J Evol Biol. 2022;35:467–478.
Received: 24 June 2021
Revised: 31 Januar y 2022
Accepted: 6 February 2022
DOI : 10.1111/j eb.13994
Genetic colour variation visible for predators and conspecifics
is concealed from humans in a polymorphic moth
Ossi Nokelainen1,2 | Juan A. Galarza1,2 | Jimi Kirvesoja1 | Kaisa Suisto1 |
This is an open access article under the terms of the Creative Commons Attribution License, which permits use, distribution and reproduction in any medium,
provided the original work is properly cited.
© 2022 The Authors. Journal of Evoluti onary Biolog y published by John Wiley & Sons Ltd on behalf of European Society for Evolutionar y Biology
Ossi Noke lainen and Juan A. Gal arza author s are contributed equally to thi s work.
1Department of Biological and
Environmental Science, University of
Jyväskylä, Jyväskylä, Finland
2Organismal and Evolutionary Biology
Research Program, Faculty of Biological
and Environmental Sciences, University
of Helsinki, Helsinki Universit y, Helsinki,
Ossi Nokelainen, Depar tment of Biological
and Environmental Science, Universit y of
Jyväskylä, P.O. Box 35, Jyväskylä 4 0014,
This work was supported by the Academy
of Finland to JM (#320438) and the grant
(#21000 038821) to ON.
The definition of colour polymorphism is intuitive: genetic variants express discretely
coloured phenotypes. This classification is, however, elusive as humans form subjective
categories or ignore differences that cannot be seen by human eyes. We demonstrate
an example of a ‘cryptic morph’ in a polymorphic wood tiger moth (Arctia plantaginis),
a phenomenon that may be common among well- studied species. We used pedigree
data from nearly 20,000 individuals to infer the inheritance of hindwing colouration.
The evidence supports a single Mendelian locus with two alleles in males: WW and
Wy produce the white and yy the yellow hindwing colour. The inheritance could not
be resolved in females as their hindwing colour varies continuously with no clear link
with male genotypes. Next, we investigated if the male genotype can be predicted
from their phenotype by machine learning algorithms and by human observers. Linear
discriminant analysis grouped male genotypes with 97% accuracy, whereas humans
could only group the yy genotype. Using vision modelling, we also tested whether the
genotypes have differential discriminability to humans, moth conspecifics and their
bird predators. The human perception was poor separating the genotypes, but avian
and moth vision models with ultraviolet sensitivity could separate white WW and
Wy males. We emphasize the importance of objective methodology when studying
colour polymorphism. Our findings indicate that by- eye categorization methods may
be problematic, because humans fail to see differences that can be visible for relevant
receivers. Ultimately, receivers equipped with different perception than ours may im-
pose selection to morphs hidden from human sight.
aposematism, Arctia plantaginis, discriminant analysis, multispectral imaging, polymorphism,
wood tiger moth
NO KELA INEN Et A L.
1 | INTRODUCTION
Colour polymorphism, the occurrence of multiple discrete colour phe-
notypes within a population (Ford, 1945; Huxley, 1955; White & Kemp,
2016), is a flagship topic of evolutionary biology (Gray & McKinnon,
2007; McKinnon & Pierotti, 2010; Svensson, 2017). The study of co-
lour polymorphism has traditionally been a very popular topic among
evolutionary biologists (Brakefield & Liebert, 1985; Cain & Sheppard,
1954; Fisher & Ford, 1947; Kettlewell, 1955), because as a visible
trait, colouration enables scientists to study evolution in action in a
tractable manner. Importantly, colouration is a composite trait that
has multiple fitness- linked functions (Cuthill et al., 2017), including
thermoregulation (e.g. Stuart- Fox et al., 2017), immune defence (e.g.
Freitak et al., 2005), sexual signalling (e.g. Tibbetts et al., 2017) and
avoiding predation either through camouflage, mimicry or warning sig-
nalling (e.g. Ruxton et al., 2019). The diversity of colours, as well as co-
lour polymorphism, is therefore valuable to understand the processes
generating and maintaining genetic variation in the wild.
Classically, a genetic polymorphism is defined as: ‘the occurrence
together of two or more discontinuous forms of a species in the
same habitat in such proportions that the rarest of them cannot be
maintained merely by recurrent mutation’ (Ford, 1945, 1965; Huxley,
1955). This definition has remained virtually unchanged over the last
75 years (Nokelainen et al., 2018; Svensson, 2017; White & Kemp,
2016). While the concept of colour polymorphism may be rather
intuitive, its quantification is not, mainly because the difference
between colour variants is not always clear- cut. For example, phe-
notypic plasticity can provide discrete appearances as if they were a
result of polymorphism (Price, 2006), such as the density- dependent
colour change (i.e. polyphenism) in a desert locust (Schistocerca
gregaria) (Sword et al., 2000). On the other hand, sometimes a ge-
netically polymorphic trait may show overlapping phenotypic dis-
tribution (Kappers et al., 2018; Nokelainen et al., 2018), such as the
extravagant colour polymorphism of the Hawaiian happy- face spi-
der (Theridion grallator) that shows high phenotypic variation among
populations (Gillespie & Oxford, 2009).
In Lepidoptera, one of the potential caveats of early polymor-
phism studies is that colouration was mostly quantified through
human vision and thus included a source of subjectivity (Brakefield
& Liebert, 1985; Endler, 1990; Fisher & Ford, 1947). This may not
always be a problem as humans are good in categorizing colours
across a broad visible wavelength spectrum from 400 to 70 0 nm
(Bergeron & Fuller, 2017) and have excellent visual acuity (Caves
et al., 2018). However, our perception excludes the near ultravio-
let part of the spectrum (300– 400 nm), which is important to many
animals in signalling (Kelber et al., 2003; Osorio & Vorobyev, 2005).
Also, humans may not be able to detect nuances in patterns or judge
colour polymorphism based only on a single key trait (e.g. hindwing
‘base’ colour), or see subjective categories where they do not exist.
Colouration must therefore be objectively quantified using either
spectrometry or multispectral imaging approaches (Endler, 1990,
Troscianko & Stevens, 2015, van den Berg et al., 2020). As such
there is a clear need for studies that can link the phenotypes to their
genotypes, because this can illuminate our understanding of how
selection of allele frequencies that constitute the genotypes operate
in the wil d (C uthill et al., 2017; Svens son, 2017; Tib bet ts et al., 2017).
We investigated genotype– phenotype associations in the wood
tiger moth (Arctia plantaginis), a widely distributed member of the
Erebidae family (Rönkä et al., 2016) found across the Northern
hemisphere (Hegna et al., 2015). It is known that its polymorphic
hindwing colour is heritable in males (Nokelainen et al., 2013) and
in females (Lindstedt et al., 2016). In general, males have either yel-
low or white hindwings (Nokelainen et al., 2012, 2013; Suomalainen,
1938). Female hindwings, on the other hand, vary continuously in
the yellow- orange- red range (Lindstedt et al., 2011).
