Global Invasion History of the Tropical Fire Ant: A Stowaway on the First Global Trade Routes

Abstract

Biological invasions are largely thought to be contemporary, having recently increased sharply in the wake of globalization. However, human commerce had already become global by the mid-16th century when the Spanish connected the New World with Europe and Asia via their Manila galleon and West Indies trade routes. We use genetic data to trace the global invasion of one of the world's most widespread and invasive pest ants, the Tropical Fire Ant, Solenopsis geminata. Our results reveal a pattern of introduction of Old World populations that is highly consistent with historic trading routes suggesting that Spanish trade introduced the Tropical Fire Ant to Asia in the 16th century. We identify southwestern Mexico as the most likely source for the invasive populations, which is consistent with the use of Acapulco as the major Spanish port on the Pacific Ocean. From there, the Spanish galleons brought silver to Manila, which served as a hub for trade with China. The genetic data document a corresponding spread of S. geminata from Mexico via Manila to Taiwan and from there, throughout the Old World. Our descriptions of the worldwide spread of S. geminata represent a rare documented case of a biological invasion of a highly invasive and globally distributed pest species due to the earliest stages of global commerce.This article is protected by copyright. All rights reserved.

Full text

Global invasion history of the tropical fire ant: a
stowaway on the first global trade routes
DIETRICH GOTZEK,*†
1
HEATHER J. AXEN,‡
1
ANDREW V. SUAREZ,* SARA HELMS CAHAN‡
and DEWAYNE SHOEMAKER §
*Department of Animal Biology and Department of Entomology, University of Illinois at Urbana-Champaign, Urbana, IL 61801,
USA, †Department of Entomology, National Museum of Natural History, Smithsonian Institution, Washington, DC 20013,
USA, ‡Department of Biology, University of Vermont, Burlington, VT 05405, USA, §CMAVE, USDA-ARS, Gainesville, FL
32608, USA
Abstract
Biological invasions are largely thought to be contemporary, having recently increased
sharply in the wake of globalization. However, human commerce had already become
global by the mid-16th century when the Spanish connected the New World with Eur-
ope and Asia via their Manila galleon and West Indies trade routes. We use genetic
data to trace the global invasion of one of the world’s most widespread and invasive
pest ants, the tropical fire ant, Solenopsis geminata. Our results reveal a pattern of
introduction of Old World populations that is highly consistent with historical trading
routes suggesting that Spanish trade introduced the tropical fire ant to Asia in the 16th
century. We identify southwestern Mexico as the most likely source for the invasive
populations, which is consistent with the use of Acapulco as the major Spanish port
on the Pacific Ocean. From there, the Spanish galleons brought silver to Manila, which
served as a hub for trade with China. The genetic data document a corresponding
spread of S. geminata from Mexico via Manila to Taiwan and from there, throughout
the Old World. Our descriptions of the worldwide spread of S. geminata represent a
rare documented case of a biological invasion of a highly invasive and globally dis-
tributed pest species due to the earliest stages of global commerce.
Keywords: biogeography, fire ant, global trade, invasion biology, Solenopsis geminata
Received 20 September 2014; revision received 24 November 2014; accepted 6 December 2014
Introduction
Human-mediated changes to the environment and glob-
alization of trade have brought biological invasions into
the public, political and scientific limelight (Vitousek
et al. 1997; Bright 1999; Py
sek & Richardson 2010;
Lowry et al. 2013). Several recently introduced species
have become notorious invaders and highly invasive
and damaging pests, costing billions to control and tak-
ing a toll on human health and the environment
(Pimentel et al. 2000, 2005). Human migration and trade
have always aided dispersal of other organisms as com-
mensals. Indeed, Darwin (1859) had already recognized
the transformation of ecosystems by invasions during
his 19th century travels. Some of the earliest docu-
mented cases of dispersal via human migration are the
spread of a skink and gecko during the colonization of
Polynesia starting from 1500 BCE (Fisher 1997; Austin
1999) and the movement of house mice (Rajabi-Maham
et al. 2008) and domestic animals following the Neo-
lithic expansion (Larson et al. 2007; Sacks et al. 2013).
However, historical records or archaeological evidence
for species invasions is often lacking or difficult to
acquire, and most evidence for commerce-mediated
invasions is recent. This bias is undoubtedly augmented
by human-mediated introductions having increased
sharply with the rise in global transportation and trade
(Meyerson & Mooney 2007; Westphal et al. 2007; Hulme
Correspondence: Dietrich Gotzek, Fax: +1-217-244-1224;
E-mail: dgotzek@uga.edu
1
These authors contributed equally to the work.
©2014 John Wiley & Sons Ltd
Molecular Ecology (2015) 24, 374–388 doi: 10.1111/mec.13040

2009). Although such global transport is largely
assumed to be contemporary, Europe, Africa and Asia
were already well connected through a vibrant commer-
cial network, and trade had become truly global in the
16th century when the Spanish established trade routes
across the Pacific and Atlantic Oceans (Flynn & Gir
al-
dez 2004). The Manila galleons and West Indies fleet
connected Spain to the Asian markets via the New
World. The extent to which early global trade may have
been important for setting up many current biogeo-
graphic patterns is unknown, as we lack studies that
examine range expansions at this time (but see Aplin
et al. 2011).
Ants are ideal models to test whether the emergence
of global trade had a far-reaching biogeographic impact,
as they are readily moved via ship transport, making
them highly susceptible to human dispersal. There are a
number of widespread ant species with global or near
global distributions (McGlynn 1999). Their native
ranges are often unknown, as is accurate and detailed
historical documentation of their spread. Many of these
species are known to have reached a near global distri-
bution by the 19th century (Wetterer 2005, 2008, 2010,
2011, 2012; Wetterer et al. 2009), raising the possibility
that they were distributed along the same routes and
during the same times. If so, comparative analyses of
invasion histories should recover the predominant trade
routes during the time of invasion and highlight the
importance of human commerce as a main vector for
human-mediated dispersal.
One such globally distributed ant species is the tropi-
cal fire ant (TFA), Solenopsis geminata (Fabricius 1804).
While TFA is not as well studied as its congener, the Red
Imported Fire Ant (S. invicta), TFA is often the most com-
mon, abundant and conspicuous ant in disturbed areas
and has a much wider distribution. Having colonized
virtually all tropical regions in the world (Wetterer 2011),
it is one of the most widespread tropical and subtropical
ants. Like many other widely distributed early invasives,
there has been uncertainty regarding the native range of
TFA. It has long been known outside the New World
(Wetterer 2011); a red variant (subspecies S. geminata
rufa, now a junior synonym of the nominal form (Etter-
shank 1966)) was initially considered to be indigenous to
Asia and distinct from the New World forms (Creighton
1930). However, this red form was later found to occur
from Florida to Panama, which suggests a New World
origin for S. g. rufa (Creighton 1930). Trager (1991)
hypothesized that the Old World S. g. rufa derived from
a single introduction event based on the high uniformity
of the Asian and Pacific populations, but noted that the
dark West African specimens of the TFA more resembled
forms from the southeastern U.S.A. and the Caribbean.
Clearly, without exact knowledge of an invasive species’
population and invasion history, it is difficult to identify
native ranges.
Tropical fire ant is a pioneer species (Perfecto 1991), a
generalist keystone predator (Risch & Carroll 1982), and
readily occupies urban and agro-ecosystems (Perfecto
1991; Holway et al. 2002; Perfecto & Vandermeer 2011).
This ant also is well known as an agricultural pest (La-
kshmikantha et al. 1996; Holway et al. 2002) and is
thought to be one of the infamous Hispaniolan plague
ants of the early 16th century (Wilson 2005). Damage to
crops is mostly indirect, by tending honeydew-produc-
ing aphids and other Hemiptera (Risch & Carroll 1982;
Carroll & Risch 1984), but foragers also are known to
girdle citrus trunks (Wolcott 1933), deter specialist poll-
inators (Carroll & Risch 1984) and damage irrigation
tubing (Chang & Ota 1976). Aside from its negative
impacts on agriculture, TFA can impact both vertebrate
(Travis 1938; Kroll et al. 1973; Moulis 1996; Plentovich
et al. 2009) and invertebrate (Lake & O’Dowd 1991;
Way et al. 1998; Geetha et al. 2000) faunas. The latter
has led it to be described as an important biocontrol
agent of invertebrate pest species (Way et al. 2002; Way
& Heong 2009).
Research on invasive populations of TFA has lagged
far behind that of the other invasive ant species, partic-
ularly its congener S. invicta (Tschinkel 2006). System-
atic study and understanding of biological invasions are
not only important to better control, manage and pre-
vent establishment of invasive species, but also repre-
sent ideal model systems for the study of important
questions in evolution and ecology (e.g. rapid evolution,
genetics of adaptation to new environments) (Sax et al.
2007; Suarez & Tsutsui 2008; Simberloff et al. 2013).
Molecular genetic methods have become an important
tool for the study of biological invasions (Estoup & Gu-
illemaud 2010; Fitzpatrick et al. 2012; Kirk et al. 2013),
allowing inference of parameters otherwise difficult or
impossible to obtain by other means. Here, we present
the first population genetic study using microsatellites
and mtDNA sequences of the TFA on a global scale.
We describe the population structure of this ant, iden-
tify the most likely source population, estimate times
and infer routes of invasion, and document recent,
human-mediated long distance dispersal. Our data indi-
cate that TFA spread in the wake of the first global
trade network in the 16th century.
Materials and methods
Data generation
Sample collection. A total of 192 TFA colonies were sam-
pled from across the current geographic distribution of
Solenopsis geminata (Fig. 1, Table S1, Supporting
©2014 John Wiley & Sons Ltd
GLOBAL INVASION OF THE TROPICAL FIRE ANT 375

