Polarization Contrast Vision in Octopus

Unlike humans, numerous animals are differentially sensitive to the e-vector orientation of linearly polarized light; such a capability is known as ‘polarization sensitivity’. These animals include various terrestrial and marine invertebrates (e.g. Bandai et al. 1992; Nilsson et al. 1987; Phillipsborn and Labhart, 1990; Schwind, 1984, 1991; Snyder and Laughlin, 1975; Rossel, 1993; see reviews by Waterman, 1981, 1984) as well as several species of fish (Cameron and Pugh, 1991; Hawryshyn, 1992), amphibians (Auburn and Taylor, 1979) and birds (Martin, 1991; Phillips and Moore, 1992). Animals use this sensitivity for navigation (Wehner, 1976; Waterman, 1981, 1984; Rossel, 1993), for spatial orientation (Hawryshyn, 1992; Waterman, 1988) and for finding large bodies of water (Schwind, 1991). Can polarization sensitivity be used for object detection or identification? This is a fundamental question regarding visual perception. Objects such as fish (Cameron and Pugh, 1991; Hawryshyn, 1992) and plants (Land, 1993; Wehner and Bernard, 1993) reflect partially linearly polarized light in a significantly different pattern from their depolarized light appearance. However, no animal has yet been shown to detect or recognize objects using polarization sensitivity.

Like other cephalopods (Jander et al. 1963; Saidel et al. 1983), octopuses are sensitive to the orientation of the e-vector of linearly polarized light and thus possess polarization sensitivity (Moody, 1962; Moody and Parriss, 1960, 1961; Rowell and Wells, 1961). The anatomical basis for this sensitivity is the orthogonal arrangement of the microvilli in the photoreceptive rhabdoms throughout the retina (Moody, 1962; Moody and Parriss, 1961; Tasaki and Karita, 1966a,b). Nevertheless, the biological function of this sensitivity is as yet unknown.

While previous researchers (Moody, 1962; Moody and Parriss, 1960, 1961; Rowell and Wells, 1961) showed that octopuses are sensitive to the orientation of polarization of linearly polarized light coming from a single source, they did not determine whether these animals can detect and analyze polarization patterns within a single imaged object. The ability to analyze such polarization patterns is essential if polarization vision is to be used for object detection or recognition.

Octopuses can readily be trained to choose a target marked with a particular pattern (shape, brightness contrast, etc.) in preference to another, and are able to transfer this learning to select a target similar but not identical to the learned one (Wells, 1978). In this study, we made use of such learning to examine the ability of octopuses to discriminate between targets solely on the basis of polarization vision. This was performed by training them to select between targets differing only in the presence or absence of a polarization contrast within them.

Three questions were asked. (a) Can octopuses detect the contrast produced by variations in the orientation of polarization within an object? (b) Can octopuses recognize specific targets on the basis of their polarization patterns? (c) What is the minimal polarization contrast that can be recognized by octopuses?

Octopuses of two species, Octopus vulgaris Lamarck and Octopus briareus Robson, were obtained from a commercial supplier. Individual octopuses were maintained in 58 cm×30 cm×30 cm tanks, visually separated from each other, supplied with running, ultraviolet-sterilized sea water at 21 °C. Tanks were exposed to a 13 h:11 h light:dark cycle, provided through dim fluorescent illumination from above. As animals arrived in the laboratory at different times, individuals experienced varying numbers of target presentations.

Behavioral experiments involving training present a dilemma: because octopuses are very curious animals, they will behave ‘incorrectly’ every once in a while. Indeed, even after prolonged training, there may be a residual error rate of about 20 % (Moody and Parris, 1961, 1962; Wells, 1978). Additionally, the response to training varies among individuals (Mather and Anderson, 1993) and the use of a strong punishment may cause them not to react at all. We chose not to punish negative behavior and therefore to accept a certain level of negative results.

