Current Biology 24, 1–6, August 4, 2014 ª2014 Elsevier Ltd All rights reserved http://dx.doi.org/10.1016/j.cub.2014.06.027

ReportBehavioral Analysisof Cuttlefish Traveling Wavesand Its Implications for Neural Control

Andres Laan,1 Tamar Gutnick,1 Michael J. Kuba,1

and Gilles Laurent1,*1Department of Neural Systems and Coding, Max PlanckInstitute for Brain Research, Max-von-Laue Strasse 4,60438 Frankfurt, Germany

Summary

Traveling waves (from action potential propagation to swim-ming body motions or intestinal peristalsis) are ubiquitousphenomena in biological systems and yet are diverse inform, function, and mechanism. An interesting such phe-nomenon occurs in cephalopod skin, in the form of movingpigmentation patterns called ‘‘passing clouds’’ [1]. Thesedynamic pigmentation patterns result from the coordinatedactivation of large chromatophore arrays [2]. Here, we intro-duce a new model system for the study of passing clouds,Metasepia tullbergi, inwhichwave displays are very frequentand thus amenable to laboratory investigations. The mantleof Metasepia contains four main regions of wave travel,each supporting a different propagation direction. The fourregions are not always active simultaneously, but thosethat are show synchronized activity and maintain a constantwavelength and a period-independent duty cycle, despite alarge range of possible periods (from 1.5 s to 10 s). Thewave patterns can be superposed on a variety of otherongoing textural and chromatic patterns of the skin. Finally,a travelingwavecanevendisappear transiently and reappearin a different position (‘‘blink’’), revealing ongoing but invis-ible propagation. Our findings provide useful clues aboutclasses of likely mechanisms for the generation and propa-gation of these travelingwaves. They rule out wave propaga-tion mechanisms based on delayed excitation from a pace-maker [3] but are consistent with two other alternatives,such as coupled arrays of central pattern generators [3] anddynamic attractors on a network with circular topology [4].

Results

Soft-bodied cephalopod skin is covered with hundreds tomillions of elastic pigmented cells called chromatophores,whose size can be rapidly and individually altered by neuralactivation of radial muscles [1]. Coordinated size changesacross arrays of chromatophores are necessary to generatethe large variety of visual displays observed in squid, cuttle-fish, and octopus [2]. One such display is the ‘‘passing cloud,’’a dark band that travels across the body of the animal [1]. Thetropical cuttlefish Metasepia tullbergi (Appellof, 1886) provedto be an excellent model organism to study these travelingwaves due to the animal’s small size, the high occurrence ofwave displays, and the slow speed of the animal (convenientfor filming). A keeping permit was obtained by the VeterinaryDepartment of the city of Frankfurt according to section 11of the German Animal Welfare Act.

Four Main Regions of Wave TravelWe filmed five adult, freely behaving M. tullbergi in their hometank from above at a rate of 50 frames per second (fps) at high-definition (HD) resolution. Based on the observed wave dis-plays (Movie S1 available online), the left half and right half ofthe body of M. tullbergi can each be divided into four regions(Figures 1A and 1B, regions 1–4). The regions are bilaterallysymmetric, and each region supports a distinct travelingwave with a unique region-specific direction of travel (Fig-ure 1A, arrows). The waves do not cross region boundaries.Within a region, the direction of travel is constant over timeand across animals, although the travel velocity varies withineach animal (Figures 1C and 1D). In each region, the wavealways appears as a pigmented moving band with steeplyfading boundaries. The wavelength of the display (see detailsbelow) is approximately commensurate with the total lengthof travel. Hence, each region usually supports only one pig-mented moving band, except when that band approachesthe distal edge of its region of travel; at that time, a secondband appears at the proximal edge.The wave in the anterior region (Figure 1A, regions 1 and 10)

emerges at the anterior rim of the mantle and travels posteri-orly. When it occurs bilaterally, the anterior wave forms a sin-gle, continuous, jagged, C-shaped band extending from onefin line to the other, on each side of the animal (Figure 1A, greenzone). The medial arc of this band travels on the mediodorsalmantle (region 10), where it eventually collides with the bound-ary of region 3 and disappears. The disappearance of themedial arc of the anterior wave breaks the band into twoisolated lateral bands, one on each side of the animal; eachband continues traveling until it reaches the region 2 boundaryand vanishes.The body ofM. tullbergi has a relatively smoothmediodorsal

surface, separated from its downward sloping laterodorsalsurfaces by two longitudinal ridges of papillae (one on eachside of the animal; Figures 1A and 1B). A second zone ofwave initiation is located slightly lateral to this ridge (Figure 1A,cyan bands). From there and on each side of the body, twowaves begin their propagation simultaneously: one travelslaterally toward the fin line (region 2) while the other propa-gates medially (region 3) toward the dorsal midline. Propaga-tion in region 2 ends when the wave reaches the fin line.In region 3, the centripetal waves on each side of the bodycollide at the midline.The last region of wave propagation (Figure 1A, region 4) is

located in the posterior half of the mantle, with the initiationpoint at the tip (Figure 1A, red). From there, a right-side bandand left-side band simultaneously begin to move anteriorlytoward the head. The left and right bands meet at the midline,thus forming a traveling V shape. As observed with the anteriorC-shaped band, the V breaks into two separate bands whenits apex collides with the posterior boundary of region 3(Figure 1B).

