Resource scarcity is a common stress in nature and has a major impact on microbial physiology. This review highlights microbial acclimations to resource scarcity, focusing on resource investment strategies for chemoheterotrophs from the molecular level to the pathway level. Competitive resource allocation strategies often lead to a phenotype known as overflow metabolism; the resulting overflow byproducts can stabilize cooperative interactions in microbial communities and can lead to cross-feeding consortia. These consortia can exhibit emergent properties such as enhanced resource usage and biomass productivity. The literature distilled here draws parallels between in silico and laboratory studies and ties them together with ecological theories to better understand microbial stress responses and mutualistic consortia functioning.
Resources and cellular function are inextricably linked
Nutrient scarcity is a major selective pressure for most life including microorganisms. Natural selection has resulted in microorganisms which utilize limiting resources in competitive manners. Microbial acclimation to nutrient scarcity is a fundamental biological process which influences mass and energy flows through a continuum of biological scales, including individual microbial cells, microbial consortia, and ecosystems [1–4].
Scarcity of resources, such as anabolic and catabolic nutrients, can influence the elemental and macromolecular composition of microorganisms [5–9]. Organisms grown in nitrogen- or phosphorous-limited environments contain relatively smaller amounts of the respective elements in their biomass due to shifts in macromolecular composition such as protein to RNA ratios [5,6]. Resource scarcity has influenced the evolution of macromolecular sequences including DNA, RNA, and protein [10,11]. For example, codon usage in the marine bacterium Pelagibacter is biased toward low nitrogen-containing nucleotides, likely because of evolution under extreme nitrogen scarcity . Microbial proteins involved in the transport of carbon, nitrogen, or sulfur sources typically contain less of the respective element than an average protein, thereby reducing the quandary of investing a limiting resource into enzymes for its acquisition [5,13,14]. Competitive use of scarce resources extends to relatively minor cellular metabolite pools . For example, production of the iron-containing cofactor ferredoxin is reduced when iron availability is scarce while production of the iron-free cofactor flavodoxin increases . The competitive investment of scarce resources also influences the aggregate investment into both metabolite pools and enzymes used to drive cellular function; it has been proposed that the investment is minimized [17–19].
Resource scarcity influences regulation and function of metabolic networks. The typical bacterial proteome accounts for 50–70% of the total cellular mass with metabolic enzymes accounting for 15–20% of this macromolecular pool [20,21]. Natural selection has enriched populations for microorganisms that allocate scarce resources to functions that affect outcomes of competition. These include highly efficient mass and energy utilization strategies that are appropriate for nutritional availabilities [22–26]. Competitive cellular functions, such as the complete oxidation of a substrate using the TCA cycle as opposed to partial oxidation through glycolysis, require a larger resource investment into enzymes and metabolite pools. The availability of appropriate resources therefore constrains whether a certain cellular function can be realized. The tradeoff between resource availability and biological functioning has been conceptualized by ecologists using the ‘Darwinian Demon’ thought experiment . A Darwinian Demon is a ‘superspecies’ that is unconstrained by resource limitations and can optimize all aspects of fitness simultaneously. If such an organism was feasible, it would outcompete all other organisms, leading to a monoculture. However, manifestation of the Darwinian Demon is precluded by the requirement that resources allocated to one function cannot typically be allocated elsewhere. This limits how many discrete processes can be optimized at a given time.
Acclimation to nutrient scarcity can dramatically shift phenotype
Acquisition of mass and energy from the environment is a prerequisite for life. Imbalances or perturbations in the acquisition of mass and energy are common occurrences due to feast and famine events associated, for example, with diel or seasonal cycles. Microorganisms have evolved competitive strategies to maximize fitness during these events including strategies that excrete large quantities of organic byproducts. These phenotypes are widely observed in prokaryotes and eukaryotes and have been called overflow metabolism, Crabtree effect, or Warburg effect dependent on observation in bacteria, fungi, or mammalian cells (reviews and further data can be found in refs [23,24,28–32]). A recent set of chemostat experiments examined acclimations of Escherichia coli cultures grown under three separate nutrient limitations: (1) glucose, the energy source, and an essential macronutrient, (2) ammonium, an essential macronutrient, and (3) iron, an essential micronutrient [1,33]. Each nutrient limitation was examined at four distinct levels of stress using chemostats operated at different dilution rates  (Figure 1). Chemostat nutrient stress is controlled via the bioreactor dilution rate, which is equal to the culture-specific growth rate, and is not determined via the concentration of the limiting substrate in the growth medium .
