Differentiation and molecular heterogeneity of inhibitory and excitatory neurons associated with midbrain dopaminergic nuclei

Midbrain dopaminergic (DA) and dorsal raphe serotonergic (5-HT) neurons regulate responses to reward and punishment, associative learning, mood, and coordination of voluntary movements. Function of these monoaminergic pathways is controlled by inhibitory and excitatory neurons, including important populations residing locally within the monoaminergic nuclei as well as in the adjacent midbrain and anterior hindbrain (Russo and Nestler, 2013; Lammel et al., 2014; Proulx et al., 2014; Morello and Partanen, 2015). Dysfunction of monoaminergic circuits is thought to lead to a multitude of psychiatric and neurophysiological disorders, such as schizophrenia, depression, addiction and Parkinson's disease. Understanding these complex disorders requires knowledge of the various components of these networks. However, compared with monoaminergic neurons themselves, very little is known about the development and heterogeneity of their regulators.

Inhibitory signals to DA neurons, located in the ventral tegmental area (VTA), substantia nigra pars compacta (SNpc) and retrorubral field (RRF), arrive from neighbouring GABAergic neurons in the VTA, substantia nigra pars reticulata (SNpr) and rostromedial tegmental nucleus (RMTg, also called the tail of VTA) (Perrotti et al., 2005; Jhou et al., 2009a,,b; Kaufling et al., 2009; Zhou and Lee, 2011; Cohen et al., 2012; Lammel et al., 2012; Margolis et al., 2012; Yetnikoff et al., 2015). Collectively, these GABAergic neurons are referred to as dopaminergic neuron-associated GABAergic (D-GABA) neurons (Lahti et al., 2013). It has been suggested that the RMTg integrates signals from different parts of the brain, including the lateral habenula (LHb), converting them into inhibitory inputs to both DA and 5-HT networks (Lavezzi and Zahm, 2011; Barrot et al., 2012; Bourdy and Barrot, 2012; Proulx et al., 2014; Sego et al., 2014).

Excitatory glutamatergic inputs to DA neurons come from a number of brain regions, including the laterodorsal tegmental nucleus (LDTg) in the anterior hindbrain (Geisler et al., 2007; Lammel et al., 2012). Recent studies have also identified glutamatergic neurons embedded in the DA nuclei themselves and these may provide local excitatory neurotransmission (Morales and Root, 2014). Moreover, the interpeduncular nucleus (IPN) below the DA and RMTg nuclei can relay excitatory input to monoaminergic neurons, in particular the 5-HT system (Groenewegen et al., 1986; Proulx et al., 2014; Antolin-Fontes et al., 2015).

The GABAergic and glutamatergic neurons within the DA nuclei also have targets other than the DA neurons. GABAergic SNpr is the main output source of the basal ganglia and projects to brain regions important for activation of voluntary movements. Furthermore, some VTA GABAergic and glutamatergic neurons project to the forebrain and regulate pathways involved in associative learning (Fields et al., 2007; Brown et al., 2012; Kabanova et al., 2015).

Our previous work has shown that the majority of D-GABA neurons are born outside the midbrain in the anterior rhombomere 1 (r1) (Achim et al., 2012). The r1 generates multiple neuronal subtypes in different dorsoventral domains characterized by specific combinations of transcription factors (Waite et al., 2012; Lahti et al., 2013). Expression of Nkx6-1 in the ventricular zone and Gata2 and Tal1 (also known as Scl) in the corresponding mantle zone was suggested to specify the ventrolateral r1 domain contributing to D-GABA neurons (Achim et al., 2012). Because of molecular similarity to the spinal cord V2 domain, we call this region rhombencephalic V2 (rV2) (previously referred to as the GABA I domain; Lahti et al., 2013). Of the rV2 factors, Gata2 is dispensable for development of D-GABA neurons, whereas inactivation of Tal1 leads to a specific loss of the majority of GABAergic neurons in the SNpr (Kala et al., 2009; Achim et al., 2012,, 2013). However, these studies left open the exact role of Tal1 as well as its co-factors and downstream effectors during neuronal differentiation in rV2.

Here, we characterized the diversity of neurons generated in the ventral r1. We show that, in addition to GABAergic neurons of the SNpr, VTA and RMTg, the rV2 domain gives rise to glutamatergic neurons in the SNpc, VTA, IPN and LDTg. Thus, our findings reveal that these neurons, all components of the DA and 5-HT regulatory circuits, share a common developmental history. Importantly, the balance between rV2 inhibitory and excitatory neurons is controlled by Tal1, which cooperates with both Gata2 and Gata3. Although these factors are activated independently of each other, together they promote GABAergic and suppress glutamatergic identity as postmitotic selector genes. Furthermore, we discovered a set of transcription factors, which are activated downstream of Gata/Tal selectors, and are expressed differentially in D-GABA neuron subgroups, providing novel molecular markers for them and exposing heterogeneity within and between these nuclei.

