High speed coding for velocity by archerfish retinal ganglion cells
© Kretschmer et al.; licensee BioMed Central Ltd. 2012
Received: 6 February 2012
Accepted: 24 May 2012
Published: 18 June 2012
Archerfish show very short behavioural latencies in response to falling prey. This raises the question, which response parameters of retinal ganglion cells to moving stimuli are best suited for fast coding of stimulus speed and direction.
We compared stimulus reconstruction quality based on the ganglion cell response parameters latency, first interspike interval, and rate. For stimulus reconstruction of moving stimuli using latency was superior to using the other stimulus parameters. This was true for absolute latency, with respect to stimulus onset, as well as for relative latency, with respect to population response onset. Iteratively increasing the number of cells used for reconstruction decreased the calculated error close to zero.
Latency is the fastest response parameter available to the brain. Therefore, latency coding is best suited for high speed coding of moving objects. The quantitative data of this study are in good accordance with previously published behavioural response latencies.
Archerfish have the ability to down aerial prey by shooting precisely aimed jets of water according to size, distance and weight of the prey . As archerfish are swarm fish and compete for food, they need to calculate the impact point of shot down prey on the water surface, turn towards it, and match the swimming speed in order to be the first reaching the impact point. This calculation has to take place in the first few milliseconds of the prey falling and includes the trajectory of the prey, as well as the needed turning angle and the distance to the estimated point of impact. Behavioural experiments show latencies down to 40 ms for the first motor reaction after presentation of a falling prey . Fast turning and acceleration might be achieved by the archerfish’s C-start escape network, which involves large, reticulospinal neurons associated with the Mauthner cells, and which shows fast responses to visual stimuli for life saving purposes [3–5].
Throughout different species retinal ganglion cells are well known to process motion-related information like speed [6, 7] or direction [8–10]. In turtle and rabbit retina, for example, the firing rates of direction sensitive ganglion cells depend on a combination of both direction and speed of a movement [11, 12]. Many of these studies focused on changes in the firing rate [12–14], where the number of spikes induced in a certain period of time conveys the necessary information. More recently, spike timing [15, 16] with temporally precisely fired spikes increasingly attracted attention for understanding the neural code [17–20].
As archerfish show very short behavioural latencies to moving stimuli they are ideally suited to compare rate coding versus latency coding. In this study we, therefore, analysed archerfish ganglion cell responses to stimuli moving with various velocities and tested stimulus reconstruction quality using a maximum likelihood estimation procedure. We found that velocity reconstruction was superior using first spike latency in the reconstruction procedure, compared to first inter-spike interval and to spike rate.
Generally, the recorded ganglion cells showed precise spike timing with short latencies after movement onset when stimulated with the different velocities. This is depicted in the example shown in Figure 1. The raster plots show that all units elicited only a few spikes within a small time window (Figures 1 A & D). In this example spontaneous activity was completely absent, and at movement offset no responses were observed. The population PSTH of all units illustrates the sharp response peaks for all velocities (Figures 1 B & E).
Comparison of stimulus reconstruction with latency, first inter-spike interval and rate
To see which response component conveys most information about the 22 velocities, we calculated latencies, the first inter spike interval (ISI) and the rate in a 100 ms time window after stimulus onset. In addition to the latencies calculated with respect to the known stimulus onset (absolute latencies) we also calculated the population responses to the various velocities and determined the latencies of the single ganglion cell responses with reference to these population response onsets. This will be termed relative latencies in the following.
Velocity estimation with the maximum a posteriori (MAP) estimator on basis of the respective response component is shown in Figures 2 E – H. Estimations using absolute latencies yielded nearly perfect stimulus reconstruction (Figure 2 E). Errors occurred mainly in mistaking the different velocities producing similar latencies, e.g. -20 m/s and -1 m/s or the same velocities for different directions, e.g. -20 and 20 m/s. This has an influence on the centre of mass for all estimations, which is mainly shifted to lower speeds both for downward and upward movements. The best estimation by a single trial was able to correctly reconstruct all velocities (Figure 2 I, red line).
In contrast to absolute latency, the overall estimations based on first ISI (Figures 2 F & J) or rate (Figures 2 G & K) were not able to reconstruct the stimulus velocity. The best trial yielded five correct velocity estimations out of 22 for the rate (red line in Figure 2 K), but none for the first ISI (Figure 2 J). In addition, the centres of mass showed clear direction dependence in case of the rate (Figure 2 G), yet within each direction velocity could not be discriminated. Weak signs of direction discrimination were also found for the first ISI, but the centres of mass were always in the region of estimated positive direction, independent of whether the stimulus direction was upwards or downwards (Figure 2 F).
