Velocity prediction in corrective saccades during smooth-pursuit eye movements in monkey

Summary It has been noted in a variety of studies in both humans and monkeys that saccades made during smooth pursuit eye movements are usually quite accurate. Since saccades are known to be planned on the basis of neuronal information existing at some interval of time before the actual onset of the movement, it is generally accepted that some sort of prediction or use of visual motion velocity is combined with static position error in the execution of these saccades to moving targets. However, statistical treatment of this response in humans has provided evidence for alternative mechanisms, including a strategy of saccading ahead in the direction of target motion without any incorporation of actual speed information about target motion in the response. We reinvestigated this question quantitatively in the monkey on a large data base of saccades. We found evidence that supports the hypothesis that information about target speed per se is used in this species in the production of saccades to moving targets. Multiple linear regression analysis supported the hypothesis that information about the position error and the target velocity that exists at about 100 ms prior to the saccade onset are both required to provide a statistical explanation of saccade size during pursuit eye movements under the conditions of our experiments.     

Motion processing for saccadic eye movements in humans

Summary 1. We studied the latencies and amplitudes of saccades to moving targets in normal human subjects. Targets underwent ramp or step-ramp motions. The goal was to determine how the saccadic system uses information about target velocity. 2. For simple ramp motion saccadic latency decreased as target speed increased. A threshold distance model, which assumes that the target has to move a minimum distance before saccadic processing starts, provided a good fit to the responses of all four subjects and explains discrepancies between previously published findings. 3. A double step experiment showed that target position may have some effect on saccadic amplitude when sampled 70 ms before saccade onset, but it must be sampled at least 140 ms before onset for an accurate saccade to occur. 4. Saccades to simple ramp targets approximated the target position 55 ms before saccade onset. Based on our double step results, this is more compensation than possible by a simple position estimate and implies extrapolation of target motion by the saccadic system. The lack of complete compensation may be due to an underestimate of the target speed and/or of the saccadic latency. 5. A delayed-saccade paradigm resulted in saccades with a longer, constant latency and allowed longer viewing of target motion. These saccades accounted for all but 20 ms of target motion, suggesting that with more processing time of target motion a better extrapolation may be generated. 6. In a step-ramp paradigm the target stepped in one direction, then moved smoothly in the opposite direction. Saccades in this paradigm could be made in either the direction of the step or in the direction of target motion: the direction and latency were determined solely by the time at which the target crossed the fixation point. This time must be calculated from target speed and position, implying that the saccadic system must use speed information to adjust latency or to cancel unnecessary saccades.     

Initiation of smooth-pursuit eye movements to first-order and second-order motion stimuli

Since normal human subjects can perform smooth-pursuit eye movements only in the presence of a moving target, the occurrence of these eye movements represents an ideal behavioural probe to monitor the successful processing of visual motion. It has been shown previously that subjects can execute smooth-pursuit eye movements to targets defined by luminance and colour, the first-order stimulus attributes, as well as to targets defined by derived, second-order stimulus attributes such as contrast, flicker or motion. In contrast to these earlier experiments focusing on steady-state pursuit, the present study addressed the course of pre-saccadic pursuit initiation (less than 100 ms), as this early time period is thought to represent open-loop pursuit, i.e. the eye movements are exclusively driven by visual inputs proceeding the onset of the eye movement itself. Eye movements of five human subjects tracking first- and second-order motion stimuli had been measured. The analysis of the obtained eye traces revealed that smooth-pursuit eye movements could be initiated to first-order as well as second-order motion stimuli, even before the execution of the first initial saccade. In contrast to steady-state pursuit, the initiation of pursuit was not exclusively determined by the movement of the target, but rather due to an interaction between dominant first-order and less-weighted second-order motion components. Based on our results, two conclusions may be drawn: first and specific for initiation of smooth-pursuit eye movements, we present evidence supporting the notion that initiation of pursuit reflects integration of all available visual motion information. Second and more general, our results further support the hypothesis that the visual system consists of more than one mechanism for the extraction of first-order and second-order motion.     

Predictive behavior of optokinetic eye movements

Summary Optokinetic nystagmus is often thought of as a primitive oculomotor response, while smooth pursuit is thought of as a higher one. We have used conditions that are usually thought of as eliciting optokinetic responses; i.e., large-field stimuli confined to the retinal periphery, and instructions to subjects to respond passively. In spite of this, the responses showed predictive behavior similar to that described for smooth pursuit.     

