INTRODUCTION

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION APPENDIX REFERENCES

The vestibuloocular reflex (VOR) has been extensively employed to investigate sensory-motor transformations and motor learning. Most studies have been performed from a simple linear system point of view, evaluating gain and phase in response to a sinusoidal or step stimulus, although various kinds of nonlinearities have been reported. A few studies have examined nonlinear dynamics of VOR motor learning by adapting the VOR using high-acceleration (Clendaniel et al. 2001) or high-frequency stimuli (Raymond and Lisberger 1996). Clendaniel et al. (2001) documented the differential VOR adaptation of the acceleration step and velocity plateau segments of a step of head motion.

Up-down asymmetries in vertical (V) eye movements have been well documented (Darlot et al. 1981; Demer 1992; Matsuo and Cohen 1984; Matsuo et al. 1979; Murasugi and Howard 1989). Although the peak velocity of upward and downward vestibular-evoked slow phases are roughly equal in magnitude, the duration of downward slow phases is generally shorter (Matsuo and Cohen 1984). Optokinetic reflex (OKR) upward slow phases have a closer relationship to stimulus velocity and a generally more regular and higher nystagmus beat frequency than their downward counterparts (Matsuo and Cohen 1984). These are another type of nonlinearity that may provide insight into the control mechanisms of the VVOR. Currently, we questioned whether these up-down asymmetries extend to the ability to modify the up-down VOR gains differentially. The frequency dependency of VOR adaptation has also been documented (Collewijn and Grootendorst 1979; Godaux et al. 1983; Hirata et al. 2000; Lisberger et al. 1983; Powell et al. 1991; Raymond and Lisberger 1996). Training using a particular frequency causes changes in VOR gain not only at the trained frequency but at adjacent frequencies as well. The magnitude of the VOR gain change at the trained frequency is always greater than that at the untrained frequency, and training at a higher frequency induces VOR gain changes in a wider frequency range around the trained frequency (Raymond and Lisberger 1996). Currently we further questioned whether VOR adaptation could modify low-high-frequency VOR gains differentially.

Herein we demonstrate that gains of the up- and downward VOR and of the low- and high-frequency VOR can be modified in opposite directions simultaneously (one to high gain, the other to low gain). We also demonstrate that the gain controls of up and downward VOR and those of low- and high-frequency VOR are not completely independent. These experimental results confine the adaptive site(s) responsible for vertical VOR motor learning to those that can process up and downward or low- and high-frequency head signals separately or specifically but not completely independently. The necessity of the cerebellar circuit to achieve these adaptations is discussed based on anatomy of VVOR. A preliminary report has been presented (Hirata et al. 2000).

	METHODS

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION APPENDIX REFERENCES

Subjects and experimental setup

Five adult squirrel monkeys weighing between 750 and 900 g were utilized for these experiments. Animals were placed in a primate chair daily for several weeks before any surgery was performed to acclimatize them to the experimental setup. For head fixation, a stainless steel bolt was secured to the occiput using small, stainless steel screws and dental cement. For eye-movement recording by the scleral search coil technique, a prefabricated eye coil constructed of Teflon-insulated stainless steel wire (Cooner) was implanted under the conjunctiva at the limbus of either eye and sutured to the sclera, and the twisted ends of the coil wire led to an occipital plug. All surgery was performed in a sterile operating suite using induction by ketamine and inhalation anesthesia using isoflourane. All wounds were treated daily with antibiotics. The Animal Welfare and Use Committee of Washington University approved all procedures and experiments.

Animals were seated in a primate chair with their heads fixed in the center of a magnetic field generated by two sets of field coils driven in quadrature. The eye-coil output was led to a phase-locked detector the output of which gave signals proportional to horizontal and vertical eye position. The chair was placed in the center of a white cylindrical screen 1 m in diameter (extending 36 cm above and 50 cm below the animal's head) on which black random dots were projected. This is the optokinetic stimulus (OKS). The right side of the animal was placed down so that rotation of the chair and the OKS about an earth vertical axis produced up- or downward nystagmus. The axis of the chair and the OKS rotations were aligned with the animal's interaural axis. Vertical and horizontal eye velocities were calibrated assuming that their gain during VOR in light at 0.5 Hz is unity (Page 1983). Horizontal and vertical eye position, OKS velocity, and chair velocity were continuously digitized at a sampling frequency of 200 Hz with the use of a CED 1401 interface (Cambridge Electronic Design) for display and storage using the Spike-2 program.

Animals were exposed in the following visual-vestibular interaction paradigms to modify their VOR gains.

Up-down asymmetrical visual-vestibular interaction

Three monkeys (1-3) were used. In this paradigm, the vestibular chair velocity was an 0.5-Hz sinusoid with an amplitude of 40°/s [40 sin(t)] while the OKS movement was a full-rectified 0.5-Hz sinusoid with the same amplitude [|40 sin(t)| or |40 sin(t)|]. When OKS is |40 sin(t)|, it moves in-phase with upward head motion and is 180° out-of-phase during downward head motion (Fig. 1A). Thus to reduce image slip on the retina, the animal's VOR in response to upward head motion should be suppressed while its VOR in response to downward head motion should be enhanced. Conversely, when the OKS is |40 sin(t)| (Fig. 1B), the VOR in response to upward head motion should be enhanced while that to downward head motion should be suppressed. The former paradigm is called up suppression-down enhancement (Up-Sup/Dn-Enh) and the latter, up enhancement-down suppression (Up-Enh/Dn-Sup).

View larger version (43K): [in this window] [in a new window]   Fig. 1. Examples of eye velocity during up-down asymmetrical visual-vestibular mismatch paradigms from monkey 1. A: up head suppression-down head enhancement (Up-Sup/Dn-Enh) paradigm in which the optokinetic stimulus (OKS) moves in phase with upward head motion and out of phase with downward head motion during sinusoidal head movement with a frequency of 0.5 Hz and amplitude of 40°/s. B: up head enhancement-down head suppression (Up-Enh/Dn-Sup) paradigm in which the OKS moves out of phase with upward head motion and in phase with downward head motion during the same head movement as in A. Note that VOR eye movements are suppressed during upward head motion and enhanced during downward head motion in A, whereas in B they are enhanced during upward head motion and suppressed during downward head motion. Only 2 cycles of the stimulus are shown. Monkeys had not been trained to pursue the visual target.

To test dependencies between up- and downward VOR adaptation, one of the monkeys (1) was trained by the following up or down alone paradigms: in the up suppression-down dark (Up-Sup/Dn-Dk) paradigm, the chair and the OKS movements are the same as those in the Up-Sup/Dn-Enh paradigm except that the OKS projection light was blanked (i.e., no OKS, chair rotates in dark) during downward head motion. Thus to reduce image slip on the retina, the animal's VOR in response to upward head motion only should be suppressed while no modification is required for its VOR in response to downward head motion. Similarly, in up enhancement-down dark (Up-Enh/Dn-Dk) paradigm, the chair and the OKS move in the same way as in the Up-Enh/Dn-Sup paradigm except that the OKS projection light was blanked during downward head movement. Likewise, in up dark-down suppression (Up-Dk/Dn-Sup) and up dark-down enhancement (Up-Dk/Dn-Enh) paradigms, the OKS projection light was blanked during upward head rotation while the OKS moves in phase (Up-Dk/Dn-Sup) or 180° out of phase (Up-Dk/Dn-Enh) with the chair, respectively.

