Paired Associative Stimulation Fails to Induce Plasticity in Freely Behaving Intact Rats

Animals and surgical preparation

All animal procedures were performed in accordance with the regulations of the Université Laval animal care committee. Nine Long–Evans rats and one Sprague Dawley rat (all male) were housed under 12 h inverse daylight cycle with food and water available ad libitum. Animals were individually housed to prevent implant damage. One-hundred fifty experimental sessions were planned in 10 animals during the dark (active) phase of their daylight cycle (coinciding with our work day) to investigate the effectiveness of 15 interstimulus intervals (ISIs), including control stimulation protocols. However, due to rare implant failure, some ISIs were not tested in all the rats. All ISI conditions were tested in a minimum of five animals, and an average of seven to eight (Extended Data Fig. 4-1). Our PAS intervention typically targeted the extensor carpi radialis (ECR) muscle. However, in some animals, we used pairs of EMG wires implanted in more proximal locations (biceps or trapezius muscles). The distribution of muscles tested within the full dataset is shown in Extended Data Figure 4-2.

Chronic PAS implantation surgery

During aseptic surgeries performed under isoflurane anesthesia, rats were implanted with 1 × 1 mm custom square arrays of four 80/20 platinum-iridium electrodes, each 75 μm in diameter and an approximate impedance of 20 kΩ. The electrodes were inserted 1.5 mm deep into the caudal forelimb area (CFA) of the primary motor cortex (M1) by a stereotaxic craniectomy, centered at 0.7 mm anterior and 3.5 mm lateral relative to bregma, and the surrounding exposed dura matter was covered with silicone gel for protection (Fig. 2B). This allowed us to perform intracortical stimulation using an isolated constant current stimulator (model 2100, A-M Systems). Three pairs of PFA-coated multistranded stainless steel wire electrodes (A-M Systems) were inserted into the contralateral ECR, biceps brachii, and trapezius muscles (the latter two serving as alternative muscles in case the ECR electrode failed). EMG and cortical electrodes were presoldered to either an InVivo1 MS12P or a SAMTEC 2 × 7 connector, which were secured to the skull with dental cement and six bone screws as anchors. A posterior skull screw served as the ground electrode for cortical monopolar stimulation. An additional reference electrode for the EMG measurement was embedded subcutaneously in the upper back. Animals recovered undisturbed for a week after implantation prior to testing and were given time to familiarize themselves with being connected. EMG electrodes were then tested for recording quality, and two electrodes in each cortical array with the lowest stimulation intensity required for motor evoked potentials (MEPs) in the target muscle were determined prior to data collection (barring electrode failure, the same two were used in a bipolar configuration for all of the experiments for that rat).

Study design and experimental paradigm

We used a repeated-measures randomized block design (same rat tested on all ISI conditions in a randomized order) to test the effect of STDP timing condition on the change in integral of the averaged MEP response after the PAS experiment.

We tested nine rats with chronic implants (the implant for one rat failed prior to data collection). Each rat was to be tested once in each condition. To account for possible order effects inherent to a within-subjects design, the order of testing conditions was randomly assigned using the randperm function in MATLAB (Extended Data Fig. 4-1). One condition (ISI, −15 ms) was added at the end for six rats that had all data collection completed, based on another study (Zhang et al., 2018) that showed a promising timing condition and was published while data collection was in progress. For rats with which data collection had not started yet, a rerandomization was performed to integrate this new condition. For each rat, each test was separated by ∼24 h to minimize carryover effects between previous paired stimulation interventions. A posteriori analyses verified that there was no cumulative effect of PAS (Extended Data Fig. 4-4).

We tested the following four control conditions: three PAS controls, involving (1) cortical stimulation only, (2) peripheral stimulation only, and (3) no stimulation, as well as (4) one extralong ISI timing control involving paired stimulation of the motor cortex and the contralateral peripheral muscle offset by +505 ms. We reasoned that if timing during paired stimulation was the driving factor behind plasticity, and not the pairing of stimulation per se, this condition should have a null effect comparable to the previous control conditions.

