Downloaded from http://rspb.royalsocietypublishing.

org/ on August 30, 2017

Body size affects the evolution of hidden

rspb.royalsocietypublishing.org colour signals in moths
Changku Kang1, Reza Zahiri2 and Thomas N. Sherratt1
1
Department of Biology, Carleton University, Ottawa, Ontario K1S 5B6, Canada
2
Canadian Food Inspection Agency, Ottawa, Ontario K1A 0Y9, Canada
Research
CK, 0000-0003-3707-4989
Cite this article: Kang C, Zahiri R, Sherratt
TN. 2017 Body size affects the evolution of Many cryptic prey have also evolved hidden contrasting colour signals which
are displayed to would-be predators. Given that these hidden contrasting
hidden colour signals in moths. Proc. R. Soc. B
signals may confer additional survival benefits to the prey by startling/
284: 20171287. intimidating predators, it is unclear why they have evolved in some species,
http://dx.doi.org/10.1098/rspb.2017.1287 but not in others. Here, we have conducted a comparative phylogenetic analy-
sis of the evolution of colour traits in the family Erebidae (Lepidoptera), and
found that the hidden contrasting colour signals are more likely to be found
in larger species. To understand why this relationship occurs, we present a
Received: 8 June 2017 general mathematical model, demonstrating that selection for a secondary
Accepted: 21 July 2017 defence such as deimatic display will be stronger in large species when
(i) the primary defence (crypsis) is likely to fail as its body size increases
and/or (ii) the secondary defence is more effective in large prey. To test the
model assumptions, we conducted behavioural experiments using a robotic
moth which revealed that survivorship advantages were higher against wild
Subject Category: birds when the moth has contrasting hindwings and large size. Collectively,
Evolution our results suggest that the evolutionary association between large size and
hidden contrasting signals has been driven by a combination of the need for
Subject Areas: a back-up defence and its efficacy.
evolution, behaviour, taxonomy and
systematics

Keywords: 1. Introduction
deimatic display, secondary defence, Adult Lepidoptera are renowned for their diversity in wing colour patterns,
anti-predator adaptation, Lepidoptera, reflecting a variety of anti-predatory strategies [1–3]. While many species have
predator– prey, startle display evolved concealing colours (crypsis [4] and disruptive patterns [5]), some species
have evolved a chemical/physical defence and associated warning colours
(aposematism [2]) and others have evolved a resemblance to these aposematic
species (mimicry [6]). This functional colour pattern classification is usually
Author for correspondence: based on the appearance of the forewings, but there also exists substantial
Changku Kang colour variation on the body parts that are normally hidden from view, generally
e-mail: changku.kang@gmail.com in the hindwings and/or abdomen [7].
Perhaps the most notable example of interspecific variation in hindwing
colour can be found in the species with seemingly cryptic forewings. Some species
have ‘tricks up their sleeves’, evolving highly contrasting hindwings that have
rarely been explored in terms of their evolution and function. One explanation
for the adaptive significance of contrasting hindwings is that they are displayed
under an imminent predatory threat to frighten predators—a deimatic display
(also called startle display) [7,8]. Empirical evidence suggests that this visual dei-
matic display is effective in deterring attacks by avian predators [9], which begs
the question why deimatic signals have not evolved in more species. One possi-
bility is that secondary defences such as deimatic display are more readily
selected in those prey species where the primary defence (e.g. crypsis) is more
liable to fail. An alternative complementary explanation is that secondary
defences are more likely to arise in those prey species that possess attributes
Electronic supplementary material is available that make the secondary defence particularly effective. Larger bodied insects
tend to be more profitable to avian predators [10] and more likely to be discovered
online at https://dx.doi.org/10.6084/m9.
and attacked by them compared with smaller insects [11,12], and it is also possible
figshare.c.3844768. that deimatic display in a larger species is more effective in deterring attack

& 2017 The Author(s) Published by the Royal Society. All rights reserved.
 Downloaded from http://rspb.royalsocietypublishing.org/ on August 30, 2017

because they may be intrinsically more intimidating. Since and used the average contrast over the three photos as the forew- 2
previous studies have revealed that the evolution of key anti- ing–hindwing contrast value of each species. Next, we classified

rspb.royalsocietypublishing.org
predator adaptations including eyespots [12] and high fidelity each species into high-contrasting and low-contrasting groups
mimicry [13] was more likely in larger species, we have inves- using k-means clustering (k ¼ 2) [20].
tigated the role of body size in explaining the variation in We acknowledge that because we used the images from
putative deimatic signals within a phenotypically diverse the web, we had no control of the photographic conditions,
family of Lepidoptera which includes the underwing (Catocala) which could affect the RGB values of the wings. To reduce
moths, celebrated for their contrasting deimatic signals [7]. this variation, we obtained the photos principally from one
reliable source (The Barcode of Life Data Systems, boldsystem-
s.org). Also, we consider that the variation in photographic
2. Phylogenetic analysis conditions should affect our results less because our contrast
We studied the family Erebidae (Lepidoptera) to examine measurements were conducted within an image rather than

