Abstract
Microbial decomposition of soil organic matter is a key component of the global carbon cycle. As Earth’s climate changes, the response of microbes and microbial enzymes to rising temperatures will, though emission of additional CO2, largely determine the soil carbon feedback to climate. However, while increasing attention focuses on physiological and ecological mechanisms of microbial responses, the role of evolutionary adaptation to warming has been little studied. To address this gap, we developed an eco-evolutionary model of a soil microbe-enzyme system under warming. Constraining the model with observations from five contrasting biomes reveals that evolution will likely aggravate soil carbon losses to the atmosphere, a positive feedback to climate change. The model reveals a strong latitudinal gradient in evolutionary effects, driven mostly by initial temperature, from small evolutionary effects at low (warm) latitudes to large effects at high (cold) latitudes. Accounting for evolutionary mechanisms will likely be critical for improving projections of Earth system responses to climate change.
Introduction
Microorganisms are key drivers of global biogeochemical cycles1. In terrestrial ecosystems, soil microbes decompose organic matter, returning carbon to the atmosphere as carbon dioxide (CO2)2. In vitro and in situ experiments suggest that changes in microbial decomposition with warming are an important feedback to climate3–5. Soil microbial populations may respond to increasing temperature through physiological mechanisms such as individual metabolic adjustment6,7 and ecological mechanisms such as shifts in population abundance or community composition8,9. Given the short generation time, large population sizes and standing genetic variation of many microbial organisms, evolutionary adaptive responses of microbial populations to warming are also likely10,11. However, how microbial evolutionary adaptation may contribute to carbon-climate feedbacks is unknown12.
Key to microbial decomposition of soil organic matter is the production by microbes of extracellular enzymes (exoenzymes), that diffuse locally in the soil and bind to soil organic matter compounds13. Because the fitness cost of exoenzyme production14 (reduced allocation to growth, Fig. 1a) is paid by individual microbes whereas fitness benefits (larger resource pool) are enjoyed by microbial collectives15, we expect genetic variation in exoenzyme production16 to be under strong selection15,17. Our objective is to evaluate how exoenzyme production responds to selection under environmental warming, and how the evolutionary response of exoenzyme production impacts the response of soil organic carbon stock (SOC). To this end, we develop and analyze a novel eco-evolutionary model, modified from that of ref. 1818 (Fig. 1a), to take microbial evolutionary adaptation into account.
In this novel eco-evolutionary model, the focal microbial adaptive trait is the fraction of assimilated carbon allocated to exoenzyme production19,20, or ‘exoenzyme allocation fraction’, hereafter denoted by φ. The balance of assimilated carbon, 1-φ, is allocated to microbial growth. Allocated carbon is then converted to exoenzymes or microbial biomass with, respectively, enzyme production efficiency γZ and microbial growth efficiency γM, while the balance is respired and lost as CO2 in order to produce the metabolic energy needed for enzyme or biomass production (see Fig. 1a and Methods). Competition between microbial strains differing in enzyme allocation fraction φ drives adaptive evolution. The model predicts the adapted value, φ*, of the exoenzyme allocation fraction at any given temperature; the adaptive response of the exoenzyme allocation fraction to temperature rise; and how this response impacts the decomposition rate and SOC stock. By comparing the full eco-evolutionary (ECOEVO) response of SOC stock to the purely ecological (ECO) response in absence of evolution (in which φ is a fixed parameter that does not change), we can evaluate the contribution of microbial evolutionary adaptation (EVO effect) to the direction and magnitude of the SOC stock response to climate warming (Fig. 1b, c).
Results
Scenarios of microbial temperature dependence
Microbial decomposition is predicted to respond to warming due to the temperature sensitivity of intra- and extra-cellular enzymatic activity21–23. In our baseline ‘kinetics-only’ scenario of temperature-dependent decomposition, we assume that microbial uptake parameters (maximum uptake rate and half-saturation constant) and exoenzyme kinetics parameters (maximum decomposition rate and half-saturation constant) increase with temperature5,24. We consider two additional scenarios for the influence of temperature on decomposition. In the microbial mortality scenario, the microbial death rate also increases with temperature25. This could be due to a higher risk of predation or pathogenic infection at higher temperatures, or faster microbial senescence due to higher protein turnover25. In the microbial growth efficiency (MGE) scenario, MGE (the fraction of carbon allocated to growth that actually contributes to microbial biomass, as opposed to being released as CO2 via growth respiration) decreases with temperature18,26, which could be due to higher maintenance costs at higher temperature27.
