Article

Mapping the adaptive landscape of Batesian

mimicry using 3D-printed stimuli

https://doi.org/10.1038/s41586-025-09216-3 Christopher H. Taylor1 ✉, David James George Watson1, John Skelhorn2, Danny Bell1,
Simon Burdett1, Aoife Codyre1, Kathryn Cooley1, James R. Davies1,3, Joshua Joseph Dawson1,
Received: 27 March 2024
Tahiré D’Cruz1, Samir Raj Gandhi1,4, Hannah J. Jackson1, Rebecca Lowe1, Elizabeth Ogilvie1,
Accepted: 29 May 2025 Alexandra Lei Pond1, Hallie Rees1, Joseph Richardson1, Joshua Sains1, Francis Short1,
Christopher Brignell5, Gabrielle L. Davidson6,7, Hannah M. Rowland8,9, Mark East10,
Published online: xx xx xxxx
Ruth Goodridge10, Francis Gilbert1 & Tom Reader1
Open access

Check for updates

In a classic example of adaptation, harmless Batesian mimics gain protection from
predators through resemblance to one or more unpalatable models1,2. Mimics vary
greatly in accuracy, and explaining the persistence of inaccurate mimics is an ongoing
challenge for evolutionary biologists3,4. Empirical testing of existing hypotheses is
constrained by the difficulty of assessing the fitness of phenotypes absent among
extant species, leaving large parts of the adaptive landscape unexplored5—a problem
affecting the study of the evolution of most complex traits. Here, to address this,
we created mimetic phenotypes that occupy hypothetical areas of trait space by
morphing between 3D images of real insects (flies and wasps), and tested the
responses of real predators to high-resolution, full-colour 3D-printed reproductions
of these phenotypes. We found that birds have an excellent ability to learn to
discriminate among insects on the basis of subtle differences in appearance, but this
ability is weaker for pattern and shape than for colour and size traits. We found that
mimics gained no special protection from intermediate resemblance to multiple
model phenotypes. However, discrimination ability was lower in some invertebrate
predators (especially crab spiders and mantises), highlighting that the predator
community is key to explaining the apparent inaccuracy of many mimics.

Batesian mimics gain protection when predators treat them as defended observe extant species, we see only small sections of the adaptive
‘models’ despite being palatable prey1,2. As this deception of preda- landscape, and miss an opportunity to examine the fitness of phe-
tors relies on a degree of perceived similarity, increasing resemblance notypes that do not currently exist. One successful solution to this
should give a higher probability of misidentification. Yet mimics vary issue is to manipulate existing phenotypes, for example, by painting
greatly in accuracy3,6, raising the question of what stops ever-greater or covering real organisms to change their appearance18, or creating
mimetic accuracy from evolving7. Numerous theoretical explanations8 artificial replicas19, but the manipulations involved tend to be limited
have proposed functional trade-offs affecting mimetic appearance9,10 in range and realism19–21.
and factors that might cause relaxed selection for accuracy4 such as To overcome these limitations, we generated stimuli combining
predators’ inability to detect differences between mimics and models11 the relevance and realism of working with real insects, full three-
or reduced motivation to discriminate12. Many of these hypotheses are dimensional (3D) representation and the power to manipulate fine
untested experimentally in realistic systems, and there is no consensus details of visual phenotypes. Hoverflies (Syrphidae) are a classic study
about the causes of variation in mimetic accuracy. system to provide reference points for our stimuli, with Batesian mim-
The expected outcomes of selection on visual adaptations depend icry of wasps (Vespidae) varying across species from near-perfect,
on the specific characteristics of signallers and receivers in each through approximate, to non-existent mimicry7 (Extended Data
study system13. Different visual receivers interpret the same colour Table 1). We used 3D scans of real model (wasp) and mimic (hoverfly)
patterns in different ways14; some features are more easily associ- species as starting points to define axes of variation within multivariate
ated with a reward than others by a given receiver15; and systems with phenotypic space, and generated gradients of mimetic similarity by
more stimulus types elicit more generalized responses16,17. Thus, while smoothly manipulating visual traits (shape, colour, pattern and size)
experiments using abstract stimuli can illuminate general principles, along those axes. We then used additive manufacturing (3D printing)
we must test these principles in realistic systems. However, when we to turn these images into physical stimuli, enabling us to explore the
1
School of Life Sciences, University of Nottingham, Nottingham, UK. 2Biosciences Institute, Faculty of Medical Sciences, Newcastle University, Newcastle upon Tyne, UK. 3School of Biological
Sciences, University of Bristol, Bristol, UK. 4The Jolly Geographer, Babraham, UK. 5School of Mathematical Sciences, University of Nottingham, Nottingham, UK. 6Department of Psychology,
University of Cambridge, Cambridge, UK. 7School of Biological Sciences, University of East Anglia, Norwich, UK. 8Predators and Toxic Prey Research Group, Max Planck Institute for Chemical
Ecology, Jena, Germany. 9Department of Evolution, Ecology and Behaviour, Institute of Infection, Veterinary and Ecological Sciences, University of Liverpool, Liverpool, UK. 10Faculty of Engineering,
University of Nottingham, Nottingham, UK. ✉e-mail: [email protected]

Nature | www.nature.com | 1
Article
a Photogrammetry Generation of intermediates Additive manufacture

3D-textured Diffeomorphic Transects of

Specimens Shape Stimuli
meshes mapping mimetic accuracy

Distance
Pattern
transform Fly (M100) Wasp (V100)
Fly (M100) Wasp (V100)
Linear
Size
interpolation

RGB M50/V50
Colour
interpolation

Legs/
Simplified
wings/
version
antennae

b c d e f Legs Wings Antennae

g
11.8 mm Black Yellow
R G B R G B
Fly (M100)
53 50 62 190 154 88
11.3 mm

M50/V50 57 49 51 188 158 84

10.8 mm

Wasp (V100) 61 49 41 187 162 80

Fig. 1 | Overview of the methods used to generate artificial mimetic stimuli. intermediates were then recombined into a single 3D object, and finally printed
Illustrations are based on an axis of similarity from the fly M. meridiana (M100) using additive manufacturing. b–g, How each component of the phenotype
to the wasp V. vulgaris (V100) through a 50% intermediate (M50/V50). a, Flowchart (shape (b), pattern (c), size (d) and colour (e) of the body, plus simplified
of the methodology. Real insect specimens were scanned by photogrammetry representations of legs, wings and antennae (f)) is interpolated, and the
to produce coloured 3D meshes. These were split into multiple components, resulting axis of mimetic accuracy (g).
which were each varied smoothly to generate intermediate values. These

fitness landscape of mimetic accuracy beyond the examples seen in Table 1). We selected three intermediate points on each axis corre-
nature. sponding to equally spaced values of shape, colour, pattern and size.
To test key hypotheses about the existence of inaccurate Batesian For example, S25/V75 indicates a stimulus of 25% based on S. ribesii and
mimics, we first tested the extent to which wild predators discriminate 75% based on V. vulgaris. To the human eye, the most accurate of these
highly accurate, yet imperfect, mimics from their models. We then mimics appears considerably more like V. vulgaris than any existing
tested whether the presence of more than one model species affords hoverfly. The axis M. meridiana to V. vulgaris, viewable digitally in 3D,
increased protection to intermediate mimics (the multiple-models is provided in Supplementary Data 1.
hypothesis12). We tested the relative signal salience22 of shape, colour, The focal predators were wild, free-living great tits (Parus major) in
pattern and size by varying those traits independently and testing which Madingley Wood, Cambridge, UK. Great tits are generalist predators
are under the strongest selection for accuracy. Finally, we tested the of Hymenoptera and Diptera25. We trained them to forage from feeding
eye-of-the-beholder hypothesis11 by comparing the responses of a range stations presenting arrays of small opaque dishes covered by openable
of insectivores towards the same set of mimetic stimuli. lids that concealed a mealworm (Tenebrio molitor). We then fixed 3D
stimuli to the lids to signal the reward status: half of the dishes displayed
a non-mimetic fly stimulus (M100) and contained a mealworm, and half
Discrimination ability displayed a model wasp stimulus (V100) with no reward (Fig. 2a). The
In the few tests of predator discrimination between real models and dishes were reset daily for a new session with a randomized arrange-
mimics, birds consistently distinguish images or specimens of wasps ment of stimuli. Birds visited the feeding stations repeatedly during
from any tested hoverfly, including some seemingly accurate Bate- a session and could select among any unopened dishes at any time.
sian mimics23,24. However, using real specimens or images can never We assumed that the birds attempted to maximize their rate of food
reveal the possible protection of a hypothetical mimic even closer to consumption26 to minimize opportunity costs and predation risk, start-
the model phenotype. The degree to which the fundamental limits ing with the dishes that they perceived most likely to be rewarding. We
of predator perception and cognition constrains decisions to attack therefore estimated the level of protection of each stimulus according
mimicry complexes is therefore uncertain. to how early or late in the sequence that dish was opened.
We generated stimuli (Fig. 1) along three axes, with each axis using During the first day of training, the birds showed no bias towards
the common wasp Vespula vulgaris to represent the aversive model at either stimulus (47% for M100 and 53% for V100 among the first 15
one end point, denoted V100 (100% based on V. vulgaris). Each axis used dishes per feeder; binomial test, P = 0.38; n = 45), suggesting that pre-
a different fly (Diptera) taxon as the other end point: the non-mimic vious encounters with real wasps or flies did not influence foraging
Mesembrina meridiana (M100), intermediate mimic Syrphus ribesii choices (this was winter, when the birds had probably not encoun-
(S100) and accurate mimic Chrysotoxum spp.3 (C100; Extended Data tered wasps for several months). After 3 weeks of training, the birds

2 | Nature | www.nature.com
 a Training phase We next presented the full range of stimuli from all three axes, with
approximately one-third being unrewarding V100 dishes, one-fifth
M100 V100
being rewarding M100 dishes and the remainder being rewarding mimic
dishes sampled from the other axis points (Fig. 2c and Supplementary
Video 1). Birds generalized immediately from their learned preference
towards the fly dishes, targeting the least-accurate stimuli first and
15 15
moving on to the more accurate mimics when no other options were
b available (Extended Data Fig. 1b). With further opportunity for learning,
the birds increased their discrimination between mimics and models
1 D D D (Extended Data Fig. 1b,c). After around 10 days, the responses stabilized
E
and the birds discriminated all mimics from models, targeting V100
Level of protection

C C
dishes significantly later than all other stimulus types, and avoiding
BC BC
0 B the 75% wasp-like mimics more than those 50% and below (n = 3,071
AB AB AB presentations; Fig. 2b and Extended Data Table 2).
A
All mimetic stimuli were prioritized over wasp stimuli, with a trend
–1 of increasingly wasp-like stimuli receiving greater protection (Fig. 2b),
despite all mimics being associated with the same reward. This aligns
with signal detection theory, which predicts that, when multiple prey
–2 626126124 126126126 126 126123 126124124 1,068 types are available, predators should respond more cautiously towards
perceived signals that are more likely to have originated from a model27.
Fly 75% fly 50:50 75% wasp Wasp
Moreover, the optimal response will be more cautious when models are
c Testing phase more abundant, and/or when models carry a more severe cost28. For
our birds, the consequence of incorrectly targeting a model is a small
M100 M75V25 M50V50 M25V75
opportunity cost (if, for example, another bird gets to a mealworm
Mesembrina

first): ethical and practical constraints mean that higher costs would
be difficult to implement. In nature, wasps potentially impose a more
10 2 2 2 severe cost (a sting), but also bring potential nutritional rewards, so
the resulting protection would depend on quantifying these outcomes.
S100 S75V25 S50V50 S25V75 V100 In a separate validation experiment, we investigated the extent to
which our printed stimuli were treated like real insects by the birds. We
Syrphus

again trained them to discriminate flies from wasps, then substituted

half of the printed stimuli with dead specimens of real flies and wasps.
2 2 2 2 17 Birds targeted the printed flies first (the learned reward in the training
phase), followed by the real flies (novel but rewarding), with printed
C100 C75V25 C50V50 C25V75 and real wasps (both unrewarding) receiving the same level of protec-
Chrysotoxum

