Gbx2 regulates thalamocortical axon guidance by modifying the LIM and Robo codes

The assembly of neural circuits in the central nervous system exhibits remarkable precision. Significant progress has been made in our understanding of the extrinsic signaling mechanisms that regulate axon pathfinding, as well as intrinsic determinants, such as transcription factors, that control the identity of neurons (O’Donnell et al., 2009). However, far less is understood about the transcriptional control of intrinsic responsiveness to guidance cues that lead to proper wiring of the brain.

The thalamus plays a pivotal role in integrating and processing visual, auditory, sensory and motor information to and from the cortex (Jones, 2007). During embryogenesis, thalamic axons, known as thalamocortical axons (TCAs), have to navigate along a complex path to reach the cerebral cortex. In mouse embryos, TCAs first grow rostrally through the prethalamus, reaching the border between the diencephalon and telencephalon, where they make a sharp turn to enter the ventral telencephalon at embryonic day (E) 12.5 (López-Bendito and Molnar, 2003; Garel and Rubenstein, 2004). By E14.5, TCAs reach the junction between the ventral and dorsal telencephalon, and then turn dorsally and extend to the intermediate zone and subplate below the cortical plate. The axons wait in the subplate before invading the overlying cortical plate and finally terminate in cortical layers IV and VI soon after birth (O’Leary and Koester, 1993). Each thalamic nucleus connects with a specific set of cortical areas, and subsequently TCAs from a given thalamic nucleus form a topographic map within a specific cortical area. It has been shown that TCAs are sorted in the ventral telencephalon, thus leading to the establishment of their topography (Métin and Godement, 1996; Molnar et al., 1998; Seibt et al., 2003; Powell et al., 2008). Genetic studies have demonstrated that Slit-Robo signaling plays an important role in the navigation of TCAs in the ventral telencephalon (Bagri et al., 2002; López-Bendito et al., 2007; Bielle et al., 2011a; Bielle et al., 2011b). Furthermore, it has been shown that Slit1 is involved in establishing the topography of TCAs in the ventral telencephalon (Bielle et al., 2011b). Robo2 is broadly expressed in the developing thalamus, whereas the expression of Robo1 is mainly restricted to the medial part of the thalamus (Bagri et al., 2002; López-Bendito et al., 2007). The differential expression of Robo1 and Robo2 in distinct groups of thalamic neurons may be important for the navigation and topographic sorting of TCAs.

Members of the LIM-homeodomain (LIM-HD) and basic helix-loop-helix (bHLH) transcription factor families are expressed in different domains of the thalamus, suggesting that these proteins may form combinatorial transcriptional codes to impart neuronal identity and connectivity to thalamic neurons (Nakagawa and O’Leary, 2001). It has recently been shown that LIM-HD factor Lhx2 directly binds to putative regulatory elements in the Robo1 and Robo2 loci, and negatively regulates their transcription (Marcos-Mondejar et al., 2012). However, how Lhx2 generates the different expression patterns of Robo1 and Robo2 in the thalamus by the same inhibitory regulation remains unclear. Expression of Gbx2 (gastrulation brain homeobox gene 2) is initiated in postmitotic neurons of the mouse thalamus at E10.5 (Bulfone et al., 1993). Deletion of Gbx2 causes severe defects in the histogenesis of the thalamus and an almost complete loss of TCAs (Miyashita-Lin et al., 1999; Hevner et al., 2002). We have recently shown that, although all thalamic neurons express Gbx2 during development, the onset and duration of Gbx2 expression are highly variable among different thalamic nuclei (Chen et al., 2009). These observations underscore the importance of Gbx2 in the regulatory network that coordinates thalamic neuron specification and connectivity.

In this study, we have examined Gbx2 function in controlling TCA development. We demonstrate that Gbx2 determines the initial directional outgrowth of TCAs into the prethalamus and their subsequent pathfinding to the cortex by regulating their responsiveness to guidance cues. We show that the guidance errors of Gbx2-deficient TCAs were associated with mis-regulation of Robo1 and Robo2 expression. We have identified Lmo3 as a putative direct transcriptional target of Gbx2. Our genetic and molecular biological data collectively demonstrate that Gbx2 regulates the LIM transcriptional codes comprising Lhx2, Lhx9 and Lmo3, which subsequently control the differential expression of Robo1 and Robo2 in the thalamus.