Each experiment followed a fixed schedule (Fig. 3, inset). After connecting the rat to the hardware interface, we completed three 5 min “probes” to assess the corticomotor excitability prior to the PAS intervention (see following section). The probe was completed when 30 stimulations were delivered or 5 min had elapsed, whichever came first. Probes were separated by 10 min each. After each PAS intervention, three post-PAS probes were completed in the same manner to assess the excitability of the corticomotor system after paired stimulation. We allotted 2 min for wire switching and software changes, immediately before and after each PAS intervention.

• Download figure
• Open in new tab
• Download powerpoint

Figure 3.

Example experiment-level result: MEPs recorded in the right ECR of one rat, obtained from electrical cortical stimulation for each probe. Shaded areas in light blue and red indicate 1 SD about the mean. Inset, Each session began with three 5 min probes in which we performed closed-loop EMG-dependent cortical stimulation to assess baseline MEP amplitudes, each separated by 10 min. The PAS session itself, involving 300 pairs of stimuli to the cortex and the muscle at a rate of 0.5 Hz, took ∼10 min. This was followed by three post-PAS probes so we could assess corticospinal excitability up to 30 min after paired stimulation for each interstimulus interval. After each experiment, we manually verified all MEPs using custom software and excluded traces with movement artifacts or noisy EMG signals.

Probe assessment of corticomotor excitability

To assess corticomotor excitability before and after PAS, we compared the size of MEPs obtained from cortical stimulation using a closed-loop stimulation protocol (Fig. 2C). As demonstrated by Darling et al. (2006), cortical stimulation during a low level of muscle contraction (5% or 10% of their maximal voluntary contraction) reduces MEP variability, compared with fully relaxed conditions. Drawing inspiration from this, we designed a protocol to stimulate during low levels of muscle contraction in the target muscle. To this end, the EMG activity in the target muscle was continuously measured, and cortical stimulation was triggered in real time if the activity reached within 2–12 SDs above the baseline (defined as the mean value of the rectified EMG signal measured over 2 s when the limb was fully relaxed, during sleep, or during sustained rest behavior with no weight bearing in each rat), and if the EMG activity was on the rising phase (i.e., contraction was being initiated during free behavior as opposed to when the muscle was relaxing from a previously larger contraction). The baseline calibration was performed on each rat prior to data collection and recalculated as necessary if we suspected a change in baseline noise. This allowed us to customize the excitability assessment to each rat to adjust for slight differences in electrode placement or impedance across days. However, the EMG assessment window was never changed within a PAS experiment. The 2–12 SD range above baseline effectively restricted the conditions for stimulation within a low to moderate level of voluntary activation of the corticospinal system achieved during free behavior (walking/grooming/exploring). This approach provides the means to stimulate under consistent conditions of corticospinal activity, in an animal model where behavioral instructions cannot be clearly provided such as in human studies. To do this, we used the envelope function in MATLAB, which calculated the peak envelope of the filtered data with a moving spline over the downsampled local maxima of the previous 32 data points. Cortical stimulation was contingent on the EMG envelope crossing the predetermined activity threshold. The variability was still high albeit reduced after applying the closed-loop stimulation protocol, so we averaged all three baseline measurements during the statistical analysis to obtain an overall assessment of corticomotor excitability prior to the PAS intervention. We recorded from different muscles, depending on the location where we obtained the best quality MEPs (Extended Data Fig. 4-2). We always used the same muscle involved in the PAS intervention to provide the closed-loop control for the corticomotor excitability assessments. When available, we chose the ECR as the PAS target muscle, but fell back on the biceps or trapezius, respectively, if electrode malfunctions or failure to evoke MEPs in the more distal muscles prevented their use. In one rat, we used a monopolar EMG recording configuration resulting in an EMG signal contaminated with cross talk from cardiac activity. We manually adjusted the upper and lower limits of the EMG window, enabling probe stimulation in a manner that better reflected a low-amplitude muscle contraction.