tion (Extended Data Fig. 2). Thus, the birds distinguished between at
least some of the printed stimuli and their real-life equivalents—which,
considering our other results (Fig. 2), is unsurprising—but generalized
2 2 2 2 from printed stimuli to real insects, indicating that they recognize a
commonality between the two.
Fig. 2 | Discrimination ability of great tits. a, The design of the training phase.
Predictions of optimal discrimination levels are only meaningful if
The solid blue border represents the rewarded stimulus, and the dashed red
predators possess the sensory and cognitive abilities to achieve those
border represents no reward. The code above is a unique label for each stimulus
levels. Our birds could detect and remember very subtle differences
type, with the letters indicating the species used as an end point and the numbers
indicating the weighting. The numbers below indicate how many of that stimulus
between mimic and model appearance, and used these differences to
type appeared on a single feeding station, within an array of 30 dishes. b, The select rewarding over unrewarding prey. These results align with theo-
levels of protection received by different stimulus types resulting from great retical predictions that, given enough experience29, signal receivers
tit behaviour during the testing phase. Level of protection is the rank order of should have the ability to discriminate between tiny differences. Our
attack for that stimulus within a session, logit transformed. A higher level of findings provide a crucial baseline for studies of mimicry, rejecting the
protection indicates that a stimulus was attacked later in the sequence, or not argument that inaccurate mimics are already sufficiently accurate to
at all. The bold points show the mean for all feeders; faint points show the be indistinguishable from their models, given favourable conditions
means for individual feeders along with 95% confidence intervals based on the for the predator. This implies that, if a bird chooses to avoid inaccu-
t-distribution. The black line shows the Mesembrina axis, the blue line shows rate mimics, it may be driven more by its motivation28,30 than a lack of
the Syrphus axis and the orange line shows the Chrysotoxum axis. Capital letters ability. Moreover, certain factors might increase the level of sensory
indicate groupings that show no significant difference after a Tukey post hoc
or cognitive difficulty of a discrimination task, such as separation of
test (P > 0.05). Numbers give the sample size (number of dishes). c, Design of the
prey in time or space preventing side-by-side comparison31, prey move-
testing phase. Models from three axes were presented together at each feeding
ment32 or the existence of multiple model species, as we explore next.
station, starting from three different fly species and all ending at V. vulgaris.
The total array size was 49 dishes. Note that M100 has a larger sample size than
S100 and C100 so that the birds faced some rewarding stimuli familiar from the
training phase.
Multiple models
Evidence from studies of Müllerian mimicry suggests that more com-
plex prey communities cause predators to use broader generalization
consistently targeted M100 first and either opened V100 dishes last or and a more conservative foraging strategy16,33. The addition of a second
not at all (M100 88% V100 12% among first 15 dishes per feeder across model species to a mimicry system may increase the fitness of inter-
the final 3 days of training; binomial test, P < 0.0001; n = 135; Extended mediate Batesian mimics, although these ‘jacks of all trades’ may not
Data Fig. 1a). necessarily outperform perfect mimics of either model12. Despite the

Nature | www.nature.com | 3
Article
clear existence of model diversity in nature34, almost all experimental a b 0.2

Level of protection
studies of Batesian mimicry use a single model phenotype (but see
M100 A100 V100 0
refs. 35,36), and a key prediction from the multiple-models hypothesis
remains untested regarding whether a mimic intermediate between two –0.2
model phenotypes gains greater protection than one with an equivalent
1M: 25 0 24 –0.4 1,824 941 795
level of accuracy to one model, but further removed from the second. 1,244 348 311 191 145 188
2M: 25 12 12
To test this, we presented two distinct model (that is, unrewarding) 0 50 100 150
stimuli to the birds: the common wasp V. vulgaris (V100) and a solitary Distance to a model
wasp Argogorytes mystaceus (A100; Extended Data Table 1). The latter c
is also defended by a sting and displays black-and-yellow warning col-
ours, but differs from the common wasp in appearance. We generated
an axis including three stimuli intermediate between the two models 0.5

(A75/V25, A50/V50 and A25/V75) and a further two stimuli at each

end extrapolated along the same trajectory (A125/V−25, A150/V−50,

Level of protection
A−25/V125 and A−50/V150). From the multiple-models hypothesis,
0
the intermediate stimuli should receive greater protection than the
extrapolated stimuli, despite equivalent similarity to a single model,
due to their resemblance to the second model species. For example,
A50/V50 and A−50/V150 are both 50 units from the model V100, but
–0.5
A50/V50 is much closer to the second model A100 (50 units) than A−50/
V150 is (150 units). If there is any additive effect of mimicry to the two
models, the intermediate A50/V50 should receive greater protection
916 242 244 915 242 278 244 909 211 275
than the extrapolated A−50/V150. –1.0 616 188 145 191 142 186 1,244 162 169
We trained wild, free-living great tits and blue tits (Cyanistes caer-
uleus; the latter making a low proportion of foraging visits; Methods) A150 A100 A50 A–50
to avoid model stimuli using the same approach as in the discrimina- V–50 V50 V100 V150
tion ability experiment. Six feeding stations were divided among two
treatments: three with a single unrewarding model stimulus V. vulgaris
V100 (1M treatment) and three including a second unrewarding model 2 or 3 0 2 or 3 20 2 or 3
stimulus A. mystaceus A100 (2M treatment; Fig. 3a). The inclusion of 2 or 3 10 2 or 3 10 2 or 3
the 1M treatment acted as a control, enabling us to compare directly
whether the addition of a second model stimulus in 2M alters the protec- M100 A125 A75 A25 A–25
V–25 V25 V75 V125
tion received by any of the mimetic stimuli. Both treatments included
a rewarding non-mimic fly stimulus M. meridiana (M100). Once birds
were consistently targeting the M100 stimuli first (Extended Data
Fig. 3a), we introduced the intermediate and extrapolated stimuli as 10 2 or 3 2 or 3 2 or 3 2 or 3
10 2 or 3 2 or 3 2 or 3 2 or 3
rewarding (Batesian) mimics, alongside existing stimuli.
At the start of the testing phase, preference for M100 stimuli was
Fig. 3 | Testing the multiple-models hypothesis. a, The design of the training
strong, but this weakened over approximately 10 days before reaching phase. The solid blue border represents rewarded stimulus, and the dashed red
an asymptote with lower levels (but not an absence) of discrimination border represents no reward. The code above is a unique label for each stimulus
among stimuli (Extended Data Fig. 3b,c). Protection received by the type, with the letters indicating the species used as an end point and the numbers
various mimetic stimuli declined with increasing distance from the indicating the weighting (which is negative in some cases). The numbers below
nearest model (n = 5,987 presentations; distance term ΔAICc = −28.9, indicate how many of that stimulus type appeared on a single feeding station,
d.f. = 1; Fig. 3b and Extended Data Table 3). This pattern was similar for both one-model (1M, top) and two-model (2M, bottom) treatments. b, The
across both treatments (treatment × distance term ΔAICc = 1.4, d.f. = 1) levels of protection received by different stimulus types resulting from great
and there was no increase in protection for intermediate as opposed tit behaviour during the testing phase. Stimuli are grouped according to
to extrapolated stimuli (intermediate × distance term ΔAICc = 3.2, phenotypic distance to the nearest model. Level of protection is the rank order
d.f. = 2). of attack for that stimulus within a session, logit transformed. A higher level of
protection indicates that a stimulus was attacked later in the sequence, or not
The early decline in predator selectivity suggests that the birds were
at all. The 1M treatment is shown in black and 2M in orange. The points show the
no longer as motivated to discriminate among the numerous stimulus
mean for all feeders along with the 95% confidence intervals based on the
types introduced in the testing phase. This contrasts with the results
t-distribution. The numbers give the sample size (number of dishes). c, The
from our discrimination ability experiment, in which the birds increased
same data as in b, but showing all stimuli separately. The bold points show
selectivity early in the testing phase as they improved their recognition the mean for all feeders and the faint points show the means for individual
of stimuli. Differences between responses towards M100 and V100 feeders along with the 95% confidence intervals based on the t-distribution.
stimuli (common to both experiments) were stronger in the discrimi-
nation ability experiment (Fig. 2) compared with in both treatments of
the multiple-models experiment (Fig. 3c), despite the latter offering no experiment, including lower variation in certain key features associ-
other fly-like stimuli. Assuming comparable predator populations, this ated with Hymenoptera such as the narrow waist and long antennae.
may indicate that the birds found the mimics in the multiple-models Despite the relatively low levels of selectivity, we still observed vari-
experiment sufficiently challenging to discriminate that they were less ation among stimuli in their level of protection. We found no evidence
motivated to try to do so16. One challenge may have been the inclusion that a mimic gains extra protection by an intermediate resemblance to
of extrapolated mimics, meaning that the model(s) were no longer at multiple models, compared with a mimic with equivalent accuracy to
one extreme of the axis. More generally, as both axis end points were only a single model. However, our test is based on a single popula-
based on Hymenoptera, these mimics may have been perceived, on tion of predators encountering all stimuli. In theory, multiple models
average, as more wasp-like than the stimuli in the discrimination ability could provide an additive benefit in cases where different predators

4 | Nature | www.nature.com
have learned to avoid different models, for example, due to allopatry strong support for a positive association between the degree of colour
or separate phenologies9,12. Further experiments incorporating geo- mimicry and the latency to attack (n = 910 presentations, 30 chicks;
graphical and/or temporal variation in mimetic communities would colour appears as an additive effect in all five top-ranked models with
be required to test this. ΔAICc < 2; Supplementary Data 2). We also found some support for an
Although we found no evidence for a selective advantage for jack- influence of stimulus size on chick behaviour (size appears as an addi-
of-all-trades mimics, the addition of an extra model species is still tive effect in three of the five top-ranked models with ΔAICc < 2, and
important to the evolution of mimicry. In the two-model treatment, a further one of five as a nested effect when colour is good or perfect;
we see less variation among mimics in the level of protection received, Supplementary Data 2).
because there are more phenotypes that are close to a model, and there Our results reaffirm that colour should be under strong selection
are therefore more ways of achieving the same level of accuracy. In by birds for mimetic accuracy, being more salient than other visual
nature, aversive models frequently exist as part of a Müllerian mimicry traits22. Hoverfly and wasp colours are typically distinct enough to be
ring of species with a shared warning signal34. In that context, there theoretically discriminated by birds, but many are highly similar and
may be sizeable regions of phenotypic space in which a Batesian mimic could, under natural conditions, be indistinguishable40. In those cases,
would achieve similar levels of mimetic similarity to one member of the our results would predict selection to act on size as the next most sali-
mimicry ring or another, and potentially experience relaxed selection ent visual trait. Our experiment demonstrates a strong propensity of
for further improvements in accuracy. chicks to respond to subtle size differences, even though our stimuli
only differed by 2 mm at most (body lengths: wasp 12 mm, fly 14 mm).
The ability to recognize Batesian mimics from their size is known in
Trait salience some garden birds41, albeit involving larger differences. Pattern and
The above experiments varied all components of visual phenotype shape appear to have weaker effects on predator behaviour, implying
simultaneously, but some elements of appearance may have had more that these traits should experience relaxed selection22, which would
influence on predator behaviour than others22,37. Typically, colour explain some elements of inaccurate appearance in mimics.
is highly salient to birds, taking precedence over (overshadowing)
other visual traits such as shape when choosing prey19,22. However,
the informativeness of a trait is context-dependent and specific to the Invertebrate predators
trait values under comparison38. If a salient mimetic trait has already Many studies of Batesian mimicry, as with the experiments above, use
evolved close resemblance to the model, predators may discriminate birds as focal predators16,18,22,23,41. The eye-of-the-beholder hypothesis11,23
using other traits37. We must therefore consider trait values that are states that mimics that appear inaccurate to one receiver might be
relevant to the study system when determining which traits are under perceived as more accurate to another. Consequently, the selective
the strongest selection. We sought to identify traits under the strongest landscape for mimetic phenotypes depends on the suite of predators
selection for mimicry in the context of avian discrimination between encountering a given mimic—the multiple-predators hypothesis42.
wasps and flies. If some traits overshadow others, this could explain Despite being important predators of many mimetic prey43, and likely
imperfection in otherwise conspicuous traits. to attend to different aspects of a mimetic signal compared with verte-
We generated experimental stimuli based on an axis from the non- brates42, invertebrates are under-represented as predators in mimicry
mimetic fly Tachina fera to the wasp V. vulgaris. We varied four com- studies. Invertebrates including praying mantises44, jumping spiders45
ponents of the appearance independently from one another, such and crab spiders46 can learn to avoid aposematic prey, and will general-
that shape, colour, pattern and size could separately be fly-like (poor ize this avoidance to mimics, but other studies of invertebrate taxa have
mimicry), wasp-like (perfect mimicry) or intermediate (good mimicry; shown limited visual discrimination47. There is little evidence about
equivalent to 50% in the discrimination ability experiment). We gener- how discerning these predators might be among mimics of varying
ated 31 mimetic phenotypes (Extended Data Table 4), with each mimic accuracy (although see ref. 48) and none have compared the responses
given combinations of poor and perfect traits, or good and perfect (but of multiple taxa to the same stimuli. Determining the extent to which
never poor and good, to limit the number of experimental subjects and perceptions of mimetic accuracy vary among invertebrates and differ
presentations required). from those of vertebrates is therefore of interest.
To exert tighter control over the predator learning experience and We assessed the ability of several invertebrates to discriminate
facilitate the use of a larger number of stimulus types than in our between fly and wasp stimuli: praying mantises (Mantidae), jump-
wild-bird experiments, we conducted the experiment in the labora- ing spiders (Phidippus audax) and crab spiders (Synema globosum).
tory. We trained newly hatched chicks Gallus gallus domesticus in Owing to difficulties in training these predators to associate inedible
binary-choice trials to associate fly stimuli (poor in all traits) with a stimuli with a separate reward, we trained them to associate the wasp
hidden mealworm reward and wasp stimuli (perfect in all traits) with stimuli V100 with a negative experience, leading to the same broad
no reward (Fig. 4a). Once the chicks showed a preference for the fly outcome of training: the spiders associated a non-mimetic fly (M100)
dish in at least 80% of presentations, we tested how they generalized with a more-positive outcome than V100. We then tested their response
their response to the various novel mimic (probe) stimuli in single (Extended Data Table 6) towards stimuli from an axis running from
presentations (Fig. 4b and Supplementary Video 2). M100 to V100 (as used in the discrimination ability experiment), with
The chicks did not reject any presented prey, but did show signifi- the aversive experience repeated in the case of V100 to reinforce the
cantly greater latency to attack the unrewarded stimuli (wasp: median learning (Supplementary Video 3).
1.28 s (1.139 lower quartile, 1.48 upper quartile), n = 545 presentations, All three invertebrate taxa discriminated among the stimuli based on
30 chicks) compared with the rewarded stimuli (fly: 0.86 s (0.76 lower appearance (phenotype term in mantis models: n = 40 presentations,
quartile, 1.04 upper quartile), n = 544 presentations, 30 chicks; stimu- 8 individuals, ΔAICc = −36.4, d.f. = 4 (Fig. 5a); jumping spider: n = 57
lus term from linear mixed model: 0.43 s, t = 24.5, P < 0.0001). Even presentations, 9 individuals, ΔAICc = −26.8, d.f. = 4 (Fig. 5b); crab spi-
hesitations of fractions of a second, as seen here, could determine prey der: n = 50 presentations, 50 individuals, ΔAICc = −11.7, d.f. = 4 (Fig. 5c);
capture versus escape in a natural context39, and therefore influence Extended Data Table 5). There was a sharp increase in protection from
the selective pressures experienced32. The latency to attack was also mantises between M75/V25 and M50/V50, after which point protec-
significantly affected by stimulus type among the novel probe stimuli tion reached similar levels to the aversive V100 phenotype. Jumping
(Fig. 4c, Extended Data Fig. 4 and Supplementary Data 2). We tested spiders showed a similar pattern but with a more gradual increase in
both additive and nested models to predict chick response and found protection, only reaching wasp-like levels of protection at M25/V75.

