The role of the diencephalon in the guidance of thalamocortical axons in mice

A striking feature of the axonal tracts that interlink the component parts of the nervous system is the high degree of order with which they map the array of neurons in one structure onto the array of neurons in their target. Often, the order of axons at the target closely mirrors that at the source. An excellent example is the mapping of thalamic neurons onto their cerebral cortical targets via the thalamocortical pathway (Fig. 1A). Thalamic neurons located at one end of the thalamus in a dorsolateral region called the dorsal lateral geniculate nucleus (dLGN) innervate the caudal (visual) part of cortex; neurons located at the other end of the thalamus in more rostral-medial regions known as the ventrolateral (VL) and ventromedial (VM) nuclei innervate more rostral cortical regions, including motor and frontal cortex; neurons located in between – in the ventromedial posterior (VMP) nuclei – innervate central (somatosensory) cortex (Fig. 1A) (Amassian and Weiner, 1966; Bosch-Bouju et al., 2013; Jones, 2007; Tlamsa and Brumberg, 2010). The mechanisms that generate this orderly topographic mapping remain poorly understood.

The maintenance of order among thalamic axons as they grow is likely to contribute to the generation of orderly topographic mapping in the mature thalamocortical tract. During embryogenesis, thalamic axons exit the thalamus from about E12.5 onwards (Auladell et al., 2000; Braisted et al., 1999; Tuttle et al., 1999), approximately coincident with the cessation of neurogenesis in this structure (Angevine, 1970; Li et al., 2018). They then cross the adjacent prethalamus and turn laterally out of the diencephalon and into the ventral telencephalon where they traverse two consecutive instructive regions – the corridor (López-Bendito et al., 2006) and the striatum – before entering the cortex. There is evidence that the maintenance of spatial order among thalamocortical axons (TCAs) crossing the ventral telencephalon requires interactions between the axons and signals released by cells they encounter in this region (Bielle et al., 2011; Bonnin et al., 2007; Braisted et al., 1999; Dufour et al., 2003; Molnár et al., 2012; Powell et al., 2008). The importance of earlier interactions within the diencephalon remains unclear.

Here, we have tested the effects of mutating the gene for the Pax6 transcription factor, which is essential for normal diencephalic patterning (Caballero et al., 2014; Clegg et al., 2015; Parish et al., 2016; Pratt et al., 2000; Stoykova et al., 1996; Warren and Price, 1997), on the topographic mapping of TCAs onto the cortex. Pax6 starts to be expressed in the anterior neural plate well before TCAs start to form (Walther and Gruss, 1991). As the forebrain develops from the anterior neural plate, Pax6 expression becomes localized in: (1) cortical progenitors that generate the target neurons for TCAs; (2) diencephalic (thalamic and prethalamic) progenitors; and (3) prethalamic (but not thalamic) postmitotic neurons (Quintana-Urzainqui et al., 2018; Stoykova et al., 1996; Warren and Price, 1997). We discovered that deletion of Pax6 from mouse embryos at the time when thalamic axons are starting to grow results in the development of a thalamocortical projection containing mapping errors. Axons from dorsolateral thalamus are misrouted medially and end up projecting abnormally rostrally in the cortex. We went on to explore the reasons for this defect.

We first used conditional mutagenesis to test whether misrouting is due to the loss of Pax6 from prethalamic neurons, as previous work has shown that: (1) Pax6 is not required in the cortex for normal TCA topography (Piñon et al., 2008); and (2) Pax6 is neither expressed nor required autonomously by thalamic neurons for them to acquire the ability to extend axons to the cortex (Clegg et al., 2015). We found that although loss of Pax6 from a specific set of prethalamic neurons prevented them developing their normal axonal projections to thalamus and resulted in the abnormal fasciculation of thalamic axons, it did not cause TCAs to misroute. This suggested that the thalamus itself contains important navigational cues for TCAs. We used slice culture transplants and gene expression studies to show: (1) that the thalamic environment is indeed instructive for TCA navigation; and (2) to identify molecular changes within the thalamus that likely cause the disruption in TCA topography observed upon Pax6 deletion. Our findings indicate that the normal topographic mapping of TCAs requires that order be established and maintained from the earliest stages of their growth by molecular cues in the thalamus itself.

Previous studies have shown that constitutive loss of Pax6 function causes a total failure of TCA development, which is hypothesized to be a secondary consequence of anatomical disruption at the interface between the diencephalon and the telencephalon (Clegg et al., 2015; Georgala et al., 2011; Jones et al., 2002). No such failure occurs if Pax6 is deleted conditionally after this anatomical link is formed (Clegg et al., 2015). We first assessed whether delayed ubiquitous Pax6 deletion, induced in CAG^CreER-TM Pax6^fl/fl embryos (referred to here as CAG^CreER Pax6 cKOs), disrupts the topography of TCA connections. We induced Cre recombinase activation by tamoxifen administration at E9.5, which caused Pax6 protein loss in CAG^CreER Pax6 cKOs from E11.5 onwards (Quintana-Urzainqui et al., 2018), which is when the generation of most thalamic neurons is starting (Li et al., 2018) and before many TCAs have begun to grow (Auladell et al., 2000; López-Bendito and Molnár, 2003). Diencephalic progenitor domains in CAG^CreER Pax6 cKOs are fully recognizable (Quintana-Urzainqui et al., 2018) and patterning seems largely unaffected in the thalamus (Fig. S1). We used both wild-type and CAG^CreERPax6^fl/+ littermate embryos as controls because the latter express normal levels of Pax6 protein (see Materials and Methods; Caballero et al., 2014; Manuel et al., 2015).

