SONG DIVERGENCE BY SENSORY DRIVE IN AMAZONIAN BIRDS

20 Pages·2010·0.61 MB·English

ORIGINAL ARTICLE
doi:10.1111/j.1558-5646.2010.01067.x
SONG DIVERGENCE BY SENSORY DRIVE
IN AMAZONIAN BIRDS
JosephA.Tobias,1 JobAben,2 RobbT.Brumfield,3,4 ElizabethP.Derryberry,3 WouterHalfwerk,5
HansSlabbekoorn,5 andNathalieSeddon1,6
1EdwardGreyInstitute,DepartmentofZoology,UniversityofOxford,UnitedKingdom
2DepartmentofBiology,UniversityofAntwerp,Belgium
3MuseumofNaturalScience,119FosterHall,LouisianaStateUniversity,BatonRouge,Louisiana70803
4DepartmentofBiologicalSciences,LouisianaStateUniversity,BatonRouge,Louisiana70803
5InstituteofBiology,LeidenUniversity,2300RALeiden,Holland
6E-mail:[email protected]
ReceivedDecember7,2009
AcceptedMay26,2010
Visualsignalsareshapedbyvariationinthesignalingenvironmentthroughaprocesstermedsensorydrive,sometimesleading
tospeciation.However,theevidenceforsensorydriveinacousticsignalsisrestrictedtocomparisonsbetweenhighlydissimilar
habitats,orsingle-speciesstudiesinwhichitisdifficulttoruleouttheinfluenceofundetectedecologicalvariables,pleiotropic
effects,orchance.Hereweassesswhetherthisformofsensorydrive—oftentermed“acousticadaptation”—cangeneratesignal
divergenceacrossecologicalgradients.BystudyingaviancommunitiesintwoAmazonianforesttypes,weshowthatsongsof
17“bamboo-specialist”birdspeciesdifferinpredictablewaysfromtheirnearestrelativesinadjacentterrafirmeforest.Wealso
demonstratethatthedirectionofsongdivergenceiscorrelatedwiththesoundtransmissionpropertiesofhabitats,ratherthan
withgeneticdivergence,ambientnoise,orpleiotropiceffectsofmassandbillsize.Ourfindingsindicatethatacousticadaptation
adds significantly to stochastic processes underlying song divergence, even when comparing between habitats with relatively
similar structure. Furthermore, given that song differences potentially contribute to reproductive isolation, these findings are
consistentwithawiderroleforsensorydriveinthediversificationoflineageswithacousticmatingsignals.
KEY WORDS: Acousticadaptation,birdsong,deterministicselection,ecologicalselection,phenotypicdivergence,sensorydrive,
signalevolution.
Divergentsignalsplayakeyroleintheestablishmentandmain- novel signals and the absence of well-defined optima (Fisher
tenance of premating reproductive isolation, and the processes 1930; Iwasa and Pomiankowski 1995). The third possibility is
driving signal divergence are therefore clearly relevant to stud- that ecological selection causes adaptive divergence in signal-
iesofspeciationandspeciesco-existence(West-Eberhard1983; ing traits by “sensory drive,” a term coined by Endler (1992)
Endler1992;Panhuisetal.2001;Boughman2002;Rundleetal. for the process by which signals evolve to minimize degrada-
2005).Severalroutestodivergencehavebeenhypothesized,but tion and maximize conspicuousness against background noise.
three are particularly prominent. The first and most parsimo- It is widely recognized that each of these factors contributes
nious is that divergent traits reflect random genetic drift and to trait divergence, but their relative influence is poorly under-
mutation(Lewontin1974;Kimura1983;LynchandHill1986). stood, particularly in the case of acoustic signals (Mundinger
The second is that sexual selection drives continuous but other- 1982; Martens 1996; Irwin et al. 2008; Ame´zquita et al.
wise unpredictable divergence because of the attractiveness of 2009).
(cid:2)C 2010TheAuthor(s).
1 Evolution
JOSEPH A. TOBIAS ET AL.
Thereisabundantevidencethatvisualsignalsandassociated Furtheruncertaintyabouttheroleofsoundtransmissionin
sensorysystemsaretosomeextent“tuned”tomatchcharacter- acoustic adaptation is generated by attributes of previous stud-
istics of the environment in a range of animals, including fish ies.Inparticular,mostsingle-speciesandsomemultispeciesap-
(Endler 1980), birds (Marchetti 1993), and reptiles (Leal and proaches are vulnerable to the influence of confounding vari-
Fleishman 2002). The potential influence of sensory drive as a ablesorchance.Thisisthecaseinanalysesofsignalphenotype
force determining the tempo and direction of signal evolution over wide geographic areas, across which local conditions vary
is further highlighted by evidence that signal design is shaped in many ways, making it impossible to be certain about the di-
by variation in lighting, turbidity, or background color (Endler rectcauseofphenotypicvariation(Schluter2000).Forexample,
and Basolo 1998; Gomez and The´ry 2004; Engstro¨m-Ost and spatial variation in signals can be caused by shifts in climate,
Candolin2007;UyandStein2007).Thesefindingshaveattracted elevation, or ambient noise profiles (e.g., Slabbekoorn and Peet
attention because they suggest that habitat heterogeneity will 2003; Kirschel et al. 2009a), or by various forms of character
causedivergentselectiononsignalsassociatedwithmatechoice, displacement (Seddon 2005; Grether et al. 2009; Kirschel et al.
potentiallyfacilitatingspeciation(Endler1992;Boughman2002). 2009b;TobiasandSeddon2009).Similarly,itisoftenproposed
Indeed,empiricalstudieshaveclearlydemonstratedthelinkbe- thatacousticadaptationexplainstherelationshipbetweensignal
tweensensorydrive,signaldivergence,andreproductiveisolation structureandvegetationdensity,butanobviousalternativepos-
(Boughman2001;Seehausenetal.2008),atwo-stepmechanism sibility is that signal variation reflects a correlated evolutionary
ofspeciationthatmaybewidespreadinanimalswithvisualmat- response (sensu Nosil et al. 2008), perhaps as a byproduct of
ingsignals. habitat-relatedselectiononbodysizeortraitsassociatedwithfor-
Thesameprocessofenvironmentallydrivendivergencemay aging(e.g.,RyanandBrenowitz1985;Podos2001;Podosetal.
apply to acoustic mating signals. If habitats vary in their sound 2004;Badyaevetal.2008).