Evolution of the enzyme allocation fraction trait
At any given temperature, the evolutionary model predicts that the adapted value of the enzyme allocation fraction, φ*, depends on four parameters (equation (5) in Methods): MGE, mortality, maximum uptake rate, and local competitive advantage to exoenzyme producers, or ‘competition asymmetry’. Among these parameters, competition asymmetry quantifies the accessibility of the exoenzyme ‘public good’ to microbes. It measures the differential availability of enzymatically produced dissolved organic carbon (DOC) to different microbial strains. Competition asymmetry is shaped by diffusion of exoenzymes and DOC, and by microbial mobility, and is thus likely influenced by soil physical properties, such as texture or moisture. For simplicity, we assume that competition asymmetry is independent of temperature.
Comparing ECO and ECOEVO responses of soil carbon to warming
In all three scenarios of temperature dependence, the pure ecological (not including evolution) equilibrium of SOC generally decreases as temperature or exoenzyme allocation fraction increases (Fig. 1b, Supplementary Fig. 4, Supplementary Note 5). This evolution-independent (ECO) response results in the lost carbon being released to the atmosphere as CO2, a positive feedback to warming consistent with previous models.
With the eco-evolutionary (ECOEVO) version of the model, and focusing on the baseline scenario of temperature dependence, we find that evolution causes the adapted enzyme allocation fraction, φ* (which is held constant in the pure ECO model runs), to always increases with increasing temperature (Fig. 1b, Supplementary Fig. 5a). Combining both results, we conclude that the ECOEVO response mediated by microbial evolution amplifies the ecology-driven loss of soil carbon due to warming (ECO response) (Fig. 1c).
The effect of evolution (EVO effect) is strong when the ECOEVO response markedly departs from the ECO response. This is predicted in cold ecosystems harboring communities of slow-growing microbes, under soil conditions that give only a small competitive edge to greater enzyme producers (Fig. 2, Supplementary Fig. 6). Strong EVO effects are robust to the other model parameters – enzyme parameters (efficiency, production) and environmental parameters (litter input, leaching) (Supplementary Figs. 6 and 7, Supplementary Note 4).
We address the robustness of strong evolutionary effects (Fig. 3a) to different scenarios of temperature dependence by focusing on values of nutrient uptake parameters, litter input, and competition asymmetry that are conducive to such large EVO effects (for example, point B2 in Fig. 2b). In the temperature-dependent mortality scenario, how the direction and magnitude of the EVO effect changes from the baseline scenario is entirely determined by the sensitivity of microbial mortality to warming. When mortality is moderately sensitive to temperature, the ECO and ECOEVO responses become more similar, resulting in a smaller EVO effect (Fig. 3a, b). When mortality is strongly sensitive to temperature, the ECOEVO response becomes weaker than the ECO response, which implies that evolution buffers the loss of soil carbon (negative EVO effect, Fig. 3c). Such EVO effects are predicted to be stronger in warmer ecosystems (higher T0, Fig. 3c; see also Supplementary Note 7).
In the temperature-dependent MGE scenario, the direction and magnitude of the ECOEVO response to warming vary dramatically with the initial temperature T0 (Fig. 3d). Driven by adaptive evolution, the enzyme allocation fraction increases strongly with warming in cold systems, hardly changes in temperate systems, and decreases markedly in warm systems (Supplementary Fig. 5d), with parallel effects on the decomposition rate (Supplementary Fig. 5h). As a consequence, at low T0, aggravation of soil carbon loss by evolution is almost as severe as in the baseline scenario, even though both ECO and ECOEVO responses are weaker in this scenario (Fig. 3a, d). At higher values of T0, the ECOEVO response becomes weaker than the ECO response or even positive (Fig. 3d), resulting in the sequestration rather than loss of soil carbon.
Model predictions using empirical data from five biomes
To illustrate how EVO effects may vary in real ecosystems, we used available data22 on the decomposition kinetic parameters in five biomes of increasing latitude and decreasing mean annual temperature (Costa Rica, California, West Virginia, Maine, and Alaska, Fig. 4). We evaluated ECO and ECOEVO responses for each biome under three levels of competition asymmetry (as quantified by the local competitive advantage to producers, c0) (Fig. 4a-h). Under our baseline scenario, EVO effects correlate strongly with mean annual temperature, even more so for low competition asymmetry (Fig. 4i). Stronger EVO effects occur in colder biomes, as found in the general analysis (Fig. 3a). In contrast, the ECO response does not correlate with mean annual temperature (Fig. 4a). As a result, a temperate biome such as Maine exhibits a weak ECO response that can be strongly amplified by evolution, whereas the warm Costa Rica biome shows a strong ECO response that is little affected by evolution.
