Individual differences in behavioural plasticities.

1

Biol. Rev. (2015), pp. 000–000. doi: 10.1111/brv.12186

Individual differences in behavioural plasticities Judy A. Stamps∗ Department of Ecology and Evolution, University of California Davis, Davis, CA 95616, U.S.A.

ABSTRACT Interest in individual differences in animal behavioural plasticities has surged in recent years, but research in this area has been hampered by semantic confusion as different investigators use the same terms (e.g. plasticity, flexibility, responsiveness) to refer to different phenomena. The first goal of this review is to suggest a framework for categorizing the many different types of behavioural plasticities, describe examples of each, and indicate why using reversibility as a criterion for categorizing behavioural plasticities is problematic. This framework is then used to address a number of timely questions about individual differences in behavioural plasticities. One set of questions concerns the experimental designs that can be used to study individual differences in various types of behavioural plasticities. Although within-individual designs are the default option for empirical studies of many types of behavioural plasticities, in some situations (e.g. when experience at an early age affects the behaviour expressed at subsequent ages), ‘replicate individual’ designs can provide useful insights into individual differences in behavioural plasticities. To date, researchers using within-individual and replicate individual designs have documented individual differences in all of the major categories of behavioural plasticities described herein. Another important question is whether and how different types of behavioural plasticities are related to one another. Currently there is empirical evidence that many behavioural plasticities [e.g. contextual plasticity, learning rates, IIV (intra-individual variability), endogenous plasticities, ontogenetic plasticities) can themselves vary as a function of experiences earlier in life, that is, many types of behavioural plasticity are themselves developmentally plastic. These findings support the assumption that differences among individuals in prior experiences may contribute to individual differences in behavioural plasticities observed at a given age. Several authors have predicted correlations across individuals between different types of behavioural plasticities, i.e. that some individuals will be generally more plastic than others. However, empirical support for most of these predictions, including indirect evidence from studies of relationships between personality traits and plasticities, is currently sparse and equivocal. The final section of this review suggests how an appreciation of the similarities and differences between different types of behavioural plasticities may help theoreticians formulate testable models to explain the evolution of individual differences in behavioural plasticities and the evolutionary and ecological consequences of individual differences in behavioural plasticities. Key words: flexibility, IIV, developmental plasticity, contextual plasticity, learning, life-cycle staging, realized plasticity, potential plasticity, temporal plasticity, ontogenetic plasticity. CONTENTS I. Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2 II. Classifying behavioural plasticities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3 (1) Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3 (2) Contextual plasticities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5 (3) Developmental plasticities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6 (4) Endogenous plasticities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6 (5) Temporal plasticities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7 (6) Sources of behavioural plasticities: how internal state, external stimuli and past experiences interact to affect current behaviour . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8 (7) Why using reversibility as a criterion for categorizing plasticities is problematic . . . . . . . . . . . . . . . . . 10 * Address for correspondence (Tel: 510 845 1603; E-mail: [email protected]). Biological Reviews (2015) 000–000 © 2015 Cambridge Philosophical Society

Judy A. Stamps

2

III. Quantifying individual differences in behavioural plasticities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (1) General concerns . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (2) Experimental designs for studying individual differences in behavioural plasticities . . . . . . . . . . . . (a) Within-individual versus replicate-individual designs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (b) Experimental designs for studying individual differences in developmental behavioural plasticities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . IV. Individual differences in behavioural plasticities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . V. Effects of prior experience on behavioural plasticities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . VI. Relationships between different types of behavioural plasticities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (1) Background and hypotheses . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (2) Using indicator traits to infer relationships between behavioural plasticities . . . . . . . . . . . . . . . . . . . . . (3) Direct tests of relationships across individuals between behavioural plasticities . . . . . . . . . . . . . . . . . . VII. Implications for the evolutionary and ecological causes and consequences of behavioural plasticities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . VIII. Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . IX. Acknowledgements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . X. References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

I. INTRODUCTION Behaviour is among the most labile of traits, and for many years biologists and psychologists have described many different ways that behaviour can vary within individuals as a function of variation in external or internal stimuli. In recent years, researchers have increasingly focused on differences across individuals or genotypes in within-individual behavioural variability, as evidenced by the number of symposia and special issues dedicated to these topics (see Ellis & Bjorklund, 2012; Foster, 2013; Kappeler et al., 2013). However, a significant barrier to progress in this area is that many of the terms used to describe within-individual behavioural variability are ambiguous. The term ‘plasticity’ itself has multiple meanings. In ecology, evolutionary biology and developmental psychology, plasticity often refers to the effects of exposure to external stimuli, experiences or environments early in life on the behaviour expressed later in life (Marler, 1997; Debat & David, 2001; Piersma & Drent, 2003; West-Eberhard, 2003; Belsky & Pluess, 2009). By contrast, neuropsychologists often use the term plasticity to refer to any effects of past experience on behaviour (Kolb & Gibb, 2014). In behavioural ecology, plasticity may refer to the extent to which an individual’s behaviour changes in different situations (Sih et al., 2004; Briffa & Bibost, 2009), or it may refer to any type of within-individual change in behaviour (Dingemanse et al., 2010; Mathot et al., 2012; Brommer, 2013a). Similarly, the meaning of the term ‘flexibility’ varies widely among authors. Examples include a reversible change in any phenotypic trait (Piersma & Drent, 2003), the ability to reversibly and adaptively switch between different behaviours in different contexts (Duckworth, 2010; Holekamp, Swanson & Van Meter, 2013), and the extent to which an individual changes its behaviour in response to a change in a well-learned stimulus situation (Coppens, de Boer & Koolhaas, 2010).

10 10 13 13 17 19 21 23 23 24 25 26 29 30 31

As a result of this semantic confusion, it is currently difficult to tell whether the results of one empirical study of behavioural plasticity are relevant to another, whether theoretical models based on assumptions appropriate for one type of behavioural plasticity are relevant to other types of behavioural plasticities, and whether or how different types of behavioural plasticities might be related to one another. Hence, the first goal herein is to review the major ways that behaviour can vary within individuals or genotypes. The proposed framework for categorizing behavioural plasticities draws on concepts that have for many years provided a foundation for research in ethology, psychology and related fields, and it avoids using criteria for classification (e.g. reversibility) that have proven problematic for categorizing labile behavioural or physiological traits. The major advantage of having terms to discriminate among different types of behavioural plasticity is that it allows us to formulate and address a number of important questions. The first of these is identifying experimental designs that are appropriate for describing and quantifying individual differences in different types of behavioural plasticities. There are several reasons why this is not as straightforward a process as one might suppose. On the one hand, the widespread use of the same statistical tools (e.g. mixed-effects models, Dingemanse & Dochtermann, 2013), and the same concepts (e.g. behavioural reaction norms, Dingemanse et al., 2010; Stamps & Groothuis, 2010a) to study many different types of behavioural plasticities seems to have encouraged some researchers to assume that different types of behavioural plasticities are more similar to one another than is actually the case. Conversely, when researchers working in different disciplines study the same type of behavioural plasticity but rely on different experimental designs, they may not notice that the literature of other disciplines contains concepts, insights and results relevant to their own. Thus, a second goal is to discuss the

Biological Reviews (2015) 000–000 © 2015 Cambridge Philosophical Society

Behavioural plasticities

3

experimental designs that have been used, or that could be used in the future, to study individual or genotypic differences in various types of behavioural plasticities. A major advantage of having terms to discriminate among behavioural plasticities is that it allows us to determine whether there is empirical support for individual or genotypic differences in specific types of behavioural plasticities. A brief overview of the literature on this topic is provided, demonstrating that, to date, significant differences among individuals or genotypes have been reported for all the major categories of behavioural plasticity. A more complicated but also more important issue is how different types of behavioural plasticities are related to one another. The first question to consider is whether a type of behavioural plasticity expressed at one age is affected by experiences earlier in life, i.e. whether various types of behavioural plasticities are themselves developmentally plastic. For instance, one might ask whether within-individual variation in courtship behaviour as a function of variation in current light intensity depends upon the light intensity an individual experienced in its rearing environment (Chapman, Morrell & Krause, 2009), or whether learning rates for a given task at a given age are affected by exposure to stressful events earlier in life (Colas-Zelin et al., 2012). The question of whether a given type of behavioural plasticity is itself developmentally plastic is important because if this is true, at least some of the variation across individuals or across genotypes in that type of behavioural plasticity at a given age could be due to differences among those individuals in experiences earlier in life (Stamps & Groothuis, 2010b; Dingemanse & Wolf, 2013; Snell-Rood, Davidowitz & Papaj, 2013). Hence, we consider the empirical evidence that various types of behavioural plasticities can be affected by experiences earlier in life. Another timely question about relationships between behavioural plasticities is whether different types of behavioural plasticities are correlated with one another across individuals or genotypes, i.e. whether some individuals are generally more plastic than others. We review hypotheses that predict correlations between different types of behavioural plasticities, i.e. based on the assumption that individuals differ with respect to their sensitivity (or responsiveness) to external stimuli, or that individuals differ with respect to their general learning ability. Empirical support for these hypotheses in animals is considered, including studies that measure relationships between different types of behavioural plasticities and the same ‘indicator trait’, or that ask whether learning rates for different tasks are correlated with one another across individuals. Finally, we consider how an appreciation of the similarities and differences between different types of behavioural plasticities may help theoreticians formulate testable models to explain the evolution of

individual differences in behavioural plasticities and the evolutionary and ecological consequences of such differences.

