Brain Research, 153 (1978) 39-53 0 Elsevier/North-Holland Biomedical
39 Press
THECOORDINATIONOFEYEANDHEADMOVEMENTDURINGSMOOTH PURSUIT
JEREMY LANMAN, Department
EMIL10 BIZZI and JOHN ALLUM*
qf’ Psychology,
Massachusetts
Institute
of Technology,
Cumbridge,
Mass. 02139 (U.S. A.)
(Accepted January 4th, 1978)
SUMMARY
Eye and head movements
during
tracking
of a smoothly
moving
visual target
were recorded in trained monkeys. The head movement clearly followed the target, although with considerable variability from cycle to cycle. The eye stayed relatively near the primary position and moved in an apparently irregular fashion; however, the sum of eye and head, or gaze, remained accurately on target despite the irregularity of the individual
eye and head movements.
When compared
with tracking
with head fixed,
head free tracking was not measurably different in accuracy. Further experiments were performed which demonstrated a role for the vestibular system in coordinating eye and head during smooth pursuit.
The results of these experiments
can be best explained
by
postulating an internal smooth pursuit command driving both eye and head movements. In the case of the eye movement, this smooth pursuit command is combined with vestibular feedback
from head movement
before being forwarded
to eye movement
centers.
INTRODUCTION
A monkey
whose head is free to turn uses a combination
of eye and head move-
ments to track a moving visual target. In this study, we have investigated which these movements are coordinated: that is, how centrally generated
the way in commands
to the motor systems of the eye and the head are integrated with afferent activity originating from visual and vestibular receptors, as well as from neck proprioceptors. Our goals were to determine
whether
tracking
accuracy, and to search for evidence smooth pursuit.
* Present address: Institut fiir Hirnforschung, 8029 Ziirich, Switzerland.
with and without
suggesting
head movement
that there are different
differ in
strategies
der Universittit Ziirich, August-Forel-Strasse
of
I, CH-
40 Most investigators who have studied the problem of smooth pursuit have kept the head rigidly fixed in order to study eye movement in isolation from head movement. These studies have produced varying claims about the accuracy of tracking. In an early study, Rashbassr7, for example, found that human subjects could precisely match target velocity with eye velocity. Puckett and Steinma@, on the other hand, using similar stimuli found that eye velocity in humans was significantly smaller than target velocity. These contrasting findings suggest that the eye pursuit system may have more than one mode of operation. If there is slip of the target image across the retina, then a velocity servo modelst*as may be appropriate where this retinal slip is the stimulus for smooth pursuit eye movements. On the other hand, if there is no slip, then a predictive control model in which eye movements are driven by stored knowledge of target velocity is more likely. We re-investigated smooth pursuit by looking at continuous records of retinal error, searching for evidence of either retinal slip or error-free pursuit. We also looked for different strategies of head movement and attempted to determine whether they were related to predictive or non-predictive modes of eye movement. MATERIALS
AND METHODS
Monkey training
Three Macaca mulatta monkeys were trained to make a visual discrimination between a horizontal bar and a vertical bar superimposed on a stimulus light. The animals were reinforced with drops of fruit juice only if they pressed a lever when the vertical bar appeared. When the lever was pressed inappropriately, the trial was ended and no new stimulus was presented for several seconds. Since the vertical bar was quite narrow (3 min arc) and was only presented for about 400 msec, the monkey had to fixate the stimulus light whenever it was presented to be in aposition to detect the briefappearance of the vertical bar. All of the monkeys readily transferred their training from stationary to moving targets, and no special effort was made to train them to m_ake smooth pursuit movements. Most records, however, were taken from monkeys which had had considerable practice in the pursuit of moving targets. Surgery
The monkeys were trained for about two months, then, under anesthesia, screws which connected to the head holder were implanted in the skull and silver-silver chloride electrodes were placed in the orbital bone of each eyea. After making the measurements, which will be discussed later, bilateral labyrinthectomy was performed on two of the monkeys by drilling through the mastoid bone to expose the canals which were then opened, scraped clean of their membraneous lining and filled with dental cement. Each of the monkeys was rotated in darkness on several occasions to ensure that no vestibulo-ocular reflex was present, and that the vestibulectomy was complete and permanent. Data acquisition
During recording sessions, the monkey was seated in a primate.chair;
its head
41 was attached by the implanted skull screws to a lightweight headholder. A low torque precision
potentiometer
magnetic
clutch was used to stop spontaneous
pectedly.