We inserted two different axonal tracers in two cortical areas in E15.5 fixed brains. DiA was placed in the visual (caudal) cortex, while DiI was placed in the somatosensory (more rostral) cortex (Fig. 1B). In controls (both wild type and Pax6^fl/+), DiA retrogradely labelled cells in dorsolateral thalamic areas (dLGN; green labelling in Fig. 1C-E), while DiI labelled cells in ventromedially located thalamic regions, identified as ventral-posterior thalamic nucleus (VP) (red labelling in Fig. 1C-E). Labelling of these two thalamic regions was clearly separated in all cases (indicated by dotted line in Fig. 1C-E). In CAG^CreER Pax6 cKOs, however, the two labelled populations overlapped (Fig. 1F-H). In these mutants, the distribution of the DiA-labelled thalamic cells (from caudal cortical injections) was not obviously changed with respect to controls. However, the DiI-labelled thalamic cells (projecting to more rostral cortical areas) showed a much wider distribution than in controls and expanded to lateral thalamic areas (compare Fig. 1C-E,C′-E′ with F-H,F′-H′), even overlapping with DiA stained cells at the dLGN (Fig. 1G,H,G′,H′). To quantitate this, we measured the area occupied by DiI and DiA within each nucleus of interest in controls versus CAG^CreER Pax6 cKOs (in transverse E15.5 sections from three different litters: four controls, three cKOs; five sections per embryo). We defined the dLGN and VP nucleus as regions of interest (ROIs) blind to DiI/DiA labelling and measured the percentage of each ROI occupied by DiI and DiA (for details, see Materials and Methods). We found highly significant increase (from virtually 0% to 7.7%) of the area occupied by DiI in the dLGN in mutants compared with controls (Fig. 1I). We also found a significant increase in DiA in VP nucleus, although its magnitude was very small (0.26%) (Fig. 1I). Areas occupied by DiI in VP or DiA in dLGN were not significantly altered. This analysis shows that CAG^CreER Pax6 cKOs display topographic errors, the main one being the misrouting of axons from dLGN neurons towards abnormally rostral cortical areas in the CAG^CreER Pax6 cKOs.

As Pax6 is expressed both in the cortex and diencephalon during TCA development, the mapping defects described above might have been due to the loss of Pax6 from the cortex. This was unlikely because a previous study showed that Pax6 is not required in the cortex for the establishment of proper topographical thalamocortical connections (Piñon et al., 2008). To confirm this, we used a cortex-specific, tamoxifen-inducible Cre line (Emx1^CreER). We administered tamoxifen at E9.5, which results in a near-complete loss of cortical Pax6 between E11.5 and E12.5 (Georgala et al., 2011; Mi et al., 2013), and performed DiI/DiA labelling at E15.5, following the same experimental design described above for the CAG^Cre line. We found that the two retrogradely labelled populations did not overlap in controls (Emx1^CreER Pax6^fl/+) or in mutants (Emx1^CreER Pax6^fl/fl) (Fig. S2A,B), suggesting that the defects of TCA mapping found in the CAG^CreER Pax6 cKOs were not attributable to cortical abnormalities.

To define the anatomical region where thalamic axons probably deviated from ordered growth, we examined the TCA bundle in CAG^CreERPax6 cKOs. This bundle was ordered and segregated into rostral/somatosensory (DiI) and caudal/visual (DiA) halves at its point of exit from the prethalamus and entry into the ventral telencephalon (arrows in Fig. 1F-H), which indicated that the misrouting of lateral TCAs from dorsolateral thalamus might happen before this point, i.e. within the diencephalon.

Within the diencephalon, the first structure that thalamic axons encounter as they leave the thalamus is the prethalamus, and its neurons normally express high levels of Pax6. Therefore, we investigated whether the defects of TCA mapping in CAG^CreER Pax6 cKOs might arise from a disordered growth of thalamic axons through the prethalamus. As a first step, we examined the effects of Pax6 deletion on the behaviour of thalamic axons as they cross the prethalamus.

As the neural cell adhesion molecule L1CAM (L1) is expressed in TCAs (Fukuda et al., 1997; Ohyama et al., 2004), we examined the distribution of L1-positive thalamic axons at E13.5 in transverse and sagittal sections through the prethalamus. In controls, axons emerging from the thalamus converge progressively as they cross the prethalamus (Fig. 2A-E) to subsequently form a single thalamocortical bundle that turns laterally and exits the prethalamus (arrows in Fig. 2A,B,E). We found that, in CAG^CreER Pax6 cKOs (Fig. 2F-L), thalamic axons prematurely converge into larger bundles as soon as they cross the thalamic-prethalamic boundary (open arrows in Fig. 2G-I,L).

To obtain a quantitative measurement of this observation, we positioned three equally spaced lines across different diencephalic levels in sagittal sections (see Materials and Methods; lines represented in Fig. 2D,E,K,L). We used Fiji software (Schindelin et al., 2012) to quantify the number of axon bundles crossing each line and the width of each individual bundle. We recognized individual bundles as each single L1-positive structure above a consistent intensity threshold (red lines in Fig. 2M,N). We found a significant decrease in the number of bundles crossing all three checkpoints (Fig. 2O). Axon bundle width strikingly increased at the Th-PTh border and the mid-PTh, with no significant change at the low-PTh checkpoint line (Fig. 2P) (see figure legend for statistical details). These data indicate that, in the absence of Pax6, TCAs begin to fasciculate prematurely in their route, forming bigger and fewer bundles as they cross the prethalamus (Fig. 2Q). We next tested the role of the prethalamus in TCA formation and the potential establishment of their topography.

The prethalamus has been proposed to host a population of neurons that extend axons to the thalamus and act as ‘pioneer guides’ for TCA navigation (Price et al., 2012; Tuttle et al., 1999). We assessed whether this population is disrupted by Pax6 loss from the prethalamus because, if it is, this might provide an explanation for phenotypes described above.

