Historical and Human-Mediated Invasions of a Tropical Archipelago.

16 Pages·2012·1.87 MB·English

vol. 180, no. 1 the american naturalist july 2012
E-Article
Revisiting the Ants of Melanesia and the Taxon Cycle:
Historical and Human-Mediated Invasions of a
Tropical Archipelago
Evan P. Economo1,* and Eli M. Sarnat2
1.MichiganSocietyofFellows,DepartmentofEcologyandEvolutionaryBiology,UniversityofMichigan,AnnArbor,Michigan48409;
2.DepartmentofEntomology,UniversityofIllinois,Urbana,Illinois61801
SubmittedApril20,2011;AcceptedFebruary1,2012;ElectronicallypublishedMay24,2012
Onlineenhancement:appendix.
strategies promote phases of range expansion, but large
abstract:Understandingthehistoricalevolutionofbiotasandthe
ranges are unstable and fragment into a complex of lo-
dynamics of contemporary human-mediated species introductions
calized species. This range contraction phase is driven by
aretwocentraltasksofbiology.Onehypothesismayaddressboth—
the taxon cycle. Taxon cycles are phases of range expansion and ecological and evolutionary shifts away from the pheno-
contractioncoupledtoecologicalandevolutionarynicheshifts.These types that initially promoted range expansion. Biogeo-
historicalinvasionprocessesresemblehuman-mediatedinvasionsin graphic dynamics thus follow regular cyclical pathways,
patternandpossiblymechanism,butboththeexistenceofhistorical
andcertainphenotypesarechannelsfordispersalandsub-
cyclesandtherolesofrecentintroductionsareinquestion.Wereturn
sequent diversification.
to the system that originally inspired the taxon cycle—Melanesian
Thehypothesisisnotablyintegrative;itlinksecological
ants—andperformnoveltestsofthehypothesis.Weanalyze(i)the
habitatdistributionsofFiji’sentireantfauna(183species),(ii)eco- and evolutionary pattern and process across scales from
logicalshiftsassociatedwiththeinsituradiationofFijianPheidole local communities to biogeographicregions.Wilson’spa-
inaphylogeneticcontext,and(iii)theecologicalstructureofamas- pers stimulated conceptual advances in the field (Brown
sive exotic ant invasion of the archipelago. Our analyses indicate
andLomolino1998;Whittaker1998;LomolinoandBrown
lineages shift toward primary habitats,higherelevation,rarity,and
2009),andtheideahasbeenappliedusefullytoothertaxa
ecologicalspecializationwithincreasinglevelofendemism,consistent
and geographic areas over the years (Ricklefs and Cox
withtaxoncyclepredictions.Themarginalhabitatsthathistorically
formed a dispersal conduit in the Pacific are now mostly replaced 1972, 1978; Roughgarden and Pacala 1989; Brown and
by human-modified habitats dominated by a colonizationpulseof Lomolino 1998; Mayr and Diamond 2001; Ricklefs and
exotic species. We proposethis mayrepresentthefirstphaseofan Bermingham2002;Cooketal.2008;Gillespieetal.2008).
incipient global cycle of human-mediated colonization, ecological
However,itwasalatertheory—theequilibriumtheoryof
shifts,anddiversification.
island biogeography (ETIB; MacArthurandWilson1963,
1967)—that became the new paradigm for biogeography.
Keywords: taxon cycle, island biogeography, invasive species, For-
TheETIBwasmuchsimplerinconstructionandnarrower
micidae,Pacificislands,Pheidole,Fiji.
in scope but had the advantage of a quantitative formu-
lation and was readily testable with available data. The
Introduction complex story of the taxon cycle was not testable in a
convincingwaywiththedataandmethodsavailableatthe
Fifty years ago, E. O. Wilson synthesized his work on the
time, even for Wilson’s ants (Whittaker 2004). The exis-
Melanesian ant fauna by proposing a hypothesis for bio-
tence of taxon cycles hasbeenrepeatedlyquestionedover
geographic dynamics in space and time: the taxon cycle
the years (Pregill and Olson 1981; Losos 1992; Ricklefs
(Wilson 1959, 1961). In general terms, taxon cycles arise
and Bermingham 2002; Dexter 2010), and today it is a
when the evolution of certain phenotypes and ecological
marginal idea in ecology and evolution.
However,inmanywaysthetaxoncycleismorerelevant
* Correspondingauthor;e-mail:[email protected].
in the modern area of synthesis in biodiversity science.
Am.Nat.2012.Vol.180,pp.E1–E16.(cid:2)2012byTheUniversityofChicago.
Ecological and evolutionary perspectives are increasingly
0003-0147/2012/18001-52993$15.00.Allrightsreserved.
