Nutrition, ecology and nutritional ecology: toward an integrated framework

Authors


*Corresponding author: E-mail: d.raubenheimer@massey.ac.nz

Summary

  • 1The science of nutritional ecology spans a wide range of fields, including ecology, nutrition, behaviour, morphology, physiology, life history and evolutionary biology. But does nutritional ecology have a unique theoretical framework and research program and thus qualify as a field of research in its own right?
  • 2We suggest that the distinctive feature of nutritional ecology is its integrative nature, and that the field would benefit from more attention to formalizing a theoretical and quantitative framework for developing this.
  • 3Such a framework, we propose, should satisfy three minimal requirements: it should be nutritionally explicit, organismally explicit, and ecologically explicit.
  • 4We evaluate against these criteria four existing frameworks (Optimal Foraging Theory, Classical Insect Nutritional Ecology, the Geometric Framework for nutrition, and Ecological Stoichiometry), and conclude that each needs development with respect to at least one criterion.
  • 5We end with an initial attempt at assessing the expansion of our own contribution, the Geometric Framework, to better satisfy the criterion of ecological explicitness.

Introduction

The range of studies that go by the label ‘nutritional ecology’ encompasses an impressive diversity of taxa, methods, concepts, interests and goals, spanning, inter alia, behaviour, morphology, developmental biology, physiology, life history, ecology and evolution, with emphasis both on function and on mechanism. Such cross-disciplinary breadth provides broad conceptual and methodological foundations, and imbues the discipline with wide-ranging relevance. But it also presents challenges. Foremost among these is that to progress beyond the status of label and qualify as a field of research in its own right (Shettleworth 2000), nutritional ecology needs an identity more distinct than a diffuse confluence of methods and interests united within the general areas of nutrition and ecology.

What would be the cornerstone of that identity? In our judgement, the single most distinctive characteristic of nutritional ecology is its propensity to probe the gaps between disparate fields, yielding integrative insights that would otherwise not be obtained. The hiatus that is most closely associated with the subject is that between field ecology (e.g. resource quality and distribution) and animal phenotypes (e.g. foraging behaviour, functional morphology, digestive physiology). Progress in bridging this gap has, however, been piecemeal and incomplete, as is evidenced by growing concern in the literature for greater integration between the study of phenotypes and ecology (e.g. Jones & Lawton 1995; Fryxell & Lundberg 1997; Olff et al. 1999; McGill et al. 2006; Schmitz 2008). We believe that nutritional ecology would be better equipped for achieving this integration if more attention was paid to developing frameworks that systematically define the panoply of salient components in organism–environment interactions and explicitly model their integration. In other words, frameworks are needed that provide a scaffold for melding nutrition and ecology into an integrated nutritional ecology.

The primary aim of this article is to state what we consider to be the necessary basic properties of such a scheme, and evaluate in relation to these some frameworks that are currently in use: Optimal Foraging Theory, Classical Insect Nutritional Ecology, the Geometric Framework for nutrition, and Ecological Stoichiometry. Our survey reveals that all four approaches have provided local foci of conceptual and/or methodological cohesion within nutritional ecology, but a truly integrative framework would involve an expansion or synthesis of existing frameworks. A second aim of this article is to address the expansion of our own contribution, the Geometric Framework, to questions of community ecology.

Nutritional ecology: components and interactions

The core components of a general conceptual framework for nutritional ecology are set out in Fig. 1. Most generally, these are the organism, the ecological environment, and the nutritional basis of the interaction between organism and environment – and here we use ‘nutritional’ in the broad sense of any property of a food that affects the animal (Westoby 1974).

Figure 1.

Conceptual scheme depicting the components of an integrative framework for nutritional ecology. The organism is considered from the viewpoints of function, mechanism, development and history (Tinbergen 1963), while the environment is partitioned into biotic and abiotic components. The nutritional interactions that take place between organism and environment involve both the effects of the environment on phenotypes (downward arrows) and the impact of phenotypes on the ecological environment (upward arrows). A nutritional ecology framework should also cope with horizontal interactions (dashed arrows), for example between biotic and abiotic components of the environment, or between mechanistic and functional aspects of phenotypes.

We believe that the representation of these three components and interactions should be explicit, in the sense that the framework can enable research to be structured so as directly to address questions pertaining to each. In general, this means that the components should be represented in models as parameters or, preferably, variables, but not as constants. Thus, models that treat foods as unitary resources, that is, do not discriminate among their constituents (Raubenheimer & Simpson 1995), or assume a priori that a single component (e.g. energy) is pre-eminent, are not nutritionally explicit, as they cannot partition the actual roles of specific food components in nutritional ecology interactions (see also Boggs 2009). Furthermore, links among the components (depicted by arrows in Fig. 1) should be bidirectional, thus enabling the research to be structured in a way that addresses causal effects in either direction or in both simultaneously (i.e. reciprocal causality – e.g. Cardinale et al. 2006).

We further note that the arrows connecting elements in Fig. 1 are but a sub-set of a more complex network of biologically relevant interactions that are potentially of interest to nutritional ecology studies. There has, for example, been a recent intensification of interest in the question of how community processes and patterns influence evolution (Johnson & Stinchcombe 2007). If the context of such a study was nutritionally explicit, then it would warrant an arrow linking organism ‘function’ directly with ‘community’, or the interaction might possibly involve ‘history’ (e.g. if phylogeography were an important component). A useful term for such a network in which elements can be viewed as interacting with other elements that occur in two or more components (e.g. ‘function’ vs. ‘community’ and/or ‘history’) is ‘heterarchical’ (Gunji & Kamiura 2004).

