Evolutionary divergence in life history traits among populations of the Lake Malawi cichlid fish Astatotilapia calliptera

Abstract During the early stages of adaptive radiation, populations diverge in life history traits such as egg size and growth rates, in addition to eco‐morphological and behavioral characteristics. However, there are few studies of life history divergence within ongoing adaptive radiations. Here, we studied Astatotilapia calliptera, a maternal mouthbrooding cichlid fish within the Lake Malawi haplochromine radiation. This species occupies a rich diversity of habitats, including the main body of Lake Malawi, as well as peripheral rivers and shallow lakes. We used common garden experiments to test for life history divergence among populations, focussing on clutch size, duration of incubation, egg mass, offspring size, and growth rates. In a first experiment, we found significant differences among populations in average clutch size and egg mass, and larger clutches were associated with smaller eggs. In a second experiment, we found significant differences among populations in brood size, duration of incubation, juvenile length when released, and growth rates. Larger broods were associated with smaller juveniles when released and shorter incubation times. Although juvenile growth rates differed between populations, these were not strongly related to initial size on release. Overall, differences in life history characters among populations were not predicted by major habitat classifications (Lake Malawi or peripheral habitats) or population genetic divergence (microsatellite‐based F ST). We suggest that the observed patterns are consistent with local selective forces driving the observed patterns of trait divergence. The results provide strong evidence of evolutionary divergence and covariance of life history traits among populations within a radiating cichlid species, highlighting opportunities for further work to identify the processes driving the observed divergence.


| INTRODUCTION
Adaptive radiation is characterized by the rapid evolution of ecologically differentiated species that share recent common ancestry (Schluter, 2000). Although life history traits can diverge among derived species within radiations (Duponchelle, Paradis, Ribbink, & Turner, 2008), the role of local life history adaptation in restricting gene flow among populations remains far less well-understood than adaptation in eco-morphological and behavioral traits. This is surprising, given that many studies have demonstrated that intraspecific variation in life history strategies is driven by local environmental variation, including the quality of the food (Segers & Taborsky, 2011), habitat availability (Rollinson & Hutchings, 2013), and key limits to reproduction including local predation regimes . It is, therefore, important to assess the role of life history traits, and potential constraints in their evolution, during the process of adaptive radiation.
Studies of life history evolution often focus on how spatial or temporal variation in the environment can drive selection to optimize fitness in traits such as offspring size, offspring number, and growth rates. It is generally assumed that larger offspring must be fitter, and empirical data often support this view (Bashey, 2008;Hutchings, 1991;Reznick, Bryga, & Endler, 1990;Riesch, Plath, & Schlupp, 2012;Sogard, 1997). Therefore, we should expect individual females to favor larger offspring wherever possible. However, the energy required for somatic maintenance means that only a portion of the resources of any female can be allocated for reproduction. Selection should, therefore, modify the balance of offspring size to offspring number depending on the resources available, and show spatial variation among habitats.
How selection operates on such traits, however, will be determined not only by readily measurable spatial contrasts in habitat characteristics, but also by habitat predictability (Morrongiello, Bond, Crook, & Wong, 2012), and the extent of plasticity in the trait (e.g., Burgess & Marshall, 2014).
Life history traits show strong covariance, so evidence that habitat predictability and resource availability can both drive selection on life history traits (e.g., Winemiller & Rose, 1992). Winemiller (2005) suggests that these factors may slow down, or prevent, local adaptation in some habitats, but will accelerate local adaptation (and adaptive radiation) in others. Given this background, in this study we used the mouthbrooding cichlid fish Astatotilapia calliptera to take the first steps to investigate population level divergence, and covariance in life history traits, within the context of cichlid adaptive radiation. The species is useful for studying life history evolution, on account of the considerable maternal care exhibited (Konings, 2007;Ribbink, 1990), with females collecting eggs after fertilization and incubating them in their mouths (Ribbink, 1990).
Astatotilapia calliptera is part of the Lake Malawi haplochromine radiation (Malinsky et al., 2017) but, unlike the other members of the flock that are lacustrine specialists it is a generalist, occupying both the littoral margins of Lake Malawi and peripheral habitats including rivers and shallow lakes. The main body of Lake Malawi is comparatively stable, with relatively minor changes in water level between seasons and over decadal timescales (Scholz et al., 2011). By contrast, peripheral water bodies are prone to both flooding in the wet season and drought or even complete habitat desiccation (e.g., Nicholson, 1998; for Lake Chilwa) in the dry season (Kingdon, Bootsma, Mwita, Mwichande, & Hecky, 1999;Pauw, Thurlow, & Van Seventer, 2010).
This strong seasonal variability in water availability leads to associated changes in habitat productivity, thermal regime, and oxygen availability. The species also represents a useful model when considering evolutionary processes during early-stage adaptive diversification.
The species has seeded a sympatric species pair within a crater lake (Malinsky et al., 2015) and has also been proposed to have taken a role in generating the main species radiation in Lake Malawi (Malinsky et al., 2017). Importantly the species exhibits population variation in male color and eco-morphological traits. These differences are associated with assortative mating suggestive of incipient speciation in both allopatry (Nichols et al., 2015;Tyers & Turner, 2013), and sympatry (along a depth cline, Malinsky et al., 2015).
In this study, we used common garden laboratory-based experiments to test whether populations of A. calliptera differ in clutch size, egg mass, brooding duration, and the speed of early growth. We also tested if the observed variation differs predictably between Lake Malawi and peripheral water bodies, and how evolutionary divergence is associated with trade-offs among life history traits. We expected that occupants of lacustrine sites should possess traits that promote intraspecific competitiveness, namely small broods of larger offspring. In contrast we expected that populations from riverine sites should possess traits that maximize productivity, namely larger broods of smaller offspring. We also assessed the role of selection relative to genetic drift by exploring the relationship between population-level phenotypic divergence of life history traits (Q ST ) and population-level genetic divergence (F ST ), estimated using allelic variation at microsatellite loci. We predicted that if strong local adaptation in life history traits was taking place, Q ST would be independent of genetic distance between sampling sites.