information). To generate statistically independent sam-
ples, only a single worker ant was used from each col-
ony. We roughly distinguish between the New World
(i.e. the Americas including the Gal
apagos) and Old
World (Africa, Asia and Australia). Samples of the three
most closely related species to S. geminata (S. xyloni,S.
amblychila, and S. aurea) were used as outgroups for the
phylogenetic analyses (Trager 1991; Pitts et al. 2005).
Mitochondrial DNA sequencing. We amplified 646 bp of
the mitochondrial gene cytochrome c oxidase 1 (CO1)
from 182 specimens of S. geminata and five outgroup
specimens following published procedures (Ross et al.
2003). Resulting sequences were checked for the pres-
ence of premature stop codons and indels and com-
pared to S. geminata mtDNA sequences from GenBank;
amplified fragments were consistent with mtDNA and
not nuclear homologs. Sequences were readily aligned
by hand and are deposited in GenBank (Table S1, Sup-
porting information).
Microsatellite genotyping. Forty-five microsatellite mark-
ers were amplified following methods outlined in previ-
ous studies (Ascunce et al. 2009, 2011). Briefly,
genotypes of a single female from 151 colonies were
determined at 45 nuclear microsatellite loci. We only
included individuals with genotypic data for >30 micro-
satellite markers in our analyses, resulting in 68 native
and 77 invasive samples retained for subsequent study.
Following best practice procedures (Bonin et al. 2004;
Hoffman & Amos 2005), we estimated microsatellite
genotyping error rate by independent allele scoring and
double-checking of random individuals, which also
helped to identify and eliminate errors that had
B
A
Fig. 1 Group memberships and projection in geographic and discriminant space. A) Geographic localities of samples. Samples are
RGB colour coded according to the first three principal components of the DAPC (using four clusters which clearly distinguish an
Old World (red), South American (dark blue), Caribbean (green) and North American (light blue) cluster. The Mesoamerican samples
are intermediate between the New World clusters. B) Scatterplot of eight clusters recognized by DAPC. The first three principal com-
ponents are shown (PC1: 26.1%, PC2: 7%, PC3: 4.8%). Cluster centroids are connected by a minimum-spanning tree.
©2014 John Wiley & Sons Ltd
376 D. GOTZEK ET AL.

occurred during scoring of alleles by hand. Loci were
screened for null alleles, Hardy–Weinberg equilibrium
and linkage disequilibrium.
Data analysis
Generally, we analysed the mtDNA sequence data and
nuclear microsatellite data separately to be able to com-
pare and contrast potential differences between mater-
nal and biparental gene flow and divergence. To
examine population genetic structure at a finer resolu-
tion in the nuclear data set, we also analysed the New
World and Old World/Australian samples separately.
Multivariate analyses of microsatellite data. As part of data
exploration, we performed several multivariate analy-
ses, which do not make strong assumptions about the
underlying genetic model (Jombart et al. 2009). We con-
ducted discriminant analysis of principal components
(DAPC), which finds principal components best sum-
marizing the differences between these clusters while
minimizing within-cluster variation (Jombart et al.
2010). DAPC was carried out with the adegenet 1.3-8
package (Jombart & Ahmed 2011) implemented in R
3.0.1 (R Development Core Team 2013). As we wanted
ana
€
ıve comparison between this method and the
Bayesian clustering methods, we assumed no prior
group membership and used sequential K-means clus-
tering (up to K=15) and model selection to infer
genetic groups. The Bayesian information criterion
(BIC) (Schwarz 1978) was used to assess the support for
the model (i.e. the number of clusters and assignment
of individuals to them), which is an efficient measure of
support (Lee et al. 2009). Although K-means clustering
is performed on transformed data using PCA, we used
all 145 principal components, thus retaining all varia-
tion of the original data. We used 10
9
iterations and 10
3
random starting centroids for each run, which both aid
in the convergence of the algorithm. To not overfit the
discriminant function, we chose the optimal number of
principal components for the DAPC using the
optim.a.score function. The a-score captures the trade-off
between the power of discrimination and overfitting
using too many principal components in the analysis by
measuring the proportion of successful reassignments
of the DAPC analysis compared to K-means clustering
(observed discrimination) and random clustering (ran-
dom discrimination). Subsequent DAPCs were con-
ducted with three principal components (explaining
37.8% of variance) and three retained discriminant func-
tions. The use of principal components also ensures that
information provided to the discriminant analysis is un-
correlated, thus removing potential effects of linkage
disequilibrium. Plotting discriminant functions with
minimum-spanning trees connecting the cluster cent-
roids allows a visual representation of affinities between
clusters.
To validate our decision to recognize substructure in
our data set, we compared stability of group member-
ship probabilities of the eight inferred clusters to ran-
dom clustering, giving an indication of how well
supported the clusters are. Low group membership
probabilities suggest that clusters are not supported by
the data. As retention of too many principal compo-
nents can result in overinflated membership probabili-
ties and hence perfect discrimination, we conservatively
used only three principal components and discriminant
functions.
We estimated how well supported the group mem-
bership was relative to collection locality. Posterior
group memberships can be used to indicate admixture
or misclassification when prior groups are used to con-
duct the DAPC. We repeated the DAPC, grouping sam-
ples by region or country of origin (Fig. 2B).
Membership assignment probabilities to each region
based on retained discriminant functions were com-
pared to the groups identified by K-means. Posterior
assignment probabilities to a priori selected groups
indicate the validity of a given group.
Bayesian clustering of microsatellite data. As the multi-
variate methods inferred cluster sizes varying up to
10-fold, we used the Bayesian clustering software
STRUCTURAMA 2.0 (Huelsenbeck et al. 2011) to assign
samples to clusters, which has been shown to outper-
form other methods (Fogelqvist et al. 2010; Hausdorf
& Hennig 2010). STRUCTURAMA uses a Dirichlet process
prior (DPP) to calculate the posterior probability of
numbers of populations in the data set (Pella & Masu-
da 2006), which has been shown to be an efficient and
powerful method to infer population structure (Huel-
senbeck & Andolfatto 2007; Onogi et al. 2011; Shrin-
garpure et al. 2011). STRUCTURAMA implements a
Hierarchical Dirichlet Process model (Teh et al. 2006)
to accommodate admixture of infinite ancestral popu-
lations while treating the number of populations as a
random variable.
As this model is computationally demanding, we ran
short exploratory analyses to assess the impact of the
model (admixture vs. no admixture), concentration prior
and hyperpriors on the analysis, with five or ten repli-
cates for each model to test for consistency between
analyses. Following the recommendations of Franc
ßois &
Durand (2010), we employed various models using Ba-
yes Factor to guide model choice. While we recognize
that the harmonic mean reported by Structurama is a
poor estimator of the marginal likelihood as its variance
is often unreasonably large, we found that K=4 was
©2014 John Wiley & Sons Ltd
GLOBAL INVASION OF THE TROPICAL FIRE ANT 377

most often chosen, regardless of model used. In addition
to the marginal likelihood (which were often very similar
between analyses), we also used low sum of squares
score for mean partition (i.e. the partition distance; Huel-
senbeck & Andolfatto 2007) to identify the best analysis.
This partition distance, which is the minimum number
of individuals that must be moved between populations
in one of the partitions to make it identical to the other
partition (Gusfield 2002), measures the stability the parti-
tioning scheme of a given analysis.
For the final model, we fixed ato give the desired
mean of the prior for the expected number of popula-
tions [E(K)=5] and modelled admixture using a
gamma distribution shape and scale parameters
(hyperpriors) set to 1 each. All analyses were con-
ducted with the program default of sampling from a
single chain run for 100 000 generations with a burnin
of 100.
General description of genetic variation. Measures of
genetic diversity and population differentiation were
separately estimated for the mtDNA and microsatellite
data sets using the groups identified by the clustering
methods. We analysed the mtDNA using MEGA 5.2.2
(Tamura et al. 2011) and DNASP 5.10.01 (Librado & Ro-
zas 2009). All sequences were included in these
analyses, not only unique haplotypes. For the nuclear
data, we calculated indices of genetic variation with
GENODIVE 2.0 (Meirmans & Van Tienderen 2004) and
GENEALEX 6.5 (Peakall & Smouse 2012). Following Meir-
mans & Hedrick (2011), we report pairwise population
F
ST
, G’’
ST
and D values (Hedrick 2005; Jost 2008;
W-13 CAS6
CAS6
CBR QMX QMX2
Malagasy
India
Taiwan
Philippines
Thailand
Hawaii
Christmas Isl.
Australia
Texas
Mexico
Caribbean
Venezuela
Florida
Trinidad&Tobago
French Guiana
Peru
Brazil
China
Mesoamerica
1.0
0.5
0.0
0.0
0.0
0.0
0.0
0.0
1.0
1.0
0.5
0.6
0.3
0.0
0.8
0.4
1.0
1.0
0.0
1.0
A
B
Fig. 2 Cluster assignment and admixture proportions as inferred by A) Structurama for K=4 and B) DAPC using collection sites as
prior clusters. The heatmap (red =1, white =0) shows proportions of successful reassignment of individuals to their original clus-
ters. Sites and their successful reassignments proportions are in rows, and individuals and their prior cluster (i.e. collection sites; blue
crosses) are in columns. Large values indicate clear-cut clusters, and low values suggest admixed or poorly supported groups. Well-
supported sites (≥0.8 reassignment proportions) are indicated in bold.
©2014 John Wiley & Sons Ltd
378 D. GOTZEK ET AL.

