OUP user menu

Physiological and pathological tremors and rhythmic central motor control

J. H. McAuley, C. D. Marsden
DOI: http://dx.doi.org/10.1093/brain/123.8.1545 1545-1567 First published online: 1 August 2000

Summary

In recent years there has been increasing interest in oscillatory neural activity in the CNS and in the role that such activity may have in motor control. It is thought that physiological tremor may be a manifestation in the periphery of such central oscillatory activity and that some pathological tremors are the result of derangement of these oscillators. This review re-evaluates both early and recent studies on physiological and pathological tremors and other peripheral oscillations in order to gain a new perspective on the nature and function of their central progenitors. This approach, namely using tremor as a `window' into the function of central oscillations, is particularly suited to human investigations because of the obvious limitations of direct central recording. It is argued that physiological tremor is likely to be multifactorial in origin, with contributions not only from CNS 10-Hz range oscillatory activity, but also from motor unit firing properties, mechanical resonances and reflex loop resonances. Different origins are likely to dominate under different conditions. While some pathological tremors appear to arise as a distortion of central or peripheral components of physiological tremor, others arise de novo, such as the pathological oscillation of 3- to 6-Hz parkinsonian tremor. CNS oscillations outside the 10-Hz range are also found to modulate limb activity in normal individuals, and oscillatory activity exists in other motor systems such as eye movements. Finally, it is shown how studies of peripheral oscillations may help develop hypotheses on the role of CNS oscillations in motor control, including the proposed `binding' function of synchronized oscillations and the possibility that motor signals could be coded by frequency of modulating oscillation as well as by synaptic connectivity.

  • tremor
  • oscillation
  • physiological
  • pathological
  • motor control
  • AMG = acoustomyography
  • MEG = magnetoencephalography

Introduction

Tremor, defined as a rapid back-and-forth movement of a body part, is a readily apparent and easily quantified motor phenomenon found both in normal individuals and as a pathological symptom. Perhaps as a result, it has historically been a common subject of study by physiologists and clinicians alike. The first electrophysiological investigation of tremor seems to have been conducted by Horsley and Schafer, who found that ongoing muscle activity, whether induced voluntarily or by electrical motor tract stimulation in animal preparations, was universally characterized by superimposed 10-Hz tremulous twitches (Horsley and Schafer, 1886). It was perhaps also these authors who first considered that tremor may, at least in part, have a neurogenic basis.

Initially, the neurogenic mechanism of tremor generation was generally felt simply to reflect the inevitable consequence of the unitary nature of muscle organization (each motor unit essentially producing a distinct twitch of contraction) (Marshall and Walsh, 1956; Allum et al., 1978) or to be a peripherally arising phenomenon due to instabilities of reflex pathways (Halliday and Redfearn, 1956). Others noted that tremor may arise through the mechanical tendency for body parts to oscillate at certain frequencies (Joyce and Rack, 1974). However, dating from relatively early in the literature, a few studies have suggested that physiological tremor may arise from rhythmic neural activity in the CNS (Marsden et al., 1967b; Mori, 1975; Elble and Randall, 1976) and that some pathological tremors may reflect a distortion or amplification of these central oscillations (Elble, 1986).

The fact that neural activity in the brain is strongly rhythmic in nature has been clear from the earliest EEG recordings, and from an elegant study of pyramidal tract and motor cortical discharges in anaesthetized animals (Adrian and Moruzzi, 1939) such oscillatory activity is shown to modulate descending pathways controlling motor output. Recent monkey studies have confirmed the presence of synchronized discharges of corticomotor neurons at certain preferred frequencies (e.g. Murthy and Fetz, 1992, 1996; Nicolelis et al., 1995), suggesting a common rhythmicity. Detailed analysis of these rhythms in man is clearly more difficult because of the relative inaccessibility of the human brain to recording techniques. Although the EEG reveals much oscillatory cortical activity, oscillations recorded from the scalp in this way are poorly localized and are not clearly related to ongoing motor activity. However, partly due to the application of new recording techniques such as magnetoenecephalography (MEG), the last few years have seen greatly increased interest in the nature of central oscillatory activity and how it might become manifest in the periphery. In addition, new hypotheses have been put forward that CNS oscillations may have an important functional role in sensory information processing (Eckhorn et al., 1988) and in the processing of motor commands (Farmer, 1998).

In the light of such recent developments in understanding central oscillations, it is clearly important to reconsider earlier tentative suggestions that physiological tremor may be a peripheral manifestation of such activity. Many observations on tremor activity provide a basis for such suggestions. Quantitative recording techniques have made it clear that physiological (normal) tremor is not a pure 10-Hz range oscillation (Halliday and Redfearn, 1956), but is best regarded as the summation of a number of oscillations across a frequency range that is also well represented in the EEG. The involuntary nature of tremor and its widespread superimposition upon various motor activities make it tempting to suggest that it may be a `by-product' of corticomotor oscillations. Since oscillations in the motor cortex have been shown to modulate descending corticospinal pathways, it is easy to imagine how a similar pattern of modulation could become manifest in the muscles' EMG and finally in movement in the form of a tremulous oscillation. The clear demonstration of a central origin would mean that tremor, far from simply being considered as a peripheral phenomenon limiting or impairing motor performance, could constitute a new investigative tool, providing a non-invasive `window' into the rhythmic nature of human motor control.

The purpose of this review is therefore to re-evaluate the central origin of tremors, not only to provide an account of current understanding of mechanisms of tremor generation, but to consider how studies of tremor, as a manifestation of central oscillations, may reveal more about the nature and function of central rhythmic activity.

Origin of physiological limb tremor

While a number of early studies have suggested that physiological tremor may derive from central oscillations (e.g. Lippold et al., 1957; Elble and Randall, 1976), it is only relatively recently that such ideas have received renewed interest. A balanced perspective on the origin of physiological tremor is gained only by considering, in addition, the large bulk of literature describing early but nevertheless valid experiments that have indicated the importance of other mechanisms in tremor generation (Table 1). The various mechanisms and their sites of action along the motor pathway may be conveniently summarized diagramatically (Fig. 1).

View this table:
Table 1

Summary of proposed origin of peak EMG and tremor frequencies in healthy human subjects*

