Current Biology 23, 168–171, January 21, 2013 ª2013 Elsevier Ltd All rights reserved
http://dx.doi.org/10.1016/j.cub.2012.11.058
Report
Artificial Selection on Relative
Brain Size in the Guppy Reveals Costs
and Benefits of Evolving a Larger Brain
Alexander Kotrschal,1 Björn Rogell,1 Andreas Bundsen,1
Beatrice Svensson,1 Susanne Zajitschek,2
Ioana Brännström,1 Simone Immler,2 Alexei A. Maklakov,1
and Niclas Kolm1,*
1Animal Ecology, Department of Ecology and Genetics
2Evolutionary Biology, Department of Ecology and Genetics
Uppsala University, Norbyvägen 18D, 752 36 Uppsala, Sweden
Summary
The large variation in brain size that exists in the animal
kingdom has been suggested to have evolved through the
balance between selective advantages of greater cognitive
ability and the prohibitively high energy demands of a larger
brain (the ‘‘expensive-tissue hypothesis’’ [1]). Despite over
a century of research on the evolution of brain size, empirical
support for the trade-off between cognitive ability and energetic costs is based exclusively on correlative evidence [2],
and the theory remains controversial [3, 4]. Here we provide
experimental evidence for costs and benefits of increased
brain size. We used artificial selection for large and small
brain size relative to body size in a live-bearing fish, the
guppy (Poecilia reticulata), and found that relative brain
size evolved rapidly in response to divergent selection in
both sexes. Large-brained females outperformed smallbrained females in a numerical learning assay designed to
test cognitive ability. Moreover, large-brained lines, especially males, developed smaller guts, as predicted by the
expensive-tissue hypothesis [1], and produced fewer
offspring. We propose that the evolution of brain size is
mediated by a functional trade-off between increased cognitive ability and reproductive performance and discuss the
implications of these findings for vertebrate brain evolution.
Results
One of the most distinct features of the human brain is its
unusually large size in relation to body size [1, 5, 6]. Yet, variation in relative brain size is extensive at all taxonomic levels
across vertebrates [7]. Theory aimed at accounting for this
variation argues that the brain has evolved through a balance
between selection for increased brain size and evolutionary
constraints [7, 8]. In particular, selection on cognitive ability
has been proposed as a key factor driving the evolution of
larger brains [6, 7, 9]. This hypothesis is supported by empirical evidence from interspecific comparisons of brain size in
relation to fitness-related behaviors believed to be associated
with cognitive ability across a variety of taxa [10–17]. Moreover, larger brains seem to confer advantages in novel or challenging environments [14, 18]. But if a larger brain provides
a selective advantage through greater cognitive ability, what
limits the evolution of increased relative brain size in natural
populations? Alongside the digestive tract, the brain is the
most energetically expensive organ in the body ([12]; for
*Correspondence: [email protected]
review, see [13]). Because of this, constraints originating
from the costs of maintaining the brain tissue have been suggested to limit brain size [1]. The original ‘‘expensive-tissue
hypothesis’’ [1] attempted to explain variation in primate brain
size through a trade-off between brain tissue and gut tissue.
However, recent comparative analyses have not supported
this hypothesis [3] and have instead suggested that the
trade-off occurs between brain size and other costly aspects
of an organism’s biology, such as investment in muscle tissue
[11], gonads [4], fat storage [3], or reproductive effort [19].
The aforementioned comparative studies suggested that
evolution of a larger brain is driven by a selective advantage
of greater cognitive ability but at the same time constrained
by trade-offs with investment in other traits. However, correlative comparative analyses make it difficult to exclude the
possibility that these patterns have arisen as a result of selection upon unknown correlated traits [2, 20]. Artificial selection,
on the other hand, is a powerful tool to provide experimental
evidence for costs and benefits of larger brain size [21, 22].
We therefore used artificial selection on relative brain size in
a live-bearing fish, the guppy (Poecilia reticulata), to provide
a direct test of the prediction that increased brain size is genetically associated with increased cognitive ability but that
a large brain is also traded off against gut size and results in
reduced reproductive performance. First, we investigated
the evolutionary response to divergent selection on relative
brain size. Second, we tested the cognitive ability of largeand small-brained individuals using an associative learning
assay designed to investigate numerical quantification, a relatively advanced form of cognition [23]. Third, we tested for the
correlated evolutionary response of gut size in response to
direct selection on brain size. Fourth, we tested whether
important proxies of reproductive fitness (offspring number,
offspring size, age at first reproduction) are affected by brain
size evolution.
We selected for large and small brain size (brain mass) relative to body size (body length; see Supplemental Experimental
Procedures available online for details) in replicated lines and
found that brain size responded rapidly to divergent selection
(Figure 1; Table S1; Figure S1). Relative brain size was already
9% larger in the upward- compared to the downward-selected
lines after two generations of selection (estimated difference
across ‘‘down’’ and ‘‘up’’ selection lines for adults [henceforth
b, presented with 95% credible intervals (CI)]: b = 0.071 [0.06;
0.08] log (mg)/log (mm), p < 0.001; Figure 1; Table S1). This
difference was already apparent in newborn fish, as indicated
by a greater optic tectum width measured from digital microscopic images (b = 0.041 [0.019; 0.061], p < 0.001; Table S1).
We used optic tectum width, an accurate predictor of overall
brain size [24, 25], as a proxy for brain size of neonates
because brains of neonates were too small to be removed
and weighed. There were no significant main effects of brain
size selection on body size in newborns or in adults (neonates:
b = 0.06 [20.17; 0.27], p = 0.60; Table S2; adults: b = 20.058
[20.26; 0.16] mm, p = 0.59; Table S2). The realized heritability
of relative brain size was substantial and congruent between
sexes: 0.48 (0.38; 0.63) in females and 0.45 (0.33, 0.59) in
males.
Costs and Benefits of Evolving a Larger Brain
169
0.9
7
Number of correct choices
0.7
Relative brain size
0.5
0.3
0.1
-0.1
6
5
4
3
2
Small
brained
-0.3
Large
brained
Females
-0.5
Small
brained
Large
brained
Males
Figure 2. Cognitive Ability Improves with Increased Brain Size
-0.7
-0.9
F0
F1
Females
F2
F0
F1
F2
Males
Figure 1. Relative Brain Size Responds Rapidly to Divergent Selection
F0 is the parental generation; F1 and F2 are the first and second brain
weight-selected generation, respectively. Depending on replicate, secondgeneration large- and small-brained females (left panel) differ by 8.0%–9.3%
(p < 0.001) in relative brain size, while second-generation large- and
small-brained males (right panel) differ by 5.0%–8.3% (p < 0.001). Depicted
are the mean and SE values for residuals of brain weight regressed on
body size within each generation and replicate. See also Figure S1 for
experimental procedures and plots of raw data, and Table S1 for detailed
results.
