Distinct roles of transcription factors Brn3a and Brn3b in controlling the development, morphology, and function of retinal ganglion cells

Brn3a and Brn3b alleles that conditionally activate alkaline phosphatase expression

To both genetically modify and selectively visualize individual Brn3a- or Brn3b-expressing RGCs, we constructed conditional alleles of the Brn3a and Brn3b genes by inserting loxP sites 5′ and 3′ of each Brn3 coding region together with a human placental alkaline phosphatase (AP) coding region immediately downstream of the 3′ loxP site (Figure 1A). A series of transcription termination signals inserted immediately upstream of the 3′ loxP site minimized any read-through into the AP coding region. Following Cre-mediated recombination, the AP coding region was faithfully expressed under the control of the adjacent Brn3 promoter (Figure 1 and Supplementary Figure 1).

In the experiments described below, we used either of two heterozygous configurations at the Brn3a or Brn3b loci: the conditional allele over the wild type allele (e.g. Brn3b^CKOAP/+) or the conditional allele over a conventional null allele (e.g. Brn3b^CKOAP/−). In the former case, Cre-mediated recombination results in AP expression in cells that are heterozygous for the Brn3 gene of interest (Brn3a^AP/+ or Brn3b^AP/+) and these cells appear to be phenotypically wild type (WT), while in the latter case Cre-mediated recombination results in AP expression in cells that carry two mutant alleles (Brn3a^AP/− or Brn3b^AP/−). Brn3a^CKOAP/+ and Brn3b^CKOAP/+ mice carrying the ubiquitously expressed Rosa26CreER(T) locus (Badea et al., 2003) exhibit AP activity in the retina and in retino-recipient areas in the brain, as well as in a variety of other brain regions (Figure 1C,E–H). Brn3a^CKOAP/+ or Brn3b^CKOAP/+ mice carrying the retina-specific Pax6αCre transgene (Marquardt et al., 2001) exhibit AP activity only in the retina and in retinorecipient areas of the brain (Figure 1D,I–L). As described by Marquardt et al. (Marquardt et al., 2001) and as seen in Figure 1D, Pax6αCre is expressed at a very low level in a wedge-shaped stripe in the dorsal and central retina; within this zone, the morphologies of well-isolated AP+ RGCs can be analyzed. The Pax6αCre transgene is expressed in the retina beginning at embryonic day (E)9.5 (Marquardt et al., 2001), three days before the onset of Brn3b expression and four days before the onset of Brn3a expression in postmitotic RGC precursors (Xiang, 1998).

RGCs expressing Brn3a and Brn3b show distinct patterns of dendritic stratification

Two approaches based on visualizing the morphologies of individual RGCs in flat-mounted retinas were used to define the wild type (WT) patterns of dendritic arborization in those RGCs that express Brn3a or Brn3b. In the first approach, we analyzed RGC morphologies following sparse and random AP labeling of neurons in mice carrying Rosa26CreER(T) and Z/AP (a transgene that can express AP in essentially any cell type following Cre-mediated recombination (Badea and Nathans, 2004; Lobe et al., 1999)) and determined which AP+ RGCs expressed Brn3a (Figure 2A–D) or Brn3b by immunostaining. In the second approach, we visualized the morphologies of well-isolated AP+ RGCs in Brn3a^CKOAP/+;Pax6αCre and Brn3b^CKOAP/+;Pax6αCre retinas (Figure 2I–O). For each RGC analyzed, the locations of the dendrites were digitized, and the area of the dendritic arbor in the plane of the retina and its vertical position and thickness in the IPL were determined (Figure 2E,F). As seen in the scatter plots for monostratified RGCs (Figure 2G,H,P), both approaches reveal a striking morphological difference between Brn3a and Brn3b expressing RGCs: while arbors of Brn3b-expressing RGCs are located throughout the full extent of the IPL, the arbors of Brn3a-expressing RGCs are confined to the outer ~60–70% of the IPL. Moreover, the data in Figure 2G show that 20/20 monostratified RGCs that did not express Brn3a stratify within the inner 30–40% of the IPL. Among Brn3a^CKOAP/+;Pax6αCre and Brn3b^CKOAP/+;Pax6αCre RGCs, 19% and 9%, respectively, are bistratified (Supplementary Figure 2).

