Ecological immunology and tolerance in plants and animals

Why should we study tolerance in animals?

Why should we be interested in studying tolerance in animals? There are several important reasons. First, as demonstrated in the plant literature, tolerance can provide a potent mechanism to protect individuals against the negative fitness consequences of parasitism. As such, studying tolerance can provide important insights into the range of strategies that hosts can employ to maximize fitness when faced with parasites (Råberg, Graham & Read 2009). For example, we may predict that individuals living in groups should invest more in defence mechanisms because of increased risk of infection (reviewed in Rolff & Siva-Jothy 2003). If we focus our experiments entirely on resistance mechanisms, we may erroneously conclude that there is no effect of group living on host defence if resistance does not differ among individuals, when in fact individuals vary in tolerance. Furthermore, without making the distinction between resistance and tolerance, researchers may choose to measure either parasite burden or host fitness to study resistance. If they find variation in host fitness, they may mistakenly conclude that there is variation in resistance; however, it is well possible that all studied animals had similar parasite burdens, yet suffered different levels of fitness loss (i.e. they varied in tolerance) (Corby-Harris et al. 2007).

Secondly, a major ecological question is why there is variation in immunity, and a central answer to this question has been that host defence is costly in terms of other life-history traits (Sheldon & Verhulst 1996; Rolff & Siva-Jothy 2003; Sadd & Schmid-Hempel 2008; Schulenburg et al. 2009). Although many studies have found trade-offs between resistance and life history traits such as longevity, reproduction, developmental time and competitive ability (Boots & Begon 1993; Kraaijeveld & Godfray 1997; Moret & Schmid-Hempel 2000; Simmons & Roberts 2005; McKean et al. 2008), not all studies have found an obvious cost of resistance (e.g. Shoemaker & Adamo 2007; Voordouw, Koella & Hurd 2008). Two potential reasons for this are that: (i) tolerance, and not resistance, is the main defence mechanism in the species studied; and/or (ii) there is a trade-off between tolerance and resistance. The plant literature has provided examples of such trade-offs (Fineblum & Rausher 1995; Stowe 1998; Baucom & Mauricio 2008b), and they might be present in animals as well (Råberg, Sim & Read 2007).

A third reason for studying tolerance is that this defence mechanism may have different consequences than resistance for the spread of pathogens in populations and the evolution of parasite virulence. Although the exact epidemiological and evolutionary consequences are as yet unclear due to a lack of models that explicitly address the coevolution of host and parasite (Little et al. 2010), existing models do point at some interesting resistance-tolerance differences. For example, models that studied host defence evolution (while not allowing parasite evolution) have suggested that hosts may evolve polymorphism in resistance, but that tolerance mechanisms are more likely to become fixed (Boots & Bowers 1999; Roy & Kirchner 2000; Miller, White & Boots 2005). In these models the evolution of resistance results in a negative epidemiological feedback because resistance reduces parasite transmission. Thus, in a population with high parasite prevalence, there is strong selection for resistance and resistant hosts will increase in number; this then leads to a decrease in overall parasite transmission and prevalence. Because host resistance is assumed to be costly, the costs of resistance start to outweigh the benefits at low parasite prevalence, and susceptible hosts may be selected for again. Tolerance, in contrast, does not reduce parasite transmission, and hence results in a positive epidemiological feedback: tolerant hosts have a selective advantage in the presence of parasites; because tolerant hosts do not reduce parasite transmission, parasite prevalence increases with a higher number of tolerant hosts. This higher parasite prevalence in turn favours tolerant hosts, so that over time all hosts should become tolerant.

Models have also suggested that tolerance – in contrast with resistance – should not result in antagonistic coevolution between hosts and parasites (e.g. Rausher 2001), but this may not necessarily be true (Little et al. 2010). Thus, although the evolution of tolerant hosts may result in commensal relationships between host and parasite (Miller, White & Boots 2006), it is important to note that such commensalism is an expressed outcome determined by both host and parasite together. Because host tolerance reduces the cost of parasite virulence to parasites (by killing its host a parasite can reduce its transmissible period: e.g. Anderson & May 1982; Bremermann & Pickering 1983; Levin 1996; Frank 1996; Mackinnon & Read 2004; Fraser et al. 2007; De Roode, Yates & Altizer 2008), parasites can in fact evolve greater intrinsic growth and virulence (Restif & Koella 2003; Miller, White & Boots 2006). When infecting a non-tolerant host, the expressed virulence of such parasites would be much higher than that before tolerance evolved (Miller, White & Boots 2006). Additionally, tolerance is also expected to result in higher parasite prevalence, and this means that even if an individual’s probability of dying of infection decreases, overall disease in the population may in fact increase.