All animal procedures described herein were approved by the Animal Care Committee at the University of Connecticut Health Center. All mouse strains were maintained on an outbred genetic background. Noon of the day on which a vaginal plug was detected was designated as E0.5 in the staging of embryos. The knock-in Gbx2^creER allele and Gbx2^floxed conditional mutant allele (Gbx2^F) have been described previously (Li et al., 2002; Chen et al., 2009). To perform conditional deletion of Gbx2, 4-6 mg of tamoxifen (Sigma, St Louis, MO, USA) in corn oil was administered to pregnant females carrying Gbx2^creER/F; R26R^lacZ/+ by oral gavage. Two other mouse lines harboring the following mutant alleles were used in the study: Lhx2^–, a null allele of the Lhx2 gene (Porter et al., 1997), and Lmo3^lacZ, which is a lacZ knock-in and also a null allele of the Lmo3 gene (Tse et al., 2004).

Embryonic mouse brains were dissected in cold phosphate-buffered saline (PBS) and fixed in 4% paraformaldehyde (PFA) for 40 minutes. Brains were cryoprotected, frozen in OCT freezing medium (Sakura Finetek) and sectioned.

Standard protocols were used for X-gal histochemistry, immunofluorescence and in situ hybridization, as described previously (Chen et al., 2010). Detailed protocols are available on the Li Laboratory website (http://lilab.uchc.edu/protocols/index.html). Primary antibodies used in the study were as follows: rabbit anti-GFP (Invitrogen), rat anti-GFP (Nacalai Tesque), rabbit anti-Lhx9 (Santa Cruz), mouse anti-Nefm (2H3-DSHB), and rabbit anti-Robo1 and rabbit anti-Robo2 (kind gifts from Dr Atsushi Tamada, Niigata University, Japan). Alexa fluorescent secondary antibodies (Invitrogen) were used.

Chimeric embryos were generated by aggregation of morula, which resulted from intercrosses between Gbx2^+/– and Gbx2^creER/+ mice, and embryonic stem (ES) cells as described previously (Nagy et al., 2003). Chimeric embryos containing Gbx2^creER/+ or Gbx2^creER/– cells were identified by the presence of EGFP fluorescence, and these two groups of embryos were further distinguished by PCR genotyping the Gbx2^– and Gbx2^creER alleles (Wassarman et al., 1997; Li et al., 2002).

EMSA was performed according to standard protocols (Buratowski and Chodosh, 2001; Kain and Ganguly, 2001). The functional EGFP:Gbx2 fusion protein was generated by fusing EGFP to the N terminus of Gbx2. The DNA probe sequence is: Lmo3-ab, 5′-GATAATTAAACTAATTAGTTCCCCTGTG; Lmo3-cd, 5′-TGCTTTGTACCTAATTAGTGTCCTTGGA (Gbx2-binding sequence is underlined). The oligo with scrambled binding sequence is: Lmo3-ef, GATAATTAAAATTCATAGTTCCCCTGTG; Lmo3-gh, TGCTTTGTACTAACATTGTGTCCTTGGA (scrambled binding sequence is underlined).

Luciferase reporter assay was performed in P19 embryonal carcinoma cells using a Dual-Luciferase Assay System (Promega). A 1.2 kb genomic DNA fragment containing two putative Gbx2-binding sites upstream of Lmo3 transcription start site was cloned into a luciferase reporter pGL3 (Promega). Full-length cDNAs for Gbx2, EGFP:Gbx2, Lhx2, Lhx5, Lhx9, Otx2, Irx1 and Lmo3 were cloned in a CMV-expression construct. Lipofectamine 2000 (Invitrogen) was used for transfection of P19 cells.

Explant co-culture was performed as described previously (Bonnin et al., 2007). Explants were prepared from E13.5 Gbx2^creER/+; R26R^RFP/+ or Gbx2^creER/–; R26R^RFP/+ embryos that were given tamoxifen at E10.5 so that the thalamic neurons and their processes were labeled with RFP. Brain explants were co-cultured with aggregates of 293T cells transfected with CMV-EGFP alone or CMV-EGFP and CMV-SLIT2 in Matrigel (BD Biosciences). After 48 hours, the cultures were mounted on slides and imaged. Axon outgrowth originating from proximal and distal quadrants was determined by measuring the total RFP fluorescence associated with the axon-occupied area using ImageJ software (NIH) as described previously (Brose et al., 1999). Statistical difference between proximal and distal areas was determined by chi-square test.