First, we explored the heritability of the hindwing colouration. It
has been suggested, although with a very limited data set originating
from a single brood, that the inheritance of male hindwing coloura-
tion follows a Mendelian one- locus two- allele model where the yel-
low allele is recessive (Suomalainen, 1938). The genetic mechanism
of female wing colouration is largely unknown and some plasticity
in female colouration has been reported (Lindstedt et al., 2010). To
confirm the mode of inheritance, we compared human- visible hind-
wing colour (and only hindwing colour as genotype proxy) frequen-
cies from 452 laboratory- reared families (i.e. with pedigree) against
those predicted by the one- locus two- allele model. Second, focussing
in the males, we tested further whether their genotype can be pre-
dicted by their phenotype. We examined if the colour morphs could
be assigned to their genotype (i.e. an information derived from the
pedigree, Box 1) by human observers through sequential and simul-
taneous sorting tasks, as well as by machine learning algorithms (i.e.
discriminant functions). We used linear discriminant function analysis
as part of the machine learning realm, as we wanted to explicitly un-
derstand what are the parameters that may allow visual separation
of the genotypes. It can be expected that the computational meth-
ods should outperform human sorting skills, because the algorithms
can take into account combined nuances in phenotypic variation, in-
cluding those beyond the human- visible spectrum (Høye et al., 2021;
Wilkins & Osorio, 2019). Lastly, we asked whether these genotype–
phenotype associations may have ecological relevance beyond the
human- visible spectrum. Using vision modelling, Henze et al. (2018)
investigated the differences in the discriminability of the wood tiger
moth colour morphs by moth conspecifics and bird predators. Here,
we used receptor- noise- limited vision modelling (Maia et al., 2013;
Vorobyev & Osorio, 1998) to test pairwise genotype chromatic con-
trasts of hindwing colour using human, avian and moth vision models.
2 | MATERIAL AND METHODS
2.1 | Moth pedigree rearing protocol
Altogether we used pedigree data from 15 generations of the wood
tiger moth reared in the laboratory over the course of 6- years. As a
gene ral rearin g proto col, two adults (male and female) are put tog eth er
in a plastic box (Huhtamäki, 1000 ml, transparent casing) with mesh
NOKEL AINEN E t AL.
for ventilation at the top and allowed to mate under natural lighting.
The eggs are laid inside the box where after ~6 days the larvae hatch
and are kept for another ~14 days inside the box as they are too deli-
cate to be moved. The larvae are then separated into rearing contain-
ers (max. 30 lar vae/container) to continue growth at approximately
25°C under natural light conditions and are fed with fresh dandelion
leaves (Tara xa cum ssp.) until pupation. The pupae are then moved
into individual jars and sex and wing colour are recorded from the
emerging adults. In males, colour classification is conventionally done
by- eye (white or yellow hindwing ‘base colour’) and in females, a six-
step (yellow- orange- red) scale is used as described by Lindstedt et al.
(2010) and Nokelainen et al. (2012): in 1 to 6 scale yellow- orange- red
gradient yellows are 1– 2, oranges 3– 4 and reds 5– 6 (Figure 1).
2.2 | Inheritance of hindwing colour based
on the pedigree
We specifically tested a Mendelian inheritance model where a
single locus with two- alleles controls the white- yellow polymor-
phism. Here, the yellow allele (y) is recessive to the dominant
white (W) allele, as suggested by Suomalainen (1938) who used
a single brood of individuals originating from Finland. Under this
model, genotypes with the W allele (i.e. WW, Wy) exhibit white
hindwings, whereas only the homozygote recessive genotype (i.e.
yy) exhibit yellow hindwings. We tested the one- locus two- allele
model (Box 1), by comparing the expected model's frequencies
to those obser ved from 452 families with pedigree data (each
offspring was considered as independent data point) using a Chi
Square test for independence. The expected frequencies under
this model with their resulting hindwing colour are presented in
Tables S1 and S2. Only families with at least 10 male and female
offspring were used to ensure reliable frequency distribution.
From the pedigree data, the parental genotypes were inferred
from the phenotype distribution of the F1 offspring. For instance,
a 100% yellow mal e off spring wo uld ind icate that both parents are
homozygous for the y allele (Box 1). The reasoning uses the same
approach as myriad classic studies of ecological genetics decipher-
ing Mendelian ratios. The key point with this pedigree back- trace
approach is that by producing consecutive generations it is pos-
sible to match the observed F1 phenotype frequencies to those
expected by Mendelian inheritance, and thus, the genotype of the
parental generation can be inferred.
2.3 | Objective genotype– phenotype associations
using image analysis techniques
The pedigree back- trace approach above was used to unravel the
inheritance of hindwing colouration using hindwing colouration as
BOX 1 Pedigree crossing design and determination of the wood tiger moth genotypes with respect to hind wing
colour. The first panel shows the classic Mendelian one locus two allele segregation (A); we expect that white is
dominant trait over yellow (Suomalainen, 1938). Each homozygous parent in the parental generation produces one
type of gamete (W or y). The following generation heterozygous offspring produces again two types of gametes. In
the second panel (B), the next generation produces offspring with a 3:1 ratio of dominant allele to recessive. The
third panel, shows the crossing design followed to mate selection lines of known genotypes and their expected
phenotype frequencies. Fifteen generations were produced over the course of 6- years. The colours in the bars
indicate the hind wing colour of the offspring. An important point with this classic approach is that by producing
consecutive generations and following the logic of expected offspring phenotype frequencies, it is possible to
back- trace pedigree and determine putative genotypes of earlier generations.
(a) (b) (c)
NO KELA INEN Et A L.
perceived by humans. To investigate genotype– phenotype associa-
tions further than simply using a human- visible hindwing colour, we
used a subset of laboratory- reared adults with a known pedigree in
our image analyses. The total sample size of the photographed in-
dividuals was 292: where of the males 37 were WW, 88 were Wy,
and 42 were yy, while of the females, 33 were WW, 68 were Wy
and 24 were yy. The image calibration and analysis broadly followed
previously established methods (Troscianko & Stevens, 2015, van
den Berg et al., 2020). Briefly, photography was undertaken with a
Samsung NX1000 digital camera converted to full spectrum with no
quartz filter to enable UV sensitivity fitted with a Nikon EL 80 mm
lens. For the photos in the human- visible range, we used a UV and
infrared (IR) blocking filter on the lens, which passes wavelengths
only between 400 and 680 nm (Baader UV/IR Cut Filter). For the UV
images, a UV pass filter was used (Baader U filter), which transmits
wavelengths between 320 and 380 nm. Grey reflectance standards,
which reflect light equally at 7% and 93% between 300 and 750 nm,
were used for image calibration. A standard light source 75W Exo-
terra Sunray (mimicking sunlight across the spectrum) was used.
To obtain colour and pattern metrics, we measured the entire
dorsal view of the forewings (FW), hindwings (HW), thorax (TH)
and abdomen (AB) of the mounted and spread adult as regions of
interest (ROI). For reflectance data, we used normalized camera re-
sponses of red, green, blue and the UV channel. To extract pattern
information, we applied a pattern analysis technique (a ‘granularity’
analysis), which decomposed the image into a series of spatial fre-
quencies (‘granularity bands’) using Fourier analysis and band pass
filtering, followed by determining the relative contribution of differ-
ent marking sizes to the overall pattern (Barbosa et al., 2008; Hanlon
et al., 2009; Stoddard & Stevens, 2010). The analysis calculated the
amount of light information (or pixel energy) corresponding to mark-
ings of different sizes, starting with small markings (we used a pixel
start size of 2) and increased in size to larger markings (we used a
pixel end size of 100). Increase in pixel step size was set to multiply
each step by 1.414, thus representing exponential growth. The lu-
minance was measured over 20 bands from lowest luminance (0) to
highest luminance (65535), the maximum dynamic range of a 32- bit
TIFF image. The luminance channel was set to longwave channel (R).