From E9.5 onwards, most cells in the prethalamus express, or are derived from, cells that express Gsx2. We used a Gsx2^Cre line (Kessaris et al., 2006) carrying an EGFP Cre reporter (Sousa et al., 2009) to visualize neurons and axons belonging to the Gsx2 lineage and we observed that prethalamic Pax6-expressing cells are included within the location of the Gsx2 lineage prethalamic population (Fig. 3A,B). Zic4 is also expressed by some prethalamic cells, with an onset of expression similar to that of Gsx2 (about E9.5; Gaston-massuet et al., 2005), and most diencephalic Zic4 lineage cells express and require Pax6 for their normal development (Li et al., 2018). Using a Zic4^Cre line (Rubin et al., 2011), we observed that prethalamic neurons derived from Zic4 lineage were located in a narrow band close to the thalamic-prethalamic border (Fig. 3C,D). We assessed whether these prethalamic populations normally send axons to the thalamus.

Gsx2-lineage GFP-positive axons extended throughout the thalamus forming ordered and parallel projections (Fig. 3E) from E12.5 onwards (Fig. S3). By contrast, Zic4-lineage prethalamic cells did not project axons to the thalamus (Fig. 3D), indicating that prethalamic pioneer axons arise from Gsx2-lineage and not from Zic4-lineage prethalamic cells.

As Gsx2 is also expressed in the ventral telencephalon (Fig. 3A), and ventral telencephalic neurons are known to project to the thalamus (López-Bendito and Molnár, 2003; Métin and Godement, 1996; Molnár et al., 1998), there was a possibility that Gsx2^Cre lineage axons innervating the thalamus actually originated from ventral telencephalic neurons. To confirm the existence of Gsx2-lineage prethalamic neurons projecting to the thalamus, we injected the neuronal tracer neurobiotin into the thalamus of E13.5 Gsx2^Cre embryos and successfully labelled prethalamic neurons (arrow in Fig. 3F). Neurobiotin-positive cells were GFP-expressing Gsx2-lineage cells (Fig. 3F-F″,G) and most of them also expressed Pax6 (arrows in Fig. 3H,H′; see summary in Fig. 3I). (Note that individual injections each involved only subregions of the thalamus, explaining why each one only labelled a discrete subset of the prethalamic neurons projecting to the thalamus.) This experiment confirmed that Gsx2-lineage cells in the prethalamus project to the thalamus.

Having established that pioneer prethalamic axons belong to the Gsx2 lineage and express Pax6 we next aimed at disrupting their formation by conditionally deleting Pax6 in Gsx2-lineage cells. We crossed mice carrying the floxed Pax6 allele and EGFP Cre reporter with the Gsx2^Cre line. Pax6 conditional deletion in Gsx2-lineage cells (Gsx2^Cre Pax6 cKOs) caused a visible reduction in the number of GFP-positive axons projecting from prethalamus to thalamus in E12.5, E13.5 and E14.5 embryos (Fig. 4A-F). To confirm that the prethalamic axons that were lost in Gsx2^Cre; Pax6^loxP/loxP embryos were Pax6 expressing, we used the DTy54 YAC reporter allele to express tauGFP in cells in which the Pax6 gene is active, irrespective of whether it is mutant or not (Tyas et al., 2006). Whereas there were many tauGFP-labelled axons running from prethalamus to thalamus in controls, there were very few in experimental embryos (Fig. 4G-J).

DiI placed in the thalamus of E13.5 CAG^CreER controls (CAG^CreER Pax6 ^fl/+) retrogradely labelled a prethalamic population (arrow in Fig. 4K). In the absence of Pax6 (CAG^CreER Pax6 cKOs), no prethalamic cell bodies were labelled by thalamic DiI injection (Fig. 4L), providing further evidence that the prethalamic pioneer population does not form correctly when Pax6 is deleted. Overall, our results show that prethalamic pioneer axons originating from Gsx2-lineage cells both express and require Pax6 to develop normal connections with the thalamus (Fig. 4T).

We then studied the TCAs of Gsx2^Cre Pax6 cKOs. Similar to the phenotype described in CAG^CreER Pax6 cKOs, Pax6 deletion in Gsx2 lineage caused abnormal premature fasciculation of axons crossing the thalamic-prethalamic border, as evidenced by L1 immunohistochemistry (Fig. 4M-R). However, unlike CAG^CreER Pax6 cKOs, Gsx2^Cre Pax6 cKOs did not show abnormal topographical projections, with no obvious overlap between thalamic retrogradely labelled populations after cortical DiA and DiI placement in caudal and more rostral cortex, respectively (Fig. 1B; Fig. 4S). We conclude that, although prethalamic pioneer axons may play a role in avoiding premature TCA fasciculation, they are not required for the establishment of accurate thalamocortical topographic mapping (Fig. 4T).

We next considered the potential importance of thalamic factors in the establishment of thalamocortical topographic order. Our results above indicate that thalamic axons might have deviated from their normal trajectories before they exited the thalamus in CAG^CreER Pax6 cKOs (arrows indicate deviant axons in Fig. 1G′,H′; no such axons were observed in the controls, Fig. 1C′,D′,E′). A misrouting of TCAs in the thalamus was also evident with L1 staining (Fig. 5A,B). The largest collections of deviant axons were observed projecting from lateral to medial thalamic regions (arrow in Fig. 5B), suggesting that the loss of Pax6 had disrupted normal navigational mechanisms operating within the thalamus.

We looked for evidence that thalamic axons are actively guided through the normal thalamus by using in vitro slice culture transplants to assess the effects of repositioning lateral or medial thalamic neurons on the routes taken by their axons. We grafted thalamic slice explants from E13.5 GFP-positive donor embryos into GFP-negative host slices and cultured them for 72 h to allow thalamic axons to regrow and navigate through the host environment. The donor grafts were positioned so that their axons had to traverse at least 200 µm of host thalamic tissue before encountering the Th-PTh boundary, allowing us to assess how the host thalamic tissue affected the trajectory of the axons emerging from the donor tissue. We isolated donor explants from either lateral or medial thalamus, and grafted them either medially or laterally into host thalamus (Fig. 5C,E,G,I).

