Timing-dependent plasticity in human primary somatosensory cortex


  • Alexander Wolters,

    1. Human Cortical Physiology Laboratory, Department of Neurology, University of Rostock, D-18147 Rostock, Germany
    Search for more papers by this author
  • Arne Schmidt,

    1. Human Cortical Physiology Laboratory, Department of Neurology, University of Rostock, D-18147 Rostock, Germany
    Search for more papers by this author
  • Axel Schramm,

    1. Human Cortical Physiology and Motor Control Laboratory, Department of Neurology, University of Würzburg, D-97080 Würzburg, Germany
    Search for more papers by this author
  • Daniel Zeller,

    1. Human Cortical Physiology and Motor Control Laboratory, Department of Neurology, University of Würzburg, D-97080 Würzburg, Germany
    Search for more papers by this author
  • Markus Naumann,

    1. Human Cortical Physiology and Motor Control Laboratory, Department of Neurology, University of Würzburg, D-97080 Würzburg, Germany
    Search for more papers by this author
  • Erwin Kunesch,

    1. Human Cortical Physiology Laboratory, Department of Neurology, University of Rostock, D-18147 Rostock, Germany
    Search for more papers by this author
  • Reiner Benecke,

    1. Human Cortical Physiology Laboratory, Department of Neurology, University of Rostock, D-18147 Rostock, Germany
    Search for more papers by this author
  • Karlheinz Reiners,

    1. Human Cortical Physiology and Motor Control Laboratory, Department of Neurology, University of Würzburg, D-97080 Würzburg, Germany
    Search for more papers by this author
  • Joseph Classen

    1. Human Cortical Physiology Laboratory, Department of Neurology, University of Rostock, D-18147 Rostock, Germany
    2. Human Cortical Physiology and Motor Control Laboratory, Department of Neurology, University of Würzburg, D-97080 Würzburg, Germany
    Search for more papers by this author

Corresponding author J. Classen: Human Cortical Physiology and Motor Control Laboratory, Department of Neurology, Bayerische Julius-Maximilians Universität, Josef-Schneider Str. 11, 97080 Würzburg, Germany. Email: classen_j@klinik.uni-wuerzburg.de


Animal experiments suggest that cortical sensory representations may be remodelled as a consequence of changing synaptic efficacy by timing-dependent associative neuronal activity. Here we describe a timing-based associative form of plasticity in human somatosensory cortex. Paired associative stimulation (PAS) was performed by combining repetitive median nerve stimulation with transcranial magnetic stimulation (TMS) over the contralateral postcentral region. PAS increased exclusively the amplitude of the P25 component of the median nerve-evoked somatosensory-evoked potential (MN-SSEP), which is probably generated in the superficial cortical layers of area 3b. SSEP components reflecting neuronal activity in deeper cortical layers (N20 component) or subcortical regions (P14 component) remained constant. PAS-induced enhancement of P25 amplitude displayed topographical specificity both for the recording (MN-SSEP versus tibial nerve-SSEP) and the stimulation (magnetic stimulation targeting somatosensory versus motor cortex) arrangements. Modulation of P25 amplitude was confined to a narrow range of interstimulus intervals (ISIs) between the MN pulse and the TMS pulse, and the sign of the modulation changed with ISIs differing by only 15 ms. The function describing the ISI dependence of PAS effects on somatosensory cortex resembled one previously observed in motor cortex, shifted by ∼7 ms. The findings suggest a simple model of modulation of excitability in human primary somatosensory cortex, possibly by mechanisms related to the spike-timing-dependent plasticity of neuronal synapses located in upper cortical layers.

