COUP-TFI mitotically regulates production and migration of dentate granule cells and modulates hippocampal Cxcr4 expression

The dentate gyrus (DG) is part of the hippocampal formation and the primary input site for excitatory projections to the hippocampus. Its development is more protracted than that of other cortical regions and involves the generation of a progenitor pool that remains active well beyond birth (Altman and Bayer, 1990a,c; Eckenhoff and Rakic, 1988; Li and Pleasure, 2005; Nowakowski and Rakic, 1981; Pleasure et al., 2000; Rakic and Nowakowski, 1981). The major cell type in the DG is the granule cell (GC), the progenitors of which originate from a restricted area of the medial pallium neuroepithelium, the DG neuroepithelium (DGN) or primary (1ry) matrix at around E13.5 in mice. DG progenitors travel along the secondary (2ry) matrix towards the pial side of the cortex and form the dentate migratory stream (DMS), which is composed of a mix of intermediate progenitor cells (IPCs) and postmitotic immature granule neurons. At the end of their migration, GCs accumulate in the DG anlage or hilus and establish a new germinative pool, called the tertiary (3ry) matrix. At this stage, the glial scaffold extends to the hippocampal fissure and pial surface, and directs the migration of dentate precursors (Urban and Guillemot, 2014). Although DG morphogenesis starts early in embryonic development, the vast majority of GCs are generated within the first two postnatal weeks and originate from the 3ry matrix (Muramatsu et al., 2007). Granule cells laminate in an outside-in gradient by which the oldest cells are positioned more superficially and the youngest more deeply, whereas neural stem cells become restricted to the subgranular zone (SGZ), which constitutes one of the two adult neurogenic niches of the mammalian brain (Altman and Bayer, 1990a,b; Altman and Das, 1965; Cowan et al., 1980).

One of the particularities of DG development is that neurogenesis and migration are not independent processes, even if controlled by transcriptional regulators and signalling pathways similar to those described for the neocortex (Hevner, 2016). Lef1, which is a mediator of Wnt signalling, controls the generation of GCs (Galceran et al., 2000), whereas Ngn2 maintains progenitors in an undifferentiated state, allowing them to amplify prior to differentiation into GCs, which is regulated by NeuroD1 (Galichet et al., 2008; Roybon et al., 2009). The IPC factor Tbr2 is required for overall GC neurogenesis and proper migration of Cajal-Retzius cells to the DG (Hodge et al., 2013, 2012). Although most of these factors are widely expressed in several regions of the cortex (Englund et al., 2005), Prox1 is specifically restricted to dentate GCs from early stages to adulthood. Prox1 controls GC amplification, differentiation and maturation (Galeeva et al., 2007; Lavado et al., 2010; Lavado and Oliver, 2007), and postmitotically defines GC identity over the hippocampal pyramidal-like phenotype (Iwano et al., 2012). Migration of GCs requires reelin signalling, expressed by Cajal-Retzius cells, and the Cxcl12/Cxcr4 pathway, which represents the major chemotactic system involved in DG morphogenesis. The ligand Cxcl12 (SDF1) is secreted by meningeal and Cajal-Retzius cells, whereas migrating GCs express the receptor Cxcr4 during prenatal stages (Berger et al., 2007). As a consequence, Cxcr4 mutant mice exhibit severe DG morphogenesis abnormalities, due mainly to defective GC neurogenesis, migration and final positioning (Bagri et al., 2002; Lu et al., 2002). However, despite these studies, relatively little is known about how factors and signalling molecules interact with each other in controlling GC differentiation and migration.

We have recently demonstrated that the nuclear receptor COUP-TFI (or Nr2f1), which acts as a strong transcriptional regulator (Alfano et al., 2014a), is required for proper hippocampal growth and function (Flore et al., 2017). COUP-TFI mutant mice possess a hypomorphic hippocampus, specifically altered in its septal/dorsal pole, and are afflicted by a selective spatial memory deficit (Flore et al., 2017). However, the cellular and molecular causes of this growth defect are not known. To address this issue, we have investigated the role of COUP-TFI during DG morphogenesis from the earliest stages of DG development to adulthood. We first show that COUP-TFI is expressed in the different populations of GC progenitors and postmitotic neurons throughout all stages of DG development. Mitotic disruption of COUP-TFI function leads to the loss of a large fraction of the GC pool and to distinct migratory defects and impaired laminar organisation, resulting in severe DG dysmorphogenesis. Cxcr4 expression is highly upregulated in the DGN and migrating progenitors of mutant DG. Similarly, forced Cxcr4 expression in the DG mimics migratory defects in wild-type embryos. Postmitotic COUP-TFI inactivation induces instead only mild and transient defects, with no changes in Cxcr4 expression. Overall, our study shows that COUP-TFI modulates Cxcr4 expression levels in GC progenitors, and unravels a novel role for this transcriptional regulator in DG morphogenesis.