| Collection and animal husbandry
Fish were collected from four Lake Malawi habitats (Chisumulu island, Mbenji island, Mpatsonjoka dambo, and Makanjila) and one peripheral habitat (Linthipe river) in January 2011 ( Figure 1, Table 1). Fish derived from a further three peripheral habitats (Enukweni, Ruvuma river, and Lake Chilwa) were taken from third-generation stocks housed in the University of Hull ( Figure 1). All fish were kept at a 12 hr: 12 hr lightdark regime and at water temperatures of between 25-28°C. The adults were fed once per day with King British tropical flake and juveniles with Interpet Liquifry No3 once a day. All tanks were equipped with UV and biological filters, aeration, synthetic aquarium foliage, and drainpipes of varying diameters that served as shelters.

| Environmental variables
Several variables were sourced for each sampling location, including water flow (flowing/not flowing), the proximity to water deeper than 20 m (close < 1 km, distant > 1 km) and altitude. Additionally, we derived interpolated monthly temperature and rainfall data from the CRU TS4.0 dataset at a resolution of 0.5° (Harris, Jones, Osborn, & Lister, 2014) for the period January 2001 to December 2015. We then used the "hydrostats" package in R (R Core Team, 2015; https://github. com/nickbond/hydrostats) to estimate environmental predictability (tightness of event to season), constancy (uniformity of event through all seasons), and contingency (repeatability of seasonal patterns), following Colwell (1974), using 10 bins of equal sizes in each calculation. To ordinate environmental similarity of sampling sites, we used a Principal Component Analysis based on a correlation matrix, in PAST 3.15 (Hammer, Harper, & Ryan, 2001).

| Experimental populations
To ensure individual phenotypes were associated with parental genotypes, stock populations were bred to generate eight G1 (1st generation) populations that maximized available genetic diversity. From these G1 populations, experimental parental fish were drawn. Two separate experiments were performed, the first was aimed at testing differences in clutch size and egg mass among populations. The second was aimed at testing for differences among populations in brood size, brooding duration (incubation time), fry length when released, and juvenile growth rates.

| Experiment 1
Response variables for the first experiment were clutch size (number of eggs), average individual egg size (g), and total egg investment (g).
Between 15 and 21 G1 females from each of the eight populations were each mated with a single male from the same population (Table 1). Multiple females from the same population were housed in compartments with one male. Tanks were checked at least once daily for females that had spawned. Eggs were stripped from mouthbrooding females by gently pressing on their cheeks and opening and closing their mouth repeatedly. Once all eggs had been removed, each female was weighed using a Mettler Toledo PB602S balance, and the total length (TL) of the female was measured. The number of eggs in the clutch was counted and the eggs dried in an oven for 12 hr at 50°C. Eggs were then weighed on a Mettler Toledo AB54-S balance.

