Organization of the central control of muscles of facial expression in man

Abstract

Surface EMGs were recorded simultaneously from ipsilateral pairs of facial muscles while subjects made three different common facial expressions: the smile, a sad expression and an expression of horror, and three contrived facial expressions. Central peaks were found in the cross-correlograms of EMG activity recorded from the orbicularis oculi and zygomaticus major during smiling, the corrugator and depressor anguli oris during the sad look and the frontalis and mentalis during the horror look. The size of the central peak was significantly greater between the orbicularis oculi and zygomaticus major during smiling. It is concluded that co-contraction of facial muscles during some facial expressions are accompanied by the presence of common synaptic drive to the motoneurones supplying the muscles involved. Central peaks were found in the cross-correlograms of EMG activity recorded from the frontalis and depressor anguli oris during a contrived expression. However, no central peaks were found in the cross-correlograms of EMG activity recorded from the frontalis and orbicularis oculi or from the frontalis and zygomaticus major during the other two contrived expressions. It is concluded that a common synaptic drive is not present between all possible facial muscle pairs and suggests a functional role for the synergy. The origin of the common drive is discussed. It is concluded that activity in branches of common stem last-order presynaptic input fibres to motoneurones innervating the different facial muscles and presynaptic synchronization of input activity to the different motoneurone pools is involved. The former probably contributes more to the drive to the orbicularis oculi and zygomaticus major during smiling, while the latter is probably more prevalent in the corrugator and depressor anguli oris during the sad look, the frontalis and mentalis during the horror look and the frontalis and depressor anguli oris during one of the contrived expressions. The strength of common synaptic drive is inversely related to the degree of separate control that can be exhibited by the facial muscles involved.

All social intercourse is coloured by facial expressions. These may be intentional, but more often they are produced without conscious control. Indeed sometimes facial expression is quite involuntary, as can be seen, for example, in the countless TV out-takes where the presenter cannot control his/her laughter despite many takes and much conscious effort. Another property of facial expressions is their universality. The psychologist Paul Ekman (1993) studied the nature of facial expression by asking subjects to describe photographs of people expressing different emotions. He showed these photographs to people from many different cultures including the isolated Fore foragers of Papua New Guinea. All subjects were able to identify enjoyment, sadness, anger, fear and disgust, and when asked to make facial expressions to depict various scenarios, they were unmistakable. These expressions are therefore termed basic facial expressions. In addition, it has been noted that many facial expressions shown by primates bear striking similarities to human expressions. This has lead to the hypothesis that facial expression might be genetically determined.

Facial expressions are produced by the synergistic or co-operative action of many different facial muscles. One mechanism that could account for such muscle synergy would be that it is produced by the branching pattern of common-stem, last-order, presynaptic input fibres innervating motoneurones supplying the different muscles, and in this way a hard-wired repertoire of muscle synergies for the different expressions is ‘encoded’. According to this hypothesis, the movement ‘code’ for the different facial expressions is contained in the anatomical distribution of last-order input fibres to motoneurones supplying the different muscles. In addition, the neurones giving rise to last-order presynaptic input fibres from separate sources may receive input from branched axons. Such a ‘hard-wired’ arrangement of facial muscle synergies would account for the fact that the basic facial expressions are universal, and ensure consistency across the population.

To determine whether the distribution of common synaptic input shared between motoneurones innervating different facial muscles could be responsible for the pattern of facial muscle synergy associated with different facial expressions, we have performed cross-correlation analysis of multi-unit surface EMG signals recorded from pairs of facial muscles. The presence of such shared synaptic input activity is indicated by a peak in the cross-correlogram constructed from the times of occurrence of motor unit pulses recorded from the different coactive muscles (Moore et al. 1970; Sears & Stagg, 1976; Kirkwood & Sears, 1978, 1991; Kirkwood et al. 1982). If our hypothesis that motoneurones innervating muscles contributing to basic facial expressions share common drive is correct, then cross-correlograms of EMG activity recorded from muscles involved in making these expressions would be expected to show a central peak. To test this hypothesis we have studied three basic facial expressions; the smile, the sad expression and the look of horror, and for comparison we have also studied three contrived expressions not known to have any subjective significance.

