Anim. Behav ., 1978, 26, 7 9 1-802

COMMUNICATION SYSTEMS AND SOCIAL INTERACTIONS IN AGOBY-SHRIMP SYMBIOSIS

BY J . LYNN PRESTON*Department of Zoology, University of Hawaii, Honolulu, Hawaii

Abstract . Interphyletic communication is quantitatively shown to occur in a behavioural, mutualisticsymbiotic association between the goby Psilogobius mainlandi and two species of shrimps, Alpheusrapax and A. rapacida, by the use of information theory and x2 analyses. Communication betweengobies and shrimps is primarily tactual . Gobies use as shelter burrows dug and maintained by shrimps .Gobies sit at the burrow entrance and warn shrimps of danger by a flick of the tail. Shrimps communi-cate their presence outside the burrow by touching gobies with their antennae . Gobies never givewarning signals in the absence of shrimps . It is postulated that the spacing of the burrows in natureis a result of the complex interaction of three communication systems : tactual communication amongshrimps and between gobies and shrimps and visual communication among gobies . It is suggestedthat the analysis techniques used here be applied to the analysis of other symbiotic associations .

Communication has been defined as a process inwhich there is information exchanged betweenanimals to the mutual adaptive advantage ofboth (Klopfer & Hatch 1968) . Apparent inter-phyletic and interclass communication havebeen described in cleaning symbioses in whichshrimps clean fishes (e.g . Feder 1966) and inwhich finches clean tortoises (MacFarland &Reeder 1974). The complexity of symbiosesbetween sea anemones and pomacentrid fish(Mariscal 1972) or hermit crabs (Balasch &Mengual 1974) suggests that communicationoccurs in these associations . The communicationwithin these symbioses has not, however, beenquantified .

In this study communication in the beha-vioural mutualistic symbiosis between a goby,a small bottom-dwelling fish, and an alpheid,or snapping, shrimp is explored quantitatively .In this association gobies and shrimps communi-cate with each other tactually and possiblychemically . Gobies communicate with oneanother visually. The alpheid shrimps communi-cate with one another tactually and possiblyvisually .Communication can be analysed at several

different levels . At the syntactic level the signals,their variants, their frequency of occurrence, ortheir modification by environmental factors aredetermined. At the semantic level the meaningsof the signals are determined. The significanceof the signals to the receiver is determined at thepragmatic level (Morris 1946 ; Marler 1961 ;Cherry 1966 ; Smith 1968).*Present address : 27 Pioneer Village, Philomath, Oregon97370, U .S .A.

791

Communication signals have been analysedat the syntactic level using the concepts ofinformation theory (e.g. Haldane & Spurway1954; Wilson 1962; Altmann 1965 ; Hazlett &Bossert 1965, 1966 ; Dingle 1969, 1972 ; Chatfield& Lemon 1970 ; Steinberg & Conant 1974). Inseveral of these studies the investigators alsoused x2 analyses to determine the effects ofsignals on responses . Although the x2 analysesalso give a syntactic interpretation of signals,the significance of signals to the receivers andperhaps even to the emitters can be inferred fromthe results . These data analysis techniques areapplied here to interphyletic communication .

The Goby-Shrimp AssociationGoby-alpheid shrimp symbiotic associations,which have been observed pantropically (Bayer& Rofen 1957 ; Luther 1958 ; Harada 1969 ;Karplus et al . 1972, and many others), in generalconform to the following description. The gobysits at the entrance of a burrow which theshrimp digs and maintains . The shrimp, whenploughing sand, exits from its burrow antennaefirst. The shrimp's antennae contact the tail ofthe goby, the shrimp continues to exit from theburrow, it drops its load of sand and withdrawsinto the burrow without turning around. If theshrimp is out of the burrow when an observerapproaches, the goby flicks its tail . In response,the shrimp generally sits still or flees into itsburrow. Depending on the nature of the dis-turbance, the goby may remain at the entranceof the burrow or may turn and flee head first intothe burrow, always after the shrimp . The amountof time between the disappearance and reappear-ance of the goby at the entrance of the burrow

792

ANIMAL

varies greatly . The goby always reappears first.Thus the goby apparently profits from theready-made shelter provided by the shrimp,and the shrimp apparently profits from thewarning signals the goby provides as it sits atthe opening of the burrow .

Exceptions to this description are gobies whichhover over burrow entrances they guard, ratherthan sitting on the bottom (Klausewitz 1960 ;Harada 1969). Karplus et al. (1972) observed theshrimp cleaning its goby symbiont .In shallow areas of Kaneohe Bay, Oahu,

Hawaii, the goby Psilogobius mainlandi Baldwin(Baldwin 1972) guarded the interspersed burrowsof Alpheus rapax Fabricius and A. rapacida deMan. Spacing between burrow entrances differedwith the substratum and the size of the occupants .Where the substratum was extremely rocky,burrows were few, far apart, large, and occupiedby large inhabitants. Where the substratumwas silty with a thick coral rubble base, burrowswere numerous and close together. Burrowentrance sizes, which reflected inhabitant sizes,differed greatly . There were many small burrowsinterspersed among larger ones . Marine resincasts showed that large burrows were similarto those described by Karplus et al . (1974) .They were about 20 cm deep with severalchambers and winding passages. The burrowslay under coral rubble and had sand floors .Burrow entrances collapsed readily when notattended by shrimps and new ones were formedin a short time .

Gobies were often observed without shrimps,but shrimps were observed without gobies only(1) on very hot days when the tide was low, atwhich time A. rapax were numerous on thesurface of the substratum, or (2) when I had justcaptured a guarding goby. Shrimps guarded bygobies were extremely difficult to capture ;those without gobies guarding were easilycaptured .