As an independent measure of these patterns of dendritic stratification, vertical sections of Brn3a^CKOAP/+;Pax6αCre and Brn3b^CKOAP/+;Pax6αCre retinas were immunostained for AP and for calretinin. Calretinin localizes to three narrow bands within the IPL, with the central band marking the junction between ON and OFF sublaminae (Morgan et al., 2006). For this analysis, we focused on regions of the retina with a high level of Cre-mediated recombination where the AP signal within any section is derived from the dendrites of a large number of RGCs. As seen in Figure 3 and Supplementary Figure 3 and in agreement with the analyses of individual RGC morphologies described above, Brn3a^AP/+ RGC dendrites are confined to the outer ~60–70% of the IPL whereas the Brn3b^AP/+ RGC dendrites populate the full width of the IPL, although with a relatively reduced representation in the innermost ~50% of the OFF substrata demarcated by the outer two bands of calretinin staining. Finally, we observed by double labeling of adult Brn3a^CKOAP/+;Pax6αCre and Brn3b^CKOAP/+;Pax6αCre with anti-AP and with anti-Brn3a or anti-Brn3b antibodies that 87% of Brn3a^AP RGCs express Brn3b and 78% of Brn3b^AP RGCs express Brn3a (Supplementary Figure 1A–I).

Brn3a^−/− RGCs exhibit defects in dendritic stratification

A comparison of vertical sections of Brn3a^CKOAP/+;Pax6αCre and Brn3a^CKOAP/−;Pax6αCre retinas shows that the absence of Brn3a leads to a loss of AP-expressing RGC dendrites in a narrow stripe centered over the boundary between ON and OFF laminae in the IPL (compare Figure 3A and B). A similar gap is seen when the dendritic stratification patterns of individual monostratified Brn3a^CKOAP/− RGCs are plotted (Figure 4A). This observation indicates that loss of Brn3a leads either to a loss of Brn3a-expressing RGCs that would normally stratify in this central layer, to a re-specification of dendritic stratification within this RGC population, or both. A comparison of the morphologies of individual Brn3a^AP/+ and Brn3a^AP/− RGCs indicates a substantial degree of re-specification among the latter group, as the ratio of bistratified: monostratified RGCs increases from 0.22 (8:35) in Brn3a^CKOAP/+;Pax6αCre retinas to 1.5 (24:16) in Brn3a^CKOAP/−;Pax6αCre retinas (P<0.0001; Figure 4B,C and Supplementary Figure 2). The absence of Brn3a is also associated with a ~30% decrease in the number of RGCs fated to express Brn3a (as determined by counting Brn3a^AP/+ and Brn3a^AP/−RGCs), with a corresponding decrement in the number expressing Brn3b (Supplementary Figure 1I–L). Thus, in the absence of Brn3a, there is both RGC respecification and loss.

