RESEARCH ARTICLE Open Access
Behavioral mechanisms and morphological
symptoms of zombie ants dying from fungal
infection
David P Hughes1,2*, Sandra B Andersen2, Nigel L Hywel-Jones3, Winanda Himaman4, Johan Billen5 and
Jacobus J Boomsma2
Abstract
Background: Parasites that manipulate host behavior can provide prominent examples of extended phenotypes:
parasite genomes controlling host behavior. Here we focus on one of the most dramatic examples of behavioral
manipulation, the death grip of ants infected by Ophiocordyceps fungi. We studied the interaction between O.
unilateralis s.l. and its host ant Camponotus leonardi in a Thai rainforest, where infected ants descend from their
canopy nests down to understory vegetation to bite into abaxial leaf veins before dying. Host mortality is
concentrated in patches (graveyards) where ants die on sapling leaves ca. 25 cm above the soil surface where
conditions for parasite development are optimal. Here we address whether the sequence of ant behaviors leading
to the final death grip can also be interpreted as parasite adaptations and describe some of the morphological
changes inside the heads of infected workers that mediate the expression of the death grip phenotype.
Results: We found that infected ants behave as zombies and display predictable stereotypical behaviors of random
rather than directional walking, and of repeated convulsions that make them fall down and thus precludes
returning to the canopy. Transitions from erratic wandering to death grips on a leaf vein were abrupt and
synchronized around solar noon. We show that the mandibles of ants penetrate deeply into vein tissue and that
this is accompanied by extensive atrophy of the mandibular muscles. This lock-jaw means the ant will remain
attached to the leaf after death. We further present histological data to show that a high density of single celled
stages of the parasite within the head capsule of dying ants are likely to be responsible for this muscular atrophy.
Conclusions: Extended phenotypes in ants induced by fungal infections are a complex example of behavioral
manipulation requiring coordinated changes of host behavior and morphology. Future work should address the
genetic basis of such extended phenotypes.
Keywords: extended phenotype behavioral manipulation, ants, fungi, convergent evolution, parasites
Background
Some parasites can adaptively take over and completely
control the behavior of their hosts leading to positive
fitness returns for parasite genes [1-4]. Host behavior is
an extended phenotype of the parasite [5]. The degree
of behavioral manipulation varies greatly across parasites
from very slight alterations of pre-existing behaviors [6]
to the expression of wholly novel behaviors that are
never seen in healthy hosts [7]. Extended phenotypes
have gained considerable prominence in community-
[8], evolutionary- [9] and behavioral ecology [10].
Early studies of extended phenotypes focused on detail-
ing behavioral changes and inferring whether they repre-
sent adaptations for parasites or should rather be
interpreted as adaptive defense mechanisms of the host or
as by-products of infection [11-13]. Recently, more inte-
grative approaches have emerged which includes a greater
focus on the mechanisms by which behavioral changes
occur. An important component is a fuller understanding
of the biology of particular study systems and the timing
* Correspondence: dhughes@psu.edu
1Departments of Entomology and Biology, Penn State University, PA 16802,
USA
Full list of author information is available at the end of the article
Hughes et al. BMC Ecology 2011, 11:13
http://www.biomedcentral.com/1472-6785/11/13
© 2011 Hughes et al; licensee BioMed Central Ltd. This is an Open Access article distributed under the terms of the Creative Commons
Attribution License (http://creativecommons.org/licenses/by/2.0), which permits unrestricted use, distribution, and reproduction in
any medium, provided the original work is properly cited.
of observation or experimentation, since parasite induced
behavioral changes are highly dynamic [14].
Here we focus on a study system that is a dramatic
example of adaptive manipulation of animal behavior by
a parasite. Worker ants infected by fungal parasites
belonging to the genus Ophiocordyceps express death
grip behavior shortly before dying for no apparent other
purpose than to assist parasite reproduction [15,16].
Worker ants are infected during foraging by spores that
attach to the cuticle. The fungus is an obligate, directly
transmitted parasite that requires ants for reproduction.
Germination and subsequent penetration of the cuticle
lead to rapidly progressing infections inside the host
body [17,18], but fungal reproduction is only possible
after the growth of a large stalk from the back of the
ant’s head followed by a propulsive release of spores
from this fruiting body [16]. The fungus inevitably kills
the ant and must do this outside the colony because
ants quickly remove dead nest-mates [19], so that dying
in the nest would not allow sufficient time for stalk
development and spore release [20]. This host death as
a developmental necessity implies that Ophiocordyceps
infections would also match the functional definition of
being a parasitoid [21].
