On the Dynamic Nature of Omnivory in a Changing World

Abstract

Nature's variability plays a major role in maintenance of biodiversity. As global change is altering variability, understanding how key food web structures maintain stability in the face of variation becomes critical. Surprisingly, little research has been undertaken to mechanistically understand how key food web structures are expected to operate in a noisy world and what this means for stability. Omnivory, for example, has been historically well studied but largely from a static perspective. Recent empirical evidence suggests that the strength of omnivory varies in response to changing conditions in ways that may be fundamental to stability. In the present article, we extend existing omnivory theory to predict how omnivory responds to variation and to show that dynamic omnivory responses are indeed a potent stabilizing structure in the face of variation. We end by synthesizing empirical examples within this framework, demonstrating the ubiquity of the theoretical mechanisms proposed across ecosystem types, spatial scales, and taxa.

Ecosystems are notably variable and subject to regular polyrhythmic swings in abiotic conditions that drive bottom-up shifts in resource density (e.g., diurnal, seasonal, decadal changes in temperature, precipitation) and create patterned mosaics of spatial heterogeneity (Mougi 2020). The intensity of resource consumption by predators can also change through time and space, leading to top-down shifts in resource density (Sommer et al. 1986). The ubiquity of these varying (nonequilibrium) conditions, which are predictable to some extent, means that organisms within ecosystems have likely adapted in numerous ways to respond to changing conditions (Levin 1998).

Despite the well-known fact that ecosystems are subject to such nonequilibrium conditions, ecological theory and empiricism have tended to focus on understanding stability and function from a static equilibrium perspective until relatively recently (Ushio et al. 2018). The equilibrium simplification is understandable as it allows elegant analysis for high diversity systems (May 1972, Allesina and Tang 2012, Gellner and McCann 2012) but potentially misses out on the dynamic responses of species that play fundamental roles in maintaining their persistence in a noisy world (Neubert and Caswell 1997, Hastings 2004, Hastings et al. 2018). Understanding the multidimensional nature of ecological stability requires examining dynamic responses from multiple perspectives (e.g., equilibrium and nonequilibrium; Ives and Carpenter 2007, Donohue et al. 2016). Importantly, these dynamic responses are measurable in empirical systems, promising the ability to develop a dynamic theory that can be linked to empirical research. Given that climate change is altering these underlying abiotic polyrhythms (Myneni et al. 1997, Cai et al. 2018) and homogenizing spatial heterogeneity (Olden et al. 2006), it is essential that theory and empiricism understand the role of responsive food web structure in mediating ecosystem stability and function.

Recent research has supported the generalist module (defined as the coupling of micro or macrohabitats in space by a mobile, generalist consumer; Figure 1a) as a clear example of a food web architecture that can be a potent stabilizing force in a variable world (McCann and Rooney 2009, McMeans et al. 2015, 2016). At any given time, a generalist consumer tends to move in space toward the most profitable habitat (Pyke et al. 1977). Given that different habitats tend to have resource dynamics that are asynchronized, then this simple adaptive behavioral response by the consumer allows it to consume the resource in the highly productive habitat while releasing the resource from consumption in the less productive habitat (Figure 1a). This asynchronous flux in predation pressure is known to enhance persistence of all species in the generalist module and contribute to the adaptive capacity (the ability of systems to alter structure in response to external variation) of whole food webs (McCann and Rooney 2009, McMeans et al. 2016).

Figure 1.

Similarities in the generalist and omnivory module, where (a) the generalist (green) consumes prey (blue and orange) across multiple habitat compartments and employ a consumptive portfolio effect harnessing the asynchronous flux of prey biomass across two habitats through time. (b) The omnivore (green) consumes prey (blue and orange) across multiple trophic levels and employ a consumptive portfolio effect harnessing the asynchronous flux of prey biomass across two trophic levels through time.

Akin to the portfolio effect in primary producers (Tilman 1999), generalist predators employ a consumptive portfolio effect (see Table 1 for definitions) by altering their foraging behavior across multiple prey sources in a manner that ensures they get a relatively steady resource supply. This stabilization is due to a mixture of bottom-up processes (e.g., differences in habitat conditions that yield nonsynchronous resource dynamics; Figure 1a) and the top-down rapid behavior of the predator capable of generating resource asynchrony and integrating over their resources in space (Figure 1a). The generalist module shows how a mobile consumer in space may play a role in mediating variation, and omnivory has qualitatively similar underlying conditions in that it is also a module composed of a generalist predator capable of foraging on alternative resources that can vary asynchronously (Figure 1b). Notably, the generalist predator can be envisioned as a spatial generalist, and omnivores can be envisioned as trophic generalists that may also be capable of reducing variation (Figure 1).

Table 1.

Definitions of key terms related to dynamic omnivory.

Key term	Definition
Omnivory	Feeding at more than one trophic level by generalist predator (P).
Consumptive portfolio effect	Statistical averaging of community biomass, where the sum of several random and independently varying population biomass’ is less variable than an average population's biomass. Adapted from a long-standing economic principle that more diversified portfolios are less volatile (Doak et al. 1998, Tilman 1999). In the present article, the consumptive portfolio effect is the average prey density available to omnivores harnessing asynchronous fluxes in consumer (C) and resource (R) biomass.
Degree of omnivory	A measure of the contribution of R to P's diet, measured as the ratio of R/(R + C) consumed by P.
	Equation: Degom = ΩaRPR/(ΩaRPR+(1−Ω)aCPC); see supplement S1 for parameter definitions.
Passive omnivore	The omnivore has a fixed preference (Ω) (scaling of its attack rate) on R relative to C. The degree of omnivory passively tracks changes in R and C densities.
	Equation: Degom = ΩaRPR/(ΩaRPR+(1−Ω)aCPC), where Ω is constant; see supplement S1 for parameter definitions.
Active omnivore	The omnivore modifies preference (scaling of attack rates on R and C respectively), depending on the ratio of ωR/(ωR+(1−ω) C), where ω measures the speed with which the omnivore's attack rates on R and C respond to changes in their availability.
	Equation: Degom = ΩaRPR/(ΩaRPR+(1−Ω)aCPC), where Ω = ωR/(ωR+(1−ω)C); see supplement S1 for parameter definitions.
Bottom-heavy omnivory	Changes in degree of omnivory are driven purely by bottom-up abiotic influences (e.g., seasonal changes in productivity) that alter densities of R and C.
Top-heavy omnivory	Changes in the degree of omnivory are driven by cascading impacts of increasing top heaviness after a change in resource availability.