Superposition of Wave and Static PatternsColeoid cephalopods use their chromatophore system for avariety of cryptic and social displays. The camouflage displaysin particular are static, multiform, and highly adaptive [1, 2].We*Correspondence: gilles.laurent@brain.mpg.de

Please cite this article in press as: Laan et al., Behavioral Analysis of Cuttlefish Traveling Waves and Its Implications for Neural Con-trol, Current Biology (2014), http://dx.doi.org/10.1016/j.cub.2014.06.027

observed that the traveling waves can be superimposed on awide variety of static body patterns (e.g., flamboyant ormottle)and textures, which are also produced by activity in the chro-matophore system and skin musculature (Figures 1E–1H).Overall, we observed that M. tullbergi can express wave dis-plays in behavioral contexts as diverse as hunting, swimming,walking,mating, resting under rocks, threat posturing, and eggmaintenance. The waves ceased to be visible mainly duringmaintained periods of quiescence (tens of minutes to hours).

Wavelength, Period, Duty Cycle, and VelocityFor quantitative analysis, we filmed M. tullbergi with a high-speed camera (100 fps) at HD resolution from a lateral view.Each video lasted 19 s. We analyzed the temporal features ofthe wave in region 1 because it is the largest and most con-spicuous of the four regions when observed from a lateralview. Within this region, we selected an easily identifiableskin element, which we call ‘‘patch’’ (see Supplemental Exper-imental Procedures and Figure S1A), on the animals’ bodies(different for each individual, function of filming angle, texturaland pattern states of the animal), and the temporal character-istics of the wave were studied using visual annotation.

Within the patch, we documented the following parameters:time of darkening onset (t0) (i.e., chromatophore expansion;see Figure S1); time of return to baseline intensity (t1); time of

onset of the following wave of chromatophore expansion (t2).We then calculated the cycle period (T; T = t2 2 t0), its vari-ability, and the duty cycle (C; C = [t1 2 t0]/T). Over a sampleof 98 cycles pooled across 28 instances and 5 animals, wefound the period to vary between 1.5 s and 10 s, i.e., a 6-foldvariation (Figure 2A; Supplemental Experimental Procedures).A similar range was observed within individual animals(Figure 2B). Despite this wide variation, the duty cycle was0.45 6 0.096 and independent of the period (Figures 2A and2B; R2 = 0.025; 95% confidence intervals, 0.17–0.22; Fisher’sz transformation was used for calculation of all correlationcoefficient confidence intervals). We next measured the wave-length (distance between two successive pigmentation bandcenters) of the moving pattern in region 1 and its ratio to themaximum distance between anterior and posterior regionboundaries (ratio of green to yellow lines, ‘‘wavelength frac-tion’’; Figure 2C). Again, this ratio did not depend on the period(Figure 2D; R2 = 0.071; 95% confidence intervals, 0.29–0.414;n = 31 cycles). In conclusion, the pattern within region 1 con-sisted of a succession of moving bands separated by a con-stant distance (the wavelength) that was smaller (by about30%) than the total region of travel. Although the width ofeach pigmented band (duty cycle) varies, it varies indepen-dently of the period (and velocity). Finally, because the periodvaries about 6-fold while the wavelength is constant, the

Figure 1. Multiple Regions of Wave Travel

(A) Schematic of the animal with depiction of thefour regions of wave travel on the dorsal mantle.The regions and wave displays are almost alwaysbilaterally symmetric. Regions of wave travel(with boundaries as thin lines) are indicated onleft side of the animal; direction of travel withineach region is denoted by arrow on right side.Each region supports only one direction of travel,and waves do not cross region boundaries. Thegreen line marks the mantle rim wave initiationzone; the cyan line indicates the middle waveinitiation zone; the crimson spot indicates thethird wave initiation zone at the mantle tip.Region names are as used in text: 1 indicatesanterior, 10 indicates anterior dorsal, 2 indicatesmiddle, 3 indicates central, and 4 indicatesposterior.(B) Still video image of one of the experimentalanimals during wave display. Intensity of back-ground is reduced, and contrast of the animal isenhanced to improve visibility. Scale bar repre-sents 1cm.(C and D) Time position intensity plots for twowave displays with two different wave propaga-tion velocities (fast in C, slow in D). Each columndepicts the pigmentation of a one-pixel-wideparasagittal strip in region 1 of the animal’sbody over time (sampling rate is 5 Hz); rowsshow the same strip along the body of the animalat successive times. Motion thus appears asdiagonal lines, whereas static patterns formvertical columns. Jaggedness is due to superim-posed motion of the animal or contractions ofthe mantle.(E–H) Single video frames of animals expressingtraveling waves on four different static bodybackgrounds: flamboyant (E); mottle (F); flam-boyant-mottle (G). Images are contrast enhancedto improve visibility. Box in (G) indicates positionof expanded view in (H). Note the pigmenteddots, corresponding to single chromatophores.Green scale bar of (H) represents 2 mm.

Current Biology Vol 24 No 152

Please cite this article in press as: Laan et al., Behavioral Analysis of Cuttlefish Traveling Waves and Its Implications for Neural Con-trol, Current Biology (2014), http://dx.doi.org/10.1016/j.cub.2014.06.027

propagation velocity also varies 6-fold. Two examples ofvelocities can be seen (as the slopes of the dark band positionover time) in Figures 1C and 1D.