Methods

Multi-unit surface EMG recordings were made from eight healthy volunteers. Not all subjects participated in all recordings. All experiments were carried out with the approval of the Joint University College and University College Hospital Committee on the Ethics of Human Research and with informed consent and conformed to the standards set by the Declaration of Helsinki.

Recordings were made from six pairs of facial muscles on the left-hand side of the face: the orbicularis oculi and zygomaticus, the corrugator and depressor anguli oris and the frontalis and mentalis, which are involved in the basic facial expressions of smiling, sadness and horror, respectively, and the frontalis and zygomaticus, frontalis and depressor anguli oris and frontalis and orbicularis oculi, which were recorded from during contrived expressions. Eight subjects were recorded from in each of the six expressions.

Multi-unit EMG recordings were made using bipolar surface electrodes attached to the skin (recording area 0.75 cm² per electrode, interelectrode distance 0.4 cm). To ensure consistent placing of the electrodes for each muscle, they were placed relative to facial landmarks on the left-hand side of the face as illustrated in Fig. 1. Also shown in Fig. 1 is a superficial dissection of the face to show the facial muscles described and an example of surface EMG recording. EMG recordings were amplified, filtered (−3 dB at 32 Hz and 8 kHz) and stored on magnetic tape (Racal Store 4) for later analysis.

Figure 1. Position of the recording electrodes.

Photographs to show the positions recorded from in each of the three expressions. A, the smile. B, the sad expression. C, the horror expression. D-F, contrived expressions. G, dissection of the face to show each of the facial muscles recorded from. H, sample of a surface EMG recording.

Recordings were made while subjects made voluntary facial expressions, as illustrated in Fig. 1. Subjects were aided in this by looking in a mirror and by viewing an oscilloscope display of EMG activity. Each recording took about 5 min, although the expressions were sustained for about 20 s at a time with short rests in between to prevent fatigue.

Analysis

TTL (transistor-transistor logic) pulse trains were generated from the multi-unit EMG signal using a level detector (NL200 Neurolog, Welwyn Garden City, UK; see Fig. 1). EMG cross-correlograms were constructed on a microcomputer (Tandon 486 with CED 1401 interface; Cambridge Electronic Design, UK), using approximately 5000 pulses per train, bin widths of 1.0 ms and pre/post-trigger sweep periods of 100 ms. A mean bin count was estimated from the first and last 75 bins and used to construct a cumulative sum (CUSUM; Ellaway, 1978) graph. The y-axis of the CUSUM graph was scaled according to Bissell (1984) such that two standard errors of the mean bin count on the y-axis was the same length as one bin width on the x-axis. Therefore, a CUSUM rising at a rate of two standard deviations of the mean bin count per bin will be represented by a graph of gradient 45 deg. This scale causes random fluctuations to appear quite small in relation to any significant features. The CUSUM graphs were used in the identification of central peaks in the EMG cross-correlograms and determining their statistical significance.

The span method (British Standards Institution, 1980) was applied to CUSUMs where peaks were suspected. This method determines the significance of maxima or minima on the CUSUM chart by determining the maximum extent to which the CUSUM deviates from a straight line joining the ends of the sequence within which a change is suspected.

For a CUSUM chart, this maximum vertical height, V_max is derived from the following equation when the original CUSUM calculations are available:

where r is the sample number corresponding to a suspected change point in the sequence i + 1 to j (i.e. i, r and j are successive ‘corners’ on the CUSUM chart). C_r, C_i and C_j are the CUSUM values associated with points r, i and j, respectively. The value m defined in the above formula is the span length. Figure 2 illustrates the span method.

Figure 2. Figure to illustrate the span method for assessing the significance of a central peak on an EMG cross-correlogram.

V_max is the maximal vertical deviation of the CUSUM; C_r, C_i and C_j are the cumulative sum (CUSUM) values associated with points r, i and j, respectively; m is the span length. See Methods.

The V_max is standardized by dividing the value obtained from this formula by the standard error of the observations. The value of the standardized V_max (V_max/σ_e) is then compared with a normogram, which provides a P value for any given span length. The normogram used was derived from page 14 of the British Standards Institution guidelines (1980).