We found that axons from lateral thalamic explants showed a strong preference to follow a lateral trajectory, although surprisingly they did not avoid the vLG, as observed in lateral TCAs in vivo. Axons from medial explants extended much broadly throughout the thalamus. In both cases, this effect was irrespective of whether they were grafted laterally or medially (Fig. 5C-J). To quantitate this, we divided the thalamus in three equal sectors (lateral, intermediate and medial; Fig. 5K) and measured the area occupied by GFP-positive elements in each of the sectors. We quantified six explants (three lateral and three medial; five sections per explant) from three different litters. On average, 95.7% of axons from lateral explants navigated through lateral-most area (Fig. 5K), while medial explants extended axons more evenly across the three regions (45.8% lateral, 31.8% intermediate and 22.4% medial; Fig. 5K). Moreover, we observed that when lateral axons were confronted with medial host thalamus, most made a sharp turn towards lateral positions before heading towards the prethalamus (Fig. 5E,F). When medial explants were grafted laterally, many of their axons turned medially (arrows in Fig. 5J).

These results indicated that different subsets of thalamic axons exhibit different chemotactic responses to the thalamic environment and therefore that thalamic axons are actively guided by mechanisms operating within the thalamus itself. To gain further insight into what these mechanisms might be, we went on to examine the expression of guidance molecules in the normal thalamus and in thalamus from which Pax6 has been removed.

Semaphorin 3a (Sema3a) and netrin 1 (Ntn1) are secreted guidance molecules whose complementary gradients of expression in the ventral telencephalon are key for the correct establishment of topographical connections between thalamus and cortex (Bielle et al., 2011; Braisted et al., 1999; Molnár et al., 2012; Powell et al., 2008; Wright et al., 2007). Interestingly, we found that their transcripts are also distributed in opposing gradients in the thalamus (Fig. S4A,B). Ntn1 is most highly expressed at rostral-medial levels (Fig. S4A), while Sema3a is most highly expressed in a more caudal-lateral aspect of the thalamus and in the lateral prethalamus (Fig. S4B). In transverse in situ hybridization of E13.5 controls, we observed that Ntn1 is expressed in a narrow rostral-medial thalamic population of neurons (arrow in Fig. 6A), while Sema3a is expressed in caudal-lateral thalamic neurons (arrows in Fig. 6B) as well as flanking the TCA bundles in the prethalamus (open arrow in Fig. 6B).

To obtain a clearer three-dimensional view of these expression patterns, we reconstructed them from serial, adjacent sections stained for Sema3a, Ntn1, Pax6 and L1 in controls (Fig. 6C, see Materials and Methods). The 3D reconstruction confirmed that Sema3a and Ntn1 form opposing gradients in the normal embryonic thalamus (Fig. 6C), with Sema3a highest at caudal-lateral thalamic levels, while Ntn1 is highest at rostral-medial thalamic levels.

We next investigated the expression patterns of the main receptors for Ntn1 and Sema3a in the thalamus of control embryos. The most interesting finding was that Unc5c, which encodes a Ntn1 receptor mediating axonal repulsion (Leonardo et al., 1997), was expressed differentially from lateral to medial across the thalamus (Fig. 6D). Laterally, almost all cells expressed high levels of Unc5c, whereas medially many cells did not (Fig. 6D′). Unc5c was largely absent from a narrow strip of cells close to and parallel with the ventricular zone. This strip coincided with the region that contained Ntn1-positive cells (compare Fig. 6D with A). Plxna1, which encodes a Sema3a receptor that mediates repulsion (Rohm et al., 2000; Takahashi et al., 1999; Tamagnone et al., 1999) was found to be distributed relatively homogenously across the thalamus (Fig. 6E, Fig. S4D).

These expression patterns suggest that, whereas all thalamic axons might be repelled by Sema3a (owing to their expression of Plxna1), only some axons might be repelled by Ntn1 (i.e. those originating laterally, which express Unc5c, and those Unc5c-expressing axons that originate medially) (Fig. 6K). This could explain why, in the grafting experiments described above, axons from lateral explants invariably navigated laterally, which would be away from medially located high levels of Ntn1. It could also explain why medial explants generated axons that were able to navigate on a broader front: some axons (those that express Unc5c) would be pushed relatively laterally by repulsion from medially expressed Ntn1; others (those that do not express Unc5c) would be able to maintain a medial trajectory through the Ntn1-expressing territory, thereby avoiding the high levels of Sema3a expressed in lateral thalamus (Fig. 6J). Other receptor-coding genes analysed (Dcc, Unc5a and Unc5d) showed little or no expression within the main body of the thalamus and are therefore unlikely to contribute to the navigation of thalamic axons within the thalamus (Fig. S4F,H,J).

We next asked whether the thalamic expression of Ntn1 and Sema3a, and their receptors change in a way that might explain the medially directed deviation of lateral axons that we observed in the thalamus of CAG^CreER Pax6 cKOs. In these embryos, we found that the medial domain of Ntn1 expression was retained and appeared enlarged. Sema3a was still expressed higher laterally, although overall levels seemed reduced (Fig. 6F,G). These patterns are reconstructed in 3D in Fig. 6H. Ntn1 and Sema3a expression in the subpallium of CAG^CreER Pax6 cKOs appeared to be unaffected (Fig. S4L-Q). The significance of changes in the expression levels of Sema3a and Ntn1 in the thalamus of CAG^CreER Pax6 cKOs was confirmed by analysing a previously published RNAseq dataset (Fig. S4C, dataset from Quintana-Urzainqui et al., 2018).

Regarding the expression of guidance receptors, fewer laterally located neurons expressed Unc5c in CAG^CreER Pax6 cKOs than in controls (compare Fig. 6I′ and D′). Significant numbers of Unc5c-negative neurons were now intermingled with Unc5c-positive neurons even in the most lateral thalamic tissue (Fig. 6I′). The thalamic expression pattern of Plxna1 did not change in the absence of Pax6 (Fig. 6J, Fig. S4E), nor did that of any of the other receptor-coding genes studied (Fig. S4F-K).