These results were compared with our previous results on behavioral experiments (Hirata and Highstein 2001) in which three monkeys (3, 4, 5) were trained by using ordinal symmetrical visual-vestibular mismatch paradigms as follows: in the VOR-suppression (VORs) paradigm, the chair and the OKS move in phase, whereas in the VOR-enhancement (VORe) paradigm, they move 180° out of phase. The chair movement is the same as in the other paradigms mentioned in the preceding text.

Low-high-frequency-independent visual-vestibular interaction

Three monkeys (1, 2, 4) were used. In this paradigm, the chair movement consists of a sum of 0.05- and 2.5-Hz sinusoids with a maximum amplitude of 40°/s [20 sin(0.1t) + 20 sin(5t)] while the OKS is a combination of the same sinusoids but with different signs [20 sin(0.1t)  20 sin(5t) or 20 sin(0.1t) + 20 sin(5t)]. When OKS is 20 sin(0.1t)  20 sin(5t) (Fig. 8A), the low-frequency components of head and OKS movements are in phase, whereas the high-frequency components are 180° out of phase. To reduce retinal slip during this paradigm, the animal's VOR in response to low-frequency head motion should be suppressed and that to high-frequency head motion should be enhanced. On the other hand, when OKS is 20 sin(0.1t)+20 sin(5t) (Fig. 8B), the low-frequency components are 180° out of phase, requiring an enhancement of the low-frequency VOR, whereas the high-frequency components are in phase requiring suppression of the high-frequency VOR. We call the former paradigm low-frequency suppression-high-frequency enhancement (Lo-Sup/Hi-Enh) and the latter low-frequency enhancement-high-frequency suppression (Lo-Enh/Hi-Sup).

To test dependencies between low- and high-frequency VOR adaptations, one of the monkeys (2) was trained by using the following low- or high-frequency alone paradigms: low-frequency suppression (Lo-Sup) and high-frequency suppression (Hi-Sup) paradigms are VORs at 0.05 and 2.5 Hz, respectively, whereas low-frequency enhancement (Lo-Enh) and high-frequency enhancement (Hi-Enh) paradigms are VORe at 0.05 and 2.5 Hz, respectively.

Animals were trained by using these paradigms for at most 4 h/day. Amphetamine sulfate (0.3 mg/kg) was given orally 30 min before the training began to maintain a constant level of alertness. For the up-down asymmetrical training, VOR in dark (VORd) at 0.5 Hz was recorded for 1 min before and 0.5, 1, 2, 3, and 4 h after the beginning of the training to measure VOR gains of up and down head movements, phase and a DC eye-velocity bias (see following text for measurement of up- and downward VOR gains, phase and the DC bias). For the low-high-frequency asymmetrical training, VORd at 0.05 and 2.5 Hz were recorded for 1 min each before and 0.5, 1, 2, 3, and 4 h after the beginning of the training to measure gains, phases, and DC eye-velocity bias at these frequencies.

Data analysis

To calculate the VOR gains, phases and the DC biases, the following regressions were applied to data recorded during the VOR in darkness (VORd)


where v_eye(t) and v_head(t) denote desaccaded vertical eye and head velocity, respectively. Saccades and postsaccadic drifts, if any, were eliminated from eye-movement traces and head-velocity traces by using an automated desaccading algorithm that was visually checked on the computer screen. A and B are regression coefficients estimated to minimize the squared sum of the error term (t). A is the VOR gain and B is the DC offset in eye velocity that was close to 0 before adaptation in most cases.  seconds is the delay time between head and eye velocity and was globally searched to obtain the minimum squared sum of (t). The phase shift was calculated as 18 and 900° for 0.05 and 2.5 Hz of the stimulus, respectively. A_up and _up(t) are the gain of VOR responding to upward head motion (the upward VOR gain), and the error term, respectively, whereas A_dw and _dw(t) are those for downward head motion (the downward VOR gain and the error). B_up-dw is the DC bias of eye velocity, and it is estimated simultaneously with A_up and A_dw to minimize squared sum of _up(t) plus that of _dw(t).

Retinal image slip was evaluated to quantify the animals' performances in each paradigm. Retinal slip velocity was calculated as OKS velocity - chair velocity - eye velocity after the traces were desaccaded, then averaged over each cycle for the initial 30 min of the training. For the Up-Sup/Dn-Enh and Up-Enh/Dn-Sup paradigms, amplitudes and phases of the retinal slip during up- and downward head movements were estimated by fitting a different 0.5-Hz sinusoidal wave to the desaccaded and averaged retinal slip velocity in each half cycle by adjusting amplitude and phase of each sinusoid. To estimate the amplitude and phase of retinal slip velocity at low- and high-frequency independently for the low-high-frequency asymmetrical paradigms, the sum of two sine waves (0.05 and 2.5 Hz) were fit to the desaccaded and averaged retinal slip velocity by adjusting the amplitude and phase of each sinusoid. Matlab (Mathworks) nonlinear optimization method was utilized for the curve fittings.

Theoretical predictions

Theoretical analyses were executed to predict how the VOR system might adapt to up-down asymmetrical visual-vestibular interaction to minimize retinal slip. Four possibilities were considered: the up and down VOR systems have a common gain control, independent gain controls, a common gain and a phase control, and independent gain and phase controls. A DC eye-velocity control element is available in all the four conditions. This is based on the fact that Y group neurons and flocculus Purkinje cells change their DC firing rates in parallel with VVOR gain (Hirata and Highstein 2001; Partsalis et al. 1995b). Optimal VOR gain or/and phase and a DC bias that minimize squared sum of retinal slip velocity during the Up-Enh/Dn-Sup and Up-Sup/Dn-Enh paradigms were estimated under each of four conditions. The same kind of theoretical analyses were executed to predict how the VOR system might adapt to the low-high-frequency asymmetrical visual-vestibular interaction paradigms. The same four possibilities were considered, namely, a common gain control, independent gain controls, a common gain and phase control, and independent gain and phase controls for low- and high-frequency VOR systems. A DC eye-velocity control element is again available in these four conditions as in the case of up-down asymmetrical adaptation. Under these conditions, optimal VOR gains and/or phases and a dc component that minimize squared sum of retinal slip velocity were estimated. See APPENDIX for further description.

	RESULTS

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION APPENDIX REFERENCES

In keeping with our previous convention (Partsalis et al. 1995a,b), eye movement during training of the VOR by visual-vestibular mismatch stimuli is denoted as a "rapid modification of the VOR;" change in VOR in the dark (VORd) tested between the inception of training and 4-6 h is denoted as an "acute learning" or "acute adaptation of the VOR." VORd changes following long-term wearing of lenses are denoted as "chronic learning" or "chronic adaptation of the VOR." VOR gain always refers to gain in the dark, and up- or downward VOR gain refers to VOR gain during upward or downward head motion (not eye), respectively.

