Modulating Phonological Working Memory With Anodal High-Definition Transcranial Direct Current Stimulation to the Anterior Portion of the Supplementary Motor Area

Abstract

Background:

Phonological working memory is key to vocabulary acquisition, spoken word recognition, real-time language processing, and reading. Transcranial direct current stimulation, when coupled with behavioral training, has been shown to facilitate speech motor output processes, a key component of nonword repetition, the primary task used to assess phonological working memory. In this study, we examined the efficacy of combining overt nonword repetition training with anodal high-definition transcranial direct current stimulation (HD tDCS) to the presupplementary motor area (preSMA) to enhance nonword repetition.

Objective:

This study investigated whether 20 min of active or sham anodal HD tDCS targeting preSMA concurrently with a nonword repetition task differentially impacted nonword repetition ability.

Method:

Twenty-eight neurotypical college-age adults (18–25 years; 19 females, eight males, one nonbinary) completed a 20-min nonword repetition training task where they received either active or sham 1-mA anodal HD tDCS to the preSMA while overtly repeating a list of four-, five-, six-, and seven-syllable English-like nonwords presented in a random order. Whole nonword accuracy and error patterns (phoneme and syllable) were measured prior to and following training.

Results:

Following training, both groups showed a decrease in nonword repetition accuracy. The drop in performance was significantly greater for the active stimulation group compared to the sham stimulation group at the four-syllable nonword length.

Discussion:

The findings suggest that targeting the speech motor component of nonword repetition through overt training and HD tDCS to the preSMA does not enhance phonological working memory ability.

Phonological working memory refers to the ability to hold in memory the sequential order of individual speech sounds and is a foundational cognitive process. This memory system plays a key role in a wide range of language skills including, but not limited to, vocabulary acquisition and development, novel word learning, maintenance and reactivation of information during sentence and discourse processing, acquisition of reading abilities, and second language learning (Adams, 1996; Adams & Gathercole, 1995; Baddeley et al., 2017; Papagno et al., 1991). Apart from being one uniform memory system, research suggests that phonological working memory can be broken into two distinct parts. In models such as Baddeley's (Baddeley & Hitch, 1994), the ability to hold sequences of speech in memory is captured by the phonological loop, which consists of two parts, the short-term phonological store, and an articulatory rehearsal (historically called the articulatory loop). The phonological store encodes initial novel phonological forms and acts as an “inner ear” remembering speech sounds in the correct sequential order. Importantly, the phonological store retains information for only a moment before the memory trace is lost due to decay. Unlike the phonological store, the subvocal articulatory rehearsal acts as an “inner voice” repeating the speech sequences in a loop to prevent their decay and allows for construction of more permanent memories (Baddeley, 2010; Gathercole, 1998). Because of its relevance to language, there has been historically great interest in evaluating individual capacities and abilities of individuals' phonological working memory systems, as well as investigating methods to improve phonological working memory with the intent to improve language as a whole.

Different tasks have been designed to assess individuals' phonological working memory span for verbal content, but performance on these tasks varies depending on the makeup of the stimuli, such as the phonological complexity of the material to be retained (i.e., the number of digits, phonemes, syllables), the lexical status of the content (e.g., words vs. nonwords), and factors such as the depth of an individual's vocabulary and their linguistic prowess (Gathercole, 1995). One way researchers have attempted to control for intrinsic factors that influence performance on verbal memory span tasks is by using nonword repetition tasks, which are generally viewed as a relatively pure index of phonological short-term memory (C. Dollaghan & Campbell, 1998; Gathercole et al., 1994). In nonword repetition tasks, individuals overtly repeat novel syllable sequences of increasing length. Because real words are easier to maintain in memory and produce a motor engram for, it is pertinent to emphasize the importance of using novel verbal stimuli when assessing phonological working memory. Real-word phoneme sequences (e.g., digits) are easily produced, even on the first attempt, and thus obscure observation of the role of underlying representations that encode the sequence of movements required to produce speech. When speakers encounter these real-word sequences, they can rely on their extant linguistic knowledge of their language's speech motor templates (C. A. Dollaghan et al., 1995). In contrast, the production of novel phonological sequences allows for the direct examination of complex speech motor sequencing. An individual's phonological working memory capacity is typically measured by the number of phonemes or syllables correctly produced at each sequence length. The assumption is that nonword repetition tasks provide a measure of phonological working memory storage due to the typical phenomenon that task performance decreases as storage requirements are increased (e.g., increasing length of nonword or syllabic complexity). As such, nonword repetition tasks are believed to provide a more parsimonious measure of core phonological processing abilities (Gathercole et al., 1994), and deficits in nonword repetition are thought to indicate a breakdown in phonological working memory and, by extension, a possible cause of language impairment.

The link between phonological working memory deficits and language disorder is seen clearly in children with developmental language disorder (DLD) where nonword repetition deficits are a hallmark feature of the disorder (Ahufinger et al., 2021; Coady & Evans, 2008; Graf et al., 2007). Performance on phonological working memory tasks accounts for significant variance in spoken word recognition and sentence comprehension and production in both typically developing children and children with DLD, irrespective of environmental factors like race and maternal education level (Adams, 1996; Ahufinger et al., 2021; Gathercole, 2006; Gillam et al., 2019; Montgomery et al., 2021). Research shows a positive correlation between phonological working memory and vocabulary and sentence comprehension abilities in children and adults with and without language disorders (Ahufinger et al., 2021; Alt, 2011; Caplan & Waters, 1999; Gillam et al., 2019). This cornerstone phonological working memory deficit persists even in older children with DLD whose language impairment appears to have resolved (Bishop et al., 1996; Coady & Evans, 2008; Graf et al., 2007; Weismer et al., 2000). The exact nonword repetition deficit profile in DLD appears to be the degree to which errors increase with greater nonword articulatory complexity and/or as the number of syllables in the nonword stimuli increases (Ahufinger et al., 2021; Bishop et al., 1996; Graf et al., 2007; Montgomery, 1995). This decrease in performance with increased articulatory complexity would suggest that speech motor demands of nonword recall may also play a role in repetition accuracy in children with DLD in addition to simple storage alone (Archibald et al., 2013).