DOI:10.1086/665996 being integrated across scales of space and time, and ad-
E2 The American Naturalist
vancesinmolecularmethods,phylogenetics,statisticalin- ulation density and increased ecological specialization.
ference, geographical information systems, and the large- Wilson(1959,1961)viewedthisdeclinephaseasoccurring
scaleaggregationofbiodiversitydatagiveusnewtoolsfor because selection pressures would adapt the colonizing
analyzingbiodiversitypatternandinferringcomplexpro- speciestothemuchlargerareaoftheinteriorforest,pos-
cesses. Additionally, historical taxon cycles resemble con- sibly pushed along by competitive pressure from newly
temporary invasions in pattern and possibly mechanism, arrived expanding taxa in the marginal habitat. As the
raisingthepossibilitythatreciprocalinsightscanbegained speciesshiftsawayfrommarginalhabitatsacrosstherange,
from the study of both (Brown and Lomolino 1998; Sax geneflowislimitedandspeciationmaytransformasingle
andBrown2000;RicklefsandBermingham2002;Ricklefs widespread species into many localized endemic allospe-
2005). cies.Eventually,thenow-endemiclineagemaypersistand
Here,forthefirsttimesinceWilson’slandmarkstudies, diversify in the interior forest and/or eventually become
we revisit the ants of Melanesia and test predictions of a extinctasitisreplacedbytheprogressionofnewlineages
historical taxon cycle with one of the mostextensivedata through the cycle.
sets ever compiled for an insular ant fauna. We find the Wilson proposed interspecific competition as the driv-
system in the midst of an ongoing massive invasion of ing mechanism behind the cycle, while Ricklefs and Cox
exotic ants into the Pacific islands (Wilson and Taylor (1972) emphasized the role of “vertical” interactions be-
1967;McGlynn1999;Holwayetal.2002).Weanalyzehow tween a species and its predators and pathogens. In this
both these recent arrivals and the native fauna are orga- latter view, a species may enter an expanding phase after
nizedacrosstheislandlandscapeoftheFijianarchipelago. a coevolutionary escape fromnaturalenemies.Theinitial
success after colonization may be due to further geo-
graphic escape from natural enemies and competitors or
The Taxon Cycle Hypothesis
even the transfer of pathogens from the colonists to the
A taxon cycle begins whena geographicallyrestrictedlin- susceptible local fauna. Finally, the decline phase would
eage evolves phenotypes or ecological strategies that ini- be explained by the counteradaptation of the local fauna
tiateaphaseofrangeexpansion.ForMelanesianants,this to the now-abundant new arrival. Other variations em-
involvesanadaptiveshiftfrominteriorprimaryforesthab- phasizing different mechanisms have appeared in the lit-
itats to marginal high-disturbance habitats such as coasts erature (Roughgarden and Pacala 1989; Taper and Case
and riverbanks in the large diverse “source” areas (i.e., 1992; Matsuda and Abrams 1994). However, in general
Southeast Asia and New Guinea). Wilson viewed the in- the mechanisms driving taxon cycles are not well under-
tensity of interspecific competition in the interior forests stood, even in the Caribbean bird system where evidence
asoccasionallydrivingaspeciestomakethisadaptiveshift for the taxon cycle pattern is most compelling (Ricklefs
into marginal habitats. These marginal habitats facilitate and Bermingham 2002).
oceanicdispersalthroughraftingandcolonizationofmore Despite the conceptual advances contributed by Wil-
remote archipelagoes. Ant phenotypes adapted to open, son’sanalysis,theevidenceforthecomplexhistoricalsce-
disturbed habitats possibly include generalized ecological nario he described was not strong (Whittaker 2004). His
strategies,flexibledietandnestingrequirements,orhigher key observation was the differences in community com-
thermal tolerance (Wilson 1959, 1961; Torres 1984a, positionalongadisturbancegradientinNewGuinea,with
1984b). marginalhabitatscomposedofwidespreadspeciesandin-
Inthesecondphase,theseexpandingspeciesareinitially teriorhabitatscomposedofendemics(seetable2andfig.
moreecologicallysuccessfulinnovelareasthantheywere 9 in Wilson 1959). However, the data were generated by
intheirsourcearea.Theyreleaseoutofthemarginalhab- unknownspatialandnumericalsamplingregimes,theob-
itats and are more successful than resident endemic spe- servedpatternswerenotstatisticallysignificantandcould
cies.Wilson(1959,1961)proposedthisinitialsuccesswas have been driven by the idiosyncrasies of a few species,
due to the competitive dominance of species originating and the species classified in the subfamily Ponerinae in
in highly diverse source areas over “naive” species that 1959areasmallfractionofthefaunaandarenowknown
evolvedinlessdiverseareas(i.e.,smallerandmoreremote to be distributed across the ant phylogeny (Brady et al.
