The intertwined population biology of two Amazonian

13 Pages·2009·0.27 MB·English

Ecology,90(6),2009,pp.1595–1607
(cid:2)2009bytheEcologicalSocietyofAmerica
The intertwined population biology of two Amazonian
myrmecophytes and their symbiotic ants
MEGAN E. FREDERICKSON1 ANDDEBORAHM.GORDON
DepartmentofBiologicalSciences,StanfordUniversity,Stanford,California94305-5020USA
Abstract. A major question in ecology is: how do mutualisms between species affect
population dynamics? For four years, we monitored populations of two Amazonian
myrmecophytes, Cordia nodosa and Duroia hirsuta, and their symbiotic ants. In this system,
weinvestigatedhowpositivefeedbackbetweenmutualisticplantsandantcoloniesinfluenced
populationprocessesattwoscales:(1)howmodularorganismssuchasplantsandantcolonies
grew,org-demography,and(2)howpopulationsgrew,orN-demography.Wefoundevidence
ofpositivefeedbackbetweenantcolonyandplantgrowthrates.Plantswithmutualisticants
(Aztecaspp.andMyrmelachistaschumanni)grewinageometricorautocatalyticmanner,such
that the largest plants grew the most. By contrast, the growth of plants with parasitic ants
(Allomerus octoarticulatus) saturated. Ant colonies occupied new domatia as fast as plants
produced them, suggesting that mutualistic ant colonies also grew geometrically or
autocatalytically to match plant growth. Plants became smaller when they lost ants. While
unoccupied, plants continued to become smaller until they had lost all or nearly all their
domatia.Hence,thelossofmutualisticantslimitedplantgrowth.C.nodosaandD.hirsutalive
longerthantheirantsymbiontsandweresometimesrecolonizedafterlosingants,whichagain
promotedplantgrowth.Plantgrowthhadfitnessconsequencesforantsandplants;mortality
andfecunditydependedonplantsize.Positivefeedbackbetweenantsandplantsallowedafew
plants and ant colonies to become very large; these probably produced the majority of
offspring in thenextgeneration.
Keywords: ant–plantinteractions;Cordianodosa;demography;domatia;Duroiahirsuta;modularity;
mutualism;myrmecophytes;Peru;populationdynamics;positivefeedback;symbiosis.
INTRODUCTION infinityinwhatMay(1982)called‘‘anorgyofreciprocal
benefaction.’’ This led many to suggest how explosive
How do interactions between species affect the
population growth might be kept under control (Goh
dynamicsofnaturalpopulations?Overnearlyacentury,
1979, Heithaus et al. 1980, Addicott 1981, May 1981,
ecologists have made substantial progress toward
1982,Boucheretal.1982,Dean1983,WolinandLawlor
understanding the population dynamics of competitive
1984, Pierce and Young 1986, Wright 1989). Three
andtrophicinteractions.Bycomparison,thepopulation
generalsolutionsemerged:resourcelimitationmayresult
dynamicsofreciprocallybeneficialinteractionsbetween
inintraspecificcompetitionthatlimitspopulationgrowth
species, or mutualisms, are not well understood (Hol-
(e.g.,Goh1979);athirdspeciessuchasapredatorora
landetal.2002),eventhoughmutualismsareubiquitous
competitor may limit populationgrowth (e.g., Heithaus
and ecologically important (Boucher et al. 1982,
et al. 1980); or there may be diminishing returns to the
Bronstein1994, Brunoet al.2003).
mutualism as populations grow (e.g., Holland et al.
Early efforts to model the population dynamics of
2002). Nonetheless, how the costs and benefits of
mutualismquicklyidentifiedanimportantissue:positive
mutualismshape populationdynamics remains an open
feedback between mutualists tends to generate runaway
empiricalquestion(Hollandetal.2002).
populationgrowth.Thefirstmodelsofmutualismbegan
Symbiotic mutualisms pose a particular problem
bytakingtheLotka-Volterracompetitionequationsand
becausetheyarespatiallystructuredandhighlyvariable
reversingthesignsofthecompetitioncoefficients(Gause
fromoneinteractiontothenext.Symbiosesareintimate
and Witt 1935). But in these models, if the effect of
associations between species that involve protracted
mutualism is sufficiently strong, populations grow to
physical contact between partners (Boucher et al. 1982,
Saffo 1992). Many mutualisms are symbiotic, such as
Manuscript received 3 January 2008; revised 1 July 2008; corals and dinoflagellates, or plants and mycorrhizal
accepted 8 September 2008. Corresponding Editor: D. H. fungi, though many are not, such as plants and
Feener,Jr. pollinators, or plants and seed dispersers. The mean-
1Present address: Society of Fellows and Museum of
fieldassumptionofmodelsbasedontheLotka-Volterra
Comparative Zoology, Harvard University, 26 Oxford
Street,Cambridge,Massachusetts02138USA. equations is a good approximation when populations
E-mail:[email protected] arewellmixed,butbreaksdownwhenindividualslivein
1595
1596 MEGANE.FREDERICKSONANDDEBORAHM.GORDON Ecology,Vol.90,No.6
spatially structured environments (Dieckmann et al. forworkersandantcolonies.Similarly,thoughbymore
2000, Pascual 2005). Symbiotic organisms interact in complexprocesses,birthsofplantshootsorantworkers
local ‘‘neighborhoods’’ often comprised of only two areconvertedintoreproduction,thebirthsofnewplants
individuals, host and symbiont. In symbioses, the ornewant colonies.
benefitsofmutualismarerepeatedlyexchangedbetween Few time-series data are available with which to
thesametwopartners,withtheresultthatsurvivaland evaluate population-dynamical models of mutualism.
reproductiondependmoreonthetraitsoftheindividual Here we report time-series data on the g- and
partners, than on their population densities. Many N-demography of two plant species and four ant
studies have investigated the effects of spatial structure species that interact as mutualists in the Amazonian
and individual variability in competitive and trophic rain forest; and for comparison, we report on an ant
interactions(Hustonetal.1988,Dieckmannetal.2000, species that is a parasite of these mutualisms. To
and references therein). Here we consider the effects of explore how positive feedback, symbiosis, and modu-
spatial structure and individual variability on the larityshapethepopulationdynamicsoftheseant–plant
population dynamicsof mutualism. interactions, we investigate how the growth rate of a
Manysymbioticmutualismsoccurbetweenorganisms plant (i.e., g-demography) changes over the course of
with modular growth. Such organisms grow by the its symbiotic association with an ant colony and
iteration of modules, permitting indeterminate growth. following the death or desertion of the ant colony.
