GROUPING OF BACTERIA BY SIMCA PATTERN-RECOGNITION ON GAS-CHROMATOGRAPHIC LIPID DATA - PATTERNS AMONG MORAXELLA AND ROD-SHAPED NEISSERIA

Using gas chromatographic lipid data we have investigated the feasibility of the SIMCA (Soft Independent Modelling of Class Analogy) pattern recognition method as a tool in bacterial taxonomy. A collection of 118 strains of Moraxella, Moraxella-like bacteria and rod-shaped Neisseria was analyzed. The relative amounts (percentage) of 14 fatty acid and two long-chained alcohols were recorded for each strain. The 118 indiviudal data profiles (consisting of 16 numbers between 0 and 60), served as input to the SIMCA program. SIMCA offers different procedures for preprocessing of the recorded data. We found that logarithmization of the entire data set without further transformations provided the best results. The 118 examined strains fell into nine major clusters, in complete correspondence with the preconceived species/group classification. Eight of these clusters apeared rather homogeneous and well separated from each other, only a single strain deviated its group model and might be considered as an outlier. No distinction between animal and human isolates from pigs was relatively heterogeneous with somewhat unclear borderlines to the two closest neighbours, Moraxella phenylpyruvica and M. nonliquefaciens. Twenty strains of "M." urethralis which is temporarily placed in genus Moraxella, formed a very homogenous cluster clearly separated from all clusters of neisseriae and the unambigous moraxellae. A new group of rod-shaped Neisseria came out closest to, but clearly distinct from Neisseria elongata. The fatty acids octadecadienoate (18:2), heptadecanoate (17:1), n-hexadecanoate (16:0), and the three hydroxylated fatty acids 3-hydroxyhexadecanoate (3-OH-16:0), 3-hydroxytetradecanoate (3-OH-14:0) and 3-hydroxydodecanoate (3-OH-12:0) contributed most strongly to the differentiation of the groups. All 16 constituents, perhaps with the sole exception of n-heptadecanoate (17:0) were found to add significantly to the distinction of taxa. Although SIMCA is constructed for recognizing patterns among closely related objects, it was found to be robust and applicable also for revealing clusters within more heterogenous strain collections. By using the program step-wise, closely related bacterial groups could be uncovered and evaluated separately. Eight additional bacterial isolates were applied for testing of the system for identification.