Proc. R. Soc. B 284: 20171287

the relationship between prey size and contrasting fore- and between images. In addition, the errors associated with dif-
hindwing colours. Based on the published phylogenetic trees ferent photographic conditions are expected to influence our
[14,15], 249 terminal taxa were sampled as representatives of results in a non-biased way.
erebid lineages. A dataset of 6 172 base pairs were obtained The contrast between forewings and hindwings provided us
from eight protein-coding genes (cytochrome c oxidase subunit I, with an objective way to classify the presence of hidden contrast-
COI; elongation factor-1a, EF-1a; ribosomal protein S5, RpS5; ing signals on hindwings. To complement the contrast analysis,
carbamoylphosphate synthase domain protein, CAD; cytosolic we performed additional analysis of fore- and hindwing colour-
malate dehydrogenase, MDH; glyceraldehyde-3-phosphate dehydro- ation using human observers by allowing subjective assessment
genase, GAPDH; isocitrate dehydrogenase, IDH; and wingless). of crypsis and conspicuousness of the wing colouration. Here,
The dataset was then analysed using model-based (maximum we categorized each species into either ‘cryptic’, ‘conspicuous’
likelihood) methods. We segregated the dataset into eight sec- (the two groups without contrasting hindwings), or ‘contrast-
tors (partitioned by gene regions). Model-based phylogenetic ing’ (cryptic, but has contrasting hindwings). This human
analysis was performed using ML (Maximum Likelihood) categorization of animal colouration can be useful because,
and a GTR (General Time Reversible) þ G model was selected unlike other image processing of colouration, humans deter-
as the most appropriate model of sequence evolution for each mine the colour properties based on the overall information
gene partition based on the Akaike Information Criterion regarding colour/shape/patterns as other animals do. Human
using FindModel (http://www.hiv.lanl.gov/content/ judgement of colouration has been employed in many other
sequence/findmodel/findmodel.html). ML analyses were con- systems to understand the evolution of animal colour signals
ducted using the default settings on the web-server RAxML III [21,22], and is a practical strategy when direct measurements
BlackBox [16]. ML bootstrap analysis with 1 000 pseudorepli- of colouration for a diverse array of species are challenging
cates [17] was also conducted with RAxML III. The [21]. However, the downside of human categorization is it can
phylogram with support values can be found in electronic be subjective and relies on a few critical assumptions including
supplementary material, figure S7. background (e.g. the assumption that moths’ natural resting
Although deimatic defences may be common in nature, it places are either leaves, tree trunks, or leaf litters [23]). Therefore,
has been described only in a limited number of species. we present the results from the contrast value as main results,
Owing to this lack of behavioural data, information about the but include methods and results of human colour categorization
presence of deimatic display could not be gathered directly in the electronic supplementary material.
for most of the species we surveyed. Instead, we used the mor- Of the 249 species investigated, colour images were
phological traits, more specifically the presence of hidden obtained for 225 species. We used the wingspan of each moth
colour signals on hindwings that contrast to forewings, to as an index of body size. Wingspan data were obtained
infer the presence of deimatic defence. In insect prey, the con- from either field guides or online databases and available
trasting colours of hindwings are usually associated with for 201 species (81% of total; electronic supplementary mate-
deimatic defence [7,18], and no other functions of contrasting rial table S3). We used a mid-range value when wingspan
hindwings are reported in nocturnal moths as far as we know. information was provided as a range.
First, to determine the presence of contrasting hindwings, We employed phylogenetic generalized least square (PGLS)
we obtained photos of each species from reliable online/offline analysis to test whether the contrast between forewings and
databases (electronic supplementary material, table S3). When- hindwings of a species covaried with its wingspan. Prior
ever possible, we included at least three different photos of to employing PGLS, four species were excluded from the
specimens in which both forewings and hindwings were clearly phylogenetic analysis to remove polytomies. We also fitted a
visible, along with one photo of the species at resting posture to chronogram (ultrametricized) to our phylogram by using pena-
ensure that hindwings are normally hidden at rest. Using these lized likelihood in the ‘chronos’ function in the ‘ape’ package to
photos, we used ImageJ (opensource software, NIH, Bethesda, time-scale the tree relative to the genetic changes in the
ML) to extract the mean red, green, and blue (RGB) values of genes chosen for phylogenetic inference [24]. Then we ran
the colour of circumscribed areas of forewings and hindwings. PGLS models assuming both the Brownian motion (BM)
When there existed clear contrasting patterns within wings, and the Ornstein–Uhlenbeck (OU) character evolution model
we measured the colour of the area that occupied the major [25] and present the results of the model with the lower
region of the wings. Otherwise, we measured the overall area AIC. We specified Pagel’s lambda correlation structure in
of wings to get RGB values. We performed this procedure for the model to improve accuracy using the ‘gls’ function in the
all three specimen photos for each species. We then converted ‘nlme’ package [26].
the mean RGB values to Lab colour space and estimated Eucli- We further explored the sequence of trait evolution by
dean colour distance between forewings and hindwings [19], examining transition rates between states to evaluate whether
 Downloaded from http://rspb.royalsocietypublishing.org/ on August 30, 2017

(a) (b) 3

mean wingspan (mm)

rspb.royalsocietypublishing.org
60

50

40

30

ng

ng
sti

sti
tra

tra
n

n
co

co

Proc. R. Soc. B 284: 20171287

n-
no
(c)

non-contrasting 0 140
contrasting wingspan (mm)

Figure 1. Examples of non-contrasting and contrasting species with their phylogenetic relationship and wingspan. (a) Colouration classification and body sizes of
moths within a molecular phylogeny for the family Erebidae. The barplot on the right side depicts the wingspan of each species. (b) The mean (+1 s.e.m.) of
the wingspan of moths in each group without controlling for phylogenetic correlation. (c) Photos of representative species that were categorized as either
non-contrasting (left; Phoberia atomaris), or contrasting (right; Eudocima phalonia). Photo credits (P. atomaris, Huemer P; E. phalonia, CK).