Nature | www.nature.com | 5
Article
a
Training Poor SPCZ
phase: Perfect SPCZ
binary
choice

16 16

b
Poor SPCZ S PZ
Good PC SPZ
Perfect SPCZ PCZ SC SZ C

Testing
phase:
sequential 6 6 1 1 1 1

c Colour: poor Colour: good Colour: perfect

C C C
Standardized time to attack

2
B B
AB
A

49 56 49 42 50 46 90
Poor Good Perfect Poor Good Perfect Poor Good Perfect
Size

Fig. 4 | Chick behavioural response to multiple traits. a, Design of the training includes multiple combinations of shape and pattern traits. A plot with each
phase. Each trial consisted of 16 binary choices between T. fera stimuli with trait combination shown separately is provided in Extended Data Fig. 4. The time
shape (S), pattern (P), colour (C) and size (Z) all poor, and V. vulgaris stimuli with to attack in seconds was standardized across trials by linear scaling such that
shape, pattern, colour and size all perfect, as indicated by the codes above each values for fly and wasp presentations match the median values across all trials,
stimulus image. The solid blue border represents the rewarded stimulus, and shown as horizontal reference lines (wasp, top; fly, bottom). The bold points
the dashed red border represents no reward. b, The design of the testing phase. show the mean values and the vertical lines show the 95% confidence intervals
Each trial consisted of 16 single presentations, including four probe stimuli, based on the t-distribution. The faint points show the results of individual
with the order randomized. Here we show one possible example of a selection trials. The capital letters indicate groupings that show no significant difference
of four probe stimuli, drawn from the options detailed in Extended Data Table 4. after a Tukey post hoc test (P > 0.05). The sample sizes (number of presentations)
c, The latency to attack the stimuli, grouped by colour and size. Each column are given at the bottom of the plot.

The trends for the crab spiders are less clear-cut, but suggest similar crab spiders, but not the jumping spiders (Fig. 5). A possible explana-
levels of protection for all stimuli of M50/V50 and above. tion is the varying visual abilities of the predators in question: they all
This demonstrates that, despite the overwhelming focus on verte- use vision to hunt, but praying mantises have limited colour vision49
brates (especially birds) as predators in mimetic systems, invertebrates in comparison to jumping spiders, which also have excellent acuity50.
can also exert selective pressure for mimicry. Mimics sharing sufficient Even if all of the predators were to detect the same visual information
similarity with the models were afforded a similar level of protection to about the stimuli, their varying behaviour could be explained by differ-
the models themselves, but the similarity required was considerably ences in foraging strategy, such as levels of risk-aversion. Regardless
lower than that observed in the discrimination ability experiment using of the underlying mechanism, the identity of the predator is clearly a
great tits. Moreover, among the three invertebrate taxa there were key factor in determining the strength of selection exerted on visual
further differences in the level of mimicry required for protection. features of mimics42.
Comparisons among such taxonomically diverse predators are
inevitably limited by the need to tailor experiments to the behaviour
and physiology of the predator in question. The mimetic phenotypes Discussion
acceptable to a predator depend, among other factors, on the benefit Testing the responses of predators to 3D hoverfly-like stimuli enabled
of attacking a mimic and the cost of attacking a model12, which are us to examine directly the protection received by various mimetic phe-
impossible to standardize fully across taxa that vary, for example, in notypes. Our pipeline for generating intermediate and extrapolated
nutritional needs and foraging strategy. Even focusing only on the phenotypes enabled us to evaluate areas of the adaptive landscape
invertebrates, which all received the same aversive stimulus, we see describing the protection received by Batesian mimics. Specifically,
variation in the protection received by M50/V50, which was accurate we have (1) generated highly accurate, but not perfect, mimics to show
enough to receive the same protection as a wasp from the mantises and that such mimics still undergo selection for accuracy by a sufficiently

6 | Nature | www.nature.com
 a and the degree to which we can manipulate their traits represent a step
change compared to previous studies using artificial prey19,20.
Level of protection

400

300
We have conducted the most extensive comparison yet of invertebrate
A A responses to the same mimetic stimuli, testing the eye-of-the-beholder
200
hypothesis11,23. It is well known that varying visual systems can lead dif-
100 B B B
ferent predators to receive different information from the same signal13,
0 8 8 8 8 8 and here we have shown this variation can explain the persistence of
Fly 75% fly 50:50 75% wasp Wasp inaccurate mimicry under selection from some predators. Among prey
b experiencing attacks from predators such as praying mantises, a wide
0
range of only moderately accurate mimetic phenotypes will receive
Level of protection

–5
protection through their mimicry. Mimics attacked by more discern-
B ing predators will experience selection for greater accuracy whereas,
B
–10
AB
in mimics exposed to multiple predators, selection will depend on the
A A
combination of different levels of discernment, and/or traits used to
–15
16 8 8 9 16 identify models and mimics by different predators42.
Among the numerous proposed explanations for inaccurate mimicry,
Fly 75% fly 50:50 75% wasp Wasp
a key distinction lies between those suggesting an advantage to inac-
c 0 curate mimics, and others predicting relaxed selection over a range of
moderately accurate phenotypes8. We find no evidence in our experi-
Level of protection

–2 ments for a selective advantage to inaccurate mimics. Instead, we find

B B that some traits, and prey of some predators, are likely to experience
–4
AB AB relaxed selection for visual mimicry.
A The persistence of inaccurate mimicry in nature is a classic problem
–6
30 30 30 30 30
in evolutionary biology, notable for an abundance of theory much
Fly 75% fly 50:50 75% wasp Wasp of which has been challenging to test experimentally. Our use of
cutting-edge 3D technology enabled us to explore the selective pres-
M100 M75 M50 M25
sures on mimetic adaptation in great depth, revealing the incredible
V25 V50 V75 V100
discriminatory ability of an insectivorous bird, as well as how model
community, trait salience and predator species limit the degree of
discrimination in other contexts. In allowing such fine manipulation
of visual phenotypes, our approach brings flexibility to the study of
Fig. 5 | Levels of protection received by different stimulus types resulting mimicry, and more widely to explore the adaptive landscapes for other
from invertebrate predators’ behaviour. a, The mantis (Mantidae spp.) complex morphological traits.
level of protection is the latency to attack in seconds. b,c, The jumping spider
(b; P. audax) and crab spider (c; S. globosum) level of protection is a count of
aggressive behaviours towards the stimulus, subtracted from 0 to match the Online content
direction of response axes in other figures, with higher values indicating greater
Any methods, additional references, Nature Portfolio reporting summa-
protection. The bold points show the mean values and the vertical lines show
ries, source data, extended data, supplementary information, acknowl-
the 95% confidence intervals based on the t-distribution for the log-transformed
edgements, peer review information; details of author contributions
positive data. Faint points show results of individual trials. The capital letters
indicate groupings that show no significant difference after a Tukey post hoc
and competing interests; and statements of data and code availability
test (P > 0.05). The numbers at the bottom of each panel give the sample size are available at https://doi.org/10.1038/s41586-025-09216-3.
(number of presentations). The solid blue border represents the neutral
stimulus, and the dashed red border represents the negative stimulus. The
code above is a unique label for each stimulus type, with the letters indicating 1. Bates, H. W. Contributions to an insect fauna of the Amazon Valley. Lepidoptera: Heliconidæ.
the species used as an end point and the numbers indicating the weighting. Trans. Linn. Soc. Lond. 23, 495–566 (1862).
2. Ruxton, G. D., Sherratt, T. N. & Speed, M. P. Avoiding Attack: The Evolutionary Ecology of
Crypsis, Warning Signals, and Mimicry (Oxford Univ. Press, 2004).
3. Leavey, A., Taylor, C. H., Symonds, M. R. E., Gilbert, F. & Reader, T. Mapping the evolution
motivated avian predator; (2) generated intermediates between two of accurate Batesian mimicry of social wasps in hoverflies. Evolution 75, 2802–2815
(2021).
model species to show that jack-of-all-trades mimics receive no increase 4. Penney, H. D., Hassall, C., Skevington, J. H., Abbott, K. R. & Sherratt, T. N. A comparative
in protection compared with ones with similar accuracy to a single analysis of the evolution of imperfect mimicry. Nature 483, 461–464 (2012).
model; and (3) varied four visual traits independently to show that 5. McLean, D. J., Cassis, G., Kikuchi, D. W., Giribet, G. & Herberstein, M. E. Insincere flattery?
Understanding the evolution of imperfect deceptive mimicry. Q. Rev. Biol. 94, 395–415
shape and pattern may be under weaker selection for accuracy than (2019).
size and, in particular, colour. 6. Pekár, S. & Jarab, M. Assessment of color and behavioral resemblance to models by
Controlled testing of these hypotheses would not be possible using inaccurate myrmecomorphic spiders (Araneae). Invertebr. Biol. 130, 83–90 (2011).
7. Gilbert, F. in Insect Evolutionary Ecology (eds Fellowes M. et al.) 231–288 (CABI, 2005).
only real insect specimens. Our 3D-printed stimuli are good but not per- 8. Kikuchi, D. W. & Pfennig, D. W. Imperfect mimicry and the limits of natural selection.
fect visual replicas of real insects, inevitably being limited by available Q. Rev. Biol. 88, 297–315 (2013).
technology in aspects such as wing transparency and movement. Similar 9. Edmunds, M. Why are there good and poor mimics? Biol. J. Linn. Soc. 70, 459–466
(2000).
to most mimicry research4,14,23,36, we focus here on explaining the visual 10. Pfennig, D. W. & Kikuchi, D. W. Competition and the evolution of imperfect mimicry.
components of mimicry. Other sensory modalities such as olfaction51 Curr. Zool. 58, 608–619 (2012).
and sound/vibration52 may also contribute to predator discrimination, 11. Cuthill, I. C. & Bennett, A. T. D. Mimicry and the eye of the beholder. Proc. R. Soc. B 253,
203–204 (1993).
especially in the case of invertebrate predators42, and could be explored 12. Sherratt, T. N. The evolution of imperfect mimicry. Behav. Ecol. 13, 821–826 (2002).
in future work using a similar framework. Although our printed stimuli 13. Endler, J. A. & Mappes, J. Predator mixes and the conspicuousness of aposematic signals.
Am. Nat. 163, 532–547 (2004).
are not perceived by birds as identical to insects, birds will generalize
14. Dell’Aglio, D. D., Troscianko, J., McMillan, W. O., Stevens, M. & Jiggins, C. D. The appearance
from them to real specimens (Extended Data Fig. 2), demonstrating of mimetic Heliconius butterflies to predators and conspecifics. Evolution 72, 2156–2166
ecological relevance. Furthermore, both the realism of our stimuli (2018).