Finally, it has been suggested that because resistance has negative consequences on parasite growth and transmission whereas tolerance does not, tolerance will be less likely to select for parasites that overcome tolerance than resistance is likely to select for parasites to overcome resistance. This simple notion has led some authors to speculate that treatments focused on increasing tolerance may be evolution-proof, and should be preferred over treatments based on resistance (Roy & Kirchner 2000; Rausher 2001; Schneider & Ayres 2008). However, there are two major caveats. First, this ‘evolution-proof’ concept requires that tolerance does not trade-off with other important resistance or life-history traits. Secondly, because tolerance increases parasite prevalence (e.g. Miller, White & Boots 2006), such treatments may result in a greater prevalence of multiple-genotype infections. This is important, because many theoretical studies have suggested that multiple-genotype infections may lead to within-host competition between parasite genotypes, and select for greater competitive ability. Because competitively ability is assumed to correlate with virulence, such selection may result in more virulent parasites (Nowak & May 1994; Van Baalen & Sabelis 1995; Frank 1996; Choisy & De Roode 2010). Although empirical studies on this topic are scarce, a few studies have now confirmed the assumed relationship between competitive ability and virulence (De Roode et al. 2005; Ben-Ami, Mouton & Ebert 2008).

What defence mechanisms have animal researchers studied so far?

As described in the previous section, tolerance can provide a potent defence mechanism with important ecological and evolutionary implications. Moreover, tolerance has been studied and used by medical professionals for a long time. For example, humans in specific age groups, or with specific genetic blood genotypes (e.g. α-thalassaemia) experience different incidence and severity of malarial disease, despite having similar parasite burdens (Rogier, Commenges & Trape 1996; Wambua et al. 2006; Müller et al. 2009), and tolerance-based treatments are already in use for several diseases, including cholera and bacterial meningitis (Schneider & Ayres 2008). However, evolutionary ecologists have largely neglected tolerance in animals, even though they have studied it extensively in plants. Instead, animal researchers have studied a large number of defence mechanisms that can all be described as resistance.

For example, theoreticians often distinguish between host traits that reduce infection probability (qualitative resistance/avoidance) or reduce parasite burdens upon infection (quantitative resistance/recovery) (Boots & Bowers 1999; Gandon & Michalakis 2000). These defence mechanisms can often be modelled using a single epidemiological parameter, without describing in-host parasite densities and immune mechanisms. Experimentalists often compare multiple clones, genotypes or full/half-sib families of a host in the laboratory and define resistance as the inverse parasite burden upon inoculation with a standardized parasite dose or genotype (Carius, Little & Ebert 2001; De Roode & Altizer 2010). Resistance is often measured as a genetic or immunological mechanism (Sheldon & Verhulst 1996; Rolff & Siva-Jothy 2003; Lambrechts, Fellous & Koella 2006; Sadd & Schmid-Hempel 2008), but can also come in the form of specific animal behaviours (Hart 1990). Since all these mechanisms reduce infection probability or parasite growth – rather than alleviate the disease symptoms on the basis of a given parasite burden – they are all examples of host resistance, not tolerance. Perhaps the only tolerance mechanism that has received considerable attention in animals is fecundity compensation: a series of plastic responses by which hosts can compensate for parasite-induced damage by increasing their reproductive effort (e.g. Michalakis 2009).

It may seem puzzling that tolerance has not been extensively considered in animal systems. One potential reason for its neglect is that resistance may be more apparent and easier to measure in animals than tolerance. As mentioned above, existing theoretical models predict that the arms races between hosts and parasites should result in a large amount of variation in resistance; in contrast, many models predict much less variation – if not a complete absence – in phenotypes that are tolerant. Because it is much easier to measure variable than fixed defence types, it may be easier to detect resistance than tolerance (Roy & Kirchner 2000; Read, Graham & Råberg 2008). Another potential reason that tolerance has received less treatment in the animal literature is that its study requires twice, or more, the number of experimental individuals as would the study of resistance (see below).