Understanding the rules that shape human central sensory representations is of considerable physiological and clinical interest. Lasting changes of synaptic efficacy, long-term potentiation (LTP) and long-term depression (LTD), have been implicated as physiological mechanisms underlying experience- and injury-driven sensory map changes in humans and animals (Merzenich et al. 1983; Pons et al. 1991). Indeed, LTP/LTD have been documented in the somatosensory cortex of experimental animals using a variety of induction protocols both in vitro (Aroniadou-Anderjaska & Keller, 1995; Castro-Alamancos et al. 1995; Castro-Alamancos & Connors, 1996; Kitagawa et al. 1997; Feldman, 2000; Heusler et al. 2000; Urban et al. 2002) and in vivo (Keller et al. 1990; Glazewski et al. 1998; Froc et al. 2000; Allen et al. 2003; Werk & Chapman, 2003). These protocols differ substantially with respect to their efficacy and physiological properties (for review see Buonomano & Merzenich, 1998; Fox, 2002). Among the protocols, spike-timing-dependent plasticity of synaptic efficacy (STDP; Song et al. 2000) is unique in that the direction of synaptic efficacy changes is determined by the sequence of pre- and postsynaptic neuronal activity (for review see Dan & Poo, 2004). In STDP, LTP is induced if the postsynaptic neurone fires an action potential after the excitatory postsynaptic potential is induced by the presynaptic neurone. In contrast, LTD is generated if the sequence of events is reversed. STDP is largely independent of the neuronal firing rate and possesses significant theoretical advantages over models of plasticity that are solely driven by average correlations between the firing of different neurones (Sejnowski, 1999; Song et al. 2000; Song & Abbott, 2001). In humans, the plasticity of sensory representations has been studied in a variety of clinical and behavioural conditions by the analysis of somatosensory-evoked potentials (SSEPs) (Elbert et al. 1995; Flor et al. 1995, 1997; Knecht et al. 1996; Tinazzi et al. 1997b, 1998, 2004; Bara-Jimenez et al. 1998; Elbert et al. 1998). Different components of SSEPs reflect different stages of somatosensory information processing (Tinazzi et al. 1998). Therefore, the analysis of SSEPs enables the site of plastic changes along the neuroaxis to be located (Tinazzi et al. 1998). SSEPs have also been used to probe excitability changes induced experimentally by various external manipulations such as direct current stimulation (Matsunaga et al. 2004), repetitive transcranial magnetic stimulation (Enomoto et al. 2001; Tsuji & Rothwell, 2002; Ragert et al. 2004), or repetitive peripheral tactile stimulation (Pleger et al. 2001). However, it is not possible, using these protocols, to test for the presence of plasticity mechanisms that exhibit properties of STDP in human primary somatosensory cortex (S1). Recently, we have introduced paired associative stimulation (PAS) in humans as a protocol shaped after animal models of associative LTP/LTD. PAS consists of pairing repetitive peripheral electrical afferent stimulation with transcranial magnetic stimulation (TMS) of the primary motor cortex (Stefan et al. 2000, 2002; Wolters et al. 2003). In this arrangement, transcranial magnetic stimulation probably activates intracortical fibres travelling ‘horizontally’ with respect to the cortical surface (Rothwell, 1997), and peripheral electrical stimulation induces activity in cortico-petal (thalamo-cortical or cortico-cortical) ‘vertical’ fibres (Kaas & Pons, 1988). Unlike other plasticity-inducing protocols, PAS allows the control of the relative timing of neuronal events induced by the two stimulation modalities. PAS targeting the primary motor cortex (M1) induces either potentiation or depression of TMS-evoked potentials (Wolters et al. 2003) depending on the interval between the two stimulation modalities, and the physiological properties of this plasticity resemble those of STDP seen in animal studies (Stefan et al. 2000, 2002; Wolters et al. 2003).

Here, we use PAS with TMS over the somatosensory cortex to test the hypothesis that a timing-dependent plasticity rule governs the induction of bidirectional plasticity in human S1. Together, our findings provide further support for the notion that timing-dependent plasticity may represent an important principle subserving neocortical plasticity in humans.



The protocol conformed to the principles of the Declaration of Helsinki and was approved by the ethics committees of the Universities of Rostock and Würzburg. Experiments were performed on 64 healthy volunteers (31 men, 33 women), aged 20–48 years (mean 26.7 ± 6.0 years) with normal results on neurological examination. All subjects gave their written informed consent to the procedures.


Electrical nerve stimulation was performed with a Grass stimulator (Type S88, Grass Instruments, West Warwick, Richmond, VA, USA) connected to a stimulus isolation unit (SIU 8T, Grass Instruments) or with an electrical stimulator (Digitimer D7AH, Digitimer, Welwyn Garden City, UK) with a standard stimulation block (cathode proximal). Transcranial magnetic stimulation (TMS) was performed using a flat figure-of-eight-shaped magnetic coil (diameter of each wing: 70 mm) connected to a Magstim 200 magnetic stimulator (Magstim, Whitland, Dyfed, UK). The coil was held tangentially to the skull with the handle pointing backwards and laterally at an angle of 45 deg to the sagittal plane.


Electroencephalographic (EEG) signals were recorded from the scalp using needle electrodes (model 9013R0312, Medtronic, Skovlunde, Denmark) or Hwato acupuncture needles (Suzhou Medical Appliance Factory, Suzhou, China) in a bipolar montage with the reference electrode placed frontocentrally at Fz according to the international 10/20 system (Klem et al. 1999). To ensure that the same EEG recording positions were used before and after PAS, electrodes remained at the recording sites during the interventional stimulation. During PAS, EEG recording electrodes were disconnected from the amplifier. Although the pressure of the coil on the underlying electrode caused slight discomfort this was well tolerated by all subjects. Surface electromyographic (EMG) activity was recorded from the right abductor pollicis brevis muscle (APB) using disposable Ag–AgCl surface electrodes (Neuroline model 725 01-SC, Medicotest, Olstykke, Denmark, or reusable Ag–AgCl electrodes (Fischer Medizintechnik, Nürnberg, Germany) with the active electrode mounted on the muscle belly and the inactive electrode placed over the base of the metacarpo-phalangeal joint of the thumb. EEG and EMG signals were amplified (model BF/IEC 601-1, Jaeger-Toennies, Freiburg, Germany, or 1902 amplifier, Cambridge Electronics Design, Cambridge, UK). EEG signals were bandpass filtered between 0.2 and 1500 Hz, and EMG signals between 5 and 2000 Hz. Data were sampled at 5000 Hz using an A/D converter (model 1401 plus, Cambridge Electronics Design) and stored in a laboratory computer for display and later offline analysis.

Experimental procedures

The principal purpose of all experiments was to determine the effect of a paired associative stimulation protocol on SSEPs recorded from the scalp positions overlying S1.

Subjects were seated in a comfortable reclining chair. The optimal position of the magnetic coil for eliciting motor-evoked potentials (MEPs) in the right APB was assessed over the left motor cortex at a moderately suprathreshold stimulation intensity, and marked directly on the scalp with a soft-tip pen. At the optimal site, termed ‘motor hot spot’, the resting motor threshold (RMT) was determined as the stimulator intensity needed to produce a response of at least 50 μV in the relaxed APB in at least 5 of 10 consecutive trials at a resolution of 1% of the maximal stimulator output (Rossini et al. 1994). Throughout the experiment, complete muscle relaxation was monitored by audio-visual feedback.

