A novel floor plate boundary defined by adjacent En1 and Dbx1 microdomains distinguishes midbrain dopamine and hypothalamic neurons

Borders are defined when territorial disputes are resolved. This is also the case in early embryonic CNS development, when neighboring progenitor zones are vying for territory, ultimately settling the conflict by agreement on a boundary (Briscoe and Small, 2015; Dasen and Jessell, 2009; Kiecker and Lumsden, 2005; Prochiantz and Di Nardo, 2015). Typically, diffusible morphogens drive the expression of transcription factors initially in a ‘fuzzy’ manner, which then through cross-repressive interactions establish sharply defined, molecularly distinct progenitor zones. Accurate settling of these disputes is crucial, as they ultimately determine the numbers and types of neurons produced, with consequent outcomes on physiology and behavior. Recent papers have relied on gene expression patterns and fate mapping, to develop more sophisticated maps of the developing CNS (Bedont et al., 2015; Puelles et al., 2013; Puelles and Rubenstein, 2015; Shimogori et al., 2010; Swanson, 2012; Thompson et al., 2014). Despite significant progress, these recent reviews acknowledge that current models are imperfect and that further evaluation using modern molecular tools is imperative.

Recently, much work has focused on the progenitor zone that gives rise to dopamine (DA) neurons in the midbrain, with a goal of deriving DA neurons from stem cells, for modeling and ultimately treating Parkinson's disease (Arenas et al., 2015; Blaess and Ang, 2015; Smidt and Burbach, 2007; Smits et al., 2006; Studer, 2012). In the developing CNS, DA neurons arise from the mesodiencephalic floor plate (mdFP), here defined by the co-expression of Shh, Foxa2 and Lmx1a rostral to the midbrain-hindbrain boundary (MHB), due to the concerted action of Shh, Fgf and Wnt signaling pathways (Arenas et al., 2015; Blaess and Ang, 2015; Joksimovic and Awatramani, 2014; Lahti et al., 2012; Luo and Huang, 2016; Smits et al., 2006; Wurst and Prakash, 2014). In terms of boundaries delimiting the DA progenitor zone, several studies have demonstrated that the caudal limit of DA neuron production is at the MHB defined by the caudal limit of Otx2 expression (Brodski et al., 2003; Joyner et al., 2000; Simeone et al., 2002). Forced ectopic expression of Otx2 in the hindbrain floor plate (FP) results in DA neuron production in the hindbrain (Ono et al., 2007), whereas loss of Otx1/2 results in expanded hindbrain at the expense of midbrain (Omodei et al., 2008; Puelles et al., 2004; Simeone et al., 2002). Along the dorsoventral (D-V) axis, several studies have demonstrated that the Shh⁺/Foxa2⁺ domain is subdivided into at least two main domains defined by Lmx1a/b and Nkx6-1/Sim1/Neurog1 (Andersson et al., 2006; Blaess et al., 2011; Hayes et al., 2011; Joksimovic et al., 2009; Nakatani et al., 2010; Nouri et al., 2015; Prakash et al., 2009). The Lmx1a/b domain is the origin of most DA neurons, whereas the Nkx6-1/Sim1/Neurog1 domain is the source of Pou4f1⁺ red nucleus (RN) neurons and other smaller populations. Knockouts of Lmx1a/b results in fewer DA neurons and ectopic Pou4f1⁺ neurons from the DA domain (Deng et al., 2011; Yan et al., 2011), whereas forced ectopic expression of Lmx1a/b result in DA neuron production from the Nkx6-1/Sim1/Neurog1 domain at the expense of RN neurons (Anderegg et al., 2013; Andersson et al., 2006; Lin et al., 2009; Nakatani et al., 2010). Conversely, forced ectopic expression of Sim1 results in the production of Pou4f1⁺ neurons from the Lmx1a/b domain, at the expense of DA neurons (Nakatani et al., 2010). Thus, many studies have characterized the caudal and dorsal boundaries of the DA progenitor zone. In contrast, rostrally, it is not even clear whether there is a boundary in the strict sense, i.e. a lineage-restricted compartment (Kiecker and Lumsden, 2005), or whether DA neuron production merely tapers off, due to lack of fibroblast growth factors or other diffusible signals.