transmissionproperties,andifindividualfitnessislinkedtolong- Culturalevolutionmayalsoaffecttheinterpretationofprevi-
distance communication, then local adaptation should theoreti- ous studies testing acoustic adaptation in birds, particularly as
callyresultinhabitat-dependentacousticsignalswithproperties most of these—including all single-species studies—have fo-
thatmaximizethereliabilityoftransmissionfromsignalerstore- cused on oscine passerines (Boncoraglio and Saino 2007). Os-
ceivers.Althoughthisconceptisnestedinthebroaderframework cine species develop songs by an imprinting-like process called
of sensory drive, it is often referred to as the acoustic adapta- learning(BeecherandBrenowitz2005).Ingeneral,thisincreases
tionhypothesis(Morton1975).Itiswidelyproposedthatacous- randomgeographicalvariationandleadstotheformationofreper-
tic adaptation plays a role in speciation (Ryan and Wilczynski toires and dialects. It also means that young individuals may
1988;SlabbekoornandSmith2002a;Fo¨rschlerandKalko2007), simply learn the sounds, or parts of sounds, that they perceive
but this has never been demonstrated. Moreover, it remains un- mostclearlyintheirnatalhabitat(Hansen1979).Forthisreason,
clear whether habitat-mediated selection has a consistent effect habitat-associated patterns of song variation may lack a genetic
on acoustic signals except when habitat differences are very basis,reflectingphenotypicplasticityratherthanacousticadap-
large. tation (Ellers and Slabbekoorn 2003; Ripmeester et al. 2010).
Densehabitatsareassociatedwithgreaterreverberationand Moreover, the emphasis on oscines reduces the extent to which
increased attenuation at higher frequencies (Marten and Marler wecandrawfirmconclusionsaboutsensorydriveinspeciation
1977;Martenetal.1977;Slabbekoornetal.2002).Thesensory becauseitremainsunclearwhethersignalplasticitypromotesor
drivehypothesisthereforepredictsthatacousticsignalswillhave delaysreproductiveisolation(LachlanandServedio2004;Seddon
lower frequency, simpler structure, and slower pace in densely andTobias2007;OlofssonandServedio2008).
foliatedhabitatscomparedtoopenhabitats.Evidenceoftheseef- These issues can be largely avoided by testing for sensory
fectsisreportedinnumeroussingle-speciesstudies(Hunterand driveinsystemswithinnateacousticsignals.However,onlytwo
Krebs 1979; Slabbekoorn et al. 2002; Slabbekoorn and Smith comparative studies have taken this approach in nonpasserine
2002b; Patten et al. 2004; Nicholls and Goldizen 2006; Dingle birds, and they report conflicting results. Specifically, Bertelli
et al. 2008; Derryberry 2009; Kirschel et al. 2009a), as well andTubaro(2002)foundthatthesongsofopen-countrytinamous
as comparative analyses (Wiley 1991; Badyaev and Leaf 1997; (Tinamidae)hadawiderbandwidththanthoseoftheirrelatives
BertelliandTubaro2002;TubaroandLijtmaer2006).However, occurring in closed habitats, which supports the predictions of
therearealsoaconsiderablenumberofstudiesthateitherrefute acousticadaptation.Meanwhile,TubaroandMahler(1998)found
oronlyweaklysupportsensorydrive(e.g.,Payne1978;Lemon thatsongsofopen-countryNewWorlddoves(Columbidae)were
etal.1981;DanielandBlumstein1998;TubaroandMahler1998; of lower pitch than their relatives in closed habitats, counter to
HyltonandGodard2001;BlumsteinandTurner2005),including the predictions of acoustic adaptation. Similarly, in a nonavian
arecentmeta-analysis(BoncoraglioandSaino2007). system,noevidenceforacousticadaptationwasfoundinastudy
2 EVOLUTION 2010
SENSORY DRIVE IN BIRDSONG EVOLUTION
ofmarmot(Sciuridae)alarmcalls(DanielandBlumstein1998), drive,wepredictthatsongvariablessuchaspitchandpacewill
whichareassumedtobeinnatesignals. varyconsistentlyacrosshabitats,inlinewithdifferencesinsound
Previoussupportforacousticadaptationalmostuniversally transmissionoptima.Conversely,ifsongdifferentiationisshaped
derives from comparisons between habitats with radically dif- principallybyneutralprocesses,theassociationsbetweenvocal
ferentstructuralcharacteristics,forexample,denseforestversus andecologicaldivergencewillbeweakorabsent,andoutweighed
open grassland (Morton 1975; Wiley 1991; Tubaro and Mahler byapositiverelationshipbetweenvocalandgeneticdivergence.
1998;BertelliandTubaro2002)orforestversusurbanenviron- Oursystemisuniquelysuitedtoansweringthesequestions
ments(Slabbekoornetal.2007).Fromtheperspectiveofspecia- forthreereasons.First,bambooforestandterrafirmeforestare
tionresearch,thisisproblematicasspeciationtendsnottoinvolve relativelysimilarinstructure,bothhavingtalltreesandadense
abruptswitchesfromdenseforesttoopengrassland,orviceversa. understory.Moreover,phylogeneticdataconfirmthatAmazonian
On the contrary, sister species usually occupy relatively similar “bamboo-specialist”birdsevolvedfromterrafirmespecies(e.g.,
habitats.Putanotherway,ecologicalnichesarephylogenetically Brumfieldetal.2007;Irestedtetal.2009).Wecanthereforebe
conserved(WiensandGraham2005),orassociatedwithahigh certainthatwearefocusingonanecologicalcontrastrelevantto
phylogeneticsignal(Losos2008),andthusifacousticadaptation speciation(i.e.,potentiallyinvolvedindrivingcladogenesis).Sec-
playsawidespreadroleinspeciationitshouldberelevantacross ond,weexcludespecieswithlearntsongsfromouranalysis(i.e.,
habitatgradientsratherthanrestrictedtohabitatextremes. oursampledoesnotincludeanyoscinepasserines,hummingbirds
Setagainsttheevidenceforsensorydrive,thereisgrowing orparrots)tomaximizethelikelihoodthatphenotypicdifferences
supportfortheroleofstochasticityintheformofunpredictable have a genetic basis (Kroodsma 1996). And third, by focusing
sexualselection,orrandomdriftandmutation.Somestudiesof on a single locality we effectively control for a wide range of
visual signals conclude that patterns of signal evolution cannot factorsknowntoinfluencesongdesign,includingclimate,eleva-
be explained by adaptation to the environment (e.g., Fleishman tion,interspecificcoevolution,andgeographicaldistancebetween
et al. 2009); others conclude that sexual selection is the most samplinglocalities.