Brn3b^−/− RGCs exhibit defects in intra-retinal axon morphology

As previously reported (Erkman et al., 1996; Erkman et al., 2000; Gan et al., 1996; Xiang, 1998), loss of Brn3b leads to a loss of >70% of RGCs (Figure 3C,D and Supplementary Figures 1, 3C,D). AP staining of Brn3b^CKOAP/−;Pax6αCre retinas shows that many surviving Brn3b^AP/− RGCs continue to transcribe the recombined Brn3b locus (Figure 4D). In earlier work, Gan et al. (Gan et al., 1999) and Wang et al.(Wang et al., 2000) observed disorganized RGC axon bundles in Brn3b^−/− retinas and aberrant axonal and dendritic processes in cultured Brn3b^−/− RGCs. We have extended this analysis by observing the aberrant morphologies of individual Brn3b^AP/− RGCs in retina flat mounts (Figure 4G–J). Although the distribution of stratification levels among Brn3b^AP/− RGC dendrites resembles that of the Brn3b^AP/+ control (compare Figures 2P and 4A), the population of Brn3b^AP/− RGCs includes an excess of cells with large areas and thick arbors (Supplementary Figure 2C), and an overall increase in mean area size for dendritic arbors of all types (compare Figures 2P and 4A). While the axons of most Brn3b^AP/− RGCs correctly target the optic disc, some axons or axon-like processes from Brn3b^AP/− cells follow aberrant trajectories (Figure 4D,G), bifurcate within the retina (Figure 4I), or give rise to dendrite-like branches that ramify within the IPL (Figure 4G–J). Among a sampling of 87 Brn3b^AP/− neurons with cell bodies in the ganglion cell layer (GCL), three lacked an axon entirely (Figure 4F), suggesting that they may have adopted, at least partially, an amacrine cell fate. The idea that amacrine cell developmental pathways are derepressed in the absence of Brn3b is strengthened by the observation that Brn3b^AP/− neurons are significantly more likely to reside in the inner nuclear layer (INL); the ratio of INL:GCL Brn3b^AP/− cell bodies is 178:105 as compared to 9:147 for Brn3b^AP/+ controls (Figure 4E). Moreover, many displaced Brn3b^AP/− cells exhibit morphologies characteristic of small-field glycinergic amacrine cells (Menger et al., 1998), and immunostain for the vesicular inhibitory amino acid transporter (Chaudhry et al., 1998) (Supplementary Figure 4). These observations are consistent with those of Qiu et al., (Qiu et al., 2008) showing that amacrine markers are expressed transiently in prenatal Brn3b^−/− RGCs. Among surviving Brn3b^AP/− RGCs, the fraction that expresses Brn3a is unaltered compared to Brn3b^AP/+ RGCs (Supplementary Figure 1I–L), implying that, for at least some RGCs that normally co-express Brn3a and Brn3b, Brn3b is not required for the maintenance of Brn3a expression.

Defects in central projections from Brn3b^−/− but not Brn3a^−/− RGCs

As revealed by AP histochemistry, Brn3a^AP/+ and Brn3b^AP/+ RGCs project to a diverse set of central targets that mediate both image-forming and non-image-forming vision: the lateral geniculate nucleus (LGN), the superior colliculus (SC), the dorsal, lateral, and medial terminal nuclei (DTN, LTN, and MTN), the olivary pretectal nucleus (OPN), and the pretectal area (PA) and adjacent nucleus of the optic tract (NOT) (Figure 1I–L; Figure 5A–J). The suprachiasmatic region and the intergeniculate leaflet (IGL) receive projections from Brn3b-expressing RGCs but not from Brn3a-expressing RGCs (Figure 1I,K; Figure 5A,F). These observations regarding the central targets of Brn3a-expressing RGCs extend an earlier analysis that utilized a Brn3a^tau-lacZ allele in which projections to the accessory optic system were not detected (Quina et al., 2005), a difference that we ascribe to the different sensitivities of the AP and tau-lacZ reporters.

In contrast to the normal or nearly normal projections to central targets from Brn3a^AP/− RGCs (Figure 5A′-E′), projections from Brn3b^AP/−RGCs are greatly reduced, mirroring the reduction in the number of Brn3b-expressing RGCs (Figure 5F′-J′ and O–Q′). Large decrements in Brn3b^AP/− projections are apparent in the OPN, PA, LGN, and SC. In the accessory optic system projections to the LTN and MTN are eliminated and the accessory optic tract (AOT) is missing (Figure 5I,I′, P–Q′). Within the suprachiasmatic region, Brn3b^AP/− axon arbors shift from a narrow lateral distribution to a central and more diffuse distribution (Figure 5F, F′). In the SC the topographic arrangement of Brn3b^AP/− fibers is minimally perturbed, as shown by the central unstained stripe that bisects the SC in the coronal plane and that corresponds to the dorso-ventral stripe of low Cre activity in the Pax6αCre retina (Figure 5O,O′). Axon tracing by intraocular injection of fluorescent cholera toxin B in Brn3b^−/− mice (Figure 5K–N′) shows a nearly normal density of projections to the SCN, but for other central targets there is a decrement in retinal projections roughly proportional to the decrement observed for the projections from Brn3b^AP/− RGCs. In Brn3b^−/− mice, the MTN receives little or no retinal input of any kind (Figure 5M and M′), implying that this input is normally derived from (or is dependent on) Brn3b-expressing RGCs.

