Fstl1 is involved in the regulation of radial glial scaffold development

To better understand the development of radial glial scaffolds and to address the function of Fstl1 during cortical morphogenesis, Fstl1 was ablated by crossing the EIIa-Cre line with Fstl1 ^fl/fl . In situ hybridization showed a dramatic reduction of Fstl1 mRNA in the VZ and the pia mater in Fstl1 ^−/− brains at E16.5 (Fig. 1a–b’). The disruption efficiency was also confirmed by quantitative real-time polymerase chain reaction (qRT-PCR) (Fig. 1c). Because the homozygous null pups died within a few hours of birth, the observations were conducted at embryonic stages. HE staining at E18.5 showed a slight reduction in the thickness of the cortical wall in the Fstl1 ^−/− embryos compared with that of the WT cortical walls, but the overall architecture of the brain did not appear to be severely affected (Fig. 1d–f’). The cortical neurons, especially the upper-layer neurons, were not tightly arranged, but the overall lamination did not appear to be obviously disturbed (Fig. 1e–f’). These findings suggest that the loss of Fstl1 affects the accurate localization of cortical neurons.

To assess which groups of neurons were affected during cortical development, the layer-specific markers Brn2 (II-III), Cux1 (layers II-IV), Ctip2 (V-VI), and Tbr1 (VI) were used. At E18.5, many Cux1⁺ and Brn2⁺ cells still resided in the intermediate zone (IZ) and subplate in the Fstl1 ^−/− mice (Fig. 2a’ and b’), but in the WT mice, the majority of the Cux1⁺ and Brn2⁺ cells had arrived at the upper cortical plate (Fig. 2a and b). Double immunostaining for Cux1 and Ctip2 also showed that at E18.5, a number of Cux1⁺ neurons remained in deeper cortical areas in the Fstl1 ^−/− mice compared with the WT mice (Fig. 2c–d’). The statistical analysis showed that at E18.5, only 32 % of the Cux1⁺ neurons had arrived in the upper area in the Fstl1 ^−/− cortex compared with 44 % in the WT cortex; in addition, 35 % of the Cux1⁺ neurons were accumulated in the lower area in the Fstl1 ^−/− cortex but only 24 % were in the WT cortex (Fig. 2e). However, the total number of Cux1⁺ neurons did not differ significantly between the Fstl1 ^−/− and WT mice (Fig. 2f). Our data showed that the migration of upper-layer projection neurons was disrupted after Fstl1 deletion. In the deeper-layer projection neurons, the distributions of Tbr1⁺ layer VI neurons and Ctip2-high expression layer Va neurons in the Fstl1 ^−/− mice were comparable with those in the WT mice (Fig. 2g–i’), and the total numbers of Tbr1⁺ neurons and Ctip2 high-expression neurons did not differ (Fig. 2j and k). The lamination of the deeper layers seemed similar between the Fstl1 ^−/− and WT mice, and there was a clear boundary between layer VI and layer Va (Fig. 2g–i’).

To examine whether the radial migration of deeper-layer neurons was delayed at earlier embryonic stage and finally recovered at E18.5, cortices were harvested at E14.5 and E15.5 and immunostained with the deeper-layer-specific markers Tbr1 and Ctip2. Our data revealed that Tbr1⁺ and Ctip2⁺ neurons in the Fstl1 ^−/− cortex were positioned appropriately in the emerging cortical plate at E14.5 and in the deeper part of the cortical plate at E15.5, comparable with the positions observed in the WT cortex, as shown in Fig. 2l–o’. Thus, the migration of deeper-layer projection neurons was not obviously disrupted. Taken together, our data showed that the migration of upper-layer projection neurons was more severely impaired at perinatal stages following Fstl1 ablation.

To test whether the ablation of Fstl1 resulted in cortical interneuron defects, we examined the distribution of interneurons derived from the medial ganglionic eminence (MGE) and caudal ganglionic eminence (CGE). In situ hybridization for the neuropeptide somatostatin (SST) and the LIM homeobox protein 6 (Lhx6) showed that the distribution of MGE-derived interneurons was unchanged in Fstl1 ^−/− mice compared to WT mice. Immunostaining for COUP-TFII also indicated that the overall migration and lamination of CGE-derived interneurons were similar to those of the WT interneurons at E18.5 (data not shown). Our data demonstrated that disruption of Fstl1 had no obvious effects on cortical interneuron development.

