The dorsal lateral geniculate nucleus of the thalamus (dLGN) is the gateway through which visual information is transmitted from the retina to the cortex; therefore, visual perception relies on the ability of dLGN neurons to faithfully relay the specific features of the visual world that are encoded by the retina. Accordingly, the connections between retinal ganglion cells (RGCs) and dLGN neurons are highly precise and organized. , For instance, RGC projections are arranged such that neighboring cells in the retina innervate neighboring cells in the dLGN creating a retinotopic map in the thalamus that is capable of maintaining the spatial features of the visual scene. Furthermore, the resolution of the retinal map is also maintained in the dLGN since at maturity few RGCs converge onto dLGN neurons; consequently, dLGN receptive fields are similar to those of RGCs in their size and structure. Another key feature of retinogeniculate organization, which is important for binocular vision, is that the axons arising from each eye are segregated within the dLGN. As a result, mature dLGN neurons are monocularly driven. , Finally, some cell-type-specific organization also exists in the dLGN since different types of dLGN neurons reside in distinct laminae and receive inputs from different subtypes of functionally distinct RGCs.
How does precise retinogeniculate circuitry get established during development? Numerous experiments have demonstrated that many aspects of retinogeniculate connectivity are initially imprecise, and that the immature circuit must subsequently undergo refinement in order to achieve its finely tuned mature form. Much of this refinement occurs before the onset of vision, , and it is now well established that this early refinement requires both spontaneous retinal activity and molecular cues that are present over development.
Retinogeniculate projections are refined during development
Anatomical studies have shown that when the axons from the two eyes initially invade the dLGN their arbors are extensively overlapped, , and functional studies have demonstrated that these overlapping projections give rise to binocularly innervated dLGN neurons. , After this initial connectivity is established there is a specific developmental window during which the axons from each eye segregate into non-overlapping territories. , This “eye-specific segregation” results in a pattern of ipsilateral and contralateral projections that is highly stereotyped both in size and placement within the dLGN. Eye-specific inputs also segregate functionally. As RGC axons become restricted to the appropriate region of the dLGN their functional synaptic connections in the inappropriate region are lost, resulting in monocularly driven dLGN neurons. , In mouse, electrophysiological recordings have demonstrated that dLGN neurons are initially weakly innervated by 12–30 RGCs which get refined down to 4–6 inputs from each eye as segregation proceeds, and finally to just 1–3 strong inputs from a single eye at maturity. While the convergence of RGC inputs onto individual dLGN neurons gets reduced, the nearest neighbor relationships among RGCs are maintained, and thusly, the retinotopic map is sharpened.
In addition to eye-specific segregation, RGC axons undergo further restriction of their arbors into functionally distinct sublaminae. For example, each major class of ganglion cell consists of two subtypes: cells that are depolarized (ON ganglion cells) or hyperpolarized (OFF ganglion cells) by light onset. At maturity these parallel ON and OFF pathways are segregated in the dLGN at the level of single neurons; individual dLGN neurons generally respond to increments or decrements in light but not both. In some species, such as the ferret, dLGN neurons receiving ON input and OFF input reside in two distinct sublaminae within each eye-specific layer, and these sublaminae develop just after eye-specific segregation and before opening. The emergence of On and Off sublaminae before the onset of vision precludes a physiological assessment of whether dLGN neurons transiently receive converging ON and OFF synaptic inputs. However, this is likely because before these On and Off sublaminae form RGC axons initially arborize over both the inner and outer half of each eye-specific layer.
Among the classes of RGCs that project to the dLGN are Y cells which process movement, X cells which are responsible for image acuity, and a heterogeneous population of W cells. Whether individual dLGN neurons initially receive synaptic input from these multiple classes of ganglion cells is unclear. In the primate, however, magnocellular and parvocellular pathways project to distinct regions of the dLGN from early stages as their axons innervate targets. , This suggests that during primate development single dLGN neurons may receive inputs from just one class of RGC. This specificity could be due, at least in part, to the generation of different retinal ganglion cell classes at different times during development, and to the fact that the axons of parvo cells reach the dLGN before those of magno cells. In addition, in primates the axons from the two eyes may initially innervate the dLGN in a somewhat eye-specific manner; thus, the primate retinogeniculate connection may require less refinement than that of other mammals.