powerful influence (e.g., Allender et al. 2003). In addition, it
is often concluded that acoustic signals may diverge by purely
nonadaptiveorstochasticprocessesratherthanecologicaldeter- Materials and Methods
minism(Mundinger1982;SearcyandAndersson1986;Martens
STUDY SITE
1996; McCracken and Sheldon 1997; Price and Lanyon 2002;
We conducted fieldwork in August 2006–December 2007 in a
Ame´zquitaetal.2009).Asarecentexample,Irwinetal.(2008)
2–3 km2 area of lowland forest (280–300 m asl) at the Centro
showed that variation in song and call structure in Phyllosco-
de Investigacio´n y Conservacio´n de R´ıo Los Amigos (CICRA),
puswarblerswasrelatedtogeographicandgeneticdistance,but
MadredeDios,SEPeru(12◦34(cid:4)07(cid:4)(cid:4)S,70◦05(cid:4)57(cid:4)(cid:4)W).Vegetation
unrelatedtohabitatstructureandmorphology.Suchfindingssug-
hereconsistsofamosaicofAmazonianhabitats(seeFig.S1for
gest that, as an evolutionary force, acoustic adaptation is rather
a Landsat image) dominated by Mauritia palm swamp, flood
weak. However, the relative roles of deterministic and stochas-
plain forest, terra firme (upland) forest, and large stands of
tic processes in acoustic signal divergence are rarely assessed,
tall Guadua bamboo—mainly patches of G. weberbaueri and
particularlyacrosshabitatcontrastsrelevanttospeciation.
G. sarcocarpa (Nelson and Bianchini 2005). This kind of bam-
In this study, we focus on a bird community at a single
booforestispervasiveinsouth-westAmazonia,whereitcoversa
Amazonian locality to investigate habitat-mediated variation in
totalareaof120,000–180,000km2,typicallyinpatchesof0.07–
birdsong—aclassiclong-distancecommunicationsignalaffected
12.6 ha (Saatchi et al. 2000; Silman et al. 2003). We focus on
by the environment through attenuation, degradation, and inter-
a comparison between bamboo forest and terra firme forest be-
ference with ambient noise (reviewed in Slabbekoorn 2004b).
cause these habitats support a diverse assemblage of specialist
Specifically, we ask whether vocal divergence between pairs of
birdspecies(Terborghetal.1990;Kratter1997).
closelyrelatedbutecologicallydivergentspecies—oneoccurring
in bamboo forest and the other in adjacent terra firme (upland)
forest—is explained by acoustic adaptation to the transmission HABITAT STRUCTURE
propertiesofhabitats.Althoughourmainaimistotestthepredic- Tounderstandthebasisofdifferencesinthesoundtransmission
tionsof(1)thesensorydrivehypothesis,wealsoassesstheextent propertiesofbambooforestandterrafirmeforest,aswellasto
towhichhabitat-mediatedvocaldivergencecanbeexplainedby quantifytherelativesimilarityofthesehabitatscomparedtoadja-
(2)avoidanceofinterferencewithambientnoise,(3)acorrelated centmanmadegrassland,weconductedbasicvegetationsurveys
evolutionaryresponseofselectiononbillsizeandbodysize,or at 15 points in each of these three habitat types. Survey points
(4) neutral genetic drift. If divergence is determined by sensory were positioned at least 30 m apart, and within a 10-m radius
EVOLUTION 2010 3
JOSEPH A. TOBIAS ET AL.
of each point we quantified aspects of habitat relevant to birds sequencereflectsafast-pacedsongandthesecondaslowerpaced
singingwithin2mofgroundlevel:meancanopyheight(±5m); song (sensu Slabbekoorn et al. 2007). The master file (44100
meanvisibility(±2m)at1.5mabovetheground;understoryveg- Hz/16bitWAV)therebyconsistedofaseriesof12pairsofartifi-
etationdensity(±10%);andtotalnumberoftreeswithdiameter cial100-msconstant-frequencytonesatsixdifferentfrequencies
atbreastheight(dbh)inthreecategories(<20,20–40,>40cm). (0.5,1.0,2.0,3.0,4.0,and5.0kHz).
Understory vegetation density was an estimate, expressed as a
proportionoftotalvolume,oftheamountofdensefoliagewithin Experiments
acircleof10-mradiusandbelow4m. We conducted sound transmission experiments at 50 locations
Weconductedaprincipalcomponentsanalysis(PCA)with (hereafter, sound points), 25 in bamboo forest and 25 in terra
Varimax rotation and Kaiser Normalization on the correlation firmeforest,withatleast30mbetweenpoints.Experimentstook
matrix of survey points to quantify overall habitat structure (all placeat0600–0800h(i.e.,duringtheperiodofpeakvocalactiv-
variableslog-transformed).Thisgeneratedtwouncorrelatedprin- ityforAmazonianbirds)inMarch–April2007.WeusedaSony
cipalcomponents(PCs)accountingfor83.4%ofthevariationin HI-MDdevicetobroadcastthemastersoundfilethroughaSME-
theoriginaldataset:PC1 (Eigenvalue=3.84)whichwaspos- AFSloudspeaker(SaulMineroffElectronics,Elmont,NY),and
veg
itivelyrelatedtomeancanopyheightandnumberoflargetrees, simultaneously recorded it as a 44.1kHz WAV file using a uni-
and PC2 (1.17) was negatively related to mean visibility and directionalSennheiserME66-K3UmicrophoneandanEdirolR-
veg
positivelyrelatedtounderstoryvegetationdensity(allfactorload- 09digitalrecorder(Roland).Ateachsoundpoint,webroadcast
ings>0.8). the master file at two transmission heights (0.1 and 2.0 m) and
recorded it at two distances (5.0 and 20.0 m). All transmission
SOUND TRANSMISSION pathwayspassedthroughuninterruptedvegetationonflatground.