We have previously generated a Gbx2^creER allele, which contains a creER-ires-EGFP sequence inserted into the 5′-untranslated region of the Gbx2 locus so that the expression of both creER and EGFP mimics the endogenous Gbx2 expression (Chen et al., 2009). In embryos carrying the Gbx2^creER allele, TCAs were specifically labeled by EGFP, providing an excellent tool with which to study TCA development. Immunohistochemistry for GFP showed that TCAs extended rostrally through the prethalamus reaching the diencephalic-telencephalic border in Gbx2^creER/+ embryos at E12.5 (Fig. 1A). Strikingly, the majority of GFP⁺ neurons projected dorsally and caudally, and few GFP⁺ axons were found in the prethalamus in E12.5 Gbx2^creER/– embryos, which are deficient for Gbx2 (Fig. 1B). In Gbx2^creER/+ embryos at E14.5, immunofluorescence for GFP and neurofilament (NF) showed that TCAs were specifically labeled by GFP and GFP immunoreactivity was absent from the fasciculus retroflexus (FR) tract, which originates from the habenula (Fig. 1C). However, in Gbx2^creER/– embryos, GFP⁺ thalamic axons were mostly misrouted and abnormally present in the FR tract (Fig. 1D). To rule out the possibility that the misrouted GFP⁺ axons resulted from ectopic expression of GFP in Gbx2^creER/– embryos, we performed retrograde labeling by inserting DiI and DiA crystals in the ventral midbrain and dorsal midline of the diencephalon, respectively. In control embryos, few thalamic neurons were labeled by either DiI or DiA (supplementary material Fig. S1A-B″,E). By contrast, in Gbx2^creER/– embryos, DiI and DiA backfilled neurons were abundantly found inside the thalamus, in additional to their normal presence in the habenula, pretectum and prethalamus (supplementary material Fig. S1C-D″,F). In agreement with the previous findings (Miyashita-Lin et al., 1999; Hevner et al., 2002), TCAs were mostly absent in the cortex of Gbx2^creER/– embryos at E16.5 (Fig. 1F). Therefore, our data demonstrate that Gbx2 is essential for determining the initial trajectory of TCAs. In the absence of Gbx2, most thalamic axons abnormally project to the ventral midbrain along the FR tract or dorsal midline of the diencephalon, resulting in the loss of TCAs.

Gbx2 expression persists in many thalamic neurons (Jones and Rubenstein, 2004; Chen et al., 2009). To investigate whether prolonged Gbx2 expression is required for TCAs to properly navigate to the cortex, we deleted Gbx2 by administering tamoxifen to Gbx2^creER/F embryos at E13.5, when most TCAs have passed the prethalamus. The conditional knockout (CKO) embryos are designated as Gbx2-CKO. Interestingly, removing Gbx2 at E9.5 or E13.5 resulted in abnormal trajectories of TCAs within the diencephalons, as found in Gbx2^creER/– embryos (Fig. 2B; Fig. 3D), indicating that Gbx2 controls the axonal trajectory of newly generated thalamic neurons at various stages. Additional guidance defects of TCAs were detected in E15.5 Gbx2-CKO embryos that were given tamoxifen at E13.5. In these mutants, many GFP⁺ TCA fibers failed to enter the ventral telencephalon and instead projected towards the ventral midline of the hypothalamus (Fig. 2D,J). Furthermore, some GFP⁺ fibers abnormally crossed over the TCA bundle at the entry of the internal capsule (Fig. 2F,J). As described previously (Lund and Mustari, 1977; Rakic, 1977), TCAs mostly remain in the subplate before invading the overlying cortical plate in control embryos at E15.5 (Fig. 2G). However, increased number of TCAs were found in the cortical plate in Gbx2-CKO mutants at E15.5 (Fig. 2H,J). Taken together, these results demonstrate that, in addition to its crucial role in regulating the initial trajectory of TCAs, Gbx2 continues to regulate the navigation of TCAs and the timing to innervate the cortical plate.

