OUP user menu

The cerebral oscillatory network of parkinsonian resting tremor

Lars Timmermann, Joachim Gross, Martin Dirks, Jens Volkmann, Hans‐Joachim Freund, Alfons Schnitzler
DOI: http://dx.doi.org/10.1093/brain/awg022 199-212 First published online: 1 January 2002


Data from experiments in MPTP monkeys as well as from invasive and non‐invasive recordings in patients with Parkinson’s disease suggest an abnormal synchronization of neuronal activity in the generation of resting tremor in Parkinson’s disease. In six patients with tremor‐dominant idiopathic Parkinson’s disease, we recorded simultaneously surface electromyograms (EMGs) of hand muscles, and brain activity with a whole‐head magnetoencephalography (MEG) system. Using a recently developed analysis tool (Dynamic Imaging of Coherent Sources; DICS), we determined cerebro‐muscular and cerebro‐cerebral coherence as well as the partial coherence between cerebral areas and muscle, and localized coherent sources within the individual MRI scans. The phase lag between the EMG and cerebral activity was determined by means of a Hilbert transform of both signals. After overnight withdrawal from medication, patients showed typical Parkinson’s disease resting tremor (4–6 Hz). This tremor was associated with strong coherence between the EMG of forearm muscles and activity in the contralateral primary motor cortex (M1) at tremor frequency but also at double tremor frequency. Phase lags between M1 activity and EMG were between 15 and 25 ms (M1 activity leading) at single, but also at double tremor frequency, corresponding well to the corticomuscular conduction time. Furthermore, significant coherence was observed between M1 and medial wall areas (cingulate/supplementary motor area; CMA/SMA), lateral premotor cortex (PM), diencephalon, secondary somatosensory cortex (SII), posterior parietal cortex (PPC) and the contralateral cerebellum at single tremor and, even stronger at double tremor frequency. Spectra of coherence between thalamic activity and cerebellum as well as several brain areas revealed additional broad peaks around 20 Hz. Power spectral analysis of activity in all central areas indicated the strongest frequency components at double tremor frequency. Partial coherence analysis and the calculation of phase shifts revealed a strong bidirectional coupling between the EMG and diencephalic activity and a direct afferent coupling between the EMG and SII and the PPC. In contrast, the cerebellum, SMA/CMA and PM show little evidence for direct coupling with the peripheral EMG but seem to be connected with the periphery via other cerebral areas (e.g. M1). In summary, our results demonstrate tremor‐related oscillatory activity within a cerebral network, with abnormal coupling in a cerebello‐diencephalic–cortical loop and cortical motor (M1, SMA/CMA, PM) and sensory (SII, PPC) areas contralateral to the tremor hand. The main frequency of cerebro‐cerebral coupling corresponds to double the tremor frequency.

  • Keywords: Parkinson’s disease; MEG; coherence, tremor; DICS
  • Abbreviations: CMA = cingulate motor area; DICS = Dynamic Imaging of Coherent Sources; EDC = extensor digitorum communis; EMG = electromyography; GPI = internal globus pallidus; M1 = primary motor cortex; MEG = magnetoencephalography; PM = premotor cortex; PPC = posterior parietal cortex; SII = secondary somatosensory cortex; SMA = supplementary motor areas; STN = subthalamic nucleus


The current pathophysiological concept of Parkinson’s disease postulates alterations of the interactions within the basal ganglia complex due to the loss of dopaminergic projections from the substantia nigra to the striatum (DeLong, 1990; Bergman et al., 1998; Obeso et al., 2000). According to this model, pathological hyperactivity of the subthalamic nucleus (STN) drives the internal globus pallidus (GPI) which leads to an inhibition of the ‘motor thalamus’ (ventro‐lateral and ventro‐anterior nuclei). Consequently, the output of the thalamus to the sensorimotor cortex is reduced, resulting in hypokinesia. The involvement of other brain areas such as the supplementary and cingulate motor areas (SMA, CMA), premotor cortex (PM), sensory cortices and the cerebellum remains unclear in the described model. However, in the last years, this pathophysiological concept of Parkinson’s disease was corroborated by the successful treatment of a variety of parkinsonian symptoms by lesioning of the STN in a primate model of Parkinson’s disease (Bergman et al., 1990) and, subsequently, by high‐frequency stimulation of the STN in Parkinson’s disease patients, resulting in a remarkable reduction of symptoms (Limousin et al., 1998; Krack et al., 2000; Volkmann et al., 2001).

Non‐invasive magnetoencephalographic (MEG) recordings using back‐averaging from electromyographic (EMG) activity identified involvement of deep diencephalic, premotor, primary motor and somatosensory areas in the tremor cycle of Parkinson’s disease (Volkmann et al., 1996). Furthermore, this study demonstrated that cortical and, partially, subcortical activity was coherent with the peripheral EMG in the frequency range of the tremor. The availability of MEG systems covering the whole scalp (Ahonen et al., 1993) and methodological advances (Gross et al., 2001) now allow investigation in more detail of the oscillatory network and mechanisms involved in Parkinson’s disease tremor. The present study addressed the question of which cortical and subcortical areas contribute to the generation of resting tremor in Parkinson’s disease and how they interact with each other. Our results obtained from non‐invasive MEG recordings in six patients with idiopathic Parkinson’s disease provide evidence for an extensive network of brain areas including motor and sensory cortices as well as diencephalic and cerebellar areas that are involved in the generation of resting tremor.


