Developmental remodeling of relay cells in the dorsal lateral geniculate nucleus in the absence of retinal input

Background The dorsal lateral geniculate nucleus (dLGN) of the mouse has been an important experimental model for understanding thalamic circuit development. The developmental remodeling of retinal projections has been the primary focus, however much less is known about the maturation of their synaptic targets, the relay cells of the dLGN. Here we examined the growth and maturation of relay cells during the first few weeks of life and addressed whether early retinal innervation affects their development. To accomplish this we utilized the math5 null (math5−/−) mouse, a mutant lacking retinal ganglion cells and central projections. Results The absence of retinogeniculate axon innervation led to an overall shrinkage of dLGN and disrupted the pattern of dendritic growth among developing relay cells. 3-D reconstructions of biocytin filled neurons from math5−/− mice showed that in the absence of retinal input relay cells undergo a period of exuberant dendritic growth and branching, followed by branch elimination and an overall attenuation in dendritic field size. However, math5−/− relay cells retained a sufficient degree of complexity and class specificity, as well as their basic membrane properties and spike firing characteristics. Conclusions Retinal innervation plays an important trophic role in dLGN development. Additional support perhaps arising from non-retinal innervation and signaling is likely to contribute to the stabilization of their dendritic form and function.


Background
The dorsal lateral geniculate nucleus (dLGN) of the mouse thalamus has become a powerful model system to understand visual circuit development [23,26,27]. It has been especially useful for delineating the mechanisms underlying the establishment of the retinogeniculate pathway. A crucial element of this pathway is the synaptic target of retinal ganglion cells (RGCs), the relay cells of dLGN. These neurons serve as the principal conduit of information between the retina and visual cortex. Additionally, dLGN relay cells are the major site of convergence for a number of non-retinal inputs that work in concert to modulate the gain of retinogeniculate transmission in a state dependent manner [5,48,49].
Despite playing such a key role in visual processing, until recently little was known about the structural and functional composition of mouse dLGN relay cells. We found that mouse dLGN relay cells have highly stereotypic dendritic architecture and are readily classified as having X -, Y-or W-like profiles [30]. The distinguishing features of their dendritic morphology develop remarkably early in postnatal life. After the first postnatal week relay neurons have highly complex dendritic fields that already begin to resemble their adult counterparts. Accompanying this growth is the rapid maturation of their active membrane properties and spike firing characteristics. Such coordination enables relay cells to receive, integrate, and transmit retinal signals accurately by the time of natural eye opening [19,28,30,36], when retinal activity switches from spontaneous to visually evoked [18,53].
What remains unexplored is an understanding of the mechanisms that contribute to the development of relay cells. A prevailing view relates to the "synaptotrophic" hypothesis, which underscores the necessity of early synapse formation as a driving force for neuronal maturation (reviewed in [14,56]). A likely candidate for dLGN relay cells is the support provided by retinal input [15]. These axons innervate the dLGN at perinatal ages, a time just after the nucleus takes shape and neuronal differentiation is completed [1,22,28]. Soon after birth newly formed axon terminals form functional synapses with dLGN cells [28,40], and by postnatal week 2 retinogeniculate synapses begin to take on adult-like profiles [5].
A number of studies have adopted a loss of function approach to assess whether early retinal input and synapse formation contribute to dLGN development. However many of the manipulations to remove or silence retinal input did not focus on the development of relay cells per se [6,25,62], or more importantly, were done well after the time of early retinal innervation and synapse formation [6,44,47,52]. Past attempts to employ a genetic form of deafferentation have also been problematic since "eyeless" phenotypes often involve a polygenic form of inheritance and are accompanied by other mutations that may have an indirect impact on neuronal development [16,55,58].
To overcome these issues we employed a relatively novel genetic form of retinal deafferentation by taking advantage of the math5 null mutant mouse (math5 −/− ). Math5 is a basic helix-loop-helix (bHLH) gene that is expressed in the retina starting at embryonic day (E) 11 and is essential for the differentiation of retinal progenitor cells into RGCs [8]. As a consequence, math5 −/− exhibits a wholesale loss (>95 %) of RGCs [9,41,60], as well as a failure of the surviving cells to form an optic nerve [9,10,61]. Thus, this form of genetic deafferentation ensures that dLGN is devoid of retinal innervation even prior to perinatal times when retinal axons normally enter the nucleus. Here we made use of this mouse along with age matched wild types (WT) to understand whether retinal innervation affects the development of dLGN relay cells.