Over the last two decades, ecologists have increasingly recognized the importance of omnivorous interactions (see Table 1 for a definition of omnivory). Early theoretical arguments that omnivory was destabilizing (Pimm and Lawton 1978) have been replaced by the nuanced understanding that, although moderate to strong interactions are indeed destabilizing, weak interactions can be powerfully stabilizing (Neutel et al. 2002, Emmerson and Yearsley 2004, Gellner and McCann 2016). Over the same period, empirical work has shifted from suggesting that omnivory is rare (e.g., Pimm and Lawton 1978, but note that weak omnivorous interactions were ignored in their methodology) to showing that omnivory is rampant throughout food webs and increases in frequency with trophic level (Thompson et al. 2007). Nonetheless, empirical investigations have largely considered omnivory as a static trait within these systems (Kratina et al. 2012), while theory has also largely focused on the equilibrium stability implications of omnivory (Pimm and Lawton 1978). However, it is increasingly recognized that changing conditions can influence omnivorous interactions in space and time (Kratina et al. 2012, Tunney et al. 2012).

In the present article, we expand on this idea that omnivory ought to change in response to environmental variation (Figure 1b). In what follows, we employ two general types of behavioral responses to changing conditions (passive and active omnivores, sensu Kalinkat et al. 2011) to bracket a large range in foraging possibilities and ask whether dynamic omnivory is robustly stabilizing under these two differing foraging responses. Furthermore, following existing omnivory theory (Tunney et al. 2012, Ward and McCann 2017), we define two general categories of mechanisms that produce omnivorous responses (bottom-heavy and top-heavy omnivory). We review and synthesize theory within this novel dynamic context (e.g., responses to perturbations) to show how different conditions drive bottom-heavy and top-heavy driven changes in omnivory and that this dynamic omnivory may be an understudied stabilizing mechanism in the face of variation. We then empirically demonstrate that both passive and active omnivores—and bottom-heavy and top-heavy mechanisms—manifest themselves in the real world by reexamining well-studied food webs within our dynamic omnivory framework. Furthermore, we demonstrate the ubiquity of dynamic omnivory by providing an extensive empirical catalogue of examples that extends across ecosystem types, trophic levels, and spatial or temporal scales. We end by arguing that, like the generalist module, this dynamic understanding of omnivory allows us to consider how omnivory contributes to the adaptive capacity of food webs and how global change will affect omnivorous interactions, potentially altering carbon transfer, stability, and production in whole food webs.

A dynamic omnivory framework

In the present article, we draw from longstanding foraging ecology that has been embedded in consumer–resource and food web models through functional and numerical responses (Abrams 1982, Chesson 1983). Consistent with much behavioral ecology literature that has found that experimental data is consistent with optimal foraging theory (Pyke et al. 1977), consumer–resource and food web theory have motivated models that maximize energy intake either formulated explicitly as an optimal foraging model (Abrams 1982, Kondoh 2003, Abrams and Matsuda 2004, Beckerman et al. 2006, 2010) or as a general preference model that approximates energy maximization in the functional and numerical response (Chesson 1983, McCann et al. 2005, Kalinkat et al. 2011). For simplicity, we employ an omnivory model that uses the preference function of Chesson (1983) adopted in many food web papers (Post et al. 2000, Faria and Costa 2010). We point out that both optimal foraging models (Krivan 2000, van Baalen et al. 2001, Kondoh 2003) and the preference models (McCann and Hastings 1997, McCann et al. 2005) used in the present article have tended to consistently find that energy maximization foraging is often stabilizing.

Omnivory, feeding on more than one trophic level (Pimm and Lawton 1978), is perhaps easiest envisioned in a simple food chain (Figure 2a; see Table 1 for definitions of all of the important dynamic omnivory terms discussed in this section). In the present article, toward a simple dynamic framework of dynamic omnivory, we employ a tritrophic level perspective, and we use extensions of the Rosenzweig–MacArthur food chain model (see supplement S1 for equations) to outline some of the key aspects of omnivory that mediate its dynamic response in nature to changing conditions. Following empirical patterns that show that omnivory tends to increase as we go up the food web (Thompson et al. 2007, Zheng et al. 2021), we focus on an omnivory module that assumes that omnivory occurs through the top predator. The omnivory model and assumptions therein are based on McCann and Hastings (1997). Consistent with empirical work, we do all dynamic theory with weak to intermediate average omnivorous interaction strengths (Thompson et al. 2009).

Figure 2.

The dynamic omnivory framework. (a) Change in tritrophic food web module under conditions of an increasing ratio of resources to consumer (R: C; i.e., increasing productivity or top heaviness). Low R: C biomass ratios are characterized by a linear food chain with an Eltonian biomass pyramid distribution (left), as the R: C ratio increases, bottom-up changes in R density increase bottom-heavy omnivory (middle), and at high R: C ratio omnivores exhibit strong top-down pressure that drive cascades that increase top-heavy omnivory (right). (b) Change in the percentage of C and R in the diet of omnivore (P) under increasing R: C conditions (i.e., increasing productivity or top heaviness). As the R: C ratio increases, the omnivores diet changes from being dominated by C (linear food chain) to being dominated by R (top-heavy omnivory). (c) Change in the degree of omnivory under increasing R: C conditions (i.e., increasing productivity or top heaviness). As R: C increases and tritrophic modules transition from a linear food chain to a top-heavy omnivory module the degree of omnivory increases. In a linear food chain (left), all omnivores exhibit no degree of omnivory. As R: C increases both OP and OA increase their degree of omnivory, with OA being higher because of the ability to rapidly respond to changing prey densities. Abbreviations: C, consumer; FC, food chain omnivore (black); OA, active omnivore (white); OP, passive omnivore (grey); R, resource.

