ORIGINAL PAPER

J. D. Pettigrew á S. P. Collin á K. Fritsches

Prey capture and accommodation in the sandlance,Limnichthyes fasciatus (Creediidae; Teleostei)

Accepted: 8 December 1999

Abstract The eyes of the sandlance, Limnichthyes fas-ciatus (Creediidae, Teleostei) move independently andpossess a refractive cornea, a convexiclivate fovea and anon-spherical lens giving rise to a wide separation of thenodal point from the axis of rotation of the eye muchlike that of a chameleon. To investigate this apparentconvergence of the visual optics in these phylogeneticallydisparate species, we examine feeding behaviour andaccommodation in the sandlance with special referenceto the possibility that sandlances use accommodation asa depth cue to judge strike length. Frame-by-frameanalysis of over 2000 strikes show a 100% success rate.Explosive strikes are completed in 50 ms over prey dis-tances of four body lengths. Close-up video con®rmsthat successful strikes can be initiated monocularly (bothnormally and after monocular occlusion) showing thatbinocular cues are not necessary to judge the length of astrike. Additional means of judging prey distance mayalso be derived from parallax information generated byrotation of the eye as suggested for chameleons. Usingphotorefraction on anaesthetised sandlances, accom-modative changes were induced with acetylcholine andfound to range between 120 D and 180 D at a speed of600±720 D s)1. The large range of accommodation(25% of the total power) is also thought to be mediatedby corneal accommodation where the contraction of aunique cornealis muscle acts to change the corneal cur-vatures.

Key words Accommodation á Visual optics áCornealis muscle á Feeding behaviour á Teleost eye

Introduction

Recent work has drawn attention to a remarkable evo-lutionary convergence between a specialised group ofsquamate reptiles and a likewise-specialised group ofteleost ®sh (Collin and Collin 1988a, b, c; Pettigrew andCollin 1995; Collin and Collin 1997; Pettigrew et al.1999). Despite the contrast between their terrestrial andmarine habitats, chameleons and sandlances share anumber of similarities in their strikingly derived life-styles. These similarities include a furtive, lightning-fastmode of prey capture, highly independent perioscopicmovements of camou¯aged eyes, and unusual visualoptics. The optical similarities are perhaps the most bi-zarre and include the following: (1) a reduction in thecontribution of the lens to the total power of the eye,(2) a wide separation of the nodal point from the axis ofrotation of the eye, (3) a deep-pit fovea, and (4) highlydeveloped accommodation.

While the evolutionary pressures that have given riseto these convergent similarities are not yet clear, it isapparent that further study of these two fascinatinggroups of vertebrates could reveal more details of theconstraints underlying the evolution of complex visualand behavioural traits. Indeed, as investigation hasproceeded, the strengths of the similarities between thetwo parallel sets of adaptations have become increas-ingly apparent. The comparison has even suggestedwhat appears to be a completely new principle for eye-design, where a predator can gain monocular movementparallax information about its prey from eye rotationinstead of from the usual eye translation (Land 1995,1999). Since eye translation requires head or bodymovement that would be revealing to prey, this princi-ple of eye design could be highly relevant to furtive,``sit-and-wait'' predators like the sandlance and cha-meleon.

J Comp Physiol A (2000) 186: 247±260 Ó Springer-Verlag 2000

J. D. Pettigrew (&) á K. FritschesVision, Touch and Hearing Research Centre,Department of Physiology and PharmacologyUniversity of Queensland, Brisbane 4072,Queensland, Australiae-mail: j.pettigrew@vthrc.uq.edu.auTel.: +61-7-3365-3842; Fax: +61-7-3365-4522

S. P. CollinDepartment of Zoology,University of Western Australia, Nedlands 6907,Western Australia, Australia

In the present study, we have concentrated on thedetails of prey capture by the sandlance, with specialreference to the possibility that sandlances use accom-modation as a depth cue to judge strike length as inchameleons. While we have not been able to carry outexperiments with spectacle lenses to support this hy-pothesis in the de®nitive way that Harkness (1977) usedin the larger, more amenable chameleon, we have suc-ceeded in providing a number of lines of evidence thatsupport a similar, monocular mode of depth judgementin sandlances and chameleons.

Materials and methods

Sandlances were taken from Wistari Reef (on the Great BarrierReef, Queensland) and Rottnest Island (Western Australia) aspreviously described (Pettigrew and Collin 1995). The slightlylarger cold-water specimens from Rottnest Island were helpful inthose experiments where the size of the eye was limiting, but we stilldid not succeed in applying any contact lenses or spectacles fordepth corrections to these small eyes.

Prey

Zooplankton was collected from waters around Heron Island andWistari Reef with a 100-lm seine net trolled at depths of 1±5 m.This prey was fed to the sandlances using a 50-ml syringe connectedto 2-mm-diameter polythene tubing. Sandlances were housed inseawater display tanks (5 cm ´ 30 cm ´ 50 cm) illuminated fromabove with a ®bre-optic light source to mimic photopic conditionsand furnished with a 2-cm-deep layer of coarse coral sand takenfrom the native habitat at the time of capture. A black velvet back-drop provided dark-®eld conditions for observing and recordingprey movements and sandlance strikes. Within 2 days of capture,sandlances had associated the switching of the ®bre-optic lightsource with feeding and repositioned themselves near the surface ofthe sand, eyes exposed, so that strikes could be elicited visually bymoving prey in the water column. Under normal conditions, preymoved around the tank at speeds from 2±10 cm s)1, propelled bythe circulation system in a similar way to the movements we ob-served in the natural habitat. In some experiments, the seawatercirculation was turned o� to observe capture of prey whosemovement was self-produced. Since the sandlances were alwayssuccessful in their feeding strikes, whether prey were carried in thewater current or self-propelled, we usually left the circulation goingduring observations.