large size might have preceded the evolution of contrasting hindwings (electronic supplementary material and figure S1).
hindwings or vice versa. If large size evolved first, we would These two lines of evidence both indicate that there exists
expect that the transition rate from large/non-contrasting to evolutionary association between large size and contrasting
large/contrasting moths to be higher than that of small/ hidden signals on hindwings. The transition rate from large/
contrasting to large/contrasting moths. On the other hand, if non-contrasting to large/contrasting (3.1) was considerably
contrasting wing patterns evolved first, then we expect the tran- higher than the transition rate from small/contrasting to
sition rate from small/contrasting to large/contrasting moths large/contrasting (6.4  10215). This supports the hypothesis
should be higher. To estimate the transition rates between that being large was a driving force for the evolution of the
states, first, we partitioned wingspan data into two categories, hidden colours on hindwings (see electronic supplementary
small and large, by using k-means clustering (k ¼ 2) [20]. Then material, figure S3 for the full transition rates between states).
we estimated transition rates between the four states (two
sizes  two colours) using the ‘fitDiscrete’ function in the
‘geiger’ package in R [27]. We fitted a custom transition
model, assuming that transitions occurred in a stepwise fashion 4. Mathematical model
in that size and colour strategy cannot evolve simultaneously Arguably the most plausible explanation for the function of
but can evolve one by one. Thus, we pre-set the transition contrasting hidden colours on hindwings is that they are
rates between small/non-contrasting and large/contrasting used for startling predators as a secondary defence [7,18].
and between large/non-contrasting and small/contrasting To help understand the phylogenetic pattern we have
to zero. detected, we developed a simple analytical model of the evol-
ution of a secondary defence, deployed only if the primary
defence fails. The mathematical model (box 1; figure 2)
suggests that if large prey are attacked more frequently
3. Phylogenetic analysis results than smaller prey on account of their higher detectability
Of the species investigated, 21% were classified as contrasting, and/or profitability [10–12], then having a costly secondary
79% as non-contrasting. Without phylogenetic control, the defence is beneficial only when prey size is beyond a certain
species with contrasting hindwings were significantly larger threshold (figure 2 dotted lines). Size dependency in the effec-
than those with non-contrasting hindwings (figure 1b, t-test: tiveness of secondary defence (with conspicuous displays in
t ¼ 23.89, N ¼ 197, p , 0.001). Similar patterns were found larger species more successful in deterring predators) reinforces
even after controlling for phylogeny. PGLS assuming the OU this relationship, pushing the size threshold for evolving
evolutionary model provided a better fit than BM models secondary defence into a larger size (figure 2, dashed lines).
(DAIC ¼ 5.3). Applying the OU model, we found a substan-
tial phylogenetic signal (estimated Pagel’s l ¼ 0.91) and an
association between contrasting hindwings and large size
(coefficient ¼ 4.53 + 2.24, t ¼ 2.03, p¼ 0.04; figure 1a). The 5. Behavioural experiments using a robotic moth
additional analysis using human colour categorization To evaluate our model assumptions, we examined whether
showed consistent patterns in that the species with contrasting survivorship advantage of having contrasting hindwings is
hindwings were larger than those without contrasting greater when prey size is large than small and tested the
 Downloaded from http://rspb.royalsocietypublishing.org/ on August 30, 2017

4
Box 1. Secondary defence evolution and body size.

rspb.royalsocietypublishing.org
Consider a palatable prey species with a single reproductive episode at the end of its life (i.e. it is semelparous). Let pp rep-
resent the probability that the primary defence of the species is successful in deterring an attack following a putative
encounter with a predator, and let ps represent the probability that the secondary defence, deployed if the primary defence
fails, allows the prey of this species to survive an attack. The probability s that an individual of this species survives a single
putative encounter with a predator is therefore:
s ¼ pp þ ð1  pp Þps : ð4:1Þ

Let predators encounter prey at random, so that putative encounters with individual prey are Poisson distributed with
mean and variance l. Under these conditions, the expected probability of an individual prey surviving a series of

Proc. R. Soc. B 284: 20171287

random encounters with predators is:
X 1  l i 
e l i
s ¼ eflðpp 1Þðps 1Þg , ð4:2Þ
i¼0
i!

where e is Euler’s number and i! is the factorial of i. The fitness of this prey in terms of their reproductive success is given by
w ¼ eflðpp 1Þðps 1Þg ðb  tÞ, ð4:3Þ
where b is the mean fecundity of the prey and t is the fecundity cost of carrying the secondary defence. We assume t . 0
because the development, maintenance, and/or deployment of secondary defence is likely to be costly [28,29].
Let us now assume that the primary defence is crypsis and that the probability of avoiding detection and subsequent
attack by a predator per putative encounter is simply pp ¼ ceaz where c is the maximum crypsis, z is body size, and a
(0) governs the rate at which this crypsis decreases with increasing body size (to a lower limit of 0). Likewise, let us
assume that the secondary defence is deimatic defence and the probability of surviving an attack by a predator is
ps ¼ deb=z ,where d is the maximum effectiveness of deimatic display and b (0) governs the rate at which effectiveness
of secondary defence increases from zero as size increases. Substituting for pp and ps in (equation (4.3)) allows us to compare
the fitness of individuals with ( ps . 0, t . 0) and without ( ps ¼ 0, t ¼ 0) a secondary defence, when the effectiveness of the
primary defence and secondary defences are independent of body size (a ¼ 0 and b ¼ 0, respectively), or dependent on body
size (a . 0, b . 0).
This analysis readily shows that when the crypsis of a prey item decreases with increasing body size, then costly sec-
ondary defences will be selected for only in prey above a threshold size (a general result, because a secondary defence
never pays in prey of negligible size because when pp is high s is already near the upper threshold, yet always pays in
very large prey when pp is low). As expected, this threshold size will also depend on how the effectiveness of the deimatic
defence varies with body size. The qualitative conclusions do not change for iteroparous prey (electronic supplementary
material and figure S7).