Three Octopus vulgaris and one Octopus briareus were trained to select targets that lacked a polarization contrast pattern, and four O. vulgaris were trained to select contrast-containing targets. The targets presented were 4 cm×4 cm chromatically neutral, linearly polarizing dichroic sheet filters (Polaroid, HN38S), held in a probe made of two pieces of clear Plexiglas (which did not measurably affect the light polarization pattern) attached to a gray rod 40–50 cm long. A wax-paper depolarizer was attached to the back of each polarizing filter. A circle, 2 cm in diameter, was cut from the middle of each filter and was replaced either at the same orientation (creating no polarization contrast) or after rotation by 90 ° (creating a contrast of polarized light). This contrast could not be detected by a human eye unless the target was viewed through a linearly polarizing filter (Fig. 1). In order to prevent the animals from learning false cues, and to ensure that they reacted only to the polarization contrast, target position (left/right), the orientations (horizontal/vertical) of the center and background parts of the filter, and the probe in which each filter was presented were varied randomly, using a total of eight probes and more than 50 polarization filters, throughout training and experiments. Additionally, probes were washed after each presentation in 70 % alcohol and afterwards in fresh water to remove any odor from previous presentations and to prevent transmission of naturally occurring diseases. Tanks were illuminated from above with dim fluorescent light. Prior to experimental sessions, a white plastic sheet was placed outside the front of the aquarium, diffusing incoming light, and an incandescent lamp was lit facing the front of the aquarium, where the targets were presented. Measurements using an imaging polarized light analyzer (Cronin et al. 1994; Shashar et al. 1995) revealed no residual polarization in the incoming light or in the underwater light field.

Since each aquarium was 58 cm in length, the central part of each target, as viewed by the octopus, was at least 4 ° in diameter, and since the motion of the animals within the tanks was not restricted, this value was usually even greater (much larger than the 10–65’ minimal visual acuity angle typical of octopuses; Sutherland, 1963; Packard, 1969; Muntz and Gwyther, 1988).

The training and testing procedures were as follows. Each session consisted of five presentations separated by 5–10 min. For the first three training sessions, each animal was presented with only the positive target, which had a small piece of shrimp attached to its back as a reward. The subsequent three training sessions also involved only the positive target; however, the reward was not present on the target, but was given after the animal had selected the target. During all further tests, the animal was presented with both positive and negative targets simultaneously. If the animal selected the positive target, its choice was immediately reinforced with food; there was no punishment for a negative choice.

The selected target was defined as the first target contacted by the octopus. This definition was necessary since octopuses would occasionally persist in touching the other target. If a choice was not made within 5 min, both targets were removed. As controls, 38 presentations with both targets placed backwards in the probes (with the depolarizer facing the animals and the polarizing filter facing the aquarium wall) were randomly included among the test presentations.

To investigate whether the octopuses could transfer their learning to new situations, and as an additional control, two animals from each group were presented with targets containing filters polarizing light on 45 °/135 ° axes, included randomly among horizontal/vertical targets. These targets also either contained or lacked a pattern of polarization contrast, and a human eye could not distinguish them from the ‘regular’ targets without the aid of an additional linear polarizing filter. In the final set of experiments, we investigated the minimal detectable difference in polarization angle between the center and the background of the target. Three animals, from the group trained to choose contrast-containing targets, were presented with pairs of targets in which one did not contain a polarization contrast, while in the other the e-vector orientation of the target’s center differed from that of the background by various angles (10 °, 20 °, 30 ° or 45 °), and their choices were recorded. Presentations of the different angles (24 presentations for each angle) and the orientation of the background of both targets were randomized. In each session, two presentations involved targets with a 90 ° contrast and three included targets with smaller angles of contrast (10 °, 20 °, 30 ° or 45 °), selected randomly.

Results were grouped according to the task under investigation and were analyzed, after excluding the cases where the animals did not select a target within 5 min, using a x2 goodness-to-fit test for two cells with equal expected frequency (Gibbons, 1976). For example, when examining the ability of octopuses to detect a polarization pattern on the 45 °/135 ° axis, results from all animals presented with this task (animals trained either to prefer or to reject a polarization contrast) were grouped, cases where animals did not react to the targets were omitted, and then the results were analyzed using the 𝒳² test.

Octopuses could distinguish between targets on the basis of the presence or absence of polarization contrast. The overall results were: correct choices 374, wrong choices 222, no action 60, P<0.001 (Fig. 2). Therefore, we conclude that octopuses can detect a pattern of polarization contrast within a single object.