Extent of compensation for variations in monkey saccadic eye movements

We investigated and quantified the ability of the primate saccadic system to generate accurate eye movements in spite of naturally occurring variations in saccadic speed and trajectory. We show that the amplitude of a series of saccades directed to the same target is positively correlated to their peak speed, i.e., the faster the saccade, the bigger its amplitude. We demonstrate that this result cannot be simply accounted for by the main sequence, and that on average the saccadic system is able to compensate for only 61% of the variability in speed. Deviations from the average trajectory are also only partially compensated: the underlying mechanism, which tends to bring the eyes back toward the desired trajectory, underperforms for small movements and overperforms for large movements. We also demonstrate that the performance of this compensatory mechanism, and the metrics of saccades in general, do not depend on the presence of visual information during the movement. By showing that deviations from the desired behavior are corrected during the saccade, our results further support the hypothesis that the innervation signal that generates saccadic eye movements is not pre-programmed but rather is dynamically adjusted during the movement. However, the compensation for deviations from the desired behavior is only partial, and the underlying mechanisms have yet to be completely understood. Although none of the current models of the saccadic system can account for our results, some of them, if appropriately modified, probably could.     

Influences of hand movements on eye movements in tracking tasks in man

Summary We investigated horizontal smooth pursuit eye movements and hand movements in tracking tasks in order to find out whether hand movements influence eye movements and if so, in what ways. Externally controlled target movements were tracked either by the eyes alone or by the eyes and right hand together. Because a possible influence might depend on the stimulus, we used two classes of target movements: sinusoidal target movements (predictable target movements) and pseudo-random target movements (unpredictable target movements). Our data show that the eye movements contained only a few small saccades when sinusoidal target movements with frequencies higher than about 1 Hz were tracked by eyes and hand together. More and larger saccades were made when the same target movements were tracked by the eyes alone. The difference in smoothness of eye movements was highly significant between the two tracking conditions. Such a difference was not found during the tracking of a pseudorandom target motion. This suggests that the influence of hand movements is related to the predictability of the stimulus. In contrast to the gain of the smooth pursuit eye movements and the maximum of the cross-correlation function, the gain of the composite eye movements did not depend on the tracking condition. The delay of the eye movements with respect to the (sinusoidal) target movements also showed no dependence on the tracking condition. Visual feedback from the tracking hand was found not to play a role in the difference in eye movements for the two tracking conditions.     

Peak velocities of visually and nonvisually guided saccades in smooth-pursuit and saccadic tasks

 Smooth pursuit typically includes corrective catch-up saccades, but may also include such intrusive saccades away from the target as anticipatory or large overshooting saccades. We sought to differentiate catch-up from anticipatory and overshooting saccades by their peak velocities, to see whether the higher velocities of visually rather than nonvisually guided saccades in saccadic tasks may be found also in saccades in pursuit. In experiment 1, 12 subjects showed catch-up, anticipatory, and overshooting saccades to comprise 70.4% of all saccades in pursuit of periodic, 30°/s constant-velocity targets. Catch-up saccades were faster than the others. Saccadic tasks were run as well, on 19 subjects, including the 12 whose pursuit data were analyzed, with target-onset, target-remaining (saccade to the remaining target when the other three extinguish), and antisaccade tasks. For 17 of the 19 subjects, antisaccade velocities were lower than for either target-onset or target-remaining tasks. Velocities for the target-remaining task were near those for target onset, indicating that target presence, not its onset, defines visually guided saccades. Error and reaction-time data suggest greater cognitive difficulty for target remaining than for target onset, so that the cognitive difficulty of typical nonvisually guided saccade tasks is not sufficient to produce their lowered velocity. To produce reliably, in each subject, catch-up and anticipatory saccades with comparable amplitude distributions, nine new subjects were asked in experiment 2 to make intentional catch-up and anticipatory saccades in pursuit, and were presented with embedded target jumps to elicit catch-up saccades, all with periodic target trajectories of 15°/s and 30°/s. Velocities of intentional anticipatory saccades were lower than velocities of intentional catch-up saccades, while velocities of intentional and embedded catch-up saccades were similar. Target-onset and remembered-target saccadic tasks were run, showing the expected higher velocity for the target-onset task in each subject. Both experiments demonstrate higher peak velocities for catch-up saccades than for anticipatory saccades, suggesting that cortical structures preferentially involved in nonvisually guided saccades may initiate the anticipatory and overshooting saccades in pursuit. Received: 1 December 1995 / Accepted: 25 February 1997     

