The research protocol was approved by the Institutional Review Board at Purdue University and adhered to Human Research Protection Program regulations and guidelines. We obtained written informed consent from parents or legal guardians, henceforth referred to as parents, at the beginning of the session. Seventy-two preschool children participated in the study, 47 CWS (38 boys and 9 girls, M = 55.9 months, SD = 7.6, range = 46–72) and 25 CWNS (18 boys and 7 girls, M = 55.6 months, SD = 6.3, range = 49–70). Parents reported that their child did not have a developmental, cardiovascular, or neurological disorder and confirmed that North American English was their child’s first and primary language. Parents also noted whether their child had normal or corrected-to-normal vision, had not consumed caffeine prior to the experiment, and were not taking medications affecting the central nervous and cardiovascular systems (e.g., depressants, stimulants, analgesics, anticoagulants etc.). All participants scored within normal limits on assessments of non-verbal intelligence (PTONI; Ehrler and McGhee, 2008) and social development (Childhood Autism Rating Scale, 2nd Edition CARS-2; Schopler et al., 2010) and passed a bilateral pure-tone hearing screening at 500, 1000, 2000, 4000, and 6000 Hz at 20 db. Finally, the two groups had comparable socioeconomic status (SES) (CWS M = 5.60, SD = 1.02; CWNS M = 6.20, SD = 0.65) determined by the parent’s highest level of education (Hollingshead, 1975). SES was evaluated on a 7-point scale (1 = less than 7th grade education to 7 = completion of a graduate or professional degree).

We diagnosed childhood stuttering using standardized and observational measures. First, we collected spontaneous speech samples from each child during two play sessions, one with their primary caregiver, and the other with the project speech-language pathologist (SLP), for a combined total of 750–1000 syllables. We calculated a weighted stuttering index (WSI) for each child based upon the frequency of part- and single-syllable word repetitions, the number of iterations, and the presence and duration of dysrhythmic phonations per 100 syllables of spontaneous speech (Ambrose and Yairi, 1999). A score of 4.0 or higher indicates stuttering. We also administered the Test of Childhood Stuttering (TOCS; Gillam et al., 2009), a norm-referenced assessment tool to each child. An index score of 84 or below on this assessment identifies the child as stuttering.

In addition, the parent and SLP with experience in fluency disorders rated the child as stuttering by assigning them a score of 2 or higher on an 8-point scale [0–1 = normal; 2–3 = mild stuttering; 4–5 moderate stuttering; 6–7 = severe stuttering]. The clinician’s rating was based upon the type, duration, and frequency of disfluencies along with the presence and severity of secondary characteristics while considering the child’s own awareness and/or anxiety about his or her disfluencies.

Three CWS had a WSI <4.0; however, we retained these children in the study because they met the criteria for stuttering based upon the TOCS and clinician and parent ratings. The average duration of stuttering (i.e., time since onset) was 21 months (SD = 11 months) according to parent report.

The experiment comprised speech and non-speech tasks that were adapted for research with young children from our earlier studies with older children and adults (Weber and Smith, 1990; Arnold et al., 2014). There were four tasks: open/close jaw wag (lower effort, non-speech condition), structured sentence production (lower effort, speech condition), picture description (higher effort, speech condition), and forceful blowing (maximal maneuver, non-speech condition). The order of the jaw wag, sentence production, and picture description tasks was counterbalanced across subjects; however, we consistently collected a baseline at the beginning (pre-baseline) and a baseline at end of the experiment (post-baseline) before the blowing task. Forceful blowing, designed to elicit a maximal response, was always collected at the end of the experiment so that the expected high levels of autonomic arousal would not influence data collected during the other experimental tasks. Before each task, we explained and demonstrated what the child would do, then had each child practice two to three trials to ensure they understood the task. During data collection, a 15 s rest preceded each of the four tasks, and a short break was taken at the end of each task. During these breaks, the participants either collected a sticker or took a turn at a game.

For the two resting baseline intervals collected at the beginning and near the end of the experiment, the children saw a picture of a child resting outdoors with her eyes closed and were encouraged, but not required, to close their eyes, be still, and rest for 1 min. For the first 13 CWS who participated in the experiment, these initial and final baselines were collected in a different manner. In these cases, baseline intervals were interspersed within the experiment before each task. However, we found that these inter-task intervals were not sufficiently long enough to allow sympathetic arousal to return to resting levels and were often contaminated by movement artifacts. Our statistical approach, described below, accounts for these missing baseline measures from the 13 CWS.