Absolute latencies versus relative latencies
These results show that velocity estimation with absolute latencies is superior to estimation with the first ISI or the spike rate. The brain, however, has no knowledge about stimulus onset – information that is contained in absolute latency determination.
Therefore, we compared velocity estimation based on the known stimulus (absolute latencies) and the population response (relative latencies).
Since the population response results from the addition of the single responses, the population response onset latencies consequently showed a similar behaviour as the average of the single ganglion cell response latencies (the population response onset latencies are shown for the single experiments in Figures 4 K –O). Consequently, the tuning curve of the relative latencies of all ganglion cells, calculated with respect to the population response onsets, were rather constant around 10 ms (Figure 2 D). Despite the fact that the tuning curve of the relative latencies of the whole population showed no velocity dependence, the accuracy of the overall velocity reconstruction was still acceptable (Figure 2 H). Compared to estimation with absolute latencies (Figure 2 E), however, the centres of mass were more deteriorated, especially at higher velocities. The best trial still yielded 18 correct estimations out of 22 (Figure 2 L).
Velocity tuning curve types for the different response components
The tuning curves of the whole population from Figures 2 A – D potentially mask ganglion cell subpopulations with differing tuning curves that might be important for stimulus reconstruction. Therefore, cluster analysis was applied to separate the single cells into distinct types of velocity tuning functions based on their individual tuning curves. The median values of the relevant response components for each velocity were used from each individual cell, respectively.
Two tuning curves types were identified based on the first ISI (Figures 3 D & E). One had very short first ISI for all velocities (Figure 3 D) and the other type had short ISI for downward movement and long first ISI for upward movements (Figure 3 F), indicating directional tuning. The latter tuning curve type showed very high variability among cells for upward movement. Both types exhibited little variability among cells for downward movements.
Four tuning curves types were separated based on rate (Figures 3 F – I). Three of them showed a similar asymmetry in variability among cells: high variability for upward movements, very precise rates among cells for downward movements (Figures 3 H – I). Two of these tuning types (Figures 3 H & I) showed directional tuning and one had a flat tuning curve (Figure 3 G). The remaining tuning curve type is also flat, however, it exhibited similar variability among cells, both for upward and downward movements and a slightly lower average rate (Figure 3 F).
Four tuning type clusters were identified for relative latencies (Figures 3 K - N). All of them showed directional tuning: one of them with longer relative latencies for upward movement (Figure 3 M), and three with longer latencies for downward movements (Figures 3 K, L & N). The latter three types were discriminated mainly based on the duration of the relative latencies: short latencies (Figure 3 N), intermediate latencies (Figure 3 K), and long latencies (Figure 3 L).
These results show that subpopulations of ganglion cells with different tuning curves exist for the various response components. By comparison of Figure 3 with Figure 2 these results also qualitatively explain the differences in stimulus reconstruction using the different response parameters. The good velocity tuning in all three tuning curve types for absolute latencies is in accordance with the good velocity reconstruction using this parameter (Figure 2 E). For rate the directional selective tuning curves of about half of the recorded cells are in accordance with the good reconstruction of movement direction (Figure 2 G), but since no velocity tuning exists in the tuning curves, velocity could not be reconstructed. In the case of the first ISI only 26 out of 109 cells showed directional tuning, with high variability for upward movements among cells. In the reconstruction this obviously led to some degree of direction reconstruction, as indicated by the centres of mass in Figure 2 F, however, with a shift in the reconstruction to upward movement. For the case of relative latencies the tuning curves are qualitatively in good agreement with the centres of mass for velocity reconstruction shown in Figure 2 H. It remains still unclear, however, how the remaining velocity reconstruction, indicated by the diagonal in Figure 2 H, is accomplished.
Comparison of individual experiments
When using absolute latencies for estimation obviously some information about stimulus velocity was retained in experiments 4 and 5 (Figures 4 D and E). Most errors occurred in mistaking different velocities producing similar latencies or mistaking velocities for opposite directions. This information was gone when using relative latencies, where only estimation of the correct direction was possible in experiment 4 (Figure 4 I) and no correct estimation at all was possible in experiment 5 (Figure 4 J). The tuning curves of the population response onsets showed clear velocity dependence in all experiments (Figure 3 K - O), but the variability in experiments 4 and 5 was considerably higher. Since the population response onset is determined by the sum of the single cell responses this led us to the suggestion that precision of single cell responses might vary between experiments.