The EOG electrodes
monkey
was used to monitor
faced a tangent
the rear. Stimulus feedback
The eye position
The
was projected
by a mirror galvanometer
from
which had position
used were: (1) constant
(0.1-I Hz), and (3) triangular from the implanted
signal to yield the direction
signal was subtracted
were simultaneously
and unex-
amplifier.
posi-
waves (0.1-l
EOG electrodes
Hz).
(accuracy
so that it was linear to I_t 30”. This signal was then combined
with the head position target position
(2) sinusoids
signal recorded
of & l”)fi was corrected
to a high impedance
of 100 Hz. The target stimuli
tions for EOG calibration,
and an electro-
suddenly
I m away, on which the stimulus
was controlled
and a bandwidth
head rotation
head movements
were connected
screen,
position
horizontal
recorded
of gaze (eye plus head positions).
from the gaze to determine
on FM tape and displayed
The
retinal error. Signals
on a Honeywell
visicorder.
Data analysis Target, gaze and head position by a PDP 11 computer.
Continuous
signals for each trial were sampled every 5 msec segments
of smooth
pursuit
(ranging
from 2 to
7.4 set in length) were selected from each trial, then retinal error and eye position minus head position)
were calculated.
The data were further movements
(gaze
analyzed
to answer
two specific questions.
of gaze, eye and head lead the target profile, indicating
Firstly,
a predictive
do
mode of
tracking, or do they lag it, suggesting a servo mode of tracking? Secondly, how do the amplitudes ofgaze, eye and head movements compare with the amplitude ofthe target’s movement: smooth
in other words, what is their gain ? No provisions
and saccadic components
In order to answer the first question, of signals (such as target-gaze,
were made to separate
of eye movements. the cross-correlation
target-head
and target-eye)
coefficient for the pairs was computed
by:
where E,,
m’ N-m
Nfr
xn yr1+r I
is the cross-correlation function ofa pair ofsignals x and y, It is the sampling interval, m is the maximum lag number, N is the number of\ample\ and l?(O) is the estimate of a signal’s autocorrelation function at lag time zero 2. The delay time, rat, at which the peak of the p(rAt) occurred, was taken as the mean lead or lag of the eye, head or gaze signal with respect to the target. To obtain
a measure
of the contribution
of head movement
to gaze movement,
and to measure gaze amplitude, the gain of each signal (gaze, eye and head) was expressed as the ratio between amplitudes of target movement and the signal at the fundamentalfrequencyofthetargetspectrum(thefrequencyatwhichthetargetmovement’s amplitude is greatest). To determine this fundamental frequency and the signal am-
42 plitudes, complete records of the position of target, head, gaze, eye and retina4 error for each segment were Fourier analyzed with the digital computer for their spectral content. The procedure for Fourier analysis was as follows : (1) Line frequency noise and FM tape noise were removed from the data by a digital low-pass filters. The fitter gain was flat until 30 Hz, at which point the gain was down 0.92 dB; at 50 Hz, attenuation was -39.6 dB. The mean was removed from the data in order to reduce low-frequency artifacts. (2) The power spectra were calculated at integer multiples of 0.098 Hz from the autocorrelation function of the filtered data. For this purpose the followingrelationship from Bendat and Piers012 was used : m-1 G,(f)
== ;- [R(O) t 2
C
C
r -~-1
R(rht)
cos (TS
c
t Qmht)
cos (zFTt)] (’
Where e,(f), an estimate of the true power spectral density function, is detined for integer multiples of 0.098 Hz and computed in the range from 0.098 Hz to 30 Hz. In the equation above, @rh) is the estimate of the autocorrelation function at tag r; m is the maximum lag number (256 was used); and fc is one half of the sampling frequency (that is, 100 Hz). The signals analyzed by the computer were a set of samples of finite duration (a truncated data series). To suppress artifacts due to this truncation, autocorrelation estimates were multiplied in the time domain by a function known as a Tukey window3. RESULTS
In agreement with previous investigator#, we found that the eye pursuit system can accurately track sinusoidal and triangular wave stimuli presented at -frequencies up to about 1 Hz over a range of &30” (the maximum range of our galvanometer), with an average gain of 0.96. This gain is calculated from combined smooth and saccadic eye movements. Step changes in target velocity are followed by changes in eye velocity with a latency of about 80 msec. (This latency is considerably smaller than that reported by Barmackr and Fuchs*, possibly because of differences in the parameters of the stimuli used.) When stimuli were presented at frequencies above 1 Hz, pursuit became quite erratic, and the monkey often failed to press the reward key appropriately. In addition, our recordings of smooth pursuit in monkeys showed frequent changes in the slope of the retinal error tracing. (Retinal error is the difference between target position and eye position.) Eye movements with velocities considerably lower than the stimulus velocity were often observed; these were indicated by steadily increasing retinal error. At other times, within the same trial, we found smootheye movements which matched the stimulus velocity so closely that no detectable retinal error was present (Fig. 4). Only occasionally did we find smooth eye movements that were faster than the stimulus movement. The smooth pursuit system apparently can cperate both with and without retinal slip. Furthermore, on the average, the combined smooth and saccadic eye movements lagged the target by approximately 20 msec (Table I).
