Augmenting LTP-Like Plasticity in Human Motor Cortex by Spaced Paired Associative Stimulation

Abstract

Paired associative stimulation (PAS_LTP) of the human primary motor cortex (M1) can induce LTP-like plasticity by increasing corticospinal excitability beyond the stimulation period. Previous studies showed that two consecutive PAS_LTP protocols interact by homeostatic metaplasticity, but animal experiments provided evidence that LTP can be augmented by repeated stimulation protocols spaced by ~30min. Here we tested in twelve healthy selected PAS_LTP responders the possibility that LTP-like plasticity can be augmented in the human M1 by systematically varying the interval between two consecutive PAS_LTP protocols. The first PAS_LTP protocol (PAS1) induced strong LTP-like plasticity lasting for 30-60min. The effect of a second identical PAS_LTP protocol (PAS₂) critically depended on the time between PAS₁ and PAS₂. At 10min, PAS₂ prolonged the PAS₁-induced LTP-like plasticity. At 30min, PAS₂ augmented the LTP-like plasticity induced by PAS₁, by increasing both magnitude and duration. At 60min and 180min, PAS₂ had no effect on corticospinal excitability. The cumulative LTP-like plasticity after PAS₁ and PAS₂ at 30min exceeded significantly the effect of PAS₁ alone, and the cumulative PAS₁ and PAS₂ effects at 60min and 180min. In summary, consecutive PAS_LTP protocols interact in human M1 in a time-dependent manner. If spaced by 30min, two consecutive PAS_LTP sessions can augment LTP-like plasticity in human M1. Findings may inspire further research on optimized therapeutic applications of non-invasive brain stimulation in neurological and psychiatric diseases.

Introduction

Motor rehabilitation after cerebral injury such as stroke depends on neural plasticity, including synaptic strengthening by long-term potentiation (LTP) [1–4]. Paired associative stimulation (PAS_LTP) of the human primary motor cortex (M1) can induce an increase in corticospinal excitability as measured by motor evoked potentials (MEPs) beyond the stimulation period [5], which resembles LTP as studied at the cellular level [6–9]. However, LTP-like plasticity induced by PAS_LTP is regulated by homeostatic metaplasticity [6, 10, 11], i.e., a higher-order form of plasticity, which keeps neuronal and network excitability in a physiological range [12–15]. This homeostatic regulation implies that repeated induction of LTP-like plasticity at short delays is suppressed, which may limit the therapeutic potential of PAS_LTP to increase corticospinal excitability.

Experiments in animals, however, show that the brain possesses powerful mechanisms, which permit continued synaptic strengthening in the context of prior LTP. One such mechanism is based on an N-methyl-D-aspartate receptor (NMDAR)-dependent form of metaplasticity by which continued synaptic strengthening is possible through activation of metabotropic glutamate receptors (mGluRs) [16]. In addition, inhibition of glycogen synthase kinase-3 beta (GSK3β) results in a ~60 min lasting blockade of subsequent induction of NMDAR-dependent long-term depression (LTD), because expression of LTD requires a high level of GSK3β activity [17]. These mechanisms ensure that information encoded by LTP is not erased during ongoing neural activity, but can be retained. On the system level of the human cortex it has been shown, that motor learning immediately following PAS_LTP is not suppressed, as would be expected in the framework of homeostatic metaplasticity, but rather facilitated [18, 19]. In addition, several other studies have occasionally reported non-homeostatic metaplasticity between two consecutive non-invasive brain stimulation protocols, mainly at short intervals of 30min or less [20–24], for review [25]. Even though these studies provide system-level evidence for non-homeostatic interactions between plasticity-inducing non-invasive brain stimulation protocols, and motor learning, respectively, the conditions favoring non-homeostatic vs. homeostatic metaplasticity in the human brain remain poorly understood.

The present study investigated the role of time between two consecutive PAS_LTP protocols for repetitive induction of LTP-like plasticity in M1 of healthy human subjects. We studied the interactions between two identical sessions of PAS_LTP spaced at inter-PAS_LTP intervals (IPI) of 10, 30, 60 and 180min. Findings show that metaplasticity in human M1 is expressed in a time-dependent manner with a window of non-homeostatic metaplasticity at an IPI of 30min. If spaced by 30min, two consecutive PAS_LTP sessions can augment LTP-like plasticity in human M1. These findings may inspire further research on optimized therapeutic applications of non-invasive brain stimulation techniques in a clinical setting.