We have previously demonstrated that Gbx2 has a non-cell-autonomous function in development of the thalamus (Chen et al., 2009). We thus investigated the cell autonomy of Gbx2 function in TCA development. Because of the mosaic nature of creER-mediated recombination induced by tamoxifen, the thalamus in Gbx2^creER/F; R26R^lacZ/+ embryos that were given tamoxifen consisted of a mixture of Gbx2^creER/F (functionally wild type) and Gbx2^creER/– (Gbx2-deficient) cells as described previously (Chen et al., 2009) (supplementary material Fig. S2). Using a RFP cre-reporter line (Madisen et al., 2010), we showed that RFP+ axons, which were derived from recombined neurons (presumably Gbx2 deficient), abnormally projected to dopaminergic neurons in the ventral midbrain via the FR tract (Fig. 3A,B). Furthermore, thalamic efferents with abnormal trajectories appeared to originate from recombined cells, whereas non-recombined cells gave rise to normal TCAs in E13.5 Gbx2^creER/F; R26R^lacZ/+ embryos that received tamoxifen at E9.5 (Fig. 3D). These observations suggest a cell-autonomous requirement for Gbx2 in controlling the trajectory of thalamic axons. Because of the lack of anti-Gbx2 antibodies, we could not directly identify the individual Gbx2-deficient thalamic neurons in the mosaic embryos. To follow the behavior of Gbx2-deficient neurons, we generated chimeric embryos by aggregating Gbx2^creER/– embryos with wild-type ES cells. In control chimeric embryos that were composed of wild-type and Gbx2^creER/+ cells, Gbx2^creER/+ neurons (GFP⁺) projected rostrally into the prethalamus at E11.5 (Fig. 3E-G). A few processes that did extend into the epithalamus traveled only for a short distance and became tangled (Fig. 3F), suggesting that TCAs are inhibited from entering the epithalamus. In mutant chimeric embryos, Gbx2^creER/– (GFP⁺) cells intermingled with wild-type (GFP^–) cells in the MZ (Fig. 3H-J), demonstrating that loss of Gbx2 does not alter cell adhesion properties. Significantly, Gbx2^creER/– thalamic axons extended dorsally into the epithalamus and pretectum, and mostly failed to grow into the prethalamus (Fig. 3H-J). Altogether, our results demonstrate that Gbx2 acts cell-autonomously to control the initial trajectory of TCAs.

The guidance errors of TCAs found in mosaic mutants suggest that loss of Gbx2 alters the intrinsic responsiveness of TCAs to guidance cues. Slit1 and Slit2, which encode secreted guidance proteins, are known to regulate TCA entry and navigation into the ventral telencephalon (Bagri et al., 2002). Explant assays have shown that Slit2 inhibits the outgrowth of thalamic axons and acts as a repellant to thalamic axons (Bonnin et al., 2007; López-Bendito et al., 2007; Braisted et al., 2009). To investigate whether loss of Gbx2 alters the responsiveness of TCAs to Slit2, we performed explant co-culture experiments. As expected, control (Gbx2^creER/+) thalamic axons displayed avoidance to Slit2-expressing cells (Fig. 3K,M). However, Gbx2^creER/– thalamic neurons mostly failed to respond to Slit2 (Fig. 3L,M). Altogether, our results demonstrate that Gbx2 regulates the intrinsic responsiveness of TCAs to guidance cues.

The guidance defects at the junction between the diencephalon and the telencephalon in Gbx2-CKO mutants share similarities to those described in mutants lacking Slit or Robo genes (Bagri et al., 2002; López-Bendito et al., 2007). Furthermore, the failure to respond to Slit2 suggests that loss of Gbx2 may alter Robo expression in the thalamus. As described previously (Bagri et al., 2002; López-Bendito et al., 2007), although both Robo1 and Robo2 are expressed in the medial region of the thalamus, only Robo2 transcripts are present in the lateral wall of the thalamus (Fig. 4A-B′). In the absence of Gbx2, Robo1 transcripts and proteins were abnormally detected in the lateral area of the thalamus, as well as in the misrouted thalamic axons in the epithalamus (Fig. 4C,C′). Robo2 transcripts and proteins were greatly reduced in the thalamus without Gbx2 (Fig. 4D,D′). In control embryos at E15.5, most of the GFP⁺ TCAs displayed only low levels of Robo1 (Fig. 4E). By contrast, the expression of Robo1 proteins was noticeably enhanced in TCAs within the cortex of Gbx2-CKO embryos (Fig. 4F). Significantly, enhanced Robo1 expression was also detected in the TCAs that precociously invaded the cortical plate in Gbx2-CKO embryos (Fig. 4F), demonstrating a cell-autonomous requirement for Gbx2 in repressing Robo1 expression. Therefore, Gbx2 is essential for establishing the distinct expression patterns of Robo1 and Robo2 in the developing thalamus.

