Estimating ancestral states of a communicative display

22 Pages·1998·0.26 MB·English

Anim.Behav.,1998,55,1685–1706
Estimating ancestral states of a communicative display: a comparative study
of Cyclura rock iguanas
EMuILIA P. MARTINS & JENNY LAMONT
DepartmentofBiology,UniversityofOregonatEugene
(Received15August1997;initialacceptance22September1997;
finalacceptance16October1997;MS.number:a7862r)
Abstract.Inthisstudywedescribethesignatureheadbobdisplaysofsevenoftheeightextantspecies
ofCycluraiguanasusingdatacollectedfromthefieldandfromcaptiveanimals.Weusedphylogenetic
comparative methods to estimate the ancestral states of several measures of the headbob displays,
includingnumberofheadbobsandthedurationofheadbobsandinter-bobpauses.Divergenceinthe
headbobdisplayamongspecieshasbeensubstantial,withsomemajorchangesoccurringwithinonlya
few(aboutsix)generations.Otherwise,resultsareconsistentwiththoseobtainedpreviouslyforother
lizards which suggest that there is an evolutionary limit on the total duration of headbob displays.
DiVerencesintheresultsobtainedusingdiVerentphylogeneticmethodssuggestthatalthoughestimates
of ancestral states are reasonably robust to violations of evolutionary assumptions, we cannot
determine the standard errors of those ancestral phenotypes accurately without more detailed
information about the types of forces (e.g. selection, drift) underlying evolutionary change in these
traits. In particular, within-species variation had a substantial impact on the standard errors of
estimatedancestralstates,andshouldbeincludedinsuchestimationswheneverpossible.Finally,our
results emphasize the importance of conserving behavioural as well as genetic diversity in trying to
preserveendangeredspeciesforpossiblereintroductionintothewild.
?1998TheAssociationfortheStudyofAnimalBehaviour
The phylogenetic comparative method has been hypothetical ancestors on a phylogeny, we can
used throughout animal behaviour, but has been determine the order of evolutionary changes and
particularly useful in generating hypotheses and uncover any evidence of long-term trends in the
guiding experimental studies of the evolution of evolutionary history of the displays. We can also
communicationsystems(e.g.Sille´n-Tullberg1988; determinethegeneraltimescaleatwhichchanges
Basolo1990;Prum1994,1997;Ryan1995;Endler are occurring, thereby providing information on
&Thery1996).Althoughcommunicativedisplays whether it would be more profitable to conduct
areoftendefinedbytheirstereotypednature,they future studies at the population, species, genus,
are also quite variable, with diVerent species, or higher taxonomic level. In the current study,
populations, and even individuals producing we use a phylogenetic comparative analysis of
remarkably diVerent displays. This variation Cyclura lizards to consider evolutionary changes
may be due to a number of factors, including indisplaystructure.Indoingso,weillustratethe
ecological, genetic and behavioural diVerences. useofanewphylogeneticcomparativemethodfor
Phylogeneticcomparativestudiesprovideuswith estimating ancestral states (Martins & Hansen
a uniquely powerful tool to begin to tease apart 1997), and discuss its application to the study of
these possibilities and to highlight species and animalbehaviour.
traits that may be particularly important in Cyclura iguanas exhibit an impressive diversity
future empirical studies (see reviews in Brooks & of social behaviour and communicative displays,
McLennan1991;Losos1996;Martins1996c).By and thus oVer an excellent opportunity to study
reconstructing the communicative behaviour of the evolution of these traits. The eight species of
Cyclurarockiguanasarefoundonseveralislands
Correspondence:E.P.Martins,DepartmentofBiology,
UniversityofOregon,EugeneOR97403,U.S.A.(email: in the Caribbean, including Cuba, Jamaica,
[email protected]). Hispaniola and the Bahamas. These iguanas are
0003–3472/98/061685+22$25.00/0/ar970722 ?1998TheAssociationfortheStudyofAnimalBehaviour
1685
1686 AnimalBehaviour,55,6
quite large (ranging from 30.6 to 74.5cm maxi- raised and lowered and the relative duration
mum male snout–vent length, Schwartz & of up-and-down motions and pauses. In other
Henderson 1991), primarily herbivorous, and species of Iguania, bobbing displays are often
endangeredduebothtohabitatdestructionandto species-typical, with the number and type of
the introduction of exotic predators (e.g. feral bobs in each display containing information
cats, dogs and pigs) which prey on eggs and about the individual, sex and species identity of
juvenilelizards(Iverson1978).StudiesofCyclura the animal as well as about the type of social
behaviourarethusofconsiderablepracticalinter- contextinwhichtheanimalisengaged(Carpenter
est to conservation agencies raising these animals & Ferguson 1977; Martins 1993a; DeCourcy &
incaptivityforfuturereintroductionintothefield. Jenssen1994).