Nature | www.nature.com | 7
Article
15. Kikuchi, D. W. & Dornhaus, A. How cognitive biases select for imperfect mimicry: a study 39. Thyselius, M., Gonzalez-Bellido, P. T., Wardill, T. J. & Nordström, K. Visual approach
of asymmetry in learning with bumblebees. Anim. Behav. 144, 125–134 (2018). computation in feeding hoverflies. J. Exp. Biol. 221, jeb177162 (2018).
16. Ihalainen, E., Rowland, H. M., Speed, M. P., Ruxton, G. D. & Mappes, J. Prey community 40. Taylor, C. H., Reader, T. & Gilbert, F. Hoverflies are imperfect mimics of wasp colouration.
structure affects how predators select for Müllerian mimicry. Proc. R. Soc. B 279, Evol. Ecol. 30, 567–581 (2016).
2099–2105 (2012). 41. Marples, N. M. Do wild birds use size to distinguish palatable and unpalatable prey types?
17. Kikuchi, D. W., Dornhaus, A., Gopeechund, V. & Sherratt, T. N. Signal categorization by Anim. Behav. 46, 347–354 (1993).
foraging animals depends on ecological diversity. eLife 8, e43965 (2019). 42. Pekár, S., Jarab, M., Fromhage, L. & Herberstein, M. E. Is the evolution of inaccurate mimicry
18. Veselý, P., Luhanová, D., Prášková, M. & Fuchs, R. Generalization of mimics imperfect in a result of selection by a suite of predators? A case study using myrmecomorphic spiders.
colour patterns: the point of view of wild avian predators. Ethology 119, 138–145 (2013). Am. Nat. 178, 124–134 (2011).
19. Corral-Lopez, A. et al. Field evidence for colour mimicry overshadowing morphological 43. Carpenter, G. D. H. & Ford, E. B. Mimicry (Methuen & Company, 1933).
mimicry. J. Anim. Ecol. 90, 698–709 (2021). 44. Berenbaum, M. R. & Miliczky, E. Mantids and milkweed bugs: efficacy of aposematic
20. Wilson, L., Lonsdale, G., Curlis, J. D., Hunter, E. A. & Cox, C. L. Predator-based selection coloration against invertebrate predators. Am. Midl. Nat. 111, 64–68 (1984).
and the impact of edge sympatry on components of coral snake mimicry. Evol. Ecol. 36, 45. Taylor, L. A., Amin, Z., Maier, E. B., Byrne, K. J. & Morehouse, N. I. Flexible color learning in
135–149 (2022). an invertebrate predator: Habronattus jumping spiders can learn to prefer or avoid red
21. Kauppinen, J. & Mappes, J. Why are wasps so intimidating: field experiments on hunting during foraging. Behav. Ecol. 27, 520–529 (2015).
dragonflies (Odonata: Aeshna grandis). Anim. Behav. 66, 505–511 (2003). 46. Morris, R. L. & Reader, T. Do crab spiders perceive Batesian mimicry in hoverflies?
22. Kazemi, B., Gamberale-Stille, G., Tullberg, B. S. & Leimar, O. Stimulus salience as an Behav. Ecol. 27, 920–931 (2016).
explanation for imperfect mimicry. Curr. Biol. 24, 965–969 (2014). 47. Rashed, A., Beatty, C. D., Forbes, M. R. & Sherratt, T. N. Prey selection by dragonflies in
23. Dittrich, W., Gilbert, F., Green, P., Mcgregor, P. & Grewcock, D. Imperfect mimicry: relation to prey size and wasp-like colours and patterns. Anim. Behav. 70, 1195–1202
a pigeon’s perspective. Proc. R. Soc. B 251, 195–200 (1993). (2005).
24. Green, P. R. et al. Conditioning pigeons to discriminate naturally lit insect specimens. 48. Bowdish, T. I. & Bultman, T. L. Visual cues used by mantids in learning aversion to
Behav. Processes 46, 97–102 (1999). aposematically colored prey. The Am. Midl. Nat. 129, 215–222 (1993).
25. Royama, T. Factors governing the hunting behaviour and selection of food by the great tit 49. Sontag, C. Spectral sensitivity studies on the visual system of the praying mantis,
(Parus major L.). J. Anim. Ecol. 39, 619–668 (1970). Tenodera sinensis. J. Gen. Physiol. 57, 93–112 (1971).
26. Pyke, G. H., Pulliam, H. R. & Charnov, E. Optimal foraging: a selective review of theory and 50. Harland, D. P., Li, D. & Jackson, R. R. in How Animals See the World: Comparative Behavior,
tests. Q. Rev. Biol. 52, 137–154 (1977). Biology, and Evolution of Vision (eds Lazareva O. F. et al.) 132–163 (Oxford Univ. Press,
27. Green, D. M. & Swets, J. A. Signal Detection Theory and Psychophysics (Wiley, 1966). 2012).
28. Getty, T. Discriminability and the sigmoid functional response: how optimal foragers 51. Manubay, J. A. & Powell, S. Detection of prey odours underpins dietary specialization in a
could stabilize model-mimic complexes. Am. Nat. 125, 239–256 (1985). Neotropical top-predator: how army ants find their ant prey. J. Anim. Ecol. 89, 1165–1174
29. Fisher, R. A. The Genetical Theory of Natural Selection (Clarendon, 1930). (2020).
30. Sherratt, T. N. State dependent risk taking by predators in systems with defended prey. 52. Moore, C. D. & Hassall, C. A bee or not a bee: an experimental test of acoustic mimicry by
Oikos 103, 93–100 (2003). hoverflies. Behav. Ecol. 27, 1767–1774 (2016).
31. Beatty, C. D. & Franks, D. W. Discriminative predation: simultaneous and sequential
encounter experiments. Curr. Zool. 58, 649–657 (2012).
Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in
32. Chittka, L. & Osorio, D. Cognitive dimensions of predator responses to imperfect mimicry.
published maps and institutional affiliations.
PLoS Biol. 5, e339 (2007).
33. Beatty, C. D., Beirinckx, K. & Sherratt, T. N. The evolution of müllerian mimicry in
Open Access This article is licensed under a Creative Commons Attribution
multispecies communities. Nature 431, 63–66 (2004).
4.0 International License, which permits use, sharing, adaptation, distribution
34. Chatelain, P. et al. Müllerian mimicry among bees and wasps: a review of current
and reproduction in any medium or format, as long as you give appropriate
knowledge and future avenues of research. Biol. Rev. 98, 1310–1328 (2023).
credit to the original author(s) and the source, provide a link to the Creative Commons licence,
35. Bosque, R. J. et al. Diversity of warning signal and social interaction influences the
and indicate if changes were made. The images or other third party material in this article are
evolution of imperfect mimicry. Ecol. Evol. 8, 7490–7499 (2018).
included in the article’s Creative Commons licence, unless indicated otherwise in a credit line
36. Akcali, C. K., Pérez-Mendoza, H. A., Kikuchi, D. W. & Pfennig, D. W. Multiple models generate
to the material. If material is not included in the article’s Creative Commons licence and your
a geographical mosaic of resemblance in a Batesian mimicry complex. Proc. R. Soc. B
intended use is not permitted by statutory regulation or exceeds the permitted use, you will
286, 20191519 (2019).
need to obtain permission directly from the copyright holder. To view a copy of this licence,
37. Sherratt, T. N., Whissell, E., Webster, R. & Kikuchi, D. W. Hierarchical overshadowing of
visit http://creativecommons.org/licenses/by/4.0/.
stimuli and its role in mimicry evolution. Anim. Behav. 108, 73–79 (2015).
38. Hanley, D. et al. Egg discrimination along a gradient of natural variation in eggshell
coloration. Proc. R. Soc. B 284, 20162592 (2017). © The Author(s) 2025

8 | Nature | www.nature.com
Methods Pattern manipulation was performed using custom scripts in R
(v.4.3.0)57. Pattern data were mapped onto the reconstructed meshes
Production of artificial stimuli for the two end points, and vertices were separated into two colour seg-
Overview. To explore predator responses to realistic, but in some cases ments using k-means clustering (k = 2) of RGB colour values. A signed
hypothetical, stimuli, we produced 3D-printed plastic insect replicas. distance map was calculated for each end-point pattern, whereby all
Some stimuli were based on real insect specimens and were matched vertices were assigned a value being the shortest possible edge dis-
as closely as possible in shape, colour, pattern and size to the assigned tance to a vertex of the opposing colour. We created new intermediate
insect. Other stimuli were produced by interpolating along a smooth distance maps by taking weighted averages of the end-point distance
gradient or axis running between two real specimens. maps, and then reverse-engineered them into binary colour patterns
by assigning all positive vertices to one segment and negative vertices
Specimens. Experimental stimuli were based on real wasp (Hyme- to the other.
noptera) and fly (Diptera) taxa chosen to represent different levels of Each segment was assigned a single RGB colour value calculated as
mimetic accuracy (Extended Data Table 1). To generate intraspecific the median of the original colour data from the vertices included in
variation, in the wild-bird experiments, we used three different that segment. Colours for intermediate patterns were calculated as
individuals from each taxon to produce separate stimuli. linear interpolations between the corresponding segments in the end
Insect specimens were collected between June 2020 and August 2021 points. Ultraviolet reflection was ignored because there is no evidence
from various locations in England using a hand net. Specimens were of such colour components in wasp or hoverfly patterns40.
euthanized by freezing at −18 °C for approximately 30 min. They were Owing to limited resolution at the printing stage, legs and antennae
then pinned through the thorax and positioned into a natural-looking for all meshes were given the same standardized shape and a uniform
posture before drying for 6–24 h. colour. The shape was based around a cylinder, with diameter 0.6 mm
for legs and 0.8 mm for antennae (thinner antennae were found to
Photogrammetry. 3D digital images of the insect specimens were be too fragile after printing). In the case of legs, articulations were
obtained by photogrammetry, using a protocol adapted from a previ- added to separate the coxa, femur, tibia and tarsus, and, for anten-
ous study53. Specimens were suspended, with the anterior uppermost, nae, the cylinder was bent into a gentle curve. Colour was taken from
on a motorized turntable (Genie Mini II; Manfrotto, Cassola, Italy), whichever of the two body colour segments most closely matched the
positioned against a white background and lit indirectly using two LED majority leg colour of real specimens. Antennal length was matched
panel lights (22 W, 5600 K; Pixapro). They were photographed using a against distances measured from the original 3D digital image, with
DSLR camera (Canon EOS 600D) and macro lens (Tamron SP 90 mm) intermediates calculated by linear interpolation.
with F20, 1/6 s exposure, ISO400. Each specimen was photographed Wings were created with a flat shape, 0.4 mm thickness, based on
from 36 different angles—three vertical camera positions at each of 12 the outline taken from photographs, which corresponded to shapes as
equally spaced turntable orientations. Wings were removed and pho- they are typically seen when the insect is at rest. In contrast to Diptera,
tographed separately (single photo at a perpendicular angle), because V. vulgaris and A. mystaceus have two pairs of wings but, at rest, the
otherwise their positioning on the body prevented important details of hindwings are hidden owing to overlap with the larger forewings (the
the abdominal pattern and shape from being accurately reconstructed. latter being folded in the case of V. vulgaris). Wing shapes for intermedi-
A 3D shape file (mesh) was built from the set of 36 photographs using ate meshes were calculated using the same deformation method as for
the software 3DSOM54, which uses the outline of the specimen in each the bodies. As our printing method was unable to recreate transpar-
photograph to carve out a 3D shape. The colour information from ent materials, all wings were assigned a uniform colour value of 50%
the photographs was then projected onto this 3D shape to give the grey. This colour matched that of the bases to which the insects were
corresponding colour pattern. attached (see below).
The various components (body, legs, antennae and wings) were
3D image processing. Except where noted, 3D images were edited combined digitally to produce a mesh of the whole insect and finally
using Blender55. Using the images obtained from photogrammetry scaled to match the body length of the relevant end point, or a value
as starting points, we constructed axes of similarity between pairs calculated by linear interpolation for intermediates. A base was added
of real insects through 3D morphological space. We defined axes of to provide an attachment point for the object as a whole, as well as
phenotypic variation along which the traits of shape, colour, pattern improving the structural integrity of the legs. This base was circular as
and size varied smoothly from one image to the other, and generated viewed from above, with a narrow post extending up into the ventral
phenotypes by picking either intermediate or (in the multiple-models side of the thorax.
experiment) extrapolated points along those axes. The four traits were An example axis (M. meridiana to V. vulgaris), viewable in 3D, is pro-
varied in parallel with each other, except in the trait salience experi- vided in Supplementary Data 1.
ment, in which they were varied independently. Details of the speci-
men images used as axis end points are given under each experiment Additive manufacturing. We printed physical 3D representations of
heading below. these digital insects on a HP Jet Fusion 580 machine using polyamide
Owing to difficulties in both processing and printing of thin and 12 powder (CB PA12) and colour cosmetic settings. Stimuli were printed
elongated structures, legs and antennae were removed digitally from at Matsuura Machinery for the discrimination ability and invertebrate
the meshes, to be added back at a later step in more simplified form. predators experiments, and at the University of Nottingham for the rest.
Wings were treated in a similar manner, having already been removed Stimuli were then given VaporFuse Surfacing treatment in a DyeMansion
from specimens before photogrammetry. Powerfuse S, which created a less grainy, slightly glossier finish.
Shape deformations were carried out using the software Defor­
metrica56, which uses control points based large deformation diffeo- Nomenclature. We refer to stimuli in the text according to the initial
morphic metric mapping. A single simplified template was projected letter of the genus of the axis end points, and the percentages by which
onto both end points, such that each retained its shape features but each was weighted when creating any intermediate form. For exam-
remapped onto new vertices that now had a direct one-to-one corres­ ple, C100 indicates a stimulus based 100% on Chrysotoxum, and M25/
pondence between the two meshes. We then calculated the deforma- V75 indicates an intermediate with M. meridiana weighted by 25% and
tion of 3D space required to transform one shape into the other and, V. vulgaris weighted by 75%. In the multiple-models experiment, some
using this, calculated intermediate shapes along the same axis. stimuli were created by extrapolating beyond the range of the two end
Article
points, using weighted averages greater than 100% or below 0%, for 42%, upper quartile 64%) but fidelity to treatment was high (median
example, A150/V−50. 81%, lower quartile 66%, upper quartile 92%). In the generalization test,
only one tagged individual visited the feeders, with 68% of its visits
Ethical approval to a single feeder. No tagging had been conducted that year, so many
The Trait Salience experiment was approved by Newcastle University tagged birds had probably died or dispersed.
AWERB committee (project ID 966). Wild-bird experiments (discrimi-
nation ability and multiple models) were approved by AWERB com- Stimuli: discrimination ability experiment. Stimuli were drawn from
mittees at University of Nottingham (project ID 260) and University three axes, all ending at V. vulgaris and starting from fly taxa with vary-
of Cambridge (ref. NR2022/60). ing levels of mimetic accuracy: M. meridiana, S. ribesii and Chrysotoxum
(Extended Data Table 1). Each axis consisted of the two end points and
Wild-bird experiments three intermediates at 25%, 50% and 75% similarity to V. vulgaris.
Field site and study organisms. Fieldwork was conducted in Madingley In the training phase, we used 15 rewarding fly (M100) and 15 unre-
Wood, Cambridgeshire, UK (52.217° N, 0.049° E), a deciduous woodland warding wasp (V100) stimuli (with 19 dishes left unused). In the testing
composed primarily of broadleaf hardwood trees. The wood has a resi- phase, we used 17 unrewarding V100 stimuli and 32 rewarding stimuli,
dent population of great tits, some of which, as part of other projects, the latter including 10 M100 stimuli as experienced in the training
have been fitted with passive integrated transponder (PIT) tags. Tags of phase, as well as two of each of 11 new phenotypes. New phenotypes
birds involved in this study were fitted between July 2018 and October were three intermediates from the M. meridiana axis (M75/V25, M50/
2022 under licence from the special methods of the BTO projects 1120 V50, M25/V75), S. ribesii (S100) and its three intermediates (S75/V25,
and 1121 held by HMR. Birds included both males and females and were S50/V50, S25/V75), and Chrysotoxum (C100) and its three intermediates
a mix of ages from first-year juveniles upwards. (C75/V25, C50/V50, C25/V25).