A wide variety of zooplankton was o�ered to the sandlances.Preferred prey were around 1 mm in diameter. We observedmarked preferences for prey type. A dramatic example concernedsome brachyuran larvae that were proli®c in one troll. When ®rstpresented, the brachyurans (around 1 mm in diameter), like thecalanoid copepods also present, were taken avidly by hungrysandlances. Feeding ceased, however, when each sandlance hadtaken a brachyuran. Inspection with a dissecting microscope re-vealed that there was some di�culty in ingesting the spine-covered,tough-shelled brachyurans, which were henceforth shunned by thesandlance in favour of the similar-sized, but obviously distin-guishable, copepods.

Single-frame analysis of strike culmination

A high shutter speed (1/2000 s) on the video camera enabled eachcomponent in a strike sequence to be captured as a sharp frame,but the slow frame rate (25 Hz) meant that only random fragmentsof a given strike sequence could be captured. To assemble a moredetailed quasi-sequence, we used the prey's position on the frame

(always clearly highlighted in dark®eld) to normalise and collectedmany examples of frames from di�erent strikes showing thesandlance in close ®nal approach to the prey. While an assumptionof constant speed at this part of the strike would be very rough,given the weak relation between strike length and duration that wefound, this procedure allowed the rough ordering of frames fromthe ®nal stages of di�erent strikes to show the erection of themaxillary cage and suction of prey into the cage. These framescould then be compared with strobe photographs (1/20,000 s)triggered by striking sandlance that interrupted an infra-red beam.

Monocular occlusion/ablation

The anterior chamber of one eye was ablated with a battery-operated portable diathermy device (FST) while the sandlance wasunder methane sulfonate anaesthesia (1:10,000). Glued occludersdid not give satisfactory results because the mechanical stimulus tothe eye seemed to disturb the sandlance. We have observed sand-lances repeatedly making the same saccadic eye movement in anapparent attempt to move a sand grain that is touching the eyeduring the normal situation when the sandlance is partially buriedunder sand grains whose dimensions approximate its own eyes, oreven head. The response to the mechanical stimulus presented byan occluder glued to the eye reminded us of this behaviour to shiftaway the source of mechanical stimulation to the eye from the sandgrain. In initial attempts to provide an alternative source of oc-clusion, we tried to damage the anterior part of the eye so that clearvision would be prevented, but found that all attempts to intervenesurgically resulted in a complete loss of the anterior chamber inthese tiny eyes. All layers of the cornea were lost and sometimes thecrystalline lens was also lost.

We found that the anterior chamber of the eye regeneratescompletely after 5 days (presumably from tiny epithelial remnants).This was a great surprise to us, as we could verify in the dissectingmicroscope that a new cornea, including the corneal lenticle, couldbe regenerated over a few days. In cases where the ablations weremore extensive and also involved the destruction of the crystallinelens, we observed the development of a thin membrane that cov-ered the open side of the optic cup. Further description of theextraordinary powers of regeneration in this species will be thesubject of a separate study (S. P. Collin and J. D. Pettigrew, un-published observations). Because of the rapid recovery of the an-terior chamber, we monitored the ablated eye daily and used dataonly from the sandlances that had not yet recovered. For thisreason, data are derived from the ®rst 2 days after monoculardeprivation.

Photorefraction

A 4-mm-square piece of front-silvered mirror was positioned usinga magnetically-mounted snake arm just in front of the objectivelens of the operating microscope (400-mm Wild-Leitz M-400) sothat it re¯ected a beam from a ®bre-optic light source toward thesandlance being viewed. The front-silvered mirror was positionedslightly away from the optical axis of the microscope objective,near the place where the illuminating light beam emanated, so thatthere was no apparent obstruction to the ®eld of view. When thebeam from the mirror intersected a sandlance eye with the appro-priate eccentric orientation, there was a clearly visible light-re¯exfrom part of the pupil. This re¯ex changed position systematicallyas lenses were introduced into the optical path as described in otherspecies (Schae�el and Howland 1987). There were also changes inthe amount by which the re¯ex ®lled the pupil as the eye changed itsorientation to the beam, suggesting that di�erent eccentricities inthe eye might have di�erent refractive states. For this reason,photorefraction observations were restricted to series of frameswhen the eye did not change its orientation. Because the orientationof the eye was critical in obtaining a good re¯ex, some experiments(e.g. the whole ®sh pharmacology experiment; see below) wereconducted with the anaesthetised sandlance inside a Petri dish-cover slip assemblage that could be rotated around all three axes.

248

In this way, we could rotate the eye until a bright re¯ex wasobtained.

We placed 20-D, 30-D and 50-D positive lenses in the opticalpath and observed/recorded the resulting changes in the proportionof the pupil taken up by the light re¯ex. There was a rough linearrelationship, as described in other small eyes, between the power ofthe introduced lens and the proportion of the pupil occupied by there¯ex (Schae�el and Howland 1987). This enabled us to calibratethe relationship between changes in the position of the edge of there¯ex and dioptric changes in the eye. In such a small eye there willbe a signi®cant refractive artefact (Glickstein and Millodot 1970).We have not tried to estimate the absolute magnitude of this ar-tefact, nor do we try to make any estimates of the resting, DC levelsof accommodation (Andison and Sivak 1996). Instead we haveestimated the change in refractive state during accommodation andthen tried to interpret that decrement onto some reasonable as-sumptions about resting accommodative state. For example,sandlance accommodation involves a shift from resting myopiatoward emmetropia, just as described for accommodation in otherteleosts (Tamura and Wisby 1963; Sivak and Howland 1973;Andison and Sivak 1996). Estimating the absolute degree of restingmyopia can therefore be based on the assumption that accommo-dation ``moves the focal plane outwards'' toward emmetropia andthat the decrement in refractive state observed is fairly close inmagnitude to that of the amount of resting myopia (in other words,one would expect that teleost accommodation would achieve em-metropia, to within a dioptre or two, lest objects at distance not bevisible at all, with the corollary that the maximal decrement inpower achieved by accommodation is close to the amount ofresting myopia ± see Results and Discussion on this point).