Attempts to boost phonological working memory have directly targeted the storage rehearsal processes that maintain the representations of speech sounds in short-term memory through the overt rehearsal of items. Approaches focusing on overt rehearsal have met with limited success, in particular with clinical populations, and even when relatively modest gains have been observed with rehearsal training, these gains have been found to persist for only short periods of time (Gathercole & Alloway, 2006). This may be in part due to the fact that repetition of novel stimuli requires more than mere storage and instead also relies upon the motor system to create a motor engram that accurately reflects the phonological makeup of the novel stimuli. To repeat novel phonological forms of words, such as “woogalamic,” the listener stores an accurate phonological representation of the nonword in memory to guide the production of the output that matches the phonological input. To correctly repeat novel words, listeners rely on more than just encoding, maintaining, and retrieving the sequential order and speech sounds. They must also have the ability to assemble an articulatory representation of the sound sequences necessary to execute the complex speech motor commands needed to produce a nonword that matches the stored phonological representation (Gathercole, 1996). Evidence that nonword repetition engages complex speech motor control is further supported by evidence from functional magnetic resonance imaging (fMRI). Neuroimaging studies examining the neural correlates of speech motor control of novel speech sequences (i.e., nonwords) show that covert repetition of novel syllable sequences results in decreased activation in lateral primary motor cortex, presupplementary motor area (preSMA), superior temporal cortex, inferior frontal gyrus, and cerebellar cortex (Rauschecker et al., 2008), areas all implicated in the creation, storage, and refinement of motor planning. Segawa et al. (2015) examined the neural mechanisms underlying the learning of new speech motor sequences where participants overtly produced novel meaningless syllables composed of both legal and illegal consonant clusters over 2 days of practice. Comparing the learned illegal sequences to novel illegal sequences following training, Segawa et al. (2015) observed greater BOLD response during production of the novel sequences in the brain regions linked to nonspeech motor sequence learning (e.g., basal ganglia, preSMA) as well as in brain regions associated with learning and maintaining speech motor–related regions. These findings suggest that the brain regions responsible for sequencing speech “chunks” were engaged for the novel as compared to learned sequences (e.g., preSMA and globus pallidus)—regions believed to form part of the basal ganglia–thalamocortical loop, where activity in the right preSMA has been implicated in selecting and inhibiting unwanted movements (Rae et al., 2014). Taken together, both Segawa et al. (2015) and Rae et al. (2014) highlight the importance of the preSMA in complex motor sequencing, which raises the question of the crucial function the preSMA plays during the planning component in tasks tapping phonological working memory (i.e., nonword repetition tasks).

To this point, fMRI and electroencephalography (EEG) show that preSMA is indeed a part of the cortical control network that supports phonological working memory (Cona & Semenza, 2017; Constantinidis, 2016; Linden et al., 2003; Marvel & Desmond, 2010; Mecklinger et al., 2000; Obeso et al., 2013). For instance, Marvel et al. (2012) observed lower activation of the preSMA region during verbal working memory tasks in subjects with opioid addiction, which corresponded to lower verbal working memory performance. Perrachione et al. (2017) showed greater activation, as measured through fMRI, in the supplementary motor area (SMA; encompassing both SMA proper and preSMA), bilateral superior temporal gyrus, and inferior frontal gyrus linked to increased phonological working memory demands on nonword repetition tasks. Furthermore, activation in the preSMA has been shown to positively correlate to increased task and cognitive demands (Cummine et al., 2017). Additionally, preSMA is highly connected to other brain regions associated with working memory, including, most significantly, the prefrontal cortex and caudate (Zhang et al., 2012). Studies also show that repetition of nonwords engages a network of premotor areas for articulatory planning and articulation and that the preSMA plays a direct supervisory role in the generation and subsequent sequencing of speech motor planning (Hartwigsen et al., 2013).

While purely behavioral measures focusing on storage have failed to produce reliable modulation of phonological working memory, there is growing interest in the use of transcranial direct current stimulation (tDCS) to affect or enhance behavioral outcomes in both the domains of working memory and speech motor control. tDCS is a noninvasive brain stimulation technique that modifies neuronal resting membrane potential and the level of spontaneous neuronal firing in the area of stimulation and surround interconnected neural networks, thus modulating subthreshold neuronal membrane potentials with anodal stimulation depolarizing the neurons, thereby increasing the probability of action potentials occurring, and cathodal stimulation hyperpolarizing neurons, thus decreasing the likelihood of action potentials occurring. Research suggests that tDCS does not induce activity in targeted regions during resting states but only during modulation of a given region during spontaneous neural activity (Biabani et al., 2018; Matsumoto & Ugawa, 2017; Thair et al., 2017). Traditional tDCS delivers direct current (DC) using large pads, most commonly between 25 and 33 cm² with the resulting stimulation across relatively broad areas of the cerebral cortex located between the anode and cathode, making focal stimulation of target cortical regions difficult to achieve without involving neighboring regions. High-definition transcranial direct current stimulation (HD tDCS), on the other hand, uses arrays of smaller, specifically designed electrodes to deliver more focal stimulation. One configuration, the 4 × 1 ring electrode array, uses a center electrode overlying the target cortical region surrounded by four return electrodes. The center electrode defines the polarity of the stimulation as either anodal or cathodal, and the radii of the return electrodes confine the area undergoing the excitability modulation. Brain modeling studies show that the area of cortex undergoing modulation using the 4 × 1 HD tDCS configuration modulates a restricted cortical area (Heimrath et al., 2015; Truong et al., 2014). By improving upon the localized specificity of tDCS, HD tDCS enables asking pointed questions about the behavior of specific brain regions. The use of tDCS to dorsolateral prefrontal cortex to improve working memory ability has been successful in both single-session (Ohn et al., 2008) and multisession (Richmond et al., 2014) designs. Concurrent use of tDCS and cognitive training has also been explored as a way to enhance verbal working memory, in particular with stimulation to the dorsolateral prefrontal cortex (DLPFC) concurrent with verbal working memory training, largely with the use of n-back tasks (Andrews et al., 2011; Ruf et al., 2017). A meta-analysis of studies examining the use of tDCS to improve working memory, primarily as measured by changes in performance on n-back working memory tasks in neurotypical adults, reveals small but significant effects of left DLPFC stimulation only. Importantly, these effects are evident only when tDCS stimulation in regions hypothesized to support working memory is coupled with working memory training at the time of stimulation (Hill et al., 2016; Mancuso et al., 2016).