The modules may or may not become physically Wealsoconsidertheg-demographyofantcolonies.We
separate units. Corals and dinoflagellates, plants and thenexaminehowchangesing-demography,especially
mycorrhizal fungi, fungi and algae, termites and their those changes that result from positive feedback
gut protozoa, plants and ants, and ants and fungi are between mutualists, affect the N-demography of ant
all examples of symbiotic mutualisms involving mod- colonies and plants.
ular organisms (Wulff 1985). Demographic processes
occur at two levels in modular organisms: individual STUDY SYSTEM
organismsgrowasmodulesarebornanddie,producing We studied two species of myrmecophytic plants,
g-demography; and populations grow as individuals Cordia nodosa Lam. (Boraginaceae) and Duroia hirsuta
are born and die, producing N-demography (Harper [PoeppigandEndl.]K.Schum(Rubiaceae),andtheirant
1977, 1981, Dirzo 1984, Watkinson and White 1986). symbionts.Myrmecophytesareplantsthatbearspecial-
Thus the population dynamics of modular organisms ized structures, called domatia, to house ants. Most
have a nested structure, with g-demography nested in myrmecophytes produce domatia constitutively (Blu¨th-
N-demography. When a symbiotic mutualism occurs gen and Wesenberg 2001), and both C. nodosa and D.
between modular organisms, the interaction tends to hirsuta produce hollow, swollen stem domatia, whether
directly affect the g-demography of the participants, ornotantsarepresent(seePlate1).BothC.nodosaand
andin sodoing,indirectly affectsN-demography. D. hirsuta produce one domatium per internode, so
Here we consider a symbiotic mutualism between a individual plants can be considered to grow as popula-
plant and an ant colony in which the ant colony nests tionsofdomatia. Plantsadddomatiaastheygrowand
inside the plant’s hollow stems. Plants are modular losedomatiawhenbranchesdieandfalloff.Hence,the
becausetheygrowbytheserialrepetitionofshoots;ant g-demography of these plants can be described as
coloniesaremodularbecausetheygrowbytheaddition changesovertimeinthenumberofdomatiaperplant.
of sterile workers. Changes in the size of plants or ant Phytoecious ants nest inside the domatia of myrme-
colonies can be positive or negative, depending on the cophytic plants (Quek et al. 2004). C. nodosa and D.
birthanddeathratesofshootsandworkers,respectively hirsuta associate with two or more phytoecious ant
(Watkinson and White 1986, Tschinkel 1993). This is specieseach.Twoantspeciesassociatewithbothplants:
the g-demography of plants and ant colonies, while Azteca depilis Emery (Dolichoderinae) and Myrmela-
N-demography involves the births and deaths of whole chistaschumanniEmery(Formicinae).Thesetwospecies
plantsandwholeantcolonies.Insymbioticmutualisms are the only ants that associate with D. hirsuta. C.
between ants and plants, the benefits of mutualism nodosa also associates with two undescribed species in
generally accrue at the level of g-demography and in the genus Azteca, and with Allomerus octoarticulatus
turn, g-demography generates N-demography. Ant Mayr (Myrmicinae). Workers are often seen walking
workers usually defend their host plant against herbi- between domatia along plant stems, but reproductives
vores (Heil and McKey 2003), which affects the are rarely observed outside domatia. All of these ant
g-demography of plants by reducing the death rates of species tend scale insects (Hemiptera: Sternorrhyncha:
plant modules/shoots. From the plant, the ant colony Coccoidea) inside plant domatia; the scale insects
receives nest sites in the form of hollow stems, which providefoodtotheantcolonyintheformofhoneydew.
should both increase the birth rate and decrease the The ant species differ in how they treat their host
death rate of ant modules/workers. These effects plants. Azteca spp. and M. schumanni behave as
translate to N-demography. First, when all the shoots mutualists, protecting their host plants against herbi-
onaplantdie,theplantdiesaswell,andthesameistrue vores (Frederickson 2005). M. schumanni also provides
June2009 ANT–PLANTPOPULATIONBIOLOGY 1597
an extra benefit; workers of this species kill hetero- For Azteca spp. inhabiting C. nodosa, ants were
specific plants (potential competitors) near their host identified only to genus because species are difficult to
plants,therebycreatinglargestandsofhostplantscalled distinguishbasedonworkermorphologyalone,andwe
‘‘devil’s gardens’’ (Frederickson et al. 2005). Al. octoar- rarelyobservedreproductives.
ticulatus behaves as a parasite. Although worker Al. In 2004–2005, we collected ant colonies at a second
octoarticulatus ants protect C. nodosa leaves against site, the Reserva Nacional Allpahuayo-Mishana
herbivores, they do not protect plant stems (Yu and (RNAM; 3.968 S, 73.48 W), located about 130 km west
Pierce 1998, Frederickson 2005). As a result, C. nodosa ofMSBSontheIquitos-NautahighwayinLoreto,Peru.
withAl.octoarticulatusantsaremorefrequentlygirdled OfthethreespeciesofAztecathatoccupyC.nodosa,we
by Trachysomus beetles than C. nodosa with other ant focused only on Azteca depilis. We collected a total of
species (Yu and Pierce 1998, Frederickson 2005). 181Allomerusoctoarticulatuscoloniesand97Az.depilis
Furthermore, Al. octoarticulatus sterilizes C. nodosa by colonies(40fromC.nodosaand57fromD.hirsuta).We
destroyingflowers (Yu andPierce 1998). alsocollectedtheantsinhabiting21D.hirsutatreesfrom
Azteca spp. and Al. octoarticulatus colonies are 10 Myrmelachista schumanni colonies. To collect the
monogynous, meaning each colony has a single egg- ants, we removed all domatia from each tree using tree
laying queen. One Azteca spp. or Al. octoarticulatus prunersandthensealedthedomatiainZiplocbags.We
colony nests in a single host tree. In contrast, M. could not prevent workers from escaping during
schumanni colonies are polygynous, meaning they have collection, but we could collect nearly all reproductives
multiple egg-laying queens, and these colonies can becausetheyrarelyleavedomatia,evenwhendisturbed.
