Perturbation of canonical and non-canonical BMP signaling affects migration, polarity and dendritogenesis of mouse cortical neurons

The cerebral cortex in the mammalian brain is composed of neurons arranged in six molecularly distinct layers, the formation of which takes place in an inside-out manner. Thus, the early-born neurons give rise to the lower layers (i.e. layers V and VI), whereas the late-born neurons form the upper layers (i.e. layers II-IV) with the exception of layer I, which comprises the earliest born Cajal-Retzius cells (Angevine and Sidman, 1961; Rakic, 1974; Marín-Padilla, 1998).

The diverse population of projection neurons in the cortex is generated from the neural progenitors present in the ventricular zone (VZ) and the subventricular zone (SVZ). Radial glia (RG) are a major class of neuronal progenitors that have long processes extending from the VZ to the pial surface, which provide support to migrating neurons. RG undergo asymmetric cell division, giving rise to daughter RG as well as intermediate progenitors (IPs), which in turn undergo symmetric cell division to give rise to newborn neurons (Noctor et al., 2004). These newborn neurons migrate through the SVZ to the intermediate zone (IZ), where they acquire a bipolar shape to start the process of radial migration (Tabata and Nakajima, 2003). The newborn neurons destined to populate the lower layers undergo glia-independent somal translocation, while the upper layer neurons undergo glia-dependent radial migration (Nadarajah and Parnavelas, 2002; Nadarajah et al., 2001). When the migrating neurons reach the cortical plate (CP), the leading process transforms into the apical dendritic arbor projecting towards the pia, while the trailing process is converted into the prospective axon (Hatanaka and Murakami, 2002).

Several signaling pathways have been implicated in regulating different aspects of cortex development. For example, fibroblast growth factors and Wnt proteins have been implicated in the regulation of patterning and proliferation of cortical progenitors (Raballo et al., 2000; Chenn and Walsh, 2002; Woodhead et al., 2006). Notch signaling has been shown to be important for maintenance of the RG population in the VZ (Gaiano et al., 2000). In addition, reelin (O'Dell et al., 2012; Jossin and Cooper, 2011; Franco et al., 2011) and retinoic acid (Choi et al., 2014; Siegenthaler et al., 2009) are known to regulate the layer-specific positioning of migrating neurons.

In addition to the signaling molecules mentioned above, bone morphogenetic proteins (BMPs) have been implicated in the regulation of development of the central nervous system (Kalyani et al., 1998; Liem et al., 1995). Studies carried out with dissociated cortical cells or with cortical explants suggest that BMPs regulate cell proliferation, apoptosis and neuronal differentiation of cortical progenitors (Li et al., 1998; Mabie et al., 1999). In addition, BMPs are also known to regulate the differentiation of astrocytes and parvalbumin-positive interneurons (Bonaguidi et al., 2005; Gross et al., 1996; Mukhopadhyay et al., 2009). Although BMPs and their receptors are expressed in the developing mouse forebrain from early stages (Furuta et al., 1997; Zhang et al., 1998; Caronia et al., 2010; Segklia et al., 2012), exploration of the role of BMP signaling in vivo in this context has been limited to studies demonstrating its role in dorsomedial patterning (Furuta et al., 1997; Hébert et al., 2002; Shimogori et al., 2004). The BMP receptor knockout mice had severe defects in early patterning of the forebrain, precluding an exploration of possible additional roles played by BMP signaling at later stages of cortical development.

To investigate the role of BMP signaling in the developing mouse cortex, we first detected the presence of active BMP signaling in the cortex at different developmental time points when cortical neurogenesis is taking place. We observed a dynamic, layer-specific activity of BMP signaling in the cortex, and based on this we hypothesized that BMP signaling is likely to regulate other aspects of cortical development besides patterning and the differentiation of astrocytes and interneurons. There is redundancy among multiple BMP ligands and receptors expressed in the mouse cortex (Caronia et al., 2010; Furuta et al., 1997), and hence complete abrogation of BMP signaling through genetic means in this context is difficult if not impossible. In order to investigate possible additional roles of BMP signaling in the developing mouse cortex, we therefore adopted an in utero electroporation-based strategy. We inhibited BMP signaling in a spatiotemporally regulated manner by delivering a dominant-negative version of BMP receptor type 1B into the mouse cortex by in utero electroporation, followed by analysis at different developmental time points. Interestingly, we found that canonical and non-canonical BMP signaling seem to differentially regulate the migration, polarity and dendritic branching of E15.5 born upper layer cortical neurons.

