Controls on turnover of marine dissolved organic matter—testing the null hypothesis of purely concentration‐driven uptake: Comment on Shen and Benner, “Molecular properties are a primary control on the microbial utilization of dissolved organic matter in the ocean”

Abstract

This comment is inspired by results of a bioassay experiment, which found different dissolved organic carbon (DOC) turnover rates depending on DOC origin (Shen and Benner 2020). Based on these experiments, it was concluded that molecular properties govern microbial utilization of DOC, that is, the freshly produced DOC from a phytoplankton bloom is inherently labile and thus quickly degraded, whereas aged DOC from greater depths is inherently refractory on experimental time scales, and thus degraded more slowly. Microbial reworking of organic substrates has been suggested as a dominant factor increasing stability of dissolved organic matter and has been termed the “microbial carbon pump” (Jiao et al. 2010). Experimental evidence supports that conclusion (see Ogawa et al. 2001). A range of mechanisms have been identified but discussion focuses on the dominant role of the two competing, but not mutually excluding mechanisms of structural recalcitrance and limitation of substrate concentration (see Jiao et al. 2011; Arrieta et al. 2015; Dittmar 2015; Walker et al. 2016; Wang et al. 2018). There are still debates on the definitions of what is refractory DOM in the ocean (Baltar et al. 2021). In light of this ongoing discussion, here we test an alternative explanation for the observations in the bioassay experiment by Shen and Benner (2020). To separate the effect of molecular properties from the control of concentration-driven uptake, we use a numerical model of bacteria–DOC interactions without systematic differences in DOC reactivities. These model assumptions enable testing the null hypothesis that the experimental observations are driven purely by concentration-driven uptake. While our intention is not to contradict the conclusions drawn in the original study, we offer an additional explanation to illustrate the full range of possibilities from molecular properties governing DOC decay (as stated in the original study) to microbial DOC interactions (this study). While both mechanisms are relevant for DOC turnover, we do not yet know to what extent which one controls the size of the global DOC inventory. Here we propose a theoretical framework on how these effects can be assessed separately. We use a simple model of a microbial network to reproduce the observations of the bioassay experiments. The network model is a numerical, zero-dimensional box model that simulates the concentration of 100 DOC compound classes and the biomass of 35 bacterial groups. It is not the number of compound classes and bacteria, but their ratio that influences the DOC concentration in a specific set of model parameter values (Mentges et al. 2019). Generally, the qualitative model behavior is not altered by the number of compound or bacteria classes (see Supporting Information Fig. S1). The model has been described in detail in Mentges et al. (2019). It is based on several general observations: (1) DOC compounds and microbial consumers are diverse; (2) one microbial group can only take up a subset of the DOC compounds present; (3) DOC is exchanged by microbes through the processes of uptake, release during growth and mortality, and transformed to CO2 during respiration; and (4) bacteria molecularly diversify DOC, that is, they release more compounds than they take up (Lechtenfeld et al. 2014; Noriega-Ortega et al. 2019). Model equations and parameters are listed in the Supporting Information Appendix S1. Note that here we approximate bacterial abundance by bacterial biomass. This simplification does not influence the qualitative model results. We deliberately take a model approach that starts very simple, for example, neglecting the potential presence of viruses or grazers in detail (mortality is summarized as one parameter in the model). We test the hypothesis whether concentration-driven uptake largely explains observations using a low number of degrees of freedom (i.e., model parameters) to describe this complex system. Increasing the complexity in a second modified model version (discussed below) enables separating the quantitative effects of concentration-drive uptake and biodiversity and chemodiversity. All microorganisms in the model can take up and release the same number of DOC compounds, but their position in the network is randomly assigned. To account for this randomness in network design, we run 100 random network realizations for each model simulation and analyze the average. The experiments of the original study are repeated in silico based on the above described model. Prior to the numerical batch experiment, the network model is spun up for 500 years until quasi-steady-state after manually adjusting the growth rate and substrate specificity to reproduce the same starting concentration as in the incubation experiment (Table 1). This tuning alters the DOC concentration at t = 0 of the experiment (i.e., the concentration after spin-up) to enable direct comparison to the original data, but does not alter the model behavior. In the original experiment, total organic carbon, not DOC, was measured, so it cannot be excluded that there was particulate material present at the beginning of the experiment. We stick