Toxicity drives facilitation between 4 bacterial species

Significance Microbial communities play a major role in our lives, but we understand little about how species within them interact. Here, we studied 4 bacterial species that could degrade toxic industrial fluids. We expected these species to compete, but instead found that they all benefited from each other: Alone, only 1 species could survive, while together they all grew and degraded the fluid. However, this result depended on the environment. Positive interactions were most common in the toxic fluid, and, if we made survival easier, for example by adding nutrients, bacteria began to compete. Our findings provide a simple intuition: In a harsh environment where single species are unable to grow, the only option becomes to work together.

and the Area Above the Curve (AAC, the integral between the control and the biotic curve) represents degradation efficiency. Briefly, 1ml of MWF emulsion was harvested at the beginning of the experiment, and on days 1-6, 8, and 12, centrifuged (16,000 rcf for 15 minutes) to remove suspended cells (we found that cellular material increases the COD, Fig. S7). Centrifugation separated the MWF into two liquid phases. The top phase was carefully pipetted and discarded, while 200 µl of the second phase was added to NANOCOLOR COD tube tests, detection range 1-15 g/l by Macherey-Nagel (ref: 985 038), heated at 160 • C for 30 mins, cooled to room temperature, and the color change quantified on a LASA 7 100 colorimeter (Hach Lange, UK). This protocol was developed with the help of Guillermo Osuna.
Adapting bacteria to MWF medium. A. tumefaciens and C. testosteroni were grown in MWF medium as described above for 7 days (28 • C, 200 rpm) in five replicate mono-cultures. After 7 days, 30 ml of fresh MWF medium was prepared and 300 µl of the week-old culture transferred into it. This was repeated every week for a total of 10 weeks. At the beginning and at the end of every week, population sizes were quantified using CFUs as described above. After 3 weeks, three replicate populations of A. tumefaciens had gone extinct (Fig. S10). After 10 weeks, one colony was isolated from the first replicate of the evolved populations of A. tumefaciens and C. testosteroni, and the interactions between them quantified.
Resource-explicit mathematical model. We consider a community of n distinct species, where the change in abundance S i of species i is determined by a growth function ρ i and mortality µ i which depend on the concentrations C N and C T of the nutrient and toxin as shown in Fig. 2A. Nutrients decrease as a function of the species' growth via the biomass yield Y i , while toxins decrease according to the species' production rate δ i of enzymes that degrade the toxin as well as a passive uptake rate κ i . A fraction f i of the collected nutrients are invested into active degradation and the rest into growth. This results in the following set of differential equations: We assume that the growth and death rates saturate with increasing nutrient or toxin concentrations as: for the nutrient and toxin with half-saturation concentrations K N , K T and maximum growth or death rate r max , m max . We implemented the model in Python v3.6 using the SciPy library v1.0 and solved with standard ODE solvers for a set of parameters and initial conditions as listed in Table S3. Fig. S17 shows how changes in these parameters and initial conditions affect the outcome of the model. To generate the heat plot in Fig. 2C, we calculated the difference in the AUC of the simulated time-series, between a simulation with initial abundance S 1 = S 2 = 1 and another with initial abundance S 1 = 1 and S 2 = 0.     Fig. S2. Pairwise co-cultures in triplicate of C. testosteroni (Ct) together with six other strains that were similarly isolated from MWF, but had presumably never interacted previously with C. testosteroni. In co-cultures, the first strain mentioned in the label is the one plotted, while the partner strain, if any, is in brackets (e.g. Ct(A. caviae) is the growth curve of Ct in the presence of A. caviae). All six strains quickly died in MWF if cultured alone, but in the presence of C. testosteroni, four out of the six grew better, as shown in Fig. S3. This suggests that the positive interactions at least between C. testosteroni and the other three strains shown in Fig. 3 are unlikely to be a product of their co-evolution, but are most likely accidental. Species identities are: (A) Aeromonas caviae, (B) Delftia acidovorans, (C) Empedobacter falsenii, (D) Klebsiella pneumoniae, (E) Shewanella putrefaciens, (F) Vagococcus fluvialis. Sabrina Riveira helped PP collect these data.
. AUC calculated based on the data in Fig. S2. Four out of the six strains had a higher AUC when grown together with C. testosteroni, while C. testosteroni also improved its growth in some cases. Note that a mono-culture of C. testosteroni was not included every time, but a number of repeats of its AUC are pooled together. This makes a statistical comparison for Ct more difficult, but supports our overall conclusion that positive interactions are likely accidental.  . Repeat of the mono-culture and pairwise co-cultures presented in the main text ( Fig. 1) but where co-cultures contained the same total inoculum volume as the mono-cultures (e.g. At200 was inoculated with 200µl of At, while At(Ct) was inoculated with 100µl of At and 100µl of Ct). We also included a treatment where we doubled the volume of the inoculum for the mono-cultures (e.g. At400). (A)-(D) Growth curves for each species, (E) AUC for each treatment and species, (F) interaction network comparing mono-cultures and co-cultures of the same total inoculum volume (e.g. comparing At200 with At(Ct) of 100 each). Arrow width and color is as described for Fig.   3. For all species, doubling the inoculum volume of the mono-culture resulted in AUCs that were significantly larger (all four P < 0.015).           Fig. 5 in the main text but rather than the COD data and the AAC, here we use the total CFU/ml data and the our growth measure (AUCs) of each monoand co-culture to observe at what species number the overall population size saturates. In (A) we ask whether the carrying capacity increases with additional species and find that in MWF it reaches its maximum at 2 species, while in MWF+AA, it is already reached at 1 species. (B) The sum of AUCs in mono-culture do not always predict the co-culture. Again, in MWF, populations are sometimes larger in co-culture than expected by the additive null model, while for MWF+AA, the null model is a good predictor for the total in co-culture. Note that the scale here is logarithmic.  Table S1. Chemical composition of media used. Table S2. P-values behind interactions in Fig. 3 are shown in Dataset S1.   . Row-wise from top left to bottom right: maximum growth rate r max , maximum death rate m max , degradation investment f , inverse of biomass yield Y , degradation rate δ, passive degradation rate κ, nutrient and toxin saturation constants K N , K T . When one parameter is changed, the rest are kept at the standard value found in Table S3.