The experimental tasks consisted of 2 speech and 2 non-speech tasks:

Jaw wag (JAW). For this non-speech task, the children continuously opened and closed their mouth for 6 s while watching an animated face opening and closing its mouth, then rested for 5 s. We collected two trials of five jaw wag intervals each.

Structured sentence production (SENT). For this speech task, the children saw pictures of familiar objects or animals inside a box or a barn and spoke aloud the simple, declarative carrier sentences, “The____is in the barn” or “The____is in the box.” We considered this task less demanding in terms of required language formulation (Arnold et al., 2014). The children produced three sentences in a row to identify what they saw in three different slides (e.g., “The cookie is in the box…The doll is in the box…The ball is in the box”) then rested for 5 s before the next sequence of three picture slides appeared. Each child completed at least two trials with five sets of three sentences in each.

Picture Description (PIC). During this speech task, the children viewed child-friendly black and white picture scenes and described what was happening in each one. One experimenter modeled the task for two scenes then had each child practice describing two scenes so that they understood the task and could describe the picture scenes using connected speech (e.g., “There’s a farmer on a tractor”, as opposed to listing items (e.g., farmer…tractor…cow) that they recognized. The experimenters controlled the rate of picture presentation to ensure that we collected several utterances per scene. If children did not respond, or produced only a brief response, the experimenter encouraged them by saying, “Tell me more about that” or “What else is going on in the picture?” Within a trial, each picture scene was separated by 5 s of rest. The children completed at least two trials of five picture scenes for a minimum of ten picture scenes. If children produced few responses or were disfluent, we presented an additional block of three picture scenes.

Forceful blowing (MAX). Participants completed a forceful blowing task included to elicit relatively high levels of sympathetic arousal. In many experiments, a Valsalva maneuver is used establish a maximal response (Boucsein, 2013). Through pilot testing, we found that young children could not reliably produce a Valsalva, so they blew on a party blower and held it extended against the experimenter’s hand for 3–5 s. Each child completed two trials of three to five blows.

SNS signals were collected with a Biopac MP150 data acquisition system running AcqKnowledge 4.4 acquisition software. The system included a Biopac GSR100C amplifier to measure electrodermal activity and a PPG100C pulse plethysmograph amplifier for BPV measures. After completing an adapted handedness inventory (Oldfield, 1971), the children were seated at a small desk while the experimenter affixed pre-gelled self-adhesive Ag/AgCl electrodes to the hypothenar eminence and thenar eminence of their non-dominant hand to record electrodermal activity. We recorded skin conductance in microSiemens (μS) between the two electrodes at a 2.5 kHz sampling rate with an initial gain of 10 μS/V and low-pass filtered at 1 Hz. Next, a photoplethysmograph transducer (Biopac TSD200) was secured around the distal phalanx of the fourth finger of the child’s non-dominant hand with a Velcro strap. An infrared emitter and photodiode detector embedded in the transducer measured the relative changes in blood flow. The BPV signal was collected with a gain of 50 at a 2.5 kHz sampling rate.

We presented the experimental stimuli (e.g., pictures, animations) in Microsoft PowerPoint on a 25 in monitor. The participants’ speech acoustic signal was collected with a Shure SM90 tabletop condenser microphone at 10 kHz, and video recordings of the experimental session were made with a Logitech HD 720p webcam. The audio and video signals were synchronized with the autonomic recordings through the Biopac MP150 acquisition system. These signals were used for off-line transcription and to ensure that segments used in subsequent analyses were free from movement artifact, unrelated comments, or redirection from the experimenter.

