Augmenting Verb-Naming Therapy With Neuromodulation Decelerates Language Loss in Primary Progressive Aphasia

Abstract

Purpose:

The purpose of the study was to evaluate Verb Network Strengthening Treatment (VNeST) paired with the transcranial direct current stimulation (tDCS) of the left inferior frontal gyrus, which was compared to VNeST paired with a sham stimulation in primary progressive aphasia (PPA).

Method:

A double-blind, within-subject, sham-controlled crossover design was used. Eight participants with PPA were enrolled. Participants were enrolled in two treatment phases, one with VNeST plus real tDCS and one with VNeST plus sham. Participants received fifteen 1-hr sessions of VNeST in each phase. Linear mixed-effects models were used to compare changes between baseline and two follow-up time points (1 week and 8 weeks posttreatment) in naming trained verbs, untrained verbs, and untrained nouns; sentence production and comprehension; and producing content units and complete utterances in discourse.

Results:

VNeST was effective for significantly improving naming trained verbs and producing more complete utterances in discourse at 1 week posttreatment in both tDCS and sham conditions. A significant tDCS advantage yielded generalization of treatment effects to untrained verbs (at 1 week and 8 weeks posttreatment), sentence production (at 1 week posttreatment), and sentence comprehension (at 8 weeks posttreatment). Untrained verb naming and sentence comprehension declined when VNeST was not augmented with tDCS.

Conclusions:

Our findings provide emerging evidence that VNeST paired with tDCS can improve word finding, and other language abilities, in people with PPA. VNeST without neuromodulation can improve trained verb naming, but untrained verbs will likely decline faster when VNeST is not augmented with tDCS. Future research is required with a larger sample size to continue investigating the potential of treating word finding with VNeST and tDCS in PPA.

Supplemental Material:

https://doi.org/10.23641/asha.27914325

Primary progressive aphasia (PPA) is a devastating neurological syndrome characterized by the progressive decline of language functioning. PPA causes a gradual worsening of language to the point where many patients ultimately can no longer use or understand language. PPA results from neurodegenerative diseases including frontotemporal lobar degeneration pathology and Alzheimer's disease pathology (Gorno-Tempini et al., 2011). Language skills are disproportionately impaired relative to cognitive skills in earlier disease stages. Cognitive skills also eventually decline in later stages, but language impairment remains the most prominent feature. There are three main variants of PPA, which are each characterized by specific language impairments and brain atrophy patterns. These include logopenic variant PPA (lvPPA), semantic variant PPA (svPPA), and nonfluent variant PPA (nfvPPA). All three variants are characterized by word finding difficulty (anomia; Gorno-Tempini et al., 2011; Hillis et al., 2002). Individuals with lvPPA have impaired single-word retrieval and impaired repetition of sentences/phrases. They may also have phonological errors in spontaneous speech/naming, with spared single-word comprehension and motor speech. Individuals with the svPPA experience both impaired naming and single-word comprehension. They may also experience impaired object knowledge and surface dyslexia/dysgraphia. Speech articulation is spared in svPPA. nfvPPA (also sometimes referred to as agrammatic PPA) is characterized by agrammatism in language production, often with concomitant apraxia of speech. nfvPPA may also have apraxia of speech in the absence of agrammatism. People with nfvPPA may also experience impaired comprehension of syntactically complex sentences, but single-word comprehension and object knowledge are initially spared. Some patients with PPA cannot be classified into one of the three main variants either because they do not reliably meet the criteria for any specific variant but still have prevalent language deficits resulting from neurodegenerative disease or because they meet criteria for more than one variant (Utianski et al., 2019). Brain atrophy is most prominent in left temporoparietal regions in lvPPA, in bilateral (left > right) ventrolateral anterior temporal lobes in svPPA, and left posterior frontal and insular regions in nfvPPA (Gorno-Tempini et al., 2011).

Centrality of Verbs to Language

When treating anomia in language-impaired populations, clinicians may choose to treat nouns or verbs, but most treatment studies in poststroke aphasia (see Lavoie et al., 2017, for a review) and PPA (see Pagnoni et al., 2021, for a review) focus on nouns. This is an important consideration as verbs differ from nouns in several important ways that often make them more complex to process than nouns. First, verbs convey argument structure. A verb's argument structure is lexical information about the arguments that describes both the number and the relationships of participants needed to complete the action. The agent and the patient/theme are examples of thematic roles where the agent plays the role of the “doer” of the action while the thematic role of the patient/theme is the “receiver” of the action. For example, the verb “write” is a transitive verb that requires two arguments, the agent (the person/entity doing the writing) and the theme (what is being written). Some verbs have more than one permissible argument structure, which adds to their complexity. Verb processing requires processing both syntactic and semantic information including information about argument structure (Boland & Tanenhaus, 1991; Shapiro et al., 1993). This is important when working with people with aphasia as verbs that have more arguments or more than one possible argument structure are more difficult to name for people with stroke-induced aphasia even when naming the single verb (Kim & Thompson, 2000; Thompson, 2003). Verbs are also overall less imageable than nouns. Concrete verbs and nouns are both imageable, but concrete nouns are still more imageable than concrete verbs (Bird et al., 2003). Imageability is important as more imageable items are easier to name in picture naming tasks. Finally, in many languages, including English, verbs are more morphologically complex than nouns (Vigliocco et al., 2006).

Verb production is critical to communication as verbs are the basic building blocks of sentences. We do not typically communicate our thoughts using isolated single words; rather, we communicate using sentences, so it is imperative to consider both verb and sentence production for individuals with PPA. Given their centrality to language and importance to sentence production, it follows that verb-naming deficits are equally detrimental to functional communication as noun naming deficits (Rofes et al., 2015).

Behavioral Treatment Studies in PPA

People with PPA have few evidence-based treatment options. Treatment studies often focus on treating naming since word-finding issues are found across all PPA variants, and word-finding difficulty is often cited as one of the most frustrating symptoms of PPA (Volkmer & Beeke, 2015). Even though verb naming is critical, most behavioral word-finding treatment studies in PPA have focused on noun naming (Beeson et al., 2011; Graham et al., 1999; Grasso et al., 2019; Henry et al., 2008, 2013, 2019; Jokel & Anderson, 2012; Jokel, Cupit, et al., 2006; Jokel, Rochon, et al., 2006; Meyer et al., 2018; Newhart et al., 2009; Savage et al., 2014). While many studies are small case studies/series, treating nouns is often successful and leads to gains in treated items immediately posttreatment (Beeson et al., 2011; Henry et al., 2008, 2013, 2019; Jokel & Anderson, 2012; Jokel, Cupit, et al., 2006; Jokel, Rochon, et al., 2006; Kim, 2017; Meyer et al., 2018; Newhart et al., 2009; Savage et al., 2014), and sometimes gains can last for several months (Beeson et al., 2011; Grasso et al., 2019; Henry et al., 2008, 2013, 2019; Jokel & Anderson, 2012) or even up to a year posttreatment conclusion (Henry et al., 2019). Several studies report generalization of treatment to untreated nouns for some patients (Beeson et al., 2011; Henry et al., 2013, 2019; Jokel & Anderson, 2012; Meyer et al., 2018; Newhart et al., 2009), while others report no evidence of generalization to untreated nouns (Savage et al., 2014). A review by Croot (2018) concluded that reported generalization effects of lexical retrieval treatment to untreated words are small and unreliable. Furthermore, investigation of generalization of treatment to sentence or discourse production is rare.