Math5 expression in WT retina and dLGN
Math5 mRNA encodes a transcription factor that specifies RGC fate [8,9,60]. Embryonically, math5 is expressed in the retina as well as the tenth cranial ganglion [8]. In the retina, math5 is developmentally regulated, first appearing at E11, continuing through birth but absent in the adult [8,9,60]. However, there are some reports of math5 expression in adult brain regions such as cerebellum and the ventral cochlear nucleus [45]. A closer examination of math5 expression in central visual targets such as dLGN is lacking. Here we examined math5 expression in the developing retina and dLGN using RT-PCR ( Fig. 1; retina: n = 2 at each age; dLGN: n = 10 per age). As expected, in WT mice, math5 was expressed in the retina between E13-P3, but absent at P13 and in the adult. Moreover, in WT dLGN we found no evidence of math5 expression at any of the ages tested (e.g., P2, 3, 14, adult). Thus any reported changes observed among developing relay cells in math5 −/− cannot be attributed to the lack of math5 in dLGN neurons, but rather is due to a direct consequence of RGC elimination.

Absence of retinal input in math5 −/−
While math5 −/− mice appear to lack an optic nerve, it is not clear whether the few remaining RCGs grow axons that enter the brain and innervate retino-recipient targets ([9, 10, 60, 61], but see [54]). To test for this possibility, the anterograde tracer CTB conjugated to different Alexa fluorescent dyes was injected into each eye of math5 −/− and WT mice (Fig. 2). This technique allows for the visualization of retinal terminal fields in central visual structures [28]. In WT mice, robust labeling of retinal terminals was apparent in all retino-recipient targets. For example in Fig. 2a, retinal axons from each eye innervated the suprachiasmatic nucleus (SCN) and formed overlapping terminal fields, whereas in dLGN they formed nonoverlapping eye specific domains (Fig. 2d). By contrast, eye injections of CTB made in math5 −/− between P2-P48 (n = 8) failed to reveal any labeled elements in regions that correspond to optic nerve, optic tract or retino-recipient targets such as SCN or dLGN (Fig. 2b, e; see also [10,61]).
To further confirm the absence of retinal innervation in the dLGN, we used immunohistochemistry to detect the type 2 vesicular glutamate transporter (VGluT2), a reliable marker for retinal terminals in dLGN [21,24,31] (Fig. 3a). In a P14 math5 −/− mouse, there was almost a complete absence of VGluT2 in dLGN (Fig. 3b). The weak and sparse labeling we did detect was similar to the labeling pattern seen after a 7-day binocular enucleation (Fig. 3c), suggesting that the trace amounts of VGluT2 in math5 −/− dLGN, were of non-retinal origin [21,24].
An ultrastructural analysis of the types of synapses found in dLGN of math5 −/− mice confirmed these findings ( Fig. 3d-e). To distinguish excitatory from inhibitory profiles, we labeled those that contained gamma-aminobutyric acid (GABA) using an antibody that was subsequently tagged with gold particles. In WT mice, retinogeniculate terminals are characterized as large non-GABAergic profiles that contain round vesicles and pale mitochondria (RLP profiles, Fig. 3d, blue) [5]. In a sample of images from the dLGN of a WT mouse, (20 images at P21), we identified 29 RLP profiles with a mean area of 0.95 ± 0.11 μm 2 . Other non-GABAergic profiles were also abundant (n = 49) and had an average area of 0.51 ± 0.4 μm 2 ; see also [5]. By contrast, in a sample of images from an age-matched math5 −/− (20 images), we failed to detect any RLP profiles. However, the overall population of non-GABAergic terminals present in math5 −/− mice (n = 46) was comparable in size to WT (WT, 0.68 ± 0.05 μm 2 vs. math5 −/− , 0.67 ± 0.08 μm 2 , Student's t-test, p = 0.97). Interestingly, in math5 −/− mice, we noted the presence of non-GABAergic terminals characterized by having round vesicles, large profiles and dark mitochondria (RLD profiles) (Fig. 3e, blue, [20]). These so-called RLD profiles, which appear to supplant RLP profiles in math5 −/− mice have been observed in enucleated, anophthalmic and microphthamic strains of mice, but their origin is yet to be determined [16,29,63]. Taken together, these results indicate that the dLGN of math5 −/− mice serves as a suitable model for studying the development of relay cells in the absence of retinal innervation and signaling.
Morphological characteristics of developing relay cells in WT and math5 −/− mice In order to examine whether the absence of retinal input influences the morphological development of relay cells we made in vitro recordings from acutely . d Retinal terminals in WT mice (blue) include distinctive pale mitochondria (white asterisks). These terminals primarily synapse (arrows) on non-GABAergic dendrites (green), which often extend small protrusions into the presynaptic retinal terminals. E) In math5 −/− mice, the dLGN contains no terminals with pale mitochondria. Instead, large profiles (blue) with dark mitochondria (black asterisks) form synaptic arrangements that are similar to retinal terminals, including contacts on non-GABAergic dendritic protrusions (green). GABAergic profiles (pink) are identified by a high density of overlying gold particles. Scale bar = 1 μm and applies to both panels prepared slices containing the dLGN and filled cells with biocytin [30]. We then conducted multi-photon laser scanning microcopy to generate 3D reconstructions. Figure 5 shows representative examples of biocytin filled relay cells at different postnatal ages in WT and math5 −/− mice. At all ages examined, math5 −/− cells had large somata, multipolar dendrites, and axons that exit the nucleus (Fig. 5, arrowheads). Qualitatively, math5 −/− relay cells appeared similar to age matched WTs. However, quantitative analysis revealed a number of differences in their growth patterns and dendritic architecture.
Relay cells in math5 −/− mice showed fluctuations in dendritic growth compared to WT, initially experiencing exuberant growth (week 1-3), followed by a progressive decline (week [4][5]. To address whether these changes were due to the lengthening and sprouting of new branches (exuberant growth), or the shrinkage and pruning of dendrites (decline), we examined overall dendritic field, the number of branches, and branching patterns.
Compared to WT, math5 −/− relay cells showed comparable changes in dendritic field area throughout the

