The nature of nutrition: a unifying framework

Additional keywords: ageing, cannibalism, conservation, Geometric Framework.

Nutrition touches, links and shapes all aspects of the biological world (Simpson and Raubenheimer 2012). It builds the components of organisms, and fuels the dynamic interactions between these components; it determines whether or not wild animals thrive, how their populations grow, decline and evolve, and how assemblages of interacting species (ecological communities) and ecosystems are structured. Nutrition also drives the affairs of humans, from individuals to global geopolitics. Climate change, population growth, urbanisation, environmental degradation and species extinctions all are in one way or another linked to the need for nutrients. In short, nutrients are the interconnecting threads in the web of life.

And yet the science of nutrition remains fragmented. Because of its direct importance to human health and food animal production, nutrition has traditionally been considered the domain of the medical and agricultural sciences. Research in these areas has produced a tremendously detailed account of the nutritional biology of a few species. By contrast, with some exceptions, nutrition in the biological sciences has tended to adopt simpler, more general approaches that are applicable across the diversity of animals. Foraging might, for example, be considered a process of acquiring energy or minimising time exposed to predators, rather than a complex balancing act of obtaining enough – but not too much – of the many nutrients that are needed for sustaining health and reproduction. The advantage of this simplified approach is that it has supported the development of powerful general frameworks for biological processes, unhindered by the staggering nutritional complexity that has been uncovered in the more applied nutritional sciences. Such unidimensional approaches to nutrition have generated valuable insights into foraging behaviour and provided a heuristic framework for thinking about ecological processes (White 1983; Stephens and Krebs 1986; Stephens et al. 2007). They have, however, contributed little to the understanding of which nutrients or combinations of nutrients influence foraging, and how animal requirements for these influence ecological processes. For this, models are needed that are nutritionally explicit, in the sense that they enable a study to address these questions directly (Raubenheimer et al. 2009).

Achieving nutritional homeostasis involves a complex interplay between multiple and changing nutrient needs and variable foods. Long-term evolutionary processes have ensured that animals are equipped with the mechanisms to deal with these complexities, but considerable challenges remain for nutritional biologists to understand these processes. Many aspects of the relationship between animals and their nutritional environments underscore this challenge. First, foods are complex mixtures of multiple components, each of which has its own functional implications for the animal. Some are necessary for maintaining good health, while others are hazardous and best avoided (e.g. the antipredator chemical defences produced by some plants and animals: Sotka et al. 2009). To further complicate issues, some toxins can be beneficial if ingested in low quantities, and even essential nutrients can be toxic if overingested (Raubenheimer and Simpson 2009). Second, a given food component (at a stipulated dose) can have multiple influences on an animal. Third, most aspects of animal function are influenced by multiple food components. Finally, and in some respects most challenging of all, food components interact in intricate ways in their effects on animals.

Disentangling this web of interconnections can be as complex and daunting as it is important for understanding the biology of animals, and for managing the complex relationships between our own species and the world that we inhabit. To succeed, an approach is needed that systematically deals with each of the challenges mentioned above. Namely, it should provide a framework within which multiple food components and animal attributes can be distinguished, and the relationships among components and attributes disentangled and then linked to individual performance, ecological outcomes and evolutionary consequences. Importantly, it should not seek to recapitulate the full complexity of nutrition, nor to oversimplify it, but rather to provide a level of complexity that best suits the research question at hand.

Such an integrative framework must be able to represent the animal, the environment, and the nutritional basis for the interaction between animal and environment (Raubenheimer et al. 2009). A second requirement is that an integrative framework should take account of the fact that the nutritional interactions between animals and their environment take place on a stage that is constructed of many food components. Third, if the framework is to be grounded within the powerful paradigm of evolutionary biology, it is imperative that the consequences for the animal of its behavioural and physiological responses to the nutritional environment can be represented. These consequences – which include such factors as reproductive outputs, development rates, and the risk of premature death (e.g. the effectiveness of the immune system) – are relevant not only to individual wellbeing but also to population sizes. A framework that takes into account consequences therefore extends its reach both to evolution and also to nutritional aspects of population ecology. A further step, to community ecology, would be achievable if the framework fulfilled a fourth requirement, of being able to incorporate the nutritional basis of interactions among multiple species in food webs.

The Geometric Framework (GF) was designed with these core requirements in mind (Raubenheimer and Simpson 1993, 1999; Simpson and Raubenheimer 1993, 1995, 2012; Simpson et al. 2004; Raubenheimer et al. 2009). It satisfies the multiple-food-components requirement using a simple device known as a nutrient space. A nutrient space is a geometric space built of two or more axes, where each axis represents a food component that is suspected to play a role in influencing the animal’s responses to its environment. In most cases these food components will be nutrients but, as discussed below, this is not invariably the case. The nutrient space provides the common context in which to describe the pertinent aspects of the animal, its environment, the interactions between animal and environment, and the consequences of these interactions. In the sections that follow we describe how this is achieved.

An aspect of the animal that is fundamental to its interactions with the environment is its nutrient requirements – the amounts and balance of nutrients that it needs to eat over a given period to gain maximal benefit (usually evaluated in terms of evolutionary fitness). Nutrient requirements are fundamental in nutritional models because they provide a reference point for predicting how an animal should respond to its environment and for understanding the relationships between foods and the performance consequences of eating those foods. For example, if we know the nutrient needs of an animal and the nutritional composition of the foods available to it, we can predict which foods it would eat and which it should avoid, and if it did otherwise we would be justified in wondering why. Likewise, if the animal persisted in eating foods that do not satisfy its nutrient needs, we would have good cause to wonder about the long-term consequences of this – would it become obese, have impaired reproduction, and/or become susceptible to infectious diseases? In GF models, the optimal nutrient requirement of an animal is represented as a point (or, over time, a trajectory) in a nutrient space, called an intake target.

Except for special cases such as salt licks, nutrients in the environment come packaged together as mixtures – foods. The diet of the animal may comprise a single food or, more usually, a combination of different amounts of several foods. Foods are modelled in a nutrient space in two ways: by the amounts of nutrients they contain and, more generally, by the balance of nutrients they contain. For many purposes, some of which will be discussed shortly, it is useful to disregard the precise amount of nutrients in a specific food item, considering rather the balance of the nutrients in the food. This general property of a food type is pictured in a geometric model as a line that passes through the origin of the graph and through a point representing any quantity of the food type. The slope of such a line indicates the ratio of the nutrient in the food. Such lines representing the balance of nutrients in foods are called nutritional rails, for reasons that will become clear below.

If foods are the most fundamental aspect of an animal’s nutritional environment, then feeding is the primary nexus of interaction with this aspect of the environment. For the animal, the important thing about feeding is that it provides a means to change – and regulate – its nutritional state. Feeding is therefore represented in GF models by the change in the nutritional state of the animal that results from eating. This change is plotted as a trajectory through nutrient space. Such feeding trajectories have the same angle as the nutritional rail for the food being eaten, because as the animal eats it gains the nutrients in the same proportion as they are present in the food being eaten. By feeding, animals are therefore channelled, like a train, along tracks in nutrient space set by the nutritional rails of the foods they select, with the distance of movement along these rails being determined by the amount of the food eaten.

The challenge for animals is to select foods that direct them to their intake target, and ensure that they eat enough to arrive there. The simplest way to do this is to eat a nutritionally balanced food, which leads directly to the target – i.e. contains the same balance of the nutrients as is needed – and is plentiful enough to enable the animal to get there. Food items that are nutritionally imbalanced do not lead to the intake target. The animal could nonetheless use such a food to navigate indirectly to the intake target, if it combined its intake with a second imbalanced food whose rail falls on the opposite side of the intake target. Combinations of nutritionally imbalanced foods that jointly enable animals to reach their intake target in this way are known as nutritionally complementary foods. The fact that many animals are able to reach their intake target by mixing nutritionally complementary foods provides a useful means for researchers to estimate the position of the intake target, based on the testable prediction that animals will have evolved regulatory mechanisms to ensure ingestion of a balanced diet (e.g. Chambers et al. 1995; Raubenheimer and Jones 2006).

