How to evade a coevolving brood parasite: egg discrimination versus egg variability as host defences.

Abstract

Arms races between avian brood parasites and their hosts often result in parasitic mimicry of host eggs, to evade rejection. Once egg mimicry has evolved, host defences could escalate in two ways: (i) hosts could improve their level of egg discrimination; and (ii) negative frequency-dependent selection could generate increased variation in egg appearance (polymorphism) among individuals. Proficiency in one defence might reduce selection on the other, while a combination of the two should enable successful rejection of parasitic eggs. We compared three highly variable host species of the Afrotropical cuckoo finch Anomalospiza imberbis, using egg rejection experiments and modelling of avian colour and pattern vision. We show that each differed in their level of polymorphism, in the visual cues they used to reject foreign eggs, and in their degree of discrimination. The most polymorphic host had the crudest discrimination, whereas the least polymorphic was most discriminating. The third species, not currently parasitized, was intermediate for both defences. A model simulating parasitic laying and host rejection behaviour based on the field experiments showed that the two host strategies result in approximately the same fitness advantage to hosts. Thus, neither strategy is superior, but rather they reflect alternative potential evolutionary trajectories.

Full text

Page 1

How to evade a coevolving brood parasite:
egg discrimination versus egg variability
as host defences
Claire N. Spottiswoode1,2,* and Martin Stevens1
1Department of Zoology, University of Cambridge, Downing Street, Cambridge CB2 3EJ, UK
2DST/NRF Centre of Excellence at the Percy FitzPatrick Institute, University of Cape Town,
Rondebosch 7701, South Africa
Arms races between avian brood parasites and their hosts often result in parasitic mimicry of host eggs, to
evade rejection. Once egg mimicry has evolved, host defences could escalate in two ways: (i) hosts could
improve their level of egg discrimination; and (ii) negative frequency-dependent selection could generate
increased variation in egg appearance (polymorphism) among individuals. Proficiency in one defence
might reduce selection on the other, while a combination of the two should enable successful rejection
of parasitic eggs. We compared three highly variable host species of the Afrotropical cuckoo finch
Anomalospiza imberbis, using egg rejection experiments and modelling of avian colour and pattern
vision. We show that each differed in their level of polymorphism, in the visual cues they used to reject
foreign eggs, and in their degree of discrimination. The most polymorphic host had the crudest discrimi-
nation, whereas the least polymorphic was most discriminating. The third species, not currently
parasitized, was intermediate for both defences. A model simulating parasitic laying and host rejection
behaviour based on the field experiments showed that the two host strategies result in approximately
the same fitness advantage to hosts. Thus, neither strategy is superior, but rather they reflect alternative
potential evolutionary trajectories.
Keywords: coevolution; egg colour; egg pattern; vision
1. INTRODUCTION
Coevolutionary arms races between parasites and their
hosts can be a significant driving force in evolution, and
in avian brood parasites have led to substantial changes
in phenotypic diversity and behaviour [1]. For example,
many brood parasites have evolved highly mimetic eggs
to evade detection by hosts [2], and manipulative begging
calls to elicit increased parental care [3]. In response,
hosts can defend themselves with a range of counter-
adaptations,includingnest
parasites [4,5], and rejection of foreign eggs [2,6] or
chicks [7]. Much work investigating coevolution at the
egg stage has focused primarily on the parasite’s perspec-
tive, because selection for improved parasitic mimicry is
clearly driven by rejection behaviour of hosts [2]. From
the host’s perspective, successful rejection of parasitic
eggs is a function both of the host’s level of egg discrimi-
nation, and of the difference in appearance between
parasitic and host eggs [6,8]. Thus, when the arms race
reaches a point where parasites have evolved mimetic
eggs, hosts may have two major defence strategies at the
egg stage [9]: first, they can improve their level of egg dis-
crimination, by refining their ability or decision rules used
to distinguish differences in egg appearance between
parasitic eggs and their own. Second, they can shift
their own egg phenotype away from that of the parasite
and other hosts, rendering parasites easier to detect.
defenceagainstbrood
Much evidence for the first mechanism, elevated host
discrimination, comes from the common cuckoo Cuculus
canorus. It has a range of host-specific races (gentes) each
with a different level of host mimicry [2], and those facing
stronger host rejection show superior mimicry [2,10].
The second potential mechanism, shifting host pheno-
types, has been modelled theoretically [11,12]. Hosts
are expected to shift their phenotypes away from the orig-
inal host and mimetic parasite phenotype, and to diversify
egg appearance through frequency-dependent selection
favouring the rare kind, rendering it harder for a parasitic
female to match any one host clutch well enough to be
accepted. Accordingly, scoring of egg appearance has
revealed that host species of the common cuckoo that
show higher levels of egg rejection also have subtly greater
variation in appearance between clutches, and lower vari-
ation within them [13,14]. However, the common cuckoo
and its hosts comprise a young system, with a relatively
recent origin of currently observed gentes [15], implying
comparatively short periods of coevolution between para-
site and hosts. In contrast, in other systems among-clutch
variation is much more phenotypically extreme (e.g. the
hosts of certain African and Asian cuckoos, and the
African cuckoo finch [16]), perhaps reflecting greater
evolutionary age. In this paper, we refer to among-
clutch variation as ‘polymorphism’, following previous
studies of phenotypic diversity as a defence against para-
sites and predators (e.g. [16–18]); here, we are referring
to extreme although continuous levels of intraspecific
variation, rather than classical discontinuous polymorph-
isms (which are also classically known to be genetically
based, cf. polyphenisms).
* Author for correspondence (cns26@cam.ac.uk).
Electronic supplementary material is available at http://dx.doi.org/
10.1098/rspb.2011.0401 or via http://rspb.royalsocietypublishing.org.
Proc. R. Soc. B (2011) 278, 3566–3573
doi:10.1098/rspb.2011.0401
Published online 13 April 2011
Received 22 February 2011
Accepted 22 March 2011
3566
This journal is q 2011 The Royal Society