Embryonic r1 generates several different subsets of GABAergic neuron precursors expressing specific combinations of transcription factors. Diversity and contribution of these precursors to brain nuclei are poorly understood. We aimed to characterize the GABAergic subtypes generated in the ventral r1, focusing on the identification of D-GABAergic precursors. As we had previously shown that the peak of D-GABA precursor generation is at embryonic day (E) 11.5-12.5 (Achim et al., 2012), after which time they start to migrate towards the midbrain, we chose the E12.5 stage for the analysis.

As shown previously (Achim et al., 2012), a subset of postmitotic GABAergic precursors in the ventrolateral rV2 domain expressed Tal1 (Fig. 1A,B). In addition, the same region expressed the functionally related factors Tal2, Gata2 and Gata3 (Fig. 1C,E). In our separate experiments, gene expression profiling of Gata2 mutant midbrain revealed putative Gata/Tal factor targets, including Gata co-factors Zfpm1 and Zfpm2 (also known as Fog1 and Fog2, respectively) as well as the transcription factors FoxP1, Sox14 and Six3 (K. Achim, L.L., M.S. and J.P., unpublished). Of these, Sox14, Zfpm1, Zfpm2 and FoxP1 were also expressed in the rV2 domain (Fig. 1D,F,G). In addition, the rV2 domain expressed Sox2 and En1 (Fig. 1G, Fig. S1B,G). These 11 transcription factors were co-expressed also at the cellular level (Fig. 1E-G; Fig. S1F-M). However, their patterns were complex and divided the GABAergic precursors in the rV2 domain into subgroups. More specifically, the mantle zone (mz) could be divided apicobasally into two domains: the more apical mz1 expressed all the factors, whereas more basal mz2 lacked Gata2, Tal2 and Zfpm1. As Gata2, Tal2 and Zfpm1 are mostly not detected in the midbrain and hindbrain neurons later in embryogenesis (Kala et al., 2009; Achim et al., 2013) (data not shown), their expression patterns might reflect dynamic gene expression changes during neuronal maturation, suggesting that postmitotic precursors in mz1 are only beginning their differentiation process. Moreover, the GABAergic precursors within mz1 and mz2 were heterogeneous. For example, in mz1, Zfpm2 colocalized with Zfpm1 and Gata3, and in mz2 with Gata3 or En1 in a subset of cells (Fig. 1F; Fig. S1F-Ka). In mz2, approximately half of the Sox2⁺ cells co-expressed FoxP1 (Fig. 1G,G‴). Furthermore, the majority of these Sox2⁺FoxP1⁺ cells expressed the transcription factor Sox14 (Fig. S1L-Ma).

In addition, subsets of GABAergic precursors in the ventral r1 express Pitx2 and Pax7 (Aroca et al., 2006; Waite et al., 2012). Pax7⁺ precursors appeared to originate from the more dorsal r1, where the ventricular zone progenitors also expressed Pax7 (Fig. S1E-Eb). This is also supported by gene expression and fate-mapping data from avian embryos (Lorente-Canovas et al., 2012). We found that some Pax7⁺ precursors co-expressed Zfpm2 but no other Gata- or Tal-related factors (Fig. S1E; data not shown). Thus, the Zfpm2⁺ cells in the rV2 domain can be divided into Zfpm2⁺Pax7⁺ and Zfpm2⁺Pax7⁻ subgroups (see above). Pitx2⁺ cells were located in the most basal part of mz2 throughout the entire r1, and did not express Gata or Tal factors or their putative targets (Fig. S1C; data not shown).

We previously showed that conditional knockout (cko) of Tal1 in the mouse leads to a loss of GABAergic neurons in the SNpr (Achim et al., 2012). In order to elucidate which of the r1 GABAergic precursor subgroups contributed to this loss of D-GABA nuclei in Tal1^cko mutants, we aimed to identify Tal1-dependent precursors. Consistent with our earlier report (Achim et al., 2012), Gad1 expression in the rV2 domain of Tal1^cko mice was downregulated in mz1 and partly also in mz2 (Fig. 1A′). In mz2, the remaining GABAergic cells represented mostly Pitx2⁺ and Pax7⁺ subtypes, which appeared to be unaffected in mutants (Fig. S1C′,E′). In Tal1^cko mz1, Tal2 and Zfpm1 were downregulated completely (Fig. 1C′,F′). Although we found that Gata2 and Gata3 were activated independently of Tal1, in Tal1^cko mice the number of Gata3⁺ cells was nevertheless significantly lower (Fig. 1E′,E‴). In addition, a significant fraction of Zfpm2⁺Pax7⁻, Sox2⁺, FoxP1⁺ and Sox2⁺FoxP1⁺ cells was absent in mutants, indicating that these subtypes are largely Tal1 dependent (Fig. 1F′,F‴,G′,G‴). Sox14 was still expressed in Tal1^cko mice, although the number of Sox14-positive cells appeared to be lower in the lateral mz2 (Fig. 1D′).