In the remainder of this section we briefly expand on the role of the organism, the environment and nutrition in the scheme.

the organism

The organism is central in nutritional ecology. As is true in many other areas in organismal biology, nutritional ecology can trace important influences to the classical ethological movement of 1930–1960's. Ethology, too, is a fundamentally integrative science, in two respects that are relevant to our discussion here. First is the emphasis in ethology on understanding animal phenotypes in relation to their ecological environment, which has likewise historically been associated with the emergence of the term ‘nutritional ecology’ (e.g. Schneider 1967; Stanley Price 1978) and has continued to be central to the identity of the field. Second, ethology's ‘manifesto’, as famously articulated by Niko Tinbergen (1963), is based on an integrative approach which urges animal behaviourists to combine in their thinking about behaviour four levels of analysis: its mechanisms, development, function and evolutionary history.

Tinbergen's framework remains hugely influential and in our opinion could make a valuable contribution to integration in nutritional ecology. Specifically, it provides a more refined depiction of the organism in nutritional ecology research, through explicitly distinguishing the links between nutritional environments on the one hand, and on the other mechanistic, developmental, functional and phylogenetic aspects of phenotypes (Fig. 1). The Tinbergen scheme was developed and is most frequently applied in the context of behaviour, but in nutritional ecology it would apply to all aspects of the phenotype, including physiology, morphology and life history.

the environment and organism-environment interactions

The component of the environment that is usually at the centre of nutritional ecology studies is food, but other biotic (e.g. predators and parasites) and abiotic (e.g. temperature, photoperiod) factors are, of course, also relevant (Slansky & Rodriguez 1987) and might even be pre-eminent (Schmitz 2008). Being focused primarily on the organism, nutritional ecology studies most commonly emphasise the downward arrows in Fig. 1, the ways that organisms respond to the ecological environment at various time-scales: behavioural and physiological responses, phenotypic plasticity (e.g. in oral and gut morphology), development and life history (e.g. age at maturity), and adaptation on an evolutionary time-scale. However, as noted above, some authors have also used ‘nutritional ecology’ for studies that proceed in the opposite direction, addressing questions of how phenotypes impact on population – (e.g. Simpson et al. 2006) and community-level processes, or the reciprocal impacts of communities and phenotypes (e.g. Schmitz 2008). We believe that nutritional ecology is well-placed to make a substantial contribution to the question of how phenotypes impact on ecological communities, particularly through dialogue with other foci of integration within population, community and ecosystems ecology (S.J. Simpson et al., under review).

nutrition

A framework that is nutritionally explicit enables the questions to be addressed: ‘which nutrient(s) and other food components are important to an animal in a given situation?’, ‘how does each of these influence the animal's (e.g. homeostatic) responses?’, and ‘what are the performance and ecological consequences for the animal of responding in the way that it does?’

There exists surprisingly little information on how specific nutrients influence the homeostatic and performance responses of animals, and even less on how these influences in turn impact on populations and communities. The reason for this is that studies are most frequently conducted within frameworks that do not systematically disentangle the roles of specific food components, or else a priori identify one component (usually nitrogen, energy or plant secondary compounds) as paramount and code this as an input to the study rather than an experimental outcome. Even where the focal component is correctly identified, an important part of the story might be overlooked in this approach. This is because foods are complex mixtures, and the impact of specific components is usually contingent on and/or exerted through other components. For example, in many animals the ingestive regulatory systems weight protein more strongly than other nutrients, with the consequence that they over-ingest other nutrients when eating low-protein foods – the ‘protein leverage’ effect (Simpson & Raubenheimer 2005). In such cases, protein would correctly be identified as the pre-eminent nutrient, and yet the major constraint on protein gain might be the inability of the animal to ingest large excesses of some other food component(s), and the major health impact due to the excess of these components that they do ingested (Raubenheimer, Lee & Simpson 2005; Boggs 2009).

We consider it a high priority in nutritional ecology to adopt nutritionally explicit frameworks which systematically identify the individual and interactive roles of different food components.

Frameworks in nutritional ecology

In this section we evaluate against the criteria set above some of the frameworks currently in use in nutritional ecology. We cannot hope to do justice within the space constraints to the diversity of modelling approaches that have been applied to specific questions in nutritional ecology, and our coverage is therefore restricted to four frameworks that we consider to be particularly relevant to the question of integration: Optimal Foraging Theory, Classical Insect Nutritional Ecology, the Geometric Framework for nutrition, and ecological stoichiometry. We believe, however, that the main points our survey illustrates are robust to the inclusion of any framework in use in nutritional ecology.

optimal foraging theory

Optimal Foraging Theory (OFT) is an evolutionarily-inspired framework that aims to ‘explain and predict’ (Pyke et al. 1977) the patterns of food choice and foraging by animals. It is based on the premise that foraging can be viewed as a process that has been optimized by natural selection to maximize fitness, and thus optimization mathematics is an appropriate tool for developing foraging models (Maynard Smith 1978). Typically, the focal variable is not fitness itself, but a ‘currency’ assumed to be a proxy for fitness, such as rate of energy gain (maximized) or predation risk (minimized). Although the optimality approach is used most frequently to model behavioural aspects of foraging, it has also been applied to physiological aspects such as food processing times and digestion efficiencies (e.g. Raubenheimer & Simpson 1998).

OFT is concerned primarily with the effects of the environment on the phenotypes of animals (i.e. the downward arrows in Fig. 1), but it has also been applied in the reverse direction, exploring how the functional characteristics of organisms influence ecological communities (e.g. Belovsky 1986; Petchey et al. 2008). OFT is, therefore, clearly a framework for studying the nutritional relations between animals and their environments, and for this reason is relevant to our consideration of nutritional ecology. The key question, however, is the extent to which in its current form OFT is sufficiently nutritionally explicit to carry out the nutritional ecology agenda.