While the Ophiocordyceps clade has a global distribu-
tion local interactions tend to be highly specific with
highly stereotyped host behaviors [16,22]. Ants of the
tribe Camponotinii (Camponotus, Polyrhachis and Echi-
nopla) are known to leave their nest to bite into leaves
before dying from infections with a representative of the
species complex Ophiocordyceps unilateralis sensu lato
[16]. (See taxonomic note in Methods). A recent study in
a Thai rainforest showed that leaf biting behavior by
infected workers of this ant species was adaptive for the
fungus because it secures a stable microclimatic niche for
the post mortem development of the stalk and the subse-
quent release of spores [20]. In this intensively studied
population infected worker ants leave their colony in the
dry, hot canopy and descend to the humid understory
where they appear to actively select leaves of saplings ca.
25 +/- 3 cm above the soil surface [20]. These parasite
manipulated ants always bite into abaxial leaf veins and
not the laminar blade, edge or upper surface (adaxial).
They also predominantly die in areas where the cadavers
of previously manipulated ants are already abundant
leading to graveyards where local densities of ants killed
by the fungus may exceed 25/m2 [23]. Graveyards of
Ophiocordyceps infected ants have also been reported
from other continents [24,25]. The healthy ants, though
nesting in the high canopy, do periodically walk in the
understory leading to new infections [23].
Behavioral manipulation of worker ants by these fungi
creates zombie ants [2,4]. Once infected ants exit their
colony to die they have no further fitness gains through
their own actions (being sterile workers they only have
inclusive fitness via helping nest-mates). In fact, by
dying within the foraging area of their own colony their
behavior may still reduce inclusive fitness [26-28]. The
term zombie ants underlines that, while the manipulated
individual may look like an ant, it represents a fungal
genome expressing fungal behavior through the body of
an ant.
Using the explicit ‘parasite’s eye view’ framework out-
lined above, we set out to test two hypotheses. First, we
hypothesized that pre-biting behavior may have an
important function to help positioning dying ants in
death biting habitats that would be optimal for subse-
quent fungal reproduction. Second, we hypothesized
that the death grip requires changes in the mandibular
muscles to transform functional mandibles into death
grip lock-jaws to secure that dead ants become perma-
nently fixed to leaves against the force of gravity.
Results
Pre-biting behavior
The ant species that is the primary host of Ophiocordy-
ceps unilateralis s.l. at our field site is Camponotus leo-
nardi [23]. This ant is canopy dwelling, rarely
descending to the forest floor and when it does it always
travels on well defined trails (Additional File 1). Trail
individuals do not forage on the forest floor and trails
normally ascend into the canopy within 3-5 m from
where they descended (suggesting that workers descend
only because breaks in the canopy necessitate a descent
to reach adjacent foraging crowns in the canopy). Unlike
ants on trails the manipulated ants in the pre-leaf biting
stage were all discovered walking alone on low vegeta-
tion, usually on saplings <50 cm above soil level and
only during the time interval 09:30-12:45 h (n = 21,
Figure 1 and 2). All 21 zombie ants that we followed
were confirmed to be infected either via dissection of
the head to reveal fungal cells or by observing the emer-
gence of O. unilateralis s.l. following death on the leaf
(Figure 3a). Post mortem fungal growth starts with
abundant hyphae emanating from the intersegmental
membranes within 2-3 days after host death and ulti-
mately leads to stalk formation from the back of the
ant’s head [20].
The host ant is diurnal at our field site [23] and
infected ants (n = 42) appeared even more restricted in
their activity as they were never observed in the early
morning or late afternoon (15:00-18:00 hrs), in spite of
our searches covering these early and late periods of the
day. The understory vegetation of our study site was
extensively searched during a year-long census program
that examined every leaf below 2 m height in 1360 m2
of forest habitat [23]. We therefore conclude that pre-
biting infected C. leonardi ants at this site were only
Hughes et al. BMC Ecology 2011, 11:13
http://www.biomedcentral.com/1472-6785/11/13
Page 2 of 10
active in the morning and that this observation was not
affected by sampling bias. The occasional trails of
healthy ants that can be found on the forest floor
(Additional File 1) were observed both during the
morning hours and in the late afternoon, with activity
on trails always ceasing around sunset, i.e. between
17:00-18:00 h.