Although two species of Cyclura are found on Thephylogeneticcomparativemethodisapow-
Hispaniola, most of the Cyclura are not sym- erfultoolforinferringtheevolutionaryhistoryof
patric, and the genus exhibits an impressive behaviouralphenotypes(seeBrooks&McLennan
diversity of social behaviour types. In some 1991;Harvey&Pagel1991;orMartins&Hansen
species(e.g.Cycluracornutastejnegeri:Wiewandt 1996 for review). Nevertheless, early techniques
1977),bothsexesdefendallormostoftheirhome (e.g. parsimony) developed primarily for the
ranges against other individuals of their sex. analysisofmolecularormorphologicalcharacters
This is the pattern found in most lizards, and is are not well suited for inferring the evolution of
consideredtobeevolutionarilyprimitive(Martins many behavioural traits. Behavioural traits are
1994). Other species (e.g. Cyclura cychlura: usually thought to be more flexible, responding
Windrow1977)exhibitpersonalspaceaggression, quickly to the action of natural and sexual selec-
in which individuals do not defend any particu- tion(butseedeQueiroz&Wimberger1993;Irwin
lar geographical range, but will aggressively 1996; Wimberger & de Queiroz 1996). More
defend the area immediately around them. Still recently,we(Martins&Hansen1997)developeda
other species (e.g. the Cuban Cyclura nubila formofphylogeneticregressionwhichcanbeused
nubila: A. Alberts, personal communication) live toestimateseveralparameters(e.g.thecorrelation
in groups of up to 20 individuals which interact between two traits, the rate of phenotypic
frequently. evolution, ancestral states) from comparative
Although Cyclura iguanas produce several data while applying any one of a variety of
visual displays including tail waves and various assumptions developed for the specific characters
body postures, we concentrated entirely on their under analysis. Using this method, we can esti-
bobbing displays, in particular, the Broadcast mate various evolutionary parameters while
Display(alsotermed‘signaturebob’or‘Assertion assuming more realistic models of behavioural
display’) for each species. Most other lizards of evolution such as the models of stabilizing selec-
the Iguania group use headbob and push-up dis- tion,fluctuatingdirectionalselectionandrandom
plays in both territorial defence and maintenance genetic drift summarized by Hansen & Martins
(e.g.Carpenter&Ferguson1977).Althoughsome (1996).
species have more than one pattern of bobbing In the current study, we collected information
display, most produce a single broadcast display on the bobbing displays of nine populations
most often, and several studies have shown that (seven of the eight species) of Cyclura, filming
thesedisplayscontaininformationabouttheindi- animalsinthefieldandincaptivityandanalysing
vidual identity of the animals (e.g. Crews 1975; videotapes gathered by other researchers. Over-
Rothblum&Jenssen1978;Martins1991).During laying these data on an independent phylogeny
these displays, animals move their heads in a (Malone & Davis, unpublished data), we infer
stereotyped fashion, usually raising and lowering the sequence of evolutionary changes underlying
them in a series of headbobs or tosses. Unlike present day diversity of communicative behav-
some smaller lizards which produce the up-and- iour. We consider a variety of assumptions
down motion by extending and flexing their legs regarding the evolutionary processes underlying
(i.e. a ‘push-up’ display), iguanas produce the phenotypic evolution and incorporate within-
motionusingprimarilythemusclesintheirnecks. species variation into the analysis using general-
In the current study, we focused on the ‘bob ized least-squares regression (Martins & Hansen
pattern’,thatis,thenumberoftimestheheadwas 1997),andthusalsoillustratetheapplicationofa
Martins&Lamont:AncestralstatesofdisplaysinCycluraiguanas 1687
new phylogenetic comparative method to the anadditional10wereproducedbyjuvenilemales,
studyofanimalbehaviour. and only one display was produced by a large,
adult male. With the exception of the display
produced by the large male, all headbob displays
METHODS
were produced by animals either approaching or
being approached by larger individuals of either
DataCollection
sex,asthisseemstobetheC.carinataequivalent
We conducted behavioural observations in the ofasignaturedisplay.
field during two short trips to the Caribbean, Also in August 1995, we recorded 17 displays
gatheredvideotapesmadebyotherresearchersin from 15 adult male C.nubila nubila during about
thefield,andobservedsomeanimalsincaptivity. 15h of observation on Isla Magueyes in Puerto
Together, the data set consists of 204 displays Rico.CycluranubilanubilaisnativetoCuba,but
produced by approximately 101 animals. All asmallgroupofanimalswasreleasedfromazoo
behavioural observations were made from a dis- on Isla Magueyes in the mid 1960s (Christian
tance of at least 5m during the main activity 1986) and have been reproducing there as a
period (0900–1200 hours and 1400–1800 hours). free-ranging population. Alison Alberts also
Whenever possible, these observations were generously provided videotapes from her
recorded on videotape and in field notes during long-term field study of C.nubila nubila on
focalanimalsamples.Videotapedrecordingswere Guanta´namoBay,Cuba(unpublisheddata)from
made using a Canon L-2 Hi-8 video camera with whichweobtainedrecordingsof40displaysfrom
a Canon 15# macro zoom lens. Hand-written 24 animals. In November 1995, we observed
notesconcentratedondescribingbobbingdisplays C.rileyi for about 13h on Green Cay (near San
in terms of the number and type of bobs in the SalvadorIsland)intheBahamas,recordingatotal
display,anddidnotincludeestimatesoftemporal of 17 displays produced by 12 individuals (11
duration. The observer’s (E.P.M.) ability to adultmalesandonehatchling)onvideotape.
