Objective characterization of cells in terms of microscopical parameters: An example from muscle histochemistry

Summary

A quantitative histochemical analysis of almost 200 mouse triceps surae muscle fibres is presented, together with some data from similar surveys of rabbit skeletal muscles. In the main work, ten parameters are considered: mean diameter, and reaction intensities (estimated as apparent absorbances) of three markers for oxidative-lipolytic metabolism, three for metabolism associated with glycolysis and three for the type of myosin. Cytoarchitecture of one deposit (succinate dehydrogenase) is also noted.

Frequency histograms for each parameter and correlation analyses for all possible pairings demonstrate that markers within the same metabolic system are not truly equivalent. Therefore, fibre typing in terms of just one marker from each system cannot be independent of the markers used. Selecting the most sharply discriminatory pair-myosin ATPase (following formalin and alkaline pretreatment) and glycogen phosphorylasea — one isolates four more-or-less discrete clusters of points. Taking succinate dehydrogenase also into account, to indicate characteristic oxidative levels within clusters, one can label the indicated ‘fibre types’ (following a well-established muscle terminology) as ‘fast, essentially glycolytic’ (FG), ‘fast, oxidative and glycolytic’ (FOG), ‘fast, essentially-oxidative’ (FO) and ‘slow, essentially oxidative’ (SO). However, with either α-glycerophosphate dehydrogenase or the periodic acid-Schiff reaction (PAS) as glycolytic marker, the FO-group is not separated from the FOG; and with oxidative level as a primary clustering criterion FG and FOG groups are continuous. An entirely different basis for classification-cytoarchitecture-also suggests four fibre types but the divisions most comparable to the FG/FOG and FOG/FO boundaries are best located somewhat differently again.

Some of the techniques of formal cluster analysis are next introduced. These are ways of searching for similarities (defined in terms of various objective criteria) in large volumes of essentially multivariate data.

Within a sample, considered representative of acceptably artefact-free fibres, virtually the same four groups as before are identified. The only difference is that what were earlier grouped as the ‘most oxidative FG’ fibres are now classed FOG; the acid-pretreated myosin ATPase reaction (previously little considered) contributes to this reclassification. At the five cluster level, a strong tendency exists for this small group, designated FG(O), to appear separately. At the two-cluster level classical distinctions in terms of high/low oxidative capacity, glycolytic capacity and myosin ATPase activity are each favoured by different similarity criteria.

Surveying the whole sample of fibres, only criteria which favour non-rambling clusters produce similar results to those above. However, an artefact in one reaction, for which there is strong internal evidence, is able to explain almost all other effects. These results do not prove the biological ‘rightness’ of a 4–5 cluster pattern, but they do demonstrate its mathematical strength and the reactions upon which it depends.

The suggestion is made that cluster analysis and related multivariate statistical methods could profitably be applied to a wide range of further problems in ‘cell taxonomy’.