Chemostat operation theory and observation of overflow metabolism under multiple nutrient-limited conditions.
The E. coli cultures acclimated to increasing levels of nutrient scarcity by secreting increasingly larger fractions of catabolized glucose carbon as organic byproducts. Organic byproducts were observed even though the cultures were grown under oxic conditions, consistent with an overflow metabolism. The type and distribution of the organic byproducts changed as a function of nutrient limitation and the degree of nutrient scarcity (Figure 1). For example, iron-limited E. coli cultures grown at extreme nutrient scarcity secreted more than 70% of the catabolized glucose carbon as organic byproducts, primarily lactate; ammonium-limited E. coli chemostat cultures secreted ∼40% of the catabolized glucose carbon as organic byproducts, primarily acetate, whereas glucose-limited cultures secreted up to 4% of the glucose as organic byproducts, primarily formate .
The elemental composition of the culture biomass also varied with nutrient limitation and degree of stress, consistent with the tenets of ecological stoichiometry [6,35]. The different nutrient scarcities resulted in different biomass degrees of reduction (oxidizible electrons per carbon mole, also a metric for biomass energy density) as well as different ratios of C : N .
Overflow metabolisms have also been observed at the high end of the specific growth rate spectrum [29,31,36]. The fast growing E. coli cultures experienced nutrient stress based on a combination of metabolite transport kinetics, growth rate-associated resource allocation to cellular machinery like ribosomes, and competing resource investment into fermentative or respiratory pathways; these mechanisms are in contrast with the absolute nutrient scarcity studied in chemostat cultures (Figure 1). However, the studies on both the high and the low end of the growth rate spectrum arrived at similar interpretations, namely that overflow metabolism represents a competitive investment strategy for select environments [29,31,33,36].
In silico analysis can identify resource investment tradeoffs in metabolic pathways
Metabolic networks, especially those observed in chemoheterotrophic generalists like E. coli, are capable of remarkable phenotypic plasticity enabled by robust biochemical networks [23,25,37–39]. Systems biology tools are needed to quantify the behavior of all but the simplest metabolic networks. Stoichiometric modeling approaches, including the related methods elementary flux mode analysis (EFMA) and flux balance analysis (FBA), can be used to predict and interpret tradeoffs in metabolic pathway investment strategies . The two stoichiometric modeling approaches use the same formalism to create metabolic models using sets of mass balanced reactions and exploit similar or even identical properties of the metabolic models to predict and explain phenotype (Klamt et al. under review). Stoichiometric modeling approaches utilize data from omics datasets to build in silico representations of metabolism. These approaches utilize molecular-level data, but require minimal assumptions regarding enzyme kinetic parameters. This is a major strength because literature-reported kinetic parameters for the same enzyme can vary over orders of magnitude, complicating their inclusion into metabolic models and the interpretation of results [41–44].
Several stoichiometric modeling studies have examined resource investment into metabolic pathways using direct accounting of anabolic resources. Elementary flux modes (EFMs) are the simplest functioning metabolic protein investment schemes, making them ideal quanta for resource investment analyses . Carlson [23,24] used EFMs and two different resource investment correlations to quantify relationships between resource investment into metabolic pathways and the efficiency of the pathways to convert substrate into biomass (Figure 2). The results were used to model fluxomics data collected from nutrient-limited E. coli cultures  and to interpret chemostat physiological data [1,33]. Another stoichiometric modeling approach developed in parallel used FBA to identify cellular phenotypes that maximized growth rate while constraining enzyme-occupied volume, based on sequence length as a proxy for molecular crowding . These two resource investment analyses are conceptually and mathematically similar and are thought to identify similar metabolic phenotypes . Additionally, cellular membrane space constraints have been used to explicitly account for resource investment constraints on phenotype . Other stoichiometric modeling studies have investigated indirect proxies for resource investment into metabolic enzymes by, for instance, minimizing total phenotype flux or minimizing the number of utilized enzymatic steps [48–52].
Process to perform resource allocation analysis of metabolic pathways.
The analysis of resource investment into metabolic pathways using stoichiometric modeling approaches has also expanded into larger scale cellular models that include not only the metabolism but also protein translation efficiency, cellular volume constraints, and macromolecular composition of the cell using resource balance analysis [53,54]. The more detailed analysis provides a multiscale basis for predicting and analyzing cellular resource investment scenarios, but also requires substantially more a priori data .