Fewer studies have focused on verb naming, and most are small case studies/series (Beales et al., 2016; Lerman et al., 2023; Macoir et al., 2015; Taylor-Rubin et al., 2022). For example, Beales et al. (2016) provided four participants with PPA (three with svPPA and one with lvPPA) with self-cueing lexical retrieval treatment of multiple word classes (verbs, nouns, and adjectives). Verb naming improved in three participants, but one participant with svPPA had baseline verb-naming ceiling effects. Investigation of generalization of treatment to sentence or discourse production is rare. However, Beales et al. found that verbs in discourse improved in two participants with svPPA. They also found that percentage of content information units were maintained in the participant with lvPPA and significantly improved in the three participants with svPPA. No significant differences were found between word classes. Similarly, Taylor-Rubin et al. (2022) investigated a treatment where participants listened to target words (nouns and verbs), read the word, and repeated the word (Reading and Repetition in Presence of a Picture treatment; Croot et al., 2015; Nickels, 2002). Four participants with PPA were enrolled (three with svPPA and one with lvPPA). Treated single-word verb naming improved in three participants (two with svPPA and one with lvPPA), and treated single-word noun naming improved in two participants with svPPA. Verb phrase (sentence) production using treated verbs and nouns improved in the two participants with svPPA, which was associated with single-word verb treatment but not single-word noun treatment. Their results suggest that because verbs are essential for sentence production, verb treatment may have a greater likelihood of generalizing treatment effects to sentence production compared to noun treatment. No participants demonstrated improvement on untreated nouns or verbs. Finally, a recent case study (Lerman et al., 2023) reported word retrieval treatment for a bilingual woman with lvPPA using Verb Network Strengthening Treatment (VNeST; Edmonds, 2016; Edmonds et al., 2009, 2014). They documented language decline in the years leading up to the study. However, language decline was decelerated as they found that language skills were maintained in both the treated and untreated languages relative to baseline performance over approximately 3 months but did not significantly improve.

Noninvasive Brain Stimulation Augmentation of Behavioral Treatment for PPA

In recent years, there has been rapid growth in the number of studies in PPA pairing behavioral language therapy with noninvasive brain stimulation techniques such as the transcranial direct current stimulation (tDCS). tDCS is a safe, noninvasive, and relatively painless form of neurostimulation where a weak electrical current is applied to the brain via an anode and a cathode. The anode and cathode are composed of electrodes inserted into saline-soaked sponges, which are placed on the scalp, and a weak electrical current is induced between the two electrodes, which affects neuronal firing in brain areas targeted by stimulation. Anodal stimulation (anodal tDCS [A-tDCS]) enhances and cathodal stimulation (cathodal tDCS) decreases the likelihood of neuronal firing (Schjetnan et al., 2013). Typically, tDCS is administered for the first ~20 min of a treatment session because once about 20 min of tDCS has been administered, cerebral excitability induced by tDCS can last over an hour after tDCS stimulation ends (Nitsche & Paulus, 2001). Many tDCS studies also use a sham stimulation as a control because many participants typically only detect a tingling sensation on the scalp for the first 30 s of real tDCS stimulation; thus, an effective sham condition can be created by providing stimulation for the first 30 s and then ramping down to no stimulation over the next 15–30 s (Gandiga et al., 2006).

Enhancing the neuronal firing rate via A-tDCS promotes mechanisms that underlie long-term potentiation (a persistent strengthening of synapses following stimulation of synaptic activity) and long-term depression (a reduction of synaptic activity; Fritsch et al., 2010), which are both critical for learning and memory. Even a single tDCS session affects patterns of resting-state connectivity in brain regions close to the stimulation sites and in distant brain regions that are connected to stimulated areas as part of a network (Keeser et al., 2011). For example, research has demonstrated that using tDCS to target a specific region within the language network, such as the left inferior frontal gyrus (IFG), alters functional activity within the entire language network (Fiori et al., 2018; Tao et al., 2021).

A recent meta-analysis concluded that pairing behavioral language therapy with noninvasive brain stimulation can lead to greater benefits than language therapy alone in PPA (Nissim et al., 2020). Similar to behavioral treatment studies, many PPA treatment studies using neuromodulation focus on single-word noun-naming treatments (Cotelli et al., 2014). For example, Cotelli et al. (2014) treated single-word noun naming in 16 individuals with nfvPPA. Eight participants received A-tDCS targeting left dorsolateral prefrontal cortex, and eight received a sham stimulation. Treated and untreated noun naming improved in both groups between baseline and immediately posttreatment conclusion and 10 weeks posttreatment conclusion, but a significant tDCS advantage (i.e., naming was significantly better in tDCS compared to sham conditions) was found for treated items immediately posttreatment. Additionally, a tDCS advantage was found immediately posttreatment for functional speech production, which only improved in the tDCS group. Another research group has conducted several studies of single-word written noun picture-naming treatment paired with A-tDCS over left IFG (de Aguiar, Zhao, Faria, et al., 2020; de Aguiar, Zhao, Ficek, et al., 2020; Ficek et al., 2018; Tsapkini et al., 2014, 2018). They used a crossover design where patients served as their own control by participating in both a sham phase and a tDCS phase. In several studies, their results revealed a tDCS advantage for producing longer lasting treatment gains and enhancing generalization to untrained items (de Aguiar, Zhao, Faria, et al., 2020; de Aguiar, Zhao, Ficek, et al., 2020; Tsapkini et al., 2014, 2018). The same group conducted an additional study of both single-word noun and verb written naming paired with anodal stimulation of the left IFG (Fenner et al., 2019). Similar to their studies of written noun naming, a tDCS advantage was found for producing longer lasting improvements of trained verbs and promoting generalization to untrained items.

In our previous study, we reported results from a case series of three people with PPA (one with lvPPA and two with nfvPPA) who participated in a tDCS study with a crossover design (Sheppard et al., 2022). Participants completed two phases of treatment, which each consisted of 15 sessions of VNeST (Edmonds, 2016; Edmonds et al., 2009, 2014) therapy that was augmented by an anodal stimulation of the left IFG in one phase and a sham stimulation in the other phase. Unlike many studies that treat word-finding deficits with single words, VNeST treats verb-naming deficits through the production of sentences around target verbs (e.g., chop) and related arguments (e.g., the chef [agent of the action], the onion [theme of the action]). VNeST is designed to strengthen semantic relationships between verbs and their thematic roles.

We found that trained verb naming was maintained but did not significantly improve in all three participants. For untrained verb naming, both participants with nfvPPA had a significant tDCS advantage at 1 week and 8 weeks posttreatment conclusion. One participant with nfvPPA also had a tDCS advantage for sentence production and sentence comprehension at 1 week posttreatment. Additionally, the participant with lvPPA experienced a sentence comprehension tDCS advantage. Both participants with nfvPPA had tDCS in the second phase of the study, so even though naming is expected to worsen over time, tDCS plus VNeST still yielded a significant improvement in untrained verbs.