Todeterminewhetherbambooandterrafirmeforesthavedistinct Volume levels for playback and recording of stimuli were stan-
acoustic environments, we quantified their sound transmission dardizedat60dBSPLat10mtoapproximatenaturalsongs(see
properties. We used the standard approach of broadcasting and Seddon and Tobias 2006) using an Adastra handheld analogue
rerecording artificial sound stimuli in the different habitats and soundlevelmeter(setat“C”weightingandfastresponse).
quantifyingtheextenttowhichsoundsaredifferentiallydegraded
byattenuationandreverberation(e.g.,Morton1975;Slabbekoorn Analysis
etal.2007). Transmission experiments produced a total of 2400 sound files
for analysis. To determine levels of degradation of transmitted
Soundstimuli stimuli in respective habitats, these files were redigitized at a
Standardized artificial sound stimuli were generated using Avi- samplingrateof25kHz,afterwhichdatawereextractedautomati-
softSASLABPro,version4.15(Specht2006).Wedesignedthe callyusingapreprogrammedprocedureinSIGNAL(Engineering
sounds so that they matched the basic structure of songs in our Design,1997).Specifically,threeamplitudemeasurements(root-
communityascloselyaspossible.Thesongsofrainforestpasser- mean-squared,RMS,valuesofsoundpressuredeviationsinvolts)
ines tend to be tonal in structure, that is, they have a restricted were taken during fixed 100-ms periods relative to the onset or
bandwidth(frequencyrange)andarerelativelyfreeofovertones offsetofatone:measurementAwastakenaftertheonsetofthe
(harmonics) (Morton 1975; Slabbekoorn 2004b). Accordingly, first tone; measurement B after the offset of the first tone; and
mean song bandwidth in our avian community is only 1.72 ± measurementCaftertheonsetofthesecondtone(seeFig.S2).
0.66kHz,andallbuttwospecies(Hemitriccusflammulatusand Thesemeasureswereusedtocalculatethreetransmissioncharac-
H.griseipectus)lackdistinctovertones.Wethereforeusedcon- teristicsasfollows:(1)Attenuation(i.e.,lossofamplitude)wasA
stant frequency tones as our sound stimuli. These tones were at20mdividedbyAat5m;(2)reverberation(i.e.,tail-to-signal
100msecindurationasthiswasclosetomeannotedurationin ratio)wasBat20mdividedbyAat20m;and(3)distortion(i.e.,
oursample(117±9msec). signal-to-signalratio)wasCat20mdividedbyAat20m.We
Tonesweresynthesizedatsixdifferentfrequencies(0.5,1.0, pooledfast-andslow-pacedstimulitocalculateattenuationand
2.0,3.0,4.0,and5.0kHz)toencompasstherangeofmaximum reverberation,asthesewerequantifiedsolelyfromthefirstnote
avianauditorysensitivity(Dooling1982).Ateachfrequency,we anditsechoes.Weuseddatafrombothnotestocalculatedistor-
generated two sequences of two 100-msec tones. One sequence tioninrelationtosignalpace,andthereforefast-andslow-paced
had a relatively short interval of 150 msec, close to the mean stimuliwerenotpooled.Distortionwasameasureoftheextent
internoteintervalinoursample(152±74msec).Theotherse- towhichreverberationsfromaprecedingnotecompromisedsig-
quencehadalongerintervalof250msec,closetothemeanmax- nalfidelity,asthesemayincrease(i.e.,distort)theamplitudeof
imuminternoteintervalinoursample(283±74msec).Thefirst successivenotes.
4 EVOLUTION 2010
SENSORY DRIVE IN BIRDSONG EVOLUTION
TESTING THE SENSORY DRIVE HYPOTHESIS knowntodevelopsongsbylearning(i.e.,parrots,hummingbirds,
Thenextstepwastoinvestigatewhetherthesoundtransmission oroscinepasserines).Allstudyspecieshadsimple,stereotyped
propertiesofhabitatsinfluencetheacousticstructureofsignals. vocalizations,withoutrepertoiresordialects,andcomefromfam-
Totestthisidea,wedeterminedwhethertherewereanyconsistent ilies without evidence of song learning (see Kroodsma 1984;
differencesintheacousticstructureofthesongsofbambooand KroodsmaandKonishi1991;SeddonandTobias2007;Seddon
terrafirmebirdsand,ifso,whetherthedifferencescorrelatedwith etal.2008).Thetotalsamplecontained33species(onespecies
thesoundtransmissionpropertiesofthetwohabitats. wascomparedtwice).Fordetailsofthesespecies,andtheirphy-
logeneticrelationships,seeFigure1andTableS1.
Studyspecies Speciespairingsweretypicallynottruesisterspecies.This
Approximately 25 species regularly, if not exclusively, occur in did not matter because we were not attempting to directly test
bambooatCICRA(Kratter1997;Tobiasetal.2008;Tobiasetal. speciation hypotheses, but rather to assess the role of ecologi-
2009). Of these, we studied 17 species with a close relative in cal selection on song divergence. Instead, we paired “bamboo
adjacent terra firme forest. We excluded species that were only specialists” with their closest terra firme relative at the study
observed vocalizing higher than 2 m above ground level (e.g., site.Theonlyexceptionsweretwoambiguouscases(pairs5and
Drymophiladevillei)becauseweonlyassessedsoundtransmis- 6inFig.1)inwhichwepairedbamboospeciesandterrafirme
sion properties of the understory (i.e., at 0.1 and 2.0 m only). speciesaccordingtosimilarityintheiracousticsignals.Songsim-
Wealsolimitedthesampletosedentary,territorialnonpasserines ilarity was quantified on the basis of overlap in peak frequency
orsuboscinepasserines,andexcludedanyspeciesfromfamilies and note pace, as determined through spectrographic analysis
Figure 1. A maximum-likelihood tree from an analysis of the ND2 gene depicting evolutionary relationships among the 33 species
used in acoustic analysis (voucher data in Table S2). Acoustic comparisons were conducted on 33 pairs of species, one of which was
foundprimarilyinGuaduabambooatourstudysite(depictedingray),andtheotherfoundprimarilyinterrafirmehabitat(depictedin
black).SpecieswereidentifiedasoccurringprimarilyinbambooorterrafirmefollowingKratter(1997)andTobiasetal.(2009).Pairwise
comparisonsbetweenbamboobirdsandtheirclosestrelativesinterrafirmearenumbered(1–17).Allspeciesarecomparedonce,except
Myrmoborusmyotherinuswhichiscomparedtwice.
EVOLUTION 2010 5
JOSEPH A. TOBIAS ET AL.
(see below for details). Combining these two species pairs on Table 1. Factorloadingsonthreeprincipalcomponentsforacous-
thebasisofsongsimilarityrestedontheassumptionthatvocally tic measures taken from the songs of bamboo and terra firme
similarspeciesaremorelikelytobecloselyrelated. forestspecialists.
