THE BEHAVIOURAL ECOLOGY OF TROPHIC -LAYING

Jennifer C. Perry B.Sc. (Hon.) Environmental Biology, University of Alberta, 2001

Thesis submitted in partial fulfillment of the requirements for the degree of

MASTER OF SCIENCE

In the Department of Biological Sciences

O Jennifer C. Perry 2004

SIMON FRASER UNIVERSITY

July 2004

All rights reserved. This work may not be reproduced in whole or in part, by photocopy or other means, without permission of the author.

Partial Copyright Licence

The author, whose copyright is declared on the title page of this work, has granted to Simon Fraser University the right to lend this thesis, project or extended essay to users of the Simon Fraser University Library, and to make partial or single copies only for such users or in response to a request from the library of any other university, or other educational institution, on its own behalf or for one of its users.

The author has further agreed that permission for multiple copying of this work for scholarly purposes may be granted by either the author or the Dean of Graduate Studies.

It is understood that copying or publication of this work for financial gain shall not be allowed without the author's written permission.

The original Partial Copyright Licence attesting to these terms, and signed by this author, may be found in the original bound copy of this work, retained in the Simon Fraser University Archive.

Bennett Library Simon Fraser University Burnaby, BC, Canada ABSTRACT

Across diverse taxa, produce infertile trophic that are consumed by offspring. The objective of this thesis research was to establish a behavioural ecology framework for the study of trophic eggs; to model their ; and to investigate their adaptive function, if any, in the ladybird beetle Harmonia axyridis. Current hypotheses suggest two adaptive functions: provisioning offspring or reducing parent-offspring conflict over sibling . On the other hand, trophic eggs may simply represent infertility - they may not be an at all, a distinction given insufficient attention in the literature. I consider trophic egg laying in the context of sibling cannibalism behaviour. A frequency-independent analysis of sibling cannibalism suggests that there is a critical value of increase in offspring survival above which cannibalism benefits both parents and offspring; then mothers should adopt tactics (e.g., trophic egg laying) to facilitate cannibalism. However, the frequency-independent approach does not incorporate game interactions. I used a genetic algorithm approach to model the co-evolution of trophic eggs, sibling cannibalism, and hatching synchrony. Results suggest that, when game interactions occur or when infertile eggs are present, cannibalistic tendencies increase dramatically, as does the maternal response to limit cannibalism. Furthermore, trophic egg laying and hatch synchrony appear to be viable tactics for facilitating egg-eating. We tested trophic egg function in H. axyridis, predicting that mothers should lay fewer trophic eggs in food-rich environments and more in low food environments where offspring gain a larger relative benefit from eating an egg. As predicted, ladybirds produced 64% more trophic eggs when provided with information that they were in a low vs. high food environment (P = 0.0093). This result demonstrates that ladybirds produce trophic eggs to increase the chance that offspring survive starvation. In a second observation set, I tested whether the spatial distribution and oviposition sequence pattern of trophic eggs is over- or under-dispersed compared to random. There was no indication of a non-random distribution. In conclusion, this research is of interest in behavioural ecology because it shows that an unusual parenting strategy, killing off some offspring to benefit others, can be adaptive. Acknowledgements

First and foremost, I thank the members of my supervising committee - Dr. Bernie

Roitberg, Dr. Bernie Crespi and Dr. Felix Breden - for their carefully considered advice as I prepared for and conducted this thesis research. 1 particularly appreciate the guidance and mentorship of my senior supervisor, Bernie Roitberg, throughout this project. I am grateful to Dr. Carl Schwarz, who provided extremely useful statistical help, and Jabus Tyerman, who wrote the randomisation test program described in Chapter 4. The manuscripts were improved by thoughtful and appreciated comments from Geoff Allen, Jay Biernaskie, Felix Breden, Kelly Campbell, Bernie Crespi, Brian Ma, Jason Peterson, Bernie Roitberg, Maxence Salomon, and Jabus Tyerman. I am obliged to laboratory assistants Andy Capadouca, Nick Charrette, Beth Ann Nyboer, Jennifer O'Neil, Eva Poon, and Andy Yang, for their assistance. Members of the Behavioural Ecology Research Group, the Entomological Society of British Columbia, and the DOGG Journal Club provided helpful advice as I planned and conducted experiments. The research was supported by the National Science and Engineering Research Council (NSERC) in graduate support and grants to Bernie Roitberg. I also acknowledge graduate support from the Department. Finally, I must express my appreciation for the Roitberg lab group. Frequent discussions with Roitbergers improved all aspects of this work, and lab interactions made this research process tremendously enjoyable. Table of Contents .. Approval ...... 11 Abstract ...... nl... Acknowledgements ...... iv Table of Contents ...... v List of Figures ...... vu .. List of Tables ...... ix Chapter 1 Overview of the Thesis ...... 1 Overview ...... 1 Literature Cited ...... 3 Chapter 2 The Behavioural Ecology of Trophic Egg Laying: Hypotheses and Evidence ...... 4 Abstract ...... 4 Introduction ...... 4 Parent and offspring interests: inclusive fitness approach ...... 5 Two sets of trophic egg hypotheses ...... 7 The parent-offspring conflict reduction hypothesis ...... 8 Cooperation hypothesis ...... 9 What evidence is required to distinguish hypotheses? ...... 10 Evidence from natural history and behavioural observations ...... 11 Evidence from behavioural and life history experiments ...... 11 Comparative analyses...... 13 Challenges for future research ...... 14 Conclusions ...... 15 Literature cited ...... 17 Chapter 3 Co-Evolution of Maternal and Offspring Effectors of Sibling Cannibalism ...... 31 Abstract ...... 31 Introduction ...... 31 Model ...... 33 Biological system ...... 33 Inclusive fitness argument ...... 34 Genetic algorithm-based simulations: the basic model ...... 35 Computer experiments ...... 36 Results and discussion ...... 38 Relative benefit of egg eating ...... 38 Trait co-evolution ...... 40 Non-evolved infertility ...... 41 Time extension ...... 42 Conclusions ...... 43 Trophic .egg laying ...... 45 Future directions ...... 46 3.5 Appendix ...... 47 Details of genetic algorithm and simulations ...... 47 3.6 Literature cited ...... 49 Chapter 4 Ladybird Mothers Mitigate Offspring Starvation Risk by Laying Trophic Eggs ...... 71 4.1 Abstract ...... 71 4.2 Introduction ...... 71 4.3 Methods I: Food level experiments ...... 74 Study system and experimental animals ...... 74 Food level experiments ...... 74 Statistical analyses ...... 76 4.4 Results I: Food level experiments ...... 77 Internal cue experiment ...... 77 External cue experiment ...... 79 4.5 Methods 11: Non-random patterns of infertile egg production ...... 80 Spatial patterns ...... 80 Oviposition patterns ...... 80 4.6 Results 11: Non random patterns of infertile egg production ...... 81 Spatial patterns ...... 81 Oviposition patterns ...... 81 4.7 Discussion ...... 81 4.8 Literature cited ...... 85 Chapter 5 Summary of the Thesis and Suggestions for Future Work ...... 101 5.1 Introduction ...... 101 5.2 Thesis summary ...... 101 5.3 Suggestions for future work ...... 102 Approaches to answering question (1) ...... 103 A suggested approach for question (2) ...... 105 A suggested approach for question (3) ...... 105 5.4 Conclusions ...... 105 5.5 Literature cited ...... 107 APPENDIX I ...... 108 I .1 Frequency of sibling cannibalism in Harmonia axyridis ...... 108 Methods ...... 108 Results and summary ...... 109 1.2 Does egg viability or hatching time depend on egg mass? ...... 109 Methods ...... 109 Results and summary ...... 110 1.3 Trophic egg production in the laboratory and field ...... 111 Methods ...... 111 Results and summary ...... 112 1.4 Male-killing bacteria and the proportion of trophic eggs laid ...... 113 Methods ...... 113 Results and summary ...... 114 1.5 Literature cited ...... 115 List of Figures

Figure 2-1 The inclusive fitness interests in sibling cannibalism of a parent and her cannibalistic offspring; r is the coefficient of relatedness between siblings. Parent-offspring conflict or cooperation depends on the survival benefits of cannibalism, relative to survival of non-cannibalistic offspring. In the range from 0-r%, neither parents nor offspring benefit from sibling cannibalism...... 22 Figure 3-1 The average number of offspring cannibalized per clutch with various combinations of cannibalism delay and hatching synchrony values. Cannibalism delay alleles correspond to the number of time units offspring wait before eating an egg. Phenotypes for hatching synchrony alleles are shown in Table 3.1. The shaded area indicates combinations of traits that yield no sibling cannibalism. The number of cannibalized offspring increases diagonally to a maximum of eight (in a clutch of 16), as indicated by the arrow...... 5 1 Figure 3-2 The evolution of cannibalism delay, hatching asynchrony and trophic egg laying, when other were prevented from evolving (Table 3.2). Least square means are from a one-way ANOVA with the factor 'relative benefit of egg eating', which was a significant factor for all three genes; P < 0.0001 in each test. Means with different letters are statistically different. The bars represent standard error. The y-axis is reduced for the trophic egg laying because average allele value never exceeded 8...... 53 Figure 3-3 Evolution of the cannibalism delay gene over 1000 generations, for different levels of relative benefit of egg-eating, when the evolution of other traits was prevented (Table 3.2). Results from one representative simulation are shown for each level of relative benefit. Mean gene value was taken for the parental population every tenth generation. With the random generation of individuals in generation zero, initial average gene values were approximately 7.5 ...... 55 Figure 3-4 Effect of infertility on the evolution of three traits. For each figure, least squares mean gene values are given for different levels of relative benefit of egg eating, given by the different lines, in the absence and presence of infertile eggs. (a) Cannibalism delay. The mean difference in gene value in the presence and absence of infertile eggs was 6.7 (4.8, 8.5) at 10% relative benefit and 3.8 (1.9, 5.6) at 25% relative benefit. The difference was not significant for any other level of benefits. For clarity, levels above 75% are not shown. (b) Hatching synchrony. The presencelabsence of infertile eggs affected the mean difference in gene values at 10% benefits (2.0 (0.9, 3.2)). The difference was not significant at any other level of relative benefit. For clarity, not all of the tested levels are shown. (c) Trophic egg-laying. Infertile eggs significantly decreased trophic egg production at the following levels of relative benefit, with the mean difference in gene value given: 0.9 (0.5, 1.3) at 75%; 0.8 (0.4, 1.2) at 90%; 1.1 (0.7, 1.5) at 100%; 0.8 (0.4, 1.2) at 110%; 0.9 (0.5, 1.3) at 125%; 0.9 (0.5, 1.3) at 150%. The y-axis range is reduced to show

vii differences between the curves. For clarity, not all levels of relative benefit are shown...... 57 Figure 3-5 Trait evolution when other traits are constrained. Levels of relative benefit of egg eating are given by the different lines. (a) The evolution of cannibalism delay in offspring when maternal traits affecting cannibalism are free to evolve or constrained (set 1, Table 3.2). Least squares mean gene values are obtained from an ANOVA with the factors relative benefit, presence of constraints, and the interaction term. Constraining traits lead to delayed cannibalism; see discussion in text. Significant increases in delay times occurred at relative benefit levels of 25% (mean difference 6.6 time units (4.5, 8.7)) and 50% (2.2 time units (0.1,4.3)). The differences were not significant at other levels of relative benefit. (b) The effect of constraints on cannibalism delay and trophic egg laying on the evolution of hatching synchrony (set 2, Table 3.2). The difference in gene value when other traits were constrained, or free to evolve, was significant at 0% relative benefit (mean difference 5.2 (1.6, 8.8)) and 150% relative benefit (mean difference 3.8 (0.2, 7.4)). (c) The effect of constrained traits (set 3, Table 3.2) on the evolution of trophic egg laying. The mean difference between the trophic egg gene when traits were constrained, or free to evolve, was 0.7 (0.1, 1.4) at 75% relative benefit; 1.6 (0.9,2.2) at 125%; and 3.2 (2.6, 3.9) at 150%. Note that the y-axis scale for this trait differs from that in graphs of the other genes...... 61 Figure 4-1 External cue experimental chamber. Ladybirds were placed in the upper compartment with filter paper as an oviposition surface, a water source, and a small mass of pea aphids. The bottom compartment contained a bean leaf pair, inserted through Parafilm@ into a vial of water, and the treatment: either no or many pea aphids...... 89 Figure 4-2 Proportion of trophic eggs produced when females were given information that there was low or high food availability through (a) the internal cue of food consumption or (b) the external cue of prey scent; see text for details. The mean proportion was smaller under the low food treatment in the internal cue but not the external cue experiment (Table 4.2). Horizontal lines represent means. Whiskers indicate the range of observations...... 9 1 Figure 1-1 Proportion of egg mass lost during the development of H. axyridis egg types. Trophic eggs lose significantly less mass than viable eggs (estimated difference and 95% CI: 0.025 (0.009,0.042)). Inviable eggs, where an embryo develops but does not emerge, lost an intermediate amount of mass. The initial mass of eggs was taken 12 hours afier oviposition; final mass was taken in the 24 hours before hatching...... 116 Figure 1-2 Hatching time in relation to egg mass within egg batch 60-1 5. Hatching time is expressed relative to the first hatching egg in the batch, which has a hatch time of zero. The initial mass was taken within 12 hours following oviposition. Heavier eggs tended to hatch sooner than lighter eggs, though clearly the trend has exceptions...... 1 18

... Vlll List of Tables

Table 2-1 Taxa that produce trophic eggs, grouped by the behavioural evidence that trophic egg laying is adaptive ...... 24 Table 2-2 Hypotheses regarding the function of trophic eggs and predictions of ecological factors or traits that should be correlated with trophic egg laying. Hypotheses and predictions are grouped by whether they consider trophic eggs a tactic for reducing sibling cannibalism or provisioning offspring...... 27 Table 2-3 Evidence required to distinguish among hypotheses of trophic egg laying: whether trophic eggs are produced by non-evolved infertility or by maternal adaptation (constraint vs. adaptive hypotheses); and, if adaptive, whether trophic eggs function to reduce parent-offspring conflict over sibling cannibalism or to provision offspring (conflict reduction vs. provisioning hypotheses). Lines of evidence are grouped by how they are obtained: through observation (natural history), behavioural studies, or comparative analyses. Evidence may be 'consistent' (if it is not predicted by and does disprove a hypothesis), 'necessary but not sufficient' (if the evidence is not found, the hypothesis cannot be true, but finding the evidence does not mean the hypothesis is correct), or strongly or weakly supportive or falsifying. If investigators cannot obtain falsifying or strongly supportive evidence, the next best option is to collect several lines of weakly falsifying or supporting evidence. The framework is developed for the scenario of one trophic egg- eating or cannibalising event, but the qualitative predictions are general to the case where more than one egg or sibling is eaten...... 29 Table 3-1 The range in hatching synchrony phenotypes, in terms of number of offspring hatching in each time period for each allele, that were tested in simulations. Mothers with each non-binary allele value produce clutches with the temporal hatching pattern indicated in the table...... 65 Table 3-2 Conditions for each set of simulations of the evolution of sibling cannibalism and maternal traits that influence cannibalism: the range of relative benefits that cannibal offspring receive; whether infertile eggs are present or absent; whether cannibalism is time-limited; and the presence of constraints, if any, on the evolution of any of the tested traits ...... 67 Table 3-3 The effect of the relative benefit of egg eating, infertile eggs, and a time extension (whereby offspring could wait for viable siblings to hatch before eating eggs) on the evolution of the three genes ...... 69 Table 4-1 Comparison of protocols for the internal and external cue experiments ...... 93 Table 4-2 Effect of food level on proportion of trophic eggs and three measures of hatching synchrony (see text) when females were given information about food conditions through feeding (internal cue) or other senses (external cue). Sample sizes were: in the internal cue experiment, 7 1 for the proportion of trophic eggs and 55 for hatching synchrony; for the external cue experiment, 38 for the proportion of trophic eggs and 17 for hatching synchrony...... 95 Table 4-3 Randomization tests for the spatial distribution of trophic eggs within egg batches produced by different females. P values were generated from simulations of possible distributions, given the spatial arrangement of each batch; values less than 0.025 indicate a clumped distribution while values greater than 0.975 indicate a uniform distribution...... 97 Table 4-4 Runs tests (Zar 1999) of the temporal distribution of infertile eggs within oviposition sequences of different females ...... 99 Table I- 1 The evidence, across studies, that female ladybirds adjust trophic egg production depending on male-killing infection status. The expected proportion hatch is a weighted mean proportion of hatching eggs from uninfected lines, based on data reported in each study. The observed proportion is also a weighted mean. The null hypothesis is that the observed hatching proportion does not differ from half of the hatching proportion in uninfected lines. If females can control the production of trophic eggs, infected females may benefit from reducing trophic egg production because hatching larvae will have dead male eggs to eat. When the lower limit of the 95% CI of the observed proportion exceeds the expected proportion, the evidence is taken to support the alternate hypothesis...... 120 Table 1-2 Evidence that females control trophic egg production from studies in which male-killing (MK) infected ladybirds were cured with antibiotic. The expected proportions for MK-infected lines before antibiotic treatment are calculated as half of the proportion of hatching eggs after antibiotic treatment. If the observed proportion is greater than the expected proportion, females appear to reduce trophic egg production when infected. For lines that were uninfected before treatment, the expected hatching proportion is the proportion observed following antibiotic treatment. If the observed proportion is greater than the expected, the antibiotic causes some reduction in trophic egg production regardless of MK infection status...... 122 Chapter 1 Overview of the Thesis

1.1 Overview

In some species, mothers produce non-developing trophic eggs that are consumed by other offspring (Crespi 1992). This trophic egg-laying behaviour - noted across diverse taxa - has rarely been studied in an evolutionary ecology context. In any given species, trophic eggs may function to reduce parent-offspring conflict over sibling cannibalism or to reduce offspring starvation risk by providing an easy meal for neonates. Trophic egg-laying, then, is one of several parental tactics that could accomplish these functions, such as parental manipulation of clutch size, egg size, or hatching synchrony.

The goals of this thesis were four-fold: To organize hypotheses about the ecology and evolution of trophic egg-laying into a behavioural ecology framework; To model the co-evolution of maternal and offspring behaviours related to trophic egg- laying; To investigate the adaptive function of trophic eggs in the ladybird Harmonia axyridis; To test for the use of a functionally equivalent tactic, hatching synchrony, in the same species.

This chapter summarizes the contents of the thesis. It does not provide a general introduction to trophic eggs; the literature review in Chapter 2 serves that function. Chapters 2, 3 and 4 were written as independent manuscripts. Chapter 5 suggests questions that should be addressed in future work. Appendix I presents additional results obtained during this thesis research.

In Chapter 2, I organize the hypotheses for the function of trophic egg into a framework for their study. I identify two categories of hypotheses, grouped by the suggested adaptive function of trophic eggs: to provision offspring or to reduce parent-offspring conflict over sibling cannibalism. Another research objective should be to determine whether trophic eggs have an adaptive function at all or simply represent infertility. I review the evidence necessary to distinguish among the hypotheses, concluding that much evidence hitherto collected lacks power to distinguish hypotheses, and suggest approaches for future work.

Chapter 3 investigates predicted patterns of co-evolution for three maternal and offspring traits: trophic egg production, hatching synchrony, and sibling cannibalism. I use stochastic simulations with a genetic algorithm component. Genetic algorithm-based models are a useful approach for studying complex, multi-trait systems with game theoretic aspects (Forrest 1993). I compare the results from the co-evolving parent-offspring game to the predictions of a frequency- independent inclusive fitness analysis based on Eickwort (1973). I also analyze the effect of relative versus absolute starvation risk, the presence of infertile eggs, and time limitations on cannibalism behaviour.

The objective of the laboratory experiments with H. anyridis, presented in Chapter 4, was to determine whether trophic eggs are adaptively produced. I tested for phenotypic plasticity in trophic egg production, with the prediction that females would increase the proportion of trophic eggs per clutch in low food conditions where offspring starvation risk was greater. Because females might benefit from a uniform distribution of trophic eggs within a clutch (if it is then more likely that offspring consume trophic eggs instead of viable siblings), I assessed whether the distribution of trophic eggs within clutches and oviposition sequences differed from random. I also tested whether females manipulated hatching synchrony to adjust the opportunity for cannibalism among offspring. Results demonstrated that the infertile eggs of H. axyridis are adaptively produced to increase the chance that offspring survive starvation.

The final chapter suggests questions that should be addressed in future trophic egg research. 1.2 Literature Cited

Crespi, B. J. 1992. Cannibalism and trophic eggs in subsocial and eusocial . Pages 176-213 in M. A. Elgar, and B. J. Crespi, eds. Cannibalism: ecology and evolution among diverse taxa. Oxford University Press, Oxford.

Eickwort, K. R. 1973. Cannibalism and in Labidomera clivicollis (Coleoptera: Chrysomelidae). American Naturalist 107:452-453.

Forrest, S. 1993. Genetic algorithms: principles of applied to computation. Science 261 :872-878. Chapter 2 The Behavioural Ecology of Trophic Egg Laying: Hypotheses and Evidence

2.1 Abstract

Trophic eggs are non-hatching eggs that are consumed by offspring. They are produced in diverse taxa with a wide variety of systems. Mothers may lay trophic eggs to reduce cannibalism among offspring or to provision offspring. In this review, we draw attention to two sets of trophic egg hypotheses: whether trophic eggs represent unavoidable infertility or a maternal adaptation; and, if an adaptation, whether they are a tactic for the reduction of sibling cannibalism or starvation risk. We review an inclusive fitness model to illustrate that the hypotheses are mutually exclusive in any given system, though each may explain trophic egg laying in some animals. We outline the types of evidence useful in distinguishing the hypotheses and highlight challenges for future research on trophic egg laying behaviour. Because trophic eggs are related to parental care, parent-offspring conflict, inclusive fitness, and sibling rivalry, they have the potential to become a useful research topic in behavioural ecology with further study.

2.2 Introduction

In considering sibling interactions across taxa, an interesting pattern becomes apparent: not infrequently, individuals make meals of their siblings - thereby eliminating some of their closest relatives. In a specialized behaviour related to such sibling consumption, mothers provide offspring with trophic eggs to eat (Table 2.1). Crespi (1992) provided the first adaptive synthesis of trophic egg laying behaviour, defining trophic eggs as a maternal adaptation of "ovarian- derived structures or fluid, homologous to fertile eggs, that cannot develop into viable offspring and are normally eaten" (p. 176). Thus, trophic eggs are not simply infertile or malformed eggs produced under some constraint - for example, sperm limitation or gamete incompatibility - even if those eggs are consumed by offspring and even if egg-eating offspring derive great benefit. Furthermore, the definition illustrates a central question for trophic egg research: it is necessary to demonstrate the maternal control of trophic egg production, in order to distinguish infertile and trophic eggs. A wide variety of invertebrate and taxa produce trophic eggs (Table 2.1). They occur in animals with diverse parental care systems, ranging from no care to (wherein mothers are devoured by offspring), and may occur with or without additional cannibalism of viable siblings. Table 2.1 creates two categories of trophic egg-laying animals. The first comprises animals with limited parental care in which trophic eggs are morphologically indistinct from viable eggs. In these systems, the crucial question is whether a maternal adaptation exists, or whether offspring simply cannibalize infertile eggs without help from their mother. In the second group, trophic eggs are morphologically different from other eggs, seemingly specialized to offspring needs, or they occur along with extensive parental care behaviours. Here it seems apparent that trophic eggs are adaptively produced, because mothers appear to make a special effort in their production or distribution to offspring (but see discussion under 'Evidence'). Researchers can therefore proceed directly to determining the adaptive function of trophic eggs. This review explores functional hypotheses for the production of trophic eggs in non- eusocial taxa. Eusocial insects are excluded despite the extensive literature from these animals (reviewed in Crespi 1992; see also Gobin et al. 1998; Masuko 2003). For example, a search for the keyword 'trophic eggs' in biological studies from 1969-2004 returned 54 articles, 45 of these in eusocial insects. They are not considered here because there are unique selective forces driving the evolution of trophic eggs in eusocial taxa (see Crespi 1992), making their function and evolution a separate topic. Our discussion begins with an inclusive fitness model of mother and offspring interests in egg cannibalism. We use the model to categorize hypotheses of the function of trophic egg laying. Next, we comment on the evidence required to distinguish the adaptive vs. constraint and conflict vs. cooperation hypotheses, and discuss noteworthy examples of current empirical approaches. The review concludes with challenges for future studies of trophic eggs and offspring. We refer to non-evolved infertile eggs produced due to constraint as infertile eggs. However, in many instances we refer to consumed eggs as trophic eggs for cases in which their adaptive nature has not been demonstrated, acknowledging here that until the relevant evidence is collected their 'trophic' status is unknown. As a shorthand, we refer to offspring feeding on trophic eggs as cannibalism.