Cost–benefit analysis of metabolic pathways predicts overflow metabolisms
The optimal relationship between resource investment and the efficiency of an enzyme-catalyzed pathway can change with resource availability, resulting in optimality surfaces (Figure 2). These surfaces have been termed cost–benefit curves, Pareto curves, efficiency frontier curves, or in some cases resource ratio curves [17,56,57,91]. The tradeoff curves can be viewed from the lens of the Darwinian Demon and the observation that allocation of resources to one function typically precludes their allocation to another; in other words, one function can only be increased by decreasing another. The analysis has analogs to industrial design strategies with the two extrema representing a high investment/low operating cost process and a low investment/high operating cost process (Figure 2) .
Many computational biology studies including kinetics-based studies [25,28] predicted overflow metabolisms when resources required for enzyme assembly were constrained. These studies consistently noted that overflow metabolisms occur along optimal tradeoff surfaces and therefore do not reflect a wasteful metabolism as the overflow metabolism has sometimes been described. These studies also further highlight that overflow metabolism can be driven by optimization for either growth rate or growth yield.
The overflow metabolism associated with oxygen-limited growth is both well studied and widely acknowledged. Scarcity of a (catabolic) resource, such as a terminal electron acceptor (oxygen), limits the bioavailability of substrate energy, constraining metabolism to substrate-level phosphorylation with metabolic intermediates like pyruvate acting as electron acceptors for the production and secretion of organic acids [58–60]. The E. coli overflow metabolisms observed during carbon-, ammonium-, and iron-limited growth can be explained with a similar resource-based argument, though the mechanism is slightly different. The scarcity of a (anabolic) resource limits the bioavailability of substrate energy, constraining metabolism to substrate-level phosphorylation which extracts energy from a molecule with minimized investment into enzymes; the resulting metabolic intermediates, which require large resource investment to oxidize, are secreted.
Overflow metabolisms are commonly observed phenomena [23,24,28–32]. A few key questions regarding overflow metabolism include (1) do all observations of overflow metabolisms reflect effective resource usage, (2) how do unstructured planktonic growth vs. structured biofilm environments alter which metabolic strategies are competitive, (3) what role does metabolite toxicity and environmental modification play in driving byproduct secretion, and (4) to what extent is the process of secreting partially oxidized byproducts simply symptomatic of laboratory conditions including media selection and therefore irrelevant to nature?
Nutrient stress acclimation strategies can be quantified using opportunity costs
Resource investment vs. function tradeoff curves provide detailed predictions of metabolic acclimations to gradients of resource scarcity . Many of the predicted E. coli acclimations discussed above have been experimentally verified during steady-state growth in chemostats and are highlighted in this section. ATP opportunity costs, based on amino acid peptide-bond formation, have also been calculated as a quantitative reference. This opportunity cost is a conservative estimate, because many amino acids are also viable cellular energy sources .
Full TCA cycle vs. glyoxylate shunt (GS): The GS has historically been associated only with the catabolism of two-carbon organic compounds like acetate [62,63]. Contrary to this assertion, the GS has been observed during glucose-limited growth experiments [1,33,63,64], as predicted from resource tradeoff analysis . The GS uses two enzymes (isocitrate lyase and malate synthase) and bypasses three TCA cycle enzyme complexes (isocitrate dehydrogenase, α-ketoglutarate dehydrogenase, and succinyl-CoA synthetase), saving an investment of ∼22 000 amino acids' worth of carbon, nitrogen, and sulfur as well as ∼66 000 ATP equivalents for peptide-bond formation.
Oxic use of pyruvate formate lyase (Pfl) vs. pyruvate dehydrogenase complex (PDHc): Pfl converts pyruvate into acetyl-CoA and formate, consists of two protein subunits, and requires a total of 1520 amino acids. In contrast, the pyruvate dehydrogenase complex converts pyruvate into acetyl-CoA, NADH, and CO2, consists of 60 protein subunits, and requires 42 096 amino acids to synthesize or ∼25 times more anabolic resource and cellular volume than Pfl and ∼120 000 more ATP equivalents for peptide bonds. Historically, Pfl has been associated only with anoxic growth [65,66], but active Pfl has been reported during micro-aerophilic growth. In addition, formate secretion has been observed during both iron-limited and glucose-limited chemostat growth, and formate has been implicated in long-term aerobic glucose-limited E. coli acclimation studies [33,58,67,68]. These trends are consistent with in silico resource allocation predictions . The oxygen-sensitive portion of Pfl can be replaced by the relatively small YfiD repair protein , potentially making it possible to maintain Pfl functionality by continuously replacing the oxygen-damaged protein. YfiD protein consists of 127 amino acids and can be synthesized de novo more than 300 times while still requiring fewer resources than a single functional PDHc.