Pairs of gobies were observed throughout theyear. Males and females could be distinguishedin the field only if the two gobies were guardingthe same burrow at the same time. Then thefemale had a white abdomen with a dark bandaround her body in the region of the anal fin .Interactions between gobies in the field were few .When they did occur, they were brief .

All goby-shrimp burrows were occupied by atleast two shrimps. A. rapacida were found inmale-female pairs only, whereas A. rapax wereobserved three to a burrow, two females andone male, or in pairs consisting of two females

BEHAVIOUR, 26, 3

or a male and a female. When pairs wereobserved, it is possible that a third shrimp wasin the burrow . No interactions between shrimpsother than their digging together were observedin the field .

MethodsData for analyses of communication betweengobies P. mainlandi and shrimps A. rapax andA. rapacida were gathered from field observa-tions. Interactions were clearly visible fromabove the water on calm days during low tide .I chose a guarding goby to observe and beganrecording interindividual (interphyletic) two-actsequences immediately . I walked slowly towardthe burrow. When the goby began to flick itstail, I stopped my approach in order to avoidfrightening it into the burrow . When the tailflicking stopped, I continued my approach untilI was stopped beside the burrow entrance . Ifthe goby at that time did not flick its tail, I mademovements until the goby withdrew or fled intothe burrow . At that point observations on thatpair ended. If two shrimps were out of theburrow at the same time, I discarded myobservations since I was unable to observethe animals in sufficient detail .A. rapax and A. rapacida are easily distin-

guished in the field since A. rapacida has a whiteabdominal band (which disappears in preserva-tive) that is not present on A. rapax . One hun-dred and six goby-A . rapax pairs were observedfor a total of 8 .7 h. Sixty-six goby-A . rapacidapairs were observed for a total of 6 .0 h. Dura-tions of observations on goby-shrimp pairsranged from 15 s to 28 min. No goby-shrimppair was observed more than once.

Data AnalysesAll acts performed by gobies and shrimps were

coded and punched on IBM computer cards foranalyses with an IBM 360/65 computer . Contin-gency tables (e.g . Table I) were made for inter-individual (interphyletic) two-act sequences hav-ing no minimum interval length . The totalnumber of times each Initial Act (signal)occurred and the total number of times eachFollowing Act (response) occurred were deter-mined. The expected number of times that anact followed another act, assuming the InitialAct had no influence on the Following Acts,was calculated (Hazlett & Bossert 1965) .

To test whether `responses' followed `signals'at random, a x2 value (x2table) was calculatedfor the entire table . Columns or rows (whichever

produced the higher degrees of freedom) withany expected value less than one were lumpedtogether.

Using the null hypothesis that responsesfollowed each Initial Act with their overallfrequency of occurrence, I applied a x2 analysis(x2roW) for each Initial Act. The observed andexpected values were from the contingencytables of interindividual two-act sequences .In these calculations, if the expected number wasless than one, both observed and expectednumbers were lumped with all other categoriesin which the expected value was less than onefor that Initial Act, that is, within one row. Ifthe lumped sum of expected values was still lessthan one, the sums of observed and expectedwere added to the first figures in the row forwhich the expected value was greater than one .If there was not such a category, the x2 was zero .Lumped figures were counted as one category(Snedecor & Cochran 1967) .

To determine whether an Initial Act wasdirective or inhibitive toward a given response,I compared the observed and expected valuesfor each two-act sequence in the contingencytables. If the observed was greater than theexpected, the signal was considered to bedirective toward the response . If the observedwas less than the expected, the signal was con-sidered to be inhibitive toward that response(Hazlett & Bossert 1965). The statistical signifi-cance of the elicitation or inhibition was deter-mined from the x2 value (X2act) for each two-act sequence . In these calculations I used theobserved and expected values for the giventwo-act sequence for one ratio. I subtractedthese figures from the total observed value for

PRESTON : INTERPHYLETIC COMMUNICATION

Table I. Interindividual Two-Act Sequences in Interactions Between Gobies and A. rapax

793

the given Initial Act for the second ratio . If theexpected value in either ratio was less than one,the x2 for that act could not be calculated .

A number of two-act sequences were recordedfor each pair of animals during observationperiods which differed in length among pairs .As a result, not all two-act sequences are inde-pendent of one another and some pairs contributemore data to the overall two-act sequence dataset than others do. Because of these problemsand other questions as to the validity of X 2rowand x 2act (Moehring 1972), x2 values wereconsidered significant only if the probabilityof chance occurrence was less than 0 .005 .

I applied a x2 analysis (x 2freq) to two by twocontingency tables to test whether the frequencyof occurrence of each act for one type of inter-acting pair (goby-A . rapax) differed significantlyfrom the frequency of occurrence of each act forthe second type of interacting pair (goby-A .rapacida) . In the contingency tables, one rowwas the observed number of times the act inquestion occurred for the two types of interactingpairs and the second row was the observednumber of times all other acts occurred for thetwo types of interacting pairs .

Information as used here (Shannon & Weaver1949) is defined as variation, or uncertainty .Information `is measured . . . by its "news"value, that is, the extent of surprise it causes tothe recipient' (Singh 1966, p . 9). Informationvalues in this study were calculated fromobserved, interphyletic, two-act sequences, usingthe following formulae (Quastler 1958 ; Hazlett& Bossert 1965 ; Dingle 1969, 1972 ; Steinberg &Conant 1974) .

