A Novel Approach to Selectively Target Neuronal Subpopulations Reveals Genetic Pathways That Regulate Tangential Migration in the Vertebrate Hindbrain

Vertebrate genes often play functionally distinct roles in different subsets of cells; however, tools to study the cell-specific function of gene products are poorly developed. Therefore, we have established a novel mouse model that enables the visualization and manipulation of defined subpopulations of neurons. To demonstrate the power of our system, we dissected genetic cascades in which Pax6 is central to control tangentially migrating neurons of the mouse brainstem. Several Pax6 downstream genes were identified and their function was analyzed by over-expression and knock-down experiments. One of these, Pou4f2, induces a prolonged midline arrest of growth cones to influence the proportion of ipsilaterally versus contralaterally settling neurons. These results demonstrate that our approach serves as a versatile tool to study the function of genes involved in cell migration, axonal pathfinding, and patterning processes. Our model will also serve as a general tool to specifically over-express any gene in a defined subpopulation of neurons and should easily be adapted to a wide range of applications.


Introduction
Understanding cell-specific regulatory mechanisms is a major challenge in the post-genome era. Particularly in mammals, the reiterated usage of the same transcription factor in distinct subsets of cells or during distinct developmental time points provides the basis to generate thousands of individual cell types with a relatively small number of genes. A single transcription factor may therefore elicit variable downstream effects depending on the context of its expression. Tissue-specific knockout strategies based e.g. on the Cre-lox-system, or promoter-driven transgenic models allow a cellspecific manipulation of genes. However, as these techniques rely on the generation of new transgenic animals for each genecombination analyzed they are laborious and time-consuming. Here, we combined a transgenic model with tissue-specific transfection protocols and organotypic cultures to enable the quick analysis of numerous genes in a cell-specific manner. As a proof of principle we applied our system to decode molecular pathways initiated by the transcription factor Pax6 which is involved in neuronal cell migration and axonal pathfinding processes.
Pax6, a homeodomain and paired domain containing transcription factor, is a major determinant of visual and olfactory sensory structures and is essential for a variety of patterning and pathfinding processes throughout the nervous system [1][2][3]. Depending on the context and area of expression Pax6 initiates varying downstream effects. Homozygous small eye (Pax6 Sey/Sey ) mouse and rat embryos, which lack functional Pax6, do neither generate eye nor nasal structures and are deficient in ventral diencephalic structures [4][5][6][7][8][9]. In the ventral hindbrain and spinal cord, Pax6 controls the dorso-ventral patterning of motorneurons and of interneurons [5,10]. In the cerebral cortex Pax6 determines the neurogenic potential of radial glial cells [11,12]. Throughout the developing nervous system, with the exception of the midbrain, Pax6 is expressed in a ventral and a dorsal pool of progenitor cells. Although the dorsal Pax6 expression domain has achieved much less attention than the ventral domain there is evidence that Pax6 plays a pivotal role in the specification and migration of neurons derived from this domain [13][14][15][16].
The dorsal domain of Pax6 positive neuronal precursors of the hindbrain includes the rhombic lip (RL) [14,16] which comprises the interface between the dorsal neuroepithelium and the roof plate. The RL is the source of several tangentially migrating neurons (see also Figure 1A) [14,[17][18][19][20][21][22][23][24]. The most notable are the neurons of the marginal migratory stream (mms; also pes) which migrate from the rhombic lip circumferentially around the medulla towards their contralateral destinations to settle in the ECN (external cuneate nuclei) and the LRN (lateral reticular nuclei) [17,25]. Owing to the superficial nature of the mms migration these neurons serve as paradigm to study neuronal migration and axonal pathfinding processes.
The highly complex neuronal circuits of the vertebrate nervous system are established during development when growing axons travel considerable distances towards their targets to generate the appropriate connections. This wiring process depends on attractive and repulsive factors which emanate from final or intermediate cellular targets and which are interpreted by cell surface receptors located on axonal growth cones [26,27]. Although the general principles were uncovered during the past years our understanding of axonal pathfinding processes is far from being complete. Current methods to analyze candidates regulating neuronal migration and axonal navigation processes are laborious and often involve the generation of transgenic animals for each gene analyzed. Non-transgenic methods, as DiI labeling of neurons or vector-driven mis-expression of gene constructs, are suitable for use with certain applications, however, they are neither cell specific nor can they be targeted to distinct neuronal subpopulations.
Here we describe a novel transgenic mouse model, which allows the specific and exclusive visualization and manipulation of subsets of neurons in the developing brain. To demonstrate the power of this system we have analyzed the role of Pax6 in migrating neurons of the brainstem. In Pax6 mutant mice migration of these neurons is distorted and some neurons differentiate at ectopic positions. Using transplantation, knock-down and over-expression experiments we show that distinct migratory features are controlled by discrete sets of Pax6 downstream genes. These results demonstrate the potential of our transgenic mouse model as a tool to study the role of Pax6 in individual neurons. Moreover, our system should be widely applicable to study virtually any gene that acts during cell determination, axonal pathfinding and/ or cell migration processes.

