Support vector regression model of wastewater bioreactor performance using microbial community diversity indices: effect of stress and bioaugmentation.

The relationship between microbial community structure and function has been examined in detail in natural and engineered environments, but little work has been done on using microbial community information to predict function. We processed microbial community and operational data from controlled experiments with bench-scale bioreactor systems to predict reactor process performance. Four membrane-operated sequencing batch reactors treating synthetic wastewater were operated in two experiments to test the effects of (i) the toxic compound 3-chloroaniline (3-CA) and (ii) bioaugmentation targeting 3-CA degradation, on the sludge microbial community in the reactors. In the first experiment, two reactors were treated with 3-CA and two reactors were operated as controls without 3-CA input. In the second experiment, all four reactors were additionally bioaugmented with a Pseudomonas putida strain carrying a plasmid with a portion of the pathway for 3-CA degradation. Molecular data were generated from terminal restriction fragment length polymorphism (T-RFLP) analysis targeting the 16S rRNA and amoA genes from the sludge community. The electropherograms resulting from these T-RFs were used to calculate diversity indices - community richness, dynamics and evenness - for the domain Bacteria as well as for ammonia-oxidizing bacteria in each reactor over time. These diversity indices were then used to train and test a support vector regression (SVR) model to predict reactor performance based on input microbial community indices and operational data. Considering the diversity indices over time and across replicate reactors as discrete values, it was found that, although bioaugmentation with a bacterial strain harboring a subset of genes involved in the degradation of 3-CA did not bring about 3-CA degradation, it significantly affected the community as measured through all three diversity indices in both the general bacterial community and the ammonia-oxidizer community (α = 0.5). The impact of bioaugmentation was also seen qualitatively in the variation of community richness and evenness over time in each reactor, with overall community richness falling in the case of bioaugmented reactors subjected to 3-CA and community evenness remaining lower and more stable in the bioaugmented reactors as opposed to the unbioaugmented reactors. Using diversity indices, 3-CA input, bioaugmentation and time as input variables, the SVR model successfully predicted reactor performance in terms of the removal of broad-range contaminants like COD, ammonia and nitrate as well as specific contaminants like 3-CA. This work was the first to demonstrate that (i) bioaugmentation, even when unsuccessful, can produce a change in community structure and (ii) microbial community information can be used to reliably predict process performance. However, T-RFLP may not result in the most accurate representation of the microbial community itself, and a much more powerful prediction tool can potentially be developed using more sophisticated molecular methods.