Coordination across RegionsWe next investigated whether the wave patterns expressedin different regions of the body are synchronized. In videoswhere both the posterior boundary of region 1 and the anteriorboundary of region 4 (i.e., each boundary formedwith region 2)were clearly visible, we measured the arrival times of the ante-rior wave at the border between regions 1 and 2 for successivewaves (times ua1 and ua2) and the arrival time of the posteriorwave (up1) at the border between regions 4 and 2, which wasclosest in time to ua2. Plotting the anterior versus posteriordelay ([ua2 2 ua1] versus [ua2 2 up1]; Figure 2E) revealed avery strong correlation (R2 = 0.96; 95% confidence intervals,0.93–0.98; n = 50 cycles; see Supplemental Experimental Pro-cedures) between the two measures. Because the posteriordelay was determined relative to events in region 1, this resultis possible only if the anterior and posterior waves remainsynchronous despite variations in period. A similar analysisbetween regions 1 on opposite sides (Figure 2F) and betweenanterior and middle regions on one side (Figure S1B) con-firmed this result. Hence, when waves occur on several re-gions simultaneously, they are perfectly synchronized.

Selective Spatial Gating of the WaveThe synchrony found among the eight body regions isremarkable not only because the directions of wave travel

Figure 2. Duty Cycle, Wavelength, and Interre-gion Synchrony Are Independent of the Period

(A) Duty cycle versus period (n = 100 cycles; datapooled across five animals). The period is vari-able by a factor of 6, from 1.6 s to 10 s, yet theduty cycle is period independent. For details onlinear regression, see Results.(B) Duty cycle versus period for one male (blue;n = 19 cycles) and one female (red; n = 28 cycles)animal, illustrating period variability within indi-vidual animals.(C) Definition of wavelength fraction: yellowarrow depicts length of travel region; green arrowshows wavelength (the distance between thecenters of the two traveling bands; see Supple-mental Experimental Procedures). Wavelengthfraction is calculated as the ratio of wavelengthto travel region length.(D) Graph of wavelength fraction versus period(n = 31 cycles).(E) Synchronization of regions 1 (anterior) and 4(posterior), indicated by the strong correlationbetween anterior and posterior delays (for defini-tion, see Results).(F) Synchronization of regions 1 on oppositesides of the body, as in (E) (see SupplementalExperimental Procedures for definition of refer-ence points).

differ across regions (Figure 1A) butalso because traveling-wave expressioncan be gated locally. Whereas regions 1and 4 always appeared to be coactive,waves in these regions often occurredwithout corresponding activity in re-gions 2 or 3 (Movie S1). The animals

remained sufficiently still for automated analysis in 16 out of88 high-speed recordings. In each recording, we selectedtwo small patches of skin (Figure 3A, region 1 [blue] and region2 [red]) andmeasured the intensity of each as functions of time(see also Supplemental Experimental Procedures). Two typicalexamples are shown in Figures 3B and 3C. We cross-corre-lated the intensity profiles in regions 1 and 2 and measuredthe maximum absolute value of the Pearson cross-correlationwithin a 4 swindow, andweobserved that those values formedtwo well-separated clusters (Figure 3D, y axis). We indepen-dently visually classified the same videos into two categories:category a contained displays in which both region 1 andregion 2 visibly supported a wave, and category b containeddisplays in which only region 1 showed wave activity. Theplot of visual classification against correlation (Figure 3D) re-vealed perfect agreement. Using the thus-validated visualannotation in videos where automated analysis proved impos-sible due to motion of the animal, we found that 25% (22 out of88) of our recordings contained an inactive middle region.High-speed recordings focused mainly on a lateral view of

the animals. To better study the control of wave expressionacross regions, we filmed HD recordings at 50 fps from above,enabling filming from the dorsal side and thus revealing all fourregions on both sides of the animal. Following the animals inthis manner over a period of 5 days (w2 hr of footage;w1,600 wave activity cycles), we documented the prevalenceof coordinated wave patterns. We used the notation 1/.../4to denote a pattern where regions between 1 and 4 onone side supported a wave (e.g., 1/3/4 means that regions

Traveling Waves in Metasepia tullbergi3

Please cite this article in press as: Laan et al., Behavioral Analysis of Cuttlefish Traveling Waves and Its Implications for Neural Con-trol, Current Biology (2014), http://dx.doi.org/10.1016/j.cub.2014.06.027

1, 3, and 4, but not region 2, supported a wave). Out of allpossible pattern combinations, we observed only patterns1/4 (58%), 1/3/4 (21%), and 1/2/3/4 (21%) (% = percentageof total observation time). Patterns such as 1/3 or 2/4 werenever observed. The wave displays mentioned above wereall symmetrical, except for a single 40 s observation of anasymmetric wave display.

In two out of five animals (both females), we observed a pre-viously unreported phenomenon, which we call ‘‘blink’’. A blinkis a transient (about 0.5 s) and local (within a region) reductionof the intensity of the traveling wave (Figure 4A; Movie S1).Blinks did not greatly affect wave propagation, even thoughwave pigmentation intensity could be so reduced as to maskthe traveling band entirely (in 26% out of a total of n = 40blinks). If two pigmented bands were visible in the same region(e.g., at the anterior and posterior boundaries of region 1) at themoment of blink initiation, both bands underwent a simulta-neous reduction in amplitude. Anecdotally, blinks coincided

with the transient appearance of a dark line on the thirdtentacle and a twitch of the upper region of the tentacles.We quantified the effect of blinks on wave propagation by

measuring the travel fraction. If a particle travels at constantvelocity the length L of a track in time t, then in time t, it travelsl = L3 t/t. If velocity is constant, the travel fraction (l/L over t/t)equals one. Analysis of 35 control passing clouds (1 s longanalysis windows) indeed found the travel fraction to be quiteclose to one (1.046 0.1, p = 0.01, significantly different from 1,two-tailed t test). With 40 passing clouds interrupted by blinks,the travel fraction was very slightly but significantly reducedto 0.93 6 0.18 (p = 0.001, significantly different from control,two-tailed t test). Thus, blinks appear to have only very moder-ate effects on passing-cloud propagation. Possible alternativeoutcomes (e.g., interruption followed by resumption: expectedtravel fraction is 0.52 because the average blink lasts 0.48 s;obliteration: expected travel fraction is 0.25 because blinksbegin on average 12 frames after the start of the 50 frameanalysis window) were never observed. Blinks did not repre-sent damage to a local patch of chromatophores becauseanimals that displayed blinks also displayed typical uninter-rupted passing clouds interspersed between blinks, and blinkinitiation position (marked as the location of the dark movingband at the beginning of its intensity reduction) tiled the entirezone of propagation (Figure 4B).

