Perhaps the most powerful and obvious explanation for quantitative chemical variation and β-chemodiversity is the fact that environments are inherently variable, in space and time and across many scales, and variability, regardless of its origin, improves the likelihood of a plant's phenotype matching its environment. This is the lottery principle, originally proposed by Williams as an explanation for sex (1975). Spatial heterogeneity in herbivory is most likely to occur for species that are not ecosystem dominants, that disperse propagules widely and that are attacked by specialist herbivores. However, because competition also favours phenotypic variability (Bell, 1982), whether the predictions of the lottery principle are borne out may depend upon the relative strengths and heterogeneity of herbivore and pathogen pressure, versus ecological tradeoffs between defence and competitive ability.
Bet-hedging describes a suite of strategies by which plants adjust the mean and variance of a trait among their offspring so as to reduce temporal (usually year-to-year) variation in their fitness (Simons, 2009). A particular pattern of offspring defence may produce high arithmetic mean fitness across years, but if that pattern fails to ensure a reasonable proportion of survivors in years of high herbivore pressure, geometric mean fitness will be low, and the long-term survival of the genotype poor. Bet-hedging can be conservative, when among-sibling variation is reduced around a mean PSM type or level that ensures survival under most herbivory conditions, or diversifying, when propagule variation is increased to maximize the chances of some offspring surviving in any given year. Bet-hedging of plant chemical defence is plausible; however, because spatial (between-sibling) variation in the intensity of herbivory encountered reduces its likelihood, while between-cohort variation increases it (Starrfelt & Kokko, 2012), it might be more likely to occur in locally common annual plants where herbivore densities vary from year to year.
The occurrence of undefended chemotypes in populations of largely defended individuals can sometimes be described as automimicry (Speed et al., 2012). In rare cases where herbivores choose food items on a plant species rather than a plant individual basis, automimics are parasites on the public good (i.e. the benefit accrued to the population as a whole from the presence of defended individuals), and their fitness should decline with their frequency in the population. Mathematical models can demonstrate evolutionarily stable strategies that allow automimicry, both within and between plants (Till-Bottraud & Gouyon, 1992), but empirical tests are rare. Lawler et al. (1999) described an ecological conditioned flavour aversion, whereby herbivores used concentrations of readily tolerated terpenes in Eucalyptus as a reliable olfactory cue to assess the concentration of antifeedant formylated phloroglucinol compounds (FPCs). This situation seems to present an opportunity for more sophisticated mimicry, where high terpene concentrations could mimic defence by FPCs without their cost. Because successful mimics must be difficult to identify, mimicry might be more successful against generalist than specialist herbivores, but will fail if pre-ingestive cues such as volatile compounds or pigments allow herbivores to reliably assess defence. Hamilton & Brown (2001) suggested that overwintering aphids use the highly variable PSMs responsible for autumn leaf colour as an honest signal of defence.
Frequency-dependent selection (FDS) and local adaptation by herbivores are key to other explanations of β-chemodiversity. Because herbivores and pathogens become locally adapted to common plant chemotypes, the fitness of plant chemotypes should decrease as their frequency in the population increases and overall parasitism (herbivory and infection by pathogens) in a population should decrease with chemotypic diversity. Over time, the frequency of rare phenotypes increases as a consequence, resulting in time-lagged oscillations between host and parasite genotypes. Mixed evidence for FDS is available from a range of predator–prey interactions; for example, Pasteels & Gregoire (2009) showed that predatory insects avoided insect prey with rare defensive secretions, but Janz et al. (2005) found no evidence that preferred host-plant frequency affected rates of oviposition by a polyphagous butterfly. However, studies addressing FDS on plant defence are uncommon (Núñez-Farfán et al., 2007). Siemens & Roy (2005) demonstrated FDS for resistance to a rust fungus in Arabis holboellii but not for herbivore resistance mediated by glucosinolates in the same species. By contrast, many examples demonstrate local adaptation by plant enemies (Kaltz & Shykoff, 1998; Hereford, 2009), even though it can be prevented by high levels of herbivore immigration (Tack & Roslin, 2010).