Is the mdFP subdivided into compartments along the anteroposterior (A-P) axis? In the early embryo, differentiating Th⁺ DA neurons are observed in the midbrain with a caudal limit at the MHB. The anterior limit of this Th⁺ DA cohort is approximately near the base of the zona limitans intrathalamica (ZLI), the boundary between presumptive developmental segmental units, prosomeres 2 and 3 (p2 and p3) (Joksimovic et al., 2009; Nouri et al., 2015; Puelles and Rubenstein, 2015). Because these neurons appear to be generated from the midbrain as well as presumptive p1, p2 and p3, some have called these mesodiencephalic DA neurons (Puelles and Rubenstein, 2015; Veenvliet and Smidt, 2014), as opposed to midbrain DA neurons (Arenas et al., 2015; Blaess and Ang, 2015). Interestingly, most of the currently known DA progenitor markers, including Shh, Wnt1, Otx2, Neurog2, Lmx1a/b, Foxa1/2 and Msx1, among others, are expressed rostrally, well beyond the ZLI, into presumptive p3 of the diencephalon (Alvarez-Bolado et al., 2012; Andersson et al., 2006; Ellisor et al., 2012; Lahti et al., 2012; Nouri et al., 2015; Puelles and Rubenstein, 2015). Why DA neurons are not produced from the entire A-P axis of the Shh⁺/Foxa2⁺/Lmx1a⁺ mdFP, despite the presence of key signaling molecules and transcription factors, is a conundrum. One possibility is that the mdFP is subdivided along the A-P axis into lineage-restricted microdomains. Such a possibility is indeed depicted in prosomere-based models of the segmentally divided embryonic CNS (Björklund and Dunnett, 2007; Puelles and Rubenstein, 2015; Rubenstein et al., 1994; Smidt and Burbach, 2007; Smits et al., 2006, 2013), but evidence for such microdomains within the Shh⁺/Foxa2⁺/Lmx1a⁺ mdFP is limited. These models of the subdivided mdFP are based on prosomeric gene expression patterns in more dorsal CNS structures, and these genes are often not expressed in the mdFP. Thus, it remains open whether the mdFP is a compartmentalized structure. In this regard, studies in more caudal regions of the CNS, such as the hindbrain, demonstrate that the FP does not have segmental divisions both in terms of gene expression and cell mixing capability (Fraser et al., 1990; Seitanidou et al., 1997; Wilkinson et al., 1989). In contrast, the neighboring hindbrain neuroepithelium is the archetype of the segmentally subdivided CNS (Kiecker and Lumsden, 2005).

To test whether the Shh⁺/Foxa2⁺/Lmx1a⁺ mdFP is subdivided into microdomains, we examined transcription factors that could potentially segregate this region along the A-P axis. Engrailed 1 (En1) is known to be expressed in the midbrain beginning at about embryonic day (E) 8.0 (Chi et al., 2003; Davis and Joyner, 1988) and encompasses the majority of the DA primordium, excluding a small number of Th⁺ neurons at the rostral-most extent (Lahti et al., 2012). Engrailed 1 and 2 (En1/2) are important for DA neuron generation and survival (Alberi et al., 2004; Alvarez-Fischer et al., 2011; Alves dos Santos and Smidt, 2011; Ellisor et al., 2012; Nordström et al., 2015; Rekaik et al., 2015; Simon et al., 2001; Veenvliet et al., 2013). Moreover, engrailed proteins are involved in various boundary formation systems such as the diencephalic-dorsal midbrain boundary (Matsunaga et al., 2000; Scholpp et al., 2003) and a D-V boundary in the limb apical ectodermal ridge (Kimmel et al., 2000). In contrast, developing brain homeobox 1 (Dbx1) expression is excluded from the ventral midbrain, but is expressed in the caudal ventral diencephalon at this stage (Causeret et al., 2011; Shoji et al., 1996; Sokolowski et al., 2016). In the spinal cord, Dbx1 has been shown to repress En1 indirectly (Karaz et al., 2016; Pierani et al., 2001). Taking these previous studies into consideration, we sought to examine these two transcription factors in more detail. We reveal two precisely established and abutting microdomains within the mdFP, their diverse neuronal descendants, and a sufficient role for En1 in expanding DA neuron production along the A-P axis.

To define the A-P extent of the mdFP and its derivatives, we examined Foxa2 expression in sagittal sections of Shh::Cre, RC::NZG embryos, in which nuclear βgal is activated in the Shh domain (in figures this reporter is referred to as R.NZG). Consistent with the literature and our previous observations (Alvarez-Bolado et al., 2012; Echelard et al., 1993; Nouri et al., 2015; Puelles and Rubenstein, 2015), βgal⁺/Foxa2⁺ cells are observed in the midbrain, and well rostral to the ZLI, in the diencephalon (Fig. 1A-B′). Rostral to the βgal⁺/Foxa2⁺ region, βgal⁺/Foxa2⁻ cells are observed. The Shh domain therefore encompasses the Foxa2 domain, but also extends significantly more anteriorly than the Foxa2 domain.

At midbrain levels, Shh/Foxa2 derivatives have been previously described, and include DA and RN populations (Blaess et al., 2011; Hayes et al., 2011; Joksimovic et al., 2009). To identify mdFP derivatives rostral to the midbrain, we examined coronal sections through the rostral midbrain/caudal diencephalon. Foxa2-labeled cells are observed in two streams apparently migrating tangentially from the ventricular zone – one mediolaterally and the other rostrally (Fig. 1C). These Foxa2⁺ cells are located rostral to Th⁺ DA neuron populations. Shh fate maps combined with Foxa2 immunolabeling reveal YFP⁺/Foxa2⁺ cells in the subthalamic (STN), supramammillary (SuM), posterior hypothalamic (PH) and premammillary (PM) nuclei, suggesting that these neuronal populations are derived from the mdFP (Fig. 1D-K″).