describe the bobbing displays in hand-written Thomas Wiewandt provided a film he made
notes was tested periodically against videotaped during a long-term field study of C.cornuta stej-
recordings,andwasfoundtobecompletelyaccu- negeri on Mona Island, Puerto Rico (Wiewandt
rate in 20 of 20 trials. In most cases, individuals 1977, 1981). This film includes 14 displays pro-
were easily distinguished on the basis of unique duced by about 10 adult animals. Because these
morphologicalfeatures,andanattemptwasmade animals are sexually monomorphic (Wiewandt
to observe as many diVerent animals as possible. 1977), it is impossible to determine from the film
Becauseadultmalesaresubstantiallymoreactive exactly how many males and females displayed.
than other age and sex groups in most species, Fromthenarration,itappearsthatdisplayswere
more data were collected from males than from obtained from both males and females in both
femalesofmostspecies,andveryfewjuvenileand aggressive interactions (males defending their
hatchlinganimalsarerepresented.Totalnumbers matesandfemalesdefendingnestsites)andcourt-
of individuals above are estimates based both on ship contexts. Thus, here, as in C.carinata, the
our ability to distinguish individuals and on the displays are not necessarily all ‘signature’ head-
geographicaldistancebetweenanimals. bobs, and may be subject to more within-species
During about 20h of observation in August variation because of the combined contexts.
1995,werecorded30displays(fouronvideotape, Finally, it is important to note that the Mona
26 in hand-written notes) produced by approxi- Islandiguanaisquitedistinctinmanywaysfrom
mately23individualC.carinataonWaterCayin the Hispaniolan C.cornuta cornuta, such that
theTurksandCaicosIslands.Cycluracarinatais these data should not be considered necessarily
ratherunusualinthatthetypicalheadbobdisplay representativeofthespeciesasawhole.
is produced more often by females and juvenile John Iverson also provided videotapes of
animals than by adult males, and because the C.cychluramadeduringhislong-termfieldstudy
displayisoftenproducedinwhatappearstobean on Allen’s Cay, Bahamas (unpublished data). A
appeasement rather than an aggressive or broad- total of 17 signature displays produced by
cast context (Iverson 1979). Nineteen of the 30 about 11 adult males was extracted from these
recordeddisplayswereproducedbyadultfemales, videotapes.
1688 AnimalBehaviour,55,6
It was not possible to observe the behaviour transferred clips of videotaped headbob displays
of two exceedingly rare species (C.collei and onto a PowerPC using Radius Videofusion hard-
C.ricordi)inthewild.Thus,videotapesofavery waretocapture15frames/softheimage.Wethen
small number of individuals were obtained at recorded the vertical height of the tip of the
the Indianapolis Zoo in March 1996. The Zoo lizard’s nose (in the case of headbob displays) or
has three juvenile (two females and one male) the highest point of an eye ridge (in the case of
C.collei, and three C.ricordi (one adult male head rolls, see description of C.cornuta displays
‘Robbie’, and two juvenile females). All of the below) for each frame of the videotape. We also
juvenile animals were quite wary and relatively recordedtheverticalheightofastationaryobject
inactive. During about 8h of observation, we on the film (usually a stationary part of the
recorded only three displays produced by the animal’sbody),andcalculatedtherelativevertical
juvenile male C.collei. In contrast, 54 displays motion of the lizard’s head as vertical deviation
were recorded from the adult male C.ricordi between the stationary spot and the tip of the
during about 6h of observation. Videotapes of lizard’snoseoreyeridge.Wethenplottedrelative
three individuals of the rare Grand Cayman vertical motion against time to obtain a Display-
Islandsubspecies,C.nubilalewisi,werealsomade Action-Pattern graph (DAPgraph, Carpenter &
at the Indianapolis Zoo. During about 3h of Grubitz 1961) for each display (e.g. Fig. 1) from
observation, we recorded 10 signature displays which measures such as the number and timing
from one adult male and one adult female. of headbobs could be recorded. Note that
Although C.pinguis can still be observed on although this procedure minimizes the eVects of
Anegada Island in the British Virgin Islands, we anysmallcameramotionduringthefilmingofthe
were unable to do so during this study. Because display, we did not record detailed information
theseanimalsarealsonotincaptivityatanyzoo about the distance from the animal at each point
(Christie1995),wedidnotobtainanyvideotaped in time nor the angle at which the camera was
recordingsatallfromthisspecies. held,andthuswereunabletodetermineabsolute
Although we tried to consider only signature vertical height of the head motion from our
headbobs,displaysinthisstudywereproducedby recordings.WehaveschematizedtheDAPgraphs
animals in very diVerent conditions, including primarilytoemphasizethelackofaccuracyinour
animals in captivity and animals that had been measurements of the vertical height of display
subjected to a large degree of human contact. components.