II. CLASSIFYING BEHAVIOURAL PLASTICITIES (1) Introduction Although many authors have suggested ways to define and categorize various types of behavioural plasticities (see Sections I and II.7), investigators working in different disciplines have understandably focused on concepts and definitions that help them address questions within their own discipline. As a result, we currently lack a comprehensive framework that covers all of the major types of within-individual variation in behaviour that have been reported in animals. The first goal of this review is to provide such a framework. In service of that goal, it is necessary to designate terms to describe different types of behavioural plasticities (Table 1). Of course, the words selected to describe different types of plasticities are a matter of taste. However, a coherent, consistent set of terms is essential for any discussion of the similarities and differences between behavioural plasticities, and how they relate to one another within and across individuals. The framework advocated herein focuses on the extent to which external or internal stimuli contribute to within-individual variation in behaviour (Table 2). It divides plasticities into two broad categories, ‘exogenous plasticity’ and ‘endogenous plasticity’, where exogenous plasticity describes how behaviour varies within agents (individuals or genotypes) in response to variation in external factors, and endogenous plasticity describes how behaviour varies within agents in response to spontaneous changes in their internal state. In turn, exogenous plasticity is divided into two categories, contextual and developmental, depending on whether behaviour changes as an immediate response to current external stimuli, or whether it varies as a function of stimuli or experiences that occurred in the past. Exogenous plasticity includes any situation in which factors external to the subjects (stimuli, experiences, environments) affect their behaviour. The distinction between contextual and developmental plasticity (Stamps & Groothuis, 2010a) is a formalization of a distinction emphasized years ago by ethologists and psychologists. Ethologists have long differentiated between stimulus–response relationships that specify the immediate effects of external stimuli on behaviour (e.g. ‘releasing mechanisms’) versus the effects of experiences in the past on the development of behaviour (Tinbergen, 1951; Hogan, 2005; Hogan & Bolhuis, 2005). Similarly, psychologists have for many years distinguished between the immediate effects of stimuli on behaviour (e.g. psychophysics and stimulus discrimination) versus the effects of experiences in the past


Judy A. Stamps

4

Table 1. Glossary of terms. Many of these terms apply not only to behaviour, but also to other components of the phenotype, including labile physiological traits. However, since this review focuses on behaviour, the definitions are illustrated using examples from behaviour Term

Definition

Agent

An individual, or a group of individuals with the same genotype, raised under the same conditions prior to the onset of an experiment An equation that describes the relationship between an agent’s behaviour and a continuously distributed independent variable The extent to which the phenotype of an agent varies as an immediate response to variation in external stimuli (e.g. stimulus–response relationships, mate preference functions) The extent to which the current phenotype of an agent varies as a function of external experiences, stimuli or environmental conditions that occurred in the past (e.g. learning, ontogenetic plasticity, life-cycle staging) Change in an agent’s phenotype as a function of its age Temporal variation in phenotype caused by ‘spontaneous’ variation in internal state, in the absence of any variation in external stimuli (e.g. circadian rhythms, IIV, age-dependent changes in behaviour) Variation in phenotype caused by variation in external stimuli, experiences or environments (e.g. contextual or developmental plasticity) The extent to which the behaviour an agent expresses in a well-learned situation changes in response to a change in the stimulus situation (e.g. routine-formation, reversal learning) Short-term unpredictable fluctuations in within-individual behaviour, often assumed to result from spontaneous fluctuations in internal state The extent to which the phenotype of an individual varies each season across its lifetime as a function of exposure to external stimuli prior to that season The extent to which experience at a specific age or life stage affects the phenotype expressed at a later age or life stage (e.g. ‘classical’ developmental plasticity, filial imprinting) The ability of an agent to generate different phenotypes as a result of exposure to different external stimuli, internal stimuli or past experiences The extent to which an agent’s phenotype changes in response to changes in specific external stimuli, internal stimuli, or past experiences Individuals with the same genotype, raised in the same environment, from a line founded by individuals randomly collected from a natural population and not subsequently subjected to artificial selection The extent to which the phenotype of an agent changes as a function of age or time (e.g. behavioural developmental trajectories for free-living animals)

Behavioural reaction norm Contextual plasticity Developmental plasticity Developmental trajectory Endogenous plasticity Exogenous plasticity Flexibility Intra-individual variability (IIV) Life-cycle staging Ontogenetic plasticity Potential plasticity Realized plasticity Replicate individual Temporal plasticity

on the behaviour expressed at the current time (e.g. learning) (Shettleworth, 2010). The concept of endogenous plasticity emphasizes that behaviour can change over time within individuals even in the absence of any temporal variation in external stimuli, as a result of spontaneous, internally driven changes in physiological and or morphological state (e.g. circadian rhythms). Of course, in some cases the factors responsible for within-individual changes in behaviour are either unknown or not of interest. In such cases, the term ‘temporal plasticity’ can be used to describe how an individual’s behaviour changes over time, without regard to variation in external stimuli or internal state that might have contributed to those changes. Another useful distinction, which applies to any type of behavioural plasticity, is between ‘potential plasticity’, which refers to the ability of a given agent to vary its phenotype in response to variation in stimuli, experiences, or environmental conditions, and ‘realized

plasticity’, which refers to the extent to which an agent actually varies its phenotype in response to variation in a specific stimulus, experience, or environmental condition. Potential plasticity is a construct that is relevant to theory on the evolution and ecological significance of individual differences in plasticity. For instance, theoreticians have suggested that agents with high potential plasticity may pay costs of maintaining the ‘machinery’ that allows them to detect, monitor, and respond to different stimuli, and that these maintenance costs of being plastic must be paid even if the agent never expresses that plasticity (DeWitt, Sih & Wilson, 1998; Callahan, Maughan & Steiner, 2008; Auld, Agrawal & Relyea, 2010). By contrast, realized plasticity is what empiricists actually measure in their experiments. Concepts similar to potential and realized plasticity have been mentioned in passing by several authors. Thus Ydenberg & Prins (2012) defined ‘flexibility’ as the ability to adjust foraging behaviour as circumstances



5

Table 2. An outline of behavioural plasticities. Major categories are illustrated with examples discussed in the text A. Exogenous plasticities (responses to external stimuli, experiences, environments) 1. Contextual plasticities (immediate effects of external stimuli on behaviour) (a) Mate preference functions (b) Stimulus–response relationships 2. Developmental plasticities (effects of past experience on current behaviour) (a) Learning (1) Habituation (2) Flexibility (b) Acclimation (c) Life-cycle staging (d) Ontogenetic plasticity (1) Effects of rearing conditions on adult behaviour (2) Filial imprinting B. Endogenous plasticities (spontaneous changes in behaviour resulting from temporal changes in internal stimuli) 1. Circadian rhythms 2. Circannual rhythms 3. Intra-individual variability (IIV) 4. Age-dependent changes in behaviour

change, but then noted that flexible individuals might not change their behaviour if their original behaviour performed well under the new set of conditions. Similarly, in a theoretical model of the effects of phenotypic plasticity on population dynamics, Gomez-Mestre & Jovani (2013) distinguished between the range of phenotypes that an individual is able to generate (plasticity–range), and the extent to which an individual’s phenotype actually changes in a given situation (plasticity–used). The distinction between potential and realized plasticity is relevant to empirical studies of individual differences in behavioural plasticities because different methods of estimating behavioural plasticities can lead to different estimates of the potential plasticity of the same set of individuals or genotypes. Since this point can be most easily appreciated by comparing the results of different experimental protocols that are currently used to describe the developmental plasticity of behaviour, discussion of this topic is deferred to Section III.2b. (2) Contextual plasticities Contextual plasticity indicates the extent to which an individual’s behaviour changes as a function of the stimuli that surround the individual at the time it expresses that behaviour (Stamps & Groothuis, 2010a). The concept of contextual plasticity is based on the premise that the neural and hormonal mechanisms that exist within an individual at a given time determine how that individual will immediately respond if presented with