Factorloadings
Songdivergence
WeusedaSennheiserME67directionalmicrophoneandaSound
PC1 PC2 PC3
Devices 720 digital recorder (file format: WAV; sampling fre-
quency: 44.1kHz) to record male songs from at least three dif- Eigenvalue 3.316 1.708 1.392
ferentindividualsforallspecies(seeTableS1forsamplesizes). %ofvariance 47.375 24.398 19.886
Maximumfrequency 0.929 0.329 −0.093
Songs were defined as the loudest, most structurally complex
Minimumfrequency 0.954 −0.094 0.037
vocalization in the repertoire of each species; in all cases, they
Peakfrequency 0.969 0.193 −0.059
wereassumedtohaveaterritorialfunctionbecausetheyelicited Duration −0.042 −0.593 0.777
aggressive responses when played to conspecifics (Tobias and Notenumber −0.039 0.196 0.977
Seddon2009;J.A.TobiasandN.Seddon,unpubl.data).Terri- Pace −0.002 0.938 0.086
torialsignalsaretypicallybroadcastoverlongdistances,andare
thereforemorelikelythanshort-rangesignals(e.g.,contactcalls
oralarmcalls)tomeettheassumptionsoftheacousticadaptation Ambientnoise
hypothesis. The hypothesis that habitat-mediated vocal divergence re-
WeusedAvisoftSASLabPro4.50(Specht2006)togenerate flects adaptation to avoid or minimize interference with noise
spectrogramsfromhigh-qualityrecordings(i.e.,lowbackground (Slabbekoorn 2004b) predicts consistent differences in the am-
noise).Thefinalsamplecontained347songs(mean=10.2±5.9 bient noise profiles of bamboo and terra firme forest habitats.
songsperspecies),whichweredescribedquantitativelyintermsof To test this idea, we took digital recordings of ambient noise at
sevenbasicacousticcharacteristics:maximumfrequency(kHz), dawn,startingatnauticaldawn(i.e.,whenthesunis12◦ below
minimum frequency (kHz), peak frequency (kHz; frequency in the horizon, as determined using US Naval Observatory data:
thesongwiththegreatestamplitude),bandwidth(kHz;maximum http://aa.usno.navy.mil).Eachrecordinglasted120min,andwas
frequencyminusminimumfrequency),songduration(s),number thereforetimedtocoincidewithpeakvocalactivityinNeotropi-
of notes, and pace (number of notes s−1). Peak frequency was calbirds(Bergetal.2006).Digitalrecordings(fileformat:WAV;
automaticallymeasuredfromamplitudespectrausingAvisoft;all sampling frequency: 44.1 kHz) were made using a Sennheiser
other measures were taken manually from spectrograms using ME62 omnidirectional microphone attached to an Edirol R09
on-screencursors.Theamplitudeofeachrecordingwasadjusted digitalrecorderwithan8GBLexarsoundcard.Themicrophone
to a standard level prior to analysis to minimize the impact of wassuspended1.5mfromtheground,andthesamesettingswere
recordingvolumeonourmeasurements.Tomaximizeresolution, used for all recordings to ensure that amplitude was consistent.
spectralcharacters(inkHz)weretakenfromnarrowbandspectro- Wecollectedaninitialsampleof30recordingsinbamboo,and
grams(bandwidth=55Hz),andtemporalcharactersweretaken 30recordingsinterrafirmeforest,withallrecordingsiteslocated
frombroadbandspectrograms(323Hz).Foreachacousticcharac- atleast200mapart.However,weremovedanyrecordingsheav-
ter,wegeneratedmeanvaluesperindividualandalsoperspecies ily affected by rain, leaving a final sample of 17 recordings in
(seeTableS1).WethenconductedaPCAonthecorrelationma- bambooand18interrafirmeforest.Raingeneratesnoiseacross
trixofspeciesmeans(log-transformed)toquantifyoverallsong abroadfrequency,andmayconcealhabitat-dependentpatternsof
structure. Three uncorrelated PCs were extracted with Varimax ambientnoise(Slabbekoorn2004a).
rotation and Kaiser Normalization, one of which reflected song Fromthese120-minrecordings,wesampled1-scutsevery
pitch(PC1 ),oneofwhichreflectedsongpace(PC2 ),and 5 min, producing a total of 24 cuts per recording, and 840 cuts
song song
oneofwhichreflectedsongdurationandnotenumber(PC3 ); in total. From these, we calculated RMS-values in Matlab 7.5
song
togethertheseaccountedfor91.7%ofthevarianceintheoriginal (Mathworks,Natick,MA)byfilteringsoundfilesin50-Hzbins
acousticdataset(Table1). in the range of 0–10.0 kHz (resulting in 200 bins per 1-s cut).
RMS-valueswerethenaveragedacrosstimeforeach50-Hzbin,
ALTERNATIVE HYPOTHESES to produce a mean ambient noise spectrum per recording site.
We explored the role of three alternative drivers of song diver- RMS-valueswerethenaveragedacrossallsitestoproducemean
gencebetweenbambooandterrafirmeforestbirds:adaptations ambientnoisespectraforthefocalhabitats.
tominimizeinterferencewithambientnoise,byproductsofmor- Totestwhethersongdivergencereflectsadaptationtomini-
phologicaladaptation,andneutralgeneticdivergence. mizeinterferencewithambientnoise,wefirstdeterminedwhether
6 EVOLUTION 2010
SENSORY DRIVE IN BIRDSONG EVOLUTION
therewereanyconsistentdifferencesintheambientnoiseprofiles by 35 cycles of 94◦C for 1 min, 50◦C annealing step for 30 s,
ofthetwohabitats.Wethenplottedthemeanpeakfrequencyof and a 72◦C extension step for 1 min. The program ended with
thesongsofbambooandterrafirmebirdsagainstthemeanambi- a final 72◦C extension step for 3 min. PCR primers and inter-
entnoisespectraofthetwohabitats.Finally,weusedchi-square nalsequencingprimersincludedH6313(JohnsonandSorenson
tests to ask whether the peak frequencies of birdsongs in each 1998),L5215(Hackett1996),H5766(Brumfieldetal.2007),and
habitatavoidedprominentbandsofambientnoisemorethanex- L5758 (Johnson and Sorenson 1998). Amplicons were purified
pectedbychance.Wedefinedprominentbandsasallnoiseabove usingPEGprecipitation,elutedin12.5mL10mMTris,andse-
acut-offof0.003RMS-values(i.e.,∼50dB). quencedusingtheABIPrismcyclesequencingprotocol(Applied
BiosystemsInc.,Carlsbad,California)andanABI3100Genetic
®
Correlatedevolution Analyzer.SequencingreactionswerepurifiedusingSephadex
Habitat-mediatedvocaldivergencemayreflectacorrelatedevo- G-50and400mL96-wellfilterplates.