islands). This leads to a unidirectional flow of lineages 2006; Moreau et al. 2006).
from diverse to depauperate geographic areas. Furthermore, the role of exotic species inthesystem,a
Afterrangeexpansionandinitialecologicalsuccess,eco- majorcomponentofIndo-Pacificantfaunas(Wilsonand
logicalandevolutionarydynamicsinthenovelarearesult Taylor 1967; McGlynn 1999), is entirely unknown. Like
inadeclinephase.Thewidespreadspeciesshiftsawayfrom historical cycles, human-mediated invasions (Elton 1958;
thehigh-disturbancecoastalhabitattotheuplandinterior Williamson 1996) often involve the geographic spread of
and evolves towards an endemic syndrome of low pop- speciesadaptedtohighdisturbance,highconnectivityhab-
The Taxon Cycle in Ants E3
itats. Like historical cycles, establishment and subsequent Or, under a purely stochastic “neutral” hypothesis (Hub-
ecological success of a new arrival may be a function of bell2001),olderlineagesshouldonaveragebemoreabun-
localdiversity(i.e.,“bioticresistance”)and/orescapefrom dant than new arrivals because stochastic extinction will
natural enemies. Furthermore, like natural colonists, in- selectivelycullspeciesthathavenot“drifted”tohighabun-
troduced species may experience an eventual decline as dance (Rosindell and Phillimore 2011).
new waves of invaders arrive or the local fauna adapts to Ifecology-endemismcorrelationsareconsistentwiththe
the new arrival. taxon cycle, there remains the question of whether they
In the Melanesian ants, there is an opportunity to ana- represent phases of a temporal sequence. For example,
lyzeapotentialhistoricaltaxoncycleandongoinghuman- species adapted to disturbed lowland habitats may have
mediated invasion in the same system. Here we ask two largerangesbecausethosehabitatsfacilitatedispersal,but
questions focusing on the Fijian archipelago using data theydonotshiftovertimeintotheuplandsandeventually
generatedbyanewcomprehensiveinventoryofthefauna acquire the syndromes associated with endemic species
(SarnatandEconomo,forthcoming).First,doesthenative (Dexter 2010). In this scenario, ecological niche is con-
fauna,whichhasbeencolonizingthearchipelagosincethe served, and species colonize interior habitats on islands
Miocene (Lucky and Sarnat 2010), show evidence of a directly from similar habitats in source areas. Higher-ele-
historical taxon cycle? Second, does this same framework vationinteriorspeciescouldhaverestrictedrangessimply
apply to the exotic species brought by Europeans in the becausedispersalistoolimitedtomaintaingeneflowbut
last several hundred years—are they simply the next col- notlimitedenoughtoprecludetheoccasionalcolonization
onization pulse? event.
Phylogeneticapproachescandifferentiatebetweenthese
alternative hypotheses by testing for postcolonization
Testing the Taxon Cycle
nicheshiftsandadaptationintohigh-elevationforestlife-
The taxon cycle hypothesis proposed by Wilson and out- styles. We test this prediction in the Fijian Pheidole (20
lined above is a complex narrative with many subcom- spp.), for which phylogenetic, ecological, morphological,
ponents, and it is beyond this or anysingle studytocon- andbehavioraldataareavailable.Wesynthesizeandextend
firm or reject it in one stroke. Moreover, it is likely that previousworkonFijianPheidole(Sarnat2008;Sarnatand
some components of the model are more correct than Moreau 2011; Sarnat and Economo, forthcoming) and
others. In this study we test predictions for the habitat interpret it from the perspective of the taxoncycle. Fijiis
distributionsoftheentireantfaunaoftheFijiarchipelago. hometooneofthemostaberrant,spectacular,andthreat-
We complement this community-level study with a case ened Pheidole forms in the world: the Pheidole roosevelti
study of ecological and evolutionary shifts in one genus, group radiation. The taxon cycle predicts this radiation
Pheidole, in a phylogenetic context. We do not test the aroseinFijifromgeneralizedancestors,andtrendstoward
assertion that lineage movement is biased from high- to ecologicalandmorphologicalspecializationareassociated
low-diversity systems or in any particular geographic with in situ habitat shifts, loss of dispersal ability, and
direction. declines in density.