A novel mouse model to visualize and manipulate subsets of neurons
The functional analysis of genes in restricted tissues often involves the generation of inducible knockout mice or mice over expressing transgenic constructs. To simplify this time-consuming process we developed an in vitro model that enables the visualization and manipulation of defined populations of neurons. To label neurons in a largely unlabelled background we searched for genes that were expressed in only a subset of neuronal precursors and in migrating neurons. Pax6 meets these criteria ideally. Pax6 is expressed in several groups of tangentially migrating neurons and their precursors as well as in a small population of radially migrating neurons and their precursors ( Figure 1A-1C) [5,10,12,14,16,28]. We adopted the Tet binary system [29] and generated YAC (yeast artificial chromosome) transgenic mice which expressed the tetracycline dependent transactivator (tTA) in all Pax6 positive cells. A 420 kb YAC spanning the human PAX6 locus (Y593) [30] was modified such that the PAX6 coding region was replaced with a cassette containing an IRES (internal ribosomal entry site) and the tTA ( Figure 1D). Previously, we and others had shown that the unmodified YAC Y593 contains all elements driving full functional PAX6 expression [30][31][32] and, in agreement with this, Tg (PAX6-tTA) mice showed a wide overlap of tTA and endogenous murine Pax6 expression ( Figure S1). Tg (PAX6-tTA) mice were entirely normal and control experiments insured that neuronal patterning and migration was unaltered.
tTA is a transcriptional activator that at moderate levels of expression is completely inert in vertebrates, yet, enables the activation of artificial constructs containing a tTA-DNA-binding element (TRE = tetracyline responsive element). To examine whether our transgenic model specifically allows the labeling of only Pax6 positive cells we introduced by electroporation reporter constructs driving the green fluorescent protein into transgenic embryos. In all instances only Pax6 positive cells, e.g. retinal precursor cells, cortical precursors, or cerebellar granule cells, expressed the reporter genes ( Figure S1). Non-transgenic embryos or Pax6 negative tissues did not induce reporter gene expression ( Figure 1H, Figure S1). Together these results demonstrate that Tg (PAX6-tTA) mice enable the targeting of reporter gene constructs specifically to Pax6 positive cells and tissues during development.
As Tg (PAX6-tTA) mice allow any gene to be targeted to Pax6 expressing cells, they are of potential value to study neuronal migration and axonal pathfinding processes and for the analysis of Pax6 downstream effects. As a proof of principle, we chose to focus on the marginal migration stream (mms). Like other tangentially migrating neurons, mms neurons use the same or similar navigational cues as do growing axons, and migration of mms neurons is severely disturbed in Pax6 mutant Pax6 Sey/Sey mice (see below). Neurons of the mms are generated at the rhombic lip and migrate circumferentially around the embryonic brainstem to generate the contralateral lateral reticular (LRN) and the external cuneate (ECN) nuclei ( Figure 1A) [16,[21][22][23]25]. Migration starts at E13.0 and is completed by E16.5. Pax6 is expressed in precursors at the rhombic lip, in all migrating neurons of the mms and during the initial period of settling in the target nuclei (Figure 1B, 1C and data not shown). Antibody staining and in situ hybridization (not shown) of Tg (PAX6-tTA) mice confirmed a complete overlap of Pax6 and tTA expression in these neurons ( Figure 1I).
To visualize migrating mms neurons in Tg (PAX6-tTA) mice, reporter constructs were introduced into neuronal precursor cells in the left rhombic lip by electroporation at E12.5 before migration had begun ( Figure 1E). Whole brainstems including the cerebellar primordium were then sustained in organotypic filter cultures for up to 14 days as an open book preparation which allowed the observation of migrating neurons with a fluorescence microscope from above ( Figure 1F, 1G). Our approach to use a binary system ensured that only Pax6 positive neurons containing tTA and a TRE reporter construct expressed the desired reporter genes. This procedure resulted in the specific labeling of mms neurons originating only from one rhombic lip. Pax6 positive neurons originating from the opposite rhombic lip remained unlabelled as were Pax6 negative (and therefore tTA negative) neurons originating from regions close to the rhombic lip. Unlabelled neurons included neurons of the submarginal migration stream (sms) which generate the inferior olive (IO) thus demonstrating the specificity of our model. To allow the

Author Summary
In mammals, many genes execute a unique set of distinctive and common functions in different cell types. Strategies to address these individual roles often involve the generation of series of transgenic animals. Here, we present a novel approach that combines a single transgenic mouse line with tissue-specific transfection protocols and organotypic cultures to enable the quick analysis of numerous genes in a cell-specific manner. As a proof of principle, we analyzed the function of transcription factors in tangentially migrating neurons of the developing vertebrate hindbrain. We identified a temporary halt in migration as a novel mechanism for neurons to decide whether to cross or not cross the midline. Our model may serve as a general tool to quickly study axonal pathfinding, neuronal cell migration, or patterning processes in a