Molecular control of two novel migratory paths for CGE-derived interneurons in the developing mouse brain

In rodents, different interneuron (IN) subtypes are produced in spatially and temporally distinct regions in the subpallium that consist of the ganglionic eminences (GEs) and preoptic area (POA) domain (Butt et al., 2005; Fogarty et al., 2007; Gelman et al., 2009; Miyoshi et al., 2010; Xu et al., 2004). The medial ganglionic eminence (MGE) represents the major source (∼70%) of cortical INs (Anderson et al., 2001; Jiménez et al., 2002; Marín, 2013; Wichterle et al., 2001). The majority of these early-born MGE-derived INs express either parvalbumin (PV; PVALB – Mouse Genome Informatics) or somatostatin (SST) and migrate dorsally to populate deep layers of the neocortex (Valcanis and Tan, 2003; Wichterle et al., 2001; Xu et al., 2004). The caudal ganglionic eminence (CGE) is responsible for producing late-born INs (∼30% of all INs) (Miyoshi et al., 2010; Nery et al., 2002; Rudy et al., 2011; Vitalis and Rossier, 2011), which preferentially integrate superficial layers of the neocortex, whereas the POA contributes to a minor pool of cortical INs (Gelman et al., 2009, 2011). The serotonergic ionotropic receptor 5HT3aR (HTR3A – Mouse Genome Informatics) is expressed in the neocortex in all CGE-derived GABAergic INs, such as bipolar calretinin (CR; CALB2 – Mouse Genome Informatics)-, vasoactive intestinal peptide (VIP)-, cholecystokinin (CCK)- and reelin (RLN; RELN – Mouse Genome Informatics)-expressing cells, including multipolar or neurogliaform INs expressing neuropeptide Y (NPY), but not in PV- and SST-positive INs (Murthy et al., 2014; Lee et al., 2010; Vucurovic et al., 2010).

Molecular determinants of the MGE, such as the Dlx homeobox genes Lhx6, Sox6 and Nkx2.1, are crucial for the specification and/or migration of the PV- and SST-positive subsets of INs (reviewed by Kessaris et al., 2014; Marín, 2013); however, less is known about the factors and molecular mechanisms controlling the fate determination and migration of CGE-derived cells. The orphan nuclear receptors COUP-TFI (also called Nr2f1) and COUP-TFII (Nr2f2) were the first transcriptional regulators found to be enriched in the CGE and required in the migration of GABAergic INs (Tripodi et al., 2004). Whereas COUP-TFII acts mainly in directing CGE-derived cells to the caudal migratory stream (CMS) (Cai et al., 2013; Kanatani et al., 2008; Yozu et al., 2005), COUP-TFI is required for the correct balance between MGE- and CGE-derived INs in the developing neocortex (Lodato et al., 2011b), but its role in migration is less known. 5HT3aR is expressed in migrating and mature CGE- and POA-derived cells (Lee et al., 2010; Vucurovic et al., 2010) and is cell-autonomously required for the migration and positioning of RLN⁺ INs in the neocortex (Murthy et al., 2014). Finally, SP8 and PROX1 are transcription factors expressed principally in subpopulations of lateral ganglionic eminence (LGE)/CGE-derived cortical INs (Ma et al., 2012; Rubin and Kessaris, 2013; Rudy et al., 2011) and a recent report has demonstrated a crucial role for PROX1 in the embryonic and postnatal acquisition of CGE-derived cortical IN properties (Miyoshi et al., 2015).

To reach their final destination, early postmitotic INs start migrating through distinct routes: the MGE-derived cells tangentially migrate dorso-laterally towards the cortex before reaching their final laminar positions by radial migration (reviewed by Guo and Anton, 2014; Marín, 2013), whereas CGE-derived cells seem to colonize the cortex primarily through its caudal pole (Yozu et al., 2005). However, CGE-derived INs are not only dispersed all over the cortex but are also found in various brain regions in the adult, including the hippocampus and amygdala (Chittajallu et al., 2013; Lee et al., 2010; Miyoshi et al., 2010; Nery et al., 2002; Vucurovic et al., 2010; Yozu et al., 2005); how they reach their final positions during development is still unclear (Tanaka and Nakajima, 2012). In addition, apart from COUP-TFII (Kanatani et al., 2008, 2015), the molecular mechanisms controlling the tangential migration of CGE-derived cells remain to be elucidated.

In this study, we show that CGE-derived INs use additional paths other than the CMS to reach their final positions. Two novel caudo-rostral streams, one located laterally and the other more medially, allow CGE-derived INs to migrate either dorsally towards the LGE or ventrally towards the POA. PROX1, SP8 and COUP-TF transcription factors are all expressed in these streams with SP8 and COUP-TFII being predominant in the lateral stream, and COUP-TFII in the medial and caudal streams. Inactivation of COUP-TFI affects the rate of SP8⁺ migrating cells in the lateral migratory stream (LMS) and COUP-TFII⁺ cells in the medial migratory stream (MMS), but has no effect on the CMS. The laminar distribution of CGE-derived INs, such as RLN⁺, single and double VIP⁺CR⁺ cells, is altered in the somatosensory cortex of adult COUP-TFI mutant mice, suggesting that COUP-TFI controls the distribution of CGE-derived INs by regulating SP8 and COUP-TFII expression within their migratory paths. Together, this study reveals the presence of two novel CGE-derived paths involved in populating distinct structures of the brain and contributes to deciphering the molecular mechanisms required in the decision-making steps of interneuron tangential migration.