There is, however, a biologically more interesting possibility, which is that animals are fundamentally different from plants. In particular, most animals can move freely, while plants cannot (Walbot 1985). This has important consequences for which defences animals and plants may employ: in particular, animals can respond to parasitic threats through behaviour, while plants need to respond physiologically.

Anti-parasitic animal behaviours can focus on avoidance of infection (‘avoidance’), reduced infection probability (‘prophylactic self-medication’) or reduced parasite burdens once infected (‘therapeutic self-medication’) (Hart 1990). For example, wood ants use a form of prophylactic self-medication, whereby they incorporate pieces of conifer resin into their nests; resin inhibits the growth of bacteria and fungi and protects the ants against the detrimental effects of these microorganisms (Christe et al. 2003; Chapuisat et al. 2007; Castella, Chapuisat & Christe 2008; Castella et al. 2008). Evidence for therapeutic self-medication is mostly based on field observations on primates ingesting plant species with anti-parasitic chemicals (Wrangham & Nishida 1983; Huffman & Seifu 1989; Huffman 1997), although recent experiments have also shown direct evidence for self-medication in several species of Lepidoptera (e.g. Singer, Mace & Bernays 2009). Although some plants may utilize volatiles produced by other plants to fend off natural enemies (Himanen et al. 2010), medication is more likely to occur in free-moving animals.

In contrast, plants have evolved a greater capacity to regenerate tissues than animals (Walbot 1985). Animals maintain adult stem cells and are able to induce stem cell proliferation in differentiated cells, allowing them to regenerate tissues and maintain homeostasis (Birnbaum & Sánchez Alvarado 2008); and some animals (e.g. lizards and planarians) are even able to regenerate whole body parts. Overall, however, the apical meristem in plants provides them with greater capacity for whole-organ regeneration, and in particular the generation of reproductive organs from somatic cells (Walbot 1985). Thus, plants have greater capacity to retain fitness when damaged than animals.

Although these major plant-animal differences may indicate that tolerance is less likely to evolve in animals than plants, this certainly does not mean tolerance is unimportant in animals. Indeed, as mentioned above, tolerance is studied and used by medical scientists, and a series of recent studies by evolutionary ecologists also demonstrates that tolerance occurs in animal systems (Corby-Harris et al. 2007; Råberg, Sim & Read 2007; Ayres, Freitag & Schneider 2008; Ayres & Schneider 2009; De Roode & Altizer 2010). Before dealing with the significance of these studies, we will turn to the plant literature to learn how plant insights can be used to advance our understanding of tolerance in animals.

How to define tolerance and the best practice of estimating tolerance

The term tolerance has been used loosely across plant sub-disciplines as the ability to ‘cope’ with various stresses such as herbicide, salinity, drought or heavy metals. In the evolutionary ecology literature, tolerance is often defined as a plant defence trait that results in the retention of fitness following damage, whether that damage is incurred by biotic sources such as herbivores or parasites, or abiotic sources such as herbicide or drought. Tolerance is often graphically depicted as a reaction norm of fitness across a gradient of increasing damage, and as such it is a phenotypically plastic trait (Stowe et al. 2000). This definition underscores how one measures tolerance: experimental individuals are first classified by their genetic relationships into groups, e.g. members of a family or members of a population, the fitness of these groups is then assayed at different levels of damage, and the estimate of tolerance for each group is subsequently described as the slope of a line representing the relationship between fitness and damage (Fig. 1a).

