Article review and critique

The ecology and evolution of seed predation by Darwin’s finches on
Tribulus cistoides on the Gal�apagos Islands

SOF�IA CARVAJAL-ENDARA ,1,10 ANDREW P. HENDRY ,1,2 NANCY C. EMERY ,3 COREY P. NEU ,4

DIEGO CARMONA ,5 KIYOKO M. GOTANDA ,6 T. JONATHAN DAVIES ,1,7 JAIME A. CHAVES ,8 AND
MARC T. J. JOHNSON 9

1Department of Biology, McGill University, 1205 Avenue Docteur Penfield, Montr�eal, Quebec H3A 1B1 Canada
2Redpath Museum, McGill University, 859 Sherbrooke Street West, Montr�eal, Quebec H3A 0C4 Canada

3Department of Ecology and Evolutionary Biology, University of Colorado Boulder, Boulder, Colorado 80309-0334 USA
4Department of Mechanical Engineering, University of Colorado Boulder, Boulder, Colorado 80309-0427 USA

5Departamento de Ecolog�ıa Tropical, Campus de Ciencias Biol�ogicas y Agropecuarias, Universidad Aut�onoma de Yucat�an, M�erida,
Yucat�an M�exico

6Department of Zoology, University of Cambridge, Cambridge, CB2 3EJ United Kingdom
7Biodiversity Research Centre, Departments of Botany, Forest and Conservation Sciences, University of British Columbia, 2212 Main

Mall, Vancouver, British Columbia V6T 1Z4 Canada
8Colegio de Ciencias Biol�ogicas y Ambientales – Extensi�on Gal�apagos, Universidad San Francisco de Quito, Campus Cumbay�a, Casilla

Postal 17-1200-841, Quito, Ecuador
9Department of Biology, University of Toronto Mississauga, Mississauga, Ontario L5L 1C6 Canada

Citation: Carvajal-Endara, S., A. P. Hendry, N. C. Emery, C. P. Neu, D. Carmona, K. M. Gotanda,
T. J. Davies, J. A. Chaves, and M. T. J. Johnson. 2020. The ecology and evolution of seed predation by
Darwin’s finches on Tribulus cistoides on the Gal�apagos Islands. Ecological Monographs 90(1):e01392. 10.
1002/ecm.1392

Abstract. Predator–prey interactions play a key role in the evolution of species traits
through antagonistic coevolutionary arms races. The evolution of beak morphology in the
Darwin’s finches in response to competition for seed resources is a classic example of evolu-
tion by natural selection. The seeds of Tribulus cistoides are an important food source for
the largest ground finch species (Geospiza fortis, G. magnirostris, and G. conirostris) in dry
months, and the hard spiny morphology of the fruits is a potent agent of selection that
drives contemporary evolutionary change in finch beak morphology. Although the effects of
these interactions on finches are well known, how seed predation affects the ecology and
evolution of the plants is poorly understood. Here we examine whether seed predation by
Darwin’s finches affects the ecology and evolution of T. cistoides. We ask whether the inten-
sity of seed predation and the strength of natural selection by finches on fruit defense traits
vary among populations, islands, years, or with varying finch community composition (i.e.,
the presence/absence of the largest beaked species, which feed on T. cistoides most easily).
We then further test whether T. cistoides fruit defenses have diverged among islands in
response to spatial variation in finch communities. We addressed these questions by examin-
ing seed predation by finches in 30 populations of T. cistoides over 3 yr. Our study reveals
three key results. First, Darwin’s finches strongly influence T. cistoides seed survival,
whereby seed predation varies with differences in finch community composition among
islands and in response to interannual fluctuations in precipitation. Second, finches impose
phenotypic selection on T. cistoides fruit morphology, whereby smaller and harder fruits
with longer or more spines exhibited higher seed survival. Variation in finch community
composition and precipitation also explains variation in phenotypic selection on fruit
defense traits. Third, variation in the number of spines on fruits among islands is consistent
with divergent phenotypic selection imposed by variation in finch community composition
among islands. These results suggest that Darwin’s finches and T. cistoides are experiencing
an ongoing coevolutionary arms race, and that the strength of this coevolution varies in
space and time.

Key words: adaptive divergence; coevolutionary arms race; geographic mosaic; phenotypic selection;
plant defense; trophic interactions.

INTRODUCTION

Antagonistic interactions play a major role in the evo-
lutionary diversification of traits that mediate species
interactions (Thompson 1999, Vamosi 2005, Paterson

Manuscript received 20 December 2018; revised 8 May 2019;
accepted 9 July 2019. Corresponding Editor: Todd M. Palmer

10 E-mail: [email protected]

Article e01392; page 1

Ecological Monographs, 90(1), 2020, e01392
© 2019 by the Ecological Society of America

et al. 2010). Plant–herbivore interactions have long been
used as a model to understand the evolution and ecology
of antagonistic interactions (Ehrlich and Raven 1964,
Fritz and Simms 1992, Agrawal 2011). Plants employ a
wide diversity of mechanical and chemical defense
strategies to avoid the negative effects of herbivores,
including seed predators (Crawley 1983, Carmona et al.
2011). In turn, herbivores and predators use a variety of
strategies to counteract plant defenses, including behav-
ioral, morphological, and physiological offensive traits
(Karban and Agrawal 2002). Selection that favors traits
that better protect plants against herbivores and preda-
tors can lead to contemporary evolutionary changes in
plant defense traits (Agrawal et al. 2012, Z€ust et al.
2012, Didiano et al. 2014). Here, we study the effect of
seed predation by Darwin’s finches on plant ecology,
and its potential role in the evolution of seed defense
traits by natural selection.
The interaction between Darwin’s finches and their food

plants on the Gal�apagos Islands is a famous andwell-studied
example of contemporary evolution (Grant and Grant
2014). Previous studies in agroup of Darwin’s finches known
as ground finches show that evolutionary changes in beak
size and shape are driven by the availability and distribution
of seeds (Lack 1947, Grant 1986, Grant and Grant 1995).
Ground finches are primarily seed predators and poor seed
dispersers; they usually crush the seeds before ingesting them,
and their feces and gut samples rarely contain viable seeds
(Buddenhagen and Jewell 2006, Guerrero and Tye 2009). In
general, ground finches are opportunistic feeders that eat a
large variety of seed species, but when resources are limited
following droughts, finches become dependent on the seeds
of a smaller number of plant species that are often harder
and more difficult to open (Grant and Grant 1995, De Le�on
et al. 2014). The ability to exploit those seeds is largely influ-
enced by beak size and shape (Lack 1947, Grant and Grant
1995, De Le�on et al. 2011). Because seeds are a major part of
their diet, and because ground finches exhibit preferences for
certain seeds, it is anticipated that finches have an important
effect on the ecology and evolution of plants on the
Gal�apagos Islands. However, despite the well-developed liter-
ature on the interactions between Darwin’s finches and
plants (Boag and Grant 1981, Schluter and Grant 1984, Price
1987, Grant and Grant 1999, De Le�on et al. 2014), the eco-
logical and evolutionary consequences of seed predation by
finches on plants remains largely unexplored.
The effects of seed predation by finches on plants on

the Gal�apagos Islands are expected to be mediated by
both climate and the strength of species interactions. Pre-
dation pressure by finches on seeds during periods with
high precipitation might be negligible owing to the high
production of seeds, and the increased availability of other
food resources such as insects (Grant and Boag 1980,
Boag and Grant 1984, Price 1985, Gibbs and Grant
1987). However, during extended droughts, when seed
production is reduced, selective seed predation by finches
(Grant 1986, De Le�on et al. 2014, Grant and Grant
2014) could greatly influence seed survival, plant

distributions, and the evolution of seed defense traits.
Selection imposed by finches on seed defense traits is
expected to play the most important role for plant species
that are commonly exploited by finches. Caltrop (Tribulus
cistoides) is one of the main food sources for some species
of ground finches during dry periods, and it is credited
with driving the evolution of beak morphology in the
Medium Ground Finch (Geospiza fortis) during periods
of drought (Grant and Grant 2006, 2014). The fruits of
T. cistoides possess morphological features thought to
provide defenses against predation, including multiple
long spines and a hard protective tissue (Grant 1981;
Fig. 1). Grant (1981) showed that, within a T. cistoides
population on Daphne Major island, fruits with two
spines were eaten more frequently than fruits with four
spines, suggesting that finches impose selection on T. cis-
toides fruit morphology. However, selection on T. cis-
toides fruits has not been assessed across years or in
populations on other islands, and the association between
fruit morphology and seed survival in response to finch
predation across the archipelago remains unclear.
An additional factor that might influence the effects of

