Bacterial dominance is due to effective utilisation of secondary metabolites produced by competitors

Interactions between bacteria govern the progression of respiratory infections; however, the mechanisms underpinning these interactions are still unclear. Understanding how a bacterial species comes to dominate infectious communities associated with respiratory infections has direct relevance to treatment. In this study, Burkholderia, Pseudomonas, and Staphylococcus species were isolated from the sputum of an individual with Cystic Fibrosis and assembled in a fully factorial design to create simple microcosms. Measurements of growth and habitat modification were recorded over time, the later using proton Nuclear Magnetic Resonance spectra. The results showed interactions between the bacteria became increasingly neutral over time. Concurrently, the bacteria significantly altered their ability to modify the environment, with Pseudomonas able to utilise secondary metabolites produced by the other two isolates, whereas the reverse was not observed. This study indicates the importance of including data about the habitat modification of a community, to better elucidate the mechanisms of bacterial interactions.

through diversification of resources, which was corroborated by subsequent study 2 . Further, continued study has indicated that there is a strong positive biodiversity-ecosystem function relationship (i.e. as the number of species within a system increases, the measure of functioning does also) in species-rich bacterial systems 2,[17][18][19] . How environment changes affect bacteria during respiratory infections has not been studied previously.
Our study was designed to test the hypothesis that the modification of an environment by P. aeruginosa would negatively affect the growth of other bacteria within the community. We approached this using simple, manipulated microcosms containing three bacterial populations implicated in different stages of CF-associated lung disease 20 ; P. aeruginosa is regarded as an terminal coloniser, in-so-far-as this species is associated with end-stage disease 21 , Staphylococcus aureus is commonly identified during paediatric samples, and is known to be excluded by the presence of P. aeruginosa 22 , and Burkholderia cepacia, which can be identified at all stages of lung infection, with differing outcomes dependent on the patient 23 . By using a simple microcosm system and detailed habitat profiling ( 1 H NMR) we will elucidate the mechanisms by which the bacteria interact and affect changes in community dominance.

Results and Discussion
Microbial interactions are dependent on diversity and time since inoculation. The link between increasing biodiversity and primary productivity is well-established in environmental bacteria, and our results confirmed this relationship in clinical isolates (Fig. 1a) was significant (β 48 = 1.49 ± 0.46 log 10 (colony forming units (cfu) ml −1 ) species −1 , R 2 = 0.29, p = 0.003) over time scales encompassing initial growth, but became non-significant (β 168 = 0.79 ± 0.39 log 10 (cfu ml −1 ) species −1 , R 2 = 0.14, p = 0.051) as the community established. This indicated that the influence of increasing diversity reduces over time, which is thought to be a function of changing resource availability. If we consider these positive biodiversity-productivity relationships in situ, then it would be expected that the more diverse a community, within a patient's respiratory tract, the greater the bacterial load. Research suggests, however, that this is not the case, with many patients having constant bacterial loads 5 . This is likely due to the disparate regions of the respiratory tract 24 . The interactions indicated in this study are therefore most likely localised to pockets within the lung as seen with other, structurally complex environments 25 . This could be a cause of localised airway damage, through recruitment of neutrophils 26 , but further work is needed to understand the implications of these ecological observations.
To understand the impact changing combinations had on the growth of the bacteria in the microcosms, we predicted the primary productivity of the microcosms based on the productivity of the monocultures (assuming no interaction so the growth would be additive 18,27 ) and compared to the observed productivity 2,18,28 . Here, values >1 represents positive interactions, 1 no interaction, and <1 negative interactions 18 . We found a significant difference (F 1,27 = 7.84, p = 0.009) in mean change (mean ± 1 s.e. throughout) in abundance (48 hours: −0.44 ± 0.47; 168 hours: −1.88 ± 0.42 log 10 (cfu ml −1 )), suggesting that antagonism increases the longer the bacteria cohabited (Fig. 1b). This is in contrast to previous studies that indicate the reduction in antagonistic interactions over time, due to niche differentiation 2 . We therefore postulated that this increase in apparent competition was due to a direct exclusion of one bacteria by another.