The performance of the group trained to choose contrast-containing targets was not significantly different from that of the group trained to prefer targets without a polarization contrast (P=0.18, Fig. 2). In these experiments, the octopuses did not consistently prefer any particular orientation of polarization in either part (center or background) of the targets presented, nor did they prefer a specific combination of center and background within a single target (Fig. 3). Additionally, animals did not discriminate between targets when the depolarizer side faced them (presented to five animals; correct choices 14, wrong choices 18, no action 6, P=0.335). These results strongly suggest that discrimination between targets was based solely on the presence or absence of polarization contrast within each target. When targets which they had not previously encountered (with a polarization pattern on an axis of 45 °/135 °) were presented to two animals from each group, they transferred their learning and continued to select the target type (patterned or patternless) to which they were trained (correct choices 14, wrong choices 4, no action 0, P<0.02; Fig. 3). This transfer of learning indicates that octopuses can recognize objects on the basis of their polarization pattern.

Octopuses were able to transfer their learning to situations in which the contrast angle was less than 90 °. In tests to determine the minimal detectable angle of contrast, octopuses could recognize a polarization contrast pattern at an e-vector difference as small as 20 ° (P<0.001 for the presentations of 20 °, 30 ° and 45 ° contrast combined, P⩽0.07 for each of these angles separately), but apparently could not recognize a 10 ° difference (Table 1).

The angular size of the targets’ central region was much larger than the minimum visual acuity angle of the octopuses (Sutherland, 1963; Packard, 1969; Muntz and Gwyther, 1988). Therefore, we infer that the octopuses were capable of detecting the polarization variation within each target. Single photoreceptors of octopuses are unable to discriminate between lights polarized at 45 ° to the orientation of microvilli in the retina (Tasaki and Karita, 1961a,b). However, intact animals can discriminate behaviorally between lights polarized at 45 ° and 135 ° (Moody and Parris, 1961), and in our experiments octopuses were able to detect polarization contrast patterns on the 45 °/135 ° as well as on the horizontal/vertical axes. Furthermore, they recognized a polarization contrast angle of 45 ° (on a vertical or horizontal background). We suggest that, in intact animals, discrimination of a 45 ° variation in polarization within a single target may be possible by using irregularities known to occur in the retinal array of rhabdomeres (Tasaki and Karita, 1961b) or by taking advantage of movements of the head or eyes.

In addition to transferring their learning of polarization contrast patterns on vertical/horizontal axes to the 45 °/135 ° axes, octopuses recognized targets containing polarization contrasts at other e-vector angles, varying from 20 ° to 45 °, which they had not encountered before. We therefore conclude that their choices were based solely on the presence or absence of differences in polarization between the center and outer parts of the targets. This is the first demonstration of visual discrimination that is based solely on polarized light patterns. Object detection based on polarization recognition may be the primary function of polarization vision in octopuses. Such a means of contrast enhancement could be used to detect and identify prey or to communicate with other octopuses. For example, our observations show that several species of crustaceans, which are preyed upon by octopuses, reflect light that is strongly polarized. The underwater polarized light field is only partially described (but see Waterman and Westell, 1956; Tyler, 1963; Horvath and Varju, 1995), and study of the polarization of light reflected from objects in the marine environment is just beginning (e.g. Cariou et al. 1990). Nevertheless, polarization-based object recognition could increase the range of detection compared with the use of unpolarized light cues alone (Lythgoe and Hemming, 1967). Polarization vision, like color vision, may also be used for enhancement of borders and for breaking camouflage (Bernard and Wehner, 1977). Under water, the wavelength spectrum reflected from an object varies with depth, while the reflected e-vector orientation remains relatively constant, and the percentage polarization of the background scattered light is high even at depths exceeding 50 m (Waterman, 1955; Tyler, 1963). Therefore, polarization sensitivity may serve as a reliable means of object identification.

Our experiments demonstrate one expression of sensitivity to linearly polarized light, namely sensitivity to features in the orientation of polarization. Polarization sensitivity, or polarization vision, is likely also to include the ability to discriminate between polarized and depolarized light, between differences in the partial polarization of the light arriving from different objects and, possibly, even the ability to integrate and compare these aspects of polarization.

We predict that the use of polarization sensitivity for object detection and identification will be found in other species of marine animals and possibly in terrestrial species as well.

We thank J. D. Robertson of the Department of Anatomy, Duke University and the Duke University Marine Laboratory Beaufort, NC, USA, for the use of his octopus maintenance and training facilities, and D. R. Murlless for his invaluable help throughout this study. G. D. Bernard, R. L. Caldwell, M. F. Land, J. B. Messenger and two anonymous referees all provided helpful comments on the manuscript. This study was supported by NSF grants no. BNS-8917183 and BIR-9317927.