These results are quantitatively attenuated but qualitatively unaffected when microbial mortality increases moderately with temperature (Fig. 4b, f, j). With a stronger effect of temperature on microbial mortality, all biomes show the evolutionary buffering effect (Fig. 4k) found in the general analysis (Fig. 3c). The intensity of evolutionary buffering is independent of the biomes’ mean annual temperature, whereas it varies significantly with competition asymmetry (Fig. 4k). Under the temperature-dependent MGE scenario, ECO and ECOEVO responses are reduced in magnitude compared to the baseline scenario (Fig. 4d, h), particularly in cold biomes. However, in these biomes, EVO effects are enhanced dramatically (Fig. 4l). Thus, in a biome as cold as Alaska, a significant ECOEVO loss of soil carbon is predicted, whereas the purely ecology-driven loss of soil carbon would be negligible (Fig. 4l).
Discussion
As global warming increases environmental temperatures, our model predicts evolution of the enzyme allocation fraction, with potentially large effects on the decomposition process and SOC stock. The size of evolutionary (EVO) effects is most sensitive to MGE, microbial mortality, activation energy of uptake maximal rate, competition asymmetry, and initial temperature (Fig. 2). Evolution often aggravates the ecological loss of soil carbon in response to warming, especially in cold biomes (Figs. 1, 3). We identified two cases in which evolutionary adaptation to warming may buffer or even revert the ecology-driven loss of soil carbon: strongly temperature-dependent microbial mortality, or temperature-dependent MGE in warm ecosystems (Fig. 3). Overall, we expect evolutionary effects to vary greatly among ecosystems that differ in biotic (microbial life history and physiology) and abiotic (temperature, soil texture and moisture) characteristics.
Implications of our findings for large geographic scales, across terrestrial ecosystems, are hinted at by specifying the model for the five contrasting biomes for which exoenzyme kinetics data are available22. We find evolutionary aggravation of soil carbon loss to be the most likely outcome, with a strong latitudinal pattern induced by temperature, from small evolutionary effects at low (warm) latitudes to large evolutionary effects at high (cold) latitudes. In all cases, competition asymmetry is a strong influence on the eco-evolutionary response of microbial decomposition to warming (Fig. 4). Soil texture and moisture, which may influence competition asymmetry, vary considerably among locations22. Predicting geographic variation in ECOEVO responses and EVO effects across large geographic scales thus calls for more empirical data on variation in ecological (competition) traits, especially how the competitive advantage to exoenzyme producers varies with soil physical properties.
Large-scale projections of soil C cycle changes on timescales that are long relative to the characteristic times of the processes described by our model also raise the issue of distinguishing between readily available SOM versus physio-chemically protected SOM, which decomposes more slowly5,28. Reinterpreting the soil carbon stock compartment of our model as readily available SOM removes the constraint of having the stock of microbial biomass, M, much smaller than SOC (C in our model)29,30. The relevant measure of soil carbon then becomes the sum of all compartments, Ctot = C + M + Z + D, called ‘total soil active carbon’, where Z is the exoenzyme concentration and D the biomass of dissolved organic carbon (equations (1a-d) in online Methods). We investigated how our results are changed when ECO and ECOEVO responses are defined with respect to Ctot, rather than C, focusing on the baseline scenario (constant MGE and constant microbial turnover). Supplementary Figure 9 exemplifies the case where microbial biomass is the main component of Ctot and total soil active carbon is higher in the presence of microbes than in their absence. This occurs because here a higher SOC leaching rate, eC, is more than compensated by the microbial higher productivity (high MGE, γM, and low death rate, dM), so that carbon is retained in the soil longer in the presence of microbes than without. In this type of high-leaching soils, the direction of the ECO response shifts from negative (carbon loss) to positive (carbon gain). Microbial adaptive evolution can amplify, buffer or reverse this ecological response depending on the local competitive superiority of enzyme producers, c0 (Supplementary Fig. 10). This is because microbial biomass, M, is the main carbon stock in these systems and purely ecological processes drive an approximately linear increase in M with temperature whereas M responds non-linearly to the increase in enzyme allocation fraction (Supplementary Fig. 9b).