well-defined population of neurons. Genotyping of transfected cultures confirms that responder constructs are exclusively activated in Tg (PAX6-tTA) and not in non-transgenic embryos (H). Double-immunolabeling of mms neurons with Pax6 and VP16 (to recognize tTA) antibodies confirm a complete overlap of Pax6 and tTA expression (I). A responder vector containing two TRE elements drives expression of two genes simultaneously: a cytoplasmic green fluorescent protein (EGFPm) and a nuclear red fluorescent protein (DsRed2nls) (J). The left and right images are green and red fluorescent images, respectively, of migrating mms neurons after 1.5DIV. K, K9, and L are low magnification fluorescence (K, L) and phase contrast (K9) images of cultures transfected with an EGFP simultaneous visualization and manipulation of neurons we designed reporter constructs containing two TRE elements ( Figure 1J). Control constructs co-expressed a cytoplasmic green fluorescent protein (EGFPm) and a nuclear red fluorescent protein (DsRed2nls) in 99% (61; n = 10) of labeled neurons demonstrating that our reporter constructs enable the co-expression of two genes in the same neurons ( Figure 1J). To enable statistical analysis of the cultures, the territories of the LRN and the ECN were delineated using visible landmarks ( Figure 1K Pax6 plays multiple roles in patterning and guiding migrating ECN and LRN neurons Pax6 mutant Pax6 Sey/Sey mice display multiple neuronal patterning and migration defects. We therefore wished to determine whether Pax6 also regulates the mms. At the anatomical level, several features of the mms are severely disturbed in Pax6 mutant Pax6 Sey/Sey embryos. Most noticeable, the initiation of migration and the midline crossing was delayed by 0.5 days (asterisks in Figure 2A, 2A9 and data not shown; see also Figure S2 and Figure 4A, 4A9 The expression patterns of Pax2, Dcx, NK1R, and DopH was unaltered indicating that there is no general developmental delay in the mutant brainstem (data not shown). In Pax6 Sey/Sey embryos some migrating mms neurons used a submarginal instead of a marginal migration path (black arrowhead in Figure 2A9; see also Figure S2 and Figure 4A9) and at E14.5 a large number of mutant neurons accumulated around the midline suggesting a reduced pace in midline crossing (black arrow in Figure 2B, 2B9). Furthermore, a subset of Pax6 positive neurons migrated along the midline into the parenchyma of the hindbrain (white arrowheads in Figure 2C, 2C9). We used a variety of markers, e.g. antibodies against the potassium channel Kcnj6, and DiI tracing of mossy fiber projections to discriminate between mms neurons (generating the ECN and LRN) and neurons of the sms (generating the IO). These experiments all indicated a complete loss of neurons in the ECN and a disorganized settling of neurons in the LRN ( Figure 2D, 2D9, 2E, 2E9; Figure S2, and data not shown). Many Kcnj6 positive neurons were even observed within the inferior olivary territory (black arrows in Figure 2E9) and dorsally to the IO at the midline (open arrow in Figure 2E9). In agreement with previous reports [16], we found a slight enlargement of the IO at E14.5 when we used the Ets transcription factor Etv1 as a IO specific marker [33] ( Figure S3 and data not shown). However, by labeling for the axon-guidance-molecule B (RgmB) no alterations in the general architecture of the IO were seen [34] ( Figure S3). This is consistent with our observation that misguided Pax6 Sey/Sey mms neurons make only a negligible contribution to the IO or settle in the periphery of the IO. In summary, these data demonstrate that migration of Pax6 Sey/Sey mms neurons is severely disrupted. Mutant neurons of the mms are initially delayed. Later, a number of neurons use a sub-marginal migration path, migration is disturbed at the midline and several neurons migrate to ectopic positions along the midline. Lastly, the normal structure of the LRN is lost, the ECN is completely missing, and the IO is enlarged. In order to dissect the complex neuronal cell migration defects observed in Pax6 Sey/Sey mice, Pax6 Sey/Sey mice were crossed to the Tg (PAX6-tTA) transgenic line. Comparison of cultures obtained from wt and from Pax6 Sey/Sey embryos confirmed the anatomical observations described above. Pax6 Sey/Sey mms cells showed an initial delay in the onset of the migration ( Figure 3A, 3B) and a disturbance at midline crossing later on. After 5 DIV (5 days in vitro) Pax6 Sey/Sey mms neurons settled randomly in the LRN ( Figure 3B9) but failed to form any ECN structures ( Figure 3B''). In contrast, wt cultures formed a well organized LRN in which cells settled in a dorsal and ventral sub-nucleus of the LRN ( Figure 3A9) and in a distinguished ECN ( Figure 3A''). To quantify the effect we counted labeled cells in cultures from wt (LRNi = ipsilateral LRN = 11267; LRNc = contralateral LRN = 266616; ECN = 86614; total number of cells = 610634; n = 50) and Pax6 Sey/Sey embryos (LRNi = 313623; LRNc = 426646; ECN = 862; total number of cells = 670671; n = 8) ( Figure 3K). These data suggested that there were no alterations in the gross number of migrating neurons between wt and Pax6 Sey/Sey embryos and confirmed the complete absence of an ECN in Pax6 Sey/Sey embryos. In both, mutant and wt tissues, a proportion of LRN neurons settled ipsilaterally ( Figure 3K). All LRN neurons, however, projected to the contralateral cerebellum in respect to their origin from one rhombic lip, explaining the observations by Bourrat and Sotelo of an ipsilateral and contralateral contribution of mossy fibers [17].