Entner–Doudoroff (ED) vs. Embden–Meyerhof–Parnas (EMP) glycolysis pathway: The ED pathway has historically been associated only with pentose sugar catabolism . The ED pathway requires ∼1500 fewer amino acids to assemble relative to the classic EMP pathway although it also extracts one fewer ATP from substrate-level phosphorylation in the conversion of glucose to pyruvate . Experimental evidence and genomic analyses strongly suggest that the ED pathway is utilized during nutrient scarcity for the catabolism of glucose, again consistent with in silico predictions [23,70,72].
Shifts in overflow metabolism byproducts as a function of nutrient scarcity: Tradeoff surface analysis predicts a series of byproducts that change with the degree of resource stress . Consistent with predictions, E. coli cultures grown under moderate levels of nutrient scarcity stress secreted acetate and formate in the presence of oxygen. The formate and acetate secretion is predicted to decrease at extreme levels of nutrient stress with the metabolism shifting toward lactate secretion. These predictions in aggregate are consistent with E. coli chemostat studies [1,29,33,64]. An aerobic metabolism that completely oxidizes glucose to CO2 requires an investment of ∼50 atoms of iron and produces ∼30 ATP per glucose; the minimal iron investment ATP strategy is a homolactic acid synthesis pathway that does not require any iron investment and produces two ATP per catabolized glucose.
The discussed opportunity costs do not explicitly consider enzyme parameters. The relationship between pathway flux, proteome, enzyme kinetic parameters, and enzyme stability has been the subject of many studies and has implications for predictive resource allocation models [73–75]. Metabolic control analysis (MCA) provides theory for developing an initial basis for predicting pathway relationships. MCA has established that flux through most metabolic pathways is a systemic property that is not typically subject to individual enzyme kinetic parameters [76–78]. MCA and numerous experimental studies have demonstrated the shortcomings of the ‘rate-limiting’ enzyme concept; pathway flux is instead a complex function of metabolite concentrations, enzyme concentrations, and regulation [76,79]. This basic tenet of MCA has been supported through many elegant studies which have shown that flux through metabolic pathways does not show a strong correlation with enzyme levels. For instance, the experimental overexpression of individual glycolysis enzymes (up to ∼1400% increase) including the ‘rate-limiting’ enzyme had no substantial effect on overall pathway flux . Instead, the pathways possessed mechanisms that allowed systemic shifts in metabolite levels and regulation to acclimate to the enzyme perturbations maintaining similar overall flux [73–76,79–81].
The resource opportunity costs discussed in this section are based on the powerful concept of an EFM. The core predictive capabilities of stoichiometric modeling approaches, including EFMA and FBA, are independent of kinetic parameters or absolute fluxes and instead identify ratios of reaction fluxes that meet user-defined criteria. The investment example in this section makes no assumptions about the absolute concentrations of enzymes in each EFM; only concentration ratios between participating enzymes are assumed. The minimum possible proteome associated with an EFM has a 1 : 1 ratio between every participating enzyme or enzyme complex. Attempting to scale individual enzyme concentrations based on in vitro measurements of enzyme parameters would neglect the effects of metabolite concentrations, regulation, and the findings from MCA which demonstrated that typically, there is no ‘rate-limiting’ enzyme. EFM enzyme concentrations could likely be perturbed randomly (± order of magnitude) around the 1 : 1 ratio without substantial effects on overall pathway flux, based on experimental and theoretical studies [73,79,80]. Additionally, enzyme stability would not influence the predicted results because EFM and FBA both assume steady states. Protein stability would require additional ATP maintenance energy to degrade and reassemble enzymes, but the amino acids and cofactors would remain inside the system boundaries.