In order to investigate the role of BMP signaling during cortical development in the mouse, we carried out a spatiotemporal analysis of active BMP signaling. Binding of BMP ligands to their cognate receptors leads to phosphorylation of the downstream effector molecule SMAD1/5/8. Phosphorylated SMAD1/5/8 (pSMAD1/5/8) enters into the nucleus, with the help of the co-Smad SMAD4, to regulate the expression of the target genes of BMP signaling (Bandyopadhyay et al., 2013). Thus, to assess the presence of active BMP signaling in the developing mouse cortex, we carried out immunohistochemistry for pSMAD1/5/8 at embryonic day (E) 11.5 and 15.5 and at postnatal day (P) 0, 6 and 21 (Fig. 1). The pSMAD1/5/8 antibody used in this study has previously been demonstrated to specifically detect active BMP signaling in the context of developing bone (Prashar et al., 2014).

We observed active BMP signaling in distinct laminae within the cortex at different developmental stages. For instance, at E11.5, pSMAD1/5/8 is present in the VZ, which is a niche for cortical progenitors (Fig. 1B), and in the newly forming cortical preplate (PP), mostly comprising postmitotic neurons marked by Tbr1 (Fig. 1B′). Similarly, at E15.5, pSMAD1/5/8 was detected in the VZ, in the SVZ, the IZ (Fig. 1C) and in the CP marked by Ctip2 (Bcl11b) expression (Fig. 1C′). During later stages of cortex development, at P0, pSMAD1/5/8 could be detected in all cortical laminae including those marked by Ctip2 expression, but at very low levels in the IZ and VZ-SVZ (Fig. 1D,D′). At P6, pSMAD1/5/8 was observed mostly in the upper layers (Fig. 1E) marked by Brn2 (Pou3f2) expression (Fig. 1E′), and was also present in a mosaic pattern in the lower layers of the cortex (Fig. 1E, arrowheads). Finally, pSMAD1/5/8 immunoreactivity could be observed as late as P21 in the upper layers of the cortex (Fig. 1F). Although some background staining was observed, especially in the meninges, the pSMAD1/5/8 immunostaining in the cortical neurons is specific, as it is present in the nucleus (Fig. 1D,F, inset). These observations suggest that BMP signaling is active in the cortex in multiple laminae and across several developmental stages, and hence might be regulating distinct phenomena during the course of development.

To investigate the possible role(s) of BMP signaling during cortical development, we inhibited BMP signaling by overexpressing a dominant-negative version of BMP receptor 1B (dnBMPR1B) in the embryonic mouse cortex. Since dnBMPR1B lacks the entire intracellular region including the C-terminal serine-threonine kinase domain, when overexpressed it competes with the endogenous type I BMP receptors, thus acting as a dominant negative (Kawakami et al., 1996) (Fig. 2A).

We observed that pSMAD1/5/8 immunoreactivity was present in the CP, SVZ and IZ at E15.5, and at P0 in all cortical layers. Based on this, we hypothesized that BMP signaling might significantly influence the development of cortical neurons in multiple layers. We chose the time points E13.5, E14.5 and E15.5, when both lower layer and upper layer neurons are born, to investigate the role of BMP signaling in their development. We used in utero electroporation to deliver a construct expressing dnBMPR1B (pCAG-dnBMPR1B) together with a GFP-expressing construct (pCAG-GFP) into the mouse forebrain at these time points (Fig. S1A, Fig. 2B). These are henceforth referred to as test animals. Since pCAG-dnBMPR1B was co-electroporated with pCAG-GFP, the majority of GFP⁺ cells are expected to also express dnBMPR1B. We also electroporated pCAG-GFP alone into the mouse forebrain at these stages, henceforth referred to as control animals.