The Pax6 Sey/Sey migration defect could be caused either by a direct cell autonomous action of Pax6 in migrating RL precursors or via an indirect non-autonomous effect, for example in the ventral domain of Pax6expression (e.g. by altering migration cues at the midline). We performed three types of experiments to discriminate between these alternatives. First, we transplanted transfected Pax6 Sey/Sey rhombic lips onto wt brainstems and vice versa ( Figure 3C, 3D). Unexpectedly, migrating Pax6 Sey/Sey neurons (in a wt host) formed a well organized LRN ( Figure 3C9), but no ECN ( Figure 3C''). In contrast, wt neurons (in a mutant host) failed to form a correctly organized LRN ( Figure 3D9), but were able to generate a normal ECN ( Figure 3D''). These data suggested, that Pax6 may act cell autonomously in generating ECN neurons, but non-autonomously in specifying the correct sub-organization of LRN neurons. To further validate this assumption we rescued the Pax6 Sey/Sey migration defect by reexpression of Pax6. We tested the two major splice variants and of these, the expression of the Pax6(-5a) isoform in Pax Sey/Sey 6 rhombic lips resulted in a full recovery of the ECN ( Figure 3E''), but a disorganized LRN ( Figure 3E9), whereas, the Pax6(+5a) variant was ineffective (not shown). Thus, in the RL Pax6 splice variants differ in their biological activity, similar to the embryonic cortex [35]. Lastly, we diminished the endogenous Pax6 mRNA by using siRNAs. Transfection of siRNAs or over expression of shRNA constructs directed against Pax6 ( Figure S4) resulted in a massive reduction of ECN cells in wt explants ( Figure 3F''), whereas, control constructs had no effect (not shown). The effect was quantified by counting labeled cells that had settled in the ECN ( Figure 3L). Taken together, these experiments demonstrate that our model system enables the simultaneous visualization and manipulation of tangentially migrating cells in the mouse brainstem. In addition, we have shown that Pax6 plays numerous distinct roles in the formation and migration of mossy fiber producing neurons. Moreover, the combination of a binary model and organotypic culture assays facilitates a quick discrimination between cell-autonomous and non-autonomous effects.

Axonal pathfinding receptors guide mms neurons
We identified several genes whose expression was altered in Pax6 Sey/Sey mms neurons ( Figure S5; see also Materials and Methods). To gain more insights into the function of these putative Pax6 downstream targets all genes were over-expressed or their expression level was diminished with shRNAs. Those genes which showed the most noticeable effects are summarized in Table 1.
The altered migration and settling behavior of Pax6 Sey/Sey ECN/ LRN neurons suggested that migration cues were changed in Pax6 Sey/ Sey embryos. The most prominent candidates are ligand/ receptor couples of the Slit/ Roboand Netrin/ Dccpathways [36,37]. Expression of Netrin1, Dcc, and Robo1,2 and 3 was unaltered in migrating Pax6 Sey/Sey mms neurons ( Figure 4A, 4A9, Figure S2, and data not shown). However, Slit1 and Slit2 which were expressed in the hypoglossal nuclei were both lost in Pax6 Sey/Sey embryos ( Figure 4B, 4B9 and Figure S2) [5,10]. Motorneurons of the hypoglossal nuclei are in close proximity to the LRN settling territories suggesting that Slit1 and Slit2 expression provided from these neurons may determine the place of LRN settlement. To test this hypothesis we performed transplantation experiments and shRNA driven knockdown of the Slit-receptor Robo3 in migrating mms neurons. Both types of experiments resulted in a disorganized LRN similar to the phenotype observed in Pax6 Sey/Sey mice ( Figure 4C-4H). The above results indicate that factors provided from the hypoglossal nucleus, (most likely Slit1 and Slit2) determine the place of LRN settlement. These data also explain the cell non-autonomous role of Pax6 during this process. Hypoglossal neurons are Pax6 negative, but are completely lost in Pax6 Sey/Sey embryos ( Figure S2) [5,10]; hence, Slit1 and Slit2 are most likely not direct targets of Pax6. Additional experiments suggest that Slit1 and Slit2 may also act as repellent to push mms neurons to the marginal migration route during the initial phase of migration (data not shown).