seed predation by finches on plants on the Gal�apagos
Islands is variation in the composition of finch communi-
ties. Ground finches are broadly distributed within the
archipelago and most of the islands harbor several species
that differ in beak size and shape. Among ground finches,
only the Large Ground Finch (G. magnirostris), the Large
Cactus Finch (G. conirostris), and the Medium Ground
Finch (G. fortis) are able to exploit T. cistoides seeds (Grant
1981, Grant and Grant 1982). These species, however, are
not uniformly distributed across the islands. The contempo-
rary faunas of some major islands have one of the large-
beaked G. magnirostris and G. conirostris species and the
small-beaked G. fortis, such as Santa Cruz and Isabela
(Fig. 2), whereas others lack the large-beaked species, such
as Floreana and San Crist�obal. This spatial variation in the
finch community could have large ecological and evolution-
ary consequences because G. magnirostris are superior at
feeding on T. cistoides seeds relative to G. fortis (Grant
1981), which could lead to divergent patterns of predation
and selection imposed on fruit morphology across the
Gal�apagos Islands.
Our study focuses on understanding the effects of seed

predation by Darwin’s finches on the ecology and evolu-
tion of T. cistoides. We asked the following three ques-
tions: (1) Does seed predation by finches vary among
populations, islands, finch community composition, and
years? We expected seed predation to vary among years;
due to variation in annual precipitation, and also in asso-
ciation with finch community composition (small-beaked
finches are expected to eat fewer seeds of T. cistoides dur-
ing wetter conditions). (2) Do finches impose selection on
T. cistoides fruit morphology, and does selection vary
among populations, islands, years, and with finch com-
munity composition? We expected the strength of selec-
tion on fruit morphology to vary over time in
correspondence with precipitation, and spatially among

Article e01392; page 2 SOF�IA CARVAJAL-ENDARA ET AL. Ecological Monographs
Vol. 90, No. 1

islands in association with finch community composition:
large-beaked finch species eat seeds more readily and
likely impose differing selection on fruit morphology
compared to communities with only small-beaked

finches. (3) Does T. cistoides fruit morphology differ
among islands with contrasting finch community compo-
sition (i.e., the presence/absence of large-beaked finches)?
We expected spatial variation in fruit morphology to

FIG. 1. (a) Tribulus cistoides fruits (schizocarps), from left to right: a green immature fruit, a mature dry fruit, and a fruit
attached to a maternal plant. (b) Two sets of dry mericarps, corresponding to two fruits of different plants, showing variation in size
and number of spines. Mericarps in the upper set are larger and have four spines while mericarps in the lower set are smaller and
have only two spines. (c) Opened mericarp to expose seed compartments, one empty compartment and three compartments with
seeds inside. (d) Geospiza fortis (Medium Ground Finch) holding a T. cistoides mericarp. Mericarps showing marks observed
(e) when seeds are eaten by finches, (f) when seeds are eaten by insects, and (g) when seeds germinate. Photo credits: Marc T. J. John-
son (a [left and middle], c, and f), Andrew P. Hendry (b), Kiyoko M. Gotanda (d and e), and Sof�ıa Carvajal-Endara (a [right] and g).

February 2020 DARWIN’S FINCHES AS AGENTS OF SELECTION Article e01392; page 3

reflect spatial variation in finch community composition,
which would be consistent with adaptive responses to
divergent selective pressure. To address these questions,
we examined variation in T. cistoides fruit morphology
and patterns of seed predation in 30 natural populations
across seven islands of the Gal�apagos archipelago over 3
yr, and performed a seed predation experiment in a popu-
lation on one of the islands. Our study is one of the first
to address the potential effect of seed predation by Dar-
win’s finches on the evolution of Gal�apagos plants. We
consider the importance of these results for understand-
ing the potential coevolutionary interactions between
Darwin’s finches and the plants whose seeds they con-
sume.

METHODS

Study site and system

The Gal�apagos archipelago is located in the Pacific
Ocean approximately 1,000 km west of the Ecuadorian
coast in South America, and it comprises 14 major

islands and many small islets (Geist 1996). We restricted
our study to seven islands that vary in finch community
composition (Fig. 2), and that harbor at least one of the
three finch species that consume T. cistoides seeds:
G. fortis, G. conirostris, and G. magnirostris. The diet of
these three finch species varies according to the size and
shape of their beaks, as well as the spatial and temporal
availability of seeds (Schluter and Grant 1984; Grant
and Grant 1999, De Le�on et al. 2014). During dry peri-
ods, especially the droughts that accompany La Ni~na
events, preferred foods are limited and, hence, T. cis-
toides seeds become a main food source for these finch
species (Grant and Grant 2014).
Tribulus cistoides (Zygophyllaceae) is a perennial pros-

trate herb native to subtropical and tropical Africa and
now is widespread in tropical and subtropical arid
coastal habitats around the world (Porter 1972). Broadly
distributed across the Gal�apagos archipelago, it is usu-
ally found in arid lowlands and coastal regions, where it
grows in discrete patches close to roads, trails, and
shorelines (Porter 1971). Tribulus cistoides plants can
flower at any time of year on the Gal�apagos Islands, but
most of its vegetative growth occurs during the wet sea-
son (from January to May), they produce fruits called
schizocarps (Fig. 1a), which contain five individual seg-
ments referred to as mericarps that typically separate
from one another as the fruit dries (Fig. 1b) (Wiggins
and Porter 1971). Each T. cistoides mericarp is a hard
fibrous structure that includes from one to seven seeds
contained within individual compartments (Fig. 1c).
Mericarps typically have four spines (two upper and two
lower sharp protuberances), but the size and position of
spines varies greatly among individual plants, and some
mericarps completely lack some or all spines (Fig. 1b).
The spiny mericarps are also a means of seed dispersal
(Porter 1972); fruits adhere easily to animals, such as the
feet of seabirds (Wiggins and Porter 1971). Ocean cur-
rents and humans are considered important vectors of
long-distance dispersal, whereby fruits travel long dis-
tances by getting attached to shoes and rubber tires
(Holm et al. 1977).
To extract the seeds, finches pick up mericarps from

the ground after they have dropped from the plant. The
finches often hold the mericarp laterally between their
mandibles, and apply pressure by closing their beak,
moving the upper and lower mandibles sideways to each
other, to crack the mericarp wall, sometimes stabilizing
the mericarp against a rock or the ground (Fig. 1d, see
Video S1). The mericarps are very durable and long lived
and this, combined with the very distinct damage left by
finch predation, makes it possible to determine which
mericarps have been depredated even months after a pre-
dation event. Specifically, finches remove the ventral sur-
face of the hard mericarp tissue protecting the seeds,
exposing the empty seed compartments from which
seeds are removed (Fig. 1c), often one compartment at a
time (Video S1) (Grant 1981). Mericarps depredated by
finches (Fig. 1e) are easily distinguished from mericarps

FIG. 2. Map showing the seven islands of the Gal�apagos
archipelago where Tribulus cistoides fruits were sampled. Black
and blue identify the islands where large-beaked ground finches
are present: the Large Ground Finch (Geospiza magnirostris) is
present on Isabela and Santa Cruz and the Large Cactus Finch
(G. conirostris) is found on Espa~nola. Orange identifies the
islands where these large-beaked finches are absent. The Med-
ium Ground Finch (G. fortis) is present in all visited islands
except in Espa~nola.

Article e01392; page 4 SOF�IA CARVAJAL-ENDARA ET AL. Ecological Monographs
Vol. 90, No. 1

consumed by insects, which make smaller circular “drill”
holes (Fig. 1f), and from mericarps from which seeds
have germinated, which are apparent as empty seed com-
partments are still partially enclosed by the mericarp
wall (Fig. 1g), without the rough damage characteristic
of seed predation by finches (Fig. 1e). Other than
finches and insects, no other common predators of
T. cistoides seeds are found on the Gal�apagos Islands.
Unopened mericarps of T. cistoides were found in the
gizzard contents of a Gal�apagos dove (Zenaida galapa-
goensis); however, T. cistoides fruits are not a typical
part of the diet of this species (Grant and Grant 1979).

