Pattern- and contrast-dependent visual response in the box jellyfish Tripedalia cystophora

Eyes come in a great variety of shapes and sizes and are used for an impressive range of light-guided behaviours, from simple light detection to highly advanced intraspecific communication (Land and Nilsson, 2012). A normal assumption is that the degree of complexity found in the eyes and the behaviour they support is closely correlated with the brainpower of the organism. It has been considered an enigma, therefore, that cubomedusae, or box jellyfish, possess very elaborate visual equipment (Wehner, 2005). All known species of box jellyfish have the same overall arrangement of the eyes. They have four sensory structures called rhopalia situated in rhopalial niches along the lower part of the bell, and each of them carries six eyes of four distinct morphological types (Hertwig and Hertwig, 1878; Claus, 1878; Berger, 1898; Werner, 1975; Yamasu and Yoshida, 1976; Martin, 2004): an upper and a lower lens eye, a pair of pit eyes and a pair of slit eyes. The lens eyes are structurally similar to vertebrate and cephalopod eyes in that they have a spherical lens with a graded refractive index, an upright hemisphere-shaped retina, a cornea, a vitreous space and, in the case of the lower lens eye, a movable iris (Nilsson et al., 2005).

Box jellyfish also display a more diverse behavioural repertoire than any other known cnidarian and several of them are light guided (Garm and Ekström, 2010). It should be kept in mind that almost all the behavioural data stem from a single species of box jellyfish, Tripedalia cystophora Conant 1897, from the Caribbean, and thus many more behaviours are bound to surface when more species are examined. Tripedalia cystophora is found between the prop roots of mangrove trees, where they prey on a single species of copepod gathering in light shafts between the roots (Stewart, 1996). The medusae use their eyes to seek out the light shafts where they passively hunt (Buskey, 2003). They rest on the bottom of the mangrove lagoon at night (Garm et al., 2012). Every morning they have to navigate back to their habitat between the roots, which they do by visually detecting the mangrove canopy through the water surface (Garm et al., 2011). The medusae are also able to detect and avoid obstacles in their surroundings (like the prop roots) and for this they probably use their lower lens eye (Garm et al., 2007b).

The morphological and optical data indicate that both lens eyes are image forming with a spatial resolution in the range of 10–20 deg (Nilsson et al., 2005). In the case of the upper lens eye, the use of spatial information is confirmed by the navigation behaviour, where they see the direction to the canopy (Garm et al., 2011). The obstacle avoidance behaviour also indicates the use of spatial information and true image formation, in that its onset has been shown to be correlated with the size of the obstacle on the retina (Garm et al., 2007b). Still, because the obstacles were dark on a bright background, the behaviour could also have been triggered by a directional drop in light intensity when approaching the obstacles, and thus be part of a simpler positive phototaxis.

Here we examined the obstacle avoidance behaviour of T. cystophora and tested whether an image of the obstacle is needed to accomplish this behaviour or whether it is a case of positive phototaxis. In a behavioural assay, the medusae were presented with a visual scene of either alternating dark and bright stripes in different orientations or an uniformly grey wall. We hypothesized that the medusae would only be able to avoid the striped walls and not the uniform walls without contrast. Further, we hypothesized that vertical stripes with high contrast would evoke the strongest response, as they would be the most visible and have the most resemblance with the natural obstacles, the prop roots.

We used adult animals (7–9 mm in bell diameter) from our cultures at the University of Copenhagen. The animals were cultured in 250 litre tanks with recycled seawater at 28°C and 30 PSU. In the culture tanks, the light:dark cycle was 8 h:16 h and the medusae reached adult size in approximately 2 months. A total of 36 medusae were used and each medusa was only used in a single assay.

The experiments were conducted in a round 3 litre tank with a diameter of 16 cm. The tank was filled with water from the culture tank to minimize stress, which can be induced by moving the animals to water with changes in salinity and/or chemical composition. The water depth was ~12 cm and the water was kept at 28°C by placing the tank on a heating plate. The wall of the tank had a changeable visual scene of either 2 cm wide grey and white stripes or a uniformly grey wall. Undiffused light came from a 11 W fluorescent bulb (OSRAM Longlife, OSRAM, Augsburg, Germany) situated approximately 50 cm above the centre of the tank, thereby creating an even illumination of the behavioural arena. The light intensity measured at the surface straight under the lamp was 76 W m⁻².