In our dynamic framework, the degree of omnivory is a measure of the contribution of the resource to the predator's diet (Table 1; see supplement S1 for the equation). We note that this simple metric equates with commonly employed empirical estimates of omnivory (e.g., stable isotopes) that also estimate the percentage of carbon that comes from different trophic levels (Cabana and Rasmussen 1996). As was discussed above, we assume that foraging preference—and, therefore, the degree of omnivory—is driven by the relative densities of potential prey items for an omnivorous predator (i.e., omnivory increases as the ratio of resources to consumers [R: C] increases; Figure 2c). In our example, we show how densities change as the level of top-down control varies as we move from a bottom-heavy Eltonian biomass pyramid (the left side of Figure 2a) to a top-heavy wasp-waisted biomass pyramid that emerges out of strong top-down control (the right side of Figure 2a). Across such a gradient in top heaviness, which arises from increasing resource productivity (K), predator attack rate (a), predator biomass conversion rate (e) or reducing predator mortality (m), the densities of resource (R) and consumer (C) in predator (P) diet change in predictable ways as was shown in Figure 2b (McCauley et al. 2018). Notably, the pyramid gets more top heavy with a significant change in the relative prey densities for the omnivorous top predator (i.e., lots of resource relative to consumer; see Figure 2a). Here, the top-down pressure has indirectly driven an increase in the degree of omnivory (the right side of Figure 2c). Furthermore, as the resource densities increase relative to consumer densities and the degree of omnivory increases, the food chain length decreases (from left to right in Figure 2a). Alternatively, if the resource to consumer ratio (R: C) decreases, omnivory decreases, and the food chain length increases (right to left in Figure 2a). The food chain dynamically expands and contracts with changes in the ratio of resources to consumers.

Within this framework, we argue there are two simple but qualitatively distinct behavioral responses of the potential omnivore to changing prey densities that ought to alter the effect of changing densities on the degree of omnivory. Importantly, these two behavioral responses bracket a continuum of possible functional responses under changing prey densities and have been used experimentally by Kalinkat and colleagues (2011). First, an omnivore may be a passive omnivore, in that it never changes its preference for the consumer or resource but consumes more or less of them depending on their relative densities (i.e., proportionally; see Table 1 for a definition and supplement S1 for details). In other words, a passive omnivore exhibits a density-independent preference. We note that even a passive feeding organism, such as a filter feeder, may still have a preference (e.g., gill sizes select for certain size prey over others; Rouillon and Navarro 2003). Alternatively, the omnivore can modify its preference continuously such that the preference increases when the resource to consumer ratio increases. This altered preference makes the omnivore an active omnivore (see Table 1 for definition and supplement S1 for the equation), in that it adjusts its preference in line with the most abundant resource, generating a nonlinear preference in the functional and numerical responses (Chesson 1983, Kalinkat et al. 2011). As an example, if there is a pulse in the resource, then the top predator may increase its foraging on the resource briefly to tap into this increase and may do so in a manner that reduces its consumption of the consumer even more than the proportional change in resource to consumer densities. As a result, an active omnivore can have a much larger change in the degree of omnivory than a passive omnivore with changing prey densities (Figure 2c).

Finally, and as we will discuss further in the next section, we define two qualitatively distinct mechanisms driving dynamic omnivory. The first occurs when the ratio of consumer and resource densities changes solely because of bottom-up abiotic influences that drive bottom-heavy biomass distribution. As an example, a pulse of nutrients may immediately fuel resource growth with other trophic levels lagging behind in response (compare the increased omnivory between region 1 and region 2; Figure 3a–3c); therefore, during the early transient period after a pulse in nutrients (region 2 in Figure 3a–3c), the increased R: C is arguably purely bottom-up driven, and such a shift in R: C promotes an omnivorous behavioral response, whether that behavioral response is passive or active (om_B > om_Eq; Figure 3b, 3c). We will refer to this as bottom-heavy omnivory (Table 1). After this early transient period, the densities of C and R eventually dynamically respond to the nutrient pulse (Figure 3a–3c, region 3). In the present example, where the transmission of energy up the food chain is driven by relatively strong interactions (i.e., a strongly top-down system, with a relatively high ratio of Kae to m; Rip and McCann 2011, Gilbert et al. 2014), the current theory suggests that this delayed transient response would yield a strong top-down cascade with a top-heavy biomass pyramid and a high R: C (Figure 3b, 3c, region 3). During this transient top-heavy phase, we would therefore expect what we will refer to as an increase in top-heavy omnivory (Table 1), because the degree of omnivory would increase because of the cascading influence of a now inflated top predator (i.e., om_T > omEq). So far, we have imagined a pulse perturbation in resources as one may expect because of seasonality, but we can also apply our rationale to a press perturbation (Figure 3d–3f) that increases K indefinitely. For example, press perturbations could also be conceptualized as spatial variation where some systems are permanently more productive than others. With these definitions, we are ready to consider the implications of dynamic omnivory on the response of food chain densities, the degree of omnivory, and the local and nonlocal stability properties of the food web.

Figure 3.