To test this postulation, the effects of each isolate on the others in co-culture was monitored by calculating the difference in abundance change from the predicted productivity for each isolate individually. Our results indicated that there was no significant change in abundance for the Pseudomonas (1.19 ± 0.004, One-sample Wilcoxon rank test (µ = 1) p = 0.783) and Staphylococcus (1.15 ± 0.004, p = 0.142) isolates, whereas the Burkholderia isolate performed significantly worse (0.46 ± 0.031, p = 0.003) when in co-cultures after 48 hours incubation (Fig. 1c). After 168 hours, all the isolates had significantly reduced in their mean observed to predicted ratio (Fig. 1d); Burkholderia (0.25 ± 0.037, p = 0.002) Pseudomonas (0.95 ± 0.030, p = 0.016), Staphylococcus (0.92 ± 0.021, p = 0.004). This result supported our postulation that the reduction in overall ecosystem function was due to the exclusion of the Burkholderia isolate by the other bacteria. With the removal of this isolate from many of the microcosms, the Pseudomonas and Staphylococcus isolates may have expended energy out-competing the Burkholderia, thereby losing any potential synergistic benefit. Whilst we observed no evidence as to the direct effects of inhibitory second metabolites there is evidence that specific antagonistic interactions do occur in similar microcosm experiments 22 . It is therefore possible that specific, non-resource mediated interactions caused the apparent extinction (isolate fell below the detection limit of 50 cfu ml −1 ), as previously observed in co-culture experiments 29 . There has been no study that investigates the relative contributions of different mechanisms of competition (i.e. ecological or chemical) despite much literature stressing the interplay 30-33 , however, there are an increasing number of studies 22,27,29,[34][35][36] , including this one, where this holistic approach would be greatly informative.
Niche overlap. Having found that the Burkholderia isolate was excluded by the other two isolates, we used 1 H NMR to understand how this was possible, and tested the mechanism of competition as either due to competitive exclusion by removal of nutrients or through the production of toxic secondary metabolites. By taking integrals of the 1 H NMR spectra at 0.05 ppm intervals (Supplementary Figs. S1 and S2), and calculating the differences between the monocultures and the blank samples we could identify relative changes in the media. At the first time  (Fig. 2d-f). Using these data, the niche overlap was calculated between the isolates. There was a significant (χ 2 3 = 30.38, p < 0.001) increase in overlap in the latter time point across the combinations, suggesting that the bacteria were competing for the same resources. This may have been due to the low diversity media (carbon sources were almost entirely amino acids 11 ) the bacteria were grown in, therefore the opportunity for niche diversification was limited. This increase in overlap was strongly, albeit non-significantly, correlated (Rho = −0.619, p = 0.115) with a reduction in the mean observed to predicted ratio in each of the microcosms (Supplementary Fig. S3). Interestingly, the Pseudomonas isolate was observed to have a greater number of negative integrals than either of the Burkholderia and Staphylococcus isolates. This is likely due to P. aeruginosa being an generalist through its ability to adapt rapidly to the lung environment 37 . With the increasing overlap in resource utilisation, this is an example of competitive exclusion in bacteria 38 , and further explains the extinction of the Burkholderia isolate.
