Discussion

Using a model of anoxic-oxic regime shifts, we found three key results: first, not all functional groups had positive diversity effects on the resilience of their preferred ecosystem state (Fig. 3a-f). Second, the effect of diversity was often smaller or absent when a functional group was on the recovery trajectory rather than on the collapse trajectory (Figure 3a-f). And third, diversity effects were often reduced or erased when multiple groups varied in traits, except when a functional group had no individual effect on resilience (Figure 3g-o). From a management perspective, our results suggest that higher biodiversity in a functional group that dominates a desired ecosystem state may prevent or delay the shift to an undesired state but does not necessarily promote recovery. A more promising approach to facilitate recovery may be to identify functional groups in the undesired state (here: phototrophic sulfur bacteria) that promote their own collapse and therefore the recovery of the desired state.

Diversity did not always increase resilience

Greater diversity in cyanobacteria or sulfate-reducing bacteria increased the resilience of the state they dominated, whereas greater diversity in phototrophic sulfur bacteria reduced resilience of their preferred state (Figure 3a-f). These contrasting diversity effects reflect differences in interactions: whereas cyanobacteria and sulfate-reducing bacteria interact in a strictly antagonistic way, phototrophic sulfur bacteria have both positive and negative effects on the other two functional groups (Figure 2). By oxidizing sulfide, phototrophic sulfur bacteria indirectly increase the level of oxygen because the lower sulfide concentration (i) promotes cyanobacteria and therefore oxygen production, and (ii) leads to less oxygen consumption during abiotic oxidation of sulfide (Supplementary Report Section 10.1). Via these two pathways, phototrophic sulfur bacteria deteriorate their own environment and promote the shift to the oxic state. High diversity in this group even resulted in failure to exclude cyanobacteria likely because the fastest-growing strain lowered the sulfide concentration to such an extent that cyanobacteria were able to grow despite low oxygen diffusivity (Supplementary Report Section 10.2).
Stimulating the growth of groups that have negative diversity effects may be used as a management tool to promote the recovery of a desired ecosystem state. Such an approach would require identifying functional groups that co-dominate an undesired ecosystem state and negatively affect it. Microbes may be particularly useful for this approach because (i) they often dominate undesired ecosystem states (e.g. turbid lake state, anoxic state), and (ii) they often interact via modification of the chemical environment (Ratzke et al. 2018; Ratzke & Gore 2018), including even microbes that drive themselves extinct by deteriorating their environment (Ratzke et al. 2018).

Diversity effects were often smaller or absent on recovery trajectories

In two functional groups, the effect of trait diversity on resilience was very small (cyanobacteria; Figure 3b) or absent (phototrophic sulfur bacteria; Figure 3 f) when the group was on the recovery trajectory rather than on the collapse trajectory. For example, except for very high diversity levels, cyanobacteria diversity had no effect on the shift from the anoxic to the oxic state likely due to the abrupt change of the inhibiting substrate: sulfide was maintained by sulfate-reducing bacteria at a constantly high level such that not even the most tolerant cyanobacteria strain was able to grow (Figure 4a, e). At the tipping point to the oxic state, sulfide dropped abruptly to a level at which all cyanobacteria strains would be able to grow, resulting in immediate dominance of the strain with lowest sulfide tolerance and highest maximum growth rate (Figure 4a). Previous work showed that evolutionary trait change can even delay recovery when organisms adapt to the alternative state and then have reduced performance once environmental conditions return to original (Chaparro-Pedraza et al. 2021). Our study adds a further mechanism for why diversity might not aid recovery of a suppressed group: the greater abruptness of change of the inhibiting substrate on recovery trajectories might hinder attempts to restore ecosystems by introducing tolerant strains.

Diversity effects were often reduced or erased when multiple groups varied in traits