Patients and clinical assessment

The study was performed in six patients with idiopathic tremor‐dominant Parkinson’s disease showing a typical unilateral tremor at rest ranging from 4 to 6 Hz (determined by EMG power spectra; for patient characteristics, see Table 1). In all patients, akinesia and rigidity were less pronounced than tremor, and there was only minimal head or trunk tremor. None of the patients has had a history of neurological diseases other than Parkinson’s disease. Patients were off Parkinson’s disease medication for at least 12 h. All patients involved in the study gave their informed consent. The study was approved by the local ethics committee and is in accordance with the Declaration of Helsinki.

View this table:
Table 1

Patient clinical characteristics

PatientSexAge (years)Disease duration (years)Hoehn–YahrCurrent PD medicationTremor sideTremor frequency (Hz)
2M57212.5Selegelin, bromocriptin (7.5 mg), l‐dopa/carbidopa (437.5 mg)Left4
3M6331.5l‐dopa/carbidopa (375 mg)Left4
4F7861Budipin (60 mg)Right4
5M3531.5l‐dopa/carbidopa (375 mg), biperiden ret. (12 mg), dihydroergocryptin (15 mg)Right5.5
6F4561.5l‐dopa/carbidopa (375 mg), biperiden ret. (12 mg)Right4.5

Hoehn & Yahr state after 12 h off medication. Medications are given in the total daily dose. Tremor side denotes the clinically predominant side of Parkinson’s disease resting tremor. Tremor frequency was assessed by peaks in EMG power spectral activity.


    Cortical activity was recorded with a whole‐head Neuromag 122™ MEG system (Ahonen et al., 1993) at 1011 Hz with a band pass filter of 0.03–330 Hz. Simultaneously, muscle activity was registered with surface EMG electrodes placed on the extensor digitorum communis (EDC) muscle. To include only EMG bursting activity, the EMG was high‐pass filtered off‐line at 60 Hz and rectified. Recordings were performed during rest when the subjects relaxed their arm and hand muscles and tremor developed spontaneously. Patients were carefully selected, having unilateral hand tremor and minimal head or trunk tremor. However, the head was stabilized within the helmet with foamed plastic patches, to reach maximal stabilization with acceptable convenience for the patient. All MEG data were inspected carefully off‐line for movement artefacts, and times contaminated with movement artefacts were cut off the data trace. Further head movement artefacts result in a characteristic distribution of corticomuscular coherence in the whole‐scalp channel plots which were checked in each data set.

    The exact position of the head with respect to the MEG sensors was determined by four indicator coils attached to the head of the individual subject. The exact positions of the coils were determined with respect to three defined anatomical landmarks using a three‐dimensional digitizer. Individual high‐resolution MRIs (1 mm slice thickness) were obtained from a 1.5 T Siemens Magnetom™ MRI scanner. The pre‐defined anatomical landmarks were identified in the individual MRI scans, and the MEG and MRI coordinate systems were aligned. Cerebral sources of the MEG signals were superimposed on the individual MRI scans.

    As described by Halliday et al. (1995) and used extensively in previous studies on MEG–EMG coherence (Conway et al., 1995; Salenius et al., 1997; Brown et al., 1998; Gross et al., 2000), we calculated coherence between MEG and EMG signals with a resolution of 0.98 Hz, as a frequently applied measure for the interdependence of two signals in the frequency domain. Coherence is the ratio of the magnitude squared cross spectra of two signals to the product of their individual autospectra (for details see Halliday et al., 1995; Gross et al., 2000; Schnitzler et al., 2000). Values can range between 0, indicating independence of two signals, and 1, indicating a perfectly linear relationship. As described in detail in a previous study (Gross et al., 2002), we used the analysis tool DICS [Dynamic Imaging of Coherent Sources (Gross et al., 2001)] which employs a spatial filter algorithm (Gross and Ioannides, 1999) and a realistic head model of the individual subject to identify brain sources that are coherent to the surface EMG. Briefly, we performed the transition from time to frequency domain by using the fast Fourier transform (FFT). The FFT was applied to all MEG and EMG signals in 1 s data segments (after applying a Hanning window), and the cross‐spectral density C was computed between all combinations of MEG and EMG signals. The complex spectrum C finally was averaged across the whole recording period. One element Ci,j of the final cross‐spectral matrix consists of the cross spectrum of signals i and j, i.e. signals seen by two different sensors. In the second step, we extracted the mean cross‐spectral density of all sensor combinations in a selected frequency band as a complex N × N matrix, where N was the number of signals (MEG and EMG). Computation of cerebro‐muscular coherence used the cross spectrum between the EMG signal and all MEG signals, whereas the cerebro‐cerebral coherence required the cross spectra of all combinations of MEG signals. The third step consisted of the application of a spatial filter in the frequency domain (Gross et al., 2001). This procedure allows the estimation of coherence between two points in the brain (cerebro‐cerebral coherence) or between a point in the brain and an external reference signal (here cerebro‐ muscular coherence). To create tomographic maps, the spatial filter was applied to a large number of voxels covering the entire brain, assigning to each voxel a specific value of coherence to a given reference point or signal. A voxel size of 6 mm was used for our study.