ReferenceFrequencyLocationConditionProposed origin
*Oscillations of similar origin are ordered first by frequency band and then by publication date.
Intrinsic muscle property
Ruegg et al., 19706–8 HzIsolated muscleContractionResonance of contractile system
Curtin et al., 19743 Hz isometric, 9 Hz isotonicIsolated muscleContractionActomyosin cross-bridge cycling
Wollaston, 1810c. 20–30 HzMuscle belly vibrationContractionMuscle fibre activity
Oster and Jaffe, 198025 HzVibration of different musclesIsometric contractionActomyosin cross-linking dynamics
Individual motor units
Allum et al., 19786–12 HzSmall hand muscle tremorIsometric contractionOverlap of slow corrective fluctuations and the firing rate of individual motor units
Dietz et al., 197610 HzHand and leg musclesIsometric contractionMotor unit firing rates; non-frequency specific hard-wired synchrony
Wee and Ashley, 198911 HzBiceps brachii vibrationsContraction by `tremulous' subjectsSubharmonics of fibre firing rates
Homberg et al., 19868–10 Hz (finger), 8–20 Hz (arm)Finger and arm muscle EMGIsometric contraction (+ force tremor)Firing rates of individual units
Gamet and Maton, 198910–16 HzBiceps brachii vibrationFatiguing isometric contractionsFiring rates of individual units
Orizio et al., 19905–10 Hz + 10–20 HzBiceps brachii vibrationStrong isometric contractionFiring rates of slow and fast motor units
Gordon and Holbourn, 194825–30 HzSingle motor unitsContractionMechanical `spikes' from unit firing
Synchronization by mechanical resonances
Yap and Boshes, 196710 HzFinger tremorRestBallistocardiac
Lakie et al., 19869 HzHand tremorOn postureMechanical
Reitsma, 199513 Hz Wrist/forearm Isometric contractionMuscle vibration
Amjad et al., 199410 Hz, 30–40 HzFinger tremorOn posture10 Hz = mechanical, 30–40 Hz = CNS
Stiles and Randall, 196725–30 HzFinger tremorMechanical perturbationMechanical resonance
Adrian, 192535–60 HzEMG of different musclesStrong contractionMechanical resonance
Synchronization by reflex loops
Halliday and Redfearn, 195610 HzFinger tremorOn posture and loadedStretch reflex loops
Sutton and Sykes, 19679 HzHand tremorSteady forceful contractionContribution from visual reflex loop
Dymott and Merton, 19689 HzHand tremorSteady forceful contractionVisual and non-visual reflex loops
Joyce and Rack, 197410–12 HzForearm tremorInertial and elastic loadsMechanical and stretch reflexes
Hagbarth and Young, 19798–10 HzHand/forearm tremorSpindle afferents on tremulous contractionStretch reflex loops
Matthews and Muir, 1980c. 10 HzBiceps brachii EMG and tremorContraction against elastic loadMechanical resonance made to match stretch reflex oscillation
Sakamoto et al., 199210 Hz, 25 HzTremor of fingersOn posturePeripheral feedback
Synchronizing CNS oscillations
Marsden et al., 1969a9 HzFingers/hand tremorOn postureMultifactorial (CNS component)
Mori, 1975c. 10 HzLower leg tremorQuiet stance, isometric dorsiflexionMotor units synchronized by neuronal networks
Elble and Randall, 19768–12 Hz, 13–22 HzFinger force tremor and EMGIsometric contractionLower frequency is synchronization by Renshaw cell activity, upper frequency represents unit firing rates
Vallbo and Wessberg, 19938–10 HzFinger muscle surface EMG and tremorSlow movementsCNS
Farmer et al., 1993a, b1–12 Hz, 16–32 Hz1DI motor unit EMGCorrelation between single motor units on weak contractionMechanical and CNS, respectively
Conway et al., 1995c. 20 Hz1DI surface EMGSteady contractionCorrelated with MEG oscillation
Piper, 1907c. 50 HzLarge muscle EMGStrong contractionSynchronization of motor units
Fex and Krakau, 1957c. 50 HzEMG of different musclesContractionActive motor unit synchronization
Hagbarth et al., 1983c. 50 Hz Piper rhythm + single units at subharmonicsEMG of different musclesSteady voluntary contractionSynchronization by CNS rhythms
Amjad et al., 199410 Hz, 30–40 HzFinger tremorOn posture10 Hz = mechanical, 30–40 Hz = CNS
McAuley et al., 199710 Hz, 20 Hz, 40 HzFinger EMG, AMG, tremorCompliant finger contractionCNS oscillations
Keidel and Keidel, 1989At rest, multiple peaks c. 10 Hz. On activity, also 20–100 HzMasseter, biceps, wrist extensors, tib. ant vibrationsAt rest and isometric contractionMotor unit firing rates, peaks due to CNS synchronization
Kirkwood et al., 1982c. 60 HzCorrelations between intercostal motor unitsBreathingBrainstem oscillator
Bruce and Goldman, 198316–40 Hz, 60–84 HzDiaphragm EMGBreathingCNS oscillators
Smith and Denny, 199020–60 Hz, 60–110 HzDiaphragm EMGSpeech, breathingDifferent CNS oscillators
Fig. 1

Diagram illustrating the various points along the motor pathway where 10-Hz range physiological tremor may arise. Feedback oscillations result from an inherent instability of negative feedback loops; while most movements are dampened by negative feedback, oscillatory activity of a period double the feedback delay time is instead enhanced. Thus, feedback delays of 50 ms favour the generation of 10-Hz range tremor. Three possible feedback loops that may exist include short latency spinal reflex arcs from afferent stretch receptors, long loop transcortical or transcerebellar reflex pathways from these receptors and central feedback from the motor neuron pool back to spinal interneurons such as Renshaw cells or back up the spinal cord. Tremor arising from these feedback oscillations and from oscillations derived from mechanical resonances will be associated with numbers of motor units firing together, since the oscillations operate via stretch receptor afferents which respond to movements of the muscle as a whole. Central oscillations will similarly tend to modulate a pool of motor neurons. Tremor from all such oscillations will therefore be characterized by synchronized motor unit activity. On the other hand, tremor arising from intrinsic motor unit properties, such as minimum firing rates, does not have to be associated with unit synchronization.

Motor unit firing properties

If physiological tremor has a neurogenic basis, then similar oscillations should be reflected in EMG signals of the muscles associated with the tremulous body part. Since tremors occurring during active contraction are most likely to be associated with a clear EMG signal, many studies have looked at the force tremor and corresponding EMG generated during such contractions. From the point of view of tremor generation, an important property of individual motor units is that they do not fire over a continuous frequency range but start firing at a minimum frequency of around 8–10 Hz (Henneman, 1979), possibly partly through the influence of spinal mechanisms such as Renshaw inhibition (Granit and Renkin, 1961). Surface EMG recordings performed during steady isometric contractions reveal what is termed an `interference pattern', reflecting the summated activity of numbers of motor units. The frequency spectra of such multi-unit recordings have a broad range of activity, but generally have a peak in a similar 8- to 10-Hz frequency range, which is shown to reflect this property of individual units (for review, see Freund and Dietz, 1978).

Active tremor arising from a summation of motor unit forces at the muscle tendon could therefore reflect these motor unit firing properties (Marshall and Walsh, 1956; Homberg et al., 1986). Although it is found to remain at around 10 Hz despite different contraction strengths and changing mean motor unit firing rates, this can be explained on the basis that newly recruited units firing at the minimum 8- to 10-Hz range firing frequency will be larger and tend to dominate the faster firing units. Moreover, force fluctuations due to faster firing units are relatively more attenuated by the muscle's mechanical properties, a phenomenon known as fusion (Marsden, 1978).

Synchronized motor unit oscillations

While most studies during isometric contraction have highlighted the role of firing frequencies of individual motor units in the generation of physiological tremor, under different experimental conditions the 8- to 12-Hz peak frequency reflects a stronger `tuned' oscillation, manifest above the `noise' of the interference pattern and resulting from synchronization of motor units so that they tend to fire together in a pervasive 8- to 12-Hz rhythm rather than at random (Fox and Randall, 1970; Mori, 1975; Elble and Randall, 1976; Elble, 1986). In one study (Elble and Randall, 1976), the majority of individual finger muscle units were actually found to fire at a completely different frequency of 13–22 Hz, so that only recordings of whole populations of units by surface EMG directly revealed the 8- to 12-Hz rhythmicity (Fig. 2A). Recordings of single units showed that, although they had a mean firing rate at the higher frequency, the discharges were not regular but grouped into couplets separated by periods corresponding to an 8- to 12-Hz rhythm. In this way, the external modulation is revealed indirectly at the single unit level.

Fig. 2

Schematic diagram illustrating how external driving oscillations may modulate EMG activity at that frequency even though individual units fire at a different frequency. (A) Because of the influence of the driving 10-Hz modulation, each of the three units shown (out of the many units of a muscle) is more likely to fire on each 10-Hz `beat' than at other times. Although the three units fire at around 20 Hz, there is no pervasive 20-Hz rhythm because firing of units at times outside the 10-Hz beats is unsynchronized and their effects on tremor of the whole limb will cancel rather than summate. The same effect may be achieved if each unit fires in couplets clustered around each 10-Hz beat (Elble and Randall, 1976). (B) The reverse effect occurs where units fire at around 10 Hz intermittently to a 20-Hz rhythm. Only firings at 20-Hz beats are synchronized and so coherence between units (Farmer et al., 1999b) is revealed at 20 Hz. Because of the staggered pattern of intermittent firing, there is no 10-Hz synchronization across units.

The difference between the two groups of studies may lie in the fact that the set-ups showing synchronized activity involved stronger contractions and allowed for more movement by the body part. [Mori simply used subjects standing on a force plate (Mori, 1975).] In addition, there may have been differences between subjects' tremulousness; Elble and Randall only included subjects who produced a rather strong, sharply tuned 10-Hz finger tremor (Elble and Randall, 1976). The upper end of the range of amplitudes of tremor in normal subjects may be regarded as enhanced physiological tremor (Young et al., 1975) and the mechanisms thought to be responsible for the increased tremor amplitude in this condition (see later) may therefore have played a greater part in these experiments.

Before describing the processes that may generate the synchronized firing of motor units, it is necessary to appreciate the variety of different synchronization patterns that may exist. These patterns depend on two main factors. (i) Units may tend to fire together simply because they receive branches of inputs from a common source, or they may fire together at a certain driving frequency due to the action of a common driving oscillation (i.e. the common source is itself an oscillation). (ii) Synchronization may be `tight' or `loose' depending on the number of synapses separating the motor neurons from their shared inputs. Each synapse will introduce temporal `jitter' in signal transmission, resulting in some independent variation between the units' firing times.