In some cases, the animal might have access neither to nutritionally balanced nor nutritionally complementary foods, but only to non-complementary nutritionally imbalanced foods. In this predicament the animal cannot reach its intake target. The animal’s response in this circumstance is known as a rule of compromise because it reflects the compromise selected by the animal between overeating some nutrients and undereating others. To derive a general description of the rule of compromise we have to measure the responses of the animal to a range of nutritionally imbalanced foods. The intake points for such an experiment jointly form an intake array, the shape of which provides a comprehensive description of the rule of compromise. The shape of the intake array may differ according to the animal in question and the nutrients being modelled.

Our use of the phrase ‘nutrient requirements’ up to this point refers to the amounts and balance of nutrients that an animal needs to eat, which we have represented in geometric models as an intake target. There is, however, an even more fundamental context in which the term ‘nutrient requirements’ can be used, and that is in relation to the amounts and balance of nutrients that the animal needs to make available to its tissues to satisfy its various demands for energy metabolism, tissue growth, storage and so forth. How does this requirement, which we call the nutrient target, differ from the intake target? It differs because animals are usually unable to allocate 100% of the nutrient they eat to useful purposes, but lose some in the faeces, urine and, in some animals, through permeable membranes on the body surface (e.g. the gills of fish). To satisfy its tissue-level requirements (the nutrient target) an animal therefore needs to compensate by eating enough to cover both the nutrient target and the constrained losses. These two components, required amount + constrained losses, jointly comprise the intake target.

Just as the intake target can be partitioned into the nutrient target + constrained losses, so too can the nutrient target be partitioned into various subtargets. When an animal reaches its nutrient target, the total pool of available nutrients needs to be divided among several functions, including growth, metabolism and reproduction (Raubenheimer and Simpson 1992, 1994, 1995). How the animal allocates these nutrients is critical to fitness and, as a result, natural selection has fashioned animal physiology to achieve a favourable strategy for investing its nutritional ‘income’ across its various requirements. Experiments show that, within limits, animals are able to regulate nutrient utilisation postingestively to defend a growth target, even when eating foods that prevent them from reaching the intake target, e.g. by adjusting the digestion and absorption of nutrients from the gut (e.g. Clissold et al. 2010) and by voiding excess ingested nutrients postabsorptively (e.g. Zanotto et al. 1993, 1997).

The above discussion concerns optimal nutrient allocation to growth, but we could equally model metabolic targets, reproductive targets and so forth. The overall point, however, is to demonstrate that the GF can be used to construct multidimensional models of nutrient budgets, which include functionally optimal nutrient allocations (targets) as well as the allocations that are actually achieved (Raubenheimer and Simpson 1995). If we had additional information about the benefits of achieving these optima (or the costs of failing to do so), this would put us in a good position to understand how diverse nutritional strategies have evolved.

If we assume that the nutritional responses of animals, including regulation to an intake target, rules of compromise and postingestive regulation, have been fashioned by natural selection, it follows that an animal that achieves its intake target will enjoy maximal Darwinian fitness. The challenge is how to include such fitness consequences in geometric models. They differ from the components we have so far integrated into the models – foods, nutrient requirements, nutrient intake and postingestive utilisation – in that fitness components are not measured in terms of nutrients, but in terms of other units such as numbers of offspring, probability of premature death, longevity and so forth. Fitness consequences cannot, therefore, be depicted in the nutrient space in the usual way.

Instead, we have borrowed from evolutionary biologists the metaphor of the ‘fitness landscape’, in which the consequences of nutrient intake are plotted in a map-like surface superimposed on the nutrient space. This enables us to envisage the intake target state as the summit of a Darwinian fitness mountain mapped onto nutrient intake space. The slope, steepness and curvature of the decline in elevation with distance from the intake target need not be the same for excesses and deficits of the same nutrient, or be the same for different nutrients, or be independent between nutrients. Fitness costs can be described to take account of these various possibilities using a mathematical function, the Taylor series expansion (Simpson et al. 2004).

Experimental examples in which the components of fitness have been mapped as landscapes onto detailed nutrient intake arrays have begun to accumulate in recent years, notably in invertebrates (e.g. Simpson et al. 2004; Lee et al. 2008a ; Maklakov et al. 2008; Jensen et al. 2012). These experiments involve maintaining individuals on one of a large number of diet compositions, and are thus laborious to conduct. Nevertheless, as we shall see below, the insights gained can be substantial.

It is clear that animals possess not one appetite system but several, and are able to regulate independently their intakes and utilisation of different nutrients to maintain intake and growth targets. We will consider the mechanisms involved in outline, highlighting the major principles and key components of regulatory systems.

Maintaining an intake target involves animals achieving three things: (1) assessing the nutritional composition of available foods, (2) assessing their own nutritional state, and (3) comparing these two to produce appropriate feeding responses. The simplest means to assess the nutritional composition of foods is to detect different nutrients by tasting them. Not surprisingly, therefore, all organisms, from bacteria to mammals, possess specialised receptors for the detection of key nutrients such as amino acids, sugars and salts. Together, these receptors provide the central nervous system of an animal with information about the nutritional composition of food before, during and after ingestion (Dethier 1976; Finger 1997; Yarmolinsky et al. 2009). Other chemoreceptors respond to deterrent and toxic compounds in the food (Schoonhoven et al. 2005; Dong et al. 2009). The presence or absence of nutrients that are not themselves tasted is ‘inferred’ by learning to associate features of a food with the consequences of ingesting it. Such associations may be positive (Sclafani 2000; Touzani and Sclafani 2005; Burke and Waddell 2011; Fujita and Tanimura 2011) or aversive (Bernays 1993; Tomé 2004; Dukas 2008), and may be made individually or in a social context (Galef and Laland 2005).

Nutritional state is assessed through systemic nutrient-sensing mechanisms and hormonal feedbacks from body reserves (Zhang et al. 1994; Crespi and Denver 2006; Morton and Schwartz 2011). Integration of information about food composition and nutritional state occurs both at the periphery, by nutrient-specific modulation of taste receptors (Simpson and Raubenheimer 2000; Newland and Yates 2008; Carleton et al. 2010), and more centrally as signals from systemic and peripheral sources converge onto the neural circuitry that controls feeding behaviour (Schwartz et al. 2000; Cota et al. 2007). Learning also plays a role, and in some cases animals are able to associate their current nutritional state with food cues previously associated with particular nutrients (e.g. Simpson and White 1990; Raubenheimer and Tucker 1997; Booth and Thibault 2000).

Postingestive regulatory responses assist in rebalancing an imbalanced nutrient intake. The gastrointestinal tract sits at the interface between the food and the internal milieu and is now appreciated to be a key site of regulation, being plastic in both structure and function in response to nutritional state (Raubenheimer and Bassil 2007; Clissold et al. 2010; Sørensen et al. 2010). Having been digested and absorbed across the gut, further rebalancing of nutrient supply can be achieved by differentially voiding excess nutrients and conserving nutrients that are in limited supply (Zanotto et al. 1997; Stock 1999; Silva 2006; Cypess et al. 2009).

Nutritional geometry has been used to relate nutrition to several different fitness-related consequences. Among the most interesting and significant are the relationships between nutrition, longevity and reproductive output.

Since a seminal publication on rats by McCay et al. (1935), the view that dietary restriction without severe malnutrition prolongs life has become the big idea in research on the biology of ageing (Weindruch and Walford 1988; Masoro 2005; Colman et al. 2009; Everitt et al. 2010). The list of species that live longer when modestly food deprived is now extensive: yeasts, nematode worms, fruit flies, rodents, monkeys and many more. But what does dietary restriction actually mean? The most widely accepted view is that eating fewer calories (called ‘caloric restriction’ or ‘calorie restriction’) is the reason for living longer, whatever the source of those calories might be, whether protein, carbohydrate or fat (Weindruch and Walford 1988; Masoro 2005; Everitt et al. 2010). Some had suggested, however, that calories may not hold the key to a long life; rather, specific nutrients such as proteins and certain amino acids may be responsible (e.g. Ross 1961; Zimmerman et al. 2003; Mair et al. 2005).

As discussed above, by restricting study animals throughout their lives to one of a sufficiently large number of dietary treatments it becomes possible to map response surfaces onto nutrient intake arrays for aspects of performance, including lifespan, rate of ageing, reproductive success and other variables. From these plots the main and interactive effects of different nutrients, as distinct from energy intake alone, can be separated and attributed (Simpson and Raubenheimer 2007; Archer et al. 2009).