Electrophysiological data acquisition and stimulation configuration

Independent paired electrical stimulation protocols were achieved through two A-M Systems 2100 stimulators, each connected to separate pins on an InVivo1 commutator through a custom-made breakout board interface. Multichannel recording was made possible by routing the EMG signal into a Brownlee Precision Model 440 Instrumentation Amplifier. Within this unit, a signal gain of 100, a bandpass filter between 50 Hz and 1.0 kHz was used for EMG, and bandpass of 1–300 Hz for local field potential (LFP) signals. A 60 Hz notch filter was applied. This output signal was split in two, with one copy being routed into a Powerlab 8/sp unit by AD Instruments and further processed with a 10 Hz high-pass filter before being saved. The second copy was routed into a National Instruments Digital to Analog Converter (DAC) SCB-68A system, which was operated via custom MATLAB software. We used the DAC system and MATLAB software to initiate all probe and PAS stimulation protocols via the trigger input ports on the A-M Systems stimulators.

Latency measurements and sign convention for spike-timing experiments

To confirm the conduction latencies, we completed a series of acute experiments under anesthesia, in rats with a similar weight and size to those used for chronic implants. First, to measure the antidromic conduction time in motoneurons between the muscle and the spinal cord, we performed an acute experiment under urethane anesthesia to record and stimulate between the spinal cord and the ECR muscle, respectively. We exposed the dorsal spinal cord between the C4 and C6 regions by performing a laminectomy and deafferented the C3 to C7 segments by cutting the dorsal roots to isolate antidromic propagation instead of conduction along afferent sensory fibers. With the dura intact, we inserted a tungsten electrode, 127 μm in diameter, into the C5 region ipsilateral to the right forelimb, 1.0 mm lateral to the midline. We also inserted a pair of EMG electrodes in the right extensor carpi radialis using the same method as in the chronic implants. Stimulation of the spinal cord C5 region using single pulses led to an isolated wrist extension in the forelimb of the rat, verifying the location of the ECR motoneuron pool for efferent connections (Tosolini and Morris, 2012). Following this, we stimulated the EMG electrodes and recorded LFPs from the electrode site in C5. Filtering parameters for the LFP recording included a bandpass of 1–300 Hz, with a 60 Hz notch filter and a gain of 100. Data were averaged across 200 stimulations. We determined the antidromic motoneuronal propagation delay to be 3 ms (Extended Data Fig. 2-1A).

With an average MEP latency of 12 ms for ECR deriving from cortical stimulation of the intact animal, and a 3 ms peripheral efferent conduction time, we estimated a latency of 9 ms for a cortical stimulation-induced descending volley to reach motoneurons, including synaptic integration time.

In a second acute experiment with a different animal under ketamine/xylazine anesthesia, we measured the time a neuronal volley requires to reach the cortex after muscle stimulation. In an intact animal, we recorded LFP responses in M1 following intramuscular stimulation (Extended Data Fig. 2-1B). We postulated that the peak of the initial negative inflection in the local field potential from contralateral muscle stimulation reflected the time at which the greatest neural activity is observed among the postsynaptic neurons in the cortex. Again, we inserted a pair of EMG electrodes into the right extensor carpi radialis, then inserted one platinum-iridium electrode 1.5 mm dorsoventrally into M1, centered at the array coordinates of the rats involved in the PAS experiments. A reference electrode ∼1 mm lateral from the first was positioned on the surface of the dura. Both cortical electrodes were connected by a common ground at the skull screw, and LFPs were measured by calculating the voltage differential between the cortical electrodes with the same LFP recording parameters described above. Stimulation was delivered to the EMG electrode in the right ECR through bipolar single pulses with a 0.2 ms duration, repeated at 0.5 Hz. Data were averaged across 360 stimulations. The afferent latency from ECR stimulation to the peak of the cortical evoked potential was 16 ms (Extended Data Fig. 2-1B).