As reported above, we discovered that CAG^CreER Pax6 cKOs show a misrouting in a medial direction of axons from the lateral thalamus (Fig. 1), and our finding that many laterally located thalamic neurons lose their expression of Unc5c in these mutants, provides a likely explanation, summarized in Fig. 6K. We propose that Unc5c-negative laterally located thalamic neurons in CAG^CreER Pax6 cKO thalamus would no longer be repelled from the medial thalamus by its high levels of Ntn1. Consequently, they would be more likely to stray, or perhaps to be pushed by relatively high lateral levels of Sema3a, towards a medial direction (Fig. 6K, right side).

Finally, we grafted GFP-positive lateral thalamic explants from CAG^CreER Pax6 cKO embryos into the lateral thalamus of GFP-negative slices from littermate control embryos, following the same experimental paradigm as before (Fig. 7A, see Fig. 5, and Materials and Methods). Our model would predict that subsets of Pax6-deficient lateral thalamic axons, presumably those that have lost their repulsion to Ntn1, would deviate towards medial areas when confronted with control thalamus. We used 11 embryos from four different litters and performed a total of 22 (bilateral) graft experiments. In all cases, we observed subsets of GFP-positive axons deviating medially (Fig. 7B, red arrows). To measure the navigation pattern of these axons, we divided the thalamus in three equal medial-lateral sectors and quantified the percentage of GFP on each of them (Fig. 7C). For this analysis, we used four explants from four different litters (five slices per explant). We found that axons spread throughout the three areas of the thalamus (45.8% laterally, 31.8% intermedially and 22.44% medially; Fig. 7C). This is in striking contrast to the almost invariably (95.7%) lateral trajectories of axons from control lateral thalamic explants grafted into control thalamic tissue (Fig. 5C-F,K). This evidence supports our model and the idea that thalamic neurons hold intrinsic information that determines the medial-lateral position of their axons when crossing the thalamus. Overall, our findings indicate that mechanisms exist within the thalamus itself to ensure that its TCAs exit in an orderly manner and that these mechanisms may play an important part in the correct topographic mapping of TCAs onto the cortex.

During embryonic development, thalamic axons undertake a long journey, navigating through several tissues before they arrive at the cortex. It is therefore crucial that their guidance is tightly regulated by mechanisms placed along the route. Previous studies have demonstrated the importance of the ventral telencephalon as an intermediate target for establishing correct topographical thalamocortical connections. Here, gradients of signalling molecules sort different subsets of TCAs towards different areas of the embryonic cortex (Antón-Bolaños et al., 2018; Bielle et al., 2011; Braisted et al., 1999; Dufour et al., 2003; Métin and Godement, 1996; Molnár et al., 2012; Vanderhaeghen and Polleux, 2004). The importance of other tissues along the route in the establishment and/or maintenance of axonal topography remained unexplored. We now show that if thalamic axons do not emerge in order from the thalamus they will connect with the wrong areas of the cortex, resulting in topographical defects. This highlights the importance of maintaining axonal order throughout the route and suggests the existence of guidance mechanisms within the diencephalon to guarantee that this happens.

Our in vitro graft experiments demonstrated that embryonic thalamic tissue is instructive for TCAs and sorts axons according to their original medio-lateral position. We went on to find a possible guidance mechanism acting within the thalamus. The thalamus expresses Ntn1 and Sema3a, some of the same guidance molecules known to guide TCAs in the ventral telencephalon (Bielle et al., 2011; Powell et al., 2008; Wright et al., 2007). What is more, there is an interesting correspondence between the regions expressing each of those molecules in the thalamus and in the ventral telencephalon. TCAs that emerge and navigate through the Sema3a-high region of the thalamus (lateral-caudal thalamus, dLGN) are steered towards the Sema3a-high region in the ventral telencephalon, while axons that emerge and navigate through the Ntn1-high region of the thalamus (ventral-medial thalamus, MP) are sorted towards Ntn1-high regions in the ventral telencephalon. This suggests that each subset of thalamic axons maintains the expression of the same combination of axon guidance receptors along the route and therefore shows the same chemotactic response when confronting gradients of signalling cues. Likewise, it indicates that the same gradients of guidance molecules are re-used at different levels of the thalamocortical pathway to maintain topographic order.

The chemotactic behaviour of TCAs with respect to Sema3a and Ntn1 gradients in the thalamus and ventral telencephalon can be explained by our observations of the expression of Sema3a and Ntn1 receptors in developing thalamus. We show that all thalamic neurons seem to express homogeneous levels of Plxna1, a receptor mediating repulsion to Sema3a (Rohm et al., 2000; Takahashi et al., 1999; Tamagnone et al., 1999), while Unc5c, a receptor mediating repulsion to Ntn1 (Leonardo et al., 1997), was found to be expressed in a lateral-high medial-low gradient. According to these observations, we propose a model in which complementary expression patterns of Sema3a and Ntn1 can establish topographical order on TCAs by a mechanism of double repulsion, in which all thalamic axons have the potential to be repelled by Sema3a but only lateral axons are additionally repelled by Ntn1. It is possible that laterally derived axons experience stronger Ntn1 repulsion the more lateral they are. Thus, lateral thalamic axons prefer to navigate through Sema3a-high, Ntn1-low regions because they might be more strongly repelled by Ntn1 than by Sema3a. Axons located in intermediate regions of the thalamus express lower levels of Unc5c; thus, they might be equally repelled by Sema3a and Ntn1, and chose to navigate across regions with moderate levels of both signalling cues. Likewise, medial axons are only repelled by Sema3a and are neutral to Ntn1: therefore they chose to navigate through Sema3a-low, Ntn1-high areas.