Page 2

Two escalating defences (discrimination and poly-
morphism)canthus contribute
rejection of parasitic eggs. While these defences are not
necessarily mutually exclusive, proficiency in one should
reduce selection on the other: the more successfully para-
sitic eggs are rejected, the weaker selection will be on
either trait. Highly polymorphic species may be effective
rejectors even with relatively crude egg discrimination,
because parasites can rarely achieve a good match. Simi-
larly, highly discriminating rejectors should receive fewer
potential benefits from evolving among-clutch variation.
It is therefore unlikely that both defences will be simul-
taneously maximized, especially as there may be costs
associated with either strategy (see §4); rather, across
host species, we would expect to find a mixture of both
defences, the sum of which enables successful rejection
of parasitic eggs. While polymorphism and discrimination
are predicted to be inversely related to one another, their
relative contributions should be determined by chance
and species-specific constraints. These predictions can
be investigated by comparing multiple host–parasite
relationships, but this requires similar visual environ-
ments, consistent quantification of egg discrimination
and variable polymorphism levels in multiple hosts.
In this study, we experimentally investigated poly-
morphism and discrimination as two potential defence
strategies in the hosts of an Afrotropical brood parasite,
the cuckoo finch Anomalospiza imberbis. We investigate
different evolutionary trajectories involving three co-
occurring species: two currently exploited hosts and a
third that shows strong egg discrimination behaviour,
but is not currently parasitized at our study site. This
system is particularly appropriate for this investigation
owing to its evolutionary age (the cuckoo finch diverged
from its closest relatives, the brood parasitic Vidua
finches, ca 20 Ma [19]), and owing to the extreme levels
of egg colour and pattern polymorphism shown by hosts
and, correspondingly, by parasites (figure 1). Parasites
lay their eggs haphazardly with respect to host morph,
and incur high degrees of loss through host rejection
[18]. Hosts are all sympatric at our study site in
Zambia, and they build similar nest types in the same
habitat, allowing their discrimination behaviour to be
directly compared without potentially confounding vari-
ation in light conditions. We quantified egg colour and
pattern in terms of current understanding of avian
visual perception, which to our knowledge has not previo-
usly been applied to multiple evolutionary trajectories in
brood parasites. This approach has two major advantages:
first, it allows visual cues to be analysed in terms of how
they relate to selection mediated by avian vision (as
opposed to subjective assessment or by analysing shapes
of reflectance spectra, neither of which necessarily relate
to avian vision). Second, simultaneously quantifying mul-
tiple different aspects of pattern and colour allows us to
identify precisely which visual cues are involved in egg
discrimination and polymorphism, and their relative
importance. These may differ among host species: we
have previously found that one host species, the tawny-
flanked prinia Prinia subflava, uses cues that reveal the
most reliable information about egg identity [18]. Alter-
natively, particular egg features might be intrinsically
better for establishing egg identity [20], or for visual
discrimination tasks in general.
tothe successful
We first quantify overall levels of phenotypic variability
among individuals (polymorphism) with respect to both
egg colour and pattern in each host and its corresponding
parasitic gens. Second, we compare host eggs with those
of their corresponding parasitic gens to assess which
traits differ most consistently between them (i.e. are on
average least mimetic). Third, we use egg rejection exper-
iments combined with visual modelling to identify which
visual traits predict rejection decisions by different hosts
(providing strong evidence that they are used as cues),
and compare these to differences between real host and
parasitic eggs. Fourth, using these experiments, we calcu-
late an overall index of degree of discrimination that is
comparable across species. Finally, for each host, we
simulate the effectiveness of egg rejection in relation to
real parasitic eggs of their corresponding gens, and thus
compare the expected selective consequences of different
defensive strategies(polymorphism
nation). Overall, we predict that hosts with high levels
of polymorphism will be less discriminating in absolute
terms compared to hosts with lower levels of polymorph-
ism, but that either strategy is an effective anti-parasite
defence.
versus discrimi-