We then used a numerical learning test to assess the cognitive ability of 48 of these fish and found an interaction between
selection and sex (generalized linear mixed model [GzLMM],
n = 47, selection: c2 = 2.12, df = 1, p = 0.145, sex: c2 = 3.47,
df = 1, p = 0.063, selection 3 sex: c2 = 5.44, df = 1, p = 0.020;
Figure 2). This interaction was caused by large-brained
females outperforming small-brained females in the learning
assay (GzLMMfemales, n = 23, selection: c2 = 7.58, df = 1, p =
0.006; Figure 2), thereby providing direct evidence for a positive association between relative brain size and cognitive
ability. Interestingly, no difference was found between males
of different brain sizes (GzLMMmales, n = 24, selection: c2 =
0.38, df = 1, p = 0.535; Figure 2).
We weighed the empty guts of fish from different lines and
found that selecting on large brain size caused a correlated
evolutionary decrease in gut size (b = 20.81 [21.14; 20.49],
p < 0.001). Gut size differed between selection lines by 20%
(CI = 0.11; 0.29) and 8% (CI = 0.007; 0.17) for males and
females, respectively (Figure 3A; Table S3). Our analysis of
the reproductive costs associated with increased brain size
showed that offspring number (b = 20.19 [20.33; 20.046],
p = 0.01; Figure 3B; Table S4), but not offspring size (b = 0.06
[20.17; 0.27], p = 0.60; Table S2) or age at first reproduction
(b = 0.71 [23.94; 5.28], p = 0.76; Table S4; for all model selection criteria, see Table S5), was affected by selection on brain
size. Offspring number was thus 19% lower in the largebrained lines as compared to the small-brained lines, which
shows that the evolution of a larger brain has a strong negative
effect on an important reproductive trait.
Large-brained females outperform small-brained females in a numerical
learning task (p = 0.006), whereas there is no difference in males (p =
0.535). Depicted are the mean and SE values for the number of times, out
of eight tests, that an individual chose the correct option (after accounting
for the number of times each individual participated in the trials) of either
two or four objects in females and males selected for large and small brain
size. See Figure S2 for scheme of the testing apparatus.
Discussion
Our results show that the evolution of relative brain size in
vertebrates can be a fast process when under strong directional selection. The realized heritability of relative brain size
was also substantial in both sexes, matching those detected
in mother-offspring studies [26]. Furthermore, our demonstration of a direct association between brain size and cognition
suggests that selection for increased cognitive ability can be
mediated through rapid evolution of brain size. Because
cognitive abilities are important to facilitate behaviors such
as finding food, avoiding predation, and obtaining a mate, individuals with increased cognitive abilities are likely to have
higher reproductive success in the wild [14]. However, the
link between a larger brain and cognitive abilities has recently
been challenged because of the high cognitive capacity of
some small-bodied and small-brained invertebrates such as
bees and ants [27]. Moreover, the field of cognitive evolution
has recently shifted toward emphasizing fine-scale structural
differences in the brain as the main feature linking brain
morphology and cognitive ability [2, 7]. Our results now show
that larger brains really can be better, at least on the withinpopulation level, and that variation in a relatively crude
measure of brain morphology, relative brain size, is directly
associated with variation in cognitive ability. Interestingly,
the effect of relative brain size on cognitive ability was only
evident in females. We offer two explanations for the sexspecific response in our experiment. First, relative brain size
may not reflect cognitive ability in males to the same extent
as in females. We find this explanation unlikely because in
most species, general brain functions are usually shared
between the sexes [28]. Second, the design of our cognitive
test may have been more suitable for testing female cognitive
ability. In the guppy, females are more active and innovative
while foraging [29], most likely reflecting the fact that female
reproductive success is mainly food limited whereas males
are limited by their access to females [30]. Because females
feed more, they may thus have had more time to associate
the cue with food in our experimental design. Moreover, in
some populations, female guppies choose their partner based
A
B
5.5
25
4.5
4
21
3.5
19
3.0
17
Male gut mass (mg)
Female gut mass (mg)
5
23
Number of offspring in first parturition
Current Biology Vol 23 No 2
170
Figure 3. Individuals Selected for Large Brain
Size Decrease Gut Size and Offspring Production
7.5
7.0
6.5
6.0
5.5
5.0
Small
brained
Large
brained
Females
Small
brained
Large
brained
Small-brained
fish
Large-brained
fish
Males
on male melanin spot coloration [31]. The female visual system
may thus be preadapted for more efficient processing of the
black symbols used in this experiment.
Our demonstration of a reduction in gut size and offspring
number in the experimental populations selected for larger
relative brain size provides compelling experimental evidence
for the cost of increased brain size. This study thereby
provides the first direct support for the expensive-tissue
hypothesis [1] and corroborates recent comparative analyses
suggesting trade-offs between brain size and different costly
tissues in mammals [3]. The original expensive-tissue hypothesis proposed that the increasingly greater incorporation of
animal products into the primate diet allowed for a smaller
gut, thereby freeing energy for brain development. The greater
cognitive abilities associated with larger brains in turn enabled
hominids to exploit even higher-quality food sources, reducing
gut size further. An alternative mechanism is that neural development of the gut is traded off against neural development of
the brain. The gut forms a highly conserved, neuron-rich
control center of the enteric nervous system that controls
digestion [32] and is sometimes referred to as the ‘‘second
brain’’ [33]. This is an important additional aspect of the function of the gut, which we suggest future research should
target to fully understand the trade-off between the brain
and the digestive system. Regardless of mechanism, in the
controlled environment of our experimental setup, diet was
kept constant. Therefore, in the absence of any cognitive
benefits related to increased brain size, the genetic trade-off
between investment in brain size and other expensive tissues,
such as the gut, might have caused the reduction in reproductive performance that we observed.
Offspring number is one of the key determinants of lifetime
reproductive success [34], and reduction in this trait is very
likely to result in fitness costs. Because of this, we propose
that the existing variation in brain size among vertebrates
has been generated through the opposing evolutionary forces
of cognitive benefits and reproductive costs. Finally, our
results might help explain the evolution of larger brains in
primates and cetaceans (whales and dolphins) in comparison
to most other mammals. Both primates and cetaceans have
unusually low fertility among mammals [35]. This decrease in
fertility may therefore be a result of either an evolutionary
increase in relative brain size or, alternatively, the change
toward a slower life history [35] that allowed these orders to
evolve their unusually large brains.
Supplemental Information
Supplemental Information includes two figures, five tables, and Supplemental Experimental Procedures and Results and can be found with this
article online at http://dx.doi.org/10.1016/j.cub.2012.11.058.