During the application of acetylcholine, we observed a dramaticincrease in the proportion of the pupil occupied by the photore-fractive light re¯ex, if care was taken to achieve the appropriateorientation of the ®sh by rotating the Petri-dish-cover slip assem-bly, and if the ®sh were su�ciently well anaesthetised that the eyedid not change position. There was a rough linear relationship, asdescribed in other small eyes, between the power of the introducedlens and the proportion of the pupil occluded by the re¯ex (How-land and Schae�el 1987). The magnitude of the change producedby the 50-D lens was ®ve times that produced by the 10-D lens androughly one-third of the maximal change produced by the appli-cation of acetylcholine. Thus, we were able to calibrate the rela-tionship between changes in the position of the edge of the re¯exand dioptric changes in the eye.

Keratometry

Three ®bre-optic light sources were placed as close to the ®sh aspossible, separated by an angular subtense of around 40°. Wevisualised the three corneal re¯exes (Pettigrew and Collin 1995) in adissecting scope. In practice it was easier to monitor the ®rst andsecond re¯exes (from the front of the corneal lenticle), and some-times we could obtain data from the third (from the back of thelenticle and anterior to the iridescent layer). We monitored theseparation of three re¯exes during eye movements prior to strikes.We estimated the change in separation of the three images as achange in magni®cation on the video image.

It was also possible to monitor the curvature of the pro®le of thelenticle directly in favourable oblique pro®le views of the eye (seePettigrew and Collin 1995). By looking at previous videotapes, wewere able to ®nd a few cases of this favourable view and so verifychanges in the front curvature of the lenticle during feeding. Theback surface could not be visualised in this way, but sometimes wesaw apparent changes in the position of the lenticle which appearedto move backward during accommodation. We can infer that anincrease in the curvature of the posterior surface of the lenticle takesplace as it is pressed against the crystalline lens at this time, since wesometimes saw close apposition in histological material.

The importance of making these observations on the livingcornea only became apparent after we discovered the cornealismuscle, at the conclusion of the study. We have not therefore had

the opportunity to fully document the changes in corneal shapeduring accommodation. This will be the subject of a full study in itsown right. In the present study, we present evidence supporting themajor role of the cornea in accommodation. This evidence alsosuggests how corneal accommodation may be achieved.

Whole-®sh pharmacology

After being anaesthetised with methane sulfonate (MS 222,1:10,000), the sandlance was placed in a plasticine cradle stuck to aPetri dish that could be rotated around three axes. A cover-slipmounted on the Petri dish prevented spillage but was opened at thetop of the preparation for the introduction of drugs, once the po-sition had been found that gave a good re¯ex. Acetylcholinechloride, (2 ´ 10)2 mmol l)1), was dropped with a Pasteur pipette(drop size approximately 50 ll) directly onto the shallow sea waterjust covering the head.

These experiments were carried out in Brisbane, on the smaller®sh from Heron Island, whose eyes were only 1 mm in axial lengthand total power around 900 D.

Light and electron microscopy

The head of two sandlances, overdosed in anaesthetic, were ®xed in4% glutaraldehyde in 0.067 mol l)1 sodium cacodylate bu�er andpost-®xed in osmium tetroxide in 0.1 mol l)1 sodium cacodylatebu�er. The eyes were then embedded in Araldite and both thickand thin sections prepared using an LKB ultramicrotome in orderto examine the morphology of the retractor lentis muscle and itsattachment to the non-spherical lens. Selected thick (1 lm) sectionswere stained with Richardson's stain, examined on a light micro-scope and photographed on Technical Pan ®lm (100 ASA). Thinsections for examination by transmission electron microscopy werestained with lead citrate and uranyl acetate and examined using aSiemen's Elmiskop 1A or a Phillips 410 electron microscope.

Scanning electron microscopy

Following anaesthesia with MS 222 (1:2000), one specimen was®xed in 4% glutaraldehyde in 0.1 mol l)1 cacodylate bu�er. Thewhole ®sh was then post®xed in 1% osmium tetroxide in0.1 mol l)1 cacodylate bu�er, dehydrated in a graded series of al-cohols and then dried in a Polaron critical point dryer. The ®sh wasmounted (dorsal side uppermost) onto a 10-mm aluminium stubwith double-sided tape. The tissue was coated with 12±15 nm ofgold-palladium in a Polaron sputter coater and placed in an oven at40 °C overnight before examination. The sandlance head wasphotographed using a Joel FSEM (®eld emission scanning electronmicroscope, Model 6300F) at an accelerating voltage of 3 kV.

Dimensions of the schematic eye

The dimensions and curvatures of the ocular components in theschematic eye were obtained from frozen sections (see Pettigrewand Collin 1995) cut in the axial (transverse) plane where the sec-tion with the greatest lens thickness was assumed to represent thegeometrical axis of the eye.

Results

General features of strikes

Field observations

The typical habitat where one could observe feedingsandlances (Fig. 1) was coarse coral bottom sand under

249

a slight overhanging shelf of rock or coral. Strikesobserved were around 20±30 cm long in any direction.The tiny prey could not be observed under these ®eldconditions, but we did not observe any features in the®eld that stood in obvious contrast to the observationsand recordings of strikes in the laboratory, as describedbelow (Figs. 2, 3).

Preferred prey

Preferred prey size was around 1 mm in diameter. Ifsmaller prey were unavailable, sandlance took 3- to4-mm Artemia, but required two or three separate bites.

Fig. 1 Scanning electron micrograph of a dorsal view of thesandlance, Limnichthyes fasciatus showing its turret-like eyes. Scalebar, 0.5 mm. (Modi®ed from Collin and Collin 1997)

Fig. 2 Composite of video frames (40 ms apart) showing two strikesequences (panels A±C and D±F) where the position of each ®sh hasbeen normalised with respect to the prey. Mouth opening andmaxillary cage erection precede the point of rendezvous with the preyby a small fraction of the frame interval (approximately 4 ms). Thesesequences were used to study accuracy of strike. When there wasambiguity about which prey was the object of the strike, the sequencecould be replayed from the time of strike to help eliminate falsetargets. The whole sequence is completed in a few video frames

250

The preferred prey size is presumably related to thesuction method of ingestion using a maxillary cage asdescribed below. There was evidence of ®ne visual dis-crimination in the form of a preference for calanoidcopepods over cyclopoid copepods (data not shown).The basis of this discrimination may be colour (cyclo-poids look blue to humans whereas calanoids look ru-fous) and sandlances have a variety of cone types (Collinand Collin 1988c). In addition, form may perhaps beinvolved in this discrimination, since cyclopoids havevery long antennae compared to calanoids.