In the haematopoietic system, Tal1 works in a complex with the Gata factors Gata1 and Gata2 (Wadman et al., 1997). As Gata2 and Gata3 are also expressed in the rV2 domain, we investigated whether they could cooperate with Tal1 and whether the residual GABAergic cells in Tal1^cko r1 could be Gata2 or Gata3 dependent. In single Gata2^cko and Gata3^cko mutants, all GABAergic markers appeared to be unaffected (Fig. S2). This supports our earlier observations of intact D-GABA neurons in E18.5 Gata2^cko and Gata3^cko midbrain (Kala et al., 2009; Achim et al., 2012,, 2013). By contrast, when both genes were inactivated together (Gata2^cko;Gata3^cko) the phenotype resembled that of the Tal1^cko mutant (Fig. 1A″-G‴; Fig. S1B″-E″). Although Tal1 itself was activated independently of Gata2 and Gata3, the maintenance of its expression appeared to fail, as Tal1 expression was downregulated in the mz2 region of the Gata2^cko;Gata3^cko mutants (Fig. 1B″). Compared with Tal1^cko r1, the number of Zfpm2⁺, Sox2⁺ and FoxP1⁺ cells in Gata2^cko;Gata3^cko r1 was reduced even further (Fig. 1F′-G‴). Again, Pitx2⁺ and Pax7⁺ cells remained unaffected, demonstrating that these subtypes represent GABAergic precursors that are independent of Tal1, Gata2 and Gata3 regulation (Fig. S1C″,E″).

Taken together, our results demonstrate that the embryonic rV2 domain contains a variety of GABAergic precursors (Fig. 1A‴), a large fraction of which redundantly require Gata2 or Gata3. Most, but not all, of these Gata2/3-dependent precursors are also Tal1 dependent and consist of Tal2⁺, Gata3⁺, Zfpm1⁺, Zfpm2⁺, En1⁺, Sox2⁺, FoxP1⁺ and Sox14⁺ cells, which express these transcription factors in different combinations. In addition, the Gata2/3-independent GABAergic neurons in the lateral mz2, mostly Pitx2⁺or Pax7⁺, appear to represent developmentally distinct subgroups of r1 GABAergic neurons.

Given the similarity of Gata2^cko;Gata3^cko and Tal1^cko mutants, we investigated whether Tal1 can affect the function of Gata factors. In addition to GABAergic precursors, Gata2 and Gata3 are expressed in 5-HT precursors located ventrally to the rV2 region and are needed for the 5-HT phenotype (Craven et al., 2004; Pattyn et al., 2004; Kala et al., 2009) (our unpublished results). We speculated that the neuronal subtype identities regulated by Gata factors – GABAergic or 5-HT, depending on the region – could be influenced by Tal1, which is largely absent from 5-HT precursors. Thus, introducing Tal1 in the Gata2/3 complex might turn on a different set of downstream genes.

To investigate this, we electroporated Hamburger–Hamilton stage (HH) 14-16 chicken ventral r1 with an Nkx2-2-mTal1 construct. This vector drives expression of mouse Tal1 under the Nkx2-2 enhancer in the serotonergic domain. Indeed, 48 h after the electroporation, mouse Tal1 overexpression resulted in the appearance of chicken Gad1 and Tal1 transcripts and reduction of the 5-HT markers 5-hydroxytryptophan and cLmx1b in the Nkx2-2 domain (Fig. 2A-D′; data not shown). Expression of cGad1 was induced specifically in cells electroporated with the Nkx2-2-mTal1 vector (Fig. 2E). Taken together, the results from gain-of-function experiments and conditional mutagenesis suggest that the presence of Tal1, Gata2 and Gata3 together is both necessary and sufficient to induce GABAergic identity in ventral r1.

We then investigated whether loss of Tal1 from rV2 would redirect GABAergic precursors into a 5-HT phenotype. No ectopic expression of 5-HT markers was detected in the Tal1^cko mice (Fig. S1N,N′). Instead, the precursors are redirected to an alternative rV2 region-specific phenotype (see below).

After discovering Tal1 and Gata2/3 -dependent GABAergic precursor subtypes and their markers in the embryonic r1, we investigated whether we could detect them later within hindbrain-derived D-GABAergic nuclei. For this, we utilized Gbx2^CreERT2;R26^TdTomato embryos, in which r1-derived tissue expressed red fluorescent protein (RFP) allowing us to follow migration of D-GABAergic precursors to ventral midbrain (Fig. 3A,G,L,Q; Fig. S3). At E18.5, all Gad1-expressing D-GABA nuclei expressed Tal1 (Fig. 3A-C), but showed differences in Sox14, Zfpm2 and Six3 expression. The RFP-negative anterior SNpr (aSNpr), identifiable by Six3 expression (Madrigal et al., 2015), lacked Zfpm2 and Sox14 expression (Fig. 3D-K). Indeed, in the rV2-domain at E12.5 Six3 was not detected (Fig. S1D-D″), consistent with a different origin of aSNpr. By contrast, the RFP-labelled and Six3-negative posterior SNpr (pSNpr) expressed Sox14 in its laterodorsal region and Zfpm2 throughout (Fig. 3D-F,L-P). The few Six3-expressing cells among Zfpm2⁺ cells in pSNpr were RFP negative and not found in the more posterior regions (Fig. 3N; data not shown). In the VTA, the lateral regions contained few Zfpm2⁺ cells detectable by immunohistochemistry (IHC) but not by in situ hybridization (ISH) (Fig. 3O; Fig. S4B). However, the parabrachial pigmented nucleus (PBP) of the medial VTA contained few Six3-expressing GABAergic cells, which were also RFP negative (Fig. 3L-N). Posteriorly, the RFP-labelled RMTg lacked both Zfpm2 and Six3 transcripts but expressed Sox14 strongly (Fig. 3D-F,Q-U).