We next examined the A-P extent of Foxa2 and Lmx1a expression in relation to DA neuron production from the mdFP (Fig. 2A-C‴). In mid-sagittal sections of E9.5-E12.5 embryos, Foxa2 and Lmx1a are co-expressed at the midline; Th⁺ DA neurons are observed in the midbrain, but are excluded from more rostral portions of the Foxa2⁺/Lmx1a⁺ domain. This observation led us to hypothesize that the mdFP is subdivided along the A-P axis into two microdomains – one giving rise to DA neurons, and a second giving rise to non-DA neurons. To investigate this hypothesis, we searched for genes in the literature, particularly transcription factors expressed early in development that could subdivide the mdFP in a meaningful manner towards early CNS patterning. Consistent with previous reports (Davis and Joyner, 1988; Simon et al., 2001), we found the transcription factor En1 expressed in the early ventral midbrain (Fig. 2D). In contrast, the transcription factor Dbx1 was expressed in the ventral diencephalon, but appeared to be excluded from the ventral midbrain (Fig. 2D′) (Causeret et al., 2011; Shoji et al., 1996). These factors appear to subdivide the mdFP along the A-P axis at E9, although at later stages a gap between these two domains is observed, probably owing to the dynamic expression of these factors (Fig. S1).

To correlate these two progenitor domains with possible postmitotic neuron markers, we examined the expression of two markers, Pitx3 and Pitx2 (Martin et al., 2004; Smidt et al., 1997). Our analysis revealed adjacent, non-overlapping populations of Pitx3⁺ DA and Pitx2⁺ non-DA neurons (Fig. 2E,E′). We also observed Pitx2⁺ expression within more dorsal midbrain regions representing non-DA populations (Martin et al., 2004; Skidmore et al., 2008). Thus, early En1 expression appears to align with the majority of the Pitx3⁺ DA domain. In contrast, the adjacent Dbx1 microdomain corresponds to a region containing non-DA, Pitx2⁺ neurons.

En1 and Dbx1 appear to mark non-overlapping regions of the mdFP. We next investigated (1) whether the En1 and Dbx1 microdomains in this specific region abut, forming a novel floor plate boundary; (2) whether there is any overlap of En1 and Dbx1 expression within progenitor cells of the mdFP at any time during early development; and (3) whether, if such overlapping cells are identified, they represent specific and novel subsets of DA neurons. To answer these questions, we designed a combined intersectional and dual fate-mapping approach to label En1 and Dbx1 microdomains, and any overlapping regions within the same embryo. We used two recombinase drivers, Dbx1::Cre (Fig. S2D,D′) (Bielle et al., 2005) and En1::Dre (Plummer et al., 2016) in combination with two separate reporter alleles RC::RLTG (Rosa-CAG-rox-FRT-loxP-tdTomato-eGFP) and RC::NZG (Rosa-CAG-loxP-PGKNeo-FRT-nlsLacZ-eGFP) (Fig. 3A) (Plummer et al., 2015; Yamamoto et al., 2009). In such embryos, the intersectional reporter RC::RLTG produces expression of tdTomato after En1::Dre recombination, and eGFP after intersectional En1::Dre and Dbx1::Cre recombination events; RC::NZG expresses nls-lacZ after Dbx1::Cre recombination events. This approach allowed us to examine at improved resolution, the establishment and juxtaposition of the En1 and Dbx1 microdomains.

We analyzed mid-sagittal E9.5 and E11.5 sections of En1::Dre, Dbx1::Cre, RC::RLTG, RC::NZG embryos. Along the A-P axis of the mdFP, tdTomato⁺ cells and βgal⁺ cells form abutting microdomains with limited intermingling (Fig. 3B,C). These results demonstrate that a progenitor boundary is formed between the En1 and Dbx1 microdomains of the mdFP. Very few intersectionally labeled GFP⁺ cells were observed near this boundary (Fig. 3B,D; Fig. S2A). Thus, although a boundary is formed, it does not appear to involve initial overlap of En1 and Dbx1 transcription factors, unlike some other boundaries (Kiecker and Lumsden, 2005).

We next correlated Th expression with differentially labeled cells in these embryos. Most Th⁺ neurons were tdTomato⁺, and further, these were aligned with the En1⁺ progenitor microdomain, which extends up to the ZLI (Fig. 3D,D′; Fig. S3). Interestingly, a small subset of Th⁺ neurons were found outside the tdTomato-labeled microdomain, consistent with a previous study (Lahti et al., 2012) (Fig. 3D′, thick arrow). These Th⁺ cells were not GFP⁺, and appeared to be βgal⁺, suggesting that they were derived from the Dbx1 domain (addressed later in this article). Taken together, these data support a model wherein the En1 microdomain of the mdFP mainly demarcates the rostral limit of DA neuron production (barring a small cohort of Th⁺ neurons), and the Dbx1 microdomain delineates the boundary of neighboring non-DA neurons.

