All participants will be provided with a home exercise program using the GRASP [26]. The GRASP protocol consists of a booklet with detailed descriptions of multiple exercises for strength, range of motion, fine motor, and goal-directed activities targeting the upper limb. The provided kit includes all required equipment to perform the exercise program. The GRASP protocol is a self-administered program with program levels (grades 1-3) individualized by a qualified occupational therapist, based on the level of impairment of the upper limb. The upper limb exercises will be performed for 1 hour per day over a 2-week period (14 sessions).

Participants will also be provided with a home tDCS kit, which will be preprogrammed for 14 sessions of active or sham stimulation using a study code to facilitate research staff blinding. Participants will be unable to modify, or observe, any settings of the home tDCS devices. To initiate stimulation, participants will simply position electrodes on the head and press the start button. Those randomized to the “active” arm of the study will receive tDCS while simultaneously performing the GRASP exercises. TDCS will be delivered for 20 minutes and occur at the same time that a participant is undertaking the 1-hour GRASP program. Therefore, the initial 20 minutes of the intervention will involve both tDCS and the GRASP, with the remaining 40 minutes involving only GRASP exercises. The electrodes will be positioned with the anode over the ipsilesional M1 and cathode over the contralateral supraorbital region. TDCS will be applied at an intensity of 1 mA for 20 minutes daily for 2 weeks (total of 14 sessions) at home. Stimulation will be ramped up from 0 mA to 1 mA over the first 30 seconds and down from 1 mA to 0 mA over the final 30 seconds.

Participants randomized to the “sham” arm of the study will receive sham tDCS while undertaking the 1-hour individualized GRASP exercises. Electrodes will be positioned in the same location as the active tDCS group. Sham tDCS mimics the sensation of stimulation without changing cortical excitability. The protocol for sham tDCS will ramp the current up from 0 mA to 1 mA over the first 30 seconds before ceasing for the following 19 minutes and 30 seconds. This approach has been shown to provide an effective sham stimulation [27].

Several strategies will be used to monitor protocol compliance and ensure the correct use of home tDCS (for a summary see Textbox 3). The number of completed tDCS sessions and daily duration of GRASP will be recorded in a treatment diary.

At the completion of the 2-week tDCS intervention, participants will be asked to complete a questionnaire to identify any adverse events and establish whether participant blinding was successful. In accordance with current recommendations [28], we will ask participants to rate on a scale of 1-4 (1, absent; 2, mild; 3, moderate; 4, severe) the presence of the following symptoms: headache, neck pain, scalp pain, tingling, itching, burning sensations, skin redness, sleepiness, trouble concentrating, acute mood change, and other symptoms. In addition, we will ask participants to what extent they believe the reported symptoms were related to using tDCS (1, none; 2, remote; 3, possible; 4, probable; 5, definite). Furthermore, we will ask participants to indicate whether they believe they received active stimulation (yes or no) to determine the effectiveness of blinding.

The primary outcome measure for this study is a change in upper limb impairment as measured with the Fugl-Meyer upper extremity (FM-UE) assessment. The FM-UE is a commonly used, validated, and reliable measure of sensorimotor impairment [30]. The FM-UE is considered as one of the most comprehensive quantitative measures of motor impairment following stroke.

In addition, upper limb function will be assessed with the action research arm test (ARAT) and grip strength. The ARAT is a valid and reliable measure of hemiplegic upper limb function [31]; it provides a quantitative measure of upper limb function for domains of grip, grasp, pinch, and gross arm movement. Grip strength is associated with motor cortical output and motor recovery [32]. We will measure grip strength using a hand dynamometer (SH5001 Saehan Hydraulic Hand Dynamometer, Saehan Co, Masan, Korea). The best (maximal) grip of 3 attempts will be recorded.

Outcome assessors will be blind to group allocation and have completed training for FM-UE and ARAT assessments through the University of California Irvine. Training outcome assessors with this approach has been shown to improve accuracy and reduce variance of the FM-UE and ARAT [33,34].

Functional connectivity between brain regions will be assessed with high-density EEG. Three minutes of EEG will be acquired using an ASA-lab EEG system (ANT Neuro, Enschede, the Netherlands). Participants will be fitted with an ANT Waveguard cap with 64 sintered Ag-AgCl monopolar electrodes in standard 10-10 positions. Signals will be sampled at 2048 Hz, amplified 20X, filtered (high pass, DC; low pass, 553 Hz) and online referenced to CPz. During data recording, participants will be seated in a comfortable chair in a quiet room. Standardized instructions will be delivered to each participant asking them to relax during the 3 minutes of data recording, keep their eyes open, refrain from speaking or moving, maintain their gaze toward a fixation point straight ahead at eye level and not actively engage in any cognitive or mental tasks. Impedance will be kept <5 kΩ while recording.

Artifact rejection will be performed prior to analysis using independent component analysis. Nonphysiological artifacts will be identified using an automated and objective method to remove assessor bias [35]. Preprocessed data of participants with a right hemisphere lesion will then be flipped about the midline so that all lesions appear in the left hemisphere. Functional connectivity between electrodes will be determined using the debiased weighted phase lag index, which is a conservative estimate of connectivity based on phase consistency and biasing against zero phase lag relationships, limiting the detection of spurious measures of connectivity [36]. Regions of interest will include a seed approximating the ipsilesional M1 (C3) and clusters of electrodes approximating the ipsilesional premotor (F5, F3, FC5, and FC3), ipsilesional parietal (CP3, CP5, P3, and P5), and contralesional M1 (C4). Frequency bands of interest are the alpha band (8-15 Hz) and beta band (16-31 Hz) as they are associated with sensorimotor function [20,37,38]. For a given frequency, a debiased weighted phase lag index value of 1 indicates maximal phase coupling, whereas a value of 0 indicates no phase coupling. Connectivity analyses will be performed in MATLAB 9.2.0 (MathWorks, Inc) using both EEGLAB [39] and FieldTrip toolboxes [40].

