Top). These areas included the right putamen and insula, which were more active the better the performance on the rhythm discrimination task. Controls exhibited significantly greater modulatory effects in bilateral insula and SMA for complex relative to simple rhythms (Middle). Within the AWS group (only), activation in the right putamen and insula was significantly modulated by discrimination performance for complex rhythms: better performance in the complex rhythm task was associated with greater activity (Bottom).

Panel D. Functional connectivity results. AWS showed greater correlated activation between left putamen and precuneus, cingulate, and right superior medial gyrus (Top Left). AWS showed greater correlated activation between the right putamen and the cerebellum during the simple rhythm condition and greater correlation with the precuneus/cingulate region for complex rhythms (Top Right). Controls showed greater correlated activation between the right putamen and the anterior cingulate for complex relative to simple rhythms (Bottom Right).

4.2. fMRI results

4.2.1. Task-based activation during complex and simple rhythm discrimination in all participants

First, to demonstrate that the rhythm discrimination task used in the current study induced similar patterns of brain activation as reported in previous studies that examined simple and complex rhythm perception, we compared activation between complex and simple rhythms in all participants (i.e., collapsed across groups; Figure 3, Panel A; Table 1, left). Consistent with previous findings that pointed to the putamen for beat-based rhythm perception (Grahn & Brett, 2007; Grahn & McAuley, 2009; Grube, Cooper, et al., 2010; Teki et al., 2011), we found that the right putamen showed significantly greater activation during simple versus complex rhythm discrimination. We also found greater activation in the left SMG for simple relative to complex rhythm discrimination.

Table 1.

Left: Within group results. Regions where activity was significantly greater in the simple compared to complex rhythm condition in each group (AWS, controls) as well as with groups combined. Right: Between group results. Regions where the AWS group exhibited significantly greater activity than the control group during rhythm discrimination task performance (simple, complex, combined condition).

Within group analysis									Between group analysis
Contrast	Group	Region	Side	x	y	z	t	voxels	Contrast	Rhythm Type	Region	Side	x	y	z	t	voxels
Simple > Complex	Combined	Putamen	R	27	−3	−9	−4.5	76	AWS > Controls	Combined	SMA	L	−9	9	57	4.2	60
		SMG	L	−57	−42	27	−4.3	55		Simple	STG	L	−60	−54	6	4.4	52
	AWS	Middle Cingulate	R	6	−30	45	−5.1	75			STG	R	51	−27	6	5.5	55
	Control	IPL	L	−54	−39	39	−4.3	97		Complex	SMA	L	−9	9	57	4.3	70

4.2.2. Between-group differences during rhythm discrimination task

As a first step in examining differences between AWS and controls, we compared a) overall activation for all rhythms (i.e., collapsed across simple and complex rhythms; Figure 3B, Top; Table 1, right) followed by b) group differences for each rhythm type separately (Figure 3B Middle [Complex] and Bottom [Simple] panels; Table 1, right). In contrast to expectations, the AWS group showed greater activation relative to controls in the left SMA when combining rhythm types in the analysis (Figure 3B, Top). This difference appears to reflect a greater contribution by group differences during complex rhythms specifically (Figure 3B, Middle).

During simple rhythm discrimination (Figure 3B, Bottom), AWS again showed greater activation than controls, but this time in the bilateral STG regions. There were no significant clusters showing greater activation for controls compared to the AWS group. The double subtraction analysis (AWS(Complex-Simple) – Controls(Complex-Simple) also did not reach statistical significance.

4.2.3. Modulatory effect of d’ on brain activity

We further examined whether task performance (d’) modulated brain activation, and whether such a relationship differed between groups. We found that there was an enhanced modulation between rhythm discrimination performance and brain activation in AWS relative to controls when both simple and complex were analyzed together in the right putamen, right caudate head, and right insula. (Figure 3C, Top; Table 2).

Table 2. Modulatory effects of BBA performance.

Brain areas showing significant correlation with better rhythm discrimination performance.