Given the critical role the phonological working memory plays in both the acquisition and real-time use of language, the ability to boost phonological working memory abilities in both typical and clinical populations would be highly desirable. Due to the preSMA's involvement with sequencing and planning of complex speech motor control, in this study, we examine the ability of HD tDCS stimulation of the preSMA to improve the speech motor control component of phonological working memory in neurotypical adults through the concurrent use of tDCS and overt nonword repetition of novel speech sequences. Using the correct production of nonwords varying in syllable length (four-, five-, six-, seven-syllable) as the measure of improvement, we asked if the speech motor control component of phonological working memory can be improved in neurologically typical adults in response to concurrent anodal HD tDCS stimulation to the preSMA during nonword repetition training.

Materials and Method

Participants

The study was a single-blinded study. Participants were college-age adults ages 18–25 years (N = 28; 19 females, eight males, one nonbinary) from the University of Texas at Dallas. All participants had normal language, were all strongly right-hand dominant, and were monolingual speakers of English. No participant had a prior or current history of diagnosis of dyslexia, DLD, or other language learning disability. Exclusionary criteria based on Richmond et al. (2014) were used, and no participant had a history of head injury, neurological injury or disease, seizure, epilepsy, and/or neurological or psychiatric disorder. No participant was currently using any medications known to alter membrane stability such as psychotropic medications (e.g., stimulants or antidepressants). Females were excluded if pregnant, or if there was a possibility of being pregnant. All participants completed written informed consent protocols in accordance with the Declaration of Helsinki as well as the guidelines of the University of Texas at Dallas Institutional Review Board, which approved the protocol. Participants received college credit for their participation in the study.

Before participating in the study, participants completed a series of standardized language and cognitive measures and completed several questionnaires to assess medical history and emotional stability. These included a demographic background questionnaire, oral mechanism assessment, and the Edinburgh Handedness Inventory (Veale, 2014). To adequately assess complex speech motor control, subjects were administered a standard oral mechanism assessment to ensure adequacy of normal oral motor control. Each participant received within normal limits scores on this assessment. Because performance on working memory tasks has been shown to be influenced by the emotional state of the participant at the time of testing, a Neuropsychological Screening Form, the Positive and Negative Affect Scale (Watson et al., 1988), the Stanford Sleepiness Scale (Shahid et al., 2012), the Beck Depression Inventory (Beck et al., 1988), and the Beck Anxiety Inventory (Beck et al., 1988) were also administered.

Given the tight link between nonword repetition, vocabulary, and verbal working memory capacity, participants' vocabulary and verbal working memory capacity were measured using the Comprehensive Receptive and Expressive Vocabulary Test–Second Edition (CREVT-2; Wallace & Hammill, 1997) and Digit Span subtest of the Wechsler Adult Intelligence Scale–Fourth Edition (WAIS-IV; Wallace & Hammill, 1997). The CREVT-2 provides norm-referenced measures of both receptive and expressive oral vocabulary. Individual working memory span was assessed using the Digit Span test of the WAIS-IV. This task exposes the participant to increasingly larger amounts of information with instructions to indicate how much of the stimulus was immediately taken in by repeating what was heard in some sort of immediate response, thus depending on short-term retention capacity. Digits Forward and Digit Backward were administered using Inquisit Software (Draine, 2004). Following completion in the study, participants also completed an HD tDCS questionnaire to assess any adverse effects of the HD tDCS stimulation that may have occurred and determine if the participants were blinded to the condition assignment (active, sham).

Participants were randomly assigned to either the sham or active anodal HD tDCS condition. Fourteen participants (nine females) were assigned to the active condition (M = 20.5, SD = 1.5), and 14 participants (10 females) were assigned to the sham condition (M = 20.5, SD = 2.3). The individuals assigned to the active and sham groups did not differ in age, years of education, CREVT receptive scores, CREVT expressive scores, or forward digit span. However, the two groups' backward digit span scores were significantly different (see Table 1).

Table 1.

Age and standardized scores for language and cognitive assessment measures for active anodal HD-tDCS stimulation (active) and control (sham) groups.

Variable	Active (n = 14)			Sham (n = 14)			p
	M	SD	Range	M	SD	Range
Age in years	20.54	1.98	18–25	20.50	2.03	18–24	.47
Years of education	14.71	1.27	12–16	15.18	1.32	13–17	.60
Digit-Forwards a	7.07	1.33	5–9	6.64	0.93	5–8	.33
Digit-Backwards b	6.57	1.28	4–8	5.50	1.29	3–8	.03
CREVT-Receptive c	72.79	33.18	45–120	80.64	33.17	45–117	.53
CREVT-Expressive d	101.14	14.00	74–125	100.71	14.62	77–121	.93

Note. For each variable, age-scaled scores have a mean of 100 and an SD of 15 (except age in years).