There are important considerations to take into account at each step in the process of estimating tolerance, and such considerations will necessarily change according to the limitations of the study system being considered. Take, for example, the first step, in which one classifies individuals into a group according to their relatedness. If the appropriate crosses have been performed among individuals in a greenhouse, common garden or laboratory environment, the estimate of tolerance can reasonably be considered to reflect a genetic component, although variation in the study population cannot be assumed unless the response of families or sibships are seen to differ (Fig. 1b). It is thus important to understand the relationship of individuals within a group so that the evolutionary potential of the trait can be assessed. Some study systems preclude the use of generated family lines – long-lived organisms such as trees, or animals that are impossible to rear in the laboratory setting. In these situations, individuals from a population can contribute to a population-level estimate of tolerance, or an investigator can assess tolerance at the species level. However, if tolerance is determined at these levels of biological organization, the estimates will be influenced by non-genetic sources of variation, such as environmental maternal and population effects, and thus the evolutionary potential of the trait cannot be clearly assessed.

The second step of estimating tolerance requires that the fitness of replicates of these groups be assayed in more than one level of the damaging environment (Stowe et al. 2000). This is important for two practical reasons: first, a single individual cannot be both damaged and non-damaged, and secondly, if fitness were measured in only a single environment, and then compared among genetic lines, one cannot be certain that a differential response in fitness between lines is a consequence of the environmental factor in question or a consequence of an unknown environmental factor (Simms 2000). Furthermore, if fitness in only one environment were considered, it would not be possible to determine if fitness differences between lines were due to tolerance to the environmental factor in question, or due to differences in general vigour, which is defined as a genotype’s ability to perform well in response to all unexamined environmental factors (see Simms 2000 for a thorough discussion). Thus, tolerance is not a trait that is as easily scored as many other traits. Because of the need to measure fitness in multiple environments, an assessment of tolerance will require double the number of experimental individuals as would, for example, resistance, or the measure of a simple phenotype. This could present an obstacle to a researcher who works on organisms that are not easily reared and maintained in the laboratory setting, as in the previous example.

Finally, the operational estimate of tolerance is commonly made in one of two ways: an estimate of the slope in a reaction norm, as previously discussed and depicted in Fig. 1, or as the difference in average fitness of replicates grown in a damage environment compared to fitness in a control environment (point tolerance; W_d − W_u). The latter estimate of tolerance is generally used when damage is measured categorically, such as when estimating tolerance given the presence/absence of herbicide or other abiotic agent (Baucom & Mauricio 2004, 2008a,b), or when using a specific level of imposed damage, such as the removal of a certain amount of leaf tissue from all individuals within the treatment environment (Tiffin & Rausher 1999). The former is most often used when damage can be scored as a continuous variable – for example, the amount of leaf material removed by insects in the field (Mauricio, Rausher & Burdick 1997). A major difference between the two common operational estimates of tolerance is that the reaction norm approach can identify potential nonlinear components of tolerance, e.g. some families overcompensate at low levels of damage yet decrease in fitness at higher levels of damage, whereas point tolerance will not elucidate this type of relationship between fitness and damage.

How each researcher operationalizes tolerance is dependent upon the study system in question, and specifically, the agent of damage under investigation. For example, if a researcher wishes to assess tolerance to parasitism in a wild-caught, natural population of mice, it makes sense to use the reaction norm approach, in that the amount of parasites present in a sample derived from stomach contents or blood is quantified rather than experimentally imposed, and the fitness of these individuals is regressed across this gradient of increasing ‘parasite load’ or ‘burden’ (see Definitions). Of course, this is dependent on the researcher’s ability to score the fitness of said mice, as well.

As discussed previously, the measure of tolerance in the above scenario would be restricted to a population level estimate, unless the researcher can assess fine-scale relatedness in a group of wild-caught organisms. It is possible that in some study systems, the genetic relatedness of individuals could be determined by the use of genetic markers, such that one might be able to determine which wild-caught individuals were members of the same sib-ship. Using the previous example of wild-caught mice, a researcher would need to accurately measure the naturally occurring parasite load within each wild-caught individual, determine the relationship of each individual to one another, and make an accurate assessment of the number of offspring each parent produced, either by counting the number of offspring in a den, or again by the use of genetic markers. However, the feasibility of this untested idea rests on the ability to correctly identify the relatedness among individuals in a population to the level of a sibship and is best attempted in organisms that do not move far from their natal home site. Thus, in many situations it may be most feasible to measure tolerance on organisms that can be reared in the lab, with damage being experimentally imposed rather than natural.