Supplementary Note S3: Differential toxin degradation allows for density-independent facilitation
The original formulation in the main text (copied below for ease of reading) is valid for n species affected by a single toxin and sharing a single nutrient. For simplicity, in the main text we assume that the two species are identical in all parameters, while in practice they will probably vary in their mortality rates m, toxin degradation rates δ, growth rate r and biomass yield Y . Recall that the we use Monod kinetics for the growth function ρ i (C N ) = r i C N /(C N + K N ), (2A) and toxin-dependent Importantly, we have shown in the main text that both positive and negative interactions may occur in an environment that includes toxins, where the positive interactions relies on increasing the total population size from mono-to co-culture. If instead the total population size is kept constant by inoculating the n species co-cultures with initial abundance 1 n , only negative interactions will be found, because the same resources will be divided amongst all strains. On the contrary, our experiments revealed positive interactions even when the total population size was held constant (Fig. S5). To better capture these positive interactions when scaling the inoculum in this manner, we modified the model to include two toxins T 1 and T 2 , which affect both strains equally, as shown in Eq. (S4a)-Eq. (S4e) and Fig. S18A, where T 1 is degraded by strain S 1 while S 2 degrades T 2 . At intermediate nutrient concentrations, we again observe two-way positive effects of being in co-culture (Fig. S18) since the degrading member of each toxin facilitates the environment for the other by reducing the concentration of one of the two toxins. The simulated biomass shows positive interactions where both species are better off in co-culture even while the total inoculated biomass is the same between mono-and co-culture. In the example run, the effect of nutrients and toxins is symmetric by the choice of equal parameters for S1 and S2, and the two species only differ in their degradation abilities for the two toxins.