^aDigit-Forwards = Wechsler Adult Intelligence Scale–Fourth Edition: Digit Forward score (Wechsler, 2008).

^bDigit-Backwards = Wechsler Adult Intelligence Scale–Fourth Edition: Digit Backward score (Wechsler, 2008).

^cCREVT-Receptive = Comprehensive Receptive and Expressive Vocabulary Test–Second Edition: Receptive Language score (Wallace & Hammill, 1997).

^dCREVT-Expressive = Comprehensive Receptive and Expressive Vocabulary Test–Second Edition: Expressive Language score (Wallace & Hammill, 1997).

Procedure

Participants were seen individually to complete the standardized test and the experimental tasks. Prior to participation in the study, participants were screened via telephone call to ensure they met all qualifying criteria. Once it was determined that an individual met the study inclusion and exclusion criteria, participants were instructed to arrive at the study with product-free clean hair and to abstain from consuming alcohol, recreational drugs, and caffeine 24 hr prior to their visit. Before starting the study, each participant affirmed they had followed the prestudy instructions.

The experimental task consisted of three phases: baseline, training (active anodal HD tDCS or sham), and posttest. Participants' baseline nonword repetition ability was assessed prior to training. Immediately following baseline nonword repetition assessment, participants were then randomly assigned to either the active (anodal HD tDCS) or sham condition and completed a nonword repetition training task lasting 20 min. Following training, nonword repetition ability was again assessed on a novel set of nonwords. Two different baseline and training nonword lists were created and were counterbalanced across participants, conditions, and phases of the study to ensure that participants were repeating novel nonword sequences, thereby preventing any potential practice/learning effects and allowing for the assessment of the participant's complex speech motor control abilities (see Figure 1). The procedure was the same across all three conditions. Participants heard a series of four-, five-, six-, and seven-syllable nonwords composed of English syllables and were asked to repeat each nonword after hearing it. To control for potential learning effects inherent in completing sets of nonwords of the same syllable length, the nonwords were presented in a fixed random order. For the baseline and posttest phases, participants were given 3 s to repeat the nonwords. However, pilot testing revealed that for participants to be able to complete a nonword repetition task continuously for the 20 min of tDCS stimulation, participants required 6 s to repeat each nonword during the training phase.

Figure 1.

Study design schematic. HD tDCS = high-definition transcranial direct current stimulation; NRT = nonword repetition task.

Stimuli were presented via laptop in a silent room via Bluetooth speaker at a comfortable listening level. Participants' responses were digitally recorded via an external microphone and an iPod and were then transcribed offline by a second coder. Participants' productions were scored at the phoneme, syllable, and word levels by a double-blinded trained research assistant (blinded to the subjects' group and the order of administration of baseline and posttest list version). Both the experimenter and the coder were blind to the participants' assignment to the active/sham conditions.

Upon completion of the study, participants completed a questionnaire inquiring about potential side effects to HD tDCS. These side effects reflect the most commonly occurring side effects resulting from transcortical electrical stimulation. In addition to side effects, the questionnaire also assessed successful blinding of the participants by asking them whether they believed they were in the active or sham group. All but one subject completed the final questionnaire form (N = 27). As such, data for the 27 subjects who completed the form are reported below.

Anodal HD tDCS

The same HD tDCS montage targeting the preSMA as used in our previous study (Berglund-Barraza et al., 2020) was used in this study (see Figure 2). The montage consisted of five electrodes; a central anodal electrode at FZ; and return electrodes placed at Fpz, F7, F8, and Cz, placed according to the 10–10 EEG system for electrode positioning (Starstim, Neuroelectrics). Central Quik-Gel Conductive Gel was inserted under each electrode to contact the scalp (3.1 cm² per electrode). For anodal stimulation, a wireless battery–driven multichannel DC stimulator (Starstim, Neuroelectrics) delivered a DC over the scalp. This design allows for focal delivery of anodal current to the targeted brain region, using a constant current of 1.0 mA while applying weaker cathodal current because it is split by a factor of 4. We developed the stimulation montage to target the preSMA by generating a model of the electric field distribution for the configured stimulation protocol with Neuroelectrics Instrument Controller software. For active HD tDCS, current ramped up over the first 30 s and maintained a continuous current at 1 mA for 20 min. For the sham condition, the same cap fitting protocol and electrode configuration were used, and the current ramped up for 30 s until it reached 1 mA and then ramped down to 0 mA over 30 s until being switched off and left off for 20 min. This ramp up induces the initial tingling sensation felt in the active condition and effectively blinds the participants as it is indistinguishable from active stimulation by study participants (Palm et al., 2013).

Figure 2.

High-definition transcranial direct current stimulation montage targeting the presupplementary motor area.

Stimuli

The stimuli for this study were taken from the “Space Alien and Nonwords” study (Gupta et al., 2004). This is a corpus of 2,500 nonwords ranging in length from one to seven syllables that all follow the phonetic rules of English. All of the nonwords in the corpus were digitally recorded by the same female speaker with a standard midwestern English accent. To avoid potential ceiling effects, for this study, the four-, five-, six-, and seven-syllable nonwords for the corpus were used. For the baseline and posttest assessment of participants' nonword repetition ability, two lists were created from the larger corpus (Versions A and B; Gupta et al., 2004), each consisting of a total of 60 words (15 at each of the four-, five-, six-, and seven-syllable length). The nonwords in each list were organized in a fixed pseudorandom order, so that the initial phoneme of a nonword did not occur more than 2 times in a row, and where no more than four nonwords of the same length occurred in a row. The lists did not differ in phonemic probability, biphone phonotactic probability, phoneme onset, and stress position (see Table 2). Participants were randomly assigned to Version A or Version B as baseline and then received the other version posttest. This resulted in half of the participants completing Version A at baseline and Version B at posttest and the other half of the participants completed Version B at baseline and then Version A at posttest.