2.3 Parent and offspring interests: inclusive fitness approach

Trophic egg production is necessarily tied to the offspring behaviour of eating eggs. A useful first step, then, is to consider when egg-eating - equivalently, sibling cannibalism - is adaptive from mother and offspring perspectives. Eickwort (1 973) outlined the inclusive fitness conditions under which offspring should eat siblings. Summarizing the argument (see also Chapter 3), allow p to be the probability of survival of an offspring that does not consume an egg. An egg-eating offspring receives an increase in survival, p', through reduced starvation risk, and has a total chance of survival (p + p'). By re-stating Hamilton's Rule (Hamilton 1964) for the spread of a selfish act (Mock and Parker 1997), sibling cannibalism is favoured when the relative benefit to the cannibal exceeds the cost to the victim, weighted by relatedness - that is, when p' > pr, which can be restated as (P'JP) > r (1)

For an individual to benefit from eating a full sibling (r = OS), that individual must increase its probability of survival by 50% or greater. Note the constraint that (p + p') 51; combining the inequalities and solving for p yields the condition that, for offspring to benefit from sibling cannibalism, p < l/(r+l) - else cannibalism cannot increase survival by more than (r)%. Note also that it is the relative benefit of egg eating that matters, not the absolute increase in survival. The argument is easily extended to find a similar condition for parental benefit from sibling cannibalism. Intuitively, parents benefit from sibling cannibalism when they gain more offspring from the increased survival of cannibalism than they lose as victims (see O'Connor's (1978) model of brood reduction). With pairwise sibling interactions (i.e.,one egg or sibling per cannibal), this occurs when P' > P (2) that is, when the cannibal's total survival (p + p') is at least doubled compared to a non- cannibal offspring. The relatedness term does not occur in (2) because mothers are equally related to cannibal and victim offspring. Combining (1) and (2) reveals predicted conditions of agreement and conflict between parents and offspring over sibling cannibalism (Fig. 2.1). At a relative benefit of egg-eating of less than intra-clutch relatedness, neither offspring nor parents derive benefit from cannibalism. A relative benefit between r and one represents conflicting interests, and parents should try to limit cannibalism. When the relative benefit exceeds one, parents should cooperate with offspring and adopt behaviours that facilitate cannibalism. With the pairwise nature of the modeled interaction, the implication is that when sibling cannibalism is favoured, mothers prefer that half their offspring consume the others! This inclusive fitness approach can be compared with two related sets of models. The first considers the evolution of clutch size in invertebrates in which offspring engage in sibling cannibalism (e.g., Godfray 1987; Parker and Mock 1987; Mock and Parker 1997). These models use different assumptions than the inclusive fitness approach (i.e., they assume declining per capita offspring fitness with increasing clutch size, that cannibals benefit from reduced competition rather than direct nutritional gain, and that offspring do not kill as many siblings as opportunity permits). Furthermore, they consider a different maternal oviposition decision - number of offspring per clutch - and it is not yet clear how trophic egg laying and clutch size decisions interact. A second set of related models is the brood reduction models developed since O'Connor (1978; see Godfray and Harper 1990; Forbes 1993; Mock and Parker 1997). They identify several important factors for maternal brood reduction decisions: how resources are divided after siblicidal brood reduction (whether siblicidal offspring benefit from reduced competition) and whether victims are random or pre-selected 'runts' (which affects how easily sibling cannibalism genes can invade). While factors like these can change the predicted optimal sibling cannibalism behaviour, the above analysis is valid as a basic model and its assumptions (of pairwise sibling interactions; that the benefit of siblicide accrues only to the cannibal; random selection of victims; flat clutch size-per capita fitness curve) may be adjusted to develop the approach further.

2.4 Two sets of trophic egg hypotheses

We draw attention to two questions that must be addressed in any adaptive investigation of trophic egg laying behaviour: is a maternal adaptation present and if so, what is the adaptive function? We consider an adaptation to be a phenotype modified by selection to serve some function; are detectable as the phenotype that yields the highest fitness from a set of actual and possible alternatives in a specified environment (Reeve and Sherman 1993), and this often illuminates the function of that phenotype (Thornhill 1990). The first distinction is one of 'adaptation versus constraint', with the constraint hypothesis suggesting that offspring consume infertile or damaged eggs, produced because egg development and oviposition are imperfect processes. The offspring behaviour of eating these structures may be an adaptation - and indeed, that is the assumption - but in refemng to the infertile eggs as trophic eggs, we imply maternal control of their production. It is therefore necessary to eliminate the hypothesis of non-adaptation (alternatively phrased, the hypothesis of offspring adaptation alone). We return to this point in the section 'Evidence'. If trophic egg laying is an adaptation, we can organize hypotheses about the function of trophic eggs using the above dichotomy of parent-offspring cooperation and conflict. We elucidate these hypotheses in the following section, then discuss predictions derived from them (summarized in Table 2.2). We then consider evidence that may be useful in distinguishing the predictions. A point to stress is that the cooperation and conflict reduction are not competing hypotheses. While they are mutually exclusive for any given (because the relative benefits of egg eating are either greater than or less than the benefit threshold for parents, Fig. 2. l), and we might be interested in distinguishing them for a given animal, both are likely true for some taxa.

The parent-offspring conflict reduction hypothesis

Crespi (1992) posited an important addition to Alexander's (1974) hypothesis that trophic eggs are an adaptive maternal behaviour: in some systems, parents may use some offspring as food for others, not as a provisioning tactic per se but to reduce the cannibalism of viable siblings when it is against a mother's interests (Fig. 2.1). When mothers benefit from preventing cannibalism among offspring, a possible tactic would be to provide offspring with trophic eggs that provide sufficient energy to potential cannibals such that they avoid eating siblings (Crespi 1992). A reduction in cannibalism might occur even without an offspring preference for trophic over viable eggs, if the presence of trophic eggs makes it less likely that cannibals will happen upon, and consume, viable siblings. A necessary condition for the hypothesis is that trophic eggs are less costly to produce than viable eggs (Crespi 1992). If trophic eggs are equally or more costly to produce than viable eggs, producing trophic eggs is equivalent to a mother eliminating a viable offspring instead of a cannibal sibling, since she could have used the resources to create a viable offspring. Moreover, trophic eggs may have to be substantially less costly to make their production worthwhile for all but very large clutch sizes. And if trophic eggs are less costly - and thus, presumably, contain fewer complex and proteins - offspring are likely to prefer viable to trophic eggs, since viable eggs would provide greater survival benefit (and, presumably, RBE exceeds the benefit threshold for offspring, if mothers are producing trophic eggs). It may be difficult for mothers to strike a balance between making a cost-effective trophic structure and making a nutritious enough structure that offspring relinquish eating their siblings. Modelling is needed in this area.

Crespi (1992) made two predictions for systems in which trophic eggs function to reduce conflict. Though not major points within Crespi's review of trophic eggs, they represent the only evolutionary discussion of this hypothesis in the literature and, thus, are noteworthy here. (1) Trophic eggs will occur more frequently in monogamous species with high within- clutch relatedness, that is, when offspring benefit less from eating siblings. It may then be easier for mothers to 'persuade' offspring to avoid eating viable siblings. The opposite pattern might also occur - trophic egg laying being more common in polygamous than monogamous species - because with lower relatedness between siblings, the range of benefits from egg eating for which parents and offspring disagree is greater (Fig. 2.1). Parent-offspring conflict over sibling cannibalism should, then, be more extensive, and conflict reduction may be a greater selective pressure on mothers than in monogamous species. (2) If trophic egg production is at an optimum for both mothers and offspring, there will be no cannibalism of viable siblings. Co-evolutionary models of conflict reduction via trophic egg- laying are required to predict whether a dual optimum would be evolutionarily stable.

Cooperation hypothesis

Alexander (1974) suggested that in some taxa, parents use some offspring as a food source for others. With the conceptual model of Figure 2. I, mothers benefit from sibling cannibalism when the relative benefit of egg-eating is greater than one. A solution is to neglect to fertilize some eggs, effectively making them into meals for developed offspring. Under the cooperation hypothesis, trophic eggs may be as costly as viable eggs, or potentially more costly if they are specialized to the needs of cannibal offspring. We also refer to the cooperation hypothesis as the 'maternal provisioning hypothesis'; although trophic eggs provision offspring under the conflict reduction hypothesis as well, in that case provisioning is the proximate, not the ultimate, selective pressure. Several extensions of the cooperation hypothesis have been made (Table 2.2).

Trophic eggs and ecological variables

Two hypotheses relate trophic eggs to features of the environment. The first is the widely cited icebox effect (a concept from Alexander, 1974; dubbed the icebox hypothesis by Polis 198 I), which predicts that trophic egg laying should occur frequently in taxa that experience high food availability during offspring production, but low food conditions during offspring provisioning or foraging (Table 2.2). Then trophic offspring are a stable food storage solution for the core brood. A second pair of hypotheses links the production of trophic eggs to poor food environments. When the mean starvation risk experienced by offspring is high, but parents have limited information about future resources, mothers should produce trophic eggs as a bet-hedging strategy (Osawa 1992; Mock and Parker 1997; Dixon 2000). If the production of trophic eggs is facultative, then we can also expect to see trophic eggs used in predictable environments: parents should adjust the proportion of trophic eggs to match the resource conditions offspring will experience (Chapter 4; Frechette and Coderre 2000; Osawa 2003).

Trophic eggs and species characteristics

Three hypotheses comment on how trophic eggs should be related to other characteristics of species (Table 2.2). 1) Constraints on other tactics. Because trophic egg production may be called an extreme provisioning strategy, it may be more likely to evolve when other possible provisioning tactics are constrained (Alexander 1974; Polis 1981 ; Polis 1984; Crespi 1992). For example, parents might provision offspring by creating larger eggs (Polis 198 1; Polis 1984; Crespi 1992; Mock and Parker 1997). Such a strategy would reduce the loss of energy associated with digestion of trophic eggs (Dixon 2000). However, studies in a variety of taxa suggest that egg size is often fixed (e.g.,

Dixon and Guo 1993; Honek 1993; Christians 2002) - constrained, perhaps, by ovariole size, which is constrained by body size and the tradeoff between ovariole size and number. There would then be an advantage to trophic egg laying if it is adjustable between environments. 2) Protein limitation. Because siblings are proteinaceous food items (Polis 1984), trophic eggs may provide a greater relative benefit if the diet is protein limited, as in herbivorous animals and the ladybird and lacewing predators of aphids (e.g., Cohen and Brummett 1997; Specty et al. 1999). While carnivorous animals are also protein-limited when prey are not accessible, this hypothesis states that species with an especially protein limited diet are more likely to evolve trophic egg provisioning tactics. 3) Differential costs of foraging. Mothers may provide trophic eggs when it is easier (Nakahira 1994; Kim and Roland 2000) or less risky (McKaye 1986) for mothers to forage than it is for offspring. Mothers might forage before offspring require food (and store resources in their bodies to produce trophic eggs later) or during offspring feeding (and produce trophic eggs if, for example, it is difficult for offspring to consume cumbersome prey items, e.g. Masuko 2003). However, it is likely that parents generally have lower foraging costs than neonate offspring, so this hypothesis may not have much power to predict which species trophic eggs will occur in.

2.5 What evidence is required to distinguish hypotheses?

In Table 2.3, we summarize the evidence that may be useful in distinguishing (I) trophic eggs from non-evolved infertility (the adaptive vs. constraint hypotheses), and (2) conflict vs. cooperation hypotheses. We now turn to justifying our conclusions on the value of the evidence. Evidence from natural history and behavioural observations

For some animals, the behavioural or life history context in which trophic eggs occur suggests that they are produced by adaptation: it indicates a level of complexity that is not expected by chance alone (Williams 1966). For example, in some species trophic eggs are clearly different from other eggs in shape or size, implying that they are specialized to offspring needs (Table 2.1, 2.3). A maternal adaptation is also implied when mothers repeatedly bring trophic eggs to offspring; for example, in some tropical frogs, tadpoles live in small closed water bodies and mothers return multiple times to lay trophic eggs, apparently the only food source for offspring (Table 2.1). If a vast number of infertile eggs are provided (e.g., tens of thousands in sand , Duellman and Trueb 1986; Dopazo and Alberch 1994), adaptation is also implied because a clearly complex maternal phenotype exists. However, mothers may deliver infertile eggs to offspring; then the adaptation is the act of bringing the infertile material to offspring rather thanproducing a trophic egg. This may occur if, for example, the level of non-evolved infertility is very high; e.g., in the coronated treefrog Anotheca spinosa, 90% of apparently fertilized eggs do not develop (Jungfer 1996; Table 2.1). It also seems reasonable to surmise that trophic eggs are an adaptation when they occur with other extreme offspring provisioning behaviours (e.g., matriphagy in some , Table 2.1); however, it is still possible that the adaptation is offspring consumption of infertile eggs and not maternal production of trophic eggs. Behavioural observations may also contribute to distinguishing the conflict reduction vs. cooperation adaptive hypotheses. If trophic eggs occur with other complex provisioning behaviours (such as extended parental care, repeated visits to offspring, or matriphagy), or if other cannibalism-influencing tactics are constrained, there is weak support for the provisioning hypothesis. If cannibalism of viable siblings is frequently noted, there may be weak evidence against the conflict-reduction hypothesis (Crespi 1992; but see comments under discussion of that hypothesis above).

Evidence from behavioural and life history experiments

Four lines of evidence may be collected from trophic egg behavioural studies (Table 2.3): the increased survival offspring experience from eating eggs; whether the proportion of trophic eggs changes with starvation risk; maternal investment into trophic eggs (relative to viable eggs); and the spatial distribution (random or non-random) of trophic eggs within clutches. We separate the evidence that offspring benefit from eating eggs at all from quantitative estimates of that benefit to call attention to the need for quantification in order to draw inferences (Table 2.3). If benefits are quantified, one or the other adaptive hypotheses (of parent-offspring conflict or cooperation) may be eliminated. However, falsifying one adaptive hypothesis does not imply that the other adaptive hypothesis is unequivocally supported, because the observed level of benefits might occur by coincidence. That is, a measurement of positive benefits does not show that mothers use any tactic to facilitate egg eating or cannibalism (a fact not always recognized in empirical studies; eg,Osawa 1992; Dixon 2000; Kudo and Nakahira 2001; Osawa 2002; Osawa 2003). Obtaining relevant quantitative estimates of the relative benefit of egg eating may be challenging: the conditions under which trophic egg laying evolved may differ from average conditions today, and offspring starvation risk is likely to vary temporally and spatially. Still, if, for example, relative benefits of egg-eating are less than one in a large number of studies under a wide range of conditions, it seems reasonable to conclude that reducing sibling cannibalism was a selective factor in the evolution of trophic eggs in the study animal. Furthermore, if inferences are to be drawn about the adaptive nature of consuming potentially viable siblings from an offspring's perspective, relatedness within clutches must be estimated (to use equation (1)). A demonstration that mothers adjust the proportion of trophic eggs to match offspring starvation risk would be strong evidence for the cooperation hypothesis (Chapter 4; Frechette and Coderre 2000; Osawa 2003). However, sibling cannibalism might be more likely in low food environments; then mothers might increase the proportion of trophic eggs as a conflict reduction strategy. This concern arises only if offspring assess food levels before eating siblings (e.g., Nakahira 1994; Evans et al. 1995); in many animals offspring have no information about resources before sibling cannibalism (e.g., in many insects, Dickinson 1992; Desbuquois et al. 2000; Frechette and Coderre 2000). A second experimental difficulty is to separate the effects of

information and physiology - to give females information that they their offspring will face a poor food environment without worsening their physiological state (Chapter 4). Some studies have found no difference in the proportion of infertile eggs produced in natural habitats that differed in offspring starvation risk (Dickinson 1992; Baur and Baur 1998) or in different laboratory conditions (Frechette and Coderre 2000). However, in Chapter 4 I found that the ladybird beetle Harmonia axyridis increased the proportion of infertile eggs produced in high starvation risk conditions; I attempted to control for the confound between and information. Quantifying the relative cost of trophic eggs, a measurement that has not been conducted in any study of trophic eggs, may help in distinguishing the hypotheses. An infertile egg produced by constraint may fail to develop because it is unfertilized (and, then, should have equal cost as a viable egg if sperm is not limiting) or because it receives too little material. If there are few infertile eggs, their reduced production cost is not sufficient evidence for an adaptive function. However, accidental under-provisioning is probably infrequent, so if there are many of these eggs, the constraint hypothesis is rejected and there is weak support for the conflict reduction hypothesis. If trophic and viable eggs are equally costly to produce, conflict reduction cannot be driving their production; they may be produced by constraint or to provision offspring. Finally, the result that trophic eggs are more metabolically expensive to produce than viable eggs -that they are specialized to offspring needs - can only mean that the cooperation hypothesis drives their production in the system of study. An experimental challenge, however, is to determine whether trophic eggs are, in total, more or less expensive to produce. For example, even if evolved to reduce conflict over cannibalism, trophic eggs may secondarily become specialized to offspring nutritional needs and may contain more proteins or lipids than other eggs. They might still have lower total cost if they are smaller. It may be advantageous to mothers to distribute trophic eggs uniformly within clutches, if this maximizes the probability that cannibal offspring consume trophic rather than viable eggs (Chapter 4). Thus, an over-dispersed spatial distribution of trophic eggs would indicate adaptive production, without distinguishing between the adaptive hypotheses. A random spatial distribution would not indicate non-adaptive production - it might simply mean that the distribution of trophic eggs is not important for provisioning offspring or limiting sibling cannibalism. In general, experimenters should aim to demonstrate maternal control of trophic egg production, because such a result provides strong support for the adaptive hypothesis. For example, experiments in the white-eyed ( ferox) show that females, after laying a clutch of fertile eggs, produce a second clutch of trophic eggs if stimulated by the presence of offspring; but if offspring are removed, the second clutch is also composed of fertile eggs (Kim and Roland 2000; Kim and Horel2000). The result demonstrates that mothers are not constrained to producing infertile eggs in this species.

Comparative analyses

Several comparative results may establish the adaptive nature of trophic egg production (Table 2.2,2.3). For example, a straightforward test within any group of homogeneous animals (e.g., the coccinellid beetles) would be to compare the percentage of infertility in species that lay eggs in batches and that exhibit sibling cannibalism to the nearest relative species that does not. If infertility is higher for species in which offspring consume the eggs, then some portion of that infertility is due to an evolved response in mothers. These results would, furthermore, establish cooperation as the functional explanation for trophic eggs - unless, as noted above for the plasticity hypothesis, sibling cannibalism is also related to those predictive variables and mothers lay trophic eggs to limit that cannibalism response. Researchers must take care that sibling cannibalism tendencies are not a confounding factor in such studies. Another consideration is that there are probably different selective forces driving the evolution of trophic eggs in different groups. For this reason, it will be best to choose a homogenous group of trophic egg laying animals for a comparative analysis (e.g., tropical frogs or aphidophagous insects; Table 2.1). There have been no cross-taxa studies of trophic egg laying to date.

2.6 Challenges for future research

Future empirical studies on trophic egg laying animals should focus on determining whether a) trophic eggs are a maternal adaptation and b) whether the functional explanation is an attempt to reduce sibling cannibalism or an attempt to ensure that egg-eating among offspring occurs. The above text highlights some empirical and theoretical challenges in collecting evidence about the adaptive value of trophic eggs. Beyond studies of particular species, there is a need to set trophic egg laying behaviour in the context of other parental care, sibling rivalry, and provisioning theory. Three areas of research will facilitate this process. First, there is a great need for theoretical studies. Currently, models of sibling cannibalism must be co-opted for trophic egg questions (but see Chapter 3). Theoretical questions include implications of a decline in per capita offspring fitness with increasing clutch size on the inclusive fitness model outlined here. It is also not clear how the cost of trophic eggs affect their use in facilitating or limiting cannibalism: how cheap do trophic eggs have to be before they can be used efficiently to reduce sibling cannibalism? Second, work is needed in considering how trophic eggs may have originated and evolved. Before a maternal adaptation of producing trophic eggs could arise, a necessary pre-adaptation is that offspring consume material in or near the egg batch. Egg cannibalism by offspring might arise to exploit (1) infertile or inviable eggs, or (2) viable siblings with delayed hatch (Crespi 1992).1n (I), mothers might have then been selected to increase the production of infertile eggs. However, evolving intentional infertility may come with imperfect control of infertility, and may be selected against because it would sometimes lead to the creation of more infertile eggs than would be adaptive for mothers (in the same way that hosts of cuckoo brood parasites may accept cuckoo eggs because rejection of parasite eggs comes with the risk of rejecting host eggs as well; Lotem et a1 1995). However, it may also be possible for mothers to increase infertility without frequent mistakes causing excessive levels of infertility. Furthermore, even if occasionally more infertile eggs are made, the fitness cost of excess infertility may not be high, because if mothers are creating infertile eggs it implies that the relative benefit to offspring exceeds 100%. In (2), trophic egg laying might arise to enhance sibling cannibalism or to limit cannibalism of viable offspring. Phylogenetic analyses within homogeneous groups in which trophic eggs may presumably serve the same adaptive function (e.g., tropical frogs; see Weygoldt 1980) may help uncover how trophic egg laying arises. It may be possible to map the relevant traits - presence of infertile or inviable eggs, sibling cannibalism of eggs with delayed hatch, and trophic egg laying

- to a phylogenetic tree. If there is enough variation in the traits, it may be possible to determine the sequence of origin of each. Third, to understand the occurrence of trophic eggs across taxa, it will be necessary to predict when trophic eggs should be employed rather than other alternative tactics that facilitate or limit sibling cannibalism (such as clutch size, egg size, managing cannibalism of viable offspring (see 'Cooperation'; reviewed by Crespi 1992), and parental care). Furthermore, we can predict the use of particular tactics in species depending on environmental condition. If parents in some taxa face environments that are homogeneous in starvation risk for offspring, then an (apparently) inflexible trait such as egg size may be used, and we should expect egg size to be negatively linked to starvation risk across taxa. On the other hand, many species will encounter environments that vary in offspring starvation risk, either across clutches or across parents. If variation in offspring starvation risk is unpredictable, we suggest that parents will often manipulate clutch size to mitigate offspring starvation risk (e.g., to produce insurance offspring). If risk is predictable, a solution is to use potentially plastic strategies, such as hatching synchrony and trophic eggs, that can be adjusted between environments that differ in risk. Finally, we may consider trophic eggs a subset of the larger adaptation of producing trophic offspring, including viable eggs that could have hatched if not eaten and hatched offspring that are killed and cannibalized.

2.7 Conclusions

Alexander (1974) proposed that trophic eggs are an example of parental manipulation of offspring - that parents use some offspring as meals for others. Parents may do so to reduce offspring starvation risk or to reduce cannibalism among offspring. These two adaptive possibilities, parent-offspring cooperation or conflict, must be tested against the hypothesis that trophic eggs are produced by unavoidable constraint. Distinguishing among the hypotheses may be difficult in practice, because much evidence that is related to trophic egg laying and egg consumption by offspring cannot falsify any hypotheses. To establish that trophic egg production is an adaptation, researchers should focus on collecting evidence that falsifies the constraint hypothesis (Table 2.3). Such evidence would include observations of the morphology of trophic eggs or the associated maternal care, or the plastic adjustment of trophic egg levels between environments that differ in starvation risk. To determine the adaptive explanation for trophic egg laying, studies should concentrate on evidence that can distinguish between the two right-most columns of Table 2.3. The strongest evidence would be a reaction norm of trophic egg production along an offspring starvation risk gradient, a demonstration that trophic eggs are more metabolically costly than normal eggs, or a comparative study relating trophic eggs to predicted variables. In the absence of strong evidence, researchers should attempt to collect several lines of equivocal evidence supporting one or the other hypotheses, preferably with evidence falsifying the other. We emphasize the mutual exclusivity of the conflict vs. cooperation hypotheses in any particular system, though both may explain trophic eggs in some systems. Awareness of this incompatibility has not been apparent in the literature (e.g., Nakahira 1994; Kim and Roland 2000; Kudo and Nakahira 2001). Trophic eggs are rarely treated in a functional context, yet they have the potential to illustrate and test basic behavioural ecological ideas. There is a great need for theoretical work and cross-species phylogenetic analyses in this area. Hypotheses for trophic egg laying behaviour relate to some of the most important topics in behavioural ecology: inclusive fitness, parent- offspring conflict, sibling rivalry and parental care. They are also an interesting example of an extreme and complex parental care behaviour - sacrificing potentially viable offspring to improve the quality of other offspring. Because trophic eggs occur in animals of diverse ecologies and life histories (Table 2. l), and because there are several alternative parental care tactics that may accomplish the same end (discussed under 'Challenges'), investigators can both identify and test predictions about the ecological factors related to trophic egg laying. 2.8 Literature cited

Alexander, R. D. 1974. The evolution of social behaviour. Annual Review of Ecology and Systematics 5:325-383.