To start addressing COUP-TFI function in DG development, we characterized its distribution in GCs at different stages. At E13.5, COUP-TFI is highly expressed in the DGN located just dorsal to the cortical hem (Fig. 1A) and, by E16.5, in the majority of proliferating and differentiating cells distributed along the three matrices (89% in 1ry, 82% in 2ry and 68% in 3ry matrix; Fig. 1B-C′). Indeed, COUP-TFI is expressed in 81% of proliferating Ki67⁺ cells, in 96% of Pax6⁺ radial glia progenitors and in 80% of Tbr2⁺ IPCs in the 1ry matrix (Fig. 1D-F′), and is highly colocalized with Prox1 in GC precursors and differentiating neurons (Fig. 1G,G′). Between birth and P14, COUP-TFI is maintained in all postmitotic GCs (Fig. 1H-J), as confirmed by a high rate of colocalization with Prox1 in GCs (Fig. 1K-M,Q), with the glutamatergic marker Tbr1 in differentiating GCs (Fig. 1N-Q), and with the neuronal marker NeuN (Rbfox3) at P7 and P14 in mature GCs (Fig. 1Q-S). Thus, COUP-TFI labels all phases of GC neurogenesis in the developing DG from cycling progenitors to postmitotic and mature neurons.

To directly investigate COUP-TFI role during DG development from its onset to its final formation, we used the COUP-TFIf/f^EmxCre conditional line, referred to throughout this study as EmxCKO (Fig. 2A,A′), in which COUP-TFI is completely absent in all cells of the developing and postnatal DG and in all Prox1⁺ GCs (Fig. 2B-D′). Although no particular growth defect was observed at E16.5, the whole EmxCKO DG shows a slightly reduced volume at birth (Fig. 2E,E′,I), in which its septal region is 51% smaller than the control (Fig. S1A), in line with the selective hippocampal reduction we have previously described (Flore et al., 2017). By P7, the total volume of the mutant DG is decreased to 68% of its normal size (Fig. 2I′) with both blades reduced in length and thickness, and accumulation of ectopic cells in the hilus (Fig. 2F′). The growth defect remains evident at P14, with a total reduction of 45%, shorter blades (Fig. 2G,G′,I″) and a more pronounced defect in the septal pole compared with the temporal pole (Fig. S1A′,A″). Growth impairments are even more exacerbated in 2-month-old mutants: the DG is only 29% of its normal size (Fig. 2H,H′,J), the upper and lower blades show a 69% and 72% reduction in size, respectively (Fig. 2J′,J″), and the volume of the DG is dramatically affected along the septo-temporal axis (Fig. S1B). However, adult EmxCKO DG, in contrast to P7 and P14 EmxCKO DG, depict a rather normal layer organisation, with a sharp hilus/granule cell layer (GCL) boundary (Fig. 2H′), suggesting that ectopic hilar cells eventually reach their final position. Overall, these data indicate very little postnatal DG growth in EmxCKO mutants (Fig. 2K), and suggest that COUP-TFI plays a major role in promoting growth and expansion of the developing DG, particularly in its septal region.