Somatosensory-evoked potentials (SSEPs) For MN-SSEPs, EEG signals were recorded in a bipolar montage with the active electrode placed at C3′ and the reference electrode at Fz according to the international 10/20 system (Mauguiere et al. 1999). The C3′ position was determined 2 cm posterior to the C3 position (Nuwer et al. 1994). Median nerve stimulation was performed using a pulse width of 300 μs (Grass stimulator) or 200 μs (Digitimer D7AH) at a frequency of 3 Hz and a stimulation intensity of 300% of the individual perceptual threshold. For each sweep the recorded time interval extended from 30 ms before to 70 ms (MN-SSEP) or 120 ms (TN-SSEP), respectively, after the stimulus artifact. An SSEP represented the average of 250 responses. For TN-SSEP, the active electrode was placed over Cz′ and the reference electrode over Fz (Mauguiere et al. 1999). In all experiments, four SSEPs were obtained each before, and immediately after the intervention (see below). Sweeps were rejected automatically online if the signal following the stimulus artefact exceeded a limit of ± 20 μV. Furthermore, individual sweeps containing artifactual signals were eliminated after offline visual analysis.

Paired associative stimulation (PAS) The PAS intervention represented a modification of a protocol published previously by our group (Stefan et al. 2000; Wolters et al. 2003). In a typical experiment, repetitive single electrical stimuli were delivered to the right median nerve at the level of the wrist at 300% of the perceptual threshold (10.4 ± 1.7 mA), each followed by TMS delivered over a position 2 cm posterior to the ‘motor hot spot’. Previous results (Okamoto et al. 2004) as well as unpublished observations from our laboratory (A. Schramm & D. Zeller) using a neuronavigation system revealed that this position overlies the postcentral gyrus. This position was found to correspond closely to C3′. TMS was applied at an intensity of 1.5 times the resting motor threshold (SI1.5RMT). Taking all experiments into consideration, SI1.5RMT amounted to 51 ± 9% of the maximal stimulator output. The interval between MN stimulation and the subsequent TMS pulse was set at the individual N20 latency. N20 latency was defined as the mean latency of the N20 component of all four baseline SSEPs. One hundred and eighty pairs were delivered at 0.1 Hz over 30 min.

After PAS, four additional MN-SSEPs were obtained. Identical stimulation parameters were used before and after intervention. The principal experimental design shared by all experiments is illustrated in Fig. 1. Variations of the experimental standard protocol are described below.

Figure 1.

Principles of experimental design
Test somatosensory-evoked potentials (SSEPs) were elicited by right median nerve stimulation (MN-SSEPs) before and after the intervention. During paired associative stimulation 180 pairs were delivered, each consisting of electrical stimulation of the right median nerve followed by TMS delivered through a magnetic coil placed over the left hemisphere 2 cm posterior to the optimal site for activating the abductor pollicis brevis (APB) muscle. In the principal experiment, the interstimulus interval was set at the N20 latency of the MN-SSEP as determined before paired associative stimulation (PAS). An inter-pair interval of 10 s was used (0.1 Hz). Rec., recording site of SSEP. Open arrowhead, active electrode; filled arrowhead, reference electrode. Stim., peripheral nerve stimulation or transcranial magnetic cortex stimulation.

Care was taken that subjects maintained a steady level of attention to the task during the SSEP recordings and during the PAS intervention.

Duration and reversibility

In 10 subjects modulation of MN-SSEPs following the intervention (PASN20) was monitored over time. Four MN-SSEPs were obtained each immediately (0 min), and at 30, 60 and 90 min after PAS.

Topographical specificity of PASN20 effects

Comparison of PASN20 effects on MN-SSEP and TN-SSEP In nine subjects, the effect of PASN20 on MN-SSEP was compared with that on the SSEP elicited by stimulation of the right tibial nerve (TN-SSEP) in the same experimental session. The sequence of MN-SSEP and TN-SSEP prior to and post intervention was counterbalanced throughout this experimental condition.

Effect of PAS employing ulnar nerve stimulation on ulnar nerve-evoked somatosensory potentials (UN-SSEP) In a separate series on six subjects, PASN20 was performed by afferent stimulation of the ulnar nerve and delivering TMS at a skull site located 2 cm posterior to the ‘motor hot spot’ of the abductor digiti minimi muscle.

Effect of varying the site of magnetic stimulation In eight subjects the effect of PASN20 intervention performed with the magnetic coil positioned over the postcentral gyrus was compared with PASN20 intervention performed with the magnetic coil positioned over the ‘motor hot spot’ of the APB muscle which corresponds to M1 (Classen et al. 1998). The sequence of these two experimental conditions was counterbalanced and sessions were separated by at least 2 days.

Effect of varying the interval between median nerve stimulation and TMS

The timing of the TMS pulse with reference to the median nerve stimulation was varied. Ten different ISIs were tested in a total of 104 separate experimental sessions on 62 different subjects. ISI was set at N20 latency +x with x as −40, −30, −20, −10, −5, −2.5, 0, +5, +10, +20 ms. These interventions were termed PASN20-40ms, PASN20-30ms, … PASN20+20ms, respectively. A minimum number of five experiments were performed for each ISI. In cases where subjects participated in more than one experiment, at least 2 days elapsed between any two sessions.