    The brain source with strongest coherence to the EMG signal at tremor and at double tremor frequency was identified. This source was defined as the reference region for further coherence analysis between brain areas. Since the coherence of a reference region with itself is always 1, the reference region was projected out of the coherence matrix and further coherent areas were identified. As described previously (Gross et al., 2002), individual maps of strongest cerebro‐muscular or cerebro‐cerebral coherence were spatially normalized, averaged and displayed on a standard brain in SPM99 (Wellcome Department of Cognitive Neurology, Institute of Neurology, University College London, UK, http://www.fil.ion.ucl.ac.uk/spm).

    The confidence limit for the cerebro‐muscular coupling was derived according to Halliday et al. (1995). For cerebro‐cerebral coherence, the confidence limit was computed from surrogate data by randomly shuffling the original time courses, thereby destroying all real coherence.

    In a further analysis, we calculated the ‘partial coherence’ as previously used by others (Halliday et al., 1995; Mima et al., 2000; Ohara et al., 2001). This calculation indicates how much of a coupling between two signals (e.g. brain areas) can be explained by an independent coupling of both signals with a third signal (e.g. an EMG). In a first step, we estimated to what extent the coherence between identified brain areas and primary motor cortex (M1) could be explained by a separated coupling of both areas with the EMG. This calculation is denoted as partial cerebro‐cerebral coherence in the following. In a second step, we calculated to what extent the coupling between different cerebral areas and the EMG can be explained by M1 activity (which we expected to show a strong cortico‐muscular coherence) in order to estimate whether there was ‘true cerebro‐muscular coupling’. This calculation is denoted as partial cerebro‐muscular coherence.

    To compute the cerebro‐muscular delay, the activity of one identified cerebral source and the rectified EMG signal were filtered with a narrow band pass filter (±2 Hz) at the frequency band of cerebro‐muscular coherence. The Hilbert transform was applied to both signals. This transformation allows a separation of phase and amplitude. Times of maximum amplitudes of the EMG which corresponded to times of maximum tremor were detected from the Hilbert amplitude. This was done to improve the signal‐to‐noise ratio and due to the natural fluctuation in the intensity of the parkinsonian resting tremor. The instantaneous phase differences between the cerebral source activity and the EMG signal at these times of maximum amplitude were computed. Phase differences between the two signals were transferred into time differences and plotted in histograms to estimate the cerebro‐muscular delay (for further details see Gross et al., 2000; Schnitzler et al., 2000).


    All six patients showed the typical Parkinson’s disease resting tremor in the frequency range of 4–6 Hz (Table 1). Power spectral analysis of the EMG activity consistently revealed a dominant peak at tremor frequency and a smaller peak at double tremor frequency (Fig. 1A and B, Table 2). Figure 1C illustrates in a single patient significant coherence between MEG sensors covering the contralateral sensorimotor cortex and the EMG at tremor and at double tremor frequency.

    Fig. 1 Analysis of cerebro‐muscular coherence in a patient with left sided parkinsonian rest tremor. (A) A 3 s trace of surface EMG activity from the left extensor digitorum communis (EDC) muscle. The EMG was high‐pass filtered with 60 Hz and rectified. Regular EMG bursts occur at the tremor frequency of ∼4.5 Hz. (B) Power spectral activity of the tremor EMG. The power peak at tremor frequency ∼4.5 Hz is larger than its first harmonic at double tremor frequency. (C) Coherence between the tremor EMG and all 122 MEG sensors as viewed from above. Significant coherence at tremor frequency and, even larger, at double tremor frequency is localized primarily over the right sensorimotor cortex. The inset magnifies the sensor with highest coherence. (D) Localization of cerebro‐muscular coherence with DICS in the individual high‐resolution MRI scan in the precentral gyrus corresponding to the hand area of M1. (E) Coherence between M1 activity and the tremor EMG. Again, significant coherence occurs at tremor frequency and, even more significant, at double tremor frequency. The dashed line in C (inset) and E indicates the 99% confidence level.

    View this table:
    Table 2

    Power spectral peaks

    Frequency3–7 7–133–77–133–7 7–133–77–133–77–133–77–133–77–133–77–13

    Power spectral activity of the hand muscle EMG (extensor digitorum communis, EDC) and the cortical sources. The tremor frequency band (3–7 Hz) and the double tremor frequency band (7–13 Hz) are shown. Asterisks indicate the peak of stronger power. The last line totals up the asterisks for the two frequency ranges to reveal the preferred frequency band in the patient group (M1, primary motor cortex; Dienc., diencephalic activity; cerebellum, cerebellum ipsilateral to the tremor side; CMA/SMA, supplementary motor areas; premotor, lateral premotor areas; SII, secondary somatosensory cortex; PPC, posterior parietal cortex).

