Seasonal environments drive convergent evolution of a faster pace-of-life in tropical butterflies

Global change can trigger shifts in habitat stability and shape the evolution of organismal life-history strategies, with unstable habitats typically favouring a faster pace-of-life. We test this hypothesis in species-rich Mycalesina butterflies that have undergone parallel radiations in Africa, Asia, and Madagascar. First, our ancestral state reconstruction of habitat preference, using ~85% of extant species, revealed that early forest-linked lineages began to invade seasonal savannahs during the Late Miocene-Pliocene. Second, rearing replicate pairs of forest and savannah species from the African and Malagasy radiation in a common garden experiment, and utilising published data from the Asian radiation, demonstrated that savannah species consistently develop faster, have smaller bodies, higher fecundity with an earlier investment in reproduction, and reduced longevity, compared to forest species across all three radiations. We argue that time-constraints for reproduction favoured the evolution of a faster pace-of-life in savannah species that facilitated their persistence in seasonal habitats.


INTRODUCTION
Habitat can act as a templet for the evolution of life-history strategies (Southwood 1977;Southwood 1988). For example, habitats that are heterogeneous in time and/or space and are thus more fluctuating in resource abundance are considered to favour the evolution of 'fast' life-history strategies including high growth rates and fecundity with reduced longevity, that allow organisms to rapidly increase population size when opportunities for reproduction are available (Pianka 1970;Southwood 1977). In contrast, selection in more stable habitats is thought to favour 'slow' (low growth rates and fecundity with increased longevity) life-history strategies. These fast and slow life-history strategies parallel the dichotomy of r and K-selected reproductive strategies (Pianka 1970;Gadgil & Solbrig 1972).
The r/K selection theory was formulated to predict the type of life-history strategies that would be favoured in colonizing populations with high capacities to increase population size, versus those found in populations that experience density-dependent mortality in stable environments (Pianka 1970;Gadgil & Solbrig 1972). The concept of r/K strategies has been criticised for providing a too simplistic overview (see Stearns 1976;Stearns 1992). For example, in many studies, a coarse description of habitat use would suffice as an explanation for the evolution of organismal life-history strategies. In contrast, age-specific demographic models (Gadgil & Bossert 1970; Charlesworth 1980) expand beyond these simple correlations and can therefore be used to help identify causal mechanisms driving the evolution of life-history strategies (Reznick et al. 1996; Reznick et al. 2002). Nevertheless, comparative analyses using species in close phylogenetic proximity with similar ecological guilds and detailed information on the temporal and spatial characteristics of their habitats can still contribute to understanding the evolution of life-history strategies (Southwood 1988;Partridge & Harvey 1988).
Evolutionary trade-offs are fundamental to optimising investment in life-history traits and can constrain the types of strategies that can be achieved (Stearns 1989; Roff & Fairbairn 2007). One such ubiquitous trade-off is between offspring size and number (Smith & Fretwell 1974). Since internal resources are finite, organisms can either produce fewer offspring of higher quality or many offspring with a relatively low probability of surviving to adulthood (Smith & Fretwell 1974; Van Noordwijk & de Jong 1986). Another well-established trade-off is between voltinism and individual growth rates (Roff 1980;Abrams et al. 1996). Insects inhabiting seasonal environments typically only breed during a narrow part of the year when resources for both juveniles and adults are abundant (Tauber et al. 1986). The ability to undergo an additional generation in a single breeding season (i.e. multi-voltinism) can dramatically accelerate population growth (Roff 1980;Kivelä et al. 2009). To achieve this, there is generally a need to accelerate individual growth rates, which can be achieved by shortening the time In this study we broaden the ancestral habitat reconstruction, with near-complete taxa sampling across all geographically independent radiations, to describe patterns of habitat respectively. We argue that the evolution of faster pace-of-life in savannah species is driven by strong time-constraints for breeding in open habitats, and that such life-history strategies contributed to population persistence in seasonal environments.