Lee et al. (2008a) set out to try such an experiment on one of the best known model systems for ageing research, the fruit fly (Drosophila melanogaster). More than 1000 mated female fruit flies were kept individually on one of 28 diets varying in the ratio and concentration of yeast to sugar and intake and egg production were measured over the lifespan of each fly. Intakes of protein and carbohydrate, which are the major energy-yielding nutrients for Drosophila, were calculated from the volumes of diet ingested. The consequences for lifespan and egg production were examined by plotting response surfaces on top of the array of protein and carbohydrate intakes. Flies lived longest on a diet containing a 1 to 16 ratio of protein to carbohydrate and lived progressively less long as the ratio of protein to carbohydrate either decreased to zero or increased beyond 1 : 16. If lifespan corresponded with the number of calories eaten, as predicted by the calorie-restriction hypothesis, the lifespan contours would run parallel with lines of equal calorie intake (isocaloric lines). However, the contours of the longevity landscape ran almost at right angles to these lines. The data therefore prove that caloric restriction cannot explain the variation in lifespan; instead, the balance of carbohydrate to protein ingested correlated most strongly with longevity.

A parallel experiment was conducted by Fanson et al. (2009) on the Queensland fruit fly (Bactrocera tryoni). The results were very similar to those for Drosophila: dietary protein to carbohydrate ratio and not energy intake was strongly associated with lifespan. A subsequent experiment using a chemically defined mixture of amino acids instead of yeast confirmed that protein rather than some other correlated component in yeast was responsible for the life-shortening effect of a high-yeast, low-sugar diet (Fanson and Taylor 2011). Others reached similar conclusions regarding the importance of the protein to carbohydrate ratio in the diet for longevity in flies (e.g. Carey et al. 2008; Skorupa et al. 2008; Ja et al. 2009; Le Rohellec and Le Bourg 2009; Vigne and Frelin 2010), crickets (Maklakov et al. 2008), ants (Dussutour and Simpson 2009; Cook et al. 2010) and honey bees (Pirk et al. 2010), and the same may well be true for mammals (Orentreich et al. 1993; Zimmerman et al. 2003; Miller et al. 2005; Simpson and Raubenheimer 2007, 2009).

Living long enough to reproduce is the raison d’être of all organisms, so it is no surprise that when resources are scarce many organisms shut down reproduction and wait it out until conditions improve and the prospects for reproduction are brighter (Zera and Harshman 2001). Since ‘waiting it out’ may lead to organisms living longer than if they had reproduced earlier in life – sometimes tens or hundreds of times longer in the case of creatures that enter a state of suspended animation (Withers and Cooper 2010) – the idea arose that reproduction and ageing trade-off against one another by competing for resources. Hence, limited resources are allocated with highest priority to maintaining and repairing the organism’s body (its ‘soma’), thereby improving the chances that it will live long enough to experience better conditions and reproduce (Kirkwood 2005). The obverse of this trade-off hypothesis is that when there are sufficient resources to support reproduction, somatic maintenance is only allocated sufficient resources to survive the reproductive period – so, not only does inhibition of reproduction cause delayed ageing, reproduction effectively shortens lifespan. A variant of this idea is that reproduction doesn’t just compete with somatic maintenance for resources it produces its own damaging side-effects that shorten lifespan (Tatar 2007; O’Brien et al. 2008). Under all of these various manifestations of the trade-off hypothesis, maximal lifespan and maximal reproductive output are mutually exclusive.

We will come back to the issue of mutual exclusiveness, but we will first consider what is meant by ‘resources’. When considering the trade-offs between reproduction and ageing, ‘resources’ and ‘energy’ are usually used interchangeably. However, as we have discussed above, energy alone cannot explain ageing and lifespan: so how do lifespan and reproduction relate to one another in a more complex representation of the nutritional landscape? Lee et al. (2008a) found that mated female Drosophila lived longest on a diet containing a 1 : 16 ratio of protein to carbohydrate (P : C). The same flies, however, laid most eggs across their lifespan when confined to a diet comprising a more protein-rich diet, 1 : 4 P : C. An increase in P : C beyond 1 : 4 resulted in a drop in lifetime egg production. These results suggest two important conclusions. First, calories cannot explain the relationships between diet, longevity and reproduction; instead, these relationships can only be understood by considering nutrient balance. Second, because the performance landscapes for lifespan and lifetime egg production had different shapes and their peaks lay in different locations, flies cannot maximise both lifespan and lifetime egg production on a single diet.

An interesting question is which peak do the flies choose to climb if given a choice? To address this, Lee et al. (2008a) offered flies one of nine complementary food choices in the form of separate yeast and sugar solutions differing in concentration. The flies converged upon a diet comprising 1 : 4 P : C, thereby maximising lifetime egg production and paying the price of a diminished lifespan. The experiment of Maklakov et al. (2008) on field crickets added a new dimension by considering reproductive effort in both females and males. A major investment in reproduction by male crickets is calling to attract females. Not only is singing energetically expensive, but in addition to luring females it may attract the unwelcome attention of predators. For females, the number of eggs laid was counted, and for males the amount of time spent singing throughout their lives was recorded on 24 diets varying in protein and carbohydrate content. Both sexes of cricket lived longest on low P : C diets, and like flies, female crickets laid most eggs on a more protein-rich diet (P : C 1 : 1). Males, however, sang most on low P : C diets. For males, the response surfaces for lifespan and singing were similar, but female crickets, like female flies, were faced with a quandary: they could not live longest and lay most eggs on the same diet.

South et al. (2011) used nutritional geometry to show that nutrient balance plays another role in defining the battle of the sexes, this time in cockroaches. These authors manipulated the intake of proteins and carbohydrates by male cockroaches, Nauphoeta cinerea, and related this to their production of sex pheromones, their success in being chosen as mates by females, and their status among competing males. Carbohydrate (but not protein) intake affected sex pheromone production and male attractiveness, but did not affect male dominance status. When males were given a choice between complementary foods, they selected a higher carbohydrate intake, thereby smelling more attractive to females.

Returning to the matter of the mutual exclusiveness of reproduction and somatic maintenance, the peaks for maximal reproduction and longevity occupied different regions in nutritional space for female insects (although they came close to aligning for male field crickets), such that females could not achieve maximal fecundity and lifespan on the same diet. The trade-off theory proposes that longevity and reproductive output cannot occur simultaneously either because they compete for the same resources, or there are direct costs to reproduction that shorten lifespan. Is this true? Grandison et al. (2009) have shown that, for flies at least, the answer is ‘no’. When fed a low P : C diet, flies laid fewer eggs and lived longer than if fed a higher P : C diet. When a mixture of essential amino acids was added to the restricted diet, flies increased their egg production but lived less long – as expected from the trade-off hypothesis. However, when the restricted diet was augmented with only methionine, flies both lived long and had high egg production. Maximal longevity and fecundity can co-exist under the appropriate nutritional conditions; which means that the trade-off hypothesis as usually formulated is not correct, nor is the variant hypothesis that there are direct costs of reproduction that shorten lifespan (see also Flatt 2011; Tatar 2011).

To this point in we have explored geometric models that include the macronutrients as focal dimensions. This is for a reason: as the data show, macronutrients can explain a good deal of the variation in the behavioural, physiological and performance responses of animals. Macronutrients are, however, clearly not the only functionally important nutritional components of foods: the constituent molecules in macronutrients (amino acids and fatty acids, for example) and micronutrients such as vitamin and minerals also play a critical role in an animal’s nutritional strategies and physiology, as do other components of foods that would not normally be considered nutrients. Examples of the latter include refractory fibre (e.g. lignin from plants and chitin from some animal foods), and a range of secondary metabolites that play various roles from toxic defences to essential antioxidants and even antibiotic components that act as medicines (Huffman 2001, 2003; Villalba and Provenza 2007).

Conventionally, those chemical compounds in food that enhance health and fitness and elicit appetitive responses (e.g. amino acids, fatty acids, sugars) are categorised as ‘nutrients’, while those that are deleterious and trigger aversive and defensive responses (e.g. alkaloids, polyphenolics, terpenoids) are classified as ‘toxins’. In fact, ‘toxin’ versus ‘nutrient’ is a loose distinction, and for many purposes a more helpful dichotomy is between the adjectival versions ‘toxic’ versus ‘nutritious’ or, more generally, ‘deleterious’ versus ‘beneficial’ (Berenbaum 1995; Raubenheimer and Simpson 2009). We will use two well established phenomena to illustrate this point: hormesis and Bertrand’s rule.