For the pattern data variables, we used dominance (i.e. maxPower—
the energy at the spatial frequency with the highest pixel energy),
diversity (i.e. propPower— maximum or peak energy value divided by
the summed energy) and marking size (i.e. maxFreq— the spatial fre-
quency with peak energy).
Prior to testing, colour metrics were filtered for correlations to
avoid multicollinearity. The following variables were retained: area,
three pattern variables (pattern size, contrast and diversity) and four
bandpass channels (UV, blue, green, red channels, i.e. uv, sw, mw, lw
respectively). All values were separately measured for the four re-
gions of interest (forewing, hindwing, thorax and abdomen). In addi-
tion, the following allometric measurements of size were calculated
by dividing areas of the ROIs: forewing to abdomen (FW/AB), fore-
wing to thorax (FW/TH), forewing to hindwing (FW/AB) and thorax
to abdomen (TH/AB).
2.4 | Discreteness of colour morphs— subjective
genotype discrimination using human observers
Next, we evaluated human sorting accuracy of male genotypes
through sequential (‘sequence’) and simultaneous (‘sorting’)
FIGURE 1 Pedigree information of the wood tiger moth
genotype crossings and their human- visible hindwing coloration.
The figure shows relative frequencies of offspring phenotypes
with respect to their parental genotype crosses. The numbers
above the bars indicate the sample size. The colours in the bars
indicate the subjective by- eye hindwing colour of the offspring.
Notice the dichotomous yellow- white hindwing categorization
in males (a), whereas the females are more converged to similar
orange coloration (b); the scale depicts the visually scaled yellow-
orange- red colour gradient used to categorize female coloration.
The images were gamma corrected for better screen imaging
and are meant to illustrate representative examples of the wood
tiger moth colour variation. In males, hindwing colour shows
statistically nonsignificant difference from predicted one locus two
allele model, whereas females deviate significantly from the same
predicted outcome of phenotypes
NOKEL AINEN E t AL.
tasks. Participants were familiar with the wood tiger moth. In
both tasks, we showed participants 10 images per genotype (i.e.
3 genotypes by 10 replicates) and asked them to sort the images
according to their genotype. As the mechanism controlling for
the hindwing colouration in females is currently unknown and
warrants further investigation, we used only males in the sorting
tasks due to their discrete hindwing colouration that allows for
In the sequential sor ting task, 12 par tici pa nt s we re asked to clas-
sify moth photographs by their genotype. The photographs were
of a mounted specimen with wings spread out. The photographs
were shown in a randomized order via the Google Docs ‘Forms’ plat-
form. The participants were asked to pay attention to the appear-
ance, wing and body colouration. The following cues derived from
the image analysis (see above) to classify the male genotypes (WW,
Wy, yy) were given as training instructions. White hindwings, large
forewing patterning and pale abdomen are typical to WW. White
hindwings and a yellow tinge in forewings and abdomen are typical
to Wy. Yellow hindwings, variable wing patterning, yellow abdominal
colour is typical to yy. The participants were instructed to classify
moths using these cues (Fig. S3).
In the simultaneous sorting task, 10 participants (a subset of the
former group) were asked to sort the genotypes into three clusters
based on similarities in their appearance; no further instructions
were given to accomplish this task. All moths were visible at the
same time and the test was done using PowerPoint slides with moth
photographs (i.e. 30 images were simultaneously presented, 3 geno-
types by 10 replicates). The percentage of correct answers was then
calculated (Fig. S4).
2.5 | Vision modelling
The vision modelling we carried out largely followed established
methods (Stevens et al., 2007, Troscianko & Stevens, 2015, van
den Berg et al., 2020). To gain insight into how well different vi-
sion systems can recognize the colour differences between geno-
types, we used a receptor- noise limited (RNL) visual discrimination
model (Vorobyev et al., 1998). We compared trichromatic human,
tetrachromatic avian (Blue tit; Cyanistes caeruleus) and trichromatic
moth (wood tiger moth) vision models. This allowed us to mecha-
nistically understand human sorting accuracy of genotypes and to
compare this with more ecologically relevant vision systems of con-
specifics (moths) and predators (birds). We used 0.05 Weber fraction
for most abundant cone type for all vision models. The cone ratios
were: avian cone ratios 1:1.92:2.68:2.7 uv:sw:mw:lw (Hart, 2001b),
human cone ratios 0.057:0.314:0.629 sw:mw:lw (Hofer et al., 2005).
For moth vision model, spectral sensitivities of cone cells (uv, sw,
mw) were obtained from (Henze et al., 2018), and cone ratios 1:1:1,
were used as the specific ratio is unknown. As we were interested
in differences in chromatic contrast (dS), we excluded the achro-
matic contrast (dL) from the vision model analysis. The vision model
yields discrimination values in ‘just noticeable differences’ (JNDs), al-
though before behavioural validation these should be considered as
predicted contrast values (dS). By definition, values lower than one
(<1 JND) are considered indistinguishable, whereas larger values are
discriminable for the receiver (Kang et al., 2015; Nokelainen et al.,
2019; Siddiqi et al., 2004).
2.6 | Statistical analyses
First, we tested whether the wood tiger moth hindwing colour
follows a simple Mendelian one- locus two- allele inheritance pat-
tern using Chi Square test for independence. We would expect
that the observed phenotype frequencies from the crossing de-
signs do not deviate significantly from the predicted phenotype
frequencies. We tested the expected versus observed subjective
colour morph frequencies separately for males (Table S1) and fe-
males (Table S2).
Second, we tested the discriminability of genotype– phenotype
associations. The success rate of cor rect genotype designation was
tested with a general linear model (GLM with a Poisson distribu-
tion), where the success rate of visually genotyping each moth was
set as the dependent variable and method of genotyping (human
sequential task, human simultaneous task, or computer algorithm)
as the explanatory variable. We used a linear discriminant function
analysis to evaluate whether the computer algorithm can outper-
form human observers in the genotype sorting task. The analysis
was carried out as a 3- group problem. The genotype (derived from
the pedigree data) was set as the predicted group membership
and regions of interest (ROI) were selected form the digital image
namely; area, pattern size, pattern contrast, pattern diversity, uv,
sw, mw, lw, forewing to abdomen, forewing to thorax, forewing to
hindwing and thorax to abdomen were set as predictor variables.
All colour and pattern metrics were investigated separately for the
forewing, hindwing, thorax and abdomen. We also used the Boruta
feature selection algorithm to provide additional information on
which feature s are th e bes t predi c tor s of th e prio r gen otyp e gro ups
(under R- package ‘Boruta’). Briefly, Boruta is a random forest al-
gorithm that compares the significance of each variable against
random noise data created from all variables of interest (Kursa &
Rudnicki, 2010). The significance is then determined based on the
relative difference against the random noise. Generally, variables,
which fall in between lower- and upper- bound significance of the
random generated noise reference are flagged as non- significant
(Table S3, Figs. S1– S2).
Third, we tested the discriminability of the genotypes using
three different vision models. For this, we conducted a linear
mixed effects model (lmer- function) with the lmerTest R- package
(Kuznetsova et al., 2017). The colour contrast (dS) was set as the
dependent variable and genotype (WW, Wy, yy), ROI (abdomen,
forewing, hindwing), vision model (avian, human, moth) and their in-
teractions were set as the explanatory variables. The moth ID was
NO KELA INEN Et A L.
set as a random variable to control for data structure. All analyses
were conducted using RStudio, version 1.1.447 and R, version 3.5.0
(R Core Team, 2018; RStudio Team, 2016).