different external stimuli. This idea is captured by an alternative term for contextual plasticity, ‘activational plasticity’ (Snell-Rood, 2013), which emphasizes that external stimuli at the current time ‘activate’ neural and hormonal mechanisms which developed as a function of stimuli and experiences earlier in an individual’s life. Examples of contextual plasticity include changes in running speed as a function of current temperature in ants (Andrew et al., 2013), or immediate changes in the pitch of avian vocalizations in response to changes in ambient noise (Potvin & Mulder, 2013). Generally speaking, it is easiest to measure contextual plasticity as a function of variation in one stimulus when other stimuli are held constant, since an individual’s response to variation in one type of external stimulus (e.g. responses to different concentrations of predator odour) may depend on the current values of other external stimuli (e.g. the presence or absence of nearby conspecifics, see ‘multi-dimensional plasticity: Westneat et al., 2011; Dosmann & Mateo, 2014). In recent years, data on contextual plasticity have come from an unexpected source: studies of preference, including studies of mate choice. In particular, no-choice (alternatively, ‘one-choice’) behavioural assays for studying preference provide direct estimates of contextual plasticity, since they rely on the same experimental protocols that are used to study any other type of contextual plasticity. In no-choice experimental designs, each subject is presented with one stimulus at a time, and its response to that stimulus is measured. This process is then repeated for different stimuli. Hence, this protocol measures each subject’s immediate response to each of a set of different stimuli, and these responses are then used to infer the extent to which that subject ’prefers’ each of those stimuli. No-choice assays have been used to assess ‘preferences’ for many different types of items, including food types, hosts, habitats and potential mates. If the stimuli from different items can be arranged along a gradient, the data from no-choice tests can be used to generate ‘preference functions’ (see also Section III.1). No-choice designs differ in a number of ways from the other important protocol used to measure preference: simultaneous two-choice (or multiple-choice) assays. In simultaneous-choice tests, each subject is presented with two (or more) stimuli at the same time, and is required to choose one of them. In this case, preference scores are typically based on the proportion of times that an individual chooses each of the stimuli. No-choice and simultaneous two-choice designs sometimes generate different estimates of preference (e.g. Peso, Telford & Backwell, 2014). This is not surprising, given that one-choice assays simply record an individual’s immediate response to each of a set of stimuli, whereas simultaneous-choice assays require the subjects to assess two or more stimuli, compare them with one another, and then make the decision which results


Judy A. Stamps

6 in choosing one of them. Recent studies suggest that simultaneous comparison and choice involve high-level neurological processes which are not required for the simpler stimulus–response relationships that are measured using no-choice protocols (e.g. see Grabenhorst & Rolls, 2011). Hence, although no-choice experimental protocols for measuring preference can provide data on individual differences in contextual plasticity, simultaneous-choice protocols are less suitable for this purpose. (3) Developmental plasticities Developmental plasticity indicates the extent to which an individual’s behaviour at the current time is affected by past experiences (Stamps & Groothuis, 2010a; Snell-Rood, 2013). The concept of developmental plasticity is based on the premise that an individual’s neural and hormonal state at the current time can be affected by the individual’s experiences in the past. Because there are so many different ways that past experiences can affect current behaviour, the term developmental plasticity as it is applied to behaviour encompasses a very wide range of phenomena, including the many forms of learning (Shettleworth, 2010; Dukas, 2013), as well as examples of ‘classical’ developmental plasticity that are familiar to those studying morphological traits, e.g. how genotypes alter their developmental trajectories in response to exposure to different environmental conditions early in life (Pigliucci, 2001; West-Eberhard, 2003; Moczek et al., 2011). Some studies of behavioural developmental plasticity focus on situations in which long periods elapse between the time the experience occurs and the time behaviour is measured (e.g. the effects of stimuli from mothers early in life on adult behaviour; Weaver et al., 2004; Champagne, 2013; Bateson, Gluckman & Hanson, 2014), or on situations in which prolonged exposure to particular types of experiential factors affects the development of behaviour (e.g. how rearing larvae at different densities affects the courtship signals they produce as adults; Zhou et al., 2008). Other examples of behavioural developmental plasticity consider phenomena that occur over much shorter periods of time, e.g. studies which consider how stimuli to which an individual was exposed a few minutes to days earlier affect that individual’s current behaviour (e.g. habituation to cues from a conspecific; Bell & Peeke, 2012), or acclimation to a novel environment (Biro, 2012; Dingemanse et al., 2012a). Researchers typically use specific terms (e.g. habituation, acclimation, discrimination learning, flexibility, reversal learning, etc.) when referring to behavioural developmental plasticities which occur over the short term (i.e. situations in which experiences in the recent past affect current behaviour), and which can occur at many different ages over the lifetime of the subjects. By contrast, the generic term ‘developmental plasticity’ is more often used to refer to situations in

which experiences early in life affect the behaviour expressed in adulthood (see Westneat, Wright & Dingemanse, 2014). To avoid potential confusion, herein the term ‘developmental plasticity’ refers to any situation in which past experience affects current behaviour, and the term ‘ontogenetic plasticity’ to situations in which experience at a given age or life stage affects the behaviour expressed at a later age or life stage (see Tables 1 and 2). Many studies of ontogenetic plasticity focus on long-term developmental plasticities, i.e. situations in which experience early in life (often during the prenatal or juvenile period) affects one or more types of behaviour expressed much later in life (often during adulthood). However, this is not always the case. For instance, the extensive literature on avian filial imprinting considers how experiences that occur during a brief period around the time of hatching affect behaviour that is expressed days to weeks later, during the early juvenile period (Bateson, 1966; Bolhuis, 1991; Harshaw & Lickliter, 2011). Similarly, psychologists working with rodents have found that exposure to social cues and stressors during the adolescent period can affect a number of behaviours expressed after maturity (Delville, Melloni & Ferris, 1998; Weintraub, Singaravelu & Bhatnagar, 2010; Hoeve et al., 2013). Much of the research on ontogenetic plasticity is motivated by the assumption that ‘age matters’, i.e. that experience at a given age or life stage often has a different effect on subsequent behaviour than if that same experience had occurred at a different age or life stage (see the concept of sensitive periods; Bateson, 1979; Knudsen, 2004). As a result, empiricists studying ontogenetic plasticities carefully consider the age(s) at which the experience occurs and the age(s) at which the behaviour is measured, even if sensitive periods per se are not the focus of their study. (4) Endogenous plasticities Of course, any type of behavioural plasticity necessarily involves changes in an individual’s internal state, since changes in neural activity, hormonal levels, gene expression, etc. are required to produce any change in behaviour. By extension, any type of exogenous plasticity necessarily involves immediate or longer-term changes in some aspect of an individual’s internal state. However, it is also clear that behaviour can spontaneously change over time within individuals even if all external stimuli are held constant (e.g. circadian rhythms, see also below). Situations in which behaviour spontaneously varies within individuals in the absence of any variation in exogenous stimuli imply that temporal changes in behaviour can occur as a result of self-generated variation in internal stimuli. The term ‘endogenous plasticity’ is used to refer to these situations. Some types of endogenous plasticity are systematic, in the sense that they can be modelled using various



7

mathematical functions. Thus, even if animals are maintained under constant conditions (little or no variation in external stimuli), their behaviour can systematically change as a function of the time of day (circadian rhythms; Rusak & Zucker, 1975; Muraro, Pirez & Ceriani, 2013). Over longer periods of time, behaviour can systematically change as a function of age or maturational state (Ba & Seri, 1995; Zou et al., 2011; Olivier et al., 2013, see also Section II.5), or as a function of time of year (circannual rhythms; Gwinner, 2003; Rani & Kumar, 2013) even if external stimuli are held strictly constant throughout the period when behaviour is being measured. Other types of internal processes generate random or pseudo-random variability in behaviour. A prime example is intra-individual variability (IIV). For many years, psychologists have studied short-term, within-individual stochastic variability in behaviour, which they called IIV (Fiske & Rice, 1955; Nesselroade, 1991; Ram & Gerstorf, 2009). Studies of animals indicate that IIV varies among individuals (Stamps, Briffa & Biro, 2012) and is repeatable (Biro & Adriaenssens, 2013). Several authors have suggested reasons why IIV might be adaptive, although empirical evidence on this point is still sparse (Stamps et al., 2012; Briffa, 2013; Westneat et al., 2014). In humans, there is growing evidence that IIV in behaviour is generated by spontaneous fluctuations in brain activity (MacDonald, Nyberg & Backman, 2006; MacDonald, Li & Backman, 2009; Papenberg et al., 2013), supporting the assumption that IIV is caused by spontaneous temporal changes in internal state, as opposed to unmeasured, uncontrolled fluctuations in external stimuli. Studies of endogenous plasticity typically begin with observations of within-individual temporal changes in behaviour that occur in the absence of changes in external stimuli. Such observations may then lead to investigation of the proximate mechanisms that are responsible for generating those changes. For instance, interest in circadian rhythms began when investigators noticed that the behaviour or physiology of their subjects varied over time, even if they were maintained under carefully controlled, apparently constant environmental conditions. Once it was clear that circadian rhythms persisted even in the absence of temporal variation in external stimuli, physiologists began to investigate the neurological and hormonal processes responsible for generating them (e.g. Refinetti, 2006; Dibner, Schibler & Albrecht, 2010). In many cases, investigators studying endogenous plasticities simply describe changes in behaviour that occur when animals are maintained under constant conditions, without attempting to identify the changes in internal state that generate those changes. For example, Kasumovic, Hall & Brooks (2012) demonstrated predictable, systematic changes across the lifetime in the daily calling effort of adult male crickets, when