lutionaryresponsetoselectiononbillsizeandbodysize,inwhich Sequences were edited and aligned manually using Se-
casethereshouldbeanassociationbetweenvocaldivergenceand quencher 4.6 (Gene Codes Corporation, Ann Arbor, MI) and
morphologicaldivergence.Intheory,lowerpitchedsongsshould translatedintoaminoacidstoverifytheabsenceofstopcodonsor
be associated with larger body mass, and slower paced songs anyanomalousresidues.Allnewsequencesweredepositedinto
shouldbeassociatedwithlargerbillsize(Podosetal.2004).To GenBank(seeTableS2).Thefinaldatasetconsistedoftheentire
testthesepredictionswecollateddataonbodymassandbillsize ND2gene(1,041basepairs)for33taxa.UsingPAUP∗4.0,wecal-
andquantifiedmorphologicaldivergencebetweenspeciespairs. culatedthecorrectedp-distancebetweeneachspeciespairbased
Data on body mass (to the nearest 0.1 g) were either collected onafinite-sitesDNAsubstitutionmodel.Thebest-fitmodelwith
frombirdscaughtin12×4mmistnetsatCICRAin2004–2008, the fewest number of estimated parameters was determined us-
or from the literature (Dunning 1993; Zimmer and Isler 2003). ingtheAICtestimplementedinModelTest(PosadaandCrandall
Similarly,billmeasurementsweretakenfromaminimumofthree 1998).LikelihoodscoresforinputintoModelTestwerecalculated
malesperspeciesforall17species,mainlyfrommistnettedindi- onaneighbor-joiningtreeusingPAUP*4.0.Weusedmaximum-
viduals,withgapsfilledusingspecimenshousedattheMuseum likelihoodmethodsimplementedinRAxML7.0.4ontheCipres
of Natural Science at Louisiana State University (LSU), Baton Portal version 1.5 (http://www.phylo.org/sub_sections/portal/)
Rouge,USA.Dialcaliperswereusedtomeasure(tothenearest (Stamatakis et al. 2008) to infer the best tree using the GTR +
0.01 mm) bill length from the tip to the meeting of the culmen (cid:2)+ I model with the data partitioned by codon position (three
andskull,andbilldepthandbillwidthattheanteriorendofthe partitions)withjointbranchlengthoptimization.
nares.WeusedPCAtoreducebillmeasuresintoasinglecompo- Individualvaluesforallacousticvariables(log-transformed;
nentrepresentingbillsize(PC1 ),whichexplained84.1%ofthe Table1)wereenteredintoaPCA,fromwhichwecalculatedsong
bill
variation(Eigenvalue=2.52)andwithwhichallthreevariables distancebetweenbamboobirdspeciesandtheirclosestsympatric
hadhighcorrelationcoefficients(>0.8). terrafirmerelativeastheEuclideandistancebetweengroup(i.e.,
species)centroids.
Neutraldivergence
If neutral genetic divergence explains song divergence in our STATISTICAL AND PHYLOGENETIC COMPARATIVE
study, there should be a positive relationship between song di- ANALYSES
vergence and genetic divergence. To estimate genetic distance PCscoresforvegetationstructure(PC1 andPC2 )werecom-
veg veg
betweenspeciespairs,wesequencedthemitochondrialprotein- paredbetweenhabitattypesusingunivariateGeneralLinearMod-
coding gene (ND2). Studies of ND2 and other mitochondrial els(GLM)followedbyTukey’sHSDposthocpairwisecompari-
protein-coding genes in birds have shown genetic variation that sontests.Theeffectofhabitat,frequency,andtransmissionheight
is consistent with neutral evolution (Arbogast et al. 2006; Weir onattenuationandreverberationwastestedusingGLMwithmain
2006). Data for 11 species were obtained from GenBank (see effects and first-order interactions; and we used the same GLM
Table S2). For the remaining species, we performed DNA se- approachtoanalyzetheeffectofhabitat,frequency,andnote-pace
quencing.TotalDNAwasextractedfrom25mgofpectoralmus- ondistortion.Totestforhabitatdifferencesinthespectraldistri-
cleusingtheDNeasykitfollowingthemanufacturer’sprotocol. butionofambientsoundenergy,weusedaGLMinwhichhabitat
Ina20mLtotalvolume,PCRamplificationscontainedapprox- andfrequency(i.e.,50-Hzbin)wereenteredasfixedeffectsand
imately 60 ng of genomic template DNA, 50 mM KCl, 10 mM log-transformedRMS-valuesasthedependentvariable.
Tris-HCl,1.5mMMgCl,0.5mMdNTPs,0.75μMofeachexter- Song structure (separate acoustic measures and PC
song
nalprimer,and0.08UPromegaTaq.Thethermocyclingprogram scores) and morphology were first compared between bamboo
consistedofaninitialdenaturingstep(94◦Cfor2min)followed specialistsandtheirclosestrelativeinadjacentterrafirmeforest
EVOLUTION 2010 7
JOSEPH A. TOBIAS ET AL.
Table 2. Phylogeneticsignalofacousticandmorphologicaltraits. Prior to parametric analyses, all variables were log-
transformedtomeetassumptionsofnormalityandhomogeneity
Variable λ LRP1 AICλ AIC
ofvariance.Nonparametrictestswereusedwhereverthedistribu-
Brownian
tionofvariablesdidnotmeetparametricassumptions.Meansare
Maximumfrequency 1.00 1.00 118.75 116.75 reported±SD,unlessstatedotherwise;P-valuesaretwo-tailed
Minimumfrequency 1.00 1.00 102.18 100.18 and corrected for ties where appropriate. Phylogenetic analyses
Peakfrequency 1.00 1.00 116.08 114.08 were conducted in R (R-Development-Core-Team 2008) using
Bandwidth 0.67 0.01 67.44 72.20
theApeandGeigerlibraries(Harmonetal.2008)aswellascode
Duration 1.00 1.00 132.21 130.21
writtenbyR.P.Freckleton;allothertestswereconductedusing
Pace 0.49 <0.0001 205.96 343.15
SPSSversion17.0(SPSS2008).
