Conceived and designed the experiments: OAM SF KPH. Performed the experiments: OAM SF. Analyzed the data: OAM SF. Contributed reagents/materials/analysis tools: KPH. Wrote the paper: OAM KPH.
The authors have declared that no competing interests exist.
Global motion detection is one of the most important abilities in the animal kingdom to navigate through a 3-dimensional environment. In the visual system of teleost fish direction-selective neurons in the pretectal area (APT) are most important for global motion detection. As in all other vertebrates these neurons are involved in the control of slow phase eye movements during gaze stabilization. In contrast to mammals cortical pathways that might influence motion detection abilities of the optokinetic system are missing in teleost fish.
To test global motion detection in goldfish we first measured the coherence threshold of random dot patterns to elicit horizontal slow phase eye movements. In addition, the coherence threshold of the optomotor response was determined by the same random dot patterns. In a second approach the coherence threshold to elicit a direction selective response in neurons of the APT was assessed from a neurometric function. Behavioural thresholds and neuronal thresholds to elicit slow phase eye movements were very similar, and ranged between 10% and 20% coherence. In contrast to these low thresholds for the optokinetic reaction and APT neurons the optomotor response could only be elicited by random dot patterns with coherences above 40%.
Our findings suggest a high sensitivity for global motion in the goldfish optokinetic system. Comparison of neuronal and behavioural thresholds implies a nearly one-to-one transformation of visual neuron performance to the visuo-motor output. In addition, we assume that the optomotor response is not mediated by the optokinetic system, but instead by other motion detection systems with higher coherence thresholds.
The ability of the visual system to detect global motion is essential for almost all animals
To perceive global motion local motion signals have to be integrated over space and time. In mammals local motion detectors like orientation and direction selective neurons in V1 are only capable to encode motion signals in spatially distinct areas due to their limited receptive field sizes
The neuronal substrate for the optokinetic response is however well investigated in a variety of vertebrate species. In mammals direction-selective neurons in the pretectal NOT and the accessory optic system (AOS), composed of the dorsal terminal nucleus (DTN), the lateral terminal nucleus (LTN) and the medial terminal nucleus (MTN) in mammals, are required for this behaviour
In tetrapods other than mammals gaze stabilization is mediated by direction-selective neurons in the pretectal nucleus lentiformis mesencephali (nLM) and the nucleus of the basal optic root (nBOR), though only the nBOR is considered as part of the AOS
In teleost fish slow phase eye movements for gaze stabilization are mediated by direction selective neurons in the pretectal area (APT)
So far the sensitivities of the optokinetic response and the optomotor response as well as their neuronal substrates for global motion detection are not described. Therefore in this study we applied the well established method of varying the coherence level of moving random dot patterns to determine and compare the thresholds for OKR and OMR as well as the neurometric function of neurons in the APT in goldfish. The data are discussed to answer the question whether OKR and OMR are served by the same or different neuronal populations.
In a first step we measured the OKR in a behavioural paradigm during stimulation with random dot stimuli of different coherence levels to ascertain the threshold of the optokinetic system for global motion detection. In a second step visual direction-selective neurons in the APT, mediating the OKR in teleost fishes, were examined with the same motion stimuli to understand the transformation of sensory inputs to corresponding motor outputs.
At last we measured the optomotor response to stimuli with different coherence levels and determined its threshold. If different thresholds for the OKR and OMR exist, this would provide evidence for different underlying circuitries in mediating the OKR and OMR.
Our study shows high global motion detection abilities of the goldfish optokinetic system in comparison to other species. And a significant higher threshold for eliciting the OMR proposes that the APT of teleost fish is probably not involved in the execution of the OMR.
Data from 19 goldfish were included in the present study. Animal size varied between 5 cm–15 cm in length and included animals of both sexes. All experiments were approved by the local authorities (Regierungspräsidium Arnsberg) and carried out in accordance with the Deutsche Tierschutzgesetz of 12 April 2001, the European Communities Council Directive of 24 November 1986 (S6 609 EEC) and NIH guidelines for care and use of animals for experimental procedures.