Precision of single cell latencies
Archerfish ganglion cells are potentially able to contribute to the high speed calculation of a prey impact point on the water surface by coding different speeds and directions with precisely timed spiking. Ganglion cell responses show a very low noise level and response rate. With latencies down to 12 ms after the onset of a moving stimulus some responses are faster than recorded latencies to light flashes of the same intensity and to ganglion cell latencies of other species to comparable moving stimuli e.g. 100 ms in turtle or 50 ms in rabbit and salamander [21, 22].
Precisely timed and directed motor reactions in archerfish can occur at latencies down to 40ms after presentation of falling prey . It is likely that this motor reaction is triggered by the Mauthner cells  because these are directly electrically coupled to motorneurons over descending interneurons and hence permit very fast signal transmission of only a few milliseconds . Mauthner cells are also involved in other visually evoked behaviour like looming, where latencies of 142 ms have been observed, and feeding [24, 25]. Compared to sound induced responses with latencies of 10 to 15 ms , these longer latencies in these visually evoked behaviours reflect the comparably slow phototransduction and probably the increased amount of pre-processing, that might be involved to calculate the transformation of size and velocity and the potential time to collision with an approaching object, e.g. a predator. Similar stimulus parameters must be estimated by the archerfish nervous system to calculate the point of impact of the prey on the water surface, but the behavioural response takes place on a much shorter timescale, suggesting a time-optimized neural code.
In this study the best velocity reconstruction could be achieved on the basis of absolute latencies. In the case of the reconstruction with rates performance was overall very poor, and in the case of reconstruction with first ISI only some reconstruction of stimulus direction was possible.
Therefore, archerfish ganglion cells might encode information about stimulus speed and direction through the precise timing of first spikes. Since the brain has no knowledge about the stimulus onset, latency can only serve as a coding parameter when information about the stimulus onset is represented by the response of a cell population. Precision of the single cell responses with respect to the population response onset becomes then an important parameter too. When taking the population responses of all recorded cells into account estimation of the 22 velocities with relative latencies was possible far above chance level and above the level achievable by estimating velocities on the basis of rates or first inter spike intervals. The results clearly depend on the amount of cells involved in the estimation and on the precision of the recorded ganglion cell response latencies.
Comparison of our single experiments shows that under favourable conditions (Experiment 2) on the average 63% (38 out of 60) correct estimation could be obtained, compared to a chance rate of 4.5%. The fact that single cell precision was less precise in experiments 4 and 5 supports the notion that spiking precision is very critical for this task. It is unclear why less precise cells were recorded in experiments 4 and 5. It could of course be that the quality of the experiments was inferior in experiments 4 and 5, but this seems unlikely to us since we recorded more cells in experiment 4 than in experiments 1 and 3. It could as well be that by chance more cells with intrinsically less precise responses were recorded in experiments 4 and 5. We took care to align the retina dorso-ventral in our experiments. However, we had no good control how far dorsal or ventral the recordings took place. Since the archerfish retina is not homogeneous, with a ventral region of increased photoreceptor and ganglion cell density  it could be that we recorded more cells from one region or the other region in the different experiments.
In a competition model, based on directionally selective ganglion cells in archerfish, simulated reaction times for the decision to move to the left or to the right were in good accordance with behavioural data . In this study directionality was based on rate, leading to about 42% directionally tuned cells, which is similar to our number of 51% directionally tuned cells when using rate as response parameter (Figure 3). Therefore rate might be an additional possibility for making the decision to move left or right. However, the authors also showed that directional selectivity based on rate is invariant to changes in velocity, the same result we obtained (Figure 2 G). The task for the archerfish is, however, to calculate the impact point of shot down prey before it reaches the water surface [2, 3]. For doing this, knowledge of the speed of the prey in addition to its direction is indispensable. From our results only latency can serve as response parameter for this task.