Data analysis

SSEPs were analysed offline. Baseline was determined as the average value over a period of 30 ms immediately before the stimulus (Regan, 1989). In several initial experiments blinded assessment of baseline was shown to be virtually identical with non-blinded assessment. The N20 amplitude was assessed as the difference between the baseline and the first negative peak occurring at a latency of around 17–21 ms after the time of median nerve stimulation, and the amplitude of the N20–P25 complex was determined as the difference between the N20 peak and the subsequent positivity. Subsequently, P25 amplitude was calculated as the difference between the amplitude of the N20–P25 complex and the N20 amplitude. In some experiments the P14 amplitude was additionally assessed from baseline to peak. The P14 component could not be reliably identified in all subjects. Therefore, all PASN20 experiments were randomly screened for the presence of a reliable P14 component until 15 consecutive experiments displaying a P14 component were identified. The latency of the N20 component in MN-SSEP was determined at the maximum negativity. The P40 component of the TN-SSEP is thought to be equivalent to the P25 component of the MN-SSEP (Yamada et al. 1996; Yamada, 2000). The P40 peak was identified at latencies of 34–46 ms and its amplitude was determined as the difference from baseline.

For each subject, the values of each parameter were averaged across the four SSEP repetitions and the means were entered into the final statistical analyses.

For display purposes, averages of four SSEPs (‘grand average’) (sets of 4 SSEPs collected at individual time points before and after intervention) were generated for individual subjects. Similarly, averages of grand averages (‘great grand averages’) were computed to illustrate group results. Great grand averages were obtained after prior alignment of grand averages to the individual N20 latency.

Data were analysed using Student's t tests and analyses of variance (ANOVA). In general, two-tailed t tests were employed. If an a priori hypothesis on the sign of the tested difference could be made, one-tailed t tests were used. Details of the ANOVA factors tested are given in Results. Effects were considered significant if P < 0.05.


Modelling of the function relating the PAS-induced magnitude of P25 amplitude change to the interval between the stimulation modalities was done using commercially available software (DataFit program, Version 8.0, Oakdale Engineering, Oakdale, PA, USA).

The following function was found empirically

display math(1)

Initial parameter estimates were determined graphically. Variance analysis was used to test the overall significance of the regression model. The null hypothesis H0 (a=b=c=d= 0) was tested against the hypothesis H1 ((a≠ 0) or (b≠ 0) or (c≠ 0) or (d≠ 0): at least one parameter is different from 0). In a variation of the regression model (‘model 2’, parameter ‘a’ set to a= 1), the null hypothesis H0 (b=c=d= 0) was tested against H1 ((b≠ 0) or (c≠ 0) or (d≠ 0)).

If not stated otherwise, all group data are given as mean ±s.d.


Principal experiment

TMS over the somatosensory cortex was timed to coincide with the individual N20 latency of the MN stimulation-evoked potential. Following PASN20, MN-SSEP was changed (Fig. 2A). All experimental sessions considered (a total of 40 experimental sessions on 35 volunteers), PASN20 led to an increase of the N20–P25 amplitude from a mean of 5.6 ± 2.9 μV to 6.0 ± 3.1 μV (P= 0.001, paired two-tailed t test; Fig. 2B), or, on average, by 6.2%. Inspection of the raw data revealed that MN-SSEP tracings started to separate from the tracings recorded before intervention at about 3 ms after the N20 peak. This observation suggested a differential effect of PASN20 on the N20 and P25 components. Therefore, the amplitude of these components was determined separately. Following PASN20, the amplitude of the N20 peak remained essentially constant (pre-PAS 2.1 ± 1.3 μV to 2.0 ± 1.1 μV, P= 0.270, paired two-tailed t test; Fig. 2B), the change amounting to, on average, –0.6%. In contrast, the amplitude of the P25 component increased from a mean of 3.5 ± 2.0 μV to 3.9 ± 2.2 μV (P < 0.001, paired two-tailed t test; Fig. 2B), or, on average, by 11.5%.

Figure 2.

Effect of PASN20 on MN-SSEP
A, example trace of one subject. Average of 4 MN-SSEPs recorded before (grey dotted trace) and after (black trace) intervention with PASN20 (paired associative stimulation with an interval of N20 latency). B, group data of 40 experiments on 35 subjects. Percentage change (relative to baseline) of different components of the N20–P25 complex. C, comparison of baseline-normalized P25 amplitude (left column) and P14 amplitude (right column). Data from 15 subjects. D, comparison of baseline-normalized P25 amplitude (left column) and sensory nerve action potential (right column). Data from 10 subjects. B–D, data show mean ±s.d. Asterisks indicate statistical significance.

To investigate the location of PAS-induced changes, the amplitude of the subcortically generated P14 component was analysed in 15 experiments. Following PASN20, the baseline-normalized amplitudes of the P14 and P25 differed significantly (P < 0.001, paired one-sided t test; Fig. 2C). P25 amplitude increased from a mean of 3.6 ± 2.0 μV pre-PAS to 4.0 ± 2.1 μV post-PAS (P= 0.001), or, on average, by 14.1%. In contrast, the amplitude of the P14 component did not change significantly (pre-PAS: 0.4 ± 0.1 μV; post-PAS: 0.3 ± 0.1 μV; P= 0.107; mean percentage change −4.8%).

