We consecutively recruited stroke patients over a 13 month period from October 2016 to November 2017. For this preliminary study, our recruitment target was 16 patients [56]. This sample size would provide us with an indication of neurophysiological effects of BFR-E as well as provide information about the feasibility and tolerability of BFR-E. Fifteen participants were recruited and 14 participants completed training (see Table 1 for participant characteristics).

Participants were included if they (a) were 18 to 80 years old, (b) had been discharged from acute hospitalisation, (c) had a first time ischemic or haemorrhagic stroke of any chronicity, (d) could walk 10 m with or without a walking aid, (e) had manual muscle test scores of $\ge$3 in the tibialis anterior (f) were able to attend the exercise sessions, (g) were medically stable, (h) were able to communicate verbally and (i) could speak Danish or English. Patients were excluded if they had (a) resting systolic blood pressure $>$160 mmHg or resting diastolic blood pressure $>$100 mmHg, (b) evidence of lower limb peripheral oedema, (c) open wounds or fragile skin, (d) cognitive deficits preventing the participants from undergoing the assessments, exercise program or informed consent process, (e) major neurological/musculoskeletal deficits that could affect training, that were unrelated to the stroke, (f) administration of botulinum toxin in the lower limb at least 6 months prior to training, (g) history of epilepsy, (h) cochlear implants, (i) a pacemaker, (j) any type of deep brain stimulator and (k) metal implants in the head or neck. Participants were not excluded from the study if they did not have motor evoked potentials using TMS as they could still provide information to answer our other aims (muscle activity during training, M-Max amplitude and NRS scores).

Surface electromyography activity (sEMG) were recorded using (Ambu Neuroline) surface electrodes. Electrodes were placed on the muscle belly of the tibialis anterior in accordance with previous recommendations [57]. sEMG data were amplified using custom built amplifiers and band pass filtered from 10 Hz to 500 Hz recorded at a sampling rate of 4 kHz. TMS was delivered using a magnetic stimulator (Magstim 200, Magstim Company Ltd, United Kingdom) using a 110 mm double cone coil. Data were collected and stimuli were controlled using custom-made software (Mr Kick II) as has been used previously [58]. Brachial blood pressure was measured using a sphygmomanometer (Riester® 55 cm $\times$ 14.5 cm) and blood flow restriction was performed using one of two blood pressure cuffs, depending on the size of the thigh (Reister® 70 $\times$ 22 cm or Reister® 100 $\times$ 26 cm).

TMS and peripheral electrical stimulation were collected prior to exercise (T0), immediately after exercise (T1), 10 min after exercise (T2) and 20 minutes after exercise (T3). During testing at each time, peripheral electrical stimulation measurements always preceded TMS measurements.

Patients provided a numerical rating for their perceptions of pain, discomfort, difficulty and fatigue during exercise with 0 indicating no pain, discomfort or fatigue and 10 indicating the highest imaginable pain, discomfort or fatigue. Patients were also asked about how safe and focused they felt during exercise with 0 indicating feelings of being unsafe and unfocussed and 10 indicating feelings of being safe/focused.

Fig. 1 provides an overview of the experimental procedures. Prior to day 1, an investigator explained the experimental procedures to participants via phone or in person and participants were sent/provided a copy of the participation information sheet and consent form. On day 1, an investigator discussed the project with the participant and acquired written informed consent. In the initial consultation and prior to testing, participants were screened against the inclusion and exclusion criteria. Demographic and clinical data were collected and included: age, sex, time since stroke, stroke location, dominant leg, height, mass, 10-meter walk test time, modified Ashworth score of the plantarflexors and dorsiflexors (in lying with the knee straight), manual muscle testing of the plantarflexors and dorsiflexors (in lying with the knee straight) and brachial blood pressure (measured on both days in lying on the less affected arm). If participants were not suitable for testing based on these measures, no further questioning/testing was performed. Following testing and after eligibility was determined, the session performed first (BFR-E or E-only) was drawn randomly from an envelope. Depending on randomization, participants performed E-only or BFR-E in the first session, with the other condition performed in the second session. Sessions were spaced 7 $\pm$ 1 day(s) apart. Seven days between sessions was chosen as the patients would unlikely change functional status in this time and unlikely have any carryover effects from the first session. As such, any post-exercise fatigue as a result of the first training session would have completely subsided by the second session.