Where the currency in OFT models is nutritional (as opposed to, e.g. time minimization or survival maximization), it usually involves energy, although other nutritional currencies are occasionally involved (e.g. protein –Berteaux et al. 1998). In this respect, OFT is a uni-dimensional approach which assumes a priori that a single food component is limiting to the animal, and elevates that component to the status of currency. Other food components, such as toxins and nutrients that are not represented as currency, are coded as constraints within which the animal has to work in its attempts to achieve the postulated foraging goal (e.g. Westoby 1974; Pulliam 1975; Belovsky 1990; Hirakawa 1995). This is often, but not always (e.g. Pulliam 1975), done using linear programming (Westoby 1974; Belovsky 1990).

OFT has clearly experienced many successes (Stephens et al. 2006), but an improved understanding of nutritional processes is not among them. We believe the reason for this to be that OFT does not comfortably fulfil the criterion of a nutritionally explicit framework. First, while it might arguably be true that at any one time an animal is limited by a single nutrient (the currency), it is an open and important question as to the dynamics and time-scale of such limitation. At the one extreme, single-nutrient limitation might be a perpetual feature of an animal's nutritional ecology, as is proposed by White (1983) to be generally the case for nitrogen in many ecosystems. At the other extreme, for an animal that switches between food types frequently, the limiting component(s) might change daily, hourly, or even within a single meal (Chambers et al. 1995). Second, energy is itself not a nutrient but a property of the macronutrient groups protein, lipid and carbohydrate. Without explicitly distinguishing among these energetic components, caloric measures present the risk that foraging aimed at maximizing one or more of these macronutrients, or optimizing their balance, is confounded with energy maximization. Finally, it is often difficult, impossible, or meaningless to distinguish between ‘constraint’ and ‘adaptive strategy’. We therefore consider it a better heuristic to view nutritional processes as a ‘network of interconnected trade-offs with a global optimum’ (Illius, Tolkamp & Yearsley 2002).

Nonetheless, in addition to its successes in furthering the understanding of animal decision making, the optimality-based approach to foraging has made a substantial contribution to the development of nutritional ecology. It set the bar for conceptual and quantitative rigour in the study of foraging, and provided a foundation which is increasingly becoming integrated with other approaches in the study of nutritional ecology (Simpson et al. 2004; Newman 2006). Additionally, in its earlier formulations, OFT provided a point of contrast against which other approaches could develop. In the present context, the most relevant of these is Classical Insect Nutritional Ecology.

classical insect nutritional ecology

The development of what we refer to as ‘Classical Insect Nutritional Ecology’ (CINE) was seeded by the convergence in the 1950's and 1960's of several strands of research which shared a common interest in the factors that govern food selection by animals. Notable among these was the work of Reginald Painter (e.g. 1936), who developed the view that variation in the nutrient composition of plants is central to the patterns of food choice and performance responses by phytophagous insects. A second line of interest, more closely associated with the field of plant–animal co-evolution, asserted that food selection in phytophagous insects is driven not by nutrients, but by plant secondary compounds (e.g. Fraenkel 1959). These discussions took place in a climate of growing interest among ecologists in the extent to which the nutritional quality of plant tissues limits herbivore populations (Schmitz 2008).

Against this background, there was clearly a need in the study of animal foraging for a paradigm that approached more directly than did OFT the question of which currencies actually drive the foraging decisions and population responses of animals (Mitchell 1981; Waldbauer & Friedman 1991) – i.e. for an approach that was nutritionally explicit. The requisite paradigm was adopted from the experimental psychology literature, where it had been shown in the work of Curt Richter and others that rats can self-select from a range of nutritionally incomplete foods a diet that sustains good performance, and can alter their patterns of food selection to compensate for surgically-induced nutritional perturbations (Galef 1991). The dietary self-selection paradigm was introduced to CINE by Gil Waldbauer and colleagues (e.g. Waldbauer et al. 1984). It has since been demonstrated using this approach that dietary self-selection is ubiquitous among animals. Some combination of the macronutrient groups protein, carbohydrate and fats are regulated independently by many (if not most) animals, and so too are particular vitamins (Markison 2001), amino acids (Markison et al. 2000; Yamamoto et al. 2000), mineral salts (Denton et al. 1993) and the macromineral calcium (Tordoff 2001) regulated by some. These data underscore the importance for nutritional ecology of adopting a framework that is nutritionally explicit.

Gil Waldbauer also made another highly influential contribution to CINE, in introducing a quantitative framework for representing the nutritional responses of animals to their foods (Waldbauer 1968). Waldbauer's ‘quantitative nutrition’ is a budgetary approach, which expresses the relationships between food intake and utilization as rates and efficiencies that can be used comparatively – for example, to compare growth in insects that have different consumption rates. The proposed ratio-based nutritional indices – relative consumption rate (RCR), approximate digestibility (AD), efficiency of conversion of ingested food (ECI), and efficiency of conversion of digested food (ECD) – rapidly became the industry standard within CINE (e.g. Scriber & Slansky 1981).

By the late 1980s the field had matured to the point where Slansky & Rodriguez (1987) could propose a general conceptual framework for research in CINE. Their recommended framework would involve: (i) determining the performance of an animal in circumstances (relating to nutrition, as well as its interactions with other factors such as temperature and predation) which maximize fitness; (ii) determining how realistic changes in these circumstances influence the animal's performance, its compensatory responses for ameliorating the impacts on performance, and the trade-offs that it encounters in responding to the altered circumstances; (iii) performing comparative studies, in which the patterns in (i) and (ii) are related more generally to phylogeny, development and ecology. Slansky and Rodriguez recommended Waldbauer's (1968) ratio-based indices as a quantitative approach for carrying out this agenda.