Contrast	Rhythm Type	Region	Side	x	y	z	t	voxels
AWS > Controls	Combined	Putamen, Caudate, Insula	R	27	21	3	4.7	170
		SMA	L	−12	21	48	4.9	57
Within AWS	Complex	Putamen, Insula	R	30	12	0	4.2	56
Within Controls	Complex> Simple	Insula	L	−30	18	−6	4.9	296
		SMA	L	−6	18	54	5.2	238
		Insula	R	36	30	6	5.3	181

In the control group (only) we saw enhanced modulation for complex versus solo rhythms in the bilateral insula and left SMA (Figure 3C, Middle). There was no significant difference between rhythm types in the AWS group.

In the complex rhythm condition (only) in AWS (Figure 3C, Bottom), activation in the right putamen was significantly associated with task performance. There were no other significant findings related to the modulatory effect of d’ on brain activity patterns.

4.2.4. Effects of stuttering severity

There were no significant correlations between stuttering severity (SSI) and brain activation associated with rhythm discrimination performance.

4.2.5. Functional connectivity during rhythm discrimination task performance

A seed-based functional connectivity analysis was performed using a priori selected left and right putamen seeds, defined using the AAL atlas. The purpose of the functional connectivity analysis was to further examine whether activity in the rhythm network areas -- identified via correlated activity with the bilateral putamen seeds – would differ between rhythm types that varied in the level of intrinsic timing required to perform the task (simple, complex rhythm discrimination).

Figure 3D (Top) shows that the AWS group exhibited greater functional connectivity between the left putamen and the precuneus/cingulate and right superior medial gyrus for complex relative to simple rhythms (See also Table 3). There were no significant functional connectivity findings using the left putamen seed in the control groups.

Table 3.

Top: Regions showing significant functional connectivity with the left putamen within the AWS group. There were no significant results for the control group. Bottom: Regions showing significant functional connectivity with the right putamen.

Left Putamen Seed
Contrast	Region	side	x	y	z	t	voxels
AWS (Complex-Simple)	Precuneus/cingulate	L/R	3	−48	36	4.1	196
	Superior medial gyrus	L/R	6	57	21	3.9	78
Right Putamen Seed
Contrast	Region	side	x	y	z	t	voxels
Controls (Complex-Simple)	Anterior cingulate	L/R	3	51	3	4.1	86
AWS (Complex-Simple)	Precuneus/cingulate	L/R	−12	−54	9	4.3	261
	Cerebellum (Crus II)	L	−6	−81	−33	−3.8	59

Greater functional connectivity between the right putamen and cerebellum was observed for the AWS group during the simple rhythm condition, whereas greater functional connectivity between right putamen and the precuneus/cingulate region was observed for complex rhythms. Additionally, the control group showed greater functional connectivity between the right putamen and the anterior cingulate for the complex relative to the simple rhythm condition (Figure 3D, Bottom, Table 3). The double subtraction analysis [AWS(Complex-Simple) – Controls(Complex-Simple)] did not reach statistical significance.

5. Discussion

In this study, participants completed an auditory rhythm discrimination task while undergoing fMRI to investigate auditory rhythm perception and associated neural activity patterns in adults who stutter (AWS) compared to age-matched controls. Of primary interest for the rhythm discrimination task was a comparison of two rhythm conditions: a simple rhythm condition, where rhythms consistently had tone onsets at regular periodic intervals to reinforce an explicit beat, and a complex rhythm condition, where rhythms were comprised of the same number and distribution of time intervals as the simple rhythm condition, but did not consistently have tone onsets marking every beat.

With respect to the behavioral results, there were three main findings. First, consistent with previous rhythm discrimination studies using the same task, there was an overall beat-based advantage – simple rhythms where beats were always reinforced by tones – were better discriminated than complex rhythms with an inconsistently (weakly) marked beat (Grahn & Brett, 2009; Wieland et al., 2015). Second, consistent with our previous work with children who stutter and an internal beat deficit hypothesis (Wieland et al., 2015) adults who stutter showed poorer rhythm discrimination performance relative to controls for complex rhythms, but not for simple rhythms. Finally, within the AWS group, greater stuttering severity was associated with worse performance on the rhythm discrimination task after controlling for individual differences in working memory capacity.