The newborn neurons generated from progenitors in the VZ migrate through the SVZ to the lower part of the IZ. Our analysis of the cortex 48 h after electroporation revealed that the distribution of GFP⁺ cells in the cortex of test animals electroporated at E15.5 was altered, as compared with that in the control animals (Fig. 2C,D). However, there was no significant alteration in the distribution of GFP⁺ cells in the test animals electroporated at E13.5 and E14.5 compared with the control (Fig. S1B-G). When we analyzed the cortices of animals electroporated at E15.5, 48 h after electroporation, we observed a considerable number of GFP⁺ cells remaining in the SVZ and in the lower part of the IZ in test animals (Fig. 2D). By contrast, in the control animals there were comparatively few GFP⁺ cells in the SVZ, and most of them had migrated to the IZ and were well distributed there (Fig. 2C). On quantification of the number of GFP⁺ cells in the VZ-SVZ and the IZ, the difference in the distribution of these cells between the test and control animals was found to be significant (control: SVZ 40.3±3.13%, IZ 59.6±3.1%; test: SVZ 65.2±5.8%, IZ 34.7±5.8%; Fig. 2E).

To investigate the molecular nature of the GFP⁺ cells present in the SVZ in the test animals, we carried out immunohistochemical analysis for Tbr2, which mostly labels cells in the SVZ. We found a significant difference in the percentage of Tbr2⁺ GFP⁺ double-positive cells in the SVZ between the test and control animals (Fig. 2F-G′, arrowheads; control 44.8±7.1%, dnBMPR1B 70±4%; Fig. 2H). Further, we analyzed the effect of inhibition of BMP signaling on cell proliferation by quantifying the proportion of proliferating cells that have remained in the cell cycle 48 h after labeling with EdU (see Materials and Methods). In the dnBMPR1B-electroporated cortex a greater proportion of the proliferating cells remained in the cell cycle as compared with the controls (control 20.6±3.3%, dnBMPR1B 35.6±5.2%; Fig. 2J, Fig. S1H-I′, arrowheads). To assess the status of BMP signaling in the GFP⁺ cells in test and control animals, we carried out immunohistochemical analysis for pSMAD1/5/8, which showed a significant decrease in the percentage of pSMAD1/5/8⁺ GFP⁺ double-positive cells in the test animals compared with the controls (control 92.7±2.5%, dnBMPR1B 29.8±0.8%; Fig. 2I).

When we analyzed the effect of inhibition of BMP signaling in the cortex at P0, in the animals that had been electroporated at E13.5 and E14.5, we found no significant difference in the distribution of GFP⁺ cells between test and control animals (Fig. 3C-F). However, a striking effect of inhibition of BMP signaling was observed in the cortex of test animals electroporated at E15.5, as compared with control. The majority of the GFP⁺ cells at this stage were located in the IZ in test animals, whereas in control animals most had migrated to the upper layer of the cortex (Fig. 3G,H). To quantify the distribution of GFP⁺ cells across the cortex, we divided the cortical section into seven equal bins, with bins 1-3 spanning the CP, bins 4 and 5 spanning the IZ, and bins 6 and 7 corresponding to the SVZ and VZ, respectively (Fig. 3I-K). In animals electroporated at E13.5 and E14.5 there was no significant difference in the numbers of GFP⁺ cells across all the bins between the test and control animals (Fig. 3I,J). By contrast, in those electroporated at E15.5, the number of GFP⁺ cells was significantly lower in bin 1 and significantly higher in bins 6 and 7 in test animals compared with control (Fig. 3K).