After the experiment, the physiological signals were exported into MATLAB (ver. 2015a). The electrodermal signal was downsampled to 250 Hz to reduce file size. We derived SCR measures from SCL via high-pass filtering (Fpass 0.07 Hz/Apass 0.5 dB). The BPV signal was downsampled to 250 Hz and digitally band pass filtered between 0.5–3 Hz with a 100-order FIR filter (Arnold et al., 2014). We downsampled signals to reduce file sizes and subsequent processing time and because the energy in these SNS signals is low frequency. A custom MATLAB program simultaneously displayed the acoustic record, BPV signal, and SCL and SCR signals on the monitor from a trial (Figure 1). The synchronized video record was displayed on an adjacent monitor. As shown in Figure 1, the experimenter extracted SCL, SCRs, and BPV segments from the longer, continuous recordings by indicating onset and offset points in the acoustic record with a cursor. Other studies have recorded SCL over longer, several minute epochs (e.g., Jones et al., 2014b). This presents challenges for studies with preschool children as movement artifact, redirection from the experimenter, or unrelated comments from the child may inflate indices of sympathetic arousal. For pre and post baselines, we extracted 5 s intervals from the 1 min records when the child was still, quiet, and electrodermal activity reached a minimum. For the JAW task, we selected 4–6, 6 s trials of accurate jaw wagging from each participant. We analyzed exclusively fluent (non-stuttered) productions from the two speaking tasks, SENT and PIC, in order to make a valid comparison of sympathetic activity between CWS and CWNS. For the SENT task, we selected 4–9 fluent speech segments. These segments included all three sentences produced in a row or in some cases, 1–2 sentences if a sentence was unusable due to movement artifact, for example. Given that SNS responses typically occur within 1–3 s of a stimulus (Lim et al., 2003; Andreassi, 2007; Dawson et al., 2007), we selected fluent utterances that were not immediately adjacent (within ∼3 s) of a stuttering-like disfluency for the speech tasks. We segmented connected speech into utterances or verbal productions bounded by grammatical closure, intonation contours, or long pauses following Systematic Analysis of Language Transcripts (SALT) conventions (Miller and Iglesias, 2006). Onset/offset indices for the tasks were adjusted so that segments selected for analysis did not contain motion and other artifact. Finally, for the MAX task, we selected the first three successful trials in which the child inhaled and blew out on the party blower holding it extended for at least 3 s.

Skin conductance changes reflect the activation of sympathetic cholinergic neurons innervating the eccrine dermal sweat glands, and they are sensitive indices of the modulation of arousal during emotional, cognitive, and physical behaviors (Venables and Christie, 1980). SCLs vary in the time span of tens of seconds, while SCRs, transient changes in skin conductance seen as rapid increases superimposed on the tonic background level, vary approximately 1–3 s (Dawson et al., 2007). Investigators have recorded SCL and SCRs during fMRI scanning while young adults used biofeedback to elevate or lower SCL (Nagai et al., 2004). Findings from this and other studies suggest that cortical systems with inputs to brainstem circuits are active in controlling background SCL and generating transient SCRs (Nagai et al., 2004; Critchley, 2005). Our results demonstrate that CWS have higher SCL during task performance compared to their fluent peers; whether this higher background SNS activity reflects higher levels of attention and focus on the tasks (Iani et al., 2004), higher performance anxiety (e.g., Simpson et al., 2001) or some other factor, we do not know. We also found a high degree of heterogeneity among the CWS in their levels of SCL during task performance, with most participants’ data overlapping that of the CWNS (Figure 3).

In an experiment designed to manipulate emotional responses to positive and negative video clips in preschool CWS and CWNS, Jones et al. (2014b) also measured SCL. The pattern of results was mixed when CWS were compared with CWNS across positive and negative viewing tasks. However, in the cases in which significant between group differences were observed, SCL was higher for CWS. We did not find significant differences in SCR amplitude or SCR frequency between the two groups, although CWS showed a trend for slightly higher SCR amplitude and frequency. There is evidence that the generation of SCRs is regulated through circuits involving cingulate, insula, right parietal lobe, motor cortex, and ventral medial prefrontal cortex (Nagai et al., 2004). Of course these imaging studies involved adult participants, and whether or not four and five-year-old children engage similar neural circuits in autonomic regulation is unknown. We cite this literature here simply to emphasize that the easily recorded signal indexing increased conductivity through eccrine sweat gland activity is, in fact, driven by complex cortical and brainstem networks.

We turn to the hypothesis advanced in the introduction that SNS activation co-occurs with normal control of movement and that excess sympathetic activation could be a contributor to motor instability in early stuttering. Most of the autonomic indices we assessed did not differ between the two groups across tasks. The fact that we observed higher levels of SCL across all tasks and that the overall pattern of relative SCL amplitude was similar for the groups of CWS and CWNS does not lend support to the idea that the neural activity driving higher SCLs is related to speech production. One hypothesis to consider is that speech specific increases in SNS activity develops later in CWS and/or that speech-specific phasic increases in SNS arousal would more likely be observed surrounding stuttering events. We addressed the latter hypothesis in a companion study (Walsh and Usler, in press). We found that stuttered utterances produced during a picture description task by CWS were associated with significantly higher phasic SCR amplitude, SCR frequency, and reduced BPV compared to fluent utterances. This suggests that transient sympathetic responses increase in intervals surrounding stuttering behaviors and therefore reflect changes in fluency states. Future studies will need to consider the potential bidirectional influences between SNS and speech neural control systems that can change dynamically during speech production driven by a variety of factors both internal and external to the speaker.