Hormesis is a concept developed in toxicology, in which the effects on biological systems (cells, tissues, organs, organisms, populations) of exposure to a substance are reversed with increasing concentration (Calabrese and Baldwin 2003). Hormetic dose–response relationships can take two forms. The most common of these is the inverted U-shape, in which low doses of a substance stimulate and high-doses inhibit beneficial biological responses (e.g. growth, fecundity or longevity). A second form is the ‘J-shaped’ curve, where low doses reduce and high doses exacerbate a deleterious response (e.g. tumour formation, mortality, or growth suppression).

Over recent decades, it has become apparent that hormetic responses are common, having been observed for a wide range of chemicals, organisms, and biological responses (Calabrese et al. 1999; Calabrese 2005). We have argued (Raubenheimer and Simpson 2009), however, that this unification will remain incomplete until it encompasses parallel developments in the nutritional sciences.

Although some researchers in the area of hormesis have recognised similarities between toxin and nutrient dose–response curves (Luckey and Stone 1960; Hayes 2007), there have been separate developments in the nutritional sciences that are directly relevant. These developments stem back to the French scientist Gabriel Bertrand, who in 1912 established a rule concerning the dose–response curve for mineral nutrients, which is essentially that too little or too much of a mineral impairs biological function.

Bertrand’s rule is believed to apply to all essential micronutrients, with the detailed shape of the curve depending on the nutrient in question and the biological context (Mertz 1981). It was generally assumed that Bertrand’s rule does not apply to macronutrients, despite evidence to the contrary – not least of which is the starvation (costs of energy shortage) to obesity (costs of energy excesses) spectrum in modern humans. Indeed, the concepts of targets and rules of compromise in the GF are predicated on there being costs to excesses as well as deficits of macro- and micronutrients. That there are indeed measurable costs to both excesses and deficits of macronutrients is now experimentally verified in several instances (Raubenheimer et al. 2005, 2009; and see previous section). We suspect that Bertrand’s rule is at least as prevalent in nutrition as is hormesis in toxicology (see also Boersma and Elser 2006).

As we have seen, there is good evidence that animals balance their intake of foods so as to avoid ingesting both excesses and deficits of specific nutrients, and in many cases a good deal is known about the mechanisms involved. The recent literature shows that animals likewise regulate their intake of toxins so as to avoid ingesting a toxic overdose – leading to the term ‘nutritional toxicology’ being coined (Torregrossa and Dearing 2009). An important question is whether animals also select foods specifically in the context of ingesting toxins at levels that are beneficial. The interesting phenomenon of ‘self-medication’ suggests that they do.

Self-medication, or zoopharmacognosy (Rodriguez and Wrangham 1993), is the phenomenon in which animals use plant secondary metabolites or other non-nutritional substances to prevent or treat disease (Huffman 2003; Villalba and Provenza 2007). Although much remains to be done in this field (Lozano 1998; Hutchings et al. 2003), research shows that poisoned or parasitised animals can specifically select foods containing non-nutrient compounds that help them to recover. Might they also select diets with a nutritional composition that neutralises toxins or reduces parasite load and associated disease? In addressing this question for parasitic infection, Hutchings et al. (2003) discussed several experiments that have shown a link between diet and immune responses. Lee et al. (2006a) found that caterpillars exposed to a highly virulent nucleopolyhedrovirus resisted infection better as the proportion of protein in the diet increased (Lee et al. 2006a , 2008b ). A similar response was reported by Peck et al. (1992), who found that mice survived better on diets containing a higher ratio of protein to carbohydrate after being inoculated with Salmonella typhimurium. In the study by Lee et al. (2006a), uninfected caterpillars performed best on a diet with lower protein concentration than the infected individuals. When allowed to self-select their diet, caterpillars that survived infection increased their relative intake of protein, compared with controls and with caterpillars that died of infection. This experiment demonstrates that these caterpillars are able to combat viral infection by modulating the macronutrient composition of their diet.

Povey et al. (2009) extended the study of Lee et al. (2006a) to another pathogen, the bacterium Bacillus subtilis, with similar results, and also found that components of the immune system of the caterpillars interacted with one another in relation to diet. Elevation of some immune components was associated with a decline in others. The traditional view of such immune trade-offs is that different components compete for a limiting pool of resources, typically energy. However, as we saw above for lifespan and reproduction, another possibility is that, rather than competing for a single limiting resource, there is something akin to internal ‘niche partitioning’ among different elements within the immune system, with each component having a different nutritional optimum within a multidimensional nutrient space.

This idea was explored by Cotter et al. (2011) in experiments on caterpillars in which individuals were restricted to one of 20 diets varying in the quantity and ratio of protein and carbohydrate. Results indicated that immune traits do in fact exist within an organismal ‘ecology’, in which each trait has its own specific nutritional requirements. This was shown by plotting response surfaces for different immune traits onto nutrient intake arrays. No one diet simultaneously optimises all immune components, but the opportunity exists for the animal to adjust its food-selection behaviour to attain a nutritional state that supports the balance of immune responses that best resists infection by a particular pathogen.

The other side to ecological immunology, from a nutritional perspective, is that parasites and pathogens rely on the host for provision of resources. Because these organisms may not share the same nutritional requirements as their hosts, there are possibilities for resource competition and manipulation between the different parties (Smith and Holt 1996; Ponton et al. 2011a , 2011b ). The complexity of the nutritional interactions between hosts and pathogens is made even greater because animals play host to entire communities of commensal and symbiotic microorganisms, which receive their nutrition from the host and in turn contribute essential nutrients (Turnbaugh et al. 2009; De Filippo et al. 2010; Kau et al. 2011). These interactions are ripe for geometric analysis (Ponton et al. 2011b ).

We have argued that for many purposes nutrient and non-nutrient components cannot be simply distinguished, because both can be nutritious, medicinal or toxic. On the other hand, there clearly are cases in which nutrients are nutritious and toxins are categorically deleterious – as is true for many defensive compounds (Sotka et al. 2009). Even when toxins can be distinguished categorically in this way, it is often the case that their effects can only be understood in the context of the background nutritional milieu. The modes of interaction of nutrients and toxins are diverse, involving intake, digestion and absorption, as well as postabsorptive effects (Slansky 1992; Sotka et al. 2009).

Application of the GF to a study of nutrient–toxin interactions shows how complex and important these interactions can be (Simpson and Raubenheimer 2001). Juveniles of Locusta migratoria were confined to one of five diets varying in protein to carbohydrate ratio (P : C) and containing either 0, 3.3, 6.7 or 10% tannic acid (TA). The striking result was that locusts were impervious to TA, even at the highest level of 10%, providing that the diet was near balanced in its P : C relative to the intake target. As the diet became either protein- or carbohydrate-biased, locusts were progressively more susceptible to TA; however, the reason why nutritional imbalance affected susceptibility to TA differed according to the direction of nutritional imbalance. As the diet became more carbohydrate-biased, insects suffered because of the increasingly powerful antifeedant effects of TA. TA mixed in a low P : C food caused locusts to eat so little that they starved to death. By contrast, food intake was not affected on high P : C diets, but instead TA had its toxic effects postingestively. Behmer et al. (2002) next explored the interactive effects of tannic acid and macronutrients in a more complex environment, in which two nutritionally complementary foods (one high in carbohydrate, the other high in protein) containing TA were provided along with a third, TA-free food. Providing that the TA-free food was carbohydrate-biased, locusts were able to maintain their macronutrient intake target by mixing this food with the TA-containing high-protein food. However, they abandoned intake target regulation if the TA-free food was protein-biased, because of their aversion to TA-containing high-carbohydrate foods.

Intake target, rules of compromise and postingestive responses are fundamental, inter-related traits, against which the adequacy of diets and nutritional environments can be calibrated and the consequences of nutritional imbalance assessed. For clarity, we have treated intake and growth targets as static points integrated across a particular period in the life of an animal. In reality they are trajectories that move in time. In the short term, the requirements of the animal change as environmental circumstances impose differing demands for nutrients and energy. At a somewhat longer timescale, targets move as the animal passes through the various stages of its life, from early growth and development to maturity, reproduction and senescence. On an even longer time-scale, nutritional traits are subject to natural selection and move as species evolve to exploit new or changing nutritional environments and to adopt differing life-history strategies. Presaging such evolutionary change in gene frequencies within populations are epigenetic effects, whereby the nutritional experiences of parents influence the behaviour and metabolism of their offspring without requiring changes in gene frequencies.

