Sox9 and nuclear factor I transcription factors regulate the timing of neurogenesis and ependymal maturation in dopamine progenitors

Dopamine, a neurotransmitter of the brain, is pivotal in orchestrating motor behaviours, cognition, memory and reward mechanisms. Dopamine transmission is carried out by a diverse population of midbrain dopamine (mDA) neurons, which exhibit distinct innervation patterns and functions (Yaghmaeian Salmani et al., 2024; Poulin et al., 2020; Garritsen et al., 2023; Azcorra et al., 2023).

How mDA neurons are generated during embryonic development has attracted significant attention, mainly due to their importance in Parkinson's disease (PD). This disease is characterised by the degeneration of mDA neurons, resulting in motor complications. Molecular understanding of mDA neuron development has enabled the generation of stem cell-derived mDA neuron precursors in vitro. Several ongoing clinical trials are testing transplants of these in vitro-generated cells, aiming at restoring DA signalling in individuals with PD (Kirkeby et al., 2023; Piao et al., 2021; Doi et al., 2020; Barker and Björklund, 2023).

Another – currently hypothetical – approach to replace mDA neurons lost in PD is the stimulation of endogenous brain repair. For instance, the loss of dopamine-mediated feedback inhibition to neighbouring astroependymal cells in salamanders can induce cell cycle re-entry, leading to regeneration of mDA neurons (Berg et al., 2011). Such ability appears to be absent in the adult mammalian midbrain, which does not harbour neural stem cells (NSCs), unlike the subventricular zone (SVZ) and the hippocampus, where NSCs reside in a state of non-permanent cell cycle arrest (quiescence) (Urbán et al., 2019).

Adult NSCs are formed during embryogenesis, when some neural progenitors are specified to enter quiescence and form the stem cell reserve (Fuentealba et al., 2015; Furutachi et al., 2015; Morales and Mira, 2019). After generating neurons and glia, the remaining progenitors exit the cell cycle and differentiate into multi-ciliated ependymal cells. These cells line the ventricles and the spinal cord central canal, and contribute to the maintenance of cerebrospinal fluid homeostasis as well as to neuroblast migration (Ji et al., 2022). In the spinal cord, ependymal cells can proliferate in vivo and contribute to glial scar formation after injury (Meletis et al., 2008; Barnabé-Heider et al., 2010; Sabelström et al., 2013; Ren et al., 2017), but in the brain the regenerative capacity of the ependymal cells remains unclear (Chiasson et al., 1999; Spassky et al., 2005; Zhang et al., 2007; Carlén et al., 2009; Shah et al., 2018). After producing mDA neurons between embryonic day (E) 10 and E14 (Bayer et al., 1995), also mDA progenitors in the midbrain become ependymal cells. However, these cells appear to retain the expression of the mDA progenitor marker Lmx1a even in the adult (Hedlund et al., 2016). This raises the question of whether the ependymal cells in different brain regions have dissimilar properties and functions.

Understanding the induction of mitotic exit and the maintenance of G0 in neural progenitors is important for several reasons. First, dysregulation of the cell cycle exit of radial glial cells leads to an imbalance in the types and amounts of cells generated, which can give rise to various neurodevelopmental disorders, such as autism (Kaushik and Zarbalis, 2016), schizophrenia (Mao et al., 2009) and ADHD (Dark et al., 2018). In addition, incomplete maturation of postmitotic radial glia into ependymal cells may result in hydrocephalus (Jiménez et al., 2001; Park et al., 2016; Hou et al., 2023). Second, the failure to maintain quiescence in the adult NSCs will lead to the depletion of the stem cell pool (Van Velthoven and Rando, 2019) and can contribute to neurodegenerative disorders (Joseph et al., 2020). Indeed, a substantial effort has been made to identify the signalling pathways and transcription factors responsible for the maintenance of stem cell quiescence (Blasco-Camarro and Fariñas, 2023). Although many of them are shared between the embryonic and adult NSCs, some serve somewhat different or even opposing functions (Urbán and Guillemot, 2014). For example, transcription factors belonging to nuclear factor I (Nfi) family promote neurogenesis in the embryonic telencephalon but inhibit it to maintain NSCs in the adult (Harris et al., 2015).

In this work, we have analysed the regulation of the timing of neurogenesis and mitotic exit in the midbrain, focusing on the mDA-neuron-generating floor plate. We characterised transcriptional differences of floor plate radial glia from different embryonic stages, from the active phases of mDA neurogenesis to the ependymal maturation. Together, our data reveal a molecular basis for the tight temporal control of mitotic exit as well as neuro- and gliogenesis in the midbrain, including the region generating mDA neurons.

In order to understand the timing of mitotic exit in the mDA progenitors, we examined whether they produce glia after having generated mDA neurons. For this, we administrated tamoxifen at E9.5 to embryos carrying Lmx1a^CreERT2 and the R26^TrapCherry (R26^TrapC) reporter. Tamoxifen treatment activated Cre in mDA progenitors, labelling them and all their progeny with red fluorescent protein (RFP) mCherry (Fig. S1A). We verified that RFP⁺ cells in the adult midbrain had neuronal morphology and expressed NeuN (Rbfox3), and that none were GFAP⁺ or Olig2⁺ (Fig. S1B,C). This indicated that, after generating mDA neurons, the mDA progenitors lack a gliogenic phase and likely exit mitosis.

To understand the transcriptional changes during the transition into cell cycle exit, we re-analysed our previously generated single-cell RNA sequencing (scRNAseq) dataset of mDA progenitors (Kee et al., 2017). This revealed clusters containing either ‘early’ (E10-E11) or ‘late’ (E12-E13) Lmx1a⁺ Corin⁺ cells that expressed the progenitor markers Hes1, Nes and Mki67, indicating mDA progenitor identity (Fig. S2A-C). The ‘late’ cluster cells expressed less Ccnd1, suggesting some of these cells may have already exited the cell cycle. The single-cell differential expression (SCDE) analysis between the ‘early’ and ‘late’ clusters yielded lists of genes either down- or upregulated in ‘late’ mDA progenitors (Fig. S2D,E, Table S1). The downregulated genes were involved in cell cycle regulation, while the list of upregulated genes contained two transcription factors previously associated with the regulation of neurogenesis, gliogenesis and stem cell and progenitor quiescence, Nfib and Sox9 (Chang et al., 2013; Scott et al., 2010; Kadaja et al., 2014; Clark et al., 2019). Indeed, when analysing all known Nfi genes and Sox9 in the scRNAseq dataset, Sox9 was already present in the ‘early’ mDA progenitor cluster, as was Nfia to some extent (Fig. S2F). In contrast, Nfib and Nfix were more enriched in the ‘late’ cluster, and Nfic was present in only a few cells.

To complement the scRNAseq dataset, Lmx1a^CreERT2-labelled RFP⁺ floor plate cells were collected from E11, E13 and E18 embryos for RNA sequencing by laser capture microdissection (LCM; Fig. 1A, Table S1). Comparison of the genes that were differentially expressed between E13 and E11 revealed upregulation of, not only Nfib, but also Nfix and, to a lesser extent, Nfia (Fig. 1B).

