‘Paradoxical Driving’ of Walking Activity in Locusts

A great deal has been learnt recently about the interconnexions and properties of single, identifiable neurones that underlie the behaviour of invertebrates (reviewed by Hoyle, 1970, 1975). With this approach, from single neurones towards behaviour, however, the identification of the neural mechanisms underlying on-going behavioural changes remains a remote prospect for Hoyle (1975) and for Burrows (1975), if not for Miller (1974).

The classical approach to the neural basis of coordination (Sherrington, 1947; von Holst, 1973) started from the opposite, behavioural end. It has been advocated afresh recently by Harmon (1964), Kennedy (1967, 1974) and Nelson (1973) and forms the rationale of the present work. Harmon (1964) noted that the enormous number of degrees of freedom in a small assembly of neurones made it possible to find almost any imaginable set of stimulus-response characteristics. Accordingly Kennedy (1967) and Nelson (1973) doubted whether single-unit studies would tell us much without a detailed, previous analysis of the behavioural system. More specifically, Kennedy (1974) argued that the ‘fixed action patterns’ or ‘motor programmes’ so far favoured by neurophysiologists were not the most promising behavioural material for neurophysiological analysis, and instead recommended behavioural sequences that are not fixed but show a large measure of stimulus dependence, and are therefore more accessible to experiment. The central nervous system’s contribution to the patterning of a behavioural sequence would then ‘show itself as changes of responsive-ness to the sensory inputs’.

This suggestion stemmed from detailed, quantitative experiments on the co-ordination of variable sequences of different behaviours in Aphis fabae, flight and settling (Kennedy, 1965,1967) or two kinds of flight (Kennedy & Ludlow, 1974). From these it was clear that continuous central interaction between the different behavioural systems generated changes in their relative excitability, notably as behavioural post-inhibitory rebound (‘antagonistic induction’) and its opposite, ‘antagonistic depression’. The ultimate object of the present work was therefore to explore the neural correlates of such behavioural interactions, and walking locusts were chosen as much more promising material than flying aphids for the eventual neurophysio-logical work. The necessary first task was again a detailed behavioural analysis, some results of which are presented here.

Larvae of Locusta migratoria migratoriodes (R. & F.) in their fifth and last instar were removed from their high-density stock cages within 24 h of ecdysis and held in 10 1, cylindrical, perspex cages, 5 larvae per cage, with an abundance of glasshouse-grown wheat seedlings as food. The cages were furnished with twigs and kept in a room maintained at 28 ± 1 °C with a 60 W tungsten bulb shared between two cages providing light and radiant heat for 12 h each day from 08.00 to 20.00 h local time.

On the third day of the instar, the time when these insects were most active, the food was removed for 3 h before a single morphologically perfect individual was selected and fixed by the thoracic sternites to a chisel-shaped metal spike, using a resin-wax mixture (60% pure beeswax, 40% XAM synthetic resin supplied by George Gurr: Steel, 1971). The locust on its spike was supported rigidly and in-dependently of the framework carrying the circular treadmill wheel on which the locust walked (Fig. 1). The spike was positioned so that the locust was walking slightly uphill with its head directly over the centre of the wheel, as this was found to favour sustained walking. The treadmill wheel was 500 mm in circumference and 14 mm wide at the rim. The treadmill was supported in a gimbal which allowed it to move freely along the longitudinal and transverse axes of the locust. Freedom of movement in the locust’s vertical axis was provided for by mounting the gimbal at one end of an arm which was itself pivoted. Thus the locust’s weight was supported by the sternites, and not by the legs. At the same time, in order to simulate the sensory inputs of a free locust, the arm carrying the treadmill was counterbalanced to provide an upward force on the legs equal to the locust’s weight. These arrangements were designed to minimize mechanical restraints on the insect since such restraints had been found to produce struggling and, generally, unpredictable behaviour (cf. Roeder, 1975).

The information from the treadmill sensors (Fig. 1) was recorded on magnetic tape along with time pulses and stimulus markers. The various types of behaviour being performed by the locust on the wheel could be identified from the patterns of rotational and lateral displacements displayed in the records, after these patterns had been classified in the light of simultaneous direct observation of the insects on a television monitoring system in infra-red and white light.

