Trade-offs Among Resilience, Robustness, Stability, and Performance and How We Might Study Them

Synopsis

Biological systems are likely to be constrained by trade-offs among robustness, resilience, and performance. A better understanding of these trade-offs is important for basic biology, as well as applications where biological systems can be designed for different goals. We focus on redundancy and plasticity as mechanisms governing some types of trade-offs, but mention others as well. Whether trade-offs are due to resource constraints or “design” constraints (i.e., structure of nodes and links within a network) will also affect the types of trade-offs that are important. Identifying common themes across scales of biological organization will require that researchers use similar approaches to quantifying robustness, resilience, and performance, using units that can be compared across systems.

Introduction

The concepts of robustness, resilience, stability, and performance transcend biology. Indeed, biologists have borrowed the terms from engineering and physics. Today they are applied to topics as disparate as corporate governance, macroeconomic trends, and interpersonal relationships (Walker et al. 2004). Use of the same concepts across broad intellectual areas makes explicit definitions essential (Box 1). In any system with constraints, there may be efforts put towards one use, trait, or process that consequently may not be put towards another. The effort, or the components being traded off, could be thought of as energy, a limited resource, space, time, information, and/or design. For instance, allocation trade-offs can occur at an individual level: if a plant uses available nitrogen to create new RuBisCO molecules, investing in photosynthetic capacity, then it cannot simultaneously invest the same nitrogen to build alkaloids that could be used to defend against herbivores (Mooney and Gulman 1982). Resource allocation trade-offs can also constrain evolutionary adaptations, as described by life history theory (e.g., offspring number vs. offspring size; Smith and Fretwell 1974). Information processing presents another type of potential trade-off; for example, faster information transfer can be cheaper, but at a cost of information integrity and accuracy (e.g., DNA-repair functions, see review in Ferenci 2016), or integration of hormonal or other signaling networks can prevent independent optimization of jointly controlled functions (Glazier 2015). Time can also be a limited resource; time spent foraging by lemurs cannot be spent avoiding predators (Karpanty 2006). Rather than physical resources, processes may be limited by design constraints of molecules (e.g., stability vs. activity of similar-sized proteins/enzymes; see review in Ferenci 2016), or by design constraints of networks, as in the structure in a gene regulation network (Kalisky et al. 2007, Poelwijk et al. 2011) or interaction network (e.g., the structure of nodes and edges in a network may affect performance and robustness, Newman 2003). Sometimes it is possible to achieve multiple functions simultaneously, but when that is not possible or when functions inherently conflict, then trade-offs must exist.

Box 1

Here we provide brief definitions, with additional detail and examples in the main text

Resilience is the ability of a system to persist and maintain acceptable function when challenged by a perturbation or disturbance. Performance may vary at different points within the domain of resilience, but is qualitatively “acceptable” relative to points outside that domain.

Two similar definitions of robustness coexist in the literature. In general, robustness is the ability of a system to withstand a perturbation without undergoing substantial changes. In networks, whether of interacting genes, proteins, or species, robustness is the lack of additional changes to a network following an initial small change such as the random deletion of a single node or link, also sometimes called “attack tolerance.”

The performance of a biological system is a rate of a process, or an amount of a product. Higher rates or greater quantities of an end product result from systems with higher performance. An advantage of defining performances by rates is that this can facilitate comparisons across systems.

A trade-off is a negative functional association between two traits (usually in biological contexts, fitness-relevant traits, e.g., Zera and Harshmann (2001), but the choice of relevant traits will depend on the scale of biological organization being studied).

A perturbation is any stressor that can cause a biological system to change function, or if large enough, to shift to an alternative state with altered function or capacity to function.

Stability Is whether or not (or the rate at which) a system returns to an equilibrium after a small perturbation. It can be quantified by the eigenvalues of a Jacobian matrix, when the behavior of a system is governed by, or approximated by, a system of linear equations. In dynamical systems studies, the return time to a stable equilibrium has been called resilience; Holling (1996) defined this as “engineering resilience” in order to distinguish this from the definition of resilience given above.

Efficiency is the ratio of one performance rate to another, or if performance is an amount, then as the production of matter or conversion of energy as a proportion of the building blocks from which it was produced.