occupy many host plants (Frederickson et al. 2005, Bags of domatia were stored in an ice chest after
Frederickson and Gordon 2007). All these ant species collection. Within 48 hours of collection, each doma-
found colonies independently; queens fly to plants to tium was cut open and the contents examined. Each
colonize them. A colony of M. schumanni begins in the domatiumwasscoredascontainingreproductives(with
same way as an Azteca spp. or Al. octoarticulatus or without workers), containing workers only, or being
colony; a queen colonizes a single host tree, lays eggs, empty. We recorded the number of female alates and
and produces workers. However, unlike the other ants, female alate pupae for each domatium containing
M.schumanniworkerskillheterospecificplantsgrowing reproductives.
near their host plant, thereby creating a bare patch For the ant species, we define g as the number of
where, over time, more D. hirsuta and C. nodosa trees workers in an ant colony, and N as the number of ant
establish. The M. schumanni colony expands to occupy colonies in the population. Similarly, for the plants, we
the new plants and continues to enlarge the patch. As definegasthenumberofdomatiaonatreeandNasthe
the M. schumanni colony grows, it adds queens, number of trees inthe population.
although it is not known by what mechanism. The M.
g-demographyof myrmecophytic plants
schumannicolonycanoccupyanynumberoftreesfrom
onetoseveralhundred;whenthecolonyoccupiesmany Weconsideredplantstobegrowingaspopulationsof
trees, the trees are clumped together to form a devil’s domatia, such that g is the number of domatia on a
t
garden. plant at time t. We used geometric growth as a null
model; we chose a discrete-time model because we
METHODS counted domatia on trees at discrete time intervals of
For four years, we monitored populations of Cordia one year. The branching pattern of many trees is
nodosaandDuroiahirsutaina4-haplotatMadreSelva evocative of geometric growth (i.e., one branch splits
BiologicalStation(MSBS;3.638S,72.248W)ontheRio to become two, which split to become four, etc.). If a
OrosaintheDepartmentofLoreto,Peru.The4-haplot plantisgrowinggeometrically,theng ¼Rg.Risthe
tþ1 t
was oriented along cardinal directions and divided into geometric growth factor; if R , 1, the plant is getting
16 503 50 m squares. In July–August of 2002, 2003, smallerbylosingmoredomatiathanitadds,andifR.
2004,and2005,weexhaustivelysearchedeachsquarefor 1, the plant is getting bigger by adding more domatia
all domatia-bearing C. nodosa and D. hirsuta plants by thanitloses.GeometricgrowthassumesthatRdoesnot
walkingeachsquarebackandforthinaseriesof2.5-m depend on g (Roughgarden 1998, Turchin 2003). This
transects.Everyyear,wenumbered,tagged,andmapped property of geometric growth is usually called ‘‘density
each new (i.e., unmarked) plant that we found, and we independence’’ in population biology. Here we call it
visited every plant that had been marked in previous ‘‘size independence’’ because the concept of population
years.Everyyear,wecountedthenumberofdomatiaon density does not translate well from N-demography to
everyplantintheplot,exceptin2005,whenwecounted g-demography.
domatiaonlyonnewplantsandonplantswithoutants. We tested whether the growth of C. nodosa and D.
Wealsorecordedwhetherplantswereproducingflowers hirsutaissizeindependentbyanalyzingtherelationship
and fruit, and which species of symbiotic ant, if any, between R and g using linear regression. We used data
occupied each plant. A plant was considered to be from the 2003–2004 census interval to calculate R, or
occupiedbyantsifworkerswereobservedontheplant. g /g , and to investigate the relationship between
2004 2003
1598 MEGANE.FREDERICKSONANDDEBORAHM.GORDON Ecology,Vol.90,No.6
R and g . We used the data from the 2003–2004 the analysis. For plants that retained ants, R was
2003
censusintervalbecausewedidnotcountthenumberof measuredasg /g .Forplantsthatlostants,Rwas
2004 2003
domatia on all plants in 2005, and sample sizes were measured as g /g, where t was the census before the
tþ1 t
larger in 2003–2004 than in 2002–2003. To investigate antcolonydiedordesertedtheplant,andtþ1wasthe
whether the nonindependence of R and g inflated the firstcensusafter.Weanalyzedthedatausingtwo-tailed
significance of the regression analyses, we created our t tests assuming unequal variances or using ANOVA
own null distributions by recalculating regressions for followedbyTukeyposthoc tests.
random re-pairings of R and g . We compared the
2003
g-demography ofphytoecious ant colonies
observed regression coefficients against the distribution
of regression coefficients generated by 10000 iterations We used the data from RNAM to investigate the
of the randomization procedure. P values were calcu- g-demographyofphytoeciousantcolonies.Becausewe
lated by counting the number of times the regression could not prevent workers from escaping during
coefficientsoftherandomizeddataweregreaterthanthe collection, we did not count the number of workers in
observed regressioncoefficient anddividingby10000. the ant colonies we collected. Instead we used the
In general, there are three possible relationships number of domatia occupied by ants as an indirect
between R and g: (1) No relationship between R and measure of colony size. We analyzed the relationship
g means that plant growth is size independent; we between the number of domatia occupied by ants and
cannot reject the null model that C. nodosa and D. the total number of domatia on a plant, square-root
hirsuta grow geometrically. If R . 1, then plants transforming bothvariables.