To further elucidate the role of Fstl1 in neuronal migration, 5-bromo-2-deoxyuridine (BrdU) birth-dating was employed. BrdU was injected at E12.5 to label the deeper-layer neurons or at E14.5 to label the upper-layer neurons. The brains were then harvested at E18.5. In the mutants, the distribution of deeper-layer BrdU⁺ neurons born at E12.5 was similar to that in the WTs (Fig. 3a and a’). In the WT mice, 46.3 ± 2.3 % of the BrdU⁺ cells that were labelled at E14.5 populated the upper part of the cortex; however, this percentage was reduced to 35.3 ± 3.0 % in the Fstl1 ^−/− mice. The BrdU⁺ cells were distributed more evenly throughout the cortex, and a population of BrdU⁺ cells remained in the deeper cortical layers and IZ in the Fstl1 ^−/− mice. In the WT mice, approximately 25.5 ± 1.6 % of the BrdU⁺ cells were distributed in the deeper layers, but this percentage increased to 33.3 ± 1.3 % in the Fstl1 ^−/− mice (Fig. 3b and b’). To further assess the migration of deeper-layer neurons at earlier developmental stages, BrdU was injected at E12.5, and the brains were then harvested at E14.5. In the mutants, the distribution of BrdU⁺ neurons born at E12.5 was similar to that in the WTs (Fig. 3c and c’). These results indicate that Fstl1 dysfunction resulted in mis-positioning of the upper- but not deeper-layer projection neurons.

To examine whether Fstl1 affects neuronal proliferation, acute BrdU labelling was performed at E14.5 and E16.5 (Fig. 3d–e’). The numbers of BrdU-labelled cells did not differ between the Fstl1 ^−/− pups and their WT littermates (E14.5, 172.4 ± 5.1 for WT, n = 5 vs 173.5 ± 7.2 for Fstl1 ^−/− , n = 5; E16.5, 69.5 ± 2.4 for WT, n = 3 vs 73.5 ± 2.1 for Fstl1 ^−/− , n = 3) (Fig. 3f and g), and this result was further confirmed by immunostaining for the pan-proliferation marker Ki67 and the mitotic phase marker phospho-histone H3 (pHH3) (Fig. 3h-i’), which suggests that the deletion of Fstl1 does not affect the proliferation of progenitor cells. Because upper-layer cortical neurons are primarily derived from intermediate progenitor cells (IPCs) rather than directly from RGCs, we next examined the number of IPCs by immunostaining for Tbr2 [13] and found no obvious differences (Additional file 1: Figure S1), which indicates that the neurogenesis of the upper-layer neurons was also not affected. Additionally, immunostaining for activated Caspase 3 was performed at E14.5, and an obvious excess of cell death was not detected (data not shown).

Because the upper-layer neurons reach their final destinations by migrating along radial glial scaffolds [14], we next investigated whether the positional defect of the upper-layer neurons in the Fstl1 ^−/− mice was due to an abnormality in the RGCs. Immunostaining for nestin and BLBP, which are standard markers for RGCs, was performed. At E12.5, the nestin-stained radial glial scaffolds in the Fstl1 ^−/− mice appeared to be similar to those in the WT mice, with many radially distributed processes extending throughout the cortical wall (Fig. 4a and a’). However, although many long, radially distributed processes spanned the entire cerebral cortex in the WT mice at E14.5, numerous short radial processes failed to extend perpendicularly through the entire cerebral cortex towards the pial BM in the Fstl1 ^−/− mice, and the radial scaffolds were not fasciculated as tightly as in the WT mice (Fig. 4b‐c’). Similar defects in the radial glial scaffolds were detected at E15.5 and E16.5 (Fig. 4d‐i’). Immunostaining for BLBP further confirmed the defects at E15.5 and E16.5 (Fig. 5a‐d’). At E18.5, the defects in the RGC processes were still apparent (Fig. 5e and e’). To examine whether the radial glial progenitor pool was affected, immunostaining for the transcription factor Pax6, which labels RGCs [13, 15], was performed, and the progenitor pool was found to be relatively normal (Fig. 5f and f').

To further study the dysmorphic radial glia in the absence of Fstl1, the radial glial scaffolds were labelled with DiI at E15.5 and observed with a standard fluorescence microscope (Fig. 6a and a’) or a confocal microscope (Fig. 6b–d’). In the WT brains, the RGC processes were oriented parallel to one another and orthogonal to the pial BM (Fig. 6b and c), and their basal endfeet were less branched (Fig. 6a–d). However, in the Fstl1 ^−/− cortices, many RGC processes were not parallel to one another (Fig. 6a’-d’). Furthermore, the branches of the basal endfeet were more intricate and were located farther away from the pial surface than those in the WT cortices (Fig. 6c–d’). Thus, the disruption of Fstl1 results in abnormal RGC development.