      Identification of cerebral sources with DICS that were coherent to the EMG consistently showed activation of the contralateral M1 (Figs 1D and 2). Group results of cerebro‐muscular coherence for patients with right sided tremor are summarized in Fig. 3A and confirm M1 as the brain area of maximum coherence to muscle, and also confirm findings in previous studies (Volkmann et al., 1996; Gross et al., 2001; Salenius et al., 2002). Interestingly, M1–EDC coherence was not only significant at tremor frequency, but was even stronger at double tremor frequency (Tables 3A and 3B). Power spectral analysis of M1 revealed the strongest peaks at the double tremor frequency (Table 2).

      Fig. 2 Localization, power spectra and spectra of cerebro‐muscular coherence in a Parkinson’s disease patient with right hand tremor. Source localization as revealed by DICS showed activity in contralateral M1 (A), PM (B), ipsilateral cerebellum (C), diencephalon (D), SII (E) and PPC (F). Note that the power spectra of all areas show a peak at double tremor frequency. Coherence between cortical and subcortical activity and the right extensor digitorum communis muscle (EDC) exhibits significant peaks at tremor frequency and, in some cases, stronger at double tremor frequency.

      Fig. 3 Mean activity SPM99 maps of spatially normalized cerebro‐muscular and cerebro‐cerebral coherence of the four patients with right sided Parkinson’s disease rest tremor. (A) Cerebro‐muscular coherence at double tremor frequency is located in the contralateral motor cortex. (B) Cerebro‐cerebral coherence was computed with the reference region in M1 and averaged for all patients. Areas of consistent coupling are CMA/SMA, SII and PPC. Note that the activity in the lateral premotor area adjoins superiorly the large SII activity. (C) After the areas of strongest coupling were projected out, the average SPM map also demonstrates diencephalic activity contralateral to the tremor hand and cerebellar activation ipsilateral to the tremor hand. Due to the poor coverage by and the large distance to the MEG sensors, localization in both areas is not as precise as at the cortical level.

      View this table:
      Table 3A

      Significant cerebro‐muscular coherence and phase lags at single tremor frequency and double tremor frequency between all identified cortical sources and the EDC (95% confidence level = 0.7%)

      CM‐Delay Delay M1Dienceph.Cerebell.CMA/SMAPremotorSIIPPC
      coh.singledouble3–77–133–7 7–133–77–133–77–133–77–133–77–133–77–13

      The tremor frequency band (3–7 Hz) and the double tremor frequency band (7–13 Hz) are shown. The first number indicates the frequency, the second number the coherence (in %). Cerebellum, diencephalic activity and PPC show the strongest coherence primarily at tremor frequency, whereas all other areas demonstrate strongest coupling at double tremor frequency. Coherence in the physiological range of 15–35 Hz was not observed (for abbreviations, see Table 2 footnotes).

        View this table:
        Table 3B

        Partial coherences between the EMG and cerebral areas with respect to activity in M1 with cerebro‐muscular phase‐lags

        coh.3–7 7–133–77–133–77–133–77–133–77–133–77–13
        16.5/20 (–18, 20)5/4.0 (13, –30)5/2.3 (–20)
        24/5.2 (–24)4/2.1 (–17)
        34/34.0 (–22, 22)4/6.5 (29, –30)8.5/1.94/25 (25)8.5/6.54/7.0 (–20)8.5/34/18.5 (–35)
        44/2.0 (–10)10/2.1
        55.5/23 (–20, 27)5.5/4 (–15)5.5/7 (–20)
        64.5/32.0 (–22, 20)4.5/7.5 (–19, 21)4.5/21.5 (– )4.5/27 (–23)4.5/17.8 (–15)9/5.5

        Positive values denote the cerebral activity leading the EMG, and negative values indicate EMG activity leading the cerebral activity, the number of milliseconds is given in parentheses. PM, SMA and the cerebellum show only minor coupling with the peripheral EMG signal once the M1 activity is taken into consideration. In contrast, there is strong bi‐directional coupling between the diencephalon and the EMG, and afferent coupling between SII as well as PPC and the periphery. The number of significant coherences in a frequency band are given in the lowest line (Σ).

          In one subject, we analysed further the nature of the cortico‐muscular coupling at double tremor frequency (Fig. 4). In a first step, we averaged M1 and EMG activity with respect to the zero phase of the first dorsal interosseus muscle (FDI). Interestingly, the agonist and antagonist muscle showed an alternating activation pattern, but also the clear representation of the double tremor frequency (Fig. 4B). The averaged M1 signal was indicative of a pure representation of the 8–12 Hz frequency. In a second step, we calculated the coherence between the two antagonistic muscles. The coherence was larger at tremor frequency than at double tremor frequency. We then calculated the partial coherence between the two muscles with respect to the M1 activity. Interestingly, the partial coherence matched the coherence at single tremor frequency but was clearly reduced at double tremor frequency. Therefore, the main coupling between the two muscles at double tremor frequency was carried by the activity in M1.

          Fig. 4 (A) Coherence and partial coherence with respect to M1 activity between two antagonistic muscles (first dorsal interosseus muscle, FDI; and extensor digitorum communis, EDC) in one of the patients. Both muscles show extensive coupling at tremor frequency and, to a lesser extent, at double tremor frequency. Interestingly, the partial coherence demonstrates that M1 activity does not contribute to the coupling at tremor frequency, but contributes significantly at double tremor frequency. (B) Phase‐triggered averages on the onset of muscle activity in the FDI. Both muscles show, in the averages, an alternating pattern of activity, but the double tremor frequency is clearly represented in both muscles. In contrast, in M1 activity, only the double tremor frequency is represented.