Other embryonic regions were also GFP⁺, including the dorsal midbrain, hindbrain neurons located in rhombomere 1, and the spinal cord (Fig. S2A-C) (Pierani et al., 2001; Shoji et al., 1996). Indeed, GFP⁺ fibers extending from the dorsal midbrain or possibly other regions are detected in the caudal ventral midbrain (Fig. 3C,D,D″; Fig. S2A), similar to other studies reporting fibers in this region originating from the dorsal midbrain (Inamata and Shirasaki, 2014). In the spinal cord, Dbx1 and En1 expression along the D-V axis has been shown to be highly dynamic, yet partly and transiently overlapping in 5-10% of cells (Pierani et al., 1999, 2001). Our spinal cord results demonstrate that even in the case of transiently overlapping gene expression domains, our intersectional method is capable of detecting large numbers of overlapping cells, and at significantly greater sensitivity than traditional methods (Fig. S2B,C). Overall, we conclude that within the mdFP, the En1 and Dbx1 microdomains abut each other forming a boundary, and further, when probed at intersectional resolution, there is minimal developmental overlap.

To fate map the En1⁺ and Dbx1⁺ microdomains, we separately analyzed postnatal day (P) 28 En1::Cre, RC::NZG or Dbx1::Cre, RC::NZG animals. Analysis of En1::Cre, RC::NZG midbrain sections revealed Th⁺/Foxa2⁺ substantia nigra pars compacta (SNc), ventral tegmental area (VTA), retrorubral field (RRF) and caudal linear (CLi) neurons co-labeled with βgal (Fig. 4B,E; SNc=99.3%± 0.4 s.e.m., VTA=98.7%±0.2 s.e.m., RRF=99.7%±0.3 s.e.m., CLi=99.1%±0.8 s.e.m.). Th⁻/Foxa2⁺ neurons located in the rostral linear (RLi) region were also co-labeled with βgal (not shown). βgal was also observed in Pou4f1⁺/Foxa2⁺ cells of the RN in En1::Cre, RC::NZG fate maps (Fig. S3B-C′). Indeed, in sagittal embryonic sections, βgal⁺/Foxa2⁺/Pou4f1⁺ cells were observed throughout the midbrain with a rostral limit near the ZLI, similar to Th⁺ DA neurons (Fig. S3A-B‴).

In Dbx1::Cre, RC::NZG fate maps, βgal did not colocalize with Th⁺/Foxa2⁺ DA neurons of the VTA, SNc, RRF or CLi (Fig. 4C,F; data not shown). Also, cells in the RLi region were βgal⁻ (data not shown). βgal⁺ cells also did not co-label with Pou4f1⁺/Foxa2⁺ RN neurons (data not shown). Finally, βgal⁺/Th⁻ cells were also observed in the vicinity of DA neurons in both En1 and Dbx1 fate maps (Fig. 4B,C,E,F, arrowheads). As these cells are Foxa2⁻, they could potentially represent interneurons that have migrated from more dorsal regions and/or the hindbrain (Lahti et al., 2016). In summary, En1⁺ but not Dbx1⁺ mdFP derivatives include DA, RN and RLi neurons. The few Dbx1 microdomain-derived Th⁺ neurons observed in Fig. 3D,D′, do not contribute to the SNc and VTA, and probably contribute to scattered Th⁺ neurons in the hypothalamic region.

En1 and Dbx1 single fate-mapping analyses were also used to determine the origin of neighboring hypothalamic populations. The glutamatergic STN is a hypothalamic population just rostral to the SNc, which has been shown to express Pitx2 (Fig. 5A-C) (Martin et al., 2004; Skidmore et al., 2008, 2012). Interestingly, we determined that the STN maintains expression of the floor plate markers Foxa2 and Lmx1a, but is negative for βgal in En1::Cre, RC::NZG fate maps (Fig. 5A,D,F). In contrast, Dbx1::Cre, RC::NZG fate maps show βgal colocalized with all Pitx2⁺, Foxa2⁺ and Lmx1a⁺ cells of the STN and parasubthalamic nucleus (PSTN) (Fig. 5A,C,E,G; data not shown for PSTN). Additionally, the STN and PSTN also express Nr4a2, an important marker of FP-derived DA neurons (data not shown).