To exclude the possibility that the increase of P25 amplitude was related to a more efficient peripheral nerve stimulation following PASN20, median nerve sensory nerve action potentials (SNAPs) were recorded simultaneously with the recording of MN-SSEPs in 10 subjects. Following PASN20, the baseline-normalized amplitudes of the P25 and SNAP differed significantly (P= 0.035, paired one-tailed t test; Fig. 2D). P25 amplitude increased from a mean of 4.6 ± 1.9 μV pre-PAS to 5.2 ± 2.1 μV post-PAS (P= 0.001), or, on average, by 14.1%. By contrast, the SNAP amplitude remained essentially constant (pre-PAS: 19.0 ± 10.5 μV; post-PAS: 19.9 ± 13.1 μV; P= 0.504; mean percentage increase, 1.4%).

Duration and reversibility

The magnitude of the P25 amplitude was monitored for 90 min. The PASN20-induced increase of the mean P25 amplitude lasted at least 30 min. P25 amplitude returned to baseline at 90 min. Repeated measures ANOVA (‘Time’ (baseline, 0, 30, 60, 90 min)) revealed a significant effect of time (F4,36= 2.837; P= 0.038). Pre-planned contrasts revealed significant differences between P25 amplitude at baseline on the one hand and at 0 min (P= 0.006) and 30 min (P= 0.020) on the other (paired one-sided t test). P25 amplitude was not significantly different from baseline at 60 and 90 min (Fig. 3).

Figure 3.

Lasting effect of PASN20-induced increase of P25 amplitude
A, example trace of one subject. Averages of 4 consecutive MN-SSEPs from the same subject, recorded before PASN20 (grey), and at different times after PASN20 (black). Horizontal line indicates P25 amplitude at baseline. B, group data of 10 subjects. Asterisks indicate times when P25 amplitude was significantly (P < 0.05) different from baseline. Data show mean ±s.d.

Topographical specificity of PASN20 effects

To examine the spatial specificity of the effect induced by combined stimulation of MN and TMS over the hand representation in S1, the change of MN-SSEP was compared to that of TN-SSEP which represents the cortical leg representation located at a distance of several centimetres from the hand representation. PASN20 led to a differential effect on MN-SSEP and TN-SSEP (Fig. 4A). The baseline-normalized amplitudes of P25 and P40 were significantly different (P= 0.044, paired two-sided t test). P25 amplitude increased from a mean of 3.9 ± 2.1 μV pre-PAS to 4.2 ± 2.2 μV post-PAS (P= 0.007, paired one-sided t test), while the P40 amplitude did not change significantly (pre-PAS: 2.7 ± 0.8 μV; post-PAS: 2.5 ± 0.9 μV; P= 0.221). This finding suggests that PASN20-induced excitability change did not spill over to a remote representation that did not receive spatially homologous information.

Figure 4.

Topographical specificity of PASN20 effects
A, comparison of effects of PASN20 on MN-SSEP and TN-SSEP. PASN20 led to an increase of P25 amplitude (left column) of MN-SSEP while it did not significantly change the amplitude of the P40 amplitude (right column) of TN-SSEP. Data from 9 subjects (mean ±s.d.). B, effect of varying the site of magnetic stimulation. PASN20 intervention was done with the magnetic coil placed over S1 (left column) or M1 (right column). Data from 8 subjects (mean ±s.d.). Asterisks indicate statistical significance.

To examine whether PAS was able to induce a change in cortical representations that differed from MN-associated regions, PAS was performed with the ulnar nerve as the afferent stimulation route, and the magnetic coil placed over a position located 2 cm posterior to the ‘motor hot spot’ of the abductor digiti minimi muscle. Following this intervention, the P25 component of the UN-SSEP increased from a mean of 2.6 ± 0.9 μV to 2.9 ± 0.8 μV (P= 0.012), or, on average, by 19.8% (data not illustrated).

To test the possibility that P25 amplitude changes may have been the result of activation of neuronal elements in M1 by current spreading from the site of magnetic stimulation, paired stimulation involving TMS over S1 was compared to paired stimulation involving TMS over M1. Repeated measures ANOVA (2 × 2, with factors ‘Stimulation site’ (M1, S1) and ‘Time’ (pre-PAS, post-PAS)) revealed a significant ‘Stimulation site’בTime’ interaction (F1,7= 26.636; P= 0.001) suggesting that the effect of PASN20 on MN-SSEP depended on the magnetic coil position during the intervention (Fig. 4B). With the magnetic coil positioned over S1, P25 amplitude increased from a mean of 4.2 ± 1.5 μV pre-PAS to 5.0 ± 1.9 μV post-PAS (P= 0.005), or, on average, by 18.6%. In contrast, following PAS with the magnetic coil positioned over M1, the P25 amplitude remained constant (pre-PAS: 4.3 ± 1.5 μV; post-PAS: 4.3 ± 1.6 μV; P= 0.969; mean percentage change −0.3%).

Effect of varying the interval between median nerve stimulation and TMS

To test the hypothesis that MN stimulation paired with TMS at appropriate intervals would lead to bidirectional changes in S1, the interstimulus interval between the median nerve stimulation and the magnetic stimulation over S1 was varied over a wide range of intervals. Repeated measures ANOVA (10 × 2, between-subject factor ‘ISI’ (−40, … 20), within-subject factor ‘Time’ (pre-PAS, post-PAS)) revealed a significant interaction term ‘ISI’בTime’ (F9,94= 5.109; P < 0.001; Fig. 5) suggesting that P25 amplitude change depended on ISI. Post hoc testing revealed that P25 amplitude increased with PASN20-5ms (P= 0.036, paired two-tailed t test), PASN20-2.5ms (P= 0.023) and PASN20 (P < 0.001). P25 amplitude decreased with PASN20-20ms (P= 0.007).