Development of Brn3a^AP/− RGC dendrites and Brn3b^AP/− RGC axons

Defects in dendritic morphology and axon targeting can arise from defects in specification or pathfinding early in development, aberrant pruning later in development, or some combination of these processes. To define the development of dendritic morphologies and axon trajectories for wild type and mutant RGCs in a cell type-specific manner, we imaged developing Brn3a^AP/+, Brn3a^AP/−, Brn3b^AP/+, and Brn3b^AP/−RGCs in late fetal and early postnatal life. In the postnatal day (P)4 retina - a time when IPL lamination is still incomplete as judged by the presence of a single band of calretinin immunoreactivity – the dendrites of Brn3a^AP/+ RGCs are confined to a discrete zone at the center of the IPL (Figure 6A–D and Supplementary Figure 5). By P11, this pattern has evolved to match that of the adult IPL. Interestingly, the distinctive gap in the distribution of Brn3a^AP/− dendrites within the central IPL is already evident at P4 (Figure 6B and Supplementary Figure 5B). These data imply that the cell-autonomous differences in dendritic stratification between Brn3a^AP/+ and Brn3a^AP/− RGCs are programmed early in IPL development, prior to the time when bipolar cells are fully developed and activity-dependent dendritic refinement is thought to act (Morgan et al., 2006; Tian and Copenhagen, 2003).

In earlier descriptions of the developing Brn3b^−/− CNS, diI labeling of RGC axons revealed erroneous trajectories at the optic chiasm and SC (Erkman et al., 2000). By specifically visualizing the central projections of Brn3b^AP/− RGCs, we observed that a small number of axons from Brn3b^AP/− RGCs project to the developing MTN at E17 and are then progressively lost over the first postnatal week (Figures 6G–H′ and 5I,I′). We also observed a distinctive defect beginning at P4 in which numerous neurites emerge from the optic tract and penetrate nearby nontarget areas (Figure 6E–F′). These neurites increase in number through P9 but are lost by adulthood. We speculate that these relatively late developmental phenotypes may reflect earlier errors in targeting or synaptogenesis by Brn3b^AP/− RGC axons.

Visual sensory defects in Brn3b^−/− mice

The widespread reduction in projections to retinorecipient areas in the Brn3b^−/− brain, and, in particular, the complete absence of projections to the LTN and MTN, suggested the possibility that Brn3b^AP/− mice might exhibit distinctive deficits in visual behavior. To test this hypothesis we quantified eye tracking in response to a pattern of moving stripes (the optokinetic reflex, OKR), the pupillary light response, and circadian photoentrainment in Brn3b^−/− and control (Brn3b^+/− and Brn3b^+/+) littermates.

In Brn3b^−/− mice the OKR to horizontally moving stimuli is reduced to roughly one-third of the WT value and the response to vertically moving stimuli is abolished (Figure 7A and B). This difference between horizontal and vertical OKRs suggests that in the Brn3b^−/− brain, the loss of projections to the MTN and LTN, which control vertical eye movements, is more severe than the loss of projections to the DTN and NOT, which control horizontal eye movements (Simpson, 1984). Brn3b^−/− mice also show a profound defect in the central control of pupil constriction (Figure 7C–E). While the innervation and competence of the pupillary muscles appear to be normal, as demonstrated by full pupil constriction in response to topical carbachol (Figure 7C), the pupillary response to changes in ambient light level is greatly reduced in sensitivity, typically delayed by tens of seconds, and unstable (Figure 7D and E). This defect is likely referable to a loss of projections to the PA and OPN (Gamlin, 2005).