The different types of synchronization are perhaps best illustrated by animal studies on intercostal motor neurons where there is both a high degree of branching of the motor neuron inputs (Sears and Stagg, 1976) and a strong driving oscillation from the medullary respiratory centre at 60–120 Hz (Cohen, 1979). Three co-existing forms of synchronization are demonstrated (Kirkwood et al., 1982). (i) Short-term synchronization results from branching of the immediate non-rhythmic inputs to the motor neurons and is so-called because the resulting peak of the cross-correlogram comparing the firing times of pairs of motor neurons is very narrow, indicating that there is an increased probability of firing together within a very short (c. 5 ms) time period (Fig. 3A and B). (ii) An external or tuned synchronization results when these inputs are themselves driven by a common driving oscillation; not only is there a central correlogram peak representing branched inputs, but there is a whole series of peaks displaced at regular intervals corresponding to the period of the driving oscillation (Fig. 3E and F). (An external synchronization does not necessarily occur only in association with branched inputs because the driving oscillation may modulate a number of inputs together and therefore have a direct synchronizing action in itself.) (iii) Finally, long-term, pre-synaptic or broad-peak synchronization results from a common input many synapses distant from the point of recording; the cumulative variability in time of transmission across each intervening synapse results in a `slurring' of the common signal. In the case of respiratory motor neurons this input arises from respiratory afferents tending to respond together to sensory information and then indirectly impinging upon the motor neuron pool (Fig. 3C and D).

Fig. 3

Cross-correlation histograms between pairs of intercostal motor nerve units in the anaesthetized cat. The cross-correlogram charts the number of times that a pair of units fire together, indicated by the central zero on the x-axis, together with paired firings separated by different intervals, shown before and after the central zero. A central peak thus indicates synchronization, i.e. the units are more likely to fire together than at a finite interval of separation. B, D and F, respectively, represent A, C and E on a wider time scale. A and B show a narrow central peak cross-correlation due to short-term synchronization resulting from branched intercostal nerve inputs. C and D show a broad central peak due to shared pre-synaptic respiratory afferents and non-respiratory afferents to the intercostal nerves at a spinal level. E and F show not only a central peak, but a whole series of similar peaks each separated by a c. 15 ms period, indicating that in this preparation the branched inputs are modulated by a central oscillation at 60 Hz. From Kirkwood et al., 1982, with permission.

This demonstration of different peripheral synchronization patterns in the respiratory system, one reflecting branched descending inputs which happen also to be driven at a certain frequency by a central oscillation and the other reflecting shared polysynaptic inputs whatever their frequency, indicates that such mechanisms could also co-exist in the limb and should not be confused with one another. For example, the demonstration of a central cross-correlation peak between the EMG activities of two limb muscles (Bremner et al., 1991) does not in itself imply that there is a shared CNS oscillation generating this synchronization (McAuley and Brown, 1995). Indeed, it has been suggested that physiological tremor could be non-specifically augmented by short-term synchronization arising through branching of descending spinal inputs to multiple limb motor neurons (Dietz et al., 1976).

However, Elble and Randall considered that the 10-Hz range peak represented synchronization driven by a specific frequency of oscillation (Elble and Randall, 1976). Such synchronization is best detected by coherence analysis, a technique analogous to cross-correlation but performed in the frequency- rather than the time-domain (Jenkins and Watts, 1968). Thus, if significant coherence is present between two units, this implies that the units share common inputs generating synchronization and, in addition, that these common inputs carry a rhythmic modulation. Anatomical branching of unmodulated inputs will result in a significant cross-correlation without significant coherence. Conversely, a coherence peak could reflect either a widespread CNS oscillation with a pervasive descending influence (giving a broad central cross-correlogram peak) or a very local CNS modulation with branching of the corticospinal tracts (giving a narrow cross-correlogram peak) (Fig. 4).

Fig. 4

Comparison of signals from pairs of motor units recorded by needle EMG from 1DI in the human during a steady `pinch' contraction. The cross-correlogram (A) shows a central peak and perhaps also smaller and broader peaks separated by about 40 ms. The corresponding coherence plot (B) shows significant coherence at around 25 Hz, confirming that the two units share the same 20-Hz range oscillation. The horizontal lines in the cross-correlogram and coherence plots are 95% confidence lines for non-random levels of correlation and coherence, respectively. The bottom frequency plot (C) shows the percentage of 37 unit pairs that showed significant coherence for each frequency bin, showing that shared 20-Hz range oscillations are a consistent feature of different 1DI motor unit pairs. From Farmer et al., 1993b, with permission.

Although early reports considered that CNS oscillations were the external influence generating the synchronized rhythms of physiological tremor, mechanical resonances and peripheral feedback resonances could also generate such entrained activity (Adrian, 1925; Stiles and Randall, 1967; Joyce and Rack, 1974; Marsden, 1978; Hagbarth and Young, 1979; Young and Shahani, 1979; Matthews and Muir, 1980; Rhatigan et al., 1986) (Fig. 1), and so such factors must be considered and investigated in any study exploring the peripheral manifestation of central oscillators.

Mechanical resonances

The mechanical properties of bone, muscle and soft tissue will have an influence on the frequencies of vibration of a body part, especially when recording at a distal extremity, such as the finger. While mechanical factors will be primarily manifest as tremor, they may nevertheless result in a corresponding EMG modulation through its transmission from peripheral afferent stretch receptors to the motor neuron pool via reflex pathways (Adrian, 1925).

The mechanical fundamental frequency of vibration, f0, of any structure is related to its physical properties by the following equation (Walsh, 1992):Mathwhere K is the stiffness or forces exerted on a structure and J is the moment of inertia. The value for the unloaded finger is around 25–27 Hz (Stiles and Randall, 1967) while at the wrist and elbow the resonances are at 9 and 2 Hz, respectively (Marsden, 1984). As indicated by the equation, these values will change if the body part is loaded or any force is applied. The new fundamental frequencies can be determined by sharply perturbing the apparatus/body part arrangement and looking at the frequencies of the waning die-down oscillations resulting from these taps (Halliday and Redfearn, 1956). So unloaded resonance at the wrist is of a frequency that could result in physiological tremor, but more proximal body parts would have to be stiffened by muscular tension if mechanical factors were to account for physiological tremor, since the latter has a broadly similar frequency in different body parts.

It must always be clear whether a mechanical resonance is actually generating an oscillation or whether the mechanics are such as merely to amplify pre-existing oscillations. Even if the fundamental frequency does not result in an observed peak in an actively contracting system, it will still have a filtering influence so that another source of vibration of frequency widely different from f0 will have to be of great power to exert a noticeable modulation. Thus, an experiment which involved loading a body part would bias strongly against the detection of higher peak frequencies, whereas using an elastic force would bias towards higher frequencies.

Feedback resonances

The peripheral stretch reflex can be regarded as a negative feedback loop. If the loop is underdamped, oscillations will tend to occur with a period of double the loop time and such oscillations may result in synchronized activity of EMG and tremor at this frequency. The loop time for the spinal segmental stretch reflex in the finger, including the time from EMG to development of a movement detectable by the afferent receptors, is about 50 ms (Marsden, 1978) and would therefore tend to create oscillation at the physiological tremor frequency of 10 Hz. As well as this peak, a series of odd harmonics could result in reflex loop modulations at 30, 50 Hz, etc. of a `white-noise' type descending input [i.e. resonance at loop times equalling 0.5 (fundamental), 1.5, 2.5 cycles]. Oscillation due to the long latency stretch reflex, which plays a relatively important role in the hand and may dominate the spinal reflex in the finger (Matthews, 1993), would be at 7 Hz (Marsden, 1978). Even when short or long latency reflex loop values do not precisely correspond with observed peak frequencies, if the loops co-exist they may well interact together or with other modulations to generate other patterns of peak oscillation frequencies (Matthews, 1993).

As for mechanical oscillations, peripheral stretch reflex oscillation frequencies are lowered with mechanical loading (Berthoz and Metral, 1970) because a high inertia results in a longer delay from production of a movement to detection by afferent receptors.

Central oscillations

Despite this potential for peripheral mechanisms to generate synchronized oscillations, other evidence exists to argue against their importance and, by inference, for the importance of synchronization by CNS oscillations. Studies on deafferented patients, who therefore lack feedback loops and the possibility of reafference of mechanical tremor, reveal that postural tremor is preserved (Marsden et al., 1967b). Moreover, during certain tremor studies involving compliant finger muscle contractions there is no effect of loading, stiffness or anaesthesia on the frequency of synchronized 10-Hz unit activity (McAuley et al., 1997). As described above, these factors would be expected to change oscillations based on feedback loop or mechanical resonances. Finally, Vallbo and Wessberg have found that apparently smooth, slow, controlled finger movements are modulated by a motion tremor consisting of regular pulses fixed at the physiological tremor frequency of 8–10 Hz (Fig. 5) (Vallbo and Wessberg, 1993). These pulses are of large amplitude, unaffected by changes in finger velocity or by loading and have a timing inconsistent with the timing of reafferent impulses (Wessberg and Vallbo, 1995).