Population sampling and experimental design

To explore impacts of seed predation by finches, we
sampled nearly 7,000 mericarps from 30 T. cistoides
populations across seven islands of the archipelago over
3 yr (2015–2017). Considering only ground finch species
that consume T. cistoides seeds, finch seed-predator
communities on three of the selected islands (Santa
Cruz, Isabela, and Espa~nola) include large-beaked finch
species (G. magnirostris or G. conirostris), whereas finch
communities on the other four islands (San Crist�obal,
Floreana, Baltra, and Seymour Norte) lack large-beaked
finch species (Fig. 2). The medium-beaked species,
G. fortis, is present on all sampled islands except
Espa~nola (Fig. 2). Sampling was performed between the
months of February and March, corresponding to the
end of the dry season and beginning of the wet season
(Fig. 3a), which is when the finches’ preferred food is
expected to be most scarce and their consumption of
T. cistoides seeds becomes highest. On four of the islands
(Santa Cruz, Isabela, San Crist�obal, and Floreana), we
repeated sampling annually from 2015 to 2017. During
this period, the archipelago experienced strong climatic
variation, including an El Ni~no event that occurred in
2015 (Stramma et al. 2016) and resulted in higher pre-
cipitation relative to the preceding and subsequent years
(Fig. 3b).
The number of T. cistoides populations sampled var-

ied among islands (one to eight populations) due to spa-
tial variation in the abundance of plants, with a
“population” considered to be a discrete patch of T. cis-
toides plants separated by at least 500 m from any other
patch. Information about the populations sampled each
year (island, geographic coordinates) is provided in
Appendix S1: Table S1. From each population, we col-
lected approximately 100 mericarps chosen haphazardly
across the area; we made every effort to select mericarps
“blindly” to avoid biases, so that mericarps represented a
random subset of the morphological traits present in the
population as much as possible. Most mericarps are
expected to be from the previous season, but it is possi-
ble that some mericarps were >1 yr old. A total of 6,391
mericarps were collected across all islands, populations,
and years. For each mericarp, we used digital calipers to
measure mericarp length (mm), width (mm), and the

distance between the tips of the upper spines (upper
spine size, mm) located toward the distal end of the
mericarp, and we noted the presence or absence of lower
spines and the number of seeds removed by finches
(Fig. 4a). To estimate the total number of seeds origi-
nally produced in each mericarp we opened and counted
the number of seeds in 752 mericarps, collected from five
populations on Santa Cruz island in 2015. We evaluated
the relationship between the number of seeds per meri-
carp and mericarp morphology by fitting the following
allometric equation: number of seeds = log(length) +
log(width) + log(length) 9 log(width). We then used this
model to predict the total number of seeds per mericarp
(R2 = 0.48).
To test whether there was variation in fruit morphol-

ogy among individual plants for selection to act upon,
we sampled mericarps from two T. cistoides populations
(AB and EG) on Santa Cruz island during February
2015 (see geographic information in Appendix S1:
Table S1). From each population, we sampled 15 indi-
vidual plants, from each of which we collected four com-
plete (i.e., uneaten) and mature fruits (schizocarps), with
each schizocarp having four to five mericarps. In total,

FIG. 3. Variation in (a) monthly and (b) annual precipita-
tion (mm) from 2014 to 2017 on Santa Cruz island. Precipita-
tion data were obtained from a meteorological station at the
Charles Darwin Research Station (CDRS).

February 2020 DARWIN’S FINCHES AS AGENTS OF SELECTION Article e01392; page 5

we sampled 583 mericarps for measurement of morpho-
logical traits including length, width, upper spine size,
presence/absence of lower spines, and mericarp mass (to
the nearest milligram using a digital balance GEM20;
Smart Weigh, Jintan, China).
To experimentally test whether finches impose selec-

tion on mericarp morphology, we performed a seed pre-
dation experiment during March 2016. First, we
collected 600 mature and intact mericarps from a T. cis-
toides population (EG) located on Santa Cruz island
(see geographic information in Appendix S1: Table S1).
We measured four traits from each mericarp (length,
width, upper spine size, and presence/absence of lower
spines), and gave each mericarp a unique mark with
indelible ink so mericarps could be individually identi-
fied. We also applied an experimental removal of spines
from a haphazard subset of the 400 mericarps by
clipping either one or both of the upper spines, which
allowed us to experimentally test the functional role of
spines in defense. The marked mericarps were then
exposed to natural finch predation on 40 circular plastic
trays (~15 cm in diameter). The trays were placed across
the area where the mericarps were collected, at least
30 cm apart from each other, and were monitored every
three days. The mericarps were recovered after 30 d.
Finally, to evaluate the relationship between mericarp

morphology and hardness, we used 102 mericarps col-
lected in 2017 from three populations on Isabela island
and seven populations on Santa Cruz island
(Appendix S1: Table S1). For each mericarp, we mea-
sured hardness (0–100 value on a Shore D scale; Pam-
push et al. 2011) using a handheld durometer (Asker,
Super Ex, Type D, Kyoto, Japan). As the structure of the
mericarp wall varies over its surface (Fig. 4b), we mea-
sured hardness at six locations on each mericarp (see
detailed information in Appendix S2: Fig. S1). In addi-
tion, on each mericarp, we measured six morphological
traits (length, width, depth, upper spine size, longest
spine length, and spine position; Fig. 4a).

Statistical analyses

All statistical analyses were performed using R v. 3.4.2
(R Development Core Team 2008).

Does seed predation by finches vary among populations,
islands, finch community composition, or years?—We
used logistic linear mixed-effects models with the func-
tion glmer in lme4 v. 1.1-14 package (Bates et al. 2015)
to model the proportion of seed predation per popula-
tion (proportion of mericarps with one or more seeds
removed by finches). This model was fit as follows: pre-
dation per population = year + finch community com-
position + year 9 finch community composition +
island + error. Year, finch community composition, and
their interaction were treated as fixed effects, whereas
island was included as a random effect. Finch commu-
nity composition was categorized as 0 on islands where
large-beaked finch species (G. magnirostris and
G. conirostris) were absent (Floreana, San Crist�obal,
Baltra, and Seymour Norte), and 1 on islands where
large-beaked finch species were present (Isabela, Santa
Cruz, and Espa~nola). To examine the association of pre-
cipitation with seed predation during our study, we fit a
similar model in which we replaced the fixed factor year
with the total annual precipitation (mm) registered dur-
ing the year that preceded each sampling. Precipitation
measurements, obtained from a meteorological station
placed on Santa Cruz island at the Charles Darwin
Research Station (0°44037.600 S, 90°50021.900 W), were
log10-transformed. We also fit the following model where
the response variable was the proportion of seeds
removed per mericarp, and mericarp was the unit of
replication: proportion of seeds removed = year + finch
community composition +
year 9 finch community composition + island + popu-
lation(island) + error. In this analysis, the proportion of
seeds consumed per mericarp was calculated as the ratio
between the number of seeds removed from the mericarp

FIG. 4. (a) Mericarp traits and morphological measurements. (b) Micro-computed tomography (lCT) image showing mericarp
wall variation over its surface.

Article e01392; page 6 SOF�IA CARVAJAL-ENDARA ET AL. Ecological Monographs
Vol. 90, No. 1

and the number of seeds predicted based on the traits of
the mericarp. We included year and finch community
composition as fixed effects, whereas island and popula-
tion were included as nested random effects, with the
parentheses denoting nested factors. Significance of fixed
effects was assessed using a type II Wald’s chi-squared
test, and the significance of random effects was assessed
with likelihood-ratio tests. P values were divided by two
because tests of the significance of random effects are
one-tailed given that variance > 0 (Littell et al. 1996).
Finally, to evaluate more directly the effect of the finch
community on seed predation per year (at the level of
population and mericarps), we fit the logistic mixed-
effects models separately for each year. We performed the
analyses described above including all islands and exclud-
ing data from the three islands that were sampled only in
2016 (Espa~nola, Baltra, and Seymour Norte).

Do finches impose selection on T. cistoides fruit morphol-
ogy and does selection vary among populations, islands,
years, or with finch community composition?—We first
confirmed that most mericarp traits examined (length,
upper spine size, presence/absence of lower spines, and
mass) exhibit substantial variation among individual
plants, with the exception of mericarp width
(Appendix S3: Table S1). Next, we measured phenotypic
selection (sensu Lande and Arnold 1983) on mericarps
sampled from natural populations using logistic mixed-
effects models in the R package lme4 v. 1.1-14 (Bates
et al. 2015) to examine the relationship between T. cis-
toides fitness (seed survival) and fruit morphology (Jan-
zen and Stern 1998). Estimates of T. cistoides seed
survival included two variables: (1) a binary response,
where 0 corresponded to a mericarp that had at least
one seed removed and 1 to a mericarp that had no seeds
removed, and (2) the proportion of seeds that survived
finch predation per mericarp, calculated based on the
estimated number of seeds per mericarp. Each of these
response variables was considered in separate models,
with mericarp traits treated as fixed effects.
Mericarp length, width, and upper spine size were

log10-transformed to improve normality and standard-
ized to a mean of 0 and a standard deviation of 1.
Because of the correlation between mericarp width and
length (r = 0.43), as well as a correlation between length
and upper spine size (r = 0.51), we also performed a
principal component analysis to obtain a principal com-
ponent axis (PC1Size) that captures covariation among
these traits. We also included mericarp lower spines as a
binary response variable, with 1 indicating the presence
of a lower spine and 0 the absence of the trait. In addi-
tion, to test if selection on mericarp traits depended on
year and finch community, we added the interaction
between mericarp traits and these two factors, with
island and population nested within island as random
effects. To enable comparisons among years, we
excluded the data from the three islands that were sam-
pled only in 2016 (Espa~nola, Baltra, and Seymour

