There are four important pieces to the developmental sequence for a neural network. First, the neurons must be able to identify and locate themselves. Second, axons must sprout and find their targets. Third, the synaptic connections must form and properly adjust themselves in relation to environmental input. And fourth, as information begins to flow the network must continually adjust itself to accommodate the characteristics of incoming data.

Neuron self-identification and self-localization are brought about by the interaction of genetic information inside the cell, with chemical gradients outside the cell. Each cell knows what it is genetically, through inheritance from the stem cell lineage. But each retinal ganglion cell needs to know where it is in relation to the fovea and the blind spot, for proper wiring to take place, and for this purpose there are horizontal and vertical gradients of marker molecules in the retina during development, that interact with the growing neurons and help position them and organize them into modules.

Sprouting an axon is akin to extending filopodia, but the axon grows up a chemical gradient and then finds its target using a different marker. The growth marker is usually released by the target cells into the extracellular environment so the axons can find it, but the targeting marker is usually expressed on the membranes of the postsynaptic cells, so once the axons arrive in the proper region they will "sprout" in such a way that they're likely to find the desired markers.

To form synaptic connections, the tips of the axons proliferate into a "growth cone" that typically creates hundreds to thousands of terminal boutons. Each bouton finds an appropriate target, and often these targets are highly localized geometrically, for example a certain class of axon may target only the outer third of the apical dendrites of layer 5 pyramidal cells. Once the synapses have formed, there is a competitive process that results in pruning, some of the early connections die off and the ones that remain are strengthened.

Finally, each synapse must decide whether to remain plastic or whether to turn off its adaptive machinery. In areas of the brain that develop with "critical periods", the adaptive machinery is often turned off at the end of the critical period, and what remains is a hard-wired system with perhaps some residual forms of plasticity within individual synapses.

In contrast to historical belief, our brains do in fact generate new neurons, constantly, all the time. The process of adult neurogenesis is best studied in the hippocampus and surrounding areas. The rate of new neuron growth in these areas has been estimated at about 1000 per day. Why are new neurons needed? What's wrong with the old ones?

Electro-Reception

Considering the brain in relation to a mapping of electrical activity, synapses are relatively easy to model as long as they're isolated. If we can take every synapse in isolation, we can conveniently use matrix multiplication to calculate the influence of synaptic weights on neural firing. However the time scales are in the msec range, and we're still looking for something more precise. One interesting avenue of investigation is the effect of electromagnetic fields on neural activity. Information about this comes from two areas of research: one is transcranial stimulation, which can be focused onto small groups of neurons but is rather non-specific within those groups, and the other is electroreception, which is a capability of some sharks, skates, and eels. Electroreception is especially interesting because it entirely involves wetware, the ionic currents indicating electric fields travel in the ocean itself, and the Ampullae of Lorenzini containing the electroreceptors have an internal milieu much like seawater.

Electro-reception works through a voltage gradient between one side of the receptor cell and the other. This in turn triggers calcium channels that generate nerve impulses. In the species Apteronotus leptorhynchus (the weakly electric brown ghost knifefish), electroreception is sensitive down to 5 nV/cm (about a billionth of a volt). Contrast this with the trans-membrane electric field in a typical human neuron, which is millions of volts per meter.

The possibility of direct electromagnetic effects on calcium channels can not be ruled out. Even more intriguing is the relationship between calcium and astrocytes, which are connected by gap junctions. Ultimately the exposure to extracellular fields is linked to currents in the extracellular milieu, and one can calculate the field generated by a neuron when it fires.