A. rapax following acts

Goby initial acts FleeIn

burrowManipulateobjects

Nochange Plough Sit Withdraw

Dorsals erect 0 4 1 3 3 0 1Flee 9 13 0 0 0 0 0Guard 14 934 109 19 939 52 691Move away 0 20 4 22 21 1 22Nip sand 5 16 2 14 12 2 13Pectorals wave 0 9 2 3 8 0 5Sit away 3 8 0 0 7 1 4Tail beat 3 0 1 1 0 0 1Tail flick 119 17 3 134 22 82 65Tail wave 0 22 3 25 15 1 38Withdraw 1 15 2 3 10 2 6

7 94

1 . Maximum information possible if all actswere equally probable: H11781, = loge Ni, whereNj is the number of responses in the behaviouralrepertoire of the receiver .2. Information Present : HB = -1 Pi 1092 Pi,

where pj is the probability of occurrence of theresponse (j).3. Conditional Information Present :

HB,A = - E P,j logepi ,,, where p ij is the overallprobability that a given two-act sequence (ij)will occur and pj I, is the probability that a givenresponse (j) will occur provided that a givensignal (i) has already occurred .4. Information Transmitted

Ht =HB -HB, A .5. Contribution of a particular act to informa-

tion transmission : h, = p, E pj,, loge pj , ;/pj(Blachman 1968 ; Steinberg & Conant 1974) .6. Normalized Transmission (Steinberg &

Conant 1974; also called coefficient of constraint,Attneave 1959 ; per cent input, Herman 1965 ;per cent uncertainty reduction, Hazlett &Estabrook 1974a, b; Rubenstein & Hazlett1974) : NT = Ht/HB, expressed as a percentage .Goby-shrimp communication analyses are

divided into two sections. For analyses in whichshrimps perform Initial Acts and gobies' actsfollow (shrimp communicates to goby), I usedacts from the entire observation time since Iobserved the behaviour of the goby at all times .For analyses in which gobies initiate the beha-viour and the shrimps' behavioural acts follow(goby communicates to shrimp), I excluded thebehaviour of the goby when the shrimp was in itsburrow since it was not possible to observe theshrimp's responses . The number of encountersoccurring between gobies and shrimps wasdetermined by the number of times shrimpsentered their burrows .

Behavioural Acts of P. mainlandiThe behavioural acts of P. mainlandi are listedin alphabetical order.

Dorsals erect : The goby spread its dorsal fins .Flee : The goby rapidly entered the burrow,

generally head first .Guard : Sitting still at the burrow entrance,

the goby sat high on its pelvic fins with itspectoral fins extended to the side, its dorsal finsfolded, and its caudal fin generally extendinginto the burrow.

In burrow : The goby was in the burrow andcould not be seen.Move away: The goby moved away from the

burrow .

ANIMAL BEHAVIOUR, 26, 3

Nip sand : The goby turned its head to theside, moved slightly forward, took a quick biteof sand, then resumed its original position. Thesand was filtered through the goby's gills andfell in small piles beneath each operculum.No change : There was no change in the

behaviour of the goby in `response' to an act bythe shrimp .

Pectorals wave : The goby moved its pectoralfins alternately in a rowing motion.

Sit away: The goby sat in any position morethan 15 cm from the burrow entrance.

Tail beat : The goby moved its folded caudalfin to one side and back to its original position,usually hitting the shrimp with it .

Tail flick : The goby flicked its folded caudalfin about three times in a manner that wasbarely visually perceptible .

Tail wave : The goby moved its spread caudalfin back and forth .Withdraw : The goby moved toward the

burrow .

Behavioural Acts of A. rapax and A. rapacidaThe behavioural acts of the alpheid shrimps arelisted in alphabetical order.

Flee : The shrimp rapidly entered its burrow,usually backward.

In burrow : The shrimp was in its burrow andcould not be seen .

Manipulate objects : The shrimp picked upobjects with its large chelipeds or walking legs .No change : There was no change in the

behaviour of the shrimp in `response' to an actby the goby .Plough : The shrimp exited from its burrow

holding its large chelipeds forward and pushingsand, pebbles, or shells in front of it .

Sit : The shrimp sat still, apparently doingnothing .Withdraw : The shrimp moved toward its

burrow, generally backward .

Results and DiscussionThe signals from one symbiont in a goby-shrimp pair decreased the amount of uncertaintyin the occurrence of responses of its symbioticpartner by 13 to 15 % (NT, Table II) . That is,interphyletic communication between goby andshrimp in these associations did occur . Theimportance of each goby signal and each shrimpsignal in their transmission of information isshown in Tables III and VI, respectively. Theeffects these signals had on the behaviour of thereceiver are shown in Tables IV, V, VII and VIII .

Strong warning signals, defined here as gobyacts directive toward `flee' in the shrimps, are`withdraw', `tail flick', `tail beat', and `flee'(Tables IV and V) . Using both qualitative andquantitative observations, I have listed thesefrom weakest to strongest signal. Gobies`withdrew' toward, but not into, the burrowswhen they sensed slight danger . `Tail flick'

X 2 table = 1909 ; df = 36 .

PRESTON: INTERPHYLETIC COMMUNICATION

Table IV. Effects of Goby Initial Acts on A . rapax

795

started when the danger was more imminent .If the shrimp did not respond to `tail flick', thegoby `tail beat' . When danger was obvious andserious or if the shrimp did not respond to`tail beat', the goby fled into its burrow . Shrimpsalways entered first .

The frequency of `tail flick' was increased bythe shrimp act `plough' and decreased by the

Table II. Information Values for Communication Between Gobies and Shrimps

Hs = Information present (bits/act) ; HH = Information transmitted (bits/act) ; NT = Normalized transmission ; Acts =Acts/signaller/encounter.