Up-down asymmetrical VOR adaptation

Figure 1 illustrates examples of eye movements (rapid modification of the VOR) during up-down asymmetrical VOR adaptation training. In Fig. 1A, stimuli require the animal to suppress eye velocity when the head moves upward and to enhance eye velocity when the head moves downward (Up-Sup/Dn-Enh). One in B is just the opposite or Up-Enh/Dn-Sup. Only two cycles of stimulus are shown. In both A and B, head velocity is a 0.5-Hz sinusoid, while the OKS moves exclusively upward in A and downward in B. In A the VOR is suppressed during upward head movement and enhanced during downward head movement. In B the converse is true as the VOR is enhanced during upward head movement and suppressed during downward head movement. Performance of all monkeys so tested was qualitatively similar, therefore monkeys are capable of the rapid modification of the VOR required by these paradigms (cf. Fig. 2).

View larger version (29K): [in this window] [in a new window]   Fig. 2. Performance in rapid vestibuloocular reflex (VOR) modification during the up-down asymmetrical paradigms and conventional symmetrical paradigms. Amplitude and phase of retinal slip velocity during up- and downward head rotations averaged over the initial 30 min of training period are indicated in the polar coordination. Results from 3 monkeys (1, 2, 3) for up-down asymmetrical paradigms and those from 3 monkeys (3, 4, 5) for symmetrical paradigms are superimposed. Note that the retinal slip induced by the suppression portion (plots in right hemisphere) of the asymmetrical paradigms is smaller than that induced by the enhancement portions (plots in left hemisphere) while no significant difference is found in retinal slip during suppression and enhancement portions of the symmetrical paradigms.

Figure 1 also suggests that animals are better able to perform the suppression half cycles of these asymmetrical paradigms than the enhancement half cycles. Namely, during suppression the actual eye velocity is close to the required eye velocity of zero. However, during the enhancement half cycles the actual eye velocity is much less than the required eye velocity of 80°/s at the peak head velocity of 40°/s. These performance features of the VOR rapid modifications during up-down asymmetrical paradigms are plotted in Fig. 2.

Figure 2 is a polar diagram of retinal slip velocity during rapid modifications of the VOR in the Up-Sup/Dn-Enh and Up-Enh/Dn-Sup paradigms. Amplitudes and phases of retinal slip velocity desaccaded and averaged over the initial 30 cycles in each paradigm were plotted as the radius and angle of the diagram, respectively. Results from three monkeys (1- 3) are superimposed. Figure 2 emphasizes that neither suppression nor enhancement is perfect. If they were to be perfect, there would be no retinal slip. It is also apparent as suggested by Fig. 1, that the retinal slip required by the suppression portion (plots in right hemisphere) of the asymmetrical paradigms is always smaller than that required by the enhancement portions (plots in left hemisphere) in individual animals.

To compare with performance of rapid VOR modification in conventional symmetrical training, retinal slip amplitude and phase during VORe and VORs at 0.5 Hz were plotted as filled symbols. The amplitude and phase of retinal slip during VORs are comparable with those of upward head motion during the Up-Sup/Dn-Enh paradigm and downward head motion during the Up-Enh/Dn-Sup paradigm. In contrast, the amplitude of retinal slip during VORe is significantly smaller than that of upward head motion during the Up-Enh/Dn-Sup and that of downward head motion during the Up-Sup/Dn-Enh paradigm. In fact, the amplitudes of retinal slip of up- and downward head motion during VORs are almost the same as those during VORe. Thus monkeys' rapid VOR modification performance during the up-down asymmetrical paradigms is degraded during the enhancement half of the asymmetrical paradigms while their performance during the suppression half is unchanged in comparison with the symmetrical paradigms.

Figure 3 illustrates examples of eye velocity during VORd before and after 4 h of up-down asymmetrical training. A is Up-Sup/Dn-Enh and B, Up-Enh/Dn-Sup. In both A and B, desaccaded eye-velocity traces averaged over 30 cycles (1 min) before and after training are superimposed with head velocity. In A, during up head motion (0-1 s), the eye velocity after training is smaller in amplitude than before training, whereas during downward head motion (1-2 s), eye velocity after training is larger. VOR gains during upward head motion before and after training are 0.96 and 0.77, respectively, whereas those of downward head motion are 0.88 and 1.06, respectively. In contrast, eye velocity after training with the Up-Enh/Dn-Sup paradigm in B is larger during upward head motion and smaller during downward head motion than before the training. In this case, VOR gains of upward head motion before and after the training are 0.97 and 1.17, respectively, whereas those of downward head motion are 0.96 and 0.66, respectively. Note that there is a slight upward DC eye-velocity shift (intersection at 1 s) in A (Up-Sup/Dn-Enh) and a downward shift in B (Up-Enh/Dn-Sup) after 4 h adaptation (cf. Fig. 5). The gray shadow indicates a period in which the number of cycles to calculate the average eye velocity was <3 due to saccades and postsaccadic drifts.

View larger version (28K): [in this window] [in a new window]   Fig. 3. Examples of VOR eye velocity in darkness before and after 4 h of Up-Sup/Dn-Enh (A) and Up-Enh/Dn-Sup training (B) from monkey 1. The eye-velocity traces were desaccaded and averaged over 30 cycles of head velocity. Note that amplitude of the eye velocity is smaller in response to upward head motion and larger in response to downward head motion after the Up-Sup/Dn-Enh training, whereas the eye-velocity amplitude is larger in response to upward head motion and smaller in response to downward head motion after the Up-Enh/Dn-Sup training. The results indicate that the up-head VOR gain decreased and the down-head VOR gain increased simultaneously after the Up-Sup/Dn-Enh training, whereas the up-head VOR increased and the down-head VOR decreased simultaneously after the Up-Enh/Dn-Sup training. Note also that dc eye-velocity bias (intersection at time 1 s) in A increased in the upward direction and that in B increased in the downward direction. , there were only a few cycles of data following the desaccading procedure.

Figure 4 illustrates the learning curves of upward and downward VOR gains, phases, and DC eye-velocity biases measured in darkness. The abscissa is the training time in hours and the ordinate is changes in upward (Up: black line) and downward (Dn: gray line) VOR gain (A and B), phase angle in degrees (C and D) and DC eye velocity in degrees per second (E and F) from the initial states (before the training). Results from three animals are superimposed. In A, C, and E, the Up-Sup/Dn-Enh training was executed three times in animal 1 (solid line) and one time each in animals 2 (broken line) and 3 (dashed line) to evaluate the repeatability of results in the same animal and their generality in different animals. Similarly in B, D, and F, the Up-Enh/Dn-Sup training was executed four times in animal 1 (solid line) and one time each in animals 2 (broken line) and 3 (dashed line). Same symbols in black and gray traces indicate samples from the same experimental session. Note that there is some variability in the learning curves indicated in A and B. Variability in ordinal symmetrical VOR learning versus time plots has previously been reported (Raymond and Lisberger 1996), and the present experiments also manifest this variability during asymmetrical learning. The initial VOR gains in A range from 0.87 to 0.96 (mean 0.92 ± 0.03 SD) for upward head motion and from 0.77 to 0.97 (0.87 ± 0.07) for downward head motion, whereas in B, they range from 0.75 to 0.97 (0.87 ± 0.09) for upward head motion and from 0.82 to 0.96 (0.86 ± 0.05) for downward head motion. The initial phase shifts in C range from 1.80 to 3.60° (0 ± 2.11 SD), and those in D, range from 1.80 to 2.70° (0.60 ± 1.77). The initial DC eye velocities in E range from 3.26 to 4.49°/s (0.93 ± 3.27), and those in F range from 5.54 to 8.29°/s (3.81 ± 5.26).