Moreover, we also observed a curious effect of inhibition of BMP signaling on the morphology of the cells that appeared to be stuck in the IZ of test animals. We observed a bifurcation in the leading process in many of the GFP⁺ cells in the test animals, whereas there was a single leading process in the cells at a comparable position in control animals (compare insets in Fig. 3G,H, arrowheads, and Fig. 3N). Further, to ascertain whether there was significant downregulation of BMP signaling, we examined the presence of pSMAD1/5/8 in GFP⁺ cells in the test and control animals. Immunohistochemical analysis revealed that the majority of GFP⁺ neurons were negative for pSMAD1/5/8 in cortical sections from test animals, in contrast to control animals (Fig. 3L, white arrowheads; control 83.8±0.7%, dnBMPR1B 32.2±1.4%; Fig. 3M). Since we observed a migration defect as well as bifurcation of the leading process in GFP⁺ cells in the test animals electroporated at E15.5 but not in those electroporated at E13.5 and E14.5, it appears that BMP signaling specifically regulates the migration of upper layer cortical neurons born at E15.5.

In order to verify that the differences observed in the distribution of GFP⁺ cells in the VZ-SVZ and IZ between control and test animals was not merely due to overexpression of BMPR1B, we electroporated a construct expressing full-length BMPR1B at E15.5 and harvested the mice at E18.5. The distribution of GFP⁺ cells in the VZ, SVZ and IZ in these cortical sections was very similar to that in the control pCAG-GFP-electroporated mouse (Fig. 3O,Q,R). This indicated that the altered distribution of GFP⁺ cells observed in the dnBMPR1B-electroporated mouse (Fig. 3P, asterisk, and Fig. 3R) was specifically due to the inhibition of BMP signaling.

Our observation that inhibition of BMP signaling in the mouse cortex affected the migration of E15.5 born upper layer cortical neurons at P0, led us to further investigate if this is simply a delay or a complete block in migration. We examined the cortex of test and control animals at P6, which had been electroporated at E15.5 (Fig. 4A-D). In cortical sections from test animals comparatively few GFP⁺ neurons had migrated to the upper layer of the cortex. Moreover, the GFP⁺ neurons that had migrated appeared to be disorganized and misaligned within the upper layer (Fig. 4B) compared with those in the control animals (Fig. 4A). In addition, the orientation of the cell body and the neuronal processes was disturbed in a significant number of GFP⁺ neurons, unlike that in the controls (compare Fig. 4A′,B′; control 9.5±1.2%, dnBMPR1B 64.9±4.3%; Fig. 4F). To determine the layer-specific positioning of these GFP⁺ neurons in the cortex, we performed immunohistochemistry for Brn2, a marker of differentiated layer II/III cortical neurons (Fig. 4C,D). These GFP⁺ neurons express Brn2, suggesting that although disorganized and misoriented they had adopted the fate of layer II/III neurons even in the absence of BMP signaling (Fig. 4C′,D′). We found a significant decrease in pSMAD1/5/8 immunoreactivity in the GFP⁺ neurons in the test animals as compared with the controls, indicating that BMP signaling is indeed downregulated (Fig. 4G, arrowheads; control 87.3±2.6%, dnBMPR1B 16.3±3.1%; Fig. 4G′).

In radially migrating neurons, the Golgi apparatus is localized to the leading process, which later on forms the apical dendrite of the differentiating neuron when it reaches the mantle zone. The localization of the Golgi in the leading process determines the orientation of the apical dendrite of the differentiating neuron towards the pia of the cortex and thus imparts polarity to the neuron (Meseke et al., 2013; O'Dell et al., 2012; Matsuki et al., 2010). On inhibition of BMP signaling, we failed to observe a clearly distinguishable apical dendrite oriented towards the pia in GFP⁺ neurons. Thus, we wanted to ascertain whether the Golgi apparatus was also mislocalized in these cells. We carried out immunohistochemistry for GM130 (Golga2), a Golgi protein, to localize the Golgi apparatus in the GFP⁺ neurons in cortical sections from test and control animals at P6 (Fig. 4H-I′, Movies 1, 2). In the majority of GFP⁺ neurons from control animals, the Golgi was localized at the base of the apical dendrite and extended into it (cells with normal Golgi localization, 92.4±2.7%; Fig. 4L), and the cells were oriented towards the pia (Fig. 4H, arrowhead). By contrast, in the test animals, in a large proportion of the GFP⁺ neurons the Golgi apparatus (Fig. 4I, arrowheads) was distributed throughout the cytoplasm (cells with abnormal Golgi localization, 34.3±9.5%; Fig. 4L). Moreover, in these cells the dendritic processes were randomly oriented with no clearly distinguishable apical dendrite (Fig. 4I,I′).