Pou4f2 controls ipsilateral versus contralateral settling of LRN neurons and causes a migrational arrest at the midline
Two POU transcription factors were among the genes whose expression pattern was altered in the mms of Pax6 Sey/Sey embryos. Pou4f2 (also: Brn3b) was strongly expressed in about 18.6% (64.6%, n = 3) of E14.5 and 23.3% (66.2%, n = 3) of E15.5 wt mms neurons but was completely lost in the Pax6 Sey/Sey mms ( Figure 5A, 5A9, 5B, 5B9). Pou4f1 (also: Brn3a) was expressed between E13.5 and E15.5 in a subset of mms neurons, but was upregulated in the E14.5 and E15.5 Pax6 Sey/Sey mms ( Figure 5C, 5C9). Expression of Pou4f1 and Pou4f2 in Pax6 Sey/Sey IO neurons was unaltered ( Figure 5A, 5A9, 5C, 5C9). Pou4f2 plays several roles in specifying and guiding retinal ganglion cells and their axons. We therefore asked whether Pou4f2 may accomplish similar tasks in rhombic lip derived neurons. Pou4f2 was only expressed in a subset of wt mms neurons. We therefore over-expressed Pou4f2 in all migrating mms neurons. Remarkably, growth cones of all Pou4f2 over-expressing neurons were arrested at the midline for about 1.5 days (60.5 days, n = 17), whereas the majority axons in control cultures crossed the midline instantly ( Figure 5D, 5F). Interestingly, in control cultures the growth cones of some neurons also appeared to be arrested at the midline: 5% (63%) at 1DIV, 15% (66%) at 2DIV, 25% (65%) at 3DIV, and 6% (63%) at 4DIV (n = 11). This correlates well to the peak of Pou4f2 expression at E14.4 and E15.5 (in cultures: 2DIV and 3DIV). Over-expression of Pou4f2 had also a noticeable effect on the settling behavior of LRN neurons. Quantification of LRN neurons at 5DIV revealed that Pou4f2 expressing LRN neurons preferably settled at the ispilateral side (LRNc/LRNi = 0.8 6 0.1, n = 17; Figure 5G, 5M) compared to control cultures in which the majority of LRN neurons settled at the contralateral side (LRNc/LRNi = 2.5 60.1, n = 50; Figure 5E, 5M). Similar relations were obtained at 6DIV and 8DIV suggesting that Pou4f2 over-expression altered the migration behavior of mms neurons and did not cause a delayed settlement of these neurons. The effect was specific to Pou4f2 and could not be mimicked by over-expression of Pou4f1, Pou4f3 or Pou6f1 ( Figure 5M and data not shown). Together these data suggest that Pou4f2 acts through a novel mechanism which induces an arrest of growth cones at the midline to regulate the ratio of ipsilaterally versus contralaterally settling neurons.
We altered expression levels of about 25 potential Pou4f2 retinal target genes [38][39][40] and of these two showed an effect on the migration behavior of mms neurons. Over-expression of Gfi1, a zinc finger transcription factor, reduced the contra-/ipsi-lateral ratio of LRN neurons ( Figure 5M). In contrast, the downregulation of Gap43 by shRNA constructs caused a higher contra-/ ipsi-lateral ratio of LRN neurons ( Figure 5M and Figure S4). Gap43 is slightly reduced in the Pax6 Sey/Sey mms ( Figure S5). In addition, mis-expression of Pou4f2 resulted in a massive downregulation of Pou4f1 in transfected, but not in control, rhombic lips ( Figure 5N), suggesting that the loss of Pou4f2 in Pax6 Sey/Sey mms neurons leads to an up-regulation of Pou4f1 ( Figure 5C, 5C9). Pou4f1 over-expression or down-regulation, however, did not alter migration behavior of mms neurons ( Figure 5M).
Pou4f2 is expressed only in a subset of Pax6 positive mms neurons suggesting that other factors together with Pax6 may co-regulate Pou4f2. In the developing retina Pou4f2 expression depends on two transcription factors: the bHLH protein Math5 and the zinc finger gene Wt1 [41][42][43][44]. Wt1 was found to be expressed in the rhombic lip, though, in a region just dorsally to the Pax6 positive domain ( Figure 5J). Math5 was neither expressed in the rhombic lip nor in migrating mms neurons, however, a close homologue, Math1, was expressed in neuronal precursors at the rhombic lip and in a subset of early migrating mms neurons [22,23] (Figure 5K). Thus, Math1, but neither Math5 nor Wt1, was the most likely candidate to regulate Pou4f2 or Pou4f1 expression in mms neurons. Consistent with this, mis-expression of Math1, but not of Wt1 (+ and -KTS splice variants) or Math5, led to a midline arrest of migrating mms neurons and a reversed settling behavior of LRN neurons In summary, our work led to the identification of a gene cascade acting in tangentially migrating neurons of the brainstem, in which Pou4f2 plays a central role to induce a previously unknown mechanism that controls midline crossing behavior. Furthermore, our results imply that our model system is applicable to quickly analyze genetic hierarchies in Pax6 positive cells and may therefore serve as a general tool.

