Thirty-one adolescents aged 11-17 years were recruited via the Northern UK NF-clinical research network. Inclusion criteria included (i)Clinical diagnosis made using the National Institute of Health diagnostic criteria(41) and/or molecular diagnosis of NF1; (ii)No history of intracranial pathology other than asymptomatic optic pathway or other asymptomatic and untreated NF1-associated white matter lesion or glioma; (iii) No history of epilepsy or any major mental illness; (iv)No MRI contraindications. Participants on pre-existing medications such as stimulants, melatonin or selective serotonin re-uptake inhibitors were not excluded from participation. The study was conducted in accordance with local ethics committee approval (Ethics reference:18/NW/0762, ClinicalTrials.gov Identifier: NCT0499142. Registered 05/08/2021; retrospectively registered, https://clinicaltrials.gov/ct2/show/NCT04991428).
Experimental Procedure - cross over intervention design
The effect of atDCS on GABA and working memory was tested using a two parallel-arm, single(participant)-blinded, sham-controlled cross-over design. Each participant had two study visits at least one week apart- one with atDCS intervention and with sham as placebo control. The order of these sessions was randomized and counter-balanced. Baseline assessments (as described below) were conducted at the first visit. Subjects were positioned comfortably in the scanner and a high-resolution T1-weighted image was acquired (see figure 5). The T1-weighted image was used to place a voxel of interest (VOI) by hand - over the DLPFC and another VOI in the occipital cortex. MRS was acquired from both voxels and participants were asked to perform a working memory task for 24 minutes (4 blocks of 6-minutes each) during which fMRI data were acquired(the results of which are not reported here). AtDCS or sham stimulation was started after the first block of working memory task and continued for 15 minutes during which the participant engaged in 2 more blocks of working memory tasks. Between each working memory block, participants were asked if they were comfortable and instructions were repeated again. Following tDCS, participants performed the final block of the working memory task. Finally, MRS was acquired again from both voxels. T2-weighted images were acquired at the first visit (after the T1 image) and reviewed by a paediatric neuroradiologist (SS) to rule out NF1 associated tumours. The sample size of 30 participants in this study, powered on the expected change of 20% in GABA following tDCS based on our previous work(42).
AtDCS was delivered via a NeuroConn DC-STIMULATOR MR with the anode placed over F3 position in the international 10–20 system and the cathode over the Cz position. Scalp was cleaned with Nuprep gel and Ten20-paste was used as a conductive medium between the scalp and the electrodes. For anodal stimulation, the current was ramped up over 15 s, held at 1 mA for 15 min and then ramped down over 15 s. For sham stimulation, the current was ramped up over 15 s and then immediately turned off. The current parameters were chosen based on our previous experience from a pilot clinical trial of safety in this cohort (clinical trials identifier: NCT03310996). The atDCS induced electrical fields are simulated in figure 6. SimNIBS 3.2 (https://simnibs.github.io/simnibs/build/html/index.html) was used to estimate the electric field induced by tDCS (43, 44). The headreco pipeline (45) was used to segment the different tissue types and create a finite element mesh corresponding to an example T1 image from an open source dataset (46). The anode and cathode were placed at F3 and Cz respectively, and the standard SimNIBS conductivity values were used.
Structural and MRS data acquisition and analysis
Scanning was performed on a Philips Achieva 3T scanner (Best, NL) using a 32-channel head coil. 3D T1-weighed magnetic resonance images were acquired sagittally with a magnetization prepared rapid acquisition gradient-echo sequence (repetition time = 8.4 ms; echo time = 3.77 ms; flip angle = 8o, inversion time = 1150 ms, 0.94 mm in-plane resolution and 150 slices of 1mm). Single voxel 1H MRS data were acquired before and after stimulation from two volumes of interest (VOI) in each participant. One VOl (40 x 20 x 24 mm) was placed in the left DLPFC and a control VOI (20 x 50 x 20mm) was placed within the posterior occipital lobe, centred on the mid-sagittal plane to cover both hemispheres (figure 5). For detection of GABA, GABA-edited MEGA-PRESS spectra (47, 48) were acquired with a repetition time of 2000 ms, echo time of 68ms, 1024 sample points collected at a spectral width of 2 kHz, as previously described (49). The DLPFC MRS took approximately 7 min to acquire, with 96 averages and OCC voxel took 3 min to acquire with 32 averages. The number of averages were chosen to approximately match spectral quality between DLPFC and OCC.
Quantification was conducted using the Advanced Magnetic Resonance (AMARES)(50) routine in the Java-based magnetic resonance user’s interface (jMRUI5.1, EU project)(51). To improve the display of the spectra, line broadening of 6Hz was used. No time-domain filtering was performed on the data before analysis by AMARES. Metabolite resonances including GABA, glutamate + glutamine (Glx) and N-acetylaspartate (NAA) were calculated relative to the unsuppressed water signal from the same voxel. To examine partial volume effects on MRS voxels of interests, the T1-weighted anatomical images were segmented into gray matter (GM), white matter (WM) and cerebrospinal fluid (CSF) using SPM8 (http://www.fil.ion.ucl.ac.uk/spm/). Voxel registration was performed using custom-made scripts developed in MATLAB by Dr. Nia Goulden, which can be accessed at http://biu.bangor.ac.uk/projects.php.en. The scripts generated a mask for voxel location by combining location information from the Philips SPAR file with orientation and location information contained within the T1 image.