continue to grow larger, becoming infinitely large in
N-demography ofplants andant colonies
infinite time and plant growth does not slow down as
plants get large. (If R , 1, then plants become smaller To evaluate the N-demography of ant colonies and
and smaller, becoming infinitely small in infinite time.) plants, we chose the following measures of recruitment
(2)ApositiverelationshipbetweenRandgmeansthat and death of ant colonies and host plants. A plant was
plant growth is positively size dependent, and we reject considered to have recruited into the population if it
thenullmodelthatplantsgrowgeometrically.Growthis boredomatiaandhadnotbeenpreviouslycensused,and
autocatalytic,andplantsgrowfaster,notslower,astrees to have died if all its aboveground tissue was dead. An
get large. If autocatalytic growth continues indefinitely, ant colony was considered to have colonized a plant
plantswillbecomeinfinitelylargeinfinitetime(Rough- between censuses if its host plant was not occupied or
garden1998).(3)AnegativerelationshipbetweenRand wasoccupiedbyacolonyofadifferentantspeciesinthe
gmeansthatplantgrowthisnegativelysizedependent. previous census. An ant colony was considered to have
Again, we reject the null model that plants grow died between censuses if the host plant that it occupied
geometrically. Instead, plant growth slows down or in one census was not occupied or was occupied by a
evenbecomes negativeas plants getlarge. colonyofadifferentantspeciesinthenextcensus.For
Weinvestigatedwhathappenstoplantgrowthwhena M.schumanni colonies that occupied manyhostplants,
plantlosesitsants.Wecomparedplantgrowthrate,R, we considered a colony to have died when all the host
between (1) plants whose ants survived the 2003–2004 plants previously occupied by the colony were unoccu-
census interval, and (2) plants whose ants died or piedoroccupiedbyotherant species.Whenonlysome
abandoned plants in any census interval. The former of the host plants previously occupied by a M.
group included plants whose ants survived through all schumannicolonywereunoccupiedoroccupiedbyother
censuses (2002–2005), and also plants that were colo- ants, we considered the M. schumanni colony to have
nized by ants between 2002 and 2003 and whose ants abandoned these trees. Our methodology underesti-
survivedto2005.Plantswhoseantssurvivedfrom2003– mates both ant colony deaths and ant colony coloniza-
2004, but died or abandoned the plant in 2004–2005, tions, because we did not count colonies that died and
were included only in the latter group, so no plant was had their host plants colonized by a new colony of the
includedintheanalysistwice.Plantslostantswhenant samespeciesbeforethenextcensus.However,sincethe
colonies died and also when M. schumanni colonies numbersof ant colonydeathsand antcolony coloniza-
abandoned trees. When an ant colony died or aban- tions are underestimated by exactly the same amount,
donedatree,theplanttypicallyremainedunoccupiedby estimatesofnetpopulationgrowthratearenotaffected.
ants for at least one census, with few exceptions Since plants were counted in the yearly census only
(Appendix E). Thus we restricted the second group to after they produced domatia, and ant colonies were
plantsthatwereoccupiedbyantsinonecensusbutdid countedonlyaftertheyproducedtheirfirstworkers,the
not have ants in the following census. In general, there ratesatwhichplantsandantcolonieswereaddedtothe
weretoofewdatapoints(i.e.,n(cid:2)5)toalsocomparethe population each year represent colonization rates
growth rateof plantsthat switchedant speciesbetween instead of true birth rates. The colonization rate of ant
censuses. Only Az. depilis frequently replaced M. colonies andplants (b) was calculatedas b¼ln(N )(cid:3)
tþ1
schumannionD.hirsutatreeswithinonecensusinterval ln(N (cid:3)D),thedeathrate(d)asd¼ln(N)(cid:3)ln(N (cid:3)D),
t t t
(n¼50); we included these trees as a separate group in andthenetgrowthrate(r)asr¼ln(N )(cid:3)ln(N),where
tþ1 t
June2009 ANT–PLANTPOPULATIONBIOLOGY 1599
TABLE1. Dependenceofplantgrowthrate,R,onplantsize,g,fortwospeciesofAmazonianmyrmecophyteswithorwithout
variousantspecies.
Hosttreeandantinteraction df Regressionequation R2 P Sizedependence
Cordianodosa
Nointeraction(noants) 45 R¼(cid:3)0.139gþ1.517 0.176 0.004 negative
Parasitic(Allomerusoctoarticulatus) 137 R¼(cid:3)0.027gþ1.643 0.115 (cid:4)0.001 negative
Mutualistic(Aztecaspp.) 53 R¼(cid:3)0.002gþ1.341 0.042 0.137 sizeindependent
Mutualistic(Myrmelachistaschumanni) 15 R¼(cid:3)0.004gþ1.310 0.027 0.541 sizeindependent
Duroiahirsuta
Nointeraction(noants) 107 R¼(cid:3)0.003gþ0.887 0.009 0.327 sizeindependent
Mutualistic(Aztecadepilis) 68 R¼0.001gþ1.097 0.052 0.058(cid:2) marginallypositive
Mutualistic(Myrmelachistaschumanni) 376 R¼0.001gþ1.002 0.058 (cid:4)0.001 positive
(cid:2)ThestatisticalsignificanceoftherelationshipbetweenRandgforD.hirsutawithAz.depiliswasmarginal.However,whenwe
excludedoneoutlier,therewasasignificantpositiverelationship,withP¼0.01.SeealsoAppendixB:Fig.B1b.
N was the number of plants or colonies at the start of To determine whether plant fecundity is size depen-
t
the census interval, N was the number of plants or dent, we analyzed the relationship between fruit
tþ1
coloniesattheendofthecensusinterval,andDwasthe production and domatia number for D. hirsuta. We
number of deaths during the census interval (Chew didnotconductthesameanalysisforC.nodosabecause
1987). This model assumes that all deaths occur before only four out of the 346 C. nodosa in our plot were
anynewplantsorcoloniesareaddedtothepopulation. observedtoproducefruitduringourcensuses.Weused
Thegeometricmeanwasusedtocalculateaveragerates thedatafromRNAMtodeterminethesizedependence
(Carey1993).Lifeexpectancyatbirthwascalculatedas of ant colony fecundity. We analyzed the relationship
the inverseofthe average deathrate(Carey 2001). between the number of female alates (including both
eclosedalatesandalatepupae)inanantcolonyandthe
Links betweeng-demography andN-demography numberofdomatiaonaplant,square-roottransforming
both variables.
Weinvestigatedthesizedependenceofmortalityand
Finally, we investigated the size and age of plants
fecundity for both plants and ant colonies. To investi-
colonized by ants. We divided plants into different age
gate the effect of plant size on plant mortality, we
classes and calculated what percentage of unoccupied
compared the initial size of plants that survived the
plantsineachageclasswascolonizedbyants.Inthefirst
2003–2004censusintervaltotheinitialsizeofplantsthat
year of our study (2002–2003), we could divide plants
diedinanycensusinterval.Tomakethesecomparisons,
intoonlytwoageclasses:plantsmorethanoneyearold
we used two-sided t tests, assuming unequal variances.
were present in the initial census in 2002, while plants
Thefirstgroupincludedplantsthatsurvivedthroughall
less than one year old recruited into the population
censuses(2002–2005)andalsoplantsthatrecruitedinto
between2002and2003. Thenextyear,wecould divide
the population between 2002 and 2003 and survived to
plantsintothreeageclasses:plantsmorethantwoyears
2005. Plants that survived from 2003–2004, but died in
old, one to two years old, and less than one year old.