Table 2.

Phonological and inflectional stress characteristics of the pre- and posttraining assessment measures.

Variable	Version A			Version B				p
	M	SD	Range	M	SD	Range	t(118)
Stress position a	1.48	0.50	1.00–2.00	1.62	0.49	1.00–2.00	−1.47	.15
Phonotactic probability	1.52	0.18	1.23–2.00	1.52	0.17	1.26–1.95	−0.02	.99
Biphone probability	1.01	0.01	1.00–1.04	1.02	0.01	1.01–1.05	−0.57	.57

^aStress position indicates whether primary inflectional stress of the nonword occurred on the penultimate or the antepenultimate syllable.

For the training phase of the study, two additional nonword lists (Training Lists 1 and 2) were created, which also consisted of four-, five-, six-, and seven-syllable nonwords from the corpus (Gupta et al., 2004). These training lists consisted of a total of 148 nonwords (37 at each four-, five-, six-, and seven-syllable length). A fixed pseudo-random order also was used for these lists, where the initial phoneme of a nonword did not occur more than twice in a row and no more than four nonwords of the same length could occur in sequence. Phonemic probability, biphone phonotactic probability, phoneme onset, and stress position of the nonwords also did not differ for the two training lists (see Table 3). Participants were randomly assigned to Training Lists 1 and 2, resulting in half of the participants in the active and half of the participants in the sham conditions being trained on List 1 and the remaining participants in each conditioning being trained on List 2 (see Figure 1).

Table 3.

Phonological and inflectional stress characteristics of the training stimuli.

Variable	Version A			Version B				p
	M	SD	Range	M	SD	Range	t(294)
Stress position a	1.52	0.50	1.00–2.00	1.51	0.50	1.00–2.00	0.12	.91
Phonotactic probability	1.54	0.16	1.22–2.03	1.52	0.14	1.26–2.11	1.06	.29
Biphone probability	1.02	0.01	1.00–1.07	1.02	0.01	1.00–1.06	0.78	.43

^aStress position indicates whether primary inflectional stress of the nonword occurred on the penultimate or the antepenultimate syllable.

Reliability

A total of 33% of the participants' total productions at both baseline and posttraining were recoded by a second coder blind to the participants' group assignment to the sham or active conditions. Results revealed that there was 93% agreement transcription agreement between the two coders at the phoneme level.

Results

Response to Anodal HD tDCS

Analysis of the responses to the anodal HD tDCS in the follow-up questionnaire revealed that the participants in the sham and active conditions were similar in their responses (see Figure 3). A Mann–Whitney test revealed that the groups did not differ in the participants' beliefs regarding which that they had been assigned to the active condition (sham: 79% participants; active: 71% participants) or sham condition (sham: 21% participants; active: 29% participants), U = 89.5, p = .92.

Figure 3.

Responses to post–high-definition transcranial direct current stimulation (HD tDCS) questionnaire on the side effects experienced during stimulation for the active and sham conditions. Responses ranged from 0 = not present to 5 = severe for each of the 10 side effects.

Characteristics of the Nonword List at Baseline and Posttest

To ensure that any observed differences in performance for the active and sham groups were not due to potential differences in the nonwords lists themselves, the percent words correctly produced for each length for the participants assigned to Version A or B at baseline was first examined. A multivariate analysis of variance for group (Version A or B) × length (four, five, six, and seven syllables) revealed no difference in the percentage of words produced correctly at each length for the participants assigned to Version A or B for four-syllable words, F(1, 27) = 1.46, p = .24, η² = .05, power = .20; five-syllable words, F(1, 27) = 0.81, p = .237, η² = .03, power = .14; six-syllable words, F(1, 27) = 0.67, p = .43, η² = .02, power = .12; and seven-syllable words, F(1, 27) = 0.65, p = .42, η² = .02, power = .12. The lack of difference in performance for participants assigned to Versions A and B means that observed differences in performance of the participants assigned to the phonological working memory training alone (sham) or phonological working memory training + concurrent anodal HD tDCS stimulation (active) group could not be attributed to differences in the nonwords in the lists (e.g., Versions A and B).

Relationship Between Nonword Repetition, Vocabulary, and Digit Span

Because nonword repetition ability is influenced by vocabulary and verbal working memory capacity (Edwards et al., 2004; Gathercole, 1995), before examining the potential enhancement of anodal HD tDCS stimulation, the relationship between vocabulary (CREVT-Expressive [CREVT-E], CREVT-Receptive [CREVT-R]) and forward and backward digit span scores and percent nonwords correct produced at baseline (Time 1) and posttest (Time 2) for each word length (four, five, six, and seven syllables) was examined. Bivariate correlational analysis revealed that CREVT-E and CREVT-R were not significantly correlated with nonword repetition performance either prior to training (Time 1) or following training (Time 2; see Tables 4 and 5). In contrast to vocabulary, both forward and backward digit spans were significantly correlated with nonword repetition performance at both Time 1 and Time 2. Due to the significant correlation, both forward and backward digit spans were used as covariates in the analyses.

Table 4.

Correlation between receptive vocabulary (CREVT-REC), expressive vocabulary (CREVT-EXP), forward and backward digit span, and percentage of words produced correctly for each nonword length (4-, 5-, 6-, and 7-syllable) at Time 1 (T1).

Variable	T1 4-S	T1 5-S	T1 6-S	T1 7-S
CREVT-REC	.06	.26	.10	.27
CREVT-EXP	.06	−.02	.02	.00
Forward Digit	.51*	.66**	.60*	.33
Backward Digit	.36	.46*	.37	.17

Note. CREVT-REC = Comprehensive Receptive and Expressive Vocabulary Test–Second Edition: Receptive Language score; CREVT-EXP = Comprehensive Receptive and Expressive Vocabulary Test–Second Edition: Expressive Language score.