An animal’s nutritional requirements will depend on the physiological demands placed upon it and the intake target should move accordingly. Another major influence over nutritional requirements is environmental temperature. Nutrient intake varies as a function of body temperature, as do metabolic rate and the efficiency with which ingested nutrients are utilised (e.g. Angilletta 2009; Miller et al. 2009; Coggan et al. 2011). In warm-blooded animals the target mixture of nutrients can vary as a function of the energy demands for thermoregulation, as seen in the increased consumption of carbohydrate but not protein in weanling rats kept at 8°C rather than 23°C (Musten et al. 1974; Simpson and Raubenheimer 1997). As we have already discussed, yet another change in environmental circumstances that will shift the intake target is the need for animals to meet the nutritional demands associated with fighting pathogens and mitigating the effects of toxins.

Changes in demand for nutrients imposed by activity, temperature, pathogens, toxins or other short-term influences will act via the suite of regulatory mechanisms briefly summarised above, but over what periods are these changes detected and tracked? In theory an animal with appropriate complementary foods available could adjust its feeding behaviour and food choices mouthful by mouthful to track its moving intake target with very high resolution; but each time the animal shifts between foods it risks attracting unwarranted attention from natural enemies and wastes both time and energy (Houston and Sumida 1985). At the other extreme, an animal might switch just once, maintaining fidelity to one food until it reaches the point where it must swap foods if it is to attain its target integrated over some longer period. However, by following this strategy the animal will suffer the cost of increased time in a state of nutritional imbalance (far from the target trajectory). The optimal switching interval for a real animal under natural conditions should reflect the relative costs of staying on an imbalanced food versus those of moving to find a complementary food (Chambers et al. 1997; Simpson et al. 2010).

As animals grow, develop, reproduce and grow old, their nutritional requirements change, both in the amount and blend of nutrients needed. For example, young rat pups require protein, carbohydrate and fat to grow, but once they reach puberty their intake of protein declines and carbohydrate and fat dominate the diet (Leibowitz et al. 1991). Male and female rats have different intake trajectories, with males consuming more protein as they grow, presumably to sustain their greater size and muscle mass. When females reach maturity, their macronutrient intake varies as a function of reproduction. During pregnancy they ingest more protein than before mating, but not more carbohydrate or fat. When suckling the resulting pups, mothers increase intake of protein and fat but not carbohydrate. After the pups are weaned, the mother reduces her protein intake to premating levels but sustains high fat intake to restore body reserves that have been depleted during reproduction (Simpson and Raubenheimer 1997).

There are two basic mechanisms for tracking developmental shifts in the intake target. Barton Browne (1995) termed these ‘demand-mediated’ and ‘non-demand-mediated’ mechanisms. The former are those mechanisms involving dynamic responses to changing nutritional needs. Hence, the animal experiences a developmental change in patterns of metabolism and growth (induced, for example, by environmental or hormonal signals) and the resulting shift in demand for nutrients causes an appropriate change in feeding behaviour. Alternatively, the developmental switch may itself cause the shift in diet that achieves a new intake target without involving nutrient feedbacks, as seen, for example, in female blow flies and mosquitoes, which upon mating start to develop eggs and become responsive to odours from carrion and live animal hosts, respectively (Barton Browne 1995). Attraction to these resources results in ingestion of high-protein food (rotting flesh or fresh blood) that supports egg development.

Recent studies on the fruit fly, Drosophila melanogaster, have begun to provide molecular insights into the relationship between demand-mediated and non-demand-mediated mechanisms associated with female reproduction. Ribeiro and Dickson (2010) found that the rapidly induced preference for protein-rich yeast in female flies after mating results not from depletion of protein reserves as eggs are developed, but as a direct result of a sex peptide that is introduced with the male’s seminal fluid during mating and stimulates special sensory neurons in the female’s reproductive tract. An additional demand-driven mechanism then modulates how much yeast is eaten, involving TOR/S6 kinase and serotonin signalling pathways in the central nervous system (Ribeiro and Dickson 2010; Vargas et al. 2010).

Some changes in the nutritional environment are sufficiently persistent that intake targets, rules of compromise and postingestive responses need to shift across generations. The main mechanism allowing the phenotype to track changing environments over generations is, of course, natural selection acting via changing gene frequencies within populations. However, it has become increasingly appreciated that effects of the nutritional environment experienced by one generation can be passed to subsequent generations without requiring changes in gene frequencies. For example, dietary-induced obesity in mothers and even grandmothers has a direct influence over the metabolism and risk of obesity in offspring (Barker 1998; McMillen and Robinson 2005; Gluckman and Hanson 2006a ). Obesity in fathers can, it seems, also affect metabolic function in offspring (Ng et al. 2010). That nutritional consequences in one generation can reverberate and become amplified across subsequent generations has profound health and socioeconomic implications (Gluckman et al. 2009).

The search for controlling mechanisms of epigenetic effects is well underway. A major focus of research has been on inherited patterns of gene activation, involving processes such as gene methylation, histone modification and involvement of micro-RNAs, but other pathways are also possible, including changes in milk production by lactating mothers and cultural influences. Another focus of interest is the extent to which epigenetic influences represent adaptive developmental responses, or are simply unavoidable changes that do not improve or might even reduce the fitness of the offspring. Responses that are specifically adaptive have been termed ‘predictive adaptive responses’, to denote the fact that the developmental trajectory is cued to take a course that is adaptive (beneficial) with respect to the predicted state of the future environment. Such responses can, however, lead to problems when there is a mismatch between the predicted future environment and that actually encountered (Gluckman and Hanson 2004, 2006b ).

As organisms evolve to exploit different nutritional niches or to track changing nutritional environments, their intake targets will evolve to match the diet. As a result, the optimal diet composition will differ substantially among organisms. The optimal diet for an aphid that feeds on sugary plant sap low in nitrogen is not the same as that for a locust that eats whole plant tissues, or a carnivore that eats predominantly meat. Comparative data on intake targets from many species can be used to explore the association between an animal’s diet, its life-history and evolutionary past. One example concerns an analysis in which the optimal protein to carbohydrate ratios (P : C) in the larval diets of 117 insect species were estimated from a combination of geometric experiments and a published compendium of diet recipes used for rearing insects in the laboratory (Simpson and Raubenheimer 1993). These ratios were superimposed upon a phylogenetic tree of the insect groups, to control for shared ancestry when associating diet composition with features of life-history. The strongest pattern to emerge from the comparative analysis was the association between a reduced P : C and the acquisition of endosymbiotic microorganisms. Species with endosymbionts have a capacity to thrive on nitrogen-poor diets, such as phloem sap (aphids), dead wood (borer beetles) and detritus (cockroaches). Indeed, such insects have evolved a mutual association with symbionts precisely because they provide the host with a novel metabolic capability – the upgrading of dietary nitrogen (Douglas 2010). Possessing symbionts allows the host to exploit low-protein resources that are inadequate for other species. The corollary of acquiring this new metabolic capacity is that diets with higher level of protein, although optimal for other species, have become nutritionally imbalanced for species with endosymbionts (Abisgold et al. 1994).

Not only is the intake target dynamic over physiological, developmental and evolutionary timescales, so too are rules of compromise. An instructive example comes from a consistent difference found in rules of compromise between host plant specialist and generalist insect herbivores (Raubenheimer and Simpson 1999, 2003; Simpson et al. 2002; Lee et al. 2002, 2003, 2006b ; Raubenheimer and Jones 2006). When rules of compromise for protein and carbohydrate were derived, generalists were consistently more willing to overeat surplus nutrient in nutritionally imbalanced diets than were specialists. A plausible explanation for this difference relates to the degree of nutritional heterogeneity experienced in the environment of specialists and generalists (Raubenheimer and Simpson 1999). A compatible prediction is that generalists should be better able to deal with excesses postingestively. This was confirmed in studies of locusts and caterpillars (Simpson et al. 2002; Lee et al. 2002, 2003, 2004, 2006b ; Raubenheimer and Simpson 2003). In each instance, nutrient-utilisation budgets indicated that generalists were better able than specialists to use excess ingested protein on high-protein, low-carbohydrate diets for energy metabolism, thereby simultaneously reducing the cost of surplus ingested protein and offsetting the carbohydrate deficit (see Lee et al. 2003).