Figure 5.

Dependence of PAS-induced effects on the interval between afferent peripheral nerve stimulation and transcranial magnetic stimulation
A, ‘great grand’ averages (see Methods) for tracings obtained at ISI = N20 latency − 20 ms (PASN20-20ms, left) and ISI = N20 latency (PASN20, right) (grey dotted trace, before intervention; black trace, after intervention). Great grand averages were generated by aligning SSEPs to the peak of the N20 component. B, for each interval 5–35 subjects were tested. Abscissa: intervals relative to the individual N20 latency. Asterisks indicate that changes of P25 amplitude are significantly different from the baseline condition. Number of experiments for each interval shown in top row. Data are mean ± standard error of the mean.

To estimate the timing of the equilibrium between facilitating and depressing effects of PAS on P25 amplitude, data were modelled by fitting parameters of eqn (1) which was found empirically. Initial parameter estimates were determined graphically. The results of two different models with four (model 1) or three (parameter ‘a’ set to a= 1, model 2) regression variables are displayed in Table 1. Equation (1) generated a fit of the data explaining 91.6% (model 1) or 86.4% (model 2) of the variance. For both models, the time of equilibrium between facilitating and depressing effects of PAS was computed to be N20 latency – 9.2 ms.

In a previous study PAS was used to target M1. The magnitude of PAS-induced changes of the motor-evoked potentials recorded from the abductor pollicis brevis muscle varied as a function of the interval between median nerve stimulation and magnetic stimulation of M1 (Wolters et al. 2003). To compare the ISI dependence of PAS targeting M1 (Wolters et al. 2003) with those observed in the present study, all ISIs from the previous study were recalibrated to the mean N20 latency of 18.8 ms found in a representative subgroup of subjects (Wolters et al. 2003). Data were modelled using eqn (1). The regression variables from the current experiments were used as initial parameter estimates. Equation (1) generated a fit of the data explaining 86.3% (model 1) or 74.3% (model 2) of the variance (Table 1). Using these regression variables the time of equilibrium of facilitating and depressing effects of PAS targeting M1 was computed to be N20 latency – 2.4 ms (model 1) or N20 latency – 2.7 ms (model 2). Therefore, the difference between the times of equilibrium with PAS targeting S1 versus PAS targeting M1 was 6.8 ms (model 1, Fig. 6) or 6.5 ms (model 2).

Figure 6.

Comparison of ISI dependence of PAS-induced excitability changes in S1 and M1
Fitted data were scaled to their relative extremes. •, data from the current experimental series. □, data from Wolters et al. (2003). S1 curve crosses the level of equilibrium at an ISI shorter by 6.8 ms than the M1 curve.


The present results have shown that peripheral stimulation of somatosensory afferents, if paired with low-frequency TMS at appropriate intervals, may induce bidirectional changes in somatosensory-evoked potentials.

Regional and laminar location of PAS-induced SSEP changes

Representational plasticity may occur at each anatomical level of the central somatosensory system (Florence & Kaas, 1995; Nicolelis et al. 1998; Tinazzi et al. 1998; Jones, 2000; Chung et al. 2002). The P14 component of MN-SSEP is known to be generated subcortically, possibly in the terminal part of the ascending lemniscal system at its arrival in the ventroposterior lateral nucleus of the thalamus (Desmedt & Cheron, 1981; Moller et al. 1986; Sonoo et al. 1997; Lee & Seyal, 1998). Following PASN20, the amplitude of the P14 component remained unchanged (Fig. 2). Therefore, PAS did not induce SSEP changes arising at a level below the thalamus.

Of the two early cortical SSEP components evaluated, changes were induced in the P25 component, while the amplitude of the N20 component remained constant (Fig. 2B). The generators of the early components of the MN-SSEP have been extensively investigated over the past decades (for review see Allison et al. 1991) with respect to their regional and laminar origin. Both the N20 and the P25 components are now widely accepted to be generated in the posterior bank of the central sulcus, corresponding to Brodmann area 3b (Allison et al. 1989, 1991; McLaughlin & Kelly, 1993; Urbano et al. 1997; Lee & Seyal, 1998; Mauguiere et al. 1999; Legatt & Kader, 2000; Balzamo et al. 2004). Therefore, changes in the P25 component probably indicate excitability changes in S1, although it is possible that a minor contribution to the P25 component arises from the anterior bank of the central sulcus in Brodmann area 4 (Huang et al. 2000; Balzamo et al. 2004). The issue of an additional contribution to changes in the P25 component by an anterior source in area 4 or by an additional radial source residing in area 1 may be clarified further only by applying multi-dipole localization algorithms to multichannel recordings (Huang et al. 2000).