3 | RESULTS
3.1 | Mendelian inheritance of the hindwing colour
Of the individuals used in the pedigree analysis 10911 were males
and 8295 were females (Figure 1, Tables S1 and S2). In the males, the
frequencies of white and yellow offspring were in close agreement
to the expected phenotype frequencies under one- locus two- allele
Mendelian inheritance where white dominates over yellow (Box 1,
Figure 1a). Thus, the one- locus two- allele inheritance mode with
dominance of the W allele over y was confirmed by the pedigree
data for males. Whether the established locus also controls hindwing
colour in the females was less clear. In females, the spread of the
hindwing colour phenotypes showed an apparent normal distribu-
tion (Figure 1b) and indicated phenotypic tendency towards orange
hindwing colouration (by- eye classification; yellow- orange- red). Out
of all crossings, the emerged females on average, were 3% yellow,
68% orange and 29% red regardless of the parental genotype. Thus,
there was no obvious correlation between the male (white- yellow)
and female (yellow- orange- red) hindwing colour within the brood
(Figure 1a- b) when using the subjective hindwing colour classifica-
tion made by human observers.
3.2 | Discriminability of genotype– phenotype
associations— human versus algorithm
A computer- based discrimination algorithm outperformed subjec-
tive sorting accuracy of humans using linear combinations of col-
our and pattern data (ANOVA: F2,48 = 7.10 , p < 0.001). In males, the
linear discriminant analysis reached 96.80% accuracy for predicting
the correct genotype membership. Within test data, 88.88% of WW,
98.11% of Wy and 100% of yy were correctly classified (Figure 2a).
The discriminability of genotypes using colour metrics was also as-
sessed using the Boruta feature extraction algorithm (Table S3). The
most important variables to separate white and yellow morphs are
all hindwing features and include: UV reflectance, short wavelength
reflectance, pattern diversity, pattern contrast and long wavelength
reflectance. The most important variables to separate white geno-
types (WW, Wy) are: thorax UV reflectance, thorax to abdomen size
ratio, abdomen UV, forewing to abdomen size ratio and forewing
In females, the linear discriminant analysis reached 65.65% ac-
curacy for predicting the correct genotype membership. Within test
data, 83.33% of WW, 58.62% of Wy and 69.56% of yy were cor-
rectly classified (Figure 2b). The most important variables to make
a distinction between white homozygotes (WW) and yellow allele
bearers (Wy or y y) are: thorax UV reflectance, abdomen marking
size, abdomen UV reflectance, forewing UV reflectance, hindwing
short wavelength reflectance (Table S3). The most important vari-
ables to separate white heterozygotes (Wy) from yellow homozy-
gotes (y y) are: abdomen marking size, hindwing short wavelengths,
FIGURE 2 Objective phenotype quantification using colour and pattern metrics. Linear discriminant analysis for wood tiger moth male
(a) and female (b) genotypes. Notice that here genotypes refer to one locus two allele model, where there are two predominantly white
morphs and one yellow in males and red, orange, yellow morphs in females. In males, the yy (yellow) genotype is clustered easily as its own
subgroup, but linear combinations also separates WW (white) and Wy (white) genotypes with good degree of certainty along a second axis
that combines parts of the full spectrum (invisible to us) as well as pattern metrics. In females, the three genotypes seemingly cluster into
three subgroups; however, there is much more grouping overlap (i.e. phenotypic convergence) than in males
NOKEL AINEN E t AL.
hindwing medium wavelengths, forewing short wavelengths and
hindwing long wavelengths.
We next focussed on discriminability of genotypes for human
observers only using males as we detected a close phenotypic
similarity in females. Human participants were not able to reliably
categorize the genotypes (Figure 3, Fig. S5). Of the male moths, par-
ticipants were only able to distinguish yellow (yy) genotype from the
whites, but not the two white male genotypes (W W, Wy). In the si-
multaneous sorting task where all moths were presented together,
participants were able to sort the homozygous and heterozygous
males only slightly better than in the sequential sorting task, yet only
60% was the highest success rate in sorting white males based on
3.3 | Chromatic discriminability of the genotypes
to ecologically relevant receivers
The discriminability of the genotypes was measured pairwise using
hindwing chromatic contrast (dS) of the two moths being compared
(Figure 4). The vision modelling results indicate that detectability of
the genotypes was different for human, bird and moth vision mod-
els, as the three- way interaction of vision model, ROI and genotype
was significant (lmer ANOVA, F4,23 028.1 = 506.11, p < 0.001). Also,
vision modelling results suggest that human perception is poor at
separating WW and Wy male genotypes, whereas avian and moth
vision systems with UV sensitivity could separate white WW and Wy
male genotypes (Figure 4). Thus, when viewed through ecologically
FIGURE 3 The success of objective
versus subjective sorting of male
genotypes. Colour metric super vised
computer algorithm (LDA, linear
discriminant analysis) outperforms human
sorting accuracy of genotypes in both
sequential (‘sequential’) and simultaneous
(‘sorting’) tasks. In sequential sorting
people were asked to classify moth
pictures by their genotypes; only basic
information of the best describing colour
metrics were given as instructions. In
simultaneous sorting, people were asked
to sort the genotypes into three clusters
based on their superficial appearance;
no further instructions were given, but
all moths were visible at the same time.
Percentage indicates the number of
correct answers by genotype
NO KELA INEN Et A L.
relevant receivers’ vision, the genotypes may have nuanced pheno-
typic differences beyond human perception (Figure 5).
4 | DISCUSSION
Our results highlight that genetic polymorphism expressed at the
phenotypic level is not always clear- cut to define as categorization
depends on the perception. With a wood tiger moth stock originating
from north Europe, we validate that male hindwing colour is geneti-
cally controlled by a single Mendelian locus, where the white allele
(W) is dominant to the yellow (y) allele. Male genotypes can be told
apart using colour metrics with high accuracy by machine learning al-
gorithms, but not by human obser vers, because white heterozygous
males can be separated from white homozygotes by differences in
ultraviolet reflectance. In turn, female genotypes are currently in-
separable by their phenotypes to us. It seems plausible that female
colour is controlled to some extent by the same genetic loci as the
male colour, although with different dominance relationships and
other interacting loci. Although we have less- extensive data about
colour variation in females, it seems that in many localities a yellow-
orange- red continuum is common (Lindstedt et al., 2011).
Generally, functional genes in the melanin biosynthetic path-
way can affect both wing scale pigmentation and morphology in
Lepidoptera (Matsuoka & Monteiro, 2018). Our preliminary pig-
ment analyses in the wood tiger moth indicate that the white pig-
mentation in the wings is produced by N- acetyldopamine (NADA)
sclerotin, whereas the yellow pigment is derived from a mix of N-
β- alanyldopamine (NBAD) sclerotin and pheomelanin and the red
pigment results from a dopamine- derived pheomelanin (Brien et al.