2. METHODS
(a) Study system
The cuckoo finch parasitizes a range of grass-dwelling
warblers of the genera Prinia and Cisticola, and several well-
defined host-specific races or ‘gentes’ specialize on different
species (C. N. Spottiswoode & M. Stevens 2011, unpublished
data). At our study site, it most commonly parasitizes the
tawny-flanked prinia (rate of attempted parasitism greater
than19%)andalsoregularlyparasitizesthered-facedcisticola
Cisticola erythrops (rate of attempted parasitism greater than
8%). A third species, the rattling cisticola Cisticola chiniana,
is common at this site but parasitism has never been recorded
here (of 95 recent and 116 historical breeding records,
the latter from 1969 to 1991; C. N. Spottiswoode & J. F. R.
Colebrook-Robjent 2011, unpublished data); nonetheless,
we cannot exclude the possibility of occasional parasitism
attempts eliminated by host rejection before we detected
them. Elsewhere in Africa, it has been recorded as a cuckoo
finch host [21] and we provide evidence that it has been a
host at our study site in the past. For brevity, we refer to all
threewarbler speciesas‘hosts’. Examplesofhostandparasitic
eggs are shown in figure 1.
We carried out fieldwork during January–March 2007–
2009, within a ca 800 ha area on and around Musumanene
Farm (168470S, 268540E) near Choma, southern Zambia.
All hosts build woven nests with a side entrance, stitched
among the broad leaves of small herbaceous shrubs (tawny-
flanked prinias and red-faced cisticolas), or tucked into the
base of a shrub or grass tussock (rattling cisticolas). Hosts
pay strong fitness costs of parasitism since cuckoo finch
hatchlings usually outcompete all host young [22]. Hosts
removed foreign eggs by puncturing then ejecting them.
(b) Field experiments
Detailed methods for field experiments are given in
Spottiswoode & Stevens [18]. Briefly, we used conspecific
eggs from other nests to experimentally parasitize hosts,
then modelled potential predictors of egg rejection (from a
candidate set of eight colour, luminance and pattern traits,
How to evade a coevolving brood parasite
C. N. Spottiswoode & M. Stevens3567
Proc. R. Soc. B (2011)

Page 3

detailed below). The potential for phenotypic mismatch in
our experimental nests differed among host species since
degree of polymorphism varied among them; however, this
does not confound our estimate of the degree of discrimi-
nation (below) since we sought to present all hosts with
more difficult rejection decisions than would be generated
by randomly placing host eggs (see also [18]). We mimicked
cuckoo finch laying behaviour by removing one host egg
when we placed an experimental egg in a nest. A different
host female was used for each experimental trial. All eggs
were photographed in RAW format alongside a 17 per cent
neutral grey card (Kodak) using a Fuji Finepix S7000 digital
camera, and these photographs were used to quantify pattern
(below). Reflectance spectra (for colour and luminance
analysis) of the removed host egg were subsequently
measured indoors (below), and the colour of the removed
egg was taken as representative of the host clutch; this was
justified by intermediate to high intra-class correlation coeffi-
cients (ICC or repeatabilities) of colour channel (CC) values
(defined below) among eggs within a clutch (electronic
supplementary material, table S1). Experimental nests were
visited daily when possible and eggs were considered
accepted if they remained for 3 days in the nest; experimental
eggs that disappeared while the rest of the clutch remained in
the nest were considered rejected.
(c) Quantifying colour and pattern attributes
We measured reflectance spectra using an Ocean Optics
USB2000 spectrophotometer, with a PX-2 pulsed xenon
light source and an R400-7-UV/VIS reflectance probe (all
Ocean Optics Inc., Dunedin, FL, USA), and with reference
to a Spectralon 99% white reflectance standard (Labsphere,
Congleton, UK). A slanted plastic sleeve held each egg at a
constant distance (5 mm) and angle (458) from the probe
tip. Five measurements were taken from the egg’s back-
ground colour (i.e. avoiding overlaid darker patterns) and
the mean analysed. Irradiance (‘ambient light’) within
nests was measured for each species in the field using a
tawny-flanked prinia Prinia subflava
red-faced cisticola Cisticola erythrops
rattling cisticola Cisticola chiniana
respective cuckoo finch gens
Figure 1. Representative host (left) and parasitic (right) eggs showing the range of polymorphism among females. Each egg
came from a different clutch. No parasitic gens is shown for the rattling cisticola because it is not currently parasitized at
our study site. (Note: not to scale.)
3568C. N. Spottiswoode & M. Stevens
How to evade a coevolving brood parasite
Proc. R. Soc. B (2011)