As it has been shown that Robo1 and Robo2 are direct transcriptional targets of Lhx2 (Marcos-Mondejar et al., 2012), we studied the interaction between Lhx2 and Gbx2 in regulating Robo1 and Robo2. We found that the expression of Lhx2 was unaltered in the thalamus lacking Gbx2 at E12.5 (data not shown), demonstrating that Gbx2 does not regulate Lhx2 at the mRNA level. The function of LIM-HD transcription factors can be negatively regulated by LIM-domain-only (LMO) proteins, which compete for LIM-HD obligate co-factor, LIM domain-binding protein (Ldb) (Bach, 2000). Significantly, loss of Gbx2 led to upregulation of Lmo3, one of the four members of mammalian LMO genes (Tse et al., 2004). Lmo3 is normally expressed in the epithalamus and pretectum, and was ectopically expressed in the thalamus of Gbx2^creER/– embryos at E12.5 (Fig. 5A,B). In E12.5 Gbx2^creER/F; R26R^lacZ/+ embryos that received tamoxifen at E10.5, ectopic Lmo3-expressing cells displayed a ‘salt-and-pepper’ pattern of expression in the thalamus (Fig. 5C). This pattern of ectopic expression is consistent with a cell-autonomous function of Gbx2 in repressing Lmo3 in mosaic mutant thalamus.

To investigate whether Gbx2 may directly regulate Lmo3 transcription, we searched for putative Gbx2 binding sequences in the evolutionarily conserved genomic DNA sequence upstream of the Lmo3 promoter using rVISTA (Loots et al., 2002; Berger et al., 2008). Two Gbx2-binding sites were identified within a highly conserved 510 bp DNA sequence located 1.2 kb upstream of the Lmo3 transcription start site (Fig. 5E). Electrophoretic mobility shift assay (EMSA) showed that Gbx2 specifically bound to these two putative binding sites (Fig. 5F; data not shown). Therefore, Gbx2 is required to inhibit Lmo3 transcription in the thalamus probably by binding to the Lmo3 promoter.

Interestingly, Gbx2 and Lhx2 proteins have almost identical DNA-binding sequences (Berger et al., 2008), and similar temporospatial expression patterns in the thalamus (Nakagawa and O’Leary, 2001; Lakhina et al., 2007). These observations raise the interesting possibility that Gbx2 and Lhx2 might have opposing functions in regulating the transcription of Lmo3 by competing for binding to the same sequence in the Lmo3 promoter. Using a luciferase reporter assay in P19 cells, we showed that Lhx2 robustly activated the Lmo3 promoter, while Gbx2 efficiently inhibited the activity of Lhx2 (Fig. 5G,H). The transactivation was specific to Lhx2 and Lhx5, whereas other homeodomain transcription factors that were tested had no activity in the same assay (Fig. 5G). Moreover, mutating any of the two putative Gbx2/Lhx2-binding sites completely abolished the activation by Lhx2 (Fig. 5G). This is in agreement with the report that LIM-HD transcriptional complexes bind to DNA with two ‘half-binding sites’ (Thaler et al., 2002; Lee et al., 2008). Finally, we showed that, in the presence of Lmo3 protein, Lhx2 failed to activate the transcription of the Lmo3 promoter, demonstrating that Lmo3 can inhibit the transcriptional activity of Lhx2 (Fig. 5G).

To determine whether Lhx2 is indeed responsible for inducing Lmo3 in the Gbx2-deficient thalamus, we generated Gbx2^creER/creER; Lhx2^–/– double mutants. In support of our model, ectopic expression of Lmo3 in the thalamus was greatly reduced in Gbx2^creER/creER; Lhx2^–/– embryos at E12.5 (Fig. 5D). Furthermore, in both Lhx2^–/– and Gbx2^creER/creER; Lhx2^–/– embryos, Lmo3 expression was mostly lost in the ventral telencephalon (Fig. 5D; data not shown), demonstrating a requirement of Lhx2 to induce Lmo3 in this region. Therefore, our results show that Gbx2 and Lhx2 have opposite (negative and positive, respectively) functions in regulating Lmo3 transcription, and Lmo3 acts as a feedback inhibitor of Lhx2 in the developing thalamus (Fig. 5I).

Based on the above model (Fig. 5I), we made the following two predictions: (1) inactivation of Gbx2 would lead to functional disruption of Lhx2 through ectopic expression of Lmo3; (2) removal of Lmo3 would restore defects caused by the disrupted Lhx2 function in Gbx2 mutants. To test the first prediction, we examined whether loss of Gbx2 or Lhx2 leads to similar changes in Robo1/2 expression. The previous study has only examined Robo1/2 expression in the Lhx2-deficient thalamus at E14.5 (Marcos-Mondejar et al., 2012). Therefore, we compared the initial expression of Robo1/2 in Gbx2 and Lhx2 mutant thalamus at E12.5. Similar to that found at later stages, ectopic expression of Robo1 was found in the lateral-most region of the thalamus lacking either Gbx2 or Lhx2 (Fig. 6A-C). Interestingly, the level of Robo2 expression was reduced in the thalamus in Lhx2^–/– embryos, similar to that found in Gbx2^creER/– embryos, at E12.5 (Fig. 6D-F). These findings are consistent with our model that Gbx2 is essential for Lhx2 function in the regulation of Robo1 and Robo2 expression in the thalamus.

