Participants were right-handed native English-speakers either self-referred (through the Clinicaltrials.gov registry: NCT05368350) or referred from the memory-specialty clinic at UT Southwestern Medical Center. Participant selection was limited to individuals in the early stages of PPA, defined as a confirmed diagnosis within the past 3 years. All participants had an initial diagnosis of PPA or PPAOS made by a neurologist, which were confirmed or further specified following comprehensive speech and language assessment. Individuals who were no longer verbally communicative were excluded at the time of initial inquiry. Functional status was assessed using the Reisberg Functional Assessment Staging Scale (FAST) (40) and those with >4 on the scale were excluded. Hence clinically these patients were at the stage of mild cognitive impairment or mild dementia, reflecting early-stage PPA. Participants who met the initial inclusion criteria (n = 11) were invited to have a first visit and completed baseline testing. Three were excluded from the data analysis: one (svPPA) was found to be more impaired than previously stated and was unable to complete baseline testing, one (nfvPPA) was excluded as English was a second language, and one (lvPPA) withdrew due to reduced tolerance of participation schedule (i.e., change in routine) and reported subjective decline in cognition after completing eight stimulation sessions. Eight PPA patients (M_age = 74.1 (range 69–79), female = 2) completed intervention and were included in the overall analysis. All participants who completed intervention provided relevant medical history, which is summarized in Supplementary material. Although EEG data from all eight participants were included in the baseline analysis, post-intervention EEG data were unavailable for PPA08 due to technical issues, resulting in seven participants with complete pre-post data.

An age- and sex-matched group of cognitively normal subjects (n = 7, M_age = 71.0 [range 67–79], female = 2) collected from prior studies [subgroup in Chiang et al. (41)] who underwent the identical EEG tasks were also included for EEG analysis to establish baseline ERP differences in PPA (PPA08 included for this baseline analysis) compared to this normal control group.

Baseline assessments included the Raven’s Progressive Colored Matrices (RPCM) (42) to evaluate nonverbal reasoning and the Western Aphasia Battery–Revised (WAB-R) (43) to assess aphasia presence and severity using Aphasia Quotient (AQ) and Language Quotient (LQ) scores. AQ scores ranged from no aphasia (AQ > 93.8) to moderate aphasia (AQ = 51–75).

Motor speech impairments, including apraxia of speech (AOS) and dysarthria, were rated using tasks from the Apraxia Battery for Adults–Second Edition (ABA-2) (44) which included repetition, speech rate tasks, oral reading, an apraxia checklist, and a narrative sample. Two trained graduate clinicians (authors KLM and CA) completed the initial ratings, which were verified by a licensed speech-language pathologist (co-first author CSD), with final consensus reached among all three raters.

Three participants were classified with nfvPPA, three with the lvPPA, and two with progressive AOS or dysarthria and minimal aphasia. Four participants had comorbid AOS or dysarthria, while two exhibited motor speech deficits with little to no aphasia. AOS was further classified as either “prosodic,” characterized by slow speech rate and reduced pitch variation, or “phonetic,” involving inconsistent phonological and articulatory errors (45). Dysarthria was categorized as either hypokinetic, marked by reduced vocal intensity, limited breath support, and increased speech rate, or spastic, characterized by strained vocal quality and difficulty initiating speech (46–48). Progressive speech apraxia may present with dysarthria in parkinsonism-plus syndrome that can eventually progress to clinical phenotypes of progressive supranuclear palsy or cortico-basal syndrome, indicating subcortical involvement (49). However, we did not have evidence due to lack of neuropathology or other ancillary testing in the research study. PPA subtype representation and associated motor speech impairments were relatively balanced across stimulation groups (see Table 1). Participant demographics, diagnostic classification, and baseline measures are provided in Table 1.