Further analysis revealed robust expression of the mdFP markers Foxa2 and Lmx1a in more ventral medial nuclei of the PH, SuM and PM regions. The Foxa2⁺ and Lmx1a⁺ PH/PM neurons co-labeled with βgal in Dbx1::Cre, RC::NZG [Fig. 5A,I,K shows the ventral premammillary (PMv) region; PH and SuM not shown], but not in En1::Cre, RC::NZG fate maps (Fig. 5A,H,J). The Foxa2⁻/Lmx1a⁻ mammillary and lateral hypothalamic area were also labeled with βgal in Dbx1::Cre, RC::NZG, but not in En1::Cre, RC::NZG fate maps (data not shown). In summary, our fate maps show that two clear and non-overlapping microdomains defined by En1 and Dbx1 exist within the mdFP, and these give rise to distinct neuronal populations.

Through separate studies in our laboratory, we identified an Slc6a3⁺ population in the PMv region. To explore the possibility that this unique population expressing DA markers was derived from the mdFP, we analyzed Slc6a3::Cre, Ai9 P30 brains. Our experiments revealed tdTomato⁺ neurons (Fig. 6A,B) co-labeled with the markers Foxa2, Lmx1a and Nr4a2, indicative of their mdFP origin (Fig. 6C). Very few tdTomato⁺ neurons were co-labeled with Th antibodies (Fig. 6C; zero to two neurons per section). Allen Brain Atlas (ABA) characterizations also revealed robust transcript expression of Slc6a3, Ddc and Slc18a2 in the PMv. Although we were able to detect robust Ddc immunolabeling, we were unable to detect Slc6a3 and Slc18a2 protein in PMv cells (Fig. S5). Taken together, these analyses reveal for the first time a Dbx1⁺ mdFP-derived neuronal population that is positive for many key DA pathway markers, although some are detectable only at the transcript level.

We next sought to examine if En1 expression was sufficient to generate DA neurons from the Dbx1 microdomain. We therefore generated conditional En1 overexpressor (En1OE) mice, also designed to co-express nuclear eGFP (Fig. 7A). We inserted a single copy of the transgene using ΦC31 integrase and attPx3 docking sites at the Rosa26 locus (Tasic et al., 2011). We first validated conditional expression of both En1 and GFP in Dbx1::Cre, En1OE mutants at E10.5 (Fig. 7B-E). By this embryonic stage in controls, En1 expression appears retracted and is mainly detectable in the caudal midbrain and rostral hindbrain (Fig. 7E, arrowhead). In mutants, ectopic En1⁺ cells were observed in the Dbx1⁺ mdFP, and they also expressed eGFP (Fig. 7C, arrow), whereas En1 expression in the midbrain and hindbrain appeared normal in mutants compared with control samples (Fig. 7C-E, arrowheads). In mutant embryos, ectopic En1 appeared of comparable or slightly lower intensity to En1 expression near the MHB (Fig. 7D, arrow versus arrowhead). In addition, other CNS regions expressing Dbx1, including the dorsal midbrain and hindbrain, also co-expressed eGFP and En1 (Fig. 7B). These results suggest that our overexpressor line expresses En1 and eGFP, and that ectopic En1 levels are within a physiological range.

Next, we compared Th expression in Dbx1::Cre, En1OE mutants with both Dbx1::Cre, RC::NZG and En1::Cre, RC::NZG controls at E12.5. Our initial findings reveal that in controls, a few Th⁺ neurons are observed outside the En1 and within the Dbx1 microdomain. In En1OE mutants, there appeared to be a significant rostral expansion of Th⁺ neurons within the Foxa2 domain (Fig. 7F-I, brackets). We then analyzed Th expression at E14.5, the time point after normal DA neuron production is complete. We used coronal sections to visualize cell distributions and their boundaries better. A small number of low Th⁺/Lmx1a⁺-expressing neurons can be detected in the Dbx1⁺ mdFP region of controls, but these do not contribute to the SNc and VTA (Fig. 3; Fig. 4; Fig. 7H,I; Fig. 8A,A′). Quantitative analysis revealed a large increase of Th⁺/Lmx1a⁺ neurons in Dbx1::Cre, En1OE mutants in comparison with controls (Fig. 8A,A′,C,C′,G). Similar analysis for the highly specific DA marker Pitx3 revealed a highly significant increase of Pitx3⁺ neurons (Fig. 8E,F,H) derived from the Dbx1⁺ mdFP. We also observed ectopic Ddc and Slc18a2 immunopositive cells in this region, but very few of these expressed detectable Slc6a3 (Fig. 8B,D; Fig. S4). However, Slc6a3 is not a pan-DA marker, and it is possible the ectopic DA neurons have a molecular profile resembling a DA subset containing undetectable levels of Slc6a3 (Lammel et al., 2008).