Thetaxoncyclepredictsthataftercolonizationandover
time,antlineages(i)shiftfromdisturbedtoprimaryforest
habitats, (ii) shift from low- to high-elevation habitats, Material and Methods
and(iii)declineincommonness.Suchshiftsshouldleave
Study System and Biodiversity Data
a signature in the distribution of species across habitats;
increasing levels of endemism should be correlated with Over the last several years we have conducted a broad
theecologicalchangesdescribedabove.Wetestthesepre- inventoryof the Fijian antfauna.Wepoint-mountedand
dictions with data on the entire ant fauna of the Fijian examinedmorethan10,000antspecimens,representative
islands. If species from different endemism classes have of∼200,000sortedspecimensthatremaininalcohol.Our
contrasting distributions across habitat types in this par- primarysourceofspecimenswasourownfieldexpeditions
ticular way, the taxon cycle hypothesis will be supported. from2002–2008,includinghandcollectionsandlittersift-
The taxon cycle will not be supported if habitat distri- ing. Each leaf litter sample constitutes a 200-m transect
bution is either unrelated to endemism class or if they of 20 1-m2 samples taken every 10 m and combined into
exhibitoneofthemanybiologicallyplausiblenonrandom asingle“maxi-Winkler”extraction.Wealsoprocessed714
patterns that would be inconsistent with the hypothesis. malaisetrapsamplescollectedduringtheNationalScience
For example, endemics could have thewidesthabitatdis- foundation Fiji Terrestrial Arthropod Survey (Evenhuis
tributionandhavethehighestabundance,perhapsbecause andBickel2005),eachofwhichrepresentsapproximately
theyarethebestadaptedtothelocalenvironmentofFiji. 2 weeks of trap time. In addition, we processed samples
E4 The American Naturalist
from canopy fogging and leaf litter conducted by Hilda possibleinacity).Thestatisticalanalysesdescribedbelow
Waqa(UniversityoftheSouthPacific)andcolleaguesand are designed to mitigate these potential complications.
litter sifting conducted by Dave Olsen and the Wildlife
Conservation Society. Using the material collected in the
surveys and additional material we examined in major Species Geographic Classification
collections (U.S. National Museum of Natural History,
The taxon cycle predicts that habitat distribution is cor-
Harvard Museum of Comparative Zoology), we pursued
related with geographic range as a lineagemovesthrough
acomprehensivetaxonomicrevisionofthefauna.Tothese
the cycle. While it would be ideal to have each species
data,weaddedrecordsfromtheextensivehandcollections
placed into a phylogenetic context including the massive
of D. Ward during field expeditions in 2004.
Indo-Australian source pools, this willnotbe available in
Each specimen was identified to species and analyzed
the near future for Fiji’s 42 genera. We classified species
inrelationtotheregionalIndo-Pacificfauna.Undescribed
intobroadcategoriesofendemismthat(accordingtothis
specieswereassignedgeographicallyconsistenttemporary
hypothesis)reflectstagesofprogressionthroughthecycle.
codes. The results of this analysis, including a discussion
Wesortedspeciesintofourgroups:exotics,widespread
of each species with images, collection records, and dis-
natives, endemic allospecies, and deep endemics. We did
tributionmapswillbepresentedinaforthcomingmono-
not use Wilson’s “stages” (Wilson 1961) because these
graph (Sarnat and Economo, forthcoming). classifications have varied in the taxoncycle literaturein-
As ants are social organisms, we ignored how many creasing the likelihood of confusion if they are reapplied
individuals were collected in a sample, and focused on a toants.(1)Exotics(25spp.)arespeciesoriginatingoutside
unique species-sample pair as our unit of data. Together, thePacificandknowntobetransportedaroundtheglobe
our sources yielded 6,548 species-sample pairs from 436 byhumancommerce.ThesepresumablyreachedFijiafter
localities that could be associated with habitat data. Of the arrival of Europeans (last 300 years). (2) Widespread
these, 1,495 were from hand collections, 1,830 fromlitter natives (26 spp.) are native to the Indo-Australian region
sifting, 3,136 from malaise traps, and 86 from canopy and occur broadly and continuously in the Pacific from
fogging.Thesamplingspanseverymajorandmanyofthe India,Indonesia,SoutheastAsia,orNewGuineatoatleast
minor Fijian islands (fig. 1). The resulting data set is no- FijiandoftenfurthereasttoPolynesia(theseareequivalent
tableforitstaxonomiccompleteness;wehavehabitatdis- to Wilson’s “stage I” species). It is possible that some of
tribution data for 177 of the 183 known species in Fiji these species are native to the western Pacific region but
(the missing six all being endemics collectedbyearlycol- wereintroducedtoFijibyearlyhumanmigrationsormore
lectors with no habitat data). Both sampling effort and recent commerce. In one case, a Pacific native, Iridomyr-
methodologiesvariedacrosshabitattypes(e.g.,urbanareas mexanceps,wasdocumentedtobeintroducedrecentlyto
were lesscollectedthanprimaryforest;littersiftingisnot Fiji (Wilson and Taylor 1967) and thus was classified as
Yasawas (YS)
Taveuni (TA)
NEW SOLOMON
GUINEA ISLANDS Koro (KR)
Ovalau (LA)
VANUATU
SAMOA Gau (GA) Lakeba (LK)
FIJI Beqa (BQ) Moala (ML)
NEW
CALEDONIA Kadavu (KV)
AUSTRALIA
1,000
Kilometers collection localities
Figure1: LocationofFijiintheregionandamapofsamplinglocalitiesacrossthearchipelago.Eachblackdotrepresentsatleastone,but
uptohundredsofcollectionevents.