43
I
Fig. I. Pursuit with combined eye and head movement. Pursuit movements of eye and head are shown together with the computed gaze and retinal error. The target, which periodically reverses direction, is superimposed on both head and gaze tracings. The head pursued the target, while the eyes remained relatively close to the center of the orbit. There is no obvious difference between the retinal error pattern recorded here and that observed in monkeys whose heads were restrained.
The hradpursuit
system
(A) He&fixed.
In an animal
with the head fixed, head torque can be measured
as an indication of intended head movements. These measurements showed that head torque is clearly related to target motion, indicating that the neck musculature was programmed to move the head even when the resulting muscle torques were rendered completely ineffective by blocking the head’s movemenP. There were also occasional
44 TABLE I The averagedgain
of eye and headrelative
to the target und the lag of gaze and head relariw
to
rhr
tafgct
The gains were computed from the amplitude of the fundamental Fourier component. and the delay was taken as the peak of the cross correlation function (see Methods). Several trials (n) over a range of amplitudes (A) and frequencies (Hz) are averaged for each monkey, both with head free and head fixed. Target
C2Cfr.c Head
M-66
Head free
Head fixed
M-73 Head free
Head fixed
n 24 A 21.6’ + 8. I 0.1-0.68 Hz 0.36 f 0.28 n 26 A .~ 23.5’ -I_9.4 0.1-0.88 Hz 0.34 t 0.28
0.83 1 0.16
n= 14 A 17.6” j_ 7.3 0.1-0.88 Hz 0.38 i- 0.31 n -~ 12 A -- 14.9” 3: 6.7 0.1-1.50 Hz 0.67 % 0.44
0.71 + 0.21
0.29 + 0.26
34 $ 34
0.94 .I: 0.06
36 -! 33
128 _f 83
discontinuities in the head torque record - suggesting that a sudden head movement may have been programmed ; however, the eyes remained on target whether or not these head torques were present. (B) Head,free. Allowing a monkey to pursue a target with a combination of eye and head movements by freeing its head had no evident effect on the totalgazeronrpared to head fixed tracking (Fig. 1). Both with head fixed and with it free, the movement of the gaze and that of the target had approximately the same amplitude, and, in both cases, the gaze lagged the target by about 20 msec (Table I and schematic Fig. 6). Although the total gaze is on target with or without head movement,~the eye movements are very much different. With head fixed, as stated earlier, the eyes very nearly match target amplitude and, on the average, lag the target by some 20 msec. With the head free to pursue the target, the eyes must now make up the difference between target and head in order to keep the gaze on target. Typically, the head may lag the target by some 75 msec and move about 85 ‘4 of the target’s amI;litude. The eyes now make upthe missing 15 % of target amplitude, and in addition must lead the target to make up for the lag of the head (Fig. 6). Over the range of frequencies tested, the primary effect of freeing the head was that movements of smooth pursuit were almost completely accompli3red by the head movement system, while the eyes tended to remain relatively stationary in the center
45
1
SEC
Fig. 2. Different strategies of head movement. The top record shows head movements which generally lag target movement (TARG) by about 200 msec. The bottom record shows head movements:which anticipate target movement.
of the orbit. For the 3 monkeys tested, the amplitude of head movement accounted for an average of 99 T
39 Press
THECOORDINATIONOFEYEANDHEADMOVEMENTDURINGSMOOTH PURSUIT
JEREMY LANMAN, Department
EMIL10 BIZZI and JOHN ALLUM*
qf’ Psychology,
Massachusetts
Institute
of Technology,
Cumbridge,
Mass. 02139 (U.S. A.)