Banks, C. J. 1956. Observations on the behaviour and mortality in Coccinellidae before dispersal from the egg shells. Proceedings of the Royal Entomology Society of London (A) 3 1:56- 60.

Baur, A., and B. Baur. 1998. Altitudinal variation in size and composition of eggs in the land snail Arianta arbustorum. Canadian Journal of Zoology 76:2067-2074.

Baur, B. 1990. Seasonal changes in clutch size, egg size, and mode of oviposition in Arianta arbustorum L. (Gastropoda) from alpine populations. Zoologischer Anzeiger 225:253- 264.

-. 1992. Cannibalism in gastropods. Pages 102-127 in M. A. Elgar, and B. J. Crespi, eds. Cannibalism: ecology and evolution among diverse taxa. Oxford University Press, Oxford.

Brown, H. D. 1972. The behaviour of newly hatched coccinellid larvae (Coleoptera: Coccinellidae). Journal of the Entomological Society of South Africa 35: 149-157.

Cohen, A.C., and D. K. Brummett. 1997. The non-abundant nutrient (NAN) concept as a determinant of predator-prey fitness. Entomophaga 4295-9 1.

Christians, J. K. 2002. Avian egg size: variation within species and inflexibility within individuals. Biological Reviews 77: 1-26.

Crespi, B. J. 1992. Cannibalism and trophic eggs in subsocial and eusocial insects. Pages 176-213 in M. A. Elgar, and B. J. Crespi, eds. Cannibalism: ecology and evolution among diverse taxa. Oxford University Press, Oxford.

Crump, M. L. 1996. Parental care among the Amphibia. Advances in the Study of Behavior 25: 109-144.

Desbuquois, C., L. Chevalier, and L. Madec. 2000. Variability of egg cannibalism in the land snail Helix aspersa in relation to the number of eggs available and the presence of soil. Journal of Molluscan Studies 66:273-28 1.

Desbuquois, C., and L. Madec. 1998. Within-clutch egg cannibalism variability in hatchlings of the land snail Helix aspersa (Pulmonata: Stylommatophora): influence of two proximate factors. Malacologia 39: 167-173.

Dickinson, J. L. 1992. Egg cannibalism by larvae and adults of the milkweed leaf beetle (Labidomera clivicollis, Coleoptera: Chrysomelidae). Ecological Entomology 17:209- 218.

Dimetry, N. Z. 197 1. The consequences of egg cannibalism in Adalia bipunctata (Coleoptera: Coccinellidae). Entomophaga 19:445-45 1. Dixon, A. F. G. 2000. predator-prey dynamics. Cambridge University Press, London.

Dixon, A. F. G., and Y. Guo. 1993. Egg and cluster size in ladybird beetles (Coleoptera: Coccinellidae): the direct and indirect effects of aphid abundance. European Journal of Entomology 90:457-463.

Dopazo, H., and P. Alberch. 1994. Preliminary results on optional and intrauterine siblicide in Salamandra salamandra populations from northern Spain. Mertensiella 4:125-138.

Duellman, W. E., and L. Trueb. 1986. Biology of . McGraw-Hill, New York.

Eickwort, K. R. 1973. Cannibalism and kin selection in Labidomera clivicollis (Coleoptera: Chrysomelidae). American Naturalist 107:452453.

Evans, T. A., E. J. Wallis, and M. A. Elgar. 1995. Making a meal of mother. Nature 376:299.

Forbes, L. S. 1993. Avian brood reduction and parent-offspring "conflict". American Naturalist 142:82-117.

Frechette, B., and D. Coderre. 2000. Oviposition strategy of the green lacewing Chrysoperla rufilabris (Neuroptera: Chrysopidae) in response to extraguild prey availability. European Journal of Entomology 97507-5 10.

Gibson, R. C., and K. R. Buley. 2004. Maternal care and obligatory oophagy in Leptodactylus fallax: a new reproductive mode in frogs. Copeia 2004: 128-135.

Gobin, B., C. Peeters, and J. Billen. 1998. Production of trophic eggs by virgin workers in the ponerine Gnamptogenys menadensis. Physiological Entomology 23:329-336.

Godfray, H. C. J. 1987. The evolution of clutch size in invertebrates. Oxford surveys in evolutionary biology 4: 1 17-154.

Godfray, H. C. J., and A. B. Harper. 1990. The evolution of brood reduction by siblicide in birds. Journal of Theoretical Biology 145: 163-175.

Hamilton, W. D. 1964. The genetical evolution of social behaviour. I, 11. Journal of Theoretical Biology 7: 1-52.

Hawkes, 0. A. M., and T. F. Marriner. 1927. A preliminary account of the life-history of Coccinella 11-punctata (L.). Transactions of the Entomological Society of London 75:47-53.

Henry, C. S. 1972. Eggs and rapagula of Ululodes and Ascaloptynx (Neuroptera: Ascalaphidae): a comparative study. Psyche 79: 1-22.

Heying, H. E. 200 1. Social and reproductive behaviour in the Madagascan poison frog, Mantella laevigata, with comparisons to the dendrobatids. Animal Behaviour 61 567-577.

Honek, A. 1993. Intraspecific variation in body size and fecundity in insects: a general relationship. Oikos 66:483-492. Jungfer, K. H. 1996. and parental care of the coronated treefrog, Anotheca spinosa (Steindachner, 1864) (Anura: Hylidae). Herpetologica 52:25-32.

Kam, Y.-C., Y.-H. Chen, T.-C. Chen, and I.-R. Tsai. 2000. Maternal brood care of an arboreal breeder, Chirixalus eiffingeri (Anura: Rhacophoridae) from Taiwan. Behaviour 137: 137- 151.

Kam, Y.-C., Z. S. Chuang, and C. F. Yen. 1996. Reproduction, oviposition-site selection, and tadpole oophagy of an arboreal nester, Chirixalus eiflngeri (Rhacophoridae), from Taiwan. Journal of Herpetology 30:52-59.

Kawai, A. 1978. Sibling cannibalism in the first instar larvae of Harmonia axyridis Pallas (Coleoptera: Coccinellidae). Kontyu 46: 14-19.

Kim, K.-W., and A. Horel. 2000. Matriphagy in the spider (Araneidae, ): an example of mother offspring interactions. Ethology 104: 102 1- 1037.

Kim, K. W., and C. Roland. 2000. Trophic egg laying in the spider, Amaurobius ferox: mother- offspring interactions and hnctional value. Behavioural Processes 50:3 1-42.

Kudo, S.-I. 2001. Intraclutch egg size variation in acanthosomatid bugs: adaptive allocation of maternal investment? Oikos 92:208-214.

Kudo, S.-I., and T. Nakahira. 2001. Trophic egg production and its hnction in a subsocial cydnid bug. Advances in Ethology 3: 197.

Lotem, A., H. Nakamura, and A. Zahavi. 1995. Constraints on egg discrimination and cuckoo- host co-evolution. Animal Behaviour 49: 1185-1209.

Masuko, K. 2003. Larval oophagy in the ant Amblyopone silvestrii (Hymenoptera, Formicidae). lnsectes Sociaux 50:3 17-322.

McKaye, K. R. 1986. Trophic eggs and parental foraging for young by the catfish Bagrus meridionalis of Lake Malawi, Africa. Oecologia 69:367-369.

Mock, D. W. 1994. Brood reduction: narrow sense, broad sense. Journal of Avian Biology 25:3- 7.

Mock, D. W., and L. S. Forbes. 1995. The evolution of parental optimism. Trends in Ecology and Evolution 10: 130-134.

Mock, D. W., and G. A. Parker. 1997. The evolution of sibling rivalry. Oxford University Press, Oxford.

Nakahira, T. 1994. Production of trophic eggs in the subsocial burrower bug Admerus triguttulus. Naturwissenschaften 8 1:413-4 14.

O'Connor, R. J. 1978. Brood reduction in birds: selection for fratricide, infanticide and suicide. Animal Behaviour 26:79-96. Osawa, N. 1989. Sibling and non-sibling cannibalism by larvae of a lady beetle Harmonia axyridis Pallas (Coleoptera: Coccinellidae) in the field. Researches in Population Ecology 31:153-160.

-. 1991. Consequences of sibling cannibalism for the fitness of mother and offspring of the ladybird beetle Harmonia axyridis. Pages 221-225 in L. Polgar, R. J. Chambers, A. F. G. Dixon, and I. Hodek, eds. Behaviour and impact of Aphidophaga. Academia, Prague.

-. 1992. Sibling cannibalism in the ladybird beetle Harmonia axyridis: fitness consequences for mother and offspring. Researches in Population Ecology 34:45-55.

-. 2002. Sex-dependent effects of sibling cannibalism on life history traits of the ladybird beetle Harmonia axyridis (Coleoptera: Coccinellidae). Biological Journal of the Linnean Society 76:349-360.

-. 2003. The influence of female oviposition strategy on sibling cannibalism in the ladybird beetle Harmonia axyridis (Coleoptera: Coccinellidae). European Journal of Entomology 100:43-48.

Parker, G. A., and D. W. Mock. 1987. Parent-offspring conflict over clutch size. Evolutionary Ecology 1: 16 1- 174.

Peck, W. B., and W. H. Whitcomb. 1970. Studies on the biology of a spider, Chiracanthium inclusum (Hentz). Bulletin of the Arkansas Experiment Station 753: 1-75.

Pfennig, D. W. 1997. Kinship and cannibalism. Bioscience 47:667-775.

Pienkowski, R. L. 1965. The incidence and effect of egg cannibalism in first instar Coleomegilla maculata lengi (Coleoptera: Coccinellidae). Annals of the Entomological Society of America 58:150-153.

Polis, G. A. 198 1. The evolution and dynamics of intraspecific . Annual Review of Ecology and Systematics l2:225-25 1.

-. 1984. Intraspecific predation and "infant killing" among invertebrates. Pages 87-104 in G. Hausfater, and S. Hrdy, eds. Infanticide: comparative and evolutionary perspectives. Aldine Publishing Company, New York.

Reeve, H.K., and P.W. Sherman. 1993. Adaptation and the goals of evolutionary research. Quarterly Review of Biology 68: 1-32.

Specty, O., , G. Febvay, S. Grenier, B. Delobel, C. Piotte, J.-F. Pageaux, A. Ferran, and J. Guillard. 1999. Nutritional plasticity of the predatory ladybeetle Harmonia axyridis (Coleoptera: Coccinellidae): comparison between natural and substitution prey. Archives of Insect Biochemistry and Physiology 52:s 1-9 1.

Springer, S. 1948. Oviphagous embryos of the sand , Carcharias taurus. Copeia 1948: 153- 157.

Stinson, C. H. 1979. On the selective advantage of fratricide in raptors. Evolution 33:1219-1225. Summers, K., and W. Amos. 1997. Behavioural, ecological, and molecular genetic analyses of reproductive strategies in the Amazonian dart-poison frog, Dendrobates ventrimaculatus. Behavioural Ecology 8:260-267.

Thompson, R. L. 1996. Larval habitat, ecology and parental investment of Osteopilus brunneus (Hylidae) in R. Powell, and R. W. Henderson, eds. Contributions to West Indian herpetology: a tribute to Albert Schwartz. Society for the Study of Amphibians and Reptiles, Ithaca, New York.

Thornhill, R. 1990. The study of adaptation. Pages 3 1-61 in M. Bekoff, and D. Jamieson, eds. Interpretation and explanation in the study of animal behaviour. Westview Press, Boulder.

Ueda, H. 1986. Reproduction of Chirixalus eijfingeri (Boettger). Scientific Report of the Laboratory for Biology, Hiroshima University 8: 109-1 16.

Valerio, C. E. 1974. Feeding on eggs by spiderlings of Achaeranea tepidariorum (Araneae, Theridiidae), and the significance of the quiescent instar in spiders. Journal of Arachnology 2:57-63.

-. 1977. Population structure in the spider Achaeranea tepidariorum (Araneae, Theridiidae). Journal of Arachnology 3: 1 85-190.

Wassersug, R. J., K. J. Krogner, and R. F. Ingr. 1981. Adaptations for life in tree holes by rhacophorid tadpoles from Thailand. Journal of Herpetology 15:4 1-52.

West, M. J., and R. D. Alexander. 1963. Sub-social behaviour in a burrowing cricket Anurogryllus muticus (De Geer). Ohio Journal of Science 63: 19-24.

Weygoldt, P. 1980. Complex brood care and reproductive behaviour in captive poison-arrow frogs, Dendrobatespumilio 0. Schmidt. Behavioural Ecology and Sociobiology 7:329- 332.

Williams, G.C. 1966. Adaptation and natural selection. Princeton University Press, Princeton.

Wourms, J. P., B. D. Grove, and J. Lombardi. 1988. The maternal-embryonic relationship in viviparous . Pages 1-134 in W. S. Hoar, and D. J. Randall, eds. physiology. Academic Press, San Diego, California. Figure 2-1 The inclusive fitness interests in sibling cannibalism of a parent and her cannibalistic offspring; r is the coefficient of relatedness between siblings. Parent-offspring conflict or cooperation depends on the survival benefits of cannibalism, relative to survival of non- cannibalistic offspring. In the range from 0-r%, neither parents nor offspring benefit from sibling cannibalism. Offspring. - cannibalizing I siblings 0 - Ph r- 100% > 100% Paornts benefit Relative increase in offs~ringsurvival through cannibalism cannibalism among offspring Table 2-1 Taxa that produce trophic eggs, grouped by the behavioural evidence that trophic egg laying is adaptive Taxon Cannibalism of viable siblings? Source 1. No clear evidence that trophic eggs are adaptive (no parental care; offspring eat non-differentiated infertile eggs) Milkweed leaf beetle, Labidomera clivicollis + Dickinson 1992 Green lacewing, Chrysoperla rufilabris + Frechette and Coderre 2000 Common house spider, Achaearanea Valerio 1974; Valerio 1977 tepidariorum Land snails + Baur 1990; Desbuquois and Madec 1998; Desbuquois et al. 2000 Prosobranch gastropods + Baur 1992 (review) Ladybird beetles, Coccinellidae 2 spot ladybird, Adalia bipunctata + Banks 1956; Dimetry 1971 Cheilomense lunata + Brown 1972 11 spot ladybird, Coccinella I I-punctata Hawkes and Marriner 1927 7 spot ladybird, Coccinella septempunctata + Banks 1956 Coleomegilla maculata lengi + Pienkowski 1965 Multi-spotted Asian ladybird, Harmonia axyridis + Chapter 4; Kawai 1978; Osawa 1989; Osawa 1991; Osawa 1992 Lioadalia flavomaculata + Brown 1972 Propylea 14-punctata + Banks 1956

2. Behavioural evidence that trophic eggs are Unique Parental Cannibalism of adaptive morphology? care? viable siblings? Burrowing cricket, Anurogryllus muticus + +*** West and Alexander 1963 Owlfly, Ascaloptynx furciger + Henry 1972 Burrower bug, Admerus triguttulus + + Nakahira 1994; Kudo and Nakahira 200 1 Spiders, Araneae White-eyed spider, Amaurobius ferox ? +** Kim and Roland 2000; Kim and Horel 2000 Australian social crab spider, Diaea ergandros + +** Evans et al. 1995 Yellow sac spider, Chiracanthium inclusum + Peck and Whitcomb 1970 Taxon Cannibalism of viable siblings? Source Amphibians Tropical frogs Dendrobatidae, Mantellidae and Leptodactylidae - + Crump 1996 (review); Summers and Amos 1997; Heying 2001; Gibson and Buley 2004 Hylidae + + +/- * Crump 1996 (review); Jungfer 1996; Thompson 1996 Rhacophoridae +/- * + Wassersug et al. 198 1; Ueda 1986; Kam et al. 1996; Kam et al. 2000 Salamander, Salamandra atra -? -? Duellman and Trueb 1986; Dopazo and Alberch 1994 Other African catfish, Bagrus- rneridionalis - + ? McKaye 1986 Sand tiger shark -? -? + springer 1948; Wourms et al. 1988 * Occurs in some species ** Matriphagy *** Matriphagy after mother's death Table 2-2 Hypotheses regarding the function of trophic eggs and predictions of ecological factors or species traits that should be correlated with trophic egg laying. Hypotheses and predictions are grouped by whether they consider trophic eggs a tactic for reducing sibling cannibalism or provisioning offspring. Hypothesislprediction Source Taxon referred toa 1. Parent-offspring conflict reduction: Crespi 1992; see also Nakahira 1994; Mock and Admerus triguttulus (Nakahira); mothers produce trophic eggs to limit or ~arker1997; Kudo 2001 ~canthosoiatidae(Kudo) prevent cannibalism among offspring Trophic eggs should be more common in Crespi 1992 monogamous species If trophic egg production is at an optimum Crespi 1992 for mothers and offspring, viable offspring should not be eaten

2. Parents use some offspring as food for Chapters 3,4; Alexander 1974; Stinson 1979; Polis Eagles (Stinson); Coccinellidae (Osawa; others, 'principle of parental manipulation' 1981; Godfray 1987; Crespi 1992; Osawa 1992; Dixon; Perry and Roitberg); Chrysoperla Mock 1994; Mock and ~orbes1995; Pfennig 1997; rufilabris (Frechette and Coderre) Dixon 2000; Frechette and Coderre 2000 % Trophic eggs are likely when.. . There is abundant food during offspring Alexander 1974; see also Mock and Forbes 1995; creation but limited food during offspring Mock and Parker 1997 provisioning/foraging (icebox hypothesis) Parents have no other way to convert Alexander 1974; Polis 1981; Polis 1984; Crespi 1992 resources to nourishment Egg size is inflexible Polis 198 1; Polis 1984; Crespi 1992; Mock and Coccinellidae (Dixon) Parker 1997; Dixon 2000 Offspring starvation risk is generally high Godfray 1987; Crespi 1992; Thompson 1996 Tropical frogs (Thompson) Offspring starvation risk is unpredictable at Osawa 1992; Mock and Parker 1997; Dixon 2000 Coccinellidae (Osawa, Dixon) oviposition Resource levels are predictable; then Chapter 4; Frechette and Coderre 2000; Osawa 2003 C. rufilabris (Frechette and Coderre); mothers should adjust production of Coccinellidae (Osawa, Perry and trophic eggs accordingly Roitberg) Offspring diet is protein limited It is easier or less risky for mothers to find McKaye 1986; Crespi 1992; Nakahira 1994; Kim and Bagrus meridionalis (McKaye); food than for offspring Roland 2000; Masuko 2003 Amaurobius ferox (Kim and Roland);

Amblvo~one,1 silvestrii (Masuko) " Listed for authors who referred to particular species; many authors made general predictions Table 2-3 Evidence required to distinguish among hypotheses of trophic egg laying: whether trophic eggs are produced by non-evolved infertility or by maternal adaptation (constraint vs. adaptive hypotheses); and, if adaptive, whether trophic eggs function to reduce parent-offspring conflict over sibling cannibalism or to provision offspring (conflict reduction vs. provisioning hypotheses). Lines of evidence are grouped by how they are obtained: through observation (natural history), behavioural studies, or comparative analyses. Evidence may be 'consistent' (if it is not predicted by and does disprove a hypothesis), 'necessary but not sufficient' (if the evidence is not found, the hypothesis cannot be true, but finding the evidence does not mean the hypothesis is correct), or strongly or weakly supportive or falsifying. If investigators cannot obtain falsifying or strongly supportive evidence, the next best option is to collect several lines of weakly falsifying or supporting evidence. The framework is developed for the scenario of one trophic egg-eating or cannibalising event, but the qualitative predictions are general to the case where more than one egg or sibling is eaten. Implications for.. . Constraint Adaptive hypothesis hypothesis Evidence Conflict reduction Provisioning 1. Natural history Trophic eggs morphologically distinct Falsifies Consistent Consistent Mother brings trophic eggs to offspring Falsifies* Consistent Consistent Other complex provisioning behaviours Weakly falsifies Consistent Weak support Cannibalism of viable siblings frequently occurs Consistent Weakly falsifies* Consistent

Other potential tactics are constrained Consistent Consistent Weak support

2. Behavioural studies Offspring benefit from eating eggs Consistent Necessary but not sufficient Necessary but not sufficient Relative benefit most often 0-100% Consistent Consistent Falsifies w > Consistent Falsifies Consistent 0 Relative benefit most often 100% Directional Plasticity: mothers produce trophic eggs when relative Falsifies Falsifies Strong support benefit>100%, and produce fewer when relative benefit

3. Cross-taxa comparisons Trophic egg laying related to temporal pattern of resource abundance Falsifies Falsifies, unless related to Strong evidence increasing sib cannibalism Trophic eggs more common in protein limited taxa Falsifies* * Falsifies, unless related to Strong evidence increasing sib cannibalism Trophic eggs more common when it is easier or less risky for mother Falsifies Falsifies, unless related to Strong evidence than offspring to find food increasing sib cannibalism * But see discussion of this point in text ** But may mean offspring adaptation is more important in these taxa Chapter 3 Co-Evolution of Maternal and Offspring Effectors of Sibling Cannibalism

3.1 Abstract

Sibling rivalry and parent-offspring conflict games are generally modelled separately, yet their interaction determines the co-evolved outcome for many games within family units. For example, when sibling cannibalism occurs in clutches of more than two offspring, parent- offspring conflict or cooperation may exist over the occurrence of cannibalism, and siblings may compete over the opportunity to cannibalize. Using a game theoretic genetic algorithm approach, we derive optimal offspring cannibalism behaviour and parental behaviours influencing cannibalism (hatching synchrony and trophic egg production), under a range of values for the relative benefit of cannibalism. The results are compared to the predictions of a frequency- independent inclusive fitness analysis that does not incorporate the parent-offspring or between sibling games. We also model the expected effect of relative versus absolute starvation risk, the presence of infertile eggs, and time limitations on cannibalism behaviour. The addition of a game component to the model led to a prediction of increased cannibalism, compared to the frequency- independent prediction, as offspring competed to eat siblings. As cannibalism among offspring increased, mothers increased behaviours to limit cannibalism. Absolute survival probability had no influence on offspring or parent behaviours; as predicted by the analytical model, the relative increase in survival from cannibalism was the important parameter. When infertile eggs were present, offspring risked eating viable siblings in competition to eat infertile eggs, enhancing parent-offspring conflict. Allowing offspring to wait to cannibalize eggs until all viable siblings hatched had no effect on evolved cannibalism behaviours, because it was never optimal for offspring to wait to cannibalize (except at low levels of relative survival benefit from cannibalism). Finally, the results demonstrate the utility of trophic egg-laying behaviour as a maternal mechanism to promote sibling cannibalism.