After thresholds had been established, T0 measurements for TMS and peripheral electrical stimulation were collected. For all participants, sEMG electrodes were affixed and sEMG data were collected during training.

During training, patients were seated comfortably. The foot was relaxed and resting on a board at an average ankle angle of 118 $\pm$ 12${}^{\circ }$ (mean $\pm$ SD) and an average knee angle of 109 $\pm$ 10${}^{\circ }$. From this position, the maximum comfortable dorsiflexion range for each subject was determined. A target was placed at this location during testing. Following this, subjects placed their foot under a TheraBand affixed to the board (for stronger patients) or no TheraBand (for weaker patients) and asked to lift the foot to the target. The target was placed within the range of motion of the ankle joint. Patients practiced 4–5 times while timing the contraction to a metronome and adjustments to the set up were made, as required. Once subjects were comfortable, the knee and ankle angle were measured to facilitate testing in the same position between days.

During training, participants performed one set of 30 repetitions and one set of 15 repetitions of dorsiflexion, with 30 seconds rest between sets. This is a reduced exercise paradigm from blood flow restriction paradigms that have been used previously [39, 40, 59, 60]. Participants concentrically contracted the TA for $\approx$1 s, held the foot at the stop for $\approx$2 s and lowered the foot over $\approx$0.5 s. Some participants fatigued and were unable to reach the target at the end of training. If this occurred, participants were asked to dorsiflex the foot as much as able in time with a metronome.

For BFR-E, the blood pressure cuff was placed around the thigh and inflated over 1 min to 0.8 $\times$ systolic blood pressure, as performed previously. Exercise commenced when the testing pressure had stabilised. Pressure was monitored during testing and adjusted if needed.

Following training, participants that were eligible for TMS and M-wave measurements received these at T1, T2 and T3. Participants that did not received these, waited for 20 min before answering the numerical rating scales of the exercise session. For these questions, patients were asked only to comment on the exercise itself and not the neurophysiological testing, if performed.

For sEMG during training, the amplitude of the sEMG measurements during exercise for each contraction were measured. Post data collection, data were smoothed using a 1st order, 1 Hz low pass Butterworth filter. Contraction onset threshold was 25 $\mu$V. A contraction was determined when the sEMG exceeded the threshold for at least 20 ms and the contraction lasted for at least 1.5 seconds. sEMG amplitude was the root mean square (RMS) measured in the 1.5 seconds following the contraction onset. The RMS amplitude was averaged in blocks of 5 contractions, for a total of nine contraction blocks (45 total contractions).

The amplitude of M-wave and MEPs was taken as the peak-to-peak amplitude ($\mu$V) of the raw EMG traces from 3 ms to 33 ms after the stimulus for the M-wave and 25 to 70 ms after the test stimulus for MEPs. These time windows encompassed the peak-to-peak responses for all subjects.

Fig. 2 shows example rectified and smoothed data for one participant during exercise in BFR-E and E-only conditions.

During exercise there was a significant main effect of contraction-block (p 0.05) but no condition $\times$ contraction block interaction (p = 0.149) and no main effect of condition (p = 0.364) or time since stroke (p = 0.402). Table 2 shows the modelled estimates of fixed effects of condition and contraction block. Pairwise comparisons for each contraction block minus contraction block 1 and condition are shown in Table 3.

Fig. 3 shows example raw data for one participant for M-max, single pulse TMS, SICI and ICF stimulation paradigms before and after BFR-E and E-only conditions.

There was no significant condition $\times$ time interaction for M-max amplitude (p = 0.463), single pulse TMS amplitude (p = 0.343), SICI amplitude (p = 0.973) or ICF amplitude (p = 0.850). Table 4 shows the modelled estimates of fixed effects of condition and time. There was no main effect of time for M-max amplitude (p = 0.136), single pulse TMS amplitude (p = 0.157), SICI amplitude (p = 0.797) or ICF amplitude (p = 0.354). There was significant main effect of condition for ICF amplitude (p = 0.009) but not M-max amplitude (p = 0.329), single pulse TMS amplitude (p = 0.521) or SICI amplitude (p = 0.698). There was no significant main effect of time since lesion for M-max amplitude (p = 0.671), single pulse TMS amplitude (p = 0.996), SICI amplitude (p = 0.822) or ICF amplitude (p = 0.783). Table 5 shows the modelled mean differences and 95% CIs of the pairwise comparisons for the main effects of time and condition.