In the aTDCS group, 28/29 DLPFC pre-intervention spectra and 24/29 post-intervention spectra were included for analyses (1 pre-intervention and 3 post intervention spectra rejected due to movement artefacts and 2 rejected due to >3SD difference in pre-post intervention NAA line width) and 29/29 pre-intervention OCC spectra and 28/29 post-intervention OCC spectra were included (1 post-intervention spectra not acquired due to technical difficulties). In the sham group, 31/ 31 DLPFC pre- intervention spectra and 25/31 DLPFC post-intervention spectra were included for analyses (5 rejected due to movement artefacts and 1 rejected due to >3SD difference in pre-post intervention NAA line width) and 29/31 pre-intervention OCC spectra and 25/31 post intervention OCC spectra were included for analyses( 2 pre and 3 post-intervention spectra not acquired due to technical difficulties, 2 post-intervention spectra rejected due to movement artefacts and 1 due to >3SD change in NAA LW). The calculation of partial volume within the VOIs provided the percentage of each tissue type within the relevant voxels. The DLPFC VOI consisted of 39% (±17%) of GM, 24% (±23) of WM, and 37% (± 31%) of CSF and the OCC VOI, 50% (±10%) of GM, 37% (±15%) of WM, and 13%(±9%) of CSF. The tissue fraction was used as a covariate in the baseline analyses of the relationship between GABA and behavioural measures.
Detailed cognitive assessments were carried out to assess working memory at baseline, at the first visit of the participant. Both verbal and visuospatial working memory were assessed using the n-back task. The task was programmed in-house using E-Prime software. Each participant completed verbal and visuospatial tasks at four levels of complexity- 0-back, 1-back, 2-back and 3-back tasks. For the verbal task, random letters were presented one at a time and the participant was asked to respond with a key-press if the letter corresponded to the letter one (1-back), two (2-back) or 3 (3-back) letters before. For the 0-back verbal task, participants were asked to press the key to the occurrence of the letter ‘X’. For the visuospatial n-back task, blue squares were presented sequentially on a black 2 x 2 grid. Participants were instructed to respond with a key press if the position of the square matched the position one (1-back), two (2-back) or 3 (3-back) positions before. For the 0-back visuospatial task, participants were asked to respond with a key press to the occurrence of an orange square. Each participant was presented with three blocks of each n-back task (24 blocks in total). All stimuli were presented for 500 ms and the inter-stimulus interval was set to 1,500 ms. Accuracy was calculated as the proportion of correctly identified hits + correct omissions within each block (correct hits + correct omissions/ total responses) averaged across each n-back condition as presented in Table 1. Response times (RT) were calculated only for time to correct response to target stimuli, averaged across each n-back condition.
Parent-rated Vineland Adaptive Behaviour Scale - third edition(52) was administered to the parents to assess child adaptive behaviour with overall functioning computed as standardized age equivalent and expressed as an Adaptive Behaviour Composite (ABC). Conners 3 rating scale(53) was used as a standardized measure for parent reported ADHD symptoms. It consists of 27 items each rated on a 4-point Likert scale (0 = not true at all to 3 = very much true) in five subscales: attention, hyperactivity, learning problems, oppositionality and peer problems. The inattention and hyperactivity subscales are reported below.
Behavioural outcome measures
At the start and end of each scanning session, while outside the scanner, participants were asked to complete the computerised Corsi block task on the Psychology Experiment Building Language (PEBL)(54). In this task, 9 identical blue blocks are presented on the screen. These blocks light up on the screen in a sequence, which starts off as a simple sequence of two blocks and increases in complexity based on participant performance. The participant is asked to mimic the sequence observed on the screen. A measure of the memory span and mean RT is reported.
Within the scanner, participants performed 4 runs of working memory tasks- one run each before and after stimulation and two during the atDCS/sham stimulation. Each run consisted of 6 blocks each of 0-back and 2-back verbal working memory task as described above. Each block was 30 s long and consisted of 9 target stimuli. Accuracy was calculated separately for 0-back and 2-back tasks(correct hits + correct omissions/ total responses). RT were calculated only for time to correct response to target stimuli.
Statistical analyses were performed in SPSS version25 and R version 1.2. Partial Pearson’s correlations were used to investigate the relationship between GABA in DLPFC and OCC and the behavioural outcomes using tissue fraction as a covariate in the analyses. The Fishers Z transformation was used to compare the correlation coefficients. Comparison of correlation coefficients was undertaken using Fisher’s transformation. Group differences in metabolites post intervention were analysed using linear regression models adjusting for baseline values of the relevant outcome as a linear covariate. A p value<0.05 was considered significant.