Immunohistochemical analysis of these factors in the embryonic midbrain showed that their expression dynamics followed an approximately similar pattern with signal first detected in the ventrolateral ventricular zone (VZ), followed by upregulation in the mDA progenitors. At E10, when the first mDA progenitors become post-mitotic, Sox9 started to be expressed in the ventral midbrain, followed by Nfia and Nfib at E11 and Nfix at E12 (Fig. 1C, Fig. S3A,C). By E13, when mDA neurogenesis begins to cease, the floor plate expressed Sox9, Nfix, Nfia and Nfib (Fig. 1C, Fig. S3A,C). In contrast, Nfic was restricted to roof plate (Fig. S3D). The expression of Nfib, Nfix and Sox9 appeared to be unchanged in the mDA progenitor domain between E14 and E15 (Fig. S3B), suggesting a plateau at G0. Supporting these results, the E18 versus E13 comparison of the dataset showed no expression changes of these factors (Fig. 1D). These results show that Nfib, Nfix, Nfia and Sox9 become gradually upregulated in the midbrain VZ during the neurogenic period and that by the end of mDA neurogenesis, they are all expressed in the floor plate.

These observations led us to speculate that Sox9 and Nfi transcription factors might regulate the transition from proliferation to G0 in the midbrain neuronal progenitors, and that the inactivation of these factors might delay or inhibit this transition.

To analyse this, we conditionally inactivated these alleles in the midbrain by recombining the floxed alleles with En1^Cre, which is expressed in the midbrain and anterior hindbrain from E8.5 onwards (Kimmel et al., 2000; Sgaier et al., 2007). As Nfia was strongly expressed in all mDA progenitors already at E11, it may be less crucial for regulating their mitotic exit and was thus excluded from these functional studies. Although Sox9 displayed a similar expression pattern, being expressed even earlier than Nfia, we chose to include this factor for two reasons: first, it had been shown to act upstream of Nfia in the developing spinal cord (Kang et al., 2012), suggesting that it might regulate Nfi transcription factors also in the midbrain, and, second, its conditional inactivation in the embryonic cerebellum results in prolonged neurogenesis (Vong et al., 2015).

We generated both single mutants and different combinations of these alleles, noting that they did not affect the expression of each other or Nfia (Fig. S3E). These results also showed that in the midbrain Sox9 was not required for the expression of Nfi genes. We detected no apparent midbrain phenotype in Nfib^cko or Nfix^cko single mutant embryos (data not shown), and they are not described further in this study. Instead, we focused on Sox9^cko, NfixNfib^dcko and Sox9NfibNfix^tcko mutants.

To analyse the proliferative state of the midbrain VZ in these mutants, we pulsed them with bromodeoxyuridine (BrdU) for 30 min before tissue collection at different embryonic stages and compared the proportion of BrdU⁺Sox2⁺ cells to all Sox2⁺ cells in the VZ. As these factors had shown different expression dynamics in the dorsoventral domains of the midbrain (Fig. 1C), we analysed separately the most dorsal tip (roof plate), the most ventral tip (mDA-generating Lmx1a⁺ floor plate), and the dorsolateral and ventrolateral midbrain.

As Nfib, Nfix and Sox9 were gradually upregulated in the midbrain (Fig. 1C), the lack of major differences in BrdU uptake between the genotypes at earliest stages analysed was unsurprising (E11 and E12; Fig. S4A). In contrast, a few days later numerous BrdU⁺ nuclei could be detected in the VZ of all mutants (Fig. 2A). Quantifying the signal across the stages showed the most drastic increase in BrdU uptake in Sox9NfibNfix^tcko VZ (Fig. 2B). The requirement of these factors differed across the dorsoventral domains, with Sox9^cko showing a significantly higher proportion of cycling cells in the ventrolateral domain compared to NfibNfix^dcko (Fig. 2B). These results indicate that the loss of Sox9, Nfib and Nfix keeps the neural progenitors longer in the cell cycle and postpones the entry into G0.

The prolonged proliferative state might also disturb or postpone the maturation of progenitors into ependymal cells. We analysed this by immunohistochemical staining for FoxJ1, a regulator of ciliogenesis in ependymal cells (Jacquet et al., 2009), and S100b, a marker for mature ependymal cells (Vives et al., 2003). We could detect fewer FoxJ1⁺ and S100b⁺ cells in all mutants compared with control at E18, with an almost complete absence of S100b⁺ cells in Sox9NfibNfix^tcko VZ (Fig. S4B-E). This suggests that, consistent with observations in the Nfix^−/− cortex (Harkins et al., 2022), the loss of Nfib, Nfix and Sox9 delays or even prevents full ependymal maturation in the midbrain.

As cell cycle exit is linked to neurogenesis and gliogenesis, we investigated whether the increased proportion of cycling cells in Nfib/x-Sox9 mutant VZ had any effect on these processes. The ventrolateral and dorsolateral progenitor domains of the midbrain generate glutamatergic, GABAergic and some cholinergic neurons (Lahti et al., 2013). At E15, the number of Pou4f1⁺ glutamatergic neurons was increased in the midbrain in all mutants compared to the controls (Fig. 2C,D). An opposite effect was seen in the production of Olig2⁺ oligodendrocyte precursors, with the highest reduction in Sox9NfibNfix^tcko embryos (Fig. 2E,F). By E18, the number of Olig2⁺ cells in all mutants was approximately 60% of that seen in controls (Fig. S4F,H), but the majority of Olig2⁺ cells in both control and Sox9NfibNfix^tcko midbrain co-expressed Sox9 at this stage (Fig. S4G,I). This suggests that the wild-type Sox9⁺ oligodendrocytes from outside the En1^Cre-recombined region repopulate the mutant midbrain. Indeed, when only Sox9⁻Olig2⁺ cells, which should predominantly be of midbrain origin, were compared in mutants to Sox9⁺Olig2⁺ cells in controls, we could see a much more drastic reduction in the Sox9NfibNfix^tcko midbrain (Fig. S4I). Furthermore, all mutants had fewer Aldh1l1⁺ cells in the midbrain marginal zone, indicating that the generation of astrocytes was also affected (Fig. S4J-L).

These results indicate that, similar to what has been shown in other tissues (Stolt et al., 2003; Deneen et al., 2006; Kang et al., 2012; Clark et al., 2019), Nfib, Nfix and Sox9 promote gliogenesis in the midbrain, and that by inducing mitotic exit in the progenitors, they also control the extent of neurogenesis.

We then turned our attention to the mDA neuron-generating floor plate. Similar to more dorsal midbrain, floor plate progenitors showed increased BrdU uptake in all mutants compared to the controls, with the strongest effect in the Sox9NfibNfix^tcko mutants (Fig. 3A,B). By E14, when mDA neurogenesis is normally complete, only few individual BrdU⁺ Sox2⁺ cells were detected in the floor plate of the control embryos. In contrast, the mutants showed higher BrdU uptake until E16 (Fig. 3B).

Furthermore, the neurogenic genes Neurog2, Ascl1 and Dll1 were upregulated in the E14 Sox9NfibNfix^tcko ventral midbrain compared to the control, indicating prolonged neurogenesis (Fig. S5A). Hes5, a marker for both neural and glial progenitors, showed only upregulation in more lateral midbrain of the mutants and was absent in the non-gliogenic floor plate. The cell cycle G1-S transition regulator cyclin D1, and PCNA, another marker for cycling cells, were also upregulated in E15 Sox9NfibNfix^tcko ventral midbrain (Fig. S5B-F). Taken together, these results show that in the absence of Sox9, Nfib and Nfix the ventral midbrain progenitors remained in the cell cycle at least until E15. However, by E18, progenitors were apparently in G0 in this region across all genotypes (Fig. 3B). This shows that the loss of Nfib, Nfix and Sox9 postponed – but did not prevent – cell cycle exit in mDA progenitors.