Table III. Contribution of Individual Goby Signals to Information Transmission

Category HB Hr NT Acts Encounters

Goby communicating to shrimpA.rapax 2 . 35 0 .32 14% 2 . 2 1059A. rapacida 2 .34 0 . 35 15% 2 . 0 1005

Shrimp communicating to gobyA. rapax 2 . 40 0 . 31 13 % 3 . 2 1059A. rapacida 2 . 21 0 .31 14% 3 . 5 1005

A. rapax following acts

Goby initial acts

Directive

Inhibitive

P for X g row

Dorsals erect

> 0.250Flee

Flee, in burrow

Plough

< 0.005Guard

Plough, in burrow

Sit, flee, no change

< 0.005Move away

No change

< 0.005Nip sand

No change

< 0.005Pectorals wave

> 0.500Sit away

> 0.500Tail beat

Flee

< 0.005Tail flick

Sit, flee, no change

Plough, manipulate, withdraw, in burrow < 0 .005Tail wave

Withdraw, no change

Plough

< 0.005Withdraw

> 0.750

Goby to A. rapax Goby to A. rapacida

Signal ht Signal hr

Tail flick 0 . 180 Tail flick 0 . 181Guard 0 . 090 Guard 0 .094Flee 0 . 012 Tail wave 0 .022Tail wave 0 . 012 Move away 0 .017Moveaway 0 . 008 Flee 0 . 012Tail beat 0 . 004 Withdraw 0 . 008Nip sand 0 . 004 Tail beat 0 . 006Dorsals erect 0 . 002 Dorsals erect 0 . 005Sit away 0 . 002 Nip sand 0 . 003Withdraw 0 . 001 Sit away 0 . 002Pectorals wave 0 . 001 Pectorals wave 0 . 001

796

ANIMAL

shrimp acts `withdraw', `flee', and `in burrow'(Tables VII and VIII). Apparently `tail flick'functions only as a warning signal. A. rapacidawas persistent in remaining out of its burrowprimarily while it was `manipulating objects' .Thus `manipulate objects' was directive toward`tail beat' .

x 2table = 2040; df = 55 .

Table V. Effects of Goby Initial Acts on A. rapacida Following Acts

Table VI. Contribution of Individual Shrimp Signals to Information Transmission

BEHAVIOUR, 26, 3

For two reasons I postulate that warningsignals `tail flick' and `tail beat' have evolvedfrom the goby act `withdraw'. The most obviousreason is that gobies `withdraw' or `flee' intotheir burrows for their own safety. These thenare positive indications of danger . When a gobyswims backwards, its usual position for `with-

Table VII. Effects of A . rapax Initial Acts on Goby Following Acts

A. rapacida following acts

Goby initial acts Directive

Inhibitive

P for x 2row

Dorsals erect No change

< 0.025Flee Flee, in burrow

Plough, withdraw

< 0.005Guard Plough

Sit, flee, no change 0 . 005Move away Withdraw, no change

Plough

< 0.005Nip sand No change

< 0.005Pectorals wave > 0 . 500Sit away > 0 . 500Tail beat No change

< 0.005Tail flick Sit, flee no change

Plough, manipulate, withdraw, in burrow < 0 . 005Tail wave Withdraw, no change

Plough, in burrow

< 0.005Withdraw Flee, no change

Withdraw

< 0.005

X2 table = 2423 ; df = 36.

Goby following acts

A. rapax initial acts Directive Inhibitive P for x 2 row

Flee Guard Tail flick, move away, no change < 0 . 005In burrow Guard, nip sand, dorsals erect, tail Tail flick, no change < 0 . 005

Manipulate objects

wave, pectorals wave, move away,sit away, withdraw, flee, in burrow

> 0 .050Plough Tail flick, no change Guard, nip sand, dorsals erect, move < 0 . 005

Sit Guard, no change

away, pectorals wave, withdraw, flee,in burrowNip sand 0 .005

Withdraw No change Guard, nip sand, tail wave, move away < 0 .005flee, withdraw, in burrow, tail flick

A. rapax to goby A. rapacida to goby

Signal hr Signal ht

In burrow 0 . 101 In burrow 0 .084Plough 0 .086 Flee 0 .074Withdraw 0 .064 Ploughi 0 .073Flee 0. 045 Withdraw 0 .065Sit 0.012 Sit 0 . 006Manipulate objects 0.004 Manipulate objects 0 . 006

draw', its tail makes a motion similar to `tailflick' . `Tail beat' looks like an exaggerated `tailflick' . Thus, `tail flick' and `tail beat' may beintention movements, in the sense of Daanje(1950), which have derived signal function in thissymbiosis .`Tail wave' and `move away' are weak

warning signals, defined here as goby actswhich were directive toward `withdraw' inshrimps (Tables IV and V) . `Move away' wasdirective toward `withdraw' only for A. rapacida.The antennae of A . rapax are longer than thoseof A. rapacida and thus may maintain contactwith a goby that `moves away' to a greaterdistance from the burrow entrance . Both shrimpspecies fled into or stayed in the burrow whenthe goby was far from the entrance .

The frequency of `tail wave' increased whenA . rapacida `ploughed', but not when A. rapaxploughed. When A. rapacida `ploughed', italmost always bumped the goby severely,generally ploughing sand in a path that wentunder the goby's tail . A. rapax, on the otherhand, seldom bumped the goby . A. rapacida withits shorter antennae may maintain contact withthe goby by using all parts of its body whereasA. rapax maintains contact primarily by usingits longer antennae . This suggests that `tailwave' may be a method by which gobies keeptheir balance. `Tail wave' also occurred in theabsence of A. rapax. Thus `tail wave' may alsobe a means by which the burrow is aerated or itmay be a signal which has no immediate effecton a shrimp but communicates the presenceof a goby at the entrance of the burrow . Perhapsit was directive toward `withdraw' only becauseit resembles warning movements .

Information present was higher when A. rapaxcommunicated to gobies than when A. rapacida

PRESTON : INTERPHYLETIC COMMUNICATION

797

communicated to gobies (Table II) . That is, thegobies' behaviour was less restricted in theirassociation with A. rapax than it was in theirassociation with A . rapacida . At the same timethe amount of information transmitted byshrimps to gobies was the same for both associ-ations. The number of shrimp acts per encounterwas slightly higher for interactions between A .rapacida and gobies, yet information transmittedper encounter was approximately the same forthe two shrimp species. These results and thefact that only one bit of information wastransferred per encounter support the idea thatone yes-no question was answered for the gobyin each encounter. This was independent of thenumber of shrimp acts performed to communi-cate the message and independent of the differ-ences in the behaviour for the two shrimpspecies. More transmission probably would bewasteful .