We next examined whether removal of Lmo3 could restore Lhx2 function as well as the expression of Robo1/2 in Gbx2 mutants. We used an Lmo3-null allele, Lmo3^lacZ (Tse et al., 2004), to generate Gbx2 and Lmo3 double mutants. In agreement with our prediction, the expression of Robo2 was mostly restored in the thalamus of Gbx2^creER/creER; Lmo3^lacZ/lacZ embryos at E13.5 (Fig. 6I,J). Unexpectedly, both Gbx2^creER/– and Gbx2^creER/creER; Lmo3^lacZ/lacZ embryos displayed similar ectopic expression of Robo1 in the lateral wall of the thalamus at E13.5 (Fig. 6G,H). Our results demonstrate that the ectopic expression of Lmo3 in Gbx2 mutants contributes to the loss of Robo2, but not to the ectopic expression of Robo1.

The failure to restore Robo1 expression in Gbx2^creER/creER; Lmo3^lacZ/lacZ embryos suggests that other factors in addition to Lmo3 and Lhx2 are involved in mediating Gbx2 function to regulate Robo1 in the thalamus. It has been reported that forced expression of Lhx2 leads to downregulated expression of Lhx9, as well as Robo1, in the thalamus (Marcos-Mondejar et al., 2012). This prompted us to investigate whether Lhx9 may act downstream of Gbx2 or Lhx2 in regulating Robo1 expression. We therefore first examined whether expression of Lhx9 is altered due to loss of either Lhx2 or Gbx2. As described previously (Nakagawa and O’Leary, 2001), Lhx9 is broadly expressed but its transcripts are clearly absent in the lateral-most area of the thalamus at E12.5 (Fig. 7A). Without Lhx2 or Gbx2, the expression domain of Lhx9 was expanded to the lateral wall of the thalamus at E12.5 (Fig. 7B,C). Immunohistochemistry revealed the same abnormal expression of Lhx9 in the thalamus lacking Lhx2 or Gbx2 (data not shown). Therefore, both Lhx2 and Gbx2 are required to inhibit Lhx9 expression in the lateral-most region of the thalamus.

To investigate the potential regulation of Robo1 by Lhx9 in the developing thalamus, we performed in situ hybridization to compare the expression of Robo1 and Lhx9 on adjacent sections of control embryos at E13.5. Compared with the expression domain of Robo1, Lhx9 transcripts were detected in cells closer to the ventricular surface (Fig. 7E-J), suggesting that Lhx9 is induced before the onset of Robo1 transcription. Robo1 and Lhx9 transcripts were mostly absent from the lateral-most area. Importantly, although the negative expression domains changed in size at different rostral-caudal levels of the thalamus, these two genes displayed an almost identical negative domain at any given position (Fig. 7E-J). In the absence of Gbx2, the expression domains of Robo1 and Lhx9 were similarly expanded to the lateral-most area of the thalamus (Fig. 7E′-J′). Finally, we found that removing Lmo3 had no effect on the ectopic expression of Robo1 and Lhx9 in Gbx2-deficient thalamus (Fig. 7E″-J″). Therefore, Lhx2 and Gbx2 both negatively regulate Lhx9. Our expression analyses suggest that Lhx9 may be responsible for the induction of Robo1, and the deregulation of Lhx9 may contribute to the ectopic expression of Robo1 in the thalamus lacking Gbx2.

In this study, by taking advantage of the EGFP expression that specifically labels TCAs in mice carrying the Gbx2^creER allele, we have uncovered a hitherto unknown TCA phenotype of Gbx2 mutants. We demonstrate that Gbx2 regulates the intrinsic responsiveness of TCAs, as least in part, by regulating the expression of Robo1 and Robo2 receptors. We identified an opposing activity between Gbx2 and Lhx2 in their direct control of Lmo3 transcription. We showed that Lmo3 acts as a feedback inhibitor of Lhx2. Using genetic and molecular biological studies, we have identified the mechanisms by which Gbx2 regulates the LIM code in controlling Robo1 and Robo2 expression in thalamic neurons.