This open-label study, pre-registered at the Clinicaltrials.gov website (NCT05368350), aimed to include patients with a diagnosis of primary progressive aphasia and/or primary progressive apraxia of speech. All primary and secondary outcome measures were administered at baseline, immediate post, and 8-weeks post HD-tDCS intervention and alternate forms were administered when available. Participants were pseudo-randomized into stimulation groups to balance for PPA subtypes and severity. Participant group assignment is summarized in Table 1. Informed consent was obtained from all participants (and their legal representatives) in accordance with Declaration of Helsinki and the protocols approved by the Institutional Review Boards of The University of Texas at Dallas (for PPA and normal control) and The University of Texas Southwestern Medical Center (for normal control).

The HD-tDCS intervention protocol included one 20-min session per day over 10 days (five consecutive daily sessions per week for 2 weeks), using a Neuroelectrics Starstim system. During HD-tDCS, participants were instructed to sit quietly awake and at rest with their eyes open. The HD-tDCS montage comprised 5 Ag/AgCl 12-mm sintered pellet electrodes filled with conductive gel for the pre-SMA (one anode 1 mA at Fz and four 0.25 mA cathodes at FP1, FP2, F7, F8) and for the LIFG (one 1 mA anode at F7 and four 0.25 mA cathodes at T7, FP1, AF3, FC5 at 0.25 mA) (Figure 1). Electrodes were positioned in a neoprene cap. All patients underwent open-label active stimulation and were assigned to either the pre-SMA or the LIFG condition. Each stimulation session was preceded by a 30 s ramp up period to 1 mA current intensity followed by a 30 s ramp down period at the end. During stimulation sessions, a brief questionnaire was administered regarding adverse reactions intermittently (3–5×/10 sessions).

We use the term word retrieval to refer to the process of accessing and producing words belonging to any grammatical class. Word retrieval deficits are not limited to difficulties with nouns and verbs (i.e., content words), which are often observed in individuals with lvPPA and svPPA, but also extend to function words such as articles, prepositions, and conjunctions, which are more commonly affected in agrammatism associated with nfvPPA (1, 3, 50). Therefore, in this study, we selected tasks that tap into multiple aspects of word retrieval, including picture naming, verbal fluency, and discourse production, to serve as primary outcome measures.

The following standardized language assessments served as primary outcome measures administered at baseline, immediate post, and 8-weeks post stimulation: The Controlled Oral Word Association Test (COWAT) using “FAS” to assess phonemic fluency, the Category fluency tasks using “animals” (51), and the Boston Naming Test (BNT-60 items) (52). Discourse analysis was conducted on connected speech samples elicited through picture description using the Cookie Theft scene from the Boston Diagnostic Aphasia Examination (BDAE) (53). Samples were analyzed for words per minute (WPM) and two discourse measures: main concepts (MC) and core lexicon (CoreLex) targets (54–56). Scoring was performed by graduate student clinicians (authors: KLM, CA) and verified by a licensed speech-language pathologist (co-first author: CSD). Discourse analysis methods and secondary outcome measures and analysis are provided in Supplemental material.

Word retrieval has been shown to engage semantic inhibition/selection mechanisms, which plays a critical role in lexicosemantic search and retrieval processes (26, 27). To examine these semantic inhibition and selection process in individuals with word retrieval impairments, we employed a Go/NoGo EEG task which engages a retrieval circuit implicated these retrieval processes (26, 27, 36, 39, 41).