The top of the treadmill wheel, where the locust walked, projected 50 mm through a slot cut in the floor of a circular black arena 1·0 m in diameter surround by a black wall 0·15 m high. The arena was surrounded in turn by a white muslin cylinder 1·02 m in diameter and 1·2 m high; on the inside of this cylinder and directly facing the locust was a vertical black stripe 0·06 m wide and 1·2 m high. The top of the muslin cylinder was closed with a black board with a central hole 0·27 m in diameter. The beam from a 250 W d.c.-powered quartz-halogen projector was reflected by a metal plane mirror through this aperture down on to the arena. The whole structure was further surrounded by blackout curtaining in a subterranean laboratory maintained at 28 ± 0-5 °C.

While the locust was being attached to its spike, in light and without anaesthesia, the treadmill was held out of reach below it, leaving the locust freely suspended above. Ten minutes later, when the locust had usually become still, the treadmill was released from outside the screening and allowed to float up to within the locust’s grasp. The experimental treatment began from that moment.

Extensive preliminary experiments were done using a variety of visual, mechanical, acoustic and olfactory inputs for the locust, in order to find some convenient external conditions in which the locust would repeatedly switch between walking activity and other forms of behaviour without any change in these conditions. Light was eventually chosen as the experimental variable, being the most easily characterized and controlled. The experimental treatments consisted of three separate light regimes, the locust always having been in light until the treatment began: (i) Continuous light for 3 h. (2) Continuous darkness for 3 h. (3) Continuous darkness for 1 h followed by 2 h of alternating 10 min periods of light and of darkness. Each individual locust was given one of the treatments only, with 10 locusts per treatment.

For the analysis presented here, only two categories of behaviour were distin-guished:

Walking- forward progression on the treadmill at speeds of at least 2 mm/s.

Non-walkingall other behaviour, including backward progression and forward progression at less than 2 mm/s (often accompanied by grooming, biting or ‘peering’), and complete immobility.

Grooming, biting, peering and even ventilatory movements of a non-walking locust sometimes caused one or more of the radial stripes of the wheel to pass across the sensor, and so they needed to be distinguished from walking. This problem was complicated by such activities sometimes persisting beyond the start of a walking bout, and disappearing only gradually as speed increased; whereas at the end of a walking bout they did not normally begin again until walking had stopped. To allow for this the rate of stripe-passing used as a criterion of walking was set higher for the start than for the finish of each bout of walking. Technically this meant that the start of walking was signalled in the records when the wheel turned at a rate at which 4 or more stripes passed the sensor in 512 ms, whereas cessation of walking was signalled when the wheel slowed until no stripes passed in 512 ms. As the records were analysed in units of one second this meant in practice that the minimum walking speed recorded over the first whole second of a walking bout was 4 mm/s, and that for the end of it 2 mm/s.

The tapes were analysed into measurements of the following behavioural quantities: Walking bout length- duration of a period of continuous walking as defined above, disregarding any pauses of less than 512 ms.

Non-walking bout length- duration of a period of continuous non-walking as defined above, between walking bouts.

Walking speed- rate of walking mm/s.

Time spent walking in seconds.

Distance walked- total distance covered (= total duration of walking bouts × mean walking speed when walking). This provided an overall measure of walking activity by incorporating the proportion of time spent walking as well as the speed when walking.

All measures of behaviour were first averaged for each of the eighteen 10 min periods of the 3 h of each individual’s treatment. A Friedman two-way analysis of variance by ranks was then performed on these data from all the individuals under each treatment in order to check whether there were statistically significant differences among the 10 min periods. If this test was positive, then a two-tailed Wilcoxon matched-pairs signed-ranks test was used to compare each 10 min period with every other, in order to identify precisely where the significant differences lay. The same procedure was used to test the significance of differences between whole hours within and between treatments. Correlations between the several behavioural components within Treatments 1 and 2 were tested by ranking the 3 h of each treatment for each measure and treating these ranks as three levels for a Friedman analysis of variance. A similar test was applied to Treatment 3 as explained under the Results.

Differences, changes and correlations mentioned among the Results below were statistically significant at the 5 % level or better, unless explicitly described as insignificant.

Fig. 2 is the record obtained from one individual locust in continuous light. The general course of this individual’s behaviour was close to the mean for the 10 locusts given the same treatment as can be seen by comparing Fig. 4 with Figs. 5 (i) and 6.