Norte) from this analysis. The full model was seed sur-
vival = PC1Size + upper spine size + presence of lower
spines + year + finch community + year 9 PC1Size +
year 9 upper spine size + year 9 presence of lower
spines + finch community 9 PC1Size + finch commu-
nity 9 upper spine size + finch community 9
presence of lower spines + island + population(island)
+ island 9 PC1Size + island 9 upper spine size +
island 9 presence of lower spine + error.
To avoid overparameterization of the models, reduced

statistical power, and elevated type II error, we used the
dredge function in the R package MuMIn v. 1.15. 6. (Bar-
ton 2016) and glmer to compare the models resulting
from all combinations of the fixed effects. This multi-
model averaging approach provides the most comprehen-
sive and powerful approach for finding the best-fitting
models based on the lowest Akaike information criterion
(AIC) values. The importance of each effect was evalu-
ated from the best-fitting model selected for each
response variable, with significance estimated using type
II Wald’s tests in the case of fixed effects, and likelihood-
ratio tests in the case of random effects as described
above. We averaged the subset of models with AIC val-
ues < 2 to estimate average coefficients for each indepen-
dent variable using the function model.avg.
To explore whether finches imposed selection on meri-

carps in the seed predation experiment, we also used
logistic mixed-effects models in which seed survival was
coded as 0 when the recovered mericarp had one or
more seeds removed, and 1 when the mericarp had no
seeds removed. This model had mericarp as the unit of
replication and took the following form: Seed sur-
vival = spine treatment + lower spines + PC1Size + tray.
Spine treatment (a categorical variable coded as 0, 1, or
2 according to the number of large spines remaining on
the mericarp), presence/absence of lower spines, and
PC1Size (i.e., mericarp size) were included as fixed effects,
and tray was included as a random effect. The signifi-
cance of each fixed effect was evaluated using a type II
Wald’s test.
To evaluate the relationship between mericarp mor-

phology and mericarp hardness, we first ran three inde-
pendent principal component analyses to collapse the
hardness measures and morphological measures into
separate, multivariate axes: PC1Global hardness included
hardness measures from all six positions on the meri-
carp, whereas PC1Local hardness included hardness mea-
sures from the three hardest positions on the mericarp
(Appendix S2: Table S1) that we expect to be most
directly involved in protecting seeds from finches preda-
tion. In this analysis, PC1Size was the first principal com-
ponent generated from all six mericarp morphological
traits (length, width, depth, upper spine size, longest
spine, and spine position). We used a mixed-effects
model to evaluate the relationship between hardness
and morphology: PC1Global hardness = PC1Size +
island + population(island), with PC1Size as fixed effect,
and island and population nested within island treated

February 2020 DARWIN’S FINCHES AS AGENTS OF SELECTION Article e01392; page 7

as random effects. We repeated this model using
PC1Local hardness as the response variable. Two additional
models were fit, replacing PC1Size with the six individual
morphological variables in the same model to
simultaneously evaluate the independent contributions
of each morphological trait variable to variation in
PC1Global hardness and PC1Local hardness. In each case, a
model-averaging procedure was used. All data were
standardized within populations to a mean of zero and
standard deviation of one prior to analysis. R2 values
were computed for mixed-effect models using the func-
tion r.squaredGLMM from the R package MuMIn v.
1.15. 6 (Barton 2016); we estimated R2 values associated
with fixed effects (R2 marginal), and R2 values associ-
ated with fixed and random effects (R2 conditional).

Does T. cistoides fruit morphology differ among islands
with contrasting finch community composition?—To evalu-
ate if finch community composition (i.e., presence/ab-
sence of large-beaked finches) influences T. cistoides
fruit morphology, we first fit a linear mixed-effects
model for each of the following traits: width, length,
upper spines size, and the PC1size separately, using the
lmer function from the lme4 v. 1.1-14 package (Bates
et al. 2015). The data were fit to the following model:
trait = finch community composition + year + finch
community composition 9 year + island + population
(island) + error, whereby parentheses indicate nested
terms. Finch community composition, year, and the
interaction between these factors were included as fixed
effects, while island and population nested within island
were modelled as random effects. The models were also
fit for each year separately to test the effect of finch com-
munity on mericarp traits in each year. Last, we ana-
lyzed presence/absence of lower spines as a response
variable; for this analysis, we fit a logistic mixed-effects
model using the function glmer implemented in the lme4
v. 1.1-14 package (Bates et al. 2015).

RESULTS

Variation in seed predation by Darwin’s finches

Seed predation by finches varied among populations,
islands, and years, as well as with finch community com-
position. The proportion of seed predation per population
differed among years (v2 = 208.60, P < 0.01) and islands
(v2 = 74.00, P < 0.01). In 2016, a year following high pre-
cipitation, we found 39% less predation than in 2015 and
45% less predation than in 2017. In addition, higher
annual precipitation, registered on Santa Cruz island, was
associated with reduced seed predation the following year
(v2 = 203.45, P < 0.01). Among islands, mericarps on Isa-
bela had 29% less predation than Santa Cruz and Flore-
ana, and 39% less predation than San Crist�obal. The
effect of finch community composition on the proportion
of seed predation varied among years (finch commu-
nity 9 year: v2 = 40.34, P < 0.01, Table 1 and Fig. 5).

Finch community composition did not influence the pro-
portion of predation in 2015 (Z = �1.20, P = 0.23) or
2017 (Z = �1.40, P = 0.16). In contrast, in 2016, T. cis-
toides experienced 32% higher predation on islands where
the large-beaked finches are present (Z = 3.32,
P < 0.01), compared to islands where they are absent.
These results excluded data from the three islands that
were sampled only in 2016 (Espa~nola, Baltra, and Sey-
mour Norte); yet similar results were obtained when all
islands were included (Appendix S4: Table S1). The pro-
portion of seeds eaten per mericarp also showed varia-
tion among years (v2 = 158.60, P < 0.001), and finch
community composition (v2 = 7.14, P = 0.008; see
Appendix S4: Table S2, Fig. S1). No effect of finch com-
munity composition was seen for the proportion of seeds
eaten per mericarp in 2016 (Z = 0.10, P = 0.809). How-
ever, in 2015 and 2017, the proportion of seeds eaten per
mericarp was 37% and 36% (respectively) lower on
islands where the large-beaked finches were present com-
pared to islands where they are absent. Overall, we found
that, on islands where large-beaked finches were absent,
predation rate per population decreased in the year fol-
lowing high precipitation, and the proportion of seeds
eaten per mericarp increased in drier years.

Phenotypic selection on T. cistoides fruit morphology

Finches imposed phenotypic selection on mericarp mor-
phology (Table 2). In samples from natural populations,
smaller mericarps (PC1Size: v

2 = 21.47, P < 0.001) with
longer upper spines (v2 = 81.20, P < 0.001) were more
likely to escape predation by finches. The presence of
lower spines also reduced predation, but the effect was
marginally nonsignificant (v2 = 3.36, P = 0.067). The pat-
tern of selection on upper spine size and on the presence
of lower spines depended on finch community composi-
tion (finch community 9 upper spine size, v2 = 9.72,
P < 0.002; finch community 9 lower spines, v2 = 6.25,
P = 0.012; Table 2). Longer upper spines and the presence
of lower spines tended to provide greater protection to
mericarps against seed predation on islands where large-
beaked finch species were absent (Fig. 6a).
Selection on mericarp upper spine size and the pres-

ence of lower spines also varied among years
(year 9 upper spine size, v2 = 11.56, P = 0.003;
year 9 lower spines, v2 = 9.83, P = 0.007). Selection for
longer upper spines was stronger in 2016 (Fig. 6b) than
in 2015 and 2017, whereas selection on the presence of
lower spines was strongest in 2015. Model-averaged
coefficients are presented in Appendix S5: Table S1. Sim-
ilar results were obtained when the proportion of seeds
that survived predation per mericarp was used as the
response variable (Appendix S5: Table S2), except that
we found stronger evidence for selection on the presence
of lower spines (v2 = 23.11, P < 0.001) and selection on
mericarp traits did not vary between years (P > 0.5).
In our short-term seed predation experiment in 2016,

we recovered 32 of the 40 trays containing mericarps.

Article e01392; page 8 SOF�IA CARVAJAL-ENDARA ET AL. Ecological Monographs
Vol. 90, No. 1

From these trays, 18.3% of the mericarps showed evi-
dence of predation by finches, 69.2% were uneaten, and
12.5% were not recovered. In our analysis, we included
only the mericarps that were recovered. No relationship
was found between number of upper spines and survival
to finch predation (v2 = 1.26, P = 0.533), but larger
mericarps were more likely to escape predation (PC1Size,
v2 = 5.09, P = 0.024), contrasting with the patterns we
observed in natural populations.