Table 6 summarises the numerical rating scores for BFR-E and E-only. There were no differences between conditions for patient perceived pain (p = 0.10), discomfort (p = 0.17), fatigue (p = 0.47), safety (p = 0.50), focus (p = 0.24) or difficulty (p = 0.16).

There were no interaction effects for single pulse stimulation, SICI and ICF. Although a main effect of condition was reported for ICF, this is a reflection of the higher ICF at baseline. Unfortunately, we could only find rMT in eight participants. For five participants the rMT could not be found on both days. Following stroke, it is common for patients to have reduced cortical excitability in the affected hemisphere compared to the unaffected hemisphere and compared to healthy controls [49]. A systematic review demonstrated large standardised main differences (SMD) and relatively narrow 95% confidence intervals (95% CI) when comparing the affected hemisphere with the unaffected hemisphere for rMT (SMD: 1.03, 95% CI: 0.84 to 1.23) and MEP amplitude (SMD: –0.96, 95% CI: –1.17 to –0.74) and the affected hemisphere with healthy controls for rMT (SMD: 1.24, 95% CI: 0.81 to 1.67) and MEP amplitude (SMD: –0.64, 95% CI: –0.93 to 0.34) [49]. These differences were apparent in both early and chronic phases of stroke [49]. Given this, the inability to determine a rMT in five participants, may not be unexpected.

When determining our target sample size of 16, we expected that although rMT may be higher and MEP amplitude maybe reduced in our cohort, we would still be able to attain a rMT. It is difficult to ascertain why our cohort had so many non-responders when patients were relatively well functioning and had a dorsiflexor MMT strength of $>$3. Non-responders to TMS have been reported by other investigators but not all. Huynh et al. [50] performed TMS to the ipsilesional cortex to the contralateral Abductor Pollicis Brevis and found that 6/31 stroke participants had no MEPs, 6 days after the lesion onset. Further, Stinear et al. [61] used TMS to assess the ipsilesional cortex to the ipsilateral Extensor Carpi Radialis and found that 9/40 patients had small or absent MEPs, 2 weeks after the lesion onset. Although reported in the above-mentioned studies, other studies adjust estimates for non-responders [62] or may not report on non-responders at all [63]. If this is the case, it is possible that the number of patients that were unresponsive to TMS in our study is actually representative of people that have a minimally responsive/unresponsive ipsilesional cortex when attempting to stimulate the contralateral tibialis anterior following stroke. Dharmadasa et al. [64] postulated that inexcitability may determine a poorer clinical profile in patients with amyotrophic lateral sclerosis as postulated in upper extremity rehabilitation following stroke [61]. Given this, the inexcitability observed in the tibialis anterior may be related to a clinical profile, associated with recovery. Although possible, our study does not have sufficient participant number or sufficient follow up time points to further explore this hypothesis.

A further consideration is that the responses from patients in our study may have been too variable to observe a consistent effect. The patients differed in the time since stroke, location of stroke and type of stroke, which can increase inter-subject variability of neurophysiological responses [46]. In our analysis, we adjusted for time since lesion (as a continuous variable) and found no effect for any neurophysiological measure. We did not add location of stroke or stroke severity as a factor as we did not have sufficient patient numbers with each to do this however a cross over design was chosen to reduce some of this variance.

Taken together, although we almost attained our target of 16 participants, due to the number of non-responders and response variability, our study was likely underpowered to determine changes/differences in TMS measurements if a difference exists. While we acknowledge that this is a significant limitation, non-publication of non-significant results, and ‘file-drawer’ publications, has been highlighted as a significant problem in TMS research [65]. As such, we decided to disseminate our results rather than not disseminate the results at all.