Notenumber 0.00 <0.0001 269.76 1141.38
PC1 1.00 1.00 90.89 88.89
song
PC2 0.88 0.04 89.98 92.17
song
PC3 0.86 0.01 92.90 96.93 Results
song
Bodymass 0.91 <0.0001 5350.14 5635.57
HABITAT STRUCTURE
PC1 1.00 1.00 91.79 89.79
bill
We found highly significant variation in habitat structure be-
1DenotesthePvalueonalikelihoodratio(LR)testcomparingλandBrow- tweenstandsofGuaduabamboo,terrafirmeforestandadjacent
nianmodelsofevolution;P<0.05(markedinboldface)indicatesthatλisa manmadegrassland(PC1veg:F2,44 =56.3,P<0.0001;PC2veg:
betterfitthanaBrownianmodel. F2,44 = 73.04, P < 0.0001). Grassland had a highly uniform
structurewithanopenunderstoryandfewtrees,andshowedno
overlapinoverallstructurewitheitherbambooorterrafirmefor-
using a paired, nonparametric approach (Wilcoxon signed-rank
est (Fig. S3). Conversely, the principal component scores were
tests). However, this approach assumes that species pairs are
more variable and partially overlapping for bamboo and terra
equally related to one another and that the phylogenetic signal
firme (Fig. S3). Nonetheless, we found that the former had a
ofacousticandmorphologicalcharactersisweak.Totestthisas-
significantlydenserunderstory,lowermeanvisibilityandfewer
sumption,weestimatedlambda(λ),whichmeasuresthedegreeto
largetreesthanthelatter(allposthocpairwisecomparisons:P<
whichtraitsvary/covaryacrossatreeinlinewithBrownianmo-
0.0001;Fig.S4).Insummary,allthreehabitatshadadistinctive
tion(Freckletonetal.2002).Aλof1correspondstotheBrownian
structure, but bamboo and terra firme forest were much more
model, λ of 0 indicates a lack of phylogenetic structure, and λ
similartooneanotherthaneitherwastograssland.
valuesbetween0and1indicatethedegreeoftraitlability(Pagel
1999).Todeterminewhetherλvaluesdepartedsignificantlyfrom
aBrownianmodel,wecomparedthefitofthetwomodelsusing HABITAT SOUND TRANSMISSION PROPERTIES
alikelihoodratiotest.Abouthalfofthevariablesdepartedfrom Wefoundasignificanteffectofsignalfrequency,heightoftrans-
a strict Brownian model (Table 2). The implication is that the missionandhabitattypeonattenuation(Table3,Fig.2).Trans-
evolution of these traits may have been influenced by shared mission height explained most variance in attenuation, and the
ancestry. interaction between transmission height and signal frequency
Tocorrectforthis,weusedtwotypesofcomparativemodels. had the strongest effect. At 0.1 m above the ground there was
First,weusedgeneralizedestimatingequations(GEEs)forcom- aprominentattenuationwindowbetween2.0and4.0kHzinboth
parativedataasdescribedinParadisandClaude(2002).Wethen habitats(Fig.2A),withbambooattenuatinghighfrequencysig-
usedthegeneralizedleastsquares(GLS)phylogeneticcompara- nals(4.0and5.0kHz)significantlymorethanterrafirmeforest.
tivemethodasdescribedinFreckletonetal.(2002).Wepresentre- This mid-frequency attenuation window, a common feature of
sultsfrombothapproachesbecausetheGEEmodelsarerobustto forested habitats, is caused by high attenuation for low-pitched
analysisofrepeatedmeasures(OverallandTonidandel2004)but signalsclosetotheground(Fig.2A;Morton1975).At2.0m,the
assumeaBrownianmotionmodelofevolution,whereasthemod- windowwasabsentandinsteadwefoundastronglinearincrease
ifiedGLSapproachsimultaneouslyestimatesandusesλtoadjust inattenuationwithincreasingfrequency(Fig.2B),withbamboo
the phylogenetic correction to reflect trait lability (Freckleton attenuating high frequency signals significantly more than terra
etal.2002).ToworkaroundtheassumptionofaBrownianmodel firmeforest.
in the GEE analyses, we transformed tree branch-lengths using Signal frequency, transmission height, and the interaction
the estimated λ for each trait and used the transformed trees in betweenfrequencyandheighthadastrongsignificanteffectonthe
analyses. We used the maximum likelihood tree (Fig. 1) as our wayinwhichtheartificialsoundstimuliweredegradedthrough
phylogenetichypothesisforcomparativeanalyses. reverberation.Specifically,wefoundthatsignalreverberationwas
8 EVOLUTION 2010
SENSORY DRIVE IN BIRDSONG EVOLUTION
Table 3. Effectsofsignalfrequency,habitattype,andtransmissionheightonthe(A)attenuation,and(B)reverberationofartificial
soundstimuli.StatisticsarefromGLMswithmainandtwo-factorinteractioneffects.Habitatreferstothehabitat(bambooorterrafirme
forest)inwhichthetransmissionexperimenttookplace;heightreferstotheheight(0.1or2.0m)ofthemicrophoneandloudspeaker
abovegroundlevel;n=25soundpointsperhabitattype.
Sourcesofvariation df MS Fratio P2
(A)Attenuation
Maineffects Frequency 5 1.556 16.227 <0.0001
Habitat 1 1.351 14.091 <0.0001
Height 1 3.585 37.395 <0.0001
Two-wayinteractions Frequency×Habitat 5 0.315 3.287 0.006
Frequency×Height 5 3.155 32.909 <0.0001
Habitat×Height 1 0.002 0.023 0.881
Finalmodel1 r2=0.35 17 1.769 18.59 <0.0001
(B)Reverberation
Maineffects Frequency 5 2.283 34.593 <0.0001
Habitat 1 0.112 1.704 0.192
Height 1 26.276 398.229 <0.0001
Two-wayinteractions Frequency×Habitat 5 0.096 1.458 0.202
Frequency×Height 5 0.321 4.871 <0.0001
Habitat×Height 1 0.104 1.581 0.209
Finalmodel1 r2=0.51 11 3.572 53.9 <0.0001
1Excludesnonsignificanttermsandinteractions.