The misorientation of dendrites and the altered polarity of the GFP⁺ cortical neurons in test animals prompted us to investigate the effect of inhibition of BMP signaling on formation of the dendritic arbor at a later time point. We analyzed GFP⁺ cells at P15 in the cortex of test and control animals, which had been electroporated at E15.5. Although the apical dendrites in the majority of GFP⁺ neurons were oriented towards the pia, the dendritic arbor of these neurons was much less elaborate in the cortical sections from test animals as compared with the control (compare Fig. 4J,K). We quantified the average number of branches emerging from GFP⁺ neurons and found that in the test animals the number of secondary as well as tertiary neurites (Fig. 4J,K, blue and green arrowheads) was significantly reduced (Fig. 4M). By contrast, primary neurites (Fig. 4J,K, red arrowheads) were not significantly reduced in number (Fig. 4M) in the test animals compared with the controls. Thus, BMP signaling seems to regulate the polarity of E15.5 born upper layer cortical neurons, which is manifested as Golgi mislocalization in the differentiating neuron at P6. In addition, BMP signaling also appears to be important for the formation of dendritic branches in these neurons at later stages.

BMPs are known to signal through both canonical and non-canonical pathways. The canonical pathway involves phosphorylation of the downstream effector SMAD1/5/8 and subsequent regulation of expression of BMP target genes. There are several non-canonical pathways downstream of BMP, one of which results in activation of Lim kinase (LIMK) bound to the cytoplasmic tail of the type II BMP receptor. This is followed by phosphorylation of cofilin by activated LIMK, ultimately resulting in modulation of actin polymerization (Foletta et al., 2003). It has been reported from in vitro studies that the dendritogenesis of cortical neurons is regulated by BMP signaling acting through the non-canonical LIMK-mediated pathway (Lee-Hoeflich et al., 2004). LIMK activation downstream of BMP signaling is dependent on two factors: (1) binding of LIMK to the cytoplasmic domain of BMPR2, which relieves an autoinhibitory effect on its catalytic domain (Lee-Hoeflich et al., 2004); and (2) phosphorylation of the activation loop of LIMK by P21-activated kinase 1 (PAK1), which interacts with BMPR1 (Podkowa et al., 2013) (see Fig. 8). We speculate that when we overexpressed dnBMPR1B it led to blockage of canonical BMP signaling by inhibition of SMAD1/5/8 phosphorylation and blockage of non-canonical BMP signaling by preventing the PAK1-mediated activation of LIMK (see Fig. 8).

Since overexpression of dnBMPR1B in the developing mouse cortex led to defects in neuronal migration, polarity and dendritic morphology, we investigated the contribution of canonical and non-canonical BMP signaling to each of these phenotypes. To selectively inhibit the Smad-dependent canonical pathway we overexpressed Smurf1, an E3 ubiquitin ligase for Smad proteins (Zhu et al., 1999). To inhibit the LIMK-mediated non-canonical BMP pathway, we overexpressed a BMPR2 lacking the LIMK-interacting domain in its cytoplasmic tail (BMPRIIΔLIMK) (Phan et al., 2010) (Fig. 5A). In order to assess the efficacy with which these constructs selectively inhibit canonical and non-canonical BMP signaling, we co-transfected HEK293T cells with either pCAG-dnBMPR1B, pCAG-Smurf1 or pCAG-BMPRIIΔLIMK along with pCAG-GFP. This was followed by immunohistochemical detection of pSMAD1/5/8 and phosphorylated cofilin (p-Cofilin) as readouts of the canonical and non-canonical pathways, respectively (see the supplementary Materials and Methods). The dnBMPR1B construct inhibited both canonical and non-canonical BMP signaling. By contrast, the pCAG-BMPRIIΔLIMK and the pCAG-Smurf1 constructs inhibited only the non-canonical or the canonical pathway, respectively (Fig. S2A).