Materials and Methods

Subjects

Written informed consent was obtained from all subjects prior to participation. None of the subjects had a history of neurological or psychiatric disease or was on CNS-active drugs at the time of the experiments and all subjects were checked for contraindications to transcranial magnetic stimulation (TMS) [26]. The study conformed to the Declaration of Helsinki and was approved by the ethics committee of the Medical Faculty of Goethe-University Frankfurt. Twenty-seven subjects were screened for a significant PAS_LTP-induced increase in motor evoked potential (MEP) amplitude ≥ 1.1 (ratio of mean MEP amplitude post-PAS_LTP / pre-PAS_LTP) [27, 28]. Sixteen subjects fulfilled this criterion (PAS_LTP responders) and were included into the study. Of those, four subjects withdrew consent after the first experiment for experiencing some discomfort by TMS. Thus, complete datasets were obtained in 12 subjects (6 females, mean (± SEM) age, 25.6 ± 1.4 years). All subjects were right-handed according to the Edinburgh handedness questionnaire [29].

Electromyography (EMG) recordings

Surface EMG recordings were obtained from the right abductor pollicis brevis (APB) muscle using Ag-AgCl electrodes in a belly-tendon montage. The raw EMG signal was amplified and band-pass filtered (20Hz–2kHz, Counterpoint Mk2 electromyograph, Dantec, Denmark), digitized at an A/D rate of 5kHz (Micro1401, Cambridge Electronic Design, UK) and stored in a laboratory computer for online display and offline analysis, using customized software (Spike2 for Windows, Version 3.05, Cambridge Electronic Design).

Transcranial magnetic stimulation (TMS)

Subjects were seated comfortably in a reclining chair. TMS was performed using a Magstim 200 magnetic stimulator (Magstim Company, UK) with a monophasic current waveform. Stimuli were applied to the hand area of left M1 through a figure-of-eight coil (inner diameter of each wing, 70mm) with the handle pointing backwards and 45° away from midline. The optimal coil position for eliciting MEPs in the right APB was marked with a soft-tipped pen on the scalp in order to ensure constant placement of the coil throughout the experiment.

At the beginning of each experiment the resting motor threshold (RMT) was measured, which was defined as the lowest intensity (indicated in percent of maximum stimulator output, MSO) that elicited small MEPs (>50 μV) in at least five out of ten consecutive trials in the resting APB. RMT was determined to nearest 1% of MSO. Thereafter, the intensity to elicit MEPs of, on average, 1mV peak-to-peak amplitude (SI_1mV) was determined in the resting APB. This stimulus intensity was then kept constant throughout a given session of the main experiment (Fig 1).

Fig 1. Time line of experiments.

In the main experiment, motor-evoked potentials (MEPs) were recorded at all time points (circles; B0, B1-B2, P1-P9) with SI_1mV, i.e. the stimulation intensity that evoked MEPs of, on average, 1mV peak-to-peak amplitude in the resting abductor pollicis brevis muscle at baseline (time point B0). Note that subjects took part in four experimental sessions in a crossover design with different intervals between the consecutive sessions of two identical LTP-like plasticity inducing PAS_LTP protocols (PAS₁, PAS₂): 10min (IPI₁₀), 30min (IPI₃₀), 60min (IPI₆₀), and 180min (IPI₁₈₀). In the control experiment (IPI_30adj) SI_1mV was readjusted at time point B2 to match baseline MEPs of 1mV peak-to-peak amplitude (SI'_1mV). Otherwise, the control experiment was identical to the IPI₃₀ condition of the main experiment.

Paired associative stimulation (PAS_LTP)

PAS_LTP was applied according to a protocol originally described by Stefan and colleagues [5] and slightly modified by our group [6, 10]. Briefly, electrical stimulation of the median nerve at the wrist of the right hand was applied through a bipolar electrode (cathode proximal), using constant-current square-wave pulses (duration, 1ms) at an intensity of three times the perceptual sensory threshold. Each stimulus was followed by single-pulse TMS of the left M1 at SI_1mV. The interstimulus interval equaled the individual N20-latency of the median nerve somatosensory-evoked cortical potential plus 2ms (mean ± SEM, 21.3 ± 0.3ms). At this or similar intervals PAS induces an LTP-like increase of MEPs in the majority of subjects [5, 6, 27, 30–32]. PAS_LTP consisted of 225 stimulus pairs applied at a frequency of 0.25Hz. The level of attention, a significant modulator of PAS_LTP effects [33], was controlled and attention was maximized to the stimulated hand by a light emitting diode (LED) attached to the right wrist which flashed randomly (0.2–1Hz) during PAS_LTP. Subjects were requested to count and report the total number of flashes as correctly as possible at the end of PAS_LTP.

Experimental design

In the main experiment, all subjects took part in four different sessions in a pseudorandomized crossover design (Fig 1). Each session consisted of two identical, consecutive PAS_LTP protocols (PAS₁, PAS₂) with a specific inter-PAS_LTP interval (IPI). Intervals were 10min (IPI₁₀), 30min (IPI₃₀), 60min (IPI₆₀) and 180min (IPI₁₈₀). Between PAS₁ and PAS₂ subjects were requested to stay awake, keep seated and not to use their stimulated hand. The order of IPI conditions was pseudo-randomized across subjects and experimental sessions were separated by at least three days to avoid carry-over effects (mean individual minimum inter-session interval 7.6 days, range 3–14 days).