Supporting this model are the experiments showing that TCAs are repelled by Nnt1 (Bielle et al., 2011; Bonnin et al., 2007; Powell et al., 2008) and thalamic growth cones show retraction in the presence of Sema3a (Bagnard et al., 2001). Moreover, Wright and colleagues reported that, in mice harbouring a mutation that makes the axons non-responsive to Sema3a, axons from the ventrobasal (VB) thalamic nucleus were caudally shifted and target the visual cortex instead of the somatosensory cortex (Wright et al., 2007). Our double repulsion model satisfactorily explains this phenotype. The VB nucleus is located in an intermediate thalamic region that would contain a substantial number of Unc5c-positive neurons. In those mutants, VB axons lose their repulsion to Sema3a but many would still be repelled by Ntn1, and therefore would steer towards Sema3a-high, Ntn1-low regions both in the thalamus and the ventral telencephalon. The behaviour of thalamic axons in our explant experiments using donor tissue from controls and CAG^CreER Pax6 cKOs, also support the model.

Other molecules known to form gradients and guide TCAs in the ventral telencephalon, such as Slit1 or ephrin A5 (Bielle et al., 2011; Dufour et al., 2003; Molnár et al., 2012; Seibt et al., 2003; Vanderhaeghen and Polleux, 2004), were not analysed in this study. It remains to be tested whether these molecules and their receptors are expressed in the thalamus in a gradient fashion and whether they follow the same rules proposed in our model.

It is important to highlight that, in this study, we considered only the medio-lateral axis of the main thalamic body, but the same or other guidance cues and receptors probably function in other directions. For example, work in mice showed that Unc5c (Bonnin et al., 2007) and DCC (Powell et al., 2008) are also highly expressed in the rostral thalamus, at a level we did not cover in our expression and tracer analyses.

Another important question is how Pax6 inactivation leads to deficits in thalamic organization, as Pax6 is only expressed in progenitors. We show here and in a previous paper that thalamic patterning does not seem to be largely affected in CAG^CreER Pax6 cKOs (Fig. S1 and Quintana-Urzainqui et al., 2018) and that the main changes in gene expression in postmitotic neurons are related to axon guidance molecules. One possibility is that Pax6 expression in thalamic progenitors affects transcriptional programmes and indirectly translates into actions on the postmitotic expression of specific axon guidance molecules.

Finally, our results have given interesting new insights into the development and the importance of the pioneer axons from the prethalamus to the thalamus. First, we found that the prethalamic neurons extending pioneer axons to the thalamus belong to a particular lineage: the Gsx2-lineage and not the Zic4-lineage. Second, although disturbing the prethalamus-to-thalamus pioneers did not stop TCAs reaching the cortex without any topographic error, it did cause them to fasciculate prematurely as they crossed the prethalamus. Growing axons often increase their fasciculation when they cross regions that are hostile to their growth. Ours and other studies have shown that the prethalamus also expresses guidance cues with potential to exert a repulsive response of TCAs (Ono et al., 2014); thus, it is possible that interactions between the developing TCAs and prethalamic pioneers somehow helps the passage of the TCAs across this region. Further work is needed to discover what the consequences are if this help is unavailable.

All animals (Mus musculus) were bred according to the guidelines of the UK Animals (Scientific Procedures) Act 1986 and all procedures were approved by Edinburgh University's Animal Ethics Committee. For conditional inactivation of Pax6, we used a tamoxifen-inducible Pax6^loxP allele (Simpson et al., 2009) and a RCE:LoxP EGFP Cre reporter allele (Sousa et al., 2009), and we combined them with different Cre lines. To generate a deletion of Pax6 throughout the embryo, we used lines carrying a CAGGCre-ER^TM allele (Hayashi and McMahon, 2002; Quintana-Urzainqui et al., 2018). As controls, we used both wild-type and CAG^CreERPax6^fl/+ littermate embryos as the latter express normal levels of Pax6 protein, almost certainly because of a feedback loop that compensates for a deletion in one allele by increasing the activity of the other (Caballero et al., 2014; Manuel et al., 2015).

To inactivate Pax6 in different parts of the prethalamus we used either a Gsx2-Cre (Kessaris et al., 2006) or the Zic4-Cre allele (Rubin et al., 2011). For cortex-specific deletion of Pax6, we used Emx1Cre-ER^T2 (Kessaris et al., 2006). The DTy54 YAC reporter allele (Tyas et al., 2006) was combined with the Gsx2-Cre allele to generate Gsx2Cre; Pax6^loxP/loxP embryos expressing tauGFP in cells in which the Pax6 gene is active.

Embryos heterozygous for the Pax6^loxP allele (Pax6^fl/+) were used as controls because previous studies have shown no detectable defects in the forebrain of Pax6^fl/+ embryos (Simpson et al., 2009). Embryos carrying two copies of the floxed Pax6 allele (Pax6^fl/fl) were the experimental conditional knockout (cKO) groups.

For thalamic explant experiments, we generated litters containing GFP-positive and -negative embryos by crossing a line of studs heterozygous for a constitutively active form of the CAGGCre-ER^TM allele and the RCE:LoxP EGFP Cre reporter allele with wild-type females. In the second set of explant experiments, we performed crossings to generate litters containing GFP-positive Pax6 cKO embryos (CAGCRE^{ER TM}; Pax6^fl/fl), and GFP-negative control embryos (homozygous for the Pax6 floxed allele but not expressing the CRE allele). Both embryos carried a RCE:LoxP EGFP allele reporting Cre activity and allowing us to select the embryos within the same litter and visualizing Pax6 cKO axons in the explants. Embryos used in the two sets of explant experiments had the same mixed genetic background (CD1, CBA and C67BL\6).

The day the vaginal plug was detected was considered E0.5. Pregnant mice were given 10 mg of tamoxifen (Sigma) by oral gavage on embryonic day 9.5 (E9.5) and embryos were collected on E12.5, E13.5, E14.5, E15.5, E16.5 or E18.5. For the DiI and DiA tracing experiments, wild-type embryos (CD1 background) were additionally used as controls.