To test photoentrainment, wheel running activity was recorded from seven Brn3b^−/− and seven Brn3b^+/− control mice subjected to 12 hour:12 hour light: dark (12:12 LD) cycles with progressively dimmer light regimens (Figure 7F, panels i, ii, iii). Two Brn3b^−/− mice free-ran with a period less than 24 hours in all light regimens (Figure 7F, right panels; “severe phenotype”). The remaining five Brn3b^−/− mice were able to photoentrain, albeit weakly and with an unstable phase relation to the imposed LD cycle, and often had what appeared to be a split rhythm (Figure 7F, center panels; “mild phenotype”). Under constant darkness (Figure 7F, panel iv), Brn3b^−/− and Brn3b^+/− mice exhibited similarly shortened periodicities (controls: 23.69 ± 0.08 hours; Brn3b^−/−: 23.71 ± 0.25 hours), implying the existence of a normally functioning circadian clock in Brn3b^−/− mice. When a 15 minute light pulse was given four hours after the onset of the subjective night (Figure 7F, panel iv), all Brn3b^+/− mice responded with a robust phase shift (−2.37 ± 0.79 hours), while Brn3b^−/− mice were unresponsive (−0.11 ± 0.27 hours; P<0.01). Mice were then subjected to a jet-lag paradigm in which their imposed LD cycle was advanced and then delayed by six hours (Figure 7F, panel v). In response to the new LD cycle, Brn3b^+/− mice rapidly resynchronized, mild phenotype Brn3b^−/− mice gradually resynchronized, and severe phenotype Brn3b^−/− mice did not resynchronize. Constant light strongly suppressed the activity of Brn3b^+/− mice, but had little affect on the activity or periodicity of Brn3b^−/− mice (Figure 7G). Finally, to directly observe masking responses independent of the circadian clock, we used an ultradian paradigm consisting of 3.5:3.5 LD cycles (Figure 7H). Brn3b^−/− mice were substantially more active in the light than their control counterparts: the percentage of daily activity that occurred within the light phase was 4.72 ± 2.65% in controls versus 39.12 ± 14.22% in Brn3b^−/− mice (P<0.05). These observations indicate that the retinal inputs to the SCN that remain in Brn3b^−/− mice (Figure 5K,K′), although substantial, are insufficient to mediate normal photoentrainment. Defects in other retinorecipient nuclei, such as the OPN or IGL, which receive input from ipRGCs (Hattar et al., 2006), project to the SCN (Moga and Moore, 1997), and are affected in Brn3b^−/− mice, may account, at least in part, for the photoentrainment defects.

Brn3 transcription factors and the specification of neuronal identity

Understanding the transcriptional regulatory networks that control neuronal diversity is one of the central aims of developmental neurobiology. In the developing spinal cord, investigations over the past two decades have shown that there are complex combinatorial, partially redundant, and stage-specific interactions among transcription factor genes that determine the distinctive properties of different groups of motor neurons. For example, the territories defined by Hox gene expression and retinoid receptor signaling control rostro-caudal diversity among motor neurons, and the combinatorial action of LIM-homeodomain genes controls axon pathfinding and target selection (Maden, 2006; Shirasaki and Pfaff, 2002). In the retina, a large number of transcription factors have been identified that control cell fate decisions within the major classes of neurons, including rods and cones (Furukawa et al., 1999; Ng et al., 2001; Oh et al., 2007), bipolar cells (Cheng et al., 2005; Chow et al., 2004), horizontal and amacrine cells (Elshatory et al., 2007; Fujitani et al., 2006; Li et al., 2004), and RGCs (Mu et al., 2008; Qiu et al., 2008). In general, the mechanisms by which these transcription factor genes produce their cellular effects –i.e. the identities and functions of their target genes – remain unknown.