3.2 Introduction

For many egg batch-laying animals, interactions among siblings early in life have important fitness consequences (Roitberg and Mange1 1993; Mock and Forbes 1995; Mock and Parker 1997). Siblings can be strong competitors for limited parental and environmental resources, yet may also make important contributions to an individual's inclusive fitness. Thus, one should observe seemingly restrained selfishness in interactions among siblings (Mock and Parker 1997). The outcome of sibling interactions affects parental fitness, and as such, parents also have an evolutionary stake in the extent of sibling rivalry. Depending on the relative benefit that an offspring receives, parental traits that facilitate or limit selfish interactions among siblings may be favoured (O'Connor 1978). Hence, the observed degree of selfishness among siblings should reflect a web of influence, involving traits that evolve in both parents and offspring. Sibling cannibalism represents an extreme of sibling antagonism. Cannibalistic individuals acquire a nutritious meal and may reduce future competition, but in doing so sacrifice a close relative. Applying Hamilton's rule (Hamilton 1967), individuals should cannibalize siblings when their own survival is increased by a factor at least as great as the coefficient of relatedness between siblings (Eickwort 1973). As such, cannibalism is most likely to occur if within-clutch relatedness is low, offspring starvation risk is high, or cannibalism provides a large survival benefit. Parental interests in sibling cannibalism depend critically on the relative survival benefit that accrues to cannibal offspring. Parents should use tactics that limit or facilitate cannibalism according to whether they gain more net offspring from the increased survival of cannibals than they lose as victims (Chapter 2, O'Connor 1978; Crespi 1992). As an example, female Harmonia axyridis ladybirds (Coleoptera: Coccinellidae) may use two oviposition behaviours that modulate the opportunity for cannibalism: hatch synchrony and the production of trophic eggs (non- developing eggs that are consumed by hatching larvae). Current understanding of siblicide as an adaptation is based on frequency independent inclusive fitness analysis (Chapter 2, Eickwort 1973; O'Connor 1978; Pfennig 1997), or game theoretic models of between-sibling interactions (Dickins and Clark 1987; Godfray and Harper 1990; Forbes 1993; Godfray 1995; Rodriguez-GironCs 1996; Mock and Parker 1997; Mock and Parker 1998) or of parent-offspring conflict over clutch size (Parker and Mock 1987). In the sibling cannibalism scenario, however, two games are played simultaneously: between siblings over which individuals do the cannibalizing (when it is beneficial to eat a sibling; Parker and Mock 1987), and between parents and offspring over the occurrence of cannibalism. With the complexity of a co-evolving, multiple-trait system, it is necessary to extend the traditional game theory approach in order to find stable optima in the large set of possible trait combinations, i.e., co-evolved gene complexes. A suitable approach, which we adopt here, is the use of simulation models with a genetic algorithm search component. A genetic algorithm (GA) is a search algorithm that finds an optimal strategy set by allowing coded alleles to compete for space on a population genome, over many simulated generations (Goldberg 1989). The approach approximates a series of negotiated interactions in time (McNamara et al. 1999), as opposed to the traditional game theory method of finding the best strategy for one player given the strategies played by others (e.g., Parker and Mock 1987). Because GAS can simulate a game over evolutionary time, and can search a very large strategy set quickly, they are an appropriate technique for complex co-evolutionary games (Forrest 1993; Robertson et al. 1998; Reinhold et al. 2002). We ask: what is the outcome of evolutionary conflict or cooperation over sibling cannibalism between mother and offspring? Using H. axyridis as a focal taxon, we model the co- evolution of sibling cannibalism and two maternal behaviours: hatching synchrony (whereby mothers may limit or encourage cannibalism among offspring) and trophic egg production (a way for mothers to facilitate egg-eating). We begin with a frequency independent, deterministic inclusive fitness analysis considering the conditions under which cannibalism benefits mothers and offspring. The predicted conditions are then tested in stochastic simulations of within-clutch cannibalistic interactions, where a GA is used to find sets of stable trait values. The simulations also model the effects of three factors: relative versus absolute starvation risk, non-evolved egg infertility, and time limitations on cannibalism. We demonstrate that if each trait is permitted to evolve alone (i.e., the parent-offspring game is prevented), then the results qualitatively match the predictions of the frequency-independent model. Adding a co-evolutionary game between mothers and offspring, however, favours increased cannibalism and maternal response to cannibalism. The presence of non-evolved infertile eggs also drives evolution for increased cannibalistic tendencies among offspring. We compare the results to data available for ladybirds and discuss the model's relevance to other taxa.

3.1 Model

Biological system

We use a model organism to focus our discussion. However, the model is generally applicable to any taxon where parent-offspring conflict and sibling competition co-occur. Harmonia axyridis, like many coccinellids, lays eggs in batches of 10-50. During hatching, larvae emerge asynchronously and there is the potential for egg consumption within the clutch. Both trophic eggs and viable siblings with delayed hatch may be consumed, with substantial fitness benefits for the cannibal (16% to 226%, Osawa 1992; reviewed by Hodek and Honek 1996). Emerged larvae are not vulnerable to cohort cannibalism. Cannibal larvae consume a portion of an egg or an entire egg, then become quiescent for approximately 24 hours. By that time, other eggs have generally hatched or been eaten by other cannibals, so consumption of more than one egg per cannibal is rare. Cannibals do not appear to discriminate between viable and infertile eggs (Banks 1956; Pienkowski 1965; Brown 1972; Osawa 1992; Majerus 1994). In H. axyridis, 15.2% (Kawai 1978) to 24.5% (Osawa 1992) of eggs are infertile.

Inclusive fitness argument

Eickwort (1973; see also Pfennig 1997, and Chapter 2) used Hamilton's rule to investigate when an individual should cannibalize a sibling. To review the argument, consider a non-cannibal with a probability of survival, p. A cannibal receives an increase in survival, p', through reduced starvation risk, and has total survival (p + p'). By re-stating Hamilton's Rule ((Hamilton 1964)) for the spread of a selfish act ((Mock and Parker 1997)), sibling cannibalism is favoured when the relative benefit to the cannibal exceeds the cost to the victim, weighted by relatedness - that is, when p' > pr. This can be restated as p'/p > r (1) For example, a cannibal must increase survival by at least 50% to profit from consuming a full sibling. Throughout the paper, we refer to the term (p'/p) * 100 as the relative benefit of egg- eating (ME). A parent favours sibling cannibalism when more net offspring are gained from the increased survival of cannibals than are lost as victims (see O'Connor's (1978) model of brood reduction). With pairwise sibling interactions, this occurs when P'>P (2) that is, parents favour cannibalism when the survival of cannibals is at least doubled through cannibalism (when RBE > unity). The model demonstrates that parent and offspring interests may overlap or conflict, depending on intra-clutch relatedness and the relative increase in survival from cannibalism. Neither parents nor offspring favour cannibalism when RBE < r. When r < ME< 1, it is in the interests of offspring to cannibalize siblings and the interests of parents to prevent cannibalism. When RBE > 1, parents and offspring agree that cannibalism should proceed. The model also shows that, for sibling cannibalism to be beneficial, a non-cannibal's probability of survival must be low. For example, if p > 213, individuals cannot be selected to eat full siblings due to the constraint that (p' + p) 51. Similarly, survival must be less than 50% for parents to favour sibling cannibalism, else (p'lp) cannot exceed one (Fig. 2-1). Genetic algorithm-based simulations: the basic model

The H. axyridis-like system contains three interdependent behaviours: the maternal behaviours that determine hatch synchrony and the proportion of trophic eggs within egg batches, and the offspring behaviour of eating eggs within the natal clutch. In the simulations, individuals are represented as strings of binary information that code for the three behaviours, with 16 possible alleles for each gene. Every individual is a female that possesses all three genes, one of which is expressed as an offspring and the other two as an adult. Details of the model and simulations are given in the chapter appendix. Hatching synchrony phenotypes range from complete synchrony with all offspring hatching in one time unit, to complete asynchrony with one offspring hatching per time unit (Table 3.1). Trophic egg alleles code for the production of zero to 15 trophic eggs. The cannibalism delay gene codes for a lag time that offspring wait after emergence before seeking an egg to eat within their clutch. Phenotypes for the cannibalism delay gene range from waiting one time unit afier emergence to waiting 16 time units, by which time all viable siblings have hatched and offspring should consume any unhatched eggs (if eggs provide a survival benefit). Because the range of alleles allowed maximal variation in phenotypes (0 to 100% trophic egg production; 0 to maximal cannibalism events permitted through hatching synchrony; minimal cannibalism delay to long enough that viable siblings survive), we are confident that number of alleles tested allowed fine enough resolution of phenotypes in simulations. In the simulations, loci assort independently. Interactions within clutches are simulated simultaneously and independently. For each clutch, the number of trophic eggs is determined by the mother's trophic-egg-gene allele and trophic status is assigned randomly to offspring. In each of 16 time units, offspring are hatched according to the maternal allele for hatch synchrony (Table 3. l), with the hatching offspring selected randomly. Offspring wait some number of time units determined by their allele for cannibalism delay, and then consume an egg if there is one available within their clutch. When time ends, whether live offspring survive starvation is determined by drawing a random number for comparison to survival parameters, which differ depending on whether the offspring has consumed an egg. Selection in the model. Fitness is assigned indirectly: successful alleles make more copies of themselves and are proportionally more represented in the pool of living offspring from which the next generation's parents are randomly drawn. We do not incorporate a clutch-level selection component (which might arise if larger surviving clutches have higher per capita survival, as in Breden and Wade 1989) because siblings disperse from egg batches and likely have little further interaction after sibling cannibalism occurs in H. axyridis. There is thus no basis for group benefits in this system, though they may be important in other systems. Mating and offspring generation are described in Appendix I. We conducted five simulation runs for each parameter set with different random number seeds. Convergent values in different runs allow more confidence that the GA has found a global optimum. The simulations are used to search for co-evolutionary outcomes under a range of ecological conditions (i.e., starvation risk and relative benefit of egg eating), considering the effect of biological factors (e.g., presence of non-evolved infertile eggs) and of constraints on trait values (e.g., when hatching cannot become perfectly synchronous). We ran simulations for 1,000 generations; average gene values stabilized after approximately 100 generations. To quantify gene evolution, we obtained the average gene value of each evolving gene in the parent population of 400 individuals, every tenth generation for 200 generations near the end of the simulation (750-940). The average cannibalism delay gene value corresponds directly to the average number of time units offspring wait before eating an egg. Similarly, the average trophic egg gene value is equal to the average number of trophic eggs produced per clutch. We analyzed the effect of factors on gene evolution with t-tests or one- or two-way ANOVAs, as appropriate, in full factorial designs with Tukey post-hoc tests to determine differences between groups. 95% confidence intervals of estimates are presented as: estimate (lower bound, upper bound). The statistics are conducted based on averages from five replicates for each parameter set - that is, sample size was not artificially large.

Computer experiments

Experiments were conducted to test the following predictions, which are derived from the analytical model. We refer to selection and evolution in the experiments and results to describe processes that are analogous to organic evolution, but generated within a search algorithm.

Relative benefit of egg eating (RBE) thresholds

The RBE thresholds for mothers and offspring to favour sibling cannibalism (1 00% and

50%, respectively, when r = 0.5) should predict evolution of the three traits when each evolves separately (Table 3.2). (Relatedness should also influence trait evolution (equation l), but was not manipulated in our models. With single mating in randomly-drawn pairs, relatedness within clutches is 0.5 (see Grafen 1991).) The relative benefits and absolute survival probabilities are given in Appendix I. We examined the evolution of traits separately, when (1) cannibalism delay was permitted to evolve, but hatching was fixed at complete asynchrony and trophic eggs were prevented (set 1, Table 3.2); or (2) hatching synchrony (set 2, Table 3.2) or (3) trophic egg laying (set 3, Table 3.2) could evolve, but the other maternal tactic was fixed and offspring were fixed for extreme cannibalism. A corollary hypothesis is that absolute survival and benefit of cannibalism should have no impact on the evolution of sibling cannibalism traits; only relative benefit of egg eating should matter. We tested this hypothesis by facilitating evolution of the cannibalism delay gene at 50% relative benefits, expressed as 10: 15 or 66:99 survival probability (set 4, Table 3.2).

Game aspects

The above simulations effectively assume that offspring or mothers are constrained in their possible evolved responses to sibling cannibalism. When all traits co-evolve, the game between mothers and offspring or between siblings may affect trait evolution. We compared gene evolution when there was multiple trait co-evolution (set 5, Table 3.2) to simulations where only one trait evolved.

Non-evolved infertility

The presence of happenstance infertile eggs (hereinafter called infertile; sets 7 and 8, Table 3.2) should create increased selection for cannibalism in offspring, if RBE > 0. The predicted maternal response is not clear: mothers may try to combat increased cannibalism among offspring or may be selected to help offspring eat infertile eggs.

Time extension

If offspring can wait until all viable offspring hatch before attacking eggs, they may be selected to delay cannibalism to avoid eating viable siblings. We therefore manipulated a parameter referred to as 'time extension'. With an extension (set 5, Table 3.2), offspring could wait until hatching was over before cannibalism. Without an extension (set 6, Table 3.2), cannibalism was not permitted after hatching ended, so a cannibalizing offspring had to risk consuming viable siblings. This parameter effectively permits offspring discrimination between infertile and viable eggs; for this reason, we examined it in combination with the above parameter of infertile eggs. We next present and discuss results of each simulation set, before drawing more general conclusions on parent-offspring conflict, sibling rivalry, maternal caution and the strength of selection. To assist the reader in interpreting results, we draw attention to the existence of multiple equivalent trait combinations (Fig. 3.1). For example, there are many combinations of cannibalism delay and hatch synchrony behaviours that achieve the outcome of no sibling cannibalism.

3.3 Results and discussion

Relative benefit of egg eating

For comparison with the frequency-independent predictions from the previously discussed analytical model, we considered simulation sets where only one trait was permitted to evolve (sets 2-3; Table 3.2). These simulations and the frequency-independent model effectively ask the same question: given a particular offspring (or maternal) behaviour and level of benefits from egg eating, what is the optimal behaviour for mothers (or offspring)? In the next section, we compare these simulations to simulations where all three traits evolved to examine the effect of co- evolution. As the frequency-independent analytical model predicts, the relative benefit of eating siblings affected the evolution of sibling cannibalism and of maternal attempts to limit (via hatching synchrony) or facilitate (via hatching asynchrony or trophic egg laying) sibling cannibalism (Fig. 3.2, 3.3). In the first one-trait simulation (set 2), we permitted sibling cannibalism to evolve, but made hatching maximally asynchronous (so that offspring risked eating viable siblings) and prevented trophic egg laying. The cannibalism delay mean phenotype decreased as RBE increased from 0% to 75% (Fig. 3.2), when it reached a minimal value. As an example of gene evolution over time, the cannibalism delay gene had an intermediate average value at the beginning of simulations following random generation of individuals (Fig. 3.3), but quickly evolved, with the direction of change dependent on the direction of selection pressure. Cannibalism delay began to decrease before the 50% threshold predicted by the inclusive fitness model (Fig. 3.2). In fact, the delay gene experienced its largest decline, from 11.6 to 7.2 time units, when RBE increased from 30% to 40%. There are at least two possible explanations for the trend. First, at RBE values near 50%, fitness is similar for individuals who eat or avoid eating siblings. The decrease in average delay at 40% RBE (Fig. 3.2) may represent weakened selection against consumed siblings, as opposed to positive selection for increased cannibalism. In support of this hypothesis, the LS mean delay gene value at 40% RBE was 7.2 (7.1, 7.4), which is close to the value generated from randomly created individuals in the initial generation (7.5). Second, the decrease in the threshold for cannibalism may be evidence of sibling rivalry. When RBE < 50%, an individual maximizes its inclusive fitness by allowing full siblings to live. If a second hatched sibling has an allele for cannibalism, however, the 'merciful' offspring does not reap the kin selection gains of mercy, and does not gain the nutritional benefits from cannibalism. It therefore loses fitness compared to a more cannibalistic sibling. In a gene pool containing a variety of cannibalism delays, alleles will continually be paired with, and tested against, others in the population. Sibling rivalry may drive evolution for higher cannibalistic tendency than would otherwise be optimal. For the second one-trait simulation, we examined evolution of the synchrony gene by minimizing cannibalism delay (i.e., offspring readily ate eggs) and preventing trophic egg laying (set 2, Table 3.2). Hatching became increasingly asynchronous as the RBE increased from 90% to 150% (Fig. 3.2). The first statistically detectable increase occurred as RBE increased from 90% to 1 lo%, a range encompassing the predicted 100% threshold for mothers to favour cannibalism among offspring. The largest increases in asynchrony gene value per increment in RBE occurred from 90% to 100% RBE (3.9 to 7.8). Two insights can be gained from this set of simulations. First, there was evidence of parent-offspring conflict and cooperation. At RBE values below loo%, mothers appeared to limit sibling cannibalism with more synchronous hatching. When RBE > loo%, mothers began to facilitate cannibalism with asynchronous hatching. Second, the strength of selection may have affected gene evolution. Asynchrony increased slightly at 90% RBE, before the predicted 100% threshold (Fig. 3.2), possibly because selection against asynchrony weakened near the threshold. Similarly, in the simulation where only trophic egg laying evolved, significantly more trophic eggs were produced at 150% RBE than 1 lo%, because the magnitude of gain from producing more trophic eggs was greater. We allowed the third gene, trophic egg laying, to evolve when offspring readily ate eggs and hatching was completely synchronous, which prevented mothers from using hatching asynchrony to facilitate cannibalism (set 3, Table 3.2). Trophic egg laying ensures that offspring can eat eggs, and thus should not occur until the RBE equals or exceeds 100%. The mean number of trophic eggs per clutch (equivalent to the mean gene value) increased from 0%-150%, with the first detectable increase as RBE increased from 50-75% (Fig. 3.2). This increased trophic egg laying when RBE < 100% was not predicted by the inclusive fitness model, and may be explained by decreased strength of selection against trophic egg laying as RBE neared 100%. Increases in trophic egg production became larger after RBE passed the 100% threshold: the largest increases were from 110% to 125% (4.1 to 5.5) and 125% to 150% (5.5 to 7.0). The relative benefit of egg eating is the increase in survival that cannibals receive compared to non-cannibalistic individuals. We tested the effect of absolute survival by allowing cannibalism to evolve (set 4, Table 3.2) at 50% RBE, with 10% or 66% probability of survival of non-cannibals. There was no effect of the absolute probability of survival on the cannibalism delay gene (F,, 8 = 0.26, P > 0.62), confirming the analytical model's prediction that the relative increase in survival is the important parameter. This result implies that we can expect similar sibling cannibalism behaviour in natural systems where the relative benefit that cannibalism provides is similar. The absolute harshness of the environment does not matter.

Trait co-evolution

When all three traits were allowed to evolve simultaneously, cannibalism delays were significantly shorter than when single traits evolved (6.2 vs. 4.0 time units, F,, 56= 97.5, P <

0.0001). The effect depended strongly on RBE (Fig. 3.5A; interaction term: F6,56 = 10.5, P < 0.0001). There are two possible interpretations of this result. First, in the co-evolving simulations, mothers increased hatching synchrony at RBE < 100% (Fig. 3.5B). The reduced opportunity for cannibalism may have created selection on offspring to reduce cannibalism delay, in order to capitalize on rare cannibalism opportunities. In effect, maternal attempts to limit cannibalism among offspring may intensify selection for cannibalistic tendencies. Second, in the one-trait simulations, offspring were forced to respond to the risk of eating viable siblings. In the co- evolving simulations, mothers limited the opportunity for sibling cannibalism - effectively weakening selection against sibling-eating individuals. The latter explanation is supported by the results: the strongest effects of constrained maternal traits are seen at 25% RBE, a level where both mothers and offspring disfavour cannibalism. Game aspects significantly affected the evolution of hatching asynchrony compared to the situation where only asynchrony evolved, with more asynchronous hatching when all traits co- evolved (5.4 vs. 6.4, F,,@= 7.3, P = 0.009; mean difference 1.0 (0.3, 1.7)). The interaction between constraints and benefits was also significant (Fig. 3.5B; F7,64 = 7.9, P < 0.0001). The maternal response is easily understood. When only asynchrony evolved, the gene value decreased at 0% benefits compared to the co-evolutionary simulation, because when cannibalism delay also evolved, delays quickly lengthened, reducing selection pressure on mothers to prevent sibling cannibalism. On the other hand, asynchrony was significantly higher at 150% benefits when only asynchrony evolved, compared to when all three genes evolved, because when trophic egg laying also evolved mothers used both tactics to facilitate egg eating. When trophic egg laying was prevented, mothers increased asynchrony to allow the same level of egg eating through increased cannibalism of viable offspring. More trophic eggs were produced in the one-gene simulation compared to when all three genes evolved (3.3 vs. 2.9, F,, 72 = 37.1, P < 0.0001; mean difference 0.4 (0.3,0.5)). The interaction between constraints and benefits was significant (Fig. 3.X; F8,72 = 46.6, P < 0.0001). Trophic egg laying increased significantly when it was the only evolving trait at 125% and 150% RBE, for a similar reason as hatching asynchrony: when it was the only trait available for facilitating cannibalism, levels increased. In summary, when all three genes were permitted to evolve simultaneously, mothers used both hatching synchrony and trophic egg laying to facilitate egg eating. When the effects of both traits were combined and when RBE > loo%, approximately half of all eggs laid were eaten - an optimal solution predicted by the analytical model. In this way, adding a game component to the simulations increased cannibalistic tendencies among offspring at lower levels of RBE. At these low levels, mothers responded to the change in offspring behaviour to limit sibling cannibalism. However, at an RBE > loo%, when mothers attempt to facilitate cannibalism, adding a game component did not change the result, and the optimality inclusive fitness model was a good predictor of maternal behaviour.

Non-evolved infertility

The presence of infertile eggs affected the evolution of all three traits. Offspring did not wait as long to eat eggs when happenstance infertile eggs were present (3.2 vs. 4.3, Table 3.3; mean difference 1.1 time units (0.80, 1.4)). The difference depended on the level of RBE (Fig. 3.4A; Table 3.3), such that infertile eggs increased cannibalistic tendencies among offspring only when RBE was below the 50% threshold for cannibalizing full sibs (significant effect at 10% and 25% levels of RBE only; Fig. 3.4A). At RBE values above the threshold, the delay until cannibalism was minimal in the presence and absence of infertile eggs. We can therefore infer that: (1) parent-offspring conflict over sibling cannibalism occurred, because mothers would gain more offspring per clutch if offspring waited until viable siblings had hatched before eating infertile eggs; (2) sibling rivalry was an important selective factor in the evolution of cannibalism delay. Sibling rivalry is evident because, using inclusive fitness reasoning, cannibal offspring should wait until viable siblings hatch and then consume infertile eggs; the only reason not to wait is so that cannibals can eat infertile eggs before siblings eat the eggs. We conclude that non-evolved infertility enhances parent-offspring conflict and increases cannibalistic tendencies among offspring. Hatching was significantly more synchronous in the presence of infertile eggs, as mothers tried to limit sibling cannibalism (6.7 vs. 6.2; Table 3.3; mean difference 0.5 (0.3, 0.7)). The response in hatch synchrony is due to a change in offspring cannibalism behaviour caused by the presence of infertile eggs: offspring were more cannibalistic when infertile eggs were present at levels of relative benefits where mothers did not favour sibling cannibalism, and mothers tried to limit the cannibalism of viable offspring. Mothers produced significantly fewer trophic eggs in the presence of infertile eggs (2.7 vs. 2.1; Table 3.3; mean difference 0.6 eggs (0.5-0.7)). The result is easily understood: with two infertile eggs per clutch, 14 viable offspring remain. Each offspring can eat only one egg, leaving 12 offspring without an egg to eat. At most, six offspring should be trophic, lower than the maximum of eight when all offspring are potentially viable. Our results suggest that non-evolved infertility enhances parent-offspring conflict over cannibalism. The increased conflict is not due to the presence of infertile eggs, per se, because mothers should prefer that offspring eat those eggs (if there is any nutritional benefit to be gained), rather than waste the resource. Rather, the increased conflict stems from offspring response to infertile eggs: cannibalistic tendencies increase because there is competition among siblings over eating the infertile eggs. Driven by sibling rivalry and the cannibal's dilemma, offspring risk eating viable siblings, even when it is not in their inclusive fitness interests to do so (i.e., at RBE of 10% and 25%; Fig. 3.4A) and certainly not in their mother's interests. Mothers respond to the increased cannibalism among their offspring by using tactics that limit cannibalism or egg eating (increased hatching synchrony; reduced trophic egg laying).

Time extension

There was generally no effect of allowing offspring to avoid eating viable siblings by waiting until all viable eggs hatched (Table 3.3). The only exception is a significant interaction between time extension and RBE for the trophic egg gene (Table 3.3). A time extension increased trophic egg production at an RBE of 1lo%, and no other benefit level (3.4 vs. 3.0, mean difference 0.4 (0.2, 0.6); Table 3.3). A possible explanation for the result is that, for simulations runs with a time extension, mothers happened to use the trophic egg tactic more than the hatching asynchrony tactic. The tactics achieve the same result of allowing egg eating, so it may be a chance result - and in fact, the P-value for the interaction was 0.04, whereas P-values for most other significant effects were < 0.0001. It appears that the time extension variable had no effect on the evolution of cannibalism delay because sibling rivalry was an important selective pressure. Offspring were selected to cannibalize siblings quickly, before other siblings ate them; the option of waiting until viable offspring hatched before eating eggs was never optimal.