Dendritic complexity
We examined the total number of dendritic branches and the pattern of branching for individual relay cells in the Z-plane by identifying primary dendrites, and their successive daughter branches [30].
To examine dendritic branching patterns we calculated the number of branch points as a function of branch order. Figure 7a depicts summary plots for these relationships. At all ages both WT and math5 −/− cells had 6-7 primary dendrites, with the highest number of branching occurring between the 3 rd -5 th orders. Week by week comparisons of branch complexity between WT and math5 −/− relay cells are shown in Fig. 7b. During week 1, math5 −/− relay cells showed increased numbers of 6 th order branches (n = 8 WT, 1.0 ± 0.4 vs. n = 8 math5 −/− , 4.0 ± 0.9; Student's t-test, p <0.01). Branch order continued to expand during week 2, so that math5 −/− cells had significantly more 6 th -10 th order dendritic segments compared to WT(Student's t-test, branch orders 6-8 p <0.0001, branch orders 9-10 p <0.01). However, increased sprouting was transient, so that by weeks 3-4 there were no differences in the total numbers of dendritic branches or branch order compared to WT cells (Figs. 6d and 7b). Moreover, by week, 5, additional losses were observed among 2-4 th order segments ( Fig. 7b; n = 5 WT vs. n = 4 math5 −/− , Student's t-test, branch orders 2-4 p <0.05).
In sum, these analyses show that the increased dendritic surface area noted in week 2 for math5 −/− relay cells is due to exuberant dendritic branching, especially among higher order segments (Figs. 6a, d and 7b). Furthermore, the reduction in dendritic surface area at week 5 is likely a consequence of attenuation in dendritic field as well as a continued loss of dendritic branches (Fig. 6a, c and 7b).
Relay cell class specificity and location in math5 −/− dLGN Recently we showed that relay cells can be divided into three classes that have distinct dendritic architecture and strong regional preferences in dLGN [30]. This classification scheme was based on a Scholl ring analysis and the computation of a dendritic orientation index (DOi) that was based on the number of intersections found in each of four axial planes [30]. Cells with a DOi between 0-0.49 had a bi-conical morphology (X-like); those with values between 0.50-0.79 had a hemispheric profile (W-like), while those between 0.80-1.0 were radially symmetric (Y-like). Using the identical approach, we analyzed the dendritic architecture of 42 relay cells in math5 −/− dLGN. Similar to our previous study, we limited our analysis to postnatal weeks 2-5, at times when total dendritic branching stabilizes ( Fig. 6d; see also [30]). Despite the transient increase in branching in week 2 and the subsequent loss in week 5, math5 −/− relay cells were of sufficient complexity to retain their identity. Additionally, the associated compression makes it difficult to compare regional preferences with their age-matched WT counterparts. Nonetheless, similar to the regional preferences of cell types noted in WT dLGN [30], a qualitative examination of cell location in the math5 −/− dLGN revealed that Y-like cells resided in a central band throughout the nucleus and W-like cells were preferentially located along the dorsal border of the dLGN (Fig. 8b).
Examples of voltage responses to current steps in math5 −/− cells are shown in Fig. 9c. Many of the voltage-gated conductances noted in WT were also present in math5 −/− age matched cells (not shown but see [19,28,30,36]). For example in math5 −/− relay cells, membrane hyperpolarization evoked a strong inward rectification. This large depolarizing sag in the voltage response reflects the activation of the mixed cation conductance (H) [28,32]. In addition, the termination of membrane hyperpolarization activated a t-type Ca 2+ conductance that produced a rebound low-threshold calcium spike (LTS), along with a burst of Na + spikes that ride the peak of this triangular depolarization. With membrane depolarization, relay cells exhibited an outward rectification that delayed spike firing and reflected the activation of a transient (A) type K + conductance [32,36,38]. Strong and sustained levels of membrane depolarization readily evoked a train of action potentials that exhibited spike frequency accommodation, an event attributed to the activation of K + conductances that produce an after hyperpolarizing response between spikes (AHP) [32,36].
Overall, these observations indicate that the intrinsic membrane properties and spike firing of relay cells remain largely unaffected by the absence of retinal innervation.

