2.1. Participants
Twenty patients with PPA were enrolled in this study. All patients met clinical and imaging-supported diagnostic criteria for PPA (1). Inclusion criteria were: i) diagnosis of PPA confirmed with fluorodeoxyglucose positron emission tomography (FDG-PET) imaging; ii) Clinical Dementia Rating scale score of 0-1 (global score); iii) being a native Spanish speaker. The exclusion criteria were as follows: i) any contraindication for TMS or magnetic resonance imaging (MRI); ii) history of epilepsy; iii) other language disorders prior to the diagnosis of PPA; iv) neuroimaging findings not suggestive of PPA. The main clinical and demographic characteristics are summarized in Table 1 and Supplementary Material Table S1. The study was conducted at the Department of Neurology of Hospital Clínico San Carlos, Madrid between February 2018 and February 2020.
2.2. Study design
This is a double-blind (participant, outcomes assessors), single-center pilot study of patients with PPA who were treated with rTMS. Simple randomization was used to separate patients into 2 groups with a 3:2 ratio: i) active-site rTMS therapy only (n = 12); ii) a cross-over group (n = 8). In the latter group, subjects were randomly allocated 1:1 to receive therapeutic rTMS followed by control-site rTMS, or vice versa. Figure 1 summarizes the study design. One patient with semantic variant PPA in the cross-over group dropped out of the study due to a family issue after the active phase was completed; therefore, the final sample consisted of 13 patients in active therapy only, and 7 in the cross-over group. Patients were classified into nonfluent variant PPA (n = 14, 6 of them in the cross-over group) and semantic variant PPA (n = 6, one in the cross-over group) (Figure 2). At the time of study inclusion, patients were evaluated with the Addenbrooke’s Cognitive Examination III (22), Progressive Aphasia Severity Scale (23), Functional Activities Questionnaire, and Frontotemporal Lobar Degeneration-modified Clinical Dementia Rating scale (24). All participants were also assessed using a comprehensive battery of speech and language tests, including phonological tasks (initial phoneme deletion, word spelling, non-word repetition), semantic tasks (semantic-association task, word-picture matching, verb-action matching, synonyms judgment), picture object naming, action naming, semantic fluency, letter fluency, action fluency, verbal repetition, and reading of different types of words. Apraxia of speech was evaluated using sequential and alternating dysdiadochokinesia, repetition of multisyllabic words, reading, and spontaneous speech. Other cognitive domains were evaluated using the following tests: Corsi-block tapping, digit span, Trail Making Test, Symbol Digit Modalities Test, Rey-Osterrieth Complex Figure (copy and memory at 3 and 30 minutes), Stroop Color-Word Interference test, Tower of London Drexel version, Visual Object and Space Perception Battery (subtests object decision, progressive silhouettes, position discrimination, and number location), and Judgment Line Orientation. This protocol is described in detail elsewhere (25).
MRI (1.5 T, SIGNA HDxt, GE Healthcare) was performed less than 3 months before the start of the first rTMS session and included the following sequences: T1-weighted 3D fast spoiled gradient-echo, 3D Cube T2-weighted fluid-attenuated inversion recovery, and diffusion-tensor imaging. FDG-PET images (Siemens Biograph TruePoint PET/CT scanner) were acquired at rest, according to the European guidelines for brain PET imaging (26). PET images were acquired before the onset and within the first week after the end of each intervention (active-site rTMS and control site rTMS). Language training and medications were not modified during the study. Eight patients (40%) were receiving active language therapy during the study, with a mean duration of 1.68 ± 0.59 hours per week in this group.