3.4 Conclusions

The frequency-independent analysis predicts RBE thresholds for optimal behaviours in mothers and offspring. The implicit assumption is that each party responds to a fixed behaviour in the other; the model asks: given that offspring cannibalize and benefits are x%, what should mothers do? Results from our GA-based simulations confirm the relevance of thresholds predicted by frequency-independent inclusive fitness theory, but also reveal further predictions because they incorporate the game theoretic aspects of sibling cannibalism. We discuss the evidence for these two games, the predicted optimal behaviours in parents and offspring (especially where these differ from the analytical predictions), and compare our qualitative predictions with natural systems. As in the frequency-independent analytical model, the simulation results describe optimal behaviours for mothers and offspring. Offspring should be very sensitive to the RBE in cannibalizing siblings (Fig. 3.2); in our model, the average cannibalism delay was statistically distinct between even small increments in RBE. If infertile eggs are present, offspring should be more cannibalistic at RBE levels as low as 10% (Fig. 3.4A). Mothers should also respond to RE3E in their response to offspring cannibalism. They should limit cannibalism (using a tactic such as hatching synchrony) when RE3E < 100% (Fig. 3.2). When RBE > loo%, mothers should help offspring to eat eggs; and in this model, hatching asynchrony and trophic egg laying were interchangeable tactics used to accomplish this end. Our results also suggest that mothers should err on the side of caution in promoting egg eating. The analytical model predicts that, when RE3E > loo%, mothers benefit if half of their offspring consume the rest. In our model, mothers promoted the cannibalism of nearly half of their clutches, but never equal to or exceeding half (Fig. 3.2). For example, the mean number of trophic eggs produced at an RBE of 150%, the highest tested, ranged from 6.3 to 7.3 in clutches of 16 offspring (when hatching asynchrony could not be used to promote egg eating). When offspring consume a maximum of one egg, the loss in fitness from overproducing trophic eggs is more severe than from underproduction. With 150% RBE, from 30% survival of non-cannibals and 75% survival of egg eaters, a mother producing 7 trophic eggs in a clutch of 16 gains (7 * .75) + (2 * .3) = 5.85 surviving offspring, while a mother producing 9 trophic eggs gains (7 * .75)

= 5.25 offspring. It is best, then, to produce 7 or 8 trophic eggs, rather than 8 or 9. When we allowed traits to co-evolve, results revealed two interacting evolutionary games occurring over sibling cannibalism: the competition among siblings to consume eggs, and the game between parents and offspring when there is parent-offspring conflict over cannibalism. Several results suggest that sibling rivalry over sibling cannibalism is a selective pressure influencing the co-evolved game outcome. For example, offspring began eating siblings before an inclusive fitness analysis predicted they should. Furthermore, when infertile eggs were present, cannibalism delays decreased (Fig. 3.4A) as offspring risked consuming viable eggs - even when it was possible to wait until viable siblings hatched. An inclusive fitness analysis would predict that offspring should, optimally, wait until viable siblings hatched before consuming eggs, but sibling rivalry caused offspring to eat eggs quickly before other cannibals did so. The existence of parent-offspring conflict over sibling cannibalism was evident from several results. When 50% < RBE < loo%, where parent-offspring conflict is expected, offspring tried to eat viable siblings to their mothers' detriment, and mothers tried to prevent cannibalism by making hatching synchronous (Fig. 3.2). The co-evolved outcome, in terms of number of viable offspring cannibalized per clutch, depended on two things: who had power to control sibling cannibalism, and the strength of selection on both parties. Simulation sets 1 and 2, where only cannibalism delay or hatching synchrony could evolve, are akin to natural systems where offspring or parents are constrained in trait evolution. For example, in predatory insects, offspring may experience strong selection for early predatory ability; selection may not be able to separate the behaviours of attacking prey or siblings soon after emergence, if the behaviours are genetically correlated. Hatching synchrony can also be constrained: mothers may be unable to make hatching perfectly synchronous, simply because egg formation and oviposition are not perfectly precise processes. The presence of constraints gives control of sibling cannibalism to the unconstrained party. Thus, in natural systems, the observation of sibling cannibalism does not imply that parents permit it (in an evolutionary sense); they may simply be unable to prevent it. The importance of the strength of selection was clearly seen in the GA model. In the analytical analysis, strength of selection is not considered: individuals are predicted to select the optimal tactic based on a simply inequality, even if it is only slightly better than the next-best option. In our results, evolved outcomes depended on the magnitude of selective advantage. For example, mothers produced significantly more trophic eggs at 125% and 150% than at 110% RBE (Fig. 3.2) - even though 1 10% is above the cannibalism facilitation threshold for mothers. In a second example, when cannibalism and hatching synchrony co-evolved, more viable cannibalized offspring occurred as offspring became more cannibalistic (RBE 50-go%), even though this worked against maternal interests and mothers could have prevented cannibalism through hatch synchrony. We suggest that mothers did not do so because, at levels of RBE approaching loo%, there was little fitness difference for mothers between facilitating or limiting cannibalism; but for offspring there was a large fitness gain to be had from eating a sibling. It is not clear what these results imply for the evolution of sibling cannibalism in natural systems, where natural selection in the long term may allow slightly optimal strategies to dominate (see Roitberg 1990; Roitberg and Mange1 1997). On the other hand, Roitberg and Mangel (1997) suggest that when fitness functions are fairly flat, variation in traits should persist. Thus, persistence may depend on the exact shape of the fitness function and strength of selection. We may predict that variation is more likely to persist in this case, where the fitness function appears flat, than where fitness functions appear steep based on the results we obtained. A final point concerns the effect of relatedness in the model. The coefficient of relatedness expresses the ratio of the difference between the frequency of a gene of concern in the recipient of an action and population gene frequency, compared to the difference between the gene frequency of the actor and population gene frequency (Grafen 1991). This ratio is always 0.5 between full siblings. However, the magnitude of the difference between both actors and recipients and the population gene frequency will decrease as genetic variation for the gene decreases. For a fixed gene, relatedness must be zero or undefined. Thus, as one allele comes to dominate the locus, efforts by bearers of the allele to help siblings have less effect in increasing the allele in the population - the benefit to the allele that is the basis of kin selection. In our model, this might mean weaker selection to avoid eating siblings - the action that promotes the spread of copies of long cannibalism delay alleles in the population - towards the end of simulations as gene values converged on one or a few alleles. The effect would only be relevant in simulations where offspring should be selected to avoid eating siblings (RBE < 50%), because the importance to alleles of promoting copies of themselves in others will be greatest in these cases.

Trophic egg laying

One goal of this study was to investigate the evolution of trophic egg laying as a tactic for facilitating egg eating. That is, when cannibalism is in mothers' interests (when RBE > loo%), mothers do best by ensuring that egg eating occurs; for example, by neglecting to fertilize an egg. This may benefit mothers even if trophic eggs cost the same to produce as viable eggs, as assumed in our model. It may seem counter-intuitive that a mother's best option may be to stop an egg's development altogether: why would she not, instead, increase hatching asynchrony so that an egg may continue development if it happens to escape cannibalism? In our simulations where both hatching asynchrony and trophic egg laying evolved, mothers used both tactics to provide eggs to offspring - it was not better to increase hatching asynchrony rather than producing trophic eggs. We suggest that mothers should produce trophic eggs in order to guarantee that egg eating occurs; if hatching asynchrony alone is used, there is some chance that delayed-hatch eggs will hatch before cannibalism occurs. Our results also suggest that, even if animals produce non-evolved infertile eggs, they may be selected to produce additional trophic eggs when Rl3E is high. If infertile eggs are initially produced by an uncontrollable process, however, it may be difficult for mothers to evolve production of additional trophic eggs (Crespi 1992). Whether the occurrence of infertile eggs inhibits the evolution of trophic eggs across systems is a question for empirical investigation. The feasibility of an adaptation whereby mothers feed some eggs to others has been treated skeptically by some authors (e.g., Dixon 2000). However, our model shows that use of trophic eggs as a provisioning tactic is reasonable. In Chapter 4,l suggest that the ladybird Harmonia axyridis produces trophic eggs to provision offspring, and demonstrate that mothers adjust trophic egg production to predicted starvation risk for offspring. In this species, trophic eggs appear identical to viable eggs to the eye, and their weight is not different from viable eggs (unpublished data). Previous work shows that the Rl3E in H. axyridis can be very high (reviewed in 'Methods'). Thus, evidence is accumulating that some portion of the apparently infertile eggs produced by this animal has evolved to provision offspring. A diverse assortment of other animals produce infertile eggs that are eaten by other offspring (reviewed in Chapter 2). Whether they are truly adaptively produced by mothers is an untested question in most systems.

Future directions

It would be of interest to test several additional parameters predicted to affect co- evolutionary outcomes regarding sibling cannibalism. First, the frequency-independent inclusive fitness model presented in this paper analyzes a painvise interaction between cannibal and victim siblings. Thus, a simulation where clutch size = 2 should exactly match the frequency- independent predictions. However, most animals that exhibit sibling cannibalism lay eggs in larger clutches, and as our results suggest, this may influence the evolution of cannibalistic tendencies through sibling rivalry. Second, the frequency-independent model predicts two variables that drive evolution of sibling cannibalism: RBE and within-clutch relatedness. Polygamous mating systems would create lower relatedness, expected to enhance selection for sibling cannibalism, while assortative mating (by any phenotypic trait) would increase relatedness and reduce selection for sibling cannibalism. Finally, in natural systems offspring may be selected to discriminate between viable and infertile eggs. It is, however, hard to predict which they should prefer eating. Perhaps offspring should avoid eating viable siblings with delayed hatch in

favour of eating infertile eggs - or perhaps offspring should consume viable siblings first to prevent their hatch! Our GA approach allows testing of a variety of ecological parameters predicted to be relevant to the evolution of sibling cannibalism.

3.5 Appendix

Details of genetic algorithm and simulations

Model organisms

In the model, haploid individuals are represented as 12-byte chromosomes. We use a gray coding translation of genotype to phenotype that allows the GA to find adjacent solutions more easily (Forrest 1993). Each chromosome has three four-byte genes (hatching asynchrony, proportion of trophic eggs per clutch, delay until cannibalism) with 16 possible alleles.

Simulations

In the initialization of simulations, a population of 400 chromosomes is created with randomly drawn byte sequences for the trophic egg and cannibalism delay genes. For the hatch synchrony gene, allele values from one to nine are randomly drawn because values of nine and above create the same number of opportunities for cannibalism (Table 3.1). The GA introduces new alleles and new allele combinations from which to search for optimal strategy sets through mutation- and crossover-like processes during reproduction. In our model, 200 pairs of individuals are randomly selected to mate and generate 16 offspring with a 1% chance of mutation at each byte locus. There is a 10% chance of crossover for each pair of offspring created. When crossover occurs, genotypes of mother and father are spliced at a randomly drawn location to create offspring. When there is no crossover, half of the offspring are clones of the mother and half of the father. Individuals do not have a sex and can be drawn as a mother or father in the next generation. Each individual has three states: alive or dead, the time period of its hatch, and the number of eggs it has consumed (0 or 1). Individuals may be killed in several ways during intra-clutch interactions: by being randomly selected as an infertile or trophic egg; by being cannibalized by an earlier hatching sibling; or by starvation, with risk weighted by 'fed' state.

Relative benefit of egg-eating

The RBE levels tested in simulations (shown here as probability of survival of a non- canniba1:cannibal) were 0% (75:75), 10% (70:77), 25% (60:75), 30% (70:91), 40% (70:98), 50% (60:90), 75% (40:70), 90% (50:95), 100% (49:98), 125% (40:90), and 150% (30:75). Not all simulations tested all values. 3.6 Literature cited

Banks, C. J. 1956. Observations on the behaviour and mortality in Coccinellidae before dispersal from the egg shells. Proceedings of the Royal Entomology Society of London (A) 3 1:56- 60.

Breden, F., and M. J. Wade. 1989. Selection within and between kin groups of the imported willow leaf beetle. American Naturalist 134:35-50.

Brown, H. D. 1972. The behaviour of newly hatched coccinellid larvae (Coleoptera: Coccinellidae). Journal of the Entomological Society of South Africa 35:149-157.

Crespi, B. J. 1992. Cannibalism and trophic eggs in subsocial and eusocial insects. Pages 176-2 13 in M. A. Elgar, and B. J. Crespi, eds. Cannibalism: ecology and evolution among diverse taxa. Oxford University Press, Oxford.

Dickins, D. W., and R. A. Clark. 1987. Games theory and siblicide in the kittiwake gull, Rissa tridactyla. Journal of Theoretical Biology 125:301-305.

Dixon, A. F. G. 2000. Insect predator-prey dynamics. Cambridge University Press, London.

Eickwort, K. R. 1973. Cannibalism and kin selection in Labidomera clivicollis (Coleoptera: Chrysomelidae). American Naturalist 107:452-453.

Forbes, L. S. 1993. Avian brood reduction and parent-offspring "conflict". American Naturalist 142:82-117.

Forrest, S. 1993. Genetic algorithms: principles of natural selection applied to computation. Science 261:872-878.

Godfray, H. C. J. 1995. Evolutionary theory of parent-offspring conflict. Nature 376: 133-138.

Godfray, H. C. J., and A. B. Harper. 1990. The evolution of brood reduction by siblicide in birds. Journal of Theoretical Biology 145: 163-175.

Goldberg, D. E. 1989, Genetic algorithms in search, optimization, and machine learning. Addison-Wesley, Reading, Massachusetts.

Grafen, A. 1991. Modelling in behavioural ecology. Pages 5-3 1 in J. R. Krebs, and N. B. Davies, eds. Behavioural Ecology, 3rded. Blackwell Science, Oxford.

Hamilton, W. D. 1964. The genetical evolution of social behaviour. I, 11. Journal of Theoretical Biology 7: 1-52.

-. 1967. Extraordinary sex ratios. Science 156:477-488.

Hodek, I., and A. Honek. 1996. Ecology of Coccinellidae. Kluwer Academic Publishers, Dordrecht, The Netherlands. Kawai, A. 1978. Sibling cannibalism in the first instar larvae of Harmonia axyridis Pallas (Coleoptera: Coccinellidae). Kontyu 46: 14-19.

Majerus, M. E. N. 1994. Ladybirds. HarperCollins, London.

McNamara, J. M., C. E. Gasson, and A. I. Houston. 1999. Incorporating rules for responding into evolutionary games. Nature 401 :368-371.

Mock, D. W., and L. S. Forbes. 1995. The evolution of parental optimism. Trends in Ecology and Evolution 10: 130-134.

Mock, D. W., and G. A. Parker. 1997. The evolution of sibling rivalry. Oxford University Press, Oxford.

-. 1998. Siblicide, family conflict and the evolutionary limits of selfishness. Animal Behaviour 56:l-10.

O'Connor, R. J. 1978. Brood reduction in birds: selection for fratricide, infanticide and suicide. Animal Behaviour 26:79-96.

Osawa, N. 1992. Sibling cannibalism in the ladybird beetle Harmonia axyridis: fitness consequences for mother and offspring. Researches in Population Ecology 34:45-55.

Parker, G. A., and D. W. Mock. 1987. Parent-offspring conflict over clutch size. Evolutionary Ecology 1:161-174.

Pfennig, D. W. 1997. Kinship and cannibalism. Bioscience 47:667-775.

Pienkowski, R. L. 1965. The incidence and effect of egg cannibalism in first instar Coleomegilla maculata lengi (Coleoptera: Coccinellidae). Annals of the Entomological Society of America 58:150-153.

Reinhold, K., J. Kurtz, and L. Engqvist. 2002. Cryptic male choice: sperm allocation strategies when female quality varies. Journal of Evolutionary Biology 15:201-209.

Robertson, I. C., W. G. Robertson, and B. D. Roitberg. 1998. A model of mutual tolerance and the origin of communal associations between unrelated females. Journal of Insect Behaviour 11:265-286.

Rodriguez-GironCs, M. A. 1996. Siblicide: the evolutionary blackmail. American Naturalist 148:lOl-122.

Roitberg, B.D.. 1990. Optimistic and pessimistic fruit flies: evaluating fitness consequences of estimation errors. Behaviour 1 14:65-82.

Roitberg, B., and M. Mangel. 1993. Parent offspring conflict in butterflies and wasps. American Naturalist 142:443-456. Figure 3-1 The average number of offspring cannibalized per clutch with various combinations of cannibalism delay and hatching synchrony values. Cannibalism delay alleles correspond to the number of time units offspring wait before eating an egg. Phenotypes for hatching synchrony alleles are shown in Table 3.1. The shaded area indicates combinations of traits that yield no sibling cannibalism. The number of cannibalized offspring increases diagonally to a maximum of eight (in a clutch of 16), as indicated by the arrow. Offspring cannibalism delay allele Figure 3-2 The evolution of cannibalism delay, hatching asynchrony and trophic egg laying, when other genes were prevented from evolving (Table 3.2). Least square means are from a one- way ANOVA with the factor 'relative benefit of egg eating', which was a significant factor for all three genes; P < 0.0001 in each test. Means with different letters are statistically different. The bars represent standard error. The y-axis is reduced for the trophic egg laying gene because average allele value never exceeded 8. Cannibalism delay f\b c

P\d , e

\ k-.. ---.f z.,, g g i? g 5---x-+- ---*

Relative benefit of egg eating Figure 3-3 Evolution of the cannibalism delay gene over 1000 generations, for different levels of relative benefit of egg-eating, when the evolution of other traits was prevented (Table 3.2). Results from one representative simulation are shown for each level of relative benefit. Mean gene value was taken for the parental population every tenth generation. With the random generation of individuals in generation zero, initial average gene values were approximately 7.5. Generation (1 0s) Figure 3-4 Effect of infertility on the evolution of three traits. For each figure, least squares mean gene values are given for different levels of relative benefit of egg eating, given by the different lines, in the absence and presence of infertile eggs. (a) Cannibalism delay. The mean difference in gene value in the presence and absence of infertile eggs was 6.7 (4.8,8.5) at 10% relative benefit and 3.8 (1.9, 5.6) at 25% relative benefit. The difference was not significant for any other level of benefits. For clarity, levels above 75% are not shown. (b) Hatching synchrony. The presencetabsence of infertile eggs affected the mean difference in gene values at 10% benefits (2.0 (0.9, 3.2)). The difference was not significant at any other level of relative benefit. For clarity, not all of the tested levels are shown. (c) Trophic egg-laying. Infertile eggs significantly decreased trophic egg production at the following levels of relative benefit, with the mean difference in gene value given: 0.9 (0.5, 1.3) at 75%; 0.8 (0.4, 1.2) at 90%; 1.1 (0.7, 1.5) at 100%; 0.8 (0.4, 1.2) at 1 10%; 0.9 (0.5, 1.3) at 125%; 0.9 (0.5, 1.3) at 150%. The y-axis range is reduced to show differences between the curves. For clarity, not all levels of relative benefit are shown. 1 Infertile eggs absent Infertile eggs present A- - Infertile eggs absent Infertile eggs present 0' I Infertile eggs absent Infertile eggs present Figure 3-5 Trait evolution when other traits are constrained. Levels of relative benefit of egg eating are given by the different lines. (a) The evolution of cannibalism delay in offspring when maternal traits affecting cannibalism are free to evolve or constrained (set 1, Table 3.2). Least squares mean gene values are obtained from an ANOVA with the factors relative benefit, presence of constraints, and the interaction term. Constraining traits lead to delayed cannibalism; see discussion in text. Significant increases in delay times occurred at relative benefit levels of 25% (mean difference 6.6 time units (4.5, 8.7)) and 50% (2.2 time units (0.1,4.3)). The differences were not significant at other levels of relative benefit. (b) The effect of constraints on cannibalism delay and trophic egg laying on the evolution of hatching synchrony (set 2, Table 3.2). The difference in gene value when other traits were constrained, or free to evolve, was significant at 0% relative benefit (mean difference 5.2 (1.6, 8.8)) and 150% relative benefit (mean difference 3.8 (0.2, 7.4)). (c) The effect of constrained traits (set 3, Table 3.2) on the evolution of trophic egg laying. The mean difference between the trophic egg gene when traits were constrained, or free to evolve, was 0.7 (0.1, 1.4) at 75% relative benefit; 1.6 (0.9,2.2) at 125%; and 3.2 (2.6, 3.9) at 150%. Note that the y-axis scale for this trait differs from that in graphs of the other genes. I Maternal traits not constrained Maternal traits constrained All traits permitted Trophic egg laying to evolve and cannibalism delay constrained All traits free Synchrony and to evolve cannibalism delay constrained Table 3-1 The range in hatching synchrony phenotypes, in terms of number of offspring hatching in each time period for each allele, that were tested in simulations. Mothers with each non-binary allele value produce clutches with the temporal hatching pattern indicated in the table. condition that offspring eat, at most, one egg. With a clutch size of 16, a maximum of eight cannibalism events can occur. Alleles closer to 16 allow maximal cannibalism at a wider range of offspring cannibalism delay phenotypes. Table 3-2 Conditions for each set of simulations of the evolution of sibling cannibalism and maternal traits that influence cannibalism: the range of relative benefits that cannibal offspring receive; whether infertile eggs are present or absent; whether cannibalism is time-limited; and the presence of constraints, if any, on the evolution of any of the tested traits Constrained evolution of.. . Simulation Relative Non- Can Cannibalism Hatching Trophic set benefit of evolved offspring delay asynchrony egg laying egg eating infertile wait for eggs viable present? eggs to hatch? No Yes Asynchronous Prevented

2 0-1 50 No Yes Minimal Prevented 3 0-1 50 No Yes Minimal Synchronous 4 50%, with No Yes Asynchronous Prevented 10% or 66% survival when no eggs were eaten 5 0-1 50 No Yes 6 0-150 No No 7 0-150 Yes Yes 8 0-150 Yes No Note: Each set consists of five replicated simulations. The values of relative benefit of egg eating tested in the model fall within the range given; see Appendix I. There were two non-evolved infertile eggs per clutch, when they were present. When traits were constrained, the fixed phenotype is listed; blank cells indicate that the trait was permitted to evolve. Table 3-3 The effect of the relative benefit of egg eating, infertile eggs, and a time extension (whereby offspring could wait for viable siblings to hatch before eating eggs) on the evolution of the three genes Cannibalism Hatching Trophic egg delay synchrony laying Source F ratio P F ratio P F ratio P Relative benefit of egg 225.0 <0.0001 71.0 <0.0001 293.8 <0.0001 eating Time extension Relative benefit * Time extension Infertile eggs Relative benefit * Infertile eggs Time extension * Infertile eggs Note: Average gene values for the parental population were taken from gener: ions 750-940, every tenth generation, then averaged across the five replicated simulations for each set. Whole model result: for cannibalism delay, F30, 169 = 75.0, P < 0.0001; for asynchrony, F30, 169 = 24.0, P < 0.0001; for trophic eggs, F30, 169 = 102.2, P < 0.0001. The three-way interaction term was not significant for any gene and was dropped from the models. Chapter 4 Ladybird Mothers Mitigate Offspring Starvation Risk by Laying Trophic Eggs

4.1 Abstract

A large proportion of ladybird eggs are apparently infertile - they are identical in appearance to viable eggs, but do not develop an embryo and are consumed by larvae hatching within the same egg batch. The predicted benefits of egg consumption for larvae are empirically well-supported. An important question, however, remains: are these infertile eggs a maternal strategy to feed offspring (i.e., trophic eggs) or did egg cannibalism evolve secondarily to exploit the infertile eggs already present for non-adaptive reasons (e.g., gamete incompatibility or sperm limitation)? We investigated the former explanation in laboratory experiments with multi-spotted Asian ladybirds (Harmonia axyridis). Female H. axyridis were assigned to low and high resource environments for brief intervals; we predicted that tactics to facilitate egg cannibalism, such as infertile egg production and hatching asynchrony, would be adopted in low food environments where starvation risk for offspring is greater. We conducted two experiments in this manner that provided ladybirds with information about resource levels in different ways, through prey feeding or scent. We also observed female oviposition patterns and tested for infertile egg distributions that departed from random. Females produced 56% more infertile eggs in the low compared to the high food environment; however, hatching synchrony did not change. We consider a potential confound between information and nutrition state unlikely in this experiment because, first, ladybirds are well able to tolerate low food for 24 hours, the duration of trials; second, females were in similar physiological condition when each trial began. Results suggest that female ladybirds use information from the amount of prey they encounter, but not from other cues about prey levels, to manipulate the proportion of trophic eggs produced, but not hatch synchrony, in a manner consistent with the adaptive hypothesis.