A discount rate is the rate at which a future return on an investment of resources is discounted, relative to their immediate value.

For a range of reasons, robustness, resilience, stability, and performance are all typically desirable goals (or desirable characteristics) of biological systems from a human point of view, at scales from molecular networks to ecosystems. In this review, we briefly review some empirical evidence across these scales that suggests there can exist trade-offs between performance on the one hand and robustness and/or resilience on the other. Identifying and understanding the mechanisms underlying these trade-offs is essential to developing predictive and generalizable models of how systems respond to perturbations and how, when necessary, one might attempt either to maximize performance, stability, robustness, or resilience or to minimize the trade-offs. Predictive and general models will improve our understanding of the unifying principles of biology and can be applied to such issues as food security (e.g., plant response to, or breeding for, and environmental change), infectious disease control (virulence vs. transmission), and ecosystem services (e.g., water retention and flood control). This paper aims to develop a framework for evaluating these goals and trade-offs and to suggest empirical and computational approaches for such evaluations.

Motivation

We live in an era of novel perturbations and changing perturbation regimes. To meet these challenges we need to (1) understand how high system performance is maintained and (2) learn to design robust and resilient biological systems that also function at high performance levels. As noted above, a key challenge is that strategies that increase resilience or robustness could entail trade-offs that reduce performance, or trade-off robustness against resilience. In addition to fundamental questions in basic biology, there are important applications that come from all scales of biological organization. For example, gene regulation networks that are robust and resilient can maintain high performance despite perturbations from, say, mutations, pathogens, or oxidative tissue stress. A robust and resilient agricultural system will maintain food and biomass production in the face of climate perturbations, while a robust and resilient biomass digester will maintain fuel production despite heterogeneous feedstock supplies. It will be critical to identify actions that can increase robustness and resilience while minimizing performance losses, or, at a minimum, to identify the shape(s) of trade-off curves so that informed decisions about optimal degrees of stability, robustness, and/or resilience are possible.

Robustness, resilience, and stability

We are using broad definitions of robustness and resilience as responses that limit the effect of a perturbation on a system (Box 1). There are many different facets of how systems respond to perturbations, and different terms are used in different fields, including different subfields of biology. We follow Holling (1973), who is generally credited with the first definition of resilience in an ecological context, in defining resilience as the ability of a system to persist and maintain function when challenged by a perturbation or disturbance. Note that this is separate from stability, the rate at which a system returns to an equilibrium, or even the existence of a single equilibrium; these aspects are termed “engineering resilience” (reviewed in Holling 1996) and can be quantified by eigenvalues of a Jacobian matrix evaluated at equilibrium, at least when the behavior of a system can be approximated by a system of linear equations.

We suggest that one way to operationalize separate measures of robustness and resilience would be describe the stability of a biological system with an energetics landscape. Resilience can be quantified by the size of the domain of perturbations within which a system will persist or maintain its function (Fig. 1). An energetics landscape is often described heuristically by a “cup and ball” diagram, using a physical analogy to the local stability of a ball resting at the bottom of a cup. Maximum entropy theory can be used to describe an energetics landscape in multiple dimensions, for systems of interactors from molecules to species (Cofré et al. 2019). The domain of resilience is easier to quantify in simple systems, but recent advances in quantifying energetics landscapes (e.g., Suzuki et al. 2021) make it possible to estimate resilience at scales from protein networks to ecological communities.

Fig. 1

Heuristic cup and ball diagrams, illustrating energetics landscapes of different systems. In panel (A), the system has a broad domain of resilience, i.e., the range of perturbations within which the system will remain and perform a function. By implication, the system may have qualitatively different function in parts of the energetics landscape outside this domain (not pictured). The plateau marked by the asterisk is within the resilience domain, but would not exhibit stability. The robustness of the system is given by the minimum energy required to move the system outside the domain of resilience. Panel (B) illustrates an energetics landscape with high resilience but low robustness, whereas panel (C) illustrates a case with high robustness but low resilience. When quantified via maximum entropy, whether biological systems have trade-offs between robustness and resilience is an open empirical question.

