results are generally consistent in showing that axons from the remaining eye are capable of occupying the entire LGN, including territory that normally would have been innervated by the enucleated eye (refs. 8 and 9; see ref.10 for review and ref.11 for a possible exception). These observations indicate that inputs from both eyes are necessary for segregation to occur, and they suggest that LGN neurons themselves are not rigidly specified with respect to the ocular identity of their retinal innervation.
Studies, both of the mammalian visual system and elsewhere in the central and peripheral nervous systems, suggest that the transformation from a mixed to a segregated state occurs during a period in which the axonal inputs destined to segregate from each other are first capable of forming functioning synaptic connections with common postsynaptic target cells (for reviews see refs.12–16). This evidence has generated the current hypothesis, considered below, that segregation is achieved via an activity-mediated competitive process requiring the formation and elimination of synaptic connections.
In the developing retinogeniculate pathway, the cellular machinery necessary to sustain activity-driven synaptic competition is present. Ultrastructural examination of identified retinogeniculate axons has demonstrated directly that ganglion cell axons from each eye form many synapses, both in territory ultimately destined to become innervated by that eye and also in the territory that will come to belong to the other eye (17). Not only are synapses present during the fetal period, but they are also capable of functional transmission (ref.18; see also below). By electrically stimulating the optic nerves and recording from LGN neurons with extracellular microelectrodes in vitro, we found that even before the onset of segregation there is functional synaptic transmission. Moreover, during the period of extensive anatomical intermixing (E40–E59 in cat), about 90% of the LGN neurons studied physiologically received convergent excitatory inputs from stimulation of both optic nerves. In contrast, in the adult, the vast majority of LGN neurons receive excitatory input from only one nerve. The most reasonable interpretation of these observations, particularly in the context of the anatomical experiments considered above, is that prior to the completion of segregation many LGN neurons indeed receive monosynaptic excitation from both nerves. The emergence of the eye-specific layers is then accompanied by a functional change in the synaptic physiology of the retinogeniculate pathway: from binocular to monocular excitation.
It is important to note that in every mammalian species studied so far the eye-specific layers form during a period in which vision is not possible: the photoreceptor outer segments are not yet present or functional (19,20). Therefore, unlike other developing systems, such as the neuromuscular junction or the primary visual cortex, in which action potentials are evoked via use-dependent activity, here, it is necessary to postulate that activity is present as spontaneously generated action potentials. Elegant experiments by Galli and Maffei (21) indicate that this is indeed the case. They made extracellular microelectrode recordings from fetal rat retinal ganglion cells in vivo and demonstrated that ganglion cells indeed can fire spontaneously. The nature of this spontaneous activity and whether it is relayed to the LGN neurons will be considered more fully below.
There are now several excellent examples in which activity-dependent competition is known to be required for the final patterning of axonal connections in the vertebrate visual system. These include the postnatal development of the system of ocular dominance columns in layer 4 of the visual cortex (14,22–24) and the experimentally induced formation of eye-specific stripes in the optic tecta of frogs (25,26) and goldfish (27,28). In each instance, blocking retinal ganglion cell activity [by means of injections of tetrodotoxin (TTX), a blocker of voltage dependent sodium channels] or blocking synaptic transmission (by the use of glutamate receptor blockers such as 2-amino-5-phosphonovaleric acid) prevents segregation of eye input (reviewed in refs.15 and 16). By analogy with these examples, it should be possible to prevent or at least delay retinogeniculate segregation by blocking retinogeniculate transmission. Blockade was achieved by implanting osmotic minipumps in utero in cat fetuses and infusing TTX intracranially for the 2-week period during which the eye-specific layers largely form (between E42 and E56; Fig. 1). Infusions of TTX (but not vehicle) prevented the segregation of ganglion cell axons into layers (29). Moreover, as shown in Fig. 2, the axons were not simply stunted or arrested in their growth, but, rather, they grew extensively and, in fact, were about 35% larger in total linear extent than were untreated axons (30).
These observations indicate that TTX can affect the development of two basic features of ganglion cell arbor morphology: the shape of the axon (normally restricted to a cylindrical terminal arbor) and its location (normally within a single eye-specific layer). One possible explanation for how TTX has exerted its effect is that it has acted in a nonspecific manner to deregulate the growth state of retinal ganglion cells, a possibility suggested by studies showing that action potentials can have an inhibitory effect on axonal growth in vitro (31,32). However, the effect cannot be entirely nonspecific since ganglion cell axons come to an abrupt halt at the LGN border, indicating that they can still respond to many cues even in the presence of TTX. In the context of all the evidence presented above, the most reasonable explanation for the alteration in axon arbor morphology following TTX treatment is that it has acted to block spontaneously generated action potentials and synaptic transmission, which in turn are required for the formation of the normal specific patterns of axonal arborization. It is as if, in the absence of activity, the normal elimination of side branches fails to occur, and, instead, each sidebranch continues to elongate to form a significant portion of the terminal arbor. If so, then the conservative remodeling of axon arbors seen during normal retinogeniculate development is