(A) In guppies selected for large and small brain
size, gut size differed by 8% in females and
20% in males (p < 0.001, after controlling for
body size). Depicted are the mean and SE values.
See Table S3 for detailed results.
(B) Pairs selected for large brain size showed
a 19% decrease in the number of offspring in
the first clutch (p = 0.01, after controlling for
female age at reproduction). Depicted are the
mean and SE values. See Table S4 for detailed
results.
Acknowledgments
We thank Gunilla Rosenqvist for providing us with the animals that formed
our experimental stock population. We acknowledge valuable comments
by Göran Arnqvist, David Berger, Arild Husby, Kurt Kotrschal, and Locke
Rowe. A.K. was funded by the Carl Tryggers Stiftelse (to N.K.) and the
Austrian Science Fund (J 3304-B24 to A.K.), B.R. was funded by the European Research Council (to A.A.M.), A.A.M. was funded by the European
Research Council and the Swedish Research Council, and N.K. was funded
by the Swedish Research Council. All experiments were performed in accordance with the ethical regulations for research involving animal subjects in
Uppsala, Sweden.
Received: October 4, 2012
Revised: November 9, 2012
Accepted: November 29, 2012
Published: January 3, 2013
References
1. Aiello, L.C., and Wheeler, P. (1995). The expensive-tissue hypothesis:
the brain and the digestive system in human and primate evolution.
Curr. Anthropol. 36, 199–221.
2. Healy, S.D., and Rowe, C. (2007). A critique of comparative studies of
brain size. Proc. Biol. Sci. 274, 453–464.
3. Navarrete, A., van Schaik, C.P., and Isler, K. (2011). Energetics and the
evolution of human brain size. Nature 480, 91–93.
4. Pitnick, S., Jones, K.E., and Wilkinson, G.S. (2006). Mating system and
brain size in bats. Proc. Biol. Sci. 273, 719–724.
5. Darwin, C. (1871). The Descent of Man, and Selection in Relation to Sex
(London: John Murray).
6. Jerison, H.J. (1973). Evolution of the Brain and Intelligence (New York:
Academic Press).
7. Striedter, G.F. (2005). Principles of Brain Evolution (Sunderland, MA:
Sinauer Associates).
8. Roff, D.A., Mostowy, S., and Fairbairn, D.J. (2002). The evolution of
trade-offs: testing predictions on response to selection and environmental variation. Evolution 56, 84–95.
9. Dunbar, R.I.M. (1998). The social brain hypothesis. Evol. Anthropol. 6,
178–190.
10. Brown, C. (2012). Tool use in fishes. Fish Fish. 13, 105–115.
11. Isler, K., and van Schaik, C. (2006). Costs of encephalization: the energy
trade-off hypothesis tested on birds. J. Hum. Evol. 51, 228–243.
12. Laughlin, S.B., de Ruyter van Steveninck, R.R., and Anderson, J.C.
(1998). The metabolic cost of neural information. Nat. Neurosci. 1,
36–41.
13. Niven, J.E., and Laughlin, S.B. (2008). Energy limitation as a selective
pressure on the evolution of sensory systems. J. Exp. Biol. 211, 1792–
1804.
14. Sol, D. (2009). Revisiting the cognitive buffer hypothesis for the evolution of large brains. Biol. Lett. 5, 130–133.
15. Tebbich, S., and Bshary, R. (2004). Cognitive abilities related to tool use
in the woodpecker finch, Cactospiza pallida. Anim. Behav. 67, 689–697.
16. van Schaik, C.P., and Pradhan, G.R. (2003). A model for tool-use
traditions in primates: implications for the coevolution of culture and
cognition. J. Hum. Evol. 44, 645–664.
Costs and Benefits of Evolving a Larger Brain
171
17. Gonzalez-Voyer, A., Winberg, S., and Kolm, N. (2009). Social fishes and
single mothers: brain evolution in African cichlids. Proc. Biol. Sci. 276,
161–167.
18. Maklakov, A.A., Immler, S., Gonzalez-Voyer, A., Rönn, J., and Kolm, N.
(2011). Brains and the city: big-brained passerine birds succeed in
urban environments. Biol. Lett. 7, 730–732.
19. Isler, K., and van Schaik, C.P. (2009). The Expensive Brain: a framework
for explaining evolutionary changes in brain size. J. Hum. Evol. 57,
392–400.
20. Martins, El. (2000). Adaptation and the comparative method. Trends
Ecol. Evol. 15, 296–299.
21. Falconer, D.S., and Mackay, T.F.C. (1996). Introduction to Quantitative
Genetics, Fourth Edition (Harlow, UK: Longmans Green).
22. Roderick, T., Wimer, R., and Wimer, C. (1976). Genetic manipulation of
neuroanatomical traits. In Knowing, Thinking, and Believing, L.
Petrinovich and J. McGaugh, eds. (New York: Plenum), pp. 143–178.
23. Agrillo, C., Dadda, M., and Bisazza, A. (2007). Quantity discrimination in
female mosquitofish. Anim. Cogn. 10, 63–70.
24. Kotrschal, A., Sundström, L.F., Brelin, D., Devlin, R.H., and Kolm, N.
(2012). Inside the heads of David and Goliath: environmental effects
on brain morphology among wild and growth-enhanced coho salmon
Oncorhynchus kisutch. J. Fish Biol. 81, 987–1002.
25. Kotrschal, A., Rogell, B., Maklakov, A.A., and Kolm, N. (2012). Sexspecific plasticity in brain morphology depends on social environment
in the guppy, Poecilia reticulata. Behav. Ecol. Sociobiol. 66, 1485–1492.
26. Cheverud, J.M., Falk, D., Vannier, M., Konigsberg, L., Helmkamp, R.C.,
and Hildebolt, C. (1990). Heritability of brain size and surface features in
rhesus macaques (Macaca mulatta). J. Hered. 81, 51–57.
27. Chittka, L., and Niven, J. (2009). Are bigger brains better? Curr. Biol. 19,
R995–R1008.
28. Gahr, M. (1994). Brain structure: Causes and consequences of brain
sex. In The Differences between the Sexes, R.V. Short and E. Balaban,
eds. (Cambridge: Cambridge University Press).
29. Laland, K.N., and Reader, S.M. (1999). Foraging innovation is inversely
related to competitive ability in male but not in female guppies.
Behav. Ecol. 10, 270–274.
30. Houde, A. (1997). Sex, Color, and Mate Choice in Guppies (Princeton,
NJ: Princeton University Press).
31. Brooks, R., and Caithness, N. (1995). Female choice in a feral guppy
population: are there multiple cues? Anim. Behav. 50, 301–307.
