Complexity of branching dendritic trees: dependence on number of trees per cell and effects of branch loss during sectioning.

We have investigated whether the complexity of dendritic trees is correlated with the number of primary dendrites per neuron (trees per cell). In estimating the average number of branches of centrifugal orders 1-5 per tree we used statistical methods to compensate for loss of parts of trees during sectioning. Limitations of these methods are discussed. Neurons from four populations, stained by the Golgi-Cox method, were examined: stellate cells from layer IV, area 17 of visual cortex, in normal and dark-reared cats; the pyramidal cells from layer V, somatosensory cortex, in two strains of rats. In all four groups of neurons the average number of branches of higher orders (3, 4, 5) per tree tended to be smaller in neurons bearing more trees. Thus all trees from a population of neurons should not be assumed to be equivalent. The decreasin high-order branches per tree tended to offset the increase in number of trees per cell. In three of the four groups these opposed tendencies maintained the average number of high-order branches per neuron nearly independent of the number of trees per cell. Natural selection may have favoured near-constancy in the number of high-order branches to reduce dispersion among neurons of one type in functional input-output rleations.