Multi-factors on biodegradation of phenanthrene in contaminated sediment slurry by Sphingomonas sp., a bacterial strain isolated from mangrove sediment

Abstract

Phenanthrene (Phe), a toxic three-ring polycyclic aromatic hydrocarbon (PAH), is often accumulated at a relatively high concentration in sediment and has been used as the model substrate in degradation studies. Over the past 30 years, although numerous genera of bacteria, fungi and algae capable of degrading PAHs have been isolated, information on biodegradation kinetics and optimization is still scarce. The present research aims to evaluate the effects of multi-factors on biodegradation of Phe by Sphingomonas sp. in contaminated mangrove sediment using the orthogonal design form L16(45), optimize the biodegradation condition, and predict the biodegradation potential using artificial neural networks. The present study also examines the sorption and partitioning behavior of Phe in mangrove sediment and their interactions with biodegradation using various mathematical models. Sphingomonas sp. was a bacterial strain isolated from mangrove sediment in Sai Keng with an ability to degrade PAHs. The study on multi-factors showed that salinity and inoculum size significantly effected Phe biodegradation, while the other factors like initial Phe concentration, carbon to nitrogen ratio and temperature had no significant effects. The rate and extent of Phe biodegradation were also significantly influenced by the sediment types and the presence of other inoculum such as Mycobacterium sp. but not the presence of fluorene and pyrene (other PAHs). The Phe biodegradation process could be best described by the first order rate model in both inoculated (with inoculation of Sphingomonas sp.) and control (without any inoculum) systems. The kinetic model under the optimal condition, C C e 0.1185t 0 = - , could also be used to predict Phe biodegradation in mangrove sediment slurry with the inoculation of Sphingomonas sp. at high Phe concentrations, up to 130 mg kg-1 with regression coefficient R2 of 0.9904. In all mangrove sediment slurry systems, the static sorption as well as the dynamic sorption (that is the Phe sorption during biodegradation by Sphingomonas sp.) could be described by Freundlich Equations. A degradation model, combining both sorption and biodegradation models, was further developed to predict the process of Phe biodegradation by Sphingomonas sp. in different sediment slurries. The model, kt H c e A GK k dt dc - - - = 0 2 (a 1) a , includes two sorption parameters, α (the constant of Phe sorption onto sediment) and 1/K (the diffusion resistance); a kinetic parameter k (the first order rate model constant of Phe biodegradation); and two sediment parameters, AH (the surface area unit of sediment) and G (the weight of sediment). These parameters were calculated and verified in three different types of sediment slurry systems (namely silty Ho Chung sediment with highest sorption and degradation, sandy Kei Ling Ha sediment with least sorption and medium degradation, and muddy Mai Po sediment with lowest degradation) at different Phe concentrations. Very high R2 values, ranging from 0.8380 to 0.9331, were obtained. The artificial neural networks (ANN), the universal approximations possessing the ability to approximate any real-value continuous function to any desired degree of accuracy, was firstly introduced to model and predict the Phe biodegradation behavior and process. The model used the biodegradation percentage as an output variable, and the effects of different environmental factors including C/N ratio, salinity, temperature, inoculum size, Phe concentration, sediment types (clay content) and degradation time as input variables. First, the model was built with four neurons in the hidden layer for biodegradation percentage according to the average quadratic error root-mean-square error (RMSE). Second, a portion of the data from the experiments (113 sets) was used to train the built ANN model. The predictive capacity of this trained network was then tested by the remaining data (another 111 sets), and the RMSE of the trained network was 0.0015 for biodegradation percentage. The network could predict the Phe biodegradation percentage within the ±5.5% range of the experimental values, suggesting that the ANN approach is a very useful technique for predicting Phe biodegradation under different conditions. The network also shows comparable trends in changes of Phe biodegradation with changes in respective input parameters (multi-factors), reflecting its excellent generalization capacity.