Thus, these data indicate that Six3, Zfpm2 and Sox14 expression characterize different D-GABA nuclei, with Zfpm2-expressing neurons contributing mostly to pSNpr and Sox14-expressing neurons to RMTg. The Six3 transcripts were found in the anterior SNpr and PBP, which appear to comprise developmentally and molecularly distinct subnuclei.

As a subset of Zfpm2⁺ cells in the E12.5 r1 co-expressed Gata3 and En1, we investigated whether these subtypes could be found in the pSNpr. Indeed, at E18.5 this nucleus contained Zfpm2⁺ cells, some of which co-expressed Gata3 and En1 (Fig. S4A-Fa). The most basal pSNpr consisted of a group of Zfpm2⁺ cells, which largely lacked both Gata3 and En1. In turn, pSNpr contained also Gata3⁺ and En1⁺ cells lacking Zfpm2, as well as Gata3⁺ cells expressing En1 transcripts (Fig. S4Ca,Fa,G-Ia). In the adult pSNpr, the expression of these factors was still detectable (data not shown).

Next, we investigated whether the difference between the amount of Zfpm2⁺ and Gata3⁺ cells in E12.5 Tal1^cko and Gata2^cko;Gata3^cko r1 was reflected in pSNpr at E18.5 (Fig. 4). Indeed, similar to the E12.5 r1, Tal1^cko mutant pSNpr still contained some Zfpm2⁺ and Gata3⁺ GABAergic cells in the most lateral domain, whereas in Gata2^cko;Gata3^cko mutants the Zfpm2⁺ cells were completely absent (Fig. 4A-A″,C-C″,D,D′). By contrast, again correlating with the E12.5 phenotype, both mutants lacked all GABAergic neurons expressing En1 (Fig. 4B-B″). In Gata2^cko and Gata3^cko single mutants, the expression of these markers in pSNpr appeared to be unaffected (data not shown).

Taken together, the pSNpr contains both Tal1-dependent neurons expressing combinations of En1, Gata3 and Zfpm2, as well as Tal1-independent subtypes expressing Zfpm2 and Gata3. All of these appear to be Gata2/3 dependent (Fig. 4D″).

As we had detected Sox14, Sox2 and FoxP1 co-expressing cells in E12.5 r1, we investigated whether Sox14⁺ RMTg neurons expressed all these factors. Indeed, all Sox14, FoxP1 and Sox2 were present in the same RMTg cells at E18.5 (Fig. 5A-I), and appeared to be specific to this subgroup of GABAergic neurons (Fig. 5M-Oa; data not shown). The Sox14- and FoxP1-negative Sox2⁺ cells turned out to be Olig2⁺ (Fig. 5J-L), indicating a glial identity (Hoffmann et al., 2014). Furthermore, the expression of Sox14, FoxP1 and Sox2 colocalized with Gbx2^CreERT2-labelled RFP⁺ cells supporting the origin of RMTg in the r1 (Fig. 5P-Ra).

In the adult brain, FoxP1⁺ cells were found between the caudal VTA and IPN, in the area corresponding to the Sox14⁺Sox2⁺FoxP1⁺ nucleus in perinatal embryos, but they lacked both Sox14 transcripts and Sox2 (data not shown). We wanted to confirm that these FoxP1⁺ cells corresponded to the RMTg. This nucleus can be currently identified by two main methods: retro- or anterograde labelling from the VTA/SNpc or LHb, respectively, or methamphetamine-induced expression of the protein encoded by the immediate early gene FosB (Perrotti et al., 2005; Jhou et al., 2009a,,b; Kaufling et al., 2009). For anterograde labelling of the RMTg, we injected mice with AAV2-GFP into the LHb (Fig. 6A). In these mice, GFP-positive projections descending from the LHb made highly specific contacts with the GABAergic FoxP1⁺ nucleus (Fig. 6B,C). Next, we administered methamphetamine to wild-type mice and compared the expression of FosB and FoxP1 in the RMTg. Whereas in NaCl-treated control animals the FoxP1⁺ cells were negative for FosB, in metamphetamine-treated animals we detected co-expression of FoxP1 and FosB in numerous cells in this region (Fig. 6D-K). These results together indicate that the FoxP1⁺ neuronal population corresponds to the RMTg.