2004–2005wereincludedinthelattergroup,sonoplant Finally, in 2004–2005, we could divide plants into four
was included twice. For plants that survived, plant size age classes: plants more than three years old, two to
was measured as the number of domatia in the 2003 threeyearsold,onetotwoyearsold,andlessthanone
census.Forplantsthatdied,plantsizewasmeasuredas yearold.
the number of domatia in the census before the plant
died. To investigate the effects of plant size on ant RESULTS
colony mortality, we performed the same analysis, g-demographyof myrmecophytic plants
exceptwecomparedplantswhoseantcoloniessurvived
Plants with mutualistic ants grew geometrically or
the 2003–2004 interval to plants that lost ants in any
autocatalyticallyforaslongasthemutualismpersisted,
census interval (see Methods: g-demography of myrme-
while other plants inevitably reached a size beyond
cophytic plants). We square-root transformed the
whichnofurthergrowthwaspossible.Linearregression
number of domatia to improve normality. For M. and randomization tests produced nearly identical
schumanni, we also investigated the effect of devil’s results (Appendix A); only linear regression results are
garden(i.e.,patch)sizeonthesurvivalofM.schumanni discussed.Allthreetypesofsizedependenceoccurredin
coloniesbycomparingthenumberoftreesoccupiedby this system (Table 1). Cordia nodosa without ants or
acolonythatsurvivedthe2003–2004censusintervalto withAllomerusoctoarticulatusexhibitednegativelysize-
the number of trees initially occupied by a colony that dependent growth; there was a significant negative
diedinanycensusinterval.Becausethenumberoftrees relationship between R and g (Table 1; Appendix B).
indevil’sgardenswasnotnormallydistributed,weused These plants tended to grow when small; the 95% CI
a nonparametrictest tomake thislastcomparison. aroundthey-interceptwas.1(noants,1.174–1.859;Al.
1600 MEGANE.FREDERICKSONANDDEBORAHM.GORDON Ecology,Vol.90,No.6
FIG.1. Plantgeometricgrowthrate,R(mean6SE),forAmazonianhosttrees(a)Cordianodosaand(b)Duroiahirsutawith
differentantpartners(MC¼Myrmelachistaschumanni,AZ¼Aztecaspp.,AO¼Allomerusoctoarticulatus)orwithoutants.Black
barsrepresentplantswithantcoloniesthatsurvivedfromonecensustothenextorplantsthatdidnothaveants.Theseplantsdid
notchangetheiroccupationstatusfromonecensustothenext.Iftheyhadantsin2003,theywereoccupiedbythesameantspecies
in2004,andiftheydidnothaveantsin2003,theystilldidnothaveantsin2004.Darkgraybarsrepresentplantswithantcolonies
thatdiedbetweencensuses,lightgraybarsrepresentplantsthatwereabandonedbyM.schumanni,andwhitebarsrepresentplants
thatwereoccupiedbyM.schumanniinonecensusandbyAz.depilisinthenextcensus.Significanceofmeansisindicated.
*P(cid:2)0.05;**P(cid:2)0.01;***P(cid:2)0.001;NS,notsignificant.
octoarticulatus, 1.388–1.899).Butbeyondacertain size, mutualism persisted, these plants increased in size, and
these plants tended to lose domatia overall. This size therewasnoevidenceofanupperlimittoplantgrowth.
(i.e.,thevalueofgatwhichR¼1)wasfourdomatiafor The growth of D. hirsuta with mutualistic ants was
C. nodosa without ants and 24 domatia for C. nodosa positively size dependent; these plants grew faster the
with Al. octoarticulatus. By contrast, Duroia hirsuta larger they became (Table 1; Appendix B). The
plants without ants did not grow when small. Instead, statistical significance was marginal for D. hirsuta with
these plants tended to lose domatia regardless of their Azteca depilis, but became highly significant when we
initial size; the 95% CI around the y-intercept was ,1 excludedoneoutlier.ForbothMyrmelachistaschuman-
(0.808–0.965). They lost domatia in an approximately ni- and Az. depilis-occupied D. hirsuta, the effect of the
geometric fashion;therewasnorelationship betweenR positive slope of the relationship between R and g was
and g (Table 1, Appendix B). Thus although the two negligibleatsmallsizesbutsubstantialforlargeplants.
plant species did not have the same growth pattern, in We calculated that, as a result of the positive feedback
theabsenceofmutualisticants,neitherC.nodosanorD. between R and g, a D. hirsuta plant with 32 domatia
hirsutacontinuedto grow afterreachinga certain size. gains only one extra domatium in a year, but a D.
The growth of plants with mutualistic ants did not hirsuta plant with 200 domatia gains 40 extra domatia,
taper off even at the largest sizes, but continued andaD.hirsutaplantwith400domatiagains160extra
unabated or even accelerated as plants grew. The domatiain ayear.
growth of C. nodosa with mutualistic ants was statisti- Plants grew for as long as they were occupied by
callyindistinguishablefromgeometricgrowth;Rdidnot mutualisticants,butbecamesmallerwhentheylostants
dependong(Table1;AppendixB).Foraslongasthe (Fig. 1). In the case of plants with Azteca spp. or Al.