To test whether the significant PAS₂-induced increase of MEP amplitude in the IPI₃₀ condition could be attributed to the PAS₁-induced increase in MEP amplitude at time point B2 immediately before PAS₂ (see below and cf. Figs 2 and 3), we conducted a control experiment (IPI_30adj), in which we readjusted MEP amplitudes at time point B2 by reducing the stimulation intensity (SI'_1mV) in order to match baseline MEPs of, on average, 1mV peak-to-peak amplitude. The readjusted stimulation intensity SI'_1mV was then used for PAS₂ and all following MEP measurements. Otherwise the control experiment was identical to the IPI₃₀ condition of the main experiment (Fig 1). Nine subjects (4 females, mean (± SEM) age, 26.1 ± 1.6 years) took part in the control experiment.

Fig 2. PAS₁-induced increases in MEP amplitude.

PAS₁ resulted in comparable immediate MEP amplitude increases (MEP_B1/MEP_B0) in all IPI conditions (data not shown). At time point B2 (MEP_B2/MEP_B0), MEP amplitude increases were present 10min (IPI₁₀) and 30min (IPI₃₀) after PAS₁, but no longer at 60min (IPI₆₀) and 180min (IPI₁₈₀) after PAS₁. Asterisks indicate significant differences from 1 (P < 0.01; one-sample two-tailed t tests). Note that MEP_B2/MEP_B0 was not significantly different between conditions IPI₁₀ and IPI₃₀ (P > 0.09; paired two-tailed t tests). Data are means ± 1 SEM from twelve subjects.

Fig 3. PAS₂-induced increases in MEP amplitude after PAS₁-priming.

A: The PAS₂-induced increase in MEP amplitude after PAS₁-priming (MEP_Pmean/MEP_B2) was significantly higher at IPI₃₀ than IPI₁₀ (#, P < 0.05; paired two-tailed t test). B: Time course of MEP_Pi/MEP_B2 (i = 1,2,…,6) over 30min after PAS₂ for IPI₁₀ (open squares), IPI₃₀ (filled squares), IPI₆₀ (open circles), and IPI₁₈₀ (filled circles). C: The cumulative effect of PAS₁ and PAS₂ on MEP amplitudes (MEP_Pmean/MEP_B0) was significantly higher for IPI₃₀ than for IPI₆₀ and IPI₁₈₀ (#, P < 0.05; paired two-tailed t tests), and showed a trend to be higher for IPI₃₀ vs. IPI₁₀ (P = 0.095; paired two-tailed t test). D: Time course of MEP_Pi/MEP_B2 (i = 1,2,…,6) over 30min after PAS₂ for IPI₁₀ (open squares), IPI₃₀ (filled squares), IPI₆₀ (open circles), and IPI₁₈₀ (filled circles). *, P < 0.05; one-sample two-tailed t tests. Data are means ± 1 SEM from twelve subjects.

Quantification of PAS_LTP effects

The PAS₁ effect was quantified by comparing 20 single-trial peak-to-peak MEP amplitudes at SI_1mV 5min after PAS₁ (time point B1) with those at baseline immediately before PAS₁ (time point B0; cf. Fig 1). The inter-trial interval was 10s ± 25% to limit anticipation of the next trial. If the PAS₁ effect (ratio MEP_B1/MEP_B0) was < 1.1, the experiment was stopped, the data were discarded and the experiment was repeated on another day, to minimize intra-individual variability [34] and ensure PAS₁-induced LTP-like plasticity in all experiments (overall, 12 repeated experiments in all subjects; 1.0 ± 0.4 repeated experiments per subject). Five minutes before the second PAS_LTP intervention (PAS₂) another block of 20 MEPs was measured (time point B2; in case of IPI₁₀ B1 was taken as B2). After PAS₂ six blocks of MEP measurements were performed at intervals of 5min (time points P1–P6) to monitor PAS₂ effects on MEPs for 30min after the end of PAS₂. In six subjects, additional MEP measurements were conducted 60min (time point P7), 120min (P8), and 180min (P9) after PAS₂ to determine the time course of the return of PAS₂-induced MEP amplitude increases (see Results) back to baseline. All measurements were recorded in the resting APB at SI_1mV, except for the control experiment (IPI_30adj), in which the TMS intensity was readjusted to SI'_1mV at time point B2 and then used for PAS₂ and all following MEP measurements. Complete voluntary relaxation was monitored by audio-visual feedback of the EMG raw signal at high gain (at 50μV ⁄ division). Trials contaminated with voluntary EMG activity were excluded from analysis (<1% of all trials).