2.3. Target personalization.
The target for active-site rTMS was selected after a first phase of the study, lasting approximately 10 weeks, in which patients received a single session of rTMS per week using different targets and protocols. Patients received excitatory and/or inhibitory protocols in 6 to 10 different brain targets according to clinical variant and neuroimaging findings. The excitatory protocol consisted of a session of 1500 pulses with 20-Hz rTMS trains with an interval of 20 seconds at 100% of the resting motor threshold. The inhibitory protocol consisted of a session of 600 pulses with 1-Hz rTMS trains with an interval of 1 second at 90% of the resting motor threshold. The main targets included the left inferior frontal gyrus, left superior frontal gyrus, right inferior frontal gyrus, left dorsolateral prefrontal cortex, left and right anterior temporal lobe, and vertex. All patients with the non-fluent variant were tested in the left inferior frontal gyrus, left superior frontal gyrus, right inferior frontal gyrus, left dorsolateral prefrontal cortex, and vertex. All patients with the semantic variant were tested in the left anterior temporal lobe, right anterior temporal lobe, left dorsolateral prefrontal cortex, left inferior frontal gyrus, and vertex. Additional targets (i.e. supplementary motor area, anterior cingulate) were tested only in some patients. These regions were selected according to the previous knowledge about the most relevant regions in language and PPA (i.e. inferior frontal gyrus, anterior temporal lobe), previous targets in other brain stimulation studies in PPA (i.e. left dorsolateral prefrontal cortex), and regions with greater hypometabolism and atrophy in some patients (i.e. left superior frontal gyrus). Patients were evaluated immediately (<15 minutes) after the rTMS session using the following language tasks: spontaneous speech, reading of 2 texts, object naming, and repetition (words and sentences). Patients with semantic variant PPA were also examined with a picture semantic association task. These tasks were developed by our group and include distinct but equivalent items according to several factors such as word frequency, length, degree of difficulty, etc. This procedure was used to avoid retest effects, and tests were different from those used as outcomes. Language testing was audio recorded and was evaluated by a blinded rater. Results were compared with baseline and between sessions to identify the most meaningful target for each patient. This target was decided by consensus between two of the investigators after evaluating the language changes scored by the blind rater. According to this strategy, patients received rTMS applied to the following areas: left inferior frontal gyrus (9 patients with nonfluent variant PPA), left superior frontal gyrus (3 patients with nonfluent variant PPA), left dorsolateral prefrontal cortex (1 patient with nonfluent variant PPA and 5 with semantic variant PPA), right superior frontal gyrus (1 patient with nonfluent variant PPA), and left anterior temporal lobe (1 patient with semantic variant PPA).
2.4. rTMS protocol: target selection and control-site stimulation
Brain stimulation was applied under neuronavigation, using a Magstim Rapid2 stimulator (Magstim, Whitland, UK) with a figure-eight coil. Patients received 15 sessions of 20-Hz rTMS trains with an interval of 20 seconds at 100% of the resting motor threshold. A total of 1500 pulses were applied in each session. Control-site stimulation was applied over the vertex (27), with the following protocol: 600 pulses at 1 Hz, with an interval of 10 seconds, at 25% of the resting motor threshold. Patients received 15 sessions on consecutive working days. All sessions were performed with online neuronavigation. In patients receiving both active- and control-site TMS, in the cross-over group, a washout period of at least 12 weeks was observed. These patients were informed that they would receive therapy using 2 different protocols.
2.5. Outcomes and safety evaluation
The primary outcome measure was the change in spontaneous speech (word count for 3 minutes using a story description task). Wordless children’s books Frog Stories by Mercer Mayer were used (e.g. Frog goes to dinner) to elicit spontaneous speech. Participants were randomly assigned to tell one of the five books of the Frog Stories collection at baseline and after treatment. These stories are considered equivalents (28). The instruction given was: “This is a wordless book. Please tell us the story describing it in detail as if we could not see the book.”. The test began when the examiner completed the instruction and finished at 3 minutes. No questions or interruptions by the examiner were allowed. The test was audiotaped and later transcribed for analysis.
Secondary outcomes were categorized as:
1) Language secondary outcomes:
i) Object naming test (96 items); the naming task was subdivided in 8 blocks. Each block comprised 12 pictures ranked by frequency of use (from most frequent to less). Items were line-drawn and included both animate and inanimate objects. Word frequency according to Zipf scale was 3.94±0.70 (29). Mean age of acquisition was 4.27±1.32 (range 2.22-7.44), using an 11-point linear scale (value of 1 indicates an age less than 2 years-old, from 2 to 10 indicate learning ages of 2 to 10 years, and a value of 11 indicates that the word was learnt at 11 years or older (30). Imageability, familiarity, and concreteness were measured on a scale of 1 to 7, with 7 being the most imageable, familiar, and concrete (31). Accordingly, mean imageability was 6.78±0.47, familiarity was 5.99±0.69, and concreteness was 6.03±0.47. Mean number of phonemes was 6.04±1.71, and mean number of syllables was 2.65±0.82. Distributions of these parameters are shown in Supplementary Material Figure 1. Drawings of the nouns were presented using a computer, with the following instruction: “I will show several pictures. When you hear a sound, a picture will appear on the computer screen. Please name the picture using only one word”. A maximum of 30 seconds was allowed for the subject’s response and the test was audiotaped. Each item had only a correct response. If the patient gave an incorrect response, the examiner said, “No, tell me another name for that” or “That has another name”. No phonological or semantic cues were given. The correct responses were not given in any case. A brief rest period was permitted between blocks, but not between items.
ii) Story reading test (a text of 100 words. The text was presented in letter Cambria 22 in a single sheet of paper. Words were morphologically simple, with a mean of 1.94 syllables per word, and the frequency of nouns and adjectives was 4.23±0.84 and frequency of verbs was 4.63±0.85 according to Zipf scale (31). Readability according to the Flesch-Szigriszt score was 73.47 (an index of how difficult a document is to read on a scale from 100, extremely easy, to 0, very complex) and the Fernández-Huerta score was 77.69 (a Spanish adaptation of the former index). Parameters of readability were calculated using the INFLESZ software (32). Thus, the text is considered as “quite easy” according to these indexes. We calculated the accuracy (defined as the number of words correctly read), and efficiency (defined as the accuracy divided by the time taken to read, expressed as a percentage).