Fig. 5

(A) Slow controlled inertia-less index finger movements at constant velocity (here without visual feedback) show regular series of pulses separated by around 10 ms easily visible in velocity and acceleration traces and just visible in the position traces. (B) Spectral analysis shows that the pulses occur at a fixed frequency around 10 Hz despite different overall velocities of finger motion. From Vallbo and Wessberg, 1993, with permission.

Given this evidence for a central origin of physiological tremor, one might expect to find physiological tremor correlates on direct CNS recording. In animal studies, the administration of harmaline, which enhances cerebral 10-Hz range oscillations, results in a corresponding recordable oscillation propagated down to spinal interneurons (Llinás and Volkind, 1973). However, in the human, most studies, dating back to Lindqvist (Lindqvist, 1941), reveal no correlation between EEG alpha rhythm frequencies and physiological limb tremor. Some studies have used coherence analysis of MEG to demonstrate synchronized 10-Hz range oscillations between different motor cortical areas, but this synchronized oscillation, unlike those at other frequencies (see later), was not propagated to the periphery. It is nevertheless possible that subcortical central oscillations could contribute to physiological tremor while remaining undetected by human recording techniques.

Summary of physiological tremor genesis

There clearly remains considerable uncertainty regarding the central origin of physiological tremor, since different workers report the importance of a variety of different mechanisms (see reviews by Marsden, 1978; Freund, 1983) and since there is a lack of direct demonstration of a corresponding central oscillation. However, many of the apparent contradictions in the literature can be resolved by taking into account the fact that tremor has been recorded under many different conditions (Table 1). Thus, physiological tremor, using a rather broad definition of the term, may be subdivided into the following categories, each category having relatively different contributions from the different peripheral and central processes that can generate such oscillations.

Rest tremor

At rest, the small amplitude tremor that occurs is unlikely to be neurogenic since, by definition, there is no neuromuscular activity. It has been attributed to the ballistocardiogram (Yap and Boshes, 1967; Marsden et al., 1969a); small mechanical perturbations from the arterial pulse set up an oscillation that is presumably perpetuated at 10 Hz by peripheral passive mechanical resonances.

Postural tremor

The 10-Hz tremor of the extremities when the limbs are outstretched (postural tremor) has been described as a mechanical resonance (Lakie et al., 1986; Amjad et al., 1994) because the frequency changes with loading or as a peripheral stretch reflex loop resonance (e.g. Halliday and Redfearn, 1956; Hagbarth and Young, 1979; Sakamoto et al., 1992). It is perhaps likely that postural tremor is multifactorial in origin because patients with neurological lesions resulting in deafferentation (removing the contribution of reflex loops) are found to have preserved but less sharply `tuned' 10-Hz range tremor (Marsden, 1967b). There may also be a component due to transmission from a more `active' tremor of proximal muscles maintaining the posture (Marsden et al., 1969a).

Isometric contraction

Tremors at around 10 Hz occurring during active contraction will clearly be associated with EMG activity and when of low amplitude, as described above, may be ascribed to the activity of single motor units (Freund and Dietz, 1978). Central studies comparing CNS and peripheral oscillations have often looked at isometric tremor, which may partly explain why central 10-Hz range oscillations have not been found to be manifest peripherally.

Higher amplitude tremors

When there is increased tremor amplitude, whether on posture or active contraction, unit synchronization from an external source appears to underlie this increased amplitude. Stress may increase tremor amplitude via peripheral β-adrenoceptors (Marsden et al., 1967a), which cause both a shortening of unit twitch times resulting in reduced fusion, and an enhancement of afferent feedback resulting in increased synchronization of units (Hagbarth and Young, 1979). Fatigue may also increase tremor amplitude by increased unit synchronization. Thus, while 10-Hz range physiological tremor may be explained in part by unit firing rates, external synchronization accounts for increased tremor amplitudes observed on strong, stressed or fatigued contraction.

Compliant contractions

The strong dampening effect on tremor of attaching the body part to a force transducer and measuring isometric force tremor can be eliminated by studying compliant (elastic) active contractions. Such contractions reveal strong synchronization, as shown by measured amplitudes being too great to be generated by single motor units and by demonstration of corresponding multi-unit EMG discharges (McAuley et al., 1997). The lack of change in frequency on limb anaesthesia or on applying inertial or elastic loads in this study indicated that the synchronization was central in origin. However, other studies using such experimental manipulations (Joyce and Rack, 1974; Matthews and Muir, 1980) revealed a shift in tremor (and EMG) peak, indicating a mechanical or stretch reflex origin. Unfortunately, most studies on peripheral oscillations were performed under isometric conditions, preventing experimental alteration of the mechanics of the body part, so there are no other studies that resolve this issue. Perhaps it is likely that synchronization in physiological tremor is multifactorial, with different relative contributions according to the subject, the task performed and the mechanical set-up.

Motion tremor

The relatively close control exerted by the CNS during movement as opposed to maintenance of posture could make tremors on motion especially suitable for study of peripheral manifestation of central rhythmic activity. The large amplitude 8- to 10-Hz tremor pulses described by Vallbo and Wessberg on slow finger motion illustrate this point and strongly indicate the presence of a CNS rhythm driving this activity (Vallbo and Wessberg, 1993).

Pathological tremors in the physiological tremor frequency range

Different pathological tremors (defined as those tremors that impair motor performance) appear likely to arise through a variety of different mechanisms, but they may be grouped into discrete categories according to their frequency, and it is possible that those tremors of similar frequency may have similar modes of generation (Fig. 6). While certain tremors, such as parkinsonian tremor and cerebellar kinetic tremor, occur in a 3- to 6-Hz range and are associated with marked abnormalities of motor control, others are more conveniently considered with physiological tremor since they occur at a similar frequency and are not associated with any other clinical deficit. This similarity could indicate that they arise as a distortion or amplification of mechanisms generating physiological tremor rather than from a completely different source.

Fig. 6

Speculative scheme whereby various central oscillations and other central lesions might interact with peripheral mechanisms to produce pathological tremor in the physiological 10-Hz frequency range and at lower and higher frequencies. Dashed lines and boxes indicate those mechanisms that are most uncertain. Peripheral mechanisms refers to those processes where the oscillatory rhythm arises peripherally, even though the primary abnormality is central. For example, a delayed antagonist burst resulting from an abnormality of cerebellar processing might act peripherally by underdamping of the body part and thus increasing its mechanical oscillatory tendency. Dystonic tremors could also arise peripherally by the increased muscle tone changing the resonant properties of those body parts.

Enhanced physiological tremor

This is best classified as `physiological tremor of pathological amplitude' (Young et al., 1975) and so, as mentioned above, factors that result in increased synchronization of motor units at the 10-Hz tremor frequency, such as stress, hyperthyroidism, ageing and fatigue, will lead to such tremor.

Essential tremor

Essential tremor bears similarity with physiological tremor in that their typical frequencies overlap. There is no additional clinical neurological deficit associated with essential tremor, but one neurophysiological study has suggested subtle deficits of motor control, such as asymmetries of ballistic movement profiles resulting from an abnormal delay in the timing of the second agonist burst (Britton et al., 1994). This would indicate that CNS structures, possibly the cerebellum, may be involved in the generation of essential tremor (Lamarre, 1975), as do the findings that it can be abolished by a suitably placed thalamic lesion (which could interrupt cerebellar inputs) (Ohye et al., 1982) and that abnormal cerebellar activation is found on functional imaging (Brooks et al., 1992).

A central oscillation in the 10-Hz range could modulate descending motor pathways to limb muscles and thus drive essential tremor by a process akin to that proposed for physiological tremor. A search for CNS oscillations of physiological and esssential tremor frequency in animal studies reveals that olivocerebellar neurons have a natural and harmaline-enhanced tendency to oscillate at 7–12 Hz (Llinás, 1991; Welsh et al., 1995). This oscillatory activity is found to become synchronized across groups of olivary neurons by means of electrotonic coupling via gap junctions. It is tempting to conjecture that such a system could contribute to the central component of 10-Hz range physiological tremor; these subcortical oscillations would of course not be detected on human EEG or MEG recordings. Possibly, a subtle abnormality of this oscillation, such as a slightly lower olivary oscillation frequency, an abnormal influence of deep cerebellar nuclei or abnormal GABA-mediated electrotonic olivary coupling, results in essential tremor.