To understand how CGE-derived INs reach their multiple targets, we used the 5HT3aR-GFP mouse transgenic line, previously reported to label the totality of CGE-derived cortical INs in embryos and in adult mice (Inta et al., 2008; Lee et al., 2010; Murthy et al., 2014; Rudy et al., 2011; Vucurovic et al., 2010). Embryonic day (E) 11.5 to E18.5 5HT3aR-GFP⁺ brains were cut in a horizontal plane allowing an overview of all three GE structures in one section (Fig. 1A). In these views, the cortical tissue or pallium is located laterally and the thalamus medially (Fig. 1B). Several sections along the dorso-ventral axis were analysed and the distribution of GFP⁺ cells compared with coronal antero-posterior sections of the same age (Fig. S1).

High levels of GFP are observed in the pallium from E11.5 to E13.5, but these cells are also positive for TBR1 (Fig. S2A-B‴), indicating that they represent glutamatergic preplate and subplate cells (Hevner et al., 2001) transiently expressing 5HT3aR-GFP. Double RLN⁺GFP⁺ cells in the marginal zone of E11.5 and E13.5 embryos represent Cajal–Retzius neurons (Fig. S2A-B‴; yellow arrowheads) maintained until postnatal stages, as previously described (Chameau et al., 2009; Engel et al., 2013). At the same time, GFP⁺ cells migrating though the intermediate and marginal zones express the early GABAergic marker calbindin (Fig. S2C-C‴), indicating that at this stage 5HT3aR-GFP⁺ interneurons invade the cortical plate.

Horizontal and coronal sections show that GFP⁺ cells are detected in the LGE and in some laterally and dorsally migrating cells (unfilled arrowheads in Fig. 1C-E and Fig. S1B-E). As MGE-derived cells do not express 5HT3aR (Lee et al., 2010; Vucurovic et al., 2010), these cells might either be generated locally and/or derive from the POA (Fig. S1B′-E′, unfilled arrowheads). Notably, E11.5 horizontal sections at ventral levels also reveal a stream of GFP⁺ cells between the CGE and MGE (Fig. 1C′, red arrowhead), which increases in intensity at E12.5 and E13.5 (Fig. 1D′,E′). In addition, a few GFP⁺ cells lateral and caudal to the CGE appear at E12.5 (Fig. 1D,D′, yellow and white arrowheads, respectively), and will form two prominent streams at E13.5, one migrating caudo-dorsally (Fig. 1E,E′; Fig. S1B″-E″) and previously described as the CMS (Yozu et al., 2005), and another one laterally between the CGE and LGE (Fig. 1E, yellow arrowheads), here termed the LMS. The other path between the CGE and MGE becomes very prominent at E13.5 (Fig. 1E,E′, red arrowheads) and is referred to as the MMS throughout this study. Thus, 5HT3aR-GFP⁺ cells seem to undertake a lateral, medial and/or caudal path before reaching the cortex by dorsal migration.

All the streams attain their highest intensity at E15.5 (Fig. 1F,F′; Fig. S1E-E″) in accordance with the peak of CGE cell production (Miyoshi et al., 2010), and leading processes in the dorsal LMS and ventral MMS are oriented rostrally (Fig. 1F,F′, white arrowheads in high magnification views) in line with a caudal-to-rostral migration. At E18.5, 5HT3aR-GFP⁺ cells have reached the neocortex, hippocampus and amygdala (Fig. 1G,G′; Fig. S1F-F″) and the olfactory bulb through the rostral migratory stream (RMS) (Fig. 1G′, arrow), as previously described (Inta et al., 2008). Taken together, our time course on horizontal and coronal sections of the 5HT3aR-GFP mouse line revealed two CGE-derived additional migratory paths besides the caudal one (Fig. 1B) in the subpallium of E11.5 to E15.5 embryos.