Temporal dynamics of P, C, and R densities following a pulse perturbation in K (perturbation at t = 200) (a–c) and a press perturbation in K (perturbation at t = 300) (d–f). Each time series depicts initial equilibrium starting conditions before perturbation in K (region 1 in light blue), transient bottom-heavy response to perturbation (region 2 in yellow; note this region has been overemphasized to make visualization easier), transient phase where system oscillates between top heavy and bottom heavy (region 3 in pink) and after equilibrated (region 4 in blue shading on right) for panels (a) and (d) a food chain, panels (b) and (e) a passive omnivore, and panels (c) and (f) an active omnivore. Example degrees of omnivory in each region for the passive and active omnivore are given by omEq, which represents degree of omnivory at equilibrium; omB, which is the degree of omnivory at low R: C ratios (bottom-heavy omnivory); and omT, which is the degree of omnivory at the maximum R: C ratio (top-heavy omnivory). Under pulse perturbation, dynamics return to the original equilibrium conditions, and under press perturbation, a new equilibrium is reached. Abbreviations: C, consumer; K, resource productivity; P, omnivore; R, resource.

Implications: omnivory and stability under changing conditions

We now consider the omnivory response of the predator, P, under a pulse perturbation of resource productivity, K (Figure 3a–3c) and press perturbations in K (mimicking permanently altered conditions in space or time; Figure 3d–3f; see supplement S1 for analysis details). Although we look at local equilibrium stability (local return time, based on the maximum eigenvalues), for all cases, we are also interested in nonequilibrium dynamics, so we restrict our analysis to dynamics that show overshoot (i.e., equilibrium has complex eigenvalues) that readily produce quasicycles from perturbations. We see both the press and pulse perturbations as directly related to common empirical measurements in the same ecosystem over time (e.g., a seasonal pulse in K), or a given ecosystem type over space (e.g., one habitat has higher production permanently as in a press). We will use these theoretical results to begin to synthesize empirical dynamic omnivory results with the goal of motivating future work on dynamic food web structure in general.

Figure 3a–3c depicts the time series of the omnivory model over four time periods: prior to a pulse addition of K (region 1), during its early transient bottom-heavy response (region 2), during its later transient response after higher trophic level densities respond (region 3), and after its return to equilibrium (i.e., back to region 1–type dynamics). In each case (i.e., food chain, passive and active omnivore; Figure 3a–3c, respectively), the time series show the predator (green), consumer (orange), and resource (blue) dynamics. As was discussed above, we point out that even in this simple pulse perturbation case there is a clear temporal bottom-heavy driven increase in maximum degree of omnivory (passive omnivore, omB = 0.166; active omnivore, omB = 0.216; region 2 in Figure 3b and 3c, respectively) followed by a change in maximum omnivory that occurs when the top predator has driven a subsequent cascading transient response that releases the resource while suppressing the consumer (passive omnivore, omT = 0.233; active omnivore, omT = 0.326; region 3 in Figure 3b and 3c, respectively). Therefore, we see both a short-term bottom-heavy omnivory response and a longer-term top-heavy response driven by the cascading impacts of the pulse perturbation.

To understand how dynamic omnivory affects stability, we look at both local metrics (i.e., return time) and nonlocal metrics (i.e., degree of overshoot, max–min; Neubert and Caswell 1997) of variation after a perturbation of K (see Table 1 for all definitions and supplement S1 for more details on these metrics). Recognize that in both cases of passive and active omnivory, all metrics of local and nonlocal stability tend to show stabilizing responses (i.e., faster return time, lower degree of overshoot and smaller max–min) to the pulse perturbation relative to the food chain, and the active omnivore demonstrates a stronger stabilizing response relative to the passive omnivore (Figure 4a–4c). We would argue that this stabilization is akin to the generalist predator discussed above, whereby the generalist predator and omnivorous predator are both harnessing the asynchronous responses of the consumer and the resource that are naturally occurring in the food chain (i.e., when the consumer is held in check, the resource increases and vice versa). Note that this dynamic stabilizing response is amplified with stronger top-down pressure in that stronger top-down pressure generates more asynchronous consumer and resource dynamics. This top-down driven asynchrony sets up conditions for the top predator to surf the different trophic levels in a manner that is stabilizing. In a sense, this effect is another manifestation of asynchrony generation driven by generalist predators in a noisy world, previously discussed for the diamond module (i.e., a generalist module with strong and weak pathways where the predator inherently drives asynchrony under stochastic or deterministic conditions; McCann and Rooney 2009). We note that theory has consistently found that the active switching tends to be more stabilizing (McCann et al. 2005), and indeed we see that active omnivory is even more stabilizing than passive as the active predator is able to respond quickly and strongly to changing densities in C and R and reduce the overshoot (Figure 4a–4c).

Figure 4.

Local and nonlocal stability metrics of food chain (FC), passive omnivory (OP) and active omnivory (OA) modules following a pulse perturbation (a–c) and press perturbation (d–f). Panels (a) and (d) show local return time after pulse perturbation, measured as 1 divided by the maximum real eigenvalue (Re|λ|). Panels (b) and (e) show the degree of overshoot of the resource (R), consumer (C), and omnivore (P) following the perturbation, and panels (c) and (f) show the difference in maximum and minimum density of resource, consumer, and omnivore.

Similarly, Figure 3d–3f depicts the time series of the omnivory model over four time periods: prior to a press addition of K (region 1), during its early transient bottom-heavy response (region 2), during its later transient response after higher trophic level densities respond (region 3), and the return to a now new equilibrium (i.e., region 4–type dynamics of elevated K). For empirical reasons discussed below, we draw our attention to the final new equilibrium state (region 4 in Figure 3d–3f) and ignore the transient response as it is consistent with Figure 3a–3c. This new equilibrium state following a press perturbation is akin to comparing two separate lakes with different abiotic conditions (e.g., total nutrient availability). We note that this final equilibrium state occurs after all the transient dynamics are complete and therefore shows the full cascading implications of density from omnivory after the press perturbation of a sustained increase in K. The press perturbation has effectively increased the top-down predation driving an overall increase in R relative to C for both the passive and active top-heavy omnivory. Again and importantly, local and nonlocal stability tends to be enhanced by omnivory, whether that omnivory is passive or active (Figure 4d–4f). Note that the stabilizing response of the active omnivorous predator appears to completely eliminate the oscillatory decay and returns extremely rapidly relative to the food chain or passive omnivore case (Figure 3d–3f). This is an example of stronger interactions driving asynchronous resource and consumer dynamics that are harnessed by the omnivorous predator employing a consumptive portfolio effect.