Discussion
Tg (PAX6-tTA) mice as a model for Pax6 function, cell migration, and axonal pathfinding processes The extraordinary complexity of cell determination, migration and wiring processes in the developing mammalian brain creates a major challenge for developmental neurobiologists. Here, we introduced a simple yet powerful technology to quickly analyze any gene potentially involved in these processes. Our model is of threefold use: first to study the function of Pax6 and of Pax6 downstream genes in their genuine environment, second to investigate genes involved in general patterning, axonal pathfinding and cell migration processes, and third to enable the analysis of tissue-specific gene functions. The Tg (PAX6-tTA) model complements and improves existing approaches and has certain benefits: it combines cell specific transfection protocols and organotypic culture assays, thus, facilitating the quick analysis of genes in a natural tissue environment. The experimental design and the binary nature of the Tg (PAX6-tTA) model is fundamentally simple and has several advantages over systems that are based purely on transgenic animals. First, the electroporation and subsequent culture of embryonic tissues allows the screening of large number of genes without the need of generating new transgenic animals for each construct. In fact, less than 10% of the constructs we have tested revealed phenotypes. Thus, only those genes showing positive results in culture assays may be used subsequently to generate transgenic lines. Of note however, some of the phenotypes reported here, for example, the midline arrest or the altered ipsi-to contra-lateral ratio, would have been missed in purely transgenic systems. Second, variation of the electroporation protocol allows transfections ranging from just a few cells to a complete Pax6 expression domain with thousands of cells. Hence, our approach allows adjustment according to the needs: either to monitor single migrating cells or to determine global patterning effects. In addition, neighboring cells and non-electroporated contra-lateral sides serve as internal controls. The usefulness of our system critically depends on the tightness of the TRE based promoter and on the ability of the constructs to express two genes simultaneously. To ascertain the tightness of our system we used repeated electroporations and high DNA concentrations (up to 5 mg/ml). Even under these extreme conditions we were never able to detect any reporter gene expression in Pax6 negative cells at any developmental stage. Thus, under the conditions used in this report the combination of Tg (PAX6-tTA) mice and TRE based promoters allow expression of reporter gene constructs only in Pax6 positive cells. It is also important to note, that our strategy to use a YAC based technology combined with an internal ribosomal entry site (IRES) resulted in moderate levels of reporter gene expression which were in the range of physiological concentrations. To ensure the simultaneous expression of two reporter genes we tested several types of TRE constructs. Only our approach, to use two consecutive TRE based promoters led to the activation of nearly equal amounts of two genes at the same time in the same cell. A bidirectional TRE element that previously had been shown to work in transgenic animals failed in our system [45]. One obvious difference is that in transgenic animals typically multiple copies of constructs are stably integrated into the genome, whereas, in our assay transfections were transient.
Pax6 loss of function phenotypes are often highly complex involving massive malformations in the affected organs. Pax6 is expressed in neuronal precursors of the telencephalon, commissural neurons in the dorsal spinal cord, in adult neuronal stem cells, the early eye cup, in the pancreas, in precursors and in migrating cells of several tangential and radial migration streams of the rhombencephalon and of the forebrain [5,6,10,11,14,28,46,47]. In addition to its technical advances, the Tg (PAX6-tTA) model represents a novel, highly versatile technology to study the function of Pax6 or any other gene in these tissues. As a paradigm, we have dissected the role of Pax6 in tangentially migrating cells of the brainstem. In principle, however, this system shall be applicable to any Pax6 positive tissue and we have initial evidence that our model allows to specifically target Pax6 positive telencephalic precursor cells, cerebellar granule cells, the developing retina, the rostral migratory stream, the pontine migration and ventral precursor cells of the brainstem and spinal cord ( Figure S1 and data not shown). With the help of this model it should therefore be possible to systematically analyze cell fate decisions and the migratory behavior of Pax6 expressing cells at any developmental stage.
Several studies have revealed that Pax6 is required for hindbrain and spinal cord development [5,7,10,14,15]. Our work adds that Pax6 also controls the determination and migration of rhombic lip derived neurons (for a summary see Table 1 and Figure 6). Pax6 functions twofold: first, Pax6 controls guidance cues which push migrating mms neurons to the marginal path and which control the settling pattern of LRN neurons. The most likely sources of these cues are the hypoglossal nuclei which are located close to the midline and in proximity to the LRN. Slit1 and Slit2 are expressed in the hypoglossal nuclei and the Slit receptor Robo3 is expressed in migrating mms neurons [48]. Slit expression provided by the hypoglossal nuclei may therefore act as repellent to push mms neurons to a marginal migration route and may also specify the settlement of neurons in the LRN. The loss of Slit-expressing hypoglossal nuclei in Pax6 Sey/Sey embryos [5,10] causes a major reduction of the repellent (a minor source of Slit is still present in midline cells). Consequently, migrating mms neurons would use a more sub-marginal migration route and settle less organized in Pax6 Sey/Sey embryos. Furthermore, Slit expression at the RL may be involved during the initial phase of mms migration. Secondly, Pax6 functions cell-autonomously in migrating mms neurons to control the determination, the timing of migration, and midline crossing. Several genes show altered expression in Pax6 Sey/Sey mms neurons ( Table 1: Pou4f1, Pou4f2, Unc5h1, Mafb, Chordin) and may convey individual aspects of migration.
We and others find that several transcription factors relay Pax6 downstream effects in dorsal brainstem neurons: Ngn1 in precursor cells ventral to the RL [16], and Pou4f1, and Pou4f2 in migrating neurons (this report). Mis-expression of Ngn1 or Ngn2 in Pax6 Sey/Sey embryos failed to rescue the migration defects observed in the Pax6 mutant (Table 1). Neither did the mis-expression or downregulation of these genes generate small eye -like migration defects in wt embryos (Table 1). On the other hand, Pou4f2, which is lost in the mms of Pax6 Sey/Sey embryos ( Figure 5B and also in the pontine migration and in the cerebellum, data not shown), alters migration behavior of mms neurons. Together these data suggest that Pou4f2 may regulate genes involved in pathfinding processes, whereas, Ngn1 acts earlier in the cell determination process.