effect of body size and hindwing colouration on the effective- motors generated a slight mechanical noise during the move-
ness of deimatic display. To manipulate size/colour of prey ment, but the sound level was low and will have affected all
and mimic the behavioural wing display of moths in a stan- treatments in a similar fashion.
dardized way, we developed a remote-controlled robotic The experiment comprised four treatments in a 2  2 ran-
moth model (Robomoth). This Robomoth opened its fore- domized block factorial design. We manipulated (1) the size
wings to reveal the hidden hindwings that were attached of both forewings/hindwings (small and large) and (2) the
underneath. The forewing colour pattern of the moths was colour of hindwings (grey and red; non-contrasting and con-
influenced by the colour pattern of Catocala species (electronic trasting colour, respectively). We tested either four (one block)
supplementary material, figure S4). The wings of the prey or eight different flocks (two blocks) in a day (see below) ensur-
were made out of printed paper. The action of wing display ing equal replication of the four treatments. Robomoth’s
was controlled from a distance of several metres away via forewings were 1  1.5 cm (width  height) for the small size
Bluetooth by one experimenter (KC) in all experiments. The and 3.3  5.4 cm for the large size. The large size was within
wings were displayed for 3 s (set in the controls) and then the range of the wing size of the contrasting species in phylo-
closed. We chose 3 s because this time period largely guaran- genetic analysis (wingspan ranges from 2 to 14 cm), the small
teed one-on-one interactions between a single bird and our wing size was smaller than the minimum observed size of con-
Robomoth. The opening/closing of wing display took less trasting species, but was within the range of the size of cryptic
than 0.1 s, which is comparable to the speed of insect wing dis- species (0.75–12.5 cm). The hindwings were slightly smaller
play measured from the videos of published articles [30,31]. than forewings and attached on the wooden box just under-
Robomoth was placed on top of a wooden box, with all neath the forewings. For hindwing colour, we used red (R ¼
electrical mechanisms hidden inside. A detailed description 255, G ¼ B ¼ 0) for the contrasting colour, which is a common
of Robomoth can be found in the electronic supplementary colour in deimatic signals [7], grey (the mean RGB values of
material and figure S4. The forewings of Robomoth were forewings, R ¼ 103, G ¼ 79, B ¼ 75) for the non-contrasting
grey coloured and contrasting against the wooden box, thus colour. The spectral reflectance curve of each colour is
clearly visible to the birds. We note here that the servo provided in electronic supplementary material, figure S5.
 Downloaded from http://rspb.royalsocietypublishing.org/ on August 30, 2017

1.0 Table 1. The description of each startle score of birds in response to the 5
robotic moth wing display.

rspb.royalsocietypublishing.org
strategy
score description
P
fitness

P+S 0 stayed in the arena and took the mealworm before the
0.5 P + S(size) display was over
secondary 1 flew off to a nearby branch, but landed back on the arena
defence
favoured and took the mealworm before the display was over
0 1 2
2 flew off from the arena and did not take the mealworm
size until the display was over