Finally, to understand the stability implications of dynamic omnivory more generally and determine the robustness of these patterns, we investigated local and nonlocal stability metrics (Neubert and Caswell 1997) across a wide parameter space for both pulse and press scenarios (see supplement S1 for details on stability analysis). We individually altered all parameters that increase top-down pressure of the predator (i.e., increasing the ratio of Kae to m) on its prey while keeping track of the local and nonlocal stability after a pulse perturbation (supplemental Figure S2) and a press perturbation (supplemental Figure S3) of resource productivity, K. As would be expected from existing omnivory theory (McCann and Hastings 1997, Gellner and McCann 2012), we found that our results (Figure 4) are robust under wide parameter spaces and that the stabilizing potential (i.e., faster return time, lower degree of overshoot, and smaller max–min) of adaptive omnivory is greatest as the system gets more top heavy (i.e., increasing K, a, or e or decreasing m; Figures S2 and S3). We point out that the above results rely on the underlying assumption that we employ weak to moderate average omnivory strengths, which tend to occur in nature and are known to be stabilizing (McCann and Hastings 1997, Emmerson and Yearsley 2004, Gellner and McCann 2012). Choosing strong mean omnivory strengths remain destabilizing even within this dynamic framework. As an example, a pulse in K that drives strong bottom-up responses in R can exaggerate increases in top predator densities when omnivorous interaction strengths are too strong, allowing for the suppression of C to local extinction. Similarly, press perturbation increases in K can make the strong P × R interaction oscillatory and less stable.

In summary, we have shown that omnivory increases through two qualitatively distinct mechanisms (bottom-heavy and top-heavy changes in density) and two differential behavioral responses of the predator (passive and active). Omnivory within this dynamic context tends to play a significant stabilizing role in the face of environmental noise, making it another potential adaptive food web structure like the generalist module. A key mechanism is the asynchronous responses of the consumer and resource (i.e., a portfolio effect) as the omnivorous predator changes in density and averages energy uptake over these waves. We now turn to empirical work to discuss this dynamic omnivory framework, a framework that is intentionally used to intersect with emerging empirical omnivory results.

Emerging empirical examples of dynamic omnivory

Since evidence of widespread omnivorous interactions became apparent, omnivory has been reasonably well studied in empirical food webs (Thompson et al. 2007). Recently, emphasis on changing conditions has increasingly documented varying levels of omnivory across space and time (Kratina et al. 2012). However, dynamic omnivory remains underappreciated in empirical food webs due to the historical difficulty in quantifying omnivorous interactions and the lack of a guiding theoretical framework. Detecting dynamic omnivory in real food webs requires measurement of an omnivores diet in response to changes in relative densities of trophically distinct prey under varying environmental conditions through space and/or time. Although this requires large amounts of data, advancements in empirical techniques, such as stable isotope analysis and telemetry, combined with more historical approaches such as stomach content analysis, have enhanced ecologist's ability to measure such responses in omnivorous behavior under changing conditions (see Table 2 for examples). In the present article, guided by our dynamic omnivory framework, we draw on existing data to propose examples of dynamic omnivory in empirical webs and demonstrate the apparent ubiquity of dynamic omnivory across ecosystem types, trophic levels, and spatial or temporal scales (Table 2). By outlining our framework in empirical food webs, we hope to motivate future research to undertake the nontrivial task of collecting such high-resolution data necessary to quantify dynamic omnivory in real systems.

Table 2.

Empirical examples of dynamic omnivory.