To confirm a caudo-rostral migratory direction of 5HT3aR-GFP⁺ cells in the LMS and MMS, the CGE of an E13.5 5HT3aR-GFP embryo was homotopically grafted into a non-GFP littermate, and horizontal organotypic slices were incubated for 2 days in vitro (DIV) (Fig. 2A). GFP⁺ cells migrating away from the grafted site take several paths, including the lateral, medial and caudal ones (Fig. 2B-B‴). Direction of their leading processes confirmed a caudal-to-rostral type of migration (Fig. 2A; Fig. 2B′-B‴, red arrows). To confirm CGE-derived caudo-rostral migration, we injected the lipophilic tracer DiI or introduced the cell tracker CMTPX loaded on resin beads into the CGE of E14.5 5HT3aR-GFP horizontal organotypic slices (Fig. S3A). Upon internalization by the cells, the fluorescent DiI will be retained and the CMTPX will be converted to a fluorescent compound (Haugland, 2002). Brain slices were then analysed for the presence of the tracer and expression of GFP in the different streams (Fig. S3B). DiI- and CMTPX-traced cells were found in several GFP⁺ cells in the CMS, LMS and MMS (Fig. S3C-E″ for DiI; Fig. S3F-L′ for CMTPX), but not in cells expressing NKX2.1, a marker for the MGE/POA (Sussel et al., 1999; Xu et al., 2008) (Fig. S3K-N′). This indicates that GFP⁺ cells migrating towards the MGE do originate from the CGE and not from the POA. Taken together, our three independent tracing approaches confirm that 5HT3aR-GFP⁺/CGE-derived cells migrate caudo-rostrally from the CGE to the LGE through the LMS and from the CGE to the MGE through the MMS (Fig. 2A; Fig. S3A).

Next, we assessed whether the different migrating populations express a unique transcription factor or a combination of CGE-specific transcription factors (Fig. 3). COUP-TF nuclear receptors are preferentially restricted to the CGE from E13.5 and maintained in CGE-derived cortical INs (Cai et al., 2013; Kanatani et al., 2008; Lodato et al., 2011b; Tripodi et al., 2004). The transcription factor SP8 is expressed in CGE-derived cortical INs and in dorsal LGE (dLGE)-derived olfactory bulb INs (Ma et al., 2012), whereas PROX1 is highly expressed in CGE-derived cortical INs and in a subpopulation of POA-derived INs (Rubin and Kessaris, 2013; Miyoshi et al., 2015).

COUP-TFI and/or COUP-TFII proteins are localized in different proportions within the GFP⁺ migrating streams at E15.5 with a high representation of single COUP-TFII-expressing cells in the LMS (41%) compared with 24% in the MMS (Fig. 3A-D). In the CMS, 29% of GFP⁺ cells express only COUP-TFII and 14% are positive for both COUP-TFI and COUP-TFII (Fig. 3D), revealing that the CMS has the greatest proportion of double COUP-TF-expressing cells compared with the other migrating streams. By contrast, SP8 and/or PROX1 are detected at a much lower level in the CMS and MMS, and are instead highly represented in the LMS, with 70% of GFP⁺ cells expressing SP8, 46% PROX1, and 38% both SP8 and PROX1 (Fig. 3E-H). Thus, whereas COUP-TF members are well expressed in all three streams, SP8⁺ and double SP8⁺PROX1⁺ cells are preferentially located in the lateral path (LMS) during the peak of CGE-derived IN production.

Previous studies have shown that SP8 is also expressed in the dLGE, which will give rise to INs of the olfactory bulb and amygdala (Jiang et al., 2013; Li et al., 2011; Ma et al., 2012; Stenman et al., 2003; Waclaw et al., 2006, 2010). To distinguish between the two sources of SP8⁺ cells migrating either ventrally or dorsally, we labelled SP8⁺GFP⁺ cells with ER81 (ETV1 – Mouse Genome Informatics), a transcription factor specific for olfactory bulb-derived INs (Stenman et al., 2003), on E15.5 coronal sections and in the adult olfactory bulb (Fig. S4). Triple-positive cells are only found in the dLGE at rostralmost levels (Fig. S4A-A‴) and in the granule layer of adult olfactory bulbs (Fig. S4E-E‴). No GFP⁺ cells within the LMS express ER81 (Fig. S4B-D‴), and double SP8⁺GFP⁺ cells are observed migrating dorsally towards the cortex (Fig. S4B′-D′, white arrowheads). These data imply that SP8⁺ cells might originate from a local dLGE source and migrate towards the olfactory bulb, but also derive from the CGE and migrate dorso-laterally along the LMS before reaching the cortex.

Next, we estimated the contribution of CGE-derived 5HT3aR-GFP⁺ cells expressing PROX1 and/or SP8 in juvenile brains by homochronically grafting CGE-derived neurons from E14.5 5HT3aR-GFP embryos into wild-type host embryos (Fig. 4A) (Vucurovic et al., 2010). The distribution of GFP⁺/CGE-grafted cells was subsequently analysed in the somatosensory cortex, hippocampus and amygdala of postnatal day (P) 21 host brains (Fig. 4B). The majority of grafted cells (85%) populated several amygdaloid nuclei, whereas 9% and 6% were localized in the cortex and hippocampus, respectively (Fig. 4B). We then investigated the proportions of GFP⁺ cells expressing either SP8 or PROX1, or both, in P21 grafted brains (Fig. 4C). Notably, 56% of 947 5HT3aR-GFP⁺ integrated cells express PROX1 and 39% SP8 in the cortex, of which almost 30% were double positive. Single PROX1⁺ cells were well represented in the hippocampus and amygdala compared with single SP8- and double-expressing cells (Fig. 4D). Thus, our data show that a high proportion of CGE-derived cells at E14.5 reach amygdalar structures, and that within these transplanted GFP⁺ cells the majority express PROX1.