Discussion
Our data from WT mice suggest that dLGN relay cells undergo two growth spurts [43]. The major elements and timing of these events are outlined in Fig. 10. The first phase takes place during postnatal week 1, as dendritic branches increase in number and grow in length to form highly stereotypic architecture and cell class specificity [30]. The second phase occurs during postnatal weeks 2-3 where there is a progressive increase in dendritic field size. During this time no additional branch elaboration occurs and the overall complexity of dendritic arbors remains stable. The timing of these growth spurts corresponds to a progressive increase in dLGN size, and like the maturation of relay cells, the nucleus assumes an adult-like profile by postnatal week 3 [25].
Our results in math5 −/− mutants indicate that the absence of retinal innervation disrupts the normal growth and maturation of dLGN relay cells (Fig. 10). Initially relay cells in math5 −/− follow a similar growth trajectory as age matched WT cells. However, they undergo an extended period of branch elaboration, showing an increase in branch number and length throughout postnatal week 2. Such exuberant branching is not maintained. In fact, the total number of branches and overall branch order declines through postnatal week 5, and leads to an overall reduction in dendritic surface area. Accompanying these dystrophic changes is shrinkage in the overall size of dLGN.
Taken together, these data suggest that retinal innervation plays an important trophic role in dLGN development. Indeed, the development of the retinogeniculate pathway seems to satisfy many of the key elements for "synaptotrophic" support (reviewed in [56] and [14]). A major tenant for such support is that dendritic development and synapse formation/ maturation are concurrent. Retinal axons arrive in dLGN at perinatal ages, and by early postnatal life fully innervate the dLGN [22,28,42]. Soon after innervation retinal axons form functional synapses with developing dLGN cells [5,28]. These newly formed synapses are of sufficient excitatory strength to drive action potentials in dLGN relay cells [5,28,33,40]. Such excitatory drive is provided by spontaneous wave like activity of RGCs that prevail prior to the onset vision [18,35,40]. Additionally, dendritic maturation of relay cells coincides with a highly active period of synaptic remodeling and maturation. Structurally, retinal profiles (RLP) expand in size and complexity, showing a dense clustering of vesicles [5]. Functionally, developing dLGN cells receive relatively weak synaptic input from as many a dozen or so RGCs [13,28,64]. By postnatal week 2 many of these inputs are eliminated and the remaining few show a substantial increase in synaptic strength, as well as a shift in NMDA to AMPA receptor composition (reviewed in [23,26,33]).
Our results in math5 −/− mutants also suggest that retinal innervation is needed for constraining and stabilizing the dendritic complexity of relay cells. It is believed that developing dendrites sample their environment and extend processes into regions where prospective synaptic afferents are found (reviewed in [14,15,57]). Perhaps the extensive dendritic branching we observed in math5 −/− relay cells reflects a compensatory response by these cells to seek potential synaptic partners. This notion is consistent with other reports showing that developing neurons alter their dendritic form in response to a disruption in afferent input (reviewed in [12,37]). Finally it is worth noting that retinal signaling is required for the continued maintenance of dendritic form. In math5 −/− relay cells, the exuberant branching observed during postnatal week 2 is eliminated and followed by a modest decline in proximal dendritic segments.
Our results suggest that RGCs provide trophic support that sustains the development of relay cells, as well as to support the overall structural integrity of dLGN. Such trophic effect for retinal axons on the growth of dLGN has been previously described in enucleation and anophthalmic studies where distortion and shrinkage of dLGN have been reported [16,25,62]. However, the molecular mechanisms underlying trophic support and maintenance of the mouse dLGN remain largely unknown. A likely candidate is the brain-derived neurotrophic factor (BDNF). This neurotrophin is synthesized in the retina, transported anterogradely by retinal afferents and can bind to their high affinity receptor tyrosine kinase (trkB) located on dLGN dendrites [2,11,15,39].
Finally it is important to note that despite the disruption in growth and maturation, relay cells in the math5 −/− dLGN still retained a high degree of branch complexity, morphological class specificity, and the full complement of active membrane properties. Such observations suggest that dendritic form and function are likely regulated by other factors unrelated to retinal innervation and signaling. One possibility is that synaptic signaling from non-retinal inputs could provide additional trophic support. Indeed the bulk of synaptic input to dLGN arise from a number of nonretinal sources, including glutamatergic neurons in layer VI of visual cortex, cholinergic nuclei of the brainstem, and GABAergic neurons within the thalamic reticular nucleus as well as intrinsic interneurons within the dLGN [5,46,49]. Many of these elements have been implicated in supporting developing dendritic form and function [3,17,34,50,59] (reviewed in [4]). Most notable are the inputs that arise from visual cortex, where the infusion of neurotrophic factors leads to an accelerated growth of relay cells [59]. Interestingly in mouse, corticogeniculate inputs arrive at late postnatal ages, well after retinal innervation [46]. Such timing suggests that these descending projections are poised to contribute to the maintenance and stability of dendritic form. In fact, in the math5 −/− , dLGN cortical inputs arrive much earlier than in WT [7,46], and thus could help explain why relay cells in these mutants retain much of their overall structural and functional integrity.

