Investigating natural attenuation of PAHs by soil microbial communities: insights by a machine learning approach

Polycyclic aromatic hydrocarbons (PAHs) due to “long-range atmospheric transport” can reach forest soil in areas far from anthropogenic activities. Information on microorganisms able to metabolize different PAHs helps in developing bioremediation strategies. We focused on the degradation of different molecular weight PAHs (phenanthrene, pyrene, and benzo[a]pyrene) in a biological natural attenuation process. Data on microbial biomass, growth, and enzyme activities, monitored along 360 days in mesocosms with forest soils (holm oak, black pine, and beech) artificially contaminated with the three PAHs, were analyzed by machine learning techniques, a powerful and novel approach in soil microbial ecology, still scantly employed. Nonlinear statistical learning methods (random forests) and linear-based classifiers (linear discriminant analysis) were applied to identify which variables were able to separate distinct groups, independently from time. Random forest was also applied to evaluate the importance of each variable in determining the PAH degradation. Nonlinear statistical learning methods more accurately found patterns related to PAH exposure in soil microbial community than the linear-based classifier. The role of fungi and their related enzymatic activities, laccase in holm oak and peroxidase in beech soil, in phenanthrene and pyrene degradation has been identified, suggesting the importance of myco-remediation in contaminated-site restoration