With respect to neural activation patterns in the two groups, the fMRI results revealed three main findings. First, AWS showed generally increased activation relative to controls in the SMA during rhythm perception especially during complex rhythm discrimination, as well as increased activation in the bilateral auditory areas during simple rhythm discrimination (Figure 3B). Second, rhythm discrimination task performance significantly modulated brain activity in both groups (Figure 3C). In controls, greater activity in the insula and SMA was associated with better performance for complex relative to simple rhythms. In AWS better performance on the rhythm discrimination task by AWS participants was associated with greater activation in the right putamen for complex rhythms only (Figure 3C bottom panel). Third, when examining functional connectivity involving the putamen, both groups showed heightened connectivity in cortical regions for complex compared to simple rhythm tasks. However, AWS exhibited an enhancement of functional connectivity between the left putamen and cerebellum during simple relative to complex rhythms. In the following sections, we discuss these behavioral and neuroimaging findings in turn and consider more broadly the potential contributions of altered rhythm perception to speech disfluencies observed in stuttering.

5.1. Adults who stutter show worse auditory rhythm discrimination for complex rhythms that lack a clear beat

Auditory rhythm discrimination shows a robust beat-based advantage where simple rhythms with a consistently marked beat are better discriminated than complex rhythms with an inconsistently marked beat (Grahn & Brett, 2007; Povel & Essens, 1985) supporting the view that internal timing abilities are taxed more when hearing complex rhythms that comprise less predictable beats than simple rhythms. Consistent with a deficit in internal timing, AWS (as a group) show moderately greater difficulty discriminating complex rhythms compared with controls. This result mirrors our previously reported findings in children who stutter (Chang et al., 2016; Wieland et al., 2015) and adds more generally to support for a rhythm and timing account of stuttering long proposed in the literature (Alm, 2004; Cooper & Allen, 1977; Hulstijn et al., 1992; Max & Yudman, 2003; Olander et al., 2010; Zelaznik et al., 1994).

Early rhythm and timing accounts of stuttering were described as individuals who stutter as having an inaccurate “neural clock” (Cooper & Allen, 1977) or a less flexible timing system that is prone to breakdown (Brown et al., 1990). The present study adds to these prior findings by providing support for the specific hypothesis that rhythm and timing deficits in stuttering are due in particular to disruption in the internal generation of a periodic beat (the internal beat deficit hypothesis). Povel and Essens (1985) explain the simple rhythm (over complex rhythm) discrimination advantage by proposing temporal patterns (i.e., rhythm) that afford induction of a periodic beat, enabling the formation of more stable internal representations and more efficient encoding, leading to both better rhythm discrimination and better rhythm production.

Further bolstering support for the internal beat deficit hypothesis, a significant portion of the variance in simple and complex rhythm discrimination within the AWS group was accounted for by stuttering severity. Stuttering severity (as measured by SSI) was negatively correlated with rhythm discrimination performance, d’, for both simple and complex rhythms after controlling for working memory (OSPAN score); adults with higher SSI scores (i.e., greater stuttering severity) show worse rhythm discrimination than adults with lower SSI scores. The work of Povel & Essens ((1985); see also (Essens & Povel, 1985)) combined with our earlier work on rhythm discrimination with children who stutter (Chang et al., 2016; Wieland et al., 2015) and the current findings of a similar pattern of impaired rhythm discrimination performance in adults who stutter, suggests that atypical rhythm discrimination (observed to a greater degree with complex rhythms that rely more heavily on internal timing) may be a prominent feature of developmental stuttering that is moreover linked to stuttering persistence.

5.2. Adults who stutter exhibit heightened activity in beat-based timing network areas during rhythm discrimination

For complex rhythms, greater activity in critical timing-linked areas such as the SMA and putamen was evident in AWS, while heightened activity in the auditory regions was observed for simple rhythms. The overall heightened activation in beat-based timing network areas during rhythm discrimination in AWS may reflect greater, though perhaps less efficient, engagement of these structures especially during complex rhythm discrimination, as behavioral performance still lagged that of the control group. Furthermore, greater connectivity between putamen and cerebellum was associated with simple rhythm discrimination. This result suggests a greater reliance by AWS on interval-based (over beat-based) timing mechanisms for rhythm discrimination compared to adults who do not stutter.