1. Diversity can be inherently beneficial for plant individuals, families, populations and communities
Intraspecific variation in PSMs was a formative example in developing the concepts of community and ecosystem genetics. Variation in condensed tannins in cottonwood (Populus) influences the trees that beavers select, which in turn affects tree fitness and stand composition. Tannins negatively affect nitrogen mineralization but are positively correlated with fine-root production (Whitham et al., 2006). Although this idea is an attractive way to link intraspecific variation in PSMs to community- and ecosystem-wide consequences, understanding the genes that underlie the trait has proved elusive despite the early availability of the Populus trichocarpa genome (Wang et al., 2013).
Biodiversity can boost community primary productivity and other ecosystem functions via well-studied facilitation, additive and dominance effects (Tilman et al., 2001), and recently interest has grown in the relationship between these functions and intraspecific genetic diversity (Sangster et al., 2008). The magnitude of genetic diversity effects on ecosystem function can match those of species diversity (Crutsinger et al., 2006). For example, genetic diversity can often reduce herbivory and damage by pathogens in agricultural systems (Cantelo & Sanford, 1984; Smithson & Lenne, 1996) and Hughes & Stachowicz (2004) showed that genetic diversity increased the resistance of a seagrass (Zostera marina) population to grazing geese. Nevertheless, increased diversity does not always reduce herbivore populations (Johnson et al., 2006; Kotowska et al., 2010). Utsumi et al. (2011) showed that aphid population size increased in more genotypically diverse plots, and proposed two explanations: source–sink effects where susceptible plants provide a source of aphids for defended neighbours, and reduced mortality of natural enemies of aphids. Although PSMs are likely to contribute to these effects, studies that have explicitly considered genetically-based diversity in secondary chemistry in this context are rare and less encouraging. Two studies (Poelman et al., 2009; Tack et al., 2012) observed that qualitative glucosinolate diversity in Brassica oleracea cultivars was associated with insect diversity, but that most insects responded similarly to the various compounds. Furthermore, Macel et al. (2002) found that qualitative diversity in pyrrolizidine alkaloids in eight Senecio species appeared to be selectively neutral in the context of herbivory by a specialist herbivore.
The approach used in the kinds of studies mentioned above may not be without problems. To maximize the chance of detecting effects, diversity is commonly sourced from provenances across large geographical areas, so the range of phenotypic diversity (i.e. the magnitude of quantitative diversity or the chemotypic diversity among families or clones) created in experimental populations often deliberately exceeds that seen in natural populations, yet the experimental spatial scale usually remains small (Tack et al., 2012). Consequently, the importance of the ecological effects of genetic and phenotypic diversity may be overestimated and Tack & Roslin (2011) suggest that they may be generally small. At the same time, the number of genetic families or clones in common gardens will sometimes be less than that that seen in natural populations, and this might accelerate the process of herbivore specialization on hosts, for example.
Quantitative diversity in chemical defence can directly reduce herbivory on individuals and populations via Jensen's inequality (Jensen, 1906). If the relationship between PSM concentration and the benefit gained by a herbivore is concave, then the net benefit to a herbivore (and hence the damage experienced by an individual exhibiting heterogeneity among its parts or a heterogenous population) will decrease with increasing heterogeneity given the same plant or population trait mean. The ability of induced defences to produce such heterogeneity among plant parts or individuals may offer a significant advantage over constitutive defences (Karban et al., 1997; Karban, 2011). Shelton (2004) produced a number of dynamic programming models parameterized with data from a generalist herbivore of Raphanus sativus defended by glucosinolates, showing that such quantitative variation, at a range of scales, should reduce herbivore fitness.
