UC Davis
UC Davis Previously Published Works
Title
The evolution of myrmicine ants: Phylogeny and biogeography of a hyperdiverse ant
clade (Hymenoptera: Formicidae)
Permalink
https://escholarship.org/uc/item/2tc8r8w8
Journal
Systematic Entomology, 40(1)
ISSN
0307-6970
Authors
Ward, PS
Brady, SG
Fisher, BL
et al.
Publication Date
2015
DOI
10.1111/syen.12090
Peer reviewed
eScholarship.org Powered by the California Digital Library
University of California
Systematic Entomology (2015), 40, 61–81 DOI: 10.1111/syen.12090
The evolution of myrmicine ants: phylogeny
and biogeography of a hyperdiverse ant clade
(Hymenoptera: Formicidae)
PHILIP S. WARD
1
, SEÁN G. BRADY
2
, BRIAN L. FISHER
3
andTED R. SCHULTZ
2
1
Department of Entomology and Nematology, University of California, Davis, CA, U.S.A.,
2
Department of Entomology, National
Museum of Natural History, Smithsonian Institution, Washington, DC, U.S.A. and
3
Department of Entomology, California Academy
of Sciences, San Francisco, CA, U.S.A.
Abstract. This study investigates the evolutionary history of a hyperdiverse clade, the
ant subfamily Myrmicinae (Hymenoptera: Formicidae), based on analyses of a data
matrix comprising 251 species and 11 nuclear gene fragments. Under both maximum
likelihood and Bayesian methods of inference, we recover a robust phylogeny that
reveals six major clades of Myrmicinae, here treated as newly dened tribes and occur-
ring as a pectinate series: Myrmicini, Pogonomyrmecini trib.n., Stenammini, Solenop-
sidini, Attini and Crematogastrini. Because we condense the former 25 myrmicine tribes
into a new six-tribe scheme, membership in some tribes is now notably different, espe-
cially regarding Attini. We demonstrate that the monotypic genus Ankylomyrma is nei-
ther in the Myrmicinae nor even a member of the more inclusive formicoid clade – rather
it is a poneroid ant, sister to the genus Tatuidris (Agroecomyrmecinae). Several
species-rich myrmicine genera are shown to be nonmonophyletic, including Pogono-
myrmex, Aphaenogaster, Messor, Monomorium, Pheidole, Temnothorax and Tetramor-
ium. We propose a number of generic synonymies to partially alleviate these problems
(senior synonym listed rst): Pheidole = Anisopheidole syn.n. = Machomyrma syn.n.;
Temnothorax = Chalepoxenus syn.n. = Myrmoxenus syn.n. = Protomognathus syn.n.;
Tetramorium = Rhoptromyrmex syn.n. = Anergates syn.n. = Teleutomyrmex syn.n. The
genus Veromessor stat.r. is resurrected for the New World species previously placed
in Messor; Syllophopsis stat.r. is resurrected from synonymy under Monomorium to
contain the species in the hildebrandti group; Trichomyrmex stat.r. is resurrected from
synonymy under Monomorium to contain the species in the scabriceps- and destruc-
tor-groups; and the monotypic genus Epelysidris stat.r. is reinstated for Monomorium
brocha. Bayesian divergence dating indicates that the crown group Myrmicinae origi-
nated about 98.6 Ma (95% highest probability density 87.9–109.6 Ma) but the six major
clades are considerably younger, with age estimates ranging from 52.3 to 71.1 Ma.
Although these and other suprageneric taxa arose mostly in the middle Eocene or earlier,
a number of prominent, species-rich genera, such as Pheidole, Cephalotes,
Strumigenys,
Crematogaster and Tetramorium, have estimated crown group origins in the late Eocene
or Oligocene. Most myrmicine species diversity resides in the two sister clades, Attini
and Crematogastrini, which are estimated to have originated and diversied extensively
in the Neotropics and Paleotropics, respectively. The newly circumscribed Myrmicini is
Holarctic in distribution, and ancestral range estimation suggests a Nearctic origin. The
Pogonomyrmecini and Solenopsidini are reconstructed as being Neotropical in origin,
Correspondence: Philip S. Ward, Department of Entomology and Nematology, University of California, Davis, CA 95616, U.S.A. E-mail:
psward@ucdavis.edu
© 2014 The Royal Entomological Society 61
62 P. S . Wa rd et al.
but they have subsequently colonized the Nearctic region (Pogonomyrmecini) and many
parts of the Old World as well as the Nearctic region (Solenopsidini), respectively. The
Stenammini have ourished primarily in the northern hemisphere, and are most likely
of Nearctic origin, but selected lineages have dispersed to the northern Neotropics and
the Paleotropics. Thus the evolutionary history of the Myrmicinae has played out on a
global stage over the last 100 Ma, with no single region being the principal generator of
species diversity.