Our graft experiments illustrate the contribution of PROX1- and/or SP8-expressing CGE-derived cells in different brain structures at a unique time point; however, CGE cells start to migrate from E11.5 (Fig. 1; Fig. S1). Thus, to have an overview of all 5HT3aR-GFP⁺ cells expressing SP8, PROX1 and/or COUP-TFII in adult brains, we first analysed the percentage and laminar distribution of single and/or double SP8- and PROX1-expressing GFP⁺ INs in the somatosensory cortex of 2-month-old 5HT3aR-GFP mice (Fig. 5). We found that the majority (57%) of GFP⁺ cortical INs is double positive for SP8 and PROX1, whereas single PROX1⁺ (SP8^–) and single SP8⁺ (PROX1^–) cells account for 27% and <5%, respectively, of all GFP⁺ cortical INs (Fig. 5A-C,K). The different subpopulations are distributed fairly evenly in all layers, although the double-expressing cells tend to be located in deeper layers and the single-expressing cells in upper layers (Fig. 5L). This indicates that in the somatosensory cortex 84% of all 5HT3aR-GFP⁺ cells express PROX1, of which more than half are double positive for SP8.

Next, we investigated the proportion of 5HT3aR-GFP⁺ cells expressing PROX1, SP8 and/or COUP-TFII together with particular cortical IN subtype markers, such as RLN, VIP and CR, in the adult somatosensory cortex (Fig. 5D-J) (Rudy et al., 2011; Vucurovic et al., 2010). The populations of SP8- or PROX1-expressing cells colocalize with VIP and RLN in roughly equal proportions (24% and 26%, respectively, for SP8; 23% and 20% for PROX1) (Fig. 5D,E,H,I,M), but are co-expressed with CR in only 13% of GFP⁺ INs, whereas double COUP-TFII⁺RLN⁺ represent 18% of all 5HT3aR-GFP⁺ in the cortex (Fig. 5F,G,J,M). Regarding the laminar distribution, we found that within the PROX1⁺GFP⁺ cell population, layer IV includes a higher percentage of CR⁺ than RLN⁺ and VIP⁺ cells, which are instead more represented in layers I-III and VI (Fig. 5N). By contrast, within the SP8⁺GFP⁺ population, layer V has the highest percentage of VIP⁺ and the lowest preference of CR⁺, whereas the RLN⁺ population is almost evenly distributed in all layers (Fig. 5N). Together, CGE-derived 5HT3aR-GFP⁺ cells expressing PROX1, SP8 or COUP-TFII are distributed along the whole radial extend of the cortex and consist mainly of RLN- or VIP-positive INs.

Moreover, 55% and 53% of GFP⁺ cells express PROX1 in the basolateral (BLA) and medial amygdala (MeA), respectively (Fig. S5A-B‴,D), compared with a minor SP8⁺ or double PROX1⁺SP8⁺ population (Fig. S5D). Finally, 38% express PROX1⁺ and 21% express SP8⁺ within the GFP⁺ cells located in the hippocampus, (Fig. S5C-D). Thus, PROX1 is highly expressed not only in the CGE-derived cells reaching the neocortex, as previously described (Miyoshi et al., 2015), but also in the 5HT3aR-GFP⁺ cells migrating to the amygdaloid complex and the hippocampus.

Next, we assessed whether transcription factors expressed in the caudo-rostral streams might be directly involved in the migration and specification of CGE-derived cortical INs. Because Sp8 conditional mice fail to show any evident interneuronal defects (Ma et al., 2012) and Prox1 conditional mice became available only very recently (Miyoshi et al., 2015), we took advantage of our previously generated COUP-TFI conditional knockout (CKO) mouse line (Lodato et al., 2011b), in which the COUP-TFI^flox/flox mouse is crossed to the Dlx5/6-Cre line to generate COUP-TFI::Dlx5/6-Cre (hereafter called TFICKO). We previously showed that absence of COUP-TFI leads to an increase of MGE-derived PV⁺ INs and resistance to GABA-dependent pharmacologically induced seizures (Lodato et al., 2011b). However, COUP-TFI gets restricted to the CGE from E13 onwards and its expression is maintained in CGE-derived IN subtypes (Lodato et al., 2011b). To assess its specific action on CGE-derived cortical IN specification and migration, we crossed the TFICKO mouse to the 5HT3aR-GFP reporter line and followed the migration and molecular specification of CGE-derived paths in horizontal sections (Fig. 6). At E13.5, when the MMS and CMS are well established, the density of postmitotic GFP⁺ cells in these two streams is decreased in TFICKO brains (compare Fig. 6A,A″,A‴ with B,B″,B‴, respectively), whereas the few cells located in the LMS and LGE are not altered (Fig. 6A′,B′). By contrast, at E15.5, during the peak of CGE production, the number of GFP⁺ cells increases by ∼50% in the LMS (Fig. 6C,C′,D,D′, E; TFICKO 154.0±23.7% versus control 100%; n=3; P=0.042) and by almost 25% in the mutant versus the control in the MMS (Fig. 6C″,D″) (Fig. 6E; TFICKO 123.8±9.9% versus control 100%; n=3; P=0.026). No statistical difference was detected in the amount of caudally migrating GFP⁺ cells (Fig. 6E, CMS).

