Interpreting 16S rDNA T-RFLP Data: Application of Self-Organizing Maps and Principal Component Analysis to Describe...

Interpreting the large amount of data generated by rapid profiling techniques, such as T-RFLP, DGGE, and DNA arrays, is a difficult problem facing microbial ecologists. This study compares the ability of two very different ordination methods, principal component analysis (PCA) and self-organizing map neural networks (SOMs), to analyze 16S-DNA terminal restriction-fragment length polymorphism (T-RFLP) profiles from microbial communities in glucose-fed methanogenic bioreactors during startup and changes in operational parameters. Our goal was not only to identify which samples were similar, but also to decipher community dynamics and describe specific phylotypes, i.e., phylogenetically similar organisms, that behaved similarly in different reactors. Fifteen samples were taken over 56 volume changes from each of two bioreactors inoculated from river sediment (S2) and anaerobic digester sludge (M3) and from a well-established control reactor (R1). PCA of bacterial T-RFLP profiles indicated that both the S2 and M3 communities changed rapidly during the first nine volume changes, and then became relatively stable. PCA also showed that an HRT of 8 or 6 days had no effect on either reactor communtity, while an HRT of 2 days changed community structure significantly in both reactors. The SOM clustered the terminal restriction fragments according to when each fragment was most abundant in a reactor community, resulting in four clearly discernible groups. Thirteen fragments behaved similarly in both reactors, eight of which composed a significant proportion of the microbial community as judged by the relative abundance of the fragment in the T-RFLP profiles. Six Bacteria terminal restriction fragments shared between the two communities matched cloned 16S rDNA sequences from the reactors related to Spirochaeta, Aminobacterium, Thermotoga, and Clostridium species. Convergence also occurred within the acetoclastic methanogen community, resulting in a predominance of Methanosarcina siciliae-related organisms. The results demonstrate that both PCA and SOM analysis are useful in the analysis of T-RFLP data; however, the SOM was better at resolving patterns in more complex and variable data than PCA ordination.