Page 4

cosine-corrected probe (details in [18]). We then calculated
the predicted photon catches of a bird’s single and double
cones [23], using sensitivity data of the blue tit Cyanistes
caeruleus because its visual system is better studied than
other bird species [24]; data are unavailable for our study
species. Repeating the modelling with other higher passerine
bird species’ sensitivity made negligible difference to the
results [18]. Double cone catch data were taken to indicate
luminance, as achromatic information in birds seems to
be provided by the double cones [25]. For colour, plotting
the standardized single cone catch data (using relative cone
catches to remove variations in absolute brightness) in
avian tetrahedral colour space [23] indicated that all host
species and parasite gentes were distributed along the same
single plane in the colour space. This was confirmed by a
principal component analysis (PCA) on a covariance matrix
of these standardized single cone data: two principal com-
ponents (PCs) explained 99.6 per cent of the variation in
egg appearance (of which PC1 corresponded to 73.8%).
We used this PCA as a basis to make an informed decision
about how to encode colour in a biologically meaningful
way based on the principle of opponent CCs, whereby
opposing colours are encoded in antagonistic neural path-
ways (similar to the ‘red-green’ and ‘blue-yellow’ CCs
in human vision; [25]) based on Komdeur et al. [26]. We
treated each CC as a ratio, expressed as
CC1 ¼
LW
ðUV þ SW þ MWÞ=3
and
CC2 ¼ðSW þ MW þ LWÞ=3
UV
;
whereLW,MW,SWandUVindicatelongwave,mediumwave,
shortwave and ultraviolet cone catches, respectively. Much of
our egg colour analysis is based on these two opponent-style
CC calculations (yielding CC1 and CC2).
To quantify pattern, we used an approach recently devel-
oped to quantify camouflage in cuttlefish [27] and mimicry
in the common cuckoo [10]. Briefly, we used a ‘granularity’
method that decomposes calibrated digital images [28] of a
pattern into the relative contribution of markings from differ-
ent spatial scales (frequencies). We first linearized and
calibrated the images of the eggs into reflectance data [28],
extracted the MW channel (approximating to brightness
[10]) and used Fourier transformation and bandpass filtering
to generate a series of seven images capturing information at
different spatial frequencies. Calculating the ‘energy’ (the
sum of the squared pixel values [10,27]) of each of these
seven images allows three measures of pattern to be derived:
(i) the image with the maximum energy corresponds to the
filter size that captured the most information, and reveals
which marking size is most prevalent (‘filter size’, an inverse
measure of marking size); (ii) the proportion of the total
energy across all images contained by the filtered image
with the highest energy (‘proportion energy’) measures how
much the main marking dominates the overall egg pattern,
with higher values indicating that one marking size domi-
nates; and (iii) the total energy (‘total energy’) contained
across all images, which is a measure of contrast between
egg patterns and the background colour [10,27]. Addition-
ally, we obtained two further measures of pattern (outlined
in [10]) by thresholding the images into a binary format
and calculating how much of the eggs were covered with
markings; (iv) ‘proportion coverage’ is the average proportion
of the egg’s surface covered by markings rather than the
background colour; and (v) ‘dispersion’ is the difference in
pattern coverage between the narrow wide regions of the
egg (details in [10] and [18]). All colour and pattern analyses
were undertaken with custom-written programs in MATLAB
(Mathworks).
(d) Quantifying levels of polymorphism
To investigate phenotypic variability and egg discrimination
among host species, we devised a single measure of pheno-
typic distance between any two eggs in the population
sample. This was defined as the Euclidean distance between
two points in 10-dimensional space, defined by four colour
variables (standardized single cone catch values), luminance
information (double cone values) and the five pattern vari-
ables (dispersion, pattern proportion, filter size, proportion
energy and total energy, as defined above). To make the
scale comparable, each of the 10 variables was standardized
by expressing it as a proportion of its maximum value
across all groups. To estimate the overall phenotypic space
occupied by each host or gens, we calculated the distance
between every egg and every other egg in the sample, gener-
ating a matrix of distances. The grand mean across all eggs is
an index of the overall degree of phenotypic variability in the
population (‘multi-dimensional phenotypic space’; MDP
space).
(e) Quantifying levels of discrimination
We used the MDP distance across all colour, luminance and
pattern attributes as a conservative measure of absolute phe-
notypic difference between host and experimental eggs. We
considered all traits because we cannot infer from our egg
rejection experiments which traits might historically have
been under selection. We used the lower 25 per cent quartile
value of rejected eggs as an index of the smallest phenotypic
difference that was detectable to hosts; we used a quartile
rather than a minimum value because the latter could be
unduly influenced by outliers (e.g. hosts that had just seen
an adult parasite), although conclusions would be unchanged
whichever measure was used. High values imply that a
foreign egg must be more different for it to be rejected by
the host; i.e. that the host shows cruder discrimination (or
greater ‘tolerance’ of dissimilar eggs [6]).
(f) Statistical analyses
Sample sizes between different hosts and gentes differed,
which could affect the outcome of the MDP space calcu-
lations. Therefore, for all analyses, we resampled the larger
sample down to the sample size of the smallest: in each case,
we drew a random subset of eggs from the population 999
times and carried out MDP space calculations on each. Stat-
istical analyses were carried out using R [29]. For bivariate
comparisons of egg traits (e.g. between host and parasite),
we used two-sample t-tests for unequal variances (Welch’s
t-test) on ranked data, as recommend by Ruxton [30]. To
model predictors of egg rejection, we compared logistic
regression models (including first-order interactions) by con-
structing generalized linear models with binomial error
structure. Final model selection was complicated by multiple
competing models with similar AIC values, so we used an
information theoretical approach to average a model set
defined by DAIC less than 2 [31], using the R package
How to evade a coevolving brood parasite
C. N. Spottiswoode & M. Stevens3569
Proc. R. Soc. B (2011)

