EMG acquisition and analysis

Surface EMG was recorded with the BrainVision Recorder software (Brain Products, Munich, Germany) using a 16-channel MR-compatible bipolar amplifier (ExG, Brain Products, Munich, Germany). Given our interest in studying the neural substrates of LLR for both flexors and extensors, EMG was recorded from two wrist muscles: Flexor Carpi Radialis (FCR) and Extensor Carpi Ulnaris (ECU).

For each muscle, after having carefully cleaned the skin with a 70% Isopropyl Alcohol solution, we placed the bottom layer of electrodes on the belly of the muscle oriented along the muscle fiber, and filled the central hole of each electrode with an abrasive Electrolyte-Gel (Abralyt HiCl, Rouge Resolution, Cardiff, UK). Contact impedance for each electrode was measured using the BrainVision Recoder software and a cotton swab dipped in abrasive gel was swirled on the skin until the measured contact impedance was lower then 10kΩ, as described in the product technical specification. We then carefully co-located the reference electrodes on top of the measurement electrodes, using a layer of electric tape applied on top of the measurement electrodes to avoid electrical contact. Finally, we placed the conductive substrate on top of the reference electrodes and connected it to ground. In order to minimize relative motion of the different components of the apparatus, we applied pre-wrap around the entire forearm.

EMG data have been processed using three different pipelines to compare the novel processing scheme presented in this paper with two standard methods. The first method (STD) relies on the assumption that MRI-related movement artifacts are negligible and so it implements the same standard pipeline used to process the data recorded outside the scanner (Fig. ^S7A). In this way, the estimate of the EMG signal x^ is considered to be x^=y. The second method (SUB) compensates for MR-related movement artifacts assuming perfect match between the interference r measured by the REF electrodes and the true interference w (Fig. ^S7B). As such, it quantifies the EMG signal as x^=y-r. Finally, the third method fully implements the pipeline described in “StretchfMRI technique” section, quantifying the EMG signal as x^=y-f^(r) (Fig. ​(Fig.11D).

The EMG signal was processed to quantify reflex responses using standard pipeline³³ modified to include the ANC and SUB pipelines. Specifically, both REF and EMG signals were initially segmented to extract the subset of data points representing perturbation-related activity recorded during the 200 ms silent window (25 ms after volume acquisition is completed), so that the first time point would correspond to the perturbation onset. The segmented signals were band-pass filtered using a 4th order Butterworth filter with cut-off frequencies fLP=20Hz, and fHP=250Hz, and fed to the later components of the filtering pipeline (Fig. ​(Fig.1D,1D, for ANC this is signal yf). The estimate of the EMG activity returned by the ANC, SUB, and STD filters was finally rectified and low-pass filtered with a 4th order Butterworth filter with cut-off frequency fENV=60Hz. To allow between-subject comparison, after filtering, we normalized the stretch-evoked EMG activity by the average EMG (EMGc¯|j) measured during the isometric contractions of the muscle j recorded prior to the beginning of each perturbation session. To determine EMGc¯|j, we used only the central 3 s of activity recorded for the subset of contractions in which the given muscle was active—i.e. only the flexion torques for the FCR and only the extension torques for the ECU. The same constant was used to normalize EMG activity measured in response to perturbations that both stretched and shortened the muscle. Finally, to extract the magnitude of the long-latency response Hi,j elicited by perturbation i on muscle j, we used the cumsum method³⁴, quantifying Hi,j as the area underlying the processed EMG signal EMGi,j(t) in the time window [50, 100] ms after the perturbation onset:

where time is expressed in ms.