Thus, we make no attempt to interpret the func- The term ‘headbob’ in this paper refers to an
tion or meaning of displays from this data set. up-and-downmotioninwhichthetipofthenose
Furthermore, because the sample size of displays israisedverticallyuntilthejawformsanangleof
withinanyonespecieswassmall,andoccasionally at least 30) with the substrate the animal is
came from measures of a single individual, cau- standing on. The nose may be held up for a
tion should be used in interpreting the fine struc- short time before again being lowered to the
ture of these displays in any detail. Fortunately, horizontal position and pulled in towards the
despite the variability among individuals and body. Each up-and-down motion (‘headbob’)
populations of lizards, interspecific diVerences in takes slightly less than 1s to complete, and may
the bobbing displays of other lizards are usually occurinaseriesoftwoormoreproducedinrapid
far greater than that within species even across succession.Theterm‘bout’willbeusedtoreferto
contexts (e.g. Martins 1993a,b). Estimates of the each series of up-and-down motions, separated
standard error of existing measures (see Results) from other headbobs by pauses of at least 0.13 s
show that the same is usually true of Cyclura (two frames of the digitized videotape). A bout
iguanas.Inotherwords,althoughthesamplesize might thus be a double headbob, triple headbob,
is small, it is suYcient to discover a number of or even a long trill of up to 15 up-and-down
interestingevolutionarypatterns. motions separated by pauses of no more than
0.07s (one frame of the videotape). Thus, each
display consists of one or more bouts of head-
VideotapeAnalysis
bobs. Iguanas often hold their mouths open,
Videotapes were digitized on a computer and raise their tails, or adopt other special body
subjected to frame-by-frame analysis. First, we postures while producing headbobs, but during
Martins&Lamont:AncestralstatesofdisplaysinCycluraiguanas 1689
C. nubila (Cuba)
C. nubila (P.R.)
C. nubila lewisi
C. cychlura
C. rileyi
C. carinata
C. collei
C. ricordi
C. cornuta
Figure1.Examplesof‘typical’displaysforthesevenspeciesofCycluraobservedinthisstudy.Horizontalaxisistime
(inseconds),andverticalaxisisaschematicdepictionoftheheightoftheanimal’snoseoreyeridge(forC.cornuta
stejnegeri)duringthedisplay.ThesefiguresarecomparabletoDAPgraphs(Carpenter&Grubitz1961).
the headbob display, the rest of the body is held (Martins&Hansen1997)leavesoutmanyofthe
quitestill. details for estimating ancestral states, we go
throughthisprocedureinsomedetailbelow.
This generalized least-squares approach
PhylogeneticComparativeAnalyses
(Martins & Hansen 1997) requires first that we
We used phylogenetic comparative analyses to describe the questions of interest in terms of
reconstructancestralstatesofvariousmeasuresof standard regression models. Thus, we estimated
the headbob pattern (headbob, pause, and total ancestral states using the model: A=WY+(cid:229),
durations,totalnumberofup-and-downmotions where A is a vector of the ancestral states, Y is a
and bob type), and also to estimate correlation vectoroftheextantspeciesphenotypesrelativeto
coeYcients between these measures. Given the the grand mean, W is a matrix derived from the
variationinheadbobdisplayspresentevenwithin phylogeny, and (cid:229) is vector of error terms (see
aspecies,parsimonymethodsdonotseemappro- Martins & Hansen 1997, equations 3, 10–12 for
priateforreconstructingancestralstates.Instead, details). Specifically, we measured the traits (Y0)
we applied a generalized least-squares approach inninepopulationsofCycluraandcalculatedthe
to estimating ancestral phenotypes (Martins & grand mean for each trait across the whole clade
Hansen 1997). This method oVers flexibility (M )usingM =(J*Var[Y]"1J)"1J*Var[Y]"1Y0,
G G
as to the assumptions underlying a phylogenetic whereJisacolumnof1s,andVar[Y]isamatrix
statistical analysis. Because the original paper of the expected similarities among extant taxa
1690 AnimalBehaviour,55,6
0.002
C. nubila (Cuba)
0.013
0.002
0.002 C. nubila (P.R.)
0.015
0.008 C. nubila lewisi
0.026 0.013
C. cychlura
0.012 0.016
C. rileyi
0.062
C. collei
0.023
C. ricordi
0.043
0.026
C. carinata
0.020
0.056
C. cornuta
Figure2.PhylogenyfromMalone&Davis(unpublisheddata)basedonparsimonyanalysisofmtDNAsequences.
Numbersrefertobranchlengthsinunitsofpercentnucleotidesequencedivergence.Bootstrapvaluesforallnodes
areover90%.