This published work has been registered in ZooBank, http://zoobank.org/urn:lsid:
zoobank.org:pub: BB6829C4-DA79-45FE-979E-9749E237590E.
Introduction
Ants are one of evolution’s great success stories, the most
species-rich and ecologically dominant of all eusocial insects
(Hölldobler & Wilson, 1990; Lach et al., 2009), yet this success
is far from evenly distributed among the major lineages of the
group. There are 16 extant subfamilies of ants (Bolton, 2014;
Brady et al., 2014), but four of these – Dolichoderinae, Formic-
inae, Myrmicinae and Ponerinae – account for almost 90% of
all known species (Bolton, 2014). Among these ‘big four’ the
most biologically diverse and prolic is the subfamily Myrmic-
inae. With approximately 6475 described species – about half
of all named ants – and many others awaiting discovery and
description, the myrmicines are a hyperdiverse clade inhabiting
most of the terrestrial surface of the Earth and encompassing a
wide range of lifestyles including generalist and specialist preda-
tors, scavengers, omnivores, granivores and herbivores (Kugler,
1979; Brown, 2000).
Myrmicine ants have been subjects of intense investigation
into colony organization, communication, demography and
ecology (Davidson, 1977; Brian, 1983; Tschinkel, 2006). They
contain a diverse array of socially parasitic species (Beibl
et al., 2005; Buschinger, 2009) and they furnish numerous
examples of symbioses with other organisms (Davidson &
McKey, 1993; Fiedler, 2006; Russell et al., 2009), including
the iconic mutualism between fungus-growing ants and their
cultivars (Mueller et al., 2001; Hölldobler & Wilson, 2011).
Genomic studies have begun to probe the genetic architecture
of prominent myrmicines in the genera Solenopsis (Wurm
et al., 2011), Pogonomyrmex (Smith et al., 2011), Acromyrmex
(Nygaard et al., 2012) and Atta (Suen et al., 2011).
Despite the attention that they have garnered, myrmicines
present a number of taxonomic challenges. There are many dif-
cult species complexes (Seifert, 2003; Schlick-Steiner et al.,
2006; Blaimer, 2012a), as well as multiple genera and tribes
whose monophyly is doubtful (Brady et al., 2006; Ward, 2011).
The members of this subfamily have received only modest cov-
erage in molecular phylogenetic studies of ants as a whole, and
results have been inconsistent between studies (Brady et al.,
2006; Moreau et al., 2006). Here we aim to address this prob-
lem by employing 11 nuclear genes and comprehensive taxon
sampling across the entire subfamily. We combine phylogenetic
inference with divergence dating and biogeographic analysis to
generate a robust picture of the evolutionary history of these
remarkable ants. These results also provide the basis for a
revised higher classication of the subfamily.
Material and methods
Taxon selection
We assembled a set of 234 species representing all 25 tribes of
Myrmicinae and 128 of the 145 extant genera (taxonomy follows
Bolton, 2014) (Table S1). One species could not be identied to
genus. For genera with widespread distributions we attempted
to sequence a species from each of the major biogeographic
regions in which the genus occurs, to facilitate ancestral range
estimation. We also sampled more intensively within several
genera whose monophyly appeared to be questionable, includ-
ing Aphaenogaster, Monomorium, Pheidole, Solenopsis, Stru-
migenys and Tetramorium. We included 17 outgroup taxa; these
were drawn mostly from the other subfamilies of ants, but we
also chose an aculeate hymenopteran wasp, Apterogyna (Bra-
dynobaenidae), as the most distant outgroup to root the tree.