Modern systems biology has a general trajectory toward incorporating an ever-greater number of assumptions based on experimental data such as transcript level or enzyme abundance to create ever more complex representations of metabolism. A few key questions in the field of metabolic modeling are (1) to what resolution and what metabolic cost has evolution selected cellular control of central metabolism, (2) what is the quantitative tradeoff between predictive capabilities of in silico models and the number of assumptions including flux, enzyme parameters, and omics data relationships, (3) what scientific questions benefit from adding more parameters or fewer parameters to a computational representation of biological processes, (4) what is the relationship between the number of assumed parameters/model complexity and the description of fundamental principles of biological organization, and finally (5) under what culturing conditions does predicted metabolic optimality exceed the resolution of natural selection?
Overflow metabolism provides a mechanism for consortia cross-feeding
Most naturally occurring microbes exist in multispecies communities where they co-operate or compete for resources. Cross-feeding is a common consortium interaction motif in which a microbe excretes a compound that is utilized by another individual . Several synthetic consortia have been constructed to study these interactions. The systems can be categorized broadly based on anabolic excretions like amino acids [83–86], catabolic excretions like organic acids [87,88], and mixed systems that involve both amino acids and organic acids [89,90]. The presented work focuses on catabolic exchanges based on the relationship to overflow metabolism.
Resource allocation tradeoffs can facilitate cross-feeding; stress acclimation of one trophic level can lead to overflow metabolisms, creating resource niches for organisms in a second trophic level . By consuming the byproduct, the secondary consumer population uses resources that would otherwise be neglected while removing chemicals that could represent both a thermodynamic constraint and the inhibitory byproduct . This positive feedback mechanism enables the consortia to more effectively deplete environmental resources, increase productivity, and can even allow metabolism of substrates that would not be feasible by a monoculture [88,92].
Consortia with increased efficiency have evolved and been engineered in E. coli [68,83,87,93]. One example is an evolved cross-feeding E. coli consortium that is reproducibly selected during long-term, glucose-limited chemostat experiments. Over the span of ∼100 generations, the monocultures spontaneously evolved into acetate cross-feeding consortia with ∼15% increase in biomass productivity and enhanced nutrient depletion properties when compared with the inoculum monoculture (reviewed in refs [68,93]). A synthetic consortium also demonstrated enhanced biomass productivity and resource depletion when two populations were engineered for division of labor; one population catabolized glucose and produced acetate and the second population consumed the acetate . The engineered consortium showed 20% improvement in biomass productivity during chemostat cultivation and 50–100% improvement in biomass productivity during biofilm cultivation when compared with wild-type E. coli monocultures. In both consortium examples, the basic schema entailed one population using a substrate-level phosphorylation-focused metabolism with heavy reliance on glycolysis and the second population using an oxidative phosphorylation-focused metabolism with a heavy reliance on the TCA cycle. The ‘metagenome’ of the evolved and synthetic consortia contained the same metabolic potential as wild-type E. coli; however, the enzymes either had different regulation patterns or were partitioned to different populations.
Division of labor permits reduced aggregate resource investment into enzymes and substrate improving consortia productivity
The acetate cross-feeding consortia presented in the preceding section have higher glycolysis fluxes in one population and higher TCA fluxes in the other population when grown in chemostats . This results in an additional benefit to division of labor and cross-feeding: a better functional return on resource investment into metabolism. Enzyme flux is driven by investment into both enzyme (E) and substrate (S) pools (Figure 3); substrate pools during steady-state growth represent a substantial resource investment. Free metabolite pools and enzymatic protein investment both account for ∼5% of E. coli dry mass [15,20]. An investment analysis to drive enzyme flux, quantified in terms of carbon moles (Cmoles), is shown in Figure 3. A target cellular function, quantified by an enzyme flux, can be achieved through a combination of investment into enzyme pools and substrate pools. The minimum aggregate enzyme and substrate carbon investment are highlighted in Figure 3. Relatively small fluxes are achieved with minimal aggregate investment using a ‘push’ metabolism where more resource is invested into the substrate pool than the enzyme pool (region labeled in Figure 3, top right panel). High fluxes are achieved with minimum aggregate investment through a ‘pull’ strategy where more resource is invested into the enzyme pool than the substrate pool. Increasing flux through a reaction does not require a linear increase in resource investment (Figure 3); instead, higher fluxes are relatively less resource intensive [17–19,94,95].
Optimal investment into enzyme and substrate pools leads to resource savings at high fluxes and contributes to consortia benefits.