The N20 component of the MN-SSEP is believed to reflect the passive source current for active depolarizing sinks on the cell bodies and proximal apical dendrites of pyramidal cells in layer 4, the input layer of the cortex (Allison et al. 1991). Conversely, the P25 component probably reflects the depolarization of the superficial portion of apical dendrites located in cortical layers 2/3 (Mitzdorf, 1985; Vaughan & Arezzo, 1988; Allison et al. 1991; McLaughlin & Kelly, 1993; Nicholson Peterson et al. 1995). This concept is underlined by electrical recordings in animals along the depth of the somatosensory cortex indicating that information is serially relayed from layer 4 to layer 2/3 and that this takes 2–4 ms (Armstrong-James et al. 1992; Ahissar et al. 2001). This consideration may suggest that PAS-induced cortical excitability changes were due to modulation of excitatory neuronal activity in upper cortical layers of S1. Alternatively, changes of cortical excitability could be due to alterations in (tonic) inhibition impinging on cortical pyramidal cells. As some of this inhibition is under subcortical thalamic control, it is important to consider the possibility that PAS-induced excitability changes may in fact have been generated subcortically, rather than locally within the cortex. Virtually all thalamo-cortical projections to area 3b terminate in layer 4 and lower layer 3 (Jones, 1986). Therefore, given the serial nature of information transmission from layer 4 to layer 2/3, both N20 and P25 components should be similarly affected by changes in thalamo-cortical activity. Hence, the differential modulation of the N20 component and the P25 component renders a subcortical origin of cortical excitability changes unlikely.

Our observation concurs with previous findings that combined peripheral and cortical stimulation may lead to facilitation of subsequently recorded MN-SSEPs (Tsuji & Rothwell, 2002). Just as in the present study, SSEP changes were absent for the subcortical P14 component. Interestingly, inspection of Fig. 2 of the same study suggests selective modulation of P25 amplitude (Tsuji & Rothwell, 2002). These authors used a pairing protocol in which the magnetic coil was placed over M1, a position which was found ineffective (at ISI = N20 latency) in the present study (cf. Fig. 4B). One may hypothesize that in their study SSEP changes have been induced in M1 and transmitted to S1 via changes in tonic afferent activity. Alternatively, and more likely, SSEP changes may have been induced in S1 through TMS stimulus spread towards S1. Tsuji & Rothwell (2002) used repetitive motor-point stimulation of the first interosseus muscle, delivered as a pulse train of 500 ms duration, as afferent stimulus. Although only the first of the pulses actually preceded TMS delivered after 25 ms, a train of 10 additional afferent stimuli, following within 500 ms, may have boosted a weak effect that would have been induced in S1 by the pairing of the first pulse with local TMS-evoked effects. Alternatively, comparatively greater activation of muscle spindle afferents may have played a role. More studies are needed to explain this difference between our results and those of Tsuji & Rothwell.

Physiological nature of PAS-induced SSEP changes

The evidence reviewed so far argues for local PAS-induced changes in upper cortical layers of Brodmann area 3b. Possible cellular mechanisms may be inferred by additional physiological properties of these changes. PAS-induced plasticity of the P25 component evolved rapidly (after an intervention of only 30 min), persisted for a considerable length of time (at least 30 min), and was reversible. Topographical specificity of the induced changes was suggested by the fact that MN-SSEPs were altered while TN-SSEPs remained unchanged (although the P40 amplitude in TN-SSEP has been shown to be modifiable by other manipulations; Tinazzi et al. 1997a). Further, PAS-induced plasticity was distinctly timing dependent. Effects occurred within a narrow window of ISIs. Finally, slightly changing the interstimulus interval between the afferent pulse and the magnetic cortical pulse (from N20 latency – 20 ms to N20 latency − 5 ms) caused the P25 amplitude to change in the opposite direction. This surprising dependence on the timing of PAS-induced plasticity in S1 resembles STDP in animal somatosensory cortex. In a study on cortical slices taken from rat barrel cortex, single excitatory postsynaptic potentials (EPSPs) were paired with single postsynaptic action potentials evoked by current injection through the recording electrode (Feldman, 2000). LTP was observed consistently when the EPSP led the postsynaptic action potential by short (3–15 ms) intervals. In contrast, LTD occurred when the action potential led the EPSP by 0–50 ms. Just 40 pairings were sufficient to induce robust LTP (Feldman, 2000). Together, these physiological similarities may indicate that modulations of synaptic efficacy by mechanisms resembling STDP in animal studies also underlie PAS-induced excitability changes. Interestingly, while LTP/LTD was induced at vertical inputs to layer 2/3 pyramidal cells, layer 4 neurones did not express synaptic plasticity (Feldman, 2000). Therefore, PAS-induced plasticity shares with spike-timing-dependent LTP/LTD in S1 not only its timing properties but possibly also its laminar location in upper cortical layers. Our observations concur with several studies in experimental animals suggesting that synapses in upper cortical layers may have a special role in rapidly induced sensory map plasticity (Diamond et al. 1993, 1994; Glazewski & Fox, 1996; Huang et al. 1998) while modulation of synaptic plasticity in layer 4, as a rule, is largely restricted to an early developmental period (Fox, 1992) and does not depend on the firing order of pre- and postsynaptic spikes (Egger et al. 1999). While the physiological properties of PAS-induced plasticity are suggestive of a synaptic origin, timing-dependent changes of intrinsic neuronal excitability (Daoudal & Debanne, 2003; Zhang & Linden, 2003; Li et al. 2004) may represent pre- or postsynaptic mechanisms involved synergistically in generating somatosensory plasticity.

Afferent signals may interact with late TMS-induced cortical neuronal events

For STDP to be operative the postsynaptic neuronal events must follow presynaptic events to induce enhancement of synaptic efficacy, and the sequence of neuronal events must be reversed for depression. The shortest interval leading to a significant enhancement of P25 amplitude was 5 ms shorter than the individual N20 latency. At this ISI the afferent information had not yet reached the somatosensory cortex. Possibly TMS-induced (postsynaptic) neuronal activity could follow that induced by afferent MN stimulation-induced (presynaptic) activity if a Hebbian interaction between the two events took place subcortically, e.g. in somatosensory thalamus via TMS-activated (Bestmann et al. 2004) cortico-thalamic projections. However, as outlined above, this possibility would appear to be inconsistent with the conclusion (based on the differential modulation of N20 amplitude and P25 amplitude) that PAS-induced facilitation of the P25 component had been generated locally, within the cortex.