          Starting from M1 as the reference point, we calculated cerebro‐cerebral coherences stepwise without any assumptions. We found consistently strong coherences between M1 and several other areas, including deep diencephalic areas, contralateral cerebellum (ipsilateral to the tremor hand), lateral PM and SMA/CMA, as well as SII and PPC (see Figs 2 and 3). The consistency of these findings was confirmed by the average SPM99 map of cerebro‐cerebral coherence with the reference point in M1 in the four patients with right sided tremor. The average map corroborated the findings in the single subjects by demonstrating the involvement of PPC, SII, CMA/SMA (Fig. 3B), deep diencephalic activity and the contralateral cerebellum (Fig. 3C). A detailed spectral analysis of cerebro‐cerebral coherence (Table 4A) revealed that, although tremor frequency was present in the coupling between all areas, the double tremor frequency was considerably stronger in almost all cases. In some interactions involving the diencephalic sources, we observed coupling in the 20 Hz range (Table 4A).

          View this table:
          Table 4A

          Cerebro‐cerebral coherence in a selection of areas

          Frequency3–77–13∼203–7 7–13∼203–77–133–77–133–77–133–77–13
          SD1.0/0.61.9/1.70.9/1.90.6/11.00.8/14.31.5/11.11.3/3.01.5/14.90.3/2.31.3/3.9– /2.00.8/3.20.6/0.91.2/1.5

          The tremor frequency band (3–7 Hz), the double tremor frequency band (7–13 Hz) and coherence at ∼20 Hz are shown. Combinations with low signal to noise ratio resulting in no clearly distinctive peaks are marked with x. Frequency and strength of significant coherence (frequency/coherence in %; 95% confidence level = 0.7). For abbreviations, see Table 2 footnotes

            In a further step, we analysed, using partial coherence, whether the cerebro‐cerebral coherence could be explained solely by coupling of all central areas with the peripheral EMG activity (Table 3B). Interestingly, we observed significant coupling between SII as well as PPC and M1 in only four out of the six patients. Further, coupling between M1 and the cerebellum was significant in only half of the patients using the partial coherence analysis (Table 4B). In contrast, the coupling between M1 and the diencephalon, SMA/CMA and the PM was represented consistently and strongly in the partial coherence analysis.

            View this table:
            Table 4B

            Partial coherence between M1 and other cerebral areas with respect to the peripheral EMG

            AreaM1–cerebell.M1–dienceph.M1–CMA/SMA M1–PMM1–SIIM1–PPC
            Frequency3–77–133–7 7–13∼203–77–13∼203–77–13∼203–77–133–77–13

            Note the relatively weaker partial coherence between M1 activity and the cerebellum, SII and the PPC.

              In a further step, we analysed the coherence between the EMG signal and all other areas detected previously by cerebro‐cerebral coherence analysis. As demonstrated before for M1–EMG coupling, spectral peaks of coherence were found at tremor and at double tremor frequency between hand muscle EMG and brain activity. Cerebellum, diencephalon and PPC revealed stronger cerebro‐muscular coherence at tremor frequency, whereas M1, CMA/SMA, PM and SII showed stronger coupling at double tremor frequency (Tables 3A and 3B, Fig. 2).

              To rule out a contamination of these cerebro‐muscular coherences by a tight connection of the cerebral areas with M1 (which has a strong coupling with the peripheral EMG), we calculated the partial coherence between the signals in all cerebral areas and the EMG with respect to the M1 activity (Table 3B). This analysis revealed that some areas, such as SMA/CMA, the cerebellum and the PM, have a relatively sparse direct coupling to the EMG, whereas other areas, such as the diencephalon, SII and PPC, show more frequently a significant coupling with the peripheral signal (Table 3B).

              Calculation of the phase shift between M1 activity and the EMG corresponded to a delay of 18–25 ms at tremor frequency (assessed in four patients) and 15–17 ms (M1 leading the EMG in both) at double tremor frequency, in agreement with cerebro‐muscular conduction times to forearm muscles (Rothwell et al., 1991). We also calculated the phase shifts between SII, PPC, diencephalon and the EMG signal since these three areas showed significant direct coupling with the peripheral signal in the partial coherence analysis. Phase shifts to the EMG signal were between –10 and –23 ms in SII and between –15 and –35 ms in the PPC (EMG signal leading the cerebral signal in both, Table 3B). Interestingly, the phase shifts between the diencephalic activity and the EMG showed two peaks in three of four patients with significant coupling: one phase shift with the cerebral signal leading the EMG signal and one phase shift with the diencephalic activity following the EMG signal (Table 3B).


              The present study demonstrates the involvement of a variety of cortical and subcortical brain areas in Parkinson’s disease resting tremor. We showed that cerebro‐muscular coherence during rest tremor is established primarily in the single and double tremor frequency range. We described coupling in an extensive network of cortical and subcortical brain areas preferentially in the double tremor frequency range, but also at ∼20 Hz in interactions involving the diencephalon. We now discuss our results with respect to involvement of different brain areas. The findings are summarized in Fig. 6.