Nevertheless, by E18 this prolonged neurogenic window in the floor plate resulted in, on average, 30% and 60% more mDA neurons in NfibNfix^dcko and Sox9NfibNfix^tcko embryos, respectively (Fig. 3C,D). The increase was detectable already at E14 (Fig. S4H,I) and was particularly prominent in the caudal ventral tegmental area (VTA) (Fig. 3C, Fig. S5G). As we saw delayed cell cycle exit and prolonged neurogenic marker expression in the floor plate, we speculated that some of these extra mDA neurons could be generated after the normal mDA neurogenic window. To investigate this possibility, we administered BrdU in drinking water to pregnant females from E14 to E18, E15 to E18, and E16 to E18, and analysed the number of BrdU-labelled mDA neurons (Fig. 3E). As the number of mDA neurons had shown a modest but not statistically significant increase in Sox9^cko embryos compared to controls (Fig. 3D), we decided to focus only on NfibNfix^dcko and Sox9NfibNfix^tcko embryos. In the controls, we could find rare BrdU⁺Pitx3⁺TH⁺ neurons in the caudal VTA in the E14-E18 labelling scheme but none in later labelling windows (Fig. 3E,F), supporting earlier findings that the majority of mDA neurons are generated by E14 (Bayer et al., 1995). In contrast, both NfibNfix^dcko and Sox9NfibNfix^tcko contained numerous BrdU⁺ mDA neurons, especially in the E14-E18 labelling group, but also some in the E15-18 labelling window (Fig. 3E,F). These labelled cells showed highest enrichment in the caudal VTA. However, in the E16-E18 labelling window no BrdU⁺ mDA neurons were present in either mutant. These results suggest that the loss of Nfib/x and Sox9 delays the cell cycle exit and prolongs the neurogenic window, contributing to the higher number of both actively cycling mDA progenitors, and mDA neurons. Analysis of the more dorsal regions showed that the Sox9NfibNfix^tcko dorsal midbrain still generated neurons at E16-E18 (Fig. S5J), consistent with the 30-min BrdU pulse labelling (Fig. 2B).

One possible reason for the cessation of mDA neurogenesis in mutants is that Sox9, Nfib and Nfix might have additional roles in postmitotic mDA differentiation, or that some environmental cues required for mDA neurogenesis are no longer present in later embryonic stages. Either one of these scenarios might result in mDA progenitors failing to mature into TH⁺ neurons.

Indeed, ectopic Sox2⁺ cells appeared below the floor plate VZ in Sox9^cko, but particularly in Sox9NfibNfix^tcko mutants (Fig. S6A) from E14 onwards. In contrast, in NfibNfix^dcko mutants we could not detect a similarly drastic phenotype.

These ectopic cells might result from a structural collapse of the apical junctions in the most ventral VZ. At E13, before the appearance of these ectopic progenitors, the adherens junction marker β-catenin appeared normally expressed in the Sox9NfibNfix^tcko midbrain (Fig. S6B). Two days later, when ectopic progenitors could be seen in both Sox9^cko and in Sox9NfibNfix^tcko, β-catenin was lost specifically in NfibNfix^dko and in Sox9NfibNfix^tcko but remained surprisingly unaffected in Sox9^cko (Fig. S6C). Moreover, the tight junction marker ZO1 (Tjp1) was normally expressed at E15 in the Sox9NfibNfix^tcko VZ (Fig. S6D). These results indicate that the loss of cell-to-cell adhesion may contribute to, but cannot be the sole cause of, these ectopic cells.

The ectopic cells lacked Pitx3, TH and Otx2, but retained expression of GFAP, Sox6 and FoxA2, and did not take up BrdU (Fig. S6E). Although Sox6 and FoxA2 were expressed in both mDA progenitors and neurons, GFAP was absent in mDA neurons, which suggests these cells could be mDA progenitors that have entered G0 but which are unable to complete their maturation into mDA neurons.

As the inactivation of Sox9 by En1^Cre results in death immediately after birth (Vong et al., 2015), analysis of the adult brain phenotype of Sox9^cko or Sox9NfibNfix^tcko mutants was impossible. However, NfibNfix^dcko mutants do survive, although most die before weaning. A previous study described a lack of foliation in the E18 Nfib^−/− cerebellum (Steele-Perkins et al., 2005). Consistent with this, En1^Cre-mediated conditional loss of Nfib led to a major cerebellar malformation (Fig. S7A), resulting in severe movement control problems. However, the presence of BrdU⁺ nuclei within the mutant cerebellum suggests prolonged neurogenesis and/or gliogenesis in this region (Fig. S7B).

Some of the NfibNfix^dcko mutants survived to adulthood, enabling analysis of the aqueduct, although long-term experiments were impossible. We noticed that although, similar to controls, the midbrain ependymal cells in these mutants expressed S100b, FoxJ1 and vimentin, the region above the dorsal aqueduct the region above the dorsal aqueduct contained a large cluster of Sox2⁺ cells, but they were likely postmitotic due to the lack of cyclin D1 expression and BrdU uptake (Fig. S7C).

Next, we aimed to understand whether Nfib, Nfix and Sox9 not only induce but also maintain the G0 state in midbrain ependymal cells. Indeed, the expression of these factors was retained in the adult midbrain aqueduct (Fig. 4A).

To inactivate these genes in postmitotic ependymal precursors, we employed a Pdyn^Cre mouse line (Krashes et al., 2014), as its expression begins in the most ventral midbrain VZ at around E15 to E16 (Fig. S7D), a stage when almost no BrdU uptake was detected in the control floorplate (Fig. 3B). In the adult aqueduct, the expression of RFP in Pdyn^Cre R26^TdTomato (hereafter R26^TdT) mice was seen in the most ventral and dorsal cells (Fig. S7D, Fig. 4B). As expected, no mDA neurons were labelled with this Cre line, demonstrating that Pdyn^Cre becomes active only after mDA progenitors have exited the cell cycle (Fig. S7E). Recombination of floxed alleles of Nfib, Nfix and Sox9 with this Cre line led to the absence of these factors in the ventral aqueduct (Fig. 4B).

Although we saw no BrdU uptake in the mutant cells of the aqueduct (Fig. S7G; see below), ectopic nuclei within the ventral RFP⁺ vimentin⁺ fibres were occasionally detected (Fig. 4D). Although these fibres were found also in controls and are described in greater detail further below, we could only find the ectopic nuclei in Pdyn^Cre Sox9NfibNfix^tcko mutants. Because of their rare appearance, we could not determine their precise birth date by BrdU-pulse labelling. The lack of S100b in the Pdyn^Cre Sox9NfibNfix^tcko aqueduct indicated a maturation failure of the FoxJ1⁺ ependymal precursors (Fig. 4E). Together with the observations from the En1^Cre mutants, these results suggest that Nfib, Nfix and Sox9 are required for the induction of cell cycle exit and for the maturation of ependymal precursors, but they are dispensable for the maintenance of G0 state.

Previous work suggested that, similar to DA-feedback-regulated regeneration of mDA neurons in salamanders (Berg et al., 2011), evolutionary traces of this mechanism could still operate in the mammalian brain (Hedlund et al., 2016).

Intrigued by this idea, we hypothesised that if midbrain ependymal cells were not permanently in G0 but only quiescent, DA signalling might maintain the expression of quiescence-promoting transcription factors, such as Nfi and Sox9. Thus, in the absence of DA signalling, the expression of these factors would be downregulated. We thus generated local lesions of periaqueductal DA neurons by injecting 6-OHDA close to the aqueduct. Although this method successfully removed most TH⁺ cell bodies and fibres near the aqueduct, it had no discernible impact on the expression of Nfi genes or Sox9 (Fig. S7F).