with the term “DOC” in the following, as our model does not differentiate between dissolved and suspended matter. We present simulations for the shallowest and deepest depth of the original study to cover the full range of variability. The main difference between surface and subsurface in the model experiment is the amount of external DOC that is constantly supplied to the network during the spin-up phase of the model. The amount of DOC supplied for these different scenarios is the same as in the first description of the model by Mentges et al. (2019). The amount of supply is proportional to the bacterial biomass in the model, but does not influence DOC concentration in steady state (Mentges et al. 2019, 2020). As a consequence, bacterial biomass is higher at the starting point of in silico experiments at the sea surface compared to the subsurface. At the starting point of the virtual incubations (t = 0), specific DOC compounds are added to the network representing the DOC sources of the original bioassay experiment. Although all compounds have the same properties (i.e., the same number of consumers with the same uptake parameters), they differ by their randomly assigned position in the network (Fig. 1). This position is defined as the shortest pathlength to one of the externally supplied compounds (Fig. 1). Phytoplankton-derived DOC compounds are represented as externally supplied compounds in the model and have a pathlength of 0 (red dots in Fig. 1). Solid-phase extracted deep-sea DOC (“C18-DOC” in Shen and Benner 2020) is microbially processed and represented as those compounds with the largest pathlength in the network (yellow dots in Fig. 1). The main difference between fresh and processed DOC compounds is thus not represented in their molecular properties in the model, but in the abundance of their consumers at the initial point of the virtual experiment, which is derived from the model spin-up matching the starting concentration of the experiment. The same bulk amount of DOC as in the bioassays is distributed equally among all DOC compounds of the respective pathlength for the virtual experiments. After t = 0, the supply of DOC is stopped and the model is run for 200 d. The specific assumption of this network model is that all DOC compounds are on average equally available to a part of the microbial community. This assumption allows us to test the null hypothesis that only concentration-driven uptake controls the degradation of DOC. Such an assumption is justified by the huge genomic potential and the enormous microbial diversity in the ocean (Sogin et al. 2006; Salazar and Sunagawa 2017), together with recent discoveries that the genomic potential of an abundant bacterial lineage predicts degradation of dissolved organic matter previously considered as “refractory” (Landry et al. 2017). Our model set-up does not deny the existence of distinguishable molecular compounds, but is built on the assumption that their reactivity is randomly distributed, corresponding to the null hypothesis. The exact reactivity spectrum of the complex mixture of natural DOM is not fully understood, but experimental based on the solid-phase extractable fraction of DOM and modeling studies indicate that log-normal distributions of reactivity are likely (Mostovaya et al. 2017; Zakem et al. 2021). We test the model outcome of different randomly assigned reactivities of DOC compounds. Compound-specific reactivity factors are drawn from Gaussian and log-normal distributions, and compared to the scenario where all compounds have the same reactivity. Overall model behavior is only marginally dependent on the reactivity spectrum of DOC compounds. Steady-state concentrations of Gaussian or log-normal reactivity spectra can equally be described by an average reactivity of all compounds (Fig. 2). Hence, we use the simplified version with equal reactivity to test the null hypothesis that no systematic differences in molecular properties are needed to explain the observations of the bioassay experiment. Even in the very basic configuration, where all microorganisms and compounds have exactly the same properties, different turnover rates as observed in the bioassay experiments are reproduced. First, we consider the surface scenario (Fig. 3a): Without supply of DOC after the beginning (t = 0) of the numerical experiment, the concentration of DOC declines only very slowly, because the bacterial biomass of the active degraders becomes very low (Fig. 3a, blue). This result is comparable to the control experiment in the original study. The addition of phytoplankton derived DOC is simulated by adding the same amount of DOC as in the experiment, equally distributed among those DOC compounds that are externally supplied to the network. After addition of freshly supplied DOC at t = 0 and no external supply afterwards, DOC concentration declines fast, and then slows down (Fig. 3a, red line). Here, some deviations from observations occur as the model underestimates the kinetics of initial DOC decay. To simulate the addition of processed DOC (C-18 DOC), we distribute the amount added in the bioassay experiment among all compounds that have the longest distance to the externally supplied compounds in the network at t = 0 (e.g., yellow DOC-nodes in network in Fig. 1). DOC declines slowly, and the decline accelerates towards the end (Fig. 3a, yellow line). The observed dynamic results from the time lag that the less