To decipher the cellular mechanisms underlying the mutant morphogenetic defects, we evaluated the capacity of GC progenitors to properly expand at E16.5. The proliferation marker Ki67 indicates a 28% reduction in the number of cycling cells in the ventricular zone (VZ) of the 1ry matrix and a 51% reduction in the number of migrating progenitors in the EmxCKO DG septal primordium (Fig. 3A-A″), whereas no differences were found in the 1ry matrix of the temporal DG (Fig. S1C). This was also confirmed by a significant reduction in the number of proliferating cells in S-phase, as observed after a short pulse of EdU in E16.5 embryos (Fig. 3B-B″). To further understand the types of GCs affected in EmxCKO mutants, E16.5 DG was stained either with EdU injection and Pax6, to label proliferating progenitors, including most of the radial glia cells (Gotz et al., 1998) (Fig. 3C-C″), or with Tbr2 and the replication marker Mcm2, to label proliferating IPCs (Fig. 3D-D″). The amount of cycling progenitors and IPCs was drastically reduced in mutant mice (decreased by 43% for progenitors; and by 82%, 77% and 86% in the 1ry, 2ry and 3ry matrices, respectively, for IPCs) (Fig. 3C″,D″). By injecting EdU 24 h before sacrifice and labelling EdU⁺ cells with the proliferation marker Ki67 at E16.5, we calculated the amount of cells that were proliferating at the time of injection (EdU⁺) but had since exited the cell cycle (Ki67⁻) (Fig. 3E-E″). This cell-cycle exit index (EdU⁺Ki67⁻/EdU⁺) shows a 34% increase in EmxCKO 1ry matrix compared with control, indicating that these cells have precociously exited their cell cycle. Moreover, among the Tbr2⁺ population, 50% more Tbr2⁺Prox1⁺ cells were detected in the future DG (Fig. 3F-F″), supporting precocious differentiation of IPCs into GCs in mutant newborns. Finally, E14.5 embryos, pulse-chased with EdU and labelled with Prox1 at P0, show the same proportion of EdU⁺Prox1⁺ cells among all EdU⁺ cells in control and mutant DG (Fig. 3G-G″), confirming that early progenitors do normally adopt a neurogenic granular fate in EmxCKO mutants. Hence, our data indicate that COUP-TFI regulates expansion of both Pax6⁺ radial glia and Tbr2⁺ IPC pools in a timely manner during DG development (Fig. 3H).

Next, we evaluated whether abnormalities in the IPC pool are protracted during postnatal stages in mutant mice (Fig. 4). A decreased number of proliferating IPCs (Mcm2⁺Tbr2⁺ cells) are maintained in the 3ry matrix of EmxCKO DG at P0 (Fig. 4A-B″), indicating that even during the growth phase, DG cells maintain reduced proliferative capacities. Accordingly, 12% more Prox1⁺ GCs were observed in P7 EmxCKO DG (Fig. 4C,C′,F) and most probably become NeuN⁺ neurons (Fig. 4D,D′,F′). The proportion of NeuN⁺ neurons in the GCL is not affected in P14 EmxCKO DG, despite its reduced size (Fig. 4E,E′,F″), and granule neurons acquire a mature morphology, as highlighted by the reporter line Thy1-eYFP-H (Fig. 4G-H′). This suggests that COUP-TFI acts preferentially during the early prenatal phases of granule cell proliferation and differentiation rather than during the late postnatal phase of cell differentiation and maturation.

We hypothesized that, besides impairments in cell proliferation, the reduced postnatal growth observed in mutant DG might also be due to abnormal migration. Tbr2⁺ cells normally engage two distinct migratory paths at birth: one around the DG pole of the 3ry matrix to form the transient subpial neurogenic zone (SPZ) and one towards the hilus (Fig. 5A) (Hodge et al., 2013). In both regions, cells undergo further rounds of expansion and form the second germinative zone. In EmxCKO mutants, Tbr2⁺ IPCs abnormally migrate along the DMS: cells are more spread out compared with the narrow stream in controls, and more Tbr2+ ectopic cells are found migrating towards the prospective SPZ of the hilus (Fig. 5A,A′). By P7, numerous Tbr2⁺ cells are grouped at the lower blade tip in controls (Fig. 5B), but not in mutant DG, where ectopic cells are found either along the 2ry matrix or in the lower blade molecular layer (Fig. 5B′). At P14, Tbr2⁺ cells are aligned along the future adult SGZ niche (Fig. 5C), whereas mutant cells are abnormally dispersed within the GC and molecular layers (Fig. 5C′), as also confirmed by the altered radial distribution of Mcm2⁺ and Tbr2⁺ cells (Fig. 5D,D′). Ectopically located cells fail to properly laminate the GCL, as substantial amounts of Prox1⁺ cells are widely dispersed in the hilus or differentiate in loco, as confirmed by the presence of ectopic cellular aggregates in the molecular layer and/or in the area that resembles the DMS (Fig. 5E-F′). These clusters mature into Tbr1⁺ and NeuN⁺ granule neurons and are still evident at P14 (Fig. 5G′,H′). On the contrary, no GC lamination defects can be found in the temporal pole of EmxCKO DG, confirming a preferential alteration in septal DG (Fig. S1D,D′). Thus, COUP-TFI seems to control the migration of GCs along the DMS and within the septal DG.