^*Correlation is significant at the .05 level (two-tailed).

^**Correlation is significant at the .01 level (two-tailed).

Table 5.

Correlation between receptive vocabulary (CREVT-REC), expressive vocabulary (CREVT-EXP), forward and backward digit span, and percentage of words produced correctly for each nonword length (4-, 5-, 6-, and 7-syllable) at Time 2 (T2).

Variable	T2 4-S	T2 5-S	T2 6-S	T2 7-S
CREVT-REC	.30	.34	.06	.32
CREVT-EXP	.01	.18	−.15	.23
Forward Digit	.52**	.59**	.35	.49*
Backward Digit	.35	.38**	.32	.53**

Note. CREVT-REC = Comprehensive Receptive and Expressive Vocabulary Test–Second Edition: Receptive Language score; CREVT-EXP = Comprehensive Receptive and Expressive Vocabulary Test–Second Edition: Expressive Language score.

^*Correlation is significant at the .05 level (two-tailed).

^**Correlation is significant at the .01 level (two-tailed).

Anodal HD tDCS Stimulation to the preSMA

The percent words correctly produced at Time 1 and Time 2 and the difference scores (Time 2 minus Time 1) for each of the four nonword lengths for the active and sham groups are shown in Table 6. To determine whether any observed differences in nonword repetition ability for the sham and active groups following training were due to HD tDCS stimulation, and not due to potential differences in the two groups' nonword repetition prior to training, the percent nonwords correctly produced at each nonword length (four, five, six, and seven syllables) at Time 1 was examined.

Table 6.

The percentage of nonwords correctly produced for each syllable length at Time 1, Time 2, and the difference score (Time 2 – Time 1) for the active and sham groups.

Variable	Active (n = 14)			Sham (n = 14)
	M	SD	Range	M	SD	Range
Time 1	%			%
4-syllable	75	20	33–100	69	17	33–93
5-syllable	41	22	0–73	33	22	7–80
6-syllable	16	12	0–47	13	09	0–27
7-syllable	4	04	0–13	7	12	0–40
Time 2
4-syllable	66	22	13–100	71	19	20–93
5-syllable	39	24	7–93	32	21	0–73
6-syllable	13	14	0–40	11	15	0–53
7-syllable	3	05	0–13	2	03	0–7
Difference (Time 2 – Time 1)
4-syllable	−9.5	13	−3.3–1.3	2.5	14	−2.0–2.7
5-syllable	−2.4	12	−2.0–2.0	−1.0	18	−4.0–3.3
6-syllable	−3.4	11	−2.7–1.3	−1.3	11	−2.0–2.7
7-syllable	−1.0	04	−0.7–0.7	−5.0	10	−3.3–0.7

Percent nonwords correctly produced at each length was compared for the active and sham groups at Time 1 using univariate analyses of covariance (ANCOVAs), with forward and backward digit spans as covariates. The results revealed that the two groups did not differ in the percent of nonwords correctly produced at Time 1 for any nonword length (four syllables: F(1, 27) = 0.06, p = .80, η² = .003, power = .05; five syllables: F(1, 27) = 0.00, p = .95, η² = .00, power = .0; six syllables: F(1, 27) = 0.009, p = .92, η² = .00, power = .05; and seven syllables: F(1, 27) = 1.70, p = .24, η² = .06, power = .24). Thus, the participants in the active and sham groups did not differ in their ability to correctly produce nonwords at each of the four syllable lengths prior to training.

To measure change in performance following training, the percent nonwords correctly produced at each nonword length at Time 2 (posttest) was subtracted from the percent nonwords correctly produced at each nonword length at Time 1 (baseline). This resulted in a difference score at each nonword length for each participant (for a graphical representation of difference scores across syllable length for the active and sham groups, see Figure 4). An ANCOVA with forward and backward digit scores as the covariates was first conducted comparing the difference scores (T2 − T1) for total percent nonwords correctly for the two groups. The analysis revealed that the active and sham groups differed significantly, F(1, 24) = 4.93, p = .036, η² = .17, power = .56, in the change in total percent nonwords at Time 2 as compared to Time 1. Follow-up analyses were then conducted to examine the effect of anodal HD tDCS stimulation at each of the four nonword lengths. Univariate ANCOVAs comparing the difference in percent nonwords correctly produced at each length following training for the active and sham groups revealed a significant in accuracy of nonwords produced for four-syllable nonwords, F(1, 24) = 6.36, p = .019, η² = .21, power = .68, where the active group produced significantly fewer nonwords correctly following anodal HD tDCS stimulation to preSMA concurrent with the nonword repetition training as compared to the sham group. There was no significant difference in the two groups' nonword repetition accuracy difference scores for five-syllable nonwords, F(1, 24) = 0.001, p = .97, η² = .00, power = .05; six-syllable nonwords, F(1, 24) = 0.38, p = .54, η² = .02, power = .09; or seven-syllable nonwords, F(1, 24) = 1.6, p = .21, η² = .06, power = .22. The results of this analysis revealed unexpectedly that anodal HD tDCS stimulation to the preSMA concurrent with nonword repetition training of novel nonwords resulted in a decrease in complex speech motor control following training, as compared to the sham group.

Figure 4.

Average difference scores (Time 2 – Time 1) for each syllable length for the sham and active conditions. Error bars represent the standard error.

We next examine the error patterns for the active and sham groups for the four-syllable nonwords to determine if the two groups differed in percentage of errors at the phoneme (e.g., naib vs. naif ) and syllable levels (e.g., taudge vs. judge). Univariate ANCOVAs, with forward and backward digit spans as covariates, revealed that, at Time 1, for four-syllable nonwords, the active and sham groups did not differ in percent phonemes correctly produced, F(1, 27) = 0.001, p = .95, η² = .00, power = .05 (sham: M = 94.5, SD = 4.7; active: M = 95.4, SD = 4.9) or in percent syllables correctly produced, F(1, 27) = 0.001, p = .95, η² = .00, power = .05 (sham: M = 88.8, SD = 7.7; active: M = 90.9, SD = 7.9). These results indicate that the error patterns at Time 1 were the same for active and sham groups.