Sequencing and sequence annotation
We sequenced fragments of 11 nuclear genes, including
two ribosomal genes, 18S rDNA and
28S rDNA, and nine
protein-coding genes: abdominal-A (Abd-A), elongation factor
1-alpha F1 copy (EF1aF1), elongation factor 1-alpha F2 copy
(EF1aF2), long wavelength rhodopsin (LW Rh), arginine kinase
(ArgK), topoisomerase 1 (Top1), ultrabithorax (Ubx), wingless
(Wg) and rudimentary (CAD). Protocols for extraction, ampli-
cation and sequencing follow Ward & Downie (2005), Brady
et al. (2006), Ward et al. (2010), and Ward & Sumnicht (2012).
Sequences were collated in Sequencher v4.10.1 (Gene
Codes Corporation, Ann Arbor, MI, U.S.A.), aligned with
Clustal X v1.81 (Thompson et al., 1997), and manually edited
and concatenated with MacClade v4.08 (Maddison & Mad-
dison, 2005). Alignment was straightforward for exons of
protein-coding genes and for most 18S rDNA sequence. We
excluded ambiguously aligned regions of 18S and 28S, introns
of the protein-coding genes, and autapomorphic indels in exons
© 2014 The Royal Entomological Society, Systematic Entomology, 40, 61–81
Phylogeny and evolution of myrmicine ants 63
Tab le 1 . Sequence characteristics of the 251-taxon data matrix (233
ingroup taxa).
Gene No. of sites VS PS VS, ingroup PS, ingroup
Abd-A 600 262 221 246 206
EF1aF2 517 206 193 204 190
LW Rh 458 279 250 269 239
ArgK 673 332 297 309 285
Top1 880 477 424 464 419
Ubx 627 301 254 286 232
Wg 409 251 218 224 197
EF1aF1 359 141 129 138 128
18S 1850 224 106 184 85
28S 1653 480 271 425 223
CAD 961 605 532 592 519
All genes 8987 3558 2895 3341 2723
VS = variable sites; PS = parsimony-informative sites.
of Abd-A, ArgK, CAD and Wg. The resulting matrix that was
used for all subsequent analyses contains 8987 bp (3558 variable
sites, 2895 parsimony-informative sites). This matrix (251 taxa
by 11 genes) contains no missing gene fragments. Of the 2761
fragments, 499 were previously published (Ward & Downie,
2005; Brady et al., 2006; Ward, 2007a; Branstetter, 2009;
Lucky & Sarnat, 2010; Ward et al., 2010; Brady et al., 2014);
the remaining 2262 sequences were generated for this study
(GenBank accession numbers KJ859686-KJ861947). Sequence
characteristics of each gene are summarized in Table 1. The
aligned data matrix has been deposited in TreeBase (study
accession 15764).
Phylogenetic analyses
Data partitions
We employed an iterative strategy to arrive at the partitioning
scheme for analysing the 11-gene, concatenated dataset. In a
rst round of analyses we partitioned the data into 25 partitions
based on the variability of codon-position sites within each gene.
Four genes (Abd-A, EF1aF1, EF1aF2, Ubx) were each divided
into two partitions consisting of (i) codon positions 1 and 2 and
(ii) codon position 3, resulting in eight partitions; ve genes
(LW Rh, ArgK, Top1, Wg, CAD) were assigned site-specic
models in which each codon position formed a separate parti-
tion, resulting in 15 partitions; and two nonprotein-coding genes
(18S, 28S) were each assigned a single partition, resulting in two
partitions. The choice of nucleotide substitution model for each
partition (Table S2) was determined using the Akaike Informa-
tion Criterion (AIC) (Posada & Buckley, 2004) as implemented
in JModelTest v2.0.2 (Posada, 2008). The tree resulting from
a Bayesian analysis (MrBayes v3.2.1; Ronquist et al., 2012)
of these data employing these partitions and models was then
submitted to PartitionFinder v1.0.1 (Lanfear et al., 2012) along
with 29 predened blocks, encompassing the rst, second and
third codon positions for each of the nine protein-coding genes
and the two nonprotein-coding genes (18S, 28S). Under the
Bayesian Information Criterion (BIC), the ‘greedy’ algorithm,
and ‘models = all,’ PartitionFinder identied 21 partitions
Tab le 2 . The 21 partitions and models identied by PartitionFinder
and used in the maximum likelihood and Bayesian analyses of the
concatenated, unmodied 8987-bp dataset.