However, we still wanted to test whether the loss of DA signalling could influence ependymal cells in mice with inactivated Nfib/x and Sox9. For this, we repeated the 6-OHDA lesions in Pdyn^CreSox9NfibNfix^tcko animals followed by BrdU in drinking water for 25 days to label any cells that might re-enter the cell cycle (Fig. S7G). However, we could not detect any increased BrdU labelling in any condition.

Taken together, these results indicate that dopamine signalling appears to be nonessential for maintaining the G0 state in the adult midbrain ependymal cells.

Intrigued by the fibres extending from the ventral-most ependymal cells in Pdyn^Cre reporters, we investigated the adult aqueduct further with immunohistochemistry. Our histological analysis revealed that the aqueduct was in close contact with GFAP⁺ astrocytes (Fig. 5A,B) and we could also detect Vim⁺ fibres protruding from the ependymal cells. These fibres were seen throughout the aqueduct but were particularly prominent in the dorsal and ventral tips. The dorsal fibres intermingled with GFAP⁺ fibres from the astroependymal cells (Fig. 5A). The dorsal and ventral ependymal cells also expressed nestin, similar to the embryonic roof plate and floor plate, as does adult spinal cord, where Nes⁺ fibres can be found in the dorsal side (Fig. S8A,B; Hamilton et al., 2009). The midbrain ependymal fibres mostly contacted nearby blood vessels, but, on the ventral side, longer fibres also contacted PDGFR⁺ and Olig2⁺ cells – likely oligodendrocyte precursors (Fig. S8D,E).

The ventral-most ependymal cells expressed the mDA progenitor and neuron markers FoxA2, Sox6 and FoxP1 (Fig. 5C). They also displayed mosaic expression of the progenitor markers GFAP, PCNA and Prom1 (also known as CD133), whereas the ependymal marker FoxJ1 appeared weaker compared to more dorsal aqueduct (Fig. 5B,D). Mosaic expression of several markers suggests internal heterogeneity within this domain, which may either be temporal or represent subtypes of ependymal cells, perhaps with different properties.

To improve our understanding of the properties of midbrain ependymal cells, we next analysed them by scRNAseq. FoxJ1Cre^ERT2-recombined RFP⁺ ependymal cells were collected from lateral ventricles, the third ventricle, midbrain and spinal cord, and analysed by SmartSeq2 (Fig. 6A-C). As an outgroup, RFP⁺ cells from Nkx2-2^Cre R26^TrapC midbrains – mostly glia – were also included (Fig. 6B,C). After quality control, 1473 cells, of which 567 were midbrain ependymal cells, were included in the analysis.

A t-distributed stochastic neighbour embedding (tSNE) projection segregated the dataset into 11 distinct clusters (Fig. 6A), each identifiable by its unique transcriptional profile, verified in situ in the Allen Mouse Brain Atlas (Fig. 6D,E, Table S2). The dataset can be explored using the web-based tool CellxGene (perlmannlab.org/resources).

In the tSNE projection, the forebrain and lateral midbrain ependymal cells clustered together and shared the expression of several genes, whereas most dorsal and ventral ependymal cells formed three separate clusters (Fig. 6A,D). The identities of different clusters were annotated based on the expression of markers for distinct cell populations. Thus, cluster 8, expressing Npsr1, Sspo and Igfbp6, corresponded to the subcommissural organ (SCO), and the Prokr2⁺ Vtn⁺ cluster 9 consisted of astroependymal cells, located more caudal to SCO in the dorsal side (Fig. 6A,D,E). The former mDA progenitors in the ventral-most part of the aqueduct formed cluster 11 (Fig. 6A). Their enriched genes included opioid polypeptide hormone prodynorphin (Pdyn), a precursor of several neuropeptides involved in the modulation of multiple behaviours through κ opioid receptors (Schwarzer, 2009); the cytokine Kit ligand (Kitl; also known as stem cell factor); and the serotonin receptor Htr5b. These cells also expressed several genes and transcription factors typical of mDA progenitors, such as Cnpy1, Ntn1, FoxP1, En1 and FoxA1 (Fig. 6D,E), supporting our histological analyses (Fig. 5). This indicates that these cells might retain some mDA progenitor properties. Moreover, this unique transcriptional profile – including, for example, Pdyn expression – together with the observed glia- and blood vessel-contacting fibres, suggests that these cells could form a type of circumventricular organ (CVO; discussed further below).

Some spinal cord ependymal cells can slowly proliferate in vivo, as well as differentiate into both glia and neurons in vitro, and contribute to glial scar formation after a spinal cord injury (Meletis et al., 2008; Barnabé-Heider et al., 2010).

As we had seen several progenitor markers being expressed in the ventral-most domain of the aqueduct, we investigated whether these cells had any proliferative capacity in vivo. For this, we gave tamoxifen-treated FoxJ1^CreERT2R26^TdT mice BrdU in drinking water for 6 weeks. Although this treatment successfully labelled both SVZ neural stem cells and spinal cord ependymal cells, BrdU uptake was not seen in any midbrain ependymal cells (Fig. S8F). These results suggest that, although midbrain ependymal cells share several features with the ependymal cells in the spinal cord, they cannot proliferate in vivo, at least not under normal conditions. We speculated that some signals from the immediate surroundings might maintain the G0 state of these cells, and that removing them from this environment might trigger proliferation.

A previous study reported sphere-forming potential in different parts of the neuraxis, although those spheres were not genetically fate-mapped to verify their origin (Golmohammadi et al., 2008). To trace the midbrain ependymal cells in vitro, we collected both the ependymal cells and the surrounding parenchymal cells from tamoxifen-treated FoxJ1^CreERT2R26^TdT reporter mice (Fig. 7A). Cells were then plated both on fibronectin-treated surfaces and in free-floating cultures in neurosphere media. Bipolar RFP⁺ cells were present on the fibronectin-coated plates in cultures from the aqueduct, but not from the parenchyma (Fig. 7B).

After 2 weeks in vitro, most of the parenchymal culture cells had died without forming any spheres in either fibronectin-coated wells or free-floating cultures (Fig. 7B,D). In contrast, the aqueductal cultures formed small RFP⁺ spheres on fibronectin and larger spheres in free-floating cultures (Fig. 7B-E). They could also form secondary spheres (Fig. 7E) and could be passaged for several months.

The RFP⁺ spheres expressed several ependymal and neural progenitor markers, such as Nestin, Vimentin, Sox2, and S100beta (Fig. 7F). We compared their expression profiles by RNA sequencing to those of true SVZ-neurospheres from the same animals. Although spheres in a free-floating culture are always a mix of several cell types, we could see a clear transcriptional difference between the SVZ neurospheres and midbrain ependymal spheres (Fig. 7G, Table S3). For example, SVZ neurospheres expressed more of the neural stem cell marker Gfap, and were enriched in forebrain-specific genes such as Six3, Emx2, Dlx1 and Dlx2. The midbrain spheres in turn contained several ependymal markers typical for midbrain, such as Mab21l2, Pax3 and Cnpy1. By comparing these results to those from scRNAseq, we could see that the spheres contained genes specific for all four different midbrain ependymal types (Table S2), indicating that ependymal cells in all different domains of the aqueduct could contribute to the sphere formation.

However, although ependymal spheres showed extensive proliferative capacity, they could only produce glial cells, not neurons, in the differentiating conditions (Fig. 7H).

Taken together, these results indicate that midbrain ependymal cells retain a capacity for proliferation if placed in a suitable environment, although their differentiation capacity in vitro is limited.