The Taxon Cycle in Ants E5
exotic. Inferring the arrival dates of all nonendemics to mary forests at the extremes. The position of a locality
Fiji wouldrequire populationgeneticstudiesofeachspe- along the first gradient (elevation) is straightforward. For
cies. However, even if some Indo-Australian species have the second (human disturbance), we assigned collection
been introduced in the past 3,000 years, as a group they localities (fig. 1) to one of five coarse-ordered classifica-
shouldstillhaveoccurredlongerinFijithanexotics(pre- tions.Thesecategorieswerebasedonecologicalfieldnotes
vious 300 years). We address this possibility and impli- and if necessary corroborated by satellite imagery. The
cations in the discussion. categoriesweredesignedtobeintentionallybroadinorder
We sortedthe 121 Fiji endemicsintotwogroupsbased to minimize subjectivity: (1) port cities (Suva, Lautoka,
on their taxonomic relationship to regional faunas. Our Nadi,Savusavu):themajorurbanareasandportsofentry
maingoalwastodistinguishbetweenspeciesthatareonly for Fiji; (2) human-dominated landscapes: villages, road-
weaklydivergentfromrelativeselsewhere(andpresumably sides, agricultural areas, and pastures; (3) forest edges:
more recent colonists) and those that are strongly differ- localities at the border of native forest with agricultural
entiated from overseas relatives (presumably older colo- areas or roads; (4) disturbed forest: habitats retaining a
nists). We classified 24 species as (3) endemic allospecies: general closed-canopy forest structure but some evidence
taxadifferentiatedatthespecieslevelbutareclearlyallied of disturbance, such as selective logging, planted exotic
to a widely distributed species complex with no evidence trees(e.g.mahogany,pine),orevidenceoflater-stagesec-
ofcladogenesisinFiji.Theremaining97specieswereclas- ondary status; and (5) primary forest: closed canopy for-
sifiedas(4)deependemics:theseinclude80endemicsthat ests withoutobservableevidenceofrecenthumanornat-
arepartofadistinctiveendemicFijianradiation(26from ural disturbance. Specimens not associated with habitat
molecular evidence) and thus are likely the descendants descriptions, a particular problem for the older museum
of older colonization events. The other 17 are species are collections, were excluded from the analysis.
not part of apparent radiations in Fiji but are strongly
morphologically divergent from other congeners in the
Habitat Distribution Analysis
PacificandcouldnotbeassociatedwithotherPacificspe-
cies groups. Of these four classifications, the distinction We testthetaxoncyclepredictionsthatafterlineagescol-
between endemic allospecies and deep endemics is the onize Fiji, and as they reach deeper levels of endemism,
most subjective, but some misclassification among these they(i)shiftfromdisturbedtoprimaryhabitats,(ii)shift
groups would not be fatal to the analysis. We could have fromlowtohigherelevations,(iii)declineincommonness.
chosentolumpthemtogetherintooneendemicclassand Wetaketwodifferentapproachestocharacterizinghabitat
performed the same analysis. Finally, we classified seven distributions and testing the predictions with our com-
species as (5) quasi-endemics: a handful of species that positedataset,thefirstusingregressionanalysistomodel
occur in Fiji plus one neighboring archipelago (e.g., Sol- aggregate patterns across the entire fauna and a second
omon Islands or Samoa). Table A3, available online, lists using null-model based analysis to comparethepositions
all species and their classifications. of individual species along the habitat gradients.
The taxon cycle predicts that the first four groups rep- Regression analyses. The aggregated data set consists of
resentasequenceofprogressionthroughthecycleinFiji. speciesdetectionsacrossaseriesofsamplesandlocalities.