Generally, population response onset is a good indicator of stimulus changes, even in cases with ongoing spike activity . Therefore, it seems reasonable to use this as a time reference for relative latency determination that could be used by the nervous system. The definition of population response onset is, however, arbitrary. In the case of archerfish, with low spontaneous spike rate under our experimental conditions, relative latencies could also be calculated with respect to the first spike within a cell population. This is basically the same as lowering the threshold of the population response onset to one spike. For comparison, we therefore also estimated the stimulus velocities based on relative latencies with reference to the first spike for each stimulus presentation. Velocity estimation was, in this case, comparable to estimation based on absolute latencies (Additional file 1: Figure S1).
We do not know whether the task of velocity estimation is based on the activity of all cells in a certain retinal region or the activity of a subset of cells responding very precisely. It is also not clear whether the brain uses the population response as reference signal, and if yes whether it uses the population response from all retinal ganglion cells or from a subset of very precisely responding cells. Even the spike of the fastest responding cell could serve as a reference signal as long as the ongoing activity is low. The question is, however, how the ganglion cell activity looks like under natural viewing conditions and how reliable the different reference signal then are.
The results presented in this study show that reliable and fast velocity reconstruction can be obtained with latency as response parameter of retinal ganglion cells. This is true for absolute as well as relative latencies. The latency available for the archerfish is composed of the population response latency (20 – 30 ms at the optimal velocities) plus the relative latency (around 10 ms) and, therefore, in the range of 30 – 40 ms. Since the transmission from the optic nerve to a Mauthner cell response takes only about 4 ms , this is in good agreement with the fastest reaction times of 40 – 45 ms measured in behavioural experiments [2, 27].
All animal experiments were performed in compliance with the guidelines for the welfare of experimental animals issued by the European Communities Council Directive of 24 November 1986 (86 609 EEC) and the laws of the Federal Government of Germany (Tierschutzgesetz; BGBl. I S. 1206, 1313 and BGBl. I S. 1934). Institutional approval was obtained by the ethical committee of the University of Oldenburg.
Extracellular multi-electrode recordings from ganglion cells of the archerfish (Toxotes chatareus) retina were performed using a ten by ten silicon array (Blackrock Microsystems; Salt Lake City, UT, USA) with an inter-electrode distance of 400 μm. Animals were decapitated and the retina was removed keeping the pigment epithelium attached. The flattened retina/pigment epithelium preparation was then placed into the translucent recording chamber. In the archerfish the pecten and different coloration of the dorsal/ventral retina can serve as landmarks. Incisions, that are necessary to flatten the retina, were made along the temporal/nasal line that served then as landmarks to keep track of the orientation. In the recording chamber the retina was oriented in such a way that the final stimulation with the moving grating was dorso-ventral. During the experiment the preparation was constantly superfused with oxygenated ringer solution (120mM NaCl, 5mM KCl, 2mM CaCl2, 2mM MgCl2, 10mM glucose, 22mM NaHCO3, bubbled with 95% O2 -5% CO2 ; pH7.4). The temperature within the recording chamber was held constant at 20°C by using a temperature controlled, translucent heating chamber underneath the preparation. Ganglion cell activity was pre-amplified, sampled and stored by a 128-channel Cerebus neural signal acquisition system (Blackrock Microsystems; Salt Lake City, UT, USA).
After an experiment spikes were sorted with the Plexon Offline Sorter, Version 2.8.8 (Plexon Neurotechnology Research Systems, Dallas, Tx, USA). The supervised k-means clustering algorithm was used to cluster spike waveforms on the basis of principle component analysis. All further analysis with the resulting time stamps from the sorting procedure was done in MATLAB, Version 220.127.116.114 (The MathWorks Inc., Natick, Mass, USA).
Light stimulation was realized using a white high-power LED (LXHL-FW6C; Luxeon, San Jose, CA), with broad emission spectrum as light source. Full field light flashes (12 mW/m2 on the retina) of 50 ms duration (1.54 Hz) were used to search for ganglion cell responses while penetrating the preparation. When enough ganglion cells responded to the search flashes, light stimulation was switched to a regular grating used for stimulation with different velocities. The grating was composed of black and white bars, each with a width of 400 μm. About 80% of the cells that responded to the search flashes responded also to the movement stimulation. A few cells that responded to movement did not respond to the search flashes. A total of 109 cells responding to movement were recorded in five retinae from five different archerfish.