In Prep.). Also, differential expression in melanin- promoting and
melanin- inhibiting genes impacts black colouration in the cuticle and
in the hairs of wood tiger moth caterpillars (Galarza, 2021). Plausibly,
differential regulation in genes involved in the melanin pathway
could contribute to the colour differences between the sexes (Gazda
et al., 2020). Whether an upregulation of a single gene that causes
FIGURE 4 Wood tiger moth genotype separability through ecologically relevant vision systems compared to human vision. We tested
the discriminability of the genotypes using three different vision models (human, avian and moth vision). In all images, x- axis represents
pairwise genotype comparisons and y- axis shows chromatic contrast (dS) of the two moths being compared. The region of interest (ROI)
indicates the comparison between abdomen (AB), forewing (FW) and hindwing (HW) colour. The panels separate vision modelling results for
human (a), avian (b) and moth (c) vision models. Contrast values <1 are considered indistinguishable and values above this are increasingly
easy to distinguish (outliers not shown). The black horizontal line indicates dS = 1 corresponding to the perception threshold of the contrast
NOKEL AINEN E t AL.
pigment degradation in the other sex could take place in wood tiger
moth's melanin- based colouration is currently unknown but war-
rants further investigation.
The machine learning algorithm was more efficient at assign-
ing individuals to known group memberships based on the subtle,
but consistent phenotypic differences between the genotypes in
comparison to human observers. The finding that computational
approach outperforms human perception is not surprising, but still
the majority of the colour polymorphism research relies on classic
approach using human- visible categorization. In males, the linear
discriminant analysis separated the three- group problem with high
accuracy. Successful discrimination between the three groups of
males was expected due to differences in short and long wavelength
reflectance between the two white morphs (Henze et al., 2018;
Nokelainen et al., 2012). Also, the white homozygotes have a lower
thorax UV reflectance, smaller thorax by abdomen ratio (i.e. larger
abdomens), smaller forewing by abdomen ratio, lower abdominal
UV reflectance and less variable forewing patterning. In females,
the sorting accuracy was not any better than from expected ran-
dom chance frequency. Genotypes clustered more closely together
in phenotypic space and the three- group problem was separated
with low accuracy. It may be possible to improve this prediction ac-
curacy using different boundary selection protocol; however, it will
not change the fact that the females are phenotypically more similar
than males. The covariation of some of these phenotypic differences
is still unclear, however, it may be possible to use these phenotypic
associations in combination to predict genotypes of wild caught in-
dividuals. It will be our future task to investigate whether increasing
data over the years and developing methods (e.g. convolutional neu-
ral networks) will enhance the prediction accuracy.
From an evolutionary standpoint, it is plausible that interplay
between natural and sexual selection facilitates polymorphism in
this species (Gordon et al., 2015, 2018; Nokelainen et al., 2012;
Rönkä et al., 2020). Since male wood tiger moths, which are ac-
tively searching for females in the vegetation, have limited ability
to see differences in yellow- orange- red hues (Henze et al., 2018),
it is unlikely that sexual selection alone would be responsible for
the colouration of females, but we cannot exclude the possibil-
ity that male colouration could be used in intraspecific commu-
nication. Moreover, recent studies in other species have shown
that UV may facilitate separation of incipient species as recently
demonstrated in Colias butter flies (Fic ar rotta et al., 2022) and that
the differences in UV reflection may arise from novel duplication
of the gene producing sex- specific differences in reflectance as
in Zerene cesonia butterfly (Rodriguez- Caro et al., 2021). Previous
experiments have shown that birds learn to avoid red wood tiger
moths more effectively than yellow or white ones (Ham et al.,
2006; Lindstedt et al., 2011; Rönkä et al., 2018), but the selection
FIGURE 5 Representation of the wood
tiger moth genotypes to illustrate how
they may appear to ecologically relevant
receivers. These false image examples
show genotypes of both sexes organized
in vertical columns and human, avian and
moth false colour images in horizontal
groups. For trichromatic human vision
sw, mw and lw sensitivities were used for
blue, green and red channels respectively.
For tetrachromatic avian vision uv, sw
and lw sensitivities were used for blue,
green and red channels respectively. For
trichromatic moth vision uv, sw and mw
sensitivities were used for blue, green and
red channels respectively. Images were
corrrected for better image screening
NO KELA INEN Et A L.
for visual signals may be altered due to multimodal signalling
(Rojas et al., 2018; Winters et al., 2021). Also, avian predators may
distinguish between the nuances in colouration among the gen-
otypes as they, and wood tiger moths, perceive UV wavelengths
that are beyond human perception (Henze et al., 2018). Thus, eco-
logically relevant receivers, predators and conspecifics, may exert
different selection pressures on visual signals beyond our percep-
tion (Endler, 1978) and maintain colour polymorphism in natural
conditions (Galarza et al., 2014; Mochida, 2011; Nokelainen et al.,
2014; Rönkä et al., 2020).
Conclusively, determining genotypes based on their pheno-
typic characteristics is important in any species, because it al-
lows studying allele dynamics in the wild (e.g. as in tiger moths
(Brakefield & Liebert, 1985; Fisher & Ford, 1947; Liebert &
Brakefield, 1990), lizards (Sinervo & Calsbeek, 2006; Sinervo &
Lively, 1996) and damselflies (Le Rouzic et al., 2015; Svensson
& Abbott, 20 05)). Colour polymorphisms have several fitness
consequences in the maintenance of genetic variation (Galarza
et al., 2014; Gray & McKinnon, 2007; McKinnon & Pierotti, 2010).
They can influence intraspecific variation in mating cues (Merrill
et al., 2012; Nokelainen et al., 2012), fitness of colour morphs
in different light environments due to increased predation risk
(Nokelainen et al., 2014, 2021; Rojas et al., 2014) and divergence
in thermoregulatory capabilities (Forsman, 2000; Hegna et al.,
2013; Lindstedt et al., 2009). As it has become possible to model
the conspicuousness of different genotypes to different receivers
(Endler & Basolo, 1998; Hart, 2001a; Henze et al., 2018), we may
soon be able to estimate how their appearance shapes the fate
of allelic combinations using long- term data sets (Le Rouzic et al.,
2015; Svensson & Abbott, 2005). Ultimately, this will broaden our
understanding of how genetic variation underlying phenotypic
evolution is shaped in nature.
We thank the members of the plantaginis research group and helping
to sort the phenotypes in the quiz, as well as University of Jyväskylä
Department of Biological and Environmental Science Darwin Club
CONFLICT OF INTEREST
Authors have no conflict of interest to declare.
ON wrote the fi r st dr a f t of the man u s c r ipt, con d u c ted th e im a g e an al-
ysis and analysed frequency data; JAG devised the crossing- design
and inheritance analyses; JK helped with genotype– phenotype fre-
quencies analyses; KS reared tens of thousands wood tiger moths
over consecutive years and JM contributed substantially to project
management and manuscript editing.
The peer review history for this article is available at https://publo
ns.com/publo n/10.1111/je b.13994.
DATA AVAILAB ILITY STATE MEN T
The supporting data are archived in a public repository (https://jyx.
jyu.fi/handl e/12345 6789/79808): https://doi.org/10.17011/ jyx/
Ossi Nokelainen https://orcid.org/0000-0002-0278-6698
Barbosa, A., Mäthger, L. M., Buresch, K. C., Kelly, J., Chubb, C., Chiao,
C. C., & Hanlon, R. T. (2008). Cuttlefish camouflage: The effects
of substrate contrast and size in evoking uniform, mottle or dis-
ruptive body patterns. Vision Research, 48, 1242– 1253. https://doi.