Our study demonstrates how a trait-based approach can be used to integrate microbial evolutionary adaptation into carbon cycle models31. Our focal trait, the enzyme allocation fraction, φ, is pivotal in the decomposition process. Other microbial traits that appear in our model may also respond to selection imposed by warming, such as MGE (γM) and the exoenzyme production efficiency (γZ). The γM trait may evolve in response to variation in environmental quality (e.g. predation or infection risk), possibly trading-off with the maximum rate of uptake or correlating with the death rate; γZ may evolve in response to substrate variation20, possibly correlating with the enzyme decay rate. Given specific assumptions about factors of and constraints on the evolution of these traits, our sensitivity analyses can be used to make qualitative predictions on how the ECOEVO response and EVO effect might be altered (Fig. 2, Supplementary Fig. 6). For example, strong EVO effects may not be affected if evolving γM correlates positively with dM, whereas weak effects may become strong if γM trades-off with dM. In contrast, the low sensitivity of EVO effects to γZ suggests that the evolution of this trait may have little impact on the response of soil C to warming.
The adaptive trait-based approach we use circumvents the difficulty of prescribing critical parameters (such as the enzyme allocation fraction) and their response to environmental change by allowing these parameters to naturally emerge from biologically grounded evolutionary dynamics of the system under study. This approach can easily be extended to predict (rather than assume or prescribe) the adaptive dynamics of multiple microbial traits, such as growth-related traits and substrate-specific enzyme production traits, and their diversification into coexisting functional types. A similar approach has been used to construct a physiology-based model of feedbacks between global ocean ecosystem function and phytoplankton diversity32. Rather than assuming values for the multiple physiological traits characterizing each plankton species, that model allowed interactions among randomly parametrized species drive species sorting and emergence of the corresponding trait values. This trait-based approach for microbial communities also facilitates genomic and metagenomic data, mapped to soil microbial function, to be used in model validation 33.
Future extensions of our eco-evolutionary model will enable the mechanistic representation of below-aboveground feedbacks between soil microorganisms and vegetation, by coupling the carbon cycle with other major biogeochemical cycles, such as nitrogen and phosphorus1. Existing mathematical19,34,35 and computational models20,36,37 pave the way for such extensions. Global projections of the effect of soil microbial evolution on future climate change will become possible (e.g. Wieder et al. 2013) by coupling our eco-evolutionary model of biological decomposition with soil models that account for the chemical and physical transformations of soil carbon occurring on year-to-century timescales5,38–40.
Given the large population size and short generation time of many microorganisms, biological principles suggest that microbial evolution should be an essential component of ecosystem response to warming. We have shown here that microbial evolutionary adaptation to warming, and its impact on the decomposition of soil organic matter, can radically change soil carbon dynamics. Empirical data suggest that natural values of enzyme allocation fraction are low23,34 and fall in the range for which our model predicts large eco-evolutionary responses of decomposition to warming. In spite of an increasing effort to document and understand the ecosystem impact of microbial physiological and ecological responses to climate warming18,26,41, no Earth system model that seeks to represent the role of living organisms in climate feedbacks has yet included evolutionary mechanisms of adaptation. Our model is a critical first step. We expect projections of future climate and carbon cycle feedbacks, and their uncertainty, to be significantly impacted, from local to global scales.
Methods
We use the microbe-enzyme model of litter decomposition first introduced in ref. 1818 and extend it to describe the ecological dynamics of soil organic carbon (SOC), dissolved organic carbon (DOC), microbial biomass, and extracellular enzyme abundance, given litter input, leaching rates, and soil temperature (Fig. 1a and Supplementary Fig. 1). The effect of temperature is mediated by enzymes kinetics, with exoenzymes driving the decomposition rate, and intra-cellular enzymes involved in resource uptake and microbial biomass synthesis. As temperature changes, the model predicts how the ecological equilibrium changes. The change in equilibrium SOC is what we call the ecological (ECO) response of the ecosystem.
To investigate the effect of evolutionary adaptation on decomposition, we include microbial evolution in the ecological model. Our focus is on soil bacteria (as opposed to fungi), which typically have large population size and short generation time. We assume that microbes may vary individually in their investment in exoenzymes, measured by the fraction (denoted by φ throughout the paper) of resources allocated to enzyme production. Assuming that some of this variation has a genetic basis, we derive the selection gradient and compute the evolutionarily stable value of the enzyme allocation fraction, φ*, at any given temperature. We can then evaluate how φ* changes as temperature increases, and how the ecosystem equilibrium changes from both the direct effect of temperature rise on enzyme kinetics, and the indirect effect mediated by microbial evolutionary adaptation to warming (Fig. 1b, c).