As FoxP1, Sox2 and Sox14 transcripts were downregulated in the Tal1^cko and Gata2^cko;Gata3^cko r1 at E12.5, we wanted to investigate whether the RMTg was affected in these mutants. Indeed, in the area corresponding to the RMTg in wild-type embryos, no Gad1-expressing nucleus containing Sox2, FoxP1 or Sox14-expressing cells was found in mutants (Fig. 7). The RMTg of Gata2^cko and Gata3^cko single mutants appeared normal (Fig. S5).

In summary, our results suggest that development of the RMTg requires Tal and Gata factors. Moreover, Sox14, Sox2 and FoxP1 can be used as molecular markers for the RMTg during embryogenesis, and FoxP1 also in the adult brain.

The transcription factor profile of rV2 domain resembles the spinal cord p2/V2 domain, which generates not only GABAergic, but also Vsx2⁺ (also known as Chx10) glutamatergic neurons from Nkx6-1⁺ progenitors (Arber, 2012). To elucidate whether the rV2 domain produces glutamatergic neurons, we analysed Nkx6-1 and Vsx2 expression at E12.5. As reported previously (Waite et al., 2012), Nkx6-1 was detected in all rV2 progenitors, and in a subset of postmitotic precursors (Fig. 8A,C′). The majority of these Nkx6-1⁺ precursors co-expressed Vsx2 (Fig. 8B-C′). During development, these precursors started to separate into Nkx6-1⁺, Vsx2⁺, and Nkx6-1⁺Vsx2⁺ subpopulations, upregulated a glutamatergic marker Slc17a6 (also known as Vglut2), and some of them migrated anteriorly towards the midbrain (Fig. S6A-I; data not shown). In E16.5 and E18.5 midbrain, we found these three different subtypes sparsely dispersed in the VTA and SNpr (Fig. 8D-Fa). They were labelled in Gbx2^CreERT2;R26^TdTomato embryos and expressed Slc17a6, indicating their origin in r1 and glutamatergic neuronal identity, respectively (Fig. 8G; Fig. S6J-M).

Nkx6-1⁺ cells have been previously described in the rostral part of the avian IPN (Lorente-Canovas et al., 2012). Consistent with this, numerous glutamatergic Nkx6-1⁺Vsx2⁺ and Nkx6-1⁺ cells born in the r1 were discovered in the IPN (Fig. 8H-K; Fig. S6N-Nb). These cells formed two distinct layers. The top layer, corresponding to IPN rostral (IPNr) and IPN lateral (IPNl) subnuclei, contained mostly Nkx6-1⁺Vsx2⁺ cells and few Nkx6-1 single-positive cells (Fig. 8H-J). The underlying IPN central (IPNc) subnucleus contained Nkx6-1⁺Vsx2⁺ cells that co-expressed Sox14 (Fig. 8O,Oa). Thus, in the r1, Sox14 transcripts also appear to mark specific glutamatergic neuron subgroups, which might explain why Sox14 is still expressed in E12.5 Tal1^cko and Gata2^cko;Gata3^cko mutants when the other D-GABAergic markers are downregulated. The IPNc contained also r1-derived Pax7⁺, Zfpm2⁺, and Pax7⁺Zfpm2⁺ GABAergic neurons, which in turn lacked Nkx6-1 and Sox14 expression (Fig. 8L,N,Na,P-Qa). In the adult IPN, Nkx6-1 and Vsx2 were still expressed (Fig. 8M), as were Pax7, Zfpm2 and Sox14, although the latter was drastically downregulated (data not shown).

The LDTg contains GABAergic, glutamatergic and cholinergic neurons (Wang and Morales, 2009). At E16.5 and E18.5, we discovered Nkx6-1⁺, Vsx2⁺, and Nkx6-1⁺Vsx2⁺ glutamatergic neurons in an area lateral to the dorsal raphe (DR), corresponding to the LDTg (Fig. 8R-Ua). These neurons formed two groups: a dorsal Nkx6-1⁺ and a ventral Vsx2⁺ population (Fig. 8S,T). A small group of double-positive neurons resided between these two nuclei (Fig. 8Ta). All these neurons were labelled with Gbx2^CreERT2, indicating their origin in the r1 (Fig. S6O,Oa). In the adult, we found Vsx2⁺ and Nkx6-1⁺ cells near choline acetyltransferase (ChAT)⁺ cholinergic LDTg neurons (Fig. S6P-Ua). However, Nkx6-1⁺ neurons lacked ChAT but expressed Slc17a6, indicating their glutamatergic identity (Fig. S6V,Va).

Thus, in addition to GABAergic neurons, rV2 generates several subtypes of glutamatergic neurons, which contribute to nuclei both in the midbrain and anterior hindbrain. The transcription factors Vsx2 and Nkx6-1 could be used to identify some of these different subpopulations.