Ancestral state reconstruction of habitat preference
The evolutionary history of Mycalesina butterflies was reconstructed by classifying the habitat preference for 287 species covering over 85% of all extant taxa and representing all three parallel radiations. Using the available literature, communications with local experts, and our own extensive field experience, species presence was scored in three categories; forest habitats (A), forest-fringes (B), and open or savannah habitats (C). Forest species are restricted to rainforests or habitats with extensive canopy cover. Forest-fringe species are those found in the outskirts of closed-canopy forests or fragmented forests, but never extending into savannahs. Finally, savannah specialists were quantified as species mainly occurring in open grasslands or woodlands where vegetation is dominated by grasses with few interspersed trees.
Many species were assigned to two habitat classes (i.e. habitat class A+B or B+C; N=98), and a small number of generalist species were assigned to all three habitats (habitat class A+B+C; N=32). Allowing species to occupy multiple habitats, rather than being fixed to discrete categories, provides a continuum with a decreasing degree of habitat stability; A > A+B > B > B+C > C.
A recent phylogeny from Brattström et al. (2020) was used to reconstruct the ancestral states. The evolution of habitat preference was modelled using fitpolyMk function in phytools ver. 0.7.18 (Revell 2012) which can handle polymorphic states. Since this function can only handle a maximum of two polymorphic states, species that can be found in three habitats (i.e. A+B+C) were dropped, leaving a total of 255 species. We fitted four ordered and four unordered models to the data (i.e. an equal-rates model, a symmetric model, an all-ratesdifferent model, and a transient model, for each model type). The unordered models allow all possible transitions between habitat types, while the ordered models only allow transitions along the habitat continuum. Model performances were assessed using the Akaike Information Criteria (AIC) score and an ancestral state reconstruction was conducted for the best fitting model using Bayesian stochastic mapping (Bollback 2006) implemented in phytools using make.simmap function. We simulated 1000 character states on the Maximum Clade Credibility tree and summarised the states at each node to indicate the probability of a particular state to be ancestral.

Measuring life-history traits
Eggs were collected from each laboratory population at 24-hour intervals (between 9.30 to 10.30 hours), from small Oplismenus plants and then transferred to petri dishes lined with a filter paper and kept in climate-controlled chambers (Sanyo/Panasonic MLR-350H) at 25 0 C and 70% RH under a 12:12 hour L:D photoperiod. The eggs of each species were photographed using a Leica DFC495 camera coupled to a Leica M125 stereoscope, and the cross-sectional area of each egg measured from the resulting images in ImageJ (Schneider et  After eclosion, two to four day old virgin females (except two females which were one day old) were allowed to mate with virgin males to determine daily variation in fecundity and to compute individual fecundity curves. For B. anynana, H. iboina, and H. fraterna, we were able to select 15 mating pairs in the first trial. For B. martius, we could not obtain any mating pairs in the first trial, and we only observed a single mating after multiple trials. Therefore, males and females of this species were kept in a single cage for approximately 15 days, after which females were randomly selected for measuring fecundity. After copulation, individual females were kept in cylindrical plastic pots (11.5 x 13.5 cm) and provided with a cutting of Oplismenus grass kept in water for oviposition. The number of eggs laid by the females was assessed every 24-hours for 15 consecutive days. The females of B. martius were dissected after the monitoring period to confirm their mating status. Three non-mated individuals were excluded from the data set yielding a total sample size of four for this species. Female longevity was measured as the percentage of females that were alive after the 15-day fecundity assessment period. For B. martius we included both non-mated and mated females (N=7). All females were fed on slices of moist banana throughout the experiment.

Statistical analyses
Generalised linear models (GLM) were used to examine the effects of habitat class, genus, sex, and their interactions, on a suite of life-history traits. Habitat class was a categorical variable (forest or savannah), and genus (Bicyclus or Heteropsis) was included to account for phylogenetic relatedness between the species. Sex was excluded from the models for egg size and egg development time since this variable cannot be established at this life stage. Data on development times (egg, larval, pupal and total) were analysed using GLMs with a Poisson distribution and a log link function. GLMs with a Gaussian distribution were used to fit the data for egg size, pupal weight and growth rate. Post hoc pairwise comparisons (Tukey's HSD; α = 0.05) were carried out using emmeans R package (Lenth 2019).
Individual fecundity curves were estimated using a generalised linear mixed model with a negative binomial distribution, as implemented in the R package glmmTMB (Magnusson et al. 2019). In addition to habitat class and genus, the number of days (centred) and all interactive terms were included as fixed effects. Pupal weight, centred within each species, was added as a covariate. Finally, since the raw data suggested that the fecundity curves had non-linear distribution, we included a quadratic term for the number of days in the model. We initially fitted full models, allowing interactions between all predictors, and the minimum adequate model was found by AIC guided backward elimination. All statistical analyses were performed in R ver. 3.6.1 "Action of the Toes" (R core team 2019).

Ancestral state reconstruction for habitat preference
An ordered all-rates-different model provided the best fit for the habitat preference data (see Table S1 in Supporting Information). This model suggested that the lineages that gave rise to three geographically independent radiations were forest specialists and that more open habitats were colonised repeatedly during the Late Miocene and Pliocene (8-3 Mya; Figure 1a).
The transition matrix of the best-fit-model suggests that forest fringe habitats represent a stepping-stone towards adaptation to strictly open environments ( Figure 1b). In other words, the invasion of woodland and savannah habitats were preceded by adaptation to semi-shaded habitats. Moreover, parameter estimates suggest that back-transitions, that is transitions from open to forest, are unlikely ( Figure 1b).