View larger version (38K): [in this window] [in a new window]   Fig. 4. Learning curves of up (Up) and down (Dn) VOR gains (A and B), phase angle (C and D), and DC eye-velocity bias (E and F) in the Up-Sup/Dn-Enh and Up-Enh/Dn-Sup paradigm. Data from 3 monkeys (1- 3) are superimposed. Monkey 1 provided 3 samples (solid line), whereas monkeys 2 (broken line) and 3 (dashed line) provided 1 each for the Up-Sup/Dn-Enh paradigm. For the Up-Enh/Dn-Sup paradigm, monkey 1 provided 4 samples (solid line), whereas monkeys 2 (broken line) and 3 (dashed line) provided 1 each. Black lines indicate Upward VOR gain and gray lines, downward VOR gain. Identical symbols for black and gray traces indicate data from a same experiment. Note that up and downward VOR gains gradually changed in the opposite directions simultaneously in both A and B.

In Fig. 4A, VOR gains in response to upward head motion decreased and those to downward head motion gradually increased during the Up-Sup/Dn-Enh training, whereas in B, those to up and down head motion increased and decreased, respectively, during the Up-Enh/Dn-Sup training. Therefore monkeys are capable of changing up- and downward VOR gains in opposite directions simultaneously. This provides direct evidence that there are separate gain control mechanisms for the up- and downward VOR. The average gain change following 4 h of acute asymmetrical training was about 0.2. VOR phases remained close to zero as illustrated in C and D. There was a slight DC bias upward following Up-Sup/Dn-Enh training (E), whereas the opposite paradigm did not produce an equivalent downward bias (F).

Theoretical analyses were executed to predict how the VOR system might adapt to the Up-Sup/Dn-Enh and Up-Enh/Dn-Sup paradigms (cf. APPENDIX). Figure 5 illustrates the predictions of eye movements after adaptation to the Up-Sup/Dn-Enh (A) and Up-Enh/Dn-Sup paradigms (B) assuming that the up- and downward VOR systems share a common gain control and a DC bias control mechanism (Com. gain ctrl.). The estimated optimal gain and DC bias that minimize the error [difference between an ideal VOR producing no retinal slip (thin line) and an adapted VOR (thick line)] are 1 and 25.452°/s, respectively, for Up-Sup/Dn-Enh training and 1 and 25.452°/s, respectively, for Up-Enh/Dn-Sup training. The result predicts that if monkeys do not have separate gain control mechanisms for up and down VOR, the best strategy would be to generate a DC bias without changing the VOR gain. The direction of the dc shift after Up-Sup/Dn-Enh adaptation is upward (A) while that after Up-Enh/Dn-Sup adaptation is down (B). As can be seen, even in the optimal case, the error does not become 0. The results are summarized in Table 1, Com. gain ctrl. Com. gain/phase ctrl. in Table 1 shows that even if animals can change the VOR phase together with a common gain and a DC bias control element shared by up and downward systems, the best strategy to minimize the error would be to keep the phase shift at 0 with the gain 1 and the DC bias 25.452°/s similar to Com. gain ctrl. On the other hand, if animals have separate gain control for up and down VOR (Ind. gain ctrl. in Table 1), then the best strategy would be to decrease the VOR gain during upward head movement to 0 and increase the gain during downward head movement to 2 for Up-Sup/Dn-Enh. For Up-Enh/Dn-Sup, the best strategy would be to increase the VOR gain during up head motion to 2 and decrease gain during down head motion to 0. In both cases, the error can be minimal. Even if animals have independent phase control of the VOR as well as separate gain controls for up and down VOR (Ind. gain/phase ctrl, Table 1), the best strategy to minimize error would be to keep the phase shift at 0. 

View larger version (22K): [in this window] [in a new window]   Fig. 5. Theoretical prediction of eye velocity after the Up-Sup/Dn-Enh (A) and Up-Enh/Dn-Sup (B) training under the assumption that up- and downward VOR systems do not have separate gain control mechanisms. The optimum eye velocity (after: thick solid line) was estimated to minimize sum of retinal slip velocity over a period (2 s) during each paradigm. Ideal (thin solid line) is the ideal eye velocity that achieves 0 retinal slip in each paradigm. Before (dotted line) indicates normal VOR eye velocity before training. Note that in both A and B, the predicted optimum eye velocities do not coincide with the ideal ones, and they have an upward or downward DC eye-velocity bias in A or B, respectively. Amplitude of the predicted optimum eye velocity in both A and B is same as before training. The result predicts that if the up- and downward VOR systems share a common gain control mechanism, only up- or downward eye-velocity DC bias would be generated after the Up-Sup/Dn-Enh or Up-Enh/Dn-Sup training, respectively.

Results in Table 1 only predict the final optimal states when VOR gain adaptation was completed under the Up-Sup/Dn-Enh or Up-Enh/Dn-Sup paradigm and do not provide information during the ongoing adaptation. Table 2 summarizes phases and DC bias that minimize the error during ongoing Up-Sup/Dn-Enh and Up-Enh/Dn-Sup adaptation under the Ind. gain/phase ctrl. condition in Table 1 when up and downward VOR gains are between 0 and 2. Before the Up-Sup/Dn-Enh training, i.e., when both the up and downward VOR gains are 1, the minimum error is obtained when the dc bias is a positive value (25.452°/s) and the phase shift is 0°. From that point, the error decreases to a global minimum as up and downward VOR gains get closer to 0 and 2, respectively, while the optimum phase stays constant and the optimum DC bias decreases to 0. On the other hand, before the Up-Enh/Dn-Sup training, the error is minimized by a negative DC bias (25.452°/s) and a phase shift of 0°. The error gets closer to the global minimum as the up and downward VOR gains get closer to 2 and 0, respectively, while the optimum phase stays at 0 and the optimum DC bias increases from the negative value to 0. The optimum DC bias that minimizes the error depends on the values of the up- and downward VOR gains while the optimum phase is always 0. The results predict that a DC eye-velocity bias would be produced in the VORd while monkeys are adapting to the Up-Sup/Dn-Enh or Up-Enh/Dn-Sup paradigm as demonstrated in Fig. 4. No phase shift is expected even if the VVOR system has separate phase control mechanisms for up- and downward head movements.