Interestingly, studies also suggest that once the mature pattern of anatomical and functional connections has been established they must be actively maintained. Taken together, these data indicate that the mature retinogeniculate connection must be sculpted out of an initially over-connected circuit, and that the extent of this refinement may vary from species to species. These data also suggest that retinogeniculate development occurs over several stages beginning with the segregation of eye-specific inputs, followed by a more fine-scale refinement phase that sharpens the retinotopic map, and a final maintenance phase during which this connection remains malleable.
Activity-dependent refinement of retinogeniculate projections
Once it was appreciated that much of the initial development and refinement of retinogeniculate projections occurs before the onset of vision, these findings raised the question of whether synaptic transmission and/or neuronal activity are required for refinement. The first demonstrations that action potentials are required for eye-specific segregation came from studies in which tetrodotoxin (TTX) was used to block sodium channels in fetal cat brain. This activity blockade prevented the formation of eye-specific domains. , Further experiments found that spontaneous activity is generated in the retina during eye-specific segregation and when spontaneous retinal activity is disrupted eye-specific layers do not form. In addition, activity disruptions that prevent axon refinement also lead to larger than normal receptive fields.
After eye-specific segregation, neural activity also contributes to the refinement of ON and OFF inputs in the dLGN. In the ferret visual system either blockade of retinal activity or postsynaptic N-methyl- d -aspartate (NMDA) receptor activity disrupts the formation of ON and OFF sublaminae. , This suggests that synaptic transmission from the retina to the dLGN is necessary for at least this one aspect of retinogeniculate refinement; although the specific role of synaptic transmission and plasticity in this and other aspects of retinogeniculate refinement warrants closer investigation.
After retinal inputs are segregated within the dLGN neural activity continues to be required for maintaining the precision established during development. For example, blockade of retinal activity after eye-specific layers have formed in the ferret causes the silenced projections to lose the territory that they had previously occupied. Similarly, blockade of late stage activity in mice with TTX disrupts synaptic refinement, preventing the elimination of weak inputs and the strengthening of remaining inputs. In addition, analysis of a mutant mouse that develops excessive retinal activity after eye-specific segregation found that this increased activity is capable of desegregating retinal inputs. This mutant mouse, termed nob1 (no b wave), lacks a protein necessary for photoreceptor to bipolar cell synaptic transmission and at the time when vision through photoreceptors would normally ensue it instead develops abnormally high rates of spontaneous retinal activity. Since spontaneous retinal activity is unaltered in these mice during the period when eye-specific segregation normally occurs, their axons initially segregate normally. , However, with the onset of the abnormally high-frequency activity the eye-specific inputs desegregate, leading to increased axonal overlap. In addition to this late requirement for normal spontaneous activity, visual input also appears to play an important role in the maintenance of retinogeniculate refinement. If animals are deprived of visual input when the late stages of within-eye refinement are nearly complete, dLGN neurons will become multiply innervated again, essentially reverting to a more immature innervation state. ,
Together these experiments indicate that spontaneous activity is required for driving both eye-specific and On–Off segregation and that both spontaneous and visually evoked retinal activity are necessary for maintaining segregation.
What parameters of activity drive refinement?
Retinal activity is necessary during each stage of retinogeniculate development, but what aspects of activity are required? Theoretical studies propose that patterned activity that preferentially correlates the spiking of nearby cells but not distant cells is likely to be essential. In the visual system, patterned retinal activity is known to be present before vision, , and developmental mechanisms within the retina produce a pattern of activity that meets these theoretical criteria for segregation. For instance, both multielectrode array recordings and calcium imaging experiments have revealed that bursts of activity are correlated among neighboring RGCs and these bursts propagate in a wave-like fashion across the retina. These “retinal waves” have now been observed in many vertebrate species suggesting that this activity pattern may be a conserved mechanism governing vertebrate visual system development ( Fig. 39.1 ).