Ecological model
Based on ref. 1818 (Fig. 1a, Supplementary Fig. 1), the ecological model has four state variables measured in unit mass of carbon: soil (non decomposed) organic carbon (SOC), C; soil decomposed soluble organic carbon (DOC), D; microbial biomass, M; and exoenzyme concentration, Z. Exoenzyme production drives the decomposition process of SOC into DOC, which is the only source of carbon for microbes. The model accounts for microbial production and death, exoenzyme decay, recycling of dead microbes and degraded exoenzymes, SOC input from plant litter, and leaching of SOC and DOC.
Model equations
State variables C, D, M, Z obey equations (1a-d):.
In equation (1a), decomposition follows from Michaelis-Menten kinetics of Z binding substrate C; there is a constant input, I, of soil organic (non decomposed) carbon from aboveground litter, and a loss due to leaching at constant rate eC. In equation (1b), D is produced by decomposition and the recycling of dead microbial biomass and inactive enzymes; D is consumed by microbial uptake, and lost by leaching at constant rate eD. In equation (1c), growth of microbial biomass M is driven by the rate of DOC uptake (a Monod function of D) times the fraction of uptaken DOC turned into biomass, (1 – φ) γM, minus microbial mortality at constant rate dM. In equation (1d), enzyme variation is driven by the rate of DOC uptake times the fraction allocated to enzyme production, φ, and production efficiency, γZ, minus enzyme deactivation at constant rate, dZ.
Ecosystem equilibria
The ecological model possesses either one globally stable equilibrium, or three equilibria (one of which is always unstable) (Supplementary Fig. 2). There are thresholds φmin and φmax such that the globally stable equilibrium exists for φ < φmin or φ > φmax and is given by C = I/eC, D = 0, M = 0, Z = 0. Thus, at this equilibrium, the microbial population is extinct and no decomposition occurs. For φmin < φ < φmax, the microbial population can either go extinct (then the system stabilizes at the same equilibrium as before) or persists at or around a non-trivial equilibrium, which can be solved for analytically. Note that φmin and φmax depend on all microbial and model parameters (Supplementary Fig. 3, Supplementary Note 1).
Effect of temperature on model parameters
Decomposition is predicted to respond to warming5 due to the temperature sensitivity of enzymatic activity21,22,42. Microbial assimilation may also vary with temperature if the microbial membrane proteins involved in nutrient uptake are sensitive to warming. Following ref. 1818, we assume that exoenzyme kinetics parameters (maximum decomposition rate ) and half-saturation constant ) and microbial uptake parameters (maximum uptake rate and half-saturation constant ) follow Arrhenius relations with temperature. This defines our baseline ‘kinetics-only’ scenario of temperature-dependent decomposition:
where T is temperature in Celsius, R is the ideal gas constant, and the E parameters denote the corresponding activation energies.
We consider two additional scenarios for the influence of temperature on decomposition. In the temperature-dependent microbial mortality scenario25, the microbial death rate increases with temperature. This could be due to a higher risk of predation or pathogenic infection at higher temperatures, or faster microbial senescence due to higher protein turnover25. In this scenario, the microbial death rate dM depends on temperature according to as in ref. 2525.
In the temperature-dependent microbial growth efficiency (MGE) scenario, the MGE decreases with temperature18,25,26, possibly due to higher maintenance costs at higher temperature27. This is modeled by making the microbial growth efficiency γM vary linearly with temperature18,22,43,44: with Tref = 20 °C.
How scenarios of temperature-dependence and parameter values influence the response of equilibrium C to temperature is shown in Supplementary Fig. 4 and commented on in the Supplementary Note 4.
The enzyme allocation fraction φ is a ‘public good’ trait: as an individual microbe produces exoenzymes, it experiences an energetic cost and obtains a benefit – access to decomposed organic carbon – that depends on its own and other microbes’s production in the spatial neighborhood45,46. As a public good trait, φ is under strong direct negative selection: ‘cheaters’ that produce less or no exoenzymes, and thus avoid the cost while reaping the benefit of enzyme production by cooperative neighbors, should be at a selective advantage. In a highly diffusive environment in which exoenzymes are well mixed, φ would evolve to zero, leading to evolutionary suicide47. However, in a more realistic spatially distributed environment with limited exoenzyme diffusion, microbes with a given trait are more likely to interact with phenotypically similar microbes, which puts more cooperative microbes at a competitive advantage over less cooperative strains45,48,49. This generates indirect positive selection on trait φ.