Transfer of short-term adaptation in human saccadic eye movements

Controversy exists as to the extent to which the saccadic system, adapted in the so-called gain-shortening paradigm for a particular target configuration, transfers the resulting change in saccade metrics to saccades elicited under different circumstances. In order to further assess this problem, we investigated the properties of human saccadic eye movements after visually induced short-term adaptation under a variety of conditions. We observed that saccades both during and after the adaptation did not significantly change their main sequence properties with respect to the pre-adaptation baseline. Saccade velocity profiles remained normal throughout the experiment, and we obtained no evidence that correction saccades were gradually absorbed in the primary saccade. We found that the effect of the short-term adaptation on saccade metrics is not confined to the particular combination of initial eye position and spatial position of the visual target used to induce the adaptation response. Saccades elicited from different initial positions towards targets with the same retinotopic coordinates as in the adaptation phase yield the same level of adaptation. However, our findings indicate that adaptation is confined to a limited range of saccade vectors around the oculocentric coordinates of the adaptation target (restricted adaptation field). Smaller and larger saccades are endowed with significantly lower adaptation values. Moreover, two further experiments showed that a retinal stimulus is not a prerequisite for adaptation to express itself: First, in a double-step experiment, we dissociated the retinal stimulus vector from the required oculomotor response. Second, we also investigated the effect of visually induced adaptation on auditory evoked saccades. In both tasks the adaptation was transferred to the required motor response. Based on our findings, we conclude that short-term adaptation is expressed at a multisensory stage, where saccadic eye movements are represented as desired eye displacement vectors (motor error). Possible neurophysiological implications are discussed.     

The influence of stationary and moving textured backgrounds on smooth-pursuit initiation and steady state pursuit in humans

 We investigated the effects of stationary and moving textured backgrounds on the initiation and steady state of ocular pursuit using horizontally moving targets. We found that the initial eye acceleration was slightly reduced when a stationary textured background was employed, as compared to experiments with a homogeneous background. When a moving textured background was introduced, the initial eye acceleration was significantly larger when the target and the background moved in opposite directions than when the target and the background moved in the same direction. The use of stationary and moving textured backgrounds resulted in comparable effects on the initial eye acceleration when they were presented either as a large field or as a narrow, horizontal small field, only covering the trajectory of the target. Moreover, small-field stationary backgrounds slightly reduced the eye velocity during steady state pursuit. A small-field background moving in the opposite direction to the target distinctly reduced eye velocity, while a target and a background moving in the same direction sometimes even improved pursuit performance, when compared with a homogeneous background. The influences of small-field textured backgrounds on steady state pursuit were comparable with those of large-field backgrounds in both stationary and moving conditions. Received: 14 December 1996 / Accepted: 30 December 1996     

Fast, anticipatory smooth-pursuit eye movements appear to depend on a short-term store

Anticipatory smooth pursuit before the expected appearance of a moving target can reduce the initial retinal blur caused by the 100-ms delay of visual feedback. Humans, though, can only voluntarily generate smooth velocities up to about 5°/s without a moving target. However, previous experiments have shown that repetitive brief presentations of a moving target every few seconds appear to charge an internal store, the contents of which can later be released to generate higher velocity anticipatory movements. This store’s longevity was assessed here by repetitively presenting a moving target for 500 ms at different known intervals up to 7.2 s. Target motion at 25°/s or 50°/s was tested, with presentations in alternate directions or the same direction. Anticipatory velocity, measured 100 ms after target onset, decreased with increasing interval for all target motion conditions. A decrease was still seen when accurate timing cues were given before each presentation, suggesting that the drive for anticipatory pursuit is held in a short-term store lasting a few seconds which can enhance the low velocities produced by volition alone. The results also demonstrate that high-velocity anticipatory pursuit helps to overcome the temporal delays in the system and allows target velocity to be matched at an earlier time. Received: 27 August 1997 / Accepted: 22 December 1997     

Anticipatory smooth eye movements and predictive pursuit after unilateral lesions in human brain