In the case of the stripes (contrast and orientation experiments), three different orientations were used: vertical, 45 deg oblique and horizontal. All three orientations were tested with five different grey tones, resulting in contrasts (c) of 0.93, 0.71, 0.39, 0.27 and 0.17. The contrast was calculated as (I_w−I_g)/(I_w+I_g), where I_g is the intensity of reflected light from the grey strips and I_w is the intensity reflected from the white stripes, both corrected by the absorption spectrum of the 500 nm opsin (Govardovskii et al., 2000) present in the lower lens eye (Coates et al., 2006; Garm et al., 2007a). The intensities were measured from 350 to 700 nm using a radiospectrophotometer (ILT900W, International Light Technologies, Peabody, MA, USA) with the sensor held perpendicular to the wall at a distance of 1 cm and 1 cm below the surface.

With the uniformly dark walls (intensity experiments), five different grey tones were used one at a time. The grey tones were chosen to match the mean intensities of the white stripe and the five different grey stripes, respectively (with an accuracy of ±5%), such that the intensity of the reflected light of the darkest uniformly grey wall matched the mean of reflected light of the white and the darkest grey stripe. Light intensities were measured as for the contrast experiments.

At the onset of each behavioural assay, a medusa was placed in the centre of the arena and left to adjust for 5 min, after which it had re-extended its tentacles and swam with normal pulse rate. After this acclimation each medusa was tested with either the five striped walls with the same orientation but varying contrast (contrast and orientation experiments) or the five uniformly dark walls (intensity experiments). The visual scene was changed every 4 min and the order of contrast/darkness was randomized. The experimental series (acclimation plus five tests) lasted 25 min and was repeated eight times using eight different experimental animals in the case of vertical stripes and uniform grey tones. The experiments with oblique and horizontal stripes were repeated 10 times using 10 medusae. The swim pattern during the last 2.5 min with each visual scene was recorded from above using a video camera (Sony Handycam DCR-HC40, Sony, Tokyo, Japan). The timing equals that used in earlier experiments (Garm et al., 2007b).

The video recordings were turned into swim trajectories with a temporal resolution of 1 s using a custom program for MATLAB 2011a (MathWorks, Natick, MA, USA). The average distance to the wall was calculated from the trajectories. A temporal resolution of 0.5 s was also tested for three recordings to make sure the sampling rate was sufficient to resolve the swim pattern; no difference between the two resolutions was found, therefore the 1s resolution was used for all recordings. The number of avoidance responses for each individual and each visual scene was counted manually, and the distance to the wall at the behavioural onset was determined for each avoidance response. An avoidance response is defined as the medusa swimming towards the wall and then turning a minimum of 120 deg in two to three swim contractions with an increased pulse rate (see supplementary material Movie 1 for an example). Finally, in the contrast experiments, the distance of the avoidance response (when the medusa started turning) was used to calculate the visual angle of the stripes following the equation tan½α=a/b, where α is the visual angle, a is the width of the stripe (2 cm) and b is the distance of avoidance. All statistical tests were performed in Biostat 2008 Professional (version 5.4.0.0, AnalystSoft, Vancouver, BC, Canada) and consisted of one-way ANOVAs followed by Tukey–Kramer post hoc tests unless otherwise stated. Fisher's least significant difference (LSD) post hoc test was used in cases of uneven variance (distance during avoidance and visual angle during avoidance).

As it was not possible to take light intensity measurements from small confined areas in the mangrove habitat of the medusae, we used an indirect approach. Photos of the mangrove habitat in Puerto Rico including several prop roots and medusae were taken at noon with a standard underwater camera. A representative of these RGB pictures showing the typical habitat was chosen. The red channel was removed from the photo using the program Corel PhotoPaint (version X3, Corel Corporation, Ottawa, ON, Canada) to better match the spectral sensitivity of the medusae, and the photo was converted into 8 bit greyscale. The average pixel value (0=black, 255=white) was then determined from a rectangular area (500 pixels) of a prop root and from the neighbouring area in the water. These pixel values were used as relative estimates of the light intensity, and the contrast between the root and the water was calculated in the following way: (PV_w−PV_r)/(PV_w+PV_r), where PV_w is the pixel value from the water and PV_r is the pixel value from the root. The procedure was repeated for four roots at different distances to the camera. The absolute distances were not measured, but the relative distance was determined using the location in the photo where the root intersected the water surface.

In the experiments using grey and white stripes the medusae performed many clear obstacle avoidances. With increasing contrast from 0.17 to 0.93, the medusae responded with a stronger obstacle avoidance response for all three stripe orientations. The swim trajectories show that medusae made only a few turns and came close to the wall when contrast was low (Figs 1, 2 and 3). At c=0.93, the medusae turned frequently and stayed centred in the tank (Figs 1, 2 and 3). This is in contrast to the results from the intensity experiments with the uniform grey walls (Fig. 4), where only little avoidance behaviour was seen.