Thefifthgroup(quasi-endemics)ismoredifficulttoplace Our first question is whether the composition of species
in this sequence based on geography; they could be rem- sampled across the four endemism classes (exotic, wide-
nants of contracting widespread species or incipient ex- spread native, endemic allospecies, deep endemics)
pandingspeciesandthuswereexcludedfromtheanalyses. changes across the gradients of elevation and human dis-
In lieu of their inclusion, we can report that those five turbance.Inotherwords,ifasinglespecies(ornspecies)
species were among the rarest species in Fiji and were is sampled at agivenlocation,whataretherelativeprob-
entirely confined to forest habitats. abilities that the species belongs to each class (or the ex-
pected proportions of each class in the sample). We treat
each species detected as a “trial” associated with one of
Habitat Classification
four multinomial outcomes (the four endemism classes)
We analyze habitat distributions within the two-dimen- andusemultinomiallogisticregression(TrexlerandTravis
sional space formed by the two main ecological gradients 1993) to model how those relative probabilities (odds)
in Fiji. The first is the gradient in elevation and natural changeacrossclassesandgradients.Thesamplingwasdif-
disturbancebetweenthelowland“marginal”coastalforests fuseacrossthelandscapebutatlowintensityinindividual
andhigh-elevationbryophyteforestsatthemountainsum- sites; thus, detection probability in a sample/locality is a
mits(max.1,340monVitiLevu).Thesecondisagradient function of both the presence and local abundance of a
of human disturbance with urban environments and pri- species. The advantage of using relative probabilities and
E6 The American Naturalist
notrawcountsistheyarenotsensitivetosamplingeffort flects s successes in n trials if s species of that class were
(which varies across localities), andthelogisticregression detectedoutofatotalofnspeciesoftheclassintheentire
approachaccountsfortheinherentnoiseinsamplingdis- fauna. Here the model search space included linear and
crete outcomes. However, the disadvantageisthatwecan quadratic terms for elevation, allowing for monotonic or
detect changes only in relative, not absolute, collection unimodal responses to elevation.
probabilities, giving us an index of relative prominence. Hand collections cover both gradients of elevationand
For example, if exotics are more frequently collected in disturbance,whilethestandardizedmalaiseandlittersam-
urban environments than endemics, while endemics are plesareconcentratedintheforesthabitats.Thuswhilewe
more collected in the forest, we would conclude that one modelrelativecollectionprobabilitysimultaneouslyacross
classchangesrelativetotheotherbutcannotsayhowthe both gradients and absolute collection probability along
absolute probabilities of either class changed. To correct elevation in forest habitats only. All regressions were per-
for the fact that different endemism classeshavedifferent formed in MATLAB in a GLM framework using multi-
species numbers in the total fauna, the relative collection nomial or binomial logit links. We used Akaike Infor-
probability was divided by the total species number for mation Criterion (AIC) minimizaton for model selection
comparison across groups. While individual species may amongallsubsetsofthemodelsdescribedabove,including
varyintheirconspicuousnesstosamplingevenatthesame anintercept-onlytermrepresentingthenullhypothesisof
biologicalabundance,thereisnoreasontoexpectthatthis noendemism-ecologycorrelation(buttherecouldstillbe
variation will be a function of endemism class, each of differencesinoverallprominenceacrossclasses).Afterse-
which represent a phylogenetically, morphologically, and lecting the best model for eachanalysis,confidenceinter-
behaviorally diverse group of species. vals for the fitted probabilities were compared toidentify
We performed multinomial logistic regression for rel- regionswhereendemismclassesweresignificantlydifferent
ative collection probability using the predictor variables in prominence.
elevation, disturbance, elevation2, disturbance2, and eleva-
tion # disturbance, and all simpler models nested within
this full model with the constraint that quadratic andin-
teraction terms could be included only if the main effect
was included. The higher-order terms allow themodelto
Species-Level Analyses
detect unimodal and other complex distribution patterns
if present. Disturbance was coded as an ordinal variable
(1pportcityto5pprimaryforest).Goodargumentscan The regression analyses are designed to test for aggregate
be made for using either presence-absence in individual community-wide differences in composition among en-
samples or at different localities as our unit of data and demism class across the landscape, but obscure variation
for including all sampling methods (litter, hand,malaise) among individual species. To examine this, we calculated
or only data from methods that were used in all habitats themeanelevationanddisturbancescoreofaspeciestaken
(handcollections).Weperformedtheregressionsusingall over all observations of that species. As these means also
permutations of these factors and compared results. For depend on the distribution of sampling, we do notinter-
the analyses presented in the main text we used all sam- pret them to reflect true centroids of the species distri-
pling methods, grouped by locality (3,232 occurrence re- butions. Rather, they are useful for comparing different
cordsacross434sites),butfurtherdiscussionofsampling speciesinthesamedataset.High-elevationspeciesshould
issues,statisticaldesign,andresultsfromtheothermodels still have a higher mean than low-elevation species, even
are presented in the supplemental materials. ifthosemeansarebothbiasedinonedirection.Toaccount
Withinthelargerdataset,thereisasetofstandardized forthefactthatdifferentspeciesmaybemoreorlesslikely
malaise(np617)andlittersamples(np56)fromforest tobesampledindifferentmethods,andthesamplingfrom
habitats,inwhicheachsamplereflectsastandardizedunit different methodologies is not distributed evenly across
of effort, and the detection of one species does not pre- the landscape, we generated a null expectation for each
clude the detection of another species. Using these data, species mean and examined deviations from that expec-
we can calculate absolute collection probability, in which tation.Wecalculatedthenullexpectedhabitatorelevation
(cid:2)
eachendemismclasscanbemodeledseparatelyacrossthe meanE forspeciesiusingtheformulaE p p E,where
i i ij j
gradient. Here, instead of the relative odds of sampling E is the elevation or habitat mean of all recojrds detected
j
one class versus another, we model the expected fraction with a given method j and p is the fraction of records of
ij
of species in a class that will be detected in a sample/ species i collected with method j. The deviations of the
locality. A separate binomial logistic regression was per- observedmeansfromexpectedmeanswerecalculatedand
formed on each endemism class, where each sample re- compared across groups with a one-way ANOVA.