              Fig. 5 Cerebro‐cerebral coherence spectra in the same Parkinson’s disease patient as in Fig. 2. Cerebro‐cerebral coherence is established at tremor frequency and at double tremor frequency.

              Fig. 6 Synopsis of the cortical and subcortical network in Parkinson’s disease resting tremor. M1 is the cerebral area with strongest coupling to the peripheral EMG. The direction of the coupling as revealed by the phase lags is indicated by the arrows. The cerebellum, SMA/CMA and premotor areas do not show consistent significant coupling with the peripheral muscle, as revealed by the partial coherence analysis (only two out of six patients). The activity in SII and the PPC is of afferent nature, whereas the phase lags between activity in the diencephalon and the EMG indicate a bidirectional coupling. Dashed lines indicate inconsistent or weaker coupling (only four out of the six patients).

              Primary motor cortex (M1)

              Oscillatory signals that are not strictly sinusoidal (as are most biological signals) show harmonic and subharmonic oscillations, i.e. power in frequency bands that are multiples/fractions of the original frequency. The observed cerebro‐muscular coupling at double tremor frequency might represent the first harmonic of the tremor frequency. However, cerebro‐muscular and, even more pronounced, cerebro‐cerebral coherence were stronger in many areas at double tremor frequency than at tremor frequency. This observation suggests that the 8–12 Hz spectral component may not represent the first harmonic of tremor frequency. The M1–EMG phase lags at single and double tremor frequency are suggestive of an efferent cortico‐muscular drive. Using an isometric hold task in parkinsonian patients, Salenius et al. (2002) found a similar phase lag in the 7–10 Hz range but a negative phase lag in the 4–6 Hz range. Whether this difference is due to the different tasks or different methods of phase calculations remains unclear.

              Many patients with Parkinson’s disease show a small amplitude action tremor with a typical frequency of ∼8–12 Hz (Lance et al., 1963; Bergman and Deuschl, 2002). One might speculate that the demonstrated frequency component at ∼8–12 Hz during the parkinsonian rest tremor represents this high frequency action tremor. This consideration implies that there has to be a switch from a 4–6 Hz tremor during rest to a 8–12 Hz tremor during action. Recently, Wenzelburger et al. (2000) clearly demonstrated in a reach‐to‐grasp task that the parkinsonian rest tremor is different from the parkinsonian action tremor. However, the latter does not match the double resting tremor frequency in the individual patient, although they consider that in their study the frequency of the action tremor might be underestimated. This group therefore proposed that the high frequency parkinsonian action tremor represents an enhanced physiological tremor and is different from the resting tremor (Wenzelburger et al., 2000). Further experiments are needed to demonstrate whether the frequency and the involved brain regions of the 8–12 Hz action tremor resemble the high frequency 8–12 Hz component of the parkinsonian rest tremor.

              Interestingly, the frequency range of 8–12 Hz represents a prominent physiological phenomenon in the motor system: slow finger movements show discontinuities at ∼8–10 Hz (Vallbo and Wessberg, 1993) arising from a cerebello‐thalamo‐cortical loop oscillating at 8–10 Hz (Gross et al., 2002); force adjustments in response to load perturbations during a precision grip task take ∼80–120 ms (Johansson et al., 1992, 1999); the fastest alternating finger–hand movements show a maximum at ∼8–10 Hz (Freund, 1983); and the frequency of physiological tremor has a peripheral (Allum et al., 1978) and a central component at ∼8–12 Hz (Elble and Randall, 1976; Deuschl et al., 1998; Raethjen et al., 2000, 2002). One might therefore speculate as to whether the parkinsonian tremor of 4–6 Hz simply represents the first subharmonic of the intrinsic central 8–12 Hz in the sensorimotor system. This idea is supported by the presented data of one of our patients. We demonstrated that (i) the 8–12 Hz frequency range is represented in agonistic and antagonistic muscles and (ii) the coupling between antagonistic muscles at tremor frequency is highly independent of the M1 activity, whereas the coupling between the two antagonistic muscles at double tremor frequency depends greatly on the M1 activity (Fig. 4). One suggestion to explain these findings could be that there is a 8–12 Hz motor–cortical oscillatory activity which drives the spinal motor neuron pool and leads to the high coherence at double tremor frequency and the representation of this frequency in both muscles. The cortico‐muscular drive could be modulated by an alternating inhibition, e.g. on the spinal level, which would lead to the typical parkinsonian alternating pattern of muscular activity.

              By localizing brain activity that is coherent with the tremor EMG and by determining cerebro‐muscular delays, we identified M1 as the cortical area that is driving the spinal motor neuron pool in Parkinson’s disease resting tremor. Our findings thus far confirm and integrate the observations on Parkinson’s disease rest tremor of a previous MEG study demonstrating the involvement of the motor cortex (Volkmann et al., 1996) and the coupling between EEG and MEG sensors above the sensorimotor cortex and the EMG at single and double tremor frequencies (Tass et al., 1998; Hellwig et al., 2000; Gross et al., 2001; Salenius et al., 2002).