Fig. 3 shows in detail how the frequency distribution of this individual’s walking speeds changed with time and Fig. 4 summarizes the concurrent changes in several measures of its behaviour. Fig. 5 summarizes the results from the 10 individuals undergoing each of the three treatments, in terms of the mean distance walked in each 10 min interval. Figs. 6−8 show the mean values for each treatment of the three measured components of distance walked. It is evident that the locusts’ behaviour changed within the 3 h of each treatment and also differed markedly between treat-ments. These differences within and between treatments are analysed below.

When a locust first grasped the treadmill at the end of the pre-treatment it took up the posture described by Ellis (1951, Fig. 4) as the ‘post-prandial rest’ position, with the body lying on the substrate. After several minutes this behaviour was interrupted by brief bouts of movement of the antennae and palps independently. These episodes were in turn replaced by grooming and peering. After 5−40 min spent for the most part in the post-prandial rest position the locust pushed the treadmill down (the sensory equivalent, for the locust, of its raising its body on its legs) and started to walk slowly forwards. In 2 out of the 10 individuals, walking activity remained at a very low level and non-walking activities continued throughout the 3 h observation. In the other eight individuals the distance covered by walking (Fig. 5 (i)) gradually increased in what may be called the ramp stage of walking activity, so that the mean for the 10 locusts reached about 15 m/10 min at 70−80 min from the start and then levelled off into a plateau stage. The mean time spent in walking increased in parallel until it reached 400 s/10 min, also at 70−80 min from the start, and also levelled off. The resulting intense walking activity at 70-80 min was maintained with no further statistically significant change in the 10 min means of distance walked, or of the time spent walking, up to the end of the treatment, although there was some decline in the last hour.

In the first 10 min period the average walking bout lasted only 3 s while the average non-walking bout lasted nearly 3 min (Fig. 6 (i) and (ii)). Walking bouts then lengthened somewhat and non-walking bouts shortened greatly over the next 40 min. For the next 70 min there was an alternation between walking and non-walking bouts of similar length (means of 12-15 and 8-15 s, respectively (Fig. 2 is atypical in this respect)). Mean walking speed (Fig. 6(iii)) also increased progressively and levelled off around 34 mm/s at about 80 min from the start and did not change significantly through the second hour; but during the last hour when walking was on the decline in some individuals there were several 10 min periods of significantly lower mean walking speed. The lengths of walking and non-walking bouts showed no significant changes after the first hour.

The hourly means showed a strong positive correlation between walking speed and walking bout length, and a strong negative correlation between walking speed and non-walking bout length. The apparent inverse relationship between walking and non-walking bout lengths was not significant statistically although there was a tendency for long walking bouts and short non-walking bouts to occur during the same hour. Thus the observed time-course of behaviour in continuous light displayed some ordered relationships between the several components of the behaviour.

The locusts’ behaviour in continuous darkness was more variable than in continuous light, and although they started walking sooner they failed to sustain it (Fig. 5 (ii)). The mean distance walked reached its peak of about 12 m/10 min at 20−30 min from the start, coincident with the peak of mean time spent walking. The time spent walking fell significantly below the peak level in all the 10 min periods of the last 90 min but the distance walked did so in only three of them, reflecting the varying time course of the individuals’ behaviour.

There were no significant changes in mean walking speed (Fig. 7(iii)), which remained moderate. Walking bouts on the other hand lengthened rapidly to the remarkable mean length of 86 s (Fig. 7 (i)), far more than ever achieved in the light, but they then shortened again instead of levelling off as in the light. The walking bouts never entered a phase of alternation with non-walking bouts of similar length, as they did in the light. They showed no significant changes after the decline from the peak. The few significant changes in mean non-walking bout length fell into no pattern (Fig. 7(ii)).

There was a strong positive correlation between walking speed and walking bout length, but the tendency for higher walking speeds to be associated with short non-walking bouts was not significant; nor was the tendency for long walking bouts to be associated with short non-walking bouts. Thus there was less evidence of orderly relations between the behavioural components in continuous darkness than in continuous light.

During the dark first hour the locusts’ performance was not significantly different from that in the dark first hour of Treatment 2. During the second and third, alternating hours, however, the distance walked fell dramatically in the light periods and rose again in the dark periods (Fig. 5 (iii)) and the time spent walking fell and rose in strict parallel.