Anticipatory smooth eye movements precede expected changes in target motion. It has been questioned whether anticipatory smooth eye movements are a component of the smooth pursuit system. Five subjects with unilateral brain lesions and five control subjects were tested with predictable double-ramp stimuli to determine whether these lesions have a similar effect on horizontal, visually guided smooth pursuit, anticipatory smooth eye movements, and the predictive component of smooth pursuit. All four subjects with a brain lesion involving the parietal or parietal-frontal lobe had parallel velocity asymmetries in all three forms of smooth eye movements, with lowest velocities toward the side of the lesion. A similar uniformity and magnitude of smooth eye movement directional asymmetries were not found in control subjects. Unidirectional attenuation of these three forms of smooth eye movements provides evidence that they are part of a unified smooth eye movement system.     

Spatial attention and eye movements

We previously showed that when attention is allocated to the right or left of the fixation point, saccades directed to targets located above or below the fixation point deviate contralateral to the attention locus. In the present study, we examined how general this phenomenon is and whether the amount of saccade deviation depends on the location of attention with respect to that of the saccade target. Three experiments were carried out. In experiment 1 the location of the imperative stimulus was uncued. Its presentation exogenously directed attention to its location. In experiment 2 the location of the imperative stimulus was cued by a central cognitive cue. In this experiment attention was endogenously directed to the imperative stimulus location before its presentation (expectancy paradigm). In experiment 3 all stimulus boxes contained a possible imperative stimulus at the display presentation. A central cue, presented subsequently, indicated which of them had to be used for the saccade. In this experiment attention was endogenously directed to the imperative stimulus, but after its presentation (no-expectancy paradigm). The results showed that, regardless of how attention was directed to the imperative stimulus, the vertical saccades deviated contralateral to the attention location. The deviation was larger when attention was in the upper field and the saccade was directed upward (same hemifield condition) than when attention was in the upper field and the saccade was directed downward (opposite hemifield condition). The same relationship between the same hemifield condition and opposite hemifield condition was found when attention was in the lower field. Saccadic reaction times (SRTs) were shortest in experiment 2 and longest in experiment 3. In experiment 2, SRTs of the same hemifield condition were significantly longer than those of the opposite hemifield condition. Taken altogether, these results strongly support the notion that attention allocation in space leads to an activation of oculomotor circuits, in spite of eye immobility. The possible mechanisms responsible for saccade deviations and for greater saccade deviations when attention is in the same hemifield as the programmed ocular saccade are discussed.     

Saccades to stationary and moving targets differ in the monkey

Saccade characteristics in response to moving and stationary targets were studied in three monkeys (Macaca mulatta) that had been trained to look at a target, which after an initial jump either remained in place or moved forward or backward with constant velocity (10°/s). Eye movements were recorded using a search coil. The contribution of smooth pursuit to the saccade amplitude was small (<0.25°). Saccades having the same amplitude (5.67–6.83° for different monkeys) to forward and backward moving targets were compared. Peak velocity was higher (37–42°/s on average for different monkeys) and saccade duration was shorter (8–10 ms on average) for backward saccades than for forward saccades These differences were highly significant (t-test: P<0.001). Thus, forward and backward saccades are not on the same main sequence. This suggests that saccade dynamics are affected not only by the retinal position error but also by target motion. Further analysis revealed that saccade peak velocity mainly depends on the retinal position error, but saccade amplitude also depends on a stimulus-related velocity factor, which affects the saccade mainly during deceleration. This velocity factor could be retinal slip or target velocity, which was the same under our conditions. Our results experimentally support recent models that propose that the saccade acceleration in response to moving targets might be controlled by the superior colliculus, whereas the deceleration changes are fine-tuned by the cerebellum. This prediction must still be tested on a neuronal level.Both Yanfang Guan and Thomas Eggert contributed equally     