We hypothesized that loss of COUP-TFI would affect cell proliferation in the CGE at E13.5 resulting in a reduction of CGE-derived GFP⁺ migrating cells. Indeed, more proliferating cells are detected in the subventricular (SVZ) but not ventricular (VZ) zone of mutant embryos (Fig. 6F-H; TFICKO 146.3±9.2% versus control 100%; n=3; P=0.065), in accordance with the selective inactivation of COUP-TFI in the SVZ (Lodato et al., 2011b). This implies that CGE precursors initially proliferate more and accumulate in the SVZ before being released at E15.5 and allowed to migrate along the LMS and MMS, as observed in Fig. 6D. Thus, COUP-TFI both controls the rate of proliferating cells in the SVZ and promotes CGE migrating cells to take a caudo-rostral direction.

We next assessed whether COUP-TFI controls the expression of SP8, PROX1 and/or COUP-TFII along the different CGE-derived GFP⁺ migratory streams described in this study (Fig. 1B). E15.5 horizontal 5HT3aR-GFP sections were immunostained with SP8, PROX1 or COUP-TFII antibodies (Fig. 7A-I′) and the density of cells double positive for GFP and one of these markers was evaluated in the LMS, MMS and CMS (Fig. 7J). The number of GFP⁺SP8⁺ cells was increased twofold in the LMS of COUP-TFI mutants (Fig. 7A,A′,J; TFICKO 152.7±33.0 cells versus control 74.4±4.0 cells; n=6; P=0.019), whereas no differences were found in the PROX1⁺ and COUP-TFII⁺ GFP-expressing populations (Fig. 7B-C′,J). By contrast, only the COUP-TFII⁺GFP⁺ population was significantly increased in the MMS of COUP-TFI mutants (Fig. 7F,F′,J; TFICKO 53.1±4.9 cells versus control 34.7±6.0; n=6; P=0.045), and not the PROX1⁺ and SP8⁺ subpopulations (Fig. 7D-E′,J). As expected from the data shown in Fig. 6E, no changes in the number of GFP⁺ cells expressing SP8, PROX1 or COUP-TFII were observed in the CMS (Fig. 7G-I′,J). Together, these data indicate that COUP-TFI normally restricts the number of CGE cells taking the lateral path towards the LGE by modulating SP8 expression and the medial path towards the MGE by controlling COUP-TFII.

The different paths undertaken by CGE-derived neurons might be correlated with specific cortical IN subtypes in the adult cortex. To test this hypothesis, we first evaluated the percentage of the different subtypes of CGE-specific INs in control and COUP-TFI mutant adult somatosensory cortices, and subsequently the same subtypes expressing PROX1, SP8 or COUP-TFII within the 5HT3aR-GFP⁺ cell populations (Fig. 8A,A′; compare Fig. 5C-J for controls with Fig. 8B-I for TFICKOs). Overall, we observed that the total number of 5HT3aR-GFP⁺ cells is not drastically different in control and adult COUP-TFI CKO mice, even if mutants tend to have a slightly higher number of CGE-derived GFP⁺ INs in the somatosensory cortex (Fig. 8J). Notably, among the CGE-derived GFP⁺ cell populations, the RLN⁺ population of cells increased from 27.3±2.1% to 37.1±1.6% (n=3; P=0.002) in adult TFICKO cortices, particularly in layers II-IV and VI (Fig. 8K,L), whereas the small number of double RLN⁺SST⁺ cells remained the same within the GFP⁺ population (data not shown) (Lee et al., 2010), indicating that it is mainly the RLN⁺ population, comprising neurogliaform cells (Miyoshi et al., 2010), that is affected in the absence of COUP-TFI. Notably, although the single CR- and VIP-expressing populations are only slightly increased (Fig. 8K), the double CR/VIP-expressing GFP⁺ subpopulation is statistically larger in TFICKO cortices (Fig. 8A,A′,M,N; TFICKO 12.4±1.5% versus control 8.0±1.4%; n=3; P=0.035), particularly in layers IV and V (Fig. 8M).