The grating was generated by constantly projecting the image of a photographic slide onto the retina. The slide was composed of non-transparent and transparent stripes. Light intensity on the illuminated parts of the retina was the same as above. The edges of the stripes were initially aligned with the electrode rows (Figure 1 G). The image of the slide could then be moved with various speeds perpendicular to the bar orientation (direction of bar movement dorso-ventral on the retina) by using an x-y miniature mirror system (Datronik, Rastede, Germany) connected to a stimulus computer. The stimulus computer was synchronized with the data acquisition computer for exact stimulus on- and offset determination. Each movement stopped after 400 μm, when edges of the bars reached the neighbouring electrode rows. Movement direction changed after every second movement step. This ensured that the grating reversed in half of the stimuli. The third movement started into the opposite direction. Eleven different speeds (1 m/s, 2 m/s, 3 m/s, 4 m/s, 5 m/s, 6 m/s, 8 m/s, 10 m/s, 12 m/s, 16 m/s, 20 m/s) were used, which were calculated according to the speed distribution of natural objects at approximately 30 cm distance [2, 29]. These speeds correspond to retinal stimulus velocities of 1.21 mm/s; 2,42 mm/s; 3,63 mm/s; 4,84 mm/s; 6,05 mm/s; 7,25 mm/s; 9,67 mm/s; 12,09 mm/s; 14,50 mm/s; 19,34 mm/s; 24,18 mm/s. The different velocities were applied in random order. Since the distance of movement was constant, stimulus duration varied according to movement speed. In the text, the two different directions were named downwards (movement into dorsal direction on the retina) and upwards (movement into ventral direction on the retina), in order to refer to the movement direction of potential prey in the environment. In the figures downwards and upwards were referred to as negative and positive velocities, respectively.
Each of the 22 velocities ( = speed and direction) was presented 60 times in random order, yielding 1320 stimulus presentations per experiment. Presentations were separated by a break (600 ms minus the varying time of movement; see Figure 1) from the next stimulus. Since the grating reversed every second movement step the illuminated and dark bars were on either side of the electrodes at the start of 30 trials, respectively. We tested whether this grating reversal had an effect on the tuning curves by subdividing the data for the two grating phases. The result is shown in Additional file 2: Figure S2 for the data based on absolute latency. Since we found no difference in the velocity tuning curves between the reversed and non-reversed grating we put all data together for further analysis.
Cluster analysis of tuning curves
In order to identify different tuning curve types, individual cells were classified according to the form of their individual tuning curves. We used the k-means clustering algorithm implemented in JMP 7.01 (SAS Institute Inc., Cary, NC, USA). The feature vector was 22-dimensional, consisting of the median values of the response component under study (absolute and relative latency, first ISI, rate) at each velocity, respectively. The k-means algorithm demands a predefined number of clusters. For obtaining this number we first visually inspected 2- and 3-dimensional plots of the two and three most important principal components and subjectively decided how many clusters were predefined. In addition, the resulting clustering was inspected in diagrams showing the overlay of the respective tuning curves. Finally, the average distance of all tuning curve vectors of a given cluster to the cluster centre was calculated. The resulting numbers were then compared to the average distances for clustering with one cluster more or one cluster less. If the average distance was smallest this cluster number was accepted.
The stimulus parameters speed and direction (= velocity) were reconstructed by analyzing three parameters of the ganglion cell responses: latency of the first spike; first inter-spike interval (ISI); rate within a 100 ms window after stimulus onset. Latency of the first spike was determined for two cases: latency with respect to stimulus onset (absolute latency) and latency with respect to population response onset (relative latency). For determination of the population response onset the spike trains of all cells were added and the resulting data smoothed by a moving average. Then population response onset was arbitrarily defined for each stimulus presentation as the time point where the population response reached 2.5 times the standard deviation .
with f (X|θ) being the likelihood function, the probability of observed data X given the stimulus θ. s (θ) is the prior distribution of the stimuli.
For the computation the trials, consisting of one presentation of each velocity (yielding in total 60 trials each with 22 velocities), were separated into training sets of 59 trials and one trial used for testing. Each trial was used as test set using a “jackknife” procedure. This procedure minimizes overfitting effects. For each of the trials used for training we calculated the empirical distribution and/or the Bernoulli distribution for each recorded unit. The response to each presented velocity of the trial used for testing was then compared to one of the distribution functions, for the respective response parameter, to calculate the probability that the particular stimulus was presented. When a spike occurred within 100 ms after stimulus onset, the empirical distribution was used, whereas the Bernoulli distribution was used for responses later than 100 ms after stimulus onset. The product of all probabilities from all units was then calculated and multiplied with the prior, the frequency distribution of the different velocities. The maximum probability is then taken as the most likely stimulus velocity.