(discussed below). The standard error of this tree ((cid:229) ); and (3) error due to incorrect specifi-
S
grand mean is given by the square roots of the cation of the phylogeny ((cid:229) ). First, to include
P
elements of J*Var[Y]"1Y0. We then transform measurement error, we calculated the within-
the empirical measures (Y0) into deviations from species variance for each character and taxon for
the grand mean to form Y (=Y0"M ), and which data were available from more than one
G
calculate W as the product of the expected simi- individual.Weusedthesewithin-speciesvariances
larity between extant and ancestral phenotypes to create an error variance matrix, Var[(cid:229) ], for
M
(Var[A,Y],seebelowfordetails)andtheinverseof eachtraitwithvariancesonthediagonalsandall
Var[Y] (such that W=Var[A,Y]Var[Y]"1). The otherelementssetequaltozero.
vector A (containing the estimated ancestral Second,weusedapreliminaryphylogeny(C.L.
states) is calculated as the product of W and Y, Malone&S.K.Davis,unpublisheddata; Fig.2)
and we add the grand mean to back transform tocalculatetheexpectedsimilarityduetoshared,
each estimated ancestor to the scale of Y0. Vari- stochastic evolution (Var[(cid:229) ]). Malone & Davis’
S
ances (and therefore, standard errors) for these phylogeny is part of a larger study examining
ancestral states were estimated using generalized relationships among several genera of iguanas
leastsquares,specifically: (Sites et al. 1996) using a parsimony analysis of
mitochondrial DNA data (ND4 and three tRNA
Var[A|]=Var[A]"Var[A,Y]*Var[Y]"1Var[Y,A] genes).Itiscompletelyindependentofthebehav-
iouralinformationpresentedinthecurrentstudy,
wherethesevariancematricesarecalculatedfrom and although the phylogeny is a preliminary
the phylogeny and error variance as discussed estimate, nodes are quite well supported, with
below. bootstrap values exceeding 90% in all cases.
IncalculatingbothWandthestandarderrors, Branch lengths on the phylogeny are also avail-
we need to calculate three variance–covariance able in units of DNA sequence divergence. The
matrices describing the expected degree of simi- fact that branch lengths at the tips do not all
laritybetweendiVerenttaxonphenotypes(Var[Y], line up at one end of the phylogeny suggest that
Var[A,Y]=Var[Y,A], and Var[A]). Expected simi- there is some rate heterogeneity in the molecular
larities between taxon phenotypes are due to at sequence data, and that measures of sequence
leastthreefactors:(1)withinspeciesvariabilityor divergence provide only rough estimates of rela-
measurement error ((cid:229) ); (2) error due to shared, tive time. In the present study, however, we
M
stochastic evolution of taxa along a phylogenetic assume that sequence divergence is a reasonable
Martins&Lamont:AncestralstatesofdisplaysinCycluraiguanas 1691
estimateoftime,andthatthebehaviouralcharac- where (cid:243)2 is the present, estimated value of (cid:243)2
S(t) S
tersevolvedalongthisphylogenyusingoneoftwo and(cid:243)2 isthenextvalue.Thetraceofamatrix
S(t+1)
possible forms of phenotypic evolution: (1) such (PV ) is denoted by tr[PV ], where V is Var[(cid:229) ]
2 2 2 S
that phenotypic divergence is linearly related to dividedby(cid:243)2.Mosttimes,thisiterativeprocedure
S
phylogenetic distance; and (2) such that pheno- converged on a single value of (cid:243)2 within a few
S
typicdivergenceisexponentiallyrelatedtophylo- hundred iterations. When it did not converge, it
geneticdistance.AsshownbyHansen&Martins was usually because the values bounced between
(1996),thesetwopossibilitiesdescribetheresultof twoormorepeakvaluesthatwerefairlysimilarin
phenotypic evolution under a wide variety of magnitude (i.e. within 10) and thus had little
scenarios, including random genetic drift, direc- impactonestimatesofthestandarderror.Atvery
tional selection and stabilizing selection. Note high or very low levels of Æ, values for (cid:243)2 may
S
that since we consider a variety of possibilities never converge at all, making it impossible to
and focus on the robustness of our results across estimatethestandarderror.Notethatestimatesof
these options, small errors in the original branch (cid:243)2 cancel out of the equation for estimating
S
lengths or in the assumption that branch lengths ancestralstates.Therefore,theironlyimpactison
are reasonable estimates of relative time, are estimates of the standard errors of the ancestral
likely to have little impact on the conclusions of states. Unfortunately, with the relatively small
thisstudy. numberoftaxainthisstudy,wewerenotableto
For a linear relationship between phenotypic estimate Æ eVectively from the measured data
and phylogenetic divergence, we calculated the at all. Instead, we applied a range of possible
matrix Var[(cid:229) ] with each element corresponding Æs,asawayofdeterminingtherobustnessofour
S
to a constant ((cid:243)2) multiplied by the phylogenetic estimates to diVerent Æ values (i.e. diVerent
S
distance (d) from the root of the tree to the most assumptions about the strength of stabilizing
recent common ancestor of the taxon pair. This selection and the phenotypic response to that
modelisoftenusedtodescribephenotypicevolu- selection).