Orienting of attention and eye movements

According to the premotor theory of attention, the mechanisms responsible for spatial attention and the mechanisms involved in programming ocular saccades are basically the same. The aim of the present experiments was to test this claim. In experiment 1 subjects were presented with a visual display consisting of a fixation point and four boxes arranged horizontally and located above the fixation cross. Two of the boxes were in the left visual hemifield, two in the right. A fifth box was located on the vertical meridian below the fixation cross. Digit cues indicated in which of the upper boxes the imperative stimulus was most likely to appear. Subjects were instructed to direct attention to the cued box and to perform a saccadic eye movement to the lower box on presentation of the imperative stimulus. The trajectory of the saccades deviated contralateral to the hemifield in which the imperative stimulus was presented. This deviation was larger when the hemifield where the imperative stimulus was presented was the cued one. In experiment 2, the visual display consisted of five boxes forming a cross. The central box served as a fixation point. The cue was a small line, linked to the central box, pointing to different directions and indicating where the visual imperative stimulus would appear. In 50% of trials, the imperative stimulus was a visual stimulus presented either in one of the lateral boxes or in the central one. In the remaining 50% of trials, the imperative stimulus was a non-lateralised sound. Half the subjects were instructed to make a saccade to the upper box at the presentation of the visual imperative stimulus and to the lower box at the presentation of the acoustic stimulus. Half the subjects received the opposite instructions. The results confirmed that the saccades deviate contralateral to the hemifield of stimulus presentation in the case of visual imperative stimuli. Most importantly, they showed that the saccades deviate contralateral to the cued hemifield, also in the case of acoustic imperative stimuli. Experiment 3 was similar to experiment 2. It confirmed the results of that experiment and showed that slow ocular drifts, which are observed in the time interval between cue and imperative stimulus presentation, cannot explain the ocular deviations. Taken together, the experiments demonstrate that spatial attention allocation leads to an activation of oculomotor circuits, in spite of eye immobility.     

Discharge properties of neurons in the rostral superior colliculus of the monkey during smooth-pursuit eye movements

The intermediate and deep layers of the monkey superior colliculus (SC) comprise a retinotopically organized map for eye movements. The rostral end of this map, corresponding to the representation of the fovea, contains neurons that have been referred to as "fixation cells" because they discharge tonically during active fixation and pause during the generation of most saccades. These neurons also possess movement fields and are most active for targets close to the fixation point. Because the parafoveal locations encoded by these neurons are also important for guiding pursuit eye movements, we studied these neurons in two monkeys as they generated smooth pursuit. We found that fixation cells exhibit the same directional preferences during pursuit as during small saccades-they increase their discharge during movements toward the contralateral side and decrease their discharge during movements toward the ipsilateral side. This pursuit-related activity could be observed during saccade-free pursuit and was not predictive of small saccades that often accompanied pursuit. When we plotted the discharge rate from individual neurons during pursuit as a function of the position error associated with the moving target, we found tuning curves with peaks within a few degrees contralateral of the fovea. We compared these pursuit-related tuning curves from each neuron to the tuning curves for a saccade task from which we separately measured the visual, delay, and peri-saccadic activity. We found the highest and most consistent correlation with the delay activity recorded while the monkey viewed parafoveal stimuli during fixation. The directional preferences exhibited during pursuit can therefore be attributed to the tuning of these neurons for contralateral locations near the fovea. These results support the idea that fixation cells are the rostral extension of the buildup neurons found in the more caudal colliculus and that their activity conveys information about the size of the mismatch between a parafoveal stimulus and the currently foveated location. Because the generation of pursuit requires a break from fixation, the pursuit-related activity indicates that these neurons are not strictly involved with maintaining fixation. Conversely, because activity during the delay period was found for many neurons even when no eye movement was made, these neurons are also not obligatorily related to the generation of a movement. Thus the tonic activity of these rostral neurons provides a potential position-error signal rather than a motor command-a principle that may be applicable to buildup neurons elsewhere in the SC.     

Extraretinal inputs to neurons in the rostral superior colliculus of the monkey during smooth-pursuit eye movements.

The intermediate and deep layers of the monkey superior colliculus (SC) are known to be important for the generation of saccadic eye movements. Recent studies have also provided evidence that the rostral SC might be involved in the control of pursuit eye movements. However, because rostral SC neurons respond to visual stimuli used to guide pursuit, it is also possible that the pursuit-related activity is simply a visual response. To test this possibility, we recorded the activity of neurons in the rostral SC as monkeys smoothly pursued a target that was briefly extinguished. We found that almost all rostral SC neurons in our sample maintained their pursuit-related activity during a brief visual blink, which was similar to the maintained activity they also exhibited during blinks imposed during fixation. These results indicate that discharge of rostral SC neurons during pursuit is not simply a visual response, but includes extraretinal signals.     