Several studies have highlighted the diverse roles of Sox9 and Nfi transcription factors in embryonic neurogenesis and gliogenesis, as well as in the maintenance of the adult stem cell niche (Das Neves et al., 1999; Steele-Perkins et al., 2005; Campbell et al., 2008; Martynoga et al., 2013; reviewed by Harris et al., 2015; Sarkar and Hochedlinger, 2013). In the current work, we focus on the function of these factors in midbrain development and show that they are required to induce mitotic exit in the midbrain neuronal progenitors, and that their loss in the embryonic midbrain results in prolonged and increased neurogenesis, impaired gliogenesis and delayed ependymal maturation. We also show that ependymal cells derived from mDA progenitors have radial glia-like characteristics, contact nearby blood vessels, and express a precursor for opioid peptides, suggesting that these cells might form a CVO in the adult.

Earlier studies on Sox9 and Nfi transcription factors in the developing CNS have focused on cortex, cerebellum and spinal cord (Harris et al., 2015; Jo et al., 2014; Vong et al., 2015). Here, we focused on the embryonic midbrain, and identified these factors as being upregulated in late- versus early-stage mDA progenitors.

Based on several of our observations, we postulate that these factors are required to initiate the mitotic exit and promote ependymal maturation. First, they were gradually upregulated in the mDA progenitors during the neurogenic window and peaked as proliferation and neurogenesis ceased. Second, their conditional inactivation in the midbrain led to prolonged and increased proliferation and neurogenesis, and reduced expression of marker genes for mature ependymal cells. This effect was not restricted to the mDA domain but was seen throughout the midbrain.

This phenotype could be detected in single Sox9^cko mutants but required the inactivation of both Nfib and Nfix, suggesting functional redundancy between Nfi transcription factors similar to that observed in the developing hippocampus (Harris et al., 2016). The function of Nfia in the midbrain development remains to be investigated.

Compared to earlier results from other brain regions, our results from the midbrain show both similarities and differences. Our data from Sox9^cko midbrain resembles the observations from the cerebellum in these same mutants (Vong et al., 2015), suggesting that Sox9 regulates mitotic exit and neurogenesis in both of these neighbouring tissues. In contrast, in the developing cortex Sox9 is needed for the induction and maintenance of NSCs, and its loss results in reduced neurogenesis (Scott et al., 2010).

Nfi transcription factors, in turn, promote differentiation and suppress the self-renewal of neural progenitors in the embryonic forebrain (Piper et al., 2010, 2014; Heng et al., 2014; Harris et al., 2016). In Nfix^−/− and Nfib^−/− embryos, the number of telencephalic radial glial progenitors increases, but their maturation into intermediary progenitors is delayed, thus postponing and prolonging neurogenesis and contributing to brain malformations (Harris et al., 2016). In contrast, in the midbrain of NfibNfix^dcko mutants, the increased BrdU uptake seen in the VZ did not delay or inhibit neurogenesis but led directly to increased production of neurons, resembling the phenotype in the Sox9^cko midbrain.

This increased neurogenesis may be further exacerbated in the lateral midbrain by drastically reduced gliogenesis in the mutants. If Nfib/x and Sox9 are already required in the progenitor cells to induce glial generation, their absence might switch gliogenic cell divisions to become neurogenic. Inactivation of these factors specifically in the postmitotic glial precursors would clarify the role of Sox9 and Nfi transcription factors in midbrain gliogenesis.

Although in Sox9^cko mutants we observed increased neurogenesis in the dorsolateral midbrain and prolonged BrdU uptake in mDA progenitors, the increase in mDA number by E18 in these mutants was relatively mild. In contrast, the loss of Nfib, Nfix and Sox9 together resulted in a more severe mDA phenotype compared to NfibNfix^dcko, indicating that Sox9 does contribute to the regulation of mDA neurogenesis. Furthermore, the inactivation of Sox9 by itself or combined with Nfib and Nfix led to the accumulation of ectopic mDA progenitors after E14. Understanding the differences between these phenotypes and the transcriptional targets of Nfi transcription factors and Sox9 in mDA neurogenesis would benefit from scRNAseq characterisation of these mutants.

The generation of ectopic mDA neurons in mutants was more prominent in the caudal midbrain. Anterior mDA neurons such as those in the substantia nigra pars compacta are generated first, around E10 (Bayer et al., 1995), when Nfi factors are not yet expressed in the floor plate and the majority of the progenitors are actively cycling cells. The more caudal mDA neurons are generated at the later stages when Nfi genes and Sox9 begin to be expressed. We speculate that under normal circumstances an increasing proportion of mDA neurons would have exited the cell cycle when the more caudal mDA neurons are still being generated from the remaining cycling cells. The inactivation of Nfi genes and Sox9 would keep those ‘late’ mDA progenitors in the cell cycle longer, enabling increased generation of late-born (more caudal) mDA neurons.

In the cortex, Nfix is required for the differentiation of ependymal cells and maintenance of the adult ependymal layer (Harkins et al., 2022). Our data from the embryonic midbrain also show reduced expression of the ependymal marker genes FoxJ1 and S100b in the absence of Nfib, Nfix and Sox9. However, as we could see a normal ependymal layer in the adult NfibNfix^dcko aqueduct, the ependymal maturation in these mutants was only postponed, likely due to the delayed mitotic exit. In contrast, Pdyn^Cre-mediated inactivation of all three factors during late embryogenesis resulted in the loss of S100b. This indicates that Nfib, Nfix and Sox9 promote mitotic exit enabling the ventricular cells to start their differentiation into ependymal cells. In addition, Sox9 appears to have an additional role in activating ependymal gene expression independently of FoxJ1, at least in the ventral-most aqueduct.

Spinal cord ependymal cells have the capacity to proliferate and form both glia and neurons in vitro, and they contribute to the formation of glial scars after injury in vivo (Horner et al., 2000; Meletis et al., 2008; Ren et al., 2017). In contrast, both histological and scRNAseq analyses indicate that ependymal cells in the lateral ventricles neither proliferate nor contribute to neurogenesis or gliogenesis in vivo (Shah et al., 2018; Spassky et al., 2005). The properties of midbrain ependymal cells have not been characterised before in detail.

In this study, scRNAseq identified four distinct types of cells within the midbrain ependyma. Whereas ependymal cells of the lateral walls of the aqueduct resembled transcriptionally the ‘classic’ ependymal cells in the lateral ventricles, the dorsal and ventral tips of the aqueduct consisted of cells that shared many features with neural progenitor cells, such as the expression of nestin and GFAP. The two dorsal cell types were astroependymoglial cells and the SCO, whereas the ventral-most cell population consisted of cells that descended from mDA progenitors.

The ventral-most Lmx1a⁺ ependymal cells characterised here also expressed Gfap, Pcna, Prom1 and several mDA progenitor markers such as FoxA1/2, En1, Cnpy1, Sox6 and FoxP1. The varied expression of these factors in this domain indicates further heterogeneity – possibly temporal – to be uncovered. Despite their radial glia-like characteristics, these cells did not incorporate BrdU in vivo, nor did they express cyclin D1.

As these cells did form gliogenic spheres in vitro, they seem to reside in a state of deep quiescence, rather than being permanently in G0. This state might be maintained either by inhibitory signals from the environment or by the lack of cell cycle-promoting stimuli, or both. For example, inhibition could be based signals from SCO (Vera et al., 2013), or on contacts with the surrounding parenchyma and/or each other, as in a mouse model of hydrocephalus in which the ependymal cells start to divide in order to repair the damaged ventricular walls (Bátiz et al., 2011).