Page 5

MUMIN. We first standardized predictors to have a mean of 0
and a s.d. of 1 such that model-averaged coefficients could be
interpreted as standardized effect sizes [31].
(g) Simulating selection on parasitic eggs
We used our logistic models of conspecific egg rejection
experiments to estimate the proportion of real parasitic
eggs likely to be rejected by each host species. We simulated
a randomly laying cuckoo finch female by pairing randomly
sampled host and cuckoo finch eggs (of the corresponding
gens) 999 times, and for each pair calculating the phenotypic
difference between them for each egg trait. Because for red-
faced cisticolas the available sample of real cuckoo finch eggs
was small (n ¼ 7), to minimize the risk of identical host–
parasite combinations, we chose 50 random host eggs in
random order and compared each with a randomly chosen
parasitic egg (thus ensuring a different comparison each
time), and repeated this process five times, thus generating
250 host–parasite comparisons. We then substituted these
standardized values into the rejection models (reported in
the electronic supplementary material, table S3) to estimate
theprobabilityofeggrejectionforeachsimulatedlayingevent.
3. RESULTS
(a) How polymorphic are different hosts’ and
gentes’ eggs?
In all species, there was relatively little phenotypic vari-
ation withinclutches
material, table S1); however, species differed in the level
of among-clutch variation, i.e. polymorphism. We calcu-
lated MDP space for each host species and their
respective cuckoo finch gens (figure 2). Tawny-flanked
prinias laid the most polymorphic eggs (figure 2a;
mean+s.e. ¼ 0.651+0.007) and red-faced cisticolas
the least (0.381+0.010); rattling cisticolas (not curren-
tly a host) were intermediate (0.461+0.013). These
differences were highly significant for both the raw
(Kruskal–Wallis test, x2
2¼ 237:3, p , 0.001) and the
resampled data (p , 0.001 for 100% of resamples with
n ¼ 55 per group; resampled mean+s.e. for prinias ¼
0.651+0.002 and for red-faced cisticolas ¼ 0.382+
0.002). Correspondingly, cuckoo finch eggs laid in
tawny-flanked prinia nests were more polymorphic
(mean+s.e. ¼ 0.547+0.015) than those laid in red-
facedcisticolanests(0.404+0.038)
t0¼ 24.10, d.f. ¼ 9.04, p ¼ 0.003 on raw data; p , 0.05
for 40.2% of resamples with n ¼ 7 per group; resampled
mean+s.e. for prinia gens ¼ 0.547+0.002).
Within each species, the eight individual aspects of egg
appearance showed low levels of correlation with one
another(r2
values;tawny-flanked
0.0007–0.446,median ¼ 0.019;
range ¼ 0.0001–0.285, median ¼ 0.011; rattling cistico-
las: range ¼ 0.0001–0.277, median ¼ 0.040); the only
pair of traits that explained more than 30 per cent of
the variance in one another were luminance (double
cone catch) and CC1 values for tawny-flanked prinias.
(electronicsupplementary
(figure1b;
prinias:range
red-facedcisticolas:
(b) Which egg traits do parasites mimic best?
For each of the two currently parasitized host species, we
compared the degree of matching in each visual trait
(electronic supplementary material, table S2). A similar
test for pattern traits in tawny-flanked prinias alone has
previously been published in Spottiswoode & Stevens
[18]; it is reported again here for comparison, with a