Since behavioral manipulation alters normal behavior we
could not a priori exclude the possibility that infected
individuals would have become nocturnal. We therefore
conducted evening (after 18:00 h) and night (22:00-0:00 h)
surveys using torches, but did not find any C. leonardi
ants active in the dark. Furthermore, a collected colony of
C. leonardi that likely contained some naturally infected
ants was maintained under field laboratory conditions for
2 days and did not show any activity in the dark, suggest-
ing that behaviorally manipulated C. leonardi ants remain
only active during daylight.
We also performed 20 hours and 28 minutes of focal
observations on 12 infected ants that were found walk-
ing alone (infection status was later confirmed as
described above)(see Figure 1, individuals 10-21). These
individuals all expressed irregularly spaced whole body
convulsions (vertical bars on the periodogram in Figure
1), which often made the ant fall from the vegetation
onto the ground (denoted as stars on the periodogram).
After falling infected ants always resumed walking and
always climbed a small sapling or comparable plant,
which were abundantly present in the understory.
We never observed trail ants falling from vegetation. To
document this, we removed 13 such ants from a trail on a
liana approximately 1 m above the forest floor in the same
area where we observed the behaviorally manipulated ants.
The liana descended from the canopy and the trail
ascended into the canopy via a tree trunk less than 3 m
from where we collected the ants. The trail ants (assumed
to be uninfected) were placed on the ground and they all
quickly ascended into the canopy where we collected
them again from tree trunks ca. 1.5 m above ground. The
only exception was one trail ant that was predated upon
by a spider (none of the behaviorally manipulated ants we
observed were predated upon). Trail ants did not spend
extensive time walking in the understory. Their median
time between release and reaching the trunk on which
they ascended into the canopy was 28 minutes (range 7-
51, total observation time 6 h, 2 m). After these observa-
tions, the collected trail ants were maintained singly with-
out food and died within a few days without signs of O.
unilateralis s.l. fungal growth.
Before biting a leaf, infected ants were predominantly
walking (average proportion of time walking 0.62, range
0.11-1.00, total observation time 15 h, 35 m). They tra-
versed an average of 99 leaves (range 52-239, 8 focal
ants), which was ca. twice the number traversed by trail
ants (average 51, range 8-140, 13 focal ants). Because
trail ants were never observed walking on leaves except
at the times when we removed them from trails and
placed them on the ground we conclude that traversing
leaves is not a normal behavior. Therefore we did not
Figure 1 Zombie ant behavior. Focal animal observation
periodogram of ants infected by Ophiocordyceps unilateralis s.l. The
blue horizontal bars mark the observation period, the red triangles
mark moment of biting, the vertical bars mark spasm events and
the grey diamonds the falling off events. For four individuals that
belong to focal animals 1-9 only the biting time was recorded. The
biting time was recorded for 16 ants but only 15 triangles are
visible as two ants bite at exactly the same time (12:05). Inset
picture shows a dead ant on a leaf with the fungal stalk and spore
body that emerged from the head.
Figure 2 Synchronized manipulation of ants by fungi. A sun
position chart of the death grip. Solar altitude is represented by the
yellow bars and plotted against the y-axis and the biting times are
the red circles and plotted on the x-axis in solar time (this is true
local time accounting for longitude and different from Time in
Figure 1). The red circles are stacked to prevent overlapping. At
11:47 two ants bite so only 15 circles are visible though 16 ants
were recorded.
Hughes et al. BMC Ecology 2011, 11:13
http://www.biomedcentral.com/1472-6785/11/13
Page 3 of 10
statistically test for a difference between the numbers of
leaves traversed, as this was not biologically meaningful.
During the pre-biting phase behaviorally manipulated
ants appeared to express a random “drunkard’s walk”
such that an individual remained close to its starting
point [29] but precise trajectories were not mapped so
this remains a heuristic assessment. In all cases the
infected ants finally bit into leaves <3 m from where
they were first observed.