Discussion

Descriptions of cephalopod passing clouds (with the excep-tion of Octopus cyanea [5]) have been side notes in field orlaboratory observations of camouflage, hunting, or matingbehavior in Sepia officinalis [6, 7] or Sepia apama [8]. In thesespecies, the expression of passing clouds is sporadic, makinglaboratory studies difficult. By contrast, the nearly continuousexpression of passing clouds in M. tullbergi (and possiblyrelated species Metasepia pfefferi [9]) should greatly facilitatecontrolled and quantitative investigations of their underlyingcauses.Our descriptive analysis of passing clouds provides some

hints as to what these underlying mechanisms might andmight not be. First, because the waves propagate at a widerange of velocities (approximately 6-fold), they are unlikely tobe due to simple diffusive processes such as calcium diffusionin the mantle. Second, we observed that passing clouds prop-agate even when large fields of chromatophores remain silentin part of the wave’s propagation zone (blinks). This motion,through inactive fields, rules out the role of direct interchro-matophore-muscle coupling [10, 11] in wave propagation.Electrical or mechanical coupling between chromatophoremuscles has been suggested to underlie the propagation ofa different wave-like pattern, observed in denervated or post-mortem cephalopod skin and known as ‘‘wandering clouds’’[1, 12–14]. At least two observations made in our paperindicate that passing clouds and wandering clouds aremechanistically distinct. First, wandering clouds travel in allpossible directions in the skin, and their propagation velocityis about ten times slower ([12]; A.L., T.G., M.J.K., and G.L.,unpublished data in S. officinalis). Second, myogenic propa-gation is incompatible with our observation that the pattern’swavelength is constant while the period is variable (see discus-sion on classes of pattern-generating mechanisms below).Collectively, our results on passing clouds argue for central

rather than peripheral wave generation and hence are in favorof circuit mechanisms originating with or proximal to the

Figure 3. Wave Coordination across Regions

(A) Analysis regions (blue indicates region 1; red indicates region 2) fromwhich intensity variations during a wave display were calculated in (B)–(D).(B) Plot of region intensity variations for case where regions 1 and 2 bothsupport a traveling wave. Note the correlation.(C) Plot of region intensity variations for case where only region 1 supports atraveling wave.(D) Peak cross-correlation values between the signals of regions 1 and 2 forvisually classified categories a and b. Category a: regions 1 and 2 bothsupport a traveling wave as per visual inspection. Category b: only region1 supports a traveling wave. Note the agreement between automated andvisual categorizations.(E) Pie chart showing the relative abundance of the three observed types ofcoordinated activity patterns.

Current Biology Vol 24 No 154

Please cite this article in press as: Laan et al., Behavioral Analysis of Cuttlefish Traveling Waves and Its Implications for Neural Con-trol, Current Biology (2014), http://dx.doi.org/10.1016/j.cub.2014.06.027

chromatophore motoneuron populations. Because these mo-toneurons lie in the chromatophore lobes [15], those circuitsshould first be sought within that region. Because the passingclouds can be superposed on a variety of static displays (i.e.,on a variety of chromatophore relaxation states), it seemsunlikely that the motoneurons themselves form or take partin the central pattern generation circuits. A scheme in whichchromatophore motoneurons act as readouts for wave andstatic displays is more parsimonious. We thus hypothesizethe existence of dedicated and premotor wave-pattern-gener-ating circuits, upstream of the chromatophore motoneurons.Rhythmic bursting neurons in the Octopus vulgaris posteriorchromatophore lobes have been observed in intracellularrecordings in vivo [16] (with a period of 2–3 s, which is longerthan the duration of passing clouds reported for octopuses[approximately 1 s] but is possibly explained by experimentalcooling of the preparation).

At least three different general classes of circuits couldgenerate periodic traveling waves [3, 4] (see also Figure S2).The first consists of an oscillator connected to an array ofcommon targets (e.g., neurons or muscles). If delays betweenthe oscillator and its targets increase systematically along thearray, the appearance of a traveling wave will be created. Theconnection between the driver oscillator and the respondersneed not necessarily be direct. It can result from a signal prop-agating through a network of intrinsically excitable elements(e.g., neurons [17], smooth muscles [18], astrocytes [19], andpossibly denervated chromatophores participating in wander-ing clouds [1, 14]) in a feedforward fashion, in which everyelement excites a few of its nearest neighbors further downthe chain. However, whatever the origin of the delay, thespatial wavelength of the pattern in such a design increases

proportionally with the period of the driving oscillator. Becausethe wavelength of the Metasepia wave display is independentof the period, we can rule out all realizations of circuits basedon this principle.A second class of circuits capable of generating traveling