Conclusions
The dLGN of mouse has proven to be an important model system for visual circuit development. However there is a paucity of information regarding the development of its principal cell type, namely thalamocortical relay cells. Here we examined the postnatal growth and maturation of dLGN relay cells and tested, by utilizing math5 −/− mice, the extent to which their dendritic form and function relied on retinal innervation. We found that the absence of retinal innervation leads to an overall shrinkage of dLGN and disrupts the pattern of dendritic growth of relay cells. In math5 −/− dLGN, relay cells undergo a period of exuberant dendritic growth and branching followed by branch elimination and an overall attenuation in dendritic field size. Despite these dystrophic changes, relay cells in math5 −/− mice retained a sufficient degree of complexity and cell class specificity, as well as the full complement of membrane properties and spike firing characteristics. Thus retinal innervation plays an important trophic role in dLGN development, but that additional support perhaps arising from nonretinal innervation and signaling, contributes to stabilization of dendritic form and function.

Subjects
All procedures carried out were approved by the Institutional Animal Care and Use Committee at Virginia Commonwealth University. Mice ranging in age between the first and fourth postnatal weeks were studied. Two strains were used: pigmented wild-type mice (C57/BL6), math5 −/− on a mixed C57B6/J and 129/SvEv background provided by S. Wang [60].