Three stages of patterned wave activity have been described in the developing retina. Interestingly, at each stage the activity is mediated by distinct mechanisms and displays unique properties, such as the frequency of the activity, the area of the retina involved in the correlated activity, and the speed of the propagating wave. These studies suggest that the unique features of these stages might contribute to the specific aspects of refinement that occur during each stage. The first stage of retinal activity, stage I, consists of infrequent bursts of patterned activity which appear embryonically and are largely mediated by communication through gap junctions. The bulk of retinogeniculate refinement occurs during stage II waves. Stage II waves are mediated by acetylcholine release from starburst amacrine cells , and occur embryonically in mammals born with their eyes open and postnatally in mammals born with their eyes closed. , The final stage of patterned spontaneous activity, stage III, is mediated by glutamatergic transmission from bipolar cells, spans the period of eye opening, and is propagated, at least in part, due to glutamate spillover. Interestingly, during stage III waves, ON and OFF RGCs fire at different rates and are anti-correlated as do other subtypes of RGCs such as X, Y, and W, which could play a role in the segregation of these distinct types of inputs.
By what mechanisms do retinal waves drive changes in retinogeniculate connectivity? A common mechanism that appears to underlie the development of connection specificity throughout the nervous system is activity-dependent competition among presynaptic inputs for postsynaptic space. , This competition is driven by neural activity, but how? Evidence that eye-specific segregation involves competition among RGC inputs comes from studies in which activity was either blocked or enhanced in one eye resulting in alterations to the pattern of ipsilateral and contralateral projections. In these studies when epibatidine was used to block cholinergic transmission in just one eye, the projections from the blocked eye lost territory in the dLGN while the projections from the active eye expanded their coverage. The converse was also true; that is, when activity was increased in one eye by intraocular administration of either forskolin or the cyclic AMP analog, CPT-cAMP (which increase the size, frequency, and speed of waves), the projections from the more active eye gained territory in the dLGN at the expense of the non-enhanced eye’s projections. ,
Both experimental and theoretic studies suggest that in addition to overall activity levels, specific aspects of patterned spontaneous activity may drive segregation through Hebbian mechanisms that strengthen and eliminate synaptic inputs over time. This Hebbian model of synaptic competition predicts that ganglion cells that “fire together, wire together”. For Hebbian mechanisms to drive segregation, neighboring RGCs must fire in a correlated manner while more distant RGCs or RGCs from different eyes must exhibit less correlated spiking; thereby allowing for the strengthening of the temporally correlated neighboring inputs and the weakening of the uncorrelated inputs. Indeed, these requirements are met as a consequence of wave propagation. For example, during a wave neighboring RGCs are co-activated causing their firing to become more synchronized than that of non-neighboring RGCs. In addition, the random nature of wave initiation and the relatively long refractory period between waves results in a low probability of coincident firing of more distant ganglion cells or ganglion cells in separate eyes. This favors the segregation of inputs from different eyes, and could simultaneously remodel both the connectivity between the two eyes and refine the retinotopic map within each eye. ,
Anatomical studies in mice have shown that stage II waves are of particular importance for eye-specific segregation since this segregation occurs during stage II waves and disruption of stage II waves through binocular injections of epibatidine prevents eye-specific segregation during this period ( Fig. 39.2 ). These data show that stage II waves are required for eye-specific segregation – but are they permissive or instructive for segregation? The data regarding this question are somewhat conflicting. Studies of mice lacking the nicotinic acetylcholine receptor subunit β2 have suggested that retinal wave structure is critical for driving segregation since in these mice, which retain retinal activity but have severely altered wave structure, eye-specific refinement does not occur during stage II. On the other hand, the first study to abolish retinal waves without disrupting the overall level of retinal activity did not find any deficit in eye-specific segregation. In addition, studies of other mice with disrupted waves indicated that although these animals display defects in eye-specific segregation, when certain aspects of the waves were rescued this did not lead to the rescue eye-segregation.