Relationship between variation of fruit morphology and
hardness

Morphological variation in mericarps was associated
with variation in mericarp hardness. Mericarp hardness
varied substantially among locations on the surface of
mericarps (F5, 235 = 15.301, P < 0.001; Appendix S2:
Table S2). We detected a negative relationship between
overall mericarp hardness (PC1Global hardness) and overall
mericarp size (PC1Size; b = �0.437 � 0.102, v2 = 16.876,
P < 0.001, R2conditional = 0.397, R

2
marginal= 0.147, N = 102;

Fig. 7). We detected a similar negative relationship when
only the hardest locations on the mericarp (PC1Local hardness)
were evaluated (b

length
= �0.239 � 0.107, P = 0.02;

bwidth = �0.404 � 0.090, P < 0.0001, R2conditional =
0.335, R2marginal = 0.231, N = 102; Fig. 7). When we
replaced PC1size with the six individual morphological
variables, the best model identified a negative relation-
ship between mericarp length and width and PC1Global
hardness (blength = �0.300 � 0.096, P = 0.002; bwidth =
�0.435 � 0.084, P < 0.0001; R2conditional = 0.458,
R2marginal = 0.334; Fig. 7). Collectively, these analyses
show that smaller T. cistoides mericarps tend to be
harder than larger mericarps.

Effect of finch community composition on fruit
morphology

Mericarp morphology varied substantially among
populations, islands, and years (Fig. 8; Appendix S6:
Table S1). We found differences among islands in meri-
carp length (v2 = 11.9, P < 0.01) and upper spine size
(v2 = 5.08, P < 0.02). For instance, mericarps from Isa-
bela were shorter and had shorter upper spines than did
mericarps from the other islands. Finch community
composition was associated with the presence/absence of
lower spines in mericarps (v2 = 17.98, P < 0.01). The
presence of the large-beaked finch species was associated

TABLE 1. Logistic mixed-effects models analyzing variation in the proportion of mericarps experiencing seed predation per
population among islands and years.

Factor Estimate Z v2 P

a
Fixed effects
Finch community 2.85 0.09
Year 208.60 <0.001
Finch community 9 Year 40.34 <0.001

Random effect
Island 74.00 <0.001

b
Year 2015
Fixed effect

Finch community �0.71 (0.59) �1.20 0.23
Random effect

Island 125.97 <0.001
Year 2016
Fixed effect

Finch community 0.39 (0.12) 3.32 <0.01
Random effect

Island 0.00 0.50
Year 2017
Fixed effect

Finch community �0.47 (0.34) �1.40 0.16
Random effect

Island 32.36 <0.001

Notes: The response variable was the proportion of mericarps that had at least one seed removed by finches in each population
sample (N = 100 in most populations). Finch community composition was considered as a fixed binary factor, with 0 indicating the
absence of the large-beaked finch species Geospiza magnirostris (only G. fortis present) and 1 indicating its presence. Estimate is a
mean with SE in parentheses. In part a, the model included year, finch community composition, and the interaction between those
factors as fixed effects. The effect of island was included as a random effect. In part b, separate models were fit for each year. The v2

and P values of fixed factors were estimated using type II Wald tests and random effects were estimated using likelihood-ratio tests
with one degree of freedom. Effects significant at P < 0.05 are shown in boldface type. These models only include data from islands
sampled in multiple years (see models including all islands in Appendix S4: Table S1).

February 2020 DARWIN’S FINCHES AS AGENTS OF SELECTION Article e01392; page 9

with 67% more mericarps having lower spines. However,
finch community composition was not associated with
differences in mericarp width (v2 = 0.10, P = 0.75),
length (v2 = 0.24, P = 0.62), or upper spine size
(v2 = 0.0, P = 0.44). The effect of finch community
composition on fruit morphology also varied among
years (mericarp width, v2 = 16.56, P < 0.01; length,
v2 = 41.60, P = 0.03; upper spine size, v2 = 53.90,
P < 0.01; and lower spines, v2 = 47.03, P < 0.01). When
we examined the effect in each year, we found a signifi-
cant effect of finch community composition on the pres-
ence of lower spines in 2017 (v2 = 11.13, P < 0.01), but
no effect of finch community composition on the other
traits (mericarp width, length, and upper spine size).
Divergence patterns of PC1size were similar to those
observed for mericarp width and length (see
Appendix S6: Table S1).

DISCUSSION

Seed predation by Darwin’s finches was found to
influence ecological and evolutionary processes associ-
ated with T. cistoides. Several findings address our ini-
tial questions. First, Darwin’s finches were an important
source of mortality for T. cistoides seeds, with the inten-
sity of seed predation varying over time and space in
partial association with finch community composition.
Second, finches imposed phenotypic selection on T. cis-
toides fruit traits whereby seeds within smaller and

FIG. 5. Variation in the proportion of seed predation per population among islands, years, and with contrasting finch commu-
nity composition (represented by different colors). The data correspond to the populations sampled on the four islands (represented
by different shapes) that were visited repeatedly over three years of the study (2015–2017). The mean (dark circles) and the standard
error (dark bar) of the proportion of mericarps with one or more seeds removed by finches in populations sampled from islands
where the large-beaked finch species are absent (0), and present (1).

TABLE 2. Generalized mixed-effects model analyzing pheno-
typic selection on mericarp traits by finches.

Factor v2 P

Fixed effects
Finch community 1.44 0.229
Year 188.70 <0.001
PC1(Size) 21.47 <0.001
Upper spine size 81.20 <0.001
Lower spines 3.36 0.067
Finch community 9 Upper spine size 9.72 0.002
Finch community 9 Lower Spines 6.25 0.012
Finch community 9 Year 45.46 <0.001
Year 9 PC1(Size) 3.65 0.161
Year 9 Upper spine size 11.56 0.003
Year 9 Lower spines 9.83 0.007

Random effect
Island 0.00 0.500
Population 462.80 <0.001
Island 9 PC1(Size) 0.00 0.500
Island 9 Upper spine size 0.00 0.500
Island 9 Lower spines 5.14 0.012

Notes: The response variable seed survival is binary, with 0
indicating mericarps with one or more seed removed by finches
and 1 indicating complete mericarps with no seeds removed.
Finch community composition was considered as a fixed binary
factor, with 0 indicating the absence of the large-beaked finch
species Geospiza magnirostris (only G. fortis present) and 1 indi-
cating its presence. The v2 and P values of fixed factors were
estimated using type II Wald tests and random effects were esti-
mated using likelihood-ratio test with one degree of freedom.
Effects significant at P < 0.05 are in boldface type.

Article e01392; page 10 SOF�IA CARVAJAL-ENDARA ET AL. Ecological Monographs
Vol. 90, No. 1

harder mericarps, and with longer or more numerous
spines, often exhibited higher survival from finch preda-
tion. The details of this finch-associated selection on
defense traits varied over time in accordance with varia-
tion in precipitation and changes in finch community

composition among islands, indicating that geographic
variation in coevolutionary dynamics (sensu Thompson
2005) could be a source of phenotypic diversification in
fruit morphology. Third, one of the traits examined, the
presence of lower spines, exhibited divergence among

FIG. 6. Predicted seed survival probability for each mericarp trait estimated from logistic linear mixed-effects models in relation
to (a) finch community composition and (b) year. The response variable used was binary, with 0 indicating mericarps with one or
more seeds removed by finches, and 1 indicating complete mericarps with no seeds removed. Finch community composition was
considered as a fixed binary factor, with 0 indicating the absence of the large-beaked finch species Geospiza magnirostris (only
G. fortis present) and 1 indicating its presence. Shaded areas show 95% confidence intervals.

February 2020 DARWIN’S FINCHES AS AGENTS OF SELECTION Article e01392; page 11

islands consistent with differences in finch community
composition. Overall, our results are consistent with the
interpretation that finches impose phenotypic selection
on fruit morphological traits, and that these traits act as
plant defenses in an ongoing coevolutionary arms race
between Darwin’s finches and T. cistoides.