In the adult salamander, the midbrain ependymoglial cells are kept in a quiescent state by feedback inhibition from the surrounding mDA neurons (Berg et al., 2011). Previous work (Hedlund et al., 2016) suggested that this mechanism might also generate some proliferating cells in the mouse midbrain. However, we noted no ependymal cell re-entry into the cell cycle after local DA depletion. Neither NfibNfix^dcko nor Pdyn^CreSox9NfibNfix^tcko adult mutants showed any signs of cell cycle re-entry in the midbrain ependyma, suggesting that these factors are dispensable for the maintenance of G0 in the adult. Other mechanisms to keep these cells in a quiescent state might involve chromatin modifications or a switch of transcription factor complexes from those promoting initiation of G0 to those maintaining it.

Whether growth factor infusions or other stimuli can trigger the midbrain ependymal cells to re-enter the cell cycle remains to be investigated and should be accompanied by scRNAseq analyses and a reporter line, such as Fucci or m-Venus-p27K (Oki et al., 2014), to track any changes in the cell status.

In addition to radial glia and mDA progenitor marker expression, the ventral-most aqueductal cells contained long vimentin⁺ fibres contacting blood vessels and glial cells, resembling tanycyte-like cells in several CVOs. CVOs, one of which is the SCO, regulate various behaviours and body homeostasis (Benarroch, 2011). CVOs can send and receive various signals, such as peptides and hormones, between peripheral blood and the CNS, thus bypassing the blood–brain barrier. Some contain elongated, Nes⁺ GFAP⁺ tanycyte-like cells, which can be triggered by mitogens to incorporate BrdU (Furube et al., 2020). In addition, sphere cultures have been obtained from different anteroposterior levels of the neuraxis, although they have not been fate-mapped to any CVOs (Weiss et al., 1996; Golmohammadi et al., 2008). Taken together, some cells within CVOs possess multiple radial glia or tanycyte-like properties and a potential to re-enter the cell cycle, suggesting that they are not in permanent cell cycle arrest (Bennett et al., 2009).

Our results suggest that the ventral-most cells in the aqueduct might form a previously uncharacterised part of the CVO system. First, their radial glia/tanycyte-like morphology and gene expression profile resemble that of several CVOs. Second, in addition to their contacts with blood vessels and oligodendrocytes, the ventral-most aqueduct cells are surrounded by astrocytes, all of which could be potential signalling partners. Third, these cells express Pdyn, a precursor of dynorphins, which are opioid neuropeptides involved in, for example, addiction, learning, stress and the pain response, as well as mood regulation (Schwarzer, 2009). Together, these results indicate that these cells form a type of CVO that might respond to serotonin, or some other signalling, by secreting neuropeptides.

The main receptor for dynorphins is the κ opioid receptor (KOR), encoded by Oprk1. KOR is highly enriched in the VTA mDA neurons, and dynorphin/KOR signalling in the VTA mediates the processing of aversive stimuli and promotes stress-induced compulsive behaviour (Margolis and Karkhanis, 2019; Abraham et al., 2018). Dynorphin inputs to mDA neurons originate from various brain regions, such as the striatum, lateral hypothalamus, amygdala and the bed nucleus of stria terminalis (Margolis and Karkhanis, 2019). Although we do not yet understand the full extent and contacts of the fibres in ventral aqueduct cells, it is intriguing to speculate that some of these fibres might contact VTA DA neurons and possibly form a local input source of dynorphins to modulate their function.

Taken together, our findings provide important insights into the regulatory networks governing midbrain neurogenesis and gliogenesis, with potential implications for understanding neurodevelopmental disorders and neurodegenerative conditions such as PD. Understanding the details of mDA neuron development has been instrumental in developing protocols for engineering therapeutic mDA neurons from stem cells. Future studies will aim to explore how manipulating these pathways might offer therapeutic avenues by enabling more efficient stem cell protocols or by promoting endogenous brain repair.

En1^Cre (Kimmel et al., 2000; IMSR_JAX:007916), Pdyn^Cre (Krashes et al., 2014; IMSR_JAX:027958), Lmx1a^CreERT2 (Kee et al., 2017), FoxJ1^CreERT2 (Meletis et al., 2008), Nkx2-2^Cre (Balderes et al., 2013), Nfib flox (Hsu et al., 2011), Nfix flox (Campbell et al., 2008), Sox9 flox (Akiyama et al., 2002; IMSR_JAX:013106), R26^TdTomato (Madisen et al., 2010; IMSR_JAX:007909; referred to as R26^TdT here) and R26^TrapCherry (Hupe et al., 2014; referred to as R26^TrapC here) mouse strains and their genotyping have been described before. The mice were maintained in an outbred background. The noon of the day of the vaginal plug was considered as E0.5. Control animals or embryos were Cre-negative littermates. All experimental procedures followed the guidelines and recommendations of Swedish animal protection legislation and were approved by Stockholm North Animal Ethics board (permits 13830/18 and 16527-2023).

The embryos were collected in ice-cold DPBS (Gibco, 14190144) and fixed in freshly prepared 4% paraformaldehyde (PFA; Sigma-Aldrich, P6148) in DPBS in standard biopsy cassettes at room temperature (RT) over one or two nights depending on the embryonic stage, with PFA solution changed every day. The adult mice were deeply anaesthetised and intracardially perfused first with 37°C DPBS followed by 37°C 4% PFA in DPBS, the brain dissected out and the entire midbrain separated by coronal cuts using a brain matrix (AgnTho's, 69-2165-1) and postfixed over two nights at RT, changing PFA daily. The samples were dehydrated and processed into Histosec wax without DMSO (Merck, 101676) using a Leica TP1020 automated tissue processor. The sections were cut at 5 µm (embryos) and 6-10 µm (adults) thickness using an automated rotating waterfall microtome (Epredia, HM355S), collected on Superfrost Plus slides (Menzel-Gläser, 631-9483), and dried overnight in a vertical position at 37°C. For the free-floating sections, perfused and postfixed midbrains were cryoprotected in 30% sucrose, cut at 35 µm thickness on a sliding microtome, and collected in PBS in a 48-well plate for staining.

Fluorescence immunohistochemistry and in situ hybridisation on sections were performed as described (Tiklová et al., 2019). Probes for Dll1, Ascl1 and Hes5 have been previously described (Saarimäki-Vire et al., 2007). Probes for Neurog2 and Aldh1l1 were synthetised GeneArt Strings fragments (Invitrogen). The probe for Neurog2 was an 808 bp fragment of mouse Neurog2 cDNA, from CCCTTCTCCACCTTCCTCCT to ATGCCTATTGTCCCGCCCTT. The probe for Aldh1l1 was a 972 bp fragment of mouse Aldh1l1 cDNA, from AGCAGAGGCCGTGCGGAGCTCTT to CCACCCTGATGTGAGGAAAATAGG. The fragments were cloned into a pCR-Blunt II TOPO vector (Invitrogen, K280002) according to the manufacturer's instructions.

The free-floating adult brain sections were permeabilised in 0.3% Triton X-100 in PBS for 10 mins, blocked in 5% donkey serum (Jackson ImmunoResearch, 017-000-121) in 0.1% Triton X-100 in DPBS for 1 h at RT and incubated in primary antibody in blocking solution overnight at RT, washed several times in 0.1% Triton X-100 in PBS followed by incubation in the biotinylated secondary antibody and then stained using a DAB detection kit (Vector Laboratories, SK-4100) according to the manufacturer's instructions.

For immunocytochemistry, the cells were fixed for 30 min at RT in 4% PFA, washed several times in PBS, permeabilised in 0.3% Triton X-100 for 10 min and blocked in 5% donkey serum in 0.1% Triton X-100 in DPBS for 1 h at RT. The samples were incubated in primary antibodies overnight 4°C, washed several times in 0.1% Triton X-100 in PBS, and incubated in secondary antibodies for 2 h at RT. The whole ependymal spheres from free-floating cultures were fixed, dehydrated and processed into paraffin manually, and then sectioned and stained following the protocol for embryonic and adult brain samples.