Feeding stations. Feeding stations were placed at intervals within the Stimuli: multiple-models experiment. Stimuli were drawn from an
wood, positioned close to dense vegetation to provide cover for small axis running from A. mystaceus to V. vulgaris, representing two model
birds, and separated from each other by at least 80 m. The feeding species and related phenotypes. In addition to the end points, each axis
stations consisted of a 0.75 × 0.75 m wooden board on which a 7 × 7 included three intermediates (25%, 50% and 75%) and four extrapola-
array of 30 mm diameter Petri dishes was fixed. The board was placed tions, two beyond each end point at distances equivalent to the 25% and
on top of a 1.4 m wooden post and covered with a 0.75 × 0.75 × 0.75 m 50% intermediates. A separate non-mimetic stimulus of M. meridiana
cage made from 7 mm square galvanized wire mesh. On one side of M100 was used, with no intermediates.
the cage, approximately 0.5 m above the bottom of the cage, a 30 mm Feeders were assigned to either one model (1M) or two model (2M)
entrance hole allowed small birds to enter past a data logger antenna. treatments, with treatments spatially grouped within the study site
The antenna was linked to a data logger (Francis Scientific Instruments), to reduce the chances of an individual bird that visited multiple feed-
which logged PIT tags of any tagged birds entering. A single horizontal ers experiencing both treatments. In the training phase, we used 25
perch ran across the cage at the level of the entrance, and a further six rewarding fly (M100) and 24 unrewarding wasp stimuli, the latter
perches were placed approximately 100 mm above the surface of the being either exclusively V100 (1M treatment) or 12 × V100 and 12 × A100
board, running between rows of Petri dishes. A motion-sensitive camera (2M treatment; Fig. 3a).
trap (CY70, Ceyomur) was placed above the top of the cage pointing In the testing phase, we used 20 unrewarding wasp stimuli, either
downwards, such that the cage entrance and all Petri dishes were in view. exclusively V100 (1M) or 10 × V100 and 10 × A100 (2M), and 29 reward-
An example video showing two great tit individuals interacting with ing stimuli of which 10 were M100 and the remaining 19 were drawn in
a feeding station is provided in Supplementary Video 1. equal numbers (with rounding) from the intermediate and extrapolated
phenotypes of the A. mystaceus—V. vulgaris axis A150/V−50, A125/V−25,
Discrimination ability and multiple-model experiments. Two main A75/V25, A50/V50, A25/V75, A−25/V125, A−50/V150 (Fig. 3c).
experiments were conducted at this field site using similar methodolo-
gies, along with a third generalization test: the discrimination ability Stimuli: generalization test. Here we tested whether the birds would
experiment ran from December 2021 to May 2022, multiple models generalize their preference for flies over wasps, learned from the
from October 2022 to April 2023 and the generalization test from printed stimuli, to the real insects. In the training phase, we used 12
October to December 2023. These experiments differed in timings and rewarding fly (M100) and 12 unrewarding wasp (V100) stimuli (with
the stimuli used as explained in the relevant sections below, and a few 25 dishes left unused). In the testing phase, we swapped half of the
details relating to sample sizes as follows. printed stimuli for dead, real specimens of the same fly and wasp spe-
In the discrimination ability experiment, five feeding stations were cies, glued to circular bases identical to those used for the printed
used; two of those did not receive enough successful feeding events and stimuli. The testing phase was limited to 5 days to focus on the birds’
were dropped from the study before the testing phase, leaving three initial responses to the real specimens and minimize their opportunity
feeders. In the multiple-models experiment, a sixth feeder was added to refine their learning. The short duration also minimized damage and
and all were in use throughout the experiment. In the generalization decay of the specimens.
test, four feeders were used, three of which had been used previously
and one placed in a new location within the wood. Habituation phase (wild birds). Feeders were first provided with open
Ten tagged individual great tits during the testing phase of the dis- Petri dishes which contained a single mealworm per dish, as well as
crimination ability experiment, and eight tagged individuals in the peanuts placed on the board in between dishes (only provided during
multiple-models experiment, made more than 80 visits each, including initial stages and when visitation rates were low, to encourage birds to
five individuals present in both experiments. An unknown number of visit). Food was refilled every 2–3 days, and the whole feeding apparatus
untagged individuals also visited in both cases; trapping records indi- was sterilized using 70% ethanol spray every 2 weeks. After 3 days, trans-
cate that approximately 71% of the population were tagged in November parent lids were placed onto the Petri dishes so that mealworms were
2021 and 51% in January 2023. In the discrimination ability experiment, visible, but only accessible if the lids were opened. Over the course of
tagged birds directed most of their visits to a single feeder (median 90%, four weeks, visiting great tits learned to open the lids by flipping them
lower quartile 78%, upper quartile 95%). During the multiple-models off using their beaks. Petri dishes and lids were then painted so that the
experiment, fidelity to feeder was weaker (median 51%, lower quartile contents were not visible until the lids were flipped. Great tits continued
to search for food by flipping off the lids to obtain the mealworms and, traits in the same stimulus), resulting in 31 different trait combinations
in most cases, all 49 mealworms had been consumed after 2 days. Other (Extended Data Table 4).
bird species and small mammals were seen visiting the feeding stations
to feed on the peanuts, but rarely opened lids. In the multiple-models Habituation phase (trait salience). On the first day after arrival in the
experiment, from 12,331 lids that were opened, 401 were by blue tits laboratory, chicks received six 2 min trials in the experimental area,
C. caeruleus, which were included in analysis, considering their close foraging from eight open dishes containing mealworms T. molitor.
relatedness with great tits. Mice opened 59 lids which were excluded Chicks were first grouped in threes, then pairs, then individually (two
from analysis. Only great tits opened lids in the discrimination ability trials each). Before the last three sessions on day one, and all of the
experiment and the generalization test. following sessions, chicks were food-deprived for 60 min to ensure
motivation to forage.
Training phase (wild birds). After the habituation phase, a 3D-printed Over the course of the following 6 days, chicks received one trial each
stimulus was attached to the lid of each Petri dish. To train the birds to day during which they received 16 presentations of two dishes, each
avoid the wasp stimuli (V100 and, for multiple models 2M treatment, containing a mealworm. During a presentation, chicks were placed in
A100), no food was provided in the corresponding dishes, and meal- the main arena and had up to 30 s to obtain a mealworm. Chicks were
worms were placed only in the fly dishes (M100). Every 1–2 days we removed before being able to consume the second mealworm and
began a new session by replacing all of the lids in a new configuration, placed in the holding area for 30 s in preparation for the next presen-
randomized with respect to board position, and restocking the relevant tation. Each day, opaque lids were placed increasingly covering the
dishes with mealworms. The training phase continued for 3 weeks dishes until the lids were fully on and the mealworm completely hidden,
for the discrimination ability experiment, and 4 weeks for the gener- teaching chicks to lift off a lid to obtain a mealworm.
alization test. The training phase of the multiple-models experiment
continued for 6 weeks, which included a gap of 1 week (Extended Data Training phase (trait salience). Chicks were each given a further se-
Fig. 3a) when cold weather forced a pause in the experiment because ries of trials during which they learned to discriminate fly from wasp
heavy frost made the dishes unopenable. stimuli through paired choices. Chicks were presented with the same
two dishes as in habituation, but with one bearing a 3D-printed model
Testing phase (wild birds). The testing phase followed the same meth- of T. fera (fly, poor in every trait) and a mealworm inside, and the other
odology as the training phase, but introducing a wider range of stimuli in with a model of V. vulgaris (wasp, perfect in every trait) and no reward.
addition to those on which the birds had been trained (see the ‘Stimuli: After the chick opened one of the two lids (or 30 s elapsed, whichever
discrimination ability experiment’ and ‘Stimuli: multiple-models happened first), it was moved back into the holding area to prevent it
experiment’ sections). All of the newly introduced stimuli were accessing the other dish. The chicks then remained in the holding area
rewarded, representing mimics with varying levels of accuracy. This for 30 s before the next presentation. Each day, the chicks received 1
phase lasted 5 weeks (10 weeks for the multiple-models experiment). trial of 16 presentations.
After 5 days of training the first batch of chicks, it was noted that
Trait salience experiment some individuals showed a bias towards one of the two dish positions
Study organisms and housing (chicks). Domestic chicks (G. g. domes- (left or right, not consistent across chicks), regardless of the stimulus.
ticus; P.D. Hook Hatcheries) were acquired immediately after hatching To reduce this stereotyped behaviour and encourage learning, we sub-
and housed in a laboratory at Newcastle University. Chicks (not sexed) sequently varied dish positions among presentations, placing dishes
were housed communally in two non-concurrent batches of 36 in a floor in two out of four possible positions along a line perpendicular to the
pen measuring approximately 2 m2 with access to food (HPS Starter chicks’ starting position.
Crumb, Special Diets Services) and water ad libitum. The room was Trials continued until chicks chose the fly dish on at least 13 of the
kept at 25 °C and under a 14 h–10 h light–dark cycle. The number of 16 presentations, which took 7–11 days. We excluded 12 chicks that did
chicks was chosen with the aim of a sample size of 10–20 presentations not reach this learning threshold from further testing and analysis. We
per stimulus type, allowing for some exclusions due to failure to meet note that, as a result, our conclusions apply only to the subset of the
training criteria (see below). chicks involved in the testing phase. The presence of some individual
predators which are less selective does not prevent the majority, which
Experimental arena (chicks). The experiments took place in an arena do discriminate among prey types, from exerting selective pressure
measuring 140 × 70 × 40 cm and divided into three sections of lengths on mimetic phenotypes.
25, 90 and 25 cm, separated by mesh barriers such that each section was
visible from the others. The first section formed a buddy area to house Testing phase (trait salience). Chicks then received up to four further
two buddy chicks (from a stock of eight, rotated every hour) during all daily trials (some chicks that took longer to complete the training phase
sessions. Buddy chicks were never used for experimental testing, but spent less time in the testing phase) testing their response to intermedi-
instead ensured that experimental chicks were always able to see and ate stimuli. The structure of trials was identical to the training phase,
hear conspecifics, to reduce stress. The largest section of the arena was except that birds were given only one stimulus in each presentation. In
the experimental area, which included a removable board on which each trial, chicks received six presentations of a Petri dish containing
grey opaque food dishes, with removable lids, were mounted. The final a mealworm and topped with the same fly stimuli used in training, six
section was a holding area in which chicks were placed during 30 s gaps with no reward and topped with the wasp stimuli used in training and
between presentations. four further probe presentations. The probe stimuli were dishes con-
An example video showing a chick approaching stimuli in the taining a mealworm and topped with a novel insect, drawn at random
experimental arena is provided in Supplementary Video 2. from 31 possible trait combinations (Extended Data Table 4). Note that
possible probe stimuli included one identical to the unrewarding wasp
Stimuli (trait salience). Stimuli were based on the non-mimic T. fera stimulus (perfect in all traits) but associated with a mealworm reward,
and the model V. vulgaris. Each of four traits—shape, colour, pattern so acting as a perfect Batesian mimic.
and size—was varied independently to different levels of mimicry, being Chicks opened all dishes in the testing phase, without exception.
poor (matching T. fera), good (50% intermediate) or perfect (matching The latency to attack was measured from the moment the chick was
V. vulgaris). Stimuli were created in all possible combinations of poor released into the arena to its first peck of the dish or lids. Given the
and perfect traits, or good with perfect traits (but never poor and good speed at which chicks approached the dish (median, 1.1 s), timings were
Article
taken from video recordings slowed to 0.3× speed using the BORIS not always attack, and were punished (in the same way as the mantises)
software package58 to improve accuracy. Experimenters were not blind at the end of trials involving a wasp stimulus, regardless of whether
to stimulus type during this process. the spider had attacked the stimulus or not. Fly stimuli (M100) were
associated with no reward or cost.
Invertebrate predators experiment Training for crab spiders was performed using a condensed protocol
Study organisms and housing (invertebrates). Praying mantises of as it was not possible to maintain the wild-caught spiders in the labora-
three species (Rhombodera kirbyi (n = 5, fourth instar to adult), Poly- tory for long periods. The spiders (n = 150) did not undergo trials with
spilota aeruginosa (n = 1, subadult) and Pseudoxyops perpulchra (n = 2, presentations of wasp or fly stimuli, but simply received the punishment
third instar), all unsexed; BugzUK and LDW bugs) and jumping spiders without any previous associated stimulus or behaviour. However, this
(adult male and female P. audax obtained from Jumping Spiders Web) still provided an opportunity to associate the negative experience with
were housed individually in transparent plastic boxes (19 × 13 × 8 cm) the wasp stimulus owing to its use in the ‘punishment’ process itself.
in a laboratory at University of Nottingham. The room was kept at 26 °C
and under a 12 h–12 h light–dark cycle. They were fed crickets or meal- Testing phase (invertebrates). Praying mantises and jumping spiders
worms twice weekly, with all trials conducted 30 h after feeding. received a further nine trials using the same procedures as the training
We collected crab spiders (S. globosum, adult male and female) that phase. Five probe stimuli were presented consisting of each of the five
were sitting in wait for prey on flowers (where they hunt for pollinat- points along the axis in random order, alternating with four reinforce-
ing insects) around the Quinta de São Pedro field research station ment trials (two M100 and two V100). All attacks on (mantises) or en-
(38.568° N, 9.193° W) and surrounding areas of Sobreda, Portugal. counters with (spiders) wasp stimuli were punished as before. Owing
Individuals found with recently killed prey were not included. Spiders to restricted time in captivity, each crab spider was presented once
were kept at the Quinta de São Pedro research station, and individu- with a single stimulus, selected from the five axis points at random.
ally housed in transparent plastic universal tubes. Spiders were kept, As in the training, mantises attacked all stimuli, and the latency to
unfed, for 48 h until use, but note that the median time since the last attack was measured from when the motor was switched on to the
meal will have been longer. The room was kept at 22 °C, with no artificial mantis first striking the stimulus. Spiders rarely attacked the stimulus
light–dark cycle. (P. audax 11% of trials, S. globosum 17% of trials); thus, using latency to
attack as the primary measure of behaviour would provide poor resolu-
Experimental arena (invertebrates). Mantis and jumping spider trials tion. They did, however, display a range of positive (such as orientation
were performed inside an opaque plastic box (19 × 13 × 8 cm). A fishing towards the stimulus, approach) and negative (for example, retreat,
line was fed through two small holes at either side of the box, with one hide) behaviours in response to stimuli; a full list of the observed behav-
end attached to a counterweight maintaining tension and the other end iours is shown in Extended Data Table 6. Instances of these behaviours
attached to a bobbin. The bobbin was spun by a motor, programmed were recorded over the full trial period (P. audax, 5 min; S. globosum,
with a microcontroller board (Arduino) to rotate in a randomized pat- 3 min). Experimenters were not blinded to the stimulus type during
tern (1–2 s clockwise or anticlockwise, 0–1 s pause, then repeated in the the laboratory trial.
opposite direction). This moved stimuli left and right in rapid, jerky
motions and encouraged striking59. Stimuli were suspended by a fine Statistical analysis
steel wire loop from the fishing line, allowing them to dangle and move All analysis was carried out in R (v.4.3.0)57. We used generalized lin-
in three dimensions. ear models and generalized linear mixed models implemented in the
Crab spider trials used a similar arrangement of equipment but with package lme4 (ref. 60). In all cases, model fit was assessed visually for
an arena that was larger (69 × 38 × 41 cm) and included the addition of normality of residuals and homoscedasticity using residual plots.
a single purple milk thistle (Galactites tomentosus) to provide a perch From a defined set of candidate models, the most parsimonious was
for the spiders. The fishing line to which stimuli were attached entered selected based on lowest AICc values, with ties (a difference in AICc of
through a hole in the lid of the arena as opposed to the side, causing less than two) broken by choosing the model with the fewest degrees
stimuli to move vertically towards and away from the flower, as opposed of freedom61.
to left and right.
Example video clips showing the different predators being presented Wild-bird experiments. Within each session and feeder, we determined
with stimuli in their respective arenas are provided in Supplementary the order of dishes being opened on the basis of video data, with any
Video 3. left unopened placed at the end of the sequence. Those coding the
videos were not blinded to the stimulus identity during this process.
Stimuli (invertebrates). Stimuli were drawn from an axis running from We converted this sequence to a set of protection values from 0 to 1,
M. meridiana (fly; M100) to V. vulgaris (wasp; V100) with three inter- corresponding to the first and last dishes of the sequence respectively.
mediates: M75/V25, M50/V50 and M25/V75. This axis matches one of Thus, 0 can be considered to be the least protected as it is attacked
the three axes used in the discrimination ability experiment. Stimuli first, and 1 the most protected as it is attacked last or not at all. These
were removed from their bases as the presentation method involved values were then logit-transformed using the formula log[(x + 0.01)/
hanging down on a wire rather than resting on top of a lid. (1 − x + 0.01)] to occupy an unbounded scale, which improved normality
of residuals. The 0.01 adjustment in this formula is to ensure that 0 and
Training phase (invertebrates). Praying mantises (n = 8) and jumping 1 transform to finite values62.
spiders (n = 8) each underwent six aversive conditioning trials on sepa- For the discrimination ability experiment, initial preferences were
rate days. In the first trial, the stimulus was randomly allocated (M100 assessed on the basis of bird behaviour in the first session of the training
or V100) for each individual then, in subsequent trials, the stimulus phase only. These preferences would have depended mostly or entirely
was alternated. After being placed into the arena, animals were given on the subjects’ innate sensory biases and learning from their experi-
1 min to acclimatize before the stimulus was introduced. All mantises ence in the wild and not on their experience with the experimental
attacked the stimuli within 10 min, and were immediately punished stimuli (although some learning may have taken place from the very
after attacking wasp stimuli (V100) by being prodded firmly on the first dish onwards). We used the explanatory variables reward (binary
thorax with a separate wasp stimulus attached to the end of a thin metal variable: mealworm or no mealworm, here corresponding to fly or wasp
rod. Subjects appeared to be appropriately threatened by this punish- stimuli respectively) and feeder (categorical variable identifying the
ment, responding by moving away from the rod. Jumping spiders did feeding station, here treated as fixed due to only having three levels).
We fitted the linear model preference ~ reward × feeder and compared along the axis of similarity; that is A75/V25, A50/V50 and A25/V75
it to all four nested submodels. are intermediate) and stimulus (categorical variable treating each
To highlight trends of stimulus selection as time progressed in the position along the axis of similarity separately). All models used the
training phase, we fitted a nonlinear least squares (NLS) model to the Gaussian family and included fixed effects of edge and treatment and
levels of protection for the wasp stimulus (as there were only two a random effect of feeder. H0: no effect of stimulus on bird prefer-
stimulus types, the pattern for fly stimuli is simply an inversion of this). ences preference ~ edge + treatment + (1|feeder); H1: stimuli receive
a
We used a sigmoid learning curve defined by the formula −b ×(t − c ) protection in inverse proportion to their distance from a model pref-
1+e
based on time t measured as the number of sessions. This formula erence ~ distance_to_model + edge + treatment + (1|feeder). H2: as for
assumes that zero protection is received at time zero. H1, but intermediate mimics receive extra protection preference ~ dis-
In the testing phase, we fitted separate curves to each phenotype, tance_to_model * intermediate + edge + treatment + (1|feeder). H3:
c
using an asymptotic curve with the formula a + (b − a) × e−e × t as, in certain stimuli elicit avoidance behaviour in unique and unpredictable
contrast to in the training phase, there did not appear to be any initial ways preference ~ stimulus + edge + treatment + (1|feeder). Each of H1–3
warm-up period to the rate of learning. This approach enabled us were also tested with the addition of an interaction between treatment
empirically to parameterize the learning period according to the start- and the stimulus-related term (H1 + distance_to_model:treatment;
ing level of discrimination, rate of change and final level of discrimina- H2 + intermediate:treatment; H3 + stimulus:treatment).
tion, and therefore to identify a period of learning after which bird
preferences were relatively stable. As a result, for our main analysis of Trait salience experiment. To standardize for variation in speed of
the testing phase, we excluded the first 9 days of the testing phase while behavioural responses of chicks among individual trials, within each
birds adapted to the new set of stimuli; from day 10 onwards, behav- trial, we compared the response to probe stimuli against values for
ioural responses had reached within 10% of the asymptotic value the six fly and six wasp presentations. Within a trial, latency to attack
according to the fitted learning curves. We used the explanatory vari- was linearly scaled such that values for the response to fly and wasp
ables feeder (as for training phase), plus axis (categorical variable for presentations matched the median values from across all trials (0.855
whether the stimuli were based on Mesembrina, Syrphus or Chryso- and 1.28 s respectively). In 14 out of 105 trials, there was little (<0.1 s)
toxum), phenotype (the degree of similarity to the wasp, categorical or no delay in the mean response towards the wasp stimuli; these trials
to allow for a wide range of non-linear relationships) and edge (binary were excluded from further analysis.
variable to indicate whether a dish was on the outer perimeter of the We compared models representing several sets of hypotheses about
49-dish array, included to improve the fit as we observed that birds the relative importance of the four phenotypic traits in influencing
preferred to open dishes along the perimeter of the cage; it was not chick behaviour. We used the explanatory variables day (categorical
tested for significance). We fitted the Gaussian linear model prefer- variable for the number of days through the testing phase, allowing for
ence ~ (axis × phenotype + edge) × feeder and compared it to 32 nested changes in behaviour depending how many trials the chick has already
submodels. This approach allowed the comparison of models with or completed), batch (categorical variable for which of two groups the
without (1) an effect of phenotype, that is the degree of similarity to chick belonged to, run on different dates), first_pres (binary factor
the wasp; (2) an effect of axis (M. meridiana, S. ribesii or Chrysotoxum), indicating whether or not it was the first presentation of a trial, as we
potentially interacting with phenotype; and (3) individual variation in observed chicks to be slower on their first attempt of the trial), chick
behaviour, according to the interactions with the feeder term, since (random effect for individual ID), shape, colour, pattern and size (each
feeders represent largely separate sets of individual great tits. Tukey’s represented by a three level factor of poor (fly-like), good (intermedi-
post hoc comparisons were used to test for differences among differ- ate between fly and wasp) or perfect (wasp-like)), and interactions to
ent levels of the phenotype and axis variables. represent overshadowing, where the overshadowed trait is ignored
For the multiple-models experiment, again we fitted learning curves unless another trait, termed the main trait, is above a certain level of
using NLS, but found that patterns in the data conformed less closely to accuracy. All models used the Gaussian family and included fixed effects
simple curve definitions. In the training phase, discrimination initially of day, batch and first_pres, and a random effect of chick. H0: no effect
improved and then appeared to regress after a spell of cold weather of stimulus on chick behaviour latency_to_attack ~ day + batch + first_
(see the ‘Training phase (wild birds)’ section above), possibly owing to pres + (1|chick). H1: each trait has a separate, additive effect on behaviour
turnover of individuals. Asymptotic curves fitted to the whole training latency_to_attack ~ day + batch + first_pres + shape + colour + pattern +
phase (formula as described for the discrimination ability experiment) size + (1|chick). Nested submodels were also fit that excluded dif-
did not converge but, when fitted just to the sessions after the experi- ferent combinations of the four trait terms. H2: one trait is assigned
mental pause, converged for two of the three stimulus types. In the as overshadowing others, so that other traits are ignored unless the
testing phase, asymptotic curves fitted using NLS did not converge on a main trait is perfect (for example, latency_to_attack ~ day + batch +
solution for any stimuli. This is probably because most stimuli showed first_pres + colour + colour_perfect:shape + colour_perfect:pattern +
no clear trends with time, except for M100, which showed an increase in colour_perfect:size + (1|chick)). The model was repeated with each of
the levels of protection during the first 10 days. In the absence of fitted the four traits as the main trait, and nested submodels that excluded
curves, we used the same cut-off as in the discrimination ability experi- different combinations of the overshadowed traits were also fitted.
ment, which matched our subjective assessment of the data trends, H3: one trait is assigned as partially overshadowing others, so that
removing data from days 1–9 when modelling learned preferences. other traits are ignored unless the main trait is good or perfect (for
We compared models representing several specific hypotheses example, latency_to_attack ~ day + batch + first_pres + colour + colour_
related to great tit behaviour in the testing phase. We used the explana- good_perfect:shape + colour_good_perfect:pattern + colour_good_
tory variables edge (as in the discrimination ability experiment), feeder perfect:size + (1|chick)), with variations as described for H2. H4: all
(now fitted as a random effect as there were six feeders), treatment trait combinations have their own unique effects on chick behaviour
(categorical, one model or two models), distance to model (con- latency_to_attack ~ day + batch + first_pres + shape × colour × pattern ×
tinuous variable measuring distance to the nearest model along the size + (1|chick)).
Vespula–Argogorytes axis of similarity; in the two-model treatment,
A150/V−50, A50/V50 and A−50/V150 would all have the same distance Invertebrate predators experiment. The response variable for mantis
(50 percentage points) from a model and would be predicted to elicit behaviour was latency to attack, measured in seconds and modelled
similar responses from the predators), intermediate (categorical using a Gaussian family with log link. Jumping spiders and crab spiders
variable indicating whether a stimulus lies between the two models rarely attacked the stimuli directly so instead, response was the number
Article
of observations of positive behaviour towards the stimulus: display, 56. Bône, A. et al. Deformetrica 4: an open-source software for statistical shape analysis.
In Shape in Medical Imaging (eds. Reuter, M. et al.) 3–13 (Springer, 2018).
approach and attack, and for jumping spiders, alert and orientation. 57. R Core Team. R: a language and environment for statistical computing (R Foundation for
These responses were modelled using a Poisson family with log link. Statistical Computing, 2023).
Models included a fixed effect of phenotype (categorical variable for 58. Friard, O. & Gamba, M. BORIS: a free, versatile open-source event-logging software for
video/audio coding and live observations. Methods Ecol. Evol. 7, 1325–1330 (2016).
similarity to the wasp, as for discrimination ability above) and random 59. Rilling, S., Mittelstaedt, H. & Roeder, K. D. Prey recognition in the praying mantis.
effect of individual (except for the crab spiders, which had only one Behaviour 14, 164–184 (1959).
data point per individual). Tukey’s post hoc comparisons were used to 60. Bates, D., Mächler, M., Bolker, B. & Walker, S. Fitting linear mixed-effects models using
lme4. J. Stat. Softw. 67, 1–48 (2015).
test for differences among different levels of the phenotype variable. 61. Leroux, S. J. On the prevalence of uninformative parameters in statistical models
applying model selection in applied ecology. PLoS ONE 14, e0206711 (2019).
Reporting summary 62. Warton, D. I. & Hui, F. K. C. The arcsine is asinine: the analysis of proportions in ecology.
Ecology 92, 3–10 (2011).
Further information on research design is available in the Nature Port- 63. Online 3D Viewer (Viktor Kovacs, 2023).
folio Reporting Summary linked to this article.