To further decipher the mechanisms underlying the abnormal DG laminar organisation in mutants, we stained E13.5 to adult DG with brain lipid-binding protein (BLBP) and/or glial fibrillary acidic protein (GFAP) to follow radial glia cells that form the primary scaffold along the hilus and the secondary scaffold across the GCL. During embryonic stages, no obvious fibre morphology or orientation defects were detected in EmxCKOs (Fig. S2), similar to the neocortex (Alfano et al., 2011). At birth, GFAP⁺ fibres project radially across the hilus and progenitors migrate along their processes before reaching the SPZ (Fig. 6A,B). By contrast, GFAP⁺ fibres are abnormally oriented in mutant DG, and migrating Tbr2⁺ cells acquire a disorganized pattern within the forming dentate (Fig. 6A′-B′) that could partially explain the aberrant routes undertaken by Tbr2⁺ cells at this stage (Fig. 5A′). Next, we investigated the morphology of the secondary (or transgranular) radial glia scaffolding, which is required at postnatal stages for the GC subpial to granular transition (Brunne et al., 2010). By P7, radial glia fibres have reached the molecular layer, and their cell bodies (visualized by BLBP) accumulate along the inner GCL (Fig. 6C,D). This late scaffold is fully established at P14, and SGZ cells are now aligned at the lowermost border of the GCL (Fig. 6E). Radial glia cells appear poorly organized in P7 and P14 EmxCKO DG: their BLBP⁺ somata are abnormally spread within the GCL, and GFAP⁺ shorter processes lack the characteristic orientation (Fig. 6C′-E′). This abnormal morphology is maintained throughout adulthood, in which star-like mutant GFAP⁺ cells strikingly differ from the elongated radial processes observed in control DG (Fig. 6F-G′). Finally, we evaluated the capacity of adult stem cells to proliferate and observed a decrease of more than 50% of BrDU⁺ cells in EmxCKO DG (Fig. 6H-J). Taken together, loss of COUP-TFI in early DG progenitors affects the formation of a proper radial glia scaffold, which ultimately impinges on the migration, morphology and, possibly as a secondary consequence, final establishment of the adult neurogenic niche.

As COUP-TFI is expressed in both mitotic and postmitotic cells during DG development, we wondered whether absence of solely postmitotic COUP-TFI function would recapitulate part of the EmxCKO phenotype. To investigate this, the COUP-TFI^flox/flox mouse line was crossed to the Nex-Cre mouse, which acts in early differentiating cortical neurons (Alfano et al., 2014b; Goebbels et al., 2006) (Fig. 7A,A′). In the resulting progeny (herein referred to as NexCKO), COUP-TFI is maintained in the VZ of the 1ry matrix and in migrating Tbr2⁺ IPCs from E16.5 to P7, but depleted in pyramidal hippocampal neurons and in the majority of Prox1⁺ and Tbr1⁺ GCs (Fig. 7B-C′; Fig. S3), while being maintained in Gad67⁺ interneurons (Fig. S3D,D′). Thus, in contrast to the EmxCKO mouse mutant, COUP-TFI protein is maintained in cycling progenitors in the NexCKO mouse model, allowing us to assess its unique function in postmitotic neurons during DG morphogenesis.

Notably, mutant NexCKO DG show no significant size reductions from P0 to P14 (Fig. 7D-G; Fig. S4A-A″), even if they tend to have a slightly smaller volume than control littermates (Fig. 7G) and depict a rounded C-shape instead of the characteristic V-shape of littermate controls (Fig. 7E-F′; Fig. S4B,B′). In 2-month-old NexCKO brains, the DG volume is again slightly, but not significantly, reduced compared with control (Fig. S4C), and no statistically significant differences in the entire hippocampal volume and its septo-temporal distribution are found (Fig. S4D-D″). Overall, this indicates that COUP-TFI plays no major postmitotic role in global hippocampal morphogenesis.