Embryos were decapitated and fixed in 4% paraformaldehyde (PFA) in phosphate-buffered saline (PBS) overnight at 4°C. After washes in PBS, heads were cryoprotected by immersion in 30% sucrose in PBS, embedded in OCT compound and sectioned using a cryostat at 10 µm.

Cryosections were left to stabilize at room temperature for at least 2 h and then washed three times in PBST (1× PBS with 0.1% Triton X-100; Sigma). To block endogenous peroxidase, sections were treated with 3% H₂O₂ for 10 min. After PBS washes, antigen retrieval was performed by immersing the sections in sodium citrate buffer (10 mM, pH 6) heated at ∼90°C using a microwave for 20 min. Sections were then incubated with the rabbit polyclonal anti-Pax6 antibody (1:200, BioLegend, 901302) overnight at 4°C. The secondary antibody (goat anti-rabbit biotinylated, 1:200, Vector Laboratories, BA-1000) was incubated for 1 h at room temperature followed by a 30 min incubation with Avidin-Biotin complex (ABC kit, Vector Laboratories, PK6100). Finally, a diaminobenzidene (DAB, Vector Laboratories, SK4100) reaction was used to obtain a brown precipitate and sections were mounted in DPX media (Sigma-Aldrich, 06522).

For immunofluorescence, cryosections were incubated overnight at 4°C with the following primary antibodies: rat monoclonal anti-neural cell adhesion molecule L1 (1:500 Millipore, MAB5272, clone 324, RRID:AB_2133200), rabbit polyclonal anti-Pax6 (1:200, BioLegend, 901302; RRID:AB_2565003), goat polyclonal anti-GFP (1:200, Abcam, ab6673, RRID:AB_305643), rabbit polyclonal anti-GFP (1:200, Abcam, ab290, RRID:AB_303395). The following secondary antibodies from Thermo Fisher Scientific were incubated at room temperature for 1 h: donkey anti-rat Alexa⁴⁸⁸ (1:100, Thermo Fisher Scientific, A21208), donkey anti-rat Alexa⁵⁹⁴ (1:100, Thermo Fisher Scientific, A-21209), donkey anti-rabbit Alexa⁵⁶⁸ (1:100, Thermo Fisher, A10042), donkey anti-rabbit Alexa⁴⁸⁸ (1:100, Thermo Fisher Scientific, R37118) and donkey anti-goat Alexa⁴⁸⁸ (1:100, Invitrogen, A11055). Sections were counterstained with DAPI (Thermo Fisher Scientific, D1306) and mounted in ProLong Gold Antifade Mountant (Thermo Fisher Scientific, P36930).

In vitro transcription of digoxigenin-labelled probes was carried out using DIG RNA-labelling kit (Sigma-Aldrich, 11175025910). The following digoxigenin-labelled probes were synthetized in the lab from cDNA: Ntn1 (kindly donated by Dr Thomas Theil, (University of Edinburgh, UK); forward primer, CTTCCTCACCGACCTCAATAAC; reverse primer, GCGATTTAGGTGACACTATAGTTGTGCCTACAGTCACACACC), Sema3a (forward primer, ACTGCTCTGACTTGGAGGAC; reverse primer, ACAAACACGAGTGCTGGTAG); Plxna1 (forward primer, GACGAGATTCTGGTGGCTCT; reverse primer, CATGGCAGGGAGAGGAAGG), DCC (forward primer, AACAGAAGGTCAAGCACGTG; reverse primer, CAATCACCACGACCAACACA), Unc5a (forward primer, CTGTCAGACCCTGCTGAGT; reverse primer, GGGCTAGAGTTCGCCAGTC), Unc5d (forward primer, GGACAGAGCTGAGGACAACT; reverse primer, GTATCAAACGTGGCGCAGAT). Unc5c probe was kindly donated by Dr Vassiliki Fotaki (University of Edinburgh, UK) and Dr Suran Ackerman (UC San Diego, USA). Cryosections were processed for in situ hybridization using standard protocols. Some slides were counterstained using nuclear Fast Red (Vector Laboratories, LS-J1044-500).

For cortical injections, brains were dissected between E15.5 and E18.5, and fixed in 4% PFA in PBS at 4°C for at least 48 h. After washes in PBS, filter paper impregnated with DiI (NeuroVue Red, Molecular Targeting Technologies, FS-1002) and DiA (NeuroVue Jade, Molecular Targeting Technologies, FS-1006) was inserted approximately in the somatosensory and visual areas of the cortex, respectively. Brains were incubated at 37°C in PBS for 4 weeks to allow the diffusion of the tracers.

For thalamic injections in fixed tissue, embryos were dissected at E13.5 and fixed overnight in 4% PFA in PBS at 4°C. After PBS washes, brains were cut in half at the midline and DiI was inserted in the thalamus using a fine probe. Brains were incubated for 1 week in PBS at 37°C.

Brains were then cryoprotected in 30% sucrose, embedded in OCT compound and sectioned using a cryostat at 30 µm. Sections were counterstained with DAPI diluted 1:1000 in distilled water.

For thalamic injections in non-fixed tissue, we applied neurobiotin (Vector Laboratories, SP-1120), and an amino derivative of biotin was used as an intracellular label for neurons. The tracer in powder was held at the tip of an entomological needle (00) and recrystallized using vapour from distilled water. Brains were cut in half and the crystal was inserted in the thalamus. Brains were then immersed in continuously oxygenated Ringer's solution (124 mM NaCl, 5 mM KCl, 1.2 mM KH₂PO₄, 1.3 mM MgSO₄ 7H₂O, 26 mM NaHCO₃, 2.4 mM CaCl₂ 2H₂O, 10 mM glucose) and incubated overnight at room temperature. The tissue was fixed in 4% PFA in PBS overnight at 4°C, washed in PBS, cryoprotected in 30% sucrose and sectioned using a cryostat at 10 µm. Neurobiotin was visualized by incubating the sections with either Strep⁴⁸⁸ or Strep⁵⁴⁶.