Previous studies from our group and others provide some clues on a lifespan perspective on SNS activity related to speech production in stuttering and normally fluent speakers. Early investigations (Peters and Hulstijn, 1984; Weber and Smith, 1990) demonstrated that for AWS and AWNS, speech production tasks generally produced the highest levels of SNS arousal compared to difficult mental arithmetic (Peters and Hulstijn, 1984) and an effortful physical task (Valsalva maneuver, Weber and Smith, 1990). These results from adult participants contrast with those of most of the preschool children. Both CWS and CWNS typically produced the highest levels of sympathetic arousal during the forceful blowing task, which we designed to mimic the Valsalva maneuver. With the exception of SCR frequency (which was significantly higher for the two speech tasks, SENT and PIC, and MAX), maximal values of SCL and SCR amplitude and minimum values of BPV amp (all indicators of increased SNS activation) typically occurred in the MAX blowing task. In fact, for both groups of children, SCR amplitude almost doubled in the MAX condition compared to the two speech conditions (SENT and PIC, Figure 2). Weber and Smith (1990) show similar plots for groups of AWS and AWNS (their Figure 3), indicating SCR peaks were much smaller in amplitude compared to the preschoolers in this study and were approximately equal across the speech and Valsalva conditions. We actually observed lower SCL in both groups of adults in the Valsalva compared to the speech conditions. Taken together, these results suggest that the relatively high levels of sympathetic activation occurring in adults prior to and during speech, indicative of tight coupling between sympathetic arousal and speaking, emerges with maturation. Most preschool participants showed the highest sympathetic activation to a physically demanding task rather than to the more cognitively demanding picture description task. For a subgroup of the CWS, however, this typical pattern of highest SNS activation in the MAX task was not observed. Approximately one-quarter of CWS showed higher SCL levels in the picture naming task compared to the MAX task (Figure 3). This suggests that some CWS are atypically early in developing the adult pattern of higher levels of arousal during speaking compared to physically demanding tasks. This preliminary result is worth further exploration, as it will be important to determine whether a relationship exists between early developing speech motor/sympathetic outflow and stuttering persistence and recovery.

Also relevant to the issue of a lifespan perspective on autonomic regulation related to speech, we found no differences in pulse rate between the two groups of children. The highest PRs were observed in the two speaking conditions, while lower PRs were achieved for the non-speech conditions (Figure 4). PR reflects both sympathetic and parasympathetic control of heart rate, suggesting that CWS and CWNS did not differ in sympathetic/parasympathetic control of PR during fluent speech production. Jones et al. (2014b) used respiratory sinus arrhythmia (RSA) derived from heart rate variability as an exclusive index of parasympathetic “tone” in a study of preschool CWS and CWNS (cf. Grossman and Taylor, 2007). They found reduced baseline differences in RSA in CWS, evidence they interpreted as placing CWS at greater vulnerability to emotional reactivity. Our PR results also contrast with findings from adults. Alm (2004) reviewed the available literature on heart rate and speech production tasks in AWS and concluded that a decrease in heart rate is associated with speech tasks in AWS. Increases in heart rate are primarily associated with increased SNS arousal. Alm interpreted this finding as support for the hypothesis that speech-related anxiety produces a “freezing response” associated with parasympathetic reduction in heart rate. We note that both of these prior studies included stuttered and fluent speech in their analysis. In order to compare SNS arousal during speech production in CWS and CWNS, we focused exclusively on fluent utterances. We examined PR during fluent and stuttered speech production in a separate study (Walsh and Usler, in press). Overall, these findings in studies of adults and children motivate future developmental studies of autonomic activation during speech and non-speech tasks in typical and atypical speakers.

There are additional results from an earlier study from our laboratory (Arnold et al., 2014) in which an experimental design similar to the present study was used to compare normally fluent young adults (18 to 22 years) and school-age children (7 to 9 years) across a similar set of speech/non-speech tasks. The results of this experiment generally accord with our findings that children have higher SNS activation during a Valsalva maneuver, while young adults showed highest levels during a sentence production task. However, it is not possible to directly compare the results of this earlier experiment with the present findings, due to our strategy of using pre-task baseline measures to express relative amplitudes of the autonomic variables. It is well established that there is significant within-individual variability in autonomic measures (Dawson et al., 2007); therefore, experimenters often express task-related amplitudes relative to a baseline condition. Ideally, baseline data is collected immediately prior to performance of each experimental task, so that, for example, adaptation to the experimental environment over time would not affect the measures. This is straightforward to do in adult participants, who are able to rest quietly for a few minutes between experimental tasks (e.g., Weber and Smith, 1990). In Arnold et al. (2014) we used the same procedure, but shortened the inter-task rest periods, because children could not be still and rest for lengthier durations. As a result, some findings from this study were unintuitive and difficult to interpret. Measures of task-related changes from baseline in some cases were lower than the pretrial “rest” baseline intervals, suggesting lower SNS activation during task performance compared to rest. More likely, the pre-task rest interval was not long enough for arousal levels to return to baseline from the previous task and/or simply reflect variability in the child’s behavioral states during the inter-condition rest periods.