waves involves an array of weakly coupled oscillators thatcan phase lock their activity [20]. If there is asymmetricconnectivity [21] or a uniform gradient of intrinsic oscillatorfrequencies along the array [22], the phase delay betweenneighboring oscillators will be constant, and a periodic wavewill travel across the system in one direction and with aperiod-independent wavelength. This design is proposed, forexample, to underlie swimming body movements in fish [23]and tadpoles [24] and crawling locomotion in insect larvae[25]. It is also fully compatible with all the observed propertiesof the Metasepia wave display. If it is indeed the class ofcircuits used to drive chromatophore traveling waves incephalopods, it points toward the locomotor system (morespecifically, the network used to control wave-like undulationsof the cephalopod fin during swimming; indeed, chromato-phore and fin motion motoneurons lie intermingled in the finlobes of S. officinalis [26]) as a possible evolutionary ancestorof our hypothesized premotor pattern-generating circuit [27].A third class relies on periodic circuit topology [4]. In

its simplest form, an array of neurons with feedforward andclosed (circular) connectivity can, if excited appropriatelyand connected to a linearly arranged readout (e.g., the chro-matophore array), generate waves identical to those producedby the second class. Whereas the period of the travelingwave is controlled by modulations in oscillator frequency in asystem of the second class, it is the propagation velocity alongthe ring that allows control over the period in the third.Because circuits of the second and third type can produce

equivalent outputs, we cannot choose between them basedon behavioral evidence alone. If, however, some further exper-imentation could provide support for one circuit over theother, our behavioral findings would provide constraints topin down the details of its biophysical implementation. Forexample, if a circuit of class 2 was established, our finding ofa period-independent duty cycle would point toward multi-cellular oscillators. Oscillators created through the actions ofvoltage-activated currents within a single cell typically havea period-dependent duty cycle (consider for example thecardiac pacemaker or neuronal action potentials) [28, 29]. Bycontrast, a period-independent duty cycle is easy to establishin an oscillator composed of two mutually inhibiting half-cen-ters [30]. In the unperturbed state, one can see by symmetrythat each half-center will be active for roughly half the dutycycle, whatever the period may be, and selective input intoone half-center can be used tomodulate the duty cycle aroundthe mean, default value.Upon naive inspection, the finding of four different regions of

wave propagation, eachwith its distinct propagation direction,might suggest the presence of four distinct central controlsystems. However, the observed synchrony (across sidesand across regions on one side) suggests that all the wavepatterns may be generated by a single central control systemmapped onto the body via region-specific axonal projectionpatterns [31] in four different ways on each side. Becausewe have argued that chromatophores (or chromatophoremotoneurons) act as readouts of the premotor wave-pattern-generating circuit, the presence of selective spatial gatingcan be reconciled with the notion of a single central con-troller by postulating a region-selective presynaptic gating

Figure 4. Blinks

(A) Time-lapse illustration of a blink. The multipanel image shows selectedvideo frames (t in seconds), illustrating the transient and nearly completedisappearance of the dark band (yellow arrows) in the third panel and itsreappearance at a new position in the fourth panel.(B) Blink positions (red dots; n = 20) tile the region of wave travel. A red dotwas placed on the animal’s body to mark the position of the leading edgeof the traveling pigmented band at the beginning of a blink. Each dot repre-sents a different blink. This shows that blinks can be initiated at anymomentof wave propagation.

Traveling Waves in Metasepia tullbergi5

Please cite this article in press as: Laan et al., Behavioral Analysis of Cuttlefish Traveling Waves and Its Implications for Neural Con-trol, Current Biology (2014), http://dx.doi.org/10.1016/j.cub.2014.06.027

mechanism [32]. However, behavioral experiments cannotdefinitively rule out the presence of multiple control systemsbecause coupling between multiple control systems couldestablish their mutual synchrony [22].

Supplemental Information

Supplemental Information includes Supplemental Experimental Proce-dures, two figures, and one movie and can be found with this article onlineat http://dx.doi.org/10.1016/j.cub.2014.06.027.

Acknowledgments

This work was funded by the Max Planck Society. We thank our animalfacility staff for assistance with animal husbandry and C. Hartmann andK. Costa for comments on an early version of the manuscript. We wouldalso like to thank Kazuoki Iida from Izu Chuo Aqua Trading for his help inobtaining the animals. A.L. performed these experiments as part of thedoctoral program of the International Max Planck Research School forNeural Circuits (Max Planck Institute for Brain Research).

Received: May 16, 2014Revised: June 10, 2014Accepted: June 12, 2014Published: July 17, 2014

References

1. Messenger, J.B. (2001). Cephalopod chromatophores: neurobiologyand natural history. Biol. Rev. Camb. Philos. Soc. 76, 473–528.

2. Hanlon, R.T., Chiao, C.C., Mathger, L.M., Barbosa, A., Buresch, K.C.,and Chubb, C. (2009). Cephalopod dynamic camouflage: bridging thecontinuum between background matching and disruptive coloration.Philos. Trans. R. Soc. Lond. B Biol. Sci. 364, 429–437.

3. Ermentrout, G.B., and Kleinfeld, D. (2001). Traveling electrical waves incortex: insights from phase dynamics and speculation on a computa-tional role. Neuron 29, 33–44.

4. Xie, X., Hahnloser, R.H.R., and Seung, H.S. (2002). Double-ring networkmodel of the head-direction system. Phys. Rev. E Stat. Nonlin. SoftMatter Phys. 66, 041902.

5. Mather, J., andMather,D.L. (2004).Apparentmovement inavisualdisplay:the ‘passing cloud’ of Octopus cyanea (Mollusca: Cephalopoda). J. Zool.(Lond.) 263, 89–94.

6. Hanlon, R.T., and Messenger, J. (1988). Adaptive coloration in youngcuttlefish (Sepia officinalis L.): the morphology and development ofbody patterns and their relation to behaviour. Philos. Trans. R. Soc.Lond. B Biol. Sci. 320, 437–487.