In our case the priori distribution is uniform with an equal probability for each velocity. In total there are K trials and the number of all units is N. Let T be the random variable for the stimulus occurrence.
The likelihood function of the random variable X for each observation x i of unit i in trial k was estimated from the training dataset for all 100 ms time intervals where stimulation θ occurred. M t is defined as the set of all units that elicit spikes within the specified time interval .
Additionally we calculated a centre of mass for all estimations for one velocity to determine an overall estimation tendency (e.g. to get an indication for the overall direction preference).
with the locations c weighted by their values v and V as the sum of all values for each estimation i.
This was done ten times by randomly picking the number of units from the dataset, and the mean of these ten values was defined as the error measure.
We would like to thank Lena Köpcke for critically reading the Methods section. This work was supported by DFG (Am 70/12-1).
- Timmermans PJ, Vossen JM: Prey catching in the archer fish: does the fish use a learned correction for refraction?. Behav Processes. 2000, 52 (1): 21-34. 10.1016/S0376-6357(00)00107-8.PubMedView ArticleGoogle Scholar
- Schlegel T, Schuster S: Small circuits for large tasks: high-speed decision-making in archerfish. Science. 2008, 319 (5859): 104-106. 10.1126/science.1149265.PubMedView ArticleGoogle Scholar
- Wöhl S, Schuster S: Hunting archer fish match their take-off speed to distance from the future point of catch. J Exp Biol. 2006, 209 (Pt 1): 141-151.PubMedView ArticleGoogle Scholar
- Canfield JG: Temporal constraints on visually directed c-start responses: behavioral and physiological correlates. Brain Behav Evol. 2003, 61 (3): 148-158. 10.1159/000069751.PubMedView ArticleGoogle Scholar
- Canfield JG: Functional evidence for visuospatial coding in the mauthner neuron. Brain Behav Evol. 2006, 67 (4): 188-202. 10.1159/000091652.PubMedView ArticleGoogle Scholar
- Cleland BG, Harding TH, Tulunay-Keesey U: Response to the length of moving visual stimuli of the brisk classes of ganglion cells in the cat retina. J Physiol. 1983, 345: 27-45.PubMed CentralPubMedView ArticleGoogle Scholar
- Lee BB, Willshaw DJ: Responses of the various types of cat retinal ganglion cells to moving contours. Vision Res. 1978, 18 (7): 757-765. 10.1016/0042-6989(78)90114-1.PubMedView ArticleGoogle Scholar
- Barlow HB, Hill RM, Levick WR: Retinal ganglion cells responding selectively to direction and speed of image motion in the rabbit. J Physiol. 1964, 173: 377-407.PubMed CentralPubMedView ArticleGoogle Scholar
- Clifford CWG, Ibbotson MR: Fundamental mechanisms of visual motion detection: models, cells and functions. Prog Neurobiol. 2002, 68 (6): 409-437. 10.1016/S0301-0082(02)00154-5.PubMedView ArticleGoogle Scholar
- Taylor WR, Vaney DI: New directions in retinal research. Trends Neurosci. 2003, 26 (7): 379-385. 10.1016/S0166-2236(03)00167-X.PubMedView ArticleGoogle Scholar
- Oyster CW, Takahashi E, Collewijn H: Direction-selective retinal ganglion cells and control of optokinetic nystagmus in the rabbit. Vision Res. 1972, 12 (2): 183-193. 10.1016/0042-6989(72)90110-1. Feb 1972PubMedView ArticleGoogle Scholar
- Thiel A, Greschner M, Eurich CW, Ammermüller J, Kretzberg J: Contribution of individual retinal ganglion cell responses to velocity and acceleration encoding. J Neurophysiol. 2007, 98 (4): 2285-2296. 10.1152/jn.01342.2006.PubMedView ArticleGoogle Scholar
- Shadlen MN, Newsome WT: Noise, neural codes and cortical organization. Curr Opin Neurobiol. 1994, 4 (4): 569-579. 10.1016/0959-4388(94)90059-0.PubMedView ArticleGoogle Scholar