This is a significant methodological issue for studies of autonomic activity in children. We began the present investigation employing inter-task rest periods for baseline data, but it became apparent that this was again problematic. We changed our experimental procedures to include a longer rest interval before any experimental tasks were undertaken and controlled for baseline levels of SNS indices in our statistical models. Our results suggest that this procedural change was successful in allowing us to obtain a meaningful comparison of SNS activation across tasks. Namely, the indices of SNS activity (increases in skin conductance and decreases in BPV) follow similar patterns across tasks in both groups of children (Figures 2, 4). As expected, the resting baseline measures were a significant covariate across the SNS dependent variables (Table 1), and SNS variables were highly correlated across tasks (Figure 3). Pilot data made it clear that the high levels of autonomic activity accompanying the forceful blowing task required ample time to resolve. Therefore, we elected not to vary the order of performance of the MAX task, having participants complete it at the end of the experiment. We counterbalanced the order of the JAW, SENT, and PIC tasks across participants. Taken together, these methods yielded results that were not contaminated by task order or adaptation effects.

Several studies using parent-report measures have noted temperamental differences between groups of CWS and CWNS, although the constellation of these differences varies and effect sizes reported by Alm (2014) are small. These studies found differences in attention regulation between CWS and CWNS (Karrass et al., 2006), with studies documenting hyperfocus in CWS (Anderson et al., 2003; Eggers et al., 2010). CWS were also less able to regulate their emotions (Karrass et al., 2006) or were vulnerable to increased frustration and anger compared to CWNS (Eggers et al., 2010). Other studies using parent-report measures, however, have not found temperamental differences between CWS and CWNS (Reilly et al., 2013; Kefalianos et al., 2014, 2017). The results of Kefalianos and colleagues are particularly compelling as these measures were sampled longitudinally, beginning prior to the onset of stuttering. We also found similar temperamental profiles in groups of CWS and CWNS; thus, our results do not support the hypothesis that temperamental differences distinguish preschool CWS and CWNS. Our findings, however, do not address the possibility that early temperamental profiles would distinguish children at higher risk for persistent stuttering. Ambrose et al. (2015) noted a significant, albeit small effect, for persisting CWS to have higher negative affectivity compared to both CWS who recovered and CWNS.

Regarding communication attitudes, we found that CWS achieved significantly higher KiddyCAT scores indicating more negative communication attitudes compared to their fluent peers. Interpretation of this finding is complicated by the fact that the mean KiddyCAT scores we calculated for each group were approximately half of the means reported for CWS and CWNS for this measure (Vanryckeghem and Brutten, 2006). Thus, both groups of children self-reported relatively few negative responses on this measure. The KiddyCAT scores we noted do not accord with the scores reported by other studies administering the English version of the KiddyCAT to preschoolers (Vanryckeghem et al., 2005; Clark et al., 2012). Guttormsen et al. (2015) reviewed assessments of communication attitude in children aged 3–18 years and documented negative communication attitudes in CWS beginning in the preschool years with differences between groups of CWS and CWNS becoming more apparent with advancing age. As with the CBQ results, our KiddyCAT findings do not address the role that communication attitudes may play in persistent stuttering.

A logical question is whether indices of physiological arousal levels are related to stuttering severity or to dimensions of temperament as measured by the CBQ. Our analyses of relationships among skin conductance measures and stuttering severity revealed an unexpected, negative correlation between the WSI and SCL for the PIC task (Figure 5). The correlation suggests that children with more severe stuttering have lower levels of task-related SCL. If one examines Figure 5, however, it appears that the correlation is driven by outliers at either end of the scales. The majority of our sample of CWS had WSI scores in the 3–13 range, and the data points within this range of WSI, suggest no systematic relationship with SCL. There is also the problem of the relative undersampling of children with higher WSI scores. Thus, our results are inconclusive on this issue, however, we note that overall our findings suggest no relationship between early childhood stuttering and levels of SNS arousal during speech. We did not compute correlations among skin conductance measures and TOCS or KiddyCat scores, because these score distributions again undersampled the higher end of the two scales. We also found no significant relationships between SNS variables and the CBQ composite scores. Therefore, we conclude that there is no simple mapping between the physiological measures of sympathetic arousal in preschool children during speech and non-speech tasks and any of the behavioral variables.