However, there is other evidence to indicate that, like physiological tremor, essential tremor may be derived, at least in part, from peripheral oscillations. The phase of essential tremor oscillation is relatively easily reset by peripheral perturbations (Lee and Stein, 1981), suggesting that it could arise through some abnormality in reflex loops rather than a fixed pathological CNS oscillation. The subtle CNS mistiming of motor commands suggested by Britton and colleagues could generate tremor without the need for a driving central oscillation by resulting in overshoots after small corrective movements during posture, which could then become amplified and perpetuated by the normal peripheral feedback instability at around 10 Hz (Britton et al., 1994).

Pathological tremors at other frequencies

While pathological tremors that occur at frequencies dissimilar from that of physiological tremor may not share the same mechanisms of generation, invasive animal and human CNS recordings of pathological tremor can nevertheless provide insight into the genesis of a central oscillation at the cellular level and its manifestation in the periphery. In addition, since motor deficits are closely associated with some pathological tremors, study of the latter may provide clues to the role of central oscillations in normal motor control and how their derangement may led to these deficits.

It must first be considered, however, that like physiological tremor and possibly essential tremor, some pathological tremors could in fact arise peripherally. In a manner analogous to physiological tremor, individual motor units generating an abnormal interference pattern may be responsible for pathological tremor. As already described, minimum motor unit firing rates are normally set at around 8 Hz. This threshold level is convenient because units firing at slower rates would be completely unfused and tend to result in discrete individual twitches or fasciculations rather than summating to enable the generation of appreciable force by the whole muscle. In cerebellar kinetic tremor and in parkinsonian tremor, units can fire steadily at much lower rates of around 3–4 Hz and therefore create a tremor of completely unfused motor units at this frequency (Dietz et al., 1974). However, such 3- to 4-Hz discharges often in fact consist of couplets or groups of units firing synchronously, indicating that, as in higher amplitude physiological tremor, the rhythm derives from an external synchronization.

As mentioned for essential tremor, disordered organization of motor commands could produce a synchronized pathological tremor without a direct central progenitor by generating mechanical instabilities that enhance mechanical resonances or feedback loop resonances (Fig. 6). Relatively small increases in feedback loop gains (including central reflex loops) are capable of inducing a large amplitude tremor (Prochazka and Trend, 1998). The delayed antagonist activation that is shown to occur in cerebellar tremor, possibly arising through impaired nucleus interpositus activity (Murphy et al., 1975), could set up a mechanical instability (Diener and Dichgans, 1992); the late antagonist activation would fail to dampen the agonist's action effectively, setting up an inappropriate late overshoot that could become perpetuated by reflex loops into a repetitive oscillation. A similar mechanism has been suggested to contribute to the `sensory ataxic' tremor of benign paraproteinaemic neuropathy (Bain et al., 1996).

There is, nevertheless, much evidence to support the existence of central oscillations producing pathological tremors. The amplitude of pathological oscillations is too large to be explained solely by single motor units and indicates the presence of a synchronizing mechanism. Such synchronization has been directly demonstrated on EMG recordings within a muscle and even between antagonistic muscle pairs. Moreover, both cerebellar tremor (Gilman et al., 1976) and parkinsonian rest tremor (Pollock and Davis, 1930) are still present after deafferentation, suggesting that in these tremors synchronization arises, at least in part, directly from the CNS rather than by peripheral feedback oscillations or mechanical resonances.

Central oscillations in parkinsonian tremor

Direct CNS recordings, both in animals (Poirier et al., 1966; Lamarre, 1975; Lamarre and Joffroy, 1979) and in patients undergoing stereotaxic surgery (Jasper and Bertrand, 1966; Ohye et al., 1974; Rothwell et al., 1995), suggest that in parkinsonism the central rhythmicity arises from spontaneous 3- to 6-Hz oscillatory activity in the thalamus, probably in the ventro-oralis anterior (Voa) nucleus, because the activity here is specifically correlated with tremor activity rather than also with sensory inputs resulting from passive movement of the peripheral structures (Llinás and Pare, 1995). Therapeutic lesioning of the nucleus ventralis intermedius (Vim) (which inputs to the Voa) is effective in alleviating parkinsonian tremor (Narabayashi, 1982). Dopaminergic depletion in Parkinson's disease results in increased activity and disturbed firing patterns of neurons in the globus pallidus internus, a structure which normally has an inhibitory influence on the thalamus. The hyperpolarization generated by increased thalamic inhibition may specifically excite low-threshold calcium currents in thalamic neurons, which results in a tendency for 3- to 6-Hz oscillations (Pare et al., 1990). There are reciprocal connections between different thalamic areas via the thalamic reticular nuclei and these loops may tend to amplify the oscillations and synchronize them between different thalamic neurons in an analogous manner to the synchronization of olivocerebellar oscillations by gap junctions (Steriade et al., 1991; Jeanmonod et al., 1996). The thalamic nuclei output to the premotor and motor cortex, and so in parkinsonism the synchronized modulations generated by altered basal ganglial input to the thalamus may ultimately be transmitted to motor programmes and to the descending commands to muscles.

Central oscillations in cerebellar tremor

Cerebellar tremor also occurs at around 3–6 Hz and results from disruption of the cerebellum or cerebellar outflow, particularly from the nucleus interpositus and perhaps the dentate nucleus (Cooke and Thomas, 1976). Since many cerebellar inputs pass to the thalamus, it is possible that thalamic disinhibition—arising from loss of basal ganglial inputs, cerebellar inputs or even spinal inputs in sensory ataxic tremor—could be a common mechanism involved in the generation of tremors of 3- to 6-Hz frequency (Fig. 6). This possible overlap between parkinsonian and cerebellar tremor is perhaps illustrated by the `rubral tremor' that combines the features of rest and intention tremor and results from midbrain lesions to cerebellorubrothalamic pathways (Marsden, 1984) or from other subcortical lesions (McAuley et al., 1998). Alternatively, it has been suggested that cerebellar tremor could arise centrally through separate cerebellar circuit resonances rather than by a thalamic mechanism (Tsukahara et al., 1983).

Primary orthostatic tremor

A completely different 16-Hz range centrally originating oscillation appears to exist in primary orthostatic tremor, a rare condition that occurs mainly in the legs during postural muscle activity (Heilman, 1984; Thompson et al., 1986; Wills et al., 1994). Single motor unit studies show that the 16-Hz oscillation is not an innate abnormal motoneuronal rhythm, but the result of synchronizing of motor units by an external oscillation (Deuschl et al., 1987), and cross-correlation analysis shows that the synchronization of motor units also occurs between different muscles (Britton et al., 1992). This synchronization has a phase offset, indicating that the bursts in different muscles are not simultaneous, but have relatively fixed delays relative to one another (McAuley et al., 2000a). The lengths of delays do not reflect the various conduction times from the brain to the muscles, but instead follow a more complex pattern that depends on the particular posture adopted by the patient.

Higher frequency limb `tremulous' oscillations in normal individuals

Study of pathological tremors clearly reveals that central oscillations other than those in the 10-Hz physiological tremor range can potentially become manifest in the periphery as tremor. It is thus possible, given the broad spectrum of EEG and MEG oscillatory activity in normal individuals, that peripheral physiological oscillations at frequencies different from 10-Hz physiological tremor could similarly arise from central oscillations over this broad frequency range.

Recently, a peripheral external rhythmic synchronization of motor units has been found at 16–32 Hz (Farmer et al., 1993b), much higher than physiological tremor frequencies and broadly corresponding with the beta EEG band. This motor unit modulation is revealed by coherence analysis comparing the firing of pairs of individual finger muscle units during weak isometric contractions and is corroborated by cross-correlograms showing a central peak and smaller secondary peaks around 40 ms apart (Fig. 4). The individual units (analysed by auto-correlograms) fire at around 10 Hz. Thus, although the timing of firing is such that a group of units tends to fire together in a rhythm around 20 Hz, each individual unit only contributes to this pattern intermittently so that their overall rate is only 10 Hz.

Strong evidence suggests that a central oscillation generates this 20-Hz range synchronization. It spreads between the two hands in patients with mirror movements arising from central lesions that result in branching of descending corticospinal pathways and is disrupted after other CNS lesions that interrupt such pathways (Farmer et al., 1993a). It can be influenced by cortical magnetic stimulation in a stimulus intensity dependent manner (Mills and Schubert, 1995). Finally, unlike any studies on physiological tremor frequency oscillations, the peripheral oscillation is itself directly correlated with cortical rhythmic activity as measured directly over the motor cortex by MEG (Conway et al., 1995; Baker et al., 1997).