Issues pertinent to the estimation of tolerance

There are important considerations pertinent to the estimation of tolerance that have been discussed in the plant literature that are relevant to studies of tolerance across systems, namely, biases inherent to the use of natural or artificial damage, how fitness should be measured, and the importance of controlling for differences in the scale of measurement between fitness and damage. Each of the above can obscure or alter an assessment of variability in tolerance in the study population and/or affect the ability to accurately compare the level of tolerance across species and experiments.

The discussion in the plant literature of whether one should use natural vs. artificial damage has centred on damage via insect herbivory. Tiffin & Inouye (2000) statistically described the benefits and costs of using experimentally imposed vs. natural damage, concluding that artificial damage results in a more accurate estimate of tolerance, but the use of natural damage leads to more precise estimates, or estimates that have a higher degree of experimental reproducibility. They suggest the biases resulting from either the use of experimental or natural damage should be considered in light of the main goal of the experiment. For example, if one wishes to produce tolerance estimates that are closer to the real value of tolerance, one should impose artificial damage, and if an assessment of genetic variation in tolerance, fitness costs of tolerance, or selection is of interest, then the use of natural damage will produce greater precision of underlying estimates. This is because the presence of genetic variation in a study population is determined by an anova, and if the precision behind the estimate of tolerance is biased, so too will be the model’s residuals and the power to detect differences among families. Lehtilä (2003) disagreed with their conclusions, stating that an assumption of Tiffin and Inouye’s statistical model – that of an equal distribution of levels of herbivory among natural and artificial damage – was unrealistic, and that the use of a carefully controlled allocation of imposed damage among experimental individuals by the researcher would lead to both more accurate and precise estimate of tolerance. Inouye & Tiffin (2003) made the compelling case that this concern was more relevant to situations in which the distribution of damage across experimental individuals approximated the mean level of damage in the population, but was not as relevant to experiments in which damage exhibited a bimodal distribution, or one in which the distribution of herbivory does not concentrate closely to the mean level of herbivory in the population. This discussion is relevant to future research in that some study systems, for various reasons, might preclude assessment of naturally occurring damage and require the use of artificially imposed damage. Thus, this discussion should be helpful in guiding experimental design decisions, as it is applicable to studies investigating tolerance to parasitism, or tolerance to any agent of damage that might vary in a density-dependent manner.

Various authors have discussed or experimentally determined the influence that the measure of fitness might have on tolerance (Strauss & Agrawal 1999). The majority of plant studies use seed production as the estimate of fitness, which is a measure of female plant fitness and is distinct from the estimate of male plant fitness, or the number of seeds sired through pollen donation. Few studies have simultaneously considered tolerance based on female and male estimates of fitness. In some cases tolerance varies according to sex, yet in other systems it does not (Ashman, Cole & Bradburn 2004; Stephenson et al. 2004; Cole & Ashman 2005). How the sexes allocate resources to tolerance is still a relatively unexplored concept in studies of plant tolerance, and is as yet virtually unexamined in animal systems. Furthermore, in many study systems it will be difficult to measure fitness as reproductive output, and fitness correlates, such as biomass allocation, longevity or other measures might be used instead. In such situations it is important to initially identify the strength of the relationship between reproductive output and the measure used, and to recognize the limitations inherent to using them.

Finally, another methodological concern was examined by Wise & Carr (2008) in which the necessity of using the same scale of measurement between fitness and damage when measuring tolerance was demonstrated. For example, when quantifying the amount of damage as a proportion of the total plant, fitness should also be on the ‘multiplicative’ scale, in their example, the natural log of fitness. Using an additive scale for one trait, such as the number of seeds per plant, and a multiplicative scale for the other, such as the proportion of damage, can lead to the aberrant conclusion that variability for tolerance is present in the study population, when in fact it is not.

Major hypotheses addressed in the plant literature

Many questions concerning the evolution of tolerance have been addressed in the plant literature, from the factors that maintain tolerance, to the effect of competition on the expression of costs of tolerance, to the potential for ontogenetic changes in the level of tolerance. A discussion of all of the theoretical and experimental treatments concerning plant tolerance would require its own review, and for this reason we list some of the main hypotheses that have been addressed in Table 1, although the list and its references are by no means exhaustive. Below, we discuss a predominant focus of questions concerning tolerance studied in plant systems – the potential factors responsible for the maintenance of tolerance.