Although some authors conflate robustness and resilience, we attempt to keep them distinct. In general, robustness is the ability of a system to withstand a perturbation without undergoing substantial changes. This does not necessarily imply that that a system has a point of local stability, because the definition of returning to the exact previous equilibrium point is challenging. For example, for a diverse multi-species community local stability implies returning to the exact same abundances of every single species. A wide range of mechanisms can lead to robustness, and a system that appears unchanged at one scale may accomplish this via changes at other scales. For example, thermal homeostasis at an organismal scale can be accomplished via changes in behavior and physiology. In the context of an energetics landscape, robustness can be quantified by the minimum effort required to shift a system to a new state. Thus, resilience and robustness can, in principle, be quantified separately (Fig. 1).

In ecological networks, robustness typically refers to “robustness to coextinction” or the number of secondary species extinctions resulting from a small perturbation, the removal of a single random species. This is equivalent to “attack tolerance” in network science more broadly, i.e., the response to the knock-out of a single node in a network. This idea has broad application in biology, e.g., the removal of a single neuron from a neuronal network, or a single gene from a genetic network, even if robustness is not typically defined in this exact way for researchers in neuroscience or genetic networks. Because perturbations are defined as an alteration of the network, it is impossible to keep a definition of network robustness that requires a system to remain absolutely unchanged when perturbed. Network resilience is the ability of the network to maintain fundamental functions, such as percolation and patterns of connectivity, despite deletions of nodes or edges (Newman 2003). Thus, resilience can be quantified as the size of the domain of perturbations within which a network will still maintain an acceptable degree of function, while robustness can be quantified as the ability to avoid tipping into a new state due to propagation of losses following the initial small perturbation. Again, in other contexts robustness and resilience have both been taken to be a more general idea of a system's capability of being able to rebound from a disturbance, or maintain an original structure, function, and patterns of interactions despite perturbations (Walker et al. 2004).

Neither robustness nor resilience, for any system or scale, can be defined in isolation; robustness and resilience are measured in response to a particular type and scale of perturbation. That is, they must be defined with reference to a domain of perturbations, and a system that is highly robust or resilient to one type of perturbation may not be against a different type of perturbation. For this reason, it will also be important to develop ways to predict future perturbation regimes, including changes to multiple external factors. A biological system that has evolved to be robust or resilient due to selection from previous challenges may still be vulnerable to novel perturbations.

Performance and efficiency

The act of performing, the doing that results from life itself, is exceedingly difficult to define, conjuring Supreme Court Justice Potter Stewart's remarks on determining a legal definition of obscenity: “I know it when I see it” (Gerwirtz, 1996). At different scales of organization, “performance” means very different things, but it often refers to the rate of a process or the amount of a product. At cellular scales performance is often considered in terms of anabolic rates. At organismal scales, a common metric is the number of offspring that survive to reproduce (“fitness” in the Darwinian sense). In communities and ecosystems, processes such as carbon fixation/mineralization, nitrogen mineralization, and energy flux are common metrics of performance, or processes that result in provision of “ecosystem services” for humans.

The critical feature that all of these performances share is that their units can be expressed as rates, either individuals, grams, or moles produced or consumed per unit time. To compare system performance in meaningful ways across scales and types of study systems, we must develop ways to compare performance metrics. Using rates to evaluate performance has the advantage that rates are typically measurements on a “ratio scale,” i.e., measurements based on a true zero (c.f. Measurement Theory, which clarifies valid analyses and comparisons based on types of metrics, Houle et al. 2011). In physiological contexts, measures of performance are often a function of environmental variables, such as temperature, providing “performance curves” (Sinclair et al. 2016). In evolutionary terms, performance of a genotype can be measured via “reaction norms” of fitness as a function of environmental variables (Dobzhansky, 1937; Stearns 1989). This approach could be applied more generally; from molecules to species, there exists a range of conditions in which performance is predicted to vary from zero to most-optimized.