Details of primary and secondary antibodies can be found in Table S4.

Images in Fig. 6E were sourced from the Allen Mouse Brain Atlas (mouse.brain-map.org), from the following datasets: Pdyn, https://mouse.brain-map.org/experiment/show/71717084; Ntn1, https://mouse.brain-map.org/experiment/show/74511838; FoxA1, https://mouse.brain-map.org/experiment/show/77869794; Htr5b, https://mouse.brain-map.org/experiment/show/71247644; Car9, https://mouse.brain-map.org/experiment/show/74988271; Prokr2, https://mouse.brain-map.org/experiment/show/74511780; Npsr1, https://mouse.brain-map.org/experiment/show/70560288; S100a6, https://mouse.brain-map.org/experiment/show/77278967; Sncg, https://mouse.brain-map.org/experiment/show/72081426; Gpc3, https://mousespinal.brain-map.org/imageseries/show.html?id=100013918.

Adult mice were euthanised with CO₂, decapitated, and brain tissue collected in ice-cold DPBS. The brains were cut coronally at 1 mm thickness using a pre-cooled brain matrix (AgnTho's, 69-2165-1) and razor blades. The aqueduct and parenchymal tissue from the midbrain and subventricular zone from the lateral ventricles were microdissected from the sections using sterile 27G needles attached to 1 ml syringes.

For primary cultures of ependymal and parenchymal cells on fibronectin, the tissues were enzymatically dissociated using a papain kit (Miltenyi Biotec, 130-092-628) following the manufacturer's instructions, followed by myelin-removal using beads (Miltenyi Biotec, 130-096-731) and cells were plated on fibronectin-coated 96-well plates at 20,000 cells/well in NeuroCult Proliferation solution (Stem Cell Technologies, 05702) with heparin (Stem Cell Technologies, 07980), bFGF (Stem Cell Technologies, 78003) and EGF (Stem Cell Technologies, 78006.1) according to the manufacturer's instructions, with antibiotic-antimycotic solution (Gibco, 15240096).

For free-floating sphere cultures, the tissues were dissociated into a single-cell suspension using the NeuroCult enzymatic dissociation kit for adult mouse and rat CNS tissue (Stem Cell Technologies, 05715) and plated on ultra-low attachment 6-well plates (Corning, 3471) in NeuroCult Proliferation solution with growth factors and antibiotic-antimycotic as described above. The spheres were split using either same dissociation kit as above, or DPBS with 0.02% EDTA for 15-20 min at RT followed by trituration (Sigma-Aldrich, E8008). The ependymal spheres were split at 5- to 7-day intervals, and the neurospheres from the SVZ at 3- to 5-day intervals depending on the sphere size and appearance. For the secondary sphere formation test, the single-cell suspension was plated on round-bottomed 96-well plates on a clonal density, or cells were sorted by fluorescence-activated cell sorting (FACS) for RFP directly on a 96-well plate. The plates were monitored after the cells had settled to verify the absence of any spheres at this stage. The plates were analysed 10 days later and only spheres larger than 50 µm in diameter were counted.

Spheres were collected for RNA sequencing from both neurosphere and ependymal sphere cultures from the first passage and were snap-frozen and kept at −80°C until RNA extraction.

For differentiation, the dissociated spheres were plated on 8-chamber slides (Millipore, PEZGS0816) coated on a thin layer of Matrigel without growth factors (Corning, CLS356230). One day after plating, the media was changed to NeuroCult differentiation solution (Stem Cell Technologies, 05704) with antibiotic-antimycotic.

Tissue for LCM was collected from fresh-frozen embryos (E11: n=4; E13: n=5; E18: n=5), which were embedded in OCT, cut coronally at 10 µm, and collected on PEN membrane-coated slides (Zeiss, 415190-9042-000), which were pre-treated under UV light. The sections were fixed in cold (−20°C) 95% ethanol, OCT rinsed off in PBS for 2 mins, and sections dehydrated in 70%, 95% and absolute ethanol for 30 s each and quickly air-dried. The mDA progenitor domain was identified by direct fluorescence and captured from several sections across the midbrain using a Leica LMD7000 system with 20× magnification and keeping the laser power as low as possible. The tissue pieces were collected in 0.2 ml low-adhesive tubes (VWR, 732-0548). Lysis buffer (0.4% Triton X-100 with 2 U/µl RNAse inhibitor; Takara Bio) was added immediately after sample tube removal from the machine, the tubes spun down briefly (300 g for 1 min) and snap-frozen on dry ice. The samples were stored at −80°C until library preparation.

Ependymal and glial cells for scRNAseq were prepared from freshly dissected brains. Ependymal tissue was microdissected from the lateral ventricles, midbrain aqueduct, third ventricle and spinal cord from 3-month-old FoxJ1^CreERT2 R26^TdT animals (n=10; five females and five males). Midbrain tissue was also microdissected from 3-month-old Nkx2-2^Cre R26^TrapC mice (n=7; all females). Tissues were dissociated using a papain kit (Miltenyi Biotec, 130-092-628 for ependymal cells; Worthington Biochemical, LK003150 for Nkx2-2^Cre midbrains). FACS buffer was HBSS (Gibco) without Ca²⁺ and Mg²⁺, with 50 mM glucose added to support ependymal cells. The viability dye TO-PRO3 Ready Flow Reagent (Invitrogen, R37170) was added 15 min before sorting. Cells were sorted using a BD FACS Aria III (BD Biosciences) and collected on 96-well PCR plates. The nozzle size was 100 for FoxJ1^CreERT2 samples and 130 for the Nkx2-2^Cre samples.

RNA was extracted and cDNA libraries prepared using standard SmartSeq2 protocol using Tn5 made in the lab (Picelli et al., 2013, 2014). The quality of cDNA and tagmented cDNA was checked on a High-Sensitivity DNA chip (Agilent Bioanalyzer). LCM and scRNAseq samples were sequenced with Illumina HiSeq 2000, and spheres with Illumina NextSeq 2000.

DESeq2 package version 1.34.0 was used to analyse RNA-sequencing data, including differential gene expression, using a Negative Binomial GLM fitting per gene to account for overdispersion in the count data. Next, Wald statistical test was used to derive a P-value for each model coefficient and P-value adjustment for multiple testing correction with Benjamini and Hochberg (FDR) correction (Love et al., 2014). For neurosphere differential expression, the lfcShrink() function in DESeq2 package was used to estimate the log fold changes more accurately and shrink the exaggerated log fold changes, which were observed mainly for genes with low counts or high variability. The adaptive t prior shrinkage estimator type=‘apeglm’ was used from the ‘apeglm’ package (Zhu et al., 2019). The ‘EnhancedVolcano’ package version 1.10.0 was used to make volcano plots (Blighe et al., 2024).

For the analysis of scRNAseq, the sequenced reads were aligned to the mouse genome (mm10) merged with the eGFP sequence using Star v.2.3.1o (Dobin et al., 2013) and filtered for uniquely mapping reads. Expression values were then calculated for each Ensembl ID (release 69) based on the kilobase gene model with one million mappable reads using ‘rpkmforgenes’ (Ramsköld et al., 2009). Low-quality cells were then filtered based on the following thresholds: >28.1% uniquely mapped reads, >49.5% mapped to exons, <12.6% reads mapped to the 3′ end, >3.2% of all genes detected and at least with a 100,000 normalisation reads. From a total of 1805 cells, 332 were removed with a resulting 1473 good quality cells that were used for the downstream analysis. Many of the quality control metrics of the cells were monitored and checked as suggested by the ‘Scater’ R package (McCarthy et al., 2017). The gene Yam1, which codes for a long noncoding RNA, was removed from the dataset as the average expression of this gene in the cells compared to all the other genes was disproportionately high.