In an effort to determine the specific parameters of waves that might be necessary for eye-specific segregation the Feller laboratory analyzed a number of mutant mice with different retinal wave disruptions and varying degrees of eye-specific segregation. Based on these comparisons they concluded that several aspects of retinal waves are likely to be essential for driving eye-specific segregation, such as the amount of correlated firing that occurred in bursts with frequencies greater than 10 Hz. In addition, they found the relative difference in the correlation strengths between neighboring and distant RGCs to be critical, rather than the overall activity of the RGCs or the precise time window of RGC firing. This study suggests that retinal waves encode information that instructs retinogeniculate refinement, though this topic remains controversial. As mentioned above, in the nob1 mouse the onset of altered late-stage waves causes eye-specific inputs to desegregate. Interestingly, the altered waves in the nob1 mice display correlations that are stronger between neighboring RGCs than distant RGCs, similar to the correlation structure found by Feller and colleagues to promote eye-specific segregation. This suggests that, at least for stage III waves, the levels of activity can also affect retinogeniculate development. In the nob1 mouse it is likely that very high frequency of waves leads to increased co-incident firing of RGCs in different eyes, thus leading to the re-growth of new inputs and/or the strengthening of residual inputs. ,
Synaptic inputs change strength with segregation
Electrophysiological recordings have shown that the convergence of functional synaptic inputs onto individual dLGN neurons dramatically decreases over development. As discussed above, evidence suggests that specific aspects of retinal wave structure likely play an instructive role in retinogeniculate refinement by driving Hebbian learning rules at retinogeniculate synapses. Presumably, the correlated firing of neighboring RGCs leads to increased synaptic strength which in turn stabilizes the axon arbors containing the strengthened synapses. Conversely, uncorrelated firing leads to decreased synaptic strength which ultimately leads to the elimination of the weak synapses and the axon collaterals on which they reside. So, what are the synaptic learning rules that operate at retinogeniculate synapses, and what molecular mechanisms translate changes in synaptic strength into altered morphology? The answers to these questions are still largely unknown; however, some insight has been gained and is discussed below.
The first study to demonstrate that retinogeniculate synapses can undergo synaptic plasticity stimulated RGC afferents in a retinogeniculate slice preparation with high-frequency trains and showed that this strengthened retinogeniculate synapses in an NMDA receptor-dependent fashion. However, one caveat to this study was that the high-frequency trains were faster than those typically observed during retinal waves. Using a preparation containing intact retinal inputs to the dLGN combined with patch-clamp recordings Mooney et al demonstrated that, during segregation, retinal inputs are capable of driving postsynaptic dLGN spiking, a prerequisite for Hebbian-based learning. More recently, Butts and colleagues, using stimulation that more accurately reflects endogenous activity patterns, described a burst-timing-dependent plasticity protocol that was capable of both strengthening or weakening retinogeniculate synapses depending on whether presynaptic and postsynaptic bursts (10–20 Hz) were coincident or non-overlapping. In addition to this homosynaptic learning rule, synaptic weakening may also occur via heterosynaptic depression. In support of this idea, trains of synaptic stimuli from one eye were shown to lead to depression of the opposite eye inputs. Undoubtedly, future studies will further elucidate the types of plasticity mechanisms that influence the development of this connection. In particular, since the distinct stages of retinal activity differ in their wave parameters it will be interesting to find out whether the synaptic plasticity rules operating in the dLGN change over these stages. For instance, in ferrets it was reported that NMDAR activation is not required for eye-specific segregation, but is required for On–Off segregation. Interestingly, toward the end of the period of stage III waves, both spontaneous activity and vision were shown to preferentially drive the glutamate receptor subunit 1 into retinogeniculate synapses.