Statistical analyses

Statistical testing was performed with IBM SPSS Statistics (Version 20.0.0). To test for differences of RMT, SI_1mV, and MEP amplitude at baseline (time point B0) independent one-way repeated measures analyses of variance (rmANOVAs) with the within-subject factor IPI (IPI₁₀, IPI₃₀, IPI₆₀, IPI₁₈₀) were conducted. To test for differences of the PAS₁ effect one-way rmANOVAs with the within-subject factor IPI (IPI₁₀, IPI₃₀, IPI₆₀, IPI₁₈₀) were performed separately on mean MEP amplitudes at time points B1 and B2, respectively, normalized to the individual mean MEP amplitude at time point B0 (MEP_B1/MEP_B0; MEP_B2/MEP_B0). To test for the effects of IPI (IPI₁₀, IPI₃₀, IPI₆₀, IPI₁₈₀) and TIME (P1–P6) on PAS₂ effects, two-way rmANOVAs were performed separately on mean MEP amplitudes at time points P1–P6 normalized to the individual mean MEP amplitude at time points B0 and B2, respectively (MEP_Pi/MEP_B0, MEP_Pi/MEP_B2, i = 1,2,…,6). Normalization to B2 provides information specifically on the effects of PAS₂ while normalization to B0 provides information on the cumulative effects of PAS₁ and PAS₂.

Mauchly’s test was applied to test for non-sphericity and in case of violation of sphericity the Greenhouse-Geisser correction was used. Conditional on significant main effects or their interactions in the rmANOVAs, post hoc pairwise comparisons or one-sample two-tailed t tests were performed. The Bonferroni correction was applied to adjust for multiple comparisons.

To compare the time course of PAS₁- and PAS₂-induced changes in MEP amplitude, the MEP amplitude raw data after PAS₁ and PAS₂ were fitted independently with a linear mixed effects model with fractional polynomials at all IPIs. PAS₁ data comprised mean MEP amplitudes measured at time points B0 (time: 0min) and B2 across IPI conditions (time: 10min at IPI₁₀, 30min at IPI₃₀, 60min at IPI₆₀, and 180min at IPI₁₈₀), whereas PAS₂ data comprised mean MEP amplitudes measured at time points B2 (time: 0min), P2 (time: 10min), P6 (time: 30min), P7 (time: 60min), and P9 (time: 180min), respectively, for each IPI condition separately. Model functions for PAS₁- and PAS₂ effects on MEP amplitudes were compared using F- and t-statistics, respectively.

In all tests the significance level was set to P < 0.05. All data are expressed as means ± SEM.

Results

All subjects tolerated the experimental procedures well without any adverse effects.

Baseline excitability data (RMT, SI_1mV, MEP)

The data are summarized in Table 1. There were no significant differences between IPI conditions on RMT (F _3,33 = 0.55, P = 0.53), SI_1mV (F _3,33 = 0.46, P = 0.71), or MEP amplitude (F _3,33 = 1.07, P = 0.38) at baseline (time point B0).

Table 1. Summary of baseline excitability measures in the different IPI conditions.

Condition	RMT	SI1mV	MEP
IPI10	40.1 ± 2.1	50.8 ± 2.2	0.97 ± 0.04
IPI30	39.3 ± 1.6	51.4 ± 2.0	0.91 ± 0.05
IPI60	38.9 ± 1.8	50.5 ± 2.4	0.93 ± 0.04
IPI180	38.8 ± 1.4	49.6 ± 1.7	1.02 ± 0.05

Abbreviations: IPI, inter-PAS_LTP interval [in min, index]; MEP, motor-evoked potential [in mV]; RMT, resting motor threshold [in %maximum stimulator output, MSO]; SI_1mV, stimulation intensity that induces MEPs of 1mV peak-to-peak amplitude on average [in %MSO].

PAS₁ effects on MEP amplitudes (comparison of time points B1 and B2 vs. B0)