The behavioural change with contrast is confirmed when the average distance to the wall is calculated (Fig. 5A). With the vertical stripes and the lowest contrast, the medusae had an average distance of 2.5 cm to the wall, whereas at c=0.93 the average distance was 5.7 cm (all behavioural data are summarized in Table 1). These differences are significant between a given contrast level and all others except the neighboring levels (one-way ANOVA, F_4,35=15.1, P<0.0001; post hoc Tukey–Kramer, 0.0001<P<0.0071). The results are the same for the oblique stripes, except here there is also a significant difference between c=0.39 and c=0.71 (one-way ANOVA, F_4,45=49.2, 0.001<P<0.016). With the horizontal stripes there were no differences in the average distance to the wall between the four lowest contrasts, but the medusae stayed significantly farther away from the darkest stripes than the four others (one-way ANOVA, F_4,45=12.6, 0.0001<P<0.0003).

With higher contrast, the medusae not only stayed farther away from the wall of the tank, but they also performed more obstacle avoidances per minute (Fig. 5B, Table 1). The highest rate, 3.3 min⁻¹, was obtained with the vertical stripes at c=0.93. This was significantly higher than the rates with vertical stripes at c=0.17, 0.27 and 0.39 (one-way ANOVA, F_4,35=11.7, 0.0001<P<0.0002), and c=0.71 also resulted in a significantly higher avoidance rate than c=0.17 (P=0.011). With the oblique stripes there was also an increase with contrast (Fig. 5B), and here all differences were significant (one-way ANOVA, F_4,44=61.3, 0.0001<P<0.044) except between c=0.17 and c=0.25 (P=1). In the experiments with horizontal stripes, only c=0.93 produced a significantly higher avoidance rate than the four other contrasts (one-way ANOVA, F_4,45=16.2, 0.0001<P<0.0002).

The increasing number of avoidances with higher contrast stripes was also performed at a greater distance from the stripes (Fig. 6A, Table 1). On average, the obstacle avoidance responses were performed 2.2 cm from the wall with vertical stripes at c=0.17, but 4.6 cm from the wall at c=0.93. The differences are significant between c=0.93 and the four other contrasts (one-way ANOVA, F_4,134=8.7, P<0.0001; post hoc Fisher's LSD, 0.0001<P<0.012) and between c=0.27 and c=0.71 (P=0.015). In the case of the oblique stripes, only the three highest contrasts could be tested because of the low sample size (N=1) for c=0.17 and 0.27. Still, the avoidances were performed significantly farther away from the wall at c=0.93 than at c=0.39 and 0.71 (one-way ANOVA, F_2,125=26.8, P<0.0001; post hoc Fisher's LSD, P<0.001). With the horizontal stripes and c=0.93, the avoidances were performed 4.6 cm from the wall, and this was significantly farther away than with the four other contrasts (one-way ANOVA, F_4,81=8.5, P<0.0001; post hoc Fisher's LSD, 0.0004<P<0.023). The width of the stripes at the distance of avoidance was converted into visual angle on the retina, and the average of these angles varied from 52 deg (horizontal, c=0.17) to 25 deg (oblique, c=0.93; Fig. 6B, Table 1). This transformation of the data had no significant effect on the statistics. The smallest visual angle provoking an avoidance response was 15 deg.

When presenting the medusae with uniformly grey tank walls, the obstacle avoidance behaviour was almost completely abolished even with the darkest grey tone matching the mean intensity of the white and the darkest (black) stripe. For all five grey tones, the medusae swam with few turns and stayed most of the time in the periphery of the tank, often touching the wall (Fig. 4). This resulted in them having the same average distance to the tank wall, 2.5–2.9 cm (one-way ANOVA, F_4,35=0.63, P=0.64; Fig. 5A). Further, at the three highest intensities (matching c=0.17, 0.27 and 0.39), no avoidance behaviour was observed, and only very few avoidances were seen with the two darker walls (0.05 avoidances min⁻¹ in both cases). This slight increase with darker walls was not significant (one-way ANOVA, F_4,35=0.75, P=0.57).