7. Adamo, S.A., Ehgoetz, K., Sangster, C., and Whitehorne, I. (2006).Signaling to the enemy? Body pattern expression and its responseto external cues during hunting in the cuttlefish Sepia officinalis(Cephalopoda). Biol. Bull. 210, 192–200.

8. Norman, M.D., Finn, J., and Tregenza, T. (1999). Female impersonationas an alternative reproductive strategy in giant cuttlefish. Philos. Trans.R. Soc. Lond. B Biol. Sci. 266, 1347–1349.

9. Roper, C.F.E., and Hochberg, F.G. (1988). Behavior and systematics ofcephalopods from Lizard Island, Australia, based on color and bodypatterns. Malacologia 29, 153–193.

10. Froesch-Gaetzi, V., and Froesch, D. (1977). Evidence that chromato-phores of cephalopods are linked by their muscles. Experientia 33,1448–1450.

11. Reed, C.M. (1995). Dye coupling in the muscles controlling squid chro-matophore expansion. J. Exp. Biol. 198, 2631–2634.

12. Sanders, G.D., and Young, J.Z. (1974). Reappearance of specific colorpatterns after nerve regeneration in octopus. Philos. Trans. R. Soc.Lond. B Biol. Sci. 186, 1–11.

13. Hoffmann, F.B. (1907). Gibt es in der Muskulatur der Molluskenperiphere kontinuierlich leitende Nervennetze bei Abwesenheit vonGanglienzellen? I. Untersuchungen an Cephalopoden. Pflugers Arch.Ges. Physiol. 118, 375–413.

14. Packard,A. (1995).Organizationofcephalopodchromatophoresystems:a neuromuscular image-generator. In Cephalopod Neurobiology, J.Abbott, R. Williamson, and L. Maddock, eds. (Oxford: Oxford UniversityPress), pp. 415–429.

15. Nixon,M., and Young, J.Z. (2003). TheBrains and Lives of Cephalopods,First Edition (Oxford: Oxford University Press).

16. Miyan, J.A., and Messenger, J.B. (1995). Intracellular recordings fromthe chromatophore lobes of octopus. In Cephalopod Neurobiology,J. Abbott, R. Williamson, and L. Maddock, eds. (Oxford: OxfordUniversity Press), pp. 415–429.

17. Wong, R.O. (1999). Retinal waves and visual system development.Annu. Rev. Neurosci. 22, 29–47.

18. Karaki, H., Ozaki, H., Hori, M., Mitsui-Saito, M., Amano, K., Harada, K.,Miyamoto, S., Nakazawa, H., Won, K.J., and Sato, K. (1997). Calciummovements, distribution, and functions in smooth muscle. Pharmacol.Rev. 49, 157–230.

19. Scemes, E., and Giaume, C. (2006). Astrocyte calcium waves: what theyare and what they do. Glia 54, 716–725.

20. Wallen, P., and Williams, T.L. (1984). Fictive locomotion in the lampreyspinal cord in vitro compared with swimming in the intact and spinalanimal. J. Physiol. 347, 225–239.

21. Hagevik,A., andMcClellan,A.D. (1994).Couplingof spinal locomotornet-works in larval lamprey revealedby receptorblockers for inhibitoryaminoacids: neurophysiology and computer modeling. J. Neurophysiol. 72,1810–1829.

22. Kopell, N., and Ermentrout, B. (1986). Symmetry and phaselocking inchains of weakly coupled oscillators. Commun. Pure Appl. Math. 39,623–660.

23. Grillner, S. (1985). Neurobiological bases of rhythmic motor acts invertebrates. Science 228, 143–149.

24. Roberts, A., Soffe, S.R., Wolf, E.S., Yoshida, M., and Zhao, F.Y. (1998).Central circuits controlling locomotion in young frog tadpoles. Ann.N Y Acad. Sci. 860, 19–34.

25. Pulver, S.R., and Griffith, L.C. (2010). Spike integration and cellularmemory in a rhythmic network from Na+/K+ pump current dynamics.Nat. Neurosci. 13, 53–59.

26. Gaston, M.R., and Tublitz, N.J. (2006). Central distribution and three-dimensional arrangement of fin chromatophore motoneurons in thecuttlefish Sepia officinalis. Invert. Neurosci. 6, 81–93.

27. Striedter, G.F. (2004). Principles of Brain Evolution (Sunderland: SinauerAssociates).

28. Bazett, H.C. (1920). An analysis of the time-relations of electrocardio-grams. Heart 7, 353–370.

29. Strogatz, S. (2001). Nonlinear Dynamics and Chaos (Boulder: WestviewPress).

30. Matsuoka, K. (1987). Mechanisms of frequency and pattern control inthe neural rhythm generators. Biol. Cybern. 56, 345–353.

31. Sosulski, D.L., Bloom, M.L., Cutforth, T., Axel, R., and Datta, S.R. (2011).Distinct representations of olfactory information in different corticalcentres. Nature 472, 213–216.

32. MacDermott, A.B., Role, L.W., and Siegelbaum, S.A. (1999). Presynapticionotropic receptors and the control of transmitter release. Annu. Rev.Neurosci. 22, 443–485.

Current Biology Vol 24 No 156

Please cite this article in press as: Laan et al., Behavioral Analysis of Cuttlefish Traveling Waves and Its Implications for Neural Con-trol, Current Biology (2014), http://dx.doi.org/10.1016/j.cub.2014.06.027

Current Biology, Volume 24 Supplemental Information

Behavioral  Analysis   of  Cuttlefish  Traveling  Waves   and  Its  Implications  for  Neural  Control Andres Laan, Tamar Gutnick, Michael J. Kuba, and Gilles Laurent

0!