To further evaluate whether the IN subpopulations affected in the mutants are associated with a particular transcriptional code, we quantified the percentage of 5HT3aR-GFP⁺ subtype IN populations expressing either SP8, PROX1 or COUP-TFII and reported their laminar distribution (Fig. 8N-P). No drastic effects were detected in the GFP⁺ population expressing SP8, PROX1 or COUP-TFII or even double SP8⁺PROX1⁺ in adult TFICKO brains (Fig. 8K,N). Similarly, the overall percentage of VIP⁺, RLN⁺ or CR⁺ populations expressing SP8, PROX1 or COUP-TFII was not altered within the GFP⁺ mutant populations (Fig. 8N). However, we found that the VIP⁺ subpopulation expressing SP8 is statistically increased, whereas the double PROX1⁺CR⁺ subgroup is decreased, in layer VI of COUP-TFI CKO brains (Fig. 8O,P). No laminar differences were found in the SP8/RLN-, PROX1/RLN- and COUP-TFII/RLN-expressing populations (data not shown), indicating that the increased RLN population observed in the absence of COUP-TFI is independent of these transcription factors. Thus, COUP-TFI normally modulates the number of RLN⁺ and double CR⁺VIP⁺ CGE-derived cortical INs, and controls the balance between SP8/VIP- and PROX1/CR-expressing cells in layer VI.

Previous reports showed that cells located in the CGE can reach several telencephalic regions, such as the cerebral cortex, the hippocampus and the amygdala (Nery et al., 2002; Lee et al., 2010; Vucurovic et al., 2010; Yozu et al., 2005), but little was known about how CGE-derived INs were sorted to reach these final targets. In this study, we provide novel evidence that CGE-derived mouse INs disperse from the CGE using distinct migratory paths in the subpallium before reaching their final locations. We also show that several transcriptional regulators required in IN specification and expressed at high levels in the CGE are represented at different ratios within the subpallial streams. Perturbation of one of these transcriptional regulators, COUP-TFI, affects the two caudo-rostral streams and the laminar distribution of distinct CGE-derived IN subtypes. Thus, our study contributes to our understanding of the complex migratory patterns of CGE-derived INs during development.

Why do CGE-derived INs need to use so many streams to reach their final targets? We propose at least two main reasons based on our data and other studies. First, because the CGE not only produces INs migrating to the cortex (including piriform and enthorinal cortex), but also to the hippocampus and amygdala (our study; Nery et al., 2002; Vucurovic et al., 2010), we propose that CGE-derived cells take distinct tangential routes to reach their specific targets (Fig. 9). This is already substantiated by the CMS, which directs CGE cells to more caudal brain structures, such as the hippocampus (Chittajallu et al., 2013; Yozu et al., 2005). Here, we propose that the MMS populates the amygdala, as already suggested by short-term in vitro grafts (Vucurovic et al., 2010) and now supported by this study: (1) the majority of E14.5 CGE-grafted cells populate the amygdala in juvenile brains; (2) the massive cohort of GFP⁺ cells migrating medially is abundant at E12/E13, a stage when CGE-derived cortical INs normally represent only a small fraction (Lee et al., 2010; Miyoshi et al., 2010; Nery et al., 2002; Vucurovic et al., 2010); and (3) the MMS is highly represented by cells expressing COUP-TFII, which plays a relevant role in amygdala patterning (Tang et al., 2012) and which drives POA-derived cells to the medial amygdala (Kanatani et al., 2015).

The second reason why so many paths depart from the CGE might be a temporal one. It is known that CGE/late-migrating INs, in contrast to MGE/early-migrating INs, which populate the cortex in an inside-out mode by temporally matching the integration of projection neurons, colonize the cortex independently of their birthdate (Miyoshi et al., 2007, 2010). The use of spatially and temporally regulated paths might thus allow CGE-derived INs a sequential integration into their correct laminar locations. Indeed, the different paths arise and have their peaks at distinct ages: the MMS is the earliest to start (at E11.5), followed by the LMS at E12.5 and the CMS around E13. The distinct paths might also help in sorting MGE- versus CGE-derived cells with identical birthdate, as radial distribution during cortical plate invasion occurs several days later in CGE-derived populations than in those derived from the MGE (Miyoshi and Fishell, 2011). Hence, we propose that the laminar distribution of distinct cortical IN lineages might be also controlled in the subpallium through individual tangential routes, and not only at the cortical plate level, as previously suggested (Lodato et al., 2011a; Miyoshi and Fishell, 2011).

In contrast to the MMS, the LMS might contribute to populating the cerebral cortex, in line with its presence during the peak of CGE-derived cortical IN invasion (Lee et al., 2010; Miyoshi et al., 2010, 2015; Vucurovic et al., 2010). SP8 and/or PROX1 are expressed at high levels in the CGE and maintained in the LMS towards the dLGE, and a conspicuous amount of double SP8⁺PROX1⁺ cells are found in GFP⁺ cells of the cortex expressing typical CGE-derived IN subtype markers, such as VIP, RLN and CR. This suggests that the majority of PROX1⁺ cortical INs described recently (Miyoshi et al., 2015) might co-express SP8 and that single and double SP8/PROX1-expressing INs might have different properties and functions in the brain. Indeed, PROX1⁺ cells deriving from the CGE also contribute to the hippocampus and amygdala by using either the streams described in this study or other not yet identified paths. In addition, SP8⁺ cells are also found in the dLGE before the arrival of LMS cells (this study; Ma et al., 2012), suggesting that the dLGE and CGE represent two independent sources of SP8⁺ INs. Thus, whereas double SP8⁺ER81⁺ cells in the dLGE contribute to olfactory bulb INs (Inta et al., 2008; Jiang et al., 2013; Li et al., 2011; Waclaw et al., 2006) and to a lesser extent to amygdala INs (Waclaw et al., 2010), double SP8⁺PROX1⁺ cells in the LMS might be the main source of SP8⁺ cortical INs (Ma et al., 2012) (Fig. 9).