Among transcription factor genes implicated thus far in controlling retinal development, the Brn3 genes are distinctive in the RGC-specificity of their expression, their high mutual sequence similarity, and their role in the development of primary sensory cells in other sensory systems as revealed by their knockout phenotypes. How the last of these attributes relates to their roles in controlling RGC development is unclear. In the present work, the identification of distinctive patterns of dendritic stratification by Brn3a- and Brn3b-expressing RGCs, together with the differential innervation of the SCN and IGL by these RGCs, provides the first clear connection between morphologically defined RGC subtypes and patterns of transcription factor expression. Moreover, the observation that loss of Brn3a leads to a dramatically altered pattern of dendritic stratification, whereas loss of Brn3b largely affects axonal development, pathfinding, and target selection, argues that, despite their high degree of amino acid sequence similarity, Brn3a and Brn3b perform qualitatively different functions during RGC development.

Thus far, experiments aimed at defining genes regulated by Brn3a in the somatosensory system (Eng et al., 2004; Lanier et al., 2007) and by Brn3b in the retina (Erkman et al., 2000; Mu et al., 2004), have been performed on bulk samples, with the result that distinctions between neuronal subtypes have not been revealed. Distinguishing between subtypes is likely to be important because the actions of the different Brn3 proteins on their transcriptional regulatory targets may depend on the cellular context, as suggested, for example, by the contrast between the presence of axon pathfinding defects among Brn3a^−/− dorsal root ganglion neurons (Eng et al., 2001) and the absence of axonal pathfinding defects among Brn3a^−/− RGCs. In this example, a substantial part of the context effect may reflect redundancy with co-expressed Brn3b in RGCs. Alternately, the afferent of each DRG neuron, as the input branch of a pseudo-unipolar neurite, may, in some of its molecular properties, be developmentally homologous to the dendritic arbors of RGCs, which, as shown here, are dependent on Brn3a function.

A potentially useful model for Brn3 function within a larger transcriptional network can be found in Drosophila, where the POU-domain gene Acj6 appears to participate in a combinatorial code with multiple other transcription factors to control the dendritic specificity of olfactory projection neurons and the stereotyped branching of the axons of olfactory receptor neurons in their target glomeruli (Clyne et al., 1999; Komiyama et al., 2004; Komiyama et al., 2003; Komiyama and Luo, 2007). As Acj6 is the sole Drosophila orthologue of the vertebrate Brn3 family, it may be of interest to examine mammalian homologues of those Drosophila genes identified as part of the Acj6 regulatory network as candidates that might interact with Brn3 family members.

Brn3 control of dendritic stratification, axonal structure, and target selection

Dendritic stratification is an organizational strategy used in a variety of contexts in the CNS, including the retina, the dorsal horn of the spinal cord, the cerebral cortex, and the LGN. It has been most intensively studied in the retina, where the IPL is comprised of at least ten distinct strata and most RGCs have dendritic arbors that are either monostratified or bistratified. Correlations between the level of dendritic stratification and the physiologic response properties of individual RGCs show that different IPL strata extract (and therefore different RGCs transmit) different features of the visual stimulus (Roska and Werblin, 2001). Although the molecular mechanisms controlling IPL stratification are still largely unknown, recent work has implicated the Sidekick and Dscam members of the Ig protein superfamily in stratification specificity (Yamagata and Sanes, 2008; Yamagata et al., 2002). These cell surface proteins are expressed in largely nonoverlapping subsets of RGCs, bipolar cells, and amacrine cells, and each protein mediates homophilic but not heterophilic binding, suggesting a simple homophilic adhesion model for the mutual recognition of synaptic partners in the IPL. However, it is not currently known how synaptic partners determine which stratum to occupy within the IPL and how the two partners are programmed to express the same adhesion molecule. RGC dendritic stratification is also influenced by synaptic activity, as pharmacologic blockade of glutamate release or light deprivation in the developing retina impairs its refinement (Bisti et al., 1998; Bodnarenko and Chalupa, 1993; Tian and Copenhagen, 2003; Xu and Tian, 2007).