Patterns of temporal and spatial variation in seed
predation by Darwin’s finches

We documented temporal and spatial variation in pre-
dation on T. cistoides by finches, with predation rates
being higher in 2015 and 2017 than in 2016. Temporal
variation in seed predation is common in plants (e.g.
Hulme 1994, Kolb et al. 2007), typically being attributed
to temporal variation in biotic and abiotic factors (Hulme
and Benkman 2002). On the Gal�apagos Islands, temporal
variation is strongly influenced by cycles in precipitation,
especially those attributable to the El Ni~no Southern
Oscillation cycle. This variation in precipitation drives
plant productivity in the arid zone (Porter 1979) where
T. cistoides occurs. The abundance of preferred foods of
Darwin’s finches (seeds, fruits, and insects) increases with
higher precipitation (Grant and Boag 1980, Boag and
Grant 1984, Price 1985, Gibbs and Grant 1987), which
influences predation patterns of less-preferred seeds
(Grant 1986, De Le�on et al. 2014, Grant and Grant
2014). Indeed, it has been shown that finches generally
avoid T. cistoides in wet seasons and in particularly wet
years (Boag and Grant 1981, Grant 1981). In accordance
with these previous studies, our results suggest seed preda-
tion of T. cistoides is mediated by variation in precipita-
tion. We found that the lowest predation rate on
T. cistoides occurred in 2016, which followed a year with
high precipitation associated with an El Ni~no (wet) event;

whereas the highest predation rate occurred in 2017, fol-
lowing a year with low precipitation associated with a
La Ni~na (dry) event. However, we caution against over-
interpretation of this finding because our observations
correspond to a period of only 3 yr. In addition, temporal
variation in finch predation on T. cistoides was not uni-
form across islands, suggesting that features particular to
each island might also shape the intensity of seed preda-
tion; however, variation in precipitation within and among
islands could not be included in our analysis, given that
direct measurements of precipitation were only available
from one population on Santa Cruz island. One intriguing
possibility is that higher survival of T. cistoides seeds asso-
ciated with years of high precipitation could be reinforced
by a masting breeding strategy of T. cistoides. Although
masting is a well-known mechanism in many tree species
to increase survival of seeds by satiating predators (Janzen
1971, Silvertown 1980), this strategy has not been evalu-
ated in T. cistoides.
The observed spatial variation in seed predation is

likely also driven by a combination of biotic and
abiotic factors. For example, we expected the highest
rates of T. cistoides seed predation on islands with the
large-beaked finch species G. magnirostris and
G. conirostris. Our results were partially consistent with
this expectation. In 2016, when predation rates on T. cis-
toides were typically low, islands with large-beaked finch
species indeed showed higher predation, but no effect of
community composition was evident in other years. The
smallest finch species that feeds on T. cistoides is G. for-
tis (Grant 1981), but this finch prefers other food
sources that are available following El Ni~no years. By
contrast, large finches have less difficulty feeding on
T. cistoides and they continue to feed on mericarps even
following El Ni~no events. Thus, variation in seed
predation can only be understood by looking at the

FIG. 7. Relationships between mericarp morphology and hardness. (a) Relationship between the global hardness estimate
(PC1Global hardness) and a multivariate measure of mericarp size (PC1Size). Relationship between an estimate of hardness based on
the three hardest locations on the mericarp (PC1Local hardness) and (b) mericarp length and (c) mericarp width. Regression lines are
based on intercepts and slopes estimated from linear mixed models including island and population nested within island as random
effects (see Statistical analyses).

Article e01392; page 12 SOF�IA CARVAJAL-ENDARA ET AL. Ecological Monographs
Vol. 90, No. 1

interaction between finch community composition and
temporal variation in climate.

Selection by Darwin’s finches on fruit traits of
T. cistoides

Darwin’s finches were found to impose phenotypic
selection on T. cistoides fruit defense traits. Mericarps
sampled from natural populations that had lower
spines, longer upper spines, and that were smaller,
were more likely to survive predation. Finches might
prefer larger mericarps because larger mericarps have

more seeds and thus represent a greater reward. Inter-
estingly, we also found that mericarp size was inver-
sely associated with mericarp hardness, which
contributes to defense against ground finches (Boag
and Grant 1981, Price et al. 1984). The ability of
seeds or fruits to escape predation is generally
thought to be greater for larger seeds (or fruits), for
which leverage becomes more difficult, and harder for
seeds (or fruits), which require more bite force to
crack open (Abbott et al. 1977). However, we showed
that smaller mericarps were harder than larger meri-
carps, which suggests that small mericarps could be as

FIG. 8. Variation in Tribulus cistoides mericarp morphology. Means (circle) and standard error (bars) of each mericarp trait for
the four islands that were sampled from 2015 to 2017: Floreana (Flo), San Crist�obal (S.Cri), Isabela (Isa), and Santa Cruz (S.Cru).
Islands where large-beaked finch species are absent (only G. fortis present) are indicated in orange and islands where this species
(i.e., Geospiza magnirostris) is present are indicated in black.

February 2020 DARWIN’S FINCHES AS AGENTS OF SELECTION Article e01392; page 13

(or perhaps more) difficult to open than large meri-
carps. Since finches commonly use specialized twisting
motions to open mericarps (Grant 1981; Video S1),
instead of just a direct biting effort, smaller mericarps
also could require more precision and handling ability.
The strength of selection varied over time and space,

as has been seen in other systems (Thompson 2005,
Siepielski et al. 2009, 2013, Bell 2010). As with the
observed temporal variation in seed predation, variation
in selection on T. cistoides appeared to follow climatic
cycles, although we do not yet know the causal links
between precipitation and specific forms of selection.
Furthermore, selection for longer upper spines and the
presence of lower spines was stronger on islands where
the large-beaked finch species were absent, perhaps
because the largest beaked finch species (i.e.,
G. conirostris and G. magnirostris) are less deterred by
T. cistoides defense traits. Stated in another way, the lar-
gest species might have little difficulty opening even the
most strongly defended T. cistoides mericarps. Once
again, it is the interaction between finch community com-
position and climate that appears to determine
spatial and temporal variation in finch–T. cistoides inter-
actions.
Results from the short-term seed predation experi-

ment did not match our observations from natural pop-
ulations. In the experiment, smaller mericarps were more
likely to be preyed on and no association was evident
between the presence of spines and mericarp predation.

These divergent results are perhaps not surprising since
the experiment was conducted in only a single location
and over a relatively short period of time (30 d), whereas
the observational data capture data from many popula-
tions over multiple years. Our experimental results,
therefore, further emphasize the conditional nature of
finch seed predation, and how selection varies through
time.
Evidence for phenotypic selection by Darwin’s finches

on T. cistoides fruits suggests a potential ongoing coevo-
lutionary arms race between Darwin’s finches and
T. cistoides. Our results thus add to previous studies in
other systems showing that seed predators impose selec-
tion on fruit morphology (e.g., Coffey et al. 1999,
G�omez 2004); however, selection on T. cistoides fruit
morphology has cascading implications in the
Gal�apagos Islands (Fig. 9). Tribulus cistoides mericarps
impose selection on the size and shape of the beak of
G. fortis (Boag and Grant 1981, Boag and Grant 1984,
Grant and Grant 1999), which drives episodic bouts of
evolutionary change (Grant and Grant 2002, 2006).
During dry periods, when small and soft seeds are
scarce, larger beaked birds of G. fortis that are able to
crack T. cistoides mericarps are favored when G. mag-
nirostris are absent. However, when G. magnirostris are
present, they compete with G. fortis for T. cistoides
fruits, and cause an adaptive shift in G. fortis toward
smaller beaks (Grant and Grant 2006). Therefore, the
interaction between G. fortis and T. cistoides seems to be

FIG. 9. Ecological and evolutionary processes influencing interactions between Tribulus cistoides and Darwin’s finches. (1) Dry
periods reduce seed diversity and abundance; (2) predation increases the year following La Ni~na years (low precipitation); (3) pres-
ence of large-beaked finches increases seed predation the year following El Ni~no years (high precipitation); (4) large-beaked finches
compete with Geospiza fortis; (5) decreased seed diversity/abundance leads to greater predation on T. cistoides; (6) decreased seed
diversity/abundance reduces G. fortis population size; (7) decreased seed diversity/abundance selects for larger beaks of G. fortis
when large-beaked bird species are absent, and smaller beaks when they are present; (8) selection for longer upper spines increases
following El Ni~no years, whereas selection for the presence of lower spines and decreased size increase following La Ni~na years; (9)
absence of large-beaked finch species leads to stronger selection for longer upper spines and for the presence of lower spines; (10)
seed predation selects for longer upper spines and smaller mericarps (11); hypothesized coevolutionary arms race between
T. cistoides and G. fortis. Interactions might also operate between the evolutionary processes and population dynamics within spe-
cies (arrows not shown).

Article e01392; page 14 SOF�IA CARVAJAL-ENDARA ET AL. Ecological Monographs
Vol. 90, No. 1

driven by reciprocal evolutionary changes that are medi-
ated by temporal variation in La Ni~na/El Ni~no precipi-
tation cycles and spatial (or temporal) variation in finch
community composition. To further test this hypothesis,
future research should focus on testing whether an adap-
tive evolutionary response of T. cistoides fruits to finch
predation modifies selection patterns imposed on
finches’ beaks and affects their evolutionary trajectory.