Our data show that COUP-TF members are highly represented in all three streams. Interestingly, COUP-TFI controls the rate of CGE-migrating cells rostrally to the LGE and MGE, but not to the caudal pole. This is very different from COUP-TFII, which is crucial for driving CGE cells caudally (Kanatani et al., 2008, 2015). Thus, whereas COUP-TFI restricts the number of CGE cells migrating rostrally, COUP-TFII promotes CGE cells to take caudal directions. This suggests that COUP-TFI and COUP-TFII might have minor redundant functions in the CGE during the choice of migratory routes, possibly by controlling distinct guiding molecules and/or receptors. We also found that COUP-TFI acts on the LMS population expressing SP8 and on the MMS population expressing COUP-TFII. The regulation of COUP-TFI on Sp8 expression has been well documented in the cortex (Alfano et al., 2014b; Borello et al., 2014), and control of COUP-TFI on COUP-TFII expression levels has been recently shown in the hippocampus (Flore et al., 2016). Thus, a similar mechanism in the CGE might limit the cohort of LMS SP8⁺ cells reaching the cortex, and compensate for the lack of COUP-TFI in the MMS, as previously proposed for the hippocampus, as both COUP-TFs have similar binding sites and could thus compete for the same target genes (Alfano et al., 2014a).

Our thorough analysis in adult COUP-TFI mutants has unravelled discrete differences in the CGE-derived GFP⁺ population of cortical INs. In accordance with the increased embryological SP8⁺ LMS population, we found more SP8⁺VIP⁺ cells in layer VI of the adult mutant somatosensory cortex. In the same layer, fewer PROX1⁺CR⁺ cells were detected, suggesting that COUP-TFI plays a particular function in CGE-derived cells of layer VI. This does not imply that COUP-TFI has no effect on other IN subtypes, as it limits the number of double CR⁺VIP⁺ cells in layers IV and V and RLN⁺ INs in layers II-IV and VI. It is plausible that these changes might be mediated by altered PROX1 function because this factor plays a crucial role in the specification of these IN subtypes (Miyoshi et al., 2015), and/or by COUP-TFII highly expressed in the CR⁺ populations (Varga et al., 2015; Reinchisi et al., 2012). We could not find any significant changes in the proportion of PROX1⁺VIP⁺, PROX1⁺CR⁺, PROX1⁺RLN⁺ or COUP-TFII⁺RLN⁺ INs in mutant brains even if their values tend to be increased. We hypothesize that the small amount of IN subtypes expressing GFP and a particular molecular marker can easily vary between samples, or, as in the case for COUP-TFII, compensatory mechanisms might act between embryological and adult stages. Together, our data show that COUP-TFI acts independently in distinct MGE- and CGE-derived IN subtypes to form a proper cortical inhibitory circuit and that in its absence mice become more resistant to GABA-dependent pharmacologically induced seizures (Lodato et al., 2011b).

Overall, this work reveals two novel migratory paths for CGE-derived INs in the mouse subpallium. These routes express a distinct combination of transcription factors and contribute to the laminar specification of individual subtypes in the neocortex and to specific amygdaloid nuclei. We thus propose that the existence of spatially and temporally regulated paths in the subpallium helps to direct the different IN populations towards their brain targets.

The 5HT3aR-GFP transgenic line was provided by the GENSAT consortium and conditional IN-specific COUP-TFI CKO mice were generated as previously reported (Lodato et al., 2011b) and propagated in a C57BL/6J (Charles River) inbred strain. We used COUP-TFI^flox/+, COUP-TFI^flox/flox or Dlx5/6-Cre-positive (Cre⁺) mice as controls, because they show no phenotypic abnormalities (Stenman et al., 2003; Armentano et al., 2007). We found no phenotypic differences between genders. Genotyping was performed as previously described (Lodato et al., 2011b). Timed pregnant embryos were obtained by overnight mating and the following morning was counted as E0.5. All experiments were conducted according to French ethical regulations and this project received the approval from our local ethics committee (CIEPAL NCE/2014-209).

Mice were perfused with 4% buffered paraformaldehyde (PFA) and adult brains were post-fixed in 4% PFA for 12 h at 4°C. Decapitated heads of embryos (E11.5 to E18.5) were fixed in 4% PFA for 12 h at 4°C and then brains were post-fixed in 4% PFA for 2 h at 4°C.