Acknowledgements We thank P. Wilderspin and the staff at the University Farm & Rural Estate
(University of Cambridge) for permission to conduct fieldwork in Madingley Wood; T. Fulford,
Data availability C. Thorne, J. Beaver and the members of the Madingley ringing group for PIT tagging great tits
Data have been deposited at the NERC Environmental Information at Madingley; M. Waddle and the Comparative Biology Centre staff for technical assistance in
chick husbandry at Newcastle University; D. Starkey for pilot work with jumping spiders and
Data Centre and are available online: stimuli: ‘Scanned 3D images and
mantises; B. Richter for permission to conduct crab spider experiments at the Quinta de São
3D printable images based on combinations of features of Diptera Pedro field centre; L. Baker for coding of video data; and P. Harris for sharing their HP Jet Fusion
and Hymenoptera collected from the UK in 2021–22’ (https://doi. expertise. The project was funded by a NERC standard grant (NE/S000623/1), with additional
funding from the University of Nottingham, Leverhulme Early Career Fellowship (ECF-2018-700)
org/10.5285/05169766-7355-4c3c-8ade-091db0583f9d); wild-bird
to G.L.D. and the Max Planck Society and Royal Society (RG110122) to H.M.R.
experiments: ‘Great tit behavioural responses to 3D-printed insect
replicas, featuring combinations of traits from wasps and flies, in Mad- Author contributions Conceptualization: C.H.T., F.G. and T.R. Methodology: C.H.T., D.J.G.W.,
ingley Wood, Cambridge, UK, 2021–2023’ (https://doi.org/10.5285/ J. Skelhorn, J.R.D., C.B., G.L.D., H.M.R., M.E., R.G., F.G. and T.R. Formal analysis: C.H.T.
Investigation (wild-bird experiments): C.H.T., D.B., S.B., A.C., K.C., J.R.D., S.R.G., E.O., A.L.P. and
a1c9b0cc-5585-49c5-a38f-fe05240edccf); trait salience: ‘Chick behav-
T.R. Investigation (trait salience experiment): C.H.T. and D.J.G.W. Investigation (invertebrate
ioural responses to 3D-printed insect replicas, featuring combinations predators experiment): J.J.D., T.D., H.J.J., R.L., H.R., J.R., J. Sains and F.S. Data curation: C.H.T.
of traits from wasps and flies’ (https://doi.org/10.5285/45391184-603e- Writing—original draft: C.H.T. Writing—review and editing: all of the authors. Visualization:
C.H.T., H.J.J. and J. Sains. Supervision: J. Skelhorn, F.G. and T.R. Project administration: C.H.T.,
4284-bb3c-9c8c6bf856ab); invertebrate predators: ‘Invertebrate
J. Skelhorn and T.R. Funding acquisition: C.H.T., J. Skelhorn, C.B., R.G., F.G. and T.R.
behavioural responses to 3D-printed insect replicas, featuring com-
binations of traits from wasps and flies, in laboratory trials’ (https:// Competing interests The authors declare no competing interests.
doi.org/10.5285/ee7ba05a-449b-466e-840c-8de1d3f1d4d1). Source
Additional information
data are provided with this paper. Supplementary information The online version contains supplementary material available at
https://doi.org/10.1038/s41586-025-09216-3.
53. Nguyen, C. V., Lovell, D. R., Adcock, M. & La Salle, J. Capturing natural-colour 3D models Correspondence and requests for materials should be addressed to Christopher H. Taylor.
of insects for species discovery and diagnostics. PLoS ONE 9, e94346 (2014). Peer review information Nature thanks Thomas Sherratt and the other, anonymous,
54. 3DSOM (Creative Dimension Software, 2013). reviewer(s) for their contribution to the peer review of this work. Peer reviewer reports are
55. Blender—a 3D modelling and rendering package v. 2.9 (Stichting Blender Foundation, available.
2020). Reprints and permissions information is available at http://www.nature.com/reprints.
Extended Data Fig. 1 | Changes in preference over time in the Discrimination the data. N = 828. b Test phase, trends with time. Asymptotic curves fitted to
Ability experiment. “Level of protection” is the rank order of attack for that each phenotype. N = 1295 dishes (fly), 558 (75% fly), 553 (50-50), 552 (75% wasp),
stimulus within a session, logit transformed. Higher level of protection 1565 (wasp). c Test phase, comparing initial (yellow, sessions 1-3) and asymptotic
indicates that a stimulus was attacked later in the sequence, or not at all. Points (black, session 10 onwards, after fitted response reaches within 10% of the
show mean and vertical bars show 95% confidence intervals based on the asymptote) preferences. Mesembrina axis, N = 191 dishes (initial), 1004
t-distribution. a Training phase. Data show wasp stimuli only; fly data are an (asymptotic). Chrysotoxum axis, N = 96 (initial), 502 (asymptotic). Syrphus axis,
almost exact inversion of the data shown. Line shows a sigmoidal curve fitted to N = 95 (initial), 497 (asymptotic). For images of the stimuli, see Fig. 2 in main text.
Article