With regard to the complex rhythms, which did not have consistently marked external beats as was the case in the simple rhythm condition, thus taxing intrinsic timing abilities to a greater extent, we expected that AWS would perform worse and that we would see greater group differences in terms of brain activity within the rhythm network in this condition. The finding of heightened activity in the SMA is notable for at least two reasons. First, the SMA, along with the BG (i.e., putamen), form the medial timing circuit (Kotz et al., 2009; Merchant et al., 2013). AWS showed on one hand heightened activity in the SMA relative to controls for complex rhythm processing, while the level of performance on this task appeared to be modulated by activity in the putamen and insula (Figure 3C). We speculate that functional connectivity between SMA and putamen is aberrant in AWS, resulting in discoordination of the two regions that affects intrinsic timing processing in stuttering. When examining functional connectivity of the putamen in AWS, there was heightened connectivity in areas including the cingulate and cerebellum, but not in the SMA (Figure 3D). Second, the SMA-BG connectivity supports spontaneous speech production that is not reliant on external pacing, posited to be the core deficit in stuttering (Alm, 2004). Alm argues that deficits in this circuit is compensated by the lateral premotor cortex and cerebellum. The observed significant increase in SMA activity for AWS during complex rhythm discrimination thus seems to reflect possible anomalies in this medial SMA-BG timing circuit that may also influence spontaneous speech production in stuttering.

In the case of simple rhythm discrimination, however, AWS showed comparable behavioral performance to controls, supported by notable increases in the bilateral auditory areas as well as heightened functional connectivity between the putamen and cerebellum. The heightened STG activity for simple rhythms in AWS suggests that the functional coupling between putamen and STG that underscores the putamen’s role of extracting temporal regularity in the auditory signal (Geiser et al., 2012) is affected in AWS. The fact that the behavioral performance for simple rhythms was comparable between the groups may be explained by a compensatory response, reflected in the heightened functional connectivity we saw between the putamen and the cerebellum and cingulate areas in the case of AWS.

The putamen has been reported to support beat-based timing and are more active during performance or tracking of simple rhythms, that is, those that are easier to internalize, compared to complex rhythms (Geiser et al., 2012; Grahn & Rowe, 2009, 2013). In the current study, we observed an increased correlation between rhythm discrimination performance and right putamen activity in AWS relative to controls, particularly in the complex rhythm task. Geiser et al. (2012) showed that temporal regularity of the auditory signal - encoded by the putamen - could facilitate perception of unrelated aspects of auditory signals such as their intensity. Moreover, Geiser et al. found that the putamen and the primary and secondary auditory cortices were functionally coupled for periodic auditory signal processing: that is, greater putamen activity and lesser auditory cortex activity were observed during periodic signal processing, indicating that temporal regularity in the signal reduced processing demands in the auditory cortex. In this context, the observed greater activity in auditory cortices in AWS during simple rhythm discrimination may suggest an atypical functional coupling between the auditory cortex and putamen in AWS, possibly due to deficits in BGTC function that leads to disruptions in encoding temporal regularity in the auditory signal. If there is such a deficit, the facilitatory effect of temporal regularity, which normally reduces the need for attention and other processing burdens for auditory signal processing and thus reduces auditory cortex activity, would not occur. This was indeed the case for AWS - they exhibited heightened auditory cortex activity for simple rhythms that were characterized by their temporal regularity. A deficit in beat-based timing by AWS could indicate an incomplete internal representation of the periodic structure of the rhythm. If AWS were able to maintain a robust representation of temporal regularity, we would have expected to see results more consistent with Geiser et al. (2012) who found directionally opposite activity patterns in the basal ganglia (increased) and auditory cortex (decreased) for periodic rhythms.

5.3. Adults who stutter exhibit greater reliance on an interval-based timing network for simple rhythm discrimination

Further support for atypical BGTC function in AWS was revealed in the functional connectivity results (Figure 3D), wherein AWS show greater correlated activity between the right putamen and the left cerebellum during the simple rhythm condition. This suggests that in addition to heightened activity in areas supporting beat-based timing, AWS also rely to a greater degree than controls on regions involved in interval-based timing such as the cerebellum, to perform rhythm discrimination, even when those rhythms contain a strong periodic signal. Compared to beat-based (rhythmic) timing supported by the basal ganglia, cerebellar pathways support interval-based timing, in which the time between events is estimated, rather than the rhythm of the events itself (Breska & Ivry, 2018; Grube, Cooper, et al., 2010; Teki et al., 2011). This functional dissociation is substantiated by behavioral and neurophysiological studies that have reported beat- vs. interval-based timing deficits in clinical disorders with focal lesions or degeneration confined to either the BG (i.e., Parkinson’s disease) or the cerebellum (i.e., cerebellar degeneration) (Breska & Ivry, 2018; Grahn & Brett, 2009; Grahn & McAuley, 2009; Grube, Cooper, et al., 2010; Kotz et al., 2014; Nozaradan et al., 2017; Teki et al., 2011).