Related to the concept of performance is that of efficiency. Whereas performance is an amount produced or consumed by the entity in question over a specified time interval, efficiency is the ratio of one performance to another. For example, at the subcellular level the quantum efficiency of photosynthesis is the ratio of photons absorbed to carbon dioxide molecules fixed. At the organismal scale, growth efficiency is the ratio of biomass accumulated to biomass ingested. At the ecosystem scale, water-use efficiency is the amount of carbon gained for the amount of water lost. Although the two concepts of performance and efficiency are intimately connected, they differ both in the way described above and in another important manner. There are strong reasons from first principles to think there may be inherent trade-offs between performance and robustness and resilience, whereas efficiencies may vary positively with robustness and resilience (yet may incur other costs). For example, in studies of plant responses to grazers, there are clear trade-offs between resilience and performance; those genotypes that are more resilient (resistant to a wider range of damage levels) have lower growth rates in the absence of damage (e.g., Belsky et al. 1993). The traits that lead to resistance, such as higher lignin and lower protein levels, are causally linked to lower levels of resource acquisition and growth. In general, keeping resources in reserves may increase robustness or resilience to future perturbations, but prevent current performance from being maximized, while increasing efficiency could increase current performance and also expand the domain of robustness or resilience by increasing resource availability. On the other hand, increases in efficiency may incur costs, such as greater investment in digestive organs and enzymes required to extract a higher proportion of resources from food.

Trade-offs

Trade-offs are a universal concept across many fields of inquiry. A decision, allocation pattern, or network architecture that enhances one kind of outcome prevents efforts or resources from being used for something else, a feature of biological systems from micro to macro. Here, we offer some empirical examples of trade-offs occurring between performance and either resilience, robustness, or stability. We do not provide an extensive review, and indeed there are few well-documented examples for some kinds of trade-offs; we hope to inspire research that uncovers new examples. Information processing can present a trade-off, in which faster information transfer comes at the cost of robustness to errors over time (e.g., DNA-repair functions, see review in Ferenci 2016). At an organismal level, trade-offs are central to life history theory, where selection on one fitness-linked trait comes at the cost of another (van Noordwijk and de Jong 1986, Stearns 1989). For instance, current reproductive effort comes at a cost paid in future survival and reproduction (Williams 1966). Further, energy allocated to somatic maintenance and growth cannot be allocated to maturity maintenance and reproduction (Zera and Harshman 2001; Kooijman 2009  ; Gélin et al. 2016). Thus, during times of resource scarcity, individuals may face a trade-off between performance metrics such as reproductive output and resilience to starvation. In evolutionary contexts, trade-offs can occur when traits that increase fitness in one context may come at the cost of reducing resilience to environmental variability. For instance, there are costs and limitations that constrain a population from being well adapted to diverse environments (Moran 1992). As is clear from these examples, limiting components can be measured in units of time, space, energy, information, or physical/chemical resources (e.g., grams of nitrogen, phosphorus, or moles of amino acid), thus trade-offs resulting from these limitations are diverse and ubiquitous.

While trade-offs are universal, the currencies that impose limits can change over time, space, levels of biological organization and biological study system. The nature of the resource currency may also alter strategies biological entities take to avoid or respond to perturbations. Similarly, discount rates for performance costs (Simon 1959) can differ among systems, favoring either performance or robustness and resilience, depending on the balance of risks and future rewards (Lerdau 1992). Future discounting allows calculation of future costs and benefits in terms of their present values; the principle is based on the idea that both benefits and costs have a higher absolute value in the present than in the future (Samuelson and Nordhaus 1989), such that distant benefits are worth less than near-term benefits. The cost of investment in current structures should not be compared to future potential benefits, e.g., with lag due to ontogenetic constraints, without discounting future benefits appropriately. Similarly, the loss of species with high or low discount rates may have different cascading impacts on communities or species interactions, because different discount rates may be associated with different life history strategies. This is an area in need of more research.

Identifying characteristic time scales that could be used for standardizing across study systems and scales is a challenge for integrating biology. Because performance metrics, stability, perturbations, and the responses to them have an explicit temporal component (duration of the perturbation and length of recovery time), comparing studies at different scales of organization will require a way to standardize durations. For example, metabolic pathways may respond to short-term (“pulse”) perturbations at a scale of minutes to hours, whereas plant communities may respond to perturbations at scales of months to years.