2Bolddenotessignificanttermsandinteractions(P<0.05).
significantlygreaterat0.1than2.0m,andincreasedlinearlywith iswithinthenormalrangeofsongperchheightforallspeciesin
increasing signal frequency (Table 3, Fig. 2C,D). There was no this study except Crypturellus (pair 1), which habitually sing at
overalleffectofhabitatonreverberation,althoughpost-hoctests groundlevel.
indicate a trend toward greater reverberation in higher pitched Tosummarize,ourresultsshowthathigherfrequencysignals
signalsinbamboothaninterrafirmeforest(Fig.2D). suffer greater attenuation in bamboo than terra firme, whereas
Both signal pace and signal frequency had significant ef- faster paced songs at key frequencies suffer greater distortion
fects on distortion, with pace explaining most of the variation in terra firme than bamboo. Thus, the sensory drive hypothesis
(Table4).Specifically,wefoundthatfast-paced,higherpitched predictsthatbamboosongsshouldbelowerinpitchandfasterin
signalssufferedmoredistortionthandidslow-paced,low-pitched pacethanterrafirmesongs.
signals.Inotherwords,reverberationssignificantlyincreasedthe
amplitudeofthesecondtonerelativetothefirstforfasterpaced, TESTING THE SENSORY DRIVE HYPOTHESIS
higherpitchedsignals.Therewasnooveralleffectofhabitattype Thesepredictions were tested with acoustic analyses, which re-
ondistortion,norwasthereanysignificanteffectsofinteractions vealedsignificantdifferencesinthespectralandtemporalprop-
between pace and frequency, or pace and habitat (Table 4). In- ertiesofbambooandterrafirmeforestbirdsongs(Figs.4and5;
deed,thevalueofr2 forthefinalmodelwasverylow(0.07),in Table5).Specifically,wefoundthatthesongsofbamboobirds
line with the idea that our target habitats are relatively similar. had significantly lower peak, maximum and minimum frequen-
However,therewasasignificanteffectoftheinteractionbetween cies,andsignificantlyhigherpace,thandidthesongsoftheirclose
habitatandfrequency.Exploringthisinteractionmoreclosely,we relativesinterrafirmeforest(Table5).Theyalsoconsistedofa
foundastrongsignificanteffectofhabitatonamplitudedistortion significantly greaternumberofnotes.Theeffectsizes(Cohen’s
withinspecificfrequencybands.Mostimportantly,3kHzsignals d) for all but two acoustic measures (duration and bandwidth)
transmitted2mabovegroundinterrafirmeforestsufferedgreater were>0.2(meanunsignedeffectsize=0.25±0.11),indicating
distortionwhentheywerefasterpaced(Fig.3).Conversely,there astrongdirectionaltrendforbamboosongstohavelowerpitch
was no significant difference in levels of degradation between andhigherpace.Moreover,whenweassessedoverallsongstruc-
fast and slow paced 3 kHz signals in bamboo. This finding is tureusingPCAwefoundthatPC1 (representingsongpitch),
song
notable because 3 kHz is close to the mean peak frequency of PC2 (songpace),andPC3 (songdurationandnotenum-
song song
songsinbothhabitats(Table5),andatransmissionheightof2m ber)differedsignificantlybetweenthetwohabitattypes(Fig.5,
EVOLUTION 2010 9
JOSEPH A. TOBIAS ET AL.
Figure 2. Relationshipbetweenthefrequencyofartificialsoundstransmittedandthedegreeofattenuation(A,B)andreverberation
(C,D).Dataarepresentedfortransmissionheightsof0.1m(A,C)and2.0m(B,D)inbothbamboo(gray)andterrafirmeforest(black).
Thex-axisinallcasesisthefrequencyofthesoundtransmitted;withthepositionofbambooandterrafirmepointsdisplacedslightly
eithersideofeachsoundfrequencytoavoidoverlapandfacilitateinterpretation.They-axisin(A)and(B)is1-LG10attentuation,such
that higher values indicate higher levels of attenuation, and a value of 2 signifies that the signal is 10 times weaker at 20 m when
comparedto5m.In(C)and(D)they-axisisLG10reverberation,withavalueofzerosignifyingthattailandsignalareofequalamplitude.
Barsshow95%CIs;n=25soundpointsperhabitattype.Asterisksdenotesignificantdifferencesbetweenhabitattypesinlevelsof
signaldegradation(unpairedBonferroni-correctedt-tests,P<0.008).
Table 5). In other words, the songs of bamboo birds produced ticularly prominent; these appear as two conspicuous peaks in
songsofsignificantlylowerpitchbuthighertemporalcomplex- themeanambientnoisespectrashowninbothbamboo(Fig.6B)
itythantheirclosestrelativesinadjacentterrafirmeforest.This and terra firme forest (6C). We found that the peak frequency
significant effect of habitat on song structure held when con- of songs fell within these bands of noise in 8 of 17 (47%) of
trollingforphylogenyusingboththeGEEandGLSapproaches bamboospecies(Fig.6D)and8of16(50%)ofterrafirmeforest
(Table6). species (Fig. 6E). Thus, the proportion of songs in our com-
munity with peak frequency overlapping with noise bands did
ALTERNATIVE HYPOTHESES notdifferfromthatexpectedbychance(chi-squaretests:χ21 <
Ambientnoise 0.029,ns).
Wefoundasignificanteffectoffrequencyonambientnoisechar-
acteristics, but no effect of habitat or the interaction between Correlatedevolution
frequency and habitat (Table 7). This was corroborated by vi- Bambooandterrafirmeforestbirdsdidnotdifferconsistentlyor
sualinspectionofambientnoiseprofiles,whichshowedthatboth significantly in body mass (Fig. 7A) or bill size (Fig. 7B), sug-
habitats were characterized by continuous noise at a number of gestingthathabitat-dependentsongdivergenceisnotabyproduct
discretefrequencybands(Fig.6).Thesebandsaretypicaloftrop- of morphological adaptation (Fig. 7, Tables 5 and 6). In other
icalforestandaremainlygeneratedbycallinginsects,inpartic- words,wefoundnoevidencethathabitat-mediatedvocaldiver-
ulardifferentspeciesofcicada(Cicadidae;Slabbekoorn2004a). gencereflectsacorrelatedevolutionaryresponseofselectionon
Inourrecordings,twobands(at∼2.0and∼5.0kHz)werepar- bodysizeorbillsize.
10 EVOLUTION 2010