Ber geron , Z. T., & Fuller, R . C. (2017). Usin g human vision to detec t varia-
tion in avian coloration: How bad is it? The American Naturalist, 191 ,
269– 276. htt ps://doi.org/10.1086/695282
Brakefield, P. M., & Liebert, T. G. (1985). Studies of colour polymorphism
in some marginal populations of the aposematic jersey tiger moth
Callimorpha quadripunctaria. Biological Journal of the Linnean Society,
26, 225– 241. https://doi.org/10.1111/j.1095- 8312.1985.tb016 34.x
Cain, A. J., & Sheppard, P. M. (1954). Natural selection in Cepaea.
Genetics, 39, 89– 116. https://doi.org/10.1093/genet ics/39.1.89
Caves, E. M., Brandley, N. C., & Johnsen, S. (2018). Visual acuity and the
evolution of signals. Trends in Ecology & Evolution, 33, 1– 15. https://
Cuthill, I. C., Allen, W. L., Arbuckle, K., Caspers, B., Chaplin, G., Hauber, M.
E., Hill, G. E., Jablonski, N. G., Jiggins, C. D., Kelber, A., Mappes, J.,
Marshall, J., Merrill, R., Osorio, D., Prum, R., Roberts, N. W., Roulin,
A., Rowland, H. M., Sherratt, T. N., … Caro, T. (2017). The biology of
color. Science, 357(6350), eaan0221. https://doi.org/10.1126/scien
Endler, J. A. (1978). A predator’s view of animal color patterns. In In
Evolutionary biology (pp. 319– 364). Boston, MA: Springer.
Endler, J. A. (1990). On the measurement and classification of colour in
studies of animal colour patterns. Biological Journal of the Linnean
Society, 41, 315– 352. https://doi.org/10.1111/j.1095- 8312.1990.
Endler, J. A., & Basolo, A. L. (1998). Sensory ecology, receiver biases
and sexual selection. Trends in Ecology and Evolution, 13, 415– 420.
h t t p s : / / d o i . o r g / 1 0 . 1 0 1 6 / S 0 1 6 9 - 5 3 4 7 ( 9 8 ) 0 1 4 7 1 - 2
Ficarrotta, V., Hanly, J. J., Loh, L. S., & Francescutti, C. M. (2022). A ge-
netic switch for male UV- iridescence in an incipient species pair of
sulphur butterflies. Proceedings of the Nationa l Academy of Science s
of the United States of America, 119, 1– 7. htt ps://doi.org/10.1073/
Fisher, R. A., & Ford, E. B. (1947). The spread of a gene in natural condi-
tions in a colony of the moth Panaxia dominula. Heredity, 1, 1 4 3 – 1 7 4 .
Ford, E. B. (1945). Polymorphism. Biological Reviews, 20, 73– 88. https://
doi.org/10.1111/j.1469- 185X.1945.tb003 15.x
Ford, E. B. (1965). Genetic polymorphism. Faber and Faber Limited.
Forsman, A . (200 0). Some like it hot: Intra- population variation in behav-
ioral thermoregulation in color- polymorphic pygmy grasshoppers.
Evolutionary Ecolog y, 14, 25– 38. htt ps://doi.org/10.1023/A:10110
Freitak, D., Vanatoa, A., Ots, I., & Rantala, M. J. (2005). Formation
of melanin- based wing patterns is influenced by condi-
tion and immune challenge in Pieris brassicae. Entomologia
Experimentalis Et Applicata, 116, 237– 243. https://doi.
org /10.1111/j .1570- 7458.200 5.00 33 0. x
Galarza, J. A. (2021). Comparative transcriptomics of albino and
warningly- coloured caterpillars. Ecolog y and Evolution, 11, 7507–
7517. https://doi.or g/10.1002/ece3.7581
NOKEL AINEN E t AL.
Galarza, J. A., Nokelainen, O., Ashrafi, R., Hegna, R. H., & Mappes, J.
(2014). Temporal relationship between genetic and warning signal
var iation in the aposemati c wood tiger moth (Parasemia plantaginis).
Molecular Ecology, 23, 4939– 4957.
Gazda, M. A., Araújo, P. M., Lopes, R. J., Toomey, M. B., Andrade, P.,
Afonso, S., Marques, C., Nunes, L., Pereira, P., Trigo, S., Hill, G. E.,
Corbo, J. C., & Carneiro, M. (2020). A genetic mechanism for sex-
ual dichromatism in birds. Science, 368, 1270– 1274. https://doi.
Gillespie, R. G., & Oxford, G. S. (2009). Selection on the color polymor-
phism in Hawaiian happy- face spiders: Evidence from genetic struc-
ture and temporal fluctuations. Evolution, 52, 775– 783. https://doi.
org /10.1111/j .1558- 56 46.1998.tb037 01.x
Gordon, S. P., Burdillat, S., & Mappes, J. (2018). Phenotype- dependent
mate choice and the influence of mixed- morph lineage on the repro-
ductive success of a polymorphic and aposematic moth. Evolutionary
Ecology, 32, 4 2 7 – 4 4 1 . h t t p s : / / d o i . o r g / 1 0 . 10 0 7 / s 1 0 6 8 2 - 0 1 8 - 9 9 4 4 - 5
Gordon, S. P., Kokko, H., Rojas, B., Nokelainen, O., & Mappes, J. (2015).
Colour polymorphism torn apart by opposing positive frequency-
dependent selection, yet maintained in space. Journal of Animal
Ecology, 84, 1555– 1564. htt ps://doi.org/10 .1111/1365- 2656.12416
Gray, S. M., & McKinnon, J. S. (2007). Linking color polymorphism main-
tenance and speciation. Trends in Ecology & Evolution, 22, 71– 79.
Ham, A. D., Ihalainen, E., Lindström, L., & Mappes, J. (2006). Does colour
matter? The importance of colour in avoidance learning, memora-
bility and generalisation. Behavioral Ecology and Sociobiology, 60,
4 8 2 – 4 9 1 . h t t p s : / / d o i . o r g / 1 0 . 1 0 0 7 / s 0 0 2 6 5 - 0 0 6 - 0 1 9 0 - 4
Hanlon, R., Chiao, C.- C., Mathger, L., Barbosa, A., Buresch, K., & Chubb,
C. (2009). Cephalopod dynamic camouflage: bridging the contin-
uum between background matching and disruptive coloration.
Philosophical Transac tions of the Royal Societ y B: Biological Sciences,
364, 429– 437. https://doi.org/10.1098/rstb.2008.0270
Hart, N. S. (2001a). The visual ecology of avian photoreceptors. Progress
in Retinal a nd Eye Research, 20, 675– 703. https://doi.or g/10.1016/
S 1 3 5 0 - 9 4 6 2 ( 0 1 ) 0 0 0 0 9 - X
Hart, N. S. (2001b). Variations in cone photoreceptor abundance and the
visual ecology of birds. Journal of Comparative Physiology, 187, 685–
6 9 7 . h t t p s : / /d o i . o r g / 1 0 . 1 0 0 7 / s 0 0 3 5 9 - 0 0 1 - 0 2 4 0 - 3
Hegna, R. H., Galarza, J. A., & Mappes, J. (2015). Global phylogeography
and geographical variation in warning coloration of the wood tiger
moth (Parasemia plantaginis). Journal of Biogeography, 42, 1469– 14 81.
Hegna, R. H., Nokelainen, O., Hegna, J. R., & Mappes, J. (2013). To quiver
or to shiver: Increased melanization benefits thermoregulation, but
reduces warning signal efficacy in the wood tiger moth. Proceedings
of the Royal Society B: Biological Sciences 280, 20122812.
