Computer assisted classification and identification of actinomycetes
Three computer software packages were written in the C++ language for the analysis of numerical phenetic, 16S rRNA sequence and pyrolysis mass spectrometric data. The X program, which provides routines for editing binary data, for calculating test error, for estimating cluster overlap and for selecting diagnostic and selective tests, was evaluated using phenotypic data held on streptomycetes. The AL16S program has routines for editing 16S rRNA sequences, for determining secondary structure, for finding signature nucleotides and for comparative sequence analysis; it was used to analyse 16S rRNA sequences of mycolic acid-containing actinomycetes. The ANN program was used to generate backpropagation-artificial neural networks using pyrolysis mass spectra as input data. Almost complete 1 6S rDNA sequences of the type strains of all of the validly described species of the genera Nocardia and Tsukamurel!a were determined following isolation and cloning of the amplified genes. The resultant nucleotide sequences were aligned with those of representatives of the genera Corynebacterium, Gordona, Mycobacterium, Rhodococcus and Turicella and phylogenetic trees inferred by using the neighbor-joining, least squares, maximum likelihood and maximum parsimony methods. The mycolic acid-containing actinomycetes formed a monophyletic line within the evolutionary radiation encompassing actinomycetes. The "mycolic acid" lineage was divided into two clades which were equated with the families Coiynebacteriaceae and Mycobacteriaceae. The family Coiynebacteriaceae contained the genera Cotynebacterium, Dietzia and Turicella and the family Mycobacteriaceae the genera Gordona, Mycobacterium, Nocardia, Rhodococcus and Tsukamurella. It was clear from the 1 6S rDNA sequence data that Nocardia pinensis was misclassified in the genus Nocardia and that TsukamurelIa wratislaviensis belonged to the genus Rhodococcus. The genus Nocardia formed a distinct dade that was clearly associated with the genus Rhodococcus. Two sublines were recognised within the Nocardia dade; one consisted of Nocardia asteroides and related taxa and the other of Nocardia otitidiscaviarum and allied species. The two sublines are distinguished by nucleotide differences in helix 37-1. The type strains of all of the Nocardia species contained hexahydrogenated menaquinones with eight isoprene units in which the two end units were cyclised. Actinomycetes selectively isolated from an activated sludge plant showing extensive foaming were the subject of a polyphasic taxonomic study. The sludge isolates, which clearly belong to the genus Tsukamurella on the basis of 1 6S rRNA data, contained highly unsaturated long chain mycolic acids and unsaturated menaquinones with nine isoprene units, properties consistent with their classification in the genus Tsukamurella. Six representative isolates and marker strains of Tsukamurella paurometabola were the subject of a numerical phenetic taxonomic study. The test strains were assigned to four groups in the simple matching coefficient, unweighted pair group method with arithmetic averages analysis. The sludge isolates formed a homogeneous cluster with the three remaining clusters composed of Tsukamurella paurometabola strains. Excellent congruence was found between these numerical taxonomic data and results derived from corresponding studies based on Curie point pyrolysis mass spectrometric and whole-organism protein electrophoretic analyses. The combined data suggest that the sludge isolates form the nucleus of a new species of the genus Tsukamurella and that Tsukamure!!a paurometabola is a heterogeneous taxon. Representatives of three putatively novel streptomycete species isolated from soil were used to develop and evaluate an identification system based on Curie point pyrolysis mass spectromety and artificial neural network analysis. The test strains consisted of sixteen target Streptomyces strains and one hundred and thirty-eight actinomycetes belonging to the genera Actinomadura, Mycobacterium, Nocardia, Nocardiopsis, Saccharomonospora and Streptosporangium. It was clear from the identification results that artificial neural network analysis was superior to conventional procedure based on principal component and canonical variate statistics. The problem of misidentification of some of the non-target strains was solved by the development of a neural network topology which contained an output neuron designed to detect non-target pyrolysis mass spectrometric patterns. The pyrolysis mass spectrometry-artificial neural network system was evaluated using thirteen fresh streptomycete isolates and found to be capable of long-term identification of the target strains.