those males were maintained under constant conditions in the laboratory. At this point, however, studies of the hormonal and neurological processes that contribute to age-dependent behavioural changes in adult insects are still relatively rare (but see Anton, Dufour & Gadenne, 2007; Duportets et al., 2013), and no one has investigated the changes in internal state that contribute to spontaneous changes in the age-specific calling behaviour of crickets. (5) Temporal plasticities Behaviour often changes over time within individuals for reasons that are, at least initially, unknown. Here we will use the term ‘temporal plasticity’ to refer to within-individual changes in behaviour as a function of age, time, months, years or any other temporal variable. In contrast to the other types of plasticity described above, temporal plasticity simply describes how behaviour changes as a function of time; it says nothing about the factors responsible for those changes. That is, even when time is an independent or ‘predictive’ variable in statistical analyses of within-individual variability in behaviour (e.g. Stamps et al., 2012; Han & Brooks, 2014), we do not assume that time per se causes variation in behaviour. Instead, observations indicating that behaviour changes over time imply that some (and perhaps all) of the factors outlined in the previous paragraphs contributed to the changes over time in the behaviour of the subjects. Recently, for example Kluen & Brommer (2013) described seasonal changes in various types of behaviour (e.g. activity, neophobia) in free-living blue tits (Cyanistes caeruleus). They tested each individual using standard assays during the breeding season and during the winter, and reported differences across individuals in temporal plasticity, in this case measured as the difference between the summer and the winter scores of each individual. These individual differences in temporal plasticity could have been due to individual differences in (i) responses to minor differences in the position of the test arena during the summer and winter tests (contextual plasticity), (ii) responses to seasonal differences in temperature, day length, food regime or other factors that occurred in the weeks to months prior to the tests (developmental plasticity), (iii) circannual rhythms (endogenous plasticity), or (most likely) interactions among some or all of these factors. Data on individual differences in temporal plasticities are common in field studies (Brommer, 2013a), since many of the factors that contribute to changes in behaviour over time cannot be measured, controlled, or manipulated in free-living animals. However, psychologists face a similar problem, since human subjects cannot be reared and maintained under controlled conditions in the laboratory. As a result, psychologists have generated a number of useful terms for describing different forms of temporal plasticities at the group or


Judy A. Stamps

8 the individual level. For instance, mean-level consistency describes the extent to which the mean scores for the members of a group change over time, differential consistency describes the extent to which individual differences in behaviour are maintained over time, rank order consistency describes differential consistency when relationships among the scores of different subjects are described using ordinal rather than interval scales, and individual stability describes the extent to which the behaviour of a single individual changes over time (Caspi & Roberts, 2001; Roberts, Caspi & Moffitt, 2001; De Fruyt et al., 2006; review in Stamps & Groothuis, 2010a). Behavioural developmental trajectories (systematic changes in behaviour as a function of age) provide a prime example of a type of temporal plasticity that can be affected by any or all of the processes described above. If animals are maintained in social isolation, under strictly controlled environmental conditions over the period when behaviour is measured (e.g. Carey et al., 2006), then it would be reasonable to assume that spontaneous age-related changes in their internal state contributed to age-related changes in their behaviour (see Section II.4). However, it is also obvious that no animal can be reared or maintained in the absence of any external stimuli. Once one begins to list all of the environmental conditions (e.g. temperature, light, food regime, cage sizes) that applied in a given experiment, it becomes apparent that the behavioural developmental trajectories observed in that experiment might have been different if the same subjects had been reared or maintained under a different set of environmental conditions. For example, in larval sculpin (Myoxocephalus scorpius), the developmental trajectories for critical swimming speed differed for subjects raised at 3∘ C versus subjects reared at 6∘ C (Guan, Snelgrove & Gamperl, 2008). Thus, even if all environmental stimuli are held strictly constant, age-related changes in behaviour are not only affected by spontaneous changes over time in the subjects’ internal state (endogenous plasticity), but also by the cumulative effects of that set of environmental stimuli on their current behaviour (developmental plasticity) (see also Section III.2b). When animals are raised or maintained in groups, or are periodically exposed to conspecifics (e.g. Sussman & Ha, 2011; Han & Brooks, 2014), it is even more difficult to untangle the effects of exogenous and endogenous factors on behavioural developmental trajectories. It is clear that stimuli from conspecifics can affect behavioural developmental trajectories, as evidenced by the many empirical studies which have documented differences in the behavioural trajectories of group-reared versus isolation-reared animals (Hood & Cairns, 1989; Mimura, Nakamura & Koshiba, 2013; Ballen, Shine & Olsson, 2014; Hesse & Thunken, 2014). However, it is also clear that stimuli from conspecifics cannot be controlled to the same extent as one can

control temperature regimes, light levels or other environmental factors in laboratory studies of behavioural developmental trajectories. As a result, there is no way to know whether or how unobserved and unmeasured age-related changes in stimuli from a focal individual’s social companions affected the developmental behavioural trajectory of that individual. Raising and testing subjects in the absence of conspecifics can alleviate this problem, but this is not an option for highly social species, which are apt to develop behavioural pathologies if raised in isolation (Fone & Porkess, 2008). Of course, when behavioural developmental trajectories are measured for free-living animals (e.g. Rivera-Gutierrez, Pinxten & Eens, 2012; Petelle et al., 2013), investigators have no control over the environmental or social stimuli to which the individuals were exposed over the study period. In that situation, the best one can say is that exogenous and endogenous factors might both have contributed to observed age-related changes in behaviour. (6) Sources of behavioural plasticities: how internal state, external stimuli and past experiences interact to affect current behaviour One of the more important points about behavioural plasticities is that all of them can occur within the same individual. Later in this review we will consider questions involving relationships between different plasticities within individuals, e.g. whether behavioural plasticities expressed at a given age are themselves developmentally plastic (Section V) and whether some individuals are generally more plastic than others (Section VI). However, in order to appreciate how behavioural plasticities might be related to one another within individuals, we first need to consider, in the most general terms, how internal state, external stimuli and past experiences interact with one another to affect the behaviour expressed by an individual at a given age and time. Figure 1 provides a schematic to aid in this discussion. The external stimuli that impinge on an individual at a given moment interact with its internal state at that moment, resulting in the expression of behaviour (BehaviourT , Fig. 1). By extension, if a given individual is exposed to different external stimuli (different values of External StimuliT ), it may respond differently to them. The extent to which an individual’s immediate responses to external stimuli vary as a function of variation in those stimuli is indicated by its contextual plasticity. Conversely, an individual’s internal state may spontaneously change across time (variation in Internal StateT ), even if it is maintained under constant conditions. In that case, even if an individual is exposed to the same external stimuli on different occasions, it may express different behaviour. Within-individual variation in behaviour as a function of spontaneous variation in internal state is captured by the term endogenous plasticity. In practice, of course, spontaneous changes



Experience A

Experience B

Internal State T-2

Internal State T-1

9

External Stimuli T

Internal State T

Behaviour T

Present

Past

TIME

Fig. 1. A schematic illustrating how external stimuli and experiences interact with an individual’s internal state to affect current behaviour. The behaviour an individual expresses at the present time (T) can change as a result of variation in the external stimuli that impinge upon that individual at that time (contextual plasticity); it can also change as a result of spontaneous temporal changes in its internal state (endogenous plasticity). Finally, an individual’s response to external stimuli at the present time T can change as a function of changes in its internal state induced by the cumulative effects of various types of experiences (A or B) that occurred in the recent past (at T − 1) and the more distant past (at T − 2) (developmental plasticity). Changes in an individual’s behaviour over time (temporal plasticity) can occur as a result of any or all of these processes.