The timing at which infected ants bit into leaves was
synchronised around noon (Figure 1 &2; n = 16), sug-
gesting either a direct solar cue or an indirect one via
correlated temperature or humidity. The solar elevation
at the moment of biting was 80.28° +/- 1.32 SE, which
was close to the maximum solar elevation of 87.29° +/-
0.39 SE during our study period. Once they had bitten
leaves, ants rarely became detached and when this hap-
pened it was due to disturbance (two cases, # 15 and 18
Figure 3 Heads of manipulated ants colonized by fungi. A (top panel) is a light micrograph (LM) saggital section through the head of an O.
unilateralis s.l infected ant that was biting a leaf at the moment of fixation (i.e. alive). The small grey blobs are fungal hyphal bodies that fill the
head and mandibles. Note the spacing between the muscle fibers. The insect shows a close up of hyphal bodies around the post-pharyngeal
gland (PPG). B is the brain, Mu, Muscles and Cu is cuticle. B) is a LM of healthy muscle and C) is a LM of muscle from a behaviorally manipulated
ant that was biting a leaf and alive when removed for fixation. The small blobs between the fibers are fungal cells.
Hughes et al. BMC Ecology 2011, 11:13
http://www.biomedcentral.com/1472-6785/11/13
Page 4 of 10
in Figure 1, after very heavy rain). Biting leaves is not
part of the repertoire of healthy ants of this species.
Post-death grip behavior
After biting into leaves infected ants always died as this
is a developmental necessity for the subsequent growth
of the fungus [15,20]. It was not possible to determine
the exact time of death since obvious signs such as mus-
cle activity could be the result of fungal action, but it
did appear that ants could remain alive for as long as
six hours after biting. Video recordings of six live ants
biting leaves revealed very little behavior of interest
besides a periodic twitching of the legs (Additional File
2). The arrival of an ant of a different species close to a
biting ant provoked no responses (Additional File 2), in
contrast to healthy ants on trails, which were very
aggressive to other ant species they encountered at food
sources as well as to flying insects like wasps and flies
that landed near honey baits.
Muscular atrophy accompanies behavioral manipulation
At the moment of the death grip, when the ant is under
fungal control and biting into the major vein of a leaf
its head is filled with fungal cells (Figure 3). These cells,
called hyphal bodies, were very abundant and could be
found between the muscle fibers and surrounding the
brain and post pharyngeal gland (Figure 3), but not
inside muscles, brains or glands.
The most prominent other sign of infection, besides
the abundance of fungal cells inside the head capsule,
was that the mandibular muscles were atrophied. We
sectioned the heads of 10 ants that were biting leaves
and the pathology was the same across all 10. Mandibu-
lar muscle fibers, which normally attach to the head
capsule, often appeared to have become detached (Fig-
ure 3c) and where fibers remained attached they were
stretched (compare 3b and 3c). Ant workers have both
mandibular opening and closing muscles and these can
be discriminated in healthy ants by their typical length
of sarcomeres: 2-3 μm for opening muscles and 5-6 μm
for closing muscles (Figure 3b). However, in parasitized
ants the characteristic stretching of sarcomeres made it
impossible to accurately distinguish between these two
types of muscles. This may imply that fungal effects on
these muscles are unlikely to be cell specific at the time
of biting. Our behavioral observations revealed that the
mandibles worked normally in the hours preceding the
death grip as infected ants were observed to self groom,
cleaning their antennae and legs, which involves precise
opening and closing of the mandibles as these appen-
dages are pulled across the maxillae to be cleaned.
At the sub-cellular level (as seen with TEM) the mus-
cles of infected ants were very distinct from those of
healthy individuals (Figure 4). Striated muscles (such as
the mandible muscles) are composed of fibers that are
multinucleated cells formed as a result of cell fusion.
These fibers contain thick (myosin) and thin (actin) fila-
ments which attach during cross-bridge cycling leading
to muscle shortening. To achieve contraction mitochon-
dria and sarcoplasmic reticulum provide energy (ATP)
and ionic calcium (Ca++), respectively. At the end of
each sarcomere unit there is a z-line (sarcomeres are in
fact defined as the area between z-lines), which can be
thought of as the anchor points for muscle contraction.
Infected ants sampled during the death grip had broken
z-lines and significantly less dense sarcoplasmic reticu-
lum and mitochondria. This was determined from a
measurement of the increase of interfibrillar spaces that
appears following the loss of organelles, which in this
case are sarcoplasmic reticulum and mitochondria
(Kruskal-Wallis test, 20.25, df = 1, p < 0.0001, n = 6,
Figure 4e). Similar to the light micrographs, the trans-
mission electron micrographs showed a distinct atrophy
in the muscles of infected ants.
Despite the apparent atrophy of muscles the behavio-
rally manipulated ants were able to exert considerable
force. We removed 29 dead ants from divers