abundant consumers need to grow to substantially decrease the DOC concentration. The experiment with addition of both, fresh and processed DOC, shows a fast turnover at the beginning until Day 40, and a slower decline afterward (Fig. 3a, orange line), almost in parallel to the treatment with only processed DOC (Fig. 3a, yellow line). As in the previous experiment, the difference in DOC turnover results from the difference in concentration of the consumers at t = 0. The model also reproduces the dynamic in the subsurface experiment with deep water, that is, it captures the constant DOC concentration in the control experiment and in the treatment with processed DOM (C-18 DOM) (Fig. 3b, blue and yellow lines, respectively). Yet, there are some deviations between experiment and model for the subsurface. Fresh DOC, representing phytoplankton-derived DOC, is degraded slower on average than in the bioassay experiments (Fig. 3b, red line). This numerical experiment shows largest differences among the randomly assigned networks (Fig. 3b, large red area). In total, our basic model reproduces the experimental observations that (1) processed, deep-sea DOC is only marginally degraded in the course of 200 d despite the assumption in our model that all compounds are degradable; (2) fresh, unprocessed DOC is faster degraded than processed DOC. The model behavior, that is, different turnover times for different DOC fractions depending on substrate concentration and consumer abundance, is a general result applicable to other depths as well (Supporting Information Fig. S2). These results show that concentration-driven uptake can explain variations in DOC turnover time independent from molecular reactivity properties, underlining its role as a primary control. The basic model describes different turnover times solely based on concentration-driven uptake in this complex system with only ~ 10 degrees of freedom (parameters), which supports the hypothesis that microbial interactions are a dominant mechanism behind varying DOC degradation rates. The model is further modified to assess which process is missing in the basic version to achieve full agreement with observations. The modified model now includes a diverse microbial community that varies in the individual DOC uptake rates of the bacteria. The DOC uptake rate (linearly proportional to the growth rate in our model) is now a function of the pathlength in the network, between the externally supplied compounds (Fig. 1, red nodes) and bacteria (pathlength in our model configuration can have values 1, 3, and 5, Fig. 1). The modification is in agreement with the observation that growth rates are high at the beginning of the succession in phytoplankton blooms, resulting from the metabolic properties of bacteria that readily respond to transient nutrient pulses (Buchan et al. 2014). The effect is similar to a classical opportunist-gleaner trade-off. Therein, “opportunists” have the potential to increase their growth rate strongly with increasing substrate concentration (copiotroph feeding strategy), whereas the “gleaner” only increases its DOC uptake slowly (Litchman and Klausmeier 2008; Ferenci 2016). The modification of the maximum uptake rate is functionally similar to introducing systematic differences in compound reactivity. To assign a property to each bacterium in the network individually, the uptake is linearized as described in Mentges et al. (2020). The linearization means that instead of the nonlinear Monod function defined by maximum turnover ρ and substrate specificity κ, we use only the parameter ζ as a second-order rate constant with DOC and bacteria as the reactants. This approach is rationalized by the assumption that at low concentrations of an individual substrate, the Monod curve can be approximated by a linear relationship, expressed as ζ = ρ κ . Generally, the kinetic of substrate uptake at very low concentrations is not very well constrained by observations, due to the difficulty of culturing marine bacteria, the molecular diversity of natural substrates in the ocean and detection limits in substrate concentration analysis. Half-saturation constants and maximum uptake rates thus differ among experiments, microorganisms, substrates, and water depths, sometimes over several orders of magnitudes. For example, half-saturation constants have been shown to be in the range of 7–74 μmol L−1 for a bacterium from a fjord for glucose, and much lower (1.3–1.8 μmol L−1) for alanine (Schut et al. 1995). An even larger range has been reported in another study, that is, 2.5 nmol L−1 to 500 μmol L−1 across several coastal and shelf sampling stations in the United States and Canada (Azam and Hodson 1981). Half-saturation constants as low as 1 nmol L−1 have been found for strains of SAR11 (Noell and Giovannoni 2019). Similarly, turnover times (comparable here to ρ in the model) ranged between 0.22 and 1.82 d−1 for different phylogenetic bacterioplankton groups in an upwelling area (Teira et al. 2009), from undetectable to 5.5 d−1 in Delaware estuary (Yokokawa et al. 2004) and from 0.5 to 2.9 d−1 in other ocean areas (calculated from Eilers et al. 2000; Beardsley et al. 2003; Yokokawa et al. 2004). The parameter values in the network model fall into this range of reported growth rates and half saturation constants. In summary, we show that the observed difference in turnover time of DOM from different sources can largely be explained by