Shrimp signals apparently informed gobieswhether or not an actively digging shrimp waspresent, that is, whether or not warning signalswere necessary in the event of danger. Theshrimp act `in burrow', the main act signallingthat a warning signal would be unnecessary,contributed the most to information trans-mission from shrimp to goby (Table VI) . Whenthe shrimp was in its burrow, the goby guardedand was active in its interactions with othergobies and in feeding (Tables VII and VIII) .When the shrimp `withdrew' or `fled', warningsignals were unnecessery since the shrimp wasentering its burrow anyway. Warning signalswere, in fact, inhibited . The shrimp acts `sit' and`flee' were directive toward `guard' in the goby .This suggests that the goby sat alert todanger which it may have just signalled waspresent . The shrimp act `sit' combined with the

Table VIII. Effects of A. rapacida Initial Acts on Goby Following Acts

Goby following acts

A. rapacida initial acts Directive Inhibitive P for x grow

Flee Guard Tail flick, withdraw, no change < 0 .005In burrow Guard, nip sand, dorsals erect, move Tail flick, no change < 0 .005

Manipulate objects

away, pectorals wave, sit away, with-draw, flee, in burrowTail beat 0 .025

Plough Tail wave, tail flick, no change Guard, nip sand, withdraw, in burrow 0 .005Sit No change < 0 .005Withdraw No change Guard, nip sand, tail wave, tail flick, < 0 .005

move away, withdraw, in burrow

x2 table = 1978 ; df = 45 .

79 8

goby act `guard' may be a strategy for conceal-ment. `Tail flick' was the only act directivetoward `sit' (Tables IV and V). Shrimps andgobies are difficult to see when they are still .If the goby `guards' and the shrimp `sits', andif the pair was not seen previously, they may notbe detected by an observer . Thus warningsignals would be unnecessary for a `sitting'shrimp. `Sit' and `guard' consume less energythan `withdraw' or `flee' if the animals sit stilluntil danger passes and then continue theiractivity . Thus, `sit' and `guard' appear to behighly adaptive behavioural acts .`Guard' was the only goby act directive

toward digging behaviour in shrimps (TablesIV and V). `Guard' was directive toward `inburrow' for A. rapax but not for A. rapacida .A. rapacida often stayed out of its burrow,`manipulating objects', until it `fled' in responseto a warning signal . A. rapax, on the other hand,`manipulated objects' for shorter periods (quali-tative observations), although with the samefrequency as A. rapacida (P > 0 .250), andentered its burrow without necessarily havingbeen warned of danger. In fact, `flee' contributedquite a bit more to information transmitted fromA. rapacida to gobies than from A. rapax togobies (Table VI). For both species of shrimps,`guard' was inhibitive toward inactivity and`flee' (Tables IV and V) . This suggests that`guard', in certain contexts, communicates toshrimps that there is no danger .

Certain goby acts appear to have no signifi-cant effect on shrimp behaviour (insignificantx 2row or directive toward `no change') for one ofseveral reasons . (1) The goby act may actuallyhave had no effect on shrimp responses, i .e .it had no communicative function . (2) Thenumber of times responses occurred may havebeen too small to show the true effect of theinitial act. (3) The goby act may have had aneffect that was not immediately expressed, i .e .there may have been a latency in the character-istic response . (4) The shrimp's previous responsemay have been the act with which it wouldnormally respond to the goby's new signal .It therefore continued to perform that act andregistered `no change' in response to the goby'sact. (5) The observer was unable to detect aresponse which actually occurred, for example,a chemical response .

`Sit away' had no significant effect on thebehaviour of either A. rapacida or A. rapax(Tables IV and V), yet it showed a significanteffect when the data for the two species were

ANIMAL BEHAVIOUR, 26, 3

combined (P < 0 .025). This suggests that thenumber of times it occurred for the separatespecies of shrimps was too small to show theeffect. In the combined data analysis, `sit away'was directive toward `flee' . Although it wasrelatively unimportant in the overall trans-mission of information from goby to shrimp(Table III), `sit away' may communicate theabsence of a warning system .

Goby signals `withdraw' and `move away'had significant effects on A. rapacida responsesand no significant effects on A. rapax responses .`Tail beat' had significant effects on A . rapaxresponses and no significant effects on A .rapacida responses (Tables IV and V). For`withdraw' the combined data analysis showsthe same effect as for A. rapacida alone . `Tailbeat' and `move away' have the same effectas for A. rapax alone . This suggests that the gobyacts `withdraw' and `tail beat' actually did havean effect on the responses of both shrimpspecies. At least it indicates that the responsesof A. rapax to `withdraw' and of A. rapacidato `tail beat' were not sufficiently different toalter the significance of the signals . The shrimpstended to `flee' in response to both signals . Thegoby act `move away' apparently had an effecton the behaviour of A. rapacida but not on thebehaviour of A. rapax . That is, the response ofA. rapax was sufficiently different from that ofA. rapacida to alter the results . This follows sinceA . rapax, with its long antennae, could maintaincontact with the goby even when it `movedaway' . A . rapacida, with its shorter antennae,would lose its warning system when the goby`moved away' . A. rapacida would thus 'with-draw' into its burrow.