For example, Teki et al. (2011) found a striking separation in terms of neural activity for absolute, interval-based timing, and relative, beat-based timing (see Teki et al., Figure 5) using a task that compared regular (with a beat) vs. irregular (without a beat) rhythms. The authors conclude that an olivo-cerebellar system supports interval-based timing, whereas a striato-thalamo-cortical system supports beat-based timing, and further argue that a beat-based clock is a more efficient method of timing based on greater accuracy and speed in their behavioral results. Interestingly, Xu et al. (2006) also found inferior olive activity specifically during perception of complex rhythms compared to isochronous rhythms, which Teki et al. (2011) suggests is evidence for “a possible role for the olivocerebellar system in detecting errors in the regular operation of a beat-based timer in the striatum” (p. 3810). Relevant to the present study, then, it is possible that due to aberrant functioning of the beat-based timing system in AWS, the cerebellar pathway may attempt to compensate during the rhythm discrimination tasks, even for simple rhythms, leading to the greater correlated activity seen between the cerebellum and basal ganglia. This view is supported by more recent work in AWS that showed increased functional connectivity within the cerebellum and between the cerebellum and prefrontal cortex during paced, rhythmic speech (Frankford et al., 2021).

5.4. Relevance of non-speech rhythm processing to developmental stuttering

Because stuttering, as a disorder, is characterized by disruptions in speech production, it is perhaps not surprising that there has been limited work examining more general deficits in auditory perception. The focus of the current study on auditory rhythm discrimination in adults who stutter (an emphasis on a non-speech auditory perception ability that is not directly related to speech disfluencies) is thus an exception in that regard. Therefore, it is important to discuss our findings in the broader context of stuttering as a disorder of speech fluency that is impacted by impaired temporal prediction. From this perspective, why would people who stutter exhibit deficits in a non-speech rhythmic perception task?

The link between perception and production with regard to neural activity is well established, even for a quasi-periodic signal such as speech (for discussions, see (Kotz et al., 2018; Poeppel & Assaneo, 2020). Accumulating evidence shows overlapping brain activity in motor cortex for speech and music processing, as well as connection between the SMA, basal ganglia, and frontal cortical areas (for a discussion see (Peretz et al., 2015)). Network hubs that are common to both music (timing/rhythm) and speech processing often show co-activation and evidence of coordination of information across modalities (i.e., speech and non-speech; (Peretz et al., 2015)). Accordingly, hubs represent points of increased vulnerability, and because hubs that are shared by speech and timing networks overlap, this may be why music and speech processing deficits can co-occur (Kotz et al., 2018; Ladányi et al., 2020). Note, however, that while atypical rhythm processing in speech and language disorders may seem to have obvious connections to the speech disorder itself, overlap in neural activity across different domains does not necessarily indicate shared neural processing for speech and non-speech (e.g., musical) stimuli or tasks ((Peretz et al., 2015)). Nevertheless, several influential and prevailing theories do support overlapping networks for speech and non-speech processing (Fujii & Wan, 2014; Goswami, 2011, 2018; Jones, 1976; Patel, 2011, 2012; Schwartze et al., 2011; Tierney & Kraus, 2014).

Recently, Ladányi et al. (2020) proposed a general Atypical Rhythm Risk Hypothesis implicating impaired rhythm as a risk factor in neurodevelopmental disorders. Although the Atypical Rhythm Risk hypothesis is similar to the internal-beat deficit hypothesis (both implicate impaired rhythm as a risk factor), it is a much broader hypothesis. According to the Atypical Rhythm Risk hypothesis, disorders such as dyslexia, developmental language disorder, stuttering, developmental coordination disorder, attention deficit hyperactivity disorder, and other idiopathic speech and language disorders all show impairments in some aspects of rhythm processing. Consistent with this hypothesis, there have been a number of studies showing correlations between performance on rhythm and language tasks and short-term rhythm-based interventions have been linked to improved performance on language measures in typically developing children (for a summary, see (Ladányi et al., 2020)).