We next investigated the independency of these putative, separate gain control mechanisms. If up and down gain controls are not independent, it might be expected that a larger gain decrement in the upward VOR would accompany a smaller gain increment or even a decrement in the downward VOR gain during Up-Sup/Dn-Enh training or vice versa. To evaluate this potential dependency, experiments were designed with the usual stimulus for asymmetrical training except the visual scene was blanked during specific half cycles of the head-velocity stimulus. For example, the visual stimulus moved with the head during upward head movement causing VOR suppression, but there was no visual stimulus during downward head motion. Results of 4 h of training using the four permutations of this approach are plotted in Fig. 6. In all cases, the gain of the untrained half cycle was pulled in the direction of the change caused by the training in the trained half cycle. For example, in A, Up-Sup/Dn-Dk training, the animal was asked to suppress its VOR during upward head movement, while there was no visual stimulus during downward head movement. The VOR gain during the half cycle of upward head movement steadily decreased as did that in the untrained half cycle, even though there was no visual stimulus. In B, Up-Enh/Dn-Dk training, the animal was asked to enhance its VOR during upward head movement, but there was no visual stimulus during downward head movement. VOR gain steadily increased during both the trained and untrained half cycles. This parallel change of gain in the untrained half cycles toward that resulting from the training was seen during all four permutations (A-D). This suggests that the gain controllers for the VOR during up- and downward head movement are not completely independent but influenced each other.

View larger version (30K): [in this window] [in a new window]   Fig. 6. Learning curves of up (Up)- and downward (Dn) VOR gains in up alone or down alone training paradigm in the same format in Fig. 4, A and B. Note that VOR gains during the untrained half cycle changed in the same direction as those during the trained half cycle in all 4 permutations of the paradigm.

To further evaluate this dependency, we compared the gain changes during asymmetrical training with those produced by symmetrical training. Because dependencies between up- and downward VOR gain changes have now demonstrated, smaller magnitudes of VOR gain change during the asymmetrical training might be expected when compared with those caused by ordinal, symmetrical training. Figure 7 overlays the results of symmetrical VOR training with that of asymmetrical training illustrated in Fig. 4, A and B. In spite of the noted variability in results, the magnitude of the averaged gain change produced by symmetrical training exceeded that produced by asymmetrical training at all time points. This suggests that, although there may be separate gain control mechanisms for up- and downward head movements, they are not completely independent. There were little changes in phase and DC bias during symmetrical training (not illustrated).

View larger version (24K): [in this window] [in a new window]   Fig. 7. Comparison of learning curves in up-down asymmetrical adaptation with those of symmetrical adaptation. Up (Up)- and downward (Dn) VOR gains in asymmetrical and symmetrical adaptations are superimposed in the same format as in Figs. 4, A and B, and 6. A: up- and downward VOR gains in the Up-Sup/Dn-Enh paradigm in comparison with upward VOR gain in VORs and downward VOR gain in VORe, respectively. B: up- and downward VOR gains in the Up-Enh/Dn-Sup paradigm in comparison with upward VOR gain in VORe and downward VOR gain in VORs, respectively. Note that average VOR gain changes in the up-down asymmetrical adaptations are always smaller than those in the symmetrical adaptations.

Low-high-frequency asymmetrical adaptation

Three monkeys (1, 2, 4) were used in the asymmetrical frequency training experiments. Two of them (1 and 2) are the same animals that were used for the up-down asymmetrical training.

Figure 8 illustrates examples of eye movements in response to low-high-frequency asymmetrical visual-vestibular interaction stimuli. One cycle of the stimulus (20 s) is shown. During the Lo-Sup/Hi-Enh paradigm in A, the animal suppressed eye movements at the low-frequency (0.05 Hz) stimulus and enhanced them only partially at the high-frequency (2.5 Hz) one. During the Lo-Enh/Hi-Sup paradigm in B, eye movements at the low frequency were enhanced while those at the high frequency were only partially suppressed. The same tendency in the performance of the rapid modifications was found in the other animals, as summarized in Fig. 9.

View larger version (52K): [in this window] [in a new window]   Fig. 8. Examples of eye velocity during low-high-frequency asymmetrical visual-vestibular mismatch paradigms from monkey 1. A: low-frequency suppression-high-frequency enhancement (Lo-Sup/Hi-Enh) paradigm in which the low-frequency component (0.05 Hz) of the optokinetic stimulus (OKS) moves in phase with that of head movement and the high-frequency component (2.5 Hz) of OKS moves out of phase with that of head movement during a sum of sines of head movement consisting of 0.05- and 2.5-Hz sinusoids with the maximum amplitude of 40°/s. B: low-frequency enhancement-high-frequency suppression (Lo-Enh/Hi-Sup) paradigm in which the low-frequency component of the OKS moves out of phase with that of the head movement and the high-frequency component of the OKS moves in phase with that of head movement during the same head movement as in A. Note that the low-frequency VOR component is suppressed and enhanced in A and B, respectively. while the high-frequency component is neither suppressed nor enhanced in A and B, respectively. Sharp diffractions in the eye-velocity trace are saccades. Only 1 cycle of the stimulus is shown. Monkeys had not been trained to pursue the visual target.

View larger version (28K): [in this window] [in a new window]   Fig. 9. Performance in rapid VOR modification during the low-high-frequency asymmetrical paradigms and low- or high-frequency-alone paradigms. Amplitude and phase of retinal slip velocity at low and high frequency averaged over the initial 30 min of training period are indicated in the polar coordination. Results from 3 monkeys (1, 2, 4) for low-high-frequency asymmetrical paradigms and those from 1 monkey (2) for low- or high-frequency-alone paradigms are superimposed. Note that the retinal slip at low frequency is smaller than that at high frequency in both low-high-frequency asymmetrical paradigms and low- or high-frequency-alone paradigms. Note also that retinal slip at the low frequency in low-high-frequency asymmetrical paradigms is greater than that in the low- or high-frequency-alone paradigms, while retinal slip at the high frequency in both low-high-frequency asymmetrical and low- or high-frequency paradigms alone is comparable.

Figure 9 plots the performance of the rapid modifications of the VOR resulting from asymmetrical frequency training as retinal slip. As seen in the example in Fig. 8, retinal slip amplitudes of the high-frequency component (black symbols) are greater than those of the low-frequency component (gray symbols). In contrast to up-down asymmetrical training, retinal slip required by the suppression portions of these paradigms (plots in right hemisphere) was not less than that required by enhancement (plots in left hemisphere). Further, the phase shifts were more pronounced than during up-down training.

Retinal slip velocity amplitude and phase during VORe and VORs at high and low frequency were plotted as filled symbols to contrast them with the performance of rapid VOR modification in low-high asymmetrical frequency paradigms. Retinal slip during VORs at high frequency (Hi-Sup, filled black square) is comparable with that during Lo-Enh/Hi-Sup at high frequency (open black circles). The same is true for retinal slip during VORe at high frequency (Hi-Enh, filled black triangle) and that during Lo-Sup/Hi-Enh at high frequency (black crosses). In contrast, retinal slip during VORs at low frequency (Lo-Sup, filled gray square) and VORe at low frequency (Lo-Enh, filled gray triangle) are significantly smaller than that during the Lo-Sup/Hi-Enh (gray crosses) and Lo-Enh/Hi-Sup paradigms at low frequency (open gray circles). Further, retinal slip during the Lo-Enh/Hi-Enh paradigm in which the animal was asked to make rapid VOR modification at low and high frequencies in the same direction (high gain) simultaneously is plotted as black (high-frequency) and gray stars (low frequency). Retinal slip at low and high frequencies during this paradigm are comparable with those during Lo-Enh/Hi-Sup at high-frequency (open black circles) and those during Lo-Sup/Hi-Enh at low frequency (gray crosses).