The conclusion that Brn3a is part of a genetic regulatory network that controls IPL stratification is based on (1) the stratification differences between Brn3a-expressing and Brn3b-expressing RGCs and (2) the effect of Brn3a deletion on the distribution of RGC dendrites within IPL strata and on the ratio of bistratified to monostratified RGCs. With respect to the increased ratio of bistratified to monostratified RGCs, we note that it is possible to account for this increase either by a conversion of monostratified to bistratified RGCs or by a selective loss of monostratified RGCs. We favor the former explanation because (1) the altered stratification pattern is observed at a time (P4) when RGC dendrites are just beginning to stratify in the IPL (Figure 6 and supplementary Figure 5), thus inconsistent with late elimination of monostratified RGCs, and (2) an RGC elimination mechanism alone cannot account for the appearance of Brn3a^AP/− RGC dendrites at the level of the innermost of the three calretinin strata, a stratum that is minimally populated by Brn3a^AP/+ RGC dendrites (Figure 3).

In contrast to Brn3a^AP/− RGCs, Brn3b^AP/− RGCs show an essentially normal distribution of dendritic morphologies. However, by selectively visualizing Brn3b^AP/− RGCs we observe a wide variety of axonal defects that are consistent with in vitro culture experiments showing that Brn3b^−/− axons have polarization defects and have acquired some of the properties of dendrites (Wang et al., 2000). Earlier diI tracing analyses had indicated defects among Brn3b^AP/− axons in pathfinding at the optic chiasm and in retinotopic mapping in the colliculus (Erkman et al., 2000). By selectively visualizing Brn3b^AP/− axons, we additionally observe defasciculated processes invading inappropriate territories along the optic tract (Figure 6), but we do not see evidence for large-scale defects in retinopy in the SC (Figure 5O and O′). In light of the qualitative differences that we observe in Brn3a and Brn3b loss-of-function RGC phenotypes, it would be of interest to re-examine the retinas of mice in which Brn3a coding sequences were substituted for Brn3b coding sequences (Pan et al., 2005) to determine whether RGC morphologies are or are not affected.

Genetic analysis of mammalian neural development at single cell resolution

The conditional knockout approach presented here and in a recent study of Frizzled5 function in the thalamus (Liu et al., 2008) – in which Cre-mediated excision of a target gene’s coding region and 3′UTR also leads to the expression of a histochemical reporter - represents a new strategy for analyzing the morphology and function of individual genetically altered mammalian neurons. In its current form, this strategy is most easily applied to relatively compact genes, in which case a single gene targeting construct can be used to engineer loxP sites both 5′ and 3′ of the coding region. This strategy also requires a conventional loss-of-function allele that lacks the histochemical reporter, since sparse Cre-mediated recombination requires that the conditional allele be present in only one copy. The present work demonstrates the general utility of this approach.

Related approaches for visualizing and/or altering mammalian neurons include (1) visualizing the progeny of a progenitor cell population by activating an unlinked lacZ reporter with a target gene-CreER knock-in (Zirlinger et al., 2002), and (2) activating loxP recombination by sparse co-expression of GFP with Cre or CreER from a thy-1 transgene or following transduction by AAV or lenitivirus vectors (Kasper et al., 2007; McCarty et al., 2004). In Drosophila, individual neurons can be manipulated by using mitotic recombination to simultaneously mark a cell and render it homozygous for a target gene mutation, (‘mosaic analysis with a repressible cell marker’/’MARCM’; (Komiyama et al., 2004; Lee and Luo, 1999), an approach that has recently been extended to the mouse (‘mosaic analysis with a double reporter’/’MADM’;(Zong et al., 2005)).

Potential extensions of the strategy used here include: (1) indelibly marking the cell that has undergone Cre-mediated recombination by placing an unrelated site-specific recombinase 3′ of the target gene and crossing in a reporter controlled by the second recombinase; (2) using reporters other than AP - in particular, fluorescent or epitope-tagged fusion proteins that localize to subcellular structure such as synapses; and (3) using cell-type specific Cre lines to selectively target neuronal subpopulations defined by the intersection of the expression patterns of the recombinase and its target. Given the importance of visualizing and genetically manipulating individual identified mammalian neurons, we predict that these strategies will become widely used.