For visual horizontal wholefield stimulation different videos projected by a beamer, ranging from 0% coherence up to 100% coherence in 10% steps were used. All videos were custom made in MATLAB (7.01). Here 100% coherence means that all dots moved into one direction, whereas e.g. in a 70% coherence video only 70% of the dots moved in one direction and the remaining 30% moved randomly (please see supplementary video files:
For horizontal eye movement recordings animals were fixed within a plastic fish holder and placed in the middle of pairs of horizontal and vertical coils (
Optokinetic eye movements were recorded for 30 s in each trial. After each experiment, the search coil used was detached, exactly repositioned in the magnetic field and calibrated with a protractor. The recorded calibration and eye position signal from the search coils were analyzed off-line with a custom made MATLAB program.
Ten out of nineteen fish were measured ten times and 9 fish were measured three times in each condition (0%–100% coherence) and direction (CW and CCW). For each condition the slope of ten slow phase eye movements was calculated to evaluate the gain of the slow phase eye movement.
For each fish also an individual threshold was determined by judging from which coherence on slow phase eye movements or resetting saccades were visible in the eye traces.
Before surgery animals were first anesthetized by immersion in a bath containing 0.1% 3-aminobenzoic acid ethyl esther (MS222). Anaesthesia was further supplemented locally with 2.5% lidocaine, before a craniotomy was performed to allow access to the left tectum opticum and pretectum. Immediately following surgery the animals were immobilized with Flaxedil (0.5–1 mg, i.m.) and transferred to a transparent recording hemisphere (diameter 70 cm), where they were artificially ventilated with cooled water (19°C). Single units were recorded with glass-coated tungsten microelectrodes (impedance 1–2.5 MΩ) in the left pretectum. For localizing direction-selective neurons in the APT the visual stimulus consisted of random light dots projected into the hemisphere by a planetarium projector centred above the fish's head (for further information of the experimental setup please see
In contrast to the OKR measurements animals were allowed to swim freely within a ring shaped octagon tank with 95 cm diameter and a water depth of about 15 cm (
Again different coherence videos were projected by a beamer; the centre of rotation was positioned to the centre of the ring shaped octagon channel. All stimulus parameters were the same as for the optokinetic measurements, except that here both eyes see the same stimulus direction (back to front or front to back). With e. g. back to front stimulus movement the fish perceived a motion like during drifting backwards. To compensate this the OMR should force the fish to swim forward. Animals were tested individually by inserting one by one into the experimental tank. Animals were allowed to accustom to the tank for 30 min in the dark. The presentation of stimuli with different coherence levels was randomized. Responses of the animals were videotaped for 2 min per stimulus direction and coherence level and analyzed off-line; the whole procedure was done four times with each fish.
To quantify the OMR the experimental tank was divided into four sectors. For each condition the number of sectors which the fish passed through in the direction of the stimulus (OMR), against the direction of the stimulus and the number of stationary phases were counted. To calculate the individual coherence threshold of each animal, the lowest coherence at which the number of responses in stimulus direction was significantly higher than the number of responses against the stimulus direction was determined. To assure the behavioral threshold of the OMR we analysed our data also by the use of a receiver operating characteristic (ROC).
To evaluate the OKR the median of all gains for each coherence step, direction and each fish was calculated. Median gains were plotted against the coherence level to visualize the behaviour of slow phase eye movements. We then compared with a t-test all obtained median gain values of one coherence level with the gain values of the subsequent lower coherence. A significant difference between both coherences indicates a decrease in OKR performance. This analysis shows the systematic dependence of the gain of optokinetic eye movements on the coherence level in random dot stimuli. We never observed smooth pursuit eye movements against the stimulus direction, so there was no possibility to apply a ROC analysis. Instead to determine the threshold at the population level we compared the number of trials in which we could observe a clear OKR independent of gain and number of slow phases. A sigmoid function was fitted to the data and threshold was set arbitrarily at 50% effective trials which is a conservative estimate.