| Experiment 2
The second experiment used a hierarchical half-sib design to quantify variation among populations and families (males). Response variables were incubation time (number of days from fertilization to release), brood size (number of fry released), and fry total length (at release, at day 35, at day 70). We generated second-generation family clutches of offspring from four males, each mated with four different females, for each of the eight populations. In total we generated 128 families, and 1,595 individual offspring. This was achieved by housing females in a compartment with one dominant male. Tanks were checked at least once daily for any females that had spawned. Brooding females, easily identifiable by their pronounced gular, were removed from their tank and placed in 16.5 cm × 12.7 cm × 12.7 cm fry nets. We then checked daily to see if the female had released free-swimming fry.
On release of fry, the mother was removed from the fry net, TL measured, and then weighed using a Mettler Toledo PB602S balance.
We then calculated incubation time by counting the number of days from egg laying to fry release. The released fry were then placed in a water-filled Petri dish, counted, and photographed with a size standard. They were then returned to their fry net. All broods were limited to 32 individuals, which were chosen at random. These fry were again photographed 35-day postrelease, and 70-day postrelease. Fry total length was measured using ImageJ 1.46 (Schneider, Rasband, & Eliceiri, 2012). Average brood growth rates were calculated as the difference in mean total length of fry in the brood between time points.

| Experimental data analysis
Response variables were analyzed using General Linear Models (GLMs) in R, with Tukey's HSD post hoc comparisons using an adjusted p-value for multiple comparisons. Least-square means of focal response variables, correcting for statistically significant covariables, were calculated using the package "lsmeans" (Lenth & Hervé, 2014). In Experiment 1, we focused on clutch size, average individual egg mass and total egg investment, using female postspawning TL as a covariate. We noted that female postspawning TL had a strong linear relationship with postspawning mass (F 1,136 = 500.935; r 2 = .788; p < .001). In Experiment 2, we focussed on brood size, incubation time, average size of fry on release, using female postbrooding TL as covariate. Female postbrooding TL had a strong linear relationship with postbrooding mass (F 1,127 = 1324.575; r 2 = .913; p < .001). We also considered growth between release and day 35, growth between release and day 70, in these cases using fry rearing densities for the relevant time periods T A B L E 1 Source population sample sites, coordinates (decimal degrees), habitat, and experimental sample sizes

| Genetic data analysis
Linkage disequilibrium among loci was quantified within populations using GENEPOP 4.2 (Raymond & Rousset, 1995), employing the log-likelihood ratio statistic, 1,000 dememorizations, 100 batches, and 1,000 iterations per batch. Significant linkage disequilibrium was tested across locus pairs using Fisher's method, but no evidence was found. Observed and expected heterozygosity was calculated in Arlequin 3.5 (Excoffier, Laval, & Schneider, 2005

| Environmental variables
Lake Malawi habitats were characterized by a close proximity to deep water, non-flowing waters, relatively low attitude, warm temperatures, high precipitation, low predictability of temperatures, and high predictability of rainfall. By contrast, the peripheral habitats were characterized by absence of deep water, flowing waters, high altitudes, cold temperatures, high predictability, and contingency in temperature, but low predictability of rainfall (Figure 1c).

| Experiment 1
Clutch size (# eggs) increased with female total length (TL; Figure 2a) and differed significantly among populations ( Figure 3a; Table 3). Post hoc comparisons showed that Chisumulu clutches contained significantly fewer eggs than Chilwa, Enukweni, Linthipe, and Mbenji after correcting for female length. They also showed Mbenji clutches were larger than Makanjila and Ruvuma (Table 4). Average egg mass increased with female TL (Figure 2b) and differed significantly among populations ( Figure 3a; Table 3). Post hoc comparisons showed that Chisumulu and Ruvuma eggs were larger than all other populations (Table 4). Total egg investment increased with female TL (Figure 2c) but did not significantly differ among populations ( Figure 3c; Table 3).
There was a significant negative correlation between clutch size (# eggs) and average egg mass, after correcting for female TL the species. We found no significant differences between Lake Malawi and peripheral habitat populations in any variables (Table 3).

| Experiment 2
Brood size increased with female TL (Figure 4a) and differed among populations ( Figure 5a; Table 5). However, no pairwise comparisons of populations showed significant differences in brood size ( Figure 5a; Table 4). Incubation time was not dependent on female TL, and differed significantly among populations (Figure 5b; Table 5).
Post hoc Tukey's HSD tests revealed that Chisumulu, Ruvuma, and Chilwa all had significantly longer brooding times than the Linthipe and Makanjila populations (Table 4) Table 5). Pairwise comparisons revealed that Chisumulu fry had grown more at day 35 than those from Chilwa, Makanjila, Mbenji, Mpatsonjoka, and Ruvuma (Table 4). Additionally, Ruvuma populations had grown less than populations from Enukweni and Linthipe (Table 4). Fry growth to day 70 was negatively related to the densities of individuals in compartments (Figure 4f), and differed among populations ( Figure 5e). Pairwise comparisons revealed that Chisumulu fry had grown more to day 70 than those from Lake Chilwa and the Ruvuma river (Table 4).  (Table 5).