During these Go/NoGo tasks, subjects were instructed to push a button for certain stimuli (Go) while withholding responses for others (NoGo). The single car task (SC) involved basic categorization and used single exemplars of a car (Go) and a dog (NoGo). The object animal task (OA) involved superordinate categorization and used multiple exemplars of objects (Go) and animals (NoGo) across trials. Each task consisted of 200 trials: 160 (80%) ‘Go’ trials that required a response through button pressing and 40 (20%) ‘NoGo’ trials that required inhibition/withholding of a response (57). Concurrent EEG data were recorded from a 64-electrode Neuroscan Quikcap using a Neuroscan SynAmps2 amplifier and Scan 4.5 software (sampling rate: 1 kHz, DC-200 Hz). Trials were included for ERP analysis only if they were correct responses. Baseline correction was done by subtracting the mean amplitude of the pre-stimulus interval (−200 ms to 0 ms) from each time point. Individual ERPs were generated by averaging all the included epochs for Go and NoGo. Difference waves were also generated (NoGo minus Go epochs) for each individual. See Supplementary material for the detailed EEG preprocessing methods. We focused on N2/N200 and P3/P300 components around the midline electrodes. Electrode sites and time windows were selected a priori based on previous N200/P300 studies (38, 57, 58). For both Go and NoGo trials, we first measured peak latency between 150 and 300 ms for N200 at electrodes Fz-FCz-Cz, and between 250 and 650 ms for P3 at electrodes FCz-Cz-CPz. We used the peak latencies from all subjects’ baseline data for each group separately for each task (NC vs. PPA, SC vs. OA) and calculated mean amplitude within this window (mean latency ± 1 SD, separately for N200 and P300) to better estimate amplitude differences. We then averaged mean amplitude and peak latency across electrodes Fz, FCz, and Cz for examining N200 and across the electrodes FCz, Cz, and CPz electrodes for examining P300. This approach has been used and published in our previous studies (38, 58). Group level ERPs (NoGo – Go) are illustrated in Figure 2, and group level ERPs for Go and NoGo trials with topographies are illustrated in Supplementary Figures 1–3.

To evaluate effect of intervention, raw scores from primary outcome measures (phonemic fluency—FAS, category fluency—animals, BNT, WPM, MC, and CoreLex) were evaluated using linear-mixed effect models (LMMs) for each dependent variable with one between-subject factor of target (Group/stimulation condition: pre-SMA versus LIFG) and one within-subject factor of time (baseline versus post-treatment/follow-up) with a random intercept for subject to account for repeated-measures. LMMs were used to maximize statistical power given the small sample size within each group (n = 4) to account for within-subject variability and better assess group x time interactions. LMMs were conducted separately to evaluate immediate effects of treatment and potential maintenance of treatment gains at 8-weeks as compared to baseline measures. Standardized t-scores and z-scores were also reported for each individual participant for baseline, immediate post, and 8-weeks post in Supplementary Table 1. Clinical significance was measured using standardized assessments, normed for individuals with similar, age, sex, and levels of education. Clinically meaningful change was set at a change of at least 1 in z-score or 10 in t-score in the follow-ups compared to baseline and reported for primary outcome measures (15), which is a conservative threshold equivalent to improving at least two clinical interpretations for performance (e.g., moderately impaired to mildly impaired). All analyses were conducted using R (version 4.3.2) (59). LMMs were fit using the lmer() function from the lme4 package, Type III Wald chi-square tests were conducted to assess the significance of main effects and interactions, using the Anova() function from the car package (60). For additional fixed effects estimation, p-values and degrees of freedom estimated using the lmerTest package (version 3.1.3) (61). Post hoc comparisons were performed using the emmeans package (version 1.10.0) (62), with Tukey corrections applied to adjust for multiple comparisons. Given the small sample size, effect sizes were interpreted alongside 95% confidence intervals, and Hedges’ g was calculated to provide a bias-corrected estimate of standardized mean differences (63) and was computed using the effectsize package (version 0.8.9) in R (64). An alpha level of 0.05 was used for determining statistical significance.

For baseline performance, LMMs were administered to examine group difference (control vs. PPA), task effect (SC vs. OA) and their interaction in (1) Go-RT, (2) Go-accuracy, (3) NoGo-accuracy, (4) N2 latency, (5) N2 amplitude, (6) P3 latency, and (7) P3 amplitude [all of the ERP difference wave (NoGo – Go)] using R, with an alpha level set at 0.05 for significance. For treatment effects, LMMs were administered to examine target assignment (pre-SMA vs. LIFG), treatment effects (baseline vs. immediate-post), task effect (SC vs. OA) and their interactions in (1) Go-RT, (2) Go-RT, (3) NoGo-accuracy, (4) N2 latency, (5) N2 amplitude, (6) P3 latency, and (7) P3 amplitude [all of the ERP difference wave (NoGo – Go)].

All group means and standard deviations for primary and secondary outcome measures for each time point are provided in Table 2. In all separate LMMs, no significant between group effects were observed for any of the primary outcome measures (p > 0.05) so the following results report on the within-subject main effects of time and interactions.