We chose VNeST because it is an evidence-based word-finding treatment developed for poststroke aphasia that is quite effective at improving anomia (Edmonds, 2016; Edmonds et al., 2009, 2014). In addition to improving trained verbs, VNeST promotes generalization to improve naming untrained verbs and nouns, as well as producing and comprehending sentences (Edmonds, 2016; Edmonds et al., 2009, 2014). Additionally, evidence suggests VNeST can promote generalization at the discourse level in poststroke aphasia as evidenced by an improvement in the percentage of complete utterances (utterances that contain a subject, verb, and object and are relevant to the context) produced in a discourse task (Edmonds et al., 2009, 2014).

Current Study

The current study expands the findings from our previously reported case series study (Sheppard et al., 2022) in several ways. First, we added five additional participants (N = 8) including three participants with svPPA. Second, unlike our case series study, we investigated baseline to posttreatment changes in discourse production in addition to word- and sentence-level skills.

We used a randomized, double-blind, within-subject, sham-controlled crossover design. A crossover design was used as there is significant heterogeneity in individuals with PPA, and a crossover design allows for each participant to serve as their own control. In one phase, participants received 15 sessions of VNeST augmented with A-tDCS applied to the left IFG (VNeST + tDCS), and in the other phase, participants received 15 sessions of VNeST paired with a sham stimulation (VNeST + sham). The left IFG was chosen as the stimulation site for several reasons. First, it is a vital component of the language network associated with language and verb production (Amunts et al., 2004; Costafreda et al., 2006; Grande et al., 2012; Grodzinsky, 2000; Hagoort, 2014; Hickok & Poeppel, 2007; Horwitz et al., 2003; Schnur et al., 2009) and verb naming (Crescentini et al., 2010; den Ouden et al., 2009; Havas et al., 2015; Martin & Cheng, 2006; Thompson-Schill et al., 1998). tDCS of the left IFG has been found to increase verb and sentence naming in poststroke aphasia (Fiori et al., 2019; Marangolo, 2013) and improve written verb production in PPA (Fenner et al., 2019). Finally, stimulating the left IFG has been previously shown to activate multiple regions of interest in the language network (Fiori et al., 2018; Meinzer et al., 2012).

Our study has several aims: (a) determine whether VNeST therapy (with or without stimulation) leads to improvement in naming trained verbs, untrained verbs, untrained nouns, sentence production, sentence comprehension, or discourse measures from baseline to any posttreatment time point (1 week or 8 weeks posttreatment); (b) determine whether there is a tDCS advantage (compared to sham) for trained verbs, untrained verbs, untrained nouns, sentence production, sentence comprehension, or discourse measures (greater improvement from the baseline with VNeST + tDCS vs. VNeST + sham) at 1 week posttreatment; and (c) determine whether tDCS contributed to a longer lasting improvement for trained verbs, untrained verbs, untrained nouns, sentence production, sentence comprehension, or discourse measures (greater improvement from the baseline with VNeST + tDCS vs. VNeST + sham) at 8 weeks posttreatment.

We hypothesized that VNeST would be effective for people with PPA, regardless of stimulation. We also hypothesized that there would be a tDCS advantage for trained verbs and a tDCS advantage for the generalization of treatment benefits to naming untrained verbs, and untrained nouns, producing sentences, comprehending sentences, and producing discourse.

Method

Participants

Nine individuals consented to the study. One individual did not pass the VNeST screening step and was not enrolled in the study. The eight participants with PPA who were enrolled in the study (M_age = 68.25 years, SD = 5.73; mean education = 16.63 years, SD = 3.07) included four men and four women (see Table 1). Two participants had lvPPA, three had nfvPPA, and three had svPPA. They were an average of 3.90 years post symptom onset (SD = 2.78). Inclusion criteria included (a) a diagnosis of PPA by A.E.H., a neurologist specializing in PPA; (b) naming deficits as indicated by < 80% accuracy on the Object and Action Naming Battery (OANB; Druks & Masterson, 2000); (c) at least 10 years of education; (d) native English speaker; (e) normal or corrected-to-normal vision and hearing; and (f) medically stable. Participants were also required to pass a VNeST screening to ensure they could successfully engage in each step of VNeST therapy. A clinician took the participant through one trial of VNeST with a sample verb and determined whether they were able follow along with each step. Exclusion criteria included (a) a history of neurological or psychiatric disorder affecting the brain or behavior besides PPA; (b) a history of significant drug or alcohol abuse; (c) a history of seizures in the last 12 months; (d) a history of brain surgery or metal in the head; (e) currently taking medications that substantially lower the seizure threshold (e.g., methylphenidate); and (f) currently taking N-methyl-D-aspartate (2) antagonists, which can reduce the effects of tDCS.

Table 1.

Participant demographic and baseline testing information.

Participant	PPA subtype	Sex	Education (years)	Age (years)	Post symptom onset (years)	Stimulation order	tDCS stimulation dose	Baseline testing
								WAB-R (AQ)	BNT	PPT–Short Form (%)	KD–Short Form (%)	SR (% of correct sentences)	SOAP–canonical (%)	SOAP–noncanonical (%)	ABA-2 (Apraxia Severity)
Participant 1	Semantic	M	20	71	2	Sham only	1 mA	71.4	25	57	40	Not assessed	50	50	Not assessed
Participant 2	Semantic	M	20	73	1	tDCS only	1 mA	77	25	86	67	80%	90	80	Not assessed
Participant 3	Logopenic	F	18	72	5–10	tDCS first	1 mA	73	6	100	100	60%	75	60	Not assessed
Participant 4	Nonfluent	F	14	69	7–9	Sham first	1 mA	43.7	39	100	80	0%	55	60	Severe
Participant 5	Nonfluent	F	16	56	5	Sham first	2 mA	61	50	100	100	0%	95	60	Moderate–severe
Participant 6	Logopenic	M	12	73	1.5	Sham first	2 mA	81.9	19	93	80	40%	95	50	Not assessed
Participant 7	Nonfluent	M	19	65	4.5	tDCS first	2 mA	53.2	27	100	87	0%	80	25	Moderate–Severe
Participant 8	Semantic	F	14	67	1.75	Sham first	2 mA	48	8	93	60	20%	85	95	Not assessed

Note. PPA = primary progressive aphasia; tDCS = transcranial direct current stimulation; WAB-R = Western Aphasia Battery–Revised (Kertesz, 2007); AQ = Aphasia Quotient from the WAB-R; BNT = Boston Naming Test (Kaplan et al., 2001); PPT–Short Form = Pyramids and Palm Trees Test–Short Form (Breining et al., 2015; Howard & Patterson, 1992); KD–Short Form = Kissing and Dancing Test Short Form (Bak & Hodges, 2003); SR = Sentence Repetition test from the National Alzheimer's Coordinating Center's Frontotemporal Dementia Battery from the National Institute on Aging; SOAP = Subject-Relative, Object-Relative, Active, and Passive Test of Sentence Comprehension (Love & Oster, 2002); ABA-2 = Apraxia Battery for Adults–Second Edition (Dabul, 2000); M = male; F = female.