Ecosystem class	Ecosystem	Omnivory mechanism	Type of omnivore	Changing condition	Taxa	Metric of changing omnivory	Reference
Freshwater	Lake	Bottom heavy	Passive	Temporal changes in fish trophic position and diet composition in response to seasonal changes in resource availability	Silver carp (Hypophthalmichythys molitrix) and bighead carp (Hypophthalmichythys nobilis); zooplankton; phytoplankton	Stable isotope and stomach content analysis	Yu et al. 2019
	Stream	Bottom heavy	Passive and active	Temporal changes in amphipod and caddisfly trophic position in response to seasonal changes in aquatic and terrestrial resource availability	Amphipod (Gammarus pulex) and caddisfly larvae (Hydropsyche spp.); benthic macroinvertebrates; algae or detritus	Stable isotope analysis	Hellmann et al. 2013
	Lake	Bottom heavy	Active	Temporal changes in dolly varden diet composition in response to seasonal pulse in salmon eggs during salmon spawning migration	Dolly varden (Salvelinus malma); sockeye salmon eggs (Oncorhynchus nerka); macroinvertebrates	Stomach Content Analysis and Physiological changes in gut size	Armstrong and Bond 2013
	River	Bottom heavy	Active	Temporal changes rudd trophic position in response to seasonal changes in resource availability driven by temperature	Rudd (Scardinius erythrophthalamus); emerald shiner (Notropis atherinoides); macrophyte (Stuckenia pectinata)	Stable Isotope analysis	Guinan et al. 2015
	Lake	Bottom heavy	Active	Spatial and temporal changes in cisco diet composition in response to seasonal and spatial changes in resource availability	Cisco (Coregnous artedi); round goby (Neogobius melanostomus) and alewife (Alosa pseudoharengus); Bythotrephes longimanus	Stomach content analysis	Breaker et al. 2020
	Lake	Bottom heavy	Undetermined	Temporal changes in fish diet composition in response to seasonal changes in resource availability	Roach (Rutilus rutilus); macroinvertebrates; algae or detritus	Stomach content analysis	Persson 1983
	Floodplain	Bottom heavy	Undetermined	Temporal changes in omnivorous fish diet and trophic position in response to seasonal changes in resource availability driven by seasonal flood pulse	Omnivorous fish species (e.g., Channa spp.); prey fish; invertebrates; plant material	Stomach content and stable isotope analysis (field collection and literature synthesis)	McMeans et al. 2019
	Stream	Bottom heavy	Undetermined	Spatial change in macroinvertebrate omnivores trophic position driven by gradient in agricultural eutrophication along stream	Stream secondary consumers; stream primary consumers; stream primary producers	Stable isotope analysis	van der Lee et al. 2021
	Stream	Bottom heavy	Undetermined	Temporal changes in fish diet composition in response to seasonal changes in resource availability and temperature	Omnivorous fish species (Bryconamericus iheringii); fish and aquatic and terrestrial invertebrates; algae, plants, and detritus	Stomach content analysis	González-Bergonzoni et al. 2016
	Stream	Bottom heavy and top heavy	Undetermined	Spatial heterogeneity in trophic position of creek chub in response to changing resource availability along an agricultural land-use gradient	Creek chub (Semotilus atromaculatus); benthic invertebrates (e.g., Ephemeroptera spp.); algae	Stable isotope analysis	Champagne 2020
	Lake	Top heavy	Active	Spatial change in lake trout trophic position in response to changing prey availability driven by increasing lake size	Lake trout (Salvelinus namaycush); cisco (Coregonus artedi); zooplankton	Stable isotope and biomass analysis	Tunney et al. 2012
Ecosystem class	Ecosystem	Omnivory mechanism	Type of omnivore	Changing condition	Taxa	Metric of changing omnivory	Reference
	Mesocosm experiment	Top heavy	Undetermined	Spatial and temporal heterogeneity in prey biomass density patterns in response to differential omnivorous fish densities	Bighead carp (Aristichthys nobilis); invertebrates (Leptodora richardi); zooplankton (e.g., Daphnia); phytoplankton	Biomass density analysis (mesocosm experiment)	Zhao et al. 2016
Marine	Strait	Bottom heavy	Passive	Temporal changes in diet, stable isotope and fatty acid composition of jellyfish in response to seasonal changes in resource availability	Scyphozoan jellyfish (Pelagia noctilica); fish larvae or eggs; zooplankton; phytoplankton	Stomach content, stable isotope, and fatty acid analysis	Milisenda et al. 2018
	Arctic	Bottom heavy	Undetermined	Temporal change in benthic omnivore diet in response to seasonal changes in resource availability	Amphipod spp.; zooplankton; phytoplankton or algae	Fatty acid analysis	Werner and Auel 2005
	Fjord	Bottom heavy and top heavy	Passive and active	Spatial heterogeneity in lobster trophic position along resource productivity and availability gradient driven by kelp bed habitat density	Red rock lobster (Jasus edwardsii); mussels (e.g., Mytilus edulis galloprovincialis); kelp (Ecklonia radiata)	Field density surveys or stable isotope analysis	Jack and Wing 2011
	Mesocosm experiment	Top heavy	Undetermined	Spatial heterogeneity in shrimp diet and biomass dynamics of algae and amphipod in response to a shrimp presence/absence and resource availability gradient mesocosm experiment	Shrimp (Palaemon elegans); amphipod (Gammarus spp.); ephemperal macroalgae	Biomass density analysis (mesocosm experiment)	Eriksson et al. 2011
	Intertidal zone	Bottom heavy	Active	Spatial changes in crab diet composition and trophic position in response to spatial heterogeneity in resource availability along beach width gradient	Ghost crabs (Ocypode quadrata); mole crabs (Emerita talpoida) and coquina clams (Donax variablis), and amphipods (Talorchestia sp.); macrophyte wrack	Stable isotope analysis	Tewfik et al. 2016
	Estuary	Bottom heavy	Passive	Temporal changes in copepod diet composition in response to seasonal changes in resource biomass distributions	Copepod (Centropages hamatus and Labidocera aestiva); nauplii (Acartia tonsa and Acartia hudsonica); phytoplankton	Grazing rate and stable isotope analysis	Conley and Turner 1985
Terrestrial	Coastal river watershed	Bottom heavy	Active	Temporal changes in brown bear diet composition and habitat use in response to seasonal changes in resource availability driven by resource phenology patterns	Kodiak brown bears (Ursus arctos middendorffi); sockeye salmon (Oncorhynchus nerka); red elderberry (Sambucus racemosa)	Habitat use (aerial surveys, telemetry, cameras) and scat analyses	Deacy et al. 2017
	Mix of bogs, heaths, barrens, and coniferous and mixed forest	Bottom heavy	Active	Temporal changes in black bear diet composition and habitat use in response to spatial heterogeneity and seasonal changes in resource availability driven by calving season	American black bear (Ursus americanus); caribou calves (Rangifer tarandus); ants (family Formicidae); vegetation	Habitat use (telemetry) and scat analyses	Rayl et al. 2018
	Arctic	Bottom heavy	Undetermined	Spatial and temporal (seasonal) changes in polar bear diet composition in response to seasonal changes in resource availability driven by ice on or off	Polar bear (Ursus maritmus); animals (e.g., seals, seabirds, rodents); vegetation or algae	Scat analysis	Gormezano and Rockwell 2013
	Boreal forest	Top heavy	Passive	Spatial heterogeneity in diet composition and habitat use in response to spatial heterogeneity in resource distribution during calving season	American black bear (Ursus americanus); caribou calves (Rangifer tarandus caribou) and moose calves (Alces alces); vegetation	Movement and habitat selection analyses (telemetry)	Bastille-Rousseau et al. 2011

Passive and active omnivores in empirical webs

As was outlined in the theoretical framework, passive and active omnivores bracket a continuum of possible functional responses to changing prey densities. In nature, determining the endpoints of the gradient in passive and active omnivores is difficult because it requires rigorous data on resource densities and the response of consumer preference to changing resource densities, which tend to be rare in empirical food web data (but see Kalinkat et al. 2011). Recall from the theoretical results above that both passive and active omnivores exhibit stabilizing responses under changing conditions.