Subsequently, we co-electroporated either pCAG-Smurf1 or pCAG-BMPRIIΔLIMK together with pCAG-GFP into the developing mouse cortex in utero at E15.5. The cortices from these animals were analyzed at E17.5 (Fig. S2B-D′) and P0 (Fig. 5), to study the effect, if any, on the migration of upper layer neurons. At E17.5, the majority of GFP⁺ cells in the cortex overexpressing Smurf1 were stuck in the SVZ (Fig. S2C,D′), similar to what was observed with dnBMPR1B overexpression (Fig. 2D). However, there was no obvious defect in the migration of GFP⁺ neurons in the cortex overexpressing BMPRIIΔLIMK (Fig. S2D,D′), which was similar to the control (Fig. S2B).

When we analyzed the cortex overexpressing Smurf1 or BMPRIIΔLIMK at P0, we found that many of the cortical neurons had failed to migrate to their final positions in both cases (Fig. 5E,F), as compared with the pCAG-GFP-electroporated control (Fig. 5C, Fig. 3G). However, the defect in migration of cortical neurons in these animals was less severe than that observed with overexpression of dnBMPR1B (Fig. 5D, Fig. 3H). In order to quantify the distribution of GFP⁺ neurons in each case, we divided the cortical sections into seven equal bins, similar to that shown in Fig. 3I. In the cortex overexpressing Smurf1 or BMPRIIΔLIMK, the number of GFP⁺ cells in bin 1 was significantly lower than in control animals, but higher than in the cortex overexpressing dnBMPR1B (Fig. 5G). However, in cortices with overexpression of Smurf1 or BMPRIIΔLIMK, the number of GFP⁺ cells in bins 6 and 7 (encompassing VZ and SVZ) was significantly higher than that in the control, but only somewhat lower than that in dnBMPR1B-overexpressing cortical sections (Fig. 5G).

Immunohistochemical analysis revealed a significant decrease in the number of pSMAD1/5/8⁺ cells in cortices with Smurf1 overexpression as compared with control (Fig. 5H,H′, arrowheads; control 83.4±0.7%, Smurf1 22±5.5%; Fig. 5I). Similarly, a significant decrease in p-Cofilin in the cortices overexpressing BMPRIIΔLIMK was observed compared with the control (Fig. 5J,J′, arrowheads; control 95.9±1.7%, BMPRIIΔLIMK 23.5±1.1%; Fig. 5K). We also noted a bifurcation of the leading process in a few GFP⁺ cells in the cortex overexpressing either Smurf1 or BMPRIIΔLIMK, as previously observed with dnBMPR1B overexpression (Fig. 5C-F insets, arrowheads). These observations suggest that BMP signaling regulates the migration of E15.5 born upper layer cortical neurons through both canonical and LIMK-mediated non-canonical pathways.

We examined the cortices of the mice electroporated with pCAG-Smurf1 and pCAG-BMPRIIΔLIMK at P6, in order to determine the role of canonical versus non-canonical BMP signaling in regulating the polarity of upper layer pyramidal neurons (Fig. 6). In the majority of GFP⁺ cells overexpressing Smurf1, the Golgi apparatus was mislocalized and distributed throughout the cytoplasm of the cell body (Fig. 6C-C″, arrowhead), and the orientation of the cell body was disturbed, similar to what was observed in the GFP⁺ cells expressing dnBMPR1B (Fig. 6B-B″, arrowhead). By contrast, in the majority of GFP⁺ cells overexpressing BMPRIIΔLIMK, the Golgi was localized correctly to the base of the apical dendrite and extending into it (Fig. 6D-D″, arrowheads), similar to that observed in the control (Fig. 6A-A″, arrowhead). The specific effect of inhibition of canonical BMP signaling on the polarity of the GFP⁺ cells became clearer on quantification (Fig. 6I). It is worth mentioning that although a small proportion (16.3±2.8%) of GFP⁺ cells overexpressing BMPRIIΔLIMK did show mislocalization of the Golgi, this was significantly lower than that observed in cells overexpressing dnBMPR1B (36.8±10.4%) or Smurf1 (40.6±5.1%). We confirmed the downregulation of canonical and LIMK-mediated non-canonical BMP signaling in cortex overexpressing Smurf1 and BMPRIIΔLIMK, respectively, by examining the level of pSMAD1/5/8 and p-Cofilin. We found a significant decrease in the number of cells with pSMAD1/5/8 immunoreactivity among cells overexpressing Smurf1, as compared with control (Fig. 6E, arrowheads; control 87.3±2.6%, Smurf1 20±4.5%; Fig. 6F). Similarly, there was a significant decrease in the number of cells with p-Cofilin immunoreactivity among cells overexpressing BMPRIIΔLIMK as compared with the control (Fig. 6G, arrowheads; control 90.6±1.6%, BMPRIIΔLIMK 20.4±3%; Fig. 6H). Thus, it appears that Smad-dependent canonical BMP signaling is more important than the non-canonical LIMK-mediated BMP pathway for establishing the polarity of E15.5 born upper layer neurons.