Because Tal1 and Gata2/3 factors were required for the GABAergic identity in ventral r1, we investigated whether in the same region they also suppressed the glutamatergic identity. Indeed, when we analysed Nkx6-1⁺, Vsx2⁺, and Vsx2⁺Nkx6-1⁺ precursors in E12.5 Tal1^cko and Gata2^cko;Gata3^cko embryos, their numbers were significantly increased in both mz1 and mz2 (Fig. 9A-A″,C). At this stage, Slc17a6 was expressed in the wild-type rV2 domain at a very low level (Fig. 9B). Although the low expression level made quantification impossible, in both mutants it appeared slightly upregulated (Fig. 9B′,B″). Gata2^cko and Gata3^cko single mutants showed no apparent changes either in the number of Nkx6-1⁺ and Vsx2⁺ precursors, or in the expression of Slc17a6 (Fig. S7).

The upregulation of Nkx6-1 and Vsx2 expression in the mutant rV2 domain might represent a mere temporary transcriptional misregulation, rather than a true cell fate change. If the GABAergic neurons were truly switching to glutamatergic identity, we would expect to find more glutamatergic neurons in the areas where these neurons normally reside, including the IPN, SN, VTA and LDTg. Indeed, in E18.5 Tal1^cko and Gata2^cko;Gata3^cko mutants IPNr, IPNl, SNpr and LDTg all contained more Nkx6-1⁺, Vsx2⁺, and Nkx6-1⁺Vsx2⁺ neurons (Fig. 10A-A‴,C-D‴). No significant difference was observed in the VTA (Fig. 10B-B‴).

Taken together, the loss of GABAergic precursors in the absence of Tal1 and Gata2/3 is paralleled by an increased amount of glutamatergic precursors, which later populate specific glutamatergic nuclei in the midbrain and anterior hindbrain.

Recent studies have highlighted local GABAergic and glutamatergic neurons in the control of the DA and 5-HT pathways (Fig. 11B) (Zhou and Lee, 2011; Barrot et al., 2012; Lammel et al., 2014; Morales and Root, 2014). Our current results demonstrate that these diverse GABAergic and glutamatergic neurons originate in a specific region of ventrolateral r1 and reveal a molecular framework controlling their differentiation and heterogeneity.

Terminal selector transcription factors establish subtype identities of differentiating neurons and their expression is activated at the cell-cycle exit (Deneris and Hobert, 2014). Our earlier studies demonstrated that Gata2 and Tal2 act as post-mitotic selectors during development of midbrain and diencephalic GABAergic neurons (Kala et al., 2009; Virolainen et al., 2012; Achim et al., 2013). The midbrain-derived GABAergic precursors contribute to neuronal populations in the superior and inferior colliculi, periaqueductal grey and midbrain reticular formation. Although these midbrain precursors also express Tal1, they do not require it for their differentiation (Achim et al., 2013). Here, we show that the Nkx6-1⁺ rV2 domain in ventral r1 generates a variety of D-GABAergic and glutamatergic cell types expressing subtype-specific transcription factors (Fig. 11A). In contrast to the midbrain GABAergic neurons, Tal1 operates as a selector driving differentiation of D-GABAergic neurons. Without Tal1, postmitotic neuronal precursors in the rV2 mostly fail to activate the gene expression typical for D-GABA subpopulations and assume glutamatergic identities instead. Interestingly, this fate change appears to be complete enough to allow the superfluous glutamatergic precursors to migrate to the glutamatergic subregions of the IPN, SN and LDTg.

During haematopoietic cell differentiation, Tal1 works in a complex with Gata factors (Wadman et al., 1997). Development of D-GABA neurons was not affected by inactivation of Gata2 or Gata3 alone. By contrast, GABAergic precursors in rV2 adopted a glutamatergic identity in Gata2^cko;Gata3^cko double mutants. Although the phenotypes of Tal1^cko and Gata2^cko;Gata3^cko embryos resembled each other, they also displayed clear differences, the Gata2^cko;Gata3^cko mutants showing a more complete loss of D-GABA neurons and their precursors. Thus, these selector factors might have both shared and unique targets. Redundancy between Gata2 and Gata3 is not observed in the midbrain or many other GABAergic populations. This could be explained by tissue-specific regulatory mechanisms of Gata2 and Gata3 genes. Unlike in the midbrain, where Gata2 is required for Gata3 and Tal1 expression (Kala et al., 2009; Achim et al., 2013), in the r1 the three selector genes Gata2, Gata3 and Tal1 are all activated independently of each other. Indeed, compared with the midbrain, distinct enhancer elements have been shown to drive Gata2 expression in the r1 (Zhou et al., 2000). Gata2 and Gata3 are also important for the development of serotonergic neurons in the r1 (Craven et al., 2004; Pattyn et al., 2004; Kala et al., 2009). The Gata-regulated neuronal phenotypes, GABAergic and serotonergic, appear to result from the presence or absence of Gata co-factors, especially Tal1.