Henze, M. J., Lind, O., Mappes, J., Rojas, B., & Kelber, A. (2018). An
aposematic colour- polymorphic moth seen through the eyes of
conspecifics and predators – Sensitivity and colour discrimination
in a tiger moth. Functional Ecology, 32(7), 1797– 1809. https://doi.
org /10.1111/1365- 2435.13100
Hofer, H., Carroll, J., Neitz, J., Neitz, M., & Williams, D. R. (2005).
Organization of the human trichromatic cone mosaic. Journal of
Neuroscience, 25, 9669– 9679. https://doi.org/10.1523/JNEUR
Høye, T. T., Ärje, J., Bjerge, K., Hansen, O. L. P., Iosifidis, A., Leese, F.,
Mann, H. M. R., Meissner, K., Melvad, C., & Raitoharju, J. (2021).
Deep learning and computer vision will transform entomology.
Proceedi ngs of the National Academy of Sciences of the United States
of America, 118, 1– 10. https://doi.org/10.1073/pnas.20025 45117
Huxley, J. (1955). Morphism and evolution. Heredity, 9, 1– 52. https://doi.
Kang, C., Stevens, M., Moon, J. Y., Lee, S. I., & Jablonski, P. G. (2015).
Camouflage through behavior in moths: The role of background
matching and disruptive coloration. Behavioral Ecology, 26, 45– 54.
Kappers, E. F., de Vries, C., Alberda, A., Forstmeier, W., Both, C., &
Kempenaers, B. (2018). Inheritance patterns of plumage coloration
in common buzzards Buteo buteo do not support a one- locus two-
allele model. Biology Letters, 14, 20180007.
Kelber, A., Vorobyev, M., & Osorio, D. (2003). Animal colour vision – be-
havioural tests and physiological concepts. Biological Reviews of the
Cambridge Philosophical Society, 78, 81– 118.
Kettlewell, H. B. (1955). Recognition of appropriate backgrounds by
the pale and black phases of Lepidoptera. Nature, 175, 943– 944.
Kursa, M. B., & Rudnicki, W. R. (2010). Feature selection with the boruta
package. Journal of Statistical Softwa re, 36, 1– 13.
Kuznetsova, A., Brockhoff, P. B., & Christensen, R. H. B. (2017). lmerTest
Package: Tests in linear mixed effects models. Journal of Statistical
Software, 82, 1– 26.
Le Rouzic, A., Hansen, T. F., Gosden, T. P., & Svensson, E. I. (2015).
Evolutionar y time- ser ies analy sis reveal s the signatur e of fr equency-
dependent selection on a female mating polymorphism. The American
Naturalist, 185, E182– E196. https://doi.org/10.1086/680982
Liebert, T. G., & Brakefield, P. M. (1990). The genetics of colour polymor-
phism in the aposematic Jersey Tiger Moth Callimorpha quadripunc-
taria. Heredity, 64, 87– 92. https://doi.org/10.1038/hdy.1990.11
Lindstedt, C., Eager, H., Ihalainen, E., Kahilainen, A ., Stevens, M., &
Mappes, J. (2011). Direction and strength of selection by predators
for the color of the aposematic wood tiger moth. Behavioral Ecology,
22, 580– 587. https://doi.org/10.1093/behec o/arr017
Lindstedt, C., Lindström, L., & Mappes, J. (2009). Thermoregulation con-
strains effective warning signal expression. Evolution, 63, 469– 478.
https://doi.or g/10.1111/ j.1558- 5646.200 8. 00561.x
Lindstedt, C., Reudler, J. H., Ihalainen, E., Lindström, L., & Mappes, J.
(2010). Diet quality affects warning coloration indirectly: excretion
costs in a generalist herbivore. Evolution, 64, 68– 78. https://doi.
org /10.1111/j .1558- 56 46.2009.0 0796.x
Lindstedt, C., Schroderus, E., Lindström, L., Mappes, T., & Mappes, J.
(2016). Evolutionar y constraints of warning signals: a genetic trade-
off between the efficacy of lar val and adult warning coloration can
maintain variation in signal expression. Evolution, 70(11), 2562–
2572 . https://doi.org/10.1111/evo.13066
Maia, R., Eliason, C. M., Bitton, P. P., Doucet, S. M., & Shawkey, M. D.
(2013). pavo: An R package for the analysis, visualization and or-
ganization of spectral data. Methods in Ecology and Evolution, 4,
Matsuoka, Y., & Monteiro, A. (2018). Melanin pathway genes regulate
color and morpholog y of butterfly wing scales. Cell Reports, 24, 56–
McKinnon, J. S., & Pierotti, M. E. R . (2010). Colour polymor-
phism and correlated characters: genetic mechanisms and
evolution. Molecular Ecology, 19, 5101– 5125. https://doi.
org /10.1111/j .1365- 294X. 2010.04 84 6. x
Merrill, R. M., Wallbank, R. W. R., Bull, V., Salazar, P. C. A., Mallet, J.,
Stevens, M., & Jiggins, C. D. (2012). Disruptive ecological selec-
tion on a mating cue. Proceedings of the Royal Societ y B: Biological
Sciences, 279, 4907– 4913. https://doi.org/10.1098/rspb.2012.1968
Mochida, K. (2011). Combination of local selection pressures drives di-
versity in aposematic signals. Evolutionary Ecolog y, 25, 1017– 1028.
h t t p s : / / d o i . o r g / 1 0 . 1 0 0 7 /s 1 0 6 8 2 - 0 1 1 - 9 4 7 1 - 0
Nokelainen, O., de Moraes Rezende, F., Valkonen, J. K., & Mappes, J.
(2021). Context- dependent coloration of prey and predator deci-
sion making in contrasting light environments. Behavioral Ecology,
33(1), 77– 86. https://doi.org/10.1093/behec o/arab111
Nokelainen, O., Hegna, R. H., Reudler, J. H., Lindstedt, C., & Mappes, J.
(2 012 ). Trade- off be tween wa rni ng si gna l effi cacy and m ating s u cce ss
in the wood tiger moth. Proceedings of the Royal Society B: Biological
Sciences, 279, 257– 265. https://doi.org/10.1098/rspb.2011.0880
Nokelainen, O., Lindstedt, C., & Mappes, J. (2013). Environment-
mediated morph- linked immune and life- history responses in the
NO KELA INEN Et A L.
aposematic wood tiger moth. Journal of Animal Ecology, 82, 653–
662. ht tp s://doi.o rg /10.1111/1365- 2656.12037
Nokelainen, O., Maynes, R., Mynott, S., Price, N., & Stevens, M. (2019).
Improved camouflage through ontogenetic colour change con-
fers reduced detection risk in shore crabs. Functional Ecology 33,
65 4– 6 69.
Nokelainen, O., Stevens, M., & Caro, T. (2018). Colour polymorphism
in the coconut crab (Birgus latro). Evolutionary Ecology, 32, 75– 88.
h t t p s : / / d o i . o r g / 1 0 . 1 0 0 7 /s 1 0 6 8 2 - 0 1 7 - 9 9 2 4 - 1
Nokelainen, O., Valkonen, J., Lindstedt, C., & Mappes, J. (2014). Changes
in predator community structure shifts the efficacy of two warn-
ing signals in Arctiid moths. Journal of Animal Ecology, 83, 598– 605.
https://doi.or g/10.1111/1365- 2656.12169
Osorio, D., & Vorobyev, M. (2005). Photoreceptor sectral sensitivities
in terrestrial animals: adaptations for luminance and colour vision.