CTB injection
Injection of the anterograde tracer cholera toxin subunit beta (CTB) were performed in order to visualize retinal projections in the dLGN and to assess whether any surviving RGCs in math5 −/− had axons that exited the eye and innervated retino-recipients targets in the brain. Mice were anesthetized with isoflurane vapors. Using a glass pipette, the sclera was pierced near the ora serrata and excess vitreous fluid was drained. Using another glass pipette attached to a picospritzer, 3-8 μl of CTB (1.0 % solution dissolved in distilled water) conjugated to Alexa Fluor 488 or 594 dyes (Invitrogen) were then injected into the same opening used to drain the excess vitreous fluid. Following eye injections, animals were given a 2-day survival period to allow the tracer to travel to central visual targets such as SCN or dLGN.

Acute in vitro thalamic slice preparation
Whole cell recording and filling of relay cells were done using methods described elsewhere [5,19,28,30]. Animals were anesthetized with isoflurane and decapitated. The brain was excised and placed in a 4°C oxygenated (95 % O 2 /5 % CO 2 ) slicing sucrose solution (in mM: 26 NaHCO 3 , 23.4 sucrose, 10 MgSO 4 , 0.11 glucose, 2.75 KCl, 1.75 Na H 2 PO 4 , 0.5 CaCl 2 ). Slices (300 μm) were cut in the coronal or parasagittal planes on a vibratome (Leica VT1000S), and placed for 1 h in a 35°C oxygenated solution of artificial cerebral spinal fluid (ACSF) (in mM: 124 NaCl, 2.5 KCl, 1.25 NaH 2 PO 4 , 2.0 MgSO 4 , 26 NaHCO 3 , 10 glucose, 2 CaCl 2 ). Slices containing dLGN were selected for in vitro intracellular recording in the whole cell current clamp mode, and were perfused in an oxygenated solution of ACSF that was kept heated at 30°C. Cells were visualized with the aid of IR-DIC optics, and were patched with electrodes made of borosilicate glass filled with an internal solution (in mM: 140 K gluconate, 10 HEPES, 0.3 NaCl, 2 ATP-Mg, 0.1 GTP-Na; pH 7.25) containing 5 % biocytin. Patch electrodes were vertically pulled and had a final tip resistance of 3-7 MΩ. Electrodes were connected to an amplifier (Axoclamp 2B, Axon instruments). Different protocols of square wave current pulses were applied and the resulting voltage responses were measured. Neuronal activity was digitized with an interface unit (National Instruments) and stored on a computer. Data acquisition and analysis was done using Strathclyde Electrophysiology Software, Whole Cell Analysis Program V3.8.2.
At the end of the recording, slices were fixed overnight with 4 % paraformaldehyde (PFA) in 0.1 M phosphate buffer solution (pH = 7.2). To visualize dLGN cells filled with biocytin, slices were washed with phosphate buffer saline (PBS) (3×, 30 min), and incubated overnight at 4°C in a PBS solution containing 0.1 % Triton X-100 and 0.1 % Alexa Fluor 647 conjugated streptavidin (Invitrogen). Slices were washed with PBS, mounted with ProLong Gold with DAPI (Invitrogen), and cured overnight at room temperature.

Reconstruction of biocytin filled relay cells
Three-dimensional reconstructions and analysis were done using methods described previously [30]. Biocytin filled relay cells were imaged using a multi-photon laserscanning microscope (Zeiss LSM510 NLO Meta). A HeNe laser (633 nm) was used to excite fluorescence from biocytin filled dLGN neurons and emission was detected at a range of 651-694 nm (Meta detector). The following objective lenses were used to image targeted neurons at a scanning resolution of 2048×2048 pixels: Plan-Neofluar 40× (1.3 n.a) oil immersion objective lens, or a C-Apochromat 40× (1.2 n.a) water immersion objective lens. 3-D datasets were compiled from a sequential series of optical slices with a step size through the Z-axis of 0.48 μm (40×/1.2 n.a. lens) or 0.5 μm (40×/1.3 n.a. lens). 3-D Z-stack datasets were analyzed using Volocity software (Improvision, version 4.3.2). Image sequences were deconvolved to reduce signal noise generated from outside the focal plane of interest using Iterative restoration technique, and thresholding values were set according to signal intensity and background noise.