Analysis of Figure 3 provides an interesting side story to examine recent reports that cellular metabolism can shift fluxes substantially while enzyme levels remain relatively unchanged [73–75]. Collectively, these data suggest that the cells are utilizing a flexible base metabolic proteome and a push mechanism, along with regulation, to drive function. This self-correcting organization would permit shifts in pathway fluxes via relatively inexpensive and fast shifts in metabolite pool and regulation as opposed to large restructuring of resources invested into the proteome [73,80].
The resource savings from operating enzymes at higher fluxes as highlighted in Figure 3 may improve the fitness of cross-feeding populations compared with monocultures [39,83]. This concept is illustrated on a hypothetical chemoheterotrophic example in Figure 3 (lower panel); it is assumed that the enzymes can be described by Michaelis–Menten-type kinetics. Here, a wild-type population comprised of two cells is the reference case; this population completely oxidizes glucose, and each cell has a relative overall flux of 1 (arbitrary units of flux, e.g. mole/cell/h) and a relative resource investment requirement of 1 (arbitrary units of investment, e.g. Cmoles per cell). Therefore, total system flux and investment are each 2 (arbitrary units). Realistic quantitative fluxes and investment values are dependent on enzyme parameters; here, they are simplified for the sake of argument. The wild-type population is compared with a second population of two cells that specializes in glycolysis and therefore accumulates inhibitory organic byproducts. The cells have higher-relative glycolysis fluxes (2 arbitrary units each) consistent with an overflow metabolism. The cells only use approximately half the central metabolism enzymes when compared with the wild-type cells and operate a high glycolysis flux which reduces the overall investment (1.85 arbitrary units total system). The accumulated organic byproduct inhibits the cells, preventing the investment savings from leading to higher productivity. A third system is a cross-feeding consortium with one cell specializing in glycolysis and a second cell specializing in the TCA cycle. The flux per cell would be increased (2 arbitrary units) with this configuration; the two-fold higher fluxes require less than a two-fold increase in resource investment to achieve (1.85 arbitrary units total system) (Figure 3). The reduced resource requirement is available for enhanced growth, providing an explanation for the enhanced biomass productivity seen in the naturally evolved and synthetically constructed consortia described in the previous section. These same arguments for functional return on investment are also likely relevant, to a degree, with the other mentioned synthetic consortia [83–86,89,90].
A few key questions regarding cross-feeding interactions include (1) what number of metabolite exchanges can optimize overall system productivity, (2) how does the number of exchanges correlate to properties of enzyme-catalyzed steps like resource investment and energy-producing steps as theoretically explored in , and (3) how do these predictions in time-invariant, unstructured environments like a steady-state chemostat compare to temporally dynamic structured environments like biofilms?
Biomass turnover as a mechanism of consortia mass and energy fluxes
Synthetic and artificial consortia have become popular loci of research with many studies examining cellular interactions based on excreted metabolites like organic acids, amino acids, or quorum sensing molecules [82,87,89,90]. An extreme case of metabolite exchange, that sets an upper limit, is the total transfer of cellular biomass from one population to another via the utilization of lysed cell material. This is a common, naturally occurring food web interaction and provides a contrast with overflow metabolism-based interactions . Ecosystem primary producers fix mass and energy, and the resulting biomass eventually serves as mass and energy sources for other trophic levels. The turnover of primary producer biomass occurs based on many mechanisms including predation, viral lysis, or senescence. Hunt et al.  have quantified the limits of mass and energy transfer from a primary producer trophic level to a secondary heterotrophic consumer level using both EFMA and FBA stoichiometric modeling approaches (Figure 4). The analysis considered a range of mass and energy exchanges including the selective catabolism of individual monomer pools, catabolism of entire macromolecular pools, and the catabolism of the aggregate primary producer biomass. The different potential substrate pools have different relative values as precursors for biomass or cellular energy production based on metabolite properties, including the degree of reduction and the entry point of metabolite(s) into the biochemical network. The bioavailability of mass and energy was also dependent on environmental conditions. Under oxygen sufficient conditions, ∼2.3 autotroph cells are required to produce one heterotroph cell; the relationship changes to ∼4 autotroph cells per heterotroph cell for oxygen scarce environments. This mechanism places major constraints on abundance relationships of primary producers and consumers and therefore has a very strong influence on the allocation of a limiting resource like nitrogen or oxygen.
Extreme metabolite exchange in food webs: mass and energy flow via biomass turnover.