Gerlach et al. 2010; Meirmans & Hedrick 2011) as none
is an ideal summary statistic. We used nearly unbi-
ased estimates of heterozygosity (H
S
and H
T
) (Nei &
Chesser 1983) to reduce the bias of D and G’’
ST
values
due to small sample size of some of the inferred pop-
ulations. Statistical significance of pairwise F
ST
values
was tested using an analysis of molecular variance
with 10 000 permutations (Excoffier et al. 1992; Micha-
lakis & Excoffier 1996) with Bonferroni correction (Rice
1989).
Phylogenetic analyses. To obtain a more explicitly phylo-
genetic perspective of the relationships between clusters
using the microsatellite data, we constructed a neigh-
bour-joining tree (Saitou & Nei 1987) of interindividual
Nei’s chord distances (D
A
; Nei et al. 1983) using the
neighbor program in the PHYLIP 3.69 package (Felsenstein
2005). This is expected to reflect genealogical relation-
ships when a large number of informative markers are
used (Chakraborty & Jin 1993; Bowcock et al. 1994). We
used Nei’s chord distance as it has been shown to out-
perform other distance measures for reconstructing
phylogenetic trees using microsatellite data (Takezaki &
Nei 1996, 2008). One thousand bootstrap replicates were
used to estimate branch support.
Bayesian inference of the mtDNA genealogy was per-
formed using BEAST 1.7.5 (Drummond et al. 2012). We
estimated the best fitting partition and model of nucle-
otide substitution using the greedy heuristic search
algorithm in PARTITIONFINDER 1.1. (Lanfear et al. 2012),
which selected the HKY+I, F81 and GTR+G model for
the first, second and third codon position, respectively.
Clock and tree models were linked across partitions,
but substitution models remained unlinked. We
applied an exponential size coalescent model (Griffiths
& Tavar
e 1994) with a lognormal prior on the coales-
cent size parameter to estimate the tree. All other
priors were kept at default. As we did not extensively
sample the outgroup species and this violates the
assumption of random sampling of OTUs in coalescent
analyses (Wakeley 2008), we conducted the phyloge-
netic analyses with and without outgroups. No signifi-
cant differences were found. Ten million generations
were run, sampled every 1000th generation, of which
the first 10% were discarded as burn-in. Stationarity of
the runs was assessed in TRACER 1.5 (Rambaut & Drum-
mond 2007) by plateauing of log-likelihoods and effec-
tive sample sizes (ESS) >200. Four independent runs
were combined, all of which had similar mean log-like-
lihoods. To test for undue influence of the priors on
the posterior parameter estimates, we compared poste-
riors from analyses estimated by sampling with and
without (i.e. sampling only from the prior distribution)
data.
Assignment and exclusion tests of microsatellite
data. Assignment and exclusion tests were carried out
with GENECLASS2 (Piry et al. 2004) using Bayesian
approaches, which generally outperform distance and
frequency-based approaches (Paetkau et al. 2004). We
used the four native clusters as determined by the clus-
tering methods as reference populations. Assignment
tests were used to assign individuals collected in non-
native areas to reference clusters from the native range.
Assignment probabilities of ≥95% to a given source
population were considered to be significant support
for the native population to be the source of invasives.
In the absence of statistically significant assignment
probabilities, the native reference population showing
the highest average likelihood value was considered the
most likely source population. Results of the Bayesian
assignment tests were independent of the prior used
(Rannala & Mountain 1997; Baudouin & Lebrun 2000).
However, assignment methods assume that the actual
source population is represented among the reference
populations and can thus erroneously assign individu-
als to one or another reference population with high
probability if the true source population has not been
sampled (Paetkau et al. 2004). Exclusion tests are not
prone to such an error, as they can exclude all reference
populations as putative sources of introduced popula-
tions. To perform exclusion tests, we used the resam-
pling algorithm of Paetkau et al. (2004), as other Monte
Carlo resampling methods (Rannala & Mountain 1997;
Cornuet et al. 1999) exclude an excess of resident indi-
viduals. All simulations were conducted with 100 000
simulated individuals and an alpha level of 0.01.
Once individuals from the invasive range were
assigned to reference source populations, we tested
whether any of these were first-generation migrants. As
we could not be confident that we have sampled every
native population, we used both the likelihood of an
individual’s genotype within the population where the
individual has been sampled (L_home) and the ratio of
L_home to the highest likelihood value among all popu-
lations excluding the population where the individual
was sampled (L_home/L_max) as statistical criteria for
the detection of first-generation migrants (Paetkau et al.
2004). While the likelihood ratio L_home/L_max has
more power than the L_home statistic, it is only appro-
priate if all source populations for immigrants have
been sampled. The L_home statistic is more appropriate
when some source populations are clearly missing (Pae-
tkau et al. 2004; Piry et al. 2004).
Testing invasion scenarios using approximate Bayesian com-
putation. We used approximate Bayesian computation
(ABC; Beaumont 2010; Bertorelle et al. 2010) to compare
invasion scenarios and infer the invasion history of the
©2014 John Wiley & Sons Ltd
GLOBAL INVASION OF THE TROPICAL FIRE ANT 379

tropical fire ant. ABC is a Bayesian inference approach
that does not require the specification of a likelihood
function and can hence be used to efficiently carry out
complex model-based inferences using large numbers
of simulated data sets which are compared to the
observed data set using summary statistics. All steps of
the analyses were conducted with DIYABC 2.0 (Cornuet
et al. 2014) using only microsatellite data. To limit the
number of scenarios to test and for lack of a robust and
reliable evolutionary history of the native populations
due to rooting problems of both the microsatellite and
mtDNA trees, we first sought to determine the source
population(s) of the introduced clusters. To this end,
we tested three competing scenarios (Fig. S3, Support-
ing information), with successively more narrowly
defined putative source populations. For these analyses,
we considered the invasives to belong to one popula-
tion and successively more narrowly defined two com-
peting putative source populations, loosely following
the clusters recovered with increasing K: (A) the South
American and TexMex/Meso1/Meso2/Caribbean clus-
ters; (B) the TexMex/Meso2 and Meso1/Caribbean clus-
ters; and (C) the TexMex and Meso2 clusters. The three
competing scenarios allowed the introduced population
to derive from one or the other source population or to
be admixed from both.
Second, we tested more complex and specific inva-
sion scenarios in analyses D from which we also
derived parameter estimates of invasion times, bottle-
neck sizes and duration (Fig. S4, Supporting informa-
tion). For this analysis, we recognized the Australian,
Indo-Pacific and Meso2 clusters and we sought to dis-
tinguish between four invasion scenarios: an indepen-
dent invasion, a serial invasion, an independent
invasion from an unsampled ghost population and a
serial invasion from an unsampled ghost population
(Fig. S4A, Suporting information).
Following Cornuet et al. (2008), we considered only
the simple generalized stepwise-mutation model
(Estoup et al. 2002) to reduce the number of parameters.
We left the mutation model at default settings. We also
implemented a 5:6 female to male sex ratio (Travis
1941) and haplo-diploid locus model for all analyses,
although these settings did not substantially influence
the results (not shown). We only considered uniform
priors and the following constraints on parameters:
db <t1 <t2 <ta, Nb <N. For analyses A–C, we kept
all parameter priors at their default. We set N =[10–
10 000], t1,2 =[150–1000], ta =[10–10 000] and db =[1–
1000] for analysis D. We used mean size variance, mean
number of alleles, mean Garza-William’s M index
across loci, (dl)
2
distance between samples, mean size
variance across loci, mean number of alleles across loci
and F
ST
as summary statistics. We produced reference
tables with 10
5
simulated data sets per scenario for
analyses A–C and 10
6
data sets per scenario for analysis
D for parameter estimation. To lighten the computa-
tional burden for analyses D), we used LDA-trans-
formed summary statistics (Estoup et al. 2012). For
analyses (A–C), we used raw summary statistics.
To reveal model (scenario) and/or prior misspecifica-
tion prior to full analyses, we pre-evaluated scenario
and prior distributions using both PCA and locating
observed within simulated summary statistics (Cornuet
et al. 2010) to verify that at least one prior–scenario
combination can produce simulated data sets that are
sufficiently close to the observed data set.
We estimated posterior probabilities of competing
scenarios using polychotomous logistic regression of the
1% simulated data sets closest to the observed data set
(which is generally more discriminant than the direct
estimates). We assessed confidence in scenario choice
by computing 95% confidence intervals and type I and
II errors for the most probable scenario of analysis D.
We validated the choice of thresholds by repeating pos-
terior probability calculations with fewer (0.1% and
0.0025%) and more (10% and 0.1%) simulated data sets
(Cornuet et al. 2008; Guillemaud et al. 2010), which pro-
duced similar results.
We estimated parameters after applying a logit trans-
formation to the parameter values of the 1% simulated
data sets closest to the observed data set. Use of other
transformations (log or log-tangent transformation; Es-
toup et al. 2004; Hamilton et al. 2005) produced similar
results (not shown). We additionally measured the per-
formance of parameter estimation by calculating the
median of the absolute error divided by the true param-
eter value of the 500 pseudo-observed data sets simu-
lated using the median and mode of the posterior
distribution as point estimates (relative median absolute
errors, RMAE). Finally, we performed model checking
with all summary statistics not used for the primary
analysis using both PCA and ranking of summary sta-
tistics (Cornuet et al. 2010) to assess the goodness-of-fit
of our model/parameter/posterior combination.
Results
Population structure
Four to nine clusters were supported by K-means clus-
tering and Bayesian methods (Fig. 1, Fig. S1A, Support-
ing information). Four clusters represent the simplest
summary of the data at the highest hierarchical level
(Evanno et al. 2005), but the substantial substructure, in
both geographic and discriminant space (Fig. 1), sug-
gests eight clusters among which genetic differentiation
was pronounced and always statistically significant
©2014 John Wiley & Sons Ltd
380 D. GOTZEK ET AL.