32. Olsson, C. (2009). Autonomic innervation of the fish gut. Acta
Histochem. 111, 185–195.
33. Gershorn, M. (1998). The Second Brain (New York: Harper Perennial).
34. Roff, D.A. (1992). The Evolution of Life Histories (New York: Chapman &
Hall).
35. Jones, J.H. (2011). Primates and the evolution of long, slow life histories.
Curr. Biol. 21, R708–R717.
Current Biology, Volume 23
Supplemental Information
Artificial Selection on Relative
Brain Size in the Guppy Reveals Costs
and Benefits of Evolving a Larger Brain
Alexander Kotrschal, Björn Rogell, Andreas Bundsen, Beatrice Svensson, Susanne Zajitschek,
Ioana Brännström, Simone Immler, Alexei A. Maklakov, and Niclas Kolm
Supplemental Inventory
Figure S1: This figure accompanies Figure 1. It depicts the selection regime and the raw data of brain size
plotted against body size for two generations for the two sexes separately.
Figure S2: This Figure accompanies Figure 2 and shows the testing apparatus in which guppies were
conditioned to learn to discriminate four from two symbols.
Table S1: This table gives the parameters of models used to analyze the impact of various factors on
brain size in adult and juvenile guppies selected for large and small relative brain size.
Table S2: This table gives the parameters of models used to analyze the impact of various factors on
body size in adult and juvenile guppies.
Table S3: This table gives the parameters of models used to analyze the impact of various factors on gut
size in adult guppies.
Table S4: This table gives the parameters of models used to analyze the impact of various factors on
aspects of fecundity in female guppies.
Table S5: This table gives the specifics of the model selection process for all models used in this study
(apart from the analysis of numerical learning ability).
Supplemental Experimental Procedures and Results
Here we describe in greater detail how:
- we selected for large and small relative brain size
- we quantified body and brain size of juveniles
- we conducted the test of numerical ability
- we tested for a potential sensory bias
- we quantified gut size
- we analyzed our data
- we calculated realized heritabilities of brain size.
We furthermore provide additional results for the numerical learning test and its control.
Supplemental References
Additional references for Supplemental Information.
Figure S1, Related to Figure 1.
(A) Artificial selection procedure for large and small relative brain size in guppies. Depicted is the
experimental procedure for each of the three replicates. For generation F0, we used guppies
(Poecilia reticulata) of a large, outbred stock population to set up three replicate populations of
75 breeding pairs. Since it is impossible to determine brain size in live fish, we sacrificed the
parents after offspring production. Brains of those parents were removed and weighed. From
the 75 pairs of every population we kept the offspring of the 15 pairs with the relatively largest
and smallest brains (controlled for body size) to start (F0) the “up” and “down” selected lines in
the three replicates. Within each of these six lines we used two males and females of every
family to form 30 breeding pairs for the next generation. We randomly assigned partners, but
avoided full-sib pairs. For the F2 generation we followed the same procedure.
(B) Relative brain size responded rapidly to divergent selection in guppies. Depicted is the
relationship between body size and brain size, where F1 and F2 are the first and second brain
size-selected generations, respectively. Top row: males, bottom row: females. Filled circles
denote the lines selected for large brain size, empty circles the selection lines for small brains.
Replicates are shown together here, but were controlled for in the analysis.
Figure S2, Related to Figure 2.
Test of cognitive performance in guppies (Poecilia reticulata) selected for large and small
relative brain size. We trained 48 adults to associate a specific number of symbols (two or four)
with food. Training occurred twice per day. To test whether they learned to discriminate two
symbols from four we presented the stimulus card pairs, without food reward, on opposite
sides of their individual holding tanks. Then we lifted see-through dividers, which confined the
fish to the center of the tank, and assessed to which side the fish swam. Tests of cognitive ability
were performed after each three-day period of training resulting in a total of 48 trainings and
eight cognitive tests.
Table S1. The Effect of Body Size, Sex, and Brain Size Selection on Brain Size in Juvenile and
Adult Guppies
Adults Generation F1
Fixed terms
Intercept
Log size
Sex
Selection
Log size * Sex
Random terms
Residual
Replicate line
Replicate line * Sex
Adults Generation F2
Fixed terms
Intercept
Log size
Selection
Random terms
Female Residual
Male Residual
Replicate line
Replicate line * Sex
Juveniles Generation F4
Fixed terms
Intercept
Log size
Selection
Random terms
Residual
Replicate line
Estimate ()
-2.33
1.18
-0.8
0.052
0.26
Variance
0.0027
0.004
0.00022
Lower CI
-2.74
1.05
-1.41
0.04
0.065
Lower CI
0.0023
2.45*10-10
1.2*10-11
Upper CI
-1.94
1.3
-0.22
0.063
0.46
Upper CI
0.0031
0.0082
0.001
P
<0.001
<0.001
0.009
<0.001
0.013
P
NA
NA
NA
Estimate ()
-2.81
1.32
0.071
Variance
0.0022
0.0036
0.0001126
3.56*10-5
Lower CI
-2.9
1.29
0.06
Lower CI
0.0018
0.0029
6.08*10-15
1.95*10-12
Upper CI
-2.72
1.35
0.082
Upper CI
0.0027
0.0044
0.0002024
0.00017
P
<0.001
<0.001
<0.001
P
NA
NA
NA
NA
Estimate ()
-0.32
0.27
0.041
Variance
0.0017
0.00056
Lower CI
-0.70
0.08
0.019
Lower CI
0.0011
1.37*10-12
Upper CI
0.06
0.46
0.061
Upper CI
0.0024
0.00082
P
0.10
0.005
<0.001
P
NA
NA
Adults: For each generation, the table reports the result of a GLMM calculated in the R package
MCMCglmm. The “Intercept” term is set to the intercept of the regression of log body size on
log brain size of females in the brain selection regime for smaller brains. The other estimates
represent differences between the intercept (of small brained females) and the specified
experimental unit (according to the default contrast matrix in R, “contrast treatments”). Thus,
“Selection” is the difference in relative brain size between females in selection for smaller
(“down”) and larger (“up”) brains, “Sex” is the difference between females selected for smaller
brains and males selected for smaller brains, “Log Size” is the effect of log body size on log brain
size of individuals selected for smaller brains, the interaction “Log Size * Sex” indicates the
differences in the slopes of the regression of log body size on log brain size between the sexes.