Visual signals in the dorsolateral pontine nucleus of the alert monkey: Their relationship to smooth-pursuit eye movements

Summary The visual properties of 77 dorsolateral pontine nucleus (DLPN) cells were studied in two alert monkeys. In 41 cells, presentation of a moving random dot background pattern, while the monkeys fixated a stationary spot, elicited modulations in discharge rate that were related either to (i) the velocity of background motion in a specific direction or to (ii) only the direction of background movement. Thirty-six DLPN cells exhibited responses to small, 0.6–1.7 deg, visual stimuli. Nine such cells exhibited non-direction selective receptive fields that were eccentric from the fovea. During fixation of a stationary bluish spot, the visual responses of 27 DLPN cells to movement of a small, white test spot were characterized by two components: (1) as the test spot crossed the fovea in a specific direction, transient velocity-related increases in discharge rate occurred and (2) a maintained, smaller increase in activity was observed for the duration of test spot movement in the preferred direction. This DLPN activity associated with small visual stimuli was also observed during smooth-pursuit eye movements when, due to imperfect tracking, retinal image motion of the target produced slip in the same direction. These preliminary results suggest that the DLPN could supply the smooth-pursuit system with signals concerning the direction and velocity of target image motion on the retina.This study was supported by NSF Grant BNS-8107111, NIH Grant R01 EY04552-01, and the Smith-Kettlewell Eye Research Foundation Dedication. This paper is dedicated to Dr. Kitsuya Iwama, Emeritus Professor of Osaka University Medical School, on his retirement. The first author is grateful for the inspiration and guidance that Dr. Iwama provided during the early part of the author's education in neurophysiology.     

Neuronal activity in the dorsolateral pontine nucleus of the alert monkey modulated by visual stimuli and eye movements

Summary The activity of neurons in the dorsolateral pontine nucleus (dlpn) was studied in two awake rhesus monkeys trained to participate in a variety of visual and oculomotor tests. The visual and eye movement related responses of 73 neurons encountered in the more caudal part of the dlpn were analyzed. Thirty eight of these could be assigned to one of the three following groups. Visual-only neurons (Type 1, n = 10) responded to movement of a broad range of visual stimuli in certain preferred directions. Their receptive fields were usually large, not restricted to the contralateral visual field and always included the fovea. Visual-tracking (VT) neurons (n = 28) discharged in relation to smooth pursuit of a small target in particular preferred directions. Nine of these (Type 2) did not respond to visual stimulation during stationary fixation. Nineteen VT-cells (Type 3) discharged in relation to both visual tracking and visual stimulation. In 9 of the Type 3 neurons, the preferred directions for visual stimulation and tracking were opposite, whereas they were the same in the other 10. Visual responses of Type 3 neurons were indistinguishable from those of Type 1 neurons. Testing of an additional 9 neurons driven by either visual-tracking or pattern movement was not sufficient to allow a definite assignment to one of the groups 1, 2 or 3. The distribution of preferred directions for both visual stimulation and visual tracking was widely scattered between 0 and 360 deg. Our results suggest that the dlpn is a constituent in a cerebro-cerebellar loop important for the generation of smooth pursuit eye movements.     

Integration of visual and somatosensory target information in goal-directed eye and arm movements

 In this study, we compared separate and coordinated eye and hand movements towards visual or somatosensory target stimuli in a dark room, where no visual position information about the hand could be obtained. Experiment 1 showed that saccadic reaction times (RTs) were longer when directed to somatosensory targets than when directed to visual targets in both single- and dual-task conditions. However, for hand movements, this pattern was only found in the dual-task condition and not in the single-task condition. Experiment 1 also showed that correlations between saccadic and hand RTs were significantly higher when directed towards somatosensory targets than when directed towards visual targets. Importantly, experiment 2 indicated that this was not caused by differences in processing times at a perceptual level. Furthermore, hand-pointing accuracy was found to be higher when subjects had to move their eyes as well (dual task) compared to a single-task hand movement. However, this effect was more pronounced for movements to visual targets than to somatosensory targets. A schematic model of sensorimotor transformations for saccadic eye and goal-directed hand movements is proposed and possible shared mechanisms of the two motor systems are discussed. Received: 15 June 1998 / Accepted: 21 September 1998