Figure 10 illustrates the learning curves of VOR gains, phases, and DC eye-velocity biases at low and high frequencies measured in darkness. For both Lo-Sup/Hi-Enh and Lo-Enh/Hi-Sup training, the abscissa is the training time in hours and the ordinates are changes in VOR gain (A and B), phase angle in degrees (C and D) and DC bias in degrees per second at low (Lo: black line) or high (Hi: gray line) frequency from their initial values (before the training). Results from three animals were superimposed. In A, C, and E, the Lo-Sup/Hi-Enh training was executed two times in animal 1 (solid line) and one time each in animals 2 (broken line) and 4 (dashed line) to evaluate the repeatability of results in the same animal and their generality in different animals. Similarly in B, D, and F, the Lo-Enh/Hi-Sup training was executed two times in animal 1 (solid line) and one time each in animals 2 (broken line) and 4 (dashed line). Same symbols in black and gray traces indicate samples from the same experimental session. The initial VOR gains in A range from 0.49 to 0.89 (0.70 ± 0.20) for low-frequency (0.05 Hz) head rotation and from 0.89 to 1.11 (1.02 ± 0.11) for high-frequency (2.5 Hz) head rotation, whereas in B they range from 0.77 to 0.84. (0.81 ± 0.04) for low-frequency head rotation and from 0.86 to 1.11 (0.97 ± 0.11) for high-frequency head rotation. The initial phase shifts in C range from 0.81 to 4.86° (2.41 ± 1.86) for low frequency and from 20.3 to 20.3° (1.01 ± 17.30) for high frequency, and those in D range from 16.2 to 20.3° (2.03 ± 17.65) for low frequency and from 0.77 to 0.84° (0.81 ± 0.036) for high frequency. The initial DC eye velocities in E range from 8.39 to 5.67°/s (1.06 ± 6.63) for low frequency and from 7.70 to 19.64°/s (1.80 ± 12.15) for high frequency, and those in F range from 3.94 to 8.52°/s (0.15 ± 5.79) for low frequency and from 2.27 to 19.64°/s (4.02 ± 10.44) for high frequency.

View larger version (36K): [in this window] [in a new window]   Fig. 10. Learning curves of low- and high-frequency VOR gains, phase, and DC eye-velocity bias of Lo-Sup/Hi-Enh training and Lo-Enh/Hi-Sup training. Results from 3 monkeys (1, 2, 4) are superimposed. Monkey 1 provided 2 samples (solid line), whereas monkeys 2 (broken line) and 3 (dashed line) provided 1 each. Formats are the same as in Fig. 4, except that Lo and Hi denote low- and high-frequency component, respectively.

In A, VOR gains in response to low-frequency head rotation decreased while those to high-frequency head rotation increased as the training proceeded, indicating that monkeys are capable of changing the gain of VVOR responding to low- and high-frequency head rotations in the opposite directions simultaneously. The result is not complementary in the case of the Lo-Enh/Hi-Sup paradigm in B. While monkeys could change their high-frequency VOR gains in the proper direction during Lo-Enh/Hi-Sup training, changes in their low-frequency VOR gains are much smaller. These results indicate that there are at least two different gain control mechanisms for low- and high-frequency VOR, but they are not independent and the dependency is directionally selective. Phase shifts at low frequency did not show significant changes in both Lo-Sup/Hi-Enh and Lo-Enh/Hi-Sup paradigms, whereas those at high-frequency showed a systematic incremental phase lag in Lo-Enh/Hi-Sup training (D) and rather random changes in Lo-Sup/Hi-Enh training (C). DC eye-velocity bias did not change significantly at the low frequency in Lo-Sup/Hi-Enh training (E) and at high frequency in Lo-Enh/Hi-Sup training (F), while a slight increment and decrement were found at high frequency in Lo-Sup/Hi-Enh training and at low frequency in Lo-Enh/Hi-Sup training, respectively (see DISCUSSION).

Results of the theoretical analyses executed to predict how the VVOR system might adapt to the Lo-Enh/Hi-Sup and Lo-Sup/Hi-Enh paradigms are summarized in Table 3. If the low- and high-frequency VVOR systems share a common gain control and a DC bias control mechanism (Com. gain ctrl.), the best strategy to minimize the error is to keep the gain of the VVOR at 1 and the DC bias 0. This result predicts that if the low- and high-frequency channels of the VVOR do not have separate gain controls, neither the VOR gain at low frequency nor that at high frequency will change after exposure to these paradigms. If they have separate gain controls (Ind. gain ctrl.), then the VOR gain at low frequency becomes 2 while that at high frequency becomes 0 in the Lo-Enh/Hi-Sup paradigm to minimize the error. The opposite is true in the Lo-Sup/Hi-Enh paradigm where the VOR gain at low frequency and at high frequency become 0 and 2, respectively. In these cases, the error becomes minimal. Even if common (Com. gain/phase ctrl.) or separate phase controls (Ind. gain/phase ctrl.) are available for the low- and high-frequency VOR together with the gain and DC bias control, the phase(s) should be kept at 0 to minimize the error. These results predict that no phase shift will occur after the asymmetrical frequency training even if animals have common or separate phase controls together with common or separate gain controls unless they have a phase lag or lead as a default. Also no DC eye-velocity bias reduces the error, thus no change in the dc bias is expected under any conditions during the low-high-frequency asymmetrical training.

Results in Table 3 only predict the final optimal states when VOR gain adaptation was completed. Theoretical prediction for the ongoing adaptation under the Ind. gain/phase ctrl. condition showed that the optimum dc bias and the phase shift do not depend on the gains of the low- and high-frequency VOR. The results indicate that neither the phase shift nor DC eye-velocity bias reduces the error at any given combination of low- and high-frequency VOR gains. Therefore it is predicted that no phase shift or DC bias should be observed during the low-high-frequency asymmetrical training.

Figure 11 evaluates the potential dependency between changes in low- and high-frequency gains by plotting changes in low- and high-frequency VOR gains from their initial values induced by low- or high-frequency training alone. For example, in A the animal was exposed to VORs at 2.5 Hz (Hi-Sup) and tested periodically VORd at both 2.5 and 0.05 Hz. It is apparent that there were parallel decreases in both VOR gains, although the decreases at the untrained frequency are less. In keeping with the results of Raymond and Lisberger (1996) and Lisberger et al. (1983) the same can be seen to be true for the other three conditions tested. Namely, training at one frequency can result in a spillover of the training effects to another frequency, but the magnitude of the change at the untrained frequency are always less than those at the trained frequency.