Passive omnivores are characterized by a density-independent preference that passively forage on their trophically distinct prey sources. Filter feeders are known to exhibit linear functional responses driven by fixed preference (Jeschke et al. 2004), and so omnivorous filter feeders may be perfect examples of passive omnivores. For example, bighead carp (Hypophthalmichthys nobilis) are mobile, filter-feeding fish whose diet and trophic position varies seasonally in response to changing relative densities of zooplankton and phytoplankton (e.g., higher trophic position under high zooplankton densities in spring and autumn; Figure 5a; Yu et al. 2019). This seasonal shift that follows relative density patterns with a fixed preference would classify these big-head carp as passive omnivores. However, their potential ability to spatially track high abundances of their preferred prey indicates they may exhibit some active behaviors (Yu et al. 2019).

Figure 5.

Empirical examples of dynamic omnivory along passive–active (a, b) and bottom-heavy–top-heavy continuums (c–e). (a) Bighead carp demonstrate passive omnivory because they consume trophically distinct prey sources relative to their density through time with a fixed preference. (b) American black bears demonstrate active omnivory because they shift preference and actively forage on caribou calves during caribou calving season. (c) Seasonal changes in relative zooplankton and phytoplankton availability drive bottom-heavy shifts in degree of omnivory in amphipods. (d) Agricultural land-use change increases nutrient loading in streams and drives bottom-heavy increases in degree of omnivory in creek chub, increasing their biomass and driving top-heavy omnivory. (e) Increasing access to littoral zone (productivity gradient) increases the biomass of lake trout and drives top-heavy increases in their degree of omnivory.

Active omnivores, on the other hand, can readily shift their density-dependent preference across their trophically distinct prey sources to maximize energy intake. For example, American black bears were shown to actively alter their foraging behavior and move across the landscape to target caribou calving grounds at certain periods of the year, despite other food sources still being readily available (Rayl et al. 2018). At other times of the year, when caribou calves are not as available, the bears appear to feed more passively on plants and ants in relation to those resources’ density (Rayl et al. 2018). Within our dynamic omnivory framework, we would consider, these American black bears primarily active omnivores while exhibiting some passive behaviors. We can characterize organisms by the dominant omnivorous behavior by empirically examining how omnivores respond to varying prey densities across space and time (i.e., how omnivore diets and behaviors change across spatial and temporal variation in resource densities; Table 2). However, it is important to remember that these behaviors exist along a continuum and that many organisms will fall somewhere in between and can exhibit both active and passive behaviors (Kalinkat et al. 2011).

Bottom-heavy and top-heavy omnivory in empirical webs

As was discussed above, any mechanism that inflates resource relative to consumer densities should elicit omnivory. In our dynamic omnivory framework, such inflated R: C can arise by two qualitatively distinct mechanisms, bottom-heavy and top-heavy omnivory. Both types of mechanisms appear to operate in empirical food webs on the basis of evidence from existing literature (Table 2). For example, seasonal changes that produce pulses of nutrients can increase resource densities and alter R: C and drive changes in the predators’ degree of omnivory. As was shown in the theoretical framework this change in R: C and subsequent change in omnivory is driven by bottom-heavy biomass distribution. We see evidence of this bottom-heavy driven omnivory in Arctic marine food webs where dramatic increases in light during open-water months lead to pulses in productivity that drive higher availability of lower trophic level resources (i.e., phytoplankton). In response, omnivorous amphipod species can switch from consuming higher trophic level zooplankton under winter ice cover toward consumption of lower trophic level phytoplankton during the open water season (Figure 5c; Werner and Auel 2005, McMeans et al. 2015). In this example, a purely bottom-heavy mechanism, akin to a pulse perturbation in our theoretical system (Figure 3a–3c), appears to be driving the dynamic omnivory response to seasonal changes in resource density.

Alternatively, other conditions can increase the top heaviness of food webs through time or space and drive top-heavy dynamic omnivory as our theory suggests. For example, lake trout are an omnivorous top predator that feed on both fish and invertebrates in nearshore and offshore zones of a lake (vander Zanden et al. 1999). In lakes in which access to highly productive nearshore prey is high (e.g., small lakes), energy flow to lake trout increases, increasing the top heaviness of the food web (high nearshore access means high top heaviness; vander Zanden et al. 1999). High densities of lake trout are likely to suppress their fish prey, making omnivorous foraging on lower trophic level zooplankton beneficial (Tunney et al. 2012). Under these conditions, top-heavy omnivory is therefore expected to dominate (Figure 5e). Higher lake trout density and increased omnivory in lakes with permanently higher nearshore access is consistent with predictions from our theory that press perturbations can lead to top-heavy food webs that then fuel omnivorous responses (Figure 3d–3f).

So far, we have considered empirical examples of bottom-heavy and top-heavy omnivory in isolation. Our theory shows, however, that bottom-heavy and top-heavy omnivory can also be tied together (a bottom-up pulse leads first to bottom-heavy and then to top-heavy omnivory; Figure 3). We can see this manifest in the real world as real systems also undergo changes in top-down and bottom-up dominated periods of omnivory. For example, in temperate agricultural stream systems, there are strong seasonal changes in resource densities driving bottom-heavy omnivory (Hellmann et al. 2013), as well as strong changes in top heaviness of webs through space (driven by varying nutrient level inputs) that exhibit top-heavy mechanisms of omnivory (Figure 5d; van der Lee et al. 2021, Champagne 2020). Although we have outlined only a few specific examples in the present article, Table 2 presents a catalogue of empirical examples across ecosystem types, trophic levels, and spatial or temporal scale to demonstrate the ubiquity of dynamic omnivory in empirical food webs. We note that existing empirical examples of dynamic omnivory seem to dominate in aquatic ecosystems; however, because there is widespread evidence of omnivory in terrestrial ecosystems (Thompson et al. 2007), the lack of examples in terrestrial systems may be a factor of less work focused on examining terrestrial omnivore responses to changing conditions.