A defining feature of RGCs is their apical-basal polarity. Recent studies have indicated that disruption of the polarity of RGCs disturbs both the morphology of radial glial processes and neuronal migration [5]. The tumour suppressor APC is highly concentrated at the tips of the processes and in the somas of RGCs and is required for the maintenance and extension of the radial glial processes [7]. To test whether the polarity of RGCs was altered after Fstl1 ablation, immunostaining for APC was conducted. At E12.5, APC was highly expressed at the pially directed tips of the RGCs and on the apical surface of the VZ in both the Fstl1 ^−/− and WT mice (Fig. 7a and a’). However, at E14.5, a broader APC-enriched band was observed in the Fstl1 ^−/− subpial region compared to that in the WT region (Fig. 7b‐d), which is consistent with the radial glial endfeet defect observed through DiI labelling (Fig. 6a‐d’). Upon examination of the integrity of the pial BM, the attachment site for the RGCs’ basal endfeet [16], no abnormalities were detected in the Fstl1 ^−/− mice, as demonstrated by laminin A immunoreactivity (Additional file 2: Figure S2 and data not shown). These findings demonstrated that the endfeet of the pially directed, polarised radial glial processes could not be established correctly in the absence of Fstl1, although the pial BM itself was not affected.

Previous studies have shown that β-catenin, which is localized at the apical adherens junctions, modulates appropriate expansion of the radial progenitor population [4, 17]. ZO-1, which is a tight junction-associated protein, is enriched at the lateral membrane domain of the interphase radial glial apical domain [18]. To examine whether the apical polarity of the RGCs was affected, immunostaining for β-catenin and ZO-1 was performed. Relatively normal β-catenin and ZO-1 expression patterns were observed in the Fstl1 ^−/− brains (Fig. 7e–f’), which suggests that the apical polarity was not grossly altered. Taken together, the results showed that the ablation of Fstl1 led to disruption of the basal rather than the apical polarity of RGCs.

Because Fstl1 mRNA is found in both the pia mater and the VZ [12], we sought to determine whether FSTL1 derived from the VZ or the pia mater is more important for the maintenance of polarity in RGCs. Emx1 ^IREScre mice [19] were employed to disrupt Fstl1 only in the VZ and not in the pia mater (Fig. 8a–b’). Unexpectedly, no apparent abnormalities were observed in the cortex of Emx1 ^IREScre ; Fstl1 ^fl/fl mice (Fig. 8c and c ’). Both upper-layer and deeper-layer neurons were nicely laminated (Fig. 8d–h’). BrdU birth-dating showed that in the Fstl1 ^−/− mice, the final distribution of upper-layer neurons born at E15 was comparable with that in the WT mice (Fig. 8g and g’). At E15.5, the radial glial scaffolds, viewed by immunostaining for BLBP, appeared similar to those of the WT mice, with many radially distributed processes that spanned the entire cortical wall and were oriented parallel to one another (Fig. 8i and i’). The Pax6⁺ RGC pool was also unchanged (Fig. 8j and j’). Furthermore, neither the distribution nor the intensity of APC staining was affected (Fig. 8k and k’). These results demonstrated that conditional disruption of Fstl1 in the VZ was not sufficient to cause the phenotype observed in the Fstl1 ^−/− brains. Together, our data suggest that FSTL1 derived from the VZ is not necessary for the maintenance of the morphology of RGC processes and that FSTL1 derived from the pia mater is more important.

BMP signalling may be one of the downstream targets of FSTL1 [20, 21], and the deletion of BMP7 results in less radial glia attachment to the pial BM, reduced cortical thickness and impaired neuronal migration [22, 23]. Therefore, we first examined the activity of the canonical BMP signalling pathway by examining the level of phosphorylated SMAD1/5/8. Western blot analysis indicated that the phosphorylation levels in Fstl1 ^−/− mice were similar to those in WT mice (Additional file 3: Figure S3A). BMP7 qRT-PCR further confirmed that the transcription level of BMP was also unchanged (Additional file 3: Figure S3B). The AKT/PKB signalling pathway is another target of FSTL1. Previous studies have shown that FSTL1 protects cells from apoptosis and induces angiogenesis via phosphorylated AKT [24‐26]. However, qRT-PCR and immunoblotting assays showed that the level of AKT was not changed in Fstl1 ^−/− brains (Additional file 3: Figure S3A and 3C). Furthermore, the activity of the AKT signalling pathway, as monitored by AKT phosphorylation at Ser473, was also unchanged (Additional file 3: Figure S3A). A previous study has shown that Etv5 is a critical target of the FGF/MAPK signalling pathway and has a role in the specification of RGCs [27]. However, neither the expression pattern nor the mRNA level of Etv5 differed between the WT and Fstl1 ^−/− mice (Additional file 3: Figure S3D, D’ and E). Reelin, which is secreted by CR cells, has been reported to be important for the development of radial scaffolds [28, 29]. However, neither the transcriptional level of reelin nor the distribution of Reelin⁺ CR cells was altered in Fstl1 ^−/− mice (Additional file 3: Figure S3F to G’). Additionally, we examined the mRNA levels of Cdc42 and GSK3β, which have been reported to modulate distinct aspects of radial glial process organization and function, but no differences were detected between the WT and Fstl1 ^−/− mice (Additional file 3: Figure S3H and I). Integrins are critical for the integrity of the pial BM, to which the radial glial processes attach [30, 31]; however, the mRNA levels of Itgb1, Itga5, and Itga6 did not differ (Additional file 3: Figure S3J–L). We did not observe changes in the expression of genes in the BMP, AKT/PKB, Cdc42, GSK3β, integrin or reelin signalling pathways in the Fstl1 ^−/− mice, suggesting that FSTL1 may regulate radial glial development through a unique mechanism. Further study is required to elucidate this mechanism.