4.2 Introduction

Parents in many species face the problem of high starvation risk for offspring. Furthermore, some parents (e.g., most insects) do not interact with their offspring at all after oviposition; hence mothers in this situation are limited to starvation-reduction behaviours that occur at the egg production or deposition stage. In the extreme case, a mother's best option might be to provide food for offspring in the form of eggs themselves (Alexander 1974; Kawai 1978; Polis 1981; 1984; Crespi 1992; Mock and Forbes 1995; Mock and Parker 1997; Pfennig 1997). Mothers should sacrifice some offspring to others when they gain more offspring from the increased survival of the consumers than they lose as the consumed (Chapter 3, Crespi 1992) - thus, when starvation risk is high and eating a sibling provides a large benefit.

One way for mothers to ensure that offspring have a sibling to eat is to produce 'trophic' offspring that serve as a meal (Chapter 2, Alexander 1974; Crespi 1992; Mock and Forbes 1995).

This hypothesis provides a foundation for the study of trophic eggs - ovariole-produced structures, formed to feed offspring, that do not otherwise develop into viable offspring (Crespi

1992) - which are noted across diverse taxa (reviewed in Chapter 2), but rarely treated in an adaptive context (but see Crespi 1992). The hypothesis and definition distinguish trophic eggs from infertile eggs that did not evolve for offspring provisioning, though they may be consumed by offspring. In studying trophic eggs, it is therefore necessary to ask the functional question: is there an adaptive maternal strategy to feed offspring? The alternative hypothesis is that offspring cannibalize infertile eggs produced through some constraint (e.g., sperm limitation); hence, the approach of demonstrating that cannibal offspring benefit from consuming eggs does not distinguish between the two explanations. In some taxa, the adaptive nature of trophic eggs seems clear; for example, when they have a unique morphology, clearly different from viable eggs (e.g., West and Alexander 1963; Henry 1972), or when parents actively feed them to offspring (e.g., MacKaye 1986; Nakahira 1994; Crump 1996; Gobin et al. 1998; Kim and Roland 2000; Heying 2001; Matsuko 2003). In other taxa, trophic eggs are simply unfertilized eggs consumed by offspring in the same egg batch (e.g., Peck and Whitcomb 1970; Valerio 1974; Crespi 1992; Dickinson 1992), as in ladybird beetles (Banks 1956; Pienkowski 1965; Brown 1972; Kawai 1978; Osawa 1989; 1992; Stevens 1992; Majerus 1994). When this is the case, the adaptive nature of trophic eggs must be empirically investigated (see Chapter 2).

A second way for mothers to sacrifice some offspring to others is to manipulate hatching synchrony (O'Connor 1978; Polis 1981; Dickins and Clark 1987; Magrath 1990; Forbes et al. 2002). If cannibalism is directed towards siblings still in the defenceless egg stage, as in many insects (e.g., Crespi 1992; Branquart et al. 1997; Via 1999; Sigsgaard et al. 2002), then mothers can decrease hatching synchrony in low-food environments to allow early-hatching offspring to cannibalize siblings. Lacewing mothers appear to manipulate hatching synchrony in this way (Frechette and Coderre 2000).

One way to determine whether mothers adaptively use these two provisioning strategies - trophic eggs and hatch synchrony adjustment - is to show that they are adjusted to starvation risk, which determines the benefit of sibling cannibalism from a mother's perspective (Chapter 3, Frechette and Coderre 2000). The ladybird beetle Harmonia axyridis is a good candidate for examining flexible use of the behaviours for several reasons. First, females produce many apparently infertile eggs (15.2%, Kawai 1978; 24.5%, Osawa 1992) that are consumed by larvae hatching in the same egg batch. Second, first instar larvae face a high starvation risk due to oviposition sites that are some distance from aphid prey (Majerus 1994) and their poor predatory abilities (Hodek and Honek 1996). Furthermore, egg cannibalism dramatically reduces the risk of starvation in H. axyridis (up to 226%, Osawa 19%; see also Kawai 1978) and other ladybirds (reviewed in Hodek and Honek 1996). Finally, female ladybirds frequently disperse to new sites between bouts of egg laying (Evans 2003) and are thus likely to deposit offspring at sites that vary in resource availability. Therefore, female ladybirds should be selected to use plastic strategies to mitigate offspring starvation risk.

Two possibilities, then, are (1) that the infertile eggs of H. axyridis are trophic offspring adaptively designed to be eaten (Polis 1981, 1984; Mock 1994; Pfennig 1997; Osawa 2003) or (2) that sibling cannibalism behaviour takes advantage of an available food source. We also looked for evidence that females use hatching synchrony to alter the opportunity for sibling cannibalism among offspring (Godfray 1987; Godfray and Harper 1990). Assuming that these are plastic behaviours, we hypothesized that female ladybirds would lay clutches with fewer trophic eggs and more synchronous hatching when food resources for offspring were high, and more trophic eggs and less synchronous hatching in low food environments. We conducted two experiments to test the hypotheses that differed in how ladybirds acquired information about food levels, through direct contact with prey or indirect cues. Female ladybirds used in experiments were in similar physiological condition, minimizing a potential confound between information about resource levels and physiological state. Finally, if the infertile eggs are adaptively produced then deposition of such eggs should optimize their contribution to sibling survival. Thus, we examined the spatial distribution and order of oviposition of infertile eggs, suggesting that a uniform distribution indicates adaptive production (e.g., if it maximizes the chance that cannibals attack an infertile egg first rather than a viable offspring). Results suggest that females use trophic eggs, but not hatching synchrony, as an offspring provisioning strategy. However, the distribution of trophic eggs within clutches and oviposition sequences was not different from random.

4.3 Methods I: Food level experiments

Study system and experimental animals

Harmonia axyridis and other ladybirds produce three types of eggs: apparently infertile eggs, viable eggs and 'inviable' eggs where a larva develops but does not emerge from the egg capsule (Ng 1986). After a period of quiescence (mean: 141 min; 95% CI: 126-155; N = 76; Appendix I. 1) following hatching, larvae consume any unhatched eggs, without discriminating between viable and infertile or inviable eggs (Banks 1956; Pienkowski 1965; Brown 1972; Osawa 1992; Majerus 1994). We do not consider inviable eggs to be potentially trophic eggs because some embryo development occurs. Generally, other unhatched eggs are hatched or consumed by the time a cannibal is ready to eat a second egg. Larvae remain at the egg batch about 24 hours, then disperse in search of aphid prey (Osawa 1989), the food source for juveniles and adults.

Adult H. axyridis were obtained from Applied Bionomics (Victoria, British Columbia, Canada) and pupae were collected from the Simon Fraser University campus (Burnaby, British Columbia, Canada). Females used in experiments were maintained individually in petri dishes and supplied daily with an excess of pea aphids (Acyrthosiphon pisum reared on broad bean, Vicia faba) and a water source. They were mated to haphazardly chosen males from laboratory colonies every ten days.

Food level experiments

To observe behavioural plasticity in trophic egg and hatching synchrony behaviours, we needed to provide ladybirds with information about local food availability. Ladybirds might assess resource levels by (1) using the internal cue of their own food intake (because adults and offspring consume the same prey), andlor (2) using an external cue from other senses, e.g. the scent of aphids (Stubbs 1980; Evans and Dixon 1986) or their honeydew excretions (Carter and Dixon 1984; Evans and Dixon 1986). We conducted two experiments that provided ladybirds with information about food resources through internal and external cues, in a paired design such that females experienced both low and high food treatments in randomized order. The same procedure was employed in preparing females for all experiments. Before trials, females were kept separately in petri dishes and fed a set mass of pea aphids daily for five days (0.03 10 g * 10% if alone; 0.0445 g * 10% if kept with a male for mating), to ensure a similar physiological condition. This amount is approximately as much as an adult female can consume in 24 hours (Soares et al. 2001).

Experimental trials began on the sixth day after controlled feeding. Each female was placed in an empty petri dish (90 x 23 mm) for one hour to emphasize the transition from maintenance to experimental conditions. Females were then transferred to a new petri dish that contained the randomly-assigned treatment (low or high aphid masses), a piece of cotton wick moistened with distilled water, and filter paper as an oviposition surface. During the 24-hour trials, egg batches were removed from petri dishes every three hours and aphid levels were replenished every six hours (in the internal cue experiment, by adding three second or third instar aphids to the low food treatment and a small mass of aphids to the high food treatment; in the external cue experiment, by adding three second or third instar aphids to both treatments). After first trials, females were fed the same aphid mass daily for five days before the second trial with the reverse treatment.

During hatching (c. 80 hours after each trial), we monitored egg batches every ten minutes, recording temperature and removing emerged larvae to prevent egg cannibalism. Larvae were considered emerged when all six legs were free of the egg capsule.

The experiments are summarized in Table 4.1.

Internal cue experiment

In the first of two replications of the experiment, all ladybirds came from the supply company; in the second, 47 were purchased and 23 were collected. In an initial analysis, origin of female did not significantly affect the results and it was dropped from the final models.

We did not use a 'no aphid' treatment to signal a low food environment because preliminary trials indicated that females would not oviposit in the absence of prey. Instead, we used a 'low food' treatment of the smallest mass of aphids necessary to stimulate oviposition (0.0040-0.0070 g in preliminary trials). The 'high food' treatment was over nine times that amount, 0.05 16 g + lo%, 1.7 times the amount of pea aphids a female consumes daily (Table 4.1, Soares et al. 2001).

External cue experiment

Of the 116 females used, 89 were purchased and 27 collected. The experimental chamber was designed to allow ladybirds to smell the aphids of the high food treatment but to prevent contact with aphids. It consisted of two petri dish bottoms (90 x 23 mm) separated by two pieces of nylon mesh secured to the bottom dish with a rubber band (Fig. 4.1; Table 4.1). To minimize absorbed odors, the mesh had been washed with soap and water, rinsed with 70% ethanol, and rinsed with distilled water. The upper dish was secured to the lower with cellulose transparent tape. The lower compartment contained a 1 dram vial of distilled water with a bean leaf pair inserted through Parafilm03 (Pechiney Plastic Packaging, Wisconsin, USA) into the vial. The leaf pairs were cut with a razor and inserted through the Parafilm@ under distilled water within two hours prior to the experiment, to minimize distress odors released by the plant (Petrescu et al. 2001). For the high food treatment, aphids (0.0806 g h 10%) were placed in the lower compartment, whereas the low food treatment was no aphids (Table 4.1). Ladybirds were placed in the upper compartment with a small mass of aphids (0.0040-0.0070 g) to stimulate oviposition.

Statistical analyses

We excluded egg batches from analysis when: (1) they were laid in the first six hours of a trial, because ladybirds may have been responding to the food levels of maintenance conditions rather than to the new experimental conditions; (2) they had more than 40% non-developing eggs, because H. axyridis females can carry a male-killing bacteria (Majerus et al. 1998) and killed male embryos appear as infertile eggs (taking 40% as a conservative level for exclusion); (3) they

had fewer than 5 eggs, because small egg batches are oviposited in an atypical way - scattered rather than clumped, with eggs on their sides rather than upright; (4) females produced more than one egg batch during a trial; then we chose one randomly to avoid pseudoreplication.

We used nested analysis of variance (ANOVA) tests to analyze the effect of food level on the number of eggs per batch, the proportion of trophic eggs weighted by clutch size, and hatching synchrony, with the factors order of treatment, treatment, the order by treatment interaction, and the random factor of female nested in order. Exceptions to this form are noted. The models were built using the 'Fit Model' platform and the restricted maximum likelihood function in the program JMP 5.0 (SAS 2002). Some females produced egg batches under one treatment and not the other; these singleton data were included in analysis.

We used three indices of hatching synchrony (Frechette and Coderre 2000): (1) Total hatch time, standardized to mean batch size by the equation: Standardized hatch time = Mean batch size (25.9) * (Total hatch time)/(Number of eggs). (2) The average interval between two sequential hatching larvae. (3) The proportion of eggs per batch, weighted by batch size, that were vulnerable to sibling cannibalism (i.e., because they emerged later than 141 min after the first larva in their egg batch, and thus could have been attacked by that larva; see Study system and experimental animals and Appendix I. 1). We refer to this variable as the 'proportion of delayed hatch eggs'.

We report least squares means * standard error. As an indication of the magnitude of difference between treatments, we report the estimated difference and the 95% confidence interval of the difference (based on the Tukey honestly significant difference function) in the format: mean difference (confidence interval). For non-significant results, the confidence interval indicates our ability to detect a real difference between groups (Hoenig and Heisey 2001; Colegrave and Ruxton 2003).

4.4 Results I: Food level experiments

Internal cue experiment

In the first replication of the experiment, 155 egg batches were produced (77 low food, 78 high food) and 100 were excluded from analysis - 44 were laid in the first six hours, 59 had more than 40% trophic eggs, 13 had fewer than five eggs, and two were from the same female in the same treatment - leaving 55 batches (27 low food, 28 high food). In the second replication, 50 egg batches were laid (17 low food, 33 high food) and 34 were excluded: 15 were laid in the first six hours, 27 had more than 40% trophic eggs, three had fewer than five eggs, and seven were from the same female in the same treatment, leaving six low food and 10 high food batches. Some batches had a combination of excluded features. Results did not change when egg batches produced in the first six hours were included.

Females produced similar sized batches in both replications (first replication: 26.5 eggs * 1.6; second replication: 30.1 eggs * 2.9; mean difference: 3.6 (-3.1 - 10.3); F = 1.2;~= 0.28), so the factor 'Replication' was removed from the model. Egg batches were also of similar size under both treatments (low food: 25.7 * 2.3; high food: 26.4 * 2.1; mean difference: 0.7 eggs (-6.0 - 7.4); F = 0.05; p = 0.82).

Trophic egg-laying.

The proportion of trophic eggs was not affected by the factor 'Replication' (first run:

18.4% * 1.4%; second run: 23.1 * 2.4%; mean difference = 4.8% (-0.9% - 10.4%; F = 3.0; p = 0. lo), so it was dropped from the final model. Females produced 56% more trophic eggs in a low compared to a high food environment (Fig. 4.2a; Table 4.2). There was little effect of order (low food first, 18.1% * 1.8%; high food,

20.0% * 2.3%; F = 0.38; p = 0.54) and the interaction term was not significant (F= 0.95; p = 0.35).

Hatch synchrony

We analyzed hatch synchrony data for the first replication only for two reasons. First, order significantly affected synchrony in most models - that is, there was a day effect of duration of hatching. We interpret this to mean that temperature affected hatch synchrony, because temperature was slightly warmer during hatching in the second trial of the first replication (first trial: 24.4 "C * 0.6, N=116; second trial: 25.8 "C * 0.6, N=116). Second, we lacked hatch synchrony data for seven batches from the second replication, so excluding the nine remaining batches did not sacrifice much information. In some batches, a few larvae hatched much later than the rest; we excluded their hatch times from analysis.

Hatching was not statistically different under low and high food treatments by all three measures of synchrony (Table 4.2), although hatching was less synchronous in the low food treatment for all measures, as predicted. The wide range of values in the 95% confidence intervals indicates a low ability to detect differences (Table 4.2). There was an order effect, such that females that experienced 'low food' first produced egg batches that took significantly longer to hatch, with a trend toward longer intervals between hatching eggs (Low vs. high food: Total hatch time: 227 min * 22 vs. 143 min * 30, F = 5.10,~= 0.03; Hatching interval: 16.4 min * 2.1 vs. 9.4min*2.8,F=3.87,p=0.06). There was thus no effect of food availability on any measure of hatching synchrony, though the trend was in the predicted direction; however, more trophic eggs were produced under the low food treatment.

External cue experiment

In this experiment, females produced a similar number of egg batches under both treatments (74 low food, 70 high food). Ninety-six were laid in the first six hours, 15 had more than 40% trophic eggs, one had fewer than five eggs, and 1 1 were produced by the same females in the same treatment. Some egg batches fell into more than one excluded category, leaving 38 for analysis (1 7 low food, 21 high food). Because we were unable to monitor eggs for the entire hatching period, some hatching time data were lost, leaving 17 batches for synchrony analysis (1 1 low food, 6 high food).

Egg batches were similar in size for both low and high food treatments (low food: 24.2 * 3.3; high food: 3 1.8 St 3.3; mean difference: 7.5 (-5.1 - 20.1); F = 3.624; p = 0.15).

Trophic egg-laying

When information regarding resource levels was provided through an external cue, trophic egg production was not different between treatments (Fig. 4.2b; Table 4.2). There was no effect of order (F = 0.25; p = 0.62).

Hatch synchrony

There were too few hatching synchrony data to include all factors in the nested models of hatching synchrony in this experiment, so Order and Treatment were used. Food level did not affect total hatch time or average hatching interval (Table 4.2). There were, however, significantly more larvae that were vulnerable to cannibalism from delayed hatching (Table 4.2). Order did not significantly affect total hatch time (58 min * 44 vs. 113 min 22; F = 1.28; p = 0.28), the interval between hatching eggs (8.7 min St 2.6 vs. 6.4 min St 1.4; F =

0.64; p = 0.44), or the proportion of vulnerable eggs (-2.4% * 3.1 % vs. 3.6% rt 1.4%; F = 3.35; p

= 0.09), but there was limited ability to detect differences between the groups (mean differences: total hatching time, 54.6 (-49 - 158); interval between hatching eggs, 2.3 (-8.4 - 3.8); proportion of vulnerable eggs, 0.60 (-0.01 - 13)). Thus, there was no effect of food availability on trophic egg production in the external cue experiment. And, while there were more larvae with delayed hatch in the low food treatment, as predicted, the statistical model did not include the interaction or nested terms, limiting the confidence we can place in the result.

4.5 Methods 11: Non-random patterns of infertile egg production

Experimental animals were obtained and maintained as described under 'Methods 1'.

Spatial patterns

To examine the spatial distribution of infertile eggs, we photographed egg batches from 20 females with a digital microscope (Scalar USB Microscope M2, 50X magnification). Larvae were removed from the egg batch after all viable eggs hatched and egg cannibalism occurred. It is easy to distinguish trophic eggs from embryonated eggs because trophic eggs contain visible yolk, or traces of yolk if they have been eaten. It is unlikely that consumed infertile eggs would have developed an embryo because viable and inviable eggs show visible embryo development by the time emerged larvae are ready to eat an egg.

Following the approach of Aviles et al. (1999), we obtained the coordinates of the centre of each egg (estimated visually) from the program NIH Image 1.62 (available on the Internet at http://rsb.info.nih.gov/nih-image/), then calculated the distance in pixels between each pair of eggs in the batch. After hatching occurred, we determined the average distance between each pair of the N infertile eggs. Using a computer-coded bootstrap procedure, we sampled randomly with replacement from all pairwise distances in the batch to generate average distances between N randomly drawn eggs. We repeated the sampling 10,000 times to generate a null distribution and calculated the proportion of randomly generated values that were greater or less than observed average distance between infertile eggs, following usual randomization test procedures

(Edgington 1995). Using a = 0.05, proportions less than 0.025 indicate a clumped distribution and values greater than 0.975 indicate a uniform distribution.

Oviposition patterns

To examine the dispersion of infertile eggs within the oviposition sequence, we videotaped ovipositing females kept individually in petri dishes. When the eggs hatched, we associated each type of egg with its position in the oviposition sequence. We tested for non-random pattern with a runs test, which uses the null hypothesis of randomness in the order of occurrence of members of two groups (Zar 1999).

4.6 Results 11: Non random patterns of infertile egg production

Spatial patterns

Of the twenty egg batches collected from different females, six had at least two infertile eggs, a condition necessary to examine their distribution. The spatial distribution of infertile eggs was not different from random in any batch (Table 4.3).

Oviposition patterns

Oviposition sequences from four females were recorded. The distribution of infertile eggs was not different from random in any sequence (Table 4.4).

4.7 Discussion

This study is the first to test the hypothesis that the infertile eggs of ladybirds are an evolved maternal strategy (Polis 1981 ; 1984; Mock 1994; Pfennig 1997; Osawa 2003). We found evidence that trophic egg production is a plastic maternal behaviour, used adaptively depending on resource availability: in low food conditions, mothers produced 56% more trophic eggs than in high food conditions (Fig. 4.2; Table 4.2). Our results are consistent with models of siblicide (Chapter 3, O'Connor 1978; Mock and Parker 1997), which predict that when offspring starvation risk is great enough, and the relative benefit of eating a sibling exceeds loo%, mothers should facilitate cannibalism among their offspring. One way to ensure cannibalism is to lay a trophic egg (e.g., to neglect to fertilize an offspring). Our results are also consistent with the empirical work of Osawa (1992), which used survival data from a field experiment with H. axyridis to show that sibling cannibalism often supported the inclusive fitness interests of mothers. Our contribution is showing maternal control of a behaviour that appeared adaptive; taken together, these studies suggests some amount of infertility is an adaptation in this species - a maternally- controlled sacrifice of some offspring to improve the survival of others. This suggestion does not claim that all ladybird infertility is adaptive for mothers, for some level of infertility may be unavoidable. The potential for confound between a female's nutritional and information states merits consideration. A female might lay more infertile (trophic) eggs in a low food environment because she is in poor condition, not due to an offspring provisioning strategy. This 'condition hypothesis' suggests that the short period in the low food treatment in this study made females so physiologically weak that fertility plummeted. We consider this explanation unlikely for ladybirds, which feed on an ephemeral prey and should be well adapted to surviving a day with less food (and can survive much longer under laboratory conditions). Furthermore, to ensure similar condition, we fed all females an equal mass of aphids for five days prior to each trial.

We tested another hypothesis about the level of control of trophic egg production, predicting that trophic eggs would be uniformly distributed to maximize the probability that a cannibal attacks a trophic rather than a viable egg. The lack of evidence for any non-random pattern may indicate that, while females manipulate the proportion of trophic eggs in an egg batch, they do not control fertility on an egg-by-egg basis. It is possible that the distribution of trophic eggs is not important for provisioning offspring or preventing the cannibalism of viable eggs. However, the number of egg batches examined was small, so further work on this question is warranted.

A second way mothers may facilitate egg consumption is to decrease hatch synchrony so that emerged offspring can consume siblings with delayed hatching. We did not find a statistical difference in hatching synchrony under low and high food conditions. However, for most measures of hatching synchrony in the two experiments, the difference between low and high food groups was in the predicted direction. More work is needed to make a conclusion about the plastic use of this tactic in ladybirds.

Mothers who lay eggs and leave their offspring are limited in avenues for parental care. Still, in addition to facilitating sibling cannibalism, mothers might reduce offspring starvation risk by adjusting egg size to put more yolk resources into eggs in high starvation risk environments. However, there is no evidence of egg size plasticity in the ladybird Coccinella septempuctata (Dixon and Guo 1993), and other studies in ladybirds suggest that egg size an inflexible species trait (Stewart et al. 1991a; 1991b; Soares et al. 2001), perhaps inflexibly tied to ovariole size. If this is correct, then egg size should be linked to starvation risk across species or populations; and, in fact, a recent study of two goby fish populations with different offspring starvation risk found that mothers laid larger eggs in the high risk population (Maruyama et al. 2003). Furthermore, if egg size is highly constrained, it leaves mothers with little ability for adaptive adjustments between environments, whereas manipulating the opportunity for offspring to consume each other offers the advantage of plasticity between environments of differing starvation risk.

We therefore expect the use of the trophic egg tactic in taxa where use of other potential tactics is constrained. For example, ladybird mothers may have limited control over hatching synchrony, e.g. if it is temperature dependent, as it appeared to be in our study. With imperfect control, hatching may be too synchronous, preventing cannibalism when it works for a mother's interests. Then the solution is to ensure a meal for offspring by laying trophic eggs. As a second example of a constrained behaviour, a possible tactic for reducing offspring starvation would be to refrain from oviposition in low food environments. However, ladybirds appear to be unable to delay oviposition in poor environments: they mature eggs continually and do not resorb eggs (Evans 2003), and in our experiments, they did not reduce clutch size in low food treatments (but see Dixon and Guo 1993).