An example of how postingestive responses can change over generations in response to a shift in the nutritional environment comes from the work of Warbrick-Smith et al. (2006), who reared multiple lines of the diamondback moth, Plutella xylostella, for eight generations either on a carbohydrate-rich or a protein-rich diet. The carbohydrate-rich diet comprised either chemically defined artificial food or a high-starch mutant of the plant Arabidopsis. Over the eight generations, caterpillars progressively developed the ability to eat excess carbohydrate without laying it down as body fat, a result that provided strong evidence that storing excess fat has fitness costs. Other replicate lines were reared for eight generations in a protein-rich, carbohydrate-scarce environment, comprising either artificial diet or a low-starch Arabidopsis mutant. In contrast to insects reared under a high-carbohydrate environment, these caterpillars developed an increased propensity to store ingested carbohydrate as fat. Female moths also developed a preference for laying their eggs on the low-starch plant, whereas those selected on the high-starch Arabidopsis mutant showed no preference.

By becoming less prone to laying down body fat in a high-carbohydrate world, caterpillars could minimise the costs of obesity – but at the risk of starving for want of energy reserves if food became scarce. By contrast, in a low-carbohydrate world, retaining and storing carbohydrates as fat was favoured; but should such a phenotype be placed into a high-carbohydrate world it would be at risk of obesity. It was not known whether these changes in physiology were the result of genetic selection or accumulating epigenetic effects (see above).

So far, we have viewed the world of nutrition from the perspective of individual animals. But individuals interact with one another – in family groups, aggregations, swarms and societies – and many of these interactions involve nutrition. Next we show how an individual’s nutritional state can influence the way in which an entire group behaves.

Late spring and early summer in western North America regularly sees millions of large, flightless Mormon crickets, Anabrus simplex, forming marching bands that extend for up to 10 km in length. Even though Mormon crickets are omnivores with highly catholic food tastes, they do not strip the habitat bare as they go, hence if limitation of some nutritional resource explains mass migration then it must be more specific than edible food per se.

Experiments indicated that Mormon crickets in the marching bands were selectively feeding for protein and salt. When crickets were taken from the band and housed with access to plentiful amounts of protein-rich and carbohydrate-rich foods in a simple GF experiment, they initially ate predominantly protein-rich food, but over the next day they mixed a diet that was much more typical of other crickets and grasshoppers, containing protein and carbohydrate in near equal proportions. In other words, the crickets in the marching bands were not obligatorily protein-seeking, but rather they were selectively deprived of protein as a result of local environmental conditions. However, a highly abundant source of lightly salted protein in the midst of a cricket horde is other Mormon crickets. So, is the reason that cricket bands migrate because they go on a forced march for protein and salt, with each insect chasing the moving meal in front and escaping the cannibals behind? This macabre hypothesis makes two predictions: (1) if crickets become less mobile they should stand a higher chance of getting cannibalised, and (2) satiating crickets with protein and salt should reduce cannibalism and marching. Both of these predictions were met in experiments conducted in the field (Simpson et al. 2006).

The conclusions from these experiments and others by Sword et al. (2005) are that mass migration at the group level emerges as the indirect effect of individual nutritional responses, mediated via cannibalistic interactions between individuals. As a result, there is a two-fold incentive to move, to escape the cannibal behind and to catch the meal in front, and a Mormon cricket band travels 2 km each day rather than only tens of metres when the crickets are in low-density populations (Lorch et al. 2005). The consequence of migration is a greatly increased chance of encountering fresh nutritional resources but, meanwhile, the crickets are ‘travelling with their lunch’ (Hansen et al. 2011). Bazazi et al. (2008, 2011) subsequently showed that protein deprivation is also associated with enhanced marching in groups of cannibalistic desert locust nymphs.

Social insect colonies are far more collegial than Mormon crickets and locusts: interactions among individuals somehow coordinate the activities of the entire colony so that it acts as a nutrient-acquiring, -distributing and -digesting ‘superorganism’ (Hölldobler and Wilson 2009) that is able to meet the disparate needs of all its members. Indeed, it may well be that nutrition has provided a major feature in the evolution of advanced sociality among animals, including the ants, bees and wasps (Hunt and Nalepa 1994; Kamakura 2011).

The complication for social animals such as ants is that a minority of individuals collects the food for the colony, and these foragers have very different nutritional requirements to other members of their colony (Cassill and Tschinkel 1999; Behmer 2009). In particular, the egg-laying queen and the legless larval ants have a much higher need for protein than do workers, but they do not contribute to food collection. A forager ant who responded only to her own needs would collect carbohydrate-rich nectars and exudates in small amounts, since she is already fully grown. If all foragers behaved in this manner, the brood, queen, nurse ants and other non-foraging workers would soon starve to death. How then does the entire colony communicate its needs to the forager ants?

To answer this question, Dussutour and Simpson (2008) set about posing nutritional conundrums for an Australian species, the green-headed ant, Rhytidoponera metallica. Results indicated that ant colonies compensate over time by increasing their collection of dilute over concentrated sugar solutions. Initially, ants consume most and recruited in largest numbers to a concentrated sugar solution, as would be expected were ants maximising their rate of sugar acquisition. However, over successive days, the numbers of ants recruited and the volumes collected by each ant reversed, such that the greatest amount was consumed by colonies with the dilute solution and least by those with the concentrated solution. Ants regulated sugar collection more precisely when the number of larvae was increased than when the number of adult workers was doubled, indicating that larvae were the source of nutritional feedback from the colony to the workers.

Dussutour and Simpson (2009) next challenged colonies of green-headed ants with or without larvae to maintain intake of both protein and sugar by giving them choices between two foods varying in the ratio and concentration of protein and carbohydrate. Colonies with larvae adjusted the amounts of each food collected to maintain the ratio and amounts of protein and sugar collected remarkably constant. Colonies without larvae differed in three respects: (1) they regulated to a lower protein to carbohydrate ratio, (2) they collected less total quantity of nutrients, and (3) they were unable to respond as effectively to dilution of nutrients in the food. Hence, the presence of fast-growing, energy- and protein-demanding larvae changed both the quantity and blend of nutrients required and determined the effectiveness of nutritional regulation.

As we have seen, the success of individual organisms can be related to their intake of multiple nutrients by mapping measures of performance (lifespan, reproductive output, immunity to disease, and so on) as landscapes onto nutrient intake arrays. And the nutritional responses of individuals can, in turn, fashion the behaviour of groups and societies. We will now take this one step further to consider the consequences of individual nutrition for populations and the assemblages of species that comprise ecological communities. We will argue that trophic dynamics are in large part an emergent property of regulation of nutrient balance, and that such regulation takes place at all trophic levels.

If the nutritional responses of individual organisms represent those of the species, it ought to be possible to use performance landscapes measured from a representative sample of individuals to model the success of an entire population of that species. Ecologists measure the success of a population in terms of its population growth rate (pgr). By definition, a population will increase in an environment when pgr is greater than zero. Population responses have been described in similar geometric terms to those we have used for individual organisms, most famously by Tilman (1982), who classified population growth rate responses as surfaces in a two-resource space.

While it is undoubtedly the case that population growth rates can be defined within the same multidimensional nutritional spaces used to describe the responses of individual organisms, populations are not the simple sum of their individual parts: members within a population interact with one another and with the rest of their environment. The direct and indirect interactions occurring among individuals, both within and between species, can have profound effects on a population’s growth rate and may result in very different responses to diet composition than would be predicted from the study of individual organisms (e.g. Kytö et al. 1996). We cannot understand the nature and consequences of such interactions unless we know the differing nutritional needs, priorities and regulatory capacities of the different interacting organisms.

The most direct nutritional interaction of all is when one organism attempts to eat another, but the interaction might be ecologically significant even if the prey is not killed. It might, for example, change its subsequent behaviour to avoid the risk of further attack, or the interaction might result in the induction of defensive chemicals and immune responses to prevent or counteract future threats (Karban and Baldwin 1997). Being removed from the environment as a living organism and returned as faeces – or being changed in behaviour, chemical composition and defensive armoury – may resonate throughout the ecosystem in unexpected ways (Simpson et al. 2010).