The one-way rmANOVA for the PAS₁ effect at time point B1 (MEP_B1/MEP_B0) showed no significant difference between IPI conditions (F _3,33 = 0.71, P = 0.55), indicating similar PAS₁-induced LTP-like plasticity across all IPI conditions. MEP_B1/MEP_B0 was > 1.0 for IPI₁₀ (1.69 ± 0.13; t = 5.25, P = 0.0003), IPI₃₀ (1.49 ± 0.10; t = 5.01, P = 0.0004), IPI₆₀ (1.50 ± 0.07; t = 7.57, P < 0.0001), and IPI₁₈₀ (1.53 ± 0.13; t = 4.12, P = 0.0017). In contrast, the one-way rmANOVA for the PAS₁ effect at time point B2 (MEP_B2/MEP_B0) immediately before PAS₂ showed a significant effect of IPI (F _3,33 = 11.74, P < 0.001). MEP_B2/MEP_B0 was > 1.0 for IPI₁₀ (1.69 ± 0.13; t = 5.25, P = 0.0003) and IPI₃₀ (1.51 ± 0.12; t = 4.32, P = 0.0012), but no longer for IPI₆₀ (1.17 ± 0.09; t = 1.90, P = 0.084) or IPI₁₈₀ (0.94 ± 0.05; t = -1.20, P = 0.26) (Fig 2). Of note, there was no significant difference of MEP_B2/MEP_B0 between conditions IPI₁₀ and IPI₃₀ (P > 0.9), indicating similar persistence of PAS₁-induced LTP-like plasticity in these conditions at time point B2 immediately before PAS₂.

PAS₂ effects on MEP amplitudes (comparison of time points P1-P6 vs. B2)

The two-way rmANOVA for the PAS₂ effect on MEP amplitudes at time points P1–P6 (MEP_Pi/MEP_B2, i = 1,2,…,6) showed a significant effect of IPI (F _3,33 = 2.94, P = 0.048) and TIME (F _5,55 = 2.50, P = 0.041), whereas the interaction IPI with TIME was not significant (F _15,165 = 0.73, P = 0.75) (Fig 3A and 3B). MEP_Pmean/MEP_B2 was > 1.0 for IPI₃₀ (1.39 ± 0.15; t = 2.59, P = 0.025), but not for IPI₁₀ (0.90 ± 0.06; t = -1.72, P = 0.11), IPI₆₀ (1.20 ± 0.15; t = 1.34, P = 0.21) or IPI₁₈₀ (1.23 ± 0.16; t = 1.48, P = 0.17). Post hoc testing showed that MEP amplitudes after PAS₂ (MEP_Pmean/MEP_B2) were significantly different between conditions IPI₁₀ and IPI₃₀ (P = 0.047), while all other pairwise comparisons were not significant (all P > 0.2).

It could be argued that the significant PAS₂ effect on MEP amplitudes at IPI₃₀ was due to the increased MEP amplitude at time point B2 immediately before PAS₂ (cf. Fig 2). To address this issue, we conducted a control experiment (IPI_30adj), in which we readjusted MEP amplitudes at time point B2 immediately before PAS₂ by reducing the stimulation intensity (SI'_1mV) in order to match baseline MEPs (0.99 ± 0.05mV at time point B0 with SI_1mV = 45.8 ± 1.6%MSO; 1.01 ± 0.06mV at time point B2 with SI'_1mV = 44.2 ± 1.6%MSO, t = -0.29, P > 0.7; paired two-tailed t test). Thus, whilst PAS_LTP induced similar MEP increases at time point B1 (MEP_B1/MEP_B0) in the control (IPI_30adj, 1.31 ± 0.05; t = 6.62, P = 0.0002, one-sample two-tailed t test) and the main experiment (IPI₃₀, 1.49 ± 0.10; t = 5.01, P = 0.0004, one-sample two-tailed t test; IPI_30adj vs. IPI_30: P > 0.15, unpaired two-tailed t test), MEP_B2/MEP_B0 was significantly different between the two experiments (IPI_30adj: 1.02 ± 0.04, t = 0.38, P > 0.7; IPI₃₀: 1.51 ± 0.12; t = 4.32, P = 0.0012, one-sample two-tailed t tests; IPI_30adj vs. IPI_30: P = 0.0025, unpaired two-tailed t test; Fig 4). However, PAS₂ induced a similar increase in MEP amplitudes (MEP_Pmean/MEP_B2) in the control (IPI_30adj: 1.29 ± 0.09, t = 3.09, P = 0.015) compared to the main experiment (IPI₃₀: 1.39 ± 0.15; t = 2.59, P = 0.025; one-sample two-tailed t tests; IPI_30adj vs. IPI_30: P > 0.5, unpaired two-tailed t test; Fig 4). This finding strongly suggested that the increased MEP amplitude at time point B2 at IPI₃₀ was not relevant for the significant PAS₂-induced increase of MEP amplitudes in this condition_. This notion was further supported by the observation, that PAS₂ had significantly different effects on MEP amplitudes at IPI₃₀ vs. IPI₁₀ in the main experiment (cf. Fig 3A and 3B), although MEP amplitudes were increased to a similar level immediately before PAS₂ (MEP_B2/MEP_B0) in these two conditions (cf. Fig 2).

Fig 4. PAS₂-induced increase in MEP amplitude after PAS₁-priming in the control experiment (IPI_30adj).