after Bonferroni correction (P<0.0005; Table S3, Sup-
porting information). Stability of group membership
probabilities, derived from proportions of successful
reassignments based on retained discriminant functions
of the DAPC based on maximal substructuring, was
high (100% for the New World clusters and >82% for
the Old World clusters compared to 0–50% of random
clustering (Fig. S2, Supporting information).
Of the eight clusters, five were in the New World
(Fig. 1B) and were consistently recovered in both global
and New World-only analyses. The South American
cluster contains all specimens from the Amazon, Gui-
ana Shield, Brazilian coast and Florida. The Caribbean
cluster contains samples from Panama, central Venezu-
ela, the Dominican Republic, the Grand Turks, all sam-
ples from the Gal
apagos and one individual from
Reunion (CAS6). The TexMex cluster contains samples
from Texas and northern Mexico. The first Mesoameri-
can cluster (Meso1; containing samples from Costa Rica,
Nicaragua, Honduras, Guatemala and a singleton from
Mexico) partially overlaps the more northern second
Mesoamerican cluster (Meso2; with samples from Mex-
ico, Guatemala, Belize and a singleton from Costa Rica).
Finally, three clusters were identified in the Old
World (Fig. 1B). Minimum-spanning trees of cluster
centroids and distance in discriminant space of New to
Old World samples suggest a close affinity between
Mesoamerica and the Old World. Global analyses and
Bayesian analyses of Old World samples only differenti-
ate between the Australian cluster (containing all sam-
ples from Australia, Christmas Island and a singleton
from Reunion (CAS3)) and an Indo-Pacific cluster (con-
taining all other Old World samples). K-means cluster-
ing of Old World samples supports further
substructure. At K=3, the Indo-Pacific cluster is split
into the Indo-Pacific1 (containing samples from Hawaii,
Taiwan, India and all samples from China, Thailand
and Madagascar) and Indo-Pacific2 clusters (containing
the remaining samples from Hawaii, Taiwan and India,
all samples from the Philippines and one sample from
Mauritius). Due to the relatively poor separation in dis-
criminant space and decreasing stability of group mem-
bership relative to the New World clusters in the global
analysis, we chose to ignore finer substructure (Fig. 2B,
Fig. S2, Supporting information).
Estimates of diversity
Estimates of genetic diversity show a reduction in the
Old World clusters in both the nuclear and cytoplas-
mic genomes, pointing to an invasive origin of these
samples (Tables S3 and S4, Supporting information).
There were a total of 64 unique haplotypes in 186
mtDNA sequences generated. The putative native
populations (n=106) contain the vast majority of
mtDNA diversity (57 haplotypes; Table S2, Supporting
information), whereas the putative invasive ants
(n=75) shared a total of 4 haplotypes. One of these
(CAS6 from Reunion Island) is shared with two native
specimens collected in the Caribbean islands of Turks
and Caicos. Two specimens from Hawaii and a single-
ton from Christmas Island share the second haplotype.
The third haplotype is unique to a sample from the
Philippines. All remaining Old World samples (93%;
n=70) and a single specimen from Brazil (BrGem)
share the fourth and most common haplotype. This
haplotype is thus found throughout the Indo-Pacific
region (Madagascar, Reunion Island, Mauritius, Aus-
tralia, India, the Philippines, Thailand, China and Tai-
wan). The low number of haplotypes is reflected in
the very low estimate of mitochondrial diversity (Table
S2, Supporting information). The nuclear genome of
the Old World clusters showed similar patterns of
reduced genetic diversity relative to the native popula-
tions, but to a lesser degree (Table S2, Supporting
information).
Admixture
Bayesian posterior assignment probabilities of individu-
als to a single cluster were generally very high (>99%;
Fig. 2A), indicating that there is little evidence for
admixture between clusters, with few exceptions. Two
Meso2 samples from the Yucatan (QMX and QMX2)
contained a minority proportion (6.7% and 25.6%,
respectively) of their genome from the Meso1 cluster. A
Brazilian sample, W-13, was also admixed, with a small
proportion (7.8%) of its genome derived from the Old
World clusters. Finally, CBR from Jalisco, Mexico was
estimated to derive 23.3% of its genome from the Old
World cluster.
Reassignment of individuals to areas of origin based
on the discriminant function of the DAPC (Fig. 2B)
resulted in seven sites having high (≥80%) reassign-
ment proportions, indicating clear-cut groups. Six
strongly differentiated clusters can be visually recog-
nized mirroring the results of the clustering methods,
validating our recognition of substructure in the data
set, especially in the New World. Assignment propor-
tions for sites and clusters outside the cluster inferred
by the clustering methods were generally zero, indicat-
ing no admixture between clusters. Assignment propor-
tions were often distributed across sites within a given
cluster, indicating lack of within-cluster structure. Sub-
structure is also visible in the Old World, suggesting
that recognition of an Australian cluster is valid. CAS6
from Reunion is again not reassigned to its sampled
site.
©2014 John Wiley & Sons Ltd
GLOBAL INVASION OF THE TROPICAL FIRE ANT 381

Phylogenetic analyses
Both the nuclear NJ tree and mtDNA coalescent generally
recover the same groups as the clustering methods
(Fig. 3). Importantly, in both data sets, CBR from south-
ern Jalisco, Mexico is always the sister group to the inva-
sive clade (with high support) and CAS6 from Reunion
falls within a clade of Caribbean cluster samples (the
Dominican Republic and Venezuela). In the NJ tree
(Fig. 3A), the Philippine and most Taiwanese samples
are sister to all samples within the Indo-Pacific clade, and
the Australian cluster is situated well within the invasive
samples. Despite uncertainty in rooting and lack of sup-
port for the deeper nodes in the mtDNA tree (Fig. 3B),
the placement of haplotypes from the invasive range is
consistent and well supported. The three Old World
haplotypes form a well-supported monophyletic clade
(PP =1.0), which is always placed within a clade of North
American samples.
Assignment and exclusion tests
We attempted to assign the Old World specimens to a
native population (as identified by the clustering meth-
ods; Paetkau et al. 2004). Virtually all samples of the
invasive range were assigned to the native Meso2 clus-
ter with >95% probability (Table S4, Supporting infor-
mation), with five exceptions. CAS6 from Reunion had
a 100% assignment probability to the Caribbean cluster
and four samples had a majority assignment to the
TexMex cluster, but only one of these was significant
at 95% assignment probability. Assignment tests can
incorrectly assign individuals to a reference population
with high probability if the true source population is
not sampled. Exclusion tests do not suffer from this
problem. Exclusion probabilities for our data set were
generally very low (P<0.05), suggesting that we have
not sampled the true source population. However,
Meso2 could not be excluded as source population for
most samples (at P≥0.01). Meso2 was excluded as
source population only for CAS6, for which the Carib-
bean cluster could not be excluded (P=0.164). Both the
direct likelihood (L_home) and likelihood ratio
(L_home/L_max) methods identified CAS6 from
Reunion as a first-generation migrant with high proba-
bility (<0.0001%). No other F0 migrants were identified.
Inference of invasion scenarios using ABC
Progressive exclusion of native TFA populations as the
source for the invasive Old World populations identified
the Meso2 cluster as the most probable source. All other
scenarios where the invasives are derived from other
native populations (even through admixture) were
rejected (Fig. S3, Supporting information). Of the more
complex models using the Meso2, Australian and Indo-
Pacific clusters, a serial invasion scenario (where the Aus-
tralian cluster is derived from the Indo-Pacific cluster)
was always preferred over an independent invasion sce-
nario, indicating one initial invasion event took place
with subsequent dispersal within the Old World (Fig.
S4A, Supporting information). Scenarios including an un-
sampled ghost population were also always preferred
over scenarios in which the source population had been
AB
Fig. 3 Phylogenetic hypotheses based on
nuclear (A) and cytoplasmic (B) ge-
nomes. Cluster membership is indicated
by colour (red: Indo-Pacific & Australia;
green: South America; blue: TexMex &
Meso2; purple: Caribbean & Meso1). The
three Mexican samples which are incon-
sistently placed between the Meso1 and
Meso2 clusters are in black. Branch sup-
port is indicated by red (1.0) and black
(0.95–0.99) stars; only deeper nodes are
labelled. A) Bootstrapped (1000 repli-
cates) neighbour-joining tree using micro-
satellite Nei’s chord distances. B)
Bayesian maximum clade credibility tree
based on mtDNA sequences.
©2014 John Wiley & Sons Ltd
382 D. GOTZEK ET AL.