Vibratome sections were processed for free-floating immunofluorescence (IF), as previously described (Lodato et al., 2011b) with the following primary antibodies: anti-calbindin (rabbit, 1:500, Swant CB-38a); anti-calretinin (rabbit, 1:5000, Swant 7699/4); anti-calretinin (mouse, 1:500, Millipore MAB1568); anti-COUP-TFI (mouse, 1:100, Abcam ab41858); anti-COUP-TFII (rabbit, 1:500; Tripodi et al., 2004); anti-ER81 (rabbit, 1:2000, gift from Silvia Arber); anti-GFP (rabbit, 1:1000, Molecular Probes A11122); anti-GFP (chicken, 1:800, Abcam ab13970); anti-NKX2.1 (rabbit, 1:300, Santa Cruz sc13040); anti-PROX1 N terminus (goat, 1:700, R&D Systems AF2727); anti-PROX1 C terminus (rabbit, 1:1000, Abcam 11941-100); anti-RLN (mouse, 1:500, Millipore MAB5364); anti-SP8 (goat, 1:500, Santa Cruz sc104661); anti-TBR1 (rabbit, 1:500, Abcam ab31940) and anti-VIP (rabbit, 1:500, ImmunoStar 20077). The following secondary antibodies were used: Alexa Fluor 488-conjugated anti-chicken; Alexa Fluor 488-conjugated anti-rabbit; Alexa Fluor 488-conjugated anti-mouse; Alexa Fluor 594-conjugated anti-rabbit; Alexa Fluor 594-conjugated anti-mouse; Alexa Fluor 594-conjugated anti-goat; Alexa Fluor 594-conjugated anti-rat; Alexa Fluor 647-conjugated anti-rabbit; Alexa Fluor 647-conjugated anti-mouse; Alexa Fluor 647-conjugated anti-rat (Invitrogen; all 1:300). All experiments were repeated at least three times.

Timed pregnant females received a single intraperitoneal injection of 5-ethynyl-2′-deoxyuridine (EdU) (50 mg/kg) 1 h before sacrifice and embryos were collected at E13.5 and processed for EdU antibody staining according to the manufacturer's instructions (Click-iT EdU Imaging Kit C10340, Invitrogen). A set of three wild-type and COUP-TFI CKO embryos were examined for EdU⁺ cell distribution in the CGE, and cells in the VZ and SVZ were separately counted.

Brains were embedded in 3% low melting agarose and cut horizontally using a vibratome (Leica VT1000S) at 300 μm thickness. Slices were cultured for 2 or 4 days in Neurobasal Medium (Gibco) plus N-2 and B27 supplements (Gibco), as previously described (Tripodi et al., 2004). The Cell Tracker Red CMTPX (Molecular Probes C34552) was diluted to 1 mM and impregnated on resin beads (Bio-Rad 1401241). One bead or one DiI crystal was introduced in the CGE after 2 h of culture. After incubation, the bead was carefully removed and slices were fixed in 4% PFA for 4 h at 4°C and processed for IF, as indicated above. In co-culture experiments, the CGE explant from a GFP⁺ embryo was transplanted into the CGE of a non-GFP⁺ host littermate and incubated for 2 days, before analysis.

Homochronic transplantations were performed using E14/E14.5 donor and host embryos as previously described (Vucurovic et al., 2010). Three CGEs from E14/E14.5 5HT3aR-GFP donors were used to produce the cell suspension that was injected into the ventricles of host OF1 mice. The progeny was sacrificed at P21 and processed for IF, as described above. The distribution of 5HT3aR-GFP⁺ grafted cells was determined on coronal sections of three animals (947 GFP⁺ cells) in the somatosensory cortex, the hippocampus and the amygdala. All the data of each structure were pooled together in order to estimate a representative distribution of GFP⁺ cells in the brain.

Sections were imaged using a Leica DM6000B microscope, a Leica SPE confocal microscope or a Zeiss 710 confocal microscope and images were acquired using LAS AF (Leica) or ZEN (Zeiss) software and processed with Adobe Photoshop and ImageJ software.

For quantification of IN subpopulations at embryonic stages, a representative area in the migratory paths was selected for each animal (n=3) and single- or double-positive cells were counted. For quantification of IN subpopulations in the adult, three coronal anatomically matched sections in the somatosensory cortex along the rostro-caudal axis were selected from littermate mice. Digital boxes of fixed width were superimposed on each coronal section and subdivided into ten sampling areas (bins) with a dorso-ventral extent from the pial surface to the white matter. Deep layers (VI-IV) were assigned to bin 1 to 7 and upper layers (III-I) were assigned to bin 8 to 10, based on anatomical features. All cell detections and counts were performed using Photoshop software.

Cell quantification data and graphs were constructed using Microsoft Excel software. For each genotype, a mean value was calculated from all counted sections. Error bars represent the standard error of the mean (s.e.m.). Statistical significance was determined using two-tailed Student's t-tests (*P<0.05, **P<0.01, ***P<0.001).

We thank K. Campbell for the Dlx5/6-Cre line; Silvia Arber for the ER81 antibody; Eya Setti and Nadia Elganfoud for their technical help; and the whole Studer lab for fruitful discussions.