slightly larger sample size owing to the extraction of
additional pattern data for non-experimental nests. For
tawny-flanked prinias, we found that parasitic eggs were
significantly different from host eggs with respect to:
(i) one axis of colour (CC2: the level of UV stimulation
compared with the other three cone types was higher
in parasitic eggs than hosts’), (ii) dispersion (parasitic
egg markings were more concentrated at one pole of
the egg), (iii) proportion energy (parasitic egg markings
were more dominated by a single marking size), and (iv)
filter size (parasitic egg markings were generally larger).
For red-faced cisticolas, we found that parasitic eggs dif-
fered from host eggs with respect to: (i) the other axis of
colour (CC1: the level of LW cone stimulation compared
with the other three cones was higher in parasitic eggs
than hosts’), (ii) luminance (parasitic eggs were lighter),
(iii) dispersion (parasitic egg markings were more evenly
distributed across the egg), (iv) filter size (parasitic egg
markings were smaller), and (v) total energy (parasitic
egg markings were more contrasting).
(c) Which traits do hosts use in rejecting foreign
eggs?
Of 125 rejection trials in tawny-flanked prinias, 63 eggs
were rejected and 62 accepted. Of 59 trials in red-faced
cisticolas, 26 eggs were accepted and 33 rejected. Of
37 trials in rattling cisticolas, 14 eggs were accepted and
23 rejected. The data for tawny-flanked prinias have pre-
viously been published [18] and are reported again here
for comparison; note, however, that the current paper
uses colour differences between host and experimental
eggs based on ‘opponent-style’ channels (see §2), rather
than explicit perceptual distances as previously. This is
because here we required a measure of each individual
egg’s colour, as well as differences between eggs.
Model-averaged linear models showing predictors of
rejectionarereportedin
material, table S3. Both colour and pattern cues predicted
rejection behaviour in tawny-flanked prinias, with the two
together accounting for 26.7 per cent of the variance in
rejection behaviour. Of this explained variance, colour
accounted for 38 per cent, and three pattern traits
(dispersion, proportion energy and filter size) together
accounted for the rest. In red-faced cisticolas, we were
able to explain a similar proportion (25.7%) of the var-
iance in rejection behaviour. This was accounted for an
interaction between one CC (CC1) and total energy (pat-
tern contrast). We partitioned the data to examine how
this interaction arose (electronic supplementary material,
figure S2), and found that when the difference in total
energy was low, hosts did not use colour to discriminate
between accepted and rejected eggs (suggesting that an
unmeasured additional variable must have accounted for
rejected eggs that differed little in total energy); when
the difference in total energy (i.e. difference in pattern
contrast) was high, hosts rejected eggs with a large differ-
ence in colour but accepted eggs with a small difference in
colour. In rattling cisticolas, 44 per cent of the variance in
rejection behaviour was accounted for by two traits
electronicsupplementary
3570C. N. Spottiswoode & M. Stevens
How to evade a coevolving brood parasite
Proc. R. Soc. B (2011)