Changes in environment are indicative of community structure. To understand the interplay between environmental modification and changes in bacterial interactions, we plotted the interactive index (predicted/observed) 18 using the integrals measured in microcosms with monocultures added together as the predicted. This indicated that as the isolates grew, the number of significant interactions (change in integral > mean change ± 2 s.d.) increased significantly (χ 2 3 = 32.39, p < 0.001) from 48 (Fig. 3a) to 168 hours (Fig. 3b). To demonstrate the differences between samples, the readings from the first were subtracted from the second time point. The results indicated that the primary Principle Component Analysis (PCA) clustering (Fig. 4a) was based on the abundance of Pseudomonas (R 2 = 0.32, p = 0.006) and Staphylococcus (R 2 = 0.21, p = 0.023) at the final time point, but not Burkholderia (R 2 = 0.08, p = 0.314) (Fig. 4b). The primary PCA axis was primarily correlated to the abundance, or absence, of the Pseudomonas isolate within a microcosm (Fig. 4b), and was determined by changes in 10 integrals (Fig. 4c). The mean change in integral was plotted against the primary PCA axis for microcosms depending on Pseudomonas presence (Fig. 4d). For microcosms containing Pseudomonas, these values had a significant positive relationship (β = 16.91 ± 0.82, F 1,145 = 419.53, p < 0.001); conversely, those microcosms where Pseudomonas was absent had a significant negative relationship (β = −8.55 ± 1.08, F 1,145 = 62.96, p < 0.001). This detailed investigation of the 1 H NMR profiles suggests a potential mechanism that facilitates the domination of Pseudomonas, as evident in these microcosms. Whilst we cannot exclude the possibility of the secretion of inhibitory molecules (e.g. 29 ), our data highlight the ability of the Pseudomonas isolate to utilise secondary metabolites created by other bacteria. We do not observe a reciprocal relationship between the Burkholderia and Staphylococcus isolates and metabolites produced by the Pseudomonas isolate. Putative identification of the 1 H NMR peaks, suggest that those that significantly increase in the presence of Pseudomonas are derivatives of the amino acid serine, potentially the quorum sensing molecule homo-serine lactone 39 . As these data indicate an increase in the serine-associated peaks at the second time point compared with the first, this would equate to the Pseudomonas isolate being at stationary phase with no fresh intake in nutrients, therefore a decrease in metabolism would be beneficial 40 . Conversely, the peaks produced by the Burkholderia and www.nature.com/scientificreports www.nature.com/scientificreports/ Staphylococcus isolates, that are used in the microcosms with Pseudomonas, correspond with protons attached to a carbon atom, adjacent to an aromatic ring. There is evidence that aromatic compounds have an inhibitory effect on Pseudomonas species' competitiveness 41 . Due to the evolutionary history shared by these isolates (i.e. having come from the same environment, potentially in close proximity to one another), the Pseudomonas isolate could have evolved to metabolise these compounds produced, as a defence mechanism, by its competitors. Further work would be required to know whether this is a true mechanism. Our data strongly suggest that cross feeding occurs between the bacteria, on an ecological timescale and not limited to evolutionary times 16 , ultimately to the advantage of the Pseudomonas isolate. This media was used to mimic the chemical conditions experienced in the lung environment from which these bacteria were isolated 11,42 . Previous studies have indicated that the transcriptomic expression profiles are highly similar in synthetic cystic fibrosis medium to those observed in their natural environment 43,44 . As such, we believe that this mechanism of unidirectional cross-feeding is likely to be present in natural systems and should be investigated further.

conclusion
Our results confirmed previous results that the influence of increasing richness weakens over time in bacterial microcosms 2 , but there was a significant positive relationship between biodiversity and cell numbers. This relationship between increasing productivity and diversity is present despite interactions between bacteria are predominantly negative. Previous bacterial microcosm experiments, using environmental isolates 2,17,19 , have supported the presence of this positive relationship despite antagonistic interactions being the most prevalent. Conversely, the only other microcosm experiment that used clinically-derived isolates with a similar design to this 28 indicated that there was a greater level of synergistic interactions occurring. We consider that our results suggest an ecological mechanism as to how P. aeruginosa dominates an infectious community which is in contrast to the production of bacteriocins 45 , however, we saw no evidence of this in our study.