Previous studies have demonstrated that proper apical-basal polarity and tight anchorage to the pial BM are critical for the functions of RGCs [5, 7, 8, 32]. Radial glial development has been reported to be regulated by both intrinsic factors, such as asymmetric protein distribution, and extrinsic factors, such as growth factors and other diffusible ligands [8, 33, 34]. In recent years, several studies have shown that the meninges regulate the survival, proliferation, differentiation and neurogenic properties of RGCs by releasing diffusible factors, such as CXCL12, BMP7 and retinoic acid [9, 23, 35]. Here, we found that Fstl1 is required for the development of radial glial scaffolds and the basal polarity of RGCs. Interestingly, the disruption of FSTL1 in the VZ did not affect the morphology of radial glial scaffolds, indicating that FSTL1 derived from the pia mater rather than the VZ might play a more critical role during this process and that the role of Fstl1 in the development of radial glial scaffolds is most likely non-cell-autonomous. Additionally, previous studies have suggested that meningeal cells organize the pial BM, a critical anchor point for radial glial scaffolds. Impaired pial BM integrity leads to RGC scaffold detachment and the apoptotic death of RGCs. However, our data indicates that Fstl1 is not indispensable for the integrity of the pial BM during the overall embryonic stage. One possible explanation is that FSTL1 produced by the meninges might be critical for the interaction of RGC basal endfeet with the BM, although we cannot exclude the possibility that FSTL1 regulates the organization of the RGC scaffolds directly. Further studies that employ conditional disruption of Fstl1 in the pial BM will help to elucidate the details of the role of Fstl1 in the pial BM.

The finding that RGC proliferation was not affected by the ablation of Fstl1 was also notable. Furthermore, neither the RGC progenitor pool nor the number of IPCs was disturbed, which indicates that the differentiation of RGCs was also unaffected. Our findings demonstrate a new role of Fstl1 in the maintenance of radial glial scaffolds but not the proliferation and differentiation of RGCs and thus will increase understanding of the molecular mechanisms of RGC development.

Our results show that the deletion of Fstl1 led to mis-positioning of upper-layer projection neurons. Notably, at E18.5, the distributions of deeper-layer projection neurons were not obviously affected. The lamination of layers V and VI seemed normal. One possible explanation is that neuronal migration of all layer neurons, including the upper- and deeper-layer projection neurons, has been delayed. However, the results of our birth-dating and immunostaining of deeper-layer neurons at earlier embryonic stages demonstrated that the migration of deeper-layer neurons was normal after the ablation of Fstl1, indicating that Fstl1 is not required for the migration of deeper-layer projection neurons.

Upper-layer neurons are known to preferentially migrate to their final location in a radial-glia-dependent manner, but the deeper-layer neurons likely migrate in a radial-glia-independent manner [36, 37]. Thus, the impaired upper-layer projection neuron migration in the Fstl1 ^−/− mice was probably the consequence of the abnormalities in the RGCs. It is well known that cortical interneurons migrate tangentially from the ventral telencephalon to the neocortex in a radial-glia-independent manner during embryonic development and then reach their final destination postnatally by radial migration. Here, no distribution defects were observed in our mutants, suggesting that Fstl1 is not important for the development of interneurons. Our data will help to elucidate the mechanisms of upper-layer neuronal migration.

Recently, several studies have revealed that BMP signalling and the AKT pathway are the main downstream targets of FSTL1 [20, 21, 26, 38, 39]. However, in the present study, no changes were detected in either pathway after the ablation of Fstl1. Reelin is a key regulator of the development of RGCs [29, 40]. However, reelin expression was also unchanged. A likely explanation is that FSTL1 does not function via either the reelin-mediated pathway or the BMP or AKT pathways, and further studies are required to elucidate the molecular mechanism.