MEP amplitudes at time point B2 immediately before PAS₂ were successfully readjusted by reducing the stimulation intensity (SI'_1mV) to match baseline MEPs at time point B0 (PAS₁, MEP_B2/B0). Despite this readjustment, PAS₂ induced a similar increase in MEP amplitudes in the control (IPI_30adj) compared to the main experiment (IPI₃₀) (PAS₂, MEP_Pmean/MEP_B2). *, P < 0.05, one-sample two-tailed t tests; #, P < 0.01, unpaired two-tailed t test. Data from the control experiment are from nine subjects, means ± 1 SEM.

Cumulative PAS₁ and PAS₂ effects on MEP amplitudes (comparison of time points P1-P6 vs. B0)

The two-way rmANOVA for the cumulative PAS₁ and PAS₂ effect on MEP amplitudes at time points P1–P6 (MEP_Pi/MEP_B0, i = 1,2,…,6) showed a significant effect of IPI (F _3,33 = 11.36, P < 0.0001) and TIME (F _5,55 = 2.49, P = 0.042), whereas the interaction IPI with TIME was not significant (F _15,165 = 0.66, P = 0.82) (Fig 3C and 3D). MEP_Pmean/MEP_B0 was > 1.0 for IPI₁₀ (1.51 ± 0.14; t = 3.61, P = 0.0041), IPI₃₀ (2.04 ± 0.23; t = 4.54, P = 0.0009), and IPI₆₀ (1.28 ± 0.09; t = 2.98, P = 0.013), but not for IPI₁₈₀ (1.12 ± 0.12; t = 0.97, P = 0.35). Post hoc testing showed significantly higher MEP amplitudes after PAS₂ (MEP_Pmean/MEP_B0) for IPI₃₀ than for IPI₆₀ (P = 0.009) and IPI₁₈₀ (P = 0.01), and a trend for higher MEP amplitudes for IPI₃₀ vs. IPI₁₀ (P = 0.095). Notably, at IPI₃₀ the MEP amplitude increase 30 minutes after PAS₂ normalized to baseline (MEP_P6/MEP_B0 = 2.22 ± 0.25; Fig 3D) was significantly higher than the MEP amplitude increase 30 minutes after PAS₁ in this condition (MEP_B2/MEP_B0 = 1.51 ± 0.12; P = 0.018, paired two-tailed t test), indicating that consecutive application of PAS₁ and PAS₂ at IPI₃₀ induced significantly higher MEP increases than application of PAS₁ alone.

Modelling of PAS₁ and PAS₂ effects on MEP amplitudes

Computational modelling of the PAS₁ and PAS₂ effects on the absolute MEP amplitude raw data as a function of time revealed highly significant differences between PAS₁ and PAS₂ MEP functions (P < 0.0001) (Fig 5). Post hoc testing showed significant differences between the PAS₁ and PAS₂ model functions for all IPI conditions (P < 0.0001 each), suggesting that PAS₂ effects on MEP amplitudes were modulated by prior application of PAS₁ in all IPI conditions. Specifically, at IPI₁₀ PAS₂ prolonged the MEP increase induced by PAS₁. At IPI₃₀, PAS₂ induced an extra MEP increase, notably with a longer time course as compared to the PAS₁-induced MEP increase. In contrast, at IPI₆₀ and IPI₁₈₀ there was no significant change in MEP amplitudes after PAS₂.

Fig 5. Computational modeling of the time course of PAS₁ and PAS₂ effects.

MEP amplitudes as a function of time modeled from the experimental MEP raw data independently for PAS₁ and PAS₂ at all IPIs. PAS₁ and PAS₂ model functions, i.e. the time course of MEP amplitude changes after PAS₁ vs. those after PAS₂, were significantly different at all IPIs (P < 0.0001 each). In addition, PAS₂ effects at IPI₁₀ and IPI₃₀ were significantly different from PAS₂ effects at IPI₆₀ and IPI₁₈₀ (P < 0.0001), and PAS₂ effects at IPI₁₀ from PAS₂ effects at IPI₃₀ (P = 0.033), but not PAS₂ effects at IPI₆₀ from those at IPI₁₈₀ (P > 0.5). Experimental data are shown as mean ± SEM. Note that data for PAS₂ at time points 60min and 180min post PAS₂ are from six subjects only, whereas all other data are from twelve subjects (see Material and Methods). Y-axis, logarithmic scaling.

In addition, the time courses of PAS₂-induced MEP amplitude changes were significantly different between IPI conditions (P < 0.0001). Post hoc analysis showed that PAS₂ effects at IPI₁₀ and IPI₃₀ were significantly different from PAS₂ effects at IPI₆₀ and IPI₁₈₀ (all P < 0.0001), and PAS₂ effects at IPI₁₀ were significantly different from PAS₂ effects at IPI₃₀ (P = 0.033), whereas the comparison of PAS₂ effects at IPI₆₀ and IPI₁₈₀ showed no significant differences (P > 0.5). These differences are explained by a prolonged MEP increase at IPI₃₀ compared to all other conditions (Fig 5).