Our study should be compared with the work of Frechette and Coderre (2000), who found that female lacewings increased the hatching synchrony of egg batches in the presence of aphid prey, but did not change the proportion of infertile eggs. Infertile egg production was low in both the presence (3%) and absence (4%) of aphid prey; thus one possibility is that these animals produced too few infertile eggs to allow detection of a difference. Lacewings and ladybirds have

similar life histories - both are batch-laying aphid predators showing sibling cannibalism

behaviour - and both seem to have evolved maternal strategies to promote sibling cannibalism among offspring. Taken together, their study and ours suggest that manipulating hatching synchrony and producing trophic eggs are alternate solutions to the problem of reducing offspring starvation risk when there is no or little parental care.

We have focused on a scenario of parent-offspring agreement over sibling cannibalism, proposing that mothers help cannibal offspring by neglecting to fertilize some eggs (Mock and Forbes's (1995) trophic facilitation hypothesis; Crespi 1992). An alternative explanation for trophic egg evolution is the conflict reduction hypothesis (Crespi 1992), which proposes that trophic eggs may have evolved to reduce parent-offspring conflict over cannibalism of viable eggs, in systems where trophic eggs are less costly to produce than viable offspring. However, parent-offspring agreement is a more likely explanation for the trophic eggs of ladybirds for two reasons. First, the survival benefit to cannibals of consuming eggs are substantial (reviewed in Hodek and Honek 1996), and in most studies exceed loo%, the threshold for mothers to benefit from sibling cannibalism (Chapter 3, O'Connor 1978). Second, there is no evidence that ladybird trophic eggs are energetically cheaper to produce than viable eggs. Ladybird trophic eggs and viable eggs appear identical to the eye (until the last few hours before hatching when the developing embryo is visible in viable eggs) and there is no difference in their mass (Appendix 1.2). It thus seems reasonable to assume that they require similar energy input from mothers, though this assumption requires biochemical analysis to be tested. If trophic and viable eggs truly have equal cost, then mothers should produce trophic eggs only if sibling cannibalism is also in their interests, because producing a trophic egg has the same effect as a mother killing a potentially viable offspring. Thus, there should be no conflict over the cannibalism of viable offspring - mothers and cannibals agree that it should proceed.

Our study is the first to test the assumption that trophic eggs are an adaptive maternal strategy in a nonsocial insect. Another approach to demonstrating the adaptive nature of trophic eggs is a cross-taxa comparison of infertile egg production in species with relevant characteristics, such as those that do or do not lay eggs in batches and where larvae do or do not cannibalize siblings. 4.8 Literature cited

Alexander, R. D. 1974. The evolution of social behaviour. Annual Review of Ecology and Systematics 5:325-383.

AvilCs, L., C. Varas, and E. Dyreson. 1999. Does the African social spider Stegodyphus dumicla control the sex of individual offspring? Behavioural Ecology and Sociobiology 46:237- 243.

Banks, C. J. 1956. Observations on the behaviour and mortality in Coccinellidae before dispersal from the egg shells. Proceedings of the Royal Entomology Society of London (A) 3 1 :56- 60.

Branquart, E., J.-L. Hemptinne, C. Bauffe, and L. Benfekih. 1997. Cannibalism in Episyrphus balteatus (Dipt.: Syrphidae). Entomophaga 42: 145-152.

Brown, H. D. 1972. The behaviour of newly hatched coccinellid larvae (Coleoptera: Coccinellidae). Journal of the Entomological Society of South Africa 35: 149-157.

Carter, M. C., and A. F. G. Dixon. 1984. Honeydew: an arrestant stimulus for coccinellids. Ecological Entomology 9:383-387.

Colegrave, N., and G. D. Ruxton. 2003. Confidence intervals are a more useful complement to nonsignificant tests than are power calculations. Behavioural Ecology 14:446-447.

Crespi, B. J. 1992. Cannibalism and trophic eggs in subsocial and eusocial insects. Pages 176-213 in M. A. Elgar, and B. J. Crespi, eds. Cannibalism: ecology and evolution among diverse taxa. Oxford University Press, Oxford.

Crump, M. L. 1996. Parental care among the Amphibia. Advances in the Study of Behavior 25:109-144.

Dickins, D. W., and R. A. Clark. 1987. Games theory and siblicide in the kittiwake gull, Rissa tridactyla. Journal of Theoretical Biology 125:3Ol-305.

Dickinson, J. L. 1992. Egg cannibalism by larvae and adults of the milkweed leaf beetle (Labidomera clivicollis, Coleoptera: Chrysomelidae). Ecological Entomology 17:209- 218.

Dixon, A. F. G., and Y. Guo. 1993. Egg and cluster size in ladybird beetles (Coleoptera: Coccinellidae): the direct and indirect effects of aphid abundance. European Journal of Entomology 90:457-463.

Edgington, E. S. 1995. Randomization tests. Marcel Dekker, New York.

Evans, E. W. 2003. Searching and oviposition behaviour of female aphidophagous ladybirds (Coleoptera: Coccinellidae): a review. European Journal of Entomology 100: 1- 10.

Evans, E. W., and A. F. G. Dixon. 1986. Cues for oviposition by ladybird beetles (Coccinellidae): response to aphids. Journal of Animal Ecology 55: 1027-1034. Forbes, S., R. Grosshans, and B. Glassey. 2002. Multiple incentives for parental optimism and brood reduction in blackbirds. Ecology 83:2529-254 1.

Frechette, B., and D. Coderre. 2000. Oviposition strategy of the green lacewing Chrysoperla ruJlabris (Neuroptera: Chrysopidae) in response to extraguild prey availability. European Journal of Entomology 97:507-5 10.

Gobin, B., C. Peeters, and J. Billen. 1998. Production of trophic eggs by virgin workers in the ponerine ant Gnamptogenys menadensis. Physiological Entomology 23:329-336.

Godfray, H. C. J. 1987. The evolution of clutch size in invertebrates. Oxford surveys in evolutionary biology 4: 1 17-154.

Godfray, H. C. J., and A. B. Harper. 1990. The evolution of brood reduction by siblicide in birds. Journal of Theoretical Biology 145:163-175.

Henry, C. S. 1972. Eggs and rapagula of Ululodes and Ascaloptynx (Neuroptera: Ascalaphidae): a comparative study. Psyche 79: 1-22.

Heying, H. E. 2001. Social and reproductive behaviour in the Madagascan poison frog, Mantella laevigata, with comparisons to the dendrobatids. Animal Behaviour 6 1:567-577.

Hodek, I., and A. Honek. 1996. Ecology of Coccinellidae. Kluwer Academic Publishers, Dordrecht, The Netherlands.

Hoenig, J. M., and D. M. Heisey. 2001. The abuse of power: the pervasive fallacy of power calculations for data analysis. The American Statistician 55: 19-24.

Kawai, A. 1978. Sibling cannibalism in the first instar larvae of Harmonia axyridis Pallas (Coleoptera: Coccinellidae). Kontyu 46: 14-19.

Kim, K. W., and C. Roland. 2000. Trophic egg laying in the spider, Amaurobius ferox: mother- offspring interactions and functional value. Behavioural Processes 50:3 142.

MacKaye, K. R. 1986. Trophic eggs and parental foraging for young by the catfish Bagrus meridionalis of Lake Malawi, Africa. Oecologia 69:367-369.

Magrath, R. D. 1990. Hatching asynchrony in altricial birds. Biological Reviews 95:587-622.

Majerus, M. E. N. 1994. Ladybirds. HarperCollins, London.

Majerus, T. M. O., M. E. N. Majerus, B. Knowles, J. Wheeler, D. Bertrand, V. N. Kuznetzov, H. Ueno et al. 1998. Extreme variation in the prevalence of inherited male-killing microorganisms between three populations of Harmonia axyridis (Coleoptera: Coccinellidae). Heredity 8 1:683-691.

Maruyama, A., B. Rusuwa, and M. Yuma. 2003. Interpopulational egg-size variation of a landlocked Rhinogobius goby related to the risk of larval starvation. Environmental Biology of Fishes 67:223-230.

Matsuko, K. 2003. Larval oophagy in the ant Amblyopone silvestrii (Hymenoptera, Formicidae). Insectes Sociaux 50:3 17-322. Mock, D. W. 1994. Brood reduction: narrow sense, broad sense. Journal of Avian Biology 25:3- 7.

Mock, D. W., and L. S. Forbes. 1995. The evolution of parental optimism. Trends in Ecology and Evolution 10: 130-134.

Mock, D. W., and G. A. Parker. 1997. The evolution of sibling rivalry. Oxford University Press, Oxford.

Nakahira, T. 1994. Production of trophic eggs in the subsocial burrower bug Admerus triguttulus. Naturwissenschaften 8 1:4 13-4 14.

Ng, S. M. 1986. Egg mortality of four species of aphidophagous Coccinellidae in Malaysia. Pages 77-8 1 in I. Hodek, ed. Ecology of Aphidophaga. Academia, Prague.

O'Connor, R. J. 1978. Brood reduction in birds: selection for fratricide, infanticide and suicide. Animal Behaviour 26:79-96.

Osawa, N. 1989. Sibling and non-sibling cannibalism by larvae of a lady beetle Harmonia axyridis Pallas (Coleoptera: Coccinellidae) in the field. Researches in Population Ecology 31:153-160.

-. 1992. Sibling cannibalism in the ladybird beetle Harmonia axyridis: fitness consequences for mother and offspring. Researches in Population Ecology 34:45-55.

-. 2003. The influence of female oviposition strategy on sibling cannibalism in the ladybird beetle Harmonia axyridis (Coleoptera: Coccinellidae). European Journal of Entomology 100:43-48.

Peck, W. B., and W. H. Whitcomb. 1970. Studies on the biology of a spider, Chiracanthium inclusum (Hentz). Bulletin of the Arkansas Experiment Station 753: 1-75.

Petrescu, A., E. Mondor, and B. Roitberg. 2001. Subversion of alarm communication: do plants habituate aphids to their own alarm signals? Canadian Journal of Zoology 79:737-740.

Pfennig, D. W. 1997. Kinship and cannibalism. Bioscience 47:667-775.

Pienkowski, R. L. 1965. The incidence and effect of egg cannibalism in first instar Coleomegilla maculata lengi (Coleoptera: Coccinellidae). Annals of the Entomological Society of America 58:150-153.

Polis, G. A. 198 1. The evolution and dynamics of intraspecific predation. Annual Review of Ecology and Systematics 12:225-25 1.

-. 1984. Intraspecific predation and "infant killing" among invertebrates. Pages 87-104 in G. Hausfater, and S. Hrdy, eds. Infanticide: comparative and evolutionary perspectives. Aldine Publishing Company, New York.

SAS. 2002. JMP 5.0 user's guide.

Sigsgaard, L., M. H. Greenstone, and S. J. Duffield. 2002. Egg cannibalism in Helicoverpa armigera on sorghum and pigeonpea. BioControl47: 15 1- 165. Soares, A. O., D. Coderre, and H. Schanderl. 2001. Fitness of two phenotypes of Harmonia axyridis (Coleoptera: Coccinellidae). European Journal of Entomology 98:287-293.

Stevens, L. 1992. Cannibalism in beetles. Pages 156-175 in M. A. Elgar, and B. J. Crespi, eds. Cannibalism: ecology and evolution among diverse taxa. Oxford University Press, Oxford.

Stewart, L. A., A. F. G. Dixon, Z. Ruzicka, and G. Iperti. 1991a. Clutch and egg size in ladybird beetles. Entomophaga 36:329-333.

Stewart, L. A., J.-L. Hemptinne, and A. F. G. Dixon. 1991b. Reproductive tactics of ladybird beetles: relationships between egg size, ovariole number and developmental time. Functional Ecology 5:380-385.

Stubbs, M. 1980. Another look at prey detection by coccinellids. Ecological Entomology 5: 179- 182.

Valerio, C. E. 1974. Feeding on eggs by spiderlings of Achaeranea tepidariorum (Araneae, Theridiidae), and the significance of the quiescent instar in spiders. Journal of Arachnology 2:57-63.

Via, S. 1999. Cannibalism facilitates the use of a novel environment in the flour beetle, Tribolium castaneum. Heredity 82:267-275.

West, M. J., and R. D. Alexander. 1963. Sub-social behaviour in a burrowing cricket Anurogryllus muticus (De Geer). Ohio Journal of Science 63: 19-24.

Zar, J. H. 1999. Biostatistical analysis. Prentice Hall, New Jersey. Figure 4-1 External cue experimental chamber. Ladybirds were placed in the upper compartment with filter paper as an oviposition surface, a water source, and a small mass of pea aphids. The bottom compartment contained a bean leaf pair, inserted through Parafilm0 into a vial of water, and the treatment: either no or many pea aphids.

Figure 4-2 Proportion of trophic eggs produced when females were given information that there was low or high food availability through (a) the internal cue of food consumption or (b) the external cue of prey scent; see text for details. The mean proportion was smaller under the low food treatment in the internal cue but not the external cue experiment (Table 4.2). Horizontal lines represent means. Whiskers indicate the range of observations. I -- I 0.00 ' I I Low food High food

Treatment

No food signal High food signal Treatment Table 4-1 Comparison of protocols for the internal and external cue experiments Experiment Number of Low food High food Conditions subjects treatment treatment Internal 106 in first 0.0040-0.0070 0.05 16 g * - Females could contact cue replicate, 70 in g pea aphids 10% pea treatment aphids second replicate aphids External 116 No aphids 0.0806 g * - Females were separated cue 10% pea from treatment aphids by aphids mesh - Females could contact a small amount of pea aphids (0.0040-0.0070 g) Table 4-2 Effect of food level on proportion of trophic eggs and three measures of hatching synchrony (see text) when females were given information about food conditions through feeding (internal cue) or other senses (external cue). Sample sizes were: in the internal cue experiment, 7 1 for the proportion of trophic eggs and 55 for hatching synchrony; for the external cue experiment, 38 for the proportion of trophic eggs and 17 for hatching synchrony. Least squares mean (SE) Reproductive tactic Low food High food Mean difference F P (95% CI) Internal cue experiment Proportion oftrophic eggs

Hatching synchrony Total hatch time (min) Interval between hatching eggs (min) Proportion of delayed hatch eggs External cue experiment Proportion of trophic eggs

Hatching synchrony Total hatch time (min) Interval between hatching eggs (min) Proportion of delayed hatch eggs (0.016) (0.027) Proportions are weighted by number of eggs per batch. Total hatch time is standardized to mean number of eggs per batch. Table 4-3 Randomization tests for the spatial distribution of trophic eggs within egg batches produced by different females. P values were generated from simulations of possible distributions, given the spatial arrangement of each batch; values less than 0.025 indicate a clumped distribution while values greater than 0.975 indicate a uniform distribution. Egg batch Number of eggs P 1 18 0.03 Table 4-4 Runs tests (Zar 1999) of the temporal distribution of infertile eggs within oviposition sequences of different females Egg batch Number of eggs Test statistic P 7 58 -1.29 0.20 8 2 1 1.13 0.26 9 9 -1.63 0.10 10 32 -0.29 0.77 Chapter 5 Summary of the Thesis and Suggestions for Future Work

5.1 Introduction

Trophic egg laying - the production of infertile eggs that are consumed by offspring - occurs in diverse taxa (see Chapter 2; Crespi 1992). The goals of this thesis were to outline a behavioural ecological framework for the study of trophic eggs; to investigate the co-evolution of maternal and offspring traits related to trophic eggs; and to assess the adaptive function of trophic eggs in the ladybird Harmonia axyridis. The present chapter summarizes the thesis and suggests approaches for the further study of trophic eggs.

5.2 Thesis summary

This thesis comprised a literature review and synthesis, a model of trophic egg laying and related behaviours, and empirical study of trophic eggs in H. axyridis. Chapter 1 reviews the approach taken in each chapter. In Chapter 2, I collected hypotheses for the adaptive function of trophic eggs and reviewed the evidence necessary to distinguish among hypotheses. One set of hypotheses suggests that trophic eggs function to reduce parent-offspring conflict over sibling cannibalism (Crespi 1992): mothers might produce trophic eggs if offspring will consume them instead of viable siblings. A second set of hypotheses posits that trophic eggs are an offspring-provisioning tactic. An alternative hypothesis is that trophic eggs are not a maternal adaptation, but simply infertile eggs that mothers cannot avoid producing. If trophic eggs are to be studied in an evolutionary ecological context, the hypotheses of adaptation versus constraint, and the adaptive functions of parent-offspring conflict reduction or cooperation, must be distinguished. I reviewed the value of evidence from natural history observations, behavioural experiments, and comparative analyses. Chapter 3 presented the results of a simulation-based model of the co-evolution of trophic egg laying, hatching synchrony, and sibling cannibalism. I compared the results to predictions of a frequency-independent inclusive fitness analysis of sibling cannibalism that did not incorporate game aspects. The model predicts that competition among siblings over the opportunity to cannibalize increases cannibalistic tendencies. I also analyzed the effect of three biological factors. First, as predicted by the frequency-independent model, the relative increase in survival that cannibals receive determined the optimal behaviour for offspring and mothers. Second, when infertile eggs were present and there was even a small benefit to eating eggs, offspring risked eating viable siblings in order to consume infertile eggs. Third, allowing offspring to wait until viable siblings hatched before eating eggs had no effect on optimal behaviours. The model suggests that using trophic eggs to provision offspring is a feasible maternal tactic. In Chapter 4, I presented the results of laboratory experiments on trophic eggs and hatching synchrony in H. axyridis. I hypothesized that the infertile eggs of H. axyridis are a maternal adaptation for provisioning offspring (thus, that they are trophic eggs), and that mothers would produce more trophic eggs in low food environments where starvation risk for offspring was greater. I also predicted that mothers would create asynchronously hatching clutches in low food environments so that first-hatching offspring could consume siblings with delayed hatch. I conducted two experiments to test the hypotheses that differed in how females were given information about local food conditions. The results showed that females used information from their own consumption of aphids to adjust trophic egg production as predicted. There was no statistically significant decrease in hatching synchrony in the low food treatment; however, because the trend was in the predicted direction, more work is needed in this area. I controlled for a potential confound between the nutrition and information states of females by ensuring that females were in similar condition before the experiments and exposing them to low or high food for a short period. Finally, it might benefit mothers to over-disperse trophic eggs within clutches, if offspring are then more likely to eat a trophic egg rather than a viable sibling. However, there was no evidence for a non-random distribution of trophic eggs within clutches or oviposition sequences. Appendix I, following this chapter, contains additional data collected during this thesis research. The data demonstrate that first instar H. axyridis larvae will cannibalize sibling eggs when given the opportunity; that trophic and viable eggs do not differ in mass; and that field- collected egg batches have a lower proportion of trophic or infertile eggs than laboratory egg batches. I also review the levels of infertility observed in other studies of ladybirds infected with male-killing bacteria and suggest that female ladybirds may adjust trophic egg production depending on infection status.

5.3 Suggestions for future work

Before commenting on the study of trophic eggs in general, I note further work needed to broaden the understanding of trophic eggs in H. axyridis. Chapter 4 showed that female H. axyridis adjust the proportion of trophic eggs to local food abundance in the laboratory. This result must be repeated in the field to confirm that it occurs under natural conditions. Such a field study will be a difficult undertaking. Investigators would need to find egg batches deposited near small and large aphid colonies within the same local environment - but if large aphid colonies are available, ladybirds will not oviposit near small colonies (pers. obs.). Alternatively, investigators might compare the proportion of trophic eggs in egg batches from low- and high-food sites. However, this approach cannot eliminate the possibility that females in low-food environments are in poor condition and produce eggs with low hatchability for that reason alone (the information-nutrition confound; see Chapter 3). Another approach would be to place ladybirds on plants with small or large aphid colonies and to prevent ladybirds from departing by clipping their wings or placing nets over plants. This experimental design would also be subject to the information-nutrition confound, unless experimental trials were short to prevent female condition from deteriorating in the low-food treatment. A second prediction for the H. axyridis system relates to the senescence pattern of aphid colonies. Colonies of large mature aphids are on the wane and may disappear before foraging larvae mature to adulthood (Majerus 1994). Ladybirds prefer to oviposit near young aphid colonies (Evans 2003). Thus, the age of aphid colonies is an index of offspring starvation risk, and ladybirds should increase the proportion of trophic eggs when laying near old, compared to young, aphid colonies. The following section reviews some interesting questions for systems where the adaptive function of trophic eggs is known. Some points have been covered in previous chapters; for synthesis, they are brought together here.

Future work - both theoretical and empirical - should address the following questions. (1) What ecological circumstances and life history characteristics drive selection for trophic egg laying, rather than other functionally equivalent tactics? (2) What does the evolution of trophic eggs imply for selection on related traits? (3) What is the evolutionary origin of trophic egg laying?

Approaches to answering question (1)

From a behavioural ecological perspective, an important research objective is to understand the occurrence of trophic eggs across taxa. What ecological factors or species characteristics should be correlated with the use of trophic eggs, rather than other tactics? Alternative tactics mothers might use to facilitate or limit cannibalism among offspring include the manipulation of hatching synchrony (Chapter 3), clutch size, and egg size .(Crespi 1992). Consider the situation in which mothers are selected to reduce starvation risk for offspring. I suggest that the tactic that evolves in any particular system will depend on environmental variability and predictability. In an environment of constant offspring starvation risk across habitats and seasons, there is no need for flexibility in maternal behaviour. Egg size, for example, appears to be a fixed species characteristic with little phenotypic plasticity (Chapters 2,4). Across species, then, I predict that egg size will be inversely related to the mean starvation risk that offspring experience. When environments are heterogeneous and unpredictable, parents may manipulate clutch size to create insurance offspring. In poor circumstances, senior offspring can kill (and/or cannibalize) junior offspring, while if good food conditions occur, parents can successfully rear junior offspring. Examples of the use of this tactic include facultatively siblicidal birds, such as great egrets (Mock and Parker 1997) and blue-footed boobies (Drummond and Garcia Chavelas 1989). In environments that are heterogeneous for starvation risk, but predictable at the time of egg deposition, parents can use plastic tactics to adjust the opportunity for sibling cannibalism among offspring, such as hatch synchrony and trophic eggs. In order to use trophic egg tactics - either to reduce conflict over sibling cannibalism or to provision offspring - species must already exhibit certain characteristics. Offspring must occur in a physically confined and inescapable space for some period of their lives (Mock and Parker 1997). Offspring must also have the ability to digest eggs or siblings. We might, then, expect trophic eggs to occur more frequently in carnivorous species. The above comments are related to understanding the use of trophic egg laying rather than alternative tactics. We can also gain understanding of the selective pressures that drive trophic egg laying by comparing species that lay trophic eggs to related species that do not. If trophic eggs function to provision offspring (see Chapter 2), several predictions can be made. First,

species in which offspring face high starvation risk - where the chance of surviving starvation is 150% (Chapter 2) - should exhibit trophic egg laying more often than species with lower starvation risk. Suitable indices of starvation risk include average food resource abundance experienced by foraging offspring or food acquiring ability (e.g., success as predators) in young offspring. Spatial and temporal variability in resource abundance may make the first index difficult to quantify. A second prediction is that species that lay eggs singly should experience lower rates of infertility than batch-layers. Solitary offspring have no opportunity to consume a sibling or trophic egg and therefore mothers do not derive benefit from the increased survival of egg-eating offspring. A good group for testing these predictions is the coccinellid beetles. Starvation risk ranges from 16% to 226% (Osawa 1992) in laboratory studies. Many species, but not all species, lay trophic eggs. Some species lay eggs singly while others are batch-layers. A phylogeny for the group exists (Mawdsley 2001) and they are easily reared.

A suggested approach for question (2)

In Chapter 3, I claim that trophic egg laying is necessarily related to other maternal and offspring behaviours that affect sibling cannibalism. The suite of traits that co-evolve with trophic egg laying should be tested in a selection experiment. As an example, female H. axyridis that produce very low or high proportions of trophic egg production could be used to create sib-mated lines. Each generation would be artificially selected for further divergence in trophic egg proportion. I predict that hatch synchrony and cannibalistic tendencies in offspring will co-evolve with trophic egg laying, though the expected direction of change is not clear (see discussion in Chapter 2).