The Taxon Cycle in Ants E7
exotics widespread natives
1300 1300
1100 1100
900 900
700 700 Relative Collection R.C.P. variation-
Probability within class
500 500
max
m) 0.04
300 300
(
n 0.03
o 100 100
i 0.02
t
a 1 2 3 4 5 1 2 3 4 5
v
e 0.01
El 1300 endemic allospecies 1300 deep endemics
0.005
1100 1100
0.001 min
900 900
700 700
500 500
300 300
100 100
1 2 3 4 5 1 2 3 4 5
port city an dominated landscape forest edge sturbed forest primary forest port city an dominated landscape forest edge sturbed forest primary forest
m di m di
u u
h h
Disturbance gradient
Figure2: Responsesurfaceofthebest-fittingmultinomiallogisticregressionmodelforrelativecollectionprobability(RCP)acrossdisturbance
andelevationgradientsindicatesshiftstoforesthabitatandincreasingelevationwithincreasinglevelsofendemism.RCPistheprobability
thatasinglecollectioneventwilldetectaspeciesofagivenclass,dividedbythenumberofspeciesinthatclass(seetext).Thesizeofthe
dots represents the variation in relative odds across all endemism classes and can be compared across panels, while thecolorofthedot
reflects variation in relative oddswithinan endemismclass.The dashedline reflectsthelimitsofcollectiondata(i.e.,therearenohigh-
elevationportcities).
Pheidole Background and Phylogeny the endemic Fijian Pheidole (13 spp.) were derived from
a single ancient colonist, although more extensive taxon
We used a molecular phylogeny assembled by Sarnat and sampling of the massive source faunas to the west could
Moreau (2011) and ecological data to examine evidence reveal multiple introductions.
for a postcolonization niche shift in Fijian Pheidole. Phy- Within the endemicclade,thePheidolerooseveltigroup
logeneticevidence(SarnatandMoreau2011)suggeststhe has an unusual “spinescent” morphology of pronounced
Fijian Pheidole community assembled through atleastsix bifurcate propodeal spines and acutely angled mesonotal
independent colonizations of the archipelago, including projections,foundonlyinafewspeciesinSoutheastAsian
thethreewidespreadspecies(P.oceanica,P.fervens,P.um- and Indo-Australian Pheidole and absent from all extant
bonata),oneintroducedbyhumans(P.megacephala),and New World Pheidole (Sarnat 2008). One Fijian species (P.
a lineage represented by a single endemic, P. onifera (fig. simplispinosa) represents an “intermediate spinescent”
A4, available online). The phylogeny indicates the rest of morphology of pronounced but not bifurcate spines and
E8 The American Naturalist
deep
Figure 3
endemics
1200
m) 1000
endemic n ( 800
o
allospecies ati 600
v
e
El 400
200
0
1 2 3 4 5
1200
1000
widespread m) 800
natives n (
o 600
ati
v 400
e
El 200
0
1 2 3 4 5 1 2 3 4 5
1400
1200
m) 1000
n (
exotics o 800
ati
v 600
e
El 400
200
0
1 2 3 4 5 1 2 3 4 5 1 2 3 4 5
port primary port primary
city forest city forest
Disturbance gradient
Figure3: Eachpanelisapairwisecomparisonoftwoendemismclasses,identifyingregionsinelevation-disturbancespacewheretherelative
collectionprobabilityofoneclassissignificantlygreater(nonoverlapof95%CI)basedonthemultinomialregressionmodelvisualizedin
figure2.Theareasshadedwiththecoloroftheclassindicatethatclassisgreater.