In our view, the major general contribution of CINE was to recognize explicitly the fact that the nutritional ecology of animals is complex, involving interactions among numerous environmental factors (e.g. nutrient and non-nutrient food components, temperature etc.) and animal responses (e.g. foraging, feeding, food utilization, growth). The Slansky–Rodriguez manifesto constituted an elegant approach to conceptualizing the issue, but CINE did not produce a framework that was up to the task of quantifying and interpreting these multifarious interactions. The paradigm of dietary self-selection provided a means to demonstrate cases where animals feed non-randomly on foods differing in composition, and to identify the nutrients that are involved in the patterns of food selection. It could not, however, deal with the critical interactive effects of these nutrients on the patterns of food selection and post-ingestive and performance responses. Similarly, in introducing terms representing key homeostatic processes (intake, nutrient assimilation, growth and excretion), Waldbauer's quantitative budgetary approach emphasized the active role of the organism in nutritional ecology, but fell short as regards integration. Ostensibly, the nutritional indices that he proposed did represent an integration of different homeostatic responses, because each index includes two or more of the critical regulatory variables. However, compounding several variables into a single index usually does not reveal the relationships among them, but obscures these relationships (Raubenheimer & Simpson 1992). To be sure, Waldbauer's aim in recommending these indices was not integration, but standardization: they enabled responses (e.g. growth) to be compared across animals that differed in other relevant aspects (e.g. consumption). Unbeknownst to Waldbauer, however, a literature was subsequently to emerge demonstrating that there are statistical problems with the use of ratios for standardizing variables in this way (see Raubenheimer & Simpson 1992 and citations therein). Some explicit attempts at integration have been made by plotting ratio indices against each other (e.g. Scriber & Slansky 1981; Beaupre et al. 1993), but this too can lead to serious statistical and interpretative problems (Raubenheimer 1995; Brett 2004).

the geometric framework

To address the challenge of integration, we have developed a graphical approach, the Geometric Framework (GF), which models the key relationships among relevant variables in nutritional ecology (Raubenheimer & Simpson 1993, 1994, 1997; Simpson & Raubenheimer 1993, 1995, 1999). GF is based on the logic of state–space geometry, where relevant variables are expressed and related to each other within a geometric space defined by two or more relevant food components. The variables represented within this space might include one or more foods, the organism's current and optimal nutritional states, the impact on its nutritional state of eating each food, its body composition, the efficiency of nutrient utilization, the rates of excretion, and whatever performance consequences might be of interest. A model so constructed can be used to conceptualize problems that involve two or more food components, and to design and interpret experiments for resolving these problems. GF has been applied to a range of biological questions involving diverse taxa (see Table S1, in electronic Supporting Information).

As shown in Fig. 2a–c, the main components of the Waldbauer nutritional indices (e.g. intake, growth, nutrient utilization) are represented within GF models, as is dietary self-selection (Fig. 2d). The handling of these issues is, however, very different under GF. First, graphical models enable the interactions among the model components to be visualized, rather than subsumed within nutritional ratios. Second, representation of two or more food components within a model enables their interactive effects to be quantified. A third point of difference, and one on which we would like to briefly elaborate, is that in common with OFT – and in the spirit of the Slansky–Rodriguez manifesto – GF models explicitly incorporate the notion of functional optima.

Figure 2.

Hypothetical two-dimensional geometric models showing four budgetary scenarios. (a) Balanced diet. The intake target (IT) is the amount and balance of the two nutrients the animal needs to eat within the stipulated period to achieve maximal fitness, and the line originating at the origin is a nutritional rail representing the carbohydrate : protein balance of food Fa. Since the nutritional rail intersects IT, the animal is able to match its intake (Io) to its optimal requirements by eating this food – that is, it is a nutritionally balanced diet. The growth target (GT) shows the optimal amount of ingested protein (R(u)p) and carbohydrate (R(u)c) that should be utilized for ‘growth’ (i.e. retained in the body), while the nutrient target (NT) describes the amount of nutrient that should be ingested to optimally satisfy nutrient requirements for all fitness-enhancing functions, including components that are retained in the body (GT) and utilized for purposes that involve their dissociation (loss) from the body (e.g. respiration, useful secretions etc. – collectively represented by D(u)p and D(u)c). For an animal that is 100% efficient at converting ingested nutrient to functional gain, NT = IT. However, to the extent that there is a degree of constrained inefficiency in nutrient utilization, optimal intake needs to be over-specified by D(w)p and D(w)c. In the case modelled, NT is shaped as an asymmetrical ellipse oriented along a gradient of approximately –1, this shape reflecting the underlying cost structure for the investment of ingested nutrients (Simpson et al. 2004). Such an ellipse might, for example, reflect the fact that protein and carbohydrate are to some extent interchangeable (e.g. as sources of energy), and therefore optimal utilization requirements can be met using any combination of the two nutrients whose coordinate falls on this ellipse (further illustrated in graph b). By definition, if optimal intake is achieved (i.e. Io = IT), then observed overall utilization (Nuo) will fall on NT and observed growth (Go) will equal GT. (b). Constrained intake, with nutrient inter-conversion: Model where the animal has available only nutritionally imbalanced food Fb, which contains surplus protein relative to carbohydrate, and therefore cannot reach IT but must choose between intake scenarios [I1] (satisfies requirement for protein, but suffers a deficit of carbohydrate), [I2] (gains required level of carbohydrate, but surplus protein) and [I3] (moderate protein surplus and carbohydrate deficit). It can, however, ameliorate the impact of the ingestive constraint by judiciously allocating the ingested surplus and/or deficit among budgetary components. For reference, budgetary allocations where Io = IT (i.e. from model a) are shown by the length of the grey lines, while constrained allocations are shown by the length of the black arrows. In the case modelled, the animal has regulated intake to [I3] and has thus ingested both a surplus of protein and a deficit of carbohydrate. Assuming that D(w)c has a fixed lower limit (i.e. where Io = IT utilization efficiency of carbohydrate is at a maximum), the intake deficit of carbohydrate must be absorbed by R(u)c and/or D(u)c. In this case growth is defended (R(u)c and R(u)p are unchanged), but carbohydrate allocated to fuel energy metabolism (D(u)c) is reduced. However, a portion of the surplus ingested protein (extended portion of the arrow representing D(u)p) is deaminated and channelled into energy metabolism, thus compensating for the reduction in D(u)c. As a result, overall macronutrient utilization is defended (Nuo coincides with NT), and any fitness costs due to Io not coinciding with IT must be attributed to other factors such as the need to excrete surplus protein (increased D(w)p). (c) Constrained intake without nutrient interconversion: An alternative response to constrained intake I3. IT, NT and GT are in the same positions as in panels a. and b. However, in this case the animal is taken to be incapable of deaminating amino acids for use in energy metabolism, and as a result IT, NT and GT are more localized than in the previous examples. The ingested deficit of carbohydrate, combined with inability to reduce D(w)c, results in failure to meet the nutrient target – Nuo is displaced from NT in the carbohydrate dimension. The animal prioritizes the allocation of ingested carbohydrate to energy metabolism (maintains D(u)c), and as a consequence suffers reduced carbohydrate-derived growth (R(u)c). Furthermore, to maintain proportional body composition the level of protein allocated to growth (R(u)p) is reduced, resulting in Nuo being displaced from NT also in the protein dimension. By definition the displacement of Nuo from NT incurs fitness costs, and additionally the animal has an increased burden of surplus ingested protein to excrete (increased D(w)p). (d) Nutritionally complementary foods: Here the animal has available two nutritionally imbalanced foods, Fb and Fc. However, since the protein-carbohydrate nutritional rails for these foods fall on opposite sides of IT, the animal can nonetheless reach IT (and hence NT and GT, not shown) by mixing its intake from the two foods (i.e. these are nutritionally complementary with respect to protein and carbohydrate). One possible intake trajectory is shown by the arrows, in which the animal takes meal m1 from Fb, then for meal m2 switches foods and so takes the trajectory defined by Fc, before returning to Fb for m3 and so on. Other patterns might include frequent switches within meals, or several consecutive meals on one food followed by several on the other.

