The neuron classification problem

Abstract

A systematic account of neuron cell types is a basic prerequisite for determining the vertebrate nervous system global wiring diagram. With comprehensive lineage and phylogenetic information unavailable, a general ontology based on structure-function taxonomy is proposed and implemented in a knowledge management system, and a prototype analysis of select regions (including retina, cerebellum, and hypothalamus) presented. The supporting Brain Architecture Knowledge Management System (BAMS) Neuron ontology is online and its user interface allows queries about terms and their definitions, classification criteria based on the original literature and "Petilla Convention" guidelines, hierarchies, and relations-with annotations documenting each ontology entry. Combined with three BAMS modules for neural regions, connections between regions and neuron types, and molecules, the Neuron ontology provides a general framework for physical descriptions and computational modeling of neural systems. The knowledge management system interacts with other web resources, is accessible in both XML and RDF/OWL, is extendible to the whole body, and awaits large-scale data population requiring community participation for timely implementation.

Fig. 1. A systematic account of nervous system parts and connections

(a) Schematically the vertebrate nervous system has right and left halves with rostral and caudal ends, divided into gray matter regions (G,F,M,H,S) interconnected by fiber tracts (black arrows from black box region M, left half). Each region is actually characterized by a set of neuron types (d, p, t for region M, right half) with a stereotyped pattern of axonal projections forming the tracts, and typically also a neuron-type set generating strictly intraregional axon connections (local interneurons, not shown; see Fig. 1b right half). Mathematically, the number of projection possibilities is given by the combinations of axons and axon collaterals between pairs of neuron types from different regions. Let X = {A_i ,..., A_k } the set of gray matter regions, each having n_j; $j=\stackrel{‒}{1,k}$ neuron types. The number of neuron type pairs connected by axons or axon collaterals is $P={\sum }_{i=1}^k{\sum }_{j=1}^k{n}_i{n}_j;i\ne j$ and the number of possible combinations is $C={\sum }_{i=1}^P\frac{P!}{(P-i)!i!}$. Experimentally, physical connections are established currently with anterograde and retrograde tracer methods, which may help subdivide regions (Md, Mv). (b) Historically, disagreement is common about region boundaries, profoundly affecting description and interpretation of experimental results; here in a reference nomenclature neuron type d projects from region M to F, whereas in another nomenclature the same neuron type is described as having local connections in region F. (c) A complete ontology of nervous system regions and neuron types could be represented as two reference hierarchies meeting at the lowest level of each (see text and example in Fig. 3). (d) Finally, the global nervous system connection matrix is defined by data for each neuron type (or region) in a complete reference nomenclature (entities E₁-E_n) taken from (c).

Fig. 2. Conceptual schema of classification criteria associated with “is-a” relations in BAMS's Neuron ontology

It already represents >100 variables used in the literature to classify neurons in Neuron nomenclatures.

Fig. 3. Neuron type hierarchies for seven rat gray matter regions

Simple to complex examples all fit easily into the general schema provided in Figure 1c.

Fig. 4. Graphical display of rat retinal ganglion cell terms and relationships between terms (“the knowledge map”) in BAMS Neuron ontology

The graph was obtained from Graph Viz's Neato tool, which constructs graphs based on the Kawada-Kamai virtual physical model. It places an ideal spring between any pair of nodes such that its length is set to the shortest path distance between endpoints (Gansner and North, 1999). Thus, nodes with more edges tend to cluster. For full reference citations see BAMS.