Neuronal and OMR coherence thresholds were assessed with a neurometric function as described by Britten et al.
Afterwards the normalized area under the ROC of each coherence level was estimated and plotted against the coherence threshold (
The proportion of correct choices by the model is plotted against increasing coherence levels. The correlation level is the normalized area under the corresponding receiver operator curve (ROC). The red line corresponds to the fitted sigmoidal function. Threshold was estimated at the coherence level at which the model predicted 75% correct (dash dotted line). R2 corresponds to the coefficient of determination.
As threshold the coherence level at a proportion of 50% above chance (0.75 correct) was used. For a detailed description of threshold calculation please see Britten et al.
Typical examples of slow phase eye movements during stimulation with different coherence levels are shown in
Slow phase eye movements were not evident in all animals at a coherence level of 10%, different individuals had varying thresholds for eliciting an OKR. In thirteen animals out of nineteen slow phase eye movements were already recognized during stimulation with 10% coherence, the remaining six animals had their threshold at 20% for eliciting an OKR at all. Slow phases could however not be elicited in every 30 s test trial especially at low coherence levels. We therefore determined the percentage of successful trials at each coherence level for each fish. When plotted against the coherence level and fitted with a sigmoid function a threshold set at 50% was determined. This threshold was taken because we used the same level for the OMR and the neuronal data. As
A sigmoid function was fitted to the data and threshold was set arbitrarily at 50% effective trials. R2 corresponds to the coefficient of determination.
Taken together our observation of individual animals and the population analysis show that even stimuli containing less than 20% coherently moving dots can trigger an OKR, although the likelihood to trigger an OKR decreases with decreasing coherences.
All nineteen animals showed a robust OKR in CW (median = 0.6) and CCW (median = 0.62) direction during the presentation of a 100% coherence stimulus at a velocity of 13°/s (
Data were only taken from those recordings in which clear slow phases were visible. The median gains were fitted with an exponential function (red line). R2 corresponds to the coefficient of determination.
All in all thirty-seven direction-selective neurons with typical large receptive fields were recorded and tested with all coherence levels. Twenty-two of them had a stronger response to temporo-nasal and the remaining fifteen to naso-temporal stimulus direction as seen by the eye contralateral to the recording site. Since we only used horizontally moving stimuli we could not determine the exact preferred direction which could have been in any direction
0–2000 ms and 5000–7000 ms presentation of a stationary random dot stimulus, 2000–5000 ms stimulation in naso-temporal direction, 7000–10.000 ms stimulation in temporo-nasal direction. The beginning of the moving stimulus is marked by a vertical red line, whereas the green line marks the beginning of the stationary presentation of the random dot pattern. Black line corresponds to a Gaussian fitting of the spike train.
For each neuron a specific coherence threshold was assigned by a neurometric function which takes both the firing rate in the preferred and in the null direction into consideration (see
A threshold for the optomotor response was determined in ten individuals. The stimulus with the lowest coherence at which animals showed significantly (t-test) more responses in than against the stimulus direction was ascertained as individual threshold.
Green bar: Swimming within the stimulus direction; red bar: Swimming against the stimulus direction; gray bar: Stationary phases.
Nine out of ten animals had a coherence threshold between 40 and 50% and only in one animal a stimulus with 40% coherence was able to elicit an optomotor response.
To approve that the behavioral threshold of the OMR is not influenced by noise or by our sample size, we used in addition the same data analysis as for the neuronal data. The ROC analysis results in a behavioral threshold of 43% and confirms the actual thresholds assessed in individual fish (
The proportion of correct choices by the model is plotted against increasing coherence levels. The correlation level is the normalized area under the corresponding receiver operator curve (ROC). The red line corresponds to the fitted sigmoidal function. Threshold was estimated at the coherence level at which the model predicted 75% correct (dash dotted line). R2 corresponds to the coefficient of determination.