This is done by distinguishing estimates of optimal values for nutrient intake and utilization (e.g. the Intake Target, Nutrient Target and Growth Target) from realized values. The inclusion of functional targets in a model enables nutrient budgets to be constructed that are based on functional classification of components, rather than a methodological classification as is standard in CINE (Raubenheimer & Simpson 1995). In a methodological classification:

I=R+D( eqn 1)

where I is ingested nutrient, R that which is retained by the organism (i.e. reflected in body composition) and D is dissociated (i.e. not retained). In a functional classification both terms on the right hand side of eqn 1 are partitioned into the components that contribute to fitness and those that do not:

I=R(u) +R(w) +D(u) +D(w)( eqn 2)

where R(u) and R(w) are, respectively, components that are retained beneficially (utilized for fitness gains) and non-beneficially (e.g. surplus lipid storage in obesity), and similarly D(u) and D(w) represent dissociated nutrient that is utilized (e.g. energy metabolism, defensive secretions) and wasted (excreted in the faeces, urine or via diet-induced thermogenesis –Zanotto et al. 1997).

One advantage of distinguishing functional components in this way is that it greatly increases the predictive power of models, because homeostatic regulatory systems will tend towards behavioural and physiological responses that produce functionally favourable outcomes (e.g. Simpson et al. 2004). As illustrated in Fig. 2, it also greatly increases the analytical power of a model.

In terms of our stated criteria for models of nutritional ecology, GF clearly is nutritionally explicit, being designed to disentangle the individual and interactive effects on animals of various food components. It is also organismally explicit, being capable of addressing questions concerning the relationships of nutrition across Tinbergen's four categories – function, mechanism, ontogeny and phylogeny (Simpson & Raubenheimer 1993). GF is, at least partly, ecologically explicit, as it is designed with the fundamental goal of examining the ways that the nutritional environments of animals impact on phenotypes (downwards arrows in Fig. 1). Less well-developed, however, is its application to questions of how the phenotypes of animals impact on their ecological environments (upwards arrows in Fig. 1). We return below to the prospects for GF of modelling such questions, and thus qualifying as ecologically explicit sensu stricto.

ecological stoichiometry

Ecological Stoichiometry (ES) is ‘the study of the balance of energy and multiple chemical elements in ecological interactions’ (Elser 2006). As suggested by this definition, there are some interesting parallels between ES and GF (Raubenheimer & Simpson 2004). Like GF, ES grew out of the realization that there are complexities to biological systems that cannot be captured using models based on energy alone (Reiners 1986), and therefore frameworks are needed that model the interactions among multiple currencies – that is, both GF and ES are nutritionally explicit, multi-currency frameworks. Also like GF, ES is fundamentally integrative, overtly aiming at interrelating causes and effects across multiple biological levels from ‘molecules to ecosystems’ (Sterner & Elser 2002). A third similarity between the two approaches relates to the core tenet of ES, the mass balance equation, which applies the laws of conservation of matter to trophic exchanges in ecosystems. Although couched in the terminology of chemistry (‘stoichiometry’), mass balance equations are essentially equivalent to the nutrient budgets developed in CINE and modelled in GF (Raubenheimer & Simpson 2004). Such parallels have lead some to consider ES ‘the most recent outgrowth’ of nutritional ecology (Schmitz 2008), while McGill et al. (2006) consider both approaches to be examples of the kind of ‘studies in functional ecology that community ecologists would benefit from incorporating into their thinking’.

There are also fundamental differences between ES and GF. An important point of distinction relates to the terra firma of the two approaches. As detailed above, GF was developed as a multi-currency nutritionally explicit approach to modelling nutritional phenotypes, and the question of how well-suited it is for extension to modelling ecosystem processes remains open (more on which below). By contrast, the fundamental inspiration in ES relates to the flow of matter and energy through ecosystems (Reiners 1986). Organisms are, of course, a component of ES models – indeed, are central to these models, as they constitute the primary conduits for the flow of energy and matter through ecosystems. But the generality needed for modelling the effects of interactions among organisms on ecosystem processes has been bought by ES at the cost of simplifying aspects of phenotypes which are central to more-organism-focused approaches. An important question that arises in the present context is how these simplifications impact on the extent to which ES models can be considered sufficiently organismally explicit to qualify as a general framework for nutritional ecology.