“Estimate” is the parameter estimate; “Lower CI” and “Upper CI” denote its associated 95 %
credibility intervals. “P” is the probability of falsely rejecting the null hypothesis that the
parameter equals 0. For the random effects, P-values are not available, but the variance
component explained by each random effect is accompanied by upper and lower CIs. “Female
Residual” is the residual variance between female individuals in brain size, “Male residual” is the
male residual variance in brain size, “Replicate line” is the variance explained by replicate line,
“Replicate line * Sex” is the variance explained by the interaction between sex and replicate
line. Juveniles: The effect of log offspring body size and brain size selection on log offspring
brain size (optic tectum width) in F4 juveniles. The terms are set analogous to the adult model.
Table S2. Body Size in Juvenile and Adult Guppies Selected for Large and Small Relative
Brain Size
Adults
Fixed terms
Intercept
Selection
Sex
Random terms
Female Residual
Male Residual
Replicate line
Sex * Replicate line
Juveniles
Fixed terms
Intercept
Selection
Mother size
# Siblings
Selection * Mother size
Selection * # Siblings
Mother size * # Siblings
Selection * Mother size * # Siblings
Random terms
Residual
Estimate ()
24.84
-0.06
-8.89
Variance
140.77
0.65
0.19
0.09
Lower CI
24.30
-0.26
-9.48
Lower CI
129.98
0.52
7.9*10-16
4.72*10-11
Upper CI
25.36
0.16
-8.35
Upper CI
149.83
0.80
0.32
0.34
P
<0.001
0.571
<0.001
P
NA
NA
NA
NA
Estimate ()
7.46
0.057
0.32
0.0038
-0.32
-0.018
-0.18
0.26
Variance
0.18
Lower CI
7.31
-0.17
0.11
-0.15
-0.59
-0.25
-0.39
0.011
Lower CI
0.12
Upper CI
7.62
0.27
0.54
0.17
-0.068
0.22
0.063
0.5
Upper CI
0.25
P
<0.001
0.602
0.004
0.980
0.019
0.875
0.123
0.037
P
NA
Adults: The effect of sex and brain size selection on adult body size. The terms are set analogous
to the juvenile model. Juveniles: The effect of maternal size, number of siblings and brain size
selection on juvenile body size. The terms are set analogous to Table S1, but the “Intercept”
term here is interpreted as the mean body size of juveniles selected for smaller brains (since all
covariates are standardized to a mean of 0). “Mother size” is the effect of maternal size on
juvenile body size, “# Siblings” is the effect the number of siblings has on juvenile body size and
the interaction “Selection * Mother size” indicates that the slopes of the regression of maternal
size on juvenile body size depends on the selection treatment. This suggests that the positive
effect of maternal size on offspring size was stronger in the “down” selected guppies. The
interaction “Selection * # Siblings” shows that the slopes of the regression of number of siblings
on juvenile body size also depends on selection treatment. The interaction “Mother size * #
Siblings” denotes an interaction among the two covariates and the three way interaction
“Selection * Mother size * # Siblings” indicates that this interaction is different in the two brain
selection regimes. This interaction indicates that the negative relationship between maternal
size and number of offspring is weaker in the “up” selected guppies. Parts of these relationships
may relate to the fact that large-brained guppies had a lower reproductive investment than
small-brained individuals.
Table S3. The Effect of Body Size, Sex, and Brain Size Selection on Gut Size
Fixed terms
Intercept
Standardized size
Sex
Selection
Standardized size * Sex
Random terms
Female Residual
Male Residual
Replicate line
Sex * Replicate line
Estimate ()
23.30
4.96
-18.34
-0.81
-4.37
Variance
31.36
1.14
0.11
0.05
Lower CI
22.19
3.94
-19.43
-1.14
-5.46
Lower CI
23.06
0.91
2.86*10-12
1.0*10-11
Upper CI
24.46
6.06
-17.22
-0.49
-3.31
Upper CI
40.11
1.40
0.28
0.18
P
<0.001
<0.001
0.009
<0.001
<0.001
P
NA
NA
NA
NA
Since size and sex are highly correlated in guppies, we standardized size to a mean of 0 and a
standard deviation of 1 within each sex in order to avoid problems associated with collinearity.
Apart from the different response variable (here gut mass instead of brain mass), the terms are
set analogous to Table S1.
Table S4. Reproductive Traits
Offspring number
Fixed terms
Intercept
Age at offspring
Selection
Random terms
Residual
Replicate line
Age at first reproduction
Fixed terms
Intercept
Selection
Random terms
Residual
Estimate ()
1.93
0.3
-0.19
Variance
0.04
0.89
Lower CI
1.39
0.23
-0.33
Lower CI
0.00027
2.7*10-8
Upper CI
2.43
0.39
-0.046
Upper CI
0.086
0.92
P
0.009
<0.001
0.0105
P
NA
NA
Estimate ()
140.77
0.71
Variance
197.3
Lower CI
129.98
-3.94
Lower CI
151.7
Upper CI
149.83
5.28
Upper CI
241.7
P
<0.001
0.757
P
NA
Offspring number: The effect of age at first parturition and brain size selection on number of
offspring. The terms are set analogous to Table S1. Age at first reproduction: The effect of brain
size selection on the age of first reproduction. The terms are set analogous to Table S1.
Table S5. Model Simplification Procedure
DIC
Adult brain size – F1
Random effects
Null
+ Replicate line
+ Sex : Replicate line
+ Sex : Residual
Fixed effects
Full
- Log size * Sex * Selection
- Sex * Selection
- Log size * Selection
- Log size * Sex
Adult brain size – F2
Random effects
Null
+ Replicate line
+ Sex : Replicate line
+ Sex : Residual
Fixed effects
Full
- Log size * Sex * Selection
- Sex * Selection
- Log size * Selection
- Log size * Sex
- Sex
- Selection
Juvenile brain size
Random effects
Null
+ Replicate line
Fixed effects
Full
- Log size * Selection
- Log size
Juvenile body size
Random effects
Null
+ Replicate line
Fixed effects
- Full
- Selection * Maternal size * # siblings
Adult body size
Random effects
Null
-998.4
-1041.0 *
-1040.4
-1038.6
-1041.0
-1042.8
-1043.5
-1045.3 *
-1040.3
-1029.5
-1029.1
-1029.0
-1036.4 *
-1036.4
-1038.5
-1040.5
-1042.4
-1044.2
-1046.1 *
-916.3
-207.8
-207.56 **
-207.56
-209.2*
-203.4
77.0 *
77.4
77.0 *
80.1
1212.4
+ Replicate line
+ Sex : Replicate line
+ Sex : Residual
+ Sex residual – Sex : Replicate line
Fixed effects
Full
- Sex * Selection
- Selection
Gut mass
Random effects
Null
+ Replicate line
+ Sex : Replicate line
+ Sex : Residual
Fixed effects
Full
- Log size * Sex * Selection
- Sex * Selection
- Log size * Selection
- Selection
Number of offspring
Random effects
Null
+ Replicate line
Fixed effects
Full
- Selection * Age at first reproduction
- Selection
Age at first reproduction
Random effects
Null
+ Replicate line
Fixed effects
Full
- Selection
1212.4
1212.3
1115.7 *
1118.6
1115.7
1114.6 **
1112. *
1460.3
1460.6
1461.0
1159.6 *
1159.6
1157.7
1155.7
1153.6 *
1175.0
737.5
734.5 *
734.5
733.3 *
737.6
1204.5
1198.5 *
1198.5 *
1196.6 **
The models used to analyze body size, offspring body size, brain size, age at first offspring and
fecundity were built based on the best (lowest) Deviance Information Criterion (DIC), starting
with 1) a model with a full fixed effect formula (all fixed main effects and their interactions), to
which the random effects were added sequentially according to the simplification tables. When
the random effect formula that had the lowest DIC value had been chosen, we continued with
simplifying the fixed effect formula. The fixed effects were removed sequentially starting with
the highest order interactions and then continuing until the model only contained statistically
significant parameters. The removal procedure of fixed effects that did not improve the model
was based on obtaining the lowest DIC value. Thus, in the tables below, the random effect
formula “Null” denotes that no random effects are added to the model and a “+” denotes the
inclusion of a random effect. For the fixed effect formula, “Full” denotes a full fixed effect
formula including all higher order interactions, while a “-“ denotes the removal of a fixed effect.