The size of significant central cross-correlogram peaks was determined by using an index described by Harrison et al. (1991): E/M, where E is the number of extra counts in the central peak above the mean and M is the mean bin count. These were used to compare the peaks obtained from various correlograms. A one-way ANOVA test was used to compare E/M scores, and the post-hoc Scheffe test was applied. For the purposes of this test, flat correlograms without statistically significant peaks were given E/M scores of zero. A one-way ANOVA was also applied to the durations of the significant central peaks.

In addition, the individual correlograms from all subjects making the same expression were added together to yield the grand average correlogram for that expression. These correlograms helped to clarify whether a significant correlation is present when only a few of the individual correlograms showed significance. By summing the bins from the separate correlograms, any consistent small central peak would be amplified. CUSUM charts were constructed from these grand average correlograms and the span method described above was applied to the CUSUM charts where central peaks were suspected. For significant central peaks, the E/M ratio was also calculated.

Results

Smile expression

Statistically significant central peaks were present in EMG cross-correlograms obtained from all of the subjects between orbicularis oculi and zygomaticus major recorded during smiling (P < 0.001). An example is shown in Fig. 3A. The mean E/M value was 3.6 ± 0.8 (mean ± s.e.m.; n = 8 subjects) with a mean duration of 14 ± 1.2 ms (n = 8).

Figure 3. EMG cross-correlogram constructed from multi-unit recordings obtained from facial muscle pairs.

Approximately 5000 pulses were generated using a level detector circuit from multi-unit surface EMG signals. The y-axis gives the probability of finding a pulse in the event muscle (second of the pair) in a 1 ms bin at times before and after a pulse occurring at time 0 in the trigger muscle (first of the pair). A-F, EMG cross-correlogram and CUSUM constructed from data recorded from different pairs of muscles during different expressions. A, the orbicularis oculi and zygomaticus major muscles during smiling. A short-duration central peak is present, E/M (where E is the number of extra counts in the central peak above the mean and M is the mean bin count) = 7.2. B, the corrugator and depressor anguli oris muscles during a sad expression. A small central peak is present as indicated by the inflections of the cusum, E/M = 1.0. C, the frontalis and mentalis muscles during a horror expression. Central peak, E/M = 1.1. D, the frontalis and zygomaticus major during contrived expression 1. No central peak is present. E, the frontalis and depressor anguli oris muscles during contrived expression 2. Central peak, E/M = 1.1. F, the frontalis and orbicularis oculi during contrived expression 3. No central peak is present.

Sad expression

Central peaks were present in EMG cross-correlograms from five of the eight subjects (P < 0.05) recording from corrugator and depressor anguli oris during the sad expression. An example EMG cross-correlogram with a central peak is shown in Fig. 3B. The mean E/M value was 0.5 ± 0.2 (n = 8). For EMG cross-correlograms showing central peaks, the mean duration was 19 ± 1.5 ms (n = 5).

Horror expression

One significant central peak was identified out of the eight EMG cross-correlograms between the frontalis and mentalis muscles during a look of horror (P < 0.01). The EMG cross-correlogram containing a central peak is given in Fig. 3C. The mean E/M value was 0.1 ± 0.1 (n = 8). In the EMG cross-correlogram showing a central peak, the peak had duration of 24 ms.

Contrived expression 1 with the frontalis and zygomaticus major

No significant central peaks (P > 0.05) were identified out of the eight EMG cross-correlograms between frontalis and zygomaticus major. An example flat EMG cross-correlogram is shown in Fig. 3D.

Contrived expression 2 with the frontalis and depressor anguli oris

Three statistically significant central peaks (P < 0.05) were identified out of the eight EMG cross-correlograms between the frontalis and depressor anguli oris. An EMG cross-correlogram with a central peak is shown in Fig. 3E. The mean E/M value was 0.4 ± 0.2 (n = 8). For EMG cross-correlograms showing central peaks, the mean duration was 26 ± 5.8 ms (n = 3).

Contrived expression 3 with the frontalis and orbicularis oculi

No significant central peaks were found in any of the eight EMG cross-correlograms between the frontalis and orbicularis oculi (P > 0.05). An example flat EMG cross-correlogram is shown in Fig. 3F.