Development and molecular heterogeneity of D-GABA neurons are poorly understood. Our results show that the different D-GABA nuclei express unique molecular markers, which correlate with their embryonic origins and anatomical locations (Fig. 11A). In the rV2, these markers are activated in a Gata/Tal-dependent fashion soon after the cell cycle exit. Cells expressing Zfpm2, Sox14, En1 and Gata3 contribute to the neuronal subtypes present in the pSNpr. These subtypes also differ in their requirements for Tal1 and Gata factors, as pSNpr contains a small population of Tal1-independent neurons, which are still Gata2/3 dependent. Thus, differentiation of the D-GABA neuron subtypes might be defined by slight changes in the composition of terminal selector complexes. Compared with the pSNpr, aSNpr neurons appear to be developmentally and molecularly distinct. The aSNpr expresses Six3, which is absent in rV2. Moreover, this nucleus remains unlabelled by the Gbx2^CreERT2-line, supporting its origins outside the hindbrain (Achim et al., 2012). Indeed, recent fate-mapping and gene expression studies suggested that the Six3⁺ aSNpr might originate in the Nkx6-2⁺ domain of the ventrolateral midbrain (Madrigal et al., 2015). Although our earlier results indicated that aSNpr is Gata2 dependent (Achim et al., 2012), its dependency on Gata3 and Tal1 factors remains to be investigated.

Similar to the SNpr, VTA GABAergic neurons also appear to be heterogeneous. GABAergic neurons in the lateral VTA expressing Zfpm2 are of r1 origin and are lost in Tal1^cko mutants. By contrast, our results indicate that, similar to aSNpr, a small subset of GABAergic cells in the PBP originate outside r1 and express Six3, suggesting an additional level of heterogeneity between the subregions of the VTA. As up to 40% VTA might consist of GABAergic neurons (Lammel et al., 2014), it is likely that in addition to a few Zfpm2⁺ and Six3⁺ neurons it contains additional, for the time being uncharacterized, GABAergic subtypes. It will be of great interest to study whether the molecularly distinct SNpr and VTA GABAergic neurons are involved in different neuronal circuitries and have specific functions.

Our results suggest that the third D-GABA neuron subgroup, the RMTg, is derived from Sox2⁺FoxP1⁺Sox14⁺ precursors in rV2. Furthermore, FoxP1 expression also continues in the adult RMTg, providing a useful marker of this important cell group. Based on its gene expression and dependence on Tal1, the RMTg appears to be a relatively homogeneous cell population compared with the SNpr and VTA.

The glutamatergic neurons in the LDTg and IPN also send inputs to the DA and 5-HT nuclei (Fig. 11B). Interestingly, our results show that these excitatory neurons are developmentally related to inhibitory D-GABAergic neurons. They also originate in the rV2 domain and in the Tal1^cko and Gata2^cko;Gata3^cko mutants, in which GABAergic differentiation fails, the neuronal precursors are redirected towards the glutamatergic LDTg/IPN/SN identity. Thus, in rV2, the Gata/Tal selectors control a balance in development of the inhibitory and excitatory neurons regulating the monoaminergic systems. Our results also reveal molecular heterogeneity among the glutamatergic neurons in the IPN and LDTg. For example, the Nkx6-1⁺Vsx2⁺ IPN glutamatergic neurons can be divided into Sox14-expressing IPNc neurons and Sox14-negative IPNr/IPNl neurons. In the LDTg, Nkx6-1 and Vsx2 label primarily two different cell populations.

Recently, glutamatergic tyrosine hydroxylase (TH)-negative neurons have also been demonstrated in the VTA, SNpc/SNpr and RRF (Morales and Root, 2014; Kabanova et al., 2015). We show here that some of them are positive for Nkx6-1 and Vsx2, and their numbers in the SNpr are increased in the Tal1^cko and Gata2^cko;Gata3^cko mutants. By contrast, only a few Nkx6-1- or Vsx2-expressing neurons were found in the VTA. Thus, the VTA glutamatergic neurons might constitute a distinct cell group. However, similarly to the D-GABAergic neurons, at least some of the glutamatergic neurons within the DA nuclei appear to be derived from rV2 and differ in their transcription factor profiles from the other midbrain glutamatergic or dopaminergic neurons.

Regulation of the monoaminergic systems is a key to the control of mood, motivation and voluntary movements. Recent studies have described important functions for the brainstem GABAergic and glutamatergic neurons, both in the local regulation of the monoaminergic pathways and as projection neurons affecting forebrain targets. Our study uncovers the developmental origins, molecular mechanisms of differentiation and heterogeneity of these neurons. This will allow more detailed studies of their composition, connectivity and functions.