Characterization of the proliferation and differentiation potentials of GC progenitors in E16.5 and P0 NexCKO DG revealed no defects in the number and distribution of cycling and differentiating Tbr2⁺ IPCs (Fig. S4E-I′). However, we found a slight disorganization of GFAP⁺ fibres in the primary radial scaffold at P0 (Fig. 7H-I′), even if not as severe as in EmxCKO (Fig. 6A-B′), and several Prox1⁺ cells are abnormally distributed in the forming hilus (Fig. 7J′). Similarly, glia fibres of the transgranular scaffold are slightly affected in the NexCKO mutants at P7, and BLBP⁺ somata are more disorganized in the GCL of mutants compared with controls (Fig. 7K,K′). However, these defects are rescued by P14, by which time BLBP⁺, as well as Mcm2⁺ and Tbr2⁺, cells are now properly aligned in the inner region of the GCL (Fig. 7L-N′). Moreover, both the position and number of NeuN⁺ mature GCs are similar to those in controls in NexCKO DG (Fig. S4J,J′), indicating that the minor abnormalities observed in the NexCKO pups are fully rescued by late developmental stages. Thus, the absence of postmitotic COUP-TFI has a minor effect on the organisation of the glia scaffold that does not seem to affect differentiation and migration of granule cells, and ultimately DG growth.

Finally, to assess whether cell death could partially contribute to the morphological defects observed in EmxCKO and NexCKO mutants, we labelled E16.5 and P0 DG with active-caspase 3 (Fig. S5), which marks cells undergoing apoptosis (Porter and Janicke, 1999). Even if no cell death can be detected in COUP-TFI mutants at E16.5 (Fig. S5A-A″,B), a significant amount of dying cells are found at P0 in septal, but not temporal, regions of EmxCKO and NexCKO mutant DG (Fig. S5C-D′). A concentration of dying cells is surprisingly localized in the subiculum of NexCKO brains (Fig. S5C″), most probably leading to the scattered subiculum cell layer and contributing to the abnormal rounded shape of NexCKO-affected DG (Fig. S5E″). Thus, this differential apoptotic pattern might partially contribute to the distinct morphological defects observed in the two strains of mutant mice.

In light of the similar DG defects observed in both Cxcr4 and EmxCKO mutant mice, we hypothesized that COUP-TFI-deficient GCs might abnormally respond to chemokine Cxcr4/Cxcl12 signalling (Bagri et al., 2002; Berger et al., 2007; Li et al., 2009; Lu et al., 2002). Cxcr4 is mainly localized in migrating progenitors and immature GCs (Fig. 8A-D), whereas its ligand Cxcl12 is normally expressed by meningeal and Cajal-Retzius cells around the dentate pole and in the hippocampal fissure (Fig. S6A-C). Notably, we found a drastic upregulation of the Cxcr4 transcript in the 1ry matrix, in migrating precursors along the DMS and in the forming DG of E16.5 EmxCKO embryos (Fig. 8A′). High transcript and protein levels are maintained in cells of P0 EmxCKO hippocampi, as also confirmed by quantitative reverse-transcriptase PCR (qRT-PCR) (Fig. 8B-C′,E,E′). Expression levels are similar to controls at P7 (Fig. 8D,D′), indicating that upregulation of Cxcr4 is mainly restricted to prenatal stages. Moreover, no changes in Cxcl12 transcript levels are observed from E16.5 to P7 EmxCKO (Fig. S6A-C′) and expression of both Cxcr4 and Cxcl12 is not altered in E16.5 NexCKO DG (Fig. 8F,F′; Fig. S6D,D′). To further support the role of COUP-TFI in regulating Cxcr4 expression in progenitors, we used a complementary approach by crossing the Cre-dependent hCOUP-TFI-floxed transgenic line (Alfano et al., 2014b) with Emx1-Cre, thus allowing overexpression of COUP-TFI in all hippocampal progenitors (Fig. 8G-H′). As expected from the loss-of-function data, high COUP-TFI levels downregulate Cxcr4 in the hippocampus (Fig. 8I,I′), confirming that COUP-TFI can modulate Cxcr4 expression in progenitor cells.

COUP-TFI has also been shown to regulate Cxcr4 expression in breast cancer cells (Boudot et al., 2014), suggesting that Cxcr4 might be a direct target of COUP-TFI during cell migration. To investigate this, we examined whether COUP-TFI protein could directly bind to the regulatory regions of the Cxcr4 locus. With the help of the ECR browser (Loots and Ovcharenko, 2007; Quandt et al., 1995), we found a highly conserved COUP-TFI-binding site in the 3′UTR region of the Cxcr4 locus (Fig. 8J). Chromatin immunoprecipitation (ChIP) on P0 hippocampal cells shows specific binding of COUP-TFI on the fragment amplified from the 3′UTR of Cxcr4 (Fig. 8J′). Thus, COUP-TFI modulates Cxcr4 expression levels by binding to its DNA sequence, most probably in a direct manner, although we cannot exclude that co-factors may also participate to this regulation (Fig. 8K).