Conclusions

In the present article, we have examined the role of omnivory from a dynamic perspective. By assuming two plausible behavioral omnivory responses (i.e., passive and active), we use theory to predict temporal changes in the degree of omnivory after a perturbation and the local and nonlocal stability implications of these changes. We found that dynamic omnivory responses, whether passive or active, often act as potent stabilizers in complex ecosystems in the face of environmental variation. Importantly, active omnivores have a stronger stabilizing potential than passive omnivores do, because their density-dependent preference allows for rapid prey switching, which is known to drive stabilizing sigmoidal functional responses akin to type III (McCann 2000, Post et al. 2000). Furthermore, similar to arguments that generalist couplers can stabilize lower trophic level variation by integrating over two asynchronous habitat pathways (McCann and Rooney 2009), we showed that omnivory responses to perturbations can naturally generate asynchronous consumer and resource dynamics that the omnivore can integrate over (in a simplified sense, the omnivore harnesses a consumptive portfolio effect). Consistent with classic understanding of trophic dynamics across gradients in productivity (Oksanen et al. 1981), we argue that changes in the degree of omnivory—and, therefore, the stabilizing responses—are likely driven by a combination of bottom-up and top-down cascading changes in resource and consumer densities, both of which predictably alter the ratio of resources and consumers.

Although we employed a single chain tritrophic model, our results are consistent with mechanisms proposed in other omnivory models (McLeod and Leroux 2021). Specifically, multichain omnivory theory has found that top-heavy omnivory can increase across a gradient in productivity (K) or changes in accessibility in attack rate (a), both of which were argued to increase the R: C ratio and, therefore, omnivory (Tunney et al. 2012, Ward and McCann 2017). Importantly, this multichain omnivory appears to play a role in building up biomass in the top predator in empirical studies (e.g., lake trout; Tunney et al. 2012) leading to reductions in their preferred prey (cisco) that drives increased omnivory.

Most traditional empirical omnivory approaches have been static and focused on comparing the average strength of omnivory across species and ecosystems (Kratina et al. 2012), resulting in theoretical and empirical arguments that omnivory is now believed to be widespread (Thompson et al. 2007) but often weak (Emmerson and Yearsley 2004, Gellner and McCann 2016). Nonetheless, our empirical understanding of how the strength of omnivory responds to changing conditions is only beginning to emerge. In the present article, our theoretical dynamic omnivory framework provides us with a novel tool to empirically investigate omnivory responses of real food webs (Table 2). Specifically, we show that temporal shifts in resources across seasons have predictable implications for changes in omnivory that match theory—strong bottom-up shifts in production alter the degree of omnivory seasonally, for example. Furthermore, because our theory highlights short-term responses (that are often bottom-up driven) and long-term responses (that occur after top-down responses have equilibrated), we are able to determine spatial variation in omnivory of a species across a gradient in changing conditions (e.g., ecosystem size) that reflect the “equilibrated” omnivory responses of the same species. These empirical results again resonate with theory showing, for example, that in small strongly interacting ecosystems, top-heavy omnivory can generate significantly increased omnivory responses relative to larger systems assumed to have weaker interactions (Emmerson and Yearsley 2004, Gellner and McCann 2016).

The theoretical framework we have outlined is a starting point to understand the empirical responses of organisms and food webs to changing conditions. Given that we are in a world replete with global change driving novel temporal and spatial perturbations, the development of a theory of dynamic responses in key food web modules promises to allow us to further understand the resilience implications of changing environmental conditions (Neubert and Caswell 1997, Hastings 2004, Hastings et al. 2018). Our work adds the omnivory module to the generalist module as another fundamental food web structure that can mute variation in space and time. In a sense, the behavioral responses of predation in both cases act as adaptive capacity capable of giving resilience to diverse webs in a noisy world (McMeans et al. 2016). Further work identifying other food web structures (both low and high diversity structures) can add to this critical developing framework for adaptive food webs. Our work shows the importance of harnessing the variability of ecosystems by understanding how fundamental food web structures change in space and time in response to variation.

Furthermore, we point out that dynamic responses are empirically measurable, so they importantly facilitate the interaction of theory and empirical research—an area that has hindered the rapid development of food web research (Kratina et al. 2012). By outlining the framework and stability implications of dynamic omnivory, we hope to motivate future research to consider food web structure and behavior through this dynamic lens and to expand data collection to robustly examine these mechanisms in empirical food webs. As food webs are rewiring under global change (Bartley et al. 2019), our framework is a significant step toward a better understanding of the future stability of tomorrow's ecosystems.

Acknowledgments

This project was funded by the University of Guelph's Canada First Research Excellence Fund project “Food from Thought” (grant no. 499075), awarded to KSM and a Discovery Grant from the National Science and Engineering Research Council of Canada (grant no. 400353) awarded to KSM. Thank you to Caden McCann for the animal artwork used in the figures. The code to reproduce this study is available as an archived compendium (https://doi.org/10.5281/zenodo.5776233) and the corresponding development repository is available at the following URL: https://github.com/McCannLab/Labmnivory.

Author Biographical

Marie Gutgesell ([email protected]), Kevin McCann, Gabriel Gellner, Kevin Cazelles, Matthew Guzzo, Christopher Greyson-Gaito, Carling Bieg, Connor Warne, Charlotte Ward, Reilly O'Connor, Alexa Scott, Emily Champagne, and Brandon Graham are affiliated with the University of Guelph, in Guelph, Ontario, Canada. Bailey McMeans is a professor at the University of Toronto Mississauga, in Mississauga, Ontario, Canada.