We had previously observed that at P15 the upper layer pyramidal neurons expressing dnBMPR1B have a reduced dendritic arbor with fewer secondary and tertiary branches. To investigate the effect of canonical and non-canonical BMP signaling on dendritogenesis, we extended this analysis to P21, when dendritogenesis has progressed further and the dendritic spines can be visualized. At P21, there was a significant decrease in the number of primary, secondary and tertiary neurites in the neurons overexpressing dnBMPR1B or BMPRIIΔLIMK as compared with the control (Fig. 7A,B,D-G). However, the decrease in the number of secondary and tertiary neurites was not significant in the Smurf1-overexpressing neurons as compared with the control (Fig. 7A,C,F,G). Upon quantifying the length of the dendritic branches we found a significant difference in the length of both primary and secondary neurites in neurons overexpressing dnBMPR1B or BMPRIIΔLIMK compared with control (Fig. 7H,I). In Smurf1-overexpressing neurons there was a slight, but significant, decrease in the length of primary neurites as compared with controls, although this decrease was less than that observed with either dnBMPR1B or BMPRIIΔLIMK (Fig. 7H).

Since the dendritic spines could be clearly distinguished at P21, we quantified the number of spines per 10 µm of dendritic length. There was a significant decrease in spine density in neurons overexpressing BMPRIIΔLIMK as compared with the control (Fig. 7J,M,N) and, although to a lesser extent, also in neurons overexpressing dnBMPR1B or Smurf1 (Fig. 7J,K,L,N). These observations suggest that dendritic branch formation of E15.5 born upper layer neurons is minimally affected by inhibition of canonical BMP signaling, whereas it is severely affected by inhibition of LIMK-mediated non-canonical BMP signaling.

This study has revealed a highly dynamic spatiotemporal pattern of canonical BMP signaling during the development of the mouse cortex, strongly suggesting that this pathway regulates multiple aspects of cortical development. When we inhibited BMP signaling by overexpressing dnBMPR1B during cortical development we found that it affected the migration, polarity and dendritic morphogenesis of E15.5 born upper layer cortical neurons. As dnBMPR1B inhibits both canonical and non-canonical BMP signaling, we designed experiments to dissect the role of these two types of BMP signaling in each of the three observed phenotypes. Our data suggest that migration of E15.5 born upper layer cortical neurons is regulated by both pSmad-mediated canonical and LIMK-mediated non-canonical BMP signaling. Canonical BMP signaling seems to be the major regulator of polarity of these neurons, whereas the formation of dendritic branches seems to be mostly regulated by LIMK-mediated non-canonical BMP signaling (Fig. 8).

The migration of neurons that are born at E15.5 and mostly populate the upper cortical layers is specifically delayed upon overexpression of dnBMPR1B. By contrast, the migration of neurons born at E13.5 or E14.5 is largely unaffected upon inhibition of BMP signaling. Unlike the E13.5 born neurons, which populate the lower cortical layers, the E14.5 and E15.5 born neurons populating the upper cortical layers (constituting layer IV and layers II-III, respectively) undergo glial-dependent radial migration. However, inhibition of BMP signaling seems to affect only the migration of the upper layer neurons born at E15.5 and not that of neurons born at E14.5. This differential effect of BMP signaling