Organisms are represented in ES models primarily through their body composition, usually expressed as the ratio of key elements – nitrogen (N), phosphorus (P) and/or carbon (C). Central to predictions of ES are comparisons of the elemental composition of consumers and their resources. In accordance with the law of mass balance, a consumer can maintain its elemental composition only by feeding on foods with similar elemental composition or by specifically increasing the rates at which surplus elements are excreted – i.e. by decreasing their ‘gross growth efficiency’ (GGE) for the surplus elements. In the simplest scenario, the optimal food – considered to be that which supports maximal production while minimizing wastage (Anderson et al. 2005) – would have identical elemental composition to the body of the consumer, but biological constraints on the efficiency with which elements can be converted to body tissue (i.e. GGE < 1) preclude this. Therefore, foods should be weighted for the maximal GGE of each element, such that the optimal food is defined as follows (Anderson et al. 2005):

ideal food x : y= consumer x : y (max GGE.x/max GGE.y)(  eqn 3)

where x and y are two elements, and max GGE.x and max GGE.y are the maximum efficiency with which the consumer can convert x and y, respectively, to body tissue. If the above equality does not apply, growth and reproduction of the organism are limited by the element in short supply, and/or by the cost incurred in excreting the excessive element (Anderson et al. 2005; Boersma & Elser 2006).

An important simplification in this approach is the adoption of elements as the chosen currency. ES studies have occasionally focused on biochemicals (Anderson 1994; Anderson & Pond 2000; Anderson et al. 2004), but the vast majority involve elements – indeed, ES has been defined as the ‘biology of elements’ (Sterner & Elser 2002). Elements have the advantages for ecological studies that they are easy to measure, constitute a common denominator relevant to all organisms in ecological communities, and provide a link to inorganic fluxes within ecosystems. They have the disadvantage, however, that in terms both of function and mechanism, heterotrophs relate to their nutritional environments not via the C, N and P, but via heterogeneous molecular complexes of which these elements are components. Elemental analysis will thus predict the food choices, post-ingestive responses and functional consequences for a foraging animal only to the extent that they approximate the nutritional value of the foods. In some cases such correlations likely do apply – for example, the nitrogen content of foods has been successfully used in many studies as a proxy for its protein and amino acid content – but even here there might be complexities due to the presence of other nitrogenous compounds and the fact that proteins vary in their nitrogen content (e.g. Lourenco et al. 2002). In other contexts, however, the approximation breaks down, because two or more functionally distinct molecular complexes can yield similar elemental composition. Where this is the case, element-based analyses can fall short in predicting the responses of organisms.

For example, the fractional contribution of different carbohydrates to foods has profoundly different nutritional implications for herbivores, but is indistinguishable within standard ES models (e.g. see Anderson et al. 2004). An illustration is provided in Fig. 3, which shows data from an experiment using synthetic foods to investigate nutritional regulation in locusts (Locusta migratoria). Nutritional analysis reveals tight, target-like, homeostatic regulation of the balance and amounts of nutrients eaten, a phenomenon that profoundly influences foraging choices (see for example Raubenheimer & Jones 2006). Elemental analysis of the same data suggested regulation of nitrogen intake, as might be expected because protein was the only source of nitrogen in the diet and nitrogen intake thus provided a perfect proxy for protein intake. There is, by contrast, no apparent regulation of carbon, and elemental analysis would thus fail to detect a powerful predictor of food choice and feeding behaviour. The reason for the different results is that nutritional analysis reflects the animals’ ingestive responses in distinguishing between non-nutritional (indigestible) cellulose and nutritional sources of carbon (in this case principally sucrose, dextrin, and amino acids), whereas elemental analysis confounds carbon derived from these different sources. Different carbon sources likewise have different post-ingestive consequences. Thus virtually no carbon from ingested cellulose is retained by locusts (i.e. GGE = 0), but surplus carbon derived from digestible carbohydrates is associated with increased body fat (GGE > 0) (Raubenheimer & Simpson 1997). The broader implication is that in order to derive the GGE for carbon, and hence to evaluate the ‘suitability’ of the food in relation to the consumer's body composition (see above), separate GGE's for different carbon sources would be needed (Anderson et al. 2004). These could only be derived using biochemical analyses.

Figure 3.

Comparison of ingestive regulation of macronutrients and elements. The data, taken from Chambers et al. (1995), represent selected intake points by fifth stadium locusts, Locusta migratoria . Each point represents the mean selected intake over 6 days of locusts fed one of four food pairings. The food pairings were (%protein : %carbohydrate): 14 : 28 + 14 : 7; 14 : 28 + 28 : 14; 7 : 14 + 14 : 7; or 7 : 14 + 28 : 14. Since the animals given each pairing had to distribute their feeding between the foods in very different ways to reach the same point of intake, clustered intake points represent homeostatic regulation. Such regulation is revealed when the data are plotted in terms of macronutrient intake (a), but not in terms of elements (b).