Owing to the non-viability of postnatal Dbx1::Cre, En1OE mutants, we could only continue our analysis until E18.5/P0 to determine where ectopic DA neurons settle. Our analysis in Dbx1::Cre, RC::NZG controls revealed few Th⁺/βgal⁺ neurons, with no Pitx3 expression in the caudal hypothalamus (Fig. 9A-C, arrow). Some Th⁺ axons from more caudal DA neurons projecting through this region were observed (Fig. 9B,C). Comparable sections in Dbx1::Cre, En1OE mutants revealed a significant number of ectopic Pitx3⁺/Th⁺ neurons (Fig. 9E,F, arrows), and many Pitx3⁺/Th⁻ neurons (Fig. 9F, arrowhead). These neurons are not clustered, but rather are found scattered in the mammillary and caudal hypothalamic regions (Fig. 9D-F). Our experiments reveal that extending early En1 expression into more rostral progenitors of the Dbx1 microdomain results in significant numbers of ectopic DA neurons.

We next determined the effects of En1 overexpression on caudal hypothalamic nuclei derived from the Dbx1⁺ mdFP. In the STN/PSTN of Dbx1::Cre, En1OE mutants, we observed a ∼75% reduction in the number of Pitx2⁺/Foxa2⁺ neurons (Fig. 10A-E) compared with controls. Likewise, in the PM region of Dbx1::Cre, En1OE mutants, we observed an ∼85% loss of Foxa2⁺ neurons in comparison with controls (Fig. 10F-J). Very few, if any, Pitx2⁺ cells were observed in mutants. Thus, in En1OE mutants, ectopic DA neurons appear to be generated, in part, at the expense of STN and PM neurons.

Previous studies have demonstrated that the mdFP is the source of many neuron types, unlike the caudal FP, which is largely non-neurogenic. How mdFP progenitors are apportioned along the A-P axis has not been clearly elucidated. Here, we reveal (1) two microdomains, defined by En1 and Dbx1, which subdivide early FP progenitors along the A-P axis; (2) that these microdomains abut each other, with no developmental overlap and that boundary establishment is remarkably precise; (3) that an En1⁺ microdomain gives rise to DA, RLi and RN neurons and that a Dbx1⁺ microdomain gives rise to STN, PH, PM and SuM neurons; (4) that a subpopulation of mdFP-derived neurons express several DA-related markers within the PMv; (5) that forced expression of En1 in the Dbx1 microdomain results in ectopic DA neurons at the expense of significant numbers of STN and PM neurons (Fig. 11). These studies are a step forward in the long-standing quest to elucidate the progenitor domains of the embryonic CNS. Furthermore, these studies will also serve as a foundation for optimizing FP-based protocols to derive DA neurons from stem cells.

Morphogens and homeodomain (HD) transcription factors are essential for the patterning and segmentation of the developing CNS (Dasen and Jessell, 2009; Kiecker and Lumsden, 2005). It has been proposed that via induction of transcription factors, morphogens define the regionalization of the CNS into distinct populations of progenitor cells with partially overlapping or ‘fuzzy’ expression boundaries that are later sharpened through cross-repressive interactions (Briscoe and Small, 2015; Kiecker and Lumsden, 2005; Prochiantz and Di Nardo, 2015). In contrast to these models of cross-repressive transcription factor interactions, here we probed boundary formation with intersectional resolution to show that this boundary forms in a remarkably precise manner, with minimal developmental overlap between the two microdomains. How this boundary forms with this remarkable level of precision remains to be determined, but one possibility could be through morphogen-induced cross-inhibitory positive-feedback mechanisms (Srinivasan et al., 2014). Differential cell-adhesion has also been invoked as a mechanism for boundary formation, and could play a role (Kiecker and Lumsden, 2005). Finally, the presence of physical barriers (e.g. extracellular matrix) could also contribute to the precision of boundary formation (Kiecker and Lumsden, 2005). Further experiments will be needed to elucidate the exact mechanism underpinning the precise formation of this boundary.

Our study reveals a subdivision of the floor plate along the A-P axis into two juxtaposed microdomains, giving rise to neighboring neuronal lineages. Based on our own analysis and results from other groups, it is clear that a multitude of early factors, which are required for DA progenitor specification, are expressed on either side of this boundary along the A-P axis. Thus, in a sense, we view these neighboring cell populations and their descendent lineages as developmental ‘cousins’. We were able to identify several neuronal populations originating from either side of this boundary. The En1 microdomain gave rise to the VTA, SNc and RLi and non-dopaminergic RN neurons. The Dbx1 microdomain did not give rise to DA neurons in the SNc and VTA, and instead gave rise to tangentially migrating neurons that contribute to the STN, PH and PM. Our data thus provides an account of mdFP descendants and refines previous models (Björklund and Dunnett, 2007; Bodea and Blaess, 2015; Puelles and Rubenstein, 2015; Smits et al., 2006; Veenvliet and Smidt, 2014) in that we demonstrate that SNc and VTA DA neuron populations are not derived from the domain rostral to the ZLI (presumptive Dbx1⁺/Foxa2⁺ in p3 domain).