Two of the eight participants completed one phase of the two-phase study. One participant (Participant 1) dropped out after Phase 1 due to emerging behavioral difficulties that are often a concern as PPA progresses to more advanced stages. The other participant (Participant 2) dropped out due to medical concerns unrelated to PPA and increasing difficulty following the treatment protocol due to language decline. Data from both Participants 1 and 2 are included in analyses. Additionally, data from three participants (Participants 3–6) were included in a prior case series paper (Sheppard et al., 2022). Ultimately, the analyses included data from all eight participants. Six of these participants completed both phases of the study, one participant (Participant 1) completed one phase of treatment and 1 week posttreatment assessment, and one participant (Participant 2) completed all of Phase 1. Data from Participants 1 and 2 are included in the statistical analyses as described in the Data Analysis section. Additionally, note that Participant 7 requested to not complete discourse testing and his data are not included in discourse analyses.

This study was approved by the Johns Hopkins University School of Medicine Institutional Review Board (Protocol No. IRB00174879). All participants, or their legally authorized representative, provided informed consent to participate.

Study Design

A randomized, double-blind, sham-controlled, within-subject crossover design was used in this study. The crossover design was used to reduce individual variability, which is important due to the inherent heterogeneity in the PPA population. Participants were randomly assigned to receive either tDCS in Phase 1 followed by sham in Phase 2, or sham in Phase 1 followed by tDCS in Phase 2. Participants had 15 VNeST sessions within each phase, for a total of 30 VNeST sessions across the entire study. Participants received 15 VNeST training sessions within the tDCS phase and 15 VNeST sessions within the sham phase (see Figure 1). Each session lasted for 1 hr, and participants received three to five sessions per week, depending on their personal preference and availability. Each treatment phase was followed by an 8-week washout period, which was designed to minimize potential carryover effects from Phase 1 to Phase 2 treatment. A language evaluation was conducted before therapy, 1 week after therapy conclusion, and 8 weeks after therapy conclusion for each phase; the 8-week posttesting assessment for Phase 1 also served as the Phase 2 baseline assessment (see Figure 1). This resulted in five time points (T1–T5) where T3 served as the 8-week posttreatment time point for Phase 1 and the baseline for Phase 2 (see Figure 1).

Figure 1.

Study design. Components of Phase 1 assessment and treatment are depicted in green, and components of Phase 2 assessment and treatment are depicted in blue.

Language Testing

Overall language skills were characterized at the initial baseline time point (see Table 1). The Western Aphasia Battery–Revised (Kertesz, 2007) was used to assess overall aphasia severity. The Pyramids and Palm Trees Test–Short Form (Breining et al., 2015; Howard & Patterson, 1992) and Kissing and Dancing–Short Form (Bak & Hodges, 2003) tests were used to assess semantic knowledge of objects and actions, respectively. The Boston Naming Test (Kaplan et al., 2001) assessed object naming. Auditory sentence comprehension was assessed using the Subject-Relative, Object-Relative, Active, and Passive Test of Sentence Comprehension (Love & Oster, 2002), which included canonical sentences (active and subject-relative sentences) and noncanonical sentences (passive and object-relative sentences). A short apraxia screening was administered by a speech-language pathologist. When the screening suggested apraxia may be present, the Apraxia Battery for Adults–Second Edition (Dabul, 2000) was administered.

Primary and secondary measures were evaluated at each assessment time point (see Table 2). The primary outcome measure in this study was the accuracy of naming 15–20 trained verbs on the OANB (Druks & Masterson, 2000). For Participants 1–5, the same set of 20 verbs was trained. For Participants 6–8, verb selection was personalized to focus on 15 verbs chosen by the participant. We prioritized verbs that were named incorrectly at the baseline on the OANB and verbs that would have the most functional use for individual participants. This protocol change was made to make the therapy more personalized and functional, as research shows personalization can maximize the potential benefits of therapy (Thiessen & Brown, 2021). Additionally, we changed the protocol to focus on 15 rather than 20 verbs in order to increase the number of times each verb was treated during the study.

Table 2.

Mixed-effects linear models of change in naming accuracy.

Variable	Trained verbs					Untrained verbs					Nouns
	Estimate	SE	df	t	p	Estimate	SE	df	t	p	Estimate	SE	df	t	p
(Intercept)	38.23	9.14	25	4.19	< .001*	53.94	7.62	11	7.08	< .001*	65.99	7.13	9	9.25	< .001*
tDCS	−2.44	6.20	724	−0.39	.694	1.51	1.74	3381	0.87	.384	−2.10	2.01	6642	−1.04	.297
Time: 1 week post	12.10	5.87	716	2.06	.040*	−6.07	2.64	3374	−2.30	.022*	0.26	1.90	6634	0.14	.890
Time: 8 weeks post	8.67	6.26	722	1.39	.166	−10.23	2.79	3380	−3.66	< .001*	−7.78	2.02	6640	−3.85	< .001*
Phase	1.78	3.89	722	0.46	.648	−14.01	2.79	3382	−5.03	< .001*	−6.90	1.26	6640	−5.48	< .001*
tDCS × Time (1 week post)	5.65	8.29	716	0.68	.496	14.56	3.73	3374	3.90	< .001*	3.53	2.69	6634	1.31	.190
tDCS × Time (8 weeks post)	−4.64	8.58	720	−0.54	.589	18.38	3.84	3377	4.78	< .001*	3.90	2.78	6638	1.40	.160

Note. SE = standard error; df = degrees of freedom; tDCS = transcranial direct current stimulation.

^* p ≤ .05.

The OANB consists of line drawings of 162 nouns and 100 verbs. We were also interested in investigating potential generalization to (a) naming untrained verbs and nouns, (b) producing and comprehending sentences, and (c) producing discourse. Specifically, word-level secondary outcome variables included naming 80–85 untrained verbs and 162 nouns on the OANB (all untrained items from the OANB). Sentence-level secondary outcome variables included total correct items (out of 30) on the Sentence Priming Production Test and the Sentence Comprehension Test from the Northwestern Assessment of Verbs and Sentences (NAVS; Thompson, 2012), reflecting the ability to produce/comprehend targeted syntactic forms (i.e., active, passive, subject-relative, object-relative, subject wh-questions, and object wh-questions). Discourse-level secondary outcome variables were measured using the cookie theft picture description task from the Boston Diagnostic Aphasia Examination (Goodglass et al., 2001). Discourse-level secondary outcome variables included the total number of content units (CUs) and the total number of complete utterances produced, reflecting the ability to successfully produce relevant discourse.

Scoring of the OANB and the NAVS was completed according to the instructions in each testing manual, but one phonological error per item was acceptable for production tasks. Treating clinicians were all speech-language pathologists (S.M.S., E.B.G., E.V., or K.R.). The treating clinician scored naming and sentence assessments and were blinded to whether participants were completing the tDCS or sham phase of treatment. All assessments were video-recorded to ensure accurate scoring. In cases where the clinician was unsure how to score a particular item, they conferred with one to two other clinicians to reach an agreement. Interrater reliability for production tests (OANB, NAVS, Sentence Production Priming Test, as well as total number of CUs and total number of complete utterances produced in the picture description task) was high (Cohen's κ = .878 for the OANB; Cohen's κ = .881 for the NAVS Sentence Production Priming Test, Cohen's κ = .882 for the total number of CUs, Cohen's κ = .956 for the total number of complete utterances).