Diffusion limitation can alter resource allocation in biofilm consortia
Most naturally occurring microorganisms grow as biofilms, attached to surfaces and often surrounded by extracellular polymers . Spatially structured environments such as biofilms often constrain growth due to rate imbalances between diffusion and enzyme-catalyzed reactions . The imbalances create concentration gradients leading to spatial chemical heterogeneity and spatially dependent phenotypes including overflow metabolisms . Nutrient gradients generated by microbial growth influence the strength of competition  and the feasibility of mutualism . These gradients, in turn, tend to drive segregation of competing genotypes, and mixing of cooperators , ultimately determining the composition and function of many microbial systems .
Stoichiometric modeling provides insights into the spatial distribution of metabolic strategies, and how nutrient gradients drive dynamics in the composition and function of microbial systems [89,98]. For example, a model identified how metabolic strategies of E. coli are segregated by the oxygen and carbon gradients that arise inside a colony . Models can also illuminate dynamics in multispecies systems [82,102–104]. Phalak et al.  investigated the stratification of Staphylococcus aureus and Pseudomonas aeruginosa in chronic wounds (Figure 5). Stoichiometric models suggest that this pattern is generated by differential growth, chemotaxis, and gradients in both oxygen and inhibitors. Modeling feedback between metabolic strategies and nutrient gradients has also been used to explain the productivity of fuel cells , the stable coexistence of bacteria in the human gut even when the species have very different growth rates , and even evolutionary dynamics in multispecies systems . In all cases, it is clear that the optimal metabolic strategy for a cell changes through space and time, and there is dynamic feedback between metabolism and nutrient gradients.
In silico analysis of biofilm consortia interactions including spatially resolved examination of metabolite inhibition.
Discrete agent-based (also called individual-based) modeling has also been developed to expand from population-averaged results to investigations of localized environments within biofilms and interactions between microbes . Agent-based modeling allows attribution of different physiological and metabolic properties to distinct microbes in the model, properties which are often responsive to the varying local environments within biofilms, and gives the ability to describe adaptive behavior . For example, agent-based models have been used to predict enhanced productivity of a biofilm community when inhibitory byproducts are consumed by a scavenger population  (Figure 5). Other models have predicted the optimal relationship between cell clustering and the transport rate of exchanged molecules . These models provide a counter perspective to the continuous modeling methodologies applied with the FBA-based approaches. Methods are also being developed to integrate stoichiometric and individual-based modeling techniques and have been used to study diversification of metabolic phenotypes in P. aeruginosa biofilms as well as human microbiome consortia .
A few key questions associated with reaction–diffusion-controlled microorganism growth include (1) whether it is possible to control these systems by applying theory informed, resource gradients at the appropriate system boundaries or using technologies such as three-dimensional printing of microorganisms in hydrogels or droplet-based microfluidics to construct spatial structures that possess desirable functions, (2) to what extent can metabolic interactions be inferred from snap-shot measurements of spatial biomass distributions, and (3) at what size scale do the interactions need to be modeled to provide accurate representation of system physiology?
Resource scarcity shapes biological systems at every level, from individual microorganisms to microbial consortia to entire ecosystems. Resource investment can be analyzed at various levels, from elemental and macromolecular composition to metabolic pathway utilization to partitioning between enzymes and substrates. Strategic investment of scarce resources is predicted to be necessary for fitness and is supported by experimental observations. Continuously advancing techniques for the investigation of cost–benefit tradeoffs (e.g. stoichiometric and individual-based modeling methods), combined with ecological concepts, are yielding more insights into microbial metabolisms and interactions.
The study of microbial acclimations and adaptations to resource scarcity will lead to improved understanding of emergent properties in consortia and will enable improved engineering and control of industrial, environmental, and medical consortia. Computational modeling and analysis tools that predict the impact of resource scarcity in homogeneous planktonic cultures and diffusion-limited biofilm consortia are now available. These tools enable the rational engineering of microbial species and their cross-feeding interactions using synthetic biology to drive resource investment toward desirable outcomes such as enhanced synthesis of valuable metabolites, better microbiome function, or in the case of unwanted populations, the collapse of consortia responsible for medical infections.
This work was supported by National Science Foundation [Awards DMS-1361240, CBET-4171195], National Institutes of Health [Award U01EB019416], and Department of Defense Army Research Office ARO [Award W911NF-16-1-0463].
The authors like to acknowledge Kristopher Hunt and Hans Bernstein for their assistance with figure graphics.
The Authors declare that there are no competing interests associated with the manuscript.