Suction at the end of a strike

Even when the strike was not perfectly aimed at themoving prey, capture still ensued as the prey was suckedinto the mouth from as much as 3±4 mm o� the line ofstrike. The suction was provided by the erection of amaxillary cage whose rapidly increased volume movedseawater, plus prey, into the mouth. Single frame anal-ysis revealed that cage erection began near the end of thestrike trajectory, about 4 ms (around 4±10 mm) before

prey contact. At 25 frames s)1, one required manystrikes before one observed both the prey and the be-ginning of cage erection, but the whole sequence wasreconstructed from many strikes by using the position ofthe prey to normalise the data as described in Materialsand methods (Fig. 4).

Accuracy of strikes

We observed no misses in over 2000 strikes studied usingframe-by-frame video (Figs. 2±4). This is re¯ected in theclose relationship between strike length and prey dis-tance (Fig. 5) and between prey angle and strike angle(Fig. 6). On three occasions we saw strikes that ap-peared to be directed at a particular prey but were notconsummated by a jaw opening and cage erection. Wedid not regard these as misses, because the jaw remainedclosed throughout the strike. Since we also saw fastswimming movements along the length of the tank thatwere not apparently triggered by prey, perhaps thesethree episodes were also spontaneous.

Speed of strike

The speed of the strike is perhaps the most noteworthyfeature of the sandlance to ®rst-time observers. Strikeswere explosive and could be completed in 50 ms over

Fig. 3 Composite of three ¯ash photographs (panels A±C) showingthe various stages of a strike where a ¯ash was triggered by thesandlance interrupting an infra-red beam set above the level ofthe sand. Panel B shows the sandlance just prior to the erection of themaxillary cage. Scale bar, 2 mm

251

prey distances of four body lengths (Figs. 2, 3, 5). Therelation between the duration of a strike and the lengthof the strike is shown in Fig. 7. Here it can be seen thatstrikes are common in the region of three body lengthswith durations of 100 ms but can cover a range fromone to eight body lengths with a duration from 50±200 ms. There is only a very weak correlation betweenstrike length and strike duration, suggesting a pre-programmed, ballistic kind of mechanism where thepower of the strike and its initial velocity are scaled upas a function of strike length. There is apparentlystrong pressure to keep the duration of the strike asshort as possible. This seems to be even greater for longstrikes.

Monocular vision is su�cient for accurate strikes

Monocular occlusion

Ten monocular sandlances all showed accurate strikes(Fig. 8). On the 1st day after operation where the an-terior chamber of one eye was ablated, no strikes were

observed when prey were placed in the tank, a ®ndingthat is reminiscent of the lack of activity seen with newlycaptive ®sh. On the 2nd day, there were a small numberof strikes, but these were all successful (Fig. 8). On the3rd day, strike frequency approached the normal valueseen in hungry ®sh (around 1 strike per ®sh per min).Strike accuracy was 100%. After the 3rd day, we did notcollect data from the operated ®sh because of regener-ation of the anterior chamber.

Normal monocular strikes

We analysed 50 videotaped strikes in close-up where theposition of each eye could clearly be seen during thephase just before the sandlance left the sand. The initialdata collection for this analysis often required reframingand refocusing of the ®sh before each trial (although thiswas made unnecessary on the frequent occasions whenthe sandlance returned to exactly the same place afterthe strike) and meant that the culmination of the strikecould not be observed. In six of these cases, we observedthe launching of a strike in the direction given by the®xating eye, before the non-®xating eye had beenbrought around into alignment with the ®xating eye. Inother words, the strike was being guided, in its initiallaunch, exclusively by information provided from the®xating eye. This is shown in Fig. 9. If one accepts ourproposition that the strike is open-loop and that visual

Fig. 4 Series of ¯ash photographs (panels A±F) showing the erectionof the maxillary cage during a strike. Note in panel E that followingthe retraction of the cage, water is expelled from the gills and themouth is closed. The sandlance then turns (panel F) in order to resumeits position in the sand. Scale bar, 2 mm

252

guidance during its execution plays a small or absentrole, as well as taking into account the extremely highaccuracy of strikes in general, the information guidingthe consummation of these six strikes must have beenmonocular. Binocular vision is clearly not necessary foran accurate strike by the sandlance.

Accommodation

Photorefraction

The photorefractive re¯ex observed in the pupil changedrapidly if there was an eye position change. This madeinterpretation of accommodative changes very di�cult,since even small changes in eye position resulted indramatic changes in the re¯ex that were unconnected toaccommodative changes. While this observation has theinteresting implication that di�erent eccentricities in thesandlance eye may be associated with di�erent refractive

Fig. 6 Angle of strike as a function of the angle of prey as seen by thesandlance. The convention is that straight ahead is 0° with increasingvalues representing increasing angles of elevation; 90° is the zenith and180° is directly behind the ®sh. The strikes were viewed from the sidein a narrow display tank, which con®ned the plane of the strike to theplane of the tank, which was orthogonal to the optical axis of thecamera. Note the wide range of strike angles observed, from almostdirectly ahead (~0°) to almost directly behind (140°)

Fig. 7 Time of a strike (estimated from the number of video frames)as a function of length of strike. Note that most strikes are brief(clustered around 100 ms), whatever the strike length

Fig. 8 Strike length as a function of prey distance in monocularsandlances. Note that the accuracy of strikes is una�ected by the lossof binocular vision

Fig. 5 Analysis of 128 sandlance strikes to show the accuracy ofestimation of strike length. Assuming that the strike is ballistic, withlittle or no time for corrective feedback once it is launched, the lengthof the strike is a fair indication of the sandlance's estimation of thedistance of the prey. With an allowance for overshoot, the estimatedstrike length is tightly correlated with prey distance

253

states, it is a complicated interpretation of thecontribution of accommodation. For this reason, wedeveloped an immobilised preparation where we couldstudy accommodation directly without the complicationof eye movements.