For discourse production, all language samples were video-recorded and transcribed by research assistants with experience in language sample analysis. First, we scored the number of CUs produced using the protocol described in the work of Yorkston and Beukelman (1980). In cases where a participant repeated a CU, or a variation of a CU (e.g., mom, mother), the CU was only counted once. The percentage of complete utterances was calculated by coding the total number of utterances and the number of complete utterances. Complete utterances were defined as those that contained both a complete sentence frame and were relevant to the topic (Edmonds et al., 2009, 2014). A complete sentence frame was an utterance with a subject, verb, and object. As per the protocol described by Nicholas and Brookshire (1993), any syntactic, morphological, or phonological errors were deemed acceptable when calculating complete utterances. A relevant utterance was defined as one in which the participant referenced an object, person, or action within the pictured scene or made an inference about the scene. Complete utterances were required to meet the criteria of both a complete sentence frame and a relevant utterance. Utterance types were coded using Codes for the Human Analysis of Transcripts (MacWhinney, 2014) and calculated using a bespoke Computerized Language ANalysis (MacWhinney, 2017) command.

VNeST Language Therapy

The standard VNeST procedure (Edmonds, 2016; Edmonds et al., 2014) was slightly modified in this study as described below. A computer monitor was positioned in front of the participant while the treating clinician typed participants' responses on the screen. A target verb was given at the beginning of the trial, and the participant was asked to identify an appropriate agent (the referent completing the action) for the verb (“Who can/might (target verb) something?”) and an appropriate patient/theme (the referent receiving the action; “What could be (target verb)?”). If they could not produce an appropriate agent or theme, they were given three multiple-choice options on the screen. This procedure was repeated to form three triads for the target verb, where a triad was a sentence or phrase containing an appropriate agent–target verb–patient. The participant read each triad aloud. Including appropriate morphology and inflections (e.g., “The author writes the novel” for the triad: The author–writes–the novel) was optional but not required for moving to the next step. Next, the participant chose one triad on which to expand by answering three wh-questions (where? [at the library], when? [last year], and why? [because she loved writing]) to create an expanded sentence (“The author wrote the novel at the library last year because she loved writing”). The clinician provided clarification when appropriate if the participant did not comprehend the wh-question (e.g., Where? What location or places?). Once the expanded sentence was produced, they were next asked to recall the target verb from memory. Finally, the participant was encouraged to generate five sentences for the target verb. As described in the VNeST protocol (Edmonds, 2016; Edmonds et al., 2014), participants were provided with cueing when appropriate.

Two aspects of the VNeST protocol were modified in the current study. First, we did not ask participants to complete the semantic judgment step where participants are usually asked to decide whether a sentence has a semantically appropriate agent or patient. Second, we asked participants to generate five sentences in the final step instead of three to four agent–patient pairs. We did this to increase the number of sentences produced in a single treatment session to increase the likelihood of generalization from words to sentences in the face of neurodegeneration.

To ensure treatment fidelity throughout the study, clinicians were provided with a checklist of procedures (including a list of specific verbs for each participant) to be followed for each session and a data collection sheet. Checklists and data collection sheets were recorded during each session and were regularly reviewed by the team of clinicians. New clinicians were trained in specific treatment procedures and were observed during one to two sessions to ensure adherence to treatment protocol.

tDCS Parameters

Each participant received one phase with a real tDCS stimulation and one phase with a sham stimulation. A Soterix Medical 1×1 Clinical Trials device was used in both phases for the first 20 min of each VNeST session. After stimulation concluded, the therapy session continued until an hour of therapy time was reached. It is only necessary to provide 20 min of tDCS as the effects can last more than an hour (Nitsche & Paulus, 2001), thus providing that tDCS at the beginning of the session ensures tDCS benefits last throughout the remainder of the session. tDCS was delivered at 1 mA for Participants 1–4 and at 2 mA for Participants 5–8 due to a change in study protocol motivated by promising results reported in the work of Fenner et al. (2019) where PPA participants' written verb naming improved with a 2-mA stimulation of the left IFG. The anode consisted of a 5 × 5 cm sponge soaked in saline placed on the electrode and applied over the left IFG, corresponding to the F7 coordinate of the International 10–20 system (Homan, 1988). The cathode was placed on the right shoulder. A sham condition was created by applying stimulation to the scalp for 30 s and gradually decreasing the stimulation over the next 15 s. This is often an effective sham technique because most participants only feel the initial effects of active tDCS for the first 30 s of the stimulation (Gandiga et al., 2006).

Randomization and Blinding

Participants and research team members were blinded to whether a participant was completing the real or sham stimulation phase. A unique six-digit blinded code was provided for each phase for each participant to initiate stimulation. Codes were entered into the Soterix Medical 1×1 Clinical Trials device at the beginning of the session, which indicated whether it should induce a real or a sham stimulation. The order that participants received sham versus tDCS stimulation was randomized. Five participants (Participants 1, 4–6, and 8) received sham in Phase 1 and tDCS stimulation in Phase 2, and three (Participants 2, 3, and 7) received tDCS in Phase 1 and sham in Phase 2 (see Table 1). At the conclusion of each treatment phase, treating clinicians and participants were asked to guess whether they had received a real or sham stimulation. Their confidence in their guess was rated on a 5-point Likert scale (5 = extremely confident, 4 = considerably confident, 3 = moderately confident, 2 = slightly confident, 1 = not at all confident). Mean accuracy and confidence ratings were calculated.

Data Analysis

We investigated the differences between the baseline, 1 week posttreatment, and 8 weeks posttreatment for several measures, including our primary outcome measure (i.e., correct trained OANB verbs) as well as secondary outcome measures (i.e., correct untrained OANB verbs and nouns, accuracy on NAVS Sentence Comprehension and Production subtests, and the number of CUs and percentage of complete and relevant utterances on the cookie theft task). Each of these measures was evaluated by fitting linear mixed-effects models using the lme4 package (Bates et al., 2015) in R statistical software (Version 4.3.1; R Core Team, 2021). Mixed-effects regression models are appropriate to use in cases of incomplete data (Hedeker & Gibbons, 1997); thus, models were able to account for missing data from Participants 1 and 2 who only completed Phase 1 of the study using maximum likelihood estimation. The first model included accuracy of naming trained verbs as the dependent variable, with tDCS condition (real vs. sham), time point (baseline, 1 week posttreatment, or 8 weeks posttreatment), and a tDCS × Time Point interaction as the main predictors, while controlling for phase (Phase 1 or Phase 2) to account for potential declination in performance over time due to neurodegeneration. Additional models were identical but used untrained verb naming, trained noun naming, sentence production accuracy, or sentence comprehension accuracy as dependent variables. Discourse models used total number of CUs or percentage of complete utterances as dependent variables.