(Accepted January 4th, 1978)
SUMMARY
Eye and head movements
during
tracking
of a smoothly
moving
visual target
were recorded in trained monkeys. The head movement clearly followed the target, although with considerable variability from cycle to cycle. The eye stayed relatively near the primary position and moved in an apparently irregular fashion; however, the sum of eye and head, or gaze, remained accurately on target despite the irregularity of the individual
eye and head movements.
When compared
with tracking
with head fixed,
head free tracking was not measurably different in accuracy. Further experiments were performed which demonstrated a role for the vestibular system in coordinating eye and head during smooth pursuit.
The results of these experiments
can be best explained
by
postulating an internal smooth pursuit command driving both eye and head movements. In the case of the eye movement, this smooth pursuit command is combined with vestibular feedback
from head movement
before being forwarded
to eye movement
centers.
INTRODUCTION
A monkey
whose head is free to turn uses a combination
of eye and head move-
ments to track a moving visual target. In this study, we have investigated which these movements are coordinated: that is, how centrally generated
the way in commands
to the motor systems of the eye and the head are integrated with afferent activity originating from visual and vestibular receptors, as well as from neck proprioceptors. Our goals were to determine
whether
tracking
accuracy, and to search for evidence smooth pursuit.
* Present address: Institut fiir Hirnforschung, 8029 Ziirich, Switzerland.
with and without
suggesting
head movement
that there are different
differ in
strategies
der Universittit Ziirich, August-Forel-Strasse
of
I, CH-
40 Most investigators who have studied the problem of smooth pursuit have kept the head rigidly fixed in order to study eye movement in isolation from head movement. These studies have produced varying claims about the accuracy of tracking. In an early study, Rashbassr7, for example, found that human subjects could precisely match target velocity with eye velocity. Puckett and Steinma@, on the other hand, using similar stimuli found that eye velocity in humans was significantly smaller than target velocity. These contrasting findings suggest that the eye pursuit system may have more than one mode of operation. If there is slip of the target image across the retina, then a velocity servo modelst*as may be appropriate where this retinal slip is the stimulus for smooth pursuit eye movements. On the other hand, if there is no slip, then a predictive control model in which eye movements are driven by stored knowledge of target velocity is more likely. We re-investigated smooth pursuit by looking at continuous records of retinal error, searching for evidence of either retinal slip or error-free pursuit. We also looked for different strategies of head movement and attempted to determine whether they were related to predictive or non-predictive modes of eye movement. MATERIALS
AND METHODS
Monkey training
Three Macaca mulatta monkeys were trained to make a visual discrimination between a horizontal bar and a vertical bar superimposed on a stimulus light. The animals were reinforced with drops of fruit juice only if they pressed a lever when the vertical bar appeared. When the lever was pressed inappropriately, the trial was ended and no new stimulus was presented for several seconds. Since the vertical bar was quite narrow (3 min arc) and was only presented for about 400 msec, the monkey had to fixate the stimulus light whenever it was presented to be in aposition to detect the briefappearance of the vertical bar. All of the monkeys readily transferred their training from stationary to moving targets, and no special effort was made to train them to m_ake smooth pursuit movements. Most records, however, were taken from monkeys which had had considerable practice in the pursuit of moving targets. Surgery
The monkeys were trained for about two months, then, under anesthesia, screws which connected to the head holder were implanted in the skull and silver-silver chloride electrodes were placed in the orbital bone of each eyea. After making the measurements, which will be discussed later, bilateral labyrinthectomy was performed on two of the monkeys by drilling through the mastoid bone to expose the canals which were then opened, scraped clean of their membraneous lining and filled with dental cement. Each of the monkeys was rotated in darkness on several occasions to ensure that no vestibulo-ocular reflex was present, and that the vestibulectomy was complete and permanent. Data acquisition
During recording sessions, the monkey was seated in a primate.chair;
its head
41 was attached by the implanted skull screws to a lightweight headholder. A low torque precision
potentiometer
magnetic
clutch was used to stop spontaneous
pectedly.