The first 10 min light period suppressed walking almost completely. Each succeeding one again depressed it strongly, although less than the first. Each time that the locusts were returned to the dark after one of these inhibiting light-breaks they walked very strongly, in sharp contrast with their behaviour in the second two hours of continuous darkness under Treatment 2 (cf. Fig. 5 (iii) with 5(h)). The mean distance walked and time spent walking exceeded the mean levels reached in most of the 10 min periods of the first hour of continuous darkness. In the result, the total distance covered by these locusts in the six intermittent dark periods was much more than they had covered in the dark first hour.

The distance walked (Fig. 5 (iii)) and the time spent walking appeared to increase progressively over several light periods and also over some of the dark periods, although this trend was statistically significant only for the time spent walking during the light periods (Bartlett, 1949).

The three measured behavioural components, walking bout length, non-walking bout length and walking speed (Fig. 8), were analysed separately (as 10 min means) for each of the following three parts of the treatment: the first hour of continuous darkness; the six subsequent 10 min light periods; and the six intervening 10 min dark periods. Differences both within and between these parts of the treatment were tested statistically.

Walking bout length (Fig. 8(i)) showed no statistically significant differences within the dark first hour, or among the six subsequent dark periods. Over the six light periods the changes were smaller and less erratic. The walking bouts were shorter in the first light period than in later ones, shorter in the second than in the last two, and shorter again in the third than the last, hinting at an asymptotic increase. The walking bouts were markedly shorter in the light periods than in the intermittent dark periods.

Non-walking bouts (Fig. 8(ii)) were longer in the first of the light periods than in any later one, and longer in the first dark period than in any later one. As between the three parts of the treatment, the non-walking bouts were longer in the first light period than in the dark first hour, and longer during the light periods than in most of the intermittent dark periods.

Walking speed (Fig. 8 (iii)) did not change significantly within any one of the three parts of the treatment. As between the parts, walking speeds in the intermittent dark periods were higher than in the dark first hour; and higher also than in the intervening light periods. Thus the locusts walked faster and in longer bouts within each 10 min of darkness after a 10 min light-break, and this was why they covered a greater total distance in those six dark periods than they did in the six light-breaks. It was mainly because they walked faster in those dark periods that they walked a greater total distance in them than during the dark first hour.

The relations between the several behavioural measures during the light-dark alternation were examined by ranking the six 10 min light periods and, separately, the six 10 min dark periods. In the light periods, the correlation between walking bout length and walking speed was not significant, but there were negative correlations between non-walking bout length and walking speeds, and between walking bout length and non-walking bout length. In the dark periods, on the contrary, there were no significant correlations between these measures. Thus the behaviour was again less ordered in the dark than in the light, even during these short alternating periods.

During the first hour, the distance walked increased faster in the dark than in the light but soon declined again in the dark while it went on increasing in the light (Fig. 5 and see pp. 6 and 10). The net result of these differences was that the distance walked, walking speed and walking bout length did not differ significantly between the three treatments when averaged over the whole hour. At this time non-walking bouts were longer in the light and included much other behaviour which hardly appeared in the dark. During the second hour, the mean whole-hour walking speed and distance walked were greater in continuous light than in continuous darkness. These gross measures ceased to be significantly different in the third hour when some individuals in light were slowing down and those in the dark highly variable.

The most distinctive feature of the original records in continuous darkness was the amount of walking done in long unbroken bouts. There were 27 bouts lasting more than 2 min and 11 lasting more than 3 min (including one of 19 min) in the dark, as against nine and one ( < 4 min) in the light. Altogether, bouts in which the walking was sustained for over 60 s accounted for 29% of all the time spent walking during the 3 h in the dark, but for only 13% of it in the light (P < 0·01).

The second and third hours of Treatment 3 (alternating) were compared with those of Treatments 1 and 2, using unpaired t-tests, by comparing behaviour in the six light periods of Treatment 3 with the equivalent 10 min periods under the continuous light of Treatment 1 ; the six intervening dark periods of Treatment 3 were compared separately with the equivalent 10 min periods under the continuous dark-ness of Treatment 2. When so compared, the distance walked in the intermittent light periods of Treatment 3 proved significantly less than that in continuous light (P < 0·01); whereas the distance walked in the intermittent dark periods of Treat-ment 3 was significantly greater than in continuous darkness (P < 0·01). Thus the differential effects of light and darkness on walking activity, when each was main-tained, were not only reversed but also increased by alternating between them.