in internal state and external stimuli interact with one another to affect the behaviour expressed at a given moment. For instance, larval zebrafish (Danio rerio) exhibit circadian rhythms in their responsiveness to light: the contextual plasticity of their behavioural and neurological responses to light is much higher during the subjective day than during the subjective night (Emran et al., 2010). Importantly, an individual’s internal state at the current time is also affected by experiences it had in the past, as well as by its internal state in the past. Experiences in the immediate past (e.g. Experience B) and in the more distant past (e.g. Experience A), can, via their cumulative effects on an individual’s internal state, affect the behaviour expressed at the present time. The effects of past experience on current behaviour are the focus of studies of developmental plasticity. Typically, researchers studying learning or acclimation consider the effects of experience in the recent past on behaviour, whereas researchers studying ontogenetic plasticity often consider the effects of experience early in life on the behaviour expressed later in life. In addition, of course, an individual’s internal state at the present time is also affected by its own internal state in the past. For instance, an individual’s hormonal profile and behaviour during and after sexual maturity is affected by physiological changes that occurred months to years earlier in life (Collaer & Hines, 1995; Nelson, 2011; Juraska, Sisk & DonCarlos, 2013). Usually, however, the effects of earlier internal state on current internal state are not included under the term

developmental plasticity (but see West-Eberhard, 2005). Finally, one can simply describe how an individual’s behaviour changes as a function of age or time, without specifying or identifying the factors that might be responsible for those changes (temporal plasticity). Figure 1 illustrates why it is important to distinguish the effects of past experiences on current behaviour (developmental plasticity) from the effects of current external stimuli on behaviour (contextual plasticity). Distinguishing between these requires different experimental designs than those typically used to study the plasticity of morphological or life-history traits, in which contextual plasticity is not an option. For instance, in a widely cited study Relyea (2002) investigated the plasticity of several morphological and life-history traits and one behavioural trait (activity rate) in larval wood frogs (Rana sylvatica). He used a half-sib design, and raised members of each family in each of two types of mesocosms: in the presence of cues from predators, or in the absence of cues from predators. The activity of each group of larvae was periodically recorded within each mesocosm over the rearing period. Although there were clear differences among the families in the effects of predator cues on activity (here measured as the difference, for each family, between their mean activity scores in the predator-present and the predator-absent mesocosms), it is not possible to tell whether these differences were a result of differences among the families in contextual plasticity, developmental plasticity or both. That is, the differences in the activity rates among the families might have been due to differences among them in the cumulative effects of extended exposure to predator cues on their activity rates, differences in their immediate responses to the presence or absence of predator cues, or some combination of the two. Researchers who specifically focus on behavioural plasticities are more likely to use experimental designs that distinguish between developmental and contextual plasticity. For instance, Chapman et al. (2009) studied the effects of light environments on the courtship behaviour of guppies (Poecilia reticulata) by rearing males in two treatments (two different light intensities) and then testing the courtship behaviour of adult males from each treatment group under each of the same two light intensities. This design allowed them to determine that rearing conditions had no detectable effect on male courtship, current light conditions had strong effects on male courtship behaviour, and there were no detectable interactions between the effects of rearing conditions and current light conditions on male courtship (i.e. no detectable effects of rearing conditions on contextual plasticity). More generally, Fig. 1 emphasizes why a developmental perspective is important in any study of individual differences in behavioural plasticities. If different subjects are all exposed to the same external stimulus, differences in the behaviour they immediately express in


Judy A. Stamps

10 response to that stimulus must necessarily be a function of differences in their current internal state. In turn, differences among subjects in their internal state at a given moment must necessarily be due to differences among the subjects in their developmental history, where each individual’s developmental history is a complex product of interactions between its genes and epigenes, and the many experiences it has had since conception. By extension, differences among individuals in behavioural plasticities at a given age or life stage must also be due to differences, among those individuals, in their developmental histories prior to onset of the experiment. In recent years, appreciation of this fact, coupled with growing empirical evidence of individual differences in behavioural plasticities (see Section IV), has encouraged empirical studies of the effects of prior experience on all of the types of behavioural plasticities described in the previous paragraphs (see Section V). (7) Why using reversibility as a criterion for categorizing plasticities is problematic Several authors have suggested that different types of exogenous plasticities be categorized based on whether the effects of experience on trait values are reversible or irreversible (Piersma & Lindstrom, 1997; Piersma & Drent, 2003; Gabriel et al., 2005; Utz et al., 2014), or, more loosely, whether the effects of experience on behaviour are ‘short-term’ versus ‘long-lasting’ (Dingemanse & Wolf, 2013; Westneat et al., 2014). For instance, Piersma & Drent (2003), suggested that the term ‘phenotypic flexibility’ be used to refer to environmentally induced changes in phenotype that are reversible within individuals, and the term ‘developmental plasticity’ be used to refer environmentally induced changes in phenotype that are irreversible within individuals. The problem with applying this approach to behaviour is that it is difficult to tell whether an induced change in behaviour is irreversible. In many cases, induced changes in behaviour that were assumed to be ‘permanent’ when experimental subjects were subsequently maintained under one set of conditions later turned out to be reversible when they were maintained under different conditions. For example, it was initially assumed that the effects of prenatal stress on behaviour were permanent in humans and other mammals (Champagne, 2010a). Accumulating empirical evidence that this was not the case encouraged researchers to study postnatal experiences that might reduce or reverse adverse effects of prenatal stress on the physiology and behaviour of juveniles or adults (Schaffer, 2000; Monk, Spicer & Champagne, 2012). Similarly, it was initially assumed, based on studies of rodents housed under standard conditions, that stressful events during the juvenile period had irreversible effects on adult behaviour. However, subsequent research showed that if subjects were housed under social and/or structurally

enriched conditions following the stressful experiences, the effects of early stress on behaviour decreased or disappeared (Ruis et al., 1999; Morley-Fletcher et al., 2003; Ilin & Richter-Levin, 2009). As a practical matter, it is impossible to expose animals to all of the stimuli, experiences, and factors that might reverse changes in behaviour induced by earlier experiences. Hence, while it may be possible to show that behaviour induced by an early experience is retained if subjects are maintained under one set of conditions in the laboratory, it is much more difficult to demonstrate that the effects of an early experience could not be reversed by any of many types of experiences that those subjects might have had, if they had developed under natural conditions. Interestingly, some years ago neurobiologists and endocrinologists encountered and addressed the same problem with respect to defining plasticities based on the reversibility (or lack thereof) of induced changes in phenotypic traits. For instance, neuroscientists used to believe that brains and behaviour were shaped during formative experiences early in life, but were fixed in adulthood. However, a large body of research has now documented lifelong plasticity in neural structure and function in a variety of taxa (Zhang & Meaney, 2010; Ebbesson & Braithwaite, 2012; McEwen, 2012). Similarly, endocrinologists initially assumed that hormones early in life had ‘organizational’ effects that were irreversible later in life, but later research proved this assumption untenable (Arnold & Breedlove, 1985). Hence, in contrast to the situation with many morphological traits (e.g. body size in species with determinate growth), it seems inadvisable to categorize plasticities of behavioural or physiological traits based on assumptions about whether or not induced changes in trait values are reversible.