Proc. R. Soc. B 284: 20171287

Figure 2. Fitness of prey as a function of their body size when larger prey
suffer higher detection costs (a ¼ 1). The solid line (size-dependent primary
defence only, P) indicates the fitness of prey when they rely on crypsis with- available for some insect prey in other groups: some insects
out having any secondary defence (d ¼ 0, b ¼ 0, t ¼ 0); the dotted line display only after some form of physical contact [31,32],
( primary plus size-independent secondary defence, P þ S) indicates the fit- while others display when a predator is simply within a
ness of prey when they have a secondary defence whose effectiveness is close distance [9,33]. As above, our criterion for trigging the
independent of body size (d ¼ 0.5, b ¼ 0, t ¼ 0.2); the dashed line (primary wing display was based on proximity. At the moment of dis-
plus size-dependent secondary defence, P þ S(size)) indicates the fitness of prey play, the distance between the bird and Robomoth inevitably
when the effectiveness of the secondary defence increases as the prey size varied, but in all cases it was within approximately 10 cm. In
increases (d ¼ 0.5, b ¼ 1, t ¼ 0.2). Other parameters values: c ¼ 0.99, all cases, birds’ landings on the arena were to take
l ¼ 2, b ¼ 1. Size is always positive (z . 0). When the fitness of one par- the mealworm. Once the birds took both mealworms on the
ticular defensive strategy exceeds another then it will be favoured (marked by arena, we immediately placed new mealworms and started
arrows, dependent on how the effectiveness of the secondary defence varies the next trial. The trials were continued until either (i) the
with body size). (Online version in colour.) number of birds that had finished the trials was more than
five times the estimated flock size, or (ii) 30 trials had been
reached. We video recorded the arena throughout the testing
For field testing, we introduced Robomoth to birds in the and used the videos for analysing bird behaviours (see
wild. The field testing was conducted in Stony Swamp electronic supplementary material, video S1 for an example).
(N45.29468, W75.83728) and Mer Bleue (N45.40428, W75.55998) Once the testing of a flock was done, we scattered seeds in
in Ottawa, Canada, with free-ranging black-capped chickadees. the vicinity to keep the tested birds behind us and continued
The local chickadees we exploited were accustomed to hand- walking along the trail until we encountered a new flock. In
feeding by humans and they were readily recruited to our this manner, we ensured that the flocks tested in the same
experimental set-up (see electronic supplementary material, day were all different flocks. We never visited the same trail
table S1 for the specific locations where we recruited the birds). twice to avoid pseudo-replication and the trails were at least
First, we walked along a trail until we encountered a 500 m away from each other. Since black-capped chickadee
chickadee flock. Flock size varied; therefore, we estimated flocks show high site-fidelity and limited foraging ranges in
the number of individuals in the flock during testing and winter [32,34] (on average 0.15 km2 of feeding range for a
included it as a covariate in all fitted models (see below). flock and limited overlapping of the ranges between flocks
Once located, we started hand-feeding the chickadees meal- [34]), we assumed that the flocks we have tested were all differ-
worms while we set up a small table near the feeding spot ent flocks. To account for the uncontrollable environmental
and put a wooden board (21  31 cm) on it. Then we stopped variation which may affect the feeding decisions of the birds,
hand-feeding them, and placed two mealworms on the board we treated every four flocks tested in the same day as a
to attract the birds to our set up, replacing them once both block. The order of tested treatment was randomized every
mealworms had been consumed. After several pairs of meal- four flocks. All tests were conducted between 0900–1600 in
worms were consumed, we replaced the wooden board with November and December 2015. In total, we tested 36 different
the wooden box with Robomoth on it. We placed two meal- chickadee flocks (nine replicates per treatment) which were
worms on the box (hereafter, referred to as the ‘arena’): one always treated independently in statistical analysis (one data
near the bottom of the forewings, one on the other side of point per flock).
the arena. The latter mealworm was placed simply to keep To evaluate whether there were differences in startle
attracting birds to the arena. The moth wings were displayed responses of the birds between treatments, we scored the
either (i) just after a bird landed on the moth side of the arena level of startle response of birds following each wing display
or (ii) when a bird hopped toward the moth once it landed on by Robomoth. These levels were designed to be as unambigu-
the other side of the arena. We note here that the exper- ous as possible (table 1 and electronic supplementary material,
imenter was un-blinded about the experimental treatment video S1). To test this assertion, five different volunteers scored
due to the nature of the experiments (i.e. the need to observe a subset of videos (31 bird–moth interactions), and 100% of
the birds’ approaching the Robomoth), but we tried to mini- these scores agreed among the observers, indicating high
mize this bias by strictly following this rule about the display repeatability of the score criteria. Following this confirmation,
timing. Detailed observations on the timing of deimatic dis- one scorer (CK) scored the rest of the videos. First, we extracted
play within the family Erebidae have yet to be made, but the startle score of the initial encounter between Robomoth and
the timing of wing display in the predation sequence is the bird that arrived first from its flock and compared this score
 Downloaded from http://rspb.royalsocietypublishing.org/ on August 30, 2017

(a) (b) 6
2 2

initial startle score

mean startle score

rspb.royalsocietypublishing.org
1 1
N
C
0 0
S/N S/C L/C L/C small large
treatment
(c)
initial survivorship (d)

mean survivorship
1 1

Proc. R. Soc. B 284: 20171287

0 0
small large small large
Figure 3. The startle scores of the birds (a,b) and survivorship of the prey (c,d) in each treatment. (a) The startle scores of the first-arrived bird of a flock in response
to the robotic moth display (median, first/third quartiles, and min/max scores). (b) The mean startle scores of the initial five encounters of a flock (mean +
1 s.e.m.). (c) The survivorship of the prey at the interaction with the first-arrived bird of a flock in response to the robotic moth display (mean + 95% binomial
confidence intervals). (d ) The mean survivorship of the prey among the initial five interactions of a flock (mean + 1 s.e.m.). n ¼ 9 flocks for each treatment for all
plots. S, small, L, large, N, non-contrasting hindwing, C, contrasting hindwing. (Online version in colour.)