June2009 ANT–PLANTPOPULATIONBIOLOGY 1601
PLATE1. Cordianodosa(left)andDuroiahirsuta(right)branches,showingthestemswellings(domatia)thathouseants.Photo
credit:M.E.Frederickson.
octoarticulatusants,meanplantgrowthwassignificantly N-demographyof plantsandant colonies
greater for plants with ant colonies that survived than
On average, C. nodosa and D. hirsuta plants lived
for plants with ant colonies that died between censuses
longer than their ant colonies. Per capita annual
(Fig. 1; C. nodosa with Azteca spp., t ¼ 5.94, P ,
14 mortality of C. nodosa and D. hirsuta averaged 0.013
0.0001;C.nodosawithAl.octoarticulatus,t ¼6.78,P
135 and 0.017, respectively (Appendix D). Thus life expec-
, 0.0001; D. hirsuta with Az. depilis, t ¼ 3.20, P ¼
12 tancies at birth for C. nodosa and D. hirsuta are
0.008). Most of the C. nodosa plants that lost M.
approximately 77 years and 59 years, respectively. Ant
schumanni ants were abandoned by the colony and did
colony mortality exceeded plant mortality, averaging
nothaveantsinthenextcensus(n¼8).ThreeC.nodosa
0.128forAl.octoarticulatuscolonies,0.072forcolonies
trees were inhabited by other ants after losing M.
ofAztecaspp.inhabitingC.nodosa,0.074forAz.depilis
schumanni, and one M. schumanni colony inhabiting a
colonies inhabiting D. hirsuta, and 0.093 for M.
single,isolatedC.nodosatreedied;thesetreeswerenot
schumanni colonies (Appendix D). Thus life expectancy
includedintheanalysis.C.nodosaplantsabandonedby
is approximately 7.8 years for Al. octoarticulatus
M. schumanni grew significantly less than C. nodosa
colonies,14yearsforAztecaspp.colonies,and11years
plants that kept M. schumanni (Fig. 1; t ¼2.43, P¼
19 for M.schumanni colonies.
0.03).Similarly,mostoftheD.hirsutatreesthatlostM.
There was little evidence that one ant colony ever
schumanni ants were abandoned by the colony and did
invades and expels another colony that is already
not have ants in the next census (n¼103). However, a
established on a host plant. Ants generally colonized
large number of D. hirsuta trees were occupied by M.
plants that were unoccupied at the previous census
schumanni ants in one census, but by Az. depilis in the
(Appendix E). There was one major exception: Az.
next (n ¼ 50). Both of these groups of trees grew
depiliscolonized50D.hirsutaplantsthatwereoccupied
significantlylessthanD.hirsutathatkeptM.schumanni
by M. schumanni at the previous census (Appendix E).
(Fig. 1; ANOVA, F ¼24.00, P , 0.0001). Five M.
2,386
Thesewerelocatedmostlyaroundthemarginsofdevil’s
schumanni colonies occupying only one or a few D.
hirsuta trees died during the study; in three cases, a M. gardens. It was not possible to determine whether M.
schumannicolonyonanisolatedD.hirsutawasreplaced schumanni vacated these plants before they were
byanAz.depiliscolony betweencensuses.Again,these colonized by Az. depilis, or whether Az. depilis actively
trees were not included in the analysis of plant growth castoutM.schumannifromtheseplants.However,four
ratebecause of the verysmallsamplesizes. of the six M. schumanni colonies that died (Appendix
D2) were immediately replaced by Az. depilis colonies,
g-demography ofphytoecious ant colonies suggestingthatiftheyoccur,Az.depilistakeoversofM.
Ants occupied nearly all the available domatia on a schumanni colonies may be an important source of
host plant, no matter how large a plant became. The mortality forM.schumanni colonies.
number of occupied domatia and the total number of BecauseC.nodosaandD.hirsutalivelongerthantheir
domatiawerestronglyandsignificantlycorrelatedforall ant symbionts,theseplantscanberepeatedlycolonized
ant–plantassociations(AppendixC;C.nodosawithAl. by ants throughout their lives. Ninety-eight out of a
octoarticulatus,r2¼0.996,P,0.0001;C.nodosawithAz. total of 174 ant colonizations of plants, or 56%,
depilis,r2¼0.995,P,0.0001;D.hirsutawithAz.depilis: occurred on plants that had previously been occupied
r2¼0.991,P,0.0001;D.hirsutawithM.schumanni,r2¼ by ants, even though most plants were unoccupied
1.000,P,0.0001).Thenumberofoccupieddomatiadid immediately before colonization. In addition to the 50
notsaturateasplantsbecamelarge. D. hirsuta that were occupied by M. schumanni at the
1602 MEGANE.FREDERICKSONANDDEBORAHM.GORDON Ecology,Vol.90,No.6
FIG.2. Initialplantsize,g(mean6SE),forAmazonianhosttrees(a)Cordianodosaand(b)Duroiahirsutawithdifferentant
partners (MC¼Myrmelachista schumanni, AZ¼Azteca spp., AO¼Allomerus octoarticulatus) or without ants. Plant size is
measured as the number of domatia per tree. Explanations of bar shading are the same as in Fig. 1. Significance of means is
indicatedas*P(cid:2)0.05;**P(cid:2)0.01;***P(cid:2)0.001;NS,notsignificant.
censusbeforetheywerecolonizedbyAz.depilis,another abandoned by M. schumanni were significantly smaller
13 plants were occupied by a different ant species the than D. hirsuta that kept M. schumanni; D. hirsuta on
year before they were colonized (Appendix E). An which M. schumanni was replaced by Az. depilis were
additional32plants were colonized byants after losing intermediate in size between these two groups (Fig. 2;
antsandbeingunoccupiedinonecensus,andthreewere ANOVA, F ¼ 70.85, P , 0.0001). Only six M.
2,386
colonizedbyantsafterlosingantsandbeingunoccupied schumanni colonies died during our study, and these
in twoconsecutive censuses.
coloniesoccupiedsignificantlyfewerplantsthantheM.
schumannicoloniesthatsurvived(U ¼30,P¼0.04);in
Links betweeng-andN-demography 6,20
fourcases,M.schumannicoloniesthatdiedhadoccupied
Plantsthatdiedweresignificantlysmallerthanplants
justonehostplant.Thussmallplantsaremorelikelyto
thatsurvived.C.nodosathatdiedhad6.7761.45(mean
loseantsthanlargeplants,andthesameistrueforsmall
6SE)domatia,whileC.nodosathatsurvivedhad20.96
devil’sgardens.