tion under random genetic drift or under direc- For each character, we estimated the pheno-
tional selection when the direction of selection types of hypothetical ancestors and the standard
varies randomly over time. The constant (cid:243)2 is a errorsofthosephenotypesusingformsofVar[(cid:229) ]
S S
function of the mutation rate and other forces to create the Var[Y], Var[A,Y] and Var[A] matri-
causing taxa to diversify. For an exponential ces. We also conducted separate analyses using
relationship, we calculated each element of the sum of Var[(cid:229) ] and Var[(cid:229) ] matrices to form
S M
Var[(cid:229) ] as (cid:243)2 exp("Æt) for each pair of taxa, the Var[Y] matrix. Doing both analyses allows
S S
where Æ and (cid:243)2 are estimated constants, and t is us to illustrate the diVerence in results obtained
S
the phylogenetic distance between the two taxa. by incorporating within-species variation in the
This latter model is most often used to describe analysesversusassumingthatwithin-speciesvari-
the evolution of phenotypes subjected to a ation is negligible. The matrix Var[Y] consists
restraining force, such as stabilizing selection of the expected similarities between all possible
pushing the character towards a fixed optimum. pairs of the extant taxon phenotypes, Y, whereas
Theparameter,Æ,canbeinterpretedasameasure Var[Y,A] consists of the expected similarities
ofthestrengthofthatrestrainingforce. between extant taxon phenotypes and ancestral
Weestimated(cid:243)2 usinganiterativeFisherscor- states and Var[A] contains the expected simi-
S
ingprocedure.WebeganbycalculatingtheFisher larities between ancestral taxa. We assumed that
informationmatrix,F,asone-halfthetraceofthe within-species variation in extant taxa was inde-
matrix PV"1PV"1, where P equals V"1 (I"H pendent of such variation in ancestral taxa, and
V"1 [J V"1J]), V"1 is the inverse of the total included within-species variability only in esti-
error matrix (Var[(cid:229) ]+Var[(cid:229) ]), I is an identity mates of the expected similarity between extant
S M
matrix, H is a matrix of 1s, and J is a vector of taxa (Var[Y]). In both cases, expected similarities
1s.Wecalculatedeachnewvalueof(cid:243)2 usingthe werecalculatedaslinearorexponentialfunctions
S
followingequation: of phylogenetic distance (as discussed above).
Notethatthisanalysisdoesnottakeerrordueto
incorrect specification of the phylogeny ((cid:229) ) into
P
account,simplybecausenogoodestimatesofthis
1692 AnimalBehaviour,55,6
error were available, and randomization tests tests. This final procedure incorporates the error
(e.g. Martins 1996a) which are appropriate for duetomis-specificationofthephylogeny((cid:229) )into
P
generating confidence intervals, may not be as theanalysis.
appropriate for obtaining the best estimates of
evolutionaryparameters.Fortunately,generalized
leastsquaresensuresthatourestimateswillstillbe RESULTS
unbiased, although perhaps not quite as accurate
as we might prefer. All analyses were conducted HeadbobDisplays
usingMathematica(Wolfram1991)toimplement
Cycluracarinata
thegeneralizedleast-squaresequations.Forcom-
Mostofthe30C.carinataheadbobdisplayswe
parison, we also estimated ancestral states using
observedconsistedofaseriesofalternatingsingle
thesum-of-squared-changesparsimonyalgorithm
and double headbobs (Fig. 1). Each display usu-
(Huey&Bennett1987;Maddison1991;McArdle
allybeganwithoneortwosingleheadbobs(13/30
& Rodrigo 1994), weighting estimates by the
displays) or a fast trill of 6–15 bobs (7/30 dis-
branchlengthsonthephylogeny(asinME1Gof
plays). These were then followed by a set of
Martins & Garland 1991). Programs to conduct
betweenoneandsixdoubleheadbobs,whichwere
these and similar analyses are also available in
thenfollowedagainbyanothersingleheadbobor
COMPARE(Martins1996b).
two. In some cases (10/30 displays), the display
We also applied phylogenetic comparative
began without preamble at the set of double
analyses to estimate the correlations between
headbobs.Ofthefourdisplaysrecordedonvideo-
measuresoftheheadbobpattern.Again,weused
tapeandforwhichdurationswerethusavailable,
a generalized least-squares approach (Martins
threedidnotincludeatrillandlastedanaverage
& Hansen 1996), but now applied the model
(&se)of0.5&0.10s.Thefourthdisplayincluded
Y=(cid:226)X+(cid:229),whereYandXarevariousmeasuresof
atrillof11bobsandlasted7.4s.Thus,durations
the headbob pattern, considered individually in
calculated as averages of these four displays had
separate models. With only nine taxa, estimation
rather high standard errors (Table I). Cyclura
andpowerofevolutionaryrelationshipsaresmall.
carinata also produces head nods and tail raises
Thus, we aimed primarily to obtain a general
typicalofotheriguanas.
pictureofthepatternsofrelationshipratherthan
detailedandaccurateparameterestimatesorcon-
fidence intervals. As above, we created error Cycluracollei
covariance structures based on assumptions that TwooftheC.colleiheadbobdisplaysobserved
phenotypicdivergenceislinearlyorexponentially duringthisstudyconsistedofonesingleheadbob
relatedtophylogeneticdivergence,butconsidered each.Thethirddisplay(Fig.1)consistedofthree
only the case when Æ is assumed to equal one. single headbobs with pauses of 0.7s and 1.0s
We then estimated correlation coeYcients using betweenthem(TableI).