purely concentration-driven uptake. The mechanisms responsible for this DOC decay kinetics are primarily differences in the biomass of consumers and substrate concentrations, which are modulated by biodiversity and chemodiversity: DOC that usually enters the network externally, such as phytoplankton-derived DOC, is quickly degraded by consumers who are already present in comparably large numbers. In contrast, the C-18 DOC amendment, which is represented in the network model as an addition of degraded DOC furthest away from the externally supplied compounds, is degraded very slowly, because the microorganisms feeding on this substrate are present in low abundance. In short, the difference in reactivity between phytoplankton-derived and deep-sea DOM emerges in the numerical model from (1) the fact that degraders of processed DOC are present in lower abundance at the beginning of the experiment, hence the uptake-rate of these compounds is lower, and (2) the time a consumer needs to grow toward a biomass at which uptake of deep-sea DOM is substantial. Also the modeled bacterial biomass, which is between 0.008 and 1.1 mmol C m−3 at t = 0 in our experiments, is comparable to the range from not detectable to 3.7 mmol C m−3 in the global dataset by Buitenhuis et al. (2012). Unfortunately, there are no microbial data provided in the original study. For completeness, we show data on the change of bacterial biomass in the Supporting Information (Fig. S3). Furthermore, we extend our simulation beyond the duration of the bioassay experiment (i.e., ~ 5 years), to assess the design of future bioassay experiments (Supporting Information Fig. S4). For the surface community, we find a further decline in DOC concentration after 200 d of the experiment, which is strongest for the treatments where most DOC is added. The difference largely depends on the abundance of the bacterial community at the starting point of the experiment (t = 0). Based on this, we recommend collecting data on microbial abundance and diversity in future experiments, to provide further constraints on the hypothesis of concentration-driven uptake. In conclusion, we suggest an additional, not mutually excluding, explanation to the conclusions drawn in Shen and Benner (2020). Here, we have shown that the broad observations of the bioassay experiment are reproducible independently of molecular properties, but by concentration-driven uptake alone. According to our model, differences in feeding strategy of the microbial consumers (different parameter values for each microbial group) or, alternatively, systematic differences in compound reactivity act as a secondary control that further modulates concentration-driven degradation rates. Although our approach shows that the observed differences in turnover time depending on DOC origin are explicable by microbial DOC interactions, the approach does not allow us to rule out molecular properties as the primary control. Hence, we conclude that no final conclusion on the dominant stabilization mechanism can be drawn based on the available data of these bioassay experiments. Shen and Benner (2020) argue that molecular properties control the size of the DOC pool, which in turn may facilitate long-term fluctuations of the global DOC reservoir. Furthermore, they argue that we should expect a rather constant global DOC reservoir if microorganisms keep DOC compounds at a specific threshold concentration. The results of our study point towards a fundamentally different behavior of the global DOC pool. In contrast to the rather constant DOC pool that Shen and Benner (2020) predict for a concentration-driven uptake scenario, we and a previous study (Mentges et al. 2020) have shown that microbial physiology may have a strong impact on the size of the global DOC pool. Microorganisms are influenced by environmental parameters and show distinct biogeographies, structured by environmental conditions (Fuhrman et al. 2006; Green et al. 2008; Hanson et al. 2012; Brown et al. 2014; Sunagawa et al. 2015). Large-scale changes in environmental conditions may lead to variations in the amount of carbon stored in the global DOC reservoir due to changes in the composition and overall function of microbial communities (Mentges et al. 2020). We emphasize the need for a deeper, mechanistic understanding on the relative importance of the factors driving DOC reactivity to assess long-term of the global DOC inventory. We provide a framework which can in the design of future incubation experiments. A of molecular and microbial in experiments, and model have the potential to the relative importance of different stabilization mechanisms on and global Appendix Model description S1. Model parameter and of the basic and modified that from the model in Mentges et al. to the specific DOC concentrations at t = 0 in the bioassay experiments (Shen 2020). S1. The same model experiments the modified model) as in the main have been repeated but with a different network Fig. DOC concentration in the modified model compared to experimental data at the depth of Fig. The bacterial biomass for the simulations with the basic model and the modified model experimental is available on bacterial biomass. Fig. concentrations after 5.5 for the experiments shown in Fig. in the main The is not responsible for the or of supplied by the than missing should be to the corresponding for the and by the for and by the of and by

Topics & Concepts