The results for all six expressions are summarized in Table 1.

Table 1.

Summary of results for all six facial expressions

Facial expression	Muscle pair	Synchrony incidence (n/no. of subjects)	Synchrony index (E/M)*‡	Synchrony durationt†§ (ms) ‡	Half-widths¶ (ms) ‡
Smile	Orbicularis oculi and zygomaticus major	8/8	3.55 ± 0.75	14.25 ± 1.22	4.50 ± 0.53
Sad expression	Corrugator and depressor anguli oris	5/8	0.54 ± 0.21	19.20 ± 1.46	8.00 ± 1.05
Horror expression	Frontalis and mentalis	1/8	0.14	24	19.00
Contrived expression 1	Frontalis and zygomaticus major	0/8	—	—	—
Contrived expression 2	Frontalis and depressor anguli oris	3/8	0.42 ± 0.22	25.67 ± 5.78	9.67 ± 2.40
Contrived expression 3	Frontalis and orbicularis oculi	0/8	—	—	—

^*E/M: E, extra counts in the central EMG cross-correlogram peak; M, mean bin count

^†Synchrony duration, duration of central EMG cross-correlogram peak

^‡Means ±s.e.m.

^§Those EMG cross-correlograms that did not show a central peak were not included in calculation of the duration.

^¶Calculated according to Kirkwood & Sears (1978).

Comparison of EMG cross-correlograms from different expressions

The E/M values obtained from EMG cross-correlograms during the smile were significantly larger than those from any other expression. (ANOVA, P < 0.001, post hoc Scheffé) There were no significant differences between any other expressions. (ANOVA, post hoc Scheffé).

The durations of the significant central peaks on the EMG cross-correlograms obtained from the six expressions were significantly different (ANOVA, P < 0.05). However, the sample size was too small for post hoc analysis.

Grand averages

Since the sad expression, the horrified expression and one of the contrived expressions showed significant correlation in only some of eight subjects, the correlograms from all eight subjects were summed to give grand averages for each facial expression. These can be examined for the presence of significant central peaks using the span method (see Methods) in the same way as for the individual correlograms.

These grand average correlograms are shown in Fig. 4. Four showed significant central peaks: the smile, the sad expression, the horror expression and the contrived expression, between the frontalis and depressor anguli oris (P < 0.05). The other two EMG cross-correlograms were flat (P > 0.05). Table 2 shows the E/M values, durations and half-widths of the grand average EMG cross-correlograms.

Figure 4. EMG cross-correlograms constructed from grand averages of eight multi-unit recordings obtained from facial muscle pairs.

The y-axis gives the probability of finding a pulse in the event muscle (second of the pair) in a 1 ms bin at times before and after a pulse occurring at time 0 in the trigger muscle (first of the pair). A-F, EMG cross-correlogram and CUSUM constructed from grand average of data recorded from different muscle pairs during different facial expressions. A, the zygomaticus major and orbicularis oculi during a smile. A central peak is present as indicated by the inflections in the cusum, E/M = 2.50. B, the depressor anguli oris and corrugator during a sad expression. Central peak, E/M = 0.38. C, the mentalis and frontalis during a horror expression. Central peak, E/M = 0.34. D, the frontalis and zygomaticus major during contrived expression 1. No central peak is present. E, the frontalis and depressor anguli oris during contrived expression 2. Central peak, E/M = 0.31. F, the frontalis and orbicularis oculi during contrived expression 3. No central peak is present.

Table 2.

Results from grand average correlograms

Facial expression	Muscle pair	Central peak†	Synchrony index (E/M)*	Synchrony duration‡ (ms)	Half-widths§ (ms)
Smile	Orbicularis oculi and zygomaticus major	Significant	2.50	14	7
Sad expression	Corrugator and depressor anguli oris	Significant	0.38	28	11
Horror expression	Frontalis and mentalis	Significant	0.34	25	6
Contrived expression 1	Frontalis and zygomaticus major	Not significant	—	—	—
Contrived expression 2	Frontalis and depressor anguli oris	Significant	0.31	19	10
Contrived expression 3	Frontalis and orbicularis oculi	Not significant	—	—	—

^*E/M: E, extra counts in the central EMG cross-correlogram peak; M, mean bin count

^†Central peak, significance at 5% level

^‡Synchrony duration, duration of the central EMG cross-correlogram peak; those EMG cross-correlograms that did not show a central peak were not included in calculation of the synchrony index or duration.