Cresyl violet nissl stain
Animals were anesthetized with isoflurane vapors, and transcardially perfused with PBS solution for 5 min, followed by 4 % paraformaldehyde in 0.1 M PBS (ph = 7.2) for 15-20 min. Brains were excised and fixed overnight with 4 % PFA. Slices containing dLGN were cut on a coronal plane with a vibratome (70 μm), and left to dry overnight. Slices were washed for 3 min in 95 % and 75 % ethanol solutions, respectively. Slices were washed in dH 2 O for 1 min, before immersing them in cresyl violet stain for 20-30 s, and then were rinsed briefly with dH 2 O. Sections were washed for 3 min in 70 %, 95 %, 95 %, 100 %, 100 % ethanol solutions, respectively. Finally, slices were washed in xylene twice for 5 min. Slides were mounted with Permount, and visualized with light microscopy (Olympus 1×71, Photometrics Cool snap camera), and pictures were taken with a 10× objective lens. Images were analyzed with Metamorph software. Area measurements were obtained from 2-4 consecutive sections through the middle of the dLGN. Nissl stained cell counts were calculated from a 100μm×100μm region of interest centered in the middle of 2-3 dLGN sections. Measurements were restricted to cells in which the soma and nucleus were clearly delineated.

Enucleation
Binocular enucleation was done using methods described previously [46]. The eyes were removed after cutting the optic nerve and the ophthalmic artery. To avoid hemorrhaging, the orbit was filled with Gelfoam (Upjohn), and animals were allowed to recover on a heating pad.

Immunohistochemistry
Slices containing dLGN were processed using antibody that stains for VGluT2, a vesicular glutamate transporter found in retinal terminals [61]. On a vibratome, 40 μm thick slices were cut on the coronal plane. Before incubation, sections were rinsed in PBS, and then treated for 1 h with blocking solution (5 % NGS, 2.5 % BSA and 0.1 % Triton X-100). Sections were incubated overnight with the primary antibody at 4°C (rabbit anti-VGluT2: 1:1000, Synaptic Systems). Sections were rinsed with PBS, and were incubated in the secondary antibody (1:1000 dilution; Alexa 594 conjugated goat anti-rabbit IgG: 1:1000, Invitrogen, Cat# A11037) for 2 h at room temperature. Sections were rinsed in PBS, mounted with Prolong Gold with DAPI (Invitrogen) and cured overnight at room temperature. Sections were photographed with an upright epi-fluorescence microscope (Nikon E600, Photometrics Cool snap camera).

RT-PCR
Retina and dLGN tissue were harvested from C57/BL6 mice at different embryonic and postnatal ages using methods described elsewhere [51]. RNA was isolated using the Bio-Rad Total RNA Extraction from Fibrous and Fatty Tissue kit (Bio-Rad). Reverse transcription and cDNA generation were made using Superscript II Reverse Transcriptase First-Strand cDNA Synthesis kit (Invitrogen). The following math5 primer pairs were used: 5′-ATGGCGCTCAGCTACATCAT-3′ and 5′-GGGTCT ACCTGGAGCCTAGC-3′.

Electron microscopy
Ultrastructural analysis of dLGN was carried out as previously reported [5]. Mice (P21-22) were deeply anesthetized with isoflurane vapors and perfused transcardially with 2 % PFA/2 % glutaraldehyde in 0.1 M phosphate buffer solution. Brains were excised and cut on a coronal plane (50-100 μm thick) using a Vibratome (Leica VT100E). Sections were postfixed in 2 % osmium tetroxide, dehydrated in a graded series of ethyl alcohol and then were embedded in Durcupan resin. Ultrathin sections (70 nm) were cut, collected on Formvar-coated nickel slot grids and then were stained to reveal the presence of gamma amino butyric acid (GABA), using a polyclonal, affinity-purified rabbit anti-GABA primary antibody (cat. no. A2052, Sigma, St. Louis, MO) diluted 1:2,000, and a goat anti-rabbit IgG antibody conjugated to 15-nm colloidal gold particles diluted 1:25 (British BioCell International, Cardiff, UK). The sections were then stained with uranyl acetate and examined using a Philips CM10 electron microscope. Images of math5 −/− tissue (n = 20, P22) and WT tissue (n = 20, P21) were collected with a digitizing camera (SIA-7C; SIA, Duluth, GA). In each sample of images, all nonGABAergic terminal profiles were measured using the SIA software.