the ‘contrast’ module of COMPARE (Martins
1996b) which implements several variants of the
independent contrasts method (e.g. Felsenstein Cycluracornuta
1985;Grafen1989)forincorporatingphylogenetic Cycluracornutaareunusualinthattheyarethe
information. Although we did not include only Cyclura species that produces ‘rolls’ as well
within-species variability in this analysis, we did as headbobs (Wiewandt 1977). In a roll, one side
calculatealternativeconfidenceintervalsusingthe (leftorright)oftheheadisraised,andwhenitis
‘randtree’ module of COMPARE. To do this lowered, the opposite side of the head (left or
calculation, we generated a set of 1000 possible right) is raised. The motion is often repeated
phylogeniesassumingonlythatspeciationcanbe multiple times, raising one side of the head and
described as a standard Markovian branching thentheotherinabackandforthmotion.Inmost
process(e.g.Slowinski&Guyer1989),calculated cases, rolls were produced in long series, with
correlation coeYcients using each phylogeny, virtually no pause (less than 0.07s) between suc-
and combined the results using the procedure cessive back-and-forth motions. Rolls could not
described in Martins (1996a) to construct con- be analysed in detail without a camera observing
servative confidence intervals for our hypothesis the animal from above. By recording the relative
Martins&Lamont:AncestralstatesofdisplaysinCycluraiguanas 1693
TableI.Means(&se)ofseveralmeasuresoftheheadbobdisplayforsevenspeciesofCyclura
Total Pause Headbob
duration duration duration Number Number
(s) (s) (s) ofbouts ofbobs
C.carinata 2.20(1.735) 0.63(0.129)* 0.34(0.086) 1.97(0.215) 6.00(4.673)
C.collei 0.98(0.953) 0.87(0.133)* 0.24(0.058) 1.67(0.667) 1.67(0.667)
C.cornuta 2.83(0.447) 0.75(0.139) 0.38(0.027) 2.14(0.206) 5.50(0.489)
C.cychlura 1.98(0.399) 0.82(0.118) 0.30(0.023) 2.35(0.242) 2.88(0.193)
C.nubilanubila(Cuba) 3.22(0.406) 0.67(0.048) 0.28(0.058) 3.63(0.344) 4.85(0.540)
C.nubilanubila(P.R.) 4.23(0.711) 0.74(0.053) 0.98(0.415) 3.77(0.474) 4.29(0.520)
C.nubilalewisi 3.43(0.896) 0.59(0.104) 0.22(0.022) 3.50(0.749) 5.50(1.851)
C.ricordi 3.37(0.243) 0.30(0.021) 0.27(0.006) 3.48(0.239) 9.72(0.590)
C.rileyi 3.14(0.556) 0.22(0.060) 0.88(0.007) 3.17(0.345) 5.22(0.721)
*Standarderrorsbasedonvariationwithinonedisplaybyasinglelizard.
height of one eye ridge, though, we were able to 1982;Dugan1982)ofgreeniguanas.InanS-roll,
determine the number of times each side of the the nose is raised and lowered as in a headbob,
head was raised or lowered during each roll and but is also moved in a horizontal plane during
describe these motions in much the same way as the lowering motion. Although some individuals
wedescribedheadbobs. appeared to be producing standard headbobs in
Rollsoccurredatthebeginningofnineofthe14 additiontotheS-rolls,thetwistingmotionisnot
displays,attheendofonedisplay,andnotatall asdramaticasthefullrollingmotionproducedby
in four displays. In all but one case, rolls were C.cornuta, and can hardly be seen from a lateral
accompanied by one or two series of standard view. Thus, we scored all up-and-down motions
headbobs. There was an average (&se) of producedbyC.nubilaasS-rolls.
4.0&0.42 back and forth motions per roll Displays for the two diVerent subspecies and
(Fig. 1), and up to six up and down motions per two populations of the same subspecies did not
headbob series (mean&se=2.0&0.42). Average diVersubstantiallyinmostmeasures.AllC.nubila
durationswereobtainedbytreatingup-and-down producedseriesofaboutfiveS-rollsseparatedby
headbobs and back-and-forth rolls as equivalent pauses of about 0.7s, leading to a total duration
units (Table I). Again, these data were obtained of about 4s (Table I, Fig. 1). The one exception
only from C.c.stejnegeri. Some observations of was that headbobs produced by animals in the
C.c. cornuta at the Indianapolis Zoo show that Puerto Rican population lasted more than twice
theseanimalsalsoproduceheadrolls,butdetailed as long (mean&se=1.0&0.42) as those from
measuresofthecadenceofthesedisplayswerenot the other two populations (mean&se=0.3&0.06
available. for Cuban C.n.nubila; mean&se=0.2&0.02 for
C.n.lewisiatthezoo).