Such findings appear to be in contrast with the multi-unit 10-Hz range EMG synchronization generating force tremor, described on strong and `tremulous' contractions by Elble and Randall (Elble and Randall, 1976). In the latter condition, as mentioned earlier, units fire at up to 20 Hz contributing to a broad frequency interference pattern, while an external oscillation at 10 Hz drives groups of units together (Fig. 2A). In the condition described by Farmer and colleagues, units fire relatively slowly around 10 Hz, but because of sharing of common rhythmic inputs they also receive a statistical modulation by an external oscillation in the 20-Hz range (Farmer et al., 1993b) (Fig. 2B). The apparent conflict is perhaps best resolved by considering that mean firing rates will change with strength of contraction and that different external rhythmicities may dominate in different circumstances.

It also seems paradoxical that there has been great difficulty in demonstrating a link between easily apparent peripheral physiological tremor and central alpha range oscillations, yet there is such a clear manifestation in the periphery of central oscillations at a completely different frequency. Do these higher frequency oscillations merely represent a subtle statistical influence on EMG timings or could they actually contribute to movement? In other words, is physiological tremor, visible at 10 Hz and recorded at these frequencies, actually accompanied by additional components at higher frequencies not visible to the naked eye?

Finger muscle tremors in the 20-Hz range rather than just at the 10-Hz physiological tremor frequency have indeed been reported (Sakamoto et al., 1992; Amjad et al., 1994; McAuley et al., 1997), and recently it has been shown directly that the 20-Hz EMG central oscillation modulating mirror movements, as described by Farmer and colleagues (Farmer et al., 1993a, b), is also manifest as tremor at this frequency (Mayston et al., 1999).

An examination of older literature reveals descriptions of peripheral oscillations of even higher frequencies at around 40–50 Hz, which correspond to the gamma EEG range. Piper first described rhythmical bursts of EMG signals at around 50 Hz while recording from steadily contracting muscles (Piper, 1907), and findings of similar synchronized EMG activity have been confirmed by many studies on different muscles under different conditions (Adrian, 1925; Fex and Krakau, 1957; Komi and Vitasalo, 1976; Hagbarth et al., 1983; Bruce and Ackerson, 1986; McAuley et al., 1997). Adrian originally suggested that the EMG synchronization underlying the Piper frequency oscillation arises from the peripheral mechanical properties of the body part rather than from the CNS (Adrian, 1925). However, microneurographic recording of afferent nerves has revealed no such rhythmic behaviour correlated with EMG Piper rhythms (Hagbarth et al., 1983), suggesting that these EMG rhythms may indeed originate centrally. It is possible that the central Piper generator corresponds with a synchronized 40- to 50-Hz MEG oscillation that has been recorded from the motor cortex during certain manual tasks (Conway et al., 1995).

Another technique that has been used to investigate rhythmic muscle activity is acoustomyography (AMG), which is the recording of vibrations of the muscle belly. Since muscle contraction is triggered by EMG activity, peak frequencies of multi-unit oscillations in EMG might well be associated with similar frequency peaks in muscle vibrations which may suffer less mechanical interference than in the tremor signal recorded at the end of the limb (P. A. Merton, personal communication). The first description of muscle vibrations actually dates back to 1665 when Grimaldi (in Orizio, 1993) listened for these vibrations propagated as sound waves. Such findings can be easily repeated by listening through a stethoscope placed over a contracting muscle belly [Wollaston, 1810 (in Orizio, 1993)], whereupon a low rumbling at the lower limit of human hearing (around 40 Hz) is heard. Gordon and Holbourn subsequently showed that the vibrations were mostly due to the contractile activity of muscle motor units (Gordon and Holbourn, 1948). However, there is much debate about the validity and origin of these muscle vibrations, with some workers considering them not to be neuromuscular but instead to arise directly from the muscle (Curtin et al., 1974), and most recent quantitative studies have only found peaks of vibration in the 10-Hz physiological tremor range (Rhatigan et al., 1986; Wee and Ashley, 1989; Rouse and Baxendale, 1991; Ebrahimi-Takamjani and Baxendale, 1994). Nevertheless, some studies have revealed higher frequencies (Oster and Jaffe, 1980; Orizio et al., 1990) and one group has correlated vibrations at 20–100 Hz in a variety of strongly contracting muscles with EEG at this frequency, implying a central origin (Keidel et al., 1990).

Despite this evidence for higher frequency range oscillations that may arise from components of EEG activity in the beta and gamma bands, a clear demonstration of tremor at these frequencies is clearly lacking from most standard physiological tremor studies. On the basis that this lack of demonstration of higher frequency tremors might simply arise because the mechanical properties of muscles when under inertial or fixed loading results in severe dampening of higher frequencies, steady compliant muscle contractions against an elastic load have been investigated (McAuley et al., 1997). Elastic loading changes the mechanical properties to favour detection of higher frequencies of tremor and requires a strong yet finely controlled muscle activation. In these circumstances, in addition to a 10-Hz EMG oscillation and corresponding AMG (muscle vibration) and physiological tremor oscillation, there are also peak oscillations of EMG, AMG and tremor in the 20-Hz range and in the 40-Hz Piper frequency range (Figs 7 and 8). The frequency of all three peaks is unchanged by experimental manipulations that would alter mechanical oscillations or those derived from peripheral feedback, suggesting that they may all have a central origin. Thus, the long-described rumbling sounds heard at around 40 Hz on listening to contracting muscles appear to correspond to the centrally derived EMG Piper rhythm and in turn comprise an `invisible' but not `inaudible' component of tremor.

Fig. 7

Power spectral estimates and coherence analysis of a record made during 50% maximum voluntary contraction of 1DI against an elastic load. Clear peaks are seen at 10, 22 and 41 Hz in both the tremor power spectrum as measured by an accelerometer (A) and the simultaneously recorded rectified surface EMG power spectrum (B). Coherence analysis between these records (C) reveals broad peaks of coherence in all three frequency bands. The upper horizontal line is the 95% confidence level at which the coherence at a certain frequency is significantly greater than the mean coherence over the whole spectrum and the lower line is the 95% confidence level for non-zero coherence (0.025 for 120 blocks). The phase plot (D), indicating the phase difference between the two signals at different frequencies, shows a constant linear relation at the coherent frequencies. The linear slope represents a constant time lag of the tremor acceleration of 6.5 ms behind the EMG for all three of the frequency peaks. From McAuley et al., 1997, with permission.

Fig. 8

Similar recording and analysis of tremor and simultaneously recorded 1DI AMG. The latter was measured by a heart sounds microphone placed directly over the muscle belly of 1DI. The power spectra of finger acceleration (A) and muscle AMG (B) show similar peaks of activity in the 20- and 40-Hz bands; there is a suggestion of a small peak at 10 Hz in the AMG spectrum corresponding to a large peak in the tremor spectrum. The coherence spectrum between acceleration and AMG (C) also shows a 10-Hz peak, indicating that the doubtful AMG peak represents the same oscillation as that picked up by the accelerometer. The heart sounds microphone is not as sensitive at 10 Hz as at 40 Hz, which may explain the low power of the 10-Hz AMG peak. From McAuley et al., 1997, with permission.

Motor oscillations of other systems

Studies on rhythmic activity in the periphery have not been limited to the limbs. In fact, the relatively simple motor control and neural organization of certain other motor structures makes them good candidates for investigation of peripheral manifestations of central oscillations.

In the respiratory system, as already described, medullary inputs are modulated by strong oscillations at 60–120 Hz. This has been demonstrated in animals by cross-correlation analysis of motor nerve activity and comparison with direct medullary recording (Kirkwood et al., 1982), and in the human by surface intercostal or diaphragmatic muscle EMG recordings. Coherence analysis between such EMG activity in different respiratory muscles reveals a peak synchronized frequency at 60–120 Hz during breathing, but not during voluntary movements involving these muscles (Bruce and Goldman, 1983; Bruce and Ackerson, 1986; Smith and Denny, 1990).

The relatively stereotyped and easily quantifiable nature of eye movements makes the oculomotor system another promising candidate for further study of the rhythmic modulation of motor activity. Some reports have addressed this issue, looking for such rhythms in ocular tremor, in saccade timing and during smooth eye movements.