The EOG electrodes
monkey
was used to monitor
faced a tangent
the rear. Stimulus feedback
The eye position
The
was projected
by a mirror galvanometer
from
which had position
used were: (1) constant
(0.1-I Hz), and (3) triangular from the implanted
signal to yield the direction
signal was subtracted
were simultaneously
and unex-
amplifier.
posi-
waves (0.1-l
EOG electrodes
Hz).
(accuracy
so that it was linear to I_t 30”. This signal was then combined
with the head position target position
(2) sinusoids
signal recorded
of & l”)fi was corrected
to a high impedance
of 100 Hz. The target stimuli
tions for EOG calibration,
and an electro-
suddenly
I m away, on which the stimulus
was controlled
and a bandwidth
head rotation
head movements
were connected
screen,
position
horizontal
recorded
of gaze (eye plus head positions).
from the gaze to determine
on FM tape and displayed
The
retinal error. Signals
on a Honeywell
visicorder.
Data analysis Target, gaze and head position by a PDP 11 computer.
Continuous
signals for each trial were sampled every 5 msec segments
of smooth
pursuit
(ranging
from 2 to
7.4 set in length) were selected from each trial, then retinal error and eye position minus head position)
were calculated.
The data were further movements
(gaze
analyzed
to answer
two specific questions.
of gaze, eye and head lead the target profile, indicating
Firstly,
a predictive
do
mode of
tracking, or do they lag it, suggesting a servo mode of tracking? Secondly, how do the amplitudes ofgaze, eye and head movements compare with the amplitude ofthe target’s movement: smooth
in other words, what is their gain ? No provisions
and saccadic components
In order to answer the first question, of signals (such as target-gaze,
were made to separate
of eye movements. the cross-correlation
target-head
and target-eye)
coefficient for the pairs was computed
by:
where E,,
m’ N-m
Nfr
xn yr1+r I
is the cross-correlation function ofa pair ofsignals x and y, It is the sampling interval, m is the maximum lag number, N is the number of\ample\ and l?(O) is the estimate of a signal’s autocorrelation function at lag time zero 2. The delay time, rat, at which the peak of the p(rAt) occurred, was taken as the mean lead or lag of the eye, head or gaze signal with respect to the target. To obtain
a measure
of the contribution
of head movement
to gaze movement,
and to measure gaze amplitude, the gain of each signal (gaze, eye and head) was expressed as the ratio between amplitudes of target movement and the signal at the fundamentalfrequencyofthetargetspectrum(thefrequencyatwhichthetargetmovement’s amplitude is greatest). To determine this fundamental frequency and the signal am-
42 plitudes, complete records of the position of target, head, gaze, eye and retina4 error for each segment were Fourier analyzed with the digital computer for their spectral content. The procedure for Fourier analysis was as follows : (1) Line frequency noise and FM tape noise were removed from the data by a digital low-pass filters. The fitter gain was flat until 30 Hz, at which point the gain was down 0.92 dB; at 50 Hz, attenuation was -39.6 dB. The mean was removed from the data in order to reduce low-frequency artifacts. (2) The power spectra were calculated at integer multiples of 0.098 Hz from the autocorrelation function of the filtered data. For this purpose the followingrelationship from Bendat and Piers012 was used : m-1 G,(f)
== ;- [R(O) t 2
C
C
r -~-1
R(rht)
cos (TS
c
t Qmht)
cos (zFTt)] (’
Where e,(f), an estimate of the true power spectral density function, is detined for integer multiples of 0.098 Hz and computed in the range from 0.098 Hz to 30 Hz. In the equation above, @rh) is the estimate of the autocorrelation function at tag r; m is the maximum lag number (256 was used); and fc is one half of the sampling frequency (that is, 100 Hz). The signals analyzed by the computer were a set of samples of finite duration (a truncated data series). To suppress artifacts due to this truncation, autocorrelation estimates were multiplied in the time domain by a function known as a Tukey window3. RESULTS
In agreement with previous investigator#, we found that the eye pursuit system can accurately track sinusoidal and triangular wave stimuli presented at -frequencies up to about 1 Hz over a range of &30” (the maximum range of our galvanometer), with an average gain of 0.96. This gain is calculated from combined smooth and saccadic eye movements. Step changes in target velocity are followed by changes in eye velocity with a latency of about 80 msec. (This latency is considerably smaller than that reported by Barmackr and Fuchs*, possibly because of differences in the parameters of the stimuli used.) When stimuli were presented at frequencies above 1 Hz, pursuit became quite erratic, and the monkey often failed to press the reward key appropriately. In addition, our recordings of smooth pursuit in monkeys showed frequent changes in the slope of the retinal error tracing. (Retinal error is the difference between target position and eye position.) Eye movements with velocities considerably lower than the stimulus velocity were often observed; these were indicated by steadily increasing retinal error. At other times, within the same trial, we found smootheye movements which matched the stimulus velocity so closely that no detectable retinal error was present (Fig. 4). Only occasionally did we find smooth eye movements that were faster than the stimulus movement. The smooth pursuit system apparently can cperate both with and without retinal slip. Furthermore, on the average, the combined smooth and saccadic eye movements lagged the target by approximately 20 msec (Table I).