lacking an angled mesonotal projection. Ant spines are cent groups in the region. Furthermore, the intermediate
thought to be defense mechanisms associated with con- P. simplispinosa represents a transitional formratherthan
spicuousepigaeicforaging,andtheP.rooseveltigroupfol- a secondary loss of spinescence. Species of theP.knowlesi
lowsthisstrategyexclusively.Thespinescentspeciesinthe grouparedescendedfromthesamecolonizationeventbut
regionwereoriginallygroupedbasedonmorphologyinto retain the ancestral nonspinescent phenotype.
one subgenus, Pheidolecanthinus (Smith 1865). However, We test the prediction that the in situ evolutionofspi-
the phylogenetic analysis established that the P. roosevelti nescenceisassociatedwithapostcolonizationshifttoward
group evolved convergent spinescence from more gener- high-elevation primary forest habitats and ecologicalspe-
alized ancestors and are distantly related to other spines- cializationandmoregenerallythatecologicalnicheisevo-
The Taxon Cycle in Ants E9
Figure4: Toprow,Binomiallogisticregressionresponsesforabsolutecollectionprobabilityinstandardizedmalaiseandleaflittersamples
in forest habitats demonstrate an orderly shift to higher elevations andreducedcollectionprobabilitywithincreasinglevelofendemism.
Bottom row, Relative collection probability from multinomial regression in the same samples, as well as hand collections. Shaded areas
represent95%CI.
lutionarilylabile.Wereconstructedstatesofinternalnodes of the other models in the supplemental materials. The
of this clade for elevation mean and disturbance mean Akaike Information Criterion (AIC) minimizing model
using maximum squared-change parsimony for continu- (AICp2,680, DAICp(cid:2)0.3, intercepts-only AICp
ous characters, implemented in Mesquite (Maddison 3,992)includeddisturbance,elevation,bothsquareterms,
2006). Confidence intervals for the ancestral state were and the interaction term. Coefficients of multinomial re-
estimated using the PDAP package (Midford et al.2005). gression models can be difficult to interpret directly and
Before reconstructing ancestral states, minor additional we present them in the appendix (table A1, available on-
analyses of the phylogeny were performed to resolve in- line). We examined response surfaces of this fitted model
consistencies as described in the appendix. overelevationanddisturbanceinfigure2.Regionsofpair-
wisesignificantdifferences(nonoverlapof95%confidence
interval [CI]) between endemism classes are presented in
Results figure 3. As predicted by the taxon cycle, the different
endemismclassesareorderedindistributionfromhighto
Ecological Organization of the Fijian Ant Fauna
low disturbance (prediction [i]) and from low to high
We identified 183 species of ants (table A3) as occurring elevation (prediction [ii]). Widespread natives are most
in Fiji. Our knowledge of the Melanesian fauna has in- prominent in lowland disturbed and primary habitats,
creasedconsiderablysinceWilson’swork;inthe1961pa- consistentwiththehypothesisthatthesehabitatsfacilitate
per he listed 58 endemic Fijian species and the number dispersalandcolonization.Endemicallospeciesaremostly
now stands at 129. Of the 183 species, 177 had available limited to forest habitats and are shifted further upland,
habitat data, and 172 remained after removing the five while deep endemics are the most confined to high-ele-
quasi-endemics. vationprimaryforestsandingeneralhavelowerarelative
Allvariations(samplesorlocalities,allmethodsoronly collection probability than the other classes (prediction
hand collections) of the multinomial logistic regression [iii]).
analysis returned the same general patterns with a few The analysis of absolute collection probability inforest
minordifferences.Wepresentanddiscussthemodelusing habitatsconfirmedthispatternunderamorestandardized
all methods across localities in the main text, and results sampling regime, a linear or quadratic function of eleva-
exotics widespread natives
1200 1200
1000 1000
800 800
600 600
) 400 400
m
(
n 200 200
o
i
t
a 1 2 3 4 5 1 2 3 4 5
v
e
l endemic allospecies deep endemics
E
n 1200 1200
a
e 1000 1000
M
800 800
600 600
400 400
200 200
1 2 3 4 5 1 2 3 4 5
port primary port primary
city forest city forest
Disturbance gradient
less
disturbed
1.5
600
m 1.0
o
on frm) 400 n 0.5
deviatimean ( 200 eviatiomean 0.0
elevation null 0.0 disturbance dfrom null −−−110...505
−200
−2.0
more
−400 −2.5 disturbed
exotic wide. native. endemic allo. deep endemic exotic wide. native. endemic allo. deep endemic
E10