1.62domatia(t ¼4.93,P,0.0001).D.hirsutathatdied
19 LargerD.hirsutaplantsproducedmorefruit(in2004,
had35.5610.9domatia,whileD.hirsutathatsurvived
r2¼0.093,P,0.0001).Similarly,antcoloniesthatlived
had 81.7 6 3.82 domatia (t ¼3.96, P¼0.0003). Ant
32
inlargerplantsproducedmorefemalealates(Appendix
colonies that died inhabited significantly smaller plants
thanantcoloniesthatsurvived(Fig.2;C.nodosawithAl. F; C. nodosa with Al. octoarticulatus, r2¼0.223, P ,
octoarticulatus, t ¼3.95, P¼0.0001; C. nodosa with 0.0001;C.nodosawithAz.depilis,r2¼0.316,P¼0.0002;
116
Aztecaspp., t ¼5.44,P,0.0001; D.hirsutawithAz. D. hirsuta with Az. depilis, r2 ¼ 0.311, P , 0.0001),
21
depilis,t ¼8.09,P,0.0001).C.nodosaabandonedby exceptforD.hirsutawithM.schumanni.Thenumberof
54
M. schumanni were significantly smaller than C. nodosa femaleM.schumannialatesinaD.hirsutatreewasnot
that retained M. schumanni (Fig. 2; C. nodosa with M. correlatedwiththenumberofdomatiaonthetree(r2¼
schumanni, t ¼ 2.43, P ¼ 0.03). Similarly, D. hirsuta 0.075, P ¼ 0.23), but instead with the number of D.
19
June2009 ANT–PLANTPOPULATIONBIOLOGY 1603
hirsuta trees in the devil’s garden where the tree was the parasite, Allomerus octoarticulatus. In a previous
located(r2¼0.823,P¼0.0003). study at Madre Selva Biological Station (MSBS),
Ants generally colonized small plants, irrespective of naturally occurring C. nodosa and D. hirsuta without
plant age. With the exception of Az. depilis that ants sustained roughly twice the folivory as C. nodosa
colonized D. hirsuta previously occupied by M. schu- andD.hirsutawithants,andfolivoryincreasedseveral-
manni,plantsboreanaverageof11orfewerdomatiaat fold when ants were experimentally excluded (Fred-
the census before they were colonized by ants and erickson2005).Atadifferentsite,simulatedfolivoryon
belongedto anyageclass(Appendix E). C. nodosa led to increased mortality of domatia
(Edwards et al.2006).
DISCUSSION Studies of other ant–plant interactions have found
g-demography: positivefeedback between thatasanantcolonygrows,itprovidesmoreprotection
mutualistic antcolonies andplants from herbivory to its host plant (Duarte Rocha and
Godoy Bergallo 1992, Gaume et al. 1998, Heil et al.
Ourresultssuggestthatantsaffecttheg-demography
2001). It seems likely that the more ants can lower the
oftheirhostplants.Plantswithmutualisticantsgrewin
a geometric (Cordia nodosa) or autocatalytic (Duroia burden of herbivory, the more resources a plant can
hirsuta) manner. The growth of these plants did not devote to growth, and especially to the production of
taper off, but instead continued unabated or even domatia. Domatia are often a limiting resource for
accelerated as plants grew large, such that the largest phytoecious ants (Fonseca 1993, 1999, Gaume et al.
plantsgrewthemost.Bycontrast,plantswithoutantsor 1998), but as an ant colony grows, it increases the
plantswithparasiticantsreachedsizesbeyondwhichno amount of available resources by promoting plant
further growth was possible; plants not inhabited by growth. For ant species not involved in a mutualism,
mutualisticantstendedtobecomesmalleroncetheyhad this positive relation between colony growth and
morethan24domatiainthecaseofC.nodosawithAl. resource abundance does not hold; though a larger
octoarticulatus,morethanfourdomatiainthecaseofC. colony may be able to obtain more resources, e.g.,
nodosawithoutants,oranydomatiaatallinthecaseof because it has more foragers, larger colonies are also
D. hirsuta without ants. Ants occupied new domatia as more likely to deplete available resources. Because
fastasplantsproducedthem.Thuseithercolonygrowth phytoecious ant colonies use their mutualistic partners
keepspacewithplantgrowth,orthenumberofworkers asresources,andpromotethegrowthoftheirpartners,
occupying each domatium declines as a plant grows. If there is an opportunity for geometric or even autocat-
thefirstalternativeistrue,thenmutualisticantcolonies alytic growth.
also grow geometrically or autocatalytically to match
g-demography: whatlimitspositive feedbackbetween
the growthof C.nodosaandD. hirsuta.
mutualistic antcolonies andplants?
Weexpectedtofindthatplantandantcolonygrowth
eventually tapered off, regardless of the type of ant– Positivefeedbackbetweenmutualistsmusteventually
plantinteraction,becausethisisthewaymostmodular becounteredbyanegativeeffect,orelseplantsandant
organisms grow. Modular organisms like trees and ant colonies would grow to be infinitely large. Several
colonies usually grow quickly when small and more researchers have proposed that the growth of mutual-
slowly as they become large (Oster and Wilson 1978, istic populations may be curbed by diminishing returns
Sebens1987,Gordon1992,1995,Tschinkel1993,Ryan to the mutualism as populations grow (Breton and
and Yoder 1997). All organisms consume resources as Addicott 1992, Holland et al. 2002, Bronstein et al.
they grow, depleting the resources available locally for 2003). Fonseca (1993) suggested that as an ant colony
further growth. Since most modular organisms have inhabiting a plant grows, the protective benefit the
limited mobility, local resource depletion should cause colony provides to the plant may saturate because the
the growth of modular organisms to slow down over contribution each ant worker makes to plant defense
time.Moreover,manymodularorganismsareprevented maybegreaterinasmallthanalargecolony,whilethe
from growing to the point that resources become cost of each worker to the plant remains the same. If
limiting by other factors, such as physiological con- herbivore pressure on a plant remains constant as the
straints that increase with size (e.g., transporting water plantanditsantcolonygrow,eventuallythereshouldbe
to the top of a tall tree; Koch et al. 2004). However, enough ants in the colony to reduce herbivory on the
there was no evidence that growth attenuated at large plant to zero. Any further ant colony growth cannot
sizes forplants inhabitedbymutualistic ant species. benefittheplant,butantcolonygrowthmaystillbeara
Ourresultssuggestthatinthesemutualisms,positive cost. Diminishing returns such as this should produce
feedback between ant colony and plant growth allows negativesizedependence,andplantgrowthshouldslow
the partners to overcome factors that would otherwise as plants andant colonies becomelarge.