Partition Blocks Model
p1 Abd-A pos1, Ubx pos1 SYM+I+G
p2 Abd-A pos2, Ubx pos2 TVM+I+G
p3 Abd-A pos3 TIM+I+G
p4 EF1aF2 pos3, Top1 pos3 TrNef+I+G
p5 EF1aF2 pos1 GTR+I+G
p6 ArgK pos2, EF1aF2 pos2 TIMef+I+G
p7 LW Rh pos3, Ubx pos3 TrNef+I+G
p8 LW Rh pos1 TVM+I+G
p9 LW Rh pos2 GTR+I+G
p10 ArgK pos3 TrNef+I+G
p11 ArgK pos1 SYM+I+G
p12 Top1 pos1 SYM+I+G
p13 CAD pos2, Top1 pos2 GTR+I+G
p14 Wg pos3 GTR+I+G
p15 Wg pos1, Wg pos2 TrNef+I+G
p16 EF1aF1 pos2 JC
p17 EF1aF1 pos3 TIM+G
p18 18S, EF1aF1 pos1 TrNef+I+G
p19 28S GTR+I+G
p20 CAD pos3 GTR+I+G
p21 CAD pos1 K80+I+G
(Table 2), which were then employed in intensive Bayesian
and maximum-likelihood analyses. In cases where the model
specied by PartitionFinder was not implemented in MrBayes,
the next-most-complex available model was employed.
The choice of partitions for separate, single-gene Bayesian
analyses of each of the nuclear genes followed an iterative
scheme similar to that described above for the full concatenated
dataset. An initial Bayesian analysis was conducted employing
partitions based on variability of codon sites. The resulting
tree was submitted to PartitionFinder using ‘models = mrbayes,’
‘model-Selection = BIC,’ and ‘search = all.’ If PartitionFinder
identied a partitioning scheme different from the one employed
in the rst analysis, a second Bayesian analysis was conducted
employing the PartitionFinder scheme (Table S3).
Treatments addressing base frequency heterogeneity
For each of the three codon positions within each protein-coding
gene, as well as for 18S and 28S, each treated as single
partitions, we evaluated the homogeneity of base frequencies
across taxa using PAUP* 4.0a128 (Swofford, 2002), which
indicated that nine partitions – the third positions of each
protein-coding gene – contained signicantly heterogeneous
base frequencies (Table S4). In order to examine the effects
of base frequency heterogeneity on phylogenetic results, we
subjected each dataset to a number of treatments, including:
(i) ‘partial RY coding,’ in which, for each gene, the subset of
taxa (differing for each gene) with heterogeneous third positions
deviating signicantly from the observed averages were coded
as RY and in which the remainder of third positions were coded
normally (ACTG); (ii) ‘complete RY coding’ in which all third
positions were coded as RY; (iii) exclusion of a minimum subset
© 2014 The Royal Entomological Society, Systematic Entomology, 40, 61–81
64 P. S . Wa rd et al.
of 324 characters (not necessarily third positions) deviating
signicantly from observed base-frequency averages based on
the results of a likelihood-based method for ltering potential
nonstationary sites, to be described elsewhere (D. Swofford,
personal communication); and (iv) exclusion of 843 characters
(not necessarily third positions), the 10% deviating the most
from observed base-frequency averages, also based on the same
ltering method. All analyses were carried out in MrBayes
v3.2.1 and v3.2.2 (Ronquist et al., 2012). For treatment i, ‘partial
RY,’ decisions about whether to code a particular third-position
site as ACTG or RY were made based on analyses of output
from PAUP* 4.0a128 (Swofford, 2002) (Table S4). Characters
were recoded as RY in Mesquite v2.73 (Maddison & Maddison,
2010). Partitioning schemes for each of these treatments resulted
from the same iterative strategy described in the Data partitions
section above.
Treatments addressing possible ‘wildcard’ taxa
In order to explore the reasons for poor support in one region of
the tree (Crematogastrini), particularly with respect to a number
of unusually long-branched taxa in that region, we conducted
analyses in which subsets of 11 taxa suspected of behaving as
‘wildcards’ were excluded (Nixon & Wheeler, 1992; Kearney,
2002). Unlike the most common type of wildcard taxa, how-
ever, none of the taxa in our analyses contained missing data.
For this purpose we created and analysed two datasets consist-
ing of: (i) 248 taxa, excluding Acanthomyrmex ferox, Mayriella
ebbei and Xenomyrmex oridanus; and (ii) 243 taxa, exclud-
ing Cardiocondyla mauritanica, Cardiocondyla MY01, Cardio-
condyla thoracica, Liomyrmex gestroi, Melissotarsus insularis,
Metapone madagascarica, Metapone PG01 and Ocymyrmex cf.
fortior.