Relationship between finch community and T. cistoides
fruit morphology

Spatial variation in natural selection imposed by Dar-
win’s finches has likely already caused adaptive diver-
gence among island populations of T. cistoides. For
example, the absence of lower spines in mericarps was
associated with islands where the large-beaked finch spe-
cies were absent. However, other traits examined (meri-
carp length, width, and upper spine size) were not
consistently associated with finch community composi-
tion. The mismatch between selection pressure and pat-
terns of variation of defense traits might have several
causes. For instance, some defense traits might have low
heritability, or opposing selective pressures, such as from
dispersal and germination, which could mask the effects
of seed predation on evolution (Primack 1987, Alc�an-
tara and Rey 2003, Agrawal et al. 2013). In addition,
fluctuating selection in space and time, as in the case of
mericarp defense traits, coupled with gene flow, popula-
tion bottlenecks, and founder events could constrain the
translation of selection effect into morphological
change, as predicted by the geographic mosaic of coevo-
lution (Thompson 2005). Finally, it has been suggested
that T. cistoides was introduced into the archipelago by
humans (Porter 1967, Grant 1981), sometime after 1535,
when the islands were discovered by Spanish explorers
(Grant and Grant 2014). If true, it is possible that T. cis-
toides has not yet had sufficient time, in combination
with the above factors (e.g., founder effects, gene flow,
opposing selection forces), to locally adapt to finch com-
munity composition. However, the history of T. cistoides
on the archipelago remains unresolved.
The evolutionary response of fruit morphology in

T. cistoides to selection imposed by ground finches
depends on several factors for which we still lack
detailed information, such as heritability of fruit traits.
Gal�apagos National Park restrictions prevented us from
conducting common garden quantitative genetics experi-
ments on T. cistoides; nonetheless, we detected variation
among individual plants for almost all measured meri-
carp traits. These results are consistent with the expecta-
tion that variation in these traits is at least partially
controlled by genetic variation. Future common garden
experiments and genomic analyses would add to our
understanding of the evolution of morphological
defenses in T. cistoides. Key questions that could be
addressed in future work include: What is the genetic
structure and demographic history of T. cistoides

populations across the archipelago? Do trade-offs exist
between natural selection imposed by finches and other
potential drivers of selection on T. cistoides fruit mor-
phology, such as dispersal, germination, and establish-
ment? And, finally, is T. cistoides native to the islands or
was this plant species introduced recently to the islands,
possibly by humans?

CONCLUDING REMARKS

We report evidence that Darwin’s finches select T. cis-
toides fruits based on defense traits, and that the varia-
tion in selection patterns can be explained, in part, by
finch community structure and variation in climate. Pre-
vious work has suggested that predation on T. cistoides
mericarps is an important agent of natural selection on
finch beaks (Boag and Grant 1981, 1984, Grant and
Grant 2006). We here suggest that a reciprocal process
of natural selection by finches on mericarp morphology
is also likely and, hence, that finches and T. cistoides are
coevolving in an arms race. To inhibit finch predation,
T. cistoides invests in physical defense structures, such as
spines and mericarp hardness. In turn, the higher levels
of defense in mericarps select for finches that are able to
remove the seeds more efficiently. The specific process of
reciprocal natural selection that leads to reciprocal adap-
tation has yet to be directly documented, but the individ-
ual component interactions make this scenario highly
likely. In addition, considerable temporal variation in
selection on mericarp defenses as well as finch beak mor-
phology (Grant and Grant 2002), indicates that climatic
conditions and spatial variation in finch communities
mediate the ecological strength and evolutionary out-
comes of this finch–plant interaction.
Our work expands the understanding of ecological and

evolutionary interactions between Darwin’s finches and
the plants whose seeds they eat. A long history of research,
which includes Charles Darwin (1859), David Lack
(1947), Grant and Grant (2014), and many other past and
present researchers, has built a foundation for understand-
ing the interplay between ecology and evolution from
studying Darwin’s finches. We hope that our study sup-
ports and inspires new avenues of research into these inter-
actions.

ACKNOWLEDGMENTS

The study was performed with the logistical support of the
Gal�apagos National Park (GNP) and Charles Darwin Research
Station. The research permits to conduct the study were pro-
vided by PNG (PC-29-14, PC-29-15, PC-42-16). Research fund-
ing was kindly provided by SENESCYT-Ecuador, NEO
program McGill-STRI, Department of Biology McGill, and
Delise Alison Award from Redpath Museum to S. Carvajal-
Endara; Natural Sciences and Engineering Research Council of
Canada (NSERC) grants to A. P. Hendry and M. T. J. Johnson;
NSF DEB-1553053 to N. C. Emery; CONACyT to D. Car-
mona; NSERC Banting Postdoctoral Fellowship, Le Fonds
Qu�eb�ecois de la Recherche sur la Nature et les Technologies
postdoctoral research fellowship, Vineberg Fellowship McGill

February 2020 DARWIN’S FINCHES AS AGENTS OF SELECTION Article e01392; page 15

University, Clare Hall Whitehead Fund, Christ’s College
Gal�apagos Islands Visiting Scholarship Scheme, Phyllis and
Eileen Gibbs Travelling Research Fellowship from Newnham
College, and a British Ornithologists’ Union Research Grant to
K. M. Gotanda; GAIAS-USFQ Grant to J. A. Chaves. Also,
we thank Dieta Hanson for her assistance while performing a
pilot seed predation experiments in 2014, and to Peter and
Rosemary Grant for inspiring this work and patiently answering
our many questions.

LITERATURE CITED

Abbott, I., L. K. Abbott, and P. R. Grant. 1977. Comparative
ecology of Gal�apagos ground finches (Geospiza Gould): eval-
uation of the importance of floristic diversity and interspeci-
fic competition. Ecological Monographs 47:151–184.

Agrawal, A. A. 2011. Current trends in the evolutionary ecol-
ogy of plant defence. Functional Ecology 25:420–432.

Agrawal, A. A., A. P. Hastings, M. T. J. Johnson, J. L. Maron,
and J.-P. Salminen. 2012. Insect herbivores drive real-time
ecological and evolutionary change in plant populations.
Science 338:113–116.

Agrawal, A. A., M. T. J. Johnson, A. P. Hastings, and J. L.
Maron. 2013. A field experiment demonstrating plant life-his-
tory evolution and its eco-evolutionary feedback to seed
predator populations. American Naturalist 181:S35–S45.

Alc�antara, J. M., and P. J. Rey. 2003. Conflicting selection
pressures on seed size: evolutionary ecology of fruit size in a
bird-dispersed tree, Olea europaea. Journal of Evolutionary
Biology 16:1168–1176.

Barton, K. 2016. MuMIn: Multi-model inference, R package ver-
sion 1.15. 6. https://CRAN.R-project.org/package=MuMIn

Bates, D., M. Maechler, B. Bolker, and S. C. Walker. 2015. Fit-
ting linear mixed-effects models using lme4. Journal of Statis-
tical Software 67:1–48.

Bell, G. 2010. Fluctuating selection: the perpetual renewal of
adaptation in variable environments. Philosophical Transac-
tions of the Royal Society B 365:87–97.

Boag, P. T., and P. R. Grant. 1981. Intense natural selection in a
population of Darwin’s finches (Geospizinae) in the
Gal�apagos. Science 214:82–85.

Boag, P. T., and P. R. Grant. 1984. Darwin’s finches (Geospiza)
on Isla Daphne Major, Gal�apagos: breeding and feeding ecol-
ogy in a climatically variable environment. Ecological Mono-
graphs 54:463–489.

Buddenhagen, C., and K. J. Jewell. 2006. Invasive plant seed
viability after processing by some endemic Gal�apagos birds.
Ornitolog�ıa Neotropical 17:73–80.

Carmona, D., M. J. Lajeunesse, and M. T. J. Johnson. 2011.
Plant traits that predict resistance to herbivores. Functional
Ecology 25:358–367.

Coffey, K., C. W. Benkman, and B. G. Milligan. 1999. The
adaptive significance of spines on pine cones. Ecology
80:1221–1229.

Crawley, M. J. 1983. Herbivory: the dynamics of animal-plant
interactions. University of California Press, Berkeley, Califor-
nia, USA.

Darwin, C. R. 1859. On the origin of species by means of natu-
ral selection, or the preservation of favoured races in the
struggle for life. John Murray, London, UK.

De Le�on, L. F., J. Raeymaekers, E. Bermingham, J. Podos, A.
Herrel, and A. P. Hendry. 2011. Exploring possible human
influences on the evolution of Darwin’s finches. Evolution
65:2258–2272.