There are striking similarities in gene expression pattern between sensory neurons and RL derived neurons. We found that at least two thirds of the genes which are co-expressed with Pax6 and Pou4f2 in retinal ganglion cells are also co-expressed with these genes in mms neurons. Furthermore, genetic hierarchies seem to be analogous: in the retina Math5 controls Pou4f2, which then acts upstream of Pou4f1 [38,[41][42][43], whereas, in RL derived neurons Math1, a close homologue of Math5, initiates related pathways. General genetic pathways are conserved between retinal and RL derived neurons and our model may therefore help to elucidate some of the phenotypes observed in Pou4f1 -/and Pou4f2 -/mice. Both mouse models have revealed distinct axonal pathfinding errors [39,[49][50][51][52]. Mis-expression of Pou4f2 (or Math1) in RL derived neurons stalls growth cones at the midline for several hours. To our knowledge, this is the first report of such a midline arrest and it may thereby be a paradigm for a novel mechanism controlling midline crossing. The arrest does neither induce a growth cone collapse nor does it inhibit midline crossing per se as all neurons generate axons that cross the midline after a ''waiting period''. These axons all migrated into the cerebellum like those of control cultures. Gfi1 mis-expression and Gap43 knockdown were able to partially mimic the Pou4f2 induced phenotype, however, additional unknown targets or a combinatory code may be needed to elicit the full phenotype.
As Pou4f2 was only expressed in about 1/4 of wt mms neurons, the loss of Pou4f2 in Pax6 Sey/Sey embryos mimics only minor aspects of the Pax6 Sey/Sey phenotype. The down regulation of Pou4f2 by shRNA constructs resulted in a severe midline disturbance of neuronal processes at similar to the phenotype observed in Pax6 Sey/Sey cultures, whereas, in control cultures neuronal processes crossed the midline instantly (data not shown). Comparable phenotypes were also observed in Pax6 Sey/Sey cultures, in cultures transfected with Pax6 shRNA constructs, and in Pax6 Sey/Sey brainstem sections.
Tangentially migrating neurons follow similar navigational cues as developing axons [14,19,48,[53][54][55][56][57][58]. Hence, tangentially migrating neurons of the mms provide an excellent system to study axonal pathfinding and neuronal cell migration processes. Migrating mms neurons are easily accessible as they navigate along the pial surface. Our model takes advantage of the superficial migration of these neurons and provides a straightforward assay to specifically label and manipulate these cells without affecting their surroundings. Members of most families of guidance receptors (Netrin receptors, Slit receptors, Semaphorins, Eph receptors, and Ephrins) are expressed in migrating mms neurons [48,53,54] ( Engelkamp, unpublished) and at least two of these pathways are essential for the correct guidance of mms neurons: the Slit/ Robo - [48] and the Netrin/ Dcc-pathways [53,56,57] (see also Table 1). Our system should therefore also have important implications for the study of the signal cascades entailed in these pathways.
In summary, we have established a novel model system which allows the simultaneous visualization and manipulation of neuronal subpopulations. As a prototypical model we have focused on the role of Pax6 in migrating brainstem neurons. Yet, our results imply that our model system is applicable to a range of other cells in the developing brain and may therefore serve as a general tool to quickly study axonal pathfinding, neuronal cell migration or patterning processes.

Animals
The Small Eye allele [4] was maintained on a CD1 background. Embryos were obtained from matings of heterozygote (Pax6 Sey/+ ) mice. 0.5 denotes the morning when the vaginal plug was found. Experiments were always performed on matching pairs of control (wt) and Pax6 Sey/Sey embryos that were carefully staged. All phenotypes described were confirmed on at least six individual Pax6 Sey/Sey embryos obtained from different crossings. There was no noticeable phenotypic difference between Pax6 Sey/+ and wt embryos and therefore, in our experiments wt designates wt and Pax6 Sey/+ embryos. For brainstem cultures, matings between heterozygote Tg (PAX6-tTA) and wt CD1 mice or between heterozygote Pax6 Sey/+ / Tg (PAX6-tTA) and heterozygote Pax6 Sey/+ mice (to generate Pax6 Sey/Sey cultures) were used. Genotyping was performed by PCR with primers directed against the Tet repressor (upper: GCG-CTGTGGGGCATTTTACTTTAG; lower: CCGCCAGCCCC-GCCTCTTC). All animal procedures were carried out in accordance to the guideline approved by institutional protocols.