Our study identified mdFP-derived neuronal populations in the PMv that express key DA transcription factors and DA synthesis markers. PMv neurons have significant enrichment of ribosome-associated mRNAs for Slc6a3, Th and Ddc (Soden et al., 2016). However, levels of Th were 10% of those in VTA neurons. Previous studies have reported these neurons to be catecholamine immunoreactive in some circumstances (Zoli et al., 1993). Our study shows that this population of neurons is derived from the mdFP, expresses key transcription factors including Foxa2, Lmx1a and Nr4a2 and yet are still distinct from DA neurons in that PMv neurons are derived from the Dbx1 microdomain, do not express En1 or Pitx3, and only few of these cells express Th (Figs 5, 6). The lack of En1 or Pitx3 could in part account for the low levels of Th observed in this population, despite the presence of Nr4a2 and Foxa2, known drivers of Th expression (Kadkhodaei et al., 2009; Kim et al., 2003; Pristerà et al., 2015; Stott et al., 2013; Yi et al., 2014). In accordance with this, En1 and Pitx3 have been shown to be upstream of Th, and Pitx3 has been shown to potentiate the action of Nr4a2 on the Th promoter by releasing SMRT/HDAC-mediated repression (Jacobs et al., 2009; Veenvliet et al., 2013). Although this population expresses low levels of Th, there still remains the possibility that PMv neurons might have some DA-synthesis capacity under specific environmental or physiological conditions.

Embryonic loss of En1 leads to an early reduction of rostral DA neurons. Interestingly, gene expression profiling of these mutants reveals the upregulation of the marker Pitx2 (Veenvliet et al., 2013). Utilizing a conditional overexpressor mouse model, our data reveals that En1 is sufficient to expand DA production into the Dbx1⁺ mdFP, partly at the expense of STN and PM neurons. In this regard, to our knowledge this is the first study to produce an anterior expansion of Pitx3⁺/Th⁺ DA neurons beyond their normal expression domain. Ectopic DA neurons do not settle in the STN or PH/PM probably because they do not have the appropriate molecular profile for migration into these regions. Although the expansion of DA neurons is robust, clearly there are still some STN (∼25%) and PM (∼15%) neurons produced. Thus, a key question arises – why is the entire Dbx1⁺ mdFP not reprogrammed to generate Th⁺/Pitx3⁺ DA neurons in our En1OE model? Multiple possible explanations exist, such as the lower concentration, or absence, of Fgf8 in more rostral mdFP regions, due to its distance from the MHB source. Another possibility is the lack of a complete repertoire of transcription factors in subregions of the Dbx1 domain, or the presence of repressive factors in this region that preclude DA specification, and cannot be overridden by En1 alone. Finally, it is possible that our transgenic overexpression of En1 is at modest levels during early specification and is insufficient for complete reprogramming. Many studies on boundaries use electroporation, whereby higher levels can be achieved to overexpress transcription factors (Dasen et al., 2003; Matsunaga et al., 2000). When single-copy Rosa26-targeted transgenic approaches have been used, the results are sometimes more moderate. In the case of the MHB, when a transgenic approach was used to overexpress the key transcription factor Otx2 in r1, the MHB boundary shift was only moderate and transient (Omodei et al., 2008). Thus, it is possible that in our En1OE model, modest expression levels achieved in the early embryo were not sufficient for complete reprogramming of the Dbx1⁺ mdFP to DA neurons.

Recent advances in the generation of DA neurons from stem cells have shown promising results towards potential cell replacement therapies and modeling studies for Parkinson's disease (Cooper et al., 2012; Grealish et al., 2014; Kirkeby et al., 2012; Kriks et al., 2011; Mazzulli et al., 2016; Xi et al., 2012). Key protocols designed for the conversion of induced pluripotent stem cells (iPSCs) into DA neurons were based on previous developmental studies showing DA neurons arise from FP progenitors, and use FOXA2 and LMX1A to optimize DA neuron derivation protocols (Kirkeby et al., 2012; Kriks et al., 2011; Xi et al., 2012). However, immunolabeling studies on cultured and grafted DA neurons show that many neurons in the graft are FOXA2⁺ but not TH⁺, probably representing a non-DA fate. Gene expression analysis of these cultures showed robust upregulation of DBX1 and PITX2, in addition to DA markers (Kriks et al., 2011; Xia et al., 2016), indicative of imperfect programming. As these protocols were designed to maximize FOXA2⁺/LMX1A⁺ progenitor cells, our work would predict that both DBX1 and EN1 progenitors were generated, consequently resulting in the differentiation of DA and non-DA neurons. Specifically detailing the exact microdomains within the mdFP in vivo presents an exciting advance towards designing protocols for more homogeneous and efficient production of DA neurons from iPSCs. Achieving large fractions of authentic DA progenitors will open up the possibility of generating and characterizing molecularly distinct DA subtypes (Anderegg et al., 2015; Poulin et al., 2014), an important goal for understanding selective DA neuron vulnerability.