E13.5 embryos were dissected, embedded in 4% low melting temperature agarose (Lonza, 50100) and sectioned using a vibratome to produce 300 µm coronal slices. Lateral or medial thalamic explants were dissected from slices belonging to GFP-positive embryos and transplanted into equivalent rostral/caudal slices belonging to GFP-negative embryos (see schemas in Fig. 5). The thalamus and its different medio-lateral regions were recognized under the dissecting scope by anatomical landmarks. Slices were then cultured for 72 h in floating membranes (Whatman nuclepore track-etched membranes, WHA110414) over serum-free Neurobasal medium (Thermo Fisher Scientific, 21103049) in 60 mm centre well organ culture dishes (Falcon, 353037). Cultures were fixed in 4% PFA overnight at 4°C, cryoprotected in 30% sucrose and cryosectioned at 10 µm to be processed for immunofluorescence. All grafts were positioned in direct contact with presectioned host tissue to avoid pial growth.

We performed a total of 31 transplant experiments. In the first set of experiments, we grafted control thalamus into control tissue. We used embryos from four different litters for a total of nine different transplants (four using lateral thalamus and five using medial thalamus as donor tissue). In the second set of experiments, we grafted Pax6 cKO lateral thalamus into control thalamus, using 11 embryos from four different litters and a total of 22 individual (bilateral) transplants.

Images of transverse sections were analysed blind using Fiji software (Schindelin et al., 2012) for seven E15.5 embryos belonging to three different litters (four controls and three CAG^CreER Pax6 cKOs). For each embryo, we analysed at least five sections at different rostral-caudal levels. For each section, we isolated the DAPI channel and defined the dLGN and VP nucleus as regions of interest (ROIs) blind to the other channels (DiI and DiA). We next measured the area occupied by DiI and DiA within each ROI, defining a positive label using the automated thresholding function in Fiji software set for ‘MaxEntropy’. Data from all sections and genotypes was statistically assessed to test the effects of Pax6 inactivation on the area occupied by each dye on each nucleus. Data were fitted to a mixed linear model using the lmer() function from lme4 R package (Bates et al., 2015). ‘DiI in dLG’, ‘DiI in VP’, ‘DiA in dLG’ and ‘DiA in VP’ were set as dependant variables, with ‘genotype’ as a fixed effect and ‘litter:embryo’ factors as nested random effects. Therefore, each litter was considered a biological replicate (n=3). P-values of fixed effects were obtained using the Anova() function from car package (Fox and Weisberg, 2019).

Images of transverse sections of our explants were analysed using Fiji software (Schindelin et al., 2012). We quantified a total of ten culture thalamic explants (three from lateral control donor tissue, three from medial control donor tissue and four from lateral Pax6 cKO tissue) from seven different litters, using five slices per explant corresponding to different rostral-caudal levels of the explants. For each section we first divided the thalamus into three equal medial-lateral sectors (defined as ROIs, see Fig. 5), avoiding the area occupied by the explant itself. We then isolated the GFP channel and measured the area occupied by GFP-positive elements within each ROI, defining positive label using the automated thresholding function in Fiji software set for ‘MaxEntropy’. To test for the differences of GFP occupancy in the three different areas (lateral, intermedial and medial), the data were fitted to a mixed linear model using the lmer() function from lme4 R package (Bates et al., 2015). For each experimental group (lateral control explants, medial control explants and lateral Pax6 cKOs explants), GFP occupancy in ‘lateral’, ‘intermedial’ and ‘medial’ thalamic sectors were set as dependant variables, with ‘genotype’ as a fixed effect and ‘litter:embryo’ factors as nested random effects. Therefore, each litter was considered a biological replicate (n=3 for lateral and medial control explants, and n=4 for lateral Pax6 cKOs explants). P-values of the linear hypotheses on fixed effects were obtained using the Anova() function from car package (Fox and Weisberg, 2019).

For the DiI and thalamic explant quantifications, data were presented as box plots representing the median value and the distribution of the quartiles. Individual datapoints representing measurements of each slice were plotted overlying the box plots.

Images were analysed blind for at least three E13.5 embryos for each condition belonging to three different litters. We positioned three lines across the prethalamus: (1) at the thalamic-prethalamic border (Th-PTh), guided by prethalamic expression of Pax6; (2) at a lower prethalamic position (low-PTh), guided by the end of Pax6 prethalamic expression; and (3) at the midpoint position between the two other lines (mid-PTh). We generated a L1 intensity profile using Fiji software (Schindelin et al., 2012). Intensity profiles were then processed by tracing a line at an arbitrary (but constant for all quantifications) intensity level and quantifying the number and width of bundles crossing the line. Statistical significance was assessed applying two-tailed unpaired Student's t-test and n=3.

We used Free-D software (Andrey and Maurin, 2005) to reconstruct the structure of thalamus and prethalamus from transverse slices stained with DAPI and using antibodies against L1 and Pax6 to reveal the thalamocortical tract and the limits of the diencephalic structures, respectively. The thalamus territory was recognisable by an intense DAPI staining and Pax6-negative mantle zone, contrasting with prethalamus and pretectum, which express high levels of Pax6 in the postmitotic neurons. The location of the signalling molecules was included in the model by comparison of transversal and sagittal sections stained for Ntn1 and Sema3a and their adjacent sections processed for Pax6 and L1 with the sections used to build the model scaffold. Dots are representative of staining density. We used sagittal and transverse sections from four embryos from three different litters.

In situ hybridization and immunohistochemical images were taken with a Leica DMNB microscope coupled to a Leica DFC480 camera. Fluorescence images were taken using a Leica DM5500B automated epifluorescence microscope connected to a DFC360FX camera. Image panels were created using Adobe Photoshop CS6.

We thank Dr Martine Manuel for her invaluable help with the generation of mice lines for the explant experiments and Dr Vassiliki Fotaki for providing the Unc5c probe.

The peer review history is available online at https://dev.biologists.org/lookup/doi/10.1242/dev.184523.reviewer-comments.pdf