Mouse lines

The Pax6αCre (Marquardt et al., 2001), ROSA26CreER(T) (Badea et al., 2003), and the Brn3a (Xiang et al., 1996) and Brn3b (Gan et al., 1996) conventional knock-out lines were described previously. The Brn3a^CKOAP and Brn3b^CKOAP conditional alleles were generated by homologous recombination in mouse embryonic stem cells using standard techniques. For the targeted alleles, the following changes were made: a loxP site was inserted in the 5′UTR 42 bp 5′ (for Brn3a) or 98 bp 5′ (for Brn3b) before the initiator ATG; three repeats of the SV40 early region transcription terminator were added to the 3′UTR 48 bp (for Brn3a) or 340 bp (for Brn3b) 3′ of the Brn3 translation termination codon, followed by a second loxP site and the coding region of human placental alkaline phosphatase (AP). A positive selection cassette (PGK-Neo), flanked by FRT sites, followed the AP coding region, and was subsequently removed by crossing to mice expressing FLP recombinase in the germline.

Histology

Retina whole mounts and brain sections were stained, processed, and imaged, and RGC dendrites and axons were traced and quantified as previously described (Badea and Nathans, 2004; Badea et al., 2003). High resolution images were captured on a Zeiss Imager. Z1 fitted with an Apotome for fluorescent imaging and Axiovision software. Rabbit polyclonal anti-Brn3a and anti-Brn3b antisera were described in Xiang (Xiang et al., 1995). The sources of commercial antibodies are: sheep polyclonal anti-AP (American Research Products, Belmont, MA), rabbit polyclonal anti-calretinin and anti-calbindin (Swant, Bellinzona, Switzerland), and rat monoclonal anti-Thy-1 (BD Biosciences). Cholera Toxin B conjugates were from Molecular Probes (Eugene, OR).

Optokinetic Reflex (OKR)

The OKR apparatus and recording methodology are described in detail in Cahill and Nathans (Cahill and Nathans, 2008). In brief, a head-posted mouse was immobilized in an acrylic holder in the center of a 29.5 cm diameter vertical white cylinder. A computer-generated image of alternating black and white vertical stripes (each stripe subtending 4 degrees of visual angle) was projected into the inner walls of the drum by a rotating projector mounted on the ceiling. The striped pattern had an average brightness of 100–200 lux and was rotated at 5°/sec. 30-second stimulus presentations were alternated with 30 seconds of a uniform grey. Eye movements were captured with an infrared video imaging system (ISCAN, Cambridge, MA), and stored, processed and displayed using Microsoft Excel. The number of eye tracking movements (ETMs; defined as a saccade followed by a slow tracking movement in the opposite direction) was counted per 30-second interval. Eye movements along the vertical axis (with respect to the mouse) were recorded by rotating the acrylic mouse holder 90 degrees so that the nose of the mouse pointed vertically. With the stripes now moving dorsal-to-ventral or ventral-to-dorsal with respect to the mouse, the same rotating visual stimulus was presented only to the eye from which the infrared video recording was obtained. To produce a degree of pupil constriction in Brn3b^−/− mice that matches the constriction of WT mice, Brn3b^−/− mice were treated with several microliters of 10% pilocarpine eyedrops just before OKR testing.

Pupil Constriction

The mouse’s head was stabilized as described for the OKR. Two different protocols were used to measure the pupillary light response. In the first, the environment was darkened or illuminated with intensities of 1, 10 and 400 lux in alternating 30-second intervals, a regimen that was repeated three times, progressing from dim to bright light. In the second protocol, mice were either dark adapted or exposed to room light (~100 lux), and then maintained for one minute in complete darkness, exposed to 400 lux for five minutes, and returned to complete darkness for a final one minute. Light intensities were controlled with neutral density filters and both eyes were illuminated equivalently.

Photoentrainment and light-induced activity suppression

Mice were placed in cages equipped with a 4.5 inch running wheel and exposed to different light regimens as described in the text. Light intensity was 1000 lux unless otherwise noted. Wheel running activity was monitored using VitalView software (Mini Mitter Co., Bend, OR). Data was analyzed and actograms were generated with ClockLab software (Actimetrics, Wilmette, IL).