Although it is known that Gbx2 is essential for the formation of TCAs, the mechanism underlying the Gbx2 function has not been determined (Miyashita-Lin et al., 1999; Hevner et al., 2002). We show here that a significant proportion of thalamic efferents are misrouted caudally and dorsally in Gbx2^creER/– embryos (Fig. 1; supplementary material Fig. S1). Furthermore, conditionally deleting Gbx2 between E9.5 and E14.5 caused similar aberrant trajectories of thalamic efferents (Fig. 2B; Fig. 3B,D; data not shown). Therefore, Gbx2 plays a crucial role in establishing the initial axonal trajectory of the newly generated thalamic neurons at different embryonic stages.

Genetic studies have identified numerous molecules that are essential for TCA development (López-Bendito and Molnar, 2003; Seibt et al., 2003; Powell et al., 2008; Dwyer et al., 2011). However, mutant mice lacking any of these molecules apparently do not display the trajectory defect found in Gbx2 mutants, indicating that Gbx2 controls an unknown pathway in regulating the initial trajectory of TCAs. In the absence of Gbx2, many thalamic efferents are fasciculated with the FR tract and project to the ventral midbrain (Fig. 1B,D; Fig. 3B). The projection pattern of Gbx2-deficient thalamic neurons is remarkably similar to that of habenular neurons (Quina et al., 2009). The habenula and thalamus are both derived from the alar plate of a single diencephalic segment, called prosomere 2 (Rubenstein et al., 1994). We have previously shown that Gbx2 is exclusively expressed in the thalamus, primarily in both neural precursors that are about to exit cell cycle and postmitotic neurons (Chen et al., 2009). Therefore, Gbx2 may be important for assigning distinct characteristics, such as TCA connectivity, to postmitotic neurons arising from the alar plate of prosomere 2.

By using chimera and mosaic analyses, we demonstrate that Gbx2 acts cell-autonomously to control the extension of TCAs into the prethalamus. Furthermore, deleting Gbx2 after the TCAs have passed the prethalamus led to TCA guidance defects and accelerated invasion into the cortical plate (Fig. 2J). As conditional deletion using creER creates genetic mosaicism (supplementary material Fig. S2) (Chen et al., 2009; Sunmonu et al., 2011), the guidance defects in Gbx2-CKO mutants thus reflect a cell-autonomous function of Gbx2. These in vivo studies, together with the explant assays (Fig. 3K-M), clearly demonstrate that Gbx2 regulates the outgrowth and pathfinding of TCAs by modifying their responsiveness to guidance cues en route to the neocortex.

To control TCA guidance, Gbx2 probably regulates guidance receptors and intracellular signaling in thalamic neurons and TCAs. In the current study, we have mainly focused on Robo1 and Robo2 because of their important roles in TCA development. The guidance defects of TCAs outside the diencephalon in Gbx2-CKO mutants share similarities with those reported in mouse embryos lacking Slit1 and Slit2 or Robo1 and Robo2 (Bagri et al., 2002; Andrews et al., 2006; López-Bendito et al., 2007), as well as in Lhx2-CKO mutants that exhibit ectopic expression of Robo1 (Marcos-Mondejar et al., 2012). Therefore, altered expression of Robo1 and Robo2 probably accounts for the guidance defects found in the telencephalon in Gbx2-CKO mutants. Furthermore, studying the regulation of Robo1/2 has allowed us to define an intricate genetic network regulated by Gbx2 in the developing thalamus. Significantly, it has recently been shown that Slit/Robo interactions play an important role in the topographic sorting of TCAs in the ventral telencephalon (Bielle et al., 2011a). It has been demonstrated that differential expression of Robo1 and Robo2 (robo and lea – FlyBase), which interact with Slit signaling from the midline, determines the lateral position of post-crossing commissural and longitudinal axons in Drosophila (Rajagopalan et al., 2000; Simpson et al., 2000). Similarly, the differential expression of Robo1 and Robo2 in different thalamic nuclei may form a Robo code to interpret Slit signals in establishing the TCA topography in the ventral telencephalon. We have previously demonstrated that different thalamic nuclei display distinct and dynamic expression of Gbx2 (Chen et al., 2009). Therefore, regulating Robo1 and Robo2 receptor levels through Gbx2 activity in TCAs derived from different thalamic nuclei may provide an important mechanism with which to establish the topographic projections of TCAs.