Using these conduction latencies, we chose a set of stimulus intervals that would result in various presynaptic and postsynaptic timings relevant to the rules of spike timing-dependent plasticity, either at the cortical and/or spinal levels. The full list of ISI conditions tested can be found in Extended Data Figure 4-1. Our experimental design and results followed the convention that a positive latency means the periphery was stimulated after the cortex by that time difference. These stimulation offsets lead to physiological offsets calculated at the levels of the spinal cord and cortex; positive latencies result in presynaptic activity that preceded postsynaptic activity at the specified location.

PAS intervention

We used a PAS protocol of 300 paired stimulations to the motor cortex and designated peripheral muscle, using single pulses of biphasic electrical stimulation 0.2 ms in duration, separated by 0.5 Hz. We note that this is on the higher end in terms of the number of paired stimulations compared with previous protocols, and is delivered at a higher frequency, but we reasoned, in the absence of evidence otherwise, that any effect that may be present due to paired stimulation should be enhanced using this slightly more intensive protocol.

Cortical stimulation intensity was set at 1.25 times the threshold for an MEP and muscle stimulation at 1.5 times the threshold to elicit a visible twitch (the mean motor threshold across all experiments was 790 μA for cortical stimulation and 1.8 mA for muscle stimulation). Thresholds were operationally defined as the minimal stimulation intensity required to induce a response >50% of the time. All PAS experiments were completed in the home cages of the animals with a modified cover that enabled us to pass the tethering cable during free behavior (consisting mostly of walking, grooming, and exploring, and sometimes sleeping).

MEP measurement

Raw EMG data were saved and processed offline in LabChart Version 7, and custom scripts were written in MATLAB. We plotted all individual responses for each cortical stimulation and manually excluded trials for which there was significant excessive movement artifact and/or lack of EMG signal (this was rare, and the most likely reason was due to an intermittent connection with a faulty cable, which was repaired or replaced promptly). The resulting set of verified MEPs for each probe was collected for further analysis.

MEP amplitudes were initially quantified with the following three different methods: (1) the peak-to-peak value of individual EMG responses (the literature standard); (2) the mean value of the integral of individual rectified EMG responses, measured over a tailored time window following stimulation; and (3) the integral of the averaged rectified EMG responses over the same time window. Every individual response to cortical stimulation was first manually screened to exclude any EMG traces containing large movement artifacts or other obvious contamination. In pilot analyses from our earlier experiments (data not shown), we assessed qualitatively that the calculation method did not much impact the normalized changes in the MEPs, so we proceeded with taking the integral of the average response for the probe (method 3 above). We reasoned that this approach was most effective in capturing both unimodal and multimodal MEP responses. This decision was made prior to the pooled study data analysis. In summary, the MEP values reported here were thus calculated by first rectifying the filtered EMG signal, then averaging the activity from all stimuli within a probe postscreening, and then calculating the integral of the resulting signal (Fig. 3). The software described in this article is freely available online at https://github.com/ethierlab (Windows 10). The code is available as Extended Data, if required.

Statistical analysis

Statistical analysis and visualization were completed using SAS Software, version 9.4 for Windows, Minitab 18 for Windows, and R 3.6.1/RStudio 1.2.5019 for Windows. We completed a mixed-design ANOVA with repeated measures on the normalized data to test the main effects of ISI timing condition (CONDITION) with 15 levels (1 level for each of 11 timings and four control conditions) as well as PAS probe (SESSION) with three levels (2, 17, and 32 min after PAS). We also tested for any interaction effects between CONDITION and SESSION. A random effect on the rat was used to account for the randomized block design. The level of significance for the mixed ANOVA was fixed at p < 0.05. Type III fixed effects are reported in Table 1, obtained through the Restricted Maximum Likelihood (REML) estimation method. Data from one rat in the ISI +6 condition was removed from the statistical analysis because of poor data quality (very few MEPs in each probe). Normality was assessed on standardized residuals using graphical methods.