We next examined potential differences in the error patterns at the phoneme and syllable levels for the active and sham groups following training. Univariate ANCOVAs, with forward and backward digit spans as covariates, revealed that the difference scores did not differ for four-syllable nonwords for the active as compared to the sham groups at the phoneme level, F(1, 27) = 3.74, p = .07, η² = .13, power = .45 (sham: M = 4.3, SD = 3.2; active: M = −1.8, SD = 4.5), or at the syllable level, F(1, 27) = 3.44, p = .08, η² = .12, power = .43 (sham: M = 2.0, SD = 3.2; active: M = −1.9, SD = 4.5). Taken together, these results indicate that error pattern did not differ qualitatively for the active and shams groups but that anodal HD tDCS stimulation to the preSMA concurrent with nonword repetition training of novel nonwords resulted in an overall higher rate of incorrectly produced nonwords following training for the active group as compared to the sham group.

One question was whether anodal HD tDCS stimulation to the preSMA concurrent with nonword repetition training resulted in a higher rate of nonwords produced incorrectly at Time 2, or if the overall number of participants in the active condition whose decrease in performance was greater than in the sham condition. A Mann–Whitney test revealed that the number of participants in the active and sham groups whose performance decreased from Time 1 to Time 2 did not differ (active: 11:14; sham: 5:14; U = 133, p = .06). This indicates that training of complex speech motor control via a nonword repetition task resulted in a decrease in the ability to correctly produced novel four-syllable speech sequences having the same error pattern for some individuals regardless of whether they were assigned to the active or sham condition.

Taken together, these findings indicate that training focused specifically on complex speech motor control, specifically novel speech motor sequencing in isolation, not only does not improve nonword repetition performance in neurotypical individuals but instead appears to degrade performance. More importantly, concurrent anodal HD tDCS stimulation to preSMA appeared to increase the rate of nonwords produced incorrectly produced by each individual in the active group at posttest as compared to those individuals who received complex speech motor training without stimulation. These findings show that focused training of novel speech motor sequencing in isolation does not improve the complex speech motor control component of phonological working memory in neurotypical individuals; instead, it appears to degrade performance. Moreover, these findings from this study show that concurrent anodal HD tDCS stimulation to preSMA in neurotypical individuals compounds the negative effect.

Summary of Results and Discussion

In this study, a group of neurotypical adults received either active anodal HD tDCS or sham stimulation for 20 min while completing one session of a behavioral phonological working memory training task. Based on prior studies showing enhanced verbal working memory in response to anodal HD tDCS to the DLPFC using n-back tasks, for the participants in the active condition, we predicted improvement from baseline in phonological working memory performance above and beyond that observed for the neurotypical adults in the rehearsal-only condition. Unexpectedly, this was not the case. Specifically, we observed that phonological working memory performance was significantly worse than baseline for the active group compared to the sham group following 20 min of concurrent active anodal HD tDCS stimulation to the preSMA. Interestingly, follow-up analysis revealed that a portion of the sham group also had decreased performance following 20 min of phonological working memory training, suggesting that this form of behavioral training does not facilitate improved phonological working memory abilities in neurotypical young adults. The results of this study show that neither phonological working memory training in the form of overt rehearsal alone nor phonological working memory training combined with anodal HD tDCS to preSMA improved phonological working memory in neurotypical adults.

The goal of this study was to increase the efficiency of phonological working memory training with concurrent anodal HD tDCS stimulation and presumed activating neurons in the preSMA, with our hypothesis being that, with anodal HD tDCS stimulation delivered to the preSMA, this would increase the spontaneous neural excitability and cell firing of neurons in cortical regions actively recruited during task with the subsequent improvement of phonological working memory performance. However, the decrease from baseline performance in the active group suggests instead that anodal HD tDCS stimulation to the preSMA may have temporarily disrupted the region's ability to maintain the correct sequential information of the syllable sequences of the nonwords.

The vast majority of tDCS studies focused on enhancing cognitive performance in the form of n-back or digit span training with tDCS most often applied to DLPFC in neurotypical adults have been largely equivocal showing neither strong gains in performance nor loss of performance (Mancuso et al., 2016; Tremblay et al., 2014). It may be that the effects of HD tDCS stimulation are small and behavioral performance of the participants in prior studies may have been at or near ceiling prior to training. A unique aspect of this study was that the stimuli used to train phonological working memory were extremely complex with respect to the demands on the phonological store due to the longer syllable length of the nonwords used in this study (i.e., four to seven syllables). Furthermore, unlike most nonword repetition tasks where participants are presented with all the nonwords of a given syllable length in one block, in this study, because the nonwords were presented in a random order across the syllable lengths, it required participants to constantly update aspects of the stimuli (i.e., word length) and removed any potential benefits of predictive motor patterns to a specific syllable length. Because nonword repetition tasks have historically administered nonwords in a stepwise fashion, with all nonwords of a given length given sequentially, participants have been able to entrain to a specific word structure, thereby decreasing constraints on motor planning. This would suggest that one way to increase difficulty, and thereby reduce the risk of ceiling effect, is by randomizing the order of nonwords by length, which is what was shown in this study. The lack of effect for five-, six-, and seven-syllable nonwords appears to have been due to the random administration of syllable length, which increased complexity to the point where the participants were unable to perform at those nonword lengths (i.e., floor effect). This raises the question for future research of whether presenting the same stimuli in a stepwise syllable length fashion would lead to greater performance at all syllable lengths.