De Le�on, L. F., J. Podos, T. Gardezi, A. Herrel, and A. P. Hen-
dry. 2014. Darwin’s finches and their diet niches: the

sympatric coexistence of imperfect generalists. Journal of
Evolutionary Biology 27:1093–1104.

Didiano, T. J., N. E. Turley, G. Everwand, H. Schaefer, M. J.
Crawley, and M. T. J. Johnson. 2014. Experimental test of
plant defence evolution in four species using long-term rabbit
exclosures. Journal of Ecology 102:584–594.

Ehrlich, P. R., and P. H. Raven. 1964. Butterflies and plants: a
study in coevolution. Evolution 18:586–608.

Fritz, R. S., and E. L. Simms. 1992. Plant resistance to herbi-
vores and pathogens: ecology, evolution and genetics. Univer-
sity of Chicago Press, Chicago, Illinois, USA.

Geist, D. 1996. On the emergence and submergence of the
Gal�apagos Islands, Ecuador. Noticias de Gal�apagos 56:5.

Gibbs, H. L., and P. R. Grant. 1987. Ecological consequences
of an exceptionally strong El Ni~no event on Darwin’s finches.
Ecology 68:1735–1746.

G�omez, J. M. 2004. Bigger is not always better: conflicting selec-
tive pressures on seed size in Quercus ilex. Evolution 58:71–80.

Grant, P. R. 1981. The feeding of Darwin’s finches on Tribulus
cistoides (L.) seeds. Animal Behaviour 29:785–793.

Grant, P. R. 1986. Ecology and evolution of Darwin’s finches.
Princeton University Press, Princeton, New Jersey, USA.

Grant, P. R. 1999. Ecology and evolution of Darwin’s finches.
Second edition. Princeton University Press, Princeton, New
Jersey, USA.

Grant, P. R., and P. T. Boag. 1980. Rainfall on the Galapagos
and the demography of Darwin’s finches. Auk 97:227–244.

Grant, P. R., and K. T. Grant. 1979. The breeding and feeding
of the Gal�apagos dove, Zenaida galapagoensis. Condor
81:397–403.

Grant, B. R., and P. R. Grant. 1982. Niche shifts and competi-
tion in Darwin’s finches: Geospiza conirostris and congeners.
Evolution 36:637–657.

Grant, P. R., and B. R. Grant. 1995. Predicting microevolution-
ary responses to directional selection on heritable variation.
Evolution 49:241–251.

Grant, P. R., and B. R. Grant. 2002. Unpredictable evolu-
tion in a 30-year study of Darwin’s finches. Science
296:707–711.

Grant, P. R., and B. R. Grant. 2006. Evolution of character dis-
placement in Darwin’s finches. Science 313:224–226.

Grant, P. R., and B. R. Grant. 2014. 40 Years of evolution: Dar-
win’s finches on Daphne major island. Princeton University
Press, Princeton, New Jersey, USA.

Guerrero, A. M., and A. Tye. 2009. Darwin’s finches as seed
predators and dispersers. Wilson Journal of Ornithology
121:752–764.

Holm, L. G., D. L. Plucknett, J. V. Pancho, and J. P. Herberger.
1977. The World’s Worst Weeds. Distribution and Biology.
University Press of Hawaii, Honolulu, Hawaii, USA.

Hulme, P. E. 1994. Post-dispersal seed predation in grassland:
its magnitude and sources of variation. Journal of Ecology
82:645–652.

Hulme, P. E., and C. W. Benkman. 2002. Granivory. Pages
132–154 in C. Herrera and O. Pellmyr, editors. Plant-animal
interactions: an evolutionary approach. Blackwell Scientific
Publications, New York, New York, USA.

Janzen, D. H. 1971. Seed predation by animals. Annual Review
of Ecology and Systematics 2:465–492.

Janzen, F. J., and H. S. Stern. 1998. Logistic regression for empiri-
cal studies of multivariate selection. Evolution 52:1564–1571.

Karban, R., and A. A. Agrawal. 2002. Herbivore offense.
Annual Review of Ecology and Systematics 33:641–664.

Kolb, A., J. Ehrl�en, and O. Eriksson. 2007. Ecological and evo-
lutionary consequences of spatial and temporal variation in
pre-dispersal seed predation. Perspectives in Plant Ecology,
Evolution and Systematics 9:79–100.

Article e01392; page 16 SOF�IA CARVAJAL-ENDARA ET AL. Ecological Monographs
Vol. 90, No. 1

Lack, D. 1947. Darwin’s finches. Cambridge University Press,
Cambridge, UK.

Lande, R., and S. J. Arnold. 1983. The measurement of selec-
tion on correlated characters. Evolution 37:1210–1226.

Littell, R. C., G. A. Milliken, W. W. Stroup, and R. D. Wolfin-
ger. 1996. SAS system for mixed models. SAS Institute, Cary,
North Carolina, USA.

Pampush, J. D., D. J. Daegling, A. E. Vick, W. S. McGraw, R.
M. Covey, and A. J. Rapoff. 2011. Converting durometer data
into elastic modulus in biological materials. American Jour-
nal of Physical Anthropology 146:650–653.

Paterson, S., et al. 2010. Antagonistic coevolution accelerates
molecular evolution. Nature 464:275.

Porter, D. M. 1967. Another Tribulus adventive in the New
World. Rhodora 69:455–456.

Porter, D. M. 1971. Notes on the floral glands in Tribulus (Zygo-
phyllaceae). Annals of the Missouri Botanical Garden 58:1–5.

Porter, D. M. 1972. The genera of Zygophyllaceae in the south-
eastern United States. Journal of the Arnold Arboretum
53:531–552.

Porter, D. M. 1979. Endemism and evolution in Galapagos
Islands vascular plants. Pages 225–256 in D. Bramwell, editor.
Plants and Islands. Academic Press, London, UK.

Price, T. 1985. Reproductive responses to varying food supply
in a population of Darwin’s finches: clutch size, growth rates
and hatching synchrony. Oecologia 66:411–416.

Price, T. 1987. Diet variation in a population of Darwin’s
finches. Ecology 68:1015–1028.

Price, T. D., P. R. Grant, H. L. Gibbs, and P. T. Boag. 1984.
Recurrent patterns of natural selection in a population of
Darwin’s finches. Nature 309:787–789.

Primack, R. B. 1987. Relationships among flowers, fruits, and
seeds. Annual Review of Ecology and Systematics 18:409–430.

R Core Team. 2017. R: A language and environment for statisti-
cal computing. R Foundation for Statistical Computing,
Vienna, Austria.

Schluter, D., and P. R. Grant. 1984. Determinants of morpho-
logical patterns in communities of Darwin’s finches. Ameri-
can Naturalist 123:175–196.

Siepielski, A. M., J. D. DiBattista, and S. M. Carlson. 2009. It’s
about time: the temporal dynamics of phenotypic selection in
the wild. Ecology Letters 12:1261–1276.

Siepielski, A. M., K. M. Gotanda, M. B. Morrissey, S. E. Dia-
mond, J. D. DiBattista, and S. M. Carlson. 2013. The spatial
patterns of directional phenotypic selection. Ecology Letters
16:1382–1392.

Silvertown, J. W. 1980. The evolutionary ecology of mast seeding
in trees. Biological Journal of the Linnean Society 14:
235–250.

Stramma, L., T. Fischer, D. S. Grundle, G. Krahmann, H. W.
Bange, and C. A. Marandino. 2016. Observed El Ni~no condi-
tions in the eastern tropical Pacific in October 2015. Ocean
Science 12:861–873.

Thompson, J. N. 1999. The evolution of species interactions.
Science 284:2116–2118.

Thompson, J. N. 2005. The geographic mosaic of coevolution.
University of Chicago Press, Chicago, Illinois, USA.

Vamosi, S. M. 2005. On the role of enemies in divergence and
diversification of prey: a review and synthesis. Canadian
Journal of Zoology 83:894–910.

Wiggins, I. L., and D. M. Porter. 1971. Flora of the Gal�apagos
Islands. Stanford University Press, Stanford, California, USA.

Z€ust, T., C. Heichinger, U. Grossniklaus, R. Harrington, D. J.
Kliebenstein, and L. A. Turnbull. 2012. Natural enemies
drive geographic variation in plant defenses. Science 338:
116–119.

SUPPORTING INFORMATION

Additional supporting information may be found online at: http://onlinelibrary.wiley.com/doi/10.1002/ecm.1392/full

DATA AVAILABILITY

Data are available from the Dryad Digital Repository: https://doi.org/10.5061/dryad.qq4715f

February 2020 DARWIN’S FINCHES AS AGENTS OF SELECTION Article e01392; page 17

Copyright of Ecological Monographs is the property of John Wiley & Sons, Inc. and its
content may not be copied or emailed to multiple sites or posted to a listserv without the
copyright holder’s express written permission. However, users may print, download, or email
articles for individual use.