In the domain of developmental stuttering, separate from reports of impaired rhythm processing (Chang et al., 2016; Wieland et al., 2015) there is undeniable evidence that stuttering decreases in the presence of an external rhythm, like a metronome or choral speech (e.g., (Adams & Ramig, 1980; De Nil et al., 2003; Park & Logan, 2015; Toyomura et al., 2011, 2015). This “rhythm effect” in stuttering was explored in a recent study by Frankford et al. (2021) who found that externally (vs. internally) paced speech decreased stuttering and was accompanied by greater functional connectivity within the cerebellum, but reduced connectivity between left cerebellum and prefrontal cortex. The authors concluded that the cerebellum activity was compensatory, and that the cerebellum may be a potential target for future interventional studies. Results from the current study provide some support for this perspective, namely correlated activity seen between the cerebellum and basal ganglia during rhythm discrimination may represent a shift from beat-based to interval-based timing.

Because working memory has been proposed to be implicated in stuttering (see (Bajaj, 2007) for a review) and altered functional connectivity during a phonological working memory task has been reported (e.g., (Yang et al., 2018)), a potential alternative explanation for poorer rhythm discrimination performance by AWS compared to controls, where participants have to listen to and maintain multiple rhythms for comparison, could be a deficit in working memory capacity. This seems unlikely, however, as (1) the AWS and control groups did not differ significantly in operations span – our measure of working memory capacity and (2) both the behavioral and fMRI analyses used the operation span measure as a covariate. Thus, the poorer rhythm discrimination observed for AWS compared to controls and the robust negative correlation between stuttering severity and rhythm discrimination among AWS participants is not explained by individual differences in working memory capacity.

One potentially fruitful avenue of future research would be to consider developmental changes in rhythm discrimination as a potential marker of stuttering persistence. Future work could also involve direct comparison of adults and children who stutter using the same protocol and analyses. For example, children in our studies performed the rhythm discrimination task outside the scanner, and our neuroimaging findings were based on resting state functional connectivity. We expect the resting state results would be strengthened, for example, if older children who were capable performed the rhythm discrimination task during fMRI. Such advancements in the understanding of the involvement of intrinsic timing in developmental stuttering could pave the way for rhythm-based interventions.

5.4. Conclusion

In sum, this is the first task-based fMRI study to investigate auditory rhythm discrimination in adults who stutter. The results show (1) worse rhythm discrimination for AWS (considered as a group) compared to controls for complex rhythms, but not for simple rhythms; (2) that within the AWS group, there is a negative correlation between stuttering severity and rhythm discrimination performance for both complex and simple rhythms after controlling for individual differences in working memory capacity; (3) AWS demonstrate increased activity in putative rhythm network regions relative to controls during rhythm discrimination; (4) AWS exhibit correlations between rhythm discrimination performance (especially for complex) and activity in the right putamen and insula; and (5) there is greater correlated activity between the basal ganglia (putamen) and cerebellum during simple rhythm discrimination, as compared with the complex rhythm condition. Thus, rhythm discrimination in AWS is associated with overall heightened activity and engagement in beat-based rhythm network areas, as well as suggesting there is engagement of interval-based timing network for comparable behavioral performance on the simple rhythm discrimination task relative to adults who do not stutter. Overall, these results support the internal-beat deficit hypothesis account for developmental stuttering (Alm, 2004; Etchell et al., 2014), and are consistent with the notion of weak beat representation for adults who stutter (Grahn & McAuley, 2009). Together with similar previous findings in children who stutter (Chang et al., 2016; Chang & Zhu, 2013; Wieland et al., 2015), the present findings are consistent with the broader Atypical Rhythm Risk Hypothesis (Ladányi et al., 2020) that dysfunction in the neural networks supporting rhythm perception may be one fundamental component of neurodevelopmental disorders, including stuttering.

Data statement

Data from this study may be made available for all reasonable requests to the corresponding author.