Proceedi ngs of the Royal Society B: Biological Sciences, 272, 1745–
1752. ht tps://doi.org/10.1098/rspb.2005.3156
Price, T. D. (2006). Phenotypic plasticity, sexual selection and the evolu-
tion of colour patterns. Journal of Experimental Biology, 209, 2368–
R Core Team (2018). R: A Language and environment for statistical com-
Rodriguez- Caro, F., Fenner, J., Bhardwaj, S., Cole, J., Benson, C.,
Colombara, A. M., Papa, R., Brown, M. W., Martin, A., Range, R.
C., & Counterman, B. A. (2021). Novel doublesex duplication asso-
ciated with sexually dimorphic development of dogface butterfly
wings. Molecula r Biolog y and Evolution, 38, 5021– 5033.
Rojas, B., Burdfield- Steel, E., De Pasqual, C., Gordon, S., Hernández,
L., Mappes, J., Nokelainen, O., Rönkä, K., & Lindstedt, C. (2018).
Multimodal aposematic signals and their emerging role in mate
attraction. Frontiers in Ecology and Evolutio n, 6, 1– 24. https://doi.
Rojas, B., Rautiala, P., & Mappes, J. (2014). Differential detectability of
polymorphic warning signals under varying light environments.
Behavioural Processes, 109, 164– 172. https://doi.org/10.1016/j.
Rönkä, K., De Pasqual, C., Mappes, J., Gordon, S., & Rojas, B. (2018).
Colour alone matters: no predator generalization among morphs of
an aposematic moth. Animal Behaviour, 135, 153– 163. https://doi.
Rönkä, K., Mappes, J., Kaila, L., & Wahlberg, N. (2016). Putting Parasemia
in its phylogenetic place: a molecular analysis of the subtribe
Arctiina (Lepidoptera). Systematic Entom olog y 41, 844– 853.
Rönkä, K., Valkonen, J. K., Nokelainen, O., Rojas, B., Gordon, S.,
Burdfield- Steel, E., & Mappes, J. (2020). Geographic mosaic of se-
lection by avian predators on hindwing warning colour in a poly-
morphic aposematic moth. Ecology Letters, 23, 1654– 1663. ht tps://
RStudio Team (2016). RStu dio: Integrated Development Enviro nment for R .
Ruxton, G., Allen, W., Sherratt, T., & Speed, M. (2019). Avoiding at tack:
The evolutionary ecology of crypsis, apose matism, a nd mimicry.
Oxford University Press.
Siddiqi, A., Cronin, T. W., Loew, E. R., Vorobyev, M., & Summers, K. (2004).
Interspecific and intraspecific views of color signals in the straw-
berry poison frog Dendrobates pumilio. The Journal of E xperimental
Biology, 207, 2471– 2485.
Sinervo, B., & Calsbeek, R. (2006). The developmental, physiological,
neural and genetical causes and consequences of frequency de-
pendent selection in the wild. An nual Review of Ecolog y, Evolution,
and Systematics, 37, 581– 610. https://doi.org/10.1146/annur
Sinervo, B., & Lively, C. M. (1996). The rock- paper- scissors game and
the evolution of alternative male strategies. Nature, 380, 240– 243.
Stevens, M., Párraga, C. A., Cuthill, I. C., Partridge, J. C., & Troscianko,
T. S. (2007). Using digital photography to study animal coloration.
Biological Journal of the Linnean Society, 90, 211– 237. https://doi.
org /10.1111/j .1095- 8312 .2 007.00725.x
Stoddard, M. C., & Stevens, M. (2010). Pattern mimicry of host eggs by
the common cuckoo, as seen through a bird’s eye. Proceedings of
the Royal Society B: Biological Sciences, 277, 1387– 1393. ht tps://doi.
Stuart- Fox, D., Newton, E., & Clusella- Trullas, S. (2017). Thermal con-
sequences of colour and near- infrared reflectance. Philosophical
Transactions of the Royal Society B: Biological Sciences, 372, 1– 8.
Suomalainen, E. (1938). Die erblichkeitverhältnisse des männlichen di-
morphismus bei parasemia plantaginis. Hereditas, 24, 386– 390.
Svensson, E. I. (2017). Back to basics: Using colour polymorphisms to
study evolutionary processes. Molecular Ecology, 26(8), 2204– 2211.
https://doi.or g/10.1111/m ec .14025
Svensson, E. I., & Abbott, J. (2005). Evolutionary dynamics and population
biology of a polymorphic insect. Journal of Evolutionar y Biology, 18,
1503– 1514. https://doi.org/10.1111/j.1420- 9101.2005.00946.x
Sword, G. A ., Simpson, S. J., El Hadi, O. T. M., & Wilps, H. (2000).
Density- dependent aposematism in the desert locust. Proceedings
of the Royal Society B: Biological Sciences, 267, 63– 68. https://doi.
Tib bett s, E. A., Mullen, S. P., & Dale, J. (2017). Signal function drives ph e-
notypic and genetic diversity: the effects of signalling individual
identit y, quality or behavioural strategy. Philosophical Transactions
of the Royal Society B: Biological S ciences, 372(1724), 20160 347–
Troscianko, J., & Stevens, M. (2015). Image Calibration and Analysis
Toolbox – a free software suite for objectively measuring reflec-
tance, colour and pattern. Methods in Ecology and Evolution, 6,
1320 – 13 31. https://doi .org/10.1111/20 41- 210X.12439
van den Berg, C. P., Troscianko, J., Endler, J. A., Marshall, N. J., & Cheney,
K. L. (2020). Quantitative Colour Pattern Analysis (QCPA): A com-
prehensive framework for the analysis of colour patterns in na-
ture. Methods in Ecolog y and Evolution, 11, 316– 332. https://doi.
org /10.1111/2041- 210X .13328
Vorobyev, M., & Osorio, D. (1998). Receptor noise as a determinant of
colour threshoIds. Proceedings of the Royal Society B: Biological
Sciences, 265, 351– 358.
Vorobyev, M., Osorio, D., Bennet t, A. T. D., Marshall, N. J., & Cuthill, I.
C. (1998). Tetrachromacy, oil droplets and bird plumage colours.
Journal of Comparative Physiology, 183, 621– 633. https://doi.
White, T. E., & Kemp, D. J. (2016). Colour polymorphism. Current Biology,
26, R517– R518. https://doi.org/10.1016/j.cub.2016.03.017
Wilkins, L., & Osorio, D. (2019). Object colours, material properties and
animal signals. Journal of Experimental Biology, 222, jeb204 487.
Winters, A. E., Lommi, J., Kirvesoja, J., & Nokelainen, O. (2021).
Multimodal aposematic defenses through the predation sequence.
Frontiers in Ecology and Evolution, 9, 1– 18. https://doi.org/10.3389/
Additional supporting information may be found in the online
version of the article at the publisher’s website.
How to cite this article: Nokelainen, O., Galarza, J. A.,
Kirvesoja, J., Suisto, K., & Mappes, J. (2022). Genetic colour
variation visible for predators and conspecifics is concealed
from humans in a polymorphic moth. Journal of Evolutionary
Biology, 35, 467– 478. htt ps://doi.org/10.1111/jeb.13994