The specificity of synaptic connections is established by two consecutive processes: axonal pathfinding and synaptic cell adhesion10. A simple consideration shows that of these two processes, axonal pathfinding is more important than synaptic cell adhesion, although obviously both are essential. This consideration is that synaptic cell adhesion can only operate at a distance of ^ 100 nm, which even on the scale of the densely packed neuropil is a very short distance. Whereas over short distances axons and dendrites have equal roles in establishing synaptic specificity, over long distances axons alone mediate the specificity of connections. Thus most of the specificity of synaptic connectivity must depend on guiding an axon to its correct target, with the actual formation of synaptic connections being secondary.
Axonal pathfinding is well studied, and much has been learned about its complex molecular determinants (e.g., see refs. 11-13). Axonal guidance may even contribute to the determination of the dendritic domain to which an axonal input is directed. Many key mechanisms of axonal guidance have been established, providing the molecular basis for Sperry's pioneering chemoaffinity hypothesis10.
After axonal guidance is completed, a growth cone enters the target neuropil and becomes competent to form a synapse. In the densely populated neuropil, most axons and dendrites are not destined to form synaptic connections with each other, but chance encounters between these axons and dendrites must be extremely frequent. How does the emerging nerve terminal select the right target neuron and the right dendritic domain in the densely packed neuropil? For example, how does a thalamic input in the CA1 region of the hippocampus select between pyramidal cell dendrites in the stratum lacunosum or the dendrites of at least ten different types of interneurons (Figure 0.1; Colorplate 1)? Two not mutually exclusive hypotheses can be proposed to explain synapse selection by axons that are primed for synapse formation. First, specific recognition molecules may trigger synapse formation. Second, neurons may form promiscuous synaptic connections and then eliminate the wrong connections, i.e., act by a sampling mechanism that establishes and dissolves synapses constantly as the axon moves along. It seems likely that a combination of both hypotheses is correct.
An important clue to the nature of synapse formation comes from artificial in vitro experiments. When a single hippocampal or cortical neuron is plated on microislands of glia cells, it forms hundreds of synapses, the so-called autapses, onto themselves. Although autapses do occur physiologically, they are rare in vivo (e.g., in cortical pyramidal neurons, autapses account for <1% of the total synapses; ref. 14). Thus in the absence of appropriate targets, a neuron forms synapses onto itself by default, whereas in the presence of appropriate targets, it does not. This finding implies that synapse formation is a competitive process: when a neuron cannot find a better partner, it prefers connecting to itself instead of remaining unconnected. Thus synapse formation is not necessarily governed by a strict cell adhesion code, but at least partly operates by a mechanism that selects stable synapses based on a competition between targets (i.e., a sampling mechanism).
It is clear, however, that 'sampling and selection' of synapses is not sufficient to explain the generation of patterns of synaptic connectivity. If all neurons randomly formed synapses that were then largely eliminated by a selection mechanism, this would simply shift the specificity issue from the stage of initial synapse formation to the stage of synapse elimination. Whereas activity-dependent synapse elimination is a plausible mechanism to achieve a homogeneous distribution of synaptic inputs and outputs between two classes of neurons, it is not plausible as a mechanism that eliminates a huge number of random synapses. Apart from theoretical considerations, an experimental result that demonstrates this point is the observation that in Munc18-1 deficient mice, all synaptic activity is absent but the pattern of synaptic connectivity is nevertheless not randomized15. It seems likely that the major role of the sampling mechanism is to equalize synaptic inputs and outputs for a given class of neuron.
Was this article helpful?