between treatments. Because the initial startle score data did scores of birds over the initial five encounters were, however,
not satisfy the assumptions of parametric tests, we used per- significantly higher in larger moths (figure 3b; x21 ¼ 9:38, p ¼
mutation analysis of variance with startle score as a response 0.0021) and when their hindwing colour was contrasting
variable, colour, size, as well as their interaction as explanatory (x21 ¼ 3:94, p ¼ 0.04). We also found a significant interaction
variables, with block as a random factor and flock size as a cov- effect between colour and size (x21 ¼ 4:16, p ¼ 0.04), in that
ariate. Flock size was used as a covariate to control for potential the effect of contrasting signals on bird response behaviour
effects of flock size (e.g. being competitive and bold when there was overall more prominent in small moths than large
were more birds nearby, or being cautious when there were moths (figure 3b). This interaction almost certainly arose
fewer birds). because the classified responses of birds to wing display
To complement this analysis of the initial reaction of birds, was close to maximum in large moths even when their
we calculated the average startle scores of birds within flocks hindwings were non-contrasting, so there was little room
over the first five encounters with a displaying moth. To test for improvement.
whether these average scores varied among treatments, we Perhaps a more relevant metric from the perspective of
fitted a linear mixed model with size and colour as well as whether any putative anti-predator trait will be selected for,
their interaction as explanatory variables and block as a is the effect of the display on whether the mealworm was
random factor with flock size as a covariate. We present the eaten or not (survivorship). At the first encounter with a bird
results of this analysis to support our initial Robomoth–bird of a flock, the mealworm survivorship was higher when the
interaction analysis because this includes five times more prey was large compared with small (figure 3c; x21 ¼ 4:55,
moth–bird interactions. Clearly, a bird’s response after its p ¼ 0.03) and had contrasting hindwings compared with
first encounter with the robotic moth could be affected by non-contrasting (x21 ¼ 4:55, p ¼ 0.03). We also found strong
direct and indirect (watching others) learning [35]. evidence for a significant interaction, in that the effect of con-
To evaluate the implications of bird behaviour for prey trasting hindwing in enhancing survival was only apparent
survival, we translated each encounter between birds and for large prey, but not in small prey (x21 ¼ 7:13, p ¼ 0.007). Like-
Robomoth into a binary variable, based on whether the meal- wise, when we analysed the mean survivorship of the
worm adjacent to the moth survived until the end of the 3 s mealworms over the initial five encounters, we found that
display (startle score ¼ 2) or not (startle score less than 2). large prey survived better than small prey (figure 3d;
We then analysed (i) the survivorship of prey at the initial x21 ¼ 6:48, p ¼ 0.01). However, in this instance we found no
encounter of a flock and (ii) mean survivorship of the initial significant effect of hindwing colour (x21 ¼ 2:46, p ¼ 0.12), or
five encounters by fitting a generalized linear mixed model an interaction effect (x21 ¼ 3:63, p ¼ 0.06).
containing the same explanatory variables as above.

7. Discussion
6. Behavioural experiment results Our phylogenetic results demonstrate for the first time that
From the perspective of bird behaviour, we found no effect of body size has mediated the evolution of hidden colour signals
moth size (figure 3a; p ¼ 0.31), colour ( p ¼ 0.44), nor their in moths, with larger species showing a greater tendency to
interaction ( p ¼ 0.06) on the mean startle score of the first evolve hidden contrasting hindwings. Our model suggests
bird of a flock to encounter the moth. The average startle that contrasting hindwings likely evolved predominantly in
 Downloaded from http://rspb.royalsocietypublishing.org/ on August 30, 2017

large cryptic moths because they are more readily detected and conspicuous colours such as blue, yellow, and red (all of which 7
attacked by natural predators, generating strong selection for were categorized as conspicuous regardless of the UV colour),

rspb.royalsocietypublishing.org
back-up defences that operate when crypsis fails. However, but not on dull colours such as black and brown [39]. Therefore,
we have also found evidence that, when displayed at a close it is likely that a conspicuous colour in the visible wavelength
distance to the approaching predator, the survivorship advan- would also be categorized as conspicuous even if the UV
tage for deimatic signals is greater in large moths than small range was considered. Finally, deimatic display needs to be
moths. Both the birds’ startle scores and prey survivorship effective against multiple predators including UV-insensitive
results aligned with each other in that deimatic displays were birds, predatory invertebrates, and mammals, so having private
effective only when the prey had both large size and contrast- deimatic signals only in the UV spectrum is unlikely.
ing hindwings. Nevertheless, the analysis of the initial five To conclude, the function and evolution of animal colour-
bird–Robomoth encounters showed that small prey may also ation that is ‘hidden’ from view has been overlooked for a
gain survival benefits by having a deimatic signal. This is per- long time in the field of animal colouration research [40],