The results of this study are consistent with models of speech motor control. For example, the Gradient Order Directions Into Velocities of Articulators (GODIVA), a computer-simulated neural network model of speech motor, describes the ongoing parallel representations of speech plans through the stages of production (Bohland et al., 2010; Guenther, 2016). GODIVA provides a unified hypothesis for the mechanisms underlying phonological and phonetic processing that is localized to particular regions in the brain. In the GODIVA model, the preSMA plays an integral role in speech production. The preSMA contains rank-order selective cells whose activity increases before the last movement in a sequence. As stated in the introduction, increased activation in the preSMA is seen with increased complexity, such as increased word length and syllable structure. The preSMA is hypothesized to code for syllable type (i.e., CVC, CCVC, VCC, etc. [C = consonant; V = vowel]) as well as the positions of phonemes (not the exact phonemes) within each syllable type. Through a series of excitatory and inhibitory loops connecting the preSMA to the basal ganglia and the inferior frontal sulcus (IFS), the preSMA creates a motor plan based upon the activated syllable structure and the particular phonemes selected by the IFS. Once a syllable's motor plan has been created and the corresponding preSMA neurons reach a large-enough activation threshold, this motor plan is sent through to the SMA to begin initiation of the motor process. Perhaps the most important property for the preSMA hypothesized for this study, the preSMA activates in a gradient manner, where neurons with greater activity are those occurring earliest in the production stream, and those with less activation occurring later in the stream, with the lowest activation produced last. As one syllable's motor process is sent through to the SMA, the corresponding neurons in the preSMA are deactivated, and all other gradient-activated neurons increase in activity. Due to the gradient and temporal intricacies of neuronal firing within the preSMA, it is possible that by increasing the spontaneous neural excitability throughout the entire region, active HD tDCS stimulation disrupted the temporal activation patterns necessary for effective preSMA function. As a result, the increased constant activity created a cortical environment unable to efficiently parse out the syllabic content of the novel stimuli, thus impeding the individual's ability to correctly reproduce the nonword. Therefore, while in this study anodal HD tDCS stimulation to the preSMA appears to have disrupted the region's ability to track syllable structure and sequences, the decline in performance in the active group does provide additional evidence in support of the GODIVA model's predictions that this region is involved in motor sequence planning during speech production.

There are some limitations in this study that bear mentioning. First, because only performance on nonword repetition was measured, only near effects were assessed following training and the potential impact of far impacts in changes beyond nonword repetition (e.g., verbal working memory, general working memory) was not assessed. Given the general working memory processing that preSMA is involved with, it is possible that HD tDCS stimulation in conjunction with phonological working memory training also changed abilities in other working memory domains, and not just nonword repetition ability. Indeed, it is possible that working memory tasks that do not alter the structural makeup of the stimuli, such as n-back, may have had improvements when pairing stimulation with phonological working memory training. Future research should investigate the potential modulation of multiple forms of working memory to truly understand the effects of combining HD tDCS stimulation with phonological working memory training. Additionally, many studies show greater effects where tDCS is applied over the course of multiple sessions over and above training alone, in working memory (Ke et al., 2019; Ruf et al., 2017), learning (Simonsmeier et al., 2018), addiction (Song et al., 2019), and more. While this is indeed missing in this study's ability to extend the findings to long-term changes in working memory, because the results showed a decrease in performance, this may in fact be positively viewed. Another limitation of this study is that follow-up measures were not taken to see whether one session of anodal HD tDCS stimulation to preSMA had a long-lasting impact on nonword repetition performance. Future research could remedy these limitations by using a multisession design, paired with multiple forms of HD tDCS (anode, cathode, sham), with follow-up measurements of behavioral task performance weeks or months after stimulation.

A second limitation of this study was the narrow range of activation resulting from the HD tDCS montage that was used. Because the goal of the study was the role of the preSMA in complex speech motor sequencing, the implemented montage focused on preSMA activation. As Gathercole (2006) notes in her commentary, the act of repeating a nonword is deceptively simple. The ability to repeat multisyllable nonwords relies on a “cascade of sensory, cognitive, and motor processes” (p. 514). Future studies that employ HD tDCS montages such as that used by Motes et al. (2020), which target the preSMA and dorsal anterior cingulate cortex, should begin to address the cognitive aspects of nonword repetition such as attention and executive functions.

A final point to mention is that while these data suggest that anodal HD tDCS to the preSMA may not enhance phonological working memory in neurotypicals, it is possible that application of cathodal HD tDCS would be beneficial. In one systematic review (Gathercole, 1996), it was found that cathodal tDCS stimulation decreased corticospinal excitability, which was linked to an increase in short interval intracortical inhibition and a decrease in intracortical facilitation. Because cathodal stimulation results in increased inhibitory cortical connections, future research could investigate the potential impact of cathodal HD tDCS stimulation to the preSMA, as this increased inhibition may help a participant's ability to inhibit unnecessary and conflicting information.

This study provides further evidence of the preSMA's functional role in phonological working memory. Additionally, while the results of this study showed a nonbeneficial impact of anodal HD tDCS stimulation to the preSMA for neurotypical adults, it may still be the case that anodal HD tDCS to preSMA may improve phonological working memory in individuals with neurodevelopmental disorders where the hallmark characteristic of the deficit profile is poor phonological working memory. Finally, this study shows the feasibility of HD tDCS to be used as a tool to examine models of speech motor control, such as the GODIVA model.

Data Availability Statement

The data sets generated and/or analyzed during this study are available from the corresponding author on reasonable request.

Funding Statement

The study was funded by the University of Texas Dallas research funds to Julia L. Evans and by the National Institute on Deafness and Other Communication Disorders (Grant R01 DC005650 awarded to Julia L. Evans). The authors gratefully thank the volunteer research subjects who participated in this study.