Hyperopic shift of sandlance accommodationunder acetylcholine

Direct application of a few drops of acetylcholine(2 ´ 10)2 mol l)1) to the eye of an anaesthetised sand-lance, immobilised in clay and positioned to optimise thephotorefractive re¯ex, caused a shift of the re¯ex in ahyperopic direction. This is shown in Fig. 10. This shiftcould be partly annulled by introducing a high-powerpositive lens (e.g. 20 D). By comparing the shift in there¯ex caused by the drug with the shifts caused by in-troducing lenses of di�erent power, we could calibratethe power of the accommodative change induced ac-cording to a principle already well-tested, althoughperhaps not in eyes as small as these (Schae�el andHowland 1987). By this means, we estimated that theaccommodative power induced by the application ofacetylcholine ranged between 120 D and 180 D in dif-ferent experiments (Table 1).

The direction of the change observed is in keepingwith other observations on teleost accommodation,which have also observed a resting myopia that is re-duced during accommodation (Tamura and Wisby 1963;Sivak and Howland 1973; Andison and Sivak 1996).

This accommodative arrangement may seem strangeto humans, whose eyes rest in focus with in®nity and ac-commodate for near targets, in contrast with the sand-lance which rests with the world in focus at around acentimetre (100 D), ``venturing out'' to focus at greaterdistances mostly during feeding. Nevertheless, the sand-lance may not need to have distant focus at the rest state,relying heavily on camou¯age, buried in the sand. Wehave observed a motionless sandlance watching a hermitcrab crawl across its body a centimetre away, ®sh keenlyattending to the near ®eld in this case! (if we can judgefrom the series of saccades made to ®xate the crab). Wehave also observed prey capture at distances of a centi-metre, as well as intraspeci®c interactions at these dis-tances (the ®sh is only 2 cm long). It is di�cult for ahuman (acuity of 60 cycles/deg) to see a motionlesssandlance on the sandy bottom, so teleost predators withacuities less than humans are unlikely to be able to initiatean attack on a sandlance except from very close range.

Fig. 9 Video analysis of the initiation of a strike in which only oneeye is ®xating the prey. Note that the sandlance's left eye has notturned toward the prey (which is out of the frame and being ®xated byits right eye) until after the strike has been initiated (panel D). Whilemost strikes were normally initiated when both eyes had moved to®xate prey, the example shown here was not rare. Clearly, sandlancestrikes (which never miss) can be initiated without the need forbinocular information

b

254

Speed of accommodation

The complete range of accommodation, as measuredfrom photorefraction, was 120±180 D, and this could beaccomplished in 0.2±0.25 s (see Fig. 10). This gives aspeed of accommodation of 600±720 D s)1.

Corneal changes during accommodation

The anterior corneal light re¯exes (1 and 2) of multiplesources were observed to move apart during feeding,with a change in magni®cation that was as much as 30%.This magnitude and direction of change (i.e. a reductionin power) was con®rmed in favourable side views of thecorneal lenticle whose anterior surface increased itsradius of curvature from the resting state by 30±40%.There may also have been changes in the posterior sur-face of the corneal lenticle, although these were muchharder to observe as they were occluded by the iris. Thechanges in the posterior surface of the lenticle tended tohave an opposite sign from those of the anterior surface,changing from a relatively ¯at surface to one in whichthere was increased curvature with the concavity towardthe lens with which the lenticle seems to make closecontact during accommodation. The re¯exes of multiplesources from the posterior surface of the lenticle alsoincreased their separation during accommodation. Themagnitude of this change may have been as great as20%. There may also have been a posterior displacementof the lenticle during accommodation, although this washard to distinguish from the ¯attening.

The total power of the sandlance's corneal lenticle isaround 200±300 D or more, depending on the size of theeye (Pettigrew and Collin 1995). It would be more than300 D in the small eyes (from Heron Island sandlances)used in these experiments on corneal re¯exes and ace-tylcholine. If we ignore the possible changes in lenticledisplacement during accommodation (which would addfurther to the e�ect of decreasing the overall power ofthe eye), one can see that the combined e�ect of curva-ture changes at the back and front of the lenticle will beto reduce its power by around 50%, or at least 100±150 D.Given the rough-scale nature of these measurements andobservations, and in lieu of the more precise ones, it ispossible to see that corneal changes may account for theaccommodative changes (120±180 D) that were ob-served after stimulation with acetylcholine. This cornealcontribution contrasts with any mechanism that involvescrystalline lens displacement. Quite apart from theproblem of very slow speed of the smooth muscle of theretractor lentis, there is not enough possible range ofmovement to produce such an accommodative response,as can readily be found if one tries to come up with a

Fig. 10 Photorefraction. At rest (panel A) the sandlance is highlymyopic and possesses no photorefractive re¯ex. After the applicationof acetylcholine, the re¯ex becomes visible (panel B), showing adecrease in myopia. At the height of the maximal e�ect of the drug(panel C), the re¯ex has moved even further in the emmetropicdirection across the pupil (i.e. toward the same side as theasymmetrical illuminating beam). This accommodative change, frommyopia at rest, to hyperopia in the accommodated state, is theopposite of what is observed in most land vertebrates, but conforms towhat has been observed in other ®sh. By comparing the magnitude ofthe shift in the light re¯ex induced by a negative lens with the shift inthe re¯ex seen in this experiment, we estimate that the total range ofaccommodation in the sandlance is approximately 180 D

b

255

schematic eye to show accommodation induced by lensmovement, as opposed to corneal changes.

Accommodative apparatus

Retractor lentis muscle

All teleosts use a retractor lentis muscle to move the lensin the interests of accommodation. Our initial interestwas, therefore, focused exclusively on this muscle in thesandlance. The retractor lentis in sandlance arises fromthe choroidal ®ssure, at the ora terminalis, in a placewhere the retina is furthest from the lens (Fig. 11A). Theretractor lentis inserts into the postero-lateral edge of thelens, on the ventral margin of the steepest curvature.Shortening of the retractor lentis muscle would thereforeproduce movement of the lens that was largely directedaway from the optical axis of the eye, with only a smallcomponent of motion along the axis where it would beuseful in accommodation.