In addition to classic types of trade-offs mentioned above, other trade-offs can occur at more abstract levels, in parallel with the idea of design trade-offs as opposed to resource trade-offs. One example of a trade-off in ecological networks is the inherent trade-off between local stability and robustness (to co-extinctions) of mutualistic networks, i.e., networks of partner species interacting for mutual benefit (Fig. 2). Local stability can be defined by the eigenvalues of a system of interactions, and in particular, if the real parts of all of the eigenvalues in a dynamical system are negative, then the system is stable (i.e., will return to its original equilibrium following a small perturbation). Robert May (1973) showed that as systems become more complex, in terms of having both more nodes and more connections between nodes (i.e., “edges” or “links”), their probability of local stability is reduced. By contrast, in mutualistic systems where species are interacting for mutual benefit, having more potential mutualistic partners, and thus more connections, increases robustness to coextinctions. Thus, there is a strict trade-off between local stability and robustness to coextinction (in mutualistic systems), driven by the patterns of node and link distributions in interaction network structure, rather than by any limiting resource. The number and type of links a species can have within such a network may be constrained by adaptations necessary for efficient use of one partner species that simultaneously preclude interactions with other species (e.g., short-tongued bees may be excluded from flowers with long corollas). Similarly, interactions within gene networks may be constrained by shared use of regulatory pathways.

Fig. 2

Different connectance levels trade off between robustness and local stability in mutualistic networks. Values show the mean of 100 simulations, ± 95% confidence intervals. See text for simulation details.

Figure 2 shows the results of mutualistic network simulations comparing local stability and robustness to coextinction for two levels of connectance (the proportion of realized possible links in a network). The simulations are focused on networks of 24 species in two trophic levels of 12 species each (e.g., plants and seed dispersers, host fish and cleaner fish, and so on), with connectance set at 0.125 for the “low” value and 0.5 for the “high” value. We ran 100 simulations at each level of connectance and measured a proxy for local stability (largest real part of any eigenvalue) as well as robustness to coextinction (using the “robustness” calculated within the “grouplevel” function of the “bipartite” package for R, Dormann et al. 2009). For local stability, we assumed a full adjacency matrix with no within-guild interactions and self-suppression among the primary diagonal set to −2 (May 1973). A fully reproducible Rmarkdown report including all code is included as Supplemental data S1.

Potential challenges for developing a unifying theme

Contexts

The context within which perturbations affect systems, be they organelles, cells, organisms, or ecosystems, can have dramatic effects on performance, on robustness and resilience, and on the relationship among these three. Whether a system's response to a perturbation is generic or specific under varying contexts can ultimately alter the direction and magnitude of trade-offs. Further, how the contexts of perturbations are sensed and information about contexts is transferred remains a challenge to identify in many systems. Sometimes the contextual effect involves the availability of complementary resources. For example, a 2-week drought may have very different impacts depending on whether or not groundwater is accessible, or on the precipitation history of the site. A range of pine (Pinus) species in the western USA are far more robust to bark beetle attacks when they have adequate water than when they have been suffering a water deficit (Arango‐Velez et al. 2016). Although the principle that systems with lowered performance because of one perturbation tend to have lower robustness and resilience with respect to other perturbations is empirically well supported, a concerted effort to study system responses to multiple perturbations under varying contexts is needed (Burton et al. 2020).

Performance metrics

It is important to bear in mind that performance can be measured with quite different units, depending on the process under consideration. While this diversity of metrics for performance can complicate efforts to compare performances across different processes and different systems, it is orthogonal to the question of the relationships among performance and robustness and resilience, and thus whether trade-offs can be identified. One key for studies of these relationships is to use similar units across performance, robustness, and resilience. For example, if a plant's photosynthetic performance is measured as grams C fixed per unit mass of plant per unit time, then stability of photosynthesis could be the time constant for photosynthesis to return to its pre-perturbation rate. That is, the units used for any one analysis of robustness, resilience, or stability must correspond to the units employed in the performance measure. A mismatch of units is likely to obscure important trade-offs, as well as any positive relationships. Further, studies will need to have an unbiased determination of a reference point in time that the previous state a system must return to in order to be considered resilient.