When comparing the response to stripes with the same contrast but different orientation with the response to the corresponding grey tone, interesting differences are seen. With the two lowest contrasts, all four different experimental conditions resulted in the same general distance to the wall (one-way ANOVA, P=0.052 and 0.36, respectively; Fig. 5A). At c=0.39, only the vertical stripes kept the medusae farther away than the corresponding grey tone (one-way ANOVA, F_3,32=4.03, P=0.014). When taking one step further up in contrast, the vertical and oblique stripes gave similar results and both were significantly higher than the horizontal stripes and grey tone (one-way ANOVA, F_3,32=10.2, 0.00044<P<0.016). With the highest contrast (c=0.93), the three different stripes caused the medusae to keep the same distance to the wall, 5.7–5.9 cm (0.75<P<0.99), which in all cases was significantly farther away than the corresponding grey tone (one-way ANOVA, F_3,32=37.4, P<0.0001, P<0.0001).

A similar picture is seen with the avoidance rate (Fig. 5B). Here the vertical stripes produced a stronger response than the three other visual scenes, even at c=0.27 (one-way ANOVA, F_3,32=5.5, 0.005<P<0.03). At c=0.71, there was a higher avoidance rate with both the vertical and oblique stripes than the grey tone (one-way ANOVA, F_3,32=9.5, 0.0007<P<0.001) and with the vertical stripes compared with the horizontal stripes (P=0.038). With the highest contrast, all stripes resulted in similar responses, all of which were significantly higher than the corresponding grey tone (one-way ANOVA, F_3,32=16.8, 0.0001<P<0.0004; Fig. 5B).

When considering the average distance of the avoidances and the average visual angles, there were no significant differences between the four experimental conditions at any of the contrasts (Fig. 6).

A relative measure of contrast between the prop roots and the surrounding water as a function of distance was obtained from an underwater photo (Fig. 7A). Four roots in the picture were analyzed, with root 1 being the closest and root 4 the furthest away. In the picture adjusted to the spectral sensitivity of T. cystophora (Fig. 7B) there was a correlation between relative distance and relative contrast. Root 1 had a contrast of 0.39, for root 2 and 3 it was 0.24, and the most distant root (root 4) had a relative contrast of 0.15.

The results presented here clearly demonstrate that the visually guided obstacle avoidance behaviour described for cubomedusae (Garm et al., 2007b) is dependent on actual detection of the obstacle using spatial information and is not a mere positive phototaxis. When we presented T. cystophora medusae with a visual scene without spatial information, the behaviour disappeared even though the overall brightness of the wall equalled that of a scene with stripes resulting in many avoidances. We also showed that for all three orientations of the stripes, increasing contrast resulted in the medusae staying farther away from the wall and performing more avoidances. Finally, we found that the orientation of the obstacle influences the strength of the response, with vertical stripes resulting in the strongest response, followed by the oblique stripes with an intermediate effect and lastly the horizontal stripes with the weakest response.

Our experiments returned a surprising result. We expected the obstacle avoidance response to have a contrast threshold triggering the behaviour. That is, once a certain contrast is present on the retina, the medusa would acknowledge the presence of the obstacle and start the response. The results strongly indicate that this is not the case, because there is a gradual change of the response strength (measured as average distance to wall, avoidance rate and object size on retina) more or less proportional with the change in contrast, at least for the vertical stripes. This could be because a higher contrast means greater certainty that there is an obstacle and thus a greater ‘willingness’ to respond, but there is another possible explanation and the two are not mutually exclusive.

It would be of great advantage for the medusae if they were able to determine the distance to the obstacle and not initiate the avoidance response until they were within a certain distance of the obstacle. This would ensure that they do not perform unnecessary responses interfering with their foraging behaviour in the light shafts between the roots (Stewart, 1996; Buskey, 2003). There are several ways to visually determine the distance to an object. The most exact are also the most advanced, using parameters such as depth of focus, relative movements and relative size combined with knowledge of absolute size (Land and Nilsson, 2012). These are all mechanisms demanding acute vision and much neural processing, which are resources not available to the jellyfish. But there are also more simple ways to estimate the distance to an object in the visual scene. The medusae can take advantage of the turbidity of the water in the mangrove swamp, with visibilities often down to about a metre (Garm et al., 2011). This means that, because of light absorption and scattering, the contrast of a given object decreases steeply with distance and that contrast, therefore, can be used as a semi-reliable measure of distance. This is supported by the underwater photo of the prop roots, which are the naturally occurring obstacles (Fig. 7). Even in this habitat with complex light distribution there is still an overall decrease in contrast between the roots and the surrounding water with distance. Taken together with our behavioural results, this shows that the obstacle avoidance response has a built in mechanism for distance detection, probably enabling effective foraging between the roots while still avoiding collisions. To our knowledge these are the first behavioural data pointing to the use of this mechanism for distance evaluation in any aquatic animal.