It has been proposed that a component of EEG 8- to 12-Hz alpha rhythm is related to a similar frequency eye movement tremor occurring during ocular fixation in the dark (Lippold and Novotny, 1970; Reiman et al., 1974) and to the timing of fixation saccades (Gaarder et al., 1966). However, the link between eye oscillations and alpha rhythm is not substantiated. Other recordings of ocular microtremor during fixation have revealed periodic bursts occurring at around 100 Hz with lower frequency 10-Hz range oscillations present only in patients with brainstem pathology (Abakumova et al., 1975). Like previous studies on limb tremor, the 100-Hz bursts were thought to reflect the firing of individual motor units, the higher frequency in the eyes reflecting the much higher maximum sustained firing rates of oculomotor muscle fibres. Small peaks in ocular tremor power spectra at 40 and 80 Hz have been described by Bengi and Thomas (Bengi and Thomas, 1968), but these may be artefacts of the recording technique or a mechanical resonant frequency of the eyeball (Thomas, 1967; Boyce and West, 1968). In any case, fixation tremor is uncorrelated between the two eyes (Riggs and Ratliff, 1951), indicating that it is unlikely to be derived from oscillations in higher-order CNS structures.

In addition to the above studies on ocular fixation, rhythmic activity has been investigated during large-scale eye movements. Saccades occurring in regular rhythms with a period of around 200 ms have been demonstrated during certain artificial open-loop conditions (Young and Stark, 1963), during the fast phases of optokinetic nystagmus (Cheng and Outerbridge, 1974) and during predictive tracking of intermittently concealed targets (McAuley et al., 1999a). However, it is not clear whether this rhythmicity represents an inherent CNS `clock' (Westheimer, 1954; Wheeless et al., 1966) or simply a stochastic process related to thresholds for generation of saccades (Robinson, 1973; Carpenter, 1981, 1988). Investigation of saccadic latencies during express saccade experiments (Saslow, 1967; Fischer and Ramsperger, 1984) show a bi- or even tri-modal distribution normally attributed to processing along pathways of different latency. On the other hand, Kirschfeld has proposed that this distribution may in fact be based on a 12-Hz rhythmic central oscillator controlling saccade generation (Kirschfeld et al., 1995).

Smooth eye movements are, by their nature, normally free of any rhythmicity other than that derived from the target's motion. Nevertheless, a low amplitude 3-Hz rhythmic modulation of smooth pursuit eye movements is observed (Robinson et al., 1986) and is thought to relate to the loop time for sampling of visual feedback or sampling of efference copies of prediction-boosted eye movement velocities (Barnes and Asselman, 1991). If smooth movements are `unlocked' from target motion by intermittently obscuring the target, the resulting predictive smooth eye movements that are generated internally by the CNS sometimes reveal an additional rhythmic modulation at a higher frequency of 10 Hz (McAuley et al., 1999a). As expected for an oscillation originating in the CNS and propagated along similar binocular pathways, the same 10-Hz rhythm is present in the two eyes with zero phase offset. It is noteworthy that the smooth eye modulation is reminiscent of the similar frequency CNS modulation of smooth slow finger movements described by Vallbo and Wessberg (Vallbo and Wessberg, 1993).

The role of central oscillations and their peripheral manifestations

Considerable evidence is thus available that 10-Hz physiological tremor of the limb may derive from central, possibly olivocerebellar, rhythmic neural activity, but it is likely that peripheral factors such as motor unit firing rates and reflex and mechanical resonances also have an important contribution. In addition, there also exist peripheral oscillations at other frequencies that may be manifestations of normal cortical oscillations in the beta and gamma EEG wavebands, as well as pathological tremors from thalamic 3- to 6-Hz oscillations and from a 16-Hz oscillation in primary orthostatic tremor. Finally, investigation of other motor systems also reveals rhythmic activity, including respiratory 60-Hz range oscillations in the medulla and a 10-Hz range `physiological tremor' of eye movements.

Despite these observations indicating a widespread occurrence of oscillations in the motor system at different frequencies, there has been a surprising lack of consideration of their function. Those hypotheses that have been suggested are presented.

(i) Function at the peripheral level

Descending commands that consist of synchronized pulses could act to bring motor neurons uniformly close to firing thresholds at the same time so that the motor signal results in a more linear and uniform motor neuron output; additionally, the pulsatile nature of output could act mechanically to help overcome inertial resistances when sudden velocity changes are desired (Greene, 1972).

(ii) Pulsatile CNS motor output

The timing of discrete voluntary movements is sometimes linked to the phase of the ongoing tremor, suggesting that the CNS oscillation modulating the tremor may have a role in the timing of voluntary commands (Goodman and Kelso, 1983). Welsh and Llinás have suggested that a discontinuous (rather than infinitely continuous) timing of motor output could simplify the computational demand of motor actions (Welsh and Llinás, 1997). For example, during certain tasks it would be easier to compute only at 100-ms intervals (producing a 10-Hz oscillation) the appropriate set of commands to generate a movement than to attempt continuous dynamic control and compute commands for every instant of the motion. The resulting segmented movement would still be adequately smooth for most functions (cf. Vallbo and Wessberg, 1993).

(iii) Synchronization and binding

Recording of ensembles of Purkinje cells in awake and active rats reveals that the oscillatory modulation of the firing of groups of inferior olive units may temporarily become specifically linked together during particular phases of a licking task. It is possible that such linking of units might represent a functional co-ordination of particular combinations of muscle actions, suggesting that synchronization of pulsatile or oscillatory motor signals is important and not just their oscillatory nature (Llinás, 1991; Welsh et al., 1995). In other words, ongoing motor control may be fractionated in time into a series of discrete `ballistic' movements and also grouped in space into ever-changing functional muscle collectives.

How would this synchronization actually be useful? Sensory physiologists studying visual cortex 40-Hz range oscillations have suggested that synchronization serves a `feature linking' or `binding' function during sensory information processing (Eckhorn et al., 1988; Gray et al., 1989). Signals that have a synchronized modulation, perhaps applied at an early stage in visual processing, are treated as belonging together, even though their actual firing rates or synaptic connectivities may vary. In this way, as parallel processing occurs in different cortical areas over different lengths of time, the information can still always be identified as being associated with a single visual feature that was present at a certain point in time.

Linking or binding of motor signals to form discrete muscle collectives could be an analogous process (Welsh and Llinás, 1997; Farmer, 1998). If a separate large collection of neurons all with unique synaptic connectivities was required to control every possible task involving unique combinations of muscle actions, the number of units and the computational demand would be enormous (the so-called combinatorial explosion). Instead, smaller groups of neurons responsible for certain muscle actions could be shared so that they would be used for any task requiring that muscle action. Different tasks would therefore utilize different combinations of these shared functional groups; a common synchronized modulation would allow identification of the groups as belonging to a single task. Synchronization of modulations would allow their identification as belonging together as a single command for action at a specific point in time, despite the fact that different signals making up a command may pass along different processing pathways that have different conduction times. When a different task is desired, a different set of neuronal groups within the same overall population become synchronized or `bound' together. Finally, if binding could occur between sensory input and motor output synchronies, this would provide a means of sensorimotor integration.

Studies on rhythms manifest in the periphery are suited to investigation of a possible `binding' role, since the manifestation of central oscillations in different muscles can be measured and related specifically to the motor tasks performed by these muscles. Such exploration centres around the demonstration that (a) different peripheral structures may display the same centrally originating oscillation and (b) this sharing of oscillations is not fixed, but occurs specifically when the structures act in concert during the performance of a certain motor task.

A number of studies have looked at the oscillatory activity simultaneously present in different peripheral structures to see if they reflect a single common modulation. Unfortunately, many human studies do not reveal linking between peripheral rhythms. The tremor oscillations in the left and right hands on posture appear to be independent (Marsden et al., 1969b), as do the 20-Hz range 1DI (first dorsal interosseous) EMG activities of the two hands during simultaneous pinch-gripping (Conway et al., 1995), as well as the EMG oscillations in left and right biceps while lifting a weight in both hands (Bruce and Ackerson, 1986). Coherence analysis between the 10-, 20- and 40-Hz range surface EMG oscillations of different simultaneously contracting hand muscles (McAuley and Brown, 1995) and between oscillations in a variety of different proximal muscles analysed during postural activity (McAuley et al., 2000a) also indicates they are independent, once one has excluded cross-contaminating spread of signals. (Coherence is a normalized measure not dependent on signal amplitude so that a tiny cross-contaminating signal may nevertheless lead to strong artefactual coherence.) Perhaps linking of oscillations is absent in these studies because the tasks performed do not sufficiently require functional linking to engage mechanisms involving a common `binding' of the central signals.

In contrast, coherence analysis between single motor units of neighbouring small hand muscles contracting together in a grip task reveals a common modulation at 16–32 Hz (Farmer et al., 1993b). However, the sharing of this central rhythmic modulation of motor unit firing could relate to a `hard-wired' branching of single corticospinal axons to different but closely related muscles rather than to a reversible widespread linking of oscillations in the brain. In other words, the linking could be due to branching of corticospinal tracts from a highly localized central oscillation. Unlike human surface EMG coherence studies, multi-unit EMG correlations in the monkey bear more similarity to the single unit data in that they reveal a 20-Hz range coherence peak specifically during the hold phase of pinch-grip tasks (Baker et al., 1997). Respiratory muscle EMG recordings provide a clear demonstration of linking between the oscillations of different peripheral structures (Bruce and Ackerson, 1986; Smith and Denny, 1990). This linking is again found to be task-specific in that it occurs during breathing, but not during non-respiratory activity such as speech.