This is illustrated by the study of Hawlena and Schmitz (2010) on the North American grasshopper Melanoplus femurrubrum. This herbivore selects a diet of grasses and herbs and is preyed upon by hunting spiders, such as Pisuarina mira. Hawlena and Schmitz housed grasshoppers in the presence of spiders, which were kept in clear plastic cylinders where they were visible but unable to attack the grasshoppers. Compared with controls that were not exposed to spiders, the grasshoppers selected a diet containing 40% more carbohydrate, whereas protein intake was not affected by the presence of spiders. This shift in the intake target towards a higher carbohydrate to protein ratio was due to grasshoppers exposed to spiders having an increased metabolic rate: the ‘fear effect’. There was also a change in the chemical composition of the grasshopper’ faeces, which were lower in carbon to nitrogen (C : N) ratio when spiders were around, reflecting the higher metabolic need for carbohydrate. Finally, the body composition of spider-exposed grasshoppers differed, being higher in C : N than controls. Hawlena and Schmitz calculated that the change in the grasshoppers’ intake target when at risk of spider predation would shift the structure of the natural host plant community as a result of selective foraging by fearful grasshoppers. The changes in body and faecal chemistry and plant community structure would, in turn, affect the chemical composition of the pool of detritus in the ecosystem.

Trophic interactions may also act over longer timeframes. For example, any characteristic that influences an organism’s susceptibility to being consumed is likely to be under strong genetic selection, which, over evolutionary time, will allow such genetic influences to affect other trophic levels (Shuster et al. 2006). Traits that are heritable can result in patchy nutritional environments at multiple scales. For instance, foliar concentrations of plant secondary metabolites that are toxic to mammals are genetically determined in Eucalyptus species. Because of the spatial distribution of different genotypes of trees, herbivorous marsupials such as possums and gliders must move over distances greater than 40 m to encounter significant variation in plant secondary chemistry within a eucalypt forest (Andrew et al. 2007).

Understanding nutritional interactions and their effects requires that we know the differing nutritional requirements, priorities and regulatory abilities of the various interacting organisms. Herbivores and omnivores, including species of insects, birds and mammals, have been shown using GF designs to regulate their intake of macronutrients and some micronutrients and to make postingestive adjustments to help attain the optimal balance of nutrients to meet their various requirements. In contrast, the prevailing view has been that predators do not need to practice nutrient balancing and are unlikely to possess mechanisms to do so. Recent work has shown that this view is incorrect, with evidence for separate regulation of protein and lipid intake coming from invertebrate predators (Mayntz et al. 2005; Raubenheimer et al. 2007; Jensen et al. 2011, 2012), mink (Mayntz et al. 2009), predatory fish (Sánchez-Vázquez et al. 1999; Rubio et al. 2003), and domestic cats (Hewson-Hughes et al. 2011).

Most theories for explaining food web dynamics have been based on fluxes of mass and energy between consumed and consumers (but see DeAngelis et al. 1989; Sterner et al. 1996; Sterner and Elser 2002). We have shown several times already in this review that unidimensional models are insufficient to capture the essential features of nutritional systems, and we suspect the same is true for trophic ecology. We would argue that an important aspect of the animal that needs to be represented in models of food webs is the concept of nutritional targets – intake, nutrient and growth targets. These are fundamental in ecological interactions because targets provide a good deal of predictability about the nutritional and foraging decisions of animals. We believe that incorporating the target concept into models of ecological communities can, similarly, help to predict and understand the fluxes of nutrients in ecosystems (Raubenheimer et al. 2009).

As trophic levels are ascended, regulatory feeding and growth responses will progressively narrow the range of food compositions and, more importantly, shift the mean composition towards a higher proportion of protein (Raubenheimer et al. 2009). This predicted pattern is supported by data on invertebrates showing that the percentage of body nitrogen increases as trophic levels are ascended, whereas the ratio of carbon to nitrogen (C : N) in food versus the body of consumers (C : N food/C : N consumer) narrows progressively (Denno and Fagan 2003). Denno and Fagan (Denno and Fagan 2003; Fagan and Denno 2004) concluded from these trends in body elemental composition that carnivorous arthropods are nitrogen (protein) limited. This, they suggested, explains the tendency for many predators to demonstrate trophic omnivory – to feed not only on herbivores, but also on other carnivores (which, as we have seen, tend to have higher nitrogen concentration). However, taking into account the respiratory needs of carnivores we would argue that the reverse is true: consumers become progressively non-protein energy-limited as trophic levels are ascended. This will be especially, but not exclusively, the case when body fat stores need to be increased to meet the demands of periods of starvation, hibernation, reproduction or migration (Raubenheimer et al. 2007). It is this energy shortage that motivates feeding back down the food chain (Raubenheimer et al. 2009) and perhaps explains why populations of predatory marine birds and sea lions declined in the Gulf of Alaska after the loss of high-fat species of fish (Rosen and Trites 2000; Romano et al. 2006). Recently, Wilder and Eubanks (2010) have reiterated and elaborated upon the logic of this argument.

We have seen that it may not be possible to predict the growth rate of a population or species from the nutritional responses of individuals, because of direct and indirect interactions with other organisms within food webs. To use the terminology of niche theory, the fundamental nutritional niche of a population may not be realised (Kearney 2006).

The GF shares fundamental features with Hutchinson’s (1957) niche concept, focusing on the resources component of a more inclusive Hutchinsonian environmental space. To broaden the power of GF models to predict the performance, population dynamics and distribution of species, it will be necessary to incorporate abiotic influences as extra dimensions and to transpose this niche space onto real environments in space and time (Kearney et al. 2010). Effective models in nutritional ecology must, therefore, not only be ‘nutritionally explicit’ (able to express a situation in terms of nutrition), but also ‘organismally explicit’ and ‘ecologically explicit’ (able to incorporate organism-level and ecological factors in the model) (Raubenheimer et al. 2009). We now consider some ways in which the GF can be made more ecologically explicit, by incorporating abiotic and biotic aspects of the ecological environment, so enhancing the capacity of this approach to integrate nutrition into ecosystem-level analyses.

In addition to food resources, Hutchinson’s niche model includes dimensions for abiotic factors such as temperature, light, humidity and pH. A research field known as ‘Biophysical Ecology’ has developed to represent the ‘climatic’ niches of organisms in ‘climate space’ (Porter and Gates 1969). These models define the combinations of environmental variables that are suitable for survival and reproduction for a species, and have been applied at the scale of landscapes to predict the distribution, temperature, energy and water relations of organisms (Kearney and Porter 2009).

Abiotic factors can shape nutritional interactions in two major ways. The first is through influences on the physiology and behaviour of organisms. For example, changes in body temperature alter metabolic rate, rates of water loss, amounts and ratios of nutrients required, and tolerance to toxins and diseases. Hence, a decrease in ambient temperature results in rats increasing carbohydrate or fat consumption but not changing their protein intake (Aubert et al. 1995), and changes in temperature affect tolerance of white-throated woodrats to plant secondary metabolites (Dearing et al. 2008). Miller et al. (2009) showed that locusts developed faster but converted protein and carbohydrate less efficiently to growth if kept at 38°C rather than at 32°C. If given a choice of temperatures under conditions of food abundance, locusts selected 38°C, therefore prioritising development rate over utilisation efficiency. In a subsequent experiment, Coggan et al. (2011) allowed recently fed locusts to select a temperature within a thermal gradient, and after 50 min (the average intermeal interval) they either fed the locusts again or left them without food. Locusts initially selected a temperature of 38°C and returned to that temperature after having been fed again. However, if they were not fed at 50 min, they soon moved to a cooler region of the gradient, 32°C; they were now prioritising efficient extraction of nutrients in the previous meal over rapid growth in response to the absence of food.

The second way in which abiotic factors influence nutritional interactions is through the distribution of environmental conditions in space and time. Environmental gradients, concentrations and ratios of chemical elements in soil or water, microclimatic conditions and other habitat features will shape patterns of food availability over multiple spatial and time scales and will interact with an animal’s size, mobility and regulatory behaviour (Pincebourde and Casas 2006; Kearney and Porter 2009).