Extended Data Fig. 2 | Validation experiment testing phase. Points show mean and vertical bars show 95% confidence intervals based on the t-distribution. Capital
letters indicate groupings which show no significant difference in a Tukey post-hoc test (p > 0.05). Sample sizes (number of dishes) are shown at the base of the plot.
Extended Data Fig. 3 | Changes in preference over time in the Multiple clarity of the rest of the plot. N = 3470 dishes (M100), 769 (A100) and 2227 (V100).
Models experiment. Points show mean and vertical bars show 95% confidence b Test phase, trends with time. Asymptotic curves fitted to these data, using
intervals based on the t-distribution. a Training phase. Lines show an asymptotic the same method as the training phase, failed to converge. Instead, trend lines
curve fitted to the data from session 16 onwards; the curve for A100 did not are a moving average across five sessions (centred on the third session). Sample
converge but is shown for illustration. Session 16 was the first session after a sizes shown above each plot. c Test phase, comparing initial (yellow, session 1-3)
period of one week when no birds opened dishes. Curves fitted to the full time and asymptotic (black, session 10 onwards, chosen to match Discrimination
period all failed to converge. A100 session 9 had a small sample size (3) and as a Experiment) preferences. N = 437 dishes (initial, 1 M), 384 (initial, 2 M), 2902
result has very wide confidence intervals (−5.3, 8.3) that are not shown in full for (asymptotic, 1 M), 4334 (asymptotic, 2 M).
Article

Extended Data Fig. 4 | Chick latency to attack mimetic stimuli. Levels of Fig. 4, main text). Time to attack (seconds) has been standardized across trials
accuracy of mimetic traits are coded as 0 (fly-like/poor), 50 (intermediate/ by linear scaling such that values for fly and wasp presentations match the
good) and 100 (wasp-like/perfect) for each of shape, pattern, colour and size. median values across all trials, shown as horizontal reference lines (wasp upper,
Each panel shows a certain combination of colour and size traits, and within a fly lower). Points show mean and vertical bars show 95% confidence intervals
panel, black points show certain combinations of pattern (P) and shape (S), and based on the t-distribution. Sample sizes (number of presentations) are shown
red points show data pooled across all values for pattern and shape (as shown in at the base of each plot.
Extended Data Table 1 | Species on which 3D stimuli were based
Article
Extended Data Table 2 | Comparison of models fitted to data from the testing phase of the discrimination ability experiment

*All models are based on the initial formula (level of protection ~ edge), with the addition of the listed terms. Abbreviated terms: a = axis, e = edge, f = feeder, and p = phenotype.
Extended Data Table 3 | Comparison of models fitted to data from the multiple-models experiment

*All models are based on the initial formula (level of protection ~ edge + treatment + 1|feeder), with the addition of the listed terms.
Article
Extended Data Table 4 | Trait combinations used in the trait salience experiment
Extended Data Table 5 | Comparison of models fitted to data from the multiple predators experiment
Article
Extended Data Table 6 | Behaviours performed by P. audax and S. globosum in trials within the multiple predators
experiment
 nature portfolio | reporting summary
Corresponding author(s): Christopher Taylor
Last updated by author(s): 15/05/2025

Reporting Summary
Nature Portfolio wishes to improve the reproducibility of the work that we publish. This form provides structure for consistency and transparency
in reporting. For further information on Nature Portfolio policies, see our Editorial Policies and the Editorial Policy Checklist

Statistics
For all statistical analyses, confirm that the following items are present in the figure legend, table legend, main text, or Methods section.

n/a Confirmed
The exact sample size (n) for each experimental group/condition, given as a discrete number and unit of measurement
A statement on whether measurements were taken from distinct samples or whether the same sample was measured repeatedly
The statistical test(s) used AND whether they are one- or two-sided
Only common tests should be described solely by name; describe more complex techniques in the Methods section.

A description of all covariates tested

A description of any assumptions or corrections, such as tests of normality and adjustment for multiple comparisons
A full description of the statistical parameters including central tendency (e.g. means) or other basic estimates (e.g. regression coefficient)
AND variation (e.g. standard deviation) or associated estimates of uncertainty (e.g. confidence intervals)

For null hypothesis testing, the test statistic (e.g. F, t, r) with confidence intervals, effect sizes, degrees of freedom and P value noted
Give P values as exact values whenever suitable.

For Bayesian analysis, information on the choice of priors and Markov chain Monte Carlo settings
For hierarchical and complex designs, identification of the appropriate level for tests and full reporting of outcomes
Estimates of effect sizes (e.g. Cohen's d, Pearson's r),
), indicating how they were calculated
Our web collection on statistics for biologists contains articles on many of the points above.