Another mechanism has been demonstrated by which qualitative PSM diversity among individuals may also reduce herbivory experienced by plant individuals in some, probably rare, cases. Just as generalist herbivores include many plant species in their diet to spread the detoxification load across enzymes and pathways (Marsh et al., 2006), and balance their nutrition (Felton et al., 2009), intraspecific PSM variation might promote plant switching by specialists. Mody et al. (2007) showed that Chrysopsyche imparilis caterpillars switched regularly between individual plants of Combretum fragrans, thus shortening feeding bouts on individual plants with the result that the foliage was never markedly reduced.
The sessile nature of plants allows some scope for group selection when animals make feeding decisions at the level of the patch instead of, or in addition to, the level of the individual plant, and PSM diversity can contribute to a herbivore's decision to feed in or depart from a patch. If increased diversity is beneficial to the offspring seedlings clustered around a maternal plant, then selection may act upon these sources of variation. Stabilizing selection might also act more strongly to ensure the ideal proportion of automimics among the offspring of a maternal plant where most or all automimics and automodels are siblings because the opportunity for unrelated plants to exploit the common good is reduced. Within-family automimicry allows some offspring to avoid the cost of defence but benefit from defence by its siblings, which is now a family rather than a common good. This situation is analogous to that described by Hare & Eisner (1993) where insects produce clutches of eggs where only a proportion are defended, but the group as a whole benefits.
The production of chemically variable offspring may allow a maternal plant to ensure that some offspring can circumvent the attack by locally adapted herbivores predicted by the Janzen–Connell hypothesis (Janzen, 1970; Connell, 1971) to occur within the high-seedfall zone surrounding the parent (Langenheim & Stubblebein, 1983). A meta-analysis (Hyatt et al., 2003) suggested that seedlings were probably more affected by proximity to parent trees in tropical than in temperate forests, and so predictions about chemodiversity should differ accordingly.
Finally, an alternative perspective is to view undefended seedlings as acting to dilute the impact of herbivores on their better-defended siblings, rather than as beneficiaries of within-family diversity. These family- or group-selection mechanisms all depend upon an individual's likelihood of herbivory being influenced by the density and quality of its neighbours. There are examples of herbivores that make feeding decisions only at the level of the patch, only at the level of the individual plant or plant part and at both levels (Hjalten et al., 1993; Bergvall & Leimar, 2005; Moore et al., 2010; Emerson et al., 2012). Many of these group-selection mechanisms have been proposed previously (Denno & McClure, 1983), but whether they are valid and how they may favour chemodiversity will depend upon many circumstances including habitat complexity, seed dispersal and herbivore identity. Their importance can only be addressed by the accumulation of empirical data matching the dynamic dispersal and movement of herbivores to data about plant population structures, such as detailed spatial mapping of plant chemical defence genotypes and phenotypes.
2. PSM variation may be ecologically essential
The roles of both quanitative variation and chemodiversity in ameliorating biotic and abiotic stress may play critical roles in creating the ecological niche differentiation among individual plants that is essential to reduce competition and allow coexistence. Variation in responses to the environment among individual plants has been shown to allow species coexistence in a variety of plant communities (Clark, 2010; Jung et al., 2010). If diversity is a prerequisite for finding niche space, this demand might be met by genetic diversity or by plasticity. Few studies to date have specifically looked at the role of PSM variability and diversity in these processes, but Mraja et al. (2011) showed that the species richness and composition of plant communities can alter the concentration and composition of PSMs in P. lanceolata in experimental grasslands. It is likely that the species diversity of ecological communities and the phenotypic diversity of plant populations are mutually reinforcing, as plant defensive traits structure communities just as the broader community context shapes defence phenotypes and selection upon them (Lankau & Strauss, 2007; Poelman et al., 2008). For example, quantitative chemical variation and chemodiversity can promote biodiversity by allowing species to invade and persist in dynamic systems, whether that diversity occurs among (β-chemodiversity; Lankau & Strauss, 2007, 2008) or even within (α-chemodiversity) individuals (Iason et al., 2005).