The model with the lowest DIC value (denoted with a “*”) is presented in the supplementary
tables (S1-S4). However, sometimes the purpose of a specific model was to demonstrate nonexistent influence of a specific variable (for example, the effect of brain selection on body size).
In this case we instead present statistics from a model that have a higher DIC value (i.e. since
selection had no effect on body size and would thus have been excluded from the analysis
under a strict simplification scheme) these models are denoted with “**”.
Supplemental Experimental Procedures and Results
Artificial Selection on Relative Brain Size
We used laboratory-descendants of wild guppies (Poecilia reticulata), whose founders (> 500
individuals) were imported from Trinidad in 1998 and since then kept in large populations (>
500 individuals at any time) where they were allowed to reproduce freely. From these, four
stock populations were formed using 100 l aquaria each stocked with 100 individuals of equal
sex ratio. The parental population (F0) was formed as follows: we scanned the population tanks
daily and removed newborn offspring. These juveniles were then reared in groups of three to
five individuals in 4 l tanks with a 2 cm layer of gravel and constant aeration, on a 12:12 l:d
lighting schedule. Temperature was held at 26-27°C and fish were fed flake food and live brine
shrimp six days per week. Java moss (Taxiphyllum sp.) provided spatial structuring and hiding
opportunities while 5-10 snails (Planorbis sp.) removed food remains. We separated males from
females at the first signs of maturation (gonopodium growth in males) and reared them
separately until maturity (when males showed fully developed color pattern). Since it is not
possible to determine brain weight in live fish to select breeder individuals, we used a selection
design in which parents were paired at random and sacrificed for brain size quantification after
offspring production. The offspring of parents with the largest and smallest brains were then
used to form breeder pairs for the next generation (see below). At 82.4 ± 0.3 days of age (mean
± SE) we used 450 of the F0 fish to set up three experimental replicate populations of 75
breeding pairs each (225 breeding pairs in total), in similar tanks as described above. Guppies
are live-bearing and we divided the breeding tanks with a net divider (3 mm mesh size) to create
a zone for new-born fish. We checked for offspring daily and moved juveniles to separate tanks
in groups of up to six individuals. After offspring production (Mean age: 126 ± 0.8 days) we
euthanized the parents with an overdose of benzocaine and measured their standard length
(from the tip of the snout to the end of the caudal peduncle) to the nearest 0.01 mm using
digital calipers and placed them in 5 % buffered formalin. Brains were removed under a
stereomicroscope and weighed to the nearest 0.001 mg. To select the offspring that would form
breeder pairs for the subsequent generation (F1) we extracted the residuals from sex-specific
regressions of (log-transformed) brain weight on (log transformed) body size, and standardized
these, within sexes, to a mean of zero and a standard deviation of unity. We then added the
standardized residuals of the male and female in each breeding pair and ranked them according
to their sum. Each replicate population and selection regime combination was handled
independently. From the 75 pairs of every replicate population we kept the offspring of the 15
highest- and lowest-ranking pairs to start the “up” and “down” selected lines respectively (see
Figure S1 for a scheme of the selection procedure). We thus attained six replicate populations of
juveniles (i.e. three replicates of up- and down-selected lines respectively), which we reared in
sibling groups consisting of up to 6 offspring. We separated the sexes at first signs of maturation
in males and kept them separated till all fish reached maturity. Within each of the six lines we
used two males and two females each of the selected 15 families to form 30 breeding pairs for
the next generation (180 pairs in total; we randomly assigned partners, but avoided full-sib
pairs; F1 age at pairing 80.3 ± 0.6 days; age at sacrifice: 136 ± 0.8 days). For the next generation
(F2) we followed the same procedure as described above, again using offspring of the top and
bottom 15 breeding pairs for the “up” and “down” selected lines respectively (F2 age at pairing
102.9 ± 0.6 days, age at sacrifice: 161.7 ± 1 days). Since brain size is a plastic morphological trait
[1], we did all comparisons across the selection regimes within generations, where the three
replicate populations experienced identical conditions.
Quantification of Neonate Body and Brain Size
To determine offspring body and brain size at birth we used one offspring each of 67 F3 clutches
from all lines. We placed the newborn fish in small Petri dishes and took dorsal pictures through
a dissecting microscope using a digital camera (QImaging, Go-3). We used ImageJ (1.43u NIH) to
determine standard length (from the tip of the snout to the end of the caudal peduncle) and
width of the optic tectum which is the largest separate brain structure and also clearly visible
through the semi-transparent skull-plate of newborn fish.