A second important simplification usually adopted in ES models is the emphasis on proportional body composition as a metric against which to evaluate diet quality. For ecological studies this is convenient, as body composition can readily be measured and compared across diverse organisms. But its utility in the context of organismal biology is limited, because this approach is based on a methodological classification of budgetary terms (eqn 1) and neglects fitness-enhancing components of the ingesta that are dissociated (eqn 2 and Fig. 2). For example, ingested carbon that is used to fuel life-supporting energy metabolism is coded only implicitly, as a constraint on max GGEcarbon, and not represented as a fitness-relevant component of the ingesta in its own right (Fig. 2). Neither is it distinguished from other categories of carbon that are dissociated at max GGE, which might not contribute to fitness and should thus correctly be classified as wasted rather than utilized carbon (D(w) rather than D(u) in eqn 2) (but see Anderson et al. 2004; Hessen & Anderson 2008). Emphasis on proportional body composition also neglects a fundamental component of fitness (e.g. Kingsolver & Huey 2008), body size. Thus, standard ES models would not distinguish an animal that achieved optimal growth (e.g. Go in Fig. 2b) from an animal that had the same proportional body composition but was overall smaller (Go in Fig. 2c). Finally, emphasis on proportional body composition could lead to the mistaken conclusion that an animal with depleted fat stores (and hence lower body C : N) requires a lower proportion of fat in its diet than does a member of the same species that is in better condition.

For many ecological applications the element and the body composition simplifications might be amply justified, because they enable ecological analyses to be performed where measurements of macronutrients and functional components other than body composition might not be feasible. Their success, however, depends on the extent to which these proxy measures represent the causal variables (biochemistry and overall fitness, including both retained and dissociated components) and in many contexts this will not be the case. We suspect, in particular, that analyses involving carbon will be less reliable than those involving nitrogen and phosphorus. This is partly because, as noted above, carbon is a major dietary component that is spread across several functionally distinct biochemicals (e.g. cellulose, starch, sugars, lipids, amino acids). Furthermore, fuel for energy metabolism comprises a substantial component of ingested carbon which contributes critically to fitness but is not retained in the body (i.e. falls into D(u)), and in standard stoichiometric equations will not be distinguished from wasted carbon. Finally, the functional implications of excess dietary C are substantially more complex than is implied by simple comparisons of the composition of consumers and their resources (Hessen & Anderson 2008).

A promising development of the stoichiometric approach is dynamic energy budget (DEB) theory (Kooijman 2000). DEB models use differential equations to describe the rates at which individual organisms assimilate and utilize energy from food for maintenance, growth, reproduction and development, while taking into account constraints on the fluxes of elements. A fundamental construct within DEB theory is the ‘synthesizing unit’ (SU), which is a generalization of the classical enzyme concept to complex reactions involving more than one potentially limiting substrate. SU kinetics is used in DEB to model the process whereby ingested substrates are transformed into ‘reserves’ that are in turn transformed for growth and metabolic functions (i.e. ‘assimilation’).

DEB thus provides a more fully-specified characterization of the organism than does traditional ES, founded on fundamental physicochemical principles. These principles provide a powerful means for integrating sub-cellular, organismal and ecological processes. It remains to be seen, however, whether the DEB abstraction of organisms is sufficiently versatile to provide a useful platform for nutritional ecology research. An issue that warrants particular attention is the practical difficulties of estimating the DEB parameters (van der Meer 2006). Further, to achieve the generality aimed at in DEB models, species-specific detail is relegated to the ‘residual’, whereas in the organismally explicit approach such details are the grist that feeds the mill of generalizations. The ability of DEB to provide insights into the evolution of diverse and complex phenotypes has thus yet to be demonstrated (Nisbet et al. 2000). We are currently working with proponents of DEB to explore these issues further.

Toward an organismally explicit community ecology

Above we have alluded to the trade-off between species-level detail and generality in modelling community-level processes: GF is in terms of organismal detail more highly specified than is ES, whereas the simplified depiction in ES of nutrients (as elements) and organisms (principally body composition) reduces the burden of species-specific detail and thereby more readily provides generality. The question we wish to address in this section regards the extent to which GF models are able to enlighten population- and community-level processes.

One recent example of an application of GF to a population-level phenomenon is the demonstration that the combination of protein and salt shortage in the environment, coupled with organismal regulatory responses to these nutrients, explains mass migration driven by cannibalism in Mormon crickets, Anubis simplex (Simpson et al. 2006). The question of how GF might reveal the impact of nutritional phenotypes on communities has previously been discussed in the literature (Raubenheimer & Simpson 2004; Kearney & Porter 2006), and was recently investigated empirically. Behmer & Joern (2008) tested whether co-existence of seven species of generalist grasshoppers that feed on a similar group of host plants can be explained on the basis of niche partitioning at the level of nutrients. Their results revealed significant differences in selected protein : carbohydrate intake targets between all pairwise contrasts except one. Furthermore, peak performance (in terms of development time and growth rate) corresponded with the intake points selected by these species, thus linking nutrient selection to demographic responses. These data support the idea that competition for nutrients might drive niche partitioning in the form of divergent intake targets.

Behmer and Joern's study provides a clear illustration of how models of macronutrient regulation and its links to animal performance may contribute to an understanding of species interactions. We concur, however, with their assessment that this study represents a starting point, and substantial conceptual and empirical ground remains to be covered. It is, for example, critically important to maintain a realistic perspective on the role that nutrients play in community-level processes, and thus on the role they should play in models of these processes. In particular, ecological interactions take place between organisms (both consumers and consumed), and not between nutrients, and in most cases the proper focus for the analysis would thus be the organism. Nutrients can, however, play an indispensible role in explaining and predicting the interactions among organisms, especially if the model is sufficiently organismally explicit to embody the notions of evolutionary function and homeostasis (e.g. Behmer & Joern 2008).

In the system studied by Behmer & Joern (2008), for example, it is tempting to characterize the interactions among the component species as competition for nutrients, but in ecological terms what is actually being competed for is not nutrients but foods (in this case plants). Thus, two herbivore species that had widely dissimilar intake targets might nonetheless come into direct competition if they relied on different parts of the same plant – for example, seeds vs. leaves. In this case the unit being competed for is plants, and nutrients are relevant only in so far as they constitute the functional reason for that competition. Any model that does not consider this fails to meet the criterion of ecological explicitness, and falls into the same trap as do element-based models of organismal responses: they represent an inappropriate level of reduction.