Egg size and development
We found that forest species laid larger eggs and that their offspring took longer to complete embryogenesis (habitat, P<0.001 for both egg size and egg development time; Figure   2). The effect of habitat-use was larger in the species representing the African radiation than in those representing the Malagasy radiation (habitat:genus, P<0.05 for both traits; see Table S2 & S4). The results from our study are in line with the habitat-dependent patterns observed for the Australasian species from the Asian radiation .
Here, compared to savannah species (Mycalesis perseus), the two forest specialists (Mydosoma sirius and Mydosama terminus) laid larger eggs that took longer to develop (data from Braby & Jones (1994) are presented in Figure 2; see Supporting Information for details). Note that all three species were placed in the genus Mycalesis at the time of the original publication.

Development times and growth rates
Compared to savannah species, forest species had longer larval, pupal and total development times (habitat, P<0.001 for all traits; Figure 3a; figures for larval and pupal development time in Figure S1a & S1b). Differences in developmental times between habitat specialists were larger in the species representing the African radiation (habitat:genus, P<0.002 for all traits; see Table S5-S7). Forest species had lower individual growth rates (habitat, P<0.001) and a larger body mass (habitat, P<0.001) than savannah species (Figure 3b & 3c; Table S8 & S9). All species were sexually dimorphic for body size (all pairwise comparisons, P<0.001) with female pupae weighing on average 22% more than males, except for B. martius where males were slightly heavier than females. These results complement the data on the Australian Mycalesina butterflies from the Asian radiation where two forest species had longer total development times, lower growth rates and larger body sizes than the savannah specialist (data from  are presented in Figure S3).

Fecundity curves and longevity
In both the African and Malagasy radiations, the forest species had lower fecundity and laid eggs more uniformly across their lifespan than the savannah species (habitat, P<0.001; Figure 4). The magnitude of these differences appeared to be slightly higher in the African than in the Malagasy radiation. For example, the African savannah species (B. anynana) had higher fecundity and a more pronounced early investment than the savannah species from the Malagasy radiation (H. fraterna) (Figure 4). In contrast, the fecundity curves of both forest species were closely similar. After 15 days of fecundity measurements, 55% females of forest species and 6% of savannah species survived in the African radiation ( Figure 4). Similarly, 46% and 20% of forest and savannah species survived in the Malagasy radiation, respectively ( Figure 4). Note that the females from the forest species from Africa, B. martius, were kept in communal cages for 15 days prior to fecundity assessments (see Methods). The high proportion of females still alive after the assessment suggests that this species is potentially extremely long-lived. Similarly, in the Australian species, the savannah species had higher fecundity and a prominent early peak in fecundity compared to two forest species (data from  are presented in Figure S4). Systematic data on longevity were not available for these species.

DISCUSSION
Adaptive radiations comprise the rapid differentiation of a single common ancestor into an array of species that inhabit a variety of environments and differ in the phenotypic traits

2011
). In our study, the differences in egg size and fecundity among habitat specialists remained significant after accounting for variation in body size (Table S3 & Figure 4). As a potential caveat we note that the fecundity of the African forest species B. martius was measured 15 days post-eclosion, and the females may have laid some eggs before the fecundity assessment. However, since this species is extremely long-lived (see Results and Oostra et al. 2014a) we are confident that our fecundity measurement only represents a slight underestimate.
Apart from trade-offs, including those resulting from time-constraints, hypotheses related to adaptive foraging could potentially also explain the observed differences in egg size across species. The size of the eggs is strongly correlated to the size of the larval head capsules, which has been shown to affect the foraging ability of early instar larvae (Braby 1994     In the inset, highlighted portions of pie charts show the percentage of surviving females after 15 days of fecundity assessment. Fecundity curves for species from the Asian radiation are presented in Figure S4.

Seasonal environments drive convergent evolution of a faster pace-of-life in tropical butterflies
Sridhar Halali, Erik van Bergen, Casper J Breuker, Paul M Brakefield, Oskar Brattström Table S1: AIC scores of the four ordered and four unordered models fitted using fitpolyMk function in phytools. The best fitting model is highlighted in bold and was used for reconstructing the ancestral states presented in Figure 1 of the main text.

Extracting data for species from the Asian radiation from published studies
We extracted data for three Australian species representing the Asian radiation from published studies . When possible, we extracted values from the plots showing mean and error bars as standard error using the WebPlotDigitiser Ver. 4.2.
For some traits such as the total development time, information on confidence interval was not available. Sexes are denoted with different shapes (circles = males, squares = females).