We found that the retractor lentis was composed of aregular array of smooth muscle cells aligned along thelong axis of the muscle. At the electron microscope level,these smooth muscle cells were loaded with mitochon-dria. The intracellular space available for contractile®laments, after mitochondria and nuclei were subtracted,was less than 20%. Between the smooth muscle cells wasa thin layer of electron-dense cytoplasm, also with manydamaged mitochondria but not so well-packed as themitochondria in the smooth muscle cells (Fig. 11B). Wecould not determine the origin or a�nities of this cel-lular material.

Cornealis muscle

We missed this muscle completely at ®rst. It was foundonly after a careful search that was provoked by the re-alisation that there were too many awkward facts aboutthe retractor lentis that made it a poor candidate to ex-plain the extraordinary range and speed of the sandlance'saccommodation. These awkward facts concerning theretractor lentis included its smooth muscle (too slow toexplain the 700 D s)1 accommodation), its inappropriate

axis of shortening compared with the axis required forlens movement to achieve accommodation and the di�-culties we experienced in trying to draw a schema wherethe ¯attened lens, with its reduced power, was pulled farenough through the eye by this tiny misaligned muscle toproduce around 120±180 D of accommodation.

In contrast to the retractor lentis, the cornealis mus-cle has all the features that would be required to explainaccommodation in sandlance, not withstanding oursurprise (and delay) at providing the ®rst description ofa corneal muscle in a teleost (Figs. 11C, D, 12).

The cornealis muscle is striated (Fig. 11D). It arisesfrom the external surface of the sclera in two broadsheets overlying each other in the ``notch'' of the limbusthat marks the beginning of the choroidal ®ssure(Fig. 12). It has a tendinous insertion into the lateraledge of the corneal stroma at the point where the cornealstroma and the collagenous stroma of the sclera merge.Its axis of shortening is therefore aligned with the re-tractor lentis that lies beneath it in the globe. In contrastto the retractor lentis, whose axis seems inappropriatefor accommodation, the axis of shortening of the cor-nealis muscle would act to ¯atten the corneal curvatureand to push backwards the cornea and the lens (if this isin contact with the cornea, as we have always observedin frozen sections (see Pettigrew and Collin 1995). Inother words, putting aside the fact that corneal accom-modation is so far unknown in teleosts, the cornealismuscle seems ideally placed to subserve accommodationin the sandlance.

Discussion

The present study establishes that sandlance strikes canbe accurately guided by one eye. Not only is the preci-sion of a strike una�ected by experimental eliminationof visual input from one eye, but many normal strikesare initiated by one eye before binocular ®xation can beachieved. We think that the speed of the strike makes itunlikely that its direction and length can be under closedloop visual feedback. This presumption is supported bythe ®nding that strikes are initiated with the correctheading, so the fact that the normal strikes can be

Table 1 Range of accommodation in teleosts and a chameleon. Note that the accommodative range as a percentage of total power inteleosts is between 4% and 6.7% except for the sandlance, which has an increased range (25%) like the chameleon (45%)

Species Maximumaccommodativerange (D)

Total powerof the eye (D)

Maximum accommodativerange as a percentageof total power (%)

Reference

Sail®sh, Istiophorus albicans 2.6 54.05 4.8 Tamura and Wisby (1963)Greater amberjack, Seriola dumerili 3.1 46.51 6.7 Tamura and Wisby (1963)Dolphin®sh, Coryphaena hippurus 3.4 84.0 4.0 Tamura and Wisby (1963)Great barracuda, Sphyraena barracuda 3.1 56.2 5.5 Tamura and Wisby (1963)Black®n tuna, Thunnus atlanticus 2.4 44.8 5.4 Tamura and Wisby (1963)Skipjack tuna, Euthynnus alletteratus 4.3 73.0 5.9 Tamura and Wisby (1963)Chameleon, Chamaeleo calyptratus 45 100.0 45 Ott et al. (1998)Sandlance, Limnichthyes fasciatus 180 900.0 20 This studyGar®sh, Lepisosteus osseus oxyurus 6.5 121.4 5.4 Sivak and Woo (1975)

256

initiated using monocular cues alone leads us to inferthat binocular cues are not necessary to judge the lengthof the strike.

What monocular cues are available?

The well-known monocular distance cues are size con-stancy, perspective, interposition, texture gradients,monocular movement parallax and accommodation.With respect to size constancy, we have no direct ex-perimental evidence to support the idea that sandlancescan judge depth monocularly by converting the angularsize of familiar objects to distance using a stored repre-sentation of actual physical size. We did notice that theyquickly learnt to discriminate inedible plankton (e.g.brachyuran larvae) from edible prey of the same size (e.g.

calanoid copepods). This raises the possibility that thesize of familiar objects is learned and is used to judgedistance as a consequence of its diminution of angularsize. Can perspective, interposition, texture gradients andelevation provide monocular cues? None of these fourmechanisms are relevant to the capture of a small iso-lated prey moving in the water column. A tiny, circum-scribed prey gives no perspective information, nor is itvery likely to interpose itself between other prey on manyoccasions. While the texture gradient provided by thebottom sand could provide some depth information, thesandlance's vantage point is hardly appropriate to viewthe gradient, obscured by sand grains comparable in sizeto the sandlance's own head. Moreover, establishing theprey's location with respect to the texture gradient on thebottom would require another source of distance infor-mation, about the prey item itself, thus begging thequestion of any major role for the gradient. Elevation,shown to be useful in toads (Collett 1977), is not a usefuldistance cue to the sandlance since its prey is not con®nedto the ground plane.