Adverse Effects

The Wong–Baker FACES Pain Rating Scale (Wong & Baker, 1988) was used at the end of each treatment session to evaluate whether the participant had experienced any pain or discomfort during the session. Participants were also asked whether they experienced any discomfort or tingling on the scalp.

Results

Integrity of Blinding

Participants had 64.3% accuracy with a 2.8/5 confidence rating in determining whether they received a sham or a real stimulation. Clinicians had 28.6% accuracy with a 2.2/5 confidence rating. Participants were slightly more accurate than chance at guessing the stimulation condition with a relatively low confidence rating. Clinician accuracy was below chance with a low confidence rating. These results suggest participant and clinician blinding was effective.

Adverse Events

No serious adverse events were reported throughout the study. One participant reported mild headaches at the conclusion of two sham tDCS sessions and during two assessment sessions that followed the sham phase. She rated her pain an average of 3/10 on the Wong–Baker scale and reported that she regularly experienced headaches prior to enrolling in the study.

Naming

Table 2 includes details of mixed-effects models for naming, and results are shown in Figure 2. For trained verb naming, there was a main effect of time at 1 week posttreatment, t(716) = 2.06, p = .040, indicating that VNeST significantly improved trained verb naming by an estimated 12% increase in accuracy from the baseline, collapsed across conditions. There was no significant interaction between tDCS condition (sham vs. real tDCS) and time at 1 week posttreatment, indicating that VNeST equally improved trained verb naming between the baseline and 1 week posttreatment between sham and tDCS conditions (see Table 2). At the 8-week posttreatment time point, trained verb naming had increased an estimated 9% from the baseline (across sham and tDCS conditions), but this improvement was not significant, indicating that trained verb naming was maintained and did not decline in the face of neurodegeneration at 8 weeks posttreatment. Similar to the 1-week posttreatment time point, there was no significant difference between the tDCS and sham stimulation for trained verbs. There was no significant effect of phase on trained verb naming.

Figure 2.

Naming results. Change in naming accuracy between baseline and two follow-up time points (1 week and 8 weeks posttreatment) for transcranial direct current stimulation (tDCS) and sham stimulation. Results include naming (A) trained verbs, (B) untrained verbs, and (C) nouns. A change above 0% indicates an improvement from the baseline, and a change below 0% indicates a decline from the baseline. Asterisks immediately above bars indicate a significant main effect at that time point. Brackets indicate a significant difference between tDCS and sham at that time point. Vertical error bars indicate standard error.

For untrained verb naming, there was no significant main effect of time at 1 week or 8 weeks posttreatment (see Table 2). Overall, when collapsing across tDCS and sham conditions, untrained verb naming declined by approximately 6% from the baseline to 1 week posttreatment and an estimated 10% from the baseline to 8 weeks posttreatment. However, significant tDCS × Time interactions indicated a significant difference between tDCS and sham conditions at 1 week, t(3374) = 3.90, p < .001, and 8 weeks posttreatment, t(3377) = 4.78, p < .001. Specifically, untrained verb naming improved from the baseline in the tDCS condition but declined in the sham condition at both the 1-week and 8-week time points. The significant interactions indicated that untrained verb naming was an estimated 15% greater for tDCS compared to sham at 1 week posttreatment. This tDCS advantage was maintained at the 8-week posttreatment time point where untrained verb naming was relatively stable in the tDCS condition with a continued decline in the sham condition, resulting in approximately 18% better accuracy for tDCS compared to sham at 8 weeks posttreatment. Additionally, a significant effect of phase, t(3382) = −5.03, p < .001, showed untrained verb-naming accuracy was significantly lower in Phase 2 relative to Phase 1 across conditions.

For untrained nouns, there was no main effect of time at 1 week posttreatment, indicating that noun naming was maintained with no significant improvement or decline at 1 week posttreatment compared to the baseline across tDCS and sham conditions (see Table 2). Noun naming significantly declined at 8 weeks posttreatment, t(6640) = −3.85, p < .001, by an estimated 8% from the baseline. There was no significant difference between tDCS and sham conditions at either time point for noun naming. Additionally, noun naming was significantly worse, t(6640) = −5.48, p < .001, in Phase 2 compared to Phase 1 by an estimated 7%. Results for individual participants are available in Supplemental Material S1.

Sentence Production and Comprehension

Results from mixed-effects models of sentence production and sentence comprehension are summarized in Table 3 and depicted in Figure 3. For sentence production, there was no significant main effect of time at 1 week or 8 weeks posttreatment. However, a significant tDCS × Time interaction at 1 week posttreatment, t(1222) = 2.24, p = .025, indicated that tDCS was significantly more beneficial than sham, with tDCS yielding an estimated 11% greater mean change from baseline accuracy compared to sham at 1 week posttreatment. There was no significant tDCS × Time interaction at 8 weeks posttreatment. Finally, there was no significant main effect of phase. For sentence comprehension, no main effect of time was revealed at 1 week or 8 weeks posttreatment. There was no tDCS × Time interaction at 1 week posttreatment. However, a significant tDCS × Time Point interaction, t(1225) = 2.28, p = .023, at 8 weeks posttreatment indicated tDCS significantly improved comprehension compared to sham; accuracy improved by an estimated 13% with tDCS versus sham. No significant main effect of phase was found. Results for individual participants are available in Supplemental Material S1.

Table 3.

Mixed-effects linear models of change in sentence production and comprehension.

Variable	Sentence production					Sentence comprehension
	Estimate	SE	df	t	p	Estimate	SE	df	t	p
(Intercept)	30.32	9.1	12	3.35	.006*	71.64	7.5	16	9.58	< .001*
tDCS	−0.59	3.8	1228	−0.16	.876	−8.15	4.3	1230	−1.92	.056
Time: 1 week post	1.43	3.6	1222	0.40	.692	−1.91	4.0	1222	−0.47	.637
Time: 8 weeks post	−1.51	3.8	1225	−0.39	.694	−3.86	4.3	1228	−0.90	.366
Phase	−2.57	2.4	1229	−1.08	.281	1.45	2.7	1228	0.55	.585
tDCS × time (1 week post)	11.43	5.1	1222	2.24	.025*	10.48	5.7	1222	1.84	.067
tDCS × time (8 weeks post)	3.89	5.3	1224	0.74	.460	13.39	5.9	1225	2.28	.023*

Note. SE = standard error; df = degrees of freedom; tDCS = transcranial direct current stimulation.

^* p ≤ .05.

Figure 3.

Sentence production and comprehension results. Change in accuracy between baseline and two follow-up time points (1 week and 8 weeks posttreatment) for transcranial direct current stimulation and sham stimulation. Results include changes to (A) sentence production and (B) sentence comprehension. A change above 0% indicates an improvement from the baseline, and a change below 0% indicates a decline from the baseline. Asterisks immediately above bars indicate a significant main effect at that time point. Vertical error bars indicate standard error.