Scale

From a broader perspective, generalizing patterns of trade-offs in robustness, resilience, stability, and performance across scales of biological organization will be essential to developing unifying principles of biology. In some cases, trade-offs at one scale are shown to affect performance at another, thus quantification and significance of robustness and resilience depend on the scale at which they are measured. For example, leaf emissions of isoprene trigger cascades of cellular changes via multiple signaling pathways, with resultant trade-offs among various compounds, and these can also affect whole-plant trade-offs between abiotic stress tolerance (the resilience of plant growth to climatic perturbations), and defense against herbivores (Monson et al. 2021). As another example, the transfer of energy in trophic systems can be altered by the magnitude of stress-induced changes in herbivore physiology and resource choice, which has downstream effects on nutrient assimilation efficiency, nutrient content of soils, and plant-species composition, and energy transfer up the food chain (Hawlena and Schmitz 2010). In this case, the performance at the ecosystem level (i.e., productivity) is affected by organismal robustness to stress that results in a trade-off with individual performance traits such as survival and nutrient assimilation efficiency. A recent perspective by Agrawal (2020) posits that often scale-dependent evolutionary processes will decouple trait relationships (e.g., putative trade-offs) across scales from within to among species. We agree with this perspective that a better understanding of scaling trade-offs will require stepping out of our disciplinary bounds. The issue of scale in understanding these trade-offs also arises when we consider performance in multi-dimensional parameter space (i.e., multiple axes of performance considered simultaneously), but this hurdle is diminishing with improved simulations of experimental perturbations on complex traits (e.g., pattern formation in embryonic development).

Mechanisms underlying trade-offs among resilience, robustness, and performance

One source of trade-offs may be structural, analogous to engineering constraints. For example, a leg that is more robust to physical stress, stronger or harder to break off, may not be as capable of rapid movements and may grow more slowly, and may be less resilient. Despite the wonders of some biological materials, their physical properties will still be subject to fundamental constraints.

Redundancy is another mechanism by which systems can generate resilience or robustness under some conditions. For example, experiments have examined yeast with single transcription factor knockouts, for the full complement of all known transcription factors in the yeast genome. Interestingly, only ∼3% of the binding target genes of those transcription factor removals had any impact in terms of their level of transcription (Dean et al 2008). Thus, the transcription factor network in yeast is highly resilient to single gene knockouts. Furthermore, that resilience seems to largely be driven by redundancy: nearly all of the binding target genes each interact with multiple transcription factors, which allows for essentially uninterrupted function in the face of this perturbation, single gene knockouts (Hughes and de Boer 2013). Similarly, in the aforementioned mutualistic network example, having multiple positive interaction partners allows for systems to be more robust to the loss of a single interaction partner (e.g., Memmott et al. 2004).

Because we are focused on trade-offs, it is important to note that redundancy can have costs. As mentioned previously, increasing the complexity of a dynamical system decreases its probability of local stability, strictly defined. In addition, redundancy can also reduce performance in some cases. Redundancy can carry costs in terms of the resource use required to have a functional “back up.” For example, a bacterial cell with many redundant copies of the same or similar genes may reproduce at a slower rate (i.e., have lower performance) than a competitor with fewer redundant copies, given the time and resource costs of DNA replication. Another cost can come about if the multiple redundant actors that contribute to functioning (transcription factors, neurons, species, and so on) vary in their functional efficiencies. For example, if one bird species is by far the most effective at dispersing the seeds of a particular plant, by having a redundant assemblage of bird dispersers that include several less-effective types, a plant may suffer reduced mean seed dispersal compared to a scenario that includes only its most effective partner, yet be more resilient to loss of a disperser species. Indeed, biodiversity–ecosystem functioning theory (with substantial support from experiments) suggests that the highest-performing communities are those with the greatest complementarity among their functional aspects, i.e., when the species’ functional roles complement one another. Restated, this is when there is the least redundancy among species roles. Thus, there may be an inherent or general trade-off between performance (functioning) and redundancy. This is a promising area for future explorations. For example, if there is a bacterial species with sufficient variation in the number of transcription factors among genotypes, it would be interesting to see if functioning trades off with robustness to single transcription factor losses. This would of course require variation in both transcription factor number and identity to ensure that the identity of the transcription factors does not confound the results.