Clustering of the cells was done in two different steps. (1) The cells within each of the different tissue types were clustered individually, followed by (2) the cells from all the tissues were clustered together. The cells within each tissue type were clustered with the t-SNE+k-means algorithm. The differentially expressed genes for each of the different clusters in each tissue type were then detected using MAST (Finak et al., 2015). For the clusters in the midbrain tissue, differentially expressed genes were calculated similarly for each pairwise cluster comparison. For clustering all the tissue types together, the shared nearest neighbour algorithm was applied with five neighbours to be considered for the graph construction. This graph was then used to cluster the cells with the ‘Louvain’ algorithm (Blondel et al., 2008). Based on information from the tissue-specific clusters along with information of the marker genes, three of the clusters – one each from midbrain, forebrain and third ventricle – were merged with their respective closest cluster. Similar to the tissue-specific clusters, differential expression analysis was then carried out for each cluster using MAST. All of the above-mentioned analysis and the visualisations were carried out using ‘Seurat’ R package v.3 (Stuart et al., 2019).

The dot plot for scRNAseq data was made using Seurat package v.4 (Hao et al., 2021).

The interactive single cell dataset exploration tool was created using CellxGene (https://github.com/chanzuckerberg/cellxgene).

Gene ontology terms of the scRNAseq dataset were analysed using the Panther statistical over-representation test for biological process using default parameters (https://www.pantherdb.org).

The fluorescence images were taken on a Zeiss LSM700 confocal microscope, whole ependymal spheres were imaged using a Zeiss Observer Z.1, and free-floating sections using a Nikon Eclipse E1000. The images were processed, and brightness and contrast adjusted, either in Photoshop (Adobe) or in Fiji (ImageJ2 v.2.9.0/1.54d). If needed, an unsharp mask was also applied in Photoshop. Tiled images (Figs 3C, 5A, Figs S1A, S7D,E) were compiled using either the automated photomerge function in Photoshop, or by pairwise stitching (Preibisch et al., 2009) in Fiji. Confocal stacks were compiled into maximum intensity projections using Fiji.

All quantifications were carried out using Fiji. Quantification of BrdU uptake in ventricular cells (Figs 2, 3, Fig. S4) was analysed by measuring the proportion of BrdU⁺ signal to the Sox2⁺ signal area. The regions of interests (ROIs) were first drawn manually following the outlines of the Sox2⁺ VZ, then divided into floor plate, roof plate, ventrolateral and dorsolateral domains. The signal was analysed in each embryo from several sections at 40 µm intervals from anterior to posterior midbrain, and values of these sections averaged for each embryo for the statistical analyses.

The Olig2⁺ and Pou4f1⁺ cell numbers (Fig. 2) were counted as individual particles per section using the ‘Analyse Particles’ function after automated thresholding. Three midbrain sections at 40 µm intervals from each embryo were analysed. FoxJ1⁺, S100b⁺, PCNA⁺ and cyclin D1⁺ cells were quantified as a proportion of cells in ventricular zone using manually drawn ROIs as described above.

DA neurons at E18 were counted as individual particles from each section (Fig. 3). Sections spanning the entire mDA domain in the antero-posterior direction were collected and analysed at 40 µm intervals from each embryo (19-24 sections per embryo depending on the slight changes in the sectioning angle). First, the ROI was drawn manually using the boundaries of the TH signal for each section. Then Pitx3⁺ particle number was counted within this ROI. If the ‘Analyse Particles’ function, even combined with the ‘Watershed’ function, was unable to reliably resolve areas in which DA neurons were densely clustered together, those sections were re-counted manually. For analysing different areas of the midbrain, the same datasets were divided into three regions – anterior, middle and posterior parts – and each part compared between the genotypes. E14 DA neurons were counted from sections from anterior, middle and posterior midbrain using same counting parameters in Fiji as for the E18 data.

BrdU labelling in DA neurons (Fig. 3) was counted manually, checking each BrdU⁺ cell for signal in all channels, and again sections spanning the entire midbrain were analysed.

Statistical analyses were performed in Minitab (minitab.com) and R (v.4.4.1). The homoscedasticity was analysed using Levene's test, and normal distribution using Ryan–Joiner or Shapiro–Wilk tests. One-way ANOVA with Tukey simultaneous test for differences of means and two-tailed Student's t-test were carried out with default parameters. Welch's ANOVA with Games–Howell was used for data that were normally distributed but not homoscedastic. The details of statistical tests and P-values can are found in Table S5.

For lesioning of DA neurons in the periaqueductal grey region, mice (n=2 for each genotype and condition; males and females) were deeply anaesthetised with 2-2.5% isoflurane (KDG9623, Baxter Medical AB), the skin on top of the skull shaved and disinfected, and they were placed in a stereotaxic frame on a heating pad and given local anaesthesia with subcutaneous injection of Marcaine (169912, AstraZeneca). They were then unilaterally injected with either 2 µg (freebase) freshly prepared 6-OHDA (Sigma-Aldrich, H4381) in 0.01% ascorbic acid (Sigma-Aldrich, A92902) or the same volume of only 0.01% ascorbic acid at 0.1 µl/min. The injection coordinates were AP −3.1, ML 0.1, DV −3 relative to bregma (Paxinos and Franklin, 2013). The needle was kept in place 1 min before and 5 min after the injection. For relieving post-operative pain, the mice were given Rimadyl (14920, Zoetis) subcutaneously directly after the injection and 24 h after the operation. Periaqueductal grey lesioning did not lead to any motor defects or other health issues.

Tamoxifen (Sigma-Aldrich, T5648) was dissolved at 20 mg/ml in peanut oil (Sigma-Aldrich, P2144) and incubated at 37°C in a tube roller for 2-3 h, protected from light, until it was completely dissolved, and the solution was used fresh. For the activation of Cre in the adult mice, FoxJ1^CreERT2 R26^TdT animals received 3 mg tamoxifen by oral gavage on three separate days with a minimum of a 2-day interval. For Lmx1a^CreERT2 activation in the embryos, the pregnant females were treated with 3 mg tamoxifen by oral gavage when embryos were E9.5. BrdU (Roche, 10280878001) was given either as an intraperitoneal injection in PBS (100 mg/kg of body weight) or in drinking water (0.8 mg/ml, with added sucrose to cover the taste). The BrdU water bottle was protected from light and water changed every 3 days.

We thank Belinda Pannagel and Javier Avila-Cariño for expert technical assistance with FACS; Katarina Gradin for genotyping of mice; Juha Partanen for Nkx2-2^Cre mice and ISH probes; Jonas Frisén for FoxJ1^CreERT2 mice; Jonas Muhr for the Sox6 antibody; Danijal Topcic for technical advice with neurosphere cultures; as well as both current and former members of Thomas Perlmann and András Simon labs for discussions and comments. The computations were performed on resources provided by SNIC through Uppsala Multidisciplinary Center for Advanced Computational Science (UPPMAX) under Project SNIC 2017/7-348. We acknowledge support from Science for Life Laboratory, the National Genomics Infrastructure, NGI and Uppmax for providing assistance with massive parallel sequencing and computational infrastructure. We thank especially Lokeshwaran Manoharan for the bioinformatic analyses of the ependymal data set.

The peer review history is available online at https://journals.biologists.com/dev/lookup/doi/10.1242/dev.204421.reviewer-comments.pdf