              Abnormal activity within the sensorimotor cortex of Parkinson’s disease patients compared with controls in a variety of movement paradigms has been deduced previously from increased M1 activity in fMRI (Sabatini et al., 2000; Haslinger et al., 2001) and PET studies (Fukuda et al., 2001). During rest, Parkinson’s disease patients showed an abnormal pattern of cortical inhibition in transcranial magnetic stimulation (TMS) experiments (Valls‐Sole et al., 1994; Ridding et al., 1995). Indeed, recent recordings in the M1 of the primate model of Parkinson’s disease showed abnormal synchronization of firing, but no changes in overall firing rate (Goldberg et al., 2000). Since it is conceivable that synchronization of neuronal activity changes under altered inhibition (Llinas et al., 1999), our results of strong M1–EMG coherence at tremor and double tremor frequencies could reflect the pathological synchronization of neuronal activity, in agreement with the observed abnormal cortical inhibition at rest in Parkinson’s disease patients (Valls‐Sole et al., 1994; Ridding et al., 1995). As already suggested by Volkmann et al. (1996), this abnormal rest activity in M1 is likely to interfere with the initiation and performance of voluntary motor actions. Interestingly, recent MEG–EEG studies in Parkinson’s disease patients demonstrated an altered low frequency motor–cortical drive during isometric contraction which could be reversed by l‐dopa intake (Salenius et al., 2002) and after surgical pallidotomy (Conway et al., 1999).


              The involvement of the cerebellum in parkinsonian resting tremor currently is under debate (Deuschl et al., 2000). The involvement of the cerebellum was suggested on the basis of a PET study showing decreased cerebellar activity on thalamic stimulation associated with suppression of tremor (Deiber et al., 1993). Furthermore, stimulation of the VIM (ventral intermedial nucleus) can abolish Parkinson’s disease resting tremor (Lenz et al., 1994; Krack et al., 2000). The fact that this thalamic nucleus receives primarily cerebellar afferents and projects mainly to M1 (Inase and Tanji, 1995; Steriade et al., 1997) favours the view of a pivotal role for the cerebellar–thalamo‐cortical loop in tremor generation. On the other hand, Deuschl et al. (1999) described a patient who developed a Parkinson’s disease resting tremor even though the ipsilateral cerebellum had been totally removed 17 years before. Our study demonstrates tremor‐correlated activity within the cerebellum ipsilateral to the tremor hand. The cerebellar activity is coherent with contralateral thalamic and cortical activity and was identified by coherence analysis with a reference region in M1. Furthermore, the partial coherence analysis revealed that the coupling between the cerebellum and the EMG is significant only in two out of the six patients. This finding opposes the view that cerebellar activity is solely an afferent reflection of the peripheral signal, but suggests that cerebellar activity is integrated, probably as part of a cerebello‐thalamo‐cortical pathway, in the generation of Parkinson’s disease tremor.

              Diencephalic activity

              Due to the technical limitations of MEG, the demonstrated deep activity cannot be attributed clearly to thalamic or basal ganglia activity and was therefore termed diencephalic. The identified deep sources in the present study showed activity at tremor frequency and at double tremor frequency. Both frequencies have been observed previously in MPTP (1‐methyl‐4‐phenyl‐1,2,3,6‐tetrahydropyridine) monkeys as well as during stereotactic surgery of Parkinson’s disease patients in the STN (Bergman et al., 1994; Hutchison et al., 1998; Brown et al., 2001; Levy et al., 2000, 2001), GPI (Bergman et al., 1994; Hutchison et al., 1997; Hurtado et al., 1999; Lemstra et al., 1999; Magnin et al., 2000; Raz et al., 2000; Levy et al., 2001) and thalamus (Lenz et al., 1988; Marsden et al., 2000). Simultaneous recording in STN and GPI of Parkinson’s disease patients revealed coherent activity at 6 Hz, closely resembling the tremor frequency with the STN leading the GPI signal (Brown et al., 2001). This indicates that the tremor frequency is already represented within the STN and probably transmitted to the GPI. Interestingly, there was also an ∼20 Hz component in the STN–GPI coherence in humans (Brown et al., 2001). In this frequency band, GPI activity led the STN activity. Our analysis revealed that the diencephalic activity is strongly connected to the peripheral signal, with part of the diencephalic activity leading the EMG signal and partly following the peripheral activity. These data indicate that the coupling between the periphery and the diencephalon at single and double tremor frequency is bidirectional in patients with parkinsonian rest tremor. In most patients, we have found broad 20 Hz components not only in the diencephalic–cortical coupling but also between the cerebellum and diencephalon. These 20 Hz oscillations are abolished after l‐dopa ingestion (Brown et al., 2001). Deep brain stimulation at these frequencies induced severe exacerbation of symptoms (Demeret et al., 1999). Thus, these 20 Hz oscillations might be considered as part of the pathological synchronization in the deep brain structures. Further studies will have to focus on the question of whether frequency components at ∼20 Hz are present under physiological conditions in deep diencephalic structures associated only with pathological states. Interestingly, Volkmann et al. (1996) postulated the stabilization of intrinsic 3–6 Hz oscillations by an extrinsic loop through the nucleus ventralis intermedius thalami via transcerebellar pathways back to the motor cortex. In their study, no cerebellar activity could be detected with the method of magnetic field tomography and a 37‐channel MEG system. By using DICS (Gross et al., 2001) and a whole‐head MEG system (Ahonen et al., 1993), in the present investigation we detected the involvement of cerebellar and diencephalic activity in the cerebral oscillatory network of parkinsonian resting tremor. Our observations thus provide experimental evidence for the postulations of Volkmann et al. (1996).