We posit that plasticity should generally lead to greater system resilience and robustness, and it may—in some cases—do so in ways that reduce apparent trade-offs with system performance. Plasticity is the capability of system components to change in response to exogenous factors. For example, organismal phenotypic plasticity is some developmental, physiological, morphological, or behavioral change in response to an environmental cue. If we consider biological systems as networks, comprised of nodes and links, plasticity can manifest itself as changes both in the nodes themselves (for example, organismal phenotypic plasticity) and also in the links. For link changes, these can be driven by, for example, behavioral changes in nodes (predators changing prey preference, thus changing linkage patterns).

There is evidence that plasticity can lead to enhanced network response to perturbation. In organismal consumer-resource networks (food webs and/or mutualistic networks), plasticity in foraging (i.e., adaptive foraging or optimal foraging) can lead to changes in both the intensity of feeding and the identity of which resources are foraged upon. This plasticity, in turn, has been shown in models to improve system performance in response to perturbation, such as increased persistence of nodes (species) in these networks (e.g., Kondoh 2003; Staniczenko et al. 2010; Valdovinos et al. 2016). While this work has been framed in community ecological terms, the models used to explore these concepts should be general to many consumer-resource systems at multiple levels. These examples indicate a potentially strong mechanistic role for plasticity in driving resilience and robustness at various levels of biological organization, though more work is needed to elucidate general patterns of exactly how and why plasticity plays this role. Similarly, this kind of mechanistic understanding would increase our ability to discern when plasticity might enhance vs. detract from robustness. One potential way in which plasticity could detract from robustness is if it carries substantial costs. In organismal-level studies of plants, however, several studies have attempted to characterize the costs of plasticity and these costs have not been conclusively demonstrated to date, despite some excellent studies designed for this purpose (e.g., Auld et al. 2010).

Plasticity can allow for resource allocation changes that maintain performance. This kind of mechanism also operates at organismal and smaller scales, e.g., in biochemical or genetic pathways that allow for alternative substrates or binding targets. At an organismal scale, there is evidence that plants exhibiting greater phenotypic plasticity can respond to environmental perturbations such as droughts more successfully, in terms of survival and reproduction, but this greater plasticity often comes at an evolutionary cost, thus a trade-off exists (Levins 1968).

Future directions

Understanding the mechanisms underlying trade-offs among robustness, resilience, stability, and performance is a necessary first step towards developing coherent approaches to maximizing one outcome, e.g., performance, or robustness, or minimizing the magnitude of the trade-offs themselves. Even more critical will be the identification of common currencies and measurement units to facilitate communication and comparisons across fields; without common metrics it will be difficult to identify generalities that span levels of organization. A consistent use of “rates of production of X, dX/dt” as performance metrics is a starting point, because proportional changes in performance can be compared across systems. The scaling of appropriate time units for measuring stability, however, as a function of return times to an initial state, is an outstanding challenge. The standard eco-evolutionary approach of re-expressing times in units of generations may have broad application, for example scaling down to the timing of cell division or the transcription time of a single focal gene. Still, such scaling does not always work, and may not be applicable to metabolic pathways or interaction networks. Finally, there is exciting potential in the identification and classification of mechanisms that confer robustness and resilience to different types of perturbations, and to multiple perturbations. It will require creativity and interdisciplinary communication to recognize whether mechanisms in different systems or at different scales are analogous or homologous. The development of future theoretical frameworks that integrate across scales will likely require knowledge of underlying mechanisms in order to provide useful predictions about responses to potential perturbations beyond the range of current data. Further, to address global anthropogenic change, it becomes essential to predict responses of systems to future perturbations. This cross-scale approach to studying trade-offs is an essential first step in the reintegration of biology.

Acknowledgments

We thank the organizers and facilitators of the Atlanta “Reinvigorating Biology” workshop for promoting interesting discussions, and two anonymous reviewers for their constructive suggestions.

Funding

This work was supported by a grant during manuscript preparation from NSF IOS-2005574. ML acknowledges helpful discussions with H. Shugart.

Data availability

No original data are present in this article, and code is provided to reproduce Figure 2.

Notes

Based on a Jumpstart-Reintegrating Biology Vision Paper, developed during Town Hall meetings funded by The National Science Foundation in 2019–2020.