Antagonistic diversity effects were to be expected in a system where organisms interact via mutual inhibition. Surprisingly, however, opposing diversity effects were not necessarily caused by groups that dominated alternative ecosystem states but also by two groups that dominated the same state and interacted via facilitation. Contrary to other examples where functional groups that interact via facilitation enhance each other’s diversity effects (Eisenhauer et al. 2012), variation in sulfate-reducing bacteria erased the diversity effect of phototrophic sulfur bacteria. Apparently, when both groups varied in traits, the positive effect of sulfate-reducing bacteria on sulfide outweighed the negative effect of phototrophic sulfur bacteria, precluding the earlier shift to the oxic state. Simultaneous variation did usually not alter diversity effects when a functional group had no diversity effect on resilience, which was often the case on recovery trajectories. If it is a general phenomenon that diversity plays out less on recovery trajectories, the positive diversity effect on resilience of the dominant group would not be dampened by diversity in the suppressed group. Collectively, our results illustrate that predicting the outcome of simultaneous diversity effects of multiple functional groups can be difficult in systems where organisms interact via a multitude of complex interactions.
In natural systems, the alternative states of an ecosystem are often dominated by functional groups that strongly differ in intra- and interspecific diversity, and in how quickly new variation arises. In some of the most prominent examples of ecosystems with alternative states, macro-organisms dominate the desired state and micro-organisms the undesired state. For example, benthic macro-organisms and bacteria dominate the oxic and anoxic state of coastal ecosystems, respectively (Diaz & Rosenberg 2008), and macrophytes and phytoplankton dominate the clear and turbid state of shallow lakes, respectively (Scheffer et al. 1993). Prokaryotic and eukaryotic micro-organisms are both characterized by tremendous diversity (de Vargas et al. 2015; Locey & Lennon 2016), including diversity in functional traits (Litchman & Klausmeier 2008; Escalas et al. 2019). Furthermore, microorganisms might adapt more rapidly to new environmental conditions than macro-organisms due to their short generation times, high population sizes, and low complexity (Orr 2000; Barraclough 2015). Such greater standing variation and higher ability to adapt to stressful conditions of micro-organisms may pose an additional challenge to reversing shifts to an undesired ecosystem state dominated by micro-organisms.

Ecology, evolution, or both?

The observed replacement of strains along the gradient of oxygen diffusivity may be interpreted as changes in the abundances of species (an ecological process) or of genotypes (an evolutionary process). Furthermore, models such as ours are interpreted in both of two ways (Govaert et al. 2019): as representing ecological processes (Ceulemans et al. 2019; Wojcik et al. 2021) or eco-evolutionary feedbacks (Jones et al. 2009). To be clear, this modelling approach only includes sorting of standing heritable variation. Including mutation and recombination could be a useful next step to investigate how these evolutionary processes influence regime shifts. Chaparro-Pedraza et al. (2021) used a quantitative genetics approach to show that changes in macrophyte shade tolerance can shift the tipping point between the clear and turbid state of a lake to higher levels of stress. Despite its different approach, our study yielded similar results for two of the three functional groups. This gives some credence to interpreting models such as ours as usefully informing about the effects of evolution on ecosystem resilience.

Limitations and future directions

Naturally, our model is a simplification of natural ecosystems with assumptions that might have influenced our results. The model system focuses on three groups of bacteria; however, other functional groups (e.g. eukaryotic phytoplankton, colorless sulfur bacteria) might also have relevant effects on oxic-anoxic regime shifts (Lavik et al.2009). We demonstrated that including more than two functional groups can reveal counter-intuitive diversity effects; expanding models such as ours to include even more functional groups may therefore give further useful insights into the response of ecosystems to environmental change. Furthermore, we did not vary the values of parameters other than those of maximum growth rate and environmental tolerance. For the remaining parameters we used the same values as Bush et al. (2017) because the authors found high consistency of their modelling results and empirical observations. Also, we did not vary the shape of the trade-off between environmental tolerance and maximum growth rate. We speculate that such changes to the trade-off would have altered the pattern of strain replacement but not the position of tipping points.
An interesting avenue for future studies would be the analysis of transient dynamics, that is, the response of the system to temporal environmental change. Because we addressed questions about alternative stable states, we focused on responses of stable states, and therefore ran simulations for 1,000,000 hours at each level of oxygen diffusivity. However, in natural systems environmental conditions would not remain constant over such long timescales, and new trait variation could arise by mutation and recombination. Most of our results remained qualitatively similar when we reduced simulation length to 10,000 hours, but a further reduction in simulation length considerably changed the results (Supplementary Report Section 6). Investigation of transient dynamics in a model of shallow lakes showed that the capacity of evolution to prevent a regime shift depended on the relative rates of environmental change and trait change (Chaparro-Pedraza et al.2021). It would be interesting to test with our system if the capacity of functional diversity to increase resilience is contingent on a slow rate of environmental change. Possibly, however, the rate of change is less influential in systems such as ours because sorting of standing variation might allow a faster response to environmental change than if relevant variation has yet to arise.

Conclusions

Our results illustrate that greater trait diversity of the functional group that dominates a desired ecosystem state may increase the ability of the system to absorb environmental change before tipping to the alternative, undesired state. However, once the system has shifted to an undesired state, recovery of a desired ecosystem state might be more easily achieved by managing functional groups that dominate the undesired state than by managing groups that dominate the desired state. Trying to facilitate recovery by bringing in more stress tolerant strains of the functional group that dominates the desired state might be unsuccessful because of the abrupt change of the inhibiting substrate on the recovery trajectory. A more promising management approach could be to stimulate the growth of functional groups that co-dominate the undesired state but have negative effects on it. To identify such groups, it is useful to consider more than two functional groups in models of systems with alternative stable states. Collectively, our results highlight the importance of considering multiple interacting groups when predicting the response of ecosystems with alternative states to environmental change.