Single-gene analyses
In order to assess the possibility of conicting gene trees and
to compare gene-specic support for clades, we conducted
iterative analyses of each of the 11 genes as described above,
using partitions and models identied by PartitionFinder v1.0.1
(Lanfear et al., 2012), summarized in Table S3. All single-gene
analyses were conducted under Bayesian criteria as described
below and consisted of 10 million generations with a burn-in of
1 million generations.
Constraint analyses
One surprising result of several prior studies (Brady et al., 2006;
Moreau et al., 2006; Rabeling et al., 2008), that the species
Tatuidris tatusia is a poneroid and as such only very distantly
related to the Myrmicinae, has failed to signicantly inuence
the posterior probabilities of some myrmecologists (e.g. Baroni
Urbani & De Andrade, 2007; Keller, 2011). For this reason,
we conducted a constraint analysis to explicitly quantify the
difference in likelihoods between the unconstrained outcome,
in which (Tatuidris tatusia + Ankylomyrma coronacantha)are
poneroids, and the outcome in which they are constrained to be
members of the Myrmicinae. The latter analysis was carried out
in GARLI v2.0 as described below, using 100 search repetitions
and a minimally dened topological constraint. We included
Ankylomyrma in this constraint because of a novel result (see
below) placing it close to Tatuidris.
Bayesian analyses
We conducted Bayesian analyses using MrBayes v3.2.1 and
3.2.2 (Ronquist et al., 2012) with nucmodel = 4by4, nruns = 2,
nchains = 8, samplefreq
= 1000, and either 20 million or 40 mil-
lion generations and burn-ins of 2 and 4 million generations,
respectively; single-gene analyses consisted of 10 million gen-
erations and burn-ins of 1 million generations. For partitioned
analyses all parameters, including branch-length rate multipli-
ers, were unlinked across partitions; the only exceptions were
branch lengths and topology, which were linked. All analy-
ses were carried out using parallel processing (one chain per
CPU) on the Smithsonian NMNH LAB Topaz computing clus-
ter (Apple computers with Intel processors). To address known
problems with branch-length estimation in MrBayes (Marshall
et al., 2006; Brown et al., 2010; Marshall, 2010; Ward et al.,
2010), we set brlenspr = unconstrained:Exp(100). Burn-in, con-
vergence and stationarity were assessed using Tracer v1.5 (Ram-
baut & Drummond, 2009), by examining PSRF values and .stat
output les in MrBayes, and by using Bayes factor compar-
isons of harmonic-mean marginal likelihoods of pairs of runs
with standard error estimated using 1000 bootstrap pseudorepli-
cates in Tracer 1.5 (Rambaut & Drummond, 2009), which
employs the weighted likelihood bootstrap estimator of New-
ton & Raftery (1994) as modied by Suchard et al. (2001). For
the full concatenated dataset, stationarity was regularly obtained
within the rst few million generations, even for the various
base-frequency-heterogeneity treatments; single-gene analyses
achieved stationarity within the rst few hundred thousand gen-
erations. The results reported here are based on the combined
post-burn-in data from both runs.
Maximum likelihood
ML analyses were carried out in GARLI v2.0 (Zwickl, 2006)
using parallel processing on the Smithsonian Hydra supercom-
puter (Linux-based with AMD processors). Unconstrained and
constrained ML best-tree analyses consisted of 200 and 100
replicate searches, respectively, and deviated from the default
settings as follows: topoweight = 0.01; brlenweight = 0.002.
ML bootstrap analyses consisted of 1000 pseudoreplicates
and deviated from default settings as follows: genthreshfor-
topoterm = 5000; scorethreshforterm = 0.10; startoptprec = 0.5;
minoptprec = 0.01; numberofprecreductions = 1; treerejection-
threshold = 20.0; topoweight = 0.01; brlenweight = 0.002. In all
analyses the value for modweight was calculated as 0.0005 ×
(#subsets + 1) (D. Zwickl, personal communication).
Divergence dating
We inferred divergence dates with v1.7.5 (Drum-
mond et al., 2012) under a parallel conguration using BEA-
GLE v1.0 on the aforementioned NMNH LAB Topaz cluster.
© 2014 The Royal Entomological Society, Systematic Entomology, 40, 61–81