Discourse

Results from mixed-effects models of discourse are presented in Table 4 and depicted in Figure 4. No significant effects were found for total number of CUs produced. Analyses revealed a significant main effect of time at 1 week posttreatment, t(27) = 2.76, p = .010, for percentage of complete utterances, which indicated an estimated 15% improvement from the baseline in percentage of complete utterances in both sham and real tDCS phases. There was no significant main effect of time at 8 weeks posttreatment. There were no significant tDCS × Time interactions, indicating that tDCS and sham conditions had a similar effect for percentage of complete utterances. Results for individual participants are available in Supplemental Material S1.

Table 4.

Mixed-effects linear models for discourse measures.

Variable	No. of content units					% Complete utterances
	Estimate	SE	df	t	p	Estimate	SE	df	t	p
(Intercept)	7.11	1.4	16	5.24	< .001*	48.75	14.5	10	3.36	.007*
tDCS	1.12	0.9	28	1.27	.213	−3.92	6.3	27	−0.62	.542
Time: 1 week post	1.00	0.8	27	1.30	.205	15.26	5.5	27	2.76	.010*
Time: 8 weeks post	0.35	0.8	27	0.43	.674	−1.79	5.9	27	−0.30	.765
Phase	−0.52	0.6	28	−0.86	.397	2.05	4.4	27	0.47	.642
tDCS × time (1 week post)	−1.67	1.1	27	−1.53	.137	−5.89	7.8	27	−0.75	.457
tDCS × time (8 weeks post)	−1.01	1.2	27	−0.87	.395	7.08	8.4	27	0.85	.405

Note. SE = standard error; df = degrees of freedom; tDCS = transcranial direct current stimulation.

^* p ≤ .05.

Figure 4.

Change in percentage of complete utterances. A change between baseline and two follow-up time points (1 week and 8 weeks posttreatment) for transcranial direct current stimulation (tDCS) and sham stimulation. A change above 0% indicates improvement from the baseline, and a change below 0% indicates a decline from the baseline. Brackets indicate a significant difference between tDCS and sham at that time point. Vertical error bars indicate standard error.

Discussion

The current study used a randomized, double-blind, within-subject, sham-controlled crossover design to compare VNeST + tDCS to VNeST + sham in eight participants with PPA. Our first aim was to determine whether VNeST, with or without tDCS, is beneficial for people with PPA. We found that VNeST is effective for improving trained verb naming and production of complete utterances both with and without neuromodulation at 1 week posttreatment. However, this effect was not maintained at 8 weeks posttreatment. Our second aim was to determine whether there was a tDCS advantage at 1 week posttreatment for any language measures, and our third was to determine whether tDCS would generate longer lasting language improvements apparent at the 8-week time point. We found a tDCS advantage for generalization of VNeST to untrained verbs at 1 week posttreatment that lasted through the 8-week time point. VNeST + tDCS also led to generalization to sentence production and comprehension. It appears that VNeST was beneficial in the short term for improving trained items, but tDCS boosted treatment effects and led to long-lasting generalization at the word and sentence level. The results suggest that VNeST will improve trained verb naming regardless of whether it is augmented with tDCS. They also suggest that even if patients with PPA receive VNeST, their untrained verb naming will continuously decline unless VNeST is augmented with tDCS. VNeST augmentation with tDCS led to an improvement in untrained verb naming rather than continuous decline that was observed without tDCS.

Our first aim was to investigate the efficacy of VNeST, with or without neuromodulation, on PPA as it is a well-proven treatment for poststroke aphasia. To our knowledge, VNeST has only been investigated in PPA in our case series study (Sheppard et al., 2022) and a recent case study (Lerman et al., 2023). Recall that in our case series study, we found that VNeST led to the maintenance of trained verbs in both tDCS and sham phases, and a tDCS advantage was found for untrained verbs in two of the three participants and for sentence production in one of the three participants. In Lerman et al.'s (2023) VNeST case study with no tDCS component, they found that word production, sentence production, and narrative skills were maintained over the ~3-month study in a bilingual participant with lvPPA. The current study provides evidence that VNeST is beneficial for improving trained verb naming in people with PPA with or without neuromodulation when measured 1-week posttreatment conclusion. During the 8-week washout period, treatment gains had improved relative to the baseline, but not significantly so. However, trained verb naming was maintained and did not significantly decline in the face of neurodegeneration, which is an optimistic result as maintaining the language network is an important goal in PPA. Generalization of treatment effects to untrained items is frequently reported in VNeST studies in poststroke aphasia (Edmonds, 2016; Edmonds et al., 2014). VNeST promotes generalization of treatment effects without tDCS augmentation in poststroke aphasia, though recent evidence suggests that additional generalization occurs with tDCS augmentation (Matar et al., 2022). However, untrained verb naming declined at 1 week posttreatment and continued to decline at 8 weeks posttreatment when VNeST was not augmented with tDCS. We only saw an improvement when VNeST was paired with tDCS. Thus, in our study, tDCS maximized treatment benefits by leading to generalization in PPA and by decelerating the decline of untreated verbs. Note that these results should be interpreted with caution as we enrolled a relatively small number of participants.

Our findings for trained words aligned with Tsapkini et al. (2014), who found that naming therapy was effective when paired with either sham or real tDCS. However, several studies have found evidence of a tDCS advantage for trained items (Cotelli et al., 2014; Fenner et al., 2019; Tsapkini et al., 2018). Specifically, in the work of Tsapkini et al. (2018), a tDCS advantage was found for participants who received real tDCS in Phase 1. In the Fenner et al. (2019) and Cotelli et al. (2014) studies, a tDCS advantage was found immediately after treatment conclusion, but not at any additional time points. There are several reasons we may not have found a tDCS advantage for trained items. First, we did not assess our primary outcome measure immediately after treatment conclusion. The tDCS advantage in the work of Fenner et al. (2019) immediately posttreatment was not present 2 weeks posttreatment. Posttreatment changes were assessed at the 1-week posttreatment conclusion, so it is possible we may have seen a tDCS advantage if we had assessed participants immediately posttreatment. Also, Tsapkini et al. (2018) only found a tDCS advantage for trained items in participants who received tDCS in Phase 1. Thus, we may not have found a tDCS advantage for trained items because four of the six participants who completed both phases of the study received tDCS in Phase 2. Similarly, Cotelli et al.'s (2014) study was a between-subjects study, so participants completed only one phase. Therefore, their design did not have to account for poorer performance over time due to neurodegeneration. It appears that participants with PPA are more likely to benefit from tDCS if they receive treatment earlier in the course of PPA. Our analyses did control for phase, so it is also possible that our participants overall had more advanced PPA than prior studies. Another possibility is that, even with tDCS due to the complexity of verbs versus nouns, improving verb naming is more difficult than improving noun naming, which most prior studies in PPA have treated (Cotelli et al., 2014; Tsapkini et al., 2014, 2018). This is likely given that Fenner et al. (2019) reported an attenuated tDCS advantage for verb relative to noun naming, particularly for trained verbs.

Our results showing a tDCS advantage for generalization to untrained items align with several studies (Fenner et al., 2019; Tsapkini et al., 2014, 2018). One study (Cotelli et al., 2014) found significant improvement in untrained noun naming in both sham and tDCS groups, with no tDCS advantage. Similarly, Hung et al. (2017) found that untrained noun naming was maintained and did not decline at the posttreatment time point with a sham stimulation. However, our study (along with most other studies treating word finding with language therapy plus neuromodulation) suggests that tDCS significantly improves the likelihood of long-lasting generalization to untreated words and critically prevents the rapid decline of untreated words.