Our objectives were to examine the coherence thresholds of the optokinetic response, the optomotor response and to determine neuronal thresholds of visual direction selective neurons in the pretectal area of goldfish. We find astonishing low thresholds for the optokinetic reaction and underlying neuronal circuits. In contrast to the optokinetic reaction (10% to 20% in individual cases; 16 to 27% on the population level) and visual direction-selective neurons (<20%) is the coherence threshold for the optomotor reaction is about 2 to 4 times higher (43%).
Results which were obtained with our 100% coherence random dot stimuli are by all means comparable to former studies. Already Dieringer
On a population level we observed only for some of the tested coherences levels a significant difference between gains in CW and CCW direction. Also other studies revealed slight asymmetries in the OKR of goldfish during binocular viewing conditions
As expected gains decreased with decreasing coherence of the presented stimulus, but the thresholds reached in our study are amazingly low compared to studies with other species
Under normal conditions optokinetic stimulation always leads to slow phase eye movements following the direction of the stimulus. The animal can “decide” to follow or not to follow, but it cannot produce pursuit eye movements against the stimulus direction or when the stimulus is stationary. We never observed slow phases directed against the moving dots even at only 10% coherence. If slow phases occurred at all they would follow the direction of the coherently moving dots. If we assume that the presence of an OKR to low coherence stimuli is compatible with the threshold for the perception of the enclosed global motion our data can be compared with other studies. Of the species studied so far only humans and monkeys have better capabilities to detect global motion with a coherence threshold of 5%
At least a coherence of 10% was necessary to elicit clear direction-selective responses in neurons of the APT. Up to date no other studies have dealt with global motion capabilities of the optokinetic system and our studies showed for the first time, which signal to noise ratio is needed by neurons of the APT to detect global motion. One study by Britten et al.
Thresholds for the optomotor response were about 2 to 4 times higher than for the OKR. With a 100% coherence random dot stimulus OMR could reliably be triggered. Thus our design of an OMR stimulus seems adequate and therefore it is still highly astonishing that the actual threshold for the optomotor reaction lies at coherence levels of more than 40%.
Possibly the readiness of the fish to move the whole body during OMR is much lower than to move the eyes. In addition, real drifting in water will always generate a strong signal via the lateral line sensors which may be critical to trigger compensatory body movements. Another explanation for different thresholds of the OMR and OKR might be an imperfect read out of neuronal responses by the OMR system. The APT neurons respond well to optic flow generated by rotations. But APT neurons cannot differentiate between rotation and translation as their visual input is only mediated by the contralateral retina, i.e. occurring retinal slip during horizontal rotation or forward translation are more or less the same for monocular receptive fields. We do not know whether information from these neurons can be compiled to derive information about translational optic flow to trigger the OMR. But as long as the neuronal substrate for the OMR has not been analysed this remains hypothetical.
Due to the quite different threshold of the OMR compared to thresholds of direction-selective neurons and the OKR we presume that direction-selective neurons of the goldfish APT are not directly responsible for the OMR. Former lesion studies of the tectum opticum indicated an involvement of this structure in the OMR
Studies, which investigate the coherence threshold of tectal direction-selective neurons and further lesion studies, are needed to clarify which is indeed the neuronal substrate for the OMR.
Our study showed for the first time thresholds for global motion detection in a fish. The thresholds found in the optokinetic system, i.e. neuronal and behavioural threshold are unexpectedly low and come even close to perception thresholds of monkeys and humans. One of the possible explanations for differing thresholds for the OKR and OMR is that the OMR is not mediated by the optokinetic system, but rather by other motion detection systems.
Example of a 100% coherence random dot stimulus in clockwise direction.
(1.21 MB MPG)
Example of a 70% coherence random dot stimulus in clockwise direction.
(1.26 MB MPG)
Example of a 10% coherence random dot stimulus in clockwise direction.
(1.29 MB MPG)
Example of a 0% coherence random dot stimulus.
(1.29 MB MPG)
We are grateful to Yasemin Yavuz, Elma Makic and Annika Simon, who have done some of the behavioural experiments. And we would like to thank Winfried Juhnke, Hermann Korbmacher and Stephanie Krämer for expert technical assistance.