^§Calculated according to Kirkwood & Sears (1978).

Discussion

In this study, a short-duration central peak was shown consistently between the facial muscles orbicularis oculi and zygomaticus major while smiling. The presence of the central short-duration peak in the EMG cross-correlograms taken during smiling indicates a common drive to the two motoneurone pools.

Central peaks were obtained in five subjects in the sad expression, one in the look of horror and three subjects in the contrived expression, between the frontalis and depressor anguli oris. This indicates that there is less consistent common drive to these muscle pairs.

No central peaks were present in either the contrived expression between frontalis and zygomaticus major or the contrived expression between the frontalis and orbicularis oculi.

Possibility of cross-talk between EMG signals

A potential complication with cross-correlation analysis of surface EMG recordings is that correlation may occur as a result of cross-talk. Each surface electrode is intended to pick up the EMG signal from only one muscle. However, if the muscles under study are physically close to one another, each electrode could potentially pick up the EMG signal from both muscles. Therefore, the cross-correlograms constructed from such recordings would show correlation irrespective of any underlying synergy. In this study, it could be argued that consistent muscle synergy was found between the orbicularis oculi and zygomaticus major during smiling as a result of cross-talk due to the proximity of the muscles involved, whereas no consistent synergy was demonstrated in the other muscle pairs because the muscles were situated further apart.

In order to rule out cross-talk, the detection range of a pair of surface EMG electrodes should be considered. Fuglevand et al. (1992) performed simulations of motor unit action potentials recorded from bipolar electrodes at varying distances from the edge of the motor unit territory. They found that the EMG signal rapidly decays with increasing distance from the motor unit, such that for a bipolar electrode (electrode size 4 mm² and electrode spacing 11 mm), recording from a large motor unit (2500 muscle fibres) the EMG signal is 90 % attenuated at 8 mm and approaches background noise (< 40 μV peak to peak amplitude) at 27 mm. The simulation did not account for the larger impedances of subcutaneous tissue and skin, and therefore the authors note that these simulated values might be overestimates. They also found that varying the electrode size did not affect the range of detection, although narrowing the space between electrodes in a pair decreased the detection distance. In the present experiments, electrodes were 75 mm² in area and electrode spacing was 4 mm. When recording from the orbicularis oculi and zygomaticus the bipolar electrodes were approximately 50 mm apart. At this distance such bipolar electrodes are too far apart to be contaminated by cross-talk from the other muscle.

In addition, the recording electrodes for the orbicularis oculi and frontalis in contrived expression 3 were a similar distance apart to the electrodes used in the smile, and none of the correlograms constructed from this pair of electrodes yielded significant central peaks. One would have expected at least some correlation recorded between this muscle pair if the significant correlation for the smile expression is to be attributed to cross-talk.

In a recent study, Kilner et al. (2002) have described a novel algorithm to remove electrical cross-talk between surface EMG signals. In the present study, we have been careful to use bipolar electrodes designed to have a limited range of detection. But in the future, such ‘blind signal separation’ algorithms may be useful when electrode pairs have to be placed closer together on the face than was necessary in our experiments.

Comparison with cross-correlation studies involving other muscle pairs

The duration of the peaks produced while smiling are comparable in size and duration to those obtained by Gibbs et al. (1995) for finger muscles, such as between the first dorsal interosseus and the index extensor muscles. The peaks obtained while making the other facial expressions had longer durations, and lower E/M values. Kirkwood & Sears (1978) hypothesized that the duration of the central peak in an EMG cross-correlogram could be used to determine the nature of the common drive. They proposed that short-duration peaks represented a common-stem presynaptic fibre that innervated both motoneurone pools, whereas longer-duration peaks were more typical of presynaptic synchronization.