Software and code

Policy information about availability of computer code
Data collection 3DSOM Pro v5
Deformetrica 4.3.0
Blender 2.9

Data analysis BORIS 8.19.3

R v4.3.0 with packages tidyverse 2.0.0, lme4 1.1, nlme 3.1, emmeans 1.10, and MuMIn 1.47
For manuscripts utilizing custom algorithms or software that are central to the research but not yet described in published literature, software must be made available to editors and
reviewers. We strongly encourage code deposition in a community repository (e.g. GitHub). See the Nature Portfolio guidelines for submitting code & software for further information.

Data
Policy information about availability of data
All manuscripts must include a data availability statement This statement should provide the following information, where applicable:
- Accession codes, unique identifiers, or web links for publicly available datasets
April 2023

- A description of any restrictions on data availability

- For clinical datasets or third party data, please ensure that the statement adheres to our policy

Data have been deposited at the NERC Environmental Information Data Centre. Some datasets are currently under embargo until April/May 2025 which can be

1
lifted if accepted for publication at an earlier date.
Stimuli: “Scanned 3D images and 3D printable images based on combinations of features of Diptera and Hymenoptera collected from the UK in 2021-22”

nature portfolio | reporting summary

https://doi.org/10.5285/05169766-7355-4c3c-8ade-091db0583f9d
Wild bird experiments: “Great tit behavioural responses to 3D-printed insect replicas, featuring combinations of traits from wasps and flies, in Madingley Wood,
Cambridge, UK, 2021-2023” (under embargo until 1 April 2025)
https://doi.org/10.5285/a1c9b0cc-5585-49c5-a38f-fe05240edccf
Trait Salience: “Chick behavioural responses to 3D-printed insect replicas, featuring combinations of traits from wasps and flies” (under embargo until 1 May 2025)
https://doi.org/10.5285/45391184-603e-4284-bb3c-9c8c6bf856ab
Invertebrate Predators: “Invertebrate behavioural responses to 3D-printed insect replicas, featuring combinations of traits from wasps and flies, in laboratory
trials” (under embargo until 1 April 2025)
https://doi.org/10.5285/ee7ba05a-449b-466e-840c-8de1d3f1d4d1

Research involving human participants, their data, or biological material

Policy information about studies with human participants or human data. See also policy information about sex, gender (identity/presentation),
and sexual orientation and race, ethnicity and racism.
Reporting on sex and gender No human participants/data

Reporting on race, ethnicity, or No human participants/data

other socially relevant groupings

Population characteristics No human participants/data

Recruitment No human participants/data

Ethics oversight No human participants/data

Note that full information on the approval of the study protocol must also be provided in the manuscript.

Field-specific reporting
Please select the one below that is the best fit for your research. If you are not sure, read the appropriate sections before making your selection.

Life sciences Behavioural & social sciences Ecological, evolutionary & environmental sciences
For a reference copy of the document with all sections, see nature.com/documents/nr-reporting-summary-flat.pdf

Ecological, evolutionary & environmental sciences study design

All studies must disclose on these points even when the disclosure is negative.
Study description All experiments tested the responses of predators towards 3D-printed stimuli of varying levels of mimetic accuracy, associated with
different levels of reward or punishment.
Discrimination Ability: recorded the order in which wild great tits opened dishes bearing 3D stimuli to obtain potential mealworm
reward. Three replicate feeding stations with 49 dishes on each, reset and repeated in daily sessions for three weeks. Predictors were
3D stimulus type, presence of mealworm, position of dish (edge vs centre).
Multiple Models experiment: recorded the order in which wild great tits opened dishes bearing 3D stimuli to obtain potential
mealworm reward. Six replicate feeding stations divided between two treatments: "one model" or "two model" based on the number
of model (unrewarding) stimulus types used. 49 dishes at each feeder, reset daily over the course of six weeks. Predictors were
treatment, 3D stimulus type, presence of mealworm, position of dish (edge vs centre).
Validation test: Recorded the order in which wild great tits opened dishes bearing 3D stimuli to obtain potential mealworm reward.
Four replicate feeding stations with 24 dishes on each, reset and repeated in daily sessions for five days. Predictors were 3D stimulus
type, presence of mealworm, position of dish (edge vs centre).
Trait Salience experiment: recorded the time taken for chicks to approach and peck at dishes bearing 3D stimuli and (for some
stimuli) containing a mealworm. 1445 presentations to 30 chicks. Predictors were chick id, day within experimental run (1-4), batch
(two different batches of chicks tested on different dates), whether or not it was the first presentation of a trial, and 3D stimulus
type.
Invertebrate Predators experiment: recorded behavioural responses of praying mantises (n= 40 presentations, 8 individuals), jumping
spiders (n = 57 presentations, 9 individuals) and crab spiders (n = 50 presentations, 50 individuals) towards 3D printed stimuli.
Mantises and jumping spiders received all different types of 3D stimulus, whereas crab spiders were randomly assigned to a single
stimulus type. Predictors were stimulus type and individual id.

Research sample Discrimination Ability, Multiple Models experiments and Validation test: a population of great tits (Parus major) within Madingley
April 2023

Wood, Cambridge, UK, some PIT tagged (estimated 51-71%). Birds that visited our feeding stations were all adult plumage, but sex
and exact age were not recorded. This population was chosen as consisting of wild generalist insectivorous predators expected to
show propensity and ability to learn to obtain food from our experimental set up.
Trait Salience: domestic chicks (Gallus gallus domesticus) obtained from a commercial hatchery on the day after hatching, with
testing phase of the experiment when they were 17-20 days old. At this age they were not identifiable to sex. They were chosen as

2
 generalist insectivorous predators that could be housed in a laboratory from hatching and therefore of known prior experience
regarding food types. Invertebrate Predators: Praying mantises of three species (Rhombodera kirbyi (n = 5), Polyspilota aeruginosa (n

nature portfolio | reporting summary

= 1) and Pseudoxyops perpulchra (n = 2)), jumping spiders (Phidippus audax) and crab spiders (Synema globosum). Crab spiders
included both male and female, others were not sexed. These subjects were chosen to represent a range of visual types and
predation strategies from invertebrates that are known or suspected to prey on Diptera and Hymenoptera.

Sampling strategy Discrimination Ability and Multiple Models experiments, Validation test: sample sizes were constrained mainly by the birds' seasonal
behaviour and the number of birds present in the population that chose to visit our feeding stations.
Trait Salience and Invertebrate Predators experiments: sample sizes were calculated to give repeats of around n=10 for each stimulus
type, minimising the numbers of laboratory animals required.

Data collection Discrimination Ability and Multiple Models experiments, Validation test: video recordings of bird behaviour were taken using a
motion-sensitive trail camera and combined with manual records taken of which dishes were open at the end of a given session. Data
were recorded by CHT, DB, SB, AC, KC, JRD, SRG, EO, ALP and TR.
Trait Salience experiment: behavioural data were noted by the experimenters (CHT and DJGW) at the time of the experiment, and
videos were recorded which were later analysed by CHT and DB to determine latency to attack the dishes.
Invertebrate Predators: instances of particular behaviours were noted by the experimenters during trials and in some cases verified
using video recordings. Data were recorded by TD and RL (jumping spiders), HR (praying mantises), JD, HJJ, JR, JSa (crab spiders).

Timing and spatial scale Discrimination Ability experiment: experiment ran from December 2021 to May 2022, with the testing phase beginning on 5th April
2022. Feeders were spaced within an area approximately 150 m x 150 m.
Multiple Models experiment: ran from October 2022 to March 2023, with the testing phase beginning on 13th January 2023.
Validation test: ran from October to December 2023, with the testing phase beginning on 18th December.

Data exclusions Discrimination Ability experiment: an additional two feeders were removed from the experiment before the start of the testing phase
as they did not receive sufficient great tit visits.
Multiple Models experiment: some dishes were opened by mice, which were excluded from the analysis.
Validation test: no exclusions
Trait Salience experiment: we drew a predetermined cut-off of 13 out of 16 successes during a single trial in the training phase.
Chicks that did not reach this cut-off did not proceed to the testing phase. We also excluded testing trials where there was <0.1s in
latency to attack towards the fly and wasp stimuli, on the basis that chicks were not showing detectable discrimination among
stimulus types.
Invertebrate Predators experiment: no exclusions.

Reproducibility No extra tests of reproducibility were carried out beyond the repetition inherent in the experimental design described above.

Randomization Discrimination Ability and Multiple Models experiments, Validation test: within each feeder, a defined set of stimuli were allocated to
dishes at random. Treatments for the Multiple Models experiment were assigned to feeders according to spatial grouping, to
minimise the numbers of birds that would visit both treatment types.
Trait Salience: each trial included presentations of 6 fly and 6 wasp stimuli, as well as 4 probe types that were selected at random
from a defined pool (without replacement, to ensure roughly equal sample sizes for each type). Within a trial, order was randomised
with the constraint that each half (8 presentations) would contain 3 fly, 3 wasp and 2 probe presentations, to prevent the chicks
experiencing long runs of the same stimulus type or reward status.
Invertebrate Predators: for mantises and jumping spiders, order of trials was randomised within the testing phase. For crab spiders,
individuals were assigned to treatments at random.

Blinding Data transcribed from video recordings were carried out with reference to dish location or trial number and without knowledge of
the stimulus type in question (video resolution was not sufficient to reliably identify stimuli visually).

Did the study involve field work? Yes No

Field work, collection and transport

Field conditions Highly variable. During one period of the Multiple Models experiment data collection was paused as snow and below-zero
temperatures were causing the dishes to freeze shut. Detailed weather records were not maintained due to the long period of data
collection.

Location Madingley Wood, Cambridgeshire, UK (52.217 N, 0.049 E)

Access & import/export Permission for carrying out fieldwork was obtained from University Farm & Rural Estate (University of Cambridge) who manage the
woodland

Disturbance Disturbance from researcher visits was minimised as most data recording was done by trail cameras and each feeder was visited only
once per day, with visits taking about half an hour per feeder.
April 2023

Reporting for specific materials, systems and methods

3
We require information from authors about some types of materials, experimental systems and methods used in many studies. Here, indicate whether each material,
system or method listed is relevant to your study. If you are not sure if a list item applies to your research, read the appropriate section before selecting a response.

nature portfolio | reporting summary

Materials & experimental systems Methods
n/a Involved in the study n/a Involved in the study
Antibodies ChIP-seq
Eukaryotic cell lines Flow cytometry
Palaeontology and archaeology MRI-based neuroimaging
Animals and other organisms
Clinical data
Dual use research of concern
Plants

Animals and other research organisms

Policy information about studies involving animals; ARRIVE guidelines recommended for reporting animal research, and Sex and Gender in
Research

Laboratory animals Domestic chicks (Gallus gallus domesticus) were obtained from a commercial hatchery on the day after hatching, with testing phase
of the experiment when they were 17-20 days old. Praying mantises of three species (Rhombodera kirbyi (n = 5), Polyspilota
aeruginosa (n = 1) and Pseudoxyops perpulchra (n = 2)) and jumping spiders (Phidippus audax) were obtained from commercial
sellers at a range of ages from 3rd instar to adult.

Wild animals Wild birds were free to come and go from feeding stations at will, they were not captured at any point during the experiment. Crab
spiders (adult, exact age unknown) were captured from the wild by hand in specimen tubes around the Quinta de São Pedro field
research station (38.568 N, 9.193 W) and surrounding areas of Sobreda, Portugal, and transported on foot. They were kept for
approximately 48 h before being released in the same location.

Reporting on sex Most of our study organisms were not sexable at the time of experiment. Those where sex was identifiable consisted of a mix of
sexes, and we have no reason to assume sex biases in the other cases. We did not analyse effects of sex as this was not a variable of
interest and would only have been possible in limited sub-sets of the data.

Field-collected samples No laboratory samples collected from the field

Ethics oversight The Trait Salience experiment was approved by Newcastle University AWERB committee (project ID 966). Wild bird experiments
(Discrimination Ability and Multiple Models) were approved by AWERB committees at University of Nottingham (project ID 260) and
University of Cambridge (ref. NR2022/60).
Note that full information on the approval of the study protocol must also be provided in the manuscript.

Plants
Seed stocks Report on the source of all seed stocks or other plant material used. If applicable, state the seed stock centre and catalogue number. If
plant specimens were collected from the field, describe the collection location, date and sampling procedures.

Novel plant genotypes Describe the methods by which all novel plant genotypes were produced. This includes those generated by transgenic approaches,
gene editing, chemical/radiation-based mutagenesis and hybridization. For transgenic lines, describe the transformation method, the
number of independent lines analyzed and the generation upon which experiments were performed. For gene-edited lines, describe
the editor used, the endogenous sequence targeted for editing, the targeting guide RNA sequence (if applicable) and how the editor
was applied.
Authentication Describe any authentication procedures for each seed stock used or novel genotype generated. Describe any experiments used to
assess the effect of a mutation and, where applicable, how potential secondary effects (e.g. second site T-DNA insertions, mosiacism,
off-target gene editing) were examined.
April 2023