sampled. Scenario 3 (serial invasion from a ghost popula-
tion) had the highest posterior probability (PP =0.6575),
and its 95% CI (0.5802, 0.7348) did not overlap with the
95% CI of the next best scenario (scenario 1; Fig. S4A,
Supporting information). Type I and II errors were esti-
mated at 0.31 and 0.124, respectively.
Posterior distributions of parameters are well esti-
mated with peaked posteriors and clear differences to
prior distributions and generally low RMAE values
(Table 1, Fig. S4B, Supporting information). Estimates of
population size indicate large effective population sizes
for both native and invasive populations. Founding
propagule size estimates are also rather large, which
explains the modest loss of nuclear diversity in the
introduced populations. The rather old divergence time
of the ghost population suggests it is a native and not
an invasive population. Divergence estimates of the
invasive populations using the mode indicate the
founding event occurred approximately 241 generations
ago (95% CI 178, 938). This places the initial invasion
event in the early 16th century, based on colony repro-
ductive ages of 2 years. The Australian cluster diverged
approximately in the mid-19th century (~163 genera-
tions). The bottlenecks were estimated to be quite long
(~100 generations).
Discussion
We use nuclear and cytoplasmic genotype data for sam-
ples collected throughout the known global range of the
TFA, one of the most widely distributed pest ants, and
describe its invasion history in great detail. We show
that the Old World samples are introduced and derived
from a single New World source population. Multiple
lines of evidence support the Meso2 cluster and more
specifically, southwestern Mexico, as the likely source,
including results from clustering methods, admixture
proportions, assignment and exclusion tests, measures
of genetic diversity, tree-based methods and ABC.
Creighton (1930) was the first to recognize the connec-
tion between the Old World and Mexican populations
of the TFA, pointing out that the Old World S. g. rufa
form was especially common in Texas and Mexico. Our
analyses further suggest the invasive populations origi-
nated during one main introduction period followed by
more recent long distance dispersal events from other
source populations.
Divergence time estimates of the invasive popula-
tions from the native source population suggest that
the main founding event(s) occurred in the early 16th
century, a time of burgeoning colonialism and trade.
These results are consistent with historical data. Since
the early 15th century, Europeans explored and traded
with Africa, Asia, the Americas and Oceania. This Age
of Exploration culminated in the first truly global trade
network by way of the Spanish Manila galleon trade,
which connected the Asian and American markets
(Flynn & Gir
aldez 2004). For 250 years (from 1565 to
1815 CE), one or two Spanish galleons set sail annually
to trade New World silver for Chinese silk, porcelain
and spices in Manila (Spate 2004). Importantly, the
Spanish-Mexican colonists not only brought maize,
sweet potatoes and other crops to Asia, but addition-
ally carried rock, sand and soil for ballast (Carlton
1992), giving ample opportunity for entire TFA colo-
nies to have been inadvertently transported during
these long trans-Pacific voyages. Manila served as a
hub for Spanish trade with the Chinese Ming and Qing
Dynasties via the southern Chinese province of Fujian
and to a lesser degree with Canton (Guangdong) prov-
ince and the Molucca Spice Islands. Thus, the Spanish
Philippines were well connected with the main trade
centres of Asia, not only via the Portuguese and Dutch
trade networks, but perhaps more importantly through
an already existing vast and sophisticated Asian net-
work (Spate 2004; Bjork 2005). The expansive and
dynamic 16th–18th century trade system would have
allowed the TFA to rapidly spread to the major eco-
nomic and agricultural centres of coastal Asia, Africa,
and the Pacific and Indian Oceans.
The Manila galleons sailed from the Pacific port of
Acapulco, Oaxaca, in southwestern Mexico. Even
though we lack samples from this area, our sample
from Jalisco (CBR) is both geographically closest to
Table 1 Demographic parameter estimates, their relative med-
ian absolute errors (RMAE) and their 95% confidence interval
of scenario 3 using ABC based on simulated 1% of simulated
data sets closest to the observed values. All population sizes
(N) are effective population sizes. Nxb: bottlenecked founding
populations, db: bottleneck duration, ty: divergence time. Time
(t and db) is given in generations
Parameter Median
RMAE
(median) Mode
RMAE
(mode)
q
[2.5]
q
[97.5]
N1
(Meso2)
9230 0.141 9430 0.150 7110 9960
N2
(Austral.)
4290 0.184 1800 0.253 990 9590
N2b 589 0.306 381 0.325 116 2770
N3
(Indo-P.)
1530 0.147 1010 0.172 354 7790
N3b 357 0.371 156 0.452 52 2450
Nu
(ghost)
3920 0.245 2350 0.277 801 9530
db 123 0.452 77 0.610 11 427
t1 188 0.199 163 0.208 153 357
t2 382 0.149 241 0.176 178 938
ta 3280 0.188 2980 0.220 850 8660
©2014 John Wiley & Sons Ltd
GLOBAL INVASION OF THE TROPICAL FIRE ANT 383

Acapulco (~450 km) and genetically closest to the intro-
duced populations. This lack of sampling of popula-
tions around Acapulco and southwestern Mexico could
explain divergence of the introduced populations from
the native samples (e.g. the lack of shared mitochon-
drial haplotypes, the inference of an unsampled native
ghost population and low power of the exclusion tests).
However, low mitochondrial diversity stands in con-
trast to the relatively high nuclear genetic diversity.
Although the low mtDNA haplotype variation in the
invasive ants suggests a severe bottleneck during the
founding event or selective sweeps associated with
adaptation to the new environment, the nuclear genetic
diversity of invasives is comparable to the diversity
estimates of native populations and estimates of found-
ing propagule size based on nuclear data number in the
hundreds. Several factors may explain this observation.
First, genetic bottlenecks are expected to be more severe
for the cytoplasmic than the nuclear genome due to its
smaller effective population size (Moritz et al. 1987).
Second, founding propagules may have contained mul-
tiple reproductive queens (polygyny; Banks et al. 1973;
Mackay et al. 1990). Queens in polygyne fire ant colo-
nies are often derived from the same matriline (Ross
et al. 1996), resulting in a shared cytoplasmic genome,
while the nuclear genome is not similarly impacted.
While the social form of most introduced populations
has not been conclusively determined, field observa-
tions of nest structure and spacing suggest that polyg-
yny is common. Third, selective sweeps due to
adaptation to the newly invaded environment or endo-
symbiont-driven reductions can result in reduced mito-
chondrial diversity (Hurst & Jiggins 2005; Suarez &
Tsutsui 2008). On the other hand, loss of genetic diver-
sity can be modest or nonexistent if the founding prop-
agule is large, the bottleneck brief and population
growth rapid (Suarez & Tsutsui 2008). Moreover, histor-
ical processes operating in the source population(s)
shape the level of genetic diversity introduced during
the invasion process (Taylor & Keller 2007).
Despite the relative genetic homogeneity of the inva-
sive clusters, several inferences on the spread of the TFA
within the invasive range can be made. First, the DAPC,
minimum-spanning tree and measures of population dif-
ferentiation indicate that the Indo-Pacific2 cluster is clos-
est to the source population, suggesting that the other
non-native clusters stem from there. The Indo-Pacific2
cluster, containing all Philippine and most Taiwanese
samples, also has the highest mtDNA haplotype diversity
of the invasive clusters, as the two rare haplotypes are
found within this cluster. The NJ tree places the Philip-
pine samples sister to the remaining invasives, followed
by most Taiwanese samples. This is congruent with
direction of Spanish trade from Acapulco via Manila to
southern China and Formosa (Taiwan), where the Span-
iards briefly settled in the 17th century. Finally, all analy-
ses suggest that the Australian cluster is derived from the
Indo-Pacific cluster and does not represent an indepen-
dent invasion event. The ABC analysis estimates the
invasion of Australia and the Christmas Island occurred
more recently (188–376 years ago), which is consistent
with the relatively late (ca. 1869 CE) European settlement
of northern Australia.
Human-mediated movement of the TFA is still ongo-
ing. We identified both samples from Reunion as recent
or first-generation immigrants. While CAS3 may not be
an F0 migrant, CAS6 (belonging to the Caribbean clus-
ter) clearly is. Such long distance dispersals are a well-
known feature of invasions (Suarez et al. 2001). There
also is evidence for reintroduction of the TFA into its
native range. Two samples from Bahia, Brazil (BrGem
and W-13), have admixed genomes derived from the
invasive and South American populations. A likely
explanation for the genetic footprint of the invasive clus-
ter in Brazil is due to a secondary invasion of TFA from
the invasive range. Given that both samples were col-
lected close to the city of Salvador, the economic and
cultural hub of northeastern Brazil, it is perhaps not sur-
prising to find evidence of a secondary invasion here.
We demonstrate that the TFA achieved its current pan-
tropical distribution through Spanish trade with Asia
during the 16th–18th century. One major introduction
event from a single source population first brought the
TFA from southwestern Mexico to the Philippines and
then China, and from there, it was dispersed throughout
the Old World tropics. However, movement of this
highly invasive ant is still ongoing as indicated by the
finding of long distance, first-generation migrants from
the Caribbean to Reunion, or reintroduction from the
invasive populations to the native range in Brazil. To the
best of our knowledge, this is the first documented evi-
dence of human-mediated introduction of a highly inva-
sive pest species during the first truly global trade
network. Our results are in contrast to many other unin-
tentionally introduced social insects where data largely
supports patterns of establishment much later, ranging
from the late 19th century to mid-20th century (Tsutsui
et al. 2001; Krushelnycky et al. 2005; Ascunce et al. 2011;
Beggs et al. 2011; Evans et al. 2013). However, some inva-
sive ant distributions look surprisingly similar to that of
TFA and already appeared to be well established by the
19th century (Wetterer 2005, 2008, 2011, 2012; Wetterer
et al. 2009). Thus, it is likely that other ants were spread
along the same maritime trade routes, a conclusion sup-
ported by the fact that soil often was used as ballast
(which was common well into the early 20th century;
Carlton 1992), a habit that would suggest the transport of
soil nesting ant colonies is highly likely. Further study of
©2014 John Wiley & Sons Ltd
384 D. GOTZEK ET AL.