Discussion

We showed here that LTP-like plasticity can be augmented in human M1 when two consecutive PAS_LTP protocols are spaced by 30min. In contrast, at longer intervals (60-180min) we found a suppressive interaction between two consecutive PAS_LTP protocols. These findings support the notion of non-homeostatic and homeostatic metaplasticity, respectively, and will be discussed in detail below.

The PAS₁-induced MEP increase lasted for 30-60min, in accord with the literature, and represents a form of plasticity resembling LTP as studied at the cellular level [5, 7, 9, 35]. In contrast, the effects of PAS₂ on MEP amplitude depended critically on the interval to PAS₁. Computational modelling of the time courses of MEP amplitude (Fig 5) revealed significant differences between PAS₁ and PAS₂ effects on MEP amplitude for all IPI conditions, indicating that priming M1 by PAS₁ modulated the effects of a subsequent identical PAS₂ protocol for at least three hours.

Metaplasticity constitutes a higher-order form of plasticity, which regulates the magnitude and duration of synaptic plasticity in an activity-dependent manner [12]. Importantly, it modulates the plasticity state of neurons and networks, i.e. the induction of LTP subsequent to the priming stimulation, in the absence of synaptic plasticity induced by the priming stimulation itself. Here, we demonstrated a homeostatic interaction between PAS₁ and PAS₂ at IPI₆₀ and IPI₁₈₀ (Fig 3) in the absence of any persistent PAS₁-induced MEP increase at the time of PAS₂ (Fig 2). Further, both magnitude and/or duration of PAS₂-induced LTP-like plasticity were modulated by PAS₁ (Fig 5). Thus, these findings support the notion that the interactions between PAS₁ and PAS₂ effects described in the present study represent a form of metaplasticity similar to the ones reported at the cellular level [12, 36].

Previous studies have reported on metaplasticity between two consecutive identical non-invasive brain stimulation protocols in the human M1 [10, 20–24, 37–40]. The predominant interaction was homeostatic metaplasticity, but several occasions of non-homeostatic metaplasticity have also been reported, predominantly at short IPIs of 3-20min (for review, [25]). This non-homeostatic metaplasticity resulted in late LTD- and LTP-like changes in corticospinal excitability that were prevented by pharmacological blockade of NMDA receptors by dextromethorphan [24] and that were resistant to de-depression interventions such as voluntary contraction of the target muscle [22]. These data are compatible with the finding in the present study that two consecutive PAS_LTP protocols, if spaced by 30min, resulted in non-homeostatic augmentation of LTP-like plasticity with a prolonged duration.

These time-dependent interactions between PAS₁ and PAS₂ may be explained in the framework of a cascade model of synaptic plasticity [41]. As synapses are modified during the course of LTP, they change between discrete mechanistic states [42–44]. For example, silent synapses, i.e. synapses containing NMDARs, but being devoid of α-amino-3-hydroxy-5-methyl-4-isoxazolepropionic acid receptors (AMPARs) move to a “recently silent state” (i.e. with inserted AMPARs), in which they initially cannot be further potentiated. Additional potentiation is possible only if they move to the “active state”, which occurs at around 30min after LTP induction [42, 45]. Recent results from organotypic entorhino-hippocampal slice cultures suggest that repetitive magnetic stimulation indeed may lead to an activation of silent synapses [46]. As PAS₁ and PAS₂ were identical in our study and thus presumably modified the same set of synapses, such a cascade model of synaptic plasticity states may explain why PAS₂ induced significant LTP-like plasticity at IPI₃₀, but not at IPI₁₀. In addition, the MEP increase after PAS₂ at IPI₃₀ may have been enabled by LTP-induced suppression of subsequent LTD. In rat hippocampus NMDAR-dependent LTP inhibits GSK3β, resulting in a ~60 minutes lasting blockade of subsequent NMDAR-dependent LTD induction [17]. In contrast, the suppressive interactions between PAS₁ and PAS₂ at IPI₆₀ and IPI₁₈₀ are in line with homeostatic metaplasticity [12]. Evidence in support of the notion of a delayed onset of homeostatic metaplasticity comes from animal experiments that showed that an experience-dependent increase in the NR2A/NR2B NMDAR subunit ratio, which is associated with a rightward shift of the synaptic modification threshold that favors induction of LTD over LTP [47, 48], did not occur within the first ~60min after a priming protocol [49].