The Numerical Learning Test
To investigate cognitive ability in the different lines, we first trained fish to associate a visual
numerical cue with a food reward and then we determined the number of correct decisions
made when individuals were presented with the visual cue but without the reward. The
experimental fish were 48 mature offspring of F2 (mean age 150 ± 7 days) balanced from all
replicates from large- and small-brained lines and of both sexes. Each experimental fish was
individually kept in 15 x 40 x 15 cm Plexiglas tanks with a 2 cm layer of coarse sand and constant
aeration. To minimize isolation stress, a smaller non-mature “friend”, who was changed three
times during the experiment, accompanied each fish. We blocked visual contact between tanks
with cardboard dividers. For the association learning phase, we presented stimuli randomly on
opposite sides of the holding tanks. Stimuli consisted of white 6 x 3 cm cardboard cards with
two or four black objects. To avoid potential shape-bias, half of the objects were circles, the
other half squares. Since cumulative surface area is important for quantity discrimination in fish
[2], the cumulative surface area on both type of cards was held constant at 1 cm 2 (4 objects:
0.25 cm2 each, 2 objects: 0.5 cm2 each). The separate objects on the stimuli cards were
randomly placed in eight different positions to exclude a potential location bias. Forty-eight
stimulus card pairs were randomly chosen for every trial. Two see-through dividers were used
to confine the fish and the friend in the center, prior to each training (Figure S2). During a
training session, we placed the respective stimulus card on each side of the tank and placed
flake-food on the tank bottom on the side with four objects. To avoid potential side-biases, the
stimulus cards with different number of objects were randomly placed on either the left or right
side for each training session. Fish were kept in the center for 5 minutes, thereafter the dividers
were lifted and the fish were allowed to feed in front of the four objects for 90 minutes. All food
remains and the stimulus cards were removed after each training session. Fish were trained
twice per day on three consecutive days prior to each numerical learning ability trial. The
observer was unaware of the identity of the experimental fish both during the training phase
and during the numerical learning trial.
To test for numerical learning we performed tests every fourth day as described above and
placed the stimuli on both sides but this time without adding food. To ensure that fish were
choosing according to object number, the trial objects were all of the same size (0.375 cm 2). To
exclude the possibility of pattern preference they consisted randomly of either squares or
circles, but were not mixed within trials. The stimuli used during the test phase was slightly
different than during training to ensure stimuli number was the only aspect that could be
remembered by focal individuals between training and trial. The test procedure followed those
during the training phase. When entering the correct side the fish was rewarded with food on
the correct side. When swimming to the wrong side of the tank no food was given. When the
fish did not enter either choice area within the first 5 minutes or showed signs of severe stress
when the dividers were lifted (characterized by a “dart-and-freeze” behavior, where the fish
shoots to any position in the tank and stays there motionless), we added food and the trial was
scored as “no choice”. On the fifth day we started a new round of training as described above.
We repeated this training/trial routine eight times so that every fish was trained and trialed 48
and 8 times, respectively.
Control for Preexisting Bias for Specific Number of Objects.
Since we trained all focal individuals to associate the higher number of objects (i.e. four objects)
with food, we used 48 additional individuals, naïve to the experiment, to determine whether
untrained individuals showed a preference for either four or two objects. This was done to
assure that our results are due to the previous training, thereby excluding the possibility that
any differences found in the test may be driven by an inherent preference to feed in front of
either four or two objects. We used 4 vs. 2 objects of the intermediate size and randomly chose
either squares or circles. Following the protocol for the numerical learning assay, we confined
the focal individual to the center of the tank and placed the stimulus cards with two and four
objects on either side of the tank. However, this time we dripped flake food mixed with water
simultaneously on both sides of the tanks and noted to which side the focal individual would
swim. We tested every individual once per day on three consecutive days.
Quantification of Gut Size
To explore the predicted trade-off between brain size and gut size, we used 360 randomly
chosen fully grown and mature F3 (age: 159 ± 1.4 days) balanced for lines, replicate and sex. F3
were raised identically to previous generations. To ensure guts were empty we food-fasted
them for 24 hours [3] and then euthanized them with an overdose of benzocaine. We measured
the standard length (from the tip of the snout to the end of the caudal peduncle) to the nearest
0.01 mm using digital calipers and placed them in 5 % buffered formalin. After fixation (90.1 ±
0.1 days in formalin) we removed the gut under a stereomicroscope and weighed it to the
nearest 0.001 mg. Gut mass may be influenced by fixation time but since all samples were fixed
for the same duration before processing we are confident that this does not influence our
results. To ensure that only individuals with fully evacuated guts were included in the analysis,
we checked for food remains in the gut and excluded all individuals whose guts were not
completely empty from the analysis (♀♀: 37 small-brained, 28 large-brained; ♂♂: 3 largebrained, 3 small-brained). We note that the results were qualitatively identical also when all
individuals were included.
Statistical Analysis
The models for the variables response to selection, body size, inter-brood interval, gut mass and
fecundity were fit using a Bayesian approach implemented in the R package MCMCglmm [4, 5].
Flat priors were used for the fixed effects and locally uninformative priors were used for the
random effects, both representing little prior knowledge. Initially, the models were fit with all
possible interactions across the fixed effects. However, after evaluation of DIC values, several
parameters were removed, and the final model contained the parameters presented in the
tables (see Table S5 for model simplification procedure). After a burnin of 8 * 10 5, a sample of
the posterior distribution of 3.2 * 106 was made with a thinning interval of 800, yielding a total
posterior sample of 4000. All autocorrelations across successive posterior samples were in the
interval < 0.1 and > -0.1. In addition to the Bayesian model, we also analyzed the data using
Restricted Maximum Likelihood and found these models to yield results that were highly
congruent with the Bayesian models. Specific details of the models are described below. To
analyze the response to artificial selection on relative brain size as well as its interactions with
sex, we fit linear mixed effect models with the fixed effects selection regime (“down”, “up”), sex
(“female”, “male”) and the covariate size (standard length in mm). The models were fit for
generation F1 and F2 separately. We fit the logarithm of the response (brain weight) against the
logarithm of the co-variate (size) in order to avoid problems associated with allometry. Initially,
the model was fit with all possible interactions across the fixed effects. However, after
evaluation of DIC values, several parameters were removed, and the final model contained the
parameters presented in Table S1. Replicate line (3 levels) was added to the model as a random
effect together with the interaction variance of the replicate*sex interaction.
Divergences in body size were modeled with sex and selection regime as fixed effects
(“down” and “up”) and replicate line, the interaction of replicate line and sex, as well as sex
specific residual variances as random effects.
Fecundity (estimated by number of offspring in the first brood) was assessed under the F2
generation and analyzed using a female specific model where number of offspring was modeled
dependent on the age at first offspring, selection (“down” and “up”) as well as their interaction.
Since the interaction was non-significant (based on DIC values) it was discarded from the
analysis. Number of offspring was modeled using a Poisson distribution. Body size is known to
affect number of offspring in fishes [6]. Since body size was measured on average 36 days after
the first brood, we used age at first reproduction as a proxy for female body size at first
reproduction. Equal growth rates between the two selection regimes were assumed because all
fish were fed similar ad libitum rations and because there were no significant size differences
across the selective regimes both at birth and when sacrificed (Table S2). Age at first
reproduction should therefore present an accurate proxy of size at first reproduction. Replicate
line was added to the model as a random effect. Divergence in age at first reproduction was
modeled with selection regime as fixed effect (“down” and “up”) and replicate line as random
effect.