`Pectorals wave' has insignificant x 2 valuesin the combined analysis, as well as in theseparate analyses (Tables IV and V), for itseffect on the shrimps' behaviour . Although thefrequency of `pectorals wave' is not great, it isprobably large enough for an effect to beindicated. This is suggested by the fact that otheracts ('tail beat', `sit away') with lower frequenciesof occurrence did show significant effects onresponses. `Pectorals wave' may have had noeffect because it may not be a display withcommunicative signal value. It did, in fact,contribute the least to information transmissionfrom goby to shrimp (Table III) . It occurredwith significantly greater frequency in gobiesassociated with A. rapax than in those associatedwith A. rapacida (P < 0 . 025). Because A. rapaxis generally larger than A. rapacida, it may

consume more oxygen and thus more rapidlydeplete the oxygen supply in the burrow.`Pectorals wave' may be a method of acceleratingventilation in the goby's gill slits (Kinzer 1960),or it may be a means of aerating the burrow .On the other hand, it could have had an effecton the shrimps' behaviour that was not immedi-ately expressed. The shrimp act 'in burrow' wasdirective toward `pectorals wave' (Tables VIIand VIII) . It may have communicated to ashrimp inside the burrow that a goby waspresent at the burrow entrance . The goby mayhave performed other acts before the shrimpactually exited, in which case the effect wouldnot be indicated in this type of analysis .

`No change' is an unusual category since itindicates the continuation, rather than theinitiation, of any act. The frequency of `nochange' in the shrimp's behaviour was increasedby `dorsals erect', `move away', `nip sand', `tailbeat', `tail flick', `tail wave', and `withdraw'(Tables IV and V) .

`Dorsals erect' is a visual signal . The shrimpcould not feel that the goby's dorsal fins wereerect since they did not wave and the shrimpdid not place its antennae on them . This actwas probably not sensed by shrimps and, there-fore, directed `no change' .

`Nip sand' is a feeding and perhaps a displace-ment act which occurred when one goby inter-acted with another goby . It may be sensed by ashrimp since it is a jerky movement of the bodywhich could be detected with the shrimp'santennae. It may have communicated `no danger'since a goby that nipped sand was generallyeither feeding or interacting with another goby .The shrimp's behaviour would thus go un-changed .When `no change' followed any warning

signal, weak or strong, I think the signal wasignored, not sensed, or the shrimp was already`withdrawing', `sitting', or `fleeing' . If the shrimpwas not touching the goby with its antennaeor if the shrimp was too far away from the gobyto sense the water displacement, warning signalsprobably were not sensed. Shrimps may haveignored a signal if the goby had signalled manytimes in sequence and did not flee itself .The goby act `no change' occurred with a

significantly higher frequency in gobies associ-ated with A. rapacida than with A. rapax(P < 0 .005). This suggests that P . mainlandiwas less responsive to the movements of A .rapacida than to the movements of A . rapax .Perhaps this results from the gross body contact

PRESTON: INTERPHYLETIC COMMUNICATION

799

between A. rapacida and the goby, makingdifferent body movements of the shrimp difficultto distinguish. Thus the shrimp's behaviourmight change without the goby's sensing thechange. Or, more likely, the messages that wereof significance to the goby were the presence orabsence of an actively digging shrimp near theburrow entrance. Thus the goby respondedone way to the presence of the actively diggingshrimp and the goby's behaviour did not alterwhile the shrimp continued digging, independentof the specific act the shrimp was using .

In its interactions with gobies, A. rapacidaperformed the acts `plough' and `flee' withsignificantly higher frequencies than did A .rapax (P < 0 .050 and P < 0 .005, respectively) .A. rapax responded to the goby by `no change'in its behaviour with a significantly higherfrequency than did A . rapacida (P < 0 .005) .Although the elicitation of the responses wasnot statistically significant, A. rapacida fre-quently responded to generalized movements ofthe goby by fleeing whereas A. rapax continuedwhat it was doing in response to the samemovements.

In the analysis for gobies communicating toshrimps, information present is essentially thesame for the two species of shrimps (Table II) .Information transmitted and normalized trans-mission were only slightly higher for gobiescommunicating to A . rapacida. On average,gobies performed slightly more acts per encoun-ter when associated with A. rapax than whenassociated with A. rapacida. Although in somecases the differences between the two species aresmall, the combined results from this studysuggest that A. rapacida was more sensitive to thegobies' movements whereas the behaviour ofA. rapax appeared to be more independent fromthe gobies' behaviour . At least three possibleinterpretations of these associations can be made .

(1) The symbiosis between gobies and A .rapacida is more highly developed than thesymbioses between gobies and A. rapax sinceA . rapacida responds to generalized, as well as tospecialized, movements of gobies . A. rapax, incontrast, is less dependent on the goby's beha-viour. This interpretation would lead one tobelieve that the A . rapax-goby symbiosis is morerecently developed than the A . rapacida-gobysymbiosis.

(2) The symbiosis between gobies and A. rapaxis more highly developed since A. rapax respondsprimarily to specialized warning signals of gobieswhereas A. rapacida responds also to more

800

generalized goby movements. With this inter-pretation, one would be led to believe that theA. rapacida-goby symbiosis is more generalizedand, therefore, more recent .

(3) Because of differences in the length ofantennae of the two species of shrimps, twoslightly different behavioural symbioses mayhave developed. A. rapax with its long antennaecan dig at a relatively greater distance from agoby and still maintain contact . Its antennaedistinctly detect a difference between a general-ized movement of a goby and a `tail flick' . A .rapacida with its relatively short antennae hasmore body contact with the goby . Generalizedand specialized movements of gobies cannot beeasily distinguished by this more gross contact .With this interpretation, a statement as to whichsymbiosis is more recent or more specialized isunjustified .Not all methods of communication were

considered in these analyses . Almost certainlythere was no auditory communication betweengobies and shrimps . I would have heard theshrimps snap . In laboratory observations inwhich a hydrophone was used, I never heardgobies make sounds either when with othergobies or when with shrimps .