| Genetic differences among populations, and associations between F ST and Q ST
Four of the 49 tests of deviation from Hardy-Weinberg Equilibrium were significant, but these were not consistent across populations or loci (Table 2)  Our source localities were classified broadly into Lake Malawi and peripheral habitats, on the basis of broad differences in environmental conditions. East African riverine environments are prone to strong seasonal and interannual fluctuations in water flow rate, and complete desiccation can take place (Dettinger & Diaz, 2000). By contrast, ancient lakes appear more predictable in depth and in temperature, nutrient regime, and oxygen content. The r/K model specifically addresses the role of environmental predictability in determining the evolution of life history strategy (Pianka, 1970;Winemiller & Rose, 1992 important determinants of lifetime reproductive success are survival to reproductive age and rapid reproduction, rather than competitive ability. Traits such as increased fecundity, large dispersal distance, fast growth rate, and early sexual maturity should, therefore, be favored in unpredictable habitats (Pianka, 1970). By contrast, in predictable environments, competitive ability and immunity/predator defences should be maximized because these environments are saturated by competitors, or parasites, predators, and pathogens (Parker & Begon, 1986). This means that traits such as maternal nutrient provision, gestation time, and parental care duration should be favored in predictable environments (Pianka, 1970). In our analyzes, however, we found no consistent differences in clutch size, egg size, or parental care duration between main lake and peripheral habitat populations, suggesting that spatial contrasts in life history traits are

| Trade-offs between offspring size and offspring number
Our results demonstrate a clear and consistent trade-off between individual egg mass and clutch/brood size in A. calliptera. This is perhaps one of the best understood trait correlations, and it has been strongly linked to resource allocation (Messina & Fox, 2001;Smith & Fretwell, 1974). The trade-off occurs because nutrients available to an organism are finite, and females must optimize allocation of these resources, resulting in a negative correlation between investment in individual offspring and the total number of offspring produced (Smith & Fretwell, 1974). In poor quality environments, females should favor a small clutch of highly provisioned offspring (Goulden, Henry, & Berrigan, 1987). By contrast, in high quality environments, females should produce larger clutches of less provisioned offspring   because the greater resource availability within those habitats will allow offspring to compensate for initially small size through faster growth. It is possible that further work will identify the key limiting resources in the natural environment of the focal species, allowing tests to determine if environmental quality is a predictor of the provisioning strategy.
Alternately, or additionally, population differences offspring size and number may be a consequence of differences in age-dependent among source locations. Investment in a small number of fry in habitats with high juvenile mortality may be risky, more so if a large proportion of mortality is unpredictable. Thus, the results demonstrating a diversity of investment strategies between egg number vs. egg size could be generated by differences in the temporal patterns of survivorship among the source localities. A closer understanding of the environmental variables associated with specific traits will require more detailed information on temporal patterns of survivorship across age classes within populations, and seasonal changes in resource availability.
In addition to life history trade-offs, other factors could contribute to between population-level differences in egg and clutch size, and may constrain adaptive divergence in response to local conditions. In mouthbrooding species such as A. calliptera, females store eggs in their mouths after fertilization, meaning that buccal volume imposes limits on clutch volume (Okuda, Tayasu, & Yanagisawa, 1998;Sefc, 2011). Buccal volume is in turn limited by gill size related to re- There is evidence that A. calliptera shows substantial differences in trophic resource use and head shape across its geographic range (P. Parsons, unpublished data). Taken together, it is possible that habitat characteristics such as oxygen concentration and food resource availability may also have driven the observed differences among populations.

| Offspring size and growth
Our studies demonstrate a strong association between growth rates and rearing densities. This may be due to reduced food being available per individual, as food was not provided in overabundance.
Alternatively it could be due to growth suppression induced by other aspects of the experimental conditions (Wedemeyer, 1997), including behavioral interactions among individuals. Such density dependent effects on growth may take place in natural conditions, but at present growth rates of wild fish are unclear. Insight into relative growth rates of fry in the natural environment could be gained from daily growth ring increments on otoliths (e.g., McLeod et al., 2015).
After controlling for density-related effects on growth, we found that the length of fry on release was correlated with the size of fry after 35 and 70 days. However, we found no significant effect of the initial body size on net growth by days 35 and 70. This is suggestive of all populations exhibiting equivalent growth after the initial head start determined by maternal investment. Thus, our results do not support the model of Smith and Fretwell (1974), where the constraints of the trade-off between offspring investment and number can be mitigated by smaller offspring having an increased growth rate (compensatory growth), enabling them to rapidly match the body size of more highly invested offspring. However, it is possible that initial maternal investment will affect multiple other offspring traits throughout their lifetime (Altmann & Alberts, 2004;Crean, Monro, & Marshall, 2011;Rius, Turon, Dias, & Marshall, 2010), and will not necessarily lessen in importance with age (Donelson, Munday, & McCormick, 2009;Heath, Fox, & Heath, 1999;Segers, Berishvili, & Taborsky, 2012).

