The TREES Toolbox—Probing the Basis of Axonal and Dendritic Branching

Abstract

It has now been 100 years since Ramon y Cajal described the remarkable diversity of neuronal branching. Only recently, however, have a number of rigorous formalisms emerged providing an accurate quantitative description of axonal and dendritic morphologies. We have launched a freely distributed open-source software package, the TREES toolbox, written in Matlab (Mathworks, Natick, MA), in order to help to pool together the resources offered by a wide variety of novel approaches to studying dendritic and axonal branching that have recently become available. This package introduces a simple general description of neuronal morphology as a graph and provides the basic tools to edit, visualize and analyze neuronal trees in the basis of this description. We then implement our own approach, assuming that neuronal branching can largely be expressed by local optimization of total wiring and conduction distances. We provide the corresponding modular extendable tools to automatically reconstruct neuronal branching from microscopy image stacks and to generate synthetic branched structures. The package is complemented by an extensive user interface to facilitate the generation, visualization and editing of neuronal tree structures. The TREES toolbox is structured to make it easy for other groups to integrate their own code in order to implement their own specific applications. Accurate predictions of computation in single neurons are nowadays well known to require detailed morphological representations. Tools for compartmental modelling such as NEURON, Genesis and neuroConstruct have recently facilitated the modelling of small and large neural circuits involving detailed compartmental models of the neurons. Also, a new trend highlighting the importance of morphology for better understanding of network connectivity adds to the appeal of acquiring morphologies in their full level of detail. However, obtaining the morphologies of all neurons present in one network currently remains an insurmountable hurdle. On the other hand, a number of computational methods have