Karplus et al . (1972) note that, in a goby-shrimp symbiosis similar to the one reportedhere, the shrimp Alpheus djiboutensis is chemi-cally attracted to the goby Cryptocentruscryptocentrus . A similar chemical communica-tion may occur between P. mainlandi and itssymbioticYpartners . However, considering thespeed with which responses to warning signalstook place, I doubt that they involved chemicalcommunication. It is possible that a goby'spresence or absence from the burrow entrancewas transmitted to shrimps by chemicalmeans. Also, a goby, having sensed dangerwhile the shrimp was in its burrow, mayhave communicated danger to the shrimp bychemical means . It is clear, nevertheless, that thetactual communication system in itself canadequately explain the warning behaviour .

Karplus et al. (1972) also report that C .cryptocentrus is visually attracted to its shrimpsymbiont. P. mainlandi and its symbionts mayalso communicate visually. In the warningsystem, however, the communication of shrimpto goby is most likely to be tactual, since duringmost of their communication the shrimp isbehind the goby. Tactual communication ofshrimp to goby would free the goby's visual

ANIMAL BEHAVIOUR, 26, 3

senses for attention to predators or to intrudinggobies .

After careful observation and analyses, itseems clear that because the goby uses theburrow prepared by the shrimp, the energy thegoby would otherwise need to spend in buildinga shelter can go into guarding the burrow againstpredators and into defending its shelter againstthe takeover by another goby. With the presenceof the goby to warn it of predators, the shrimpcan put more energy into shelter preparation andmaintenance. There may, however, be otherways that the goby and shrimp benefit from oneanother.

Karplus et al. (1974) found that gobies, in thegoby-shrimp associations they studied, con-trolled to some extent the location of a burrowentrance by pushing through the substratum tothe surface where burrows were not located .Gobies also determined the direction theirshrimp would plough by the orientation of theirbody in the burrow. Gobies would orient awayfrom prospective goby opponents . Shrimpsgenerally ploughed sand in a line parallel tothe body of the goby.Judging from this and from my field and

laboratory studies, I feel that the spacing ofgoby-shrimp burrows in nature is a result of theinterweaving of three communication systems :communication between gobies and shrimps,communication between gobies, and communi-cation between shrimps .Gobies placed in a large aquarium with no

shelters increase distances between individualsby going up onto the sides of the aquariumor even jumping out. When provided withshelters, each goby sets up a territory with itsshelter as the focal point . By entering theirshelters, gobies avoid or escape from visualcontact with an opponent. Since these sheltersare provided in nature by shrimps, it can beconcluded that because of their associationwith shrimps, gobies live closer together thanthey could without shelters, but only as closetogether as their visual communication systemwill allow .

When shrimps are placed in a large aquarium,they walk around examining the substratum andgenerally touch one another before they moveapart. Thus, they frequently dig burrows closeto one another. In nature, if their burrowentrances are too close together, shrimps willnot have a goby guarding them. Because theythen lack tactual communication from gobiesabout the presence of danger, they are taken

readily by predators, as indicated by the easewith which I could capture unguarded shrimps .The result is a compromise between maximaldispersion of gobies without shelters andclumping of shrimps in the absence of predators .Thus the complex communication system in thismutualistic symbiosis not only provides the gobywith shelter and warns the shrimp of predators,but also results in optimal spacing of burrows .

The insights obtained from the use of informa-tion theory and associated x2 analyses in thisstudy of interphyletic information exchangesuggest that the use of these techniques shouldnot be restricted to intraspecific studies. Muchknowledge about other symbioses, from mutual-istic to parasitic associations, could be gainedby their use.

AcknowledgmentsThis study is Hawaii Institute of Marine BiologyContribution Number 520 . It is based on mydissertation (Moehring 1972) submitted to theGraduate School of the University of Hawaiiin partial fulfilment of requirements for thePh.D. degree . It benefited a great deal from theinsights and moral support of Ernst S . Reese .I would like to express my appreciation forthese. Earlier forms of the manuscript weregreatly improved from the advice of Michael G .Hadfield, Brian A . Hazlett, Louis M . Herman,George S. Losey, Eric M. Preston, Ernst S .Reese, Howard Schein, Albert L . Tester, andthree anonymous reviewers. Eric Preston andSteve Smith helped me with the computerprogramming. D. M. Banner identified theshrimps and W . J . Baldwin identified the goby .I would like to extend my thanks to all of thesepeople .

REFERENCESAltmann, S. A. 1965. Sociobiology of rhesus monkeys .

II : Stochastics of social communication . J.Theor. Biol., 8, 490-522.

Attneave, F . 1959. Applications of Information Theoryto Psychology . New York: Holt, Rinehart &Winston.

Balasch, J. & Mengual, V. 1974. The behavior ofDardanus arrosor in association with Calliactisparasitica in artificial habitat . Mar . Behav.Physiol., 2, 251-260.

Baldwin, W . J. 1972 . A new genus and new species ofHawaiian gobiid fish . Pac. Sci., 26,125-128 .

Bayer, F . M. & Rofen, R. P. H. 1957. Project coralfish looks at Palau . Rep. Smithson . Inst ., 1956,481-508 .

Blachman, N . M. 1968 . The amount of information thaty gives about x. I.E.E.E. Trans. Inf. Theory, 14,27-31 .

PRESTON: INTERPHYLETIC COMMUNICATION

801

Chatfield, C . & Lemon, R . E . 1970. Analysing sequencesof behavioural events . J. Theor. Biol., 29, 427-445 .

Cherry, C . 1966. On Human Communication . 2nd edn .Cambridge, Mass.: M.I.T. Press .

Daanje, A . 1950 . On locomotory movements in birdsand the intention movements derived from them .Behaviour, 3, 48-98 .

Dingle, H. 1969 . A statistical and information analysisof aggressive communication in the mantis shrimpGonodactylus bredini Manning. Anim. Behav ., 17,561-575 .