While, as previously mentioned, the number of RGCs innervating individual dLGN neurons decreases dramatically over development, the total number of synapses made onto individual dLGN neurons seems to remain fairly constant. This finding is analogous to the refinement of motor neuron inputs at the neuromuscular junction where a single motor neuron takes over the synaptic territory of multiple motor neurons. The mechanisms controlling total synapse number are not known. In addition, some mutant mice retain excess retinal inputs into adulthood, yet these extra inputs do not drive aberrantly large synaptic currents. , These studies suggest that the total synaptic drive received by dLGN neurons is under homeostatic regulation.
Molecular mechanisms involved in activity-dependent axonal segregation
A major area of research that remains is to elucidate the molecular mechanisms that translate changes in synaptic drive into altered synapse and axon morphology. The use of transgenic mouse lines has led to tremendous advancement in our understanding of these molecular mechanisms. Screening of mutant animals that display fairly normal retinal waves yet have severely disrupted eye-specific segregation has demonstrated that several types of immune-related molecules are likely to be involved in this process. The first such example was the class I major immunohistocampatibility complex (MHC). This protein is upregulated by activity in the dLGN during the time of eye-specific segregation, and MHC I knockout mice display overlapping eye inputs. The neuronal pentraxins (NPs) are another example of activity-regulated genes shown to be involved in eye-specific segregation, and interestingly, NPs are also immune-related molecules that share some sequence similarity to the short pentraxins, C-reactive protein, and serum amyloid P. Like MHC, NPs are highly expressed during retinogeniculate refinement, and animals lacking two NPs, NP1 and NP2, display overlapping eye inputs past the normal period of segregation. Both MHC and NPs may directly affect synaptic function. Another example of molecular mechanisms that may couple functional and structural changes in the dLGN comes from studies of the complement pathway. Several strains of mice with deletions of protein in the classical complement pathway have overlapping eye-specific input in the dLGN. The complement pathway is well known for mediating phagocytosis of dead or infected cells. Because complement appears to co-localize with a subset of synapses in the dLGN, one possible role for the complement pathway is that it may tag weak inputs that are slated for removal during the period of eye-specific segregation.
Molecular mechanisms guiding the formation of eye-specific axonal territories
In addition to the molecular mechanisms discussed above, activity-independent molecular mechanisms also play a role in shaping the mature pattern of retinogeniculate connections. Because the size and position of the eye-specific regions in the dLGN are highly stereotyped from animal to animal and symmetrical on both sides of the thalamus, it has long been speculated that molecular cues must determine where inputs from each finally coalesce. What cues might guide the inputs from the two eyes into their distinct territories? Interestingly, in achiasmatic sheep dogs with uncrossed eye projections, inputs from the normally crossed nasal and the uncrossed temporal retina still segregate in the dLGN, strongly indicating that molecular cues direct eye-specific segregation by marking the temporal and nasal retina. Work on retinotectal mapping has shown that the nasal and temporal retina can be distinguished by their differing expression of EphAs and both the EphAs and their binding partners, ephrin-As, play a critical role in eye-specific segregation in the dLGN. Mice lacking three of the ephrins, A2, A3, and A5 still display activity-dependent segregation of eye-specific inputs, but the eye-specific regions are fractured and do not form distinct layers in the dLGN. If these triple knockout mice are subjected to activity blockade during the period of eye-specific segregation their retinal inputs form a permanent salt and pepper pattern of intermingled inputs. In addition, retinal overexpression of EphA3 or EphA5 can lead to mistargeting of retinal inputs to the incorrect eye-specific layer despite normal retinal patterned activity. These data indicate that EphA–EphrinA interactions shape the final pattern of eye-specific inputs. These also indicate that activity and ephrin signaling act in parallel and are both required for the normal organization of retinogeniculate inputs. In addition, activity manipulations do not seem to affect ephrin expression, nor does the loss or overexpression of ephrins impact activity levels, further suggesting that these pathways act largely independently from one another. ,