Mice were maintained and sacrificed according to protocols approved by the Northwestern University Animal Care and Use Committee (IACUC). Strains include En1::Cre (Kimmel et al., 2000), En1::Dre (Plummer et al., 2016), Dbx1::Cre (Bielle et al., 2005), Shh::Cre (Harfe et al., 2004), Slc6a3::Cre (Bäckman et al., 2006), RC::NZG (Yamamoto et al., 2009), RC::RLTG (Plummer et al., 2015), RC::YFP (Srinivas et al., 2001) and Ai9 (Madisen et al., 2010). For simultaneous dual and intersectional fate mapping, En1::Dre, RC::NZG males were bred to Dbx1::Cre, RC::RLTG females. Male mice could not be used carrying both Dbx1::Cre and reporter alleles, due to the early activation of Dbx1::Cre. In the case of En1::Dre, only occasional embryos displayed germline recombination, and these were not further analyzed.

To generate the En1OE transgene, plasmid Ai9 (Addgene, 22799) was subcloned into pBT378 (Addgene, 52554). A plasmid pCS2-TAG (Addgene, 26772) was amplified for the 2A-H2B-eGFP sequence and mouse En1ORF plasmid (GE-Dharmacon, OMM5895-202525739) was amplified without a stop codon. Utilizing an In-Fusion HD Cloning Kit (Clontech, 638916) linearized pBT378/Ai9, En1ORF and 2A-H2B-eGFP were fused to obtain the final construct (5′-attB-CAG-loxP-STOP-loxP-En1ORF-2A-H2B-eGFP-WPRE-bGHpA-attB-3′ with a pBT378 backbone). The open reading frame was validated by sequencing.

Transgenic TARGATT Rosa26(attPx3) FVB males (Tasic et al., 2011), were crossed to wild-type C57BL/6 females. Full circular En1OE plasmid containing attB/attB sites and ΦC31 integrase mRNA was co-injected into the FVB/B6 fertilized eggs. One male founder from a total of 27 animals screened was identified for integration of the En1OE transgene into one Rosa26(attPx3) locus using primers Rosa10, 11, 425, 436 and 522 (Tasic et al., 2011), in addition to primers for WPRE and eGFP.

mRNA in situ hybridization was performed as previously described (Caronia-Brown et al., 2016). For immunolabeling, tissue sections were processed as previously described (Nouri et al., 2015). Antibodies used were guinea-pig anti-Lmx1a (Yong Chao-Ma, Northwestern University, Evanston, IL, USA; 1:20,000), rabbit anti-Lmx1a (Millipore, AB10533; 1:1000), goat anti-Foxa2 (Santa Cruz, SC6554; 1:50), mouse anti-Pou4f1 (Santa Cruz, sc-31984; 1:50), rabbit and sheep anti-Th (Pel-Freeze, P40101-0/P60101-150; 1:500), mouse anti-Th (Sigma, T2928; 1:1000), mouse anti-En1 (DSHB, 4G11; 1:50), chicken anti-GFP (Abcam, ab13970; 1:1500 or 1:20,000), goat anti-βgal (Biogenesis, 4600-1409; 1:1500), rabbit anti-βgal (Cappel, 55976; 1:1500), rabbit anti-Nr4a2 (Santa Cruz, SC-991; 1:50), rabbit anti-Pitx2 (Capra Science, PA 1020-100; 1:1000), rabbit anti-Pitx3 (Invitrogen/Zymed, 38-2850; 1:200), rat anti-Slc6a3 (Santa Cruz, sc-32258; 1:50), rabbit anti-Ddc (Abcam, ab3905; 1:500) and rabbit anti-Slc18a2 (EMD Millipore, AB1598P; 1:500).

Quantification is shown as mean±s.e.m., using Student's two-sample t-test, significant differences were taken at P≤0.05. All cells were counted on every third 20 µm coronal section at E14.5 and E18.5/P0 or counted on every sixth 20 µm coronal section (Th⁺ and βgal⁺/Th⁺ total) at 4 weeks postnatally in En1::Cre, RC::NZG fate maps (n=3).

All images were acquired using Leica DM5000B epifluorescence, Leica TCS SP5 confocal or Nikon A1R(A) Spectral confocal microscopes. Adobe Photoshop CC 2015 was utilized for the overlay of pseudocolored single-channel images in the same sections for some figures, and to adjust brightness and contrast for all figures shown.

While this paper was in revision, an interesting manuscript relevant to this topic was published (Kee et al., 2017).

We thank Dr Patricia Jensen for providing En1::Dre and RC::RLTG mice, Dr Yong Chao-Ma for Lmx1a antibodies, Nigel P. Anderson for technical help, Jean-Francois Poulin for sharing findings, and critiques on the manuscript. En1OE mice were generated by the TTML core. Imaging was performed at the Center for Advanced Microscopy (NCI-CCSG-P30-CA060553).