To analyze how the relative gut size changed over the brain selection lines, we fit a linear
mixed effect model with the fixed effects selection regime (“down”, “up”), sex (“female”,
“male”) and the covariate size (length in mm standardized to a mean of 0 and a standard
deviation of 1 within each sex in order to avoid collinearity between sex and size). Length rather
than mass was used as co-variate for these analyzes since some of the females were pregnant,
and mass would thus have been a biased measure. These analyzes were run for generation F3.
Initially, the model was fit with all possible interactions across the fixed effects. Replicate line (3
levels) was added to the model as a random effect. Since the sexes differed greatly in their
residual variance, we modeled different residual variances for the sexes. Divergence in juvenile
body size was modeled as dependent on the clutch size, maternal size, with selection regime as
fixed effect (“down” and “up”) and replicate line as a random effect. To analyze the size and
brain size of newborn fish we fit a linear mixed model with the fixed effects selection (“down”
and “up”), body size as a co-variate and replicate line as a random effect.
To analyze cognitive performance after the first training session, we first assessed the
number of correct choices on the first day of testing, using a probit-link generalized linear mixed
model (GzLMM) with correct choice as dependent variable. We included selective regime and
sex as fixed factors and replicate line as random factor. We then analyzed the number of correct
choices of all eight numerical learning ability trials. Since not all fish participated in every trial
we used binary probit-link generalized linear mixed models (GzLMM) to analyze the cognitive
ability with the total number of correct choices as dependent variable and the number of times
the fish participated as independent variable [7, 8]. We included selective regime and sex as
fixed factors and replicate line as random factor. These analyzes were done with SPSS 19.0, SPSS
Inc., Chicago.
All experiments were done in accordance with the ethical regulations for research involving
animal subjects in Uppsala, Sweden.
Calculation of Realized Heritabilities
We used line means to estimate realized heritability. We followed Walsh and Lynch [9] and
estimated realized heritabilities as the ratio between the cumulative selection response RC and
the cumulative selection differential SC, which are defined by
and
Here, z represents the means of the specified groupings (the overall mean in the “up” and
“down” selection lines at generation t, or the means of the artificially selected individuals
included in the analysis). In the following description “selection” refers to the selection regimes,
i.e. the populations that are under selection for either larger or smaller brains. “Selected” refers
to the group of guppies within each selection regime that were chosen to establish the following
generation. We estimated the realized heritabilities for males and females separately. Brain size
means exhibited plasticity across generations and we detected generation-dependent sex
differences in brain-body size allometry (table S1). It was thus problematic to extract the means
of interest from a single model, and we therefore ran separate models for each generation and
sex combination for the models estimating the response to selection (calculated as the
difference between the “up” and “down” populations). For the models estimating the means of
the selected individuals and the population mean (the mean of the guppies in the selection
regimes for either smaller or larger brains), we ran models specific to each generation, sex and
selection combination.
The models estimating the response to selection contained the explanatory variables
selection (“up” and “down”), log body size, and replicate line as a random effect. The models
estimating the means of the selected individuals contained the explanatory variables selected
(“yes”, “no”), log body size as a covariate and replicate line as random effect. To evaluate main
effects independently of the covariate, log body size was standardized to a mean of 0 and a
standard deviation of 1. All models were fit using MCMCglmm as previously described.
Results from Numerical Learning Test
Already on the first day after the training, we found a significant interaction between sex and
selection (GzLMM, n = 39, selection: 2 = 0.16, df = 1, P = 0.695, sex: 2 = 0.02, df = 1, P = 0.889,
selection*sex: 2 = 7.05, df = 1, P = 0.012). Large-brained females outperformed small-brained
females in the learning assay (GzLMMfemales, n = 23, selection: 2 = 5.36, df = 1, P = 0.031), while
no difference was found between males of different brain sizes (GzLMMmales, n = 16, selection:
2 = 2.26, df = 1, P = 0.155). The combined results for all trials are given in the main article.
Results from Control for Preexisting Bias for Specific Number of Objects
Analogously to the numerical learning ability trial, we first tested the bias on the first day and
then analyzed the data for all three days combined. All but one individual participated readily in
all trials. We therefore excluded this individual and, analogously to the numerical learning assay
analysis, used a binary probit-link generalized linear mixed model (GzLMM) to test for
preexisting bias towards two or four objects on the first day of testing. To then test all three
days, we used a general linear mixed model (GLMM). These analyzes were done with SPSS 19.0,
SPSS Inc., Chicago. On the first day of the control test we found no preexisting bias towards four
or two objects (GzLMM, n = 47, selection: 2 = 0.01, df = 1, P = 0.913, sex: 2 = 1.10, df = 1, P =
0.300, selection*sex: 2 = 0.012, df = 1, P = 0.913). This did not change after three days of
testing (GLMM, n = 47, selection: 2 = 1.76, df = 1, P = 0.192, sex: 2 = 0.01, df = 1, P = 0.911,
selection*sex: 2 = 0.25, df = 1, P = 0.623). We therefore conclude that the significant results
shown and discussed in the main article are not due to a preexisting bias for either two or four
objects, but rather due to an increased learning performance in the large-brained females.
Supplemental References
1.
Gonda, A., Välimäki, K., Herczeg, G., and Merilä, J. (2011). Brain development and
predation: plastic responses depend on evolutionary history. Biol. Lett. 23, 249-252.
2.
Agrillo, C., Dadda, M., and Bisazza, A. (2007). Quantity discrimination in female
mosquitofish. Animal Cognition 10, 63-70.
3.
Hofer, R., and Schiemer, F. (1981). Proteolytic activity in the digestive tract of several
species of fish with different feeding habits. Oecologia 48, 342-345.
4.
R Development Core Team (2006). R: A Language and Environment for Statistical
Computing (Vienna: R Foundation for Statistical Computing).
5.
Hadfield, J.D. (2010). MCMC Methods for Multi-Response Generalized Linear Mixed
Models: The MCMCglmm R Package. Journal of Statistical Software 33, 1-22.
6.
Roff, D.A. (1992). The Evolution of Life Histories (New York: Chapman & Hall).
7.
Kotrschal, A., and Taborsky, B. (2010). Environmental change enhances cognitive abilities
in fish. PLoS Biol. 8, e1000351.
8.
Norusis, M.J. (2007). SPSS 15.0 Advanced Statistical Procedures Companion (Chicago:
Prentice Hall).
9.
Walsh, B., and Lynch, M. (2013). Evolution and Selection of Quantitative Traits. In press,
http://nitro.biosci.arizona.edu/zbook/NewVolume_2/newvol2.html.