III. QUANTIFYING INDIVIDUAL DIFFERENCES IN BEHAVIOURAL PLASTICITIES (1) General concerns Although there are many different types of behavioural plasticities, most of them can be investigated using a limited number of experimental protocols and statistical models. For instance, mixed-effects models (described below) can be used to study contextual plasticities, developmental plasticities or temporal plasticities. Below we consider general methodological issues that apply to empirical studies of a wide range of behavioural plasticities. Not surprisingly, most empirical studies of individual differences in behavioural plasticities use different individuals as subjects. However, some types of behavioural plasticities can only be investigated using ‘replicate individuals’, where each set of subjects consists of individuals with similar genetic makeup and similar experiences



11

prior to the beginning of the experiment (see Section III.2a). Because the methods outlined in this section apply to both, here we use the term ‘agent’ to refer to either an individual animal or to a group of individuals with the same genotype, reared under the same conditions prior to the time their behaviour is first measured. Most studies of behavioural plasticities either focus on situations in which the behaviour in question can be described using a continuously distributed function (e.g. variation in latency to attack) or on situations in which behaviour can divided into two discrete categories (e.g. an animal either attacks, or it doesn’t). Of course, there are many situations in which animals can express many different types of behaviour. For instance, when presented with a conspecific, a male mouse may investigate, threaten or attack the conspecific, retreat from the conspecific, groom itself, rear up on its hind legs, or perform other types of behaviour (Nyberg, Vekovischeva & Sandnabba, 2003). Unfortunately, statistical methods for studying differences among agents in behavioural plasticities when those agents express more than two discrete types of behaviour are currently underdeveloped (e.g. see Stamps, Saltz & Krishnan, 2013). Hence, below we will focus on situations in which the dependent variable in analyses of plasticity (i.e. the behaviour) is either continuously distributed or falls into two discrete categories. The independent variables in analyses of behavioural plasticities can also be continuously distributed or discrete. In studies of temporal plasticities, the independent variable (age or time) is, by definition, continuously distributed. However, in the case of contextual plasticities or developmental plasticities, the stimuli that contribute to variation in behaviour can be either continuously distributed or discrete. For instance, consider the contextual plasticity of behaviour involved in mate choice, e.g. situations in which a given individual responds differently to cues from different potential mates. In some cases, stimuli from different mates fall naturally along a gradient, in which case, each agent’s response to different stimuli along the gradient can be summarized by a preference function (Wagner, Murray & Cade, 1995; Ritchie, 1996; Wagner, 1998; Rodriguez, Rebar & Fowler-Finn, 2013b). In other cases, potential mates produce signals within the same stimulus modality whose components vary independently of one another (e.g. odour cues composed of many different hydrocarbons; Ingleby, Hunt & Hosken, 2013), or produce multimodal signals with components from different stimulus modalities (e.g. courtship signals with both visual and auditory components; Ronald, Fernandez-Juricic & Lucas, 2012). In such cases, it may be impossible to arrange stimuli from the same set of potential mates along a continuum. Instead, in this situation each potential mate can be treated as a discrete item, in which case contextual plasticity describes the extent to which an individual’s

Fig. 2. Describing individual differences in behavioural plasticities when the independent variable is continuously distributed. In this situation, reaction norms can be used to describe the functional (mathematical) relationship between the independent variable and the behaviour scores for each individual. Illustrated for linear reaction norms for activity as a function of current temperature (contextual plasticity) for juvenile damselfish (Pomacentrus moluccensis). Modified from figure 1a in Biro et al. (2010), with permission of the authors and the journal Proceedings of the Royal Society B, © 2010 The Royal Society.

responses vary when it is presented with each of the different mates (see also Section III.2a). When an independent variable can be modelled by a continuous function, the behaviour of an agent can be described using a ‘behavioural reaction norm’ (Dingemanse et al., 2010). In this case, the shape or the slope of the reaction norm describes how an agent’s expected behaviour varies as a function of an independent variable (e.g. current external stimuli, past external stimuli, time or age). A behavioural reaction norm is a function-valued trait, i.e. a phenotypic trait that varies as a function of a continuously distributed variable (Meyer & Kirkpatrick, 2005; Stinchcombe & Kirkpatrick, 2012). In the simplest possible case, the functional (mathematical) relationship between an independent variable and each agent’s behavioural scores can be described using a straight line. In this case, the slope of each agent’s relationship between behaviour and an independent variable describes how its behaviour changes as a function of that variable (Dingemanse et al., 2010) (see Fig. 2). However, reaction norms can have other shapes. For instance, mate preference functions are often unimodal (hump-shaped), such that more complicated equations are required to describe them (Ritchie, 1996; Rodriguez et al., 2013b).


Judy A. Stamps

12

Fig. 3. Behavioural plasticity versus the slopes of behavioural reaction norms. Individual A and individual B both express high levels of temporal plasticity: the behaviour of each individual systematically changes as a function of time. However, the slopes of their reaction norms for behaviour are very different (negative for individual A; positive for individual B). Individual C is less plastic than A or B, even though its slope (0) is intermediate between theirs. Because the signs of their reaction norms differ across the individuals in this sample, the slopes of their reaction norms are not positively correlated with their levels of temporal plasticity.

A variety of different statistical techniques, including mixed-effect modelling (Dingemanse & Dochtermann, 2013), structural equation modelling (Curran, Obeidat & Losardo, 2010) and factor analysis (Matzel et al., 2003) can be used to analyse behavioural plasticities when independent variables are continuously distributed. However, the individual-level statistics generated by these methods (e.g. the slopes of behavioural reaction norms) do not necessarily reflect the relative behavioural plasticity of each of the individuals within a sample. The problem arises because ’plasticity’ refers to the extent to which behaviour varies as a function of an independent variable; it does not specify whether behaviour increases or decreases as a function of that variable (see Table 1). As a result, individuals with very different values of a statistic that is used to describe their plasticity (e.g. different slopes of their reaction norms) may actually express similar levels of plasticity, in the usual sense of that term. For example, in Fig. 3, individual A and individual B both exhibit high levels of temporal plasticity: the behaviour of both individuals systematically and substantially changes over the study period. However, they differ in the direction of their temporal plasticity: individual A’s behaviour decreases as a function of time, while individual B’s behaviour increases as a function of time. Further, individual C (with a slope of 0: no change in behaviour as a function of time) is less plastic than either A or B, even though the slope of its reaction

norm is intermediate between theirs. Thus, for this set of individuals, the slopes of their reaction norms are not correlated with their levels of behavioural plasticity. Individual differences in the sign as well as the magnitude of the slopes of behavioural reaction norms have been detected in many empirical studies (e.g. Bell & Peeke, 2012; Stamps et al., 2012; Briffa, 2013), so this problem is of more than theoretical interest. Of course, if every individual’s reaction norm slope has the same sign (e.g. Fig. 2), then individual differences in slopes can provide reasonable estimates of individual differences in plasticity. At present, however, it is unclear how to characterize the relative plasticity of different individuals in the same sample, if their reaction norm slopes vary with respect to sign as well as magnitude. One possibility might be to devise statistics analogous to the absolute value of the slopes of reaction norms, and then use those to describe the extent to which each individual’s behaviour changes as a function of the same continuously distributed independent variable (see also Section VI.1). When external stimuli are discrete and subjects are exposed to two different stimuli (or two different sets of stimuli), the plasticity of an agent is described by the difference between the agent’s mean scores for the two stimuli (Auld et al., 2010). However, there is currently no consensus on how to measure the plasticity of each agent when those agents have been exposed to three or more discrete sets of stimuli (J. R. Auld, personal communication). For instance, the effects of three different rearing environments on the olfactory behaviour of adult Drosophila melanogaster varies across genotypes (isolines), as was indicated by a significant interaction between genotype and rearing environment in an analysis of variance (anova), in which behaviour was the dependent variable (Sambandan et al., 2008) (Fig. 4). However, it is not obvious how to describe the plasticity of each genotype in this experiment. One option might be to use a min-max method: for each genotype (i) compute its mean behaviour score for each environment, (ii) determine the environments which generated the smallest and the largest mean scores, and then (iii) use the difference between these two mean scores as an index of the plasticity for that genotype. This method reflects one of the traditional meanings of the term ‘plasticity’ as indicating the range of phenotypes a genotype is capable of generating in response to an array of different environments, external stimuli or experiences (see also ‘potential plasticity’: Section II.1 and Table 1). Another option might be to use the variance of the mean scores of each genotype as its index of plasticity. This index would indicate the extent to which a genotype’s mean phenotype varied in response to an array of different environments, external stimuli or experiences. The min-max and variance methods provide comparable indices of plasticity for the eight genotypes shown in Fig. 4 (r = 0.98). However, the two methods would



13 force their data inappropriately into formats (i.e. continuous distributions, or two discrete categories) that can be readily handled by familiar statistical models. Regardless of whether the independent variables in studies of plasticity are continuous or discrete, relatively large sample sizes (number of samples per individual, number of individuals) may be required to detect individual differences in plasticity, and many samples per individual may be required to provide reasonable estimates of the behaviour expressed by each of the subjects in the data set (van de Pol, 2012; Wolak, Fairbairn & Paulsen, 2012; Biro et al., 2013). In the past, many empirical studies of behavioural plasticities have lacked sufficient statistical power to detect and describe individual differences in those plasticities, so this issue should be carefully considered by anyone contemplating research on this topic.