Indeed, even at ISI = N20 latency − 5 ms, a pre- > postsynaptic sequence of neuronal events would be present in somatosensory cortex if late polysynaptic rather than early direct or monosynaptic TMS-induced activity interacts with presynaptic MN stimulation-induced events. This speculation appears to be well supported by current models of how TMS may activate cortical elements (Amassian et al. 1987; Ziemann & Rothwell, 2000). Following a single pulse of TMS to M1, the cortex emits a train of descending action potentials (termed D-waves and I-waves) that may last longer than 10 ms (Di Lazzaro et al. 2004). While the early components of this train are thought to be generated in lower cortical layers, the later I-waves probably reflect activity generated in upper motor cortical layers through a chain of interneurones (Amassian et al. 1987; Ziemann & Rothwell, 2000). We propose that TMS may induce late polysynaptic activity in upper layers of S1 through a similar mechanism to that in M1 and that it is these late events which interact with afferent signals in upper cortical layers of S1. This hypothesis also offers an explanation for the observation, in M1, that PAS effectively increased MEP amplitudes when ISIs between MN stimulation and TMS as short as 20 ms were employed (Ziemann et al. 2004). At this interval it is conceivable that late, but not early, TMS-induced events follow those induced by afferent activity in M1. This consideration predicts that late, rather than early, I-waves are modulated by PAS targeting M1.

S1 leads M1 in timing-dependent plasticity

The function describing the dependence of modulation of the P25 amplitude on the exact interstimulus interval between the MN pulse and the TMS pulse closely resembled the function describing the previously established ISI dependence of modulation of the MEP amplitude (Wolters et al. 2003). At the ISI indicating equilibrium between enhancing and depressing effects the two curves were shifted by ∼6.8 ms. This observation concurs with studies, both in humans and in non-human primates, suggesting that a somatosensory signal generally arrives in primary motor cortex several milliseconds later than the appearance of the signal in S1 (Goldring et al. 1970; Balzamo et al. 2004; Gow et al. 2004). As there is no direct anatomical connection from area 3b to area 4 (Jones, 1986; Darian-Smith et al. 1993) a somatosensory signal from S1 destined for M1 must first be relayed in areas 1 and 2 (Jones, 1986). Although it appears that 6.8 ms may be just sufficient for the travel time along this route, we cannot dismiss the alternative possibility of an afferent pathway to M1 via its private direct thalamic input (Lemon & van der Burg, 1979; Asanuma et al. 1980).

Bidirectional timing-dependent plasticity in human S1

In humans, experimental or disease-induced deafferentation (Tinazzi et al. 1997b, 1998, 2003) leads to changes in cortical somatosensory representations as assessable by evoked potentials. It may be hypothesized that Hebbian mechanisms such as LTP/LTD may be operative in some of these changes. However, it remains unknown whether STDP of synaptic efficacy is involved. First, it is unclear whether peripheral deafferentation can change the firing behaviour of neurones such that LTP or LTD formation would be promoted. While this question cannot be addressed directly in humans, recent experiments in animals provide strong arguments. In rats, whisker clipping, a model of partial deafferentation, was shown to change the firing order of neurones across different cortical layers (Celikel et al. 2004). This study therefore linked deafferentation to neuronal firing patterns that drive spike-timing-dependent LTP/LTD. Secondly, while spike-timing-dependent LTP/LTD has previously been demonstrated in vitro, in rats (Feldman, 2000) it was hitherto unknown whether human S1 holds in vivo a mechanism for timing-dependent bidirectional modulation of local excitability. Our study, demonstrating STDP-like plasticity in S1, appears to close this gap of information. We propose that it is partly through this mechanism that tactile deafferentation gives rise to lasting changes in human cortical somatosensory representations. Analogous considerations may apply for S1 plasticity induced by tactile coactivation in humans (Pleger et al. 2001, 2003), and whisker pairing in rats (Diamond et al. 1993). While the available evidence is consistent with involvement of spike-timing-dependent LTP/LTD in naturally occurring somatosensory cortex plasticity, future experiments must address the important question of whether timing-dependent bidirectional modulation of synaptic efficacy in S1 is sufficient to produce bidirectional behavioural changes. Such a relationship was recently suggested in the visual cortex. Pairings of two differently orientated near-synchronous visual stimuli led to changes of orientation tuning whose direction depended on the temporal order of presentation (Yao & Dan, 2001; Fu et al. 2002). Therefore, plasticity driven by timing relationships may represent a general principle of mnemonic representation in neocortex with important behavioural implications. This plasticity may be accessible in vivo in different human brain regions ranging from visual (Yao & Dan, 2001; Fu et al. 2002) and somatosensory cortex (this study) to motor cortex (Wolters et al. 2003) by simple non-invasive stimulation protocols.

In summary, our findings demonstrate that associative stimulation may induce bidirectional excitability changes in human somatosensory cortex. These changes are possibly located in superficial cortical layers and resemble spike-timing-dependent long-term potentiation or depression of neuronal synapses.



We acknowledge support from the Dystonia-Foundation, USA, and the State of Bavaria, Germany. We thank V. Litvak and S. Meuth for carefully reading the manuscript.