iii) Repetition of syllables (8 syllables, e.g. “pa”, “ta”), pairs of syllables (8 pairs, e.g. “pa-ma”, “na-ba”), non-words (8 non-words, e.g. “amiteso”, “biboterana”), sentences (10 sentences of increasing length, from 3 words and 6 syllables to 10 words and 18 syllables) , and multisyllabic words (for the assessment of dysdiadochokinesia). Each type of word was scored individually. These tasks belong to the Aphasia module of the “Test Barcelona revised”, a standardized battery for neuropsychological and language assessment in Spanish (33).
2) Global cognition, evaluated with the Addenbrooke’s Cognitive Examination III (22).
3) Perception of change: Changes in the impression of clinical change reported by the patient and principal caregiver (rated from 0 = very much worse to 10 = very much improved; 5 = no change)
4) Behavioral changes, according to the Neuropsychiatric Inventory (NPI) (34).
5) Brain metabolism changes on FDG-PET images.
Outcomes were examined immediately (less than 2 weeks) before the start of TMS therapy and within the first 2 weeks after the completion of treatment. In the crossover group, outcomes were evaluated before and after the completion of each period (active-site or control-site stimulation).
To evaluate safety, patients and caregivers were asked about potential adverse effects, during each session of rTMS and at the end of the study. In addition, electroencephalography studies were performed prior to the onset of the rTMS sessions, after session 7, and at the end of the course of treatment for safety purposes. The electroencephalogram was assessed by an expert clinical neurophysiologist blind to the treatment.
2.6. Neuroimaging preprocessing and analysis
Statistical Parametric Mapping (SPM) software version 12 was used to preprocess and analyze the images. PET images were coregistered to T1-weighted MRI. Images were spatially normalized and subsequently smoothed at 8 mm full width at half maximum. Global intensity scaling was performed. A paired sample t-test was used to compare brain metabolism before and after active- and control-site rTMS. Age, sex, and site of stimulation were added as covariates. A threshold of family-wise error–corrected p-value < 0.05 was used to correct for multiple comparisons. If non-statistically significant results were obtained, we used an uncorrected p-value < 0.001 as an exploratory analysis. An extent threshold of k = 50 was used.
A two-sample t-test was used to compare each diagnostic group (nonfluent and semantic PPA) versus a group of 40 healthy controls, using age and sex as covariates. A family-wise error corrected p-value <0.05 and k=50 were used for multiple comparisons correction (Figure 3). This analysis was also conducted individually to represent the regions with brain hypometabolism for each patient and the targets used in each patient. In this case, an uncorrected p-value <0.001 and an extent threshold of k=50 voxels was used (Figure 4) (35).
2.7. Sample size
Given the lack of previous studies into the use of rTMS in patients with PPA, we estimated that a sample size of at least 15 participants was needed to obtain more than 80% power for detecting effect sizes larger than d = 0.80 in the primary outcome.
2.8. Standard protocol approvals, registrations, and patient consents
The study was conducted with the approval of our hospital’s Ethics Committee (study code: 17-247E) and in accordance with the Declaration of Helsinki and its subsequent amendments. The study was registered at clinicaltrials.gov (NCT03580954) (URL of the trial record registry: https://clinicaltrials.gov/ct2/show/NCT03580954). All patients (or their legally authorized representatives) gave written informed consent.
2.9. Statistical analysis
Statistical analysis was performed using version 20 of the IBM(R) SPSS Statistics software. Descriptive results are shown as mean ± standard deviation or frequency (percentage). Because the sample size was less than 30, we opted to use non-parametric tests. We used the Mann-Whitney U test for comparisons between 2 groups (active- and control-site rTMS), and the paired samples Wilcoxon test to evaluate changes within the same subjects before and after active-site TMS. Effect sizes were estimated with Hedges’ g and classified as small (g = 0.20), moderate (g = 0.50), or large (g = 0.80). A p-value < 0.05 (two-tailed) was considered statistically significant.
Furthermore, a mixed three-way ANOVA was performed to understand the effects of rTMS on the outcomes of the study. The type of stimulation (active-site and control-site), and the clinical variant (nonfluent and semantic) were included as inter-subject factors in the analysis. Sidak method was used to correct for multiple comparisons. Accordingly, adjusted p-values are reported. Both numerical methods (skewness and kurtosis values and the Shapiro-Wilk test), and graphical methods (Normal Q-Q Plots) were used to measure normality. The outliers were kept in the analysis because they did not materially affect the results as assessed by a comparison of the results with and without the outliers. There was homogeneity of variances, as assessed by Levene's test for equality of variances (p > 0.05).