              Sensory cortices

              The cerebro‐cerebral coherence analysis revealed the involvement of SII and the PPC in Parkinson’s disease resting tremor. We assume that both activations correspond to afferent feedback of the peripheral tremor because the calculation of phase shifts revealed that the EMG activity precedes the cerebral activity in both areas. The activity in PPC was already postulated in a recent EEG–EMG study in which one parkinsonian patient showed coherence between the EMG and electrodes above the sensorimotor cortex, and also between EMG activity and EEG electrodes over the area of the PPC (Hellwig et al., 2000). Our study confirms this observation and demonstrates that PPC is integrated in the central network of Parkinson’s disease resting tremor.

              A previous study used magnetic field tomography (MFT) on data derived from back‐averaging on EMG tremor onset and was suggestive of an involvement of primary sensorimotor cortex (SI) in the tremor cycle of Parkinson’s disease (Volkmann et al., 1996). We did not observe a coherent source in SI. However, it may have been masked by the close proximity of the strong M1 source.

              Premotor area and CMA/SMA

              The present study demonstrates the involvement of lateral PM and CMA/SMA in the central representation of Parkinson’s disease rest tremor. The participation of the premotor areas in the tremor cycle had already been shown with previous back‐averaging techniques (Volkmann et al., 1996). The data of the partial coherence analysis demonstrated that there is relatively strong coupling between M1 and SMA/CMA, as well as M1 and PM. However, a significant coupling between PM, SMA/CMA and the EMG in the further partial coherence analysis could only be observed in two of the six patients. This finding suggests that the activity in PM and SMA/CMA primarily projects to M1 which then drives the spinal motor neuron pool. Our results indicate that the coupling between PM as well as CMA/SMA and M1 occurs primarily at double tremor frequency. This suggests that PM and CMA/SMA are ‘locked’ in pathologically synchronized activity. It is conceivable that in a state of highly synchronized neuronal activity, the new formation of neuronal clusters to initiate or control a voluntary movement is disturbed and might therefore interfere with activation during movement tasks. Clinically, our finding might serve as an explanation as to why Parkinson’s disease patients show severe deficits in the initiation, conception and performance of complex movements. Further studies are needed to show how activity and coupling in these areas change during voluntary isometric contraction and movement tasks when in many patients the parkinsonian resting tremor is suppressed.


              In summary, we suggest that the cerebro‐muscular coherence at a tremor frequency of 4–6 Hz to SII and PPC reflects an afferent input that is modulated in the intrinsic 8–12 Hz frequency range of the sensorimotor system. The same is true for diencephalic activity which also receives afferent input. The cerebro‐muscular coupling of M1, and also the cerebro‐cerebral coupling of M1, PM and CMA/SMA is established primarily in the double tremor frequency range of 8–12 Hz. These three areas and, possibly, diencephalic activity drive, presumably via M1, the spinal motor neuron pool mainly at double tremor frequency. This may explain the substantial difficulties of Parkinson’s disease patients in the initiation, conception and performance of complex movements. The signal at double tremor frequency is likely to correspond to an alternating output to the extensor and flexor muscles resulting in the typical antagonistic Parkinson’s disease resting tremor.

              The present study achieved an individual neurophysiological characterization of Parkinson’s disease patients based on non‐invasive MEG recordings and application of recently developed analysis tools. Interestingly, the present cerebro‐muscular coherence analysis of Parkinson’s disease tremor shows clear differences from recent MEG findings on other involuntary movements and tremor syndromes: ‘mini‐asterixis’, a postural tremor of 6–12 Hz in hepatic encephalopathy, is characterized by excessive M1–EMG coherence exclusively at tremor frequency (Timmermann et al., 2002). Slow involuntary movements increased ∼12 Hz corticomuscular coherence (pseudochoreoathetosis) after deafferentation (Timmermann et al., 2001), and conflicting results have been reported regarding MEG–EEG–EMG coherence in essential tremor (Halliday et al., 2000; Hellwig et al., 2001). More studies are needed to show whether different movement disorders can now be discriminated neurophysiologically by means of cerebro‐muscular and cerebro‐cerebral coupling analysis.


              We wish to thank our Parkinson patients for excellent cooperation during the measurements, Markus Butz, MA for generation of realistic head models, Frank Schmitz, MA for expert help during the measurements, and Mrs E. Rädisch for technical support with the MRI scans. We also wish to thank the two expert referees for excellent and helpful comments. This work was supported by the Volkswagen‐Stiftung (I/73240) and the Deutsche Forschungsgemeinschaft (SFB 194, Z2; SFB 575, C4).


              View Abstract