CF is a genetic disease, and thus affects the individual throughout their life, with a multitude of clinical interventions occurring dependent of disease severity and progression 46 . We focused our study on potential ecological interactions between some of the most clinically relevant bacteria based on different resource utilisation profiles. The potential impact of treatment regimens on the bacterial community, and resulting changes in interactions, should be the focus of subsequent studies. Previous ecological investigations based on environmental isolates have indicated that interactions are affected by abiotic stresses 47,48 , but the magnitude of the effect is predicated on the communities' resistance, or resilience, to the perturbation 49 . In CF, antibiotic treatments can cause shifts in community structure that can influence the trajectory of the community, and the selection of resistant strains to become dominant 4,46 .
This initial study is the first to utilise 1 H NMR to postulate mechanisms that dictate interspecies competition in restricted microcosm systems. By using this analytical technique, we were able to identify the interactions between the species based on the utilisation of the media to understand why the apparent dominance of Pseudomonas was observed. Isolates of Pseudomonas have been reported to be metabolically diverse 11,28 , which has been purported as a method to identify this genus 11 . Further studies are required to validate these findings, but the incorporation of microbe dependent environmental change is vital to assigning mechanisms to the observed ecology.  Fig. S4). Each monoculture was then diluted to an optical density (λ = 600 nm) of 0.1 in fresh SCFM, and combined so that the bacterial load of each inoculum remained constant regardless of the number of species present 28 . The bacterial mixtures were inoculated 1:49 (~1 × 10 3 cfu ml −1 ) in 5 ml microcosms. The three species were combined in a fully factorial design; each species was grown as a monoculture (species richness (δ) = 1, n = 3), with each of the other separately (δ = 2, n = 3) and together (δ = 3, n = 1). Each set of microcosms included a negative (no bacteria) control microcosm and were independently replicated four times (32 microcosms in total). Microcosms were incubated statically for 168 hours with samples taken at 48 and 168 hours.
Assessment of bacterial growth. The number of colonies of the three bacterial species was recorded at 48 (first time point) and 168 hours (final time point). At each of these times, bacteria were serially diluted in sterile phosphate buffered saline (PBS, pH 7.8, Sigma-Aldrich), following which the samples were inoculated 50 onto all three selective media and incubated at 37 °C for 48 hours.
Biological NMR spectroscopy. Supernatant samples were mixed with D 2 O solvent (99% atom%D, Sigma-Aldrich) at a 5:1 ratio. NMR data were collected using a JEOL JNM-ECS400 (JEOL, Tokyo, Japan) NMR spectrometer operating on Delta software (version 5.0.4.5), using a proton resonance frequency of 400 MHz and referenced to the relevant residual solvent peak of water (δ 4.79). Spectra were acquired using a 45° flip angle and a relaxation time of 5 seconds, each spectrum was acquired using 32 scans. 1 H NMR data were collected using a pre-saturation pulse applied to the water signal to minimise its intensity in the resulting 1 H NMR spectrum. Spectra were analysed using MestReNova software (version 12.0.4). Integrals were calculated at 0.05 ppm intervals to approximate the abundance of protons at each section of the spectrum. Statistical analysis. All statistical analyses and visualisations were performed in R (v3.6.0) 51 using the vegan (v2.   52 and lmerTest (v3.1-0) 53 packages. All bacterial count data were logarithm (base = 10) transformed prior to analysis and the assumptions for parametric statistics were assessed visually. If these assumptions were not met, non-parametric equivalents were used as stated in the text. In order to account for the repeated sampling of the microcosms a linear mixed effect model (REML = T) was used with "microcosm identity" entered as a random variable. For 1 H NMR data we used subtraction to highlight differences between the two sampling times (t 2 -t 1 ). Principle Component Analysis ordinations were calculated using the prcomp() function with correlations assessed using envfit() commands with 999 permutations. All models were simplified in accordance with parsimony with stepwise removal of complex interactions until the simplest significant model remained 54 .