43
I
Fig. I. Pursuit with combined eye and head movement. Pursuit movements of eye and head are shown together with the computed gaze and retinal error. The target, which periodically reverses direction, is superimposed on both head and gaze tracings. The head pursued the target, while the eyes remained relatively close to the center of the orbit. There is no obvious difference between the retinal error pattern recorded here and that observed in monkeys whose heads were restrained.
The hradpursuit
system
(A) He&fixed.
In an animal
with the head fixed, head torque can be measured
as an indication of intended head movements. These measurements showed that head torque is clearly related to target motion, indicating that the neck musculature was programmed to move the head even when the resulting muscle torques were rendered completely ineffective by blocking the head’s movemenP. There were also occasional
44 TABLE I The averagedgain
of eye and headrelative
to the target und the lag of gaze and head relariw
to
rhr
tafgct
The gains were computed from the amplitude of the fundamental Fourier component. and the delay was taken as the peak of the cross correlation function (see Methods). Several trials (n) over a range of amplitudes (A) and frequencies (Hz) are averaged for each monkey, both with head free and head fixed. Target
C2Cfr.c Head
M-66
Head free
Head fixed
M-73 Head free
Head fixed
n 24 A 21.6’ + 8. I 0.1-0.68 Hz 0.36 f 0.28 n 26 A .~ 23.5’ -I_9.4 0.1-0.88 Hz 0.34 t 0.28
0.83 1 0.16
n= 14 A 17.6” j_ 7.3 0.1-0.88 Hz 0.38 i- 0.31 n -~ 12 A -- 14.9” 3: 6.7 0.1-1.50 Hz 0.67 % 0.44
0.71 + 0.21
0.29 + 0.26
34 $ 34
0.94 .I: 0.06
36 -! 33
128 _f 83
discontinuities in the head torque record - suggesting that a sudden head movement may have been programmed ; however, the eyes remained on target whether or not these head torques were present. (B) Head,free. Allowing a monkey to pursue a target with a combination of eye and head movements by freeing its head had no evident effect on the totalgazeronrpared to head fixed tracking (Fig. 1). Both with head fixed and with it free, the movement of the gaze and that of the target had approximately the same amplitude, and, in both cases, the gaze lagged the target by about 20 msec (Table I and schematic Fig. 6). Although the total gaze is on target with or without head movement,~the eye movements are very much different. With head fixed, as stated earlier, the eyes very nearly match target amplitude and, on the average, lag the target by some 20 msec. With the head free to pursue the target, the eyes must now make up the difference between target and head in order to keep the gaze on target. Typically, the head may lag the target by some 75 msec and move about 85 ‘4 of the target’s amI;litude. The eyes now make upthe missing 15 % of target amplitude, and in addition must lead the target to make up for the lag of the head (Fig. 6). Over the range of frequencies tested, the primary effect of freeing the head was that movements of smooth pursuit were almost completely accompli3red by the head movement system, while the eyes tended to remain relatively stationary in the center
45
1
SEC
Fig. 2. Different strategies of head movement. The top record shows head movements which generally lag target movement (TARG) by about 200 msec. The bottom record shows head movements:which anticipate target movement.
of the orbit. For the 3 monkeys tested, the amplitude of head movement accounted for an average of 99 T