1!

2!

3!

4!

5!

6!

7!

0! 1! 2! 3! 4! 5! 6! 7!

dela

y 4

(s)!

delay 1 (s)!

Figure S1!

t0! t1! t2!A!

B!

Figure S2!

Supplemental Data Supplemental figure legends Figure S1, related to Figure 2: Illustration of steps in visual annotation and its application to the evaluation of synchrony between central (3) and anterior (1) regions. (A) Determination of duty cycle and period for a reference skin element. Frames 1-3 depict the location of the travelling band with respect to the reference skin element (a blue line marks the border of the reference papilla) during events t0, t1 and t2 respectively. (B) Documentation of central and anterior region synchrony (n = 19 cycles, R2 = 0.94, 95% confidence intervals 0.84 — 0.97, Fisher’s z transformation). Figure S2, related to Discussion: A schematic of three classes of wave generating systems in biological systems. An illustration of three classes of networks capable of supporting travelling waves in biological systems (based on references [S1, S2] below). For class-one networks, lightning bolts depict intrinsically excitable elements that burst when depolarized above a threshold; Δτ represents the delay required to excite the next element down the chain; P indicates pacemaker locations in the system. Class-two networks are composed of a chain of weakly coupled pacemakers, Δφ represents the constant phase delay between neighboring elements along the chain. Class-three networks consist of a set of neurons connected in a circular network exciting each other in a feed forward fashion thus causing activity propagation along the loop. A second set of recurrent connections is required to stabilize the propagating wave profile. Supplemental movie legends Supplemental movie S1: The first segment demonstrates three types of wave activity patterns; the second segment shows the superposition of the wave on various static patterns, period variability and wavelength constancy; the third segment shows a blink, slowed down by a factor of 6 to ease viewing. Comments included within video. Supplemental Experimental Procedures Animal keeping: Five wild-caught Metasepia tullbergi (2 males, 3 females) were transported by plane to our animal holding facility in Frankfurt am Main (Germany). The animals were kept in a 100x60x40 cm tank filled with continuously recirculating artificial seawater. The tank was part of a larger system of aquaria containing a total of 3000 l of artificial seawater. Temperature was kept at 23C and illumination was provided by LED lights on a 12h/12h light-dark cycle with a 1 h dusk and dawn period. Animals were fed 4 times a day with live shrimp. The animals were maintained for a period of six months in our animal holding facility during which they remained in good health and completed a successful reproductive cycle.

Filming: All M. tullbergi were filmed while freely behaving in their home tank. For high-speed filming, we used the Sony NEX FS700 U camera at 100 frames per second with manual intensity control at HD resolution with a 100mm Canon Macro lens. The camera was mounted on a tripod and directed to an individual animal. High-speed 19-second clips were recorded from a lateral view, when the animal was moving at a sufficiently slow speed to remain in the frame and to enable accurate focusing. For filming from a dorsal view, we used a Sony HXR-NX70E camera filming at full HD with 50 fps. With the top lid of the tank removed, an experimenter manually followed an individual animal over 20- to 60 min long recording sessions. Duty cycle and period manual video annotation: videos were exported for analysis from the camera memory into Apple Final Cut Pro 10.0.9. In each video, we selected a patch in the middle part of region 1. Each patch was centered on a small, easily identifiable reference point; usually this reference was a particular papillary protrusion or a white skin blotch (on mottle backgrounds, see Figure S1A). When analyzing the video frame by frame, we annotated a frame as t0 when the reference patch began darkening; frame t1 when intensity returned to baseline; and frame t2 at the onset of the next darkening cycle. The duty cycle and period were calculated as described in main text and then converted from frame numbers into seconds for plotting. See Figure S1A for illustration of wave position during the three time points. Synchrony: Quantification of synchrony between ipsilateral regions 1 and 4 was performed as described in text, also using frame-by-frame analysis in Final Cut Pro 10.0.9. Quantification of synchrony between ipsilateral regions 1 and 3, or contralateral regions 1, was done with videos acquired at 50 fps from a top view. For left-right synchrony, the reference event (u0) was the emergence of the left wave at the mantle rim of region 1. That wave was then followed until it reached the region 1-2 border (u1) and then the closest (in time) wave arrival at the region 1 - 2 boundary on the right side was used as the final reference event (u2). Plots of u1 - u0 and u2 - u0 served to quantify synchrony. For 3 - 1 synchrony, the first two events were as described for the left-right synchrony analysis; the third event was the arrival of a wave at the papillary ridge on the left side (u2). Wavelength fraction: For some of the waves where a single period was determined (for duty cycle analysis) we calculated the wavelength fraction. We used a wavelength fraction metric to compensate for variable filming distance between clips, and for size difference across animals. Wavelength fraction was calculated in animals filmed from a lateral view as the ratio of the wavelength to the length of region 1. The length of region 1 was determined as the distance between two reference points: the midpoint of the line defining the border between region 1 and 2, and the intersection point between the mantle rim and the (extrapolated) papillary ridge; this point was also typically that at which the wave could be seen to emerge when viewed laterally. The wavelength was determined as the distance between the centers of the two dark bands along the line determined by the two reference points. Automated analysis of wave activity: sixteen high-speed videos were selected (out of a total of 88) in which the animal remained sufficiently still throughout the 19-second clip, such that an analysis square (roughly 2/3 the thickness of the dark band of the wave) placed in the middle of region 2 and a second square in the center of region 1,