With this caveat in mind, we believe that models which are both nutritionally and organismally explicit have considerable potential to enlighten community-level processes. In Fig. 4 we present an example that explores this potential by modelling in the context of food webs interactions between macronutrient balance, body composition, and energetic maintenance requirements. The model, which follows the same structure as Fig. 2, demonstrates several points:

Figure 4.

Implications for food webs of interactions between macronutrient balance, body composition, and energetic maintenance requirements. (a) Herbivore: The positions of the growth target (GT) and nutrient target (NT) for a hypothetical herbivore are indicated in a protein (P), carbohydrate/lipid (CL) space. The range of food compositions available to the herbivore is indicated by the grey segment. The position of the NT is near the middle of the distribution of plant food compositions, under the assumption that animals evolve to centre their intake requirements with respect to their nutritional environments. The line E (=D(u)p + D(u)cl) indicates the energetic requirements of the animal, and D(u)cl/D(u)p the optimal C + L : P fuel blend for meeting these requirements; in this case, principally carbohydrate and lipid, with some contribution from protein, as for many herbivores. From theory, and our locust data (Raubenheimer & Simpson 1993), it is expected that D(u)cl + D(u)p will be approximately 10 times the sum of energy at the GT; that is, there is a 10% trophic energy transfer efficiency. (b) Primary carnivore: On the assumption (again based on data, e.g. Fig. 7 in Raubenheimer & Simpson 2004) that our herbivores have been 50% efficient in regulating their body compositions relative to the range of diets available in (a), the range of food compositions offered by our population of herbivores to the next trophic level is narrowed relative to the compositions of plants available to herbivores. Importantly, the narrowing is not symmetrical. Because the GT is protein-biased (animal tissues, whatever their trophic level, contain high levels of protein), the range of herbivore body compositions is shifted towards a higher mean P : CL ratio than was seen among plants. The NT, assumed for simplicity to comprise the same sum of CL and P as the herbivore (i.e. to fall on line AB), has shifted to reflect the distribution of prey body compositions, such that now more of E must come from protein. (c) Secondary carnivore. The progression continues in the transfer from primary predator to secondary predator. Again there has been a 50% narrowing of body compositions in predators relative to their prey (primary carnivores), and again this has been asymmetrical, being constrained by the high protein content of the growth target. By the fifth trophic level (not shown) there will very little variation in prey body compositions left. The overall effect (shown in d.) is that as trophic levels are ascended, homeostatic feeding and growth responses will progressively limit the range of body compositions and shift the mean composition towards a higher protein ratio.

  • 1As trophic levels are ascended, homeostatic feeding and growth responses will progressively limit the range of body compositions and shift the mean composition towards a higher protein ratio (Fig. 4d). This would explain the observations from ES that N% rises across trophic levels, and that the ratio of carbon to nitrogen between foods and consumers (C : N resource/C : N consumer) narrows progressively moving up trophic levels (Denno & Fagan 2003; note the similarity between fig. 2c in that article and Fig. 4d in our article).
  • 2Consumers become progressively carbohydrate and fat limited across trophic levels – not nitrogen limited as suggested by ES (Denno & Fagan 2003). Thus the nutritional incentives to feed down the food chain, rather than up the food chain (Denno & Fagan 2003), become greater at higher trophic levels as metabolic energy becomes increasingly scarce.
  • 3Collectively, these nutritional effects, in conjunction with the progressive loss of energy across trophic levels (the trophic pyramid effect), might explain why food webs are typically limited to fewer than 4–6 trophic levels. We are exploring this possibility further at present.

In closing, we note that both the model in Fig. 4 and the study of Behmer & Joern (2008) represent ‘as is’ applications of GF to questions of community ecology. To achieve a broader applicability, GF would need to be extended to capture the spatial and temporal dynamics of ecological communities. This project is already underway (Simpson et al., under review).

Conclusions

We have suggested that an over-arching framework for nutritional ecology would be nutritionally, organismally and ecologically explicit, and should be heterarchically structured. Numerous frameworks have been applied within nutritional ecology but, like OFT, most are based on single-currency models and are thus not nutritionally explicit. CINE was influential in highlighting the need for a nutritionally explicit approach, although it failed to produce a modelling framework for dealing with multiple currencies. ES, in contrast, has provided a useful multi-currency tool for ecological studies, but its focus on elements and on body composition as a proxy for fitness has reduced its utility for organismal studies. DEB theory more fully specifies phenotypes than does ES, but faces challenges of parameterization and incorporating species-specific detail. GF, on the other hand, was developed as a nutritionally explicit approach for organismal biology, but its potential to contribute to community ecology has yet to be proven. While each paradigm has yielded valuable contributions in their own right, we believe that an over-arching framework for nutritional ecology could only be achieved by combining aspects of these approaches. Such attempts are already underway. For example, in ES models Anderson et al. (2005) and Boersma & Elser (2006) have considered the costs of nutrient excesses, Hessen & Anderson (2008) explored the complexities of the notion ‘excess C’, and a few studies have focussed on biochemicals rather than elements (Anderson 1994, Anderson & Pond 2000; Anderson et al. 2004). Simpson et al. (2004) incorporated into GF the a priori predictive approach on which OFT is founded, and Raubenheimer & Simpson (2004) explored similarities and contrasts between ES and GF. More recently, Behmer & Joern (2008) have applied GF in an empirical study to questions of community ecology, and in the present article we have explored some implications of geometrical analysis for theoretical ecology. We consider further expansions/syntheses of existing models to be a priority goal, which stands to benefit nutrition, ecology and nutritional ecology alike.

Acknowledgement

Authors thank Tom Anderson, Mike Kearney, Carol Boggs, Spencer Behmer and two anonymous referees for useful suggestions which improved this article.

Ancillary