Fig. 4. Describing individual differences in behavioural plasticities when values of the independent variable are discrete and several. Mean scores for the olfactory behaviour (responses to benzaldehyde) of adult females of eight genotypes (inbred lines) of Drosophila melanogaster , as a function of the type of environment in which they were reared (developmental plasticity): A, standard fly food; B, tomato medium; C, ethanol-supplemented medium. The three rearing environments differed in a number of respects, and could not be arranged along a continuum. Even so, it is clear that some genotypes (e.g. 2,5) were relatively non-plastic: their behaviour was similar regardless of the environment in which they were reared. Other genotypes (e.g. 1,3,4 and 6) were more plastic: their behaviour varied as a function of rearing environment. The min-max and the variance methods (see text) generated comparable estimates of the plasticity of these genotypes, where genotype 1 was the most plastic, and genotype 5 the least plastic. Modified from figure 2B of Sambandan et al. (2008), with permission of the authors and the journal Genetics, © 2008 by the Genetics Society of America.

generate different indices of plasticity in other situations, e.g. for a genotype that produced the same mean score for behaviour after being reared in five different environments, but generated a much higher mean score when reared in a sixth environment. In fact, it might be desirable to use more than one statistic to describe different aspects of behavioural plasticity when animals are exposed to multiple discrete stimuli. Similar issues have arisen in the literature on mate choice, where a variety of indices have been developed to describe differences among individuals and genotypes in hump-shaped mate preference functions (Gray & Cade, 1999; Rodriguez et al., 2013a). This problem deserves attention, because the lack of suitable indices of plasticity for individuals or genotypes when the presumptive causative factors naturally fall into more than two discrete categories has either discouraged empiricists from studying these types of behavioural plasticities, or encouraged them to

(2) Experimental designs for studying individual differences in behavioural plasticities (a) Within-individual versus replicate-individual designs In within-individual experimental designs, the data are based on repeated measures of the same individuals. This is the ‘default’ experimental design for studies of individual differences in behavioural plasticities. The behaviour of each individual is scored at different times, and the resulting data are used to estimate the contextual plasticity, developmental plasticity, endogenous plasticity or temporal plasticity of that individual. Replicate individuals are individuals with the same genotype and the same prior experiences. In order to generate replicate individuals for studies of individual differences in behavioural plasticities, it is important to ensure that the individuals with each genotype are, as much as possible, similar to one another at the beginning of the experiment. Because the internal state of an individual at a given time is the result of interactions between an individual’s genetic makeup and all of the experiential factors that occurred to that individual prior to that time, replicate individuals have the same (or similar) genotype, i.e. isolines, clones or (more approximately), full or half sibs. In addition, individuals with the same genotype also share many other experiential factors (e.g. parental effects, inherited epigenetic effects, sibling effects) which differ more across than within genotypes, and that can affect the development of behaviour (Stamps & Groothuis, 2010a,b). The notion that clones, isolines, strains, etc. share more than genes is nicely illustrated by cross-fostering studies in laboratory mice. Traditionally, consistent differences in the behaviour expressed by different inbred strains of mice have been attributed to differences in the genetic makeup of those strains (e.g. Jacobson & Cryan, 2007; O’Leary, Gunn & Brown, 2013). However, inbred strains also differ consistently from one another with respect to other factors, and the effects


Judy A. Stamps

14 of some of those factors on behaviour can be detected using suitable experimental protocols. For instance, mice from the C57BL/6J strain consistently express different levels of behaviour than BALB/cJ mice, when both strains are maintained and reared under the same conditions and then tested for anxiety-related and exploratory behaviour as adults (Francis et al., 2003; Jacobson & Cryan, 2007; O’Leary et al., 2013). However, when C57BL/6J embryos were implanted into the uteri of BALB mothers and then reared by BALB mothers, the behaviour of the cross-fostered offspring later in life differed from that of C57BL/6J embryos implanted into and reared by C57BL/6J mothers, but was indistinguishable from the behaviour of BALB offspring naturally born and raised by BALB mothers (Francis et al., 2003). In other words, strain differences in maternal effects, and/or strain differences in interactions between maternal and offspring phenotypes may be largely responsible for major, consistent differences in the behaviour of these two strains. A parallel line of research using Long-Evans rats demonstrated that individual differences in maternal behaviour with pronounced effects on offspring behaviour can be transmitted across generations via non-genetic (epigenetic) processes (Champagne, 2008). These results suggest that in rodents, strain differences in some important types of maternal effects might be due to differences among those strains in epigenetic rather than genetic factors. Of course, in species in which fathers interact with their offspring, paternal effects can also have a major impact on offspring behaviour (e.g. Gleason & Marler, 2013; McGhee & Bell, 2014). Intriguingly, recent studies have shown that fathers can affect their offspring’s behaviour via epigenetic pathways even if those fathers have no contact with their offspring. For instance, exposing male mice to stressful events soon after they are born affects the behavioural flexibility of those males’ offspring, after those offspring have reached adulthood (Gapp et al., 2014). In this case, evidence suggests that adverse experiences of the fathers early in life alter the methylation patterns of specific genes in their germ cells, and that these epigenetic modifications subsequently passed via their sperm to their offspring (Gapp et al., 2014). Hence, differences in the behaviour of the inbred lines, isolines, clones, etc. typically used as examples of ‘genotypes’ cannot be entirely attributed to differences in the genetic makeup of the subjects, their parents, or both. What we can safely say, however, is that individuals with a given genotype share a host of genetic and epigenetic factors that may affect the development of behaviour, and that (in the laboratory, at least) these factors are reliably replicated within each genotype from one generation to the next. The criteria for using genotypes as representatives of replicate individuals are more stringent than the criteria for using genotypes to investigate other questions. First, any genotypes used as replicate individuals

should be composed of clones, sibships, isolines etc. derived from individuals randomly collected at the same locality. This is because the goal of research on individual differences in behaviour is to describe the distribution of behavioural phenotypes that exist within a single population, as opposed to the range of phenotypes that occur within or across populations. In addition, the genotypes used as replicate individuals should not have been subjected to artificial selection for behavioural or other traits after being brought into the laboratory. One reason is that strong selection for extreme values of a particular type of behaviour is likely to affect the plasticity of that behaviour. Strong artificial selection for a given behaviour favours individuals and lineages that consistently generate extreme values of that behaviour in spite of fluctuations in the subjects’ internal state (i.e. the selection favours low endogenous plasticity), in spite of variation in external stimuli when the behaviour is assayed (i.e. the selection favours low contextual plasticity), and in spite of variation in external stimuli or internal factors earlier in life that affect the expression of behaviour (i.e. the selection favours low developmental plasticity). For instance, in wild-type male fruit flies (Drosophilia melanogaster ) aggressive behaviour is infrequently observed, usually occurs when males are located on a food resource, and is only expressed at high levels in the presence of cues from adult females (Hoffmann, 1987; Hoffmann & Cacoyianni, 1989). However, following generations of strong artificial selection for high aggressiveness, males from a ‘high aggression’ line consistently engaged in vicious escalated fights in arenas that lacked any food or cues from females (Dierick & Greenspan, 2006). In addition, artificially selected lines may exhibit relationships between phenotypic traits different than those observed across individuals in natural populations. One reason is that artificial selection on one trait often leads to changes in other traits, but the effects of artificial selection on one trait on other, non-selected, traits can vary, depending on the individuals that were originally used to found a selection line. Even if different selection lines experience apparently identical directional selection, they often differ with respect to the extent or even the direction of change in other (non-selected) traits (Mayr, 1961; Crabbe et al., 1990). In house mice, for instance, the effects of strong artificial selection for male aggressiveness had different effects on female-directed behaviour in three different lines. As one would expect, males of three lines selected for high aggressiveness (SAL, TA and NC900) were all consistently highly aggressive when interacting with other males (Caramaschi et al., 2008). However, males of the SAL line were also pathologically aggressive towards females, fiercely and repeatedly attacking both familiar and unfamiliar females in both familiar and novel areas. By contrast, males of the other two