Kearney et al. (2010) proposed that Dynamic Energy Budget (DEB) theory (Sousa et al. 2008; Kooijman 2009) could be used as a ‘computational engine’ for integrating the GF and Biophysical Ecology, allowing population responses to be modelled dynamically within environmental space. A special type of DEB model is needed to integrate DEB theory with GF – one that includes a separate reserve for each nutritional component (Kearney et al. 2010). An example of how a two-currency DEB model could be used as a computational engine for implementing GF designs to predict population responses is shown in Kearney et al. (2010), who applied a DEB model developed by Kuijper et al. (2004) to calculate egg production in a copepod as a function of ingested carbohydrate and protein. DEB could generate at the level of populations (in the absence of other biological interactions) the types of responses seen in GF studies on individuals.

An appropriately defined DEB model offers a means of implementing GF models and calculating the fundamental niche of an organism, but the problem of dealing with interactions among organisms – of calculating the realised niche – remains. Since interactions among individual organisms are the most fundamental units from which ecosystems are built, we have argued that agent-based models may provide a means of scaling from individuals to ecosystems (Simpson et al. 2010). Agent-based models (ABMs) have been used to explore patterns of resource exploitation in foraging animals (Oom et al. 2004; Grimm and Railsback 2005; Nonaka and Holme 2007), but they have yet to take account of multiple nutrient dimensions; in other words, they are not yet nutritionally explicit (Raubenheimer et al. 2009; Simpson et al. 2010). However, if ABMs were to incorporate aspects of GF models, they would offer the promise of dynamic, spatially explicit models of nutritional interactions, which could take account of an individual organism’s simultaneous membership of a group, population, community and ecosystem. From such ‘heterarchical’ models (Raubenheimer et al. 2009) may well emerge step-shifts in organisational complexity, such as have been observed already at the level of groups and populations.

Much of the information required for such a model already exists and other data could readily be collected (Simpson et al. 2010). The aim in future will be to use such a platform to explore the evolution of nutritional phenotypes, to model populations, and to help explain the structure and dynamics of species assemblages within changing environments. If successful, the outcome would be a synthesis of nutritional ecology that is dynamic as well as nutritionally, organismally, ecologically and spatially explicit.

We will end with a consideration of the role that GF can play in wildlife conservation. A challenge arises in conservation ecology when the natural habitat of wild animals is diminished or altered, and decisions need to be made about providing a nutritional environment that will sustain the species. In some cases this will involve setting priorities for conserving remnants of the natural habitat, or identifying suitable habitats for establishing translocated populations. In other, usually more extreme, cases the natural diet is not available in sufficient quantities, and therefore needs to be supplemented or replaced by alternative foods. We now provide an example where the GF has been applied in the former scenario. For an example of the latter, see Raubenheimer and Simpson (2006).

At one level, it might be argued that there is no need to perform nutrient-level analyses of the foraging habitats of wildlife, and even less so the nutritional regulatory responses of the target species. In this line of thinking animals eat foods, are pretty good at knowing what they want and how much to eat, and as long as management decisions are geared towards conserving sufficient quantities of the relevant foods, then from the nutritional point of view the job of the conservationist is complete. This logic might well work in situations where the evolved link between the distribution of foods in the environment and the animal’s nutritional regulatory system remains intact. However, where this is not the case – for example, where the habitat in which the regulatory systems evolved is diminished or altered – then there might be a mismatch between the regulatory systems and the altered environment in which they need to operate. To understand, avoid or manage this mismatch, we need to think in terms of foods as well as their nutrient and other components.

A recent example where nutritional geometry was used to understand the habitat requirements of a priority conservation species was the study of Peruvian spider monkeys, Ateles chamek, in their natural forested habitat in Bolivia (Felton et al. 2009a , 2009b ). Felton first spent five months habituating the monkeys. When they were sufficiently comfortable with her presence, she followed individual monkeys throughout their feeding day (from dawn to dusk), recording everything they ate during the 12-h observation period and estimating the amount of each food that was eaten. Representative samples of each food item were collected, and later taken back to the laboratory and their nutritional compositions analysed. From these data, 38 daily intakes of protein, fat and carbohydrate were estimated and analysed using the GF.

Several interesting results emerged. First, although primarily frugivorous, the monkeys ate a range of foods that differed widely in their macronutrient composition. When available, they specifically targeted ripe figs, Ficus spp., but when these were not available in sufficient quantities they mixed their intake from foods with high and low protein : non-protein energy balances, thereby moving in a zig-zag pattern through nutrient space. Interestingly, the macronutrient balance of the daily diet when eating figs or composing the diet from diverse foods did not differ. This suggests that figs are a nutritionally balanced food for the monkeys, and when they are not available the monkeys compose a balanced diet through mixing nutritionally complementary foods.

Another pattern of feeding was observed during the late wet season, when ripe figs were scarce but other fruits, particularly high in sugars and fats, were abundant. In this period, the monkeys ate very similar amounts of protein as when eating figs or diverse complementary foods (i.e. this most likely being the target level of protein), but in so doing ate considerably more carbohydrates and fat than observed when eating figs or mixing complementary foods. The resulting intake array was thus a vertical line, with little variation on the protein axis compared with the axis for non-protein. This is very similar to the rule of compromise that has been observed in experimental studies of humans (Simpson and Raubenheimer 2005; Gosby et al. 2011), which appear to be an important contributory factor to the epidemic of obesity that has swept many countries in recent decades.

The study of Felton et al. (2009a , 2009b ) has several implications for understanding the habitat requirements of Peruvian spider monkeys. First, the results provide an estimate of the position of the macronutrient intake target of monkeys, and suggest that figs (Ficus spp.) are nutritionally balanced with respect to the macronutrient requirements of these monkeys. This underscores the importance of regulating the commercial harvesting of Ficus timber in the habitat of spider monkeys. Second, the study suggests that the monkeys are not reliant on balanced foods, but can compose a balanced diet through mixing nutritionally complementary foods. This shows that the monkeys are to some extent flexible in their food needs, and also provides an indication of which combinations of foods are important components of their nutritional environment. Third, when exposed to foods with a low protein : non-protein energy balance spider monkeys will defend their protein intake but overeat non-protein energy. This helps us to predict how these monkeys would respond to distortions in the macronutrient composition of their foraging environment. Specifically, it suggests that prolonged exposure to foods with a low protein : non-protein energy balance is unlikely to result in protein deficit, but rather a surplus of non-protein energy and, conversely, these monkeys are unlikely to overeat protein to meet their requirements for non-protein energy.

It remains to be determined, however, whether and how being constrained to suffer macronutrient imbalance would impact on the monkeys. It is easy to imagine a situation in which energy shortage, as might be expected to result from high-protein diets, can be a limiting factor to wild primates. But might a sustained energy surplus, as predicted on low-protein diets, also impact adversely on spider monkeys? Indirect evidence raises a cautionary note. Like their human cousins, non-human primates are prone to obesity under certain conditions, which include exposure to readily-available foods high in non-protein energy (Hansen 2001). While this is most common in captive primates (e.g. in zoos), significant levels of obesity have also been observed in free-ranging populations of baboons (Altmann et al. 1993) and macaque monkeys on the island of Cayo Santiago off the coast of Puerto Rico (Schwartz et al. 1993). In both cases this was associated not with natural foods, but with access to garbage dumps (baboons) and the management practice of supplementing the natural diet with primate chow (macaques). At the very least, this warrants careful consideration of the food species that should be conserved in the habitat of spider monkeys, and provides information regarding the required macronutrient composition of any supplementary or captive diets for this species.

We began with the assertions that nutrients are the connecting threads in the web of life, and that new approaches are needed for understanding the nutritional networks that comprise and link organisms. We then set out to demonstrate how geometric analysis (the Geometric Framework) can contribute by selectively reducing the pall of nutritional complexity to those interactions that help to illuminate the problem at hand. In so doing, the Geometric Framework can help to integrate nutrition research within a framework that encompasses the different levels of biological organisation from genes to ecosystems, spans taxa from slime moulds to humans, and offers the promise of practical solutions to pressing problems in agriculture, human health and conservation ecology. The examples we have presented represent a few superficial scratches to the surface of a very large world of unresolved issues in nutritional biology.