An alternative explanation for the augmenting PAS₂ effect on MEPs at IPI₃₀ may have been gating [50] by the increased level of cortical excitability immediately prior to PAS₂ in this condition (time point B2, Fig 2). However, this explanation is rather unlikely, as (i) computational modelling of MEP time functions after PAS₂ showed significant differences between conditions IPI₁₀ and IPI₃₀ (cf. Fig 5), although the PAS₁-induced MEP increases immediately before PAS₂ were comparable in these two conditions (cf. Fig 2), and (ii) a control experiment (IPI_30adj), in which we readjusted MEP amplitudes after PAS₁ immediately before PAS₂ to match baseline values, found similar PAS₂ effects as in the main experiment (IPI₃₀), in which MEP amplitudes were not readjusted (Fig 4).

At IPI₃₀, PAS₂ induced significant MEP increases over and above those induced by PAS₁. The cumulative PAS₁ and PAS₂ increase of MEP amplitude at IPI₃₀ significantly exceeded the PAS₁ effect alone, and the cumulative effects of PAS₁ and PAS₂ at IPI₆₀ and IPI₁₈₀. These results show that LTP-like plasticity can be augmented in human M1, if consecutive PAS_LTP protocols are properly spaced within a window of non-homeostatic metaplasticity.

In contrast, homeostatic metaplasticity, i.e. the suppressive interaction between consecutive PAS_LTP protocols at longer intervals (IPI₆₀ and IPI₁₈₀), may prevent runaway LTP, enabling the human motor network to maintain its modifiability within a useful dynamic range. In line with this notion, dysfunctional homeostatic metaplasticity can result in a deficient control of activity-dependent plasticity, as seems to be the case in task-dependent dystonia [51, 52]. The mechanisms of homeostatic metaplasticity at IPI₆₀ and IPI₁₈₀ are as of yet unclear, but may involve Ca²⁺ influx through voltage-gated Ca²⁺ channels [53], or intracellular Ca²⁺ stores [54].

Studies in experimental animals have provided evidence, that an mGluR-dependent form of LTP is retained in the context of NMDAR-dependent increased synaptic strength [16, 55]. Thus, continued experience leads to increased synaptic strength over time, although NMDAR-dependent LTP is occluded during continued neural activity. It could be speculated that LTP-like plasticity induced by PAS₁ vs. PAS₂ at IPI₃₀ in our experiments was due to different underlying physiological mechanisms, i.e. an NMDAR- and an mGluR-dependent form of LTP-like plasticity, respectively. This idea is indirectly supported by the following observations: (i) LTP-like plasticity induced by a single PAS_LTP protocol can be blocked by an NMDAR antagonist [7]; (ii) the PAS₁-induced LTP-like plasticity in the present study was likely saturated as the MEP amplitude increase of ≥1.5 is among the highest reported in the literature [30, 33, 56]; (iii) the time-course of PAS₁- vs. PAS₂-induced MEP changes at IPI₃₀ in the present study was significantly different (cf. Fig 5). Thus, properly timed consecutive application of non-invasive brain stimulation protocols such as PAS_LTP may lead to augmentation of LTP-like plasticity through recruitment of different physiological mechanisms. Regardless of the underlying mechanisms, our results provide experimental evidence that consecutive applications of PAS_LTP can lead to significantly increased LTP-like plasticity as compared to a single PAS_LTP session.

At first sight, the present data are at variance with those from one previous study of our group where we found homeostatic metaplasticity between two consecutive PAS_LTP protocols if spaced by 30min [10]. However, the two studies differ in one important aspect: in this but not the previous study, we deliberately maximized LTP-like plasticity induced by PAS_LTP by including only those data with a PAS₁-induced MEP increase ≥1.1, and by ensuring high attention towards the stimulated hand, which is known to facilitate PAS_LTP-induced LTP-like plasticity [33]. As a result, the PAS₁–induced MEP increase in this study (time point B1, 1.49 ± 0.10, condition IPI₃₀) was significantly higher than in the previous study (1.14 ± 0.12; t = 2.24, P = 0.036; unpaired two-tailed t test). Previous work in mice showed that the expression of non-homeostatic metaplasticity critically depended on the level of the priming activity: only during strong but not weak stimuli, potentiated synapses could be further potentiated, specifically by induction of a switch in NMDAR and mGluR properties [16, 55]. Thus, the significantly stronger LTP-like plasticity induced by PAS₁ in this compared to our previous study [10] may explain why we observed augmentation of LTP-like plasticity in the present study only.

In summary, the present study demonstrated that LTP-like plasticity in the human M1 can be augmented by application of two consecutive identical PAS_LTP protocols if spaced by 30min, while homeostatic interaction occurred at intervals of 60-180min. These findings may inspire further research to optimize therapeutic applications of non-invasive brain stimulation in patients with neurological or psychiatric diseases to modify synaptic transmission in their disordered brain networks more effectively than hitherto possible.

Data Availability

All relevant data are within the paper.

Funding Statement

The authors received no specific funding for this work.