| Incubation time and fry length
Although we found significant differences in brooding duration among populations and that brooding duration tended to be longer in smaller broods, we found no significant relationship between incubation length and fry length. This was unexpected, given evidence that the Lake Tanganyika haplochromine cichlid Ctenochromis horei extends incubation period by approximately 4 days in the presence of predators and that this yields fry that are approximately 15% longer on release (Taborsky & Foerster, 2004). Incubation period may be related to aspects of fry morphology that were not assessed in this study. For example, we only measured fry length and it is possible that extra incubation yields higher body width and/or body mass (Gillooly, Charnov, West, Savage, & Brown, 2002), or other benefits such as increased immunity or brain development. Increased gestation associated with more extensive brain development has also been reported in mammals (Barton & Capellini, 2011;Sacher & Staffeldt, 1974). Also, since personality has been linked to several life history traits (Biro & Stamps, 2008;Niemela, Dingemanse, Alioravainen, Vainikka, & Kortet, 2013;Schuett et al., 2014;Wolf, van Doorn, Leimar, & Weissing, 2007), brood time may be related to particular personality traits that vary among A. calliptera populations. It is also not clear why incubation time should be negatively related to brood size. However, as smaller broods were persistently correlated with larger eggs and fry on release, these may require a greater duration of mouthbrooding.

| Covariance among life history traits
Our results demonstrating trade-offs and covariance between life history characteristics suggest that traits are not necessarily optimized by selection independently of each other, and that they can be intrinsically correlated in complex ways. In addition to correlations resulting from fundamental ecological trade-offs driven by resource allocation, other factors such as genomic associations between traits that are driven by pleiotropy or linkage disequilibria may be important (Roff, 2007;Stearns, 1989). In the context of evolutionary divergence and adaptive radiation, such trait covariances could also prevent populations from reaching the adaptive peaks predicted by life history theory and local ecological conditions, and may therefore slow adaptive radiation more generally. It may be the case that persistent correlations between key traits across populations have constrained adaptive diversification (and specialization) between lacustrine or peripheral environments.

| Rapid adaptive evolution in haplochromine cichlids
Exceptional rates of speciation and adaptive radiation of cichlids are often associated with strong selection on traits linked to sexual selection, habitat use or feeding ecology (Kocher, 2004;Malinsky et al., 2015;Wagner, Harmon, & Seehausen, 2012). The enhanced rates of adaptive radiation seen in lakes relative to surrounding rivers may be a consequence of both a complex adaptive landscape in lakes and that these landscapes persist for long enough (and populations are large enough) to permit adaptation in both ecological and sexually selected traits (Bridle & Jiggins, 2000;Seehausen, 2015). However, much less attention has been paid to selection on life history traits in cichlids, and how adaptive divergence in these traits is related to patterns of genetic population structure, structure, and stability of the immediate environment, as well as covariances among traits. Our results are consistent with the concept that genetic population structure is associated with divergence in life history traits, and in principle divergence of these populations may be promoted by selection acting against migrant genotypes linked to poorly adapted life history phenotypes.
The importance of life history evolution in rapid cichlid adaptive radiation is supported by comparative work on Lake Tanganyika and Lake Malawi cichlids, which demonstrates significant associations between individual egg mass and habitat use (Duponchelle et al., 2008), where pelagic species had larger eggs and lower fecundity than benthic and rock dwelling species. By contrast, our study provides no strong evidence for evolutionary divergence in life history traits among conspecific populations of cichlids linked to habitat, but it does provide evidence of persistent correlations among traits that in principle may limit the ability of populations of this focal species to reach adaptive peaks. Such constraints on adaptive diversification may partly explain why, uniquely among the Lake Malawi haplochromine species, A. calliptera retains a broad ecological niche, and is found in both riverine and lake habitats, despite their strongly contrasting ecologies. To more comprehensively understand the influence of environmental variability on life history trait evolution, we need more information on which traits covary, the underlying reasons for that covariance, and the extent to which such covariances promote or restrict rapid adaptive divergence in novel environments.