Monocular movement parallax

Prey movement versus sandlance movement

Since the sandlance stays perfectly immobile duringhunting, except for its eyes, we thought at ®rst that

Fig. 11 A Light micrograph of the retractor lentis muscle (rlm) and itsattachment to the caudal pole of the non-spherical lens (le). Theproximal end of this smooth muscle is pigmented and inserts into thebase of the iris within the caudal embryonic ®ssure. Scale bar, 0.1 mm.B Electron micrograph of the smooth muscle of the retractor lentis.Note that the smooth muscle ®bres are of two types and are packedwith mitochondria. Scale bar, 0.1 lm. C Light micrograph of theanterior eye showing the unique corneal lenticle (cl) and the insertionof the striated ciliary muscle (cm) into the corneal stroma (arrowhead).Scale bar, 0.1 mm. D Higher magni®cation of the striated ciliarymuscle. Scale bar, 15 lm

257

monocular movement parallax could be ruled out as adistance cue. It is possible that a complicated cognitivestrategy, akin to monocular movement parallax, couldbe used to infer the relative positions in space of movingprey if they were all known to be swimming at the samevelocity. Closer prey would then be expected to movemore in relation to distant prey. Distance could then becomputed from the array of moving points if there weremany prey items and if the sandlance had some way oftaking into account the fact that all prey are not headingin the same direction. It seems unlikely that such amechanism would be accurate enough to explain the

Fig. 12 A Dorsal view of the sandlance head and eyes showing theposition of the ciliary muscle (pcm) which lies beneath the conjunctiva.The ciliary muscle sits in a groove formed by the embryonic ®ssurewith its position marked by a caudal notch in the pupil. Scale bar,0.5 mm. B Sagittal section of the eye showing the two accommodativemechanisms. Note the non-spherical lens and the posterior cornea areclosely opposed. Scale bar, 0.25 mm. C High magni®cation of thecaudal limbus showing the insertion of the ciliary muscle intothe corneal stroma which is continuous with the sclera (s) surroundingthe globe. Scale bar, 0.1 mm. be basal epithelial cells; c caudal; cecorneal epithelium; ci corneal iridophore; cl corneal lenticle; cm ciliarymuscle; co conjunctiva; cs corneal stroma; d dorsal; f fovea; icminsertion of ciliary muscle; il iridescent layer; ir iris; l lateral; le lens; ppigment of iris; r rostral; re retina; rlm retractor lentis muscle; soscleral ossicle; v ventral

258

perfect feeding success rate we see, since prey are notmoving in a single tangent plane and would thereforeinject incalculable components in the Z-axis. Moreover,the water column could add unpredictable velocitycomponents that were unrelated to distance. For allthese reasons, not to mention the somewhat limitedtelencephalic resources of this species, we disregardedthe contribution from movement parallax. But new in-formation, from the visual optics of both sandlance andthe chameleon, along with a characteristically creativeinsight from Land (1995, 1999), has now raised thelikelihood that monocular parallax cues are availablefrom a completely unexpected source, rotations of aneye that has unusual visual optics.

Monocular rotational parallax

One generally thinks of monocular movement parallaxin the context of head translations by the observer that isestimating depth, e.g. the head-bobbing of some birds,insects and monkeys (Collett and Harkness 1982). Analternative means of generating parallax cues is ocularrotation, so long as the optical design and acuity areappropriate. The changes required in the optical designof the eye to make rotational parallax possible involve aseparation of the nodal point of the eye from the axis ofrotation of the eye. This feature had been noted in boththe sandlance eye and in the eye of the chameleon (Ottand Schae�el 1995) which also makes monocular depthjudgements, and was initially thought to be related tothe telescopic function of the optical couplet of lens andcornea that was thought to link these remarkably similaroptical designs. The lens is reduced in power and cornealpower is relatively increased in both systems, with theresult that the nodal point of the eye is moved forward,away from the axis of rotation of the eye. Rotation ofthe eye will therefore generate parallax information thatcould be used to measure distance if one makes someconservative assumptions about the visual acuity ofthese species (Land 1995, 1999).

Land's hypothesis could account for the extraordi-nary oculomotor behaviour of the sandlance. To check ifsandlances actually make the appropriate eye move-ments while judging prey distance, more study will berequired of strikes where both the associated eyemovements and prey are visible. In the present study, wecould not obtain enough cases where the lighting andcamera orientation allowed simultaneous views of theprey and the eyes just before a strike.

Accommodation

We could not perform the lens experiments that wouldbe needed to establish a role for accommodation in thesandlance's distance judgements, as Harkness (1977) didwith the chameleon. However, the possibility is high andthe accommodative apparatus is so highly developed insandlances that it is appropriate to consider it here.

Accommodative range

Displacement of the lens may take place in the pupillaryaxis or the pupillary plane. The range of accommodationin freshwater ®shes is from 5 D to 40 D (Nicol 1989).For ®sh with accommodation in the pupillary axis (an-tero-posterior direction), the pupil is round (i.e. in thewhite sucker, Catastomus commersonii and the gold®sh,Carassius auratus). Pupillary shape and the direction oflens movement are correlated because the lens extendswell beyond the pupil plane and the lens would collidewith the iris unless there was su�cient space between thetwo structures. This may be the case in the sandlancewhere the ¯attened lens lies juxtaposed to the iris. Thisassociation occurs in lampreys where the lens is pressedagainst the iris by the vitreous. In the case of lampreys,accommodation is produced by the contraction of anextraocular cornealis muscle which causes the cornea to¯atten pushing the lens towards the retina and makingthe eye less myopic (Kleerekoper 1972). Although of aslightly di�erent mechanism, accommodation in thesandlance may similarly be aided by a deformation ofthe hypertrophied cornea, enabling the eye to be myopicfor near objects with an increased radius of curvature ofthe cornea. After accommodation, the lens is movedtowards the retina and a focused image would fall on theconvexiclivate fovea.

The sandlance has a very large range of accommo-dation compared with other teleosts, as shown inTable 1, where data on seven teleost species are pre-sented that have accommodation around 5±7% of thetotal power of the eye, compared with 25% in the eighthspecies, the sandlance. This large accommodative rangemay be made possible by the change in shape of thecornea, which is a powerful mechanism compared withchanging the position of a lens of ®xed shape. This in-terpretation of the large accommodative range in sand-lance is supported by a comparison of accommodationin the chameleon, which has both a large range of ac-commodation (45% of the total power, Table 1) andcorneal accommodation (M. Ott, unpublished data). Aswell as exerting a comparatively greater e�ect on ac-commodative range, a corneal mechanism also confersgreater speed, because the cornealis muscle is striated,whereas the retractor lentis muscle is smooth.