View larger version (28K): [in this window] [in a new window]   Fig. 11. Learning curves of low (Lo)- and high (Hi)-frequency VOR gains in low-frequency (Lo-Sup, Lo-Enh)-alone or high-frequency-alone (Hi-Sup, Hi-Enh) training paradigm. Formats are the same as in Fig. 10. Note that VOR gains at the untrained frequency changed in the same direction as those at the trained frequency in all the 4 permutations of the paradigm.

	DISCUSSION

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION APPENDIX REFERENCES

Visual vestibular mismatch stimuli invoke visually driven mechanisms to correct inappropriate VOR behaviors. In keeping with the terminology of our previous reports (Hirata and Highstein 2001; Partsalis et al. 1995a,b), these "corrected" behaviors are termed rapid modifications of the VOR. Rapid modifications reduce retinal slip produced by the mismatch stimuli. If visual-vestibular mismatch conditions continue, the open loop gain of the VOR measured during rotation in darkness is gradually recalibrated to a new, more appropriate value over a period of hours to days. This recalibration is the phenomenon of adaptation or motor learning of the VOR (Partsalis et al. 1995a). Currently the adaptive capacity of the system during new visual-vestibular mismatch paradigms that require asymmetrical behavior and gains for up- and downward VOR or low- and high-frequency VOR was evaluated. Characteristics of the adaptation including the dependency or independency of the adaptation of the up- and downward or the low- and high-frequency VOR to these stimuli were evaluated.

Up-down asymmetrical VOR adaptation

Up-down asymmetrical visual-vestibular mismatch stimuli require the animal to cancel vestibular-evoked eye movements during one half cycle of a sinusoidal head movement and enhance them during the other half cycle. Presently we employed animals that were not trained to track a visual target, and we suppose that this lack of training resulted in eye movements that were neither completely canceled nor perfectly enhanced during these paradigms. However, animals could suppress and enhance a significant amount of their VOR to reduce retinal slip during both paradigms. Therefore monkeys were capable of making rapid modifications of the VOR during these paradigms. Quantitatively though, the enhancement is less good than that evoked by the symmetrical visual-vestibular mismatch paradigms while the suppression is comparable in both asymmetrical and symmetrical paradigms. The result suggests that suppressing the VOR is easier for monkeys than enhancing it, or their priority is on the suppression if they have to make both of the rapid VOR modifications simultaneously. During 4-h exposure, VOR gains to up- and downward head motion were gradually modified to more appropriate values and up-down asymmetrical VOR adaptation occurred. We confirmed the repeatability and generality of this learning. The magnitudes of VOR gain changes do not seem different in up- and downward head motion in both asymmetrical trainings, although there was significant difference in ability to suppress and enhance VOR in these paradigms. This provides a piece of evidence that performance in the rapid VOR modification, i.e., amount of retinal slip, does not directly account for amount of acute VOR gain change (Hirata and Highstein 2001). That the gain of the up- and downward VOR evoked during a continuous sinusoidal head movement can adapt in opposite directions is direct evidence for separate up and down VOR adaptation mechanisms. However, these adaptation mechanisms are not completely independent because experiments designed to train in one direction only resulted in gain changes in the untrained direction simultaneously, albeit of a lesser magnitude. Further, the magnitudes of the gain changes produced by asymmetrical training within a fixed time period were less than those produced by symmetrical training. A simple interpretation is that there is a limited capacity for VOR adaptation that was exceeded by the requirements of asymmetrical training.

A phase control mechanism is available in the VOR adaptation process in human (Kramer et al. 1998). The phase of the VOR during up-down asymmetrical adaptation did not show significant changes. This result is consistent with the theoretical prediction that indicated that a change in phase angle would not reduce retinal slip when adapting to the Up-Sup/Dn-Enh and Up-Enh/Dn-Sup paradigms.

Y group neurons and floccular Purkinje cells- which are candidate sites of VVOR motor learning (see following text)- often change their DC firing rates during conventional up-down symmetrical VOR adaptation. Y group neurons increase their DC firing rate as VOR gain increases (Partsalis et al. 1995b), whereas Purkinje cells decrease their DC firing rate (Hirata and Highstein 2001). Both of these cell types usually show high eye-velocity sensitivities and thus may contribute to change the DC eye-velocity bias that may cause a gaze holding problem after VOR adaptation. However, there is not usually a change in DC eye velocity following acute conventional VOR adaptation. A possible explanation for this is that the velocity-to-position integrator compensates the changes in DC eye-velocity signal in the process of symmetrical VOR adaptation. Preliminary results have shown that floccular Purkinje cells increased their DC firing rate after the Up-Sup/Dn-Enh training, whereas the DC firing rate decreased after the Up-Enh/Dn-Sup training (Hirata et al. 2000). In conventional symmetrical adaptation, the DC firing rate of Purkinje cells decreased as the VOR gain increased. Because it is the upward VOR gain that increases when the Purkinje cell DC firing rates decrease in both the symmetrical and asymmetrical adaptation, it seems that the Purkinje cell dc firing rate is correlated with the upward VOR gain. Theoretically a change in DC eye velocity could reduce retinal slip error when VOR gains are not fully adapted to the up-down asymmetrical stimuli. The simulation predicted that the DC eye velocity would increase in the upward direction during the Up-Sup/Dn-Enh training and in the downward direction during the Up-Enh/Dn-Sup training. The experimental results of the Up-Sup/Dn-Enh training were mostly consistent with this prediction, while those of the Up-Enh/Dn-Sup training showed a trend opposite to the prediction. Therefore it is likely that the observed changes in the DC eye-velocity bias are a byproduct of the asymmetrical VOR gain change for which the velocity-to-position integrator could not compensate and not a strategy to adapt to the visual vestibular mismatch conditions.

Possible neuronal sites for up-down asymmetrical VOR adaptation

The visual-vestibular mismatch conditions employed in the current study may confine the neuronal site(s) subserving VVOR adaptation. The fact that animals could modify their VOR to adapt to up-down asymmetrical visual-vestibular mismatch conditions suggests that the neuronal loci responsible for VVOR adaptation are those processing up- and downward VOR separately. In other words, the evidence confines the loci responsible for VVOR adaptation to those that can modify their sensitivities to up- and downward head movement asymmetrically.