Although potentially underpowered, it is also possible that there is little difference in TMS measurements with exercise (without BFR). Although E-only paradigms in healthy subjects have shown SICI disinhibition [66, 67, 68], some have not [69]. Further, in people with chronic stroke SICI in the abductor pollicis brevis was not different for all time points after training following a single 15 min bout of repetitive thumb abduction training [55]. Our exercise protocol was shorter compared to these paradigms and perhaps too short to induce changes in SICI. Further, our paradigm was in lower limb muscles which can alter projections to muscles [70]. Similarly to SICI, there was no significant interaction in ICF between time and BFR-E and E-only conditions. Although previous studies have shown changes in ICF following exercise in healthy people [39, 68, 69] and stroke patients [55], not all have in E-only [39] or BFR-E [39] conditions. The result of our study could mean that SICI and ICF in people with stroke are not different when comparing BFR-E and E-only. Alternatively, the lack of difference between the BFR-E and E-only could be that the exercise intensity was too low to result in sufficient fatigue in the BFR-E condition and/or occlusion pressures used were too low to influence SICI. Studies that have demonstrated altered SICI as a result of BFR-E, have used significantly higher pressures for longer periods of time ($>$200 mmHg) [71]. Such protocols however, would contribute to more discomfort and pain; potentially reducing the tolerability of BFR-E [58].

Safety and comfort are important aspects when assessing feasibility of new rehabilitation strategies before designing larger studies to test its effectiveness [72]. In our study, patients felt safe during both interventions with no difference between interventions. In addition, subjects reported minimal discomfort with no reported or observed adverse reactions. Although this is encouraging, based on our findings we do not believe that BFR-E should be used in all people with stroke, in all settings. Participants were selected based on a stringent inclusion and exclusion criteria and may not represent all people following stroke. In this study, exercise was overseen by a physiotherapist in an inpatient hospital with a neurologist on standby and a local medical emergency response team in case of emergency, which does not reflect all clinical settings. In our sample, participants perceived exercise as minimally fatiguing. However, in stroke populations with different inclusion/exclusion criteria, with more fatiguing exercises over more sessions, it is possible that adverse reactions are more likely to occur. Previous research has shown that the frequency of self-reported fatigue affects 69.5% of people 1-year following stroke [73] and is 68% more prominent in people with stroke than healthy people [74]. If similar studies are performed in the future, we recommend that this is considered when choosing exercise intensity and training setting.

Patients reported minimal pain and discomfort and no difference in pain and discomfort between BFR-E and E-only conditions. This is encouraging as people with stroke are more likely to comply with exercise that doesn’t cause pain or discomfort [75]. However, it also highlights that exercise was (perhaps) too easy or that cuff pressures were too low for most patients as some pain/discomfort is expected during BFR-E [58]. Future studies should consider increasing exercise intensity and cuff pressures during BFR-E to potentially increase the effects of BFR-E.

In this study, participants generally had few comorbidities and high mobility (i.e., could walk 10 meters with or without walking aids), which is not representative of many people following stroke who are weaker and/or lower functioning. Although the participants included in our study were relatively well-functioning, as we included a convenience-based sample, there was some heterogeneity in the included population. For example, some patients received thrombolysis on admission to acute care (subject 10 and 11 received thrombolysis, subject 8 was missing and the remaining did not receive thrombolysis) and some had depression (subject 3, 8 and 14). Unfortunately, we were not able to attain the medication status, did not measure/attain general post-stroke fatigue nor have reliable/valid data on thrombectomy. These are factors that can influence neural recovery and peripheral responses to TMS. Furthermore, although all patients included in the study had deficits as a result of the stroke (as this was one of our questions in the initial interview), we cannot definitively rule out that some of the participants may have had existing issues which further exacerbated the impairment caused by the stroke. It would have been ideal to have medication status, post stroke fatigue, thrombectomy data and pre-existing deficits. However, as we used a cross-over type design and accounted for this in our analysis, although a limitation, it is part of the difficulties when conducting research in an inpatient setting and using a convenience based sample. Exercise was performed in an inpatient hospital with significant safety measures in place, should a patient experience distress. Future research using similar paradigms must account for this to ensure safety of patients when performing BFR-E. Furthermore, our training protocol was short, over 1 session, and does not account for the amount of practice required for functional recovery after stroke nor the reality of most clinical settings.