Most previous studies assessing tDCS treatment for written single-word naming in PPA (Fenner et al., 2019; Tsapkini et al., 2014, 2018) did not assess generalization to sentence- or discourse-level production or comprehension. However, Cotelli et al. (2014) used a caregiver speech questionnaire to assess changes to functional speech production and comprehension. They found a significant tDCS advantage for improving functional speech production at the conclusion of 2 weeks of treatment, but not at the 12-week follow-up time point. Similarly, we also found a tDCS advantage soon after treatment conclusion for sentence production. We also found an advantage for sentence comprehension at 8 weeks posttreatment conclusion, which may be because we directly assessed comprehension rather than using a caregiver questionnaire that estimated rather than directly measured sentence comprehension. This may also be due to our focus on sentence production where we required more sentences produced in the final VNeST step than the typical VNeST protocol. This may also help explain the improvement in producing a higher percentage of complete utterances at 1 week posttreatment for both sham and real tDCS conditions. We did not find a similar improvement in number of CUs produced (reflecting word-level lexical-semantic abilities in discourse). Thus, VNeST did promote generalization to discourse on one measure even without tDCS augmentation.

Our study has several important clinical implications for PPA treatment. First, we found that VNeST (with and without tDCS) led to significantly improved trained verb naming and discourse skills at 1 week posttreatment. Based on our findings, training verbs with VNeST will lead to immediate improvement on trained verbs, and baseline naming will be maintained over a 2-month time period for trained verbs. Discourse measures declined but were not significantly worse from baseline, indicating baseline maintenance of discourse skills after the 8-week washout period. Our results indicate that even without neuromodulation, VNeST is an effective treatment for improving and/or maintaining verb naming and discourse skills in PPA. It also suggests that regular ongoing therapy may be particularly beneficial without several weeks between sessions.

Second, the fact that untrained verb naming significantly declined when not paired with tDCS indicates that verbs that are not targeted in traditional VNeST therapy without tDCS will likely continuously decline over a relatively short time period even when individuals are in speech therapy. Note that some studies in noun naming have found that untrained items will improve (Cotelli et al., 2014) or be maintained (Hung et al., 2017) posttreatment with a sham stimulation; however, given the neurodegenerative nature of PPA, the evidence from the current study and previous studies (Fenner et al., 2019; Tsapkini et al., 2018) suggests that untrained items are particularly vulnerable to being lost from the lexicon. Thus, clinicians who are providing traditional tDCS without neuromodulation should be particularly mindful of working with clients to select verbs for treatment that will be the most functional to help them maintain their communication skills and quality of life for as long as possible. Many clinicians do not have access to tDCS in their clinical settings, but speech-language therapy augmented by tDCS is currently available at some large clinical centers. As tDCS becomes more widely available, clinicians working with PPA should consider augmenting behavioral therapy with tDCS to promote improvement to both trained and untrained words.

Finally, two participants exited the study after completing the first phase of the study due to a combination of behavioral difficulties, language decline, and medical issues unrelated to PPA. Both participants had svPPA. Behavioral difficulties are particularly prevalent in svPPA (Tippett & Hillis, 2020), and it has also been identified as the PPA variant with the most rapid decline when measured by advancement of atrophy (Rogalski et al., 2011). Evidence suggests that participants with svPPA will benefit more from speech therapy in early stages where there are fewer cognitive and behavioral difficulties that may impede engagement in treatment (Suárez-González et al., 2021). Even though they dropped out early, both participants successfully completed the first phase of the study with 15 sessions of VNeST, and reported they enjoyed engaging in therapy. Even though patients with svPPA may be more likely to experience behavioral difficulties and more rapid language decline, we would still encourage clinicians to offer treatment to any person with svPPA who is willing and able to engage in therapy.

Future work will investigate the functional outcomes to determine how tDCS versus sham generalizes to functional communication outside of a clinic/research setting. On average, participants gained approximately three functional trained verbs from the baseline at the 1-week time point and approximately one functional trained verb (statistically equivalent to maintaining baseline functioning) from the VNeST therapy (with either tDCS or sham). As measured at the 1-week posttreatment time point, VNeST + tDCS led to the addition of approximately seven functional untrained verbs, but VNeST + sham led to the loss of approximately seven functional verbs (a difference of approximately 14 verbs). Similarly, at the 8-week time point, VNeST + tDCS was associated with the gain of approximately seven functional verbs, while VNeST + sham was associated with the loss of approximately seven functional verbs. Given the importance of maintaining words in the lexicon to slow down the rate of language loss in PPA, it is likely that these therapy-induced changes would impact communication in patients' daily lives.

Our study had several limitations. First, people with PPA represent a heterogeneous population but we only had eight participants in our study. It is important to study the impact of VNeST and tDCS in a larger group of participants. We also did not establish stable naming baselines, which would offer more reliable comparisons of baseline versus posttreatment naming. Two protocol changes were made during the course of the study, including changing the protocol to focus on 15 rather than 20 treated verbs for each participant and changing tDCS dosage from 1 mA to 2 mA. Additionally, we did not acquire neuroimaging; thus, we could not evaluate the effects of VNeST or VNeST + tDCS on neurological measures. Finally, two of the eight participants only completed one phase of the study. They both had semantic PPA, and our analyses accounted for missing data; however, our findings may have been impacted by the lack of comparison data for Phase 2. Enrolling a larger group of participants and incorporating neuroimaging in future studies will help enhance our understanding of which specific participants benefit the most from the VNeST and tDCS augmentation and the mechanisms that underlie treatment effects. This will allow for the development of personalized neuromodulatatory treatment plans that will offer patients with PPA the best possible chance of maintaining and temporarily improving language skills that are essential to quality of life.

Conclusions

In this crossover sham-controlled study evaluating the augmentation of VNeST with tDCS stimulation of the left IFG in eight participants with PPA, VNeST improved the naming of trained verbs and the production of complete utterances in discourse at 1 week posttreatment with both sham and real tDCS. However, a significant tDCS advantage was found that drove generalization of treatment benefits in untrained verbs (at 1 week and 8 weeks posttreatment), sentence production (at 1 week posttreatment), and sentence comprehension (at 8 weeks posttreatment). Naming of untreated verbs declined when VNeST was not augmented with tDCS. Results are promising as they suggest that augmenting VNeST with tDCS induces greater generalization, prevents decline, and, in some cases, induces longer lasting improvements to language abilities at the word and sentence levels. Results should be interpreted with caution as our sample size was small. Further studies with larger sample sizes are required to fully interpret the benefits of VNeST and VNeST + tDCS to treat language abilities in PPA.

Data Availability Statement

Deidentified participant data will be made available upon request to the corresponding author upon publication, subject to review by the Johns Hopkins University School of Medicine Institutional Review Board, resulting in a formal data sharing agreement.

Funding Statement

Authors E.B.G., E.V., K.R., E.M., and A.E.H. received salary support from R01 DC05375, and R.S. received salary support from R00 DC015554 during the course of this study.