In Drosophila wing disc, the Lhx2 homolog Apterous induces expression of LMO protein, which in turn represses the activity of Apterous (Milán et al., 1998). In the current study, we have identified a similar feedback mechanism between Lhx2 and Lmo3 in the developing thalamus. Furthermore, we demonstrate that Gbx2 blocks Lmo3 transcription by competing for the binding of Lhx2 to the Lmo3 promoter. Therefore, Gbx2 exerts its regulatory role in the developing thalamus in part by regulating an evolutionarily conserved Apterous-Lmo feedback loop. We show that the positive regulation of Lhx2 activity by Gbx2 is essential for Robo2 expression in the thalamus (Fig. 7K). In contrast to Lhx2, Lhx9 is negatively regulated by Gbx2 at the transcriptional level. Although definitive evidence awaits further study, our data suggest that Lhx9 may mediate Gbx2 function in regulating the expression of Robo1 (Fig. 7K). We found that the expression patterns of Lhx9 and Robo1 in the thalamus almost completely mirrored each other, not only in wild type but also in Gbx2 mutant embryos (Fig. 7E-J′). Furthermore, removing Lmo3 had no effects on the ectopic expression of either Lhx9 or Robo1 in the Gbx2-deficient thalamus (Fig. 7E″-J″). Altogether, these observations suggest that Lhx9 positively regulates Robo1 expression, and that the mis-regulation of Lhx9 in Gbx2-deficient thalamus may be responsible for the ectopic expression of Robo1 in the same region.

It has been reported that Lhx2 negatively regulates Robo1 and Robo2, probably by direct transcriptional regulation (Marcos-Mondejar et al., 2012). As Lhx2 negatively regulates the expression of Lhx9 (Fig. 7B), the mis-regulation of Lhx9 may contribute to the altered expression of Robo1 following the manipulations of Lhx2 in the previous study. In contrast to our current findings, it was shown that gain or loss of function of Lhx2 resulted in the reduction or increase, respectively, in the level of Robo2 expression. Differences in experimental approaches may contribute to this apparent discrepancy. For example, the previous study examined Robo2 expression only at E14.5, while we have focused on the initial expression of Robo2. Importantly, in the previous study, deletion of Lhx2 was achieved by tamoxifen-induced deletion using the Gbx2^creER knock-in mouse line after E10.5, whereas gain of function was performed by in utero electroporation at E13.5. As the expression of Lhx2 and Gbx2 are initiated in the thalamus around the same time at E10.5, those manipulations will only change Lhx2 expression after its initially normal expression. Furthermore, those manipulations will alter Lhx2 expression in thalamic neurons in a mosaic manner. In the current study, we examined Robo2 transcripts in the thalamus in global Lhx2-KO mutants.

In summary, our results demonstrate that Gbx2 regulates the expression of Robo2 in the thalamus by repressing Lmo3 and consequently promoting Lhx2 protein function. Meanwhile, Gbx2 is essential to repress Robo1 in the lateral-most area of the thalamus, probably by inhibiting Lhx9 transcription in this region.

Here, we have identified at least three different mechanisms through which Gbx2 modifies the LIM transcriptional code: (1) Gbx2 regulates Lhx2 function by repressing its inhibitor Lmo3; (2) Gbx2 competes for binding of Lhx2 to the same DNA sequences of the Lmo3 promoter; (3) Gbx2 negatively regulates the transcription of Lhx9. Although Gbx2 and Lhx2 antagonize each other when regulating Lmo3 expression, they both promote Robo2 expression. Therefore, the interaction between Gbx2 and Lhx2 must be highly context dependent. It is conceivable that there are many other Lmo3-like or Robo2-like transcriptional targets that are regulated by both Gbx2 and Lhx2 in the developing thalamus. The relative stoichiometries of Gbx2 and Lhx2 are thus important in conferring distinct identity and connectivity of thalamic neurons. During embryogenesis, the thalamus is gradually parcellated into multiple nuclei. Each nucleus has distinct cytoarchitecture and function, and projects to a specific set of cortical areas. Gbx2, Lhx2 and Lhx9 are expressed in discrete and overlapping domains of the thalamus, and different thalamic nuclei display distinctive expression patterns of Gbx2 (Nakagawa and O’Leary, 2001; Chen et al., 2009). Gbx2 and LIM-HD transcription factors may thus constitute the core of the regulatory network that define the diverse identity and connectivity of thalamic nuclei.

We thank Dr Wei Lun Leung for critical reading and comments of the manuscript. We thank Drs Shubha Tole and Edwin Monuki for providing Lhx2 mutant embryos and mice; Dr Atsushi Tamada for providing the antibodies against Robo1 and Robo2; and Dr Lynda Erskine for providing SLIT2 expressing vector. The monoclonal anti-NEFM antibody (2H3) was developed by T. M. Jessell and J. Dodd, and obtained through the Developmental Studies Hybridoma Bank under the auspices of the NICHD and maintained by The University of Iowa (Iowa City, IA). We thank Dr Mark Kronenberg for providing reagents and advice on EMSA and luciferase reporter assays.