remained within their corresponding region for the full duration of the clip. The selected high-speed videos were subsampled by a factor of 1/10 and analysis frames were converted from RGB to grayscale. The sum of pixel intensities within the analysis square was calculated for each frame and divided by the total number of pixels contained within the square to give the final intensity value for each analysis square in each frame. Cross-correlation was determined between the two waveforms for a window of -2 to 2 seconds (a total shift value of 4 seconds). The maximum absolute value of the correlation was used as the final score. Cross-correlation with a range of delays was used because the strength of protrusion of a papilla in the middle region was variable between videos and animals and slightly affected the phase shift values between the two waveforms. All automated video processing was performed in Matlab 2012a (The Mathworks) and its Image Processing Toolbox. Analysis of blinks: Blinks were identified by visual annotation. Around each blink, a 1-second analysis window was created such that a blink occupied roughly the middle half of the clip and that in both the first and the last frame of the analysis the region contained the band involved in the blink. The duration of the blink was determined as the time between the initiation of lightening of the band to the end of darkening of the band. The travelled distance fraction was determined as the difference between the location fractions at the first and final frames. The location fraction for a frame was given by the ratio of the distance of the leading edge of the band from the 1-2 border to the total length of travel. The reference points for the total length of travel were the same as outlined in the wavelength fraction section and the distance of the band from the 1-2 border was determined along the line connecting the two reference points. For the particular band involved in the blink, the temporal fraction was determined by dividing the 1-second duration of the clip by the travel time, which was determined as the time difference between the first emergence of the band and the first contact of the band’s leading edge with the 1-2 border. For controls, we created a separate set of 1-second clips from the same animal, in which no blink was observed over the entire travel duration. The distribution of the positions of the travelling bands at the beginning of the clip was matched between control and blink clips by creating a random pairing between each blink and control wave and then matching the position of the travelling wave between the beginnings of the blink and control clip. Within- and across-animals observations: a potential issue (“pseudoreplication”) with our dataset is that successive periods of the wave cycle within a video clip tended to be strongly correlated with each other. Indeed, analysis of the periods of successive wave cycles reveals a pairwise correlation coefficient of 0.76 (n = 69 pairs, 95 % confidence intervals of 0.64 — 0.84, Fisher’s z transformation for this and all subsequent confidence intervals). A similar analysis for successive duty cycles revealed a pairwise correlation coefficient of 0.5 (n = 69 pairs, 95 % confidence intervals of 0.3 — 0.65). The number of truly independent observations is therefore, in the strict statistical sense, less than the 98 referred to in the paper. The appropriate number can be estimated using correction procedures, but the applicability of these procedures has been questioned. We therefore adopted a reliable control to validate our analysis. The correlation coefficient of successive wave cycles between two successive video clips was much lower (R2 = 0.22, n = 29, 95 % confidence intervals of - 0.16 — 0.54) and the removal of a single clear outlier pair reduced it to close to zero (R2 = 0.05, n = 28, 95 % confidence intervals of - 0.32 — 0.41). Repeating our analysis on these 28 independent observations, using the period of only the first cycle

in each clip, showed a correlation of 0 between duty cycle and period (R2 = 0.00, n = 28, 95% confidence intervals of - 0.37 — 0.37) confirming our original assessment. Pseudoreplication between animals was not an issue because (as shown in the paper) duty cycle is period independent within animals as well as across the group and the dynamic range of individuals is well matched with the dynamic range of the full population. Similarly, for wavelength vs period analysis, successive wavelength fractions are independent (R2 = 0.01, n = 20, 95 % confidence intervals of - 0.36 — 0.38); hence every wavelength-period pair can be regarded as an independent data point for the purposes of regression analysis. For middle-central and left-right synchrony analyses and blinks, data points came from separate clips sufficiently separated in time that potential “pseudoreplication” was not an issue. The case was apparently different for anterior-posterior synchrony analysis. The number of independent clips used in anterior-posterior synchrony analysis was only 14. Repeating the correlation analysis on these 14 observations confirmed our high correlation coefficient (R2 = 0.95, 95 % confidence intervals of 0.84 — 0.98). We will argue below, by means of noise correlation analysis, that the data allows a similar conclusion to be drawn with a sample size of at least 41. Noise correlation analysis was performed as follows: for each group of observations within a single clip, the deviation of every observation was calculated from the cycle mean. A correlation plot of successive deviations revealed that they were independent (R2 = 0.01, n = 29, 95 % confidence intervals of - 0.36 — 0.38). In marked contrast to this, the correlation coefficient between simultaneous deviations (plotting anterior deviation vs posterior deviation) revealed a strong correlation (R2 = 0.8, n = 40, 95 % confidence intervals of 0.65 — 0.89). The R2 value was now lower than for the 14 independent observations, probably because the dynamic range of the cycle deviation was 6-fold lower than for the 14 independent observations and measurement noise began to encroach on our estimate (estimated ratio of measurement noise variance to cycle deviation variance is 8.2 giving 0.892 as the expected correlation coefficient under the assumption of perfect synchrony, a number that remarkably close to the 95 % confidence interval of the actual correlation coefficient). Given that noise correlation analysis shows that there is considerable useful information hidden in within-clip replicates, we argue that our use of n = 50 is statistically justified and because the collective 50 observations best tile the dynamic range, we used them all for the calculation of the most reliable estimate of the correlation coefficient. Supplemental References S1. Ermentrout, B.G., Kleinfeld, D. (2001) Travelling electrical waves in cortex: insights from phase dynamics and speculation on a computational role. Neuron. 29, 33-44. S2. Xie, X., Hahnloser, R.H.R., and Seung, H.S. (2002) Double-ring network model of the head-direction system. Phys. Rev. E. 66, 041902-1-9.