Questions concerning the maintenance of tolerance have derived from the oft-made observation that variation for tolerance exists within populations of many different study systems. This finding has intrigued plant biologists, as the general thought on the basis of existing models is that all individuals should be highly tolerant following the action of selection (see above). To explain this apparent paradox between expectations and empirical observations, researchers have suggested that the presence of allocation costs, or fitness costs, might be working to maintain the level of tolerance at intermediate values, as might correlations between tolerance and other traits, namely, resistance (Núñez-Farfán, Fornoni & Valverde 2007).

A review of the plant evolutionary ecology literature finds that fitness costs associated with tolerance are widely empirically supported, suggesting that this type of cost can act to maintain intermediate levels of tolerance in a population (Tiffin & Rausher 1999; Fornoni et al. 2004). However, costs in the form of a negative correlation between resistance and tolerance are less supported, with the majority of systems finding instead a pattern of simultaneous allocation to both forms of defence (Núñez-Farfán, Fornoni & Valverde 2007; but see Fineblum & Rausher 1995; Stowe 1998; Baucom & Mauricio 2008a,b). An alternative to the hypothesis that a correlation between tolerance and resistance maintains tolerance at an intermediate level has been the suggestion that stabilizing selection, due to a nonlinear cost or benefit as a function of allocation to defence (Tiffin & Rausher 1999) acts to maintain tolerance and/or other defensive traits; however, the empirical work finds mixed support for this idea (Núñez-Farfán, Fornoni & Valverde 2007). Although the potential for costs of tolerance has been assessed in many dicot plants (Núñez-Farfán, Fornoni & Valverde 2007), these concepts are relatively unexplored in animal systems, with the exception of malaria-infected mice, which exhibited a negative relationship between resistance and tolerance, and thus a cost of tolerance and potential constraint on its evolution (Råberg, Sim & Read 2007).

Where theory and empiricism need to meet

A major challenge in understanding the epidemiological consequences of tolerance will be to develop theory that is based on empirically feasible measures of tolerance. Most current theoretical models follow variations on classical susceptible-infected-resistant (SIR) epidemiological models (Anderson & May 1992). In these models, hosts move between susceptible, infected and recovered compartments, and are born and die on the basis of rate constants, including a transmission rate β, a parasite-induced mortality rate (virulence) α and a host recovery rate γ. In these models increased resistance can be modelled as a reduced β (Boots & Bowers 1999; Restif & Koella 2003) or increased γ (Boots & Bowers 1999; Restif & Koella 2003; Miller, White & Boots 2005; Best, White & Boots 2008) while increased tolerance can be modelled as a decreased α (Boots & Bowers 1999; Restif & Koella 2003; Miller, White & Boots 2005; Best, White & Boots 2008). Slightly different models have expressed tolerance as a reduction in host longevity (Roy & Kirchner 2000), or a smaller reduction of host fecundity (Restif & Koella 2004).

An important point is that most of these models do not explicitly model within-host parasite burdens, and simply assume that rate parameters are the same for each infected host within a class (e.g. tolerant vs. non-tolerant hosts). Thus, these models implicitly measure tolerance as a reduced parasite-induced mortality rate of infected host individuals, rather than a reaction norm of fitness as a function of parasite load. Where within-host parasite burdens have been modelled (Miller, White & Boots 2005), it has been shown that the exact shape of the relationship between parasite burden and virulence (and hence tolerance) crucially affects the evolutionary trajectory of parasite virulence evolution.

Thus, although decades of plant research have concluded that reaction norms are the best approach to measure tolerance, our current theory is based on different parameters, which are difficult to measure in natural systems. Therefore, in order to enhance our understanding of tolerance in animal systems, it is essential to develop theory based on measurable tolerance mechanisms, and to measure tolerance in ways that are relevant to theory. Without such synergy, the study of host tolerance may become as problematic as the study of virulence evolution, which has also suffered from the fact that simple theoretical assumptions have been difficult to test in real-life systems (Ebert & Bull 2003; De Roode, Yates & Altizer 2008).