the TFA and other ant species in their native and invasive
ranges (e.g. Suarez et al. 2001; Ascunce et al. 2011) will
shed light on the causes, processes and consequences of
biological invasions and allow a richer assessment of the
impact of human history on contemporary biogeographic
patterns.
Acknowledgements
We thank the following scientists and institutions for gener-
ously contributing specimens: B. Hoffman, K.L. Heong and the
International Rice Research Institute, S. Hasin, V. Framenau
and the Western Australian Museum, W. Tschinkel, J. Longino,
S.C-C. Yang, L. van Aesch, D. Cherix, L. Davis, J. Orivel, B.
Fisher, E. LeBrun, H. Herrerra, R. Arauco, K. Ross, and the
National Museum of Natural History for sharing samples. E.
Caroll kindly provided assistance in the laboratory. M. Ciolek
pointed out important literature. K. Ross, J. Wetterer and three
anonymous reviewers improved the manuscript with helpful
comments and discussions. A.V.S. and D.G. gratefully
acknowledge financial support from NSF (DEB 1020979) and
USDA APHIS (292 AG 11-8130-0068-CA).
References
Aplin KP, Suzuki H, Chinen AA et al. (2011) Multiple geo-
graphic origins of commensalism and complex dispersal his-
tory of black rats. PLoS ONE,6, e26357.
Ascunce MS, Bouwma AM, Shoemaker DD (2009) Character-
ization of 24 microsatellite markers in 11 species of fire ants
in the Genus Solenopsis (Hymenoptera: Formicidae). Molecu-
lar Ecology Resources,9, 1476–1479.
Ascunce MS, Yang C-C, Oakey J et al. (2011) Global invasion his-
tory of the fire ant Solenopsis invicta.Science,331, 1066–1068.
Austin CC (1999) Lizards took express train to Polynesia. Nat-
ure,397, 113–114.
Banks WA, Plumley JK, Hicks DM (1973) Polygyny in a colony
of the fire ant Solenopsis geminata.Annals of the Entomological
Society of America,66, 234–235.
Baudouin L, Lebrun P (2000) An operational Bayesian
approach for the identification of sexually reproduced cross-
fertilized populations using molecular markers. Proceedings
of the International Symposium on Molecular Markers for
Characterizing Genotypes and Identifying Cultivars in Horti-
culture, March 6–9, 2000, Montpellier, France. Leuven: Inter-
national Society Horticultural Science, 81–93.
Beaumont MA (2010) Approximate Bayesian computation in
evolution and ecology. Annual Review of Ecology, Evolution,
and Systematics,41, 379–406.
Beggs JR, Brockerhoff EG, Corley JC et al. (2011) Ecological
effects and management of invasive alien Vespidae. BioCon-
trol,56, 505–526.
Bertorelle G, Benazzo A, Mona S (2010) ABC as a flexible
framework to estimate demography over space and time:
some cons, many pros. Molecular Ecology,19, 2609–2625.
Bjork K (2005) The link that kept the Philippines Spanish: Mex-
ican merchant interests and the Manila Trade, 1571–1815.
Journal of World History,9,25–50.
Bonin A, Bellemain E, Bronken Eidesen P, Pompanon F, Broch-
mann C, Taberlet P (2004) How to track and assess genotyp-
ing errors in population genetics studies. Molecular Ecology,
13, 3261–3273.
Bowcock AM, Ruiz-Linares A, Tomfohrde J, Minch E, Kidd JR,
Cavalli-Sforza LL (1994) High resolution of human evolu-
tionary trees with polymorphic microsatellites. Nature,368,
455–457.
Bright C (1999) Invasive species: pathogens of globalization.
Foreign Policy,1999,50–64.
Carlton JT (1992) Blue immigrants: the marine biology of mari-
time history. Log,44, 499–509.
Carroll CR, Risch SJ (1984) The dynamics of seed harvesting in
early successional communities by a tropical ant, Solenopsis
geminata.Oecologia,61, 388–392.
Chakraborty R, Jin L (1993) Determination of relatedness
between individuals using DNA fingerprinting. Human Biol-
ogy,65, 875–895.
Chang VCS, Ota AK (1976) Fire ant damage to polyethylene
tubing used in drip irrigation systems. Journal of Economic
Entomology,69, 447–450.
Cornuet J, Piry S, Luikart G, Estoup A, Solignac M (1999) New
methods employing multilocus genotypes to select or exclude
populations as origins of individuals. Genetics,153, 1989–2000.
Cornuet J-M, Santos F, Beaumont MA et al. (2008) Inferring
population history with DIY ABC: a user-friendly approach
to approximate Bayesian computation. Bioinformatics,24,
2713–2719.
Cornuet J-M, Ravign
e V, Estoup A (2010) Inference on popula-
tion history and model checking using DNA sequence and
microsatellite data with the software DIYABC (v1.0). BMC Bio-
informatics,11, 401.
Cornuet J-M, Pudlo P, Veyssier J et al. (2014) DIYABC v2.0: a soft-
ware to make approximate Bayesian computation inferences
about population history using single nucleotide polymor-
phism. DNA sequence and microsatellite data. Bioinformatics,
30, 1187–1189.
Creighton WS (1930) The new world species of the genus
Solenopsis (Hymenop. Formicidae). Proceedings of the American
Academy of Arts and Sciences,66,39–151.
Darwin C (1859) On the Origin of Species by Means of Natural
Selection, or the Preservation of Favoured Races in the Struggle
for Life. John Murray, London.
Drummond AJ, Suchard MA, Xie D, Rambaut A (2012) Bayes-
ian Phylogenetics with BEAUTI and the BEAST 1.7. Molecular
Biology and Evolution,29, 1969–1973.
Estoup A, Guillemaud T (2010) Reconstructing routes of inva-
sion using genetic data: why, how and so what? Molecular
Ecology,19, 4113–4130.
Estoup A, Jarne P, Cornuet J-M (2002) Homoplasy and mutation
model at microsatellite loci and their consequences for popu-
lation genetics analysis. Molecular Ecology,11, 1591–1604.
Estoup A, Beaumont M, Sennedot F, Moritz C, Cornuet J-M
(2004) Genetic analysis of complex demographic scenarios:
spatially expanding populations of the cane toad, Bufo mari-
nus.Evolution,58, 2021–2036.
Estoup A, Lombaert E, Marin J-M et al. (2012) Estimation of
demo-genetic model probabilities with Approximate Bayes-
ian Computation using linear discriminant analysis on sum-
mary statistics. Molecular Ecology Resources,12, 846–855.
Ettershank G (1966) A generic revision of the world Myrmici-
nae related to Solenopsis and Pheidologeton (Hymenoptera:
Formicidae). Australian Journal of Zoology,14,73–171.
©2014 John Wiley & Sons Ltd
GLOBAL INVASION OF THE TROPICAL FIRE ANT 385

Evanno G, Regnaut S, Goudet J (2005) Detecting the number of
clusters of individuals using the software STRUCTURE: a simu-
lation study. Molecular Ecology,14, 2611–2620.
Evans TA, Forschler BT, Grace JK (2013) Biology of invasive
termites: a worldwide review. Annual Review of Entomology,
58, 455–474.
Excoffier L, Smouse PE, Quattro JM (1992) Analysis of molecu-
lar variance inferred from metric distances among DNA
haplotypes - application to human mitochondrial-DNA
restriction data. Genetics,131, 479–491.
Felsenstein J (2005) PHYLIP v.3.6. Distributed by the author. Dept.
Genome Sci, Univ. Washington, Seattle.
Fisher RN (1997) Dispersal and evolution of the Pacific Basin
Gekkonid Lizards Gehyra oceanica and Gehyra mutilata.Evolu-
tion,51, 906–921.
Fitzpatrick BM, Fordyce JA, Niemiller ML, Reynolds RG (2012)
What can DNA tell us about biological invasions? Biological
Invasions,14, 245–253.
Flynn DO, Gir
aldez A (2004) Born with a “silver spoon”: the ori-
gin of world trade in 1571. Journal of World History,6, 201–221.
Fogelqvist J, Niittyvuopio A,