Cycluracychlura
The 17 C.cychlura headbob displays obtained Cycluraricordi
fromIverson’svideotapes(unpublisheddata)con-
The headbob displays of the single C.ricordi
sisted of short bouts of mostly single, but occa-
sionallydouble,headbobs(mean&se=1.2&0.07; male consisted of long series of headbobs with
progressivelylongerpausesbetweenthem(Fig.1,
Fig. 1, Table I), with no obvious pattern to the
Table I). The display usually began with a series
choiceofsingleordoubleheadbobs.
of about seven headbobs (mean&se=6.6&0.37)
with virtually no pauses. The initial series was
Cycluranubila usually followed by an additional set of about
CycluranubilaisuniqueamongCyclurainthat three single headbobs (mean&se=2.5&0.24)
it produces an S-roll similar to the ‘roll’ (Dugan with inter-bob pauses of about 0.2s each
1982), or ‘rotary head nodding’ (Distel & Veazey (mean&se=0.2&0.02s).Frequently,thesewould
1694 AnimalBehaviour,55,6
be followed by one long pause (mean&se (total duration, headbob duration, and the
duration=0.5&0.03s), and then another series numberofboutsorheadbobsperdisplay)under-
of about two headbobs (mean&se=1.7&0.12). went three to seven major evolutionary changes
Thus, inter-bob pauses slowly increased in dura- (at least one standard error) within the genus
tionthroughoutthedisplay.Sometimes,theentire (Fig. 3). These measures of the headbob display
series of headbobs was followed immediately by increased and decreased through evolutionary
another set, and it was diYcult to distinguish the time in various lineages, but there were no long-
endofonedisplayandthestartofthenext. term trends holding steady throughout the phyl-
ogeny.Infact,themoststrikingpatternisthatall
ofthesechangesareconcentratedamongthetips
Cyclurarileyi
of the phylogeny, with none occurring near the
The 17 displays recorded for C.rileyi in this root of the tree. This suggests that change is
studyconsistedofseriesofuptosevensingleand happening quite quickly, with phenotypes rarely
double headbobs (Table I, Fig. 1). These head- remaining intact past a speciation event. Focus-
bobs were mixed, with no obvious pattern to the ing, for example, on the headbob displays of the
choiceofsingleordoubleheadbobs. two populations of C.nubila, we find major dif-
ferences in the total duration, headbob duration
and number of headbobs. The introduced popu-
PhylogeneticAnalyses
lation at Isla Magueyes (Puerto Rico) hold their
Overlaying measures of the headbob displays heads up for an unusually long period of time
onthephylogeny,wefoundthatheadbobdisplays during each headbob, leading to a much longer
have evolved frequently and dramatically among headbob display, even though it contains fewer
the Cyclura, with major changes in the display headbobs. This diVerence has happened quite
occurring in almost every branch leading to a tip recently, no longer than the time separating two
species (Fig. 3, Table II). Some aspects of the populations of animals on Cuba, and possibly as
display are less malleable than others, evolving recentlyaswhenthePuertoRicanpopulationwas
onlyonceinthegenus.Forexample,threespecies established(30yearsago).
produced series of fast up-and-down motions
(‘trills’), and in all three cases, these trills were
EvolutionaryAssumptions
produced at the beginning (rather than in the
middle or at the end) of displays. Because the Estimates of ancestral states were remarkably
three species that exhibit trills are also their own consistent, whether calculated using the linear or
closest relatives, it seems likely that trills have exponentialmodels,andregardlessofthevalueof
evolved a single time in the ancestor leading to the restraining force (Æ) in the exponential model
C.ricordi,C.carinataandC.cornuta.Incontrast, (Table II). For ancestors of the most recently
although many species of Cyclura exhibit some diverged taxa (the two C.nubila populations),
slight horizontal motion in their headbob dis- theseestimateswereoftenidenticalforallmodels,
plays,theS-rollsofC.nubilaandthelongrollsof and never diVered by more than 5% of the trait
C.cornuta are clearly quite diVerent from each value. DiVerences among results for diVerent
other and those of any other species of Cyclura. models were more apparent when considering
Given the phylogenetic distance between these estimates of ancestral states deeper in the phy-
two taxa, it would seem that the roll has evolved logeny. These estimates occasionally varied by as
at least twice within Cyclura. Because green much as 50% of the trait value. Not surprisingly,
iguanasalsouserollingmotionsintheirdisplays, standard errors of estimated ancestral states also
it may be that iguanas in general have some usually increased upon approaching the root of
evolutionary predisposition towards using hori- thephylogeny.
zontalmotionintheirhead-bobdisplays. Measuresofthefitofthestatisticalmodelsalso
Pause duration was remarkably constant depended in large part on the underlying choice
throughout the genus in an absolute sense, rang- of evolutionary assumptions (i.e. linear versus
ingbetween0.2and0.9s,andnotundergoingany exponential) and whether or not within-species
evolutionary shifts greater than one standard variationwasincluded.Mostdramatically,theres-
error in magnitude (Fig. 3c). All other measures idualsumofsquaresoftendecreasedsubstantially