limit growth. Herbivory is probably one such factor Becausewedidnotfindanyevidenceofnegativesize
limiting the growth of C. nodosa and D. hirsuta in the dependence,evenatverylargeplantsizes,wedoubtthat
absence of ants. All the ant species we studied protect the ant–plant mutualisms we studied are subject to
the leaves of their host plants against foliovores, even diminishing returns of this type. It is possible that
1604 MEGANE.FREDERICKSONANDDEBORAHM.GORDON Ecology,Vol.90,No.6
herbivorepressuredoesnotremainconstant,butinstead percentile of plant size was: 31 domatia for Al.
increases as plants grow; larger plants may be bigger octoarticulatus-occupied C. nodosa (n ¼ 164), 152
targets and provide more abundant resources for domatia for Azteca spp.-occupied C. nodosa (n ¼ 61),
herbivores. For D. hirsuta with mutualistic ants, which 40 domatia for M. schumanni-occupied C. nodosa (n¼
exhibitedpositivesize-dependentgrowth,thebenefitsof 19),231domatiaforAz.depilis-occupiedD.hirsuta(n¼
mutualism may be subject not to diminishing but to 89), and 383 domatia for M. schumanni-occupied D.
increasing returns as the plant grows. For example, hirsuta(n¼385).
twiceasmanyantworkerscouldbemorethantwiceas
g-demography:the special case
goodatdefendingtheplantagainstherbivores,perhaps
ofAllomerus octoarticulatus
because large colonies are more efficient at organizing
workthansmallcolonies(Gordon1987,1992,Anderson What causes the negative size dependence observed
andRatnieks 1999). for C. nodosa with Al. octoarticulatus? Small Al.
However,protectionagainstherbivoryisnottheonly octoarticulatus-occupied C. nodosa grew, but large Al.
benefitantscanprovidetoplants.Thebestevidencefor octoarticulatus-occupied C. nodosa lost domatia and
autocatalytic growth came from D. hirsuta trees with became smaller. Al. octoarticulatus is a parasite of C.
Myrmelachista schumanni ants. These ants provide at nodosa, but its parasitic behaviors are apparent only at
least two benefits to their host plants; not only do they large plant sizes. Al. octoarticulatus protects C. nodosa
defendtheirhostplantsagainstherbivores,theyalsokill leaves,butnotstems,againstherbivores.Asaresult,the
heterospecificplantstocreatelargepatchesofmostlyD. stemsofAl.octoarticulatus-occupiedC.nodosaaremore
hirsutatreesandtheoccasionalC.nodosa(Frederickson frequentlygirdledbyTrachysomusbeetlesthanarethose
et al. 2005). In so doing, M. schumanni alters the local of C. nodosa with other ants (Yu and Pierce 1998,
environment in many ways, reducing canopy cover Frederickson 2005). Domatia on girdled stems rot and
(Frederickson 2005) and presumably decreasing inter- falloffshortlyafterbeingattacked.Becausethebeetles
specific competition for light and other resources. The tend to girdle larger plants (Yu and Pierce 1998,
positive size dependence of D. hirsuta trees with M. Frederickson 2005), the fact that Al. octoarticulatus
schumanniantsmayberelatedtothesefactors,andnot does not protect plant stems becomes increasingly
to herbivory. Yet another explanation for the growth important toplant growth asplants grow large.
pattern of D. hirsuta with M. schumanni is that M. AnalternateexplanationforthegrowthpatternofAl.
schumanni abandons slow-growing trees. Further re- octoarticulatus-occupied C. nodosa comes from the
search and manipulative experiments are needed to sterilizationbehaviorofAl.octoarticulatus.Al.octoarti-
understand how the myriad benefits and costs of culatusdestroysC.nodosaflowers(YuandPierce1998),
mutualism change with plant and ant colony size, and butonlyoncetheplantislargeenoughtoreproduce.If
how the size dependence of benefits and costs varies plants ‘‘punish’’ ants for destroying flowers by losing
amongpartnersandamonglocalenvironments,includ- domatia,thiscouldalsoexplainwhyAl.octoarticulatus-
ing thatcreated byM.schumanni. occupied C. nodosa grow when small but lose domatia
Ultimately, the reason we did not find evidence of whenlarge.YuandPierce(1998)proposedtheopposite,
negativesizedependenceamongplantswithmutualistic however; they hypothesized that Al. octoarticulatus
ants was that plants lost their ants before resources or sterilizesC.nodosasothattheplantcontinuestoinvest
other constraints limited plant growth. Hence, the loss resources in growth, resources that would otherwise be
ofmutualisticantslimitsthegrowthofC.nodosaandD. diverted to reproduction. It is difficult to reconcile this
hirsuta.Plantsgrewwhenoccupiedbymutualisticants, hypothesis with the observed growth pattern of Al.
butbecamesmallerwhentheant colonydiedor,in the octoarticulatus-occupiedC.nodosa,andfutureworkwill
case of M. schumanni, when ant colony fragments investigatethegrowthconsequencesofAl.octoarticula-
abandoned trees. (Our data set did not allow us to tussterilization of C.nodosa.
determine whether plants became smaller and then lost
Links between g- andN-demography
ants as a result, or plants became smaller after losing
their ants, or both.) Host plants grow much like Becausetheplantsinthissystemlivelongerthantheir
investments earning compound interest. The size of the ant symbionts, they are probably occupied by several
host plant (or investment) at maturity depends on how differentantcoloniesoverthecourseof theirlives.For
fasttheplantgrows(theinterestrate)andhowlongthe the most part, ants colonized plants that were not
plant grows (the term of investment). The term of alreadyoccupiedbyants,andmostoftheseplantswere
investment, or how long the plant grows, is often a small, bearing just a few domatia. However, many of
consequenceofthelifespanofitspartner,orinthecase these small, unoccupied plants were not juveniles; as in
ofM.schumanni,theresidencetimeoftheantsonatree. manymodularorganisms(WatkinsonandWhite1986),
Because both the interest rate and the term of sizeandagearenotwellcorrelatedforC.nodosaandD.
investment vary among ant species, the maximum size hirsuta.Thereareseveralsituationsinwhichplantscan
ofC.nodosaandD.hirsutaishighlyvariable,depending become smaller (e.g., the death or abandonment of an
ontheantspeciesoccupyingtheplant.In2004,the95th antcolony,thepresenceofAl.octoarticulatusonalarge