The En1^Cre (Kimmel et al., 2000), Gbx2^CreERT2 (Chen et al., 2009), Tal1^flox (Bradley et al., 2006), Gata2^flox (Haugas et al., 2010), Gata3^flox (Grote et al., 2008) and R26^TdTtomato (Madisen et al., 2010) alleles were maintained in an outbred ICR or C57BL/6 background (R26^TdTtomato) and intercrossed to generate En1^Cre/+;Tal1^flox/flox (Tal1^cko), En1^Cre/+;Gata2^flox/flox (Gata2^cko), En1^Cre/+;Gata3^flox/flox (Gata3^cko), En1^Cre/+;Gata2^flox/flox;Gata3^flox/flox (Gata2^cko;Gata3^cko) and Gbx2^CreERT2/+;R26^TdTtomato/+ embryos. Control embryos, littermates of Tal1^cko and Gata2^cko;Gata3^cko mutants, were negative for Cre and heterozygous or wild type for the flox alleles and displayed a phenotype similar to wild-type (ICR) embryos. E0.5 was defined as noon of the day of the vaginal plug.

Tamoxifen (T5648, Sigma-Aldrich) was dissolved in fresh corn oil (C8267, Sigma-Aldrich) at 20 mg/ml before oral administration at 0.1 mg/g of body weight to pregnant R26^TdTomato females at noon of E8.5.

The animal experiments were approved by the National Committee of Experimental Animal Research in Finland and carried out according to the National Institutes of Health (NIH) guidelines. AAV work was approved by the National Board for Gene Technology in Finland.

Embryos were fixed in 4% paraformaldehyde (PFA; Sigma-Aldrich P6148) in PBS. For adult brains, mice were intracardially perfused first with PBS and then with 4% PFA, and postfixed in 4% PFA. All samples were dehydrated and embedded in Histosec polymer wax (Merck Millipore) using a Leica tissue processor ASP300 and sectioned at 5-6 µm.

Non-radioactive mRNA ISH was carried out as described (Wilkinson and Green, 1990), with modifications (Lahti et al., 2012). Fluorescent ISH signal was visualized with anti-DIG-POD (1:1000, Roche) and TSA labelling kits (1:100, FITC or Cy3 dye; PerkinElmer). The DIG-labelled RNA probes were transcribed from plasmids previously described (Jukkola et al., 2006; Kala et al., 2009; Achim et al., 2013). Additional probes were mouse Sox14 (IMAGp998A2414391Q), Six3 (IMAGE 5719986) and cTal1 (a gift from David Rowitch, University of California, San Francisco, CA, USA).

IHC was performed as described (Jukkola et al., 2006) using the primary antibodies listed in Table S1. All secondary antibodies were Alexa Fluor conjugated (1:400, Invitrogen) and nuclei were visualized with DAPI (4′-6′-diamidino-2-phenylindole; Sigma-Aldrich).

C57BL/6JOlaHsd female mice (16 weeks old) were anaesthetized with isoflurane, attached to the stereotaxic frame (Stoelting, Wood Dale, IL, USA), and small holes were drilled into the skull. Each animal received a unilateral 0.5 µl injection at a speed of 0.5 µl/min using a microinjector (Stoelting, Wood Dale, IL, USA) and microsyringe (NanoFil 33 G, World Precision Instruments) of the ssAAV2-eGFP viral vector (8.1×10¹² vg/ml, UNC vector core, Chapel Hill, NC, USA) to the following coordinates: AP −1.5, DV −2.8, ML −1.2, at a 10° angle. The coordinates were obtained and empirically refined from the mouse brain atlas (Paxinos and Franklin, 2012). Two weeks after the injections, the mice were intracardially perfused and the brains were collected.

Male ICR mice (16 weeks old) received two injections of either methamphetamine hydrochloride (Sigma-Aldrich; 10 mg/kg of body weight, s.c., n=4) or corresponding volume of vehicle (0.9% NaCl, s.c., n=2), at 2-h intervals. The mice were housed individually after the first injection. Two hours after the second injection, the mice were intracardially perfused and the brains were collected.

Anterior ventral hindbrain of HH14-16 embryos were electroporated with a plasmid containing GFP cDNA under the Nkx2-2 enhancer (Nkx2-2-GFP; a gift from Johan Ericson, Karolinska Institute, Stockholm, Sweden) with or without a vector in which mouse Tal1 cDNA was placed under the Nkx2-2 enhancer (Nkx2-2-mTal1). Electroporated embryos were incubated at 38°C for 48 h before collection for analysis (n=7 in both groups).

Micrographs were taken using an Olympus BX63 connected to a DP72 camera, and processed with Adobe Photoshop CS6. Brightness, contrast and sharpness were adjusted to recapitulate the original samples as they were seen through the microscope. For each experiment, a minimum of three embryos representing each stage and genotype were analysed. For cell quantification, a minimum of three sections from each area analysed were counted per embryo, except for E12.5 embryos for which the analysis comprised the entire r1. The results were analysed using Student's t-test (*P<0.05, **P<0.01, ***P<0.001).

We thank Eija Koivunen, Outi Kostia, Eeva Partanen and Laura Lopez-Blanch for expert technical assistance; and Francesca Morello, Anna Kirjavainen, Esa Korpi and Tapio Heino for discussions and comments.