Two neuronal loci have been demonstrated to be potentially responsible for VVOR adaptation; one is in and/or upstream from the flocculus and the other in the dorsal Y group of the vestibular nuclei (Hirata and Highstein 2001; Partsalis et al. 1995a,b). Y group neurons receive vestibular signals from inter neurons in the superior vestibular nucleus. There are two types of inter neuron: one receiving its input signal from the anterior canal and the other from the posterior canal (Blazquez et al. 2000). These neurons carry pure vestibular signals. The simplest idea to achieve the up-down asymmetrical VOR adaptation is that if pathways from each of the two types of inter neurons to Y group are modified in the opposite direction (1 is strengthened and the other 1 is weakened), the adaptation might occur. However, these inter neurons exhibit a symmetrical response to vertical sinusoidal head rotation. For example, the inter neuron receiving anterior canal input increases its firing rate during downward head rotation and decreases it during upward head rotation (Blazquez et al. 2000). Therefore modifying the synaptic efficacy between the inter neuron and Y group neuron can only produce a symmetrical change in activity in Y group neurons and cannot possibly produce an up-down asymmetrical response unless the stimulus causes the inter neurons and/or Y group neurons fire in their saturation ranges. Neither inter neurons (Blazquez et al. 2000) nor Y group neurons (Partsalis et al. 1995a,b) saturate in the stimulus range currently employed for up-down asymmetrical training. Therefore the theory that could explain Y group neuronal modulation and resultant eye movements before and after the ordinal symmetrical adaptation cannot explain the up-down asymmetrical VOR adaptation. However, two different approaches [chemical inactivation of the flocculus (Partsalis et al. 1995b) and system identification (Hirata and Highstein 2001)] have demonstrated that Y group is a neuronal site responsible for VVOR adaptation, thus there may be another neuronal mechanism that enables the up-down asymmetrical adaptation there.

The flocculus forms multi-layered neural networks consisting of mossy fiber input, granular cell layer and Purkinje cell layer. Other than this basic structure, there are feedback and feed-forward loops formed by inhibitory neurons (stellate cells, basket cells and Golgi cells). Theoretical studies (Albus 1971; Marr 1969; Schweighofer et al. 2001) demonstrated that cerebellar circuitry can learn various dynamic motor behaviors not just conventional gain control. There have been several types of synaptic plasticity found in the cerebellar circuitry (Hansel et al. 2001) including long-term depression at parallel fiber-Purkinje cell synapses (Ito 1989) and potentiation of the mossy fiber-granular cell synapses (D'Angelo et al. 1999). In naïve animals, floccular Purkinje cells show only a slight modulation during VORd, but following the conventional symmetrical visual-vestibular mismatch training, the amplitude of the modulation changes in the direction that would support the observed VOR gain change (Hirata and Highstein 2001; Watanabe 1984). In limited data in hand, Purkinje cell modulation actually changes asymmetrically after Up-Sup/Dn-Enh or Up-Enh/Dn-Sup training (Hirata et al. 2000). Y group neurons receive this asymmetrical inhibition and then could produce asymmetrical commands to move the eye by combining it with the symmetrical input from the inter neurons. The mechanism by which this asymmetrical signal is generated in the cerebellar circuitry is the subject of an ongoing investigation.

A behavioral study in human has demonstrated that it is possible to induce a unidirectional change in the VOR gain in the horizontal system (Aoki et al. 1998). Yakushin et al. (2000) have demonstrated that in the horizontal VOR system, lesions of the nucleus of the optic tract (NOT) significantly reduced or abolished the monkeys' ability to adapt the gain of the contraversive VOR but not of the ipsiversive VOR. This result suggests that there are independent gain-control mechanisms for ipsi- and contraversive VORs in the horizontal system. A possible underlying mechanism for this ipsi-contra asymmetrical effect following a NOT lesion might be that contraversive retinal slip information encoded in the NOT and conveyed through climbing fibers via inferior olive (IO) is not available in flocculus to induce LTD as the authors discussed. The same kind of mechanism might be considered for the vertical system because most of the units in the lateral terminal nucleus (LTN) that send their output to the IO exhibit asymmetries in preferred (upward) and nonpreferred (downward) directions (Mustari and Fuchs 1989). Yakushin et al. (2000) also demonstrated that only contraversive VOR eye velocities are affected after unilateral inactivation of the NOT in the animals whose VOR were adapted acutely to low gain. The evidence suggests that the NOT is related to memory storage of the contraversive VOR. The same type of asymmetrical effect could be expected for the vertical system after LTN inactivation in VVOR adapted animals, but very little is known about vestibular signals in the primate LTN to discuss this possibility.

Low-high-frequency asymmetrical VOR adaptation

The indirect pathway of the OKR is considered to convey lower frequency (~0.1 Hz) components of the OKR, while the direct pathway is thought to convey higher frequency (>0.1 Hz) components (Waespe and Henn 1987). Contrary to the fact that only the indirect pathway changed its characteristics after the HVOR adaptation (Demer 1981; Lisberger et al. 1981), the gain of VOR at higher frequencies such as 0.5 Hz can be modified by employing visual-vestibular interaction stimuli at these frequencies that most likely invoke only the OKR direct pathway among the visual pathways. No matter what the stimulus frequency, adaptation of the VOR gain is relatively frequency specific with less adaptation at other than the adapting frequency (Collewijn and Grootendorst 1979; Godaux et al. 1983; Lisberger et al. 1983; Powell et al. 1991; Raymond and Lisberger 1996). The evidence suggests that separate, multiple frequency channels might exist for VOR adaptation. Presently we examined inter dependency among these frequency channels by using new visual-vestibular interaction paradigms that require two frequency channels to adapt in opposite directions simultaneously.

Low-high-frequency asymmetrical visual-vestibular mismatch stimuli, the Lo-Sup/Hi-Enh and Lo-Enh/Hi-Sup, require animals to cancel their VOR in response to low-frequency sinusoidal head rotation and enhance their VOR in response to high-frequency head rotation or vice versa. Our monkeys had difficulties to make this rapid VOR modification in both the Lo-Sup/Hi-Enh and Lo-Enh/Hi-Sup conditions at both low and high frequencies. Animals had the same difficulty to make rapid VOR modifications in symmetrical VORe and VORs paradigms at high -frequency but could do better at low frequency in symmetrical training than in asymmetrical training. Therefore the performance of rapid VOR modification at low frequency is degraded when animals have to make rapid VOR modification at both low and high frequencies in opposite directions simultaneously while that at high frequency is not affected by the low-frequency component. In fact, performance at low frequency is degraded even when animals have to a make rapid VOR modification at low and high frequencies in the same direction simultaneously (Lo-Enh/Hi-Enh paradigm).

After 4 h exposure to the Lo-Sup/Hi-Enh paradigm, the animals' VOR gains to low- and high-frequency head rotation were eventually modified to more appropriate values for head movements at each frequency. After 4 h of the Lo-Enh/Hi-Sup training, animals could also change their low- and high-frequency VOR gains in the appropriate direction, but the magnitude of the changes were smaller than those after the Lo-Sup/Hi-Enh training. We demonstrated the repeatability and generality of these phenomena. As seen in the up-down asymmetrical adaptations, these adaptation mechanisms are not completely independent because experiments designed to train at one frequency only resulted in gain changes at the untrained frequency simultaneously as demonstrated in previous reports. Our result from restricted samples indicates that VORs training at the high frequency (Hi-Sup) and VORe training at the low frequency (Lo-Enh) induced greater changes in VOR gain at the untrained frequency than their counter part paradigms (Hi-Enh, Lo-Sup) did. This means that the dependency between low and high frequencies are directionally asymmetrical: high-frequency gain adaptation exerts more influence on the low frequency gain when it is trained toward low gain than when trained toward high gain, and low gain adaptation has more influence on high-frequency gain when it is trained toward high gain than when trained toward low gain. This may explain why the Lo-Enh/Hi-Sup pa