A primary technical aim in this work was to reduce artificial restraints to the minimum compatible with precise recording of the behaviour and, eventually, of concomitant neural events. How far this aim was achieved could be judged by com-paring the ‘marching’ behaviour of fixed locusts on the treadmill in light with accounts of the free marching behaviour of grouped locusts in daylight in nature (Kennedy, 1939, 1945; Ellis & Ashall, 1957) or in illuminated laboratory arenas (Ellis, 1951). One major difference was observed: progression by long, low leaps is common in free-marching locusts, whereas leaping was very rare on the treadmill. Possibly this was associated with the square edges and precise width of the treadmill which were designed for the maximum ‘railroading’ effect. This ‘railroading’ may also have helped to compensate for the lack of optomotor stimulation from moving neighbours which is known to promote marching in groups of free locusts (Clark, 1949 ; Kennedy, 1951; Ellis, 1953; Ellis & Hoyle, 1954).

Otherwise, the treadmill locusts reproduced the marching behaviour of free locusts remarkably well, provided they were kept in continuous light, despite the much simplified and unchanging environment. The dropping out of other forms of behaviour while walking gradually rose to and then continued at a high intensity on the treadmill in continuous light (the ‘ramp’ and ‘plateau’ in Fig. 5(i)), mimicked the sequence of behaviour that has been observed each morning in nature. This sequence has been attributed to the increase in light intensity and temperature (Kennedy, 1939; Ellis & Ashall, 1957; Ellis, 1963). Since the temperature and other conditions were held constant in the treadmill laboratory, it follows that a similar behavioural sequence can be generated internally. The sequence of behaviour in continuous darkness (Figs. 5 (ii) and 7) showed that light was necessary for high-intensity march-ing to develop and persist; but this did not depend on any changes in the lighting or other conditions.

At first sight some of the results seem paradoxical, but it is these very results that suggest how the changes of behaviour were brought about in a constant environment.

The first paradox appears when the continuous light and continuous dark treatments are compared (see Fig. 9). Although walking began earlier and more strongly in continuous darkness than in continuous light, it eventually became stronger and lasted longer in the light; in darkness it soon declined and became very erratic (p. 10).

The second paradox is that although the walking activity eventually became stronger and more persistent in the light, long bouts of it were commoner in the dark (p. 12). In the dark, many walking bouts were long from the beginning (Fig. 7) and little activity of other kinds occurred even during long non-walking periods. In the light, on the other hand, walking was preceded and at first interrupted by long pauses in which other activities (grooming, peering, biting, etc.) commonly occurred (Fig. 6).

The increase in walking activity by the majority of individuals in the light came about through shortening of the non-walking bouts, lengthening of the walking bouts and increase of the walking speed (Fig. 6). The non-walking bouts made the build-up of walking activity slower in the light, yet the eventual outcome was a higher level of walking activity. During the subsequent maintenance of this intense walking activity frequent interruptions continued although many were now very brief (Fig. 2). Thus the high level of walking activity achieved in the light was associated throughout with frequent temporary inhibition of walking. In the dark there were fewer such interruptions when the walking activity was at its height (p. 10) yet it did not reach so high a level and soon declined.

The results therefore suggest that in continuous light the walking activity was both built up and maintained by repeated post-inhibitory rebound, or ‘antagonistic induction’ as described in aphids (Kennedy, 1967, 1974). This behavioural phenomenon is analogous in form to post-inhibitory rebound in spinal reflexes (Sherrington, 1947) and in single neurones (Otani & Bullock, 1950; Bullock & Horridge, 1965, p. 207; cf. Perkel & Mulloney, 1974). Maynard (1961) coined the apt expression ‘paradoxical driving’ for a case where neuronal post-inhibitory excitation was cumulative, and this term would be no less apt at the very different level of our whole animals’ behaviour.

This interpretation of the overall behavioural sequence in continuous light is consistent with the numerous instances of rebound of walking speed observed after a pause, as seen in the sample record in Fig. 2; but a full bout-by-bout analysis of the information on this and other points in all the records is required and will be given elsewhere. Meanwhile the main results given here provide quantitative evidence of post-inhibitory rebound, albeit on a somewhat longer time scale.