For immediate effects, there was a significant time x condition interaction in phonemic fluency (FAS) outcome measure revealed from a Type III Wald chi-square test, (χ²(1) = 8.31, p = 0.004), indicating that changes over time differed by stimulation condition. Post hoc comparisons revealed that participants receiving LIFG stimulation produced, on average, 8 more words immediately following treatment (SE = 1.9, p = 0.006), corresponding to a large, statistically reliable effect (Hedges’ g = 2.98, 95% CI [0.60, 5.35]). No significant change was observed from those receiving pre-SMA stimulation (mean difference = −0.25, SE = 1.9, p = 0.900), indicating the improvement was specific to LIFG stimulation. There were no significant main effects of time (χ²(1) = 0.017, p = 0.895) or condition (χ²(1) = 0.046, p = 0.830). No other effects were significant for category fluency, BNT, WPM, CoreLex, or MC in terms of immediate post-treatment difference from baseline and time x condition interaction (all ps > 0.05).

For longer-term effects at 8-week follow-up, phonemic fluency no longer showed a significant time x condition interaction (p = 0.305), indicating both groups performed similarly. However, a significant improvement was observed in main concept (MC) scores (χ²(1) = 15.46, p < 0.001), indicating that participants improved from baseline to 8-weeks post treatment regardless of stimulation condition. Post hoc comparisons confirmed a significant increase in MC scores from baseline to 8 weeks (mean difference = 4.6, SE = 1.12, p = 0.006) with a corresponding large and statistically reliable effect-size (Hedges’ g = 2.06, 95% CI [0.40, 3.73]). No significant main effect of condition or interaction was observed (ps > 0.15). There was no significant main effect of condition on MC scores (χ²(1) = 0.12, p = 0.735) and no significant interaction (χ²(1) = 2.06, p = 0.151). No other significant main effects or interactions were observed for phonemic or category fluency, BNT, WPM, or CoreLex measures in terms of 8-week change from baseline (all ps > 0.05).

Given that statistical significance cannot be equated to clinically meaningful change, we calculated the percentage of PPA patients who made clinically meaningful improvement at the 2 follow-up time points (Table 3; see individual data in Supplementary Table 2). None showed clinically meaningful change in BNT, consistent with statistical findings. In addition, only the LIFG group (25% at immediate follow-up and 50% at 8-week follow-up) showed clinically meaningful change in phonemic fluency (COWAT), again consistent with statistical findings. In contrast, despite no statistical significance in category fluency (Animals), 2 patients in the LIFG group showed clinically meaningful change at immediate post follow-up and 2 each in the pre-SMA and LIFG groups showed clinically meaningful change at 8-weeks follow-up. We were unable to calculate this for WPM and discourse measures because standardized scores cannot be derived.

For RT, LMMs showed significant main effects of task (χ²(1) = 6.41, p = 0.011, Hedges’ g = 1.42, 95% CI [0.51, 2.32]) and group (χ²(1) = 4.94, p = 0.026, Hedges’ g = 2.27, 95% CI [0.03, 4.51]), while interaction between task and group was not significant (χ²(1) = 0.03, p = 0.866). Go-RT was faster in control than PPA (425 ± 140.8 < 594 ± 140.9 ms) and faster in SC than OA (457 ± 151 < 562 ± 151 ms). For other measures (accuracy and ERP measures), LMMs showed significant task effects in P3 latency (χ²(1) = 8.94, p = 0.003, Hedges’ g = 0.96, 95% CI [0.12, 1.81]); SC, 493 ms, shorter than OA, 537 ms. No other main effects were significant for any other measures (ps > 0.05). P3 latency data also showed a reliable trend in task x group interaction (χ²(1) = 3.01, p = 0.083) and post hoc comparisons revealed P3 latency significantly differed between tasks in the NC group, indicating OA > SC P3 latency in the NC group (mean difference = 73.1 ms, SE = 24.5, p = 0.011) which showed a large and statistically reliable effect size (Hedges’ g = 1.60, 95% CI [0.12, 1.81]), with no task differences in P3 latencies observed in the PPA group (p = 0.525). There was also a significant task x group interaction in N2 latency (χ²(1) = 4.28, p = 0.039). This was explained by SC > OA N2 latency in NC (293 vs. 256 ms) but SC < OA N2 latency in PPA (264 vs. 288 ms), although these differences were not significant in post-hoc comparisons (ps > 0.05). All other 2-way and 3-way interactions were not significant for any other measures (all ps > 0.05). Group level EEG task performance and ERP measures are reported in Supplementary Table 3. ERP waveforms are demonstrated in Figures 2, 3.