15

high-aggression lines were far less offensive towards females (Caramaschi et al., 2008). A second issue with using artificially selected lines to infer relationships between phenotypic traits is that the effects of artificial selection on correlated traits can also vary depending on which of the traits was the subject of selection. In honeybees, artificial selection on one type of learning (latent inhibition, LI) led to changes in another type of learning (reversal learning): the strong-LI line eventually had slower reversal learning scores than did the weak-LI line (Chandra, Hosler & Smith, 2000). However, when researchers instead selected on reversal learning, the opposite pattern emerged: in this case, the fast-reversal line ended up with stronger latent inhibition than the slow-reversal line (Ferguson, Cobey & Smith, 2001). Because of these issues, associations between traits in artificially selected lines need not necessarily be comparable to the association between those same traits in non-selected individuals from the same population. For instance, Baugh et al. (2012) reported strong, significant differences in the acute corticosterone responses of two lines of great tits (Parus major ) artificially selected for fast versus slow rates of exploratory behaviour (see Section VI.2 and Table 3), but were unable to detect any covariation between the exploratory scores and acute corticosterone responses among non-selected birds from the same population. For the above reasons, it is inadvisable to use lines derived from different populations or artificially selected lines as replicate individuals in studies of individual differences of behavioural plasticities. However, artificially selected lines derived from the same population may provide useful pilot data and insights into individual differences in behavioural plasticities, or relationships between behavioural plasticities and other traits (see Table 3). Inbred strains of mice and rats are even less suitable as replicate individuals in studies of behavioural plasticities. Different inbred strains of laboratory rodents originated from ancestors taken from different populations, subspecies and even continents, so one cannot assume that the distribution of trait values across different strains mirrors the distribution of trait values within any single natural population. In addition, many of those strains were subsequently subjected to strong artificial selection for a variety of different traits after they were brought into captivity (mice: Lyon & Searle, 1989; Atchley & Fitch, 1991; Petkov et al., 2004; rats: Thomas et al., 2003; Russell Lindsey & Baker, 2006). Assuming that suitable replicate individuals are available, there are several reasons to use them to investigate individual differences in behavioural plasticities. First, some questions about individual differences in developmental plasticity can only be answered using replicate individuals. In particular, studies of individual differences in ontogenetic plasticity necessarily use

replicate individuals as the subjects in ‘common garden’ experiments, since a given individual can only have one type of experience when it is a given age. Since both replicate individuals and individual animals have been used to study various types of developmental plasticities, further discussion of this topic is deferred to Section III.2b. Replicate individuals are also required to study individual differences in contextual plasticity for types of behaviour that a given individual only expresses once in its life. For instance, dispersal from an individual’s natal habitat is a unique event, so studies of individual differences in the contextual plasticity of natal dispersal necessarily rely on replicate individual designs. Thus, siblings (offspring from the same clutch) were used to study the contextual plasticity of natal dispersal within populations of salt marsh wolf spiders (Pardosa purbeckensis), in a study which demonstrated differences among families on the immediate effects of wind velocity on the probability of dispersing via ‘ballooning’ (Bonte, Bossuyt & Lens, 2007). Replicate individuals can also be used to avoid the period and carryover effects that often confound studies of contextual plasticity. Within-individual designs for contextual plasticity necessarily require measuring the same individual more than once, but an individual’s response to the same stimulus can change over time as a result of changes in its internal state across the sequence of tests (‘period effects’), and exposure to one stimulus can affect an individual’s response to other stimuli later in the series of assays (‘carryover’ or ‘order’ effects) (Diaz-Uriarte, 2001; Bell, 2013). A variety of methods have been devised to control experimentally or statistically for period and carryover effects. However, these methods assume that period and carryover effects do not differ across the individuals in the sample (Diaz-Uriarte, 2001, 2002; Dochtermann, 2010; Bell, 2013); violation of this assumption can lead to biased estimates of individual differences in contextual plasticity (Bell, 2013). Given accumulating empirical evidence that the endogenous plasticities and developmental plasticities which contribute to period and carryover effects can differ across individuals (see Section IV), this assumption is problematic. One way to avoid period and carryover effects in studies of contextual plasticity is to use replicate individuals, each of which is tested once at the same age, at the same time of day, and under the same set of set of external stimuli. For instance, different isolines of Drosophila spp. have been used to study genotypic differences in the contextual plasticity of mate-choice behaviour. In such studies, focal females with different genotypes are exposed to stimulus males from a standard set of genotypes. Each focal female’s response to a male is only tested once (avoiding any period or carryover effects). But since large numbers of females with the same genotype can be exposed to each of the male genotypes,



SL

— Pig (Sus scrofa)

—

Rat (Rattus norvegicus) SL, males

House mouse (Mus musculus) Wild-type, males

SL, males

SL, males SL, males SL, males

SL, males SL, males

Ruiz-Gomez et al. (2011)

Basic et al. (2012)

Bolhuis et al. (2004)

Coppens et al. (2012, 2013)

Benus et al. (1987)

Benus et al. (1990a)

Benus et al. (1988) Benus et al. (1989) Benus (1999)

Nyberg et al. (2003) Nyberg et al. (2004)

Isolation-induced aggressiveness: high (TA), low (TNA)

Attack latencies: short (S, SAL), long (L, LAL)

Avoidance learning: fast (RHA), slow (RLA)

Resistance test: high resisting (HR), low resisting (LR)

Post-stress cortisol levels: high (HR), low (LR)

Tonic immobility test: HL = high latencies, LL = low latencies to recovery

Habituation rate Ontogenetic plasticity

Learning ability Flexibility Flexibility Flexibility Flexibility Learning rate Ontogenetic plasticity

Learning rate Flexibility Flexibility

Learning rate Flexibility Ontogenetic plasticity

Acclimation to social isolation Learning rate Flexibility Flexibility Flexibility

Learning rate Flexibility

Time spent in light side of light-dark box Effect of social rearing2,3 on aggressiveness towards males3

Trials to criterion in maze Change in cues in/around a maze Reversal learning: Y-shaped living structure Change in sex of intruder Change in light:dark cycle Avoidance conditioning Effect of handling1,2 on attack latencies3

Avoidance learning: shuttlebox Change in reward schedule (‘impulsiveness’) Extinction rate

Learning a T maze Reversal learning: T maze Effect of enriched housing1 on response to novel object1

Learning a T maze Effect of change in food location on foraging Effect of novel object on foraging Effect of novel object on foraging

Resumption of normal feeding

Problem-solving task Reversal learning

Acoustic discrimination task Acoustic discrimination task Reversal learning

Repeated exposure to an initially novel room Avoidance of aposomatic prey Problem solving: lever and string pulling Effect of food restriction1 on aggressiveness3

Change in food location: field

Change in food location Reversal learning

Description of experiment

∗ Results

for ontogenetic plasticities: the age at which experience occurred, and the age at which behaviour was tested. 1 = juvenile, 2 = adolescent, 3 = adult. might have been due to ceiling or floor effects (see text).

Rainbow trout (Oncorhynchus mykiss) SL

Ruiz-Gomez et al. (2008)

1,2,3 :

—

Zebra finch (Taeniopygia guttata)

Habituation rate Learning rate Learning rate Ontogenetic plasticity

— SL — SL, males Black-capped chickadee (Poecile atricapillus) — — Learning rate Learning rate Flexibility

Flexibility

—

Exploratory speed: fast (F), slow (S)

Flexibility Flexibility

Exploratory speed: fast (F), slow (S)

Great tit (Parus major) Males —

Type of behavioural plasticity

Indicator trait

Species

Brust, Wuerz & Kruger (2013)

Guillette et al. (2009) Guillette et al. (2011)

Verbeek et al. (1994) Titulaer, van Oers & Naguib (2012) van Overveld & Matthysen (2010, 2013) Dingemanse et al. (2012b) Exnerova et al. (2010) Cole, Cram & Quinn (2011) Carere et al. (2005)

Reference

TA > TNA TA < TNA∗

S≈L L>S LAL > SAL LAL > SAL LAL > SAL LAL < SAL LAL > SAL∗

RLA < RHA RLA > RHA RLA > RHA

LR ≈ HR LR > HR LR > HR

HR ≈ LR HR > LR HR > LR HR < LR

HR > LR

HL ≈ LL HL > LL

SF

SF S≈F SF S < F (males) S > F (females) S

Individual differences in cocaine addiction: maladaptive behavioural traits.

Individual Differences in Behavioural Decision Weights Related to Irregularities in Cochlear Mechanics.

Social isolation and individual differences: behavioural and dopaminergic responses to psychomotor stimulants.

Individual differences in associative learning.

Individual Differences in School Children.

Individual differences in helping dispositions.

Individual differences in drug response.

Individual differences in sleep patterns.

Individual Differences in Behavioural Despair Predict Brain GSK-3beta Expression in Mice: The Power of a Modified Swim Test.

Inter-individual differences in heart rate variability are associated with inter-individual differences in mind-reading.

Individual and sex-specific differences in intrinsic growth rate covary with consistent individual differences in behaviour.

Accounting for individual behavioural variation in studies of habitat selection.

Subcortical correlates of individual differences in aptitude.

ERPs reveal individual differences in morphosyntactic processing.

The darkness within: individual differences in stress.

Procedural Learning and Individual Differences in Language.

Adaptive training and individual differences in perception.

Individual differences in response conflict adaptations.

Individual differences in adult foveal visual asymmetries.

Individual differences in bridging inference processes.

Individual differences in human circadian rhythms.

Trends in the psychophysiology of individual differences.

Individual differences in locomotor activity and sensitization.

Individual differences in adaptive coding of face identity are linked to individual differences in face recognition ability.