Interestingly, the contrast dependency varied with the orientation of the stripes. At the highest contrast, c=0.93, there was no difference between vertical, oblique and horizontal stripes, but differences were seen in the general distance to the wall and the avoidance rate at lower contrasts. The medusae responded the strongest to the vertical stripes, and already at c=0.27 this scene provoked more avoidances than any of the other visual scenes. At c=0.71, both the vertical stripes and the oblique stripes resulted in the medusae staying farther away from the wall than the horizontal stripes and the grey tone. Finally, the response to the horizontal stripes did not differ from that to the grey tones until we used the highest contrast. This shows that the more vertical an obstacle (or contrast line), the stronger the medusae will react to it, and this is in good concordance with the most often encountered obstacles, the prop roots. As seen in Fig. 7, the roots typically have an orientation varying between 45 deg oblique and vertical. Almost no horizontal lines are seen in the natural habitat. This is a good example of a matched filter (Wehner, 1987), where the visual system is matched to the natural visual scene.

There are two ways orientation filters can be implemented in the visual system of T. cystophora. One possibility is that all contrast lines are detected equally by the eye independent of orientation and then some contrast line orientations are filtered away by the central nervous system (CNS). The other possibility is that the retina is better at detecting some orientations of contrast lines than others. The latter is favoured by our behavioural data, as the medusae responded with equal strength to all three orientations at the highest contrast, indicating than any high contrast object can trigger the response. The separation of the different orientation could then be accomplished by directional contrast enhancement, such that vertical contrast lines would be enhanced and horizontal not. Support for such enhancement through lateral inhibition is offered by the presence of synapses between neighbouring photoreceptors in the lens eyes (Gray et al., 2009). It would require that the synapses specifically inhibit horizontal neighbours, though, and whether this is the case is not known.

The obstacle avoidance behaviour is probably controlled by the lower lens eyes (Garm et al., 2007b). From our earlier morphological and optical modelling of T. cystophora we know that both the upper and lower lens eyes allow for spatial resolution (Nilsson et al., 2005). The slit eyes might also acquire spatial information, but in the vertical plane only (Garm et al., 2008). In the case of the upper lens eyes it was shown that the animals do indeed use the spatial information when they navigate from the mangrove lagoon to their habitat between the prop roots (Garm et al., 2011). With the present work we have now shown that T. cystophora also requires spatial resolution in order to avoid obstacles. The data again point to the lower lens eyes controlling the behaviour. The upper lens eyes and the pit eyes point upwards, observing Snell's window (Garm et al., 2011), and do not see the underwater roots. The only other eyes observing the underwater world are the slit eyes, but they should preferably detect horizontal lines. Further, the minimum size of the obstacle on the retina able to evoke a response (15 deg) nicely matches the calculated resolution of the lower lens eye, varying between 10 and 20 deg depending on the area of the retina (Nilsson et al., 2005). The possible image formation in the slit eyes, seeing the world in horizontal bands, is intriguing and still awaits proof from behavioural experiments.

Cnidarians are often accused of being brainless (Wehner, 2005), but there is no doubt that at least hydromedusae and cubomedusae possess a CNS (Passano, 1976; Mackie, 2004; Skogh et al., 2006; Garm et al., 2007c). In cubomedusae, the CNS is composed of four parallel rhopalial nervous systems interconnected by a ring nerve (Satterlie, 2002; Garm et al., 2007c; Satterlie, 2011). Electrophysiological experiments and morphological examinations indicate that the visual processing mostly takes place in the rhopalial nervous systems (Satterlie and Nolen, 2001; Parkefelt et al., 2005; Garm and Mori, 2009; Parkefelt and Ekström, 2009). In the adult medusa, only approximately 1000 neurons are found here besides the photoreceptors (Skogh et al., 2006). This limited number of neurons has to process spatial information from at least the two lens eyes and possibly also the slit eyes. Considering the amount of neuronal power often dedicated to visual processing (Thorpe et al., 1996; Masland, 2012), this is somewhat surprising. Such a system stresses the need for the abovementioned matched filters, which ensures that irrelevant information is removed and that only the essential information is processed by the CNS. These filters are often already applied in the very periphery at the sensors (Barth, 2000), and the suggested lateral inhibition in the retina-enhancing vertical stripes would be a clear example of this. In vision, matched filters may result in so-called special-purpose eyes (Land and Nilsson, 2006), where the animal has several eye types, each specialized in taking up a narrow spectrum of information supporting one or a few behaviours only. The visual system of box jellyfish is a textbook example of special-purpose eyes, and this is probably one of the explanations for how they support an elaborate behavioural repertoire with their sparse CNS.

The authors are pleased with the fruitful and constructive discussions in the Sensory Biology Group at the University of Copenhagen.