Dingle, H . 1972 . Aggressive behavior in stomatopodsand the use of information theory in the analysisof animal communication . In : Behavior of MarineAnimals, Vol . 1 . Invertebrates (Ed. by H . E . Winn& B. L . Olla), pp. 126-156. New York: PlenumPress .

Feder, H. M. 1966 . Cleaning symbiosis in the marineenvironment. In : Symbiosis I. (Ed. by S. M .Henry), pp . 327-380. New York: Academic Press.

Haldane, J . B . S. & Spurway, H. 1954. A statisticalanalysis of communication in Apis mellifera anda comparison with communication in otheranimals . Insectes soc., 1, 247-283 .

Harada, E . 1969 . On the interspecific association of asnapping shrimp and gobioid fishes . Publ. SetoMar. Biol. Lab., 16, 315-334.

Hazlett, B . A. & Bossert, W . H. 1965. A statisticalanalysis of the aggressive communications systemsof some hermit crabs . Anim. Behav ., 13, 357-373 .

Hazlett, B. A. & Bossert, W. H. 1966. Additionalobservations on the communications systems ofhermit crabs. Anim. Behav., 14, 546-549 .

Hazlett, B. A. & Estabrook, G. F. 1974a . Examinationof agonistic behavior by character analysis . I .The spider crab Microphrys bicornutus. Behaviour,48, 131-144.

Hazlett, B . A. & Estabrook, G . F . 1974b. Examinationof agonistic behavior by character analysis . II .Hermit crabs. Behaviour, 49, 88-110.

Herman, L. M . 1965 . Study of the single channel hypo-thesis and input regulation within a continuous,simultaneous task situation . Q . J1 . exp . Psychol.,17, 37-46.

Karplus, I ., Szlep, R . & Tsurnamal, M . 1972 . Associativebehavior of the fish Cryptocentrus cryptocentrus(Gobiidae) and the pistol shrimp Alpheus djibou-tensis (Alpheidae) in artificial burrows . MarineBiol., 15, 95-104.

Karplus, I ., Szlep, R . & Tsurnamal, M . 1974. Theburrows of Alpheid shrimp associated withGobiid fish in the Northern Red Sea . MarineBiol., 24, 259-268 .

Karplus, I ., Tsurnamal, M. & Szlep, R. 1972 . Analysisof the mutual attraction in the association of thefish Cryptocentrus cryptocentrus (Gobiidae) andthe shrimp Alpheus djiboutensis (Alpheidae) .Marine Biol., 17, 275-283 .

Kinzer, J .1960. Zur Ethologie and Biologie der Gobiiden,unter besonderer Beruucksichtigung der Schwarz-grundel Gobius jozo L . Zool. Beitr. N.F., 6, 207-290.

Klausewitz, W. 1960 . Fische aus dem Roten Meer . IV .Einige systematisch and okologisch bemerkens-werte Meergrundeln (Pisces, Gobiidae). Sencken-berg . biol., 41, 149-162 .

802

ANIMAL BEHAVIOUR, 26, 3

Klopfer, P. H. & Hatch, J. J . 1968 . Experimental con-siderations. In : Animal Communication : Techniquesof Study and Results of Research (Ed. by T. A.Sebeok), pp . 31-43. Bloomington, Ind . : IndianaUniversity Press .

Luther, W. 1958. Symbiose von Fischen (Gobiidae)mit einem Krebs (Alpheus djiboutensis) im RotenMeer. Z. Tierpsychol., 15, 175-177.

MacFarland, C. G. & Reeder, W. G. 1974 . Cleaningsymbiosis involving Galapagos tortoises . Z .Tierpsychol., 34, 464-483 .

Mariscal, R. N. 1972 . Behavior of symbiotic fishes andsea anemones . In : Behavior of Marine Animals .Vol . 2. Vertebrates (Ed. by H . E. Winn & B. L .Olla), pp. 327-360. New York : Plenum Press .

Marler, P. 1961 . The logical analysis of animal communi-cation . J. Theor. Biol., 1, 295-317.

Moehring, J. L. 1972. Communication systems of agoby-shrimp symbiosis . Ph.D. thesis, Universityof Hawaii (Honolulu .)

Morris, C. W. 1946 . Signs, Language, and Behavior .New York : Prentice Hall .

Quastler, H. 1958 . A primer on information theory . In :Symposium on Information Theory in Biology(Ed. by H . P. Yockey), pp. 3-49. New YorkPergamon Press .

Rubenstein, D. I. & Hazlett, B . A . 1974. Examinationof the agonistic behaviour of the crayfish Orcon-

ectes virilis by character analysis . Behaviour, 50,193-216 .

Shannon, C. E. & Weaver, W. 1949. The MathematicalTheory of Communication . Urbana, Ill . : Universityof Illinois Press .

Singh, J. 1966 . Great Ideas in Information Theory,Language, and Cybernetics. New York : DoverPublications, Inc.

Smith, W. J. 1968. Message-meaning analyses. In :Animal Communication : Techniques of Study andResults of Research (Ed. by T. A. Sebeok), pp .44-60. Bloomington, Ind . : Indiana UniversityPress .

Snedecor, G. W. & Cochran, W. G. 1967 . StatisticalMethods. 6th edn. Ames, Iowa : Iowa StateUniversity Press .

Steinberg, J . B . & Conant, R . C . 1974 . An informationalanalysis of the intermale behaviour of the grass-hopper Chortophaga viridtfasciata . Anim. Behav .,22, 617-627 .

Wilson, E. O. 1962 . Chemical communication amongworkers of the fire ant Solenopsis saevissima (Fr .Smith) . 2. An information analysis of the odourtrail . Anim . Behav ., 10, 148-158 .

(Received 7 November 1974 ; revised 19th May 1976 ;2nd revision 22 November 1976 ; MS. number: A1636)