Kirkwood & Sears (1978) maintain that only very narrow central peaks (with a half-width of less than 2.1 ms) can be attributed to common-stem presynaptic input with certainty. However, as with the results obtained by Gibbs et al. (1995), it is possible that the longer duration of the peaks obtained in this study are due to the fact that multi-unit EMG recordings were made; differences in pre- and postsynaptic conduction delays might widen the otherwise narrow peak. Therefore, branched last-order inputs might account for at least some of the findings, especially in the smile (half-width = 4.5 ms), where the peak durations were shortest, while the other peaks (half-widths = 8, 19 and 9.7 ms for sad, horror and contrived 2 expressions, respectively) have durations suggesting the presence of presynaptic synchronization.

For finger muscles, the amplitude of the central cross-correlogram peaks varied with the distance between muscles (Bremner et al. 1991a,b). For example, smaller peaks were found when recording from the first and fourth dorsal interossei, which act on the second and fourth fingers, than between the first and second interossei, which act on adjacent fingers. Two hypotheses to explain these findings were put forward. The first was that those muscles that were anatomically close also had motoneurone pools that resided close together in the spinal cord. Therefore, the common input could be envisaged to be the result of the ‘spread’ of last-order inputs directed principally to one motoneurone pool, to both motoneurone pools. The finding that distant muscles are less synchronous could be explained by their having more distant motoneurone pools and therefore less overlap of input. The spread of last-order inputs may therefore represent an inherent imprecision in the capacity of last-order input circuitry to distinguish between individual motoneurone pools. The second, and more attractive hypothesis, supposes that far from being a source of imprecision in the neuronal wiring, common drive to several motoneurone pools can be thought of as conveying a single motor command, such as to spread the fingers in the case of abductor muscles or to open or close the fist in the case of extensors and flexors, respectively. In this way the distribution of last-order branched inputs to motoneurone pools is seen as orchestrating a motor command.

The structure of the facial nucleus

The motoneurone pools controlling the muscles of facial expression reside in a brainstem nucleus called the facial motor nucleus. This nucleus lies in the caudal pontine tegmentum, and contains a musculotopic arrangement of subnuclei, each innervating a small group of facial muscles. Figure 5 shows the arrangement of these subnuclei along with their names and the facial muscles they supply (adapted from Holstege et al. 1984). The diagram is based on the facial nucleus of a cat, so the large dorsomedial subnucleus, which controls auricular muscles, would be much smaller in humans. In addition, the lateral and ventrolateral subnuclei innervating the upper and lower lip muscles would be much more extensive in humans, reflecting the greater need for complex movements in this area for speech and facial expression.

Figure 5. Diagrammatic representation of the structure of the cat facial motor nucleus.

This schematic is based on a diagram by Holstege et al. (1984) showing the five subnuclei: the lateral subnucleus, which innervates the upper lip muscles; the ventrolateral subnucleus, which innervates the lower lip muscles (both larger in humans); the ventromedial subnucleus, which supplies the platysma; the dorsomedial subnucleus, which innervates the ear muscles (probably smaller in humans); the intermediate subnucleus, which innervates the muscles around the eye.

The organization of the facial nucleus has a direct bearing on the hypotheses stated above. If the source of common drive is in the imprecision of inputs to motoneurone pools of the facial nucleus, then those muscles supplied by adjacent motoneurones would exhibit more common drive than those muscles innervated by motoneurones further apart. Therefore, the structure of the facial nucleus should determine the pattern of common drive. If, on the other hand, the source of common drive is a functional arrangement that realizes and coordinates a motor command, then the structure of the facial nucleus should have no relationship to the pattern of common drive. In this case the pattern of common drive should follow the configurations in which the muscle pairs are routinely used.

The results of the present study seem to conform to the latter hypothesis, that common input represents the realization of a motor command rather than imprecision in the distribution of inputs to the motoneurone pools, because all of the muscle pairs studied that produced significant central peaks in the EMG cross-correlogram are innervated by neurones lying in separate subnuclei of the facial nerve nucleus, making overlap unlikely. Furthermore, the muscle pair involved in smiling (the orbicularis oculi and zygomaticus major) have the strongest common synaptic drive, despite their motoneurones residing in different subnuclei. The muscle pair involved in the third contrived expression (the frontalis and orbicularis oculi), however, show no common drive despite at least some of their motoneurones lying in the same subnucleus. This makes the hypothesis that common drive arises from imprecision in the distribution of inputs to the motoneurone pools unlikely.