Proc. R. Soc. B 284: 20171287

haps not that surprising. To isolate the effects of signal efficacy, but this study provides comprehensive evidence of the evol-
the benefit (a mealworm) was the same for large winged prey utionary pressure that has shaped the diversity in hindwing
as small winged prey, and the Robomoth was always conspi- colouration in moths. Our study highlights the fact that
cuous against the panel. In nature however, smaller moths large size is a key element that (i) has generated a selection
may be not only harder to detect but also less profitable to pressure to evolve a secondary defence and (ii) makes dei-
attack, which (as our model demonstrates) may reduce the matic display more effective. Body size has been found to
advantage of adopting a costly defence. Therefore, it is likely be pivotal in influencing a range of primary defences from
that the need for a secondary defence, and efficacy of that mimicry [13] to eyespots [12], and we now have evidence
defence act in concert to promote the evolution of contrasting that it affects the nature of secondary defences that evolve,
hindwings in large species. or at least contrasting hindwing patterns typically associated
Although the presence of hidden contrasting hindwing with such defences.
colouration does not necessarily represent the presence of dei-
matic defence, we argue that hidden contrasting hindwings in
nocturnal moths have evolved primarily as an anti-predator Ethics. The bird experiments had ethical approvals from both Carleton
defence to startle predators, not least because deimatic display University Animal Care Committee (no. 103372) and National
has been evidenced in several underwing moths with contrast- Capital Commission (no. 17564).
ing hindwings (Catocala and Phyllodes genus) and deduced Data accessibility. The data and phylogenetic tree used for phylogenetic
from wing damage patterns seen in other Catocala species analysis are available in electronic supplementary material. The be-
havioural experiment data are available from the Dryad Digital
[7,27], but also because an association between contrasting
Repository: http://dx.doi.org/10.5061/dryad.7rt02 [41].
hindwings and deimatic display has previously been docu-
Authors’ contributions. C.K. and T.N.S. conceived the project, designed and
mented in other taxonomic groups (e.g. cryptic praying performed the behavioural experiment, and developed the mathematical
mantis [18]). An alternative explanation for the function of models. R.Z. provided and validated the phylogenetic tree and analysis.
contrasting hindwings is that it evolved as a sexual signal. C.K. and T.N.S. performed the statistical analysis and took the lead in
However, most nocturnal moths use non-visual sexual traits writing the manuscript, with contributions from all co-authors.
[36], and sexual signalling by colouration has not been reported Competing interests. We declare we have no competing interests.
as far as we know. Therefore, we consider that the most plaus- Funding. This research was supported by Natural Sciences and Engin-
ible evolutionary force that has shaped the contrasting eering Research of Canada (NSERC) to T.N.S. C.K. was supported by
a NSERC Discovery Accelerator Supplement (to T.N.S.) and a grant
hindwings in nocturnal moths is predation threats and these from the Research Foundation National Research Foundation of
hidden signals are used to startle predators as a secondary Korea (NRF-2014R1A6A3A03053952). R.Z. was supported by
defence. We also acknowledge that our contrast measurements Genomics Research and Development Initiative (GRDI) and the
(and also judgement of colouration by humans) necessarily Canadian Food Inspection Agency through a Genomics Applications
Partnership Program grant (GAPP 6102, Genome Canada).
neglected the colour information in the ultraviolet (UV)
range which can be perceived by moths’ natural predators Acknowledgements. We thank Rehal M. for the technical assistance in
developing a robotic moth. We also appreciate Kim Y. for her help
[37]. However, most of moths’ natural resting substrates reflect in the field research, Yip Y. for her assistance in image analysis and
negligible light in the UV spectrum [38]. In addition, in adult Arbuckle K. and Stuart-fox D. for their comments on the manuscript.
Lepidoptera, UV colours are mostly present on already Holen Ø. provided helpful comments on the model.

References
1. Cott HB. 1940 Adaptive coloration in animals. 5. Cuthill IC, Stevens M, Sheppard J, Maddocks T, Catocala (Lepidoptera: Noctuidae). Evolution 32,
London, UK: Methuen. Párraga CA, Troscianko TS. 2005 Disruptive 424–434. (doi:10.2307/2407609)
2. Ratcliffe JM, Nydam ML. 2008 Multimodal warning coloration and background pattern 8. Umbers KDL, Lehtonen J, Mappes J. 2015 Deimatic
signals for a multiple predator world. Nature 455, matching. Nature 434, 72 –74. (doi:10.1038/ displays. Curr. Biol. 25, R58– R59. (doi:10.1016/j.
96 –99. (doi:10.1038/nature07087) nature03312) cub.2014.11.011)
3. Ruxton GD, Sherratt TN, Speed MP. 2004 Avoiding 6. Kapan DD. 2001 Three-butterfly system provides a 9. Olofsson M, Eriksson S, Jakobsson S, Wiklund C.
attack, New York, NY: Oxford University Press. field test of Mullerian mimicry. Nature 409, 2012 Deimatic display in the European swallowtail
4. Stevens M, Merilaita S. 2011 Animal camouflage: 338 –340. (doi:10.1038/35053066) butterfly as a secondary defence against attacks
mechanisms and function. New York, NY: Cambridge 7. Sargent TD. 1978 On the maintenance of stability in from great tits. PLoS ONE 7, e47092. (doi:10.1371/
University Press. hindwing diversity among moths of the genus journal.pone.0047092)
 Downloaded from http://rspb.royalsocietypublishing.org/ on August 30, 2017

10. Mänd T, Tammaru T, Mappes J. 2007 Size 20. Hartigan JA, Manchek AW. 1979 Algorithm AS 136: 31. Kang C, Moon J, Sherratt TN, Lee S-I, Jablonski PG. 8
dependent predation risk in cryptic and conspicuous A k-means clustering algorithm. J. R. Stat. Soc. 2017 Multiple lines of anti-predator defence in the

rspb.royalsocietypublishing.org
insects. Evol. Ecol. 21, 485–498. (doi:10.1007/ Ser. C. Appl. Stat. 28, 100–108. (doi:10.2307/ spotted lanternfly, Lycorma delicatula (Hemiptera,
s10682-006-9130-z) 2346830) Fulgoridae). Biol. J. Linn. Soc. 120, 115– 124.
11. Remmel T, Tammaru T. 2009 Size-dependent 21. Arbuckle K, Speed MP. 2015 Antipredator defenses (doi:10.1111/bij.12847)
predation risk in tree-feeding insects with different predict diversification rates. Proc. Natl Acad. Sci. 32. Weise CM, Meyer JR. 1979 Juvenile dispersal and
colouration strategies: a field experiment. J. Anim. USA 112, 13 597 –13 602. (doi:10.1073/pnas. development of site-fidelity in the black-capped
Ecol. 78, 973–980. (doi:10.1111/j.1365-2656.2009. 1509811112) chickadee. Auk 96, 40 –55. (doi:10.2307/4085398)
01566.x) 22. Summers K, Clough ME. 2001 The evolution of 33. Sargent TD. 1973 Studies on the Catocala
12. Hossie TJ, Skelhorn J, Breinholt JW, Kawahara AY, coloration and toxicity in the poison frog family (noctuidae) of southern New England. IV.
Sherratt TN. 2015 Body size affects the evolution (Dendrobatidae). Proc. Natl Acad. Sci. USA 98, A preliminary analysis of beak-damaged specimens,
of eyespots in caterpillars. Proc. Natl Acad. 6227 –6232. (doi:10.1073/pnas.101134898) with discussion of anomaly as a potential anti-