The trait value φ* at which negative and positive selections balance is the evolutionarily stable microbial strategy, given by where c0 measures the competitive advantage, due to spatially local interactions, of any given strain over a slightly less cooperative strain, or ‘competition asymmetry’ (Supplementary Note 3). The parameter c0 is likely to depend on the diffusivity of DOC, which may itself vary with soil properties such as texture or water content.
As temperature rises from T0 to T, the direction and magnitude of the microbial adaptive response is measured by Δφ* = φ* (T) ‒ φ* (T0), which depends on the scenario of temperature dependence. The eco-evolutionary (ECOEVO) response of SOC is given by where C(T, φ) denotes ecological equilibrium C at temperature T, given enzyme allocation fraction φ. The ECOEVO response is to be compared with the purely ecological (ECO) response: in which the enzyme allocation fraction is fixed at its T0-adapted value, φ*(T0) (Fig. 1c).
We measure the magnitude of the evolutionary (EVO) effect as the difference between the ECOEVO response averaged over the temperature range (T0, T) and the ECO response averaged over the same temperature range, normalized by the ECO response:
This evaluation allows us to compare EVO effects across systems that differ in the magnitude of their ECO response. In all simulations we use T = T0 + ΔT where ΔT = 5 °C. In general, the ECO and ECOEVO responses are monotonic, close-to-linear functions of T over the considered temperature ranges (T0, T0 + ΔT), which makes all our comparative analyses almost insensitive to our choice of ΔT.
Parameters default values and variation range
Under the temperature-dependent kinetics-only scenario, the ecological model (equations (1) and (2)) includes seven microbial parameters , six enzyme parameters and four environmental parameters (I, eC, eD, T). Our set of default parameter values is derived from Allison et al. (2010) (Supplementary Table 1). The enzyme allocation fraction default value is 10% at 20 °C34. For the dependence of enzyme kinetics parameters on temperature, and , we selected the Arrhenius equations that best fit data from California22 (mean annual T = 17 °C) and match values at 20 °C (0.42 and 600, respectively)18. For the uptake kinetic parameters, we obtained by selecting that best fits the Arrhenius equation in ref. 1818 with and we obtained by selecting and that best fit the linear relation used in ref. 1818. To parametrize the temperature-dependent mortality (dM,ref, Tref) and MGE (γM,ref, m, Tref) models, we used values from ref. 2525 and tested two values of EdM (EdM = 0 is the enzyme only temperature-dependent model). For greater realism, we used a higher value of the exoenzyme deactivation rate (twice the value used in ref. 1818) and constrained the range of all parameters in order to enhance stability and produce relative stock sizes that are consistent with empirical data, so that at equilibrium M is about 1% of C, Z is about 1% of M and D is limiting (hence close to 0 at equilibrium29,30),
We analysed the model sensitivity by varying parameters over two orders of magnitude (as in ref. 1818) – except γM and γZ for which we used the whole range over which the non-trivial ecosystem equilibrium is stable (Supplementary Table 2). To assess the significance of our findings for real ecosystems, we focused on five biomes for which empirical data22 could be used to constrain the model. The five biomes contrast strongly in their initial temperature, T0, and decomposition kinetics, and for which we selected the Arrhenius equations (2) that best fit the relations used in ref. 2222 (Supplementary Table 3).
Author contribution
R.F. conceived the study. All authors developed the model. E.A. performed the analysis. E.A. and R.F. wrote the first version of the manuscript. All authors finalized the paper.
Competing interests
The authors declare no competing interests.
Additional information
Supplementary information is available for this paper at https://doi.org/xxx
Correspondence and requests for materials should be addressed to E.A. or R.F.
Acknowledgements
We thank Rachel Gallery, Pierre-Henri Gouyon, Moira Hough, Hélène Leman, Laura Meredith, and Mitch Pavao-Zuckerman for discussion. E.A. was supported by fellowships from Ecole Doctorale Frontières du Vivant and MemoLife Laboratory of Excellence (PIA-10-LBX-54). S.R.S. was supported by a grant from the Genomic Science Program of the United States Department of Energy (DOE) (DE-SC0016440) and the University of Arizona’s Agnese Nelms Haury Program in Environment and Social Justice. R.F. acknowledges support from FACE Partner University Fund, CNRS Mission pour l’Interdisciplinarité, and PSL University (IRIS OCAV and PSL-University of Arizona Mobility Program).