For task performance, LMMs revealed only main effects of task in Go-RT (p = 0.031, Hedges’ g = 1.35, 95% CI [0.45, 2.26]) showing faster RT in SC compared to OA (541 ms versus 628 ms, respectively). No other main effects or interactions, including treatment effects, were significant for other task performance measures (all ps > 0.05). For ERP measures, LMMs revealed main effects of treatment in P3 amplitude of the difference wave (χ²(1) = 8.84, p = 0.003); showing P3 amplitude was reduced from baseline at 1.17 mV to post-treatment at 0.57 mV. No other main effects were significant in any other measures (all ps > 0.05). There was also a reliable trend in treatment x target assignment interaction in P3 amplitude (χ²(1) = 3.38, p = 0.066). Post hoc comparisons showed a significant decrease of P3 amplitude in the LIFG group from baseline to post-treatment (from 1.39 mV to 0.18 mV, p = 0.018, Hedges’ g = 1.49, 95% CI [0.19, 2.80]) but no significant change in the pre-SMA group (from 0.96 mV to 0.96 mV, p = 0.991). No other two-way or three-way interactions were significant for any other measures. Group level task performance and ERP measures are reported in Supplementary Table 3 and ERP waveforms are illustrated in Figure 2. To further evaluate this treatment effect in P3 amplitude, we compared qualitatively control data to PPA data separately for the LIFG and pre-SMA groups (Figure 2). The LIFG group demonstrated larger difference between Go and NoGo P3 amplitude at baseline, while this difference was reduced and qualitatively more similar to normal at post-treatment testing (Figures 2, 3). In contrast, the pre-SMA group did not demonstrate evidence of a change between Go and NoGo P3 amplitude from baseline to post-treatment follow-up (Figures 2, 3).

This pilot study aimed to assess not only preliminary treatment effects but also the feasibility, tolerability, and safety of administering HD-tDCS at two target locations (pre-SMA, LIFG) in individuals with PPA. Safety and tolerability of anodal HD-tDCS stimulation at 1 mA was monitored across treatment sessions. All participants received active stimulation and across 80 treatment sessions (8 participants × 10 sessions), adverse effects were generally mild and transient. No serious adverse events such as seizures or psychotic symptoms were reported during our study. Adverse events were documented via questionnaire, with the most frequently reported effects being itching (16 out of 80 treatments; 20%) and tingling (15 out of 80 treatments; 19%). Sleepiness was reported in 5 sessions (6%), and headache occurred in 3 sessions (4%). Dizziness and migraine each occurred in 3 sessions (4%), with 2 out of 3 migraines localized to the stimulation site. Nausea was reported in 2 sessions (3%). Isolated instances of acute mood changes, burning sensations, skin redness, and mid-nocturnal insomnia were each reported in one treatment session (1%) across different participants. All adverse effects were self-limited, resolved without medical intervention, and no participants discontinued treatment due to discomfort. Although one participant showed reduced tolerance to complete the study, this was attributed to difficulty tolerating changes in routine rather than any reported discomfort related to stimulation. Because the participant withdrew from the study, any potential subjective cognitive decline related to stimulation could not be established. These findings align with a systematic review of 209 tDCS studies (65), which reported no significant differences in adverse event rates between active and sham stimulation. The most commonly reported effects (i.e., itching, tingling, headache, discomfort, and burning sensation) occurred at comparable rates with those observed in the present study.

The datasets presented in this study can be found in online repositories. The names of the repository/repositories and accession number(s) can be found in https://osf.io/b73cn/.