Is there functional significance to the present finding that there is strong consistent common drive to the muscles involved in smiling and not to those involved in the other expressions? Insight into this question can be gained by supposing that muscles with little or no common drive might be regarded as requiring independent control, whereas in those having a large degree of common drive, co-contraction is obligatory.

From common observation of facial expressions it is clear that the raising of the corners of the lips is invariably interpreted as some kind of smile. There does not appear to be a situation where individuals raise the corners of their lips as part of a completely unrelated facial expression. In this case, it would seem sensible to have branched presynaptic inputs also drive orbicularis oculi, which is another component of the smile. However, when the muscles involved in making the sad expression (corrugator and depressor anguli oris) are considered, these muscles do not only contribute to a sad expression but also to other voluntary facial expressions. The corrugator alone gives a frown, seen when a subject is given a hard task to do, or when he/she is puzzled. Therefore, it is plausible that the reason for the inconsistency and weakness of the common drive found between corrugator and depressor anguli oris during a sad look was due to the fact that these muscles are often required to contract independently or in combination with other facial muscles. Having a common drive would therefore hinder the ability to make these other expressions, whereas it would benefit muscles such as zygomaticus major, which is invariably part of a smile.

However, one problem remains for this hypothesis. If common drive does indeed represent the orchestration of an often-used motor command with functional significance, then why was there some significant correlation in the second contrived expression? Since the other two contrived expressions produced no common drive at all in any of the subjects, this could not represent a background level of common drive between all muscle pairs involved in facial expression; the kind of imprecision of inputs hypothesized above. Furthermore, since the central peaks in the EMG cross-correlograms for this expression were similar in size and duration to those of the sad and horror expressions it raises the possibility that in fact, this muscle pair was involved in a real facial expression. Subjectively, this contrived expression looks the most plausible real expression of the three. Or perhaps this muscle pair might form part of a facial expression that has been overlooked. More experimentation would be required to explore this further.

Innervation of the facial nucleus

Studies involving subjects who have suffered supranuclear (central) lesions (Hopf et al. 1992; Töpper et al. 1995; Trepal et al. 1996; Urban et al. 1998) show two distinct types of impairment of facial expression, voluntary and emotional. In the former, the subject is unable to voluntarily contract affected facial muscles, for example when asked to bare his/her teeth. However, these patients are perfectly able to give a spontaneous emotional expression such as a smile, even on the paralysed side. This palsy results from a lesion in the facial area of the motor cortex or in the corticobulbar tract. In the latter type of impairment, the ability to voluntarily contract facial muscles is not affected; however, these patients show marked impairment of function of these muscles when they form part of an emotional expression such as smiling or crying. This type of palsy results from lesions in the hypothalamus, mesencephalic or pontine tegmentum.

Therefore, it appears that there are two distinct pathways to the facial nucleus, the first controlling voluntary facial expression, which has been shown to involve direct connections from the motor cortex via the corticobulbar pathway (Kuypers, 1958; Holstege, 1991), and the second mediated by the hypothalamus. The nature of this second pathway is less certain and many structures have been implicated including the basal ganglia, thalamus, temporal lobes, frontal white matter, supplementary motor cortex, limbic system and thalamus (Hopf, 1992; Urban et al. 1998). It is also known from the lesion studies that the two pathways remain separate until at least the lower pons.

In the present study, subjects were asked to make voluntary facial expressions. This would activate the voluntary pathway, involving the motor cortex. The findings suggest that there is a large proportion of common input to the orbicularis oculi and zygomaticus major during smiling, but not a significant amount of common input during sad and horrified expressions. The reason for the paucity of common drive during sad and horrified expressions could therefore be related to the fact that it is unusual to voluntarily make these expressions. They are generally emotionally triggered and therefore mediated by the emotional pathway. The strong common drive found for the smile might represent the fact that unlike the other facial expressions, smiling is often a voluntary action performed as a social requirement. Therefore, the fact that smiling is more regularly required as a voluntary action might be reflected in the fact that there is a hard-wired motor command to produce it.