Agren JA, Savolainen O, Lascoux
M (2010) Cryptic population genetic structure: the number of
inferred clusters depends on sample size. Molecular Ecology
Resources,10, 314–323.
Franc
ßois O, Durand E (2010) Spatially explicit Bayesian cluster-
ing models in population genetics. Molecular Ecology
Resources,10, 773–784.
Geetha V, Ajay N, Viswananthan G, Narenda A (2000) The
effect of urbanisation on the biodiversity of ant fauna in and
around Bangalore. Journal of Ecobiology,12, 115–122.
Gerlach G, Jueterbock A, Kraemer P, Deppermann J, Harmand
P (2010) Calculations of population differentiation based on
Gst and D: forget Gst but not all of statistics!.Molecular Ecol-
ogy,19, 3845–3852.
Griffiths RC, Tavar
e S (1994) Sampling theory for neutral
alleles in a varying environment. Philosophical Transactions of
the Royal Society of London B,344, 403–410.
Guillemaud T, Beaumont MA, Ciosi M, Cornuet J-M, Estoup A
(2010) Inferring introduction routes of invasive species using
approximate Bayesian computation on microsatellite data.
Heredity,104,88–99.
Gusfield D (2002) Partition-distance: a problem and class of
perfect graphs arising in clustering. Information Processing
Letters,82, 159–164.
Hamilton G, Currat M, Ray N, Heckel G, Beaumont M, Excof-
fier L (2005) Bayesian estimation of recent migration rates
after a spatial expansion. Genetics,170, 409–417.
Hausdorf B, Hennig C (2010) Species delimitation using domi-
nant and codominant multilocus markers. Systematic Biology,
59, 491–503.
Hedrick PW (2005) A standardized genetic differentiation mea-
sure. Evolution,59, 1633–1638.
Hoffman JI, Amos W (2005) Microsatellite genotyping errors:
detection approaches, common sources and consequences for
paternal exclusion. Molecular Ecology,14, 599–612.
Holway DA, Lach L, Suarez AV, Tsutsui ND, Case TJ (2002)
The causes and consequences of ant invasions. Annual Review
of Ecology and Systematics,33, 181–233.
Huelsenbeck JP, Andolfatto P (2007) Inference of population
structure under a dirichlet process model. Genetics,175,
1787–1802.
Huelsenbeck JP, Andolfatto P, Huelsenbeck ET (2011) STRUCTU-
RAMA: Bayesian inference of population structure. Evolution-
ary Bioinformatics,7,55–59.
Hulme PE (2009) Trade, transport and trouble: managing inva-
sive species pathways in an era of globalization. Journal of
Applied Ecology,46,10–18.
Hurst GDD, Jiggins FM (2005) Problems with mitochondrial
DNA as a marker in population, phylogeographic and phy-
logenetic studies: the effects of inherited symbionts. Proceed-
ings of the Royal Society B,272, 1525–1534.
Jombart T, Ahmed I (2011) ADEGENET 1.3-1: new tools for the analy-
sis of genome-wide SNP data. Bioinformatics,27, 3070–3071.
Jombart T, Pontier D, Dufour A-B (2009) Genetic markers in the
playground of multivariate analysis. Heredity,102, 330–341.
Jombart T, Devillard S, Balloux F (2010) Discriminant analysis
of principal components: a new method for the analysis of
genetically structured populations. BMC Genetics,11, 94.
Jost L (2008) Gst and its relatives do not measure differentia-
tion. Molecular Ecology,17, 4015–4026.
Kirk H, Dorn S, Mazzi D (2013) Molecular genetics and genom-
ics generate new insights into invertebrate pest invasions.
Evolutionary Applications,6, 842–856.
Kroll JC, Arnold KA, Gotic RF (1973) An observation of preda-
tion by native fire ants on nestling barn swallows. Wilson
Bulletin,85, 478–479.
Krushelnycky PD, Loope LL, Reimer NJ (2005) The ecology,
policy and management of ants in Hawaii. Proceedings of the
Hawaiin Entomological Society,37,1–25.
Lake PS, O’Dowd DJ (1991) Red crabs in rain forest, Christmas
Island: biotic resistance to invasion by an exotic snail. Oikos,
62,25–29.
Lakshmikantha BP, Lakshminarayan NG, Ali TMM, Veeresh
GK (1996) Fire-ant damage to potato in Bangalore. Journal of
the Indian Potato Association,23,75–76.
Lanfear R, Calcott B, Ho SYW, Guindon S (2012) PARTITIONFIND-
ER: combined selection of partitioning schemes and substitu-
tion models for phylogenetic analyses. Molecular Biology and
Evolution,29, 1695–1701.
Larson G, Albarella U, Dobney K et al. (2007) Ancient DNA,
pig domestication, and the spread of the neolithic into Eur-
ope. Proceedings of the National Academy of Sciences USA,104,
15276–15281.
Lee C, Abdool A, Huang C-H (2009) PCA-based population
structure inference with generic clustering algorithms. BMC
Bioinformatics,10, S73.
Librado P, Rozas J (2009) DNASP v5: a software for comprehen-
sive analysis of DNA polymorphism data. Bioinformatics,25,
1451–1452.
Lowry E, Rollinson EJ, Laybourn AJ et al. (2013) Biological
invasions: a field synopsis, systematic review, and database
of the literature. Ecology and Evolution,3, 182–196.
Mackay WP, Porter S, Gonzalez D et al. (1990) A comparison
of monogyne and polygyne populations of the tropical fire
ant, Solenopsis geminata (Hymenoptera: Formicidae), in Mex-
ico. Journal of the Kansas Entomological Society,63, 611–615.
McGlynn TP (1999) The Worldwide transfer of ants: geographi-
cal distribution and ecological invasions. Journal of Biogeogra-
phy,26, 535–548.
Meirmans PG, Hedrick PW (2011) Assessing population struc-
ture: FST and related measures. Molecular Ecology Resources,
11,5–18.
©2014 John Wiley & Sons Ltd
386 D. GOTZEK ET AL.

Meirmans PG, Van Tienderen PH (2004) GENOTYPE and GENO-
DIVE: two programs for the analysis of genetic diversity of
asexual organisms. Molecular Ecology Notes,4, 792–794.
Meyerson LA, Mooney HA (2007) Invasive alien species in an era
of globalization. Frontiers in Ecology and Evolution,5, 199–208.
Michalakis Y, Excoffier L (1996) A generic estimation of popu-
lation subdivision using distances between alleles with
special reference for microsatellite loci. Genetics,142, 1061–
1064.
Moritz C, Dowling TE, Brown WM (1987) Evolution of animal
mitochondrial DNA: relevance for population biology and
systematics. Annual Review of Ecology Evolution and Systemat-
ics,18, 269–292.
Moulis RA (1996) Predation by the imported fire ant (Solenopsis
invicta) on loggerhead sea turtle (Caretta caretta) nests on
Wassaw National Wildlife Refuge, Georgia. Celonian Conser-
vation Biology,36, 439–472.
Nei M, Chesser R (1983) Estimation of fixation indices and
gene diversities. Annals of Human Genetics,47, 253–259.
Nei M, Tajima F, Tateno Y (1983) Accuracy of estimated phylo-
genetic trees from molecular data. Journal of Molecular Evolu-
tion,19, 153–170.
Onogi A, Nurimoto M, Morita M (2011) Characterization of a
Bayesian genetic clustering algorithm based on a Dirichlet
process prior and comparison among Bayesian clustering
methods. BMC Bioinformatics,12, 263.
Paetkau D, Slade R, Burden M, Estoup A (2004) Genetic assign-
ment methods for the direct, real-time estimation of migra-
tion rate: a simulation-based exploration of accuracy and
power. Molecular Ecology,13,55–65.
Peakall R, Smouse PE (2012) GENALEX 6.5: genetic analysis in
Excel. Population genetic software for teaching and research
–an update. Bioinformatics,28, 2537–2539.
Pella J, Masuda M (2006) The Gibbs and split-merge sampler
for population mixture analysis from genetic data with
incomplete baselines. Canadian Journal of Fisheries and Aquatic
Sciences,63, 576–596.
Perfecto I (1991) Dynamics of Solenopsis geminata in a tropical
fallow field after ploughing. Oikos,62, 139–144.
Perfecto I, Vandermeer J (2011) Discovery dominance tradeoff:
the case of Pheidole subarmata and Solenopsis geminata (Hyme-
noptera: Formicidae) in neotropical pastures. Environmental
Entomology,40, 999–1006.
Pimentel D, Lach L, Zuniga R, Morrison D (2000) Environmen-
tal and economic costs of nonindigenous species in the Uni-
ted States. BioScience,50, 53.
Pimentel D, Zuniga R, Morrison D (2005) Update on the
environmental and economic costs associated with alien-
invasive species in the United States. Ecological Economics,52,
273–288.
Piry S, Alapetite A, Cornuet J-M, Paetkau D, Baudouin L,
Estoup A (2004) GENECLASS2: a software for genetic assign-
ment and first-generation migrant detection. Journal of Hered-
ity,95, 536–539.
Pitts JP, McHugh J, Ross KG (2005) Cladistic analysis of the fire
ants of the Solenopsis saevissima species-group (Hymenoptera:
Formicidae). Zoologica Scripta,34, 493–505.
Plentovich S, Hebshi A, Conant S (2009) Detrimental effects of
two widespread invasive ant species on weight and survival
of colonial nesting seabirds in the Hawaiin Islands. Biological
Invasions,11, 289–298.
Py
sek P, Richardson DM (2010) Invasive species, environmen-
tal change and management, and health. Annual Review of
Environment and Resources,35,25–55.
R Development Core Team (2013) R: A Language and Environ-
ment for Statistical Computing. R Foundation for Statistical
Computing, Vienna, Austria.
Rajabi-Maham H, Orth A, Bonhomme F (2008) Phylogeography
and postglacial expansion of Mus musculus domesticus
inferred from mitochondrial DNA coalescent, from Iran to
Europe. Molecular Ecology,17, 627–641.
Rambaut A, Drummond AJ (2007) TRACER v1.4, Available from
http://beast.bio.ed.ac.uk/Tracer
Rannala B, Mountain J (1997) Detecting immigration by using
multilocus genotypes. Proceedings of the National Academy of
Sciences USA,94, 9197–9201.
Rice WR (1989) Analyzing tables of statistical tests. Evolution,
43, 223–225.
Risch SJ, Carroll CR (1982) Effect of a keystone predaceous ant,
Solenopsis geminata, on arthropods in a tropical agroecosys-
tem. Ecology,63, 1979–1983.
Ross KG, Vargo EL, Keller L (1996) Social evolution in a new
environment: the case of introduced fire ants. Proceedings of
the National Academy of Sciences USA,93, 3021–3025.
Ross KG, Krieger MJ, Shoemaker DD (2003) Alternative genetic
foundations for a key social polymorphism in fire ants.
Genetics,165, 1853–1867.
Sacks BN, Brown SK, Stephens D, Pedersen NC, Wu J-T,
Berry O (2013) Y chromosome analysis of dingoes and
Southeast Asian Village dogs suggests a neolithic continen-
tal expansion from Southeast Asia followed by multiple
Austronesian dispersals. Molecular Biology and Evolution,30,
1103–1118.
Saitou N, Nei M (1987) The Neighbor-Joining Method: a new
method for reconstructing phylogenetic trees. Molecular Biol-
ogy and Evolution,4, 406–425.
Sax DF, Stachowicz JJ, Brown JH et al. (2007) Ecological and
evolutionary insights from species invasions. Trends in Ecol-
ogy and Evolution,22, 465–471.
Schwarz GE (1978) Estimating the dimension of a model.
Annals of Statistics,6, 461–464.
Shringarpure S, Won D, Xing EP (2011) STRUCTHDP: automatic
inference of number of clusters and population structure
from admixed genotype data. Bioinformatics,27, i324–i332.
Simberloff D, Martin J-L, Genovesi P et al. (2013) Impacts of
biological invasions: what’s what and the way forward.
Trends in Ecology and Evolution,28,58–66.
Spate OHK (2004) The Spanish Lake. ANU E Press, Canberra.
Suarez AV, Tsutsui ND (2008) The evolutionary consequences
of biological invasions. Molecular Ecology,17, 351–360.
Suarez AV, Holway DA, Case TJ (2001) Patterns of spread in
biological invasions dominated by long-distance jump dis-
persal: insights from Argentine ants. Proceedings of the
National Academy of Sciences USA,98, 1095–1100.
Takezaki N, Nei M (1996) Genetic distances and reconstruction
of phylogenetic trees from microsatellite DNA. Genetics,144,
389–390.
Takezaki N, Nei M (2008) Empirical tests of the reliability of
phylogenetic trees constructed with microsatellite DNA.
Genetics,178, 385–392.
Tamura K, Peterson D, Peterson N, Stecher G, Nei M, Kumar S
(2011) MEGA5: molecular evolutionary genetics analysis using
©2014 John Wiley & Sons Ltd
GLOBAL INVASION OF THE TROPICAL FIRE ANT 387