The long-term effect of continuous light was to promote walking and that of continuous darkness was to depress it, but the 10 min alternation of light and dark (Treatment 3) showed the reverse effects: walking was inhibited in the light periods and resumed in the dark ones (Fig. 9). Moreover, the inhibiting light had a strongly excitatory after-effect on walking. When darkness returned after the first 10 min light-break, walking was not merely resumed but much strengthened compared with what it had been in the first hour of darkness before the light-break (p. 10), a striking example of rebound. As the light-breaks continued through the 2nd and 3rd hours this intense walking activity in the dark periods also continued, contrasting with the reduced activity in continuous darkness (Fig. 9); and, even in the light-breaks themselves, the time spent walking crept up (p. 10), which did not happen in continuous light.

Thus the diverse results of the three treatments can be unified on the basis of behavioural post-inhibitory rebound. In particular, they suggest that the main mechanism by which light promoted vigorous walking in these experiments was the indirect one of paradoxical driving. That is to say it apparently did so by exciting other reflex systems (biting, etc.) antagonistic to walking, which by inhibiting the walking system temporarily, caused it to rebound, reiteratively.

The behavioural results do not exclude the possibility that light had some central excitatory effect on the walking system, in parallel with its stronger central inhibitory effect, although only the inhibitory effect was immediately apparent in the behaviour (cf. Kennedy, 1974; Kennedy & Ludlow, 1974). But such neurophysiological conclusions cannot be drawn from behavioural evidence alone. So far, post-inhibitory rebound at the neural and behavioural levels are no more than analogous.

At the behavioural level the evident effect of the light was to elicit non-walking forms of behaviour and inhibit walking temporarily. In the aphid (Kennedy, 1974) it was found that a given kind of activity may be depressed rather than strengthened as the after-effect of its being inhibited temporarily. Which of these two consequences occurred (i.e. antagonistic induction or antagonistic depression) depended on the relative excitation of the two antagonistic systems at the time : after inhibition the less excited one was more likely to be depressed than induced, and conversely. In the present locust experiments the ‘railroading’ effect of the narrow treadmill, plus the slight upward slope presented to the locust, appeared to exert a continuous excitatory effect on walking, and this was presumably what tipped the balance in most individuals toward antagonistic induction rather than depression of walking, so that the average curve of walking activity for the 10 individuals under continuous light rose to a high level (Fig. 9). In the two locusts that walked very little (p. 6), the balance may have been tipped the other way in favour of antagonistic depression of walking.

An alternative hypothesis of the build-up of walking activity under continuous or intermittent light might be that this was simply a case of progressive ‘arousal’, meaning a cumulative increase in walking excitation, due to, say, hunger, and not dependent on central interaction in the form of post-inhibitory rebound. On this hypothesis, the inhibitory inputs received in the light (as evidenced by the early long pauses and other forms of behaviour) would be doing no more than retarding the on-going process of arousal. But they must have been doing more, because the ‘arousal’ went further in continuous light than in continuous darkness, and was notably boosted by the light-breaks in the alternating treatment. The concept of arousal would become applicable to these sequences only if it were stretched to include post-inhibitory rebound as a consequence of the central interaction of antagonistic systems. That has not so far been included among the heterogeneous set of mechanisms subsumed under ‘arousal’ (Andrew, 1974; Brady, 1975).

Similarly, the sustained level of walking activity on the ‘plateau’ under continuous light cannot be attributed simply to a steady level of excitation in the walking system alone, once arousal was complete. Rather, the frequent interruptions of walking in the light, the regularity of the alternation between walking and non-walking (p. 7) and the more orderly relations between the various behavioural measures in the light than in the dark (pp. 8 and 10), suggest that the behavioural stability on the ‘plateau’ was a dynamic one, an oscillation between two, highly excited antagonistic systems each alternately inhibited and induced.

In the ‘plateau’ stage when walking was most intense and the non-walking bouts had shortened, activities other than walking hardly appeared. Nevertheless sensory inputs capable of eliciting non-walking activities’ were present at that time since those other activities were elicited during the previous ‘ramp’ stage, and the environ-ment remained constant throughout. Thus the inhibitory effect on walking of such antagonistic inputs was presumably still present during the ‘plateau’ stage and sufficient to stop the walking frequently. This is what appears to have sustained the high level of walking activity by paradoxical driving.

We thank Dr J. Brady and Dr R. F. Chapman for their comments on the manuscript; Miss S. Green and Mr A. R. Ludlow for their advice and assistance with the statistical treatments; and Dr S. Young for advice on instrumentation.