A suggested approach for question (3)

The only considerations of the evolutionary origins of trophic eggs occur in Crespi (1992), Weygoldt (1980) in reference to the Dendrobatidae, and Chapter 2 of this thesis. As discussed in Chapter 2, trophic eggs may have arisen via increased production of already-present infertile eggs or from a novel process to stop egg development. A study of the genetic and biochemical differences between trophic eggs, other infertile eggs, and viable eggs may suggest the evolutionary precursors from which they arose, yet no biochemical or genetic work has been done on trophic eggs outside of eusocial insects. For example, chromosomal analysis could determine whether trophic eggs are not fertilized, or fertilized but halted at an early stage of development (as in the trophic eggs of fire , Voss et al. 1987). Biochemical differences between trophic and viable eggs could be examined by conducting a percentage fatkarbohydratelprotein analysis; the result might show similarities between trophic eggs and infertile or viable eggs.

5.4 Conclusions

In Chapter 2, I emphasized the need for progress in theory of the ecology and evolution of trophic egg laying, and the need for empirical studies to demonstrate that purported trophic eggs are not simply infertile eggs produced by constraint. I attempted to pursue these goals in my own study of trophic eggs in Chapters 3 and 4. An important point, emphasized in chapter 1, is that there are likely varied adaptive functions and evolutionary origins of trophic eggs across systems. No single answer exists to any of the questions of interest I have suggested here. Thus, research programs should concentrate on answering questions for particular trophic egg-laying groups, such as the dendrobatid frogs or the coccinellid beetles. Further work in the ecology and evolution of trophic eggs is warranted because the topic is of general interest to ecological theory. Answers to questions about the evolution and adaptive function of trophic eggs may shed light on other areas in behavioural and evolutionary ecology research, such as parent-offspring conflict, forms of parental care, maternal effects, sibling rivalry, and brood reduction theory. In general terms, the study of trophic eggs shows that an unusual parenting strategy, killing off some offspring to benefit others, can be adaptive. 5.5 Literature cited

Crespi, B. J. 1992. Cannibalism and trophic eggs in subsocial and eusocial insects. Pages 176-213 in M. A. Elgar, and B. J. Crespi, eds. Cannibalism: ecology and evolution among diverse taxa. Oxford University Press, Oxford.

Drummond, H., and C. Garcia Chavelas. 1989. Food shortage influences sibling aggression in the blue-footed booby. Animal Behaviour 37:806-8 19.

Evans, E. W. 2003. Searching and oviposition behaviour of female aphidophagous ladybirds (Coleoptera: Coccinellidae): a review. European Journal of Entomology 100: 1-10.

Kawai, A. 1978. Sibling cannibalism in the first instar larvae of Harmonia axyridis Pallas (Coleoptera: Coccinellidae). Kontyu 46: 14-19.

Majerus, M. E. N. 1994. Ladybirds. HarperCollins, London.

Mawdsley, J. R. 2001. Mitochondria1 cytochrome oxidase I DNA sequences and the phylogeny of Coccinellidae (Insecta: Coleoptera: Cucujoidea). Journal of the New York Entomological Society 109:304-308.

Mock, D. W., and G. A. Parker. 1997. The evolution of sibling rivalry. Oxford University Press, Oxford.

Osawa, N. 1992. Sibling cannibalism in the ladybird beetle Harmonia axyridis: fitness consequences for mother and offspring. Researches in Population Ecology 34:45-55.

Voss, S. H., J. F. McDonald, J. H. D. Bryan, and C. H. Keith. 1987. Abnormal mitotic spindles: developmental block in fire ant trophic eggs. European Journal of Cell Biology 45:9-15.

Weygoldt, P. 1980. Complex brood care and reproductive behaviour in captive poison-arrow frogs, Dendrobatespumilio 0. Schmidt. Behavioural Ecology and Sociobiology 7:329- 332. APPENDIX I

This Appendix summarizes additional data collected during and relevant to this thesis research. Each section below describes the procedure followed for an experiment or observation set, presents the results, and briefly comments on the relevance to the rest of the thesis (see Chapter 1).

1.1 Frequency of sibling cannibalism in Harmonia qridis

Cannibalism is a ubiquitous behaviour in coccinellid beetles (e.g., Majerus 1994). However, the frequency of sibling cannibalism within species has seldom been studied. Because trophic egg-laying is related to sibling cannibalism, the question arose: how often mothers faced with cannibalism among offspring? I assayed the tendency of newly emerged larvae to consume a sibling egg.

Methods

I created pairs of sibling eggs, consisting of taking an egg from each of two batches laid by the same female. Females were maintained under laboratory conditions as described in Chapter 4. Eggs were collected at least one day after oviposition, when they can be handled without damage. I glued eggs upright (as they occur normally) in the center of the bottom of 5 cm petri dishes, using a dilute mixture of UHS glue and water. Pairs of eggs were in contact with each other, as they would occur in a normal egg batch. Developing eggs turn grey several hours before hatching. When the first-laid egg of each pair turned grey, I monitored the pair every 15 minutes and noted the hatch time. Hatched larvae could consume the egg next to them or disperse away, i.e., move at least 3 cm from the egg pair. I monitored eggs until cannibalism or dispersal occurred. The assays were conducted at 24 C * 1. I monitored 76 egg pairs from eleven paired egg batches from eight females, with a mean of 10 egg pairs from each female (range 1, 3 1). Four paired egg batches were tested from one female. To obtain the mean latency time until cannibalism, I averaged the latency time of each female and calculated the weighted mean of this average. Results and summary

All 76 larvae consumed an adjacent egg upon hatching. The result suggests that cannibalism is frequent in first instar H. axyridis. Thus, we may expect mothers to evolve tactics to manage cannibalism among offspring. The weighted mean latency to cannibalism was 14 1 minutes h 12.9 (S.E.). In Chapter 4,I compared the mean latency time to the mean total hatch time for egg batches to check whether larvae delay cannibalism until after viable eggs have hatched.

1.2 Does egg viability or hatching time depend on egg mass?

Trophic eggs may be produced to reduce parent-offspring conflict over sibling cannibalism or to provision offspring. As argued in Chapter 2, the former explanation is possible only if trophic eggs are less costly to produce than viable eggs - for example, if they contain less yolk than viable eggs. To test whether egg mass was related to egg type (i.e., viable, inviable, or trophic), I weighed eggs and followed their development. The experiments discussed in Chapter 4 tested whether female ladybirds manipulated hatching synchrony within batches of offspring. Females might control synchrony by differentially provisioning eggs, given that egg mass affects development time across coccinellid species (Stewart et al. 1991). Here, I tested whether egg mass affected development time within H. axyridis.

Methods

(a) I collected egg batches from H. axyridis females maintained under laboratory conditions as described in Chapter 4.1 weighed individual eggs to the nearest 0.001 pg using a cahnTM microbalance, taking care to avoid damaging eggs. Eggs were weighed approximately 12 hours after oviposition ('initial mass') and again in the 24 hours preceding hatch ('final mass'), which occurs approximately 78 hours after oviposition. 1 weighed 288 eggs from 11 egg batches laid by eight females (mean batch size * SE, 26 * 3.4; mean eggs per female, 36 * 8.7): a single batch from five females and two batches from three females. In preliminary analysis, weight did not differ greatly among batches within females, so the factor 'egg batch' was dropped from the final models. I tested whether egg mass (independent variable) affected embryo development and the likelihood of egg hatching with logistic regression, in which 'initial mass' and 'female' were main factors. Reported statistics for these tests are in the form (Wald Chi Square, df, P). I report means * S.E. Ladybirds produce viable eggs, infertile trophic eggs where no embryo develops, and inviable eggs where an embryo develops but does not emerge from the egg capsule. I tested whether egg type affected proportionate mass lost between initial and final weighings, using a two-way ANOVA that also included the factor 'female'. (b) To examine the relationship between egg mass and development time, I haphazardly selected four weighed egg batches in which to observe hatching. Batches are referred to in the form 'female-batch number'. When eggs were close to hatching (evident from the colour change several hours before hatch), I monitored them every 10 minutes and noted the time that larvae emerged (i.e., when all six appendages were free of the egg capsule). I weighed larvae two hours after emergence, when they are sclerotized enough to handle without damage. I tested whether egg mass affected development time by the mean square successive difference test (Zar 1999), which tests the null hypothesis of randomness in the order of continuous data collected serially. The test generates a C statistic; the null hypothesis is rejected if C equals or exceeds the appropriate critical C value obtained from Table B.30 in Zar (1 999, p. App 180).

Results and summary

Across females, the mean initial and final masses of ladybird eggs were 0.222 pg * 0.001 and 0.203 p + 0.001, respectively. The initial mass of eggs did not affect embryo development (0.13, 1, P < 0.72) or hatching likelihood (0.90, 1, P < 0.34). The eight females tested produced eggs that did not differ

significantly in likelihood of embryo development (12.6, 7, P = 0.080; minimum and maximum proportion of non-developing eggs per female: 0.07,0.67), but eggs from different females

differed in likelihood of hatching (17.7, 7, P = 0.013; minimum and maximum proportions of non-hatching eggs per female: 0.10, 0.78).

Egg type affected the proportion of mass lost (Figure I. 1; F Ratio 6.4, df 2, P = 0.002). Females differed in the proportion of weight loss that their eggs experienced (F Ratio 4.2, df 7, P

= 0.0002; minimum and maximum LS mean, 0.06,O. 10). Thus, the trophic eggs of ladybirds do not differ in initial mass from viable or inviable eggs. However, viable eggs lose more mass during development than do trophic eggs, probably due to yolk metabolism in developing embryos. Some females were more likely than others to produce inviable eggs; the reason is not known. The initial mass of eggs was not related to hatching order in three of four tested batches (27-4: C, 0.04; C,",, 0.45; 32-33: C, -0.23; C,,, 0.49; 60-16: C, -0.25; C,",, 0.41). In batch 60-15, heavier eggs hatched significantly earlier than lighter eggs (C, 0.49; C,",, 0.41; Fig. 1.2). The amount of weight loss that eggs experienced was not related to hatching order (27-4: C, 0.1 1; C,~,,0.45; 32-33: C, 0.41; C,~,,0.49; 60-15: C, -0.36; Cca, 0.41; 60-16: C, -0.12; Cc,it, 0.41). There is some indication from these data that mothers can control hatch synchrony via the proximate mechanism of within-clutch variation in egg mass. Mothers might increase the opportunity for sibling cannibalism among offspring by differentially provisioning eggs within batches; then heavier, earlier hatching offspring can cannibalize lighter, later hatching siblings. However, with the small sample size and inconsistent relationship between egg mass and hatching order, more work is needed on this topic.

1.3 Trophic egg production in the laboratory and field

Some portion of the infertile eggs of H. axyridis observed in this thesis may be an artefact of laboratory conditions. For example, most laboratory settings involve periodic feeding, rather than the continuous food supply available in nature; this may disrupt the addition of yolk to eggs (Dixon 2000). To test whether infertile egg production was higher under laboratory conditions, I compared the proportion of trophic eggs in field- and laboratory-collected egg batches.

Methods

I collected 24 H. axyridis egg batches from aphid-infested trees on the Simon Fraser University campus in Burnaby, BC, in July 2003. Egg batches were maintained in laboratory conditions (described in Chapter 4) until hatch, which occurred within three days. The proportion of trophic eggs was noted for each batch. Hatched larvae were reared to the second instar, when they can be identified as H. axyridis. Cannibalism likely occurred within most batches. I recorded cannibalized eggs that showed no sign of development as trophic eggs, because viable eggs begin to turn grey by the time hatched larvae are ready to eat an egg. For comparison, I randomly selected 24 egg batches from H. axyridis females collected from the field as adults and maintained under laboratory conditions. Egg batches had to contain at least five eggs to be selected, because batches containing fewer eggs are oviposited in an atypical manner. Field-collected batches all contained at least five eggs. Female H. axyridis can be infected with a male-killing bacteria that prevents the development of male eggs (Majerus et al. 1998). Killed eggs would appear as trophic or infertile eggs. To eliminate the possibility that infection status influenced the result, I conducted a second analysis excluding egg batches with more than 40% non-developing eggs. Twenty-three field-collected batches had at least 60% of embryos develop. For comparison, I randomly selected 23 egg batches with at least 60% embryo development from laboratory-maintained females. I compared the mean proportion of trophic eggs among field- and laboratory-collected batches ('egg origin') with one-way ANOVA in JMP 5.0. In separate analyses, I considered the response variables as the weighted proportion of trophic eggs per egg batch or the number of eggs per batch. Least square means are reported * SE. Estimated differences are reported with the 95% confidence interval of the difference.

Results and summary

When egg batches with more than 40% trophic eggs were included, field-collected egg batches were composed of 14.9% * 3.4% trophic eggs, while laboratory-collected batches had 21.3% * 3.9% trophic eggs. The difference was not statistically significant (F Ratio 1.5, df 1, P = 0.23). The confidence interval (-4.1%, 16.8%) around the estimated difference of 6.3% is wide, indicating that the variation in the data is too large for the sample size of this study to allow detection of a difference, if one existed. When egg batches containing more than 40% trophic eggs were excluded, the difference in proportion of trophic eggs between laboratory and field collected batches decreased (laboratory: 13.3% * 2.5%; field: 12.5% * 2.3%; F Ratio 0.1, df 1, P = 0.81). The relatively smaller confidence interval around the estimated difference (3.4% (-7.6%, 6.0%)) indicates a better ability to detect a difference that actually existed. One of 24 field-collected batches and six of 24 randomly selected laboratory batches contained more than 40% trophic eggs. The higher incidence of egg batches with over 40% trophic eggs in laboratory batches is statistically significant (h2= 14.7, df 1, P < 0.001). The data therefore suggest that laboratory conditions do not generally cause higher rates of infertile egg production, but do cause some females to produce a high proportion of trophic eggs (> 40%). The effect may be related to the greater diet diversity in the field than in the laboratory or to the restricted number of mates. Considering these results, some portion of the infertile eggs produced in the experiments described in Chapter 4 were likely caused by laboratory conditions. This factor is not likely to bias the observed increase in trophic egg proportion in low food conditions. Egg batches were significantly smaller in the laboratory than in the field when egg batches with more than 40% trophic eggs were included (laboratory: 20.2 eggs * 2.2; field: 27.0 eggs %

2.2; F Ratio 5.0, df 1, P = 0.03). When these egg batches were excluded, the difference was not statistically significant (laboratory: 23.1 eggs * 2.0; field: 27.0 * 2.0; F Ratio 1.8, df I, P = 0.18). The estimated difference was 3.8 eggs (-1.9, 9.5). It seems that, for some females, laboratory conditions induce smaller batch sizes and higher rates of infertility.

1.4 Male-killing bacteria and the proportion of trophic eggs laid

Female ladybirds are frequently infected with cytoplasmic male-killing (MK) agents that kill male embryos at an early stage of development (e.g., Majerus et al. 1998; Hurst et al. 1999). Non-developed male eggs are eaten by their hatching sisters. An additional fraction of ladybird eggs normally fails to develop and may be a maternal adaptation to feed offspring (trophic eggs; see Chapter 4). The null expectation for the percent hatch in egg batches from infected females is therefore half of the normal hatch rate (because, with the 5050 sex ratio observed in ladybirds (Majerus 1994), half of normally viable eggs would not develop). If female ladybirds could detect their infection status, and if they control trophic egg production, then infected females might benefit from reducing trophic egg production because half of their egg batches will be effectively 'trophic'. I review studies of MK bacteria in coccinellids that report hatching success in infected and uninfected females to test whether infected females reduce trophic egg production. The null hypothesis is that the proportion of hatching eggs from infected females will not differ from half of the hatching rate in eggs of uninfected females.

Methods

Six studies report hatchability in MK-infected and uninfected ladybirds (Tables I. 1 and 1.2). Some data from Hurst et al. (1999) could not be used because only one uninfected female line was followed, so a confidence interval could not be calculated. A hrther three studies report hatch rates of MK-infected lines, but do not report rates for non-infected lines for comparison (Hurst et al. 1993; Majerus et al. 2000; Schulenburg et al. 2001). Two studies report hatch rates and number of eggs for field-collected, laboratory-reared lines of infected and uninfected females (Table 1.1). For these studies, I calculated a weighted mean proportion hatch for uninfected lines and divided it in half to obtain the expected proportion hatch for infected lines. I then tested whether the 95% CI of the weighted mean proportion hatch in uninfected females encompassed the predicted value. Hurst et al. (1999) report hatch rates for field-collected ladybirds maintained in the laboratory and for subsequent crosses within infected and uninfected lines. I analyzed these datasets separately. Two additional studies report hatch rates, but not number of eggs. For these studies, I followed an equivalent procedure without weighting the mean proportion hatch. Another two studies report hatch rates for individual females before and after an antibiotic treatment that cures MK infection. I used a paired t-test to compare pre-antibiotic hatch rate with half of the post-antibiotic hatch rate. It is possible that the antibiotic treatment influenced hatch rate independently of the cure of MK infection. I compared pre- and post-antibiotic hatch rates for uninfected females (which were reported in both studies) using a paired t-test. The rates should not differ if antibiotics do not affect females other than curing MK infection.

Results and summary

In four of the seven tested datasets, females appeared to reduce trophic egg production when infected with MK agents (Tables I. 1,1.2). When cured of MK infection by antibiotic treatment, females increase trophic egg production (Table 1.2) as predicted if females control trophic egg production and are sensitive to their infection status. However, in one of the studies the antibiotic appeared to affect hatching rate independent of the cure of MK infection. The results do not allow the conclusion that females can detect MK infection status or that females control trophic egg production. However, that the trend appears in four of seven datasets is suggestive. If the hypothesis of reduction in trophic egg production with MK infection is correct, then this is another line of evidence demonstrating that trophic eggs are a maternal adaptation in ladybirds (see Chapter 2). An alternative hypothesis is that infection confers some fertility benefit such that non-adaptively infertile eggs are eliminated in infected females. This might occur because the vertically transmitted MK agents have an interest in promoting host fertility. More work is required in this respect. 1.5 Literature cited

Dixon, A. F. G. 2000. Insect predator-prey dynamics. Cambridge University Press, London.

Hurst, G. D. D., C. Bandi, L. Sacchi, A. G. Cochrane, D. Bertrand, I. Karaca, and M. E. N. Majerus. 1999. Adonia variegata (Coleoptera: Coccinellidae) bears maternally inherited Flavobacteria that kill males only. Parasitology 1 18: 125-134.

Hurst, G. D. D., T. C. Hammarton, J. J. Obrycki, T. M. 0. Majerus, L. E. Walker, D. Bertrand, and M. E. N. Majerus. 1996. Male-killing bacterium in a fifth ladybird beetle, Coleomegilla maculata (Coleoptera: Coccinellidae). Heredity 77: 177-185.

Hurst, G. D. D., M. E. N. Majerus, and L. E. Walker. 1992. Cytoplasmic male killing elements in Adalia bipunctata (Linnaeus) (Coleoptera: Coccinellidae). Heredity 69:84-91.

-. 1993. The importance of cytoplasmic male killing elements in natural populations of the two spot ladybird, Adalia bipunctata (Linnaeus) (Coleoptera: Coccinellidae). Biological Journal of the Linnean Society 49: 195-202.

Hurst, G. D. D., E. L. Puwis, J. J. Sloggett, and M. E. N. Majerus. 1994. The effect of infection with male-killing Rickettsia on the demography of female Adalia bipunctata L. (two spot ladybird). Heredity 73:309-3 16.

Majerus, M. E. N. 1994, Ladybirds. HarperCollins, London.

Majerus, M. E. N., J. H. G. V. D. Schulenburg, and I. A. Zakharov. 2000. Multiple causes of male-killing in a single sample of the two-spot ladybird, Adalia bipunctata (Coleoptera: Coccinellidae) from Moscow. Heredity 84:605-609.

Majerus, T. M. O., M. E. N. Majerus, B. Knowles, J. Wheeler, D. Bertrand, V. N. Kuznetzov, H. Ueno et al. 1998. Extreme variation in the prevalence of inherited male-killing microorganisms between three populations of Harmonia axyridis (Coleoptera: Coccinellidae). Heredity 8 1:683-691.

Matsuka, M., H. Hashi, and I. Okada. 1975. Abnormal sex-ratio found in the lady beetle, Harmonia axyridis Pallas (Coleoptera: Coccinellidae). Applied Entomology and Zoology 10:84-89.

Schulenburg, J. H. G. V. D., M. Habig, J. J. Sloggett, K. M. Webberley, D. Bertrand, G. D. D. Hurst, and M. E. N. Majerus. 2001. Incidence of male-killing Rickettsia spp. (a- Proteobacteria) in the ten-spot ladybird beetle Adalia decempunctata L. (Coleoptera: Coccinellidae). Applied and Environmental Microbiology 67:270-277.

Stewart, L. A., J.-L. Hemptinne, and A. F. G. Dixon. 1991. Reproductive tactics of ladybird beetles: relationships between egg size, ovariole number and developmental time. Functional Ecology 5:380-385.

Zar, J. H. 1999, Biostatistical analysis. Prentice Hall, New Jersey. Figure 1-1 Proportion of egg mass lost during the development of H. axyridis egg types. Trophic eggs lose significantly less mass than viable eggs (estimated difference and 95% CI: 0.025 (0.009,0.042)). Inviable eggs, where an embryo develops but does not emerge, lost an intermediate amount of mass. The initial mass of eggs was taken 12 hours after oviposition; final mass was taken in the 24 hours before hatching. I I Viable Inviable Trophic Figure 1-2 Hatching time in relation to egg mass within egg batch 60-15. Hatching time is expressed relative to the first hatching egg in the batch, which has a hatch time of zero. The initial mass was taken within 12 hours following oviposition. Heavier eggs tended to hatch sooner than lighter eggs, though clearly the trend has exceptions. Relative hatch time (minutes) Table 1-1 The evidence, across studies, that female ladybirds adjust trophic egg production depending on male-killing infection status. The expected proportion hatch is a weighted mean proportion of hatching eggs from uninfected lines, based on data reported in each study. The observed proportion is also a weighted mean. The null hypothesis is that the observed hatching proportion does not differ from half of the hatching proportion in uninfected lines. If females can control the production of trophic eggs, infected females may benefit from reducing trophic egg production because hatching larvae will have dead male eggs to eat. When the lower limit of the 95% CI of the observed proportion exceeds the expected proportion, the evidence is taken to support the alternate hypothesis. Species N eggs from N eggs Expected Observed Evidence for Source uninfected from proportion proportion hypothesis of lines (N MK hatch hatch in female lines) lines infected lines control of (N (95% CI) trophic eggs lines) Harmonia 3598 (33) 4762 0.376 0.413 (0.383, Supports Majerus axyridis (30) 0.442) et al. (1 998) Harmonia 3968 (12) 3598 0.341a 0.360 (0.261, Does not Matsuka axyridis (12) 0.458) support et al. (1 975) Coleomegilla 1206 (20) 1056 0.432 0.533 (0.444, Supports Hurst et maculata (6) 0.590) al. (1 996) Coleomegilla 1440 (12) 1590 0.340 0.477 (0.42 1, Supports Hurst et maculata (1 1) 0.533) al. (1 996) Adalia - (31) - (31) O.43ga 0.423a Does not Hurst et bipunctata (0.40 1, support al. 0.445) (1 994) a Proportions not weighted because sample sizes not given. Field-collected lines. Laboratory crosses of field-collected lines. Table 1-2 Evidence that females control trophic egg production from studies in which male- killing (MK) infected ladybirds were cured with antibiotic. The expected proportions for MK- infected lines before antibiotic treatment are calculated as half of the proportion of hatching eggs after antibiotic treatment. If the observed proportion is greater than the expected proportion, females appear to reduce trophic egg production when infected. For lines that were uninfected before treatment, the expected hatching proportion is the proportion observed following antibiotic treatment. If the observed proportion is greater than the expected, the antibiotic causes some reduction in trophic egg production regardless of MK infection status. Study N eggs N eggs Mean Expected P-value Evidence for from lines from lines proportion proportion (Test hypothesis of before after hatched (SE) statistic) female antibiotic antibiotic before control of

treatment treatment treatment trophic eggs- - (N lines) (N lines) (SE) (a)Adonia variegata (Hurst et al. 1999) MK- 1084(6) 1014(6) 0.401 0.349 0.014 Supports infected (0.032) (0.044) (t5=3.3) lines Uninfected 591 (5) 728 (5) 0.901 0.854 0.04 1 Does not lines (0.01 6) (0.023) (t4=2.3) support (b)Adalia bipunctata (Hurst et al. 1992) MK- - (9) - (9) 0.464 0.341 0.004 Supports infected (0.038) (0.020) (te=3.2) lines Uninfected - (5) - (5) 0.826 0.850 0.287 Supports lines (0.036) (0.032) (t4=0.7)