Proc. R. Soc. B 284: 20171287

Sci. USA 112, 6664 –6669. (doi:10.1073/pnas. 23. Endler JA. 1984 Progressive background in moths, predator function of hindwing diversity. J. Lepid.
1415121112) and a quantitative measure of crypsis. Biol. J. Linn. Soc. 27, 175 –192.
13. Penney HD, Hassall C, Skevington JH, Abbott KR, Soc. 22, 187– 231. (doi:10.1111/j.1095-8312.1984. 34. Odum EP. 1942 Annual cycle of the black-capped
Sherratt TN. 2012 A comparative analysis of the tb01677.x) chickadee: 3. Auk 59, 499 –531. (doi:10.2307/
evolution of imperfect mimicry. Nature 483, 24. Paradis E, Claude J, Strimmer K. 2004. APE: analyses 4079461)
461–464. (doi:10.1038/nature10961) of phylogenetic and evolution in R language. 35. Mason JR, Reidinger RF. 1982 Observational
14. Zahiri R, Holloway JD, Kitching IJ, Lafontaine JD, Bioinformatics 20, 289–290. (doi:10.1093/ learning of food aversions in red-winged blackbirds.
Mutanen M, Wahlberg N. 2012 Molecular bioinformatics/btg412) Auk 99, 548 –554. (doi:10.2307/4085934)
phylogenetics of Erebidae (Lepidoptera, Noctuoidea). 25. Martins EP, Hansen TF. 1997 Phylogenies and the 36. Hansson BS. 1995 Olfaction in Lepidoptera. Experientia
Syst. Entomol. 37, 102–124. (doi:10.1111/j.1365- comparative method: a general approach to 51, 1003–1027. (doi:10.1007/BF01946910)
3113.2011.00607.x) incorporating phylogenetic information into the 37. Endler JA. 1978 A predator’s view of animal color
15. Zaspel JM, Zahiri R, Hoy MA, Janzen D, Weller SJ, analysis of interspecific data. Am. Nat. 149, patterns. In Evolutionary biology (eds MK Hecht, WC
Wahlberg N. 2012 A molecular phylogenetic 646 –667. (doi:10.1086/286013) Steere, B Wallace) pp. 319 –364. New York, NY:
analysis of the vampire moths and their fruit- 26. Revell LJ. 2010 Phylogenetic signal and linear Springer.
piercing relatives (Lepidoptera: Erebidae: Calpinae). regression on species data. Methods Ecol. Evol. 1, 38. Roberts AD, Ustin SL, Ogunjemiyo S, Greenberg J,
Mol. Phylogenet. Evol. 65, 786 –791. (doi:10.1016/j. 319 –329. (doi:10.1111/j.2041-210X.2010.00044.x) Dobrowski SZ, Chen J, Hinckley TM. 2004 Spectral
ympev.2012.06.029) 27. Harmon LJ, Weir JT, Brock CD, Glor RE, Challenger and structural measures of northwest forest
16. Stamatakis A, Hoover P, Rougemont JA. 2008 W. 2008 GEIGER: investigating evolutionary vegetation at leaf to landscape scales. Ecosystems 7,
A rapid bootstrap algorithm for the RAxML web radiations. Bioinformatics 24, 129–131. (doi:10. 545–562. (doi:10.1007/s10021-004-0144-5)
servers. Syst. Biol. 57, 758–771. (doi:10.1080/ 1093/bioinformatics/btm538) 39. Eguchi E, Meyer-Rochow VB. 1983 Ultraviolet
10635150802429642) 28. Harvell CD. 1990 The ecology and evolution of photography of forty-three species of Lepidoptera
17. Felsenstein J. 1985 Confidence limits on inducible defenses. Q. Rev. Biol. 65, 323 –340. representing ten families. Annot. Zool. Jpn. 56, 10–18.
phylogenies: an approach using the bootstrap. (doi:10.1086/416841) 40. Umbers KDL, Bona SD, White TE, Lehtonen J,
Evolution 39, 783–791. (doi:10.2307/2408678) 29. Van Buskirk J. 2000 The costs of an inducible Mappes J, Endler JA. 2017 Deimatism: a neglected
18. Edmunds M. 1972 Defensive behaviour in Ghanaian defense in anuran larvae. Ecology 81, 2813–2821. component of antipredator defence. Biol. Lett. 13,
praying mantids. Zool. J. Linn. Soc. 51, 1 –32. (doi:10.2307/177343) 20160936. (doi:10.1098/rsbl.2016.0936)
(doi:10.1111/j.1096-3642.1972.tb00771.x) 30. Umbers KDL, Mappes J. 2015 Postattack deimatic 41. Kang C, Zahiri R, Sherratt TN. 2017 Data from: Body
19. Wyszecki G, Stiles WS. 1982 Color science: concepts display in the mountain katydid, Acripeza reticulata. size affects the evolution of hidden colour signals in
and methods, quatitative data and formulae. Anim. Behav. 100, 68 –73. (doi:10.1016/j.anbehav. moths. Dryad Digital Repository. (http://dx.doi.org/
New York, NY: Wiley. 2014.11.009) 10.5061/dryad.7rt02)