- Thiel A, Greschner M, Ammermüller J: The temporal structure of transient on/off ganglion cell responses and its relation to intra-retinal processing. J Comput Neurosci. 2006, 21 (2): 131-151. 10.1007/s10827-006-7863-x.PubMedView ArticleGoogle Scholar
- Theunissen F, Miller JP: Temporal encoding in nervous systems: a rigorous definition. J Comput Neurosci. 1995, 2 (2): 149-162. 10.1007/BF00961885.PubMedView ArticleGoogle Scholar
- Oram MW, Xiao D, Dritschel B, Payne KR: The temporal resolution of neural codes: does response latency have a unique role?. Philos Trans R Soc Lond B Biol Sci. 2002, 357 (1424): 987-1001. 10.1098/rstb.2002.1113.PubMed CentralPubMedView ArticleGoogle Scholar
- Gollisch T, Meister M: Rapid neural coding in the retina with relative spike latencies. Science. 2008, 319 (5866): 1108-1111. 10.1126/science.1149639.PubMedView ArticleGoogle Scholar
- Johansson RS, Birznieks I: First spikes in ensembles of human tactile afferents code complex spatial fingertip events. Nat Neurosci. 2004, 7 (2): 170-177. 10.1038/nn1177.PubMedView ArticleGoogle Scholar
- Averbeck BB, Latham PE, Pouget A: Neural correlations, population coding and computation. Nat Rev Neurosci. 2006, 7 (5): 358-366. 10.1038/nrn1888.PubMedView ArticleGoogle Scholar
- Meister M, Berry MJ: The neural code of the retina. Neuron. 1999, 22 (3): 435-450. 10.1016/S0896-6273(00)80700-X.PubMedView ArticleGoogle Scholar
- Greschner M, Bongard M, Rujan P, Ammermüller J: Retinal ganglion cell synchronization by fixational eye movements improves feature estimation. Nat Neurosci. 2002, 5 (4): 341-347. 10.1038/nn821.PubMedView ArticleGoogle Scholar
- Berry MJ, Brivanlou IH, Jordan TA, Meister M: Anticipation of moving stimuli by the retina. Nature. 1999, 398 (6725): 334-338. 10.1038/18678.PubMedView ArticleGoogle Scholar
- Fetcho JR: Excitation of motoneurons by the mauthner axon in goldfish: complexities in a "simple" reticulospinal pathway. J Neurophysiol. 1992, 67 (6): 1574-1586.PubMedGoogle Scholar
- Preuss T, Osei-Bonsu PE, Weiss SA, Wang C, Faber DS: Neural representation of object approach in a decision-making motor circuit. J Neurosci. 2006, 26 (13): 3454-3464. 10.1523/JNEUROSCI.5259-05.2006.PubMedView ArticleGoogle Scholar
- Canfield JG, Rose GJ: Activation of mauthner neurons during prey capture. J Comp Physiol A. 1993, 172: 611-618. 10.1007/BF00213683.View ArticleGoogle Scholar
- Ullmann JFP, Moore BA, Temple SE, Fernandez-Juricic E, Collin SP: The retinal wholemount technique: A window to understanding the brain and behaviour. Brain Behav Evol. 2012, 79: 26-44. 10.1159/000332802.PubMedView ArticleGoogle Scholar
- Tsvilling V, Donchin O, Shamir M, Segev R: Archer fish fast hunting maneuver may be guided by directionally selective retinal ganglion cells. E J Neurosci. 2012, 35: 436-444. 10.1111/j.1460-9568.2011.07971.x.View ArticleGoogle Scholar
- Zottoli SJ, Hordes AR, Faber DS: Localization of optic tectal input to the ventral dendrite of the goldfish Mauthner cell. Brain Res. 1987, 401: 113-121. 10.1016/0006-8993(87)91170-X.PubMedView ArticleGoogle Scholar
- Segev R, Schneidman E, Goodhouse J, Berry MJ: Role of eye movements in the retinal code for a size discrimination task. J Neurophysiol. 2007, 98 (3): 1380-1391. 10.1152/jn.00395.2007.PubMedView ArticleGoogle Scholar
- Greschner M, Thiel A, Kretzberg J, Ammermüller J: Complex spike-event pattern of transient on-off retinal ganglion cells. J Neurophysiol. 2006, 96 (6): 2845-2856. 10.1152/jn.01131.2005.PubMedView ArticleGoogle Scholar
This article is published under license to BioMed Central Ltd. This is an Open Access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/2.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.