Direct cortical recordings in animals suggest that linking may occur between oscillations in cortical areas controlling different peripheral structures (Murthy and Fetz, 1992, 1996; Nicolelis et al., 1995). The linking appears to be task specific in that 25-Hz range oscillations in the monkey only spread over the cortical surface during the performance of complex motor activity, but remain localized if simple over-learnt movements are made (Murthy and Fetz, 1992, 1996).

The finding of 10-Hz range peripheral oscillations modulating smooth anticipatory eye movements (McAuley et al., 1999a) similar to those modulating smooth slow finger movements (Vallbo and Wessberg, 1993; Wessberg and Vallbo, 1995) provides an excellent opportunity to investigate linking between more complex hand–eye co-ordinated movements. Previous studies have already suggested high-level interactions between the visual system and physiological tremor (Merton et al., 1967; Sutton and Sykes, 1967; Dymott and Merton, 1968) and there would be no problems due to cross-contamination of signals or hard-wired branching between such dissimilar and spatially separate motor systems. Linking does indeed occur between the eye and finger oscillations specifically when they move together in tracking a visual target and never when the eye and finger move independently (McAuley et al., 1999b) (Fig. 9). Like the eye oscillations themselves, the linking is not always manifest during simultaneous tracking. This somewhat variable presence, like the variable presence of many other synchronized tremors, could indicate (a) that the oscillations are simply an epiphenomenon superimposed on related motor signals, (b) that it is only the peripheral manifestation that is inconstant or (c) that linking of the rhythmic activity superimposed upon motor commands is only one potential strategy for hand–eye co-ordination.

Fig. 9

Traces of a short section of simultaneous eye and finger tracking of a horizontal target sinusoid, which is intermittently obscured during certain phases of the cycle to bring out smooth anticipatory eye movements. Target, eye and finger positions and corresponding velocities are shown. Modulations of a period around 120 ms are visible in both the eye and finger velocity traces. A constant phase relationship between the two modulations is suggested by the dotted vertical lines aligned with the eye modulation peaks; each finger peak leads the corresponding eye peak by roughly 40–50 ms. The head acceleration recording shows that the shared oscillation is not an artefact resulting from head tremor. The common oscillations apparent in these traces are corroborated by coherence analysis performed over the whole trial. From McAuley et al., 1999b, with permission.

(iv) Frequency and phase coding

The particular frequency of oscillation could in itself be important in motor control, as well as the fact that synchronized pulses enable the formation of muscle collectives (McAuley et al., 1997). Since the frequencies of oscillation are generally found to be rather variable, such labelling might be most appropriate for distinguishing separate muscle collectives and for identifying different classes of task rather than for uniquely identifying a certain task. For example, when a single peripheral structure (with a single area of representation on the motor cortex) is simultaneously involved in two separate tasks such as posture and finely controlled motion, two separate frequency `wavebands' could code for the two types of task.

While linking by synchronized firing has already been clearly demonstrated centrally by olivary recordings of rats during licking activity (Welsh et al., 1995), the 6-Hz frequency of this synchronized modulation is directly tied to the lick frequency, indicating that the pattern could merely reflect that different units fire at different phases of each lick. To investigate the importance of the frequency of the oscillation as well as the synchronization that it produces, attention should be directed at motor activity that is not inherently rhythmic (i.e. where the CNS progenitor of the oscillation may be coding for rather than directly generating the activity). Demonstration of multiple and changing oscillation frequencies would provide evidence for a frequency coding process, especially if such changes were task-specific.

The 20-Hz range EMG activity that is correlated with a similar MEG oscillation tends to occur mainly during the hold phase of human (B. A. Conway, D. M. Halliday, S. F. Farmer and J. R. Rosenberg, personal communication) and monkey (Baker et al., 1997) pinch-grip tasks. During the movement phase, 10-Hz correlated EMG activity is more dominant (cf. Vallbo and Wessberg, 1993), although this has not been correlated with similar MEG activity. Localized 20-Hz range MEG rhythms in the human rolandic area, existing in conjunction with 10-Hz rhythms, were also found by Salmelin and Hari during, or in preparation for, simple thumb movements, although corresponding peripheral oscillations were not investigated (Salmelin and Hari, 1994). A study on more complex visually guided finger movement tremor also shows frequency changes dependent on the task performed (van Galen et al., 1990).

Taking these findings of shifts in oscillation frequency together with other disparate studies performed under different conditions that highlight different tremor or EMG frequencies, it may be hypothesized that the three ranges of normal peripheral oscillations at around 10, 20 and 40 Hz could all have a role relating to frequency-coding of signals during the processing of commands for motor activity. In the 10-Hz range, there is a high-level influence of visual feedback, a task-specific linking between eye and finger tracking and occurrence especially during movement. The 20-Hz range oscillations seem to relate instead to posture, preparation or hold tasks. Finally, the 40-Hz Piper rhythm frequencies could possibly relate to those oscillations that are thought to be involved in sensory information processing, since the latter are also at this frequency. The potential for co-existence of these oscillations is illustrated by their simultaneous presence during compliant finger muscle contractions requiring both maintenance of a steady posture and fine control of finger position (McAuley et al., 1997).

Pathological studies also suggest that different frequencies have specific roles, since in addition to developing abnormal 3- to 6-Hz tremor, parkinsonian patients off medication appear to lack the normal peaks in forearm EMG activity at frequencies above the 10-Hz range (Brown et al., 1997). The Piper rhythm peak EMG and tremor activity during compliant finger muscle contraction similarly diminishes when the patients are off medication (with severe parkinsonian symptoms) compared with when they are on medication (with symptoms relieved) (McAuley et al., 2000b). It is therefore possible that the Piper rhythm has a role in motor control that becomes disrupted in parkinsonism.

The fact that there is often a phase shift between oscillations of a certain frequency in different structures makes it possible that phase as well as frequency information could be used in coding motor signals. CNS recordings of rat hippocampal spatial memory cells reveal they are found to fire regularly, with a phase shift relative to background 7- to 12-Hz EEG that is specific for the animal's spatial location and which changes for different locations (O'Keefe and Recce, 1993). Similarly, peripheral recordings in different muscles of the strong 16-Hz oscillation of primary orthostatic tremor reveal complex yet consistent patterns of relative phase lag that are specific for certain postures (McAuley et al., 2000a).

Conclusions

Since the earliest descriptions of rhythms of the muscles, tremors of the body and oscillations of the nervous system, a vast literature has accumulated on the subject of central rhythmic activity and its physiological and pathological manifestation in the periphery. However, much confusion still surrounds the precise nature and role of this widespread characteristic of neural behaviour, although evidence is accumulating to suggest that oscillations may represent binding or frequency coding of CNS activity.

This review has attempted to illustrate the potential of studying rhythmic CNS activity through investigation of physiological tremor and other peripheral oscillations. Recording peripherally is technically easier, especially for humans in whom complex motor activity is much more conveniently controlled than in laboratory animals. Moreover, functional localization of central oscillations is difficult, often requiring simultaneous animal recordings of cortical activity and pyramidal output, whereas if an oscillation is present in a muscle during its activity, it is more likely to have a role, if any, in the processing of that activity rather than in some completely different task.

However, there are also readily apparent limitations of peripheral recordings. Since only the final descending output is observed, the possibility remains that rhythmic oscillations are either peripheral artefacts or non-functional. Linking of oscillations might merely reflect that common anatomical pathways are involved at some unknown site during a stage of processing of motor commands when such commands become part of a single task. Thus, the way to take investigation of the role of oscillations further may be to combine peripheral and central recordings. Direct recording of oscillations in the brain suggests their central nature and can identify those that are not strongly manifest in the periphery, either because they only weakly modulate the pyramidal cells of the motor cortex or because they are of too high a frequency for peripheral detection. They may also localize oscillations and sites of linking of oscillations to different cerebral structures. Experiments on the periphery initially identify which oscillations are most likely to be functionally important and, if combined simultaneously with central studies, can effectively enhance the `resolution' of central rhythm recording by enabling their association with specific peripheral structures and activities.

Acknowledgments

I wish to thank Dr Simon Farmer and Dr John Rothwell for comments on revisions to the manuscript.

Footnotes

  • Deceased September 28, 1998

References

View Abstract