Single-pulse transcranial magnetic stimulation will be used to quantify corticomotor excitability of the ipsilesional M1. Monophasic (posterior to anterior current flow) transcranial magnetic stimulation pulses will be delivered with a Magstim 200 stimulator (Magstim, Whitland, United Kingdom) via a figure-of-eight coil (70-mm wing diameter). The coil will be placed tangentially over the scalp with the handle pointing 45° posterolateral. Surface electromyography will be used to record motor-evoked potentials (MEPs) from the first dorsal interosseous muscle of the paretic hand with electrodes positioned in a belly-tendon montage. Suprathreshold stimuli will be delivered over the ipsilesional hemisphere to identify the optimal position for evoking an MEP from the first dorsal interosseous muscle of the paretic hand. For participants where MEPs cannot be evoked even at maximal stimulator output as a result of the stroke, we will document that a measure of corticospinal excitability was not obtainable at that experimental session. For participants where MEPs can be evoked, the optimal site will be marked on the scalp using a felt-tip marker to ensure consistent coil placement during subsequent data collection. Resting motor threshold will then be determined and is defined as the minimum stimulus intensity required to evoke an MEP of at least 50 μV in at least 5 of 10 consecutive trials. Thirty stimuli, not contaminated by pre-stimulus electromyography, will then be obtained at 120% resting motor threshold (interstimulus interval 4.5-5.5 sec) with the average, peak-to-peak amplitude, determined as a reliable measure of corticospinal excitability [41].

MRI will be performed at the Clinical Research and Imaging Centre located at the South Australian Health and Medical Research Institute with a Siemens 3T MAGNETOM Skyra scanner (Siemens, Erlangen, Germany). Standard MRI safety screening will be performed to ensure included participants are safe for MRI. At the preintervention MRI session, the imaging protocol will have a duration of 45 minutes and include T1 MPRAGE and T2-weighted fluid-attenuated inversion recovery images, diffusion tensor imaging, resting state, and task functional MRI (fMRI). At the postintervention MRI session, the imaging protocol will have a duration of 30 minutes and include T1-weighted images, resting state fMRI, and task fMRI.

The imaging protocols are as follows: T1-weighted images (MPRAGE, voxel 1 mm × 1 mm × 1 mm, repetition time (TR)=2300 ms, echo time (TE)=2.98 ms, flip angle=9°), T2-weighted fluid-attenuated inversion recovery images (voxel 1 mm × 1 mm × 1 mm, TR=5000 ms, TE=393 ms), diffusion MRI (voxel 2 mm × 2 mm × 2 mm, TR=4200 ms, b-value=0 and 2000 s mm), resting state fMRI (voxel 2.4 mm × 2.4 mm × 2.4 mm, TR=735 ms, TE=36 ms, 2 repeats of 6-min duration, 490 measurements for each), and task fMRI (voxel 2 mm × 2 mm × 2.5 mm, TR=3000 ms, TE=30 ms, 4.44-min duration). During the task fMRI, participants will be presented with a visual cue to squeeze a stress ball in their paretic or nonparetic hand, with blocks alternating every 30 seconds and repeated 4 times per hand.

Preprocessing and statistical analyses of MRI data will be carried out using tools from the FMRIB Software Library [42]. For all voxel-wise analyses, images from participants with lesions of the right hemisphere will be flipped about the midline after registration to standard space so that all lesions appear in the left hemisphere. fMRI data will be preprocessed and analyzed using the FMRIB Expert Analysis Tool [43]. Preprocessing steps will include high-pass temporal filtering at 0.01 Hz, spatial smoothing, motion correction, and removal of nonbrain tissue. Task fMRI data will be analyzed with a boxcar regressor, which will model task and rest blocks for first-level statistical maps for each participant. Higher level mixed-effects analysis will then be run using FMRIB Local Analysis of Mixed Effects [43] to test correlations with improvement in clinical scores and compare activation maps across groups.

For resting state fMRI, nuisance regressors of no-interest (cerebrospinal fluid, white matter, head motion, and physiological noise) will be modeled and removed. We will then calculate the mean time course of the blood-oxygenation level dependent signal in all voxels of the ipsilesional M1 region of interest. This time series will then be entered separately as an explanatory variable in the general linear model to determine for each participant the voxels where blood-oxygenation level dependent signal is temporally correlated with the ipsilesional M1. Connectivity between the ipsilesional M1 and other regions of interest (contralesional M1, ipsilesional dorsal premotor cortex, ipsilesional ventral premotor cortex, ipsilesional supplementary motor area, and ipsilesional posterior parietal cortex) will be determined with a Pearson’s correlation coefficient. Connectivity of the ipsilesional M1 will be compared before and after tDCS with a general linear model.

Structural connectivity will be analyzed with the FMRIB Diffusion Toolbox [44]. For each participant, the mean fractional anisotropy (FA) of the posterior limb of the internal capsule will be determined. An asymmetry index will be calculated as follows: FA = (FA–FA) / (FA+FA).

Lesion volume will be defined by manually tracing the lesion in FLS view. Lesions will be traced from each participant’s T1-weighted image using the coregistered T2-weighted image as a reference for lesion extension.