Generation of Tg (PAX6-tTA) mice YAC Y593 [30] was modified such that exons 8 to 11 of the Pax6 gene were replaced by homologous recombination with a construct containing the following elements in 59 to 39 order: Pax6k30 -IRES -tTA -loxP -LYS2 -loxP -Pax6k32. Pax6k30 and Pax6k32 corresponded to the sequences 29.792 to 30.296 and 31.587 to 32.095 of the Pax6 cosmid cFAT5 (NCBI accession no. Z95332), respectively, and were generated via PCR. The IRES (internal ribosomal entry site) was derived from pIRES-EGFP (Invitrogen), however, the original ATG-11 start codon was reconstituted to enhance translational initiation. tTA (Tet-Onsystem), a fusion of the tetracycline repressor and the activation domain of VP16, was derived from pUHD15-1neo (Clontech). The LYS2 gene from S. cerevisiae was derived from pAF107, which was obtained from B. Dujon, Institute Pasteur, Paris, France [59]. LoxP sequences were generated via PCR. All constructs were sequence verified. Homologous recombination in yeast was performed using standard techniques. The integrity of the recombined YAC was then verified by PCR and southern blotting. Preparation of the YAC DNA and the generation of transgenic mice were as described [30].

In situ hybridisation
In situ hybridization was performed on free floating vibratome sections as previously described [14]. Probes for Math1 [60], Neurod1 and Neurod2 [61], Pax6 [47], Unc5h3 [62] and Slit1, Slit2 [63] were In Pax6 Sey/Sey embryos the mms migration is delayed, some neurons migrate sub-marginally, neurons are arrested at the midline, the ECN is lost, the LRN is disrupted, and the IO is enlarged [16]. Additionally, the aes is reduced resulting in an absence of the pontine nuclei and granule cells of the cerebellum migrate ectopically into the inferior culliculus [14]. To search for genes expressed in mms neurons a large scale in situ hybridization screen was performed and a summary of the expression data of selected genes is shown in ( obtained from H. Zoghbi, A. Bartholomä , R. Hill, S. Ackerman, and M. Little, respectively. Probes for Dcc and RgmB were as published [34,64]; other probes were obtained by RT-PCR. The PCR products were subcloned and their identities were confirmed by sequencing. The general staining patterns of all probes matched published expression patterns. Probes were as follows: Etv1 (bp 853-1820 of NM_007960); Fgfr2 (bp 343-1192 of NM_201601); Pou4f1 (bp 1321-2199 of NM_011143), Pou4f2 (bp 216 -1762 of S68377); Robo3 (bp 3648-4673 of NM_011248). Several genes which are down-or up-regulated in Pax6 Sey/Sey embryos were identified with the help of a large scale in situ screen using .300 putative candidates. Individual probes are available on request.

Recombinant constructs
The vector for the co-expression of two constructs in Tg (PAX6-tTA) mice contained the following elements in 59 to 39 order: MCSI -TRE -P min CMV -IntronA -BGHPolyA -MCSII -TRE -SV40PolyA; MCS = multiple cloning sites; TRE = 7 repeats of the tetracycline responsive element, P min CMV = minimal CMV promoter, and IntronA were from ptetOi-MCS (obtained from Martin Spiegel, Tübingen); SV40polyA and BGHPolyA = polyadenylation signals (derived from pTetOi-MCS and pRc/ CMV, Invitrogen, respectively). Fluorescent markers to label migrating cells were a modified EGFP or DsRed2 (Clontech). Full length clones for Gfi1, Math1, Math5, Ngn1, and Ngn2 were obtained from the German Resource Center for Genome Research (RZPD) and sequence verified; clones for all other genes were obtained by RT-PCR and confirmed by sequencing. Fusions with a triple HAtag or the engrailed repressor domain (EnR) were generated by PCR. shRNA constructs were generated in the psiSTRIKE vector (Promega) using the Promega Web tool for designing the hairpin oligonucleotides. In the psiSTRIKE vector shRNAs are expressed under control of the U6 RNA polymerase promoter. Efficiency of shRNA knockdown was demonstrated in HEK293 cells using the psiCHECK/ Dual Luciferase system according to the manufacturers protocol (Promega). All constructs were sequence verified.

Image analysis
Images were taken at a Zeiss Axiophot microscope equipped with a Spot camera, at a confocal Zeiss LSM microscope, or at a Leica MZ12 equipped with a camera device. Images were processed using the MetaView software (Universal Imaging Corporation) and Adobe Photoshop. To perform statistical analysis the position of the ECN and the LRN were determined in wt un-manipulated cultures by in situ RNA staining of Pax6 and Kcnj6. The resulting territories were then overlaid onto the electroporated cultures with the help of three landmarks: a) the position of the rhombic lip; b) the position of the floor plate; and c) the position of the superior and inferior olivary complexes, which both are visible in phase contrast images of the cultures. This procedure allowed classifying 97% of labeled neurons on the contralateral side and 90% on the ipsilateral side as either ECN or LRN neurons. The remaining 3% (or 10% for the ipsilateral side) of labeled cells were scattered neurons mainly in between the ECN and the LRN. Quantification of growth cones arrested at the midline in wt cultures was done by counting all growth cones in a 25 mm wide territory at the midline. Continuous observations of cultures implied that mms growth cones traveled at an average speed of at least 500 mm/day, suggesting that within any 25 mm interval only 5% of growth cones should be detected if migration would not pause. Quantification of the volume of the inferior olive was done with AxioVision (Zeiss).