Mechanism of accommodation

As already outlined, the extraordinary range and speedof accommodation in sandlance is hard to explain interms of the smooth muscle and inappropriate axis ofthe retractor lentis muscle. Instead, we propose thataccommodation is largely corneal and that contractionof the cornealis muscle acts to ¯atten the cornea andreduce the refractive state from around 120±180 D my-opic to zero, as we observed in the experiment whereacetylcholine was applied directly to the eye. There maybe three components to this: an increase in the radius of

259

curvature of the anterior surface of the lenticle, a de-crease in the radius of curvature of the posterior surfaceas it is pulled against the crystalline lens, and a posteriordisplacement of lens and lenticle. All three changeswould act in the same direction, to reduce the power.The striated muscle of the cornealis would help to ex-plain the high speed of accommodation (in contrast tothe much slower speed of contraction one expects from asmooth muscle like the retractor lentis). This assignmentalso helps to explain the success of topical application ofacetylcholine, which might be expected to have less e�ecton a muscle located inside the eye, such as the retractorlentis. Finally, a corneal role for accommodation isconsistent with the large component of the eye's powerin the cornea and the large excursions of the lens thatwould be required to account for the range of accom-modation that we have observed. If the overall opticalstrategy involves an increase in the role played by thecornea at the expense of the lens, so that there is maxi-mal separation between nodal point and axis of rotation,then it is not surprising that accommodation is largelycorneal in this teleost.

Acknowledgements This work was supported by National Healthand Medical Research Council and Australian Research Councilgrants to J. D. Pettigrew and S. P. Collin. S. P. Collin was an ARCQueen Elizabeth II Research Fellow. We thank the Great BarrierReef Marine Park Authority and the Fisheries Department ofWestern Australia for permission to collect sandlances at HeronIsland and Rottnest Island, respectively. We are grateful to JoshWallman and Justin Marshall for discussions and to H. BarryCollin for assistance with the electron microscopy. We also thankJurgen Tautz, David Sandeman and Renate Sandeman of theTheodor-Boveri-Institut fuÈ r Biowissenschaften (Biozentrum) derUniversitaÈ t, WuÈ rzburg for the use of a high-speed video camera.Charlie Braekevelt provided valuable information about sarco-meres. Barry Hutchins of the West Australian Museum helped withthe collection of specimens and provided helpful taxonomic advice.The assistance of Simon Nevin, Rita Collins, Tom Steginga,Charles Nelson, Stephen Walsh, Tara Kurrajong and Chloe Cal-listemon at various stages of the study is gratefully acknowledged.

References

Andison ME, Sivak JG (1996) The naturally occurring accommo-dative response of the oscar, Astronotus ocellatus, to visualstimuli. Vision Res 36: 3021±7

Collett TS (1977) Stereopsis in toads. Nature (Lond) 267: 349±351Collett TS, Harkness L (1982) Depth vision in animals. In: Ingle

DJ, Goodale MA, Mans®eld JW (eds) Analysis of visualbehaviour. MIT Press, Boston, pp 111±176

Collin HB, Collin SP (1988a) The cornea of the sandlance,Limnichthyes fasciatus (Creeiidae). Cornea 7: 190±203

Collin SP, Collin HB (1988b) Topographic analysis of the retinalganglion cell layer and optic nerve in the sandlanceLimnichthyes fasciatus (Creeiidae, Perciformes). J Comp Neurol278: 226±241

Collin SP, Collin HB (1988c) The morphology of the retina andlens of the sandlance, Limnichthyes fasciatus (Creeiidae). ExpBiol 47: 209±218

Collin SP, Collin HB (1997) The head and eye of the sandlance,Limnichthyes fasciatus: a ®eld emission scanning electron mi-croscopy study. Clin Exp Optomol 80: 133±138

Harkness L (1977) Chameleons use accommodation cues to judgedistance. Nature (Lond) 267: 346±349

Glickstein M, Millodot M (1970) Retinoscopy and eye size. Science168: 605±6

Howland HC, Braddick O, Atkinson J, Howland B (1983) Opticsof photorefraction: orthogonal and isotropic methods. J OptSoc Am 73: 1701±8

Kleerekoper H (1972) The sense organs. In: Hardisty MW, PotterIC (eds) The biology of lampreys, vol 2. Academic Press, NewYork, pp 373±404

Land MF (1995) Fast focus telephoto eye. Nature (Lond) 373:658±659

Land MF (1999) The sandlance eye breaks all the rules. Curr Biol9: R286±R288

Nicol JAC (1989) The eyes of ®shes. Clarendon Press, OxfordOtt M, Schae�el F (1995) A negatively powered lens in the cha-

meleon. Nature (Lond) 373: 692±694Ott M, Schae�el F, Kirmse W (1998) Binocular vision and ac-

commodation in prey-catching chameleons. J Comp Physiol A182: 319±330

Pettigrew JP, Collin SP (1995) Terrestrial optics in an aquatic eye:the sandlance, Limnichthyes fasciatus (Creediidae, Teleostei).J Comp Physiol A 177: 397±408

Pettigrew JD, Collin SP, Ott M (1999) Convergence of specialisedbehaviour, eye movements and visual optics in the sandlance(Teleostei) and the chameleon (Reptilia). Curr Biol 9: 421±424

Schae�el F, Howland HC (1987) Corneal accommodation in chickand pigeon. J Comp Physiol A 160: 375±384

Sivak J, Howland HC (1973) Accommodation in the northern rockbass (Ambloplites rupestris rupestris) in response to naturalstimuli. Vision Res 13: 2059±64

Sivak JG, Woo GCS (1975) Accommodative lens movementin holosteans (Amia calva and Lepisosteus osseus oxyurus) andin the sea lamprey (Petromyzon marinus). Can J Zool 53:516±520

Tamura T, Wisby WJ (1963) The visual sense of pelagic ®shesespecially the visual axis and accommodation. Bull Mar SciGulf Carib 13: 433±448

260