maximum likelihood, evolutionary distance, and maximum
parsimony methods. Molecular Biology and Evolution,28,
2731–2739.
Taylor DR, Keller SR (2007) Historical range expansion deter-
mines the phylogenetic diversity introduced during contem-
porary species invasion. Evolution,61, 334–345.
Teh YW, Jordan MI, Beal MJ, Blei DM (2006) Hierarchical Di-
richlet processes. Journal of the American Statistical Association,
101, 1566–1581.
Trager JC (1991) A revision of the fire ants, Solenopsis geminata
group (Hymenoptera: Formicidae: Myrmicinae). Journal of the
New York Entomological Society,99, 141–198.
Travis BV (1938) The fire ant (Solenopsis spp.) as a pest of
quail. Journal of Economic Entomology,31, 649–652.
Travis BV (1941) Notes on the biology of the fire ant Solenopsis
geminata (F.) in Florida and Georgia. Florida Entomologist,24,
15–22.
Tschinkel WR (2006) The Fire Ants. Belknap Press, Cambridge.
Tsutsui ND, Suarez AV, Holway DA, Case TJ (2001) Relation-
ships among native and introduced populations of the
Argentine ant (Linepithema humile) and the source of intro-
duced populations. Molecular Ecology,10, 2151–2161.
Vitousek PM, D’Antonio CMD, Loope LL, Rejm
anek M, West-
brooks R (1997) Introduced species: a significant component
of human-caused global change. New Zealand Journal of Ecol-
ogy,21,1–16.
Wakeley J (2008) Coalescent Theory: An Introduction. Roberts &
Company Publishers, Greenwood Village, Colorado.
Way M, Heong K (2009) Significance of the tropical fire ant
Solenopsis geminata (Hymenoptera: Formicidae) as part of the
natural enemy complex responsible for successful biological
control of many tropical irrigated rice pests. Bulletin of Ento-
mological Research,99, 503–512.
Way MJ, Islam Z, Heong KL, Joshi RC (1998) Ants in tropical
irrigated rice: distribution and abundance, especially of
Solenopsis geminata (Hymenoptera: Formicidae). Bulletin of
Entomological Research,88, 467–476.
Way MJ, Javier G, Heong KL (2002) The role of ants, especially
the fire ant, Solenopsis geminata (Hymenoptera: Formicidae),
in the biological control of tropical upland rice pests. Bulletin
of Entomological Research,92, 431–437.
Westphal MI, Browne M, MacKinnon K, Noble I (2007) The
link between international trade and the global distribution
of invasive alien species. Biological Invasions,10, 391–398.
Wetterer JK (2005) Worldwide distribution and potential
spread of the long-legged ant, Anoplolepis gracilipes (Hyme-
noptera: Formicidae). Sociobiology,45,77–97.
Wetterer JK (2008) Worldwide spread of the longhorn crazy
ant, Paratrechina longicornis (Hymenoptera: Formicidae). Myr-
mecological News,11, 137–149.
Wetterer JK (2010) Worldwide spread of the Pharaoh ant, Mo-
nomorium pharaonis (Hymenoptera: Formicidae). Myrmecologi-
cal News,13, 115–129.
Wetterer JK (2011) Worldwide spread of the tropical fire ant,
Solenopsis geminata (Hymenoptera: Formicidae). Myrmecologi-
cal News,14,21–35.
Wetterer JK (2012) Worldwide spread of the african big-headed
ant, Pheidole megacephala (Hymenoptera: Formicidae). Myrme-
cological News,17,51–62.
Wetterer JK, Wild AL, Suarez AV, Roura-Pascual N, Espadaler
X (2009) Worldwide spread of the Argentine ant, Linepithema
humile (Hymenoptera: Formicidae). Myrmecological News,12,
187–194.
Wilson EO (2005) Early ant plagues in the New World. Nature,
433, 32.
Wolcott GN (1933) Recent experiments in the control of two
Puerto Rican ants. Journal of the Department of Agriculture
Puerto Rico,17, 223–239.
D.G. and H.J.A. conceived the study; D.G., H.J.A. and
D.D.S. performed research; and D.G., H.J.A., A.V.S.,
S.H.C. and D.D.S. wrote the manuscript.
Data accessibility
Sampling information can be found in Table S1, Sup-
porting information. All data, scripts and input files are
deposited in DataDryad (doi:10.5061/dryad.256kh) and
DNA sequences are additionally deposited in GenBank
(Accession nos: KP145683 –KP145854).
Supporting information
Additional supporting information may be found in the online ver-
sion of this article.
Fig. S1. A) Number of clusters inferred for the complete data-
set. Left: BIC scores for K=1-15 using K-means clustering.
Right: Posterior probabilities using the DPP implemented in
Structurama. B) Number of clusters for inferred for Old World
samples only.
Fig. S2. Stability of group membership probabilities for maxi-
mal clustering (K=8) of the complete dataset.
Fig. S3. (A) Three scenarios compared in ABC analyses A –C
to identify the source population. (B) Posterior probabilities of
competing scenarios (see SI Figure S3A) of ABC.
Fig. S4. (A) Scenarios and posterior probabilities (with 95% CI)
used in analyses D based on logistic regression estimates. (B)
Prior and posterior density distributions of demographic
parameters for scenario 3 (serial invasion from unsampled
ghost population) of analysis D.
Table S1. List of samples, cluster assignment, collection local-
ity and coordinates, and data type.
Table S2. Estimates of genetic diversity (and standard error
for microsatellites and standard deviation for mtDNA in
brackets) within clusters for nuclear and cytoplasmic
genomes.
Table S3. Degree of genetic differentiation between clusters
(F
ST
, Jost’s D
est
, Hedrick’s G”
ST
).
Table S4. Assignment and Exclusion tests of the invasive Old
World samples to the New World clusters.
©2014 John Wiley & Sons Ltd
388 D. GOTZEK ET AL.