Finally, as loss of Cxcr4 function seems to affect GC migration (Lu et al., 2002), we overexpressed Cxcr4 in wild-type hippocampal cells by in utero electroporation to evaluate whether high Cxcr4 levels would also affect their migration, as observed in EmxCKO mutants. First, we confirmed that our pCIG2-Cxcr4-IRES-GFP construct is indeed producing high levels of Cxcr4 protein in regions in which Cxcr4 is normally expressed at low levels (Fig. S7). Then, we quantified the distribution of electroporated GFP⁺ cells in CA1 and DG regions, and showed that cells containing high Cxcr4 depict a significant delayed migration in both areas, when compared with cells electroporated with a control pCIG2-IRES-GFP plasmid (Fig. 9). Indeed, a higher proportion of GFP⁺ cells is detected in the VZ of the CA1 and in the 1ry DG matrix, whereas a lower proportion of GFP⁺ cells is instead observed in the CA1 pyramidal cell layer and 3ry DG matrix, respectively (Fig. 9B′,D′). Hence, abnormally high levels of Cxcr4 alter hippocampal cell migration in a cell-autonomous way.

The DG is one of the few brain areas where progenitors proliferate and generate neurons throughout life. Understanding the molecular and cellular mechanisms governing its morphological development is thus crucial for a full comprehension of its function, including its capacity to generate new neurons throughout adulthood. In this study, we unravel for the first time a key function of the transcriptional regulator COUP-TFI in DG development. We have previously shown that COUP-TFI is required for proper growth and morphogenesis of the hippocampal septal pole and its connectivity to the entorhinal cortex (Flore et al., 2017), but the origin and mechanisms of these defects have not been elucidated. Here, we show that COUP-TFI is highly expressed in the dentate primordium and maintained in all precursors during migration and differentiation, thus representing one of the few transcription factors expressed throughout all stages of DG morphogenesis. Then, we directly assessed its function in DG progenitors or postmitotic granule cells and showed that COUP-TFI is required mainly in GC expansion and migration in the septal DG pole, but not in GC specification and maturation. Using loss- and gain-of-function approaches, we showed that COUP-TFI modulates Cxcr4 expression leading to abnormal migration of hippocampal precursors (summarized in Fig. 10), and that increased levels of Cxcr4 affect cell migration. Overall, our study contributes to the further dissection of the molecular signature required in the formation and morphogenesis of the DG.

It is well established that distinct developmental steps are required in DG morphogenesis during pre- and early postnatal development: proper patterning of the DGN by the nearby cortical hem, correct amplification of progenitors, organisation of the glia scaffolding and transformation of the tertiary matrix before full differentiation of DG neurons (reviewed by Li and Pleasure, 2005). All these events are controlled by the sequential expression of several transcription factors and signalling pathways (reviewed by Sugiyama et al., 2013; Urban and Guillemot, 2014). Our study reveals that COUP-TFI plays multiple roles during DG development. On the one hand, it is required for overall growth of the septal DG by specifically regulating the progenitor pool during pre- and perinatal stages of development. The reduced progenitor population will then impact on the establishment of the secondary germinative zone in the DG, thus compromising its ultimate expansion and growth. However, even if in reduced numbers, GCs can properly differentiate and mature in the absence of COUP-TFI, and the DG acquires an almost normal shape although smaller in size. This suggests that after exiting the cell cycle, the differentiation capability of a COUP-TFI-deficient cell is not altered, and that the major function of COUP-TFI is to temporally maintain the progenitor pool in a proliferative state.

On the other hand, COUP-TFI controls the behaviour of migratory progenitors along the DMS at prenatal stages and postnatally, within the hilus and GCL. In its absence, migrating progenitors undertake abnormal paths and form aggregates of cells that will differentiate in loco instead of reaching their target destinations (summarized in Fig. 10). This, together with increased apoptosis and a reduced progenitor pool, will strongly affect the growth and morphogenesis of the postnatal DG. Accordingly, the upper blade, where the earliest-born GCs are located, is more affected than the lower blade, in line with early migratory impairments. Even during the late outside-in transgranular radial migration, COUP-TFI-deficient cells are abnormally positioned within the GCL, leading to impairments in the distribution and proliferative capacity of SGZ cells. Very mild defects are observed in mutant differentiating GCs (Fig. 10), indicating that COUP-TFI acts primarily within the progenitor DG population. We also confirmed a stronger defect in the septal rather than temporal pole of COUP-TFI EmxCKO mutants, which is in agreement with the severe septal hippocampal growth defect described in adult mutants (Flore et al., 2017).

Several studies using Cxcl12- and Cxcr4-deficient animals have revealed major functions for these molecules in the prenatal development of the DG (Bagri et al., 2002; Lu et al., 2002). Cxcr4 is mainly expressed in rapidly dividing granule progenitors and precursors, and in immature GCs (Berger et al., 2007). In Cxcr4-deficient animals, progenitors in the DMS and DG are markedly decreased and differentiate prematurely, and postmitotic cells are found ectopically along the migratory route (Bagri et al., 2002; Lu et al., 2002). This phenotype is also partially reproduced in Tbr2 mutant mice, in which Cxcr4 expression levels are strongly decreased (Hodge et al., 2013), confirming that Cxcl12/Cxcr4 signalling is mainly required during early phases of DG development. In this study, we show by two independent methods that abnormally high levels of Cxcr4 also lead to delayed aberrant migration, suggesting that altered receptor expression can affect the response of these cells to normal hippocampal Cxcl12 signalling. In addition, Cxcr4 and GFAP expression significantly overlap between P3 and P8 during the reorganization of the hilus and in the forming GCL, indicating additional potential roles for Cxcr4 in the laminar organization of the DG during the first two postnatal weeks (Berger et al., 2007). Aberrant migration, abnormal radial scaffolding and premature neurogenesis are found in COUP-TFI mutant DG (Fig. 10), similar to mice devoid of Cxcl12, Cxcr4 or Tbr2 (Hodge et al., 2013; Li et al., 2009; Lu et al., 2002). Thus, our data suggest that, in progenitor cells, COUP-TFI controls the size and migration of the dentate progenitor pool by normally repressing Cxcr4 expression, and reveals that multiple transcriptions factors are required to properly maintain precise levels of the receptor Cxcr4 in DG progenitors during development.

COUP-TFI is well recognized as a key transcriptional regulator in areal and laminar organisation of neocortical development through its control of the radial migration of late-born neurons and its specification of sensory pyramidal neuron fate primarily in early postmitotic cells (Alfano et al., 2011, 2014a,b). Here, we show that COUP-TFI acts predominantly in progenitor cells by regulating the size of the progenitor pool and allowing proper migration of GC precursors. How can we explain this discrepancy in the mitotic versus postmitotic role of COUP-TFI between DG and neocortical development?

DG morphogenesis starts around mid-gestation in the mouse and substantially differs from other cortical regions. In the neocortex, neurogenesis and cell migration are two distinct processes occurring at different times and in distinct radial compartments (Florio and Huttner, 2014), whereas DG progenitors proliferate and produce new neurons as they migrate to the hilar region, where they continue their GC production until they reach their final residence in the SGZ (Hevner, 2016; Li and Pleasure, 2005). Thus, cell migration is prolonged in the DG when compared with cortical development, and cell proliferation and migration are two highly linked processes in the DG, whereas in the neocortex migrating cells do not proliferate. Thus, it is plausible that proliferation and migration closely influence each other in the DG and/or are controlled by similar mechanisms. In addition, as progenitors divide while migrating, they are continuously exposed to signalling molecules along their paths. COUP-TFI has been shown to be an important regulator of cell migration in vivo during forebrain development (Alfano et al., 2011; Touzot et al., 2016; Tripodi et al., 2004; Zhou et al., 2015), but also in vitro and in cancerous cells (Adam et al., 2000; Boudot et al., 2014; Le Dily et al., 2008). Here, we show that it modulates the correct expression levels of the chemokine receptor gene Cxcr4 during GC neurogenesis and migration. Thus, by maintaining a key role in cell migration, COUP-TFI is able to play different functions in distinct cell types according to the biological process in which it is implicated.

Generation of COUP-TFI conditional and overexpressing mice, as well as the Thy1-eYFP-H line have been previously described (Alfano et al., 2014b; Armentano et al., 2007; Harb et al., 2016). Midday on the day of vaginal plug formation was considered as embryonic day 0.5 (E0.5). All experiments were conducted according to French ethical regulations and received the approval from our local ethics committee (CIEPAL NCE/2014-209).

Postnatal mice were perfused with 4% paraformaldehyde (PFA) in PBS and post-fixed in 4% PFA at 4°C, for either 2 h for immunofluorescence or immunohistochemitry, or overnight for in situ hybridization experiments. Cryosections (16 µm) were processed for immunofluorescence or immunohistochemistry, as previously described (Alfano et al., 2014b) by overnight incubation at 4°C with the primary antibodies, followed by 2 h at room temperature with secondary antibodies (Table S1). Immunohistochemistry slides followed the standard avidin-biotin complex reaction procedure (Vector Laboratories), and staining was revealed using DAB Peroxidase Substrate (Vector) following the manufacturer's instructions.

Cxcr4 and Cxcl12 antisense RNA probes were labelled using a DIG-RNA Labelling Kit (Roche). In situ hybridizations were carried out on 16 µm cryosections as previously described (Alfano et al., 2014b).

Timed-pregnant females were injected intraperitoneally with 200 µl (2.5 g/ml) of EdU (FisherScientific) and revealed by using the EdU Click-It Alexa Fluor 647 kit (FisherScientific). Slides were stained for immunofluorescence with the appropriate antibodies just prior to EdU revelation. For adult neurogenesis, BrdU (Sigma) was administered intraperitoneally to 2-month-old mice at 100 g/kg for six consecutive days and sacrificed the day after.

Cryosections (16 µm) were post-fixed in 4% PFA for 10 min and incubated in the staining solution (0.025% thionin, 0.025% Cresyl Violet, 100 mM sodium acetate, 8 mM acetic acid, in deionized H₂O) for 5 min at room temperature. Visualization was carried out in the de-coloration solution (80% ethanol, 20% deionized H₂O and few drops of acetic acid). Vibratome sections (50 µm) from 2-month-old adult mice were processed as previously described (Flore et al., 2017).

Pictures were taken using an epifluorescence microscope (Zeiss Imager.M2) or a confocal microscope (Zeiss 710) for immunofluorescence and with a bright-field microscope (Leica DM6000B) equipped with a colour camera for immunohistochemistry, in situ hybridization and Nissl staining.

Total RNA (1 mg) was reverse-transcribed using Superscript III First-Strand Synthesis System for RT-PCR (Invitrogen). Amplified cDNA was quantified using KAPA SYBR FAST Master Mix (Kapa Biosystems) on a LightCycler II 480 (Roche) (see supplementary Materials and Methods for further details).

Hippocampi were dissected from 14 wild-type mice at P0 and chromatin immunoprecipitation was performed as previously described (Harb et al., 2016). Anti-COUP-TFI antibody (Alfano et al., 2011) was used for immunoprecipitation. Binding of COUP-TFI to the Cxcr4 sequence was tested by PCR amplification using primers designed to recognize the putative binding site (Table S2).

In utero electroporation was performed on E14.5 hippocampal neuroepithelium as previously described (Pacary and Guillemot, 2014) by trying to target the DGN and using the following parameters: four 40 V pulses, P(on) 50 ms, P(off) 1 s. The following plasmids were used: pCIG2-IRES-GFP (Heng et al., 2008) (n=7) or pCIG2-Cxcr4-IRES-GFP (n=9). The latter one was produced by cloning a PCR-amplified Cxcr4 DNA sequence into the EcoRI and XmaI sites of the pCIG2-IRES-GFP plasmid. Brains were collected at E18.5 and processed as for immunofluorescence.

DG volumes of at least n=3 EmxCKO or NexCKO were calculated and compared with their respective littermate controls. The volume of the adult DG and/or hippocampus was evaluated with the NIH ImageJ Software and analysed by a two-way ANOVA for repeated measures, and Duncan's post-hoc test. Cell counts were performed on at least three consecutive rostral sections for each analysed brain using the counting tool of Adobe Photoshop CS6. A spreadsheet software and a two-tailed paired Student's t-test were used to analyse statistical significance between mutant and their controls (*P<0,05; **P<0,01; ***P<0,001). All graphs represent mean+s.e.m. A detailed description can be found in supplementary Materials and Methods.

We thank K. R. Jones for the Emx1-Cre, K. A. Nave for the Nex-Cre and M. Tsai for the CAGGS-loxP-STOP-loxP-hCOUP-TFI mouse lines, as well as M. C. Tiveron for the Cxcl12 and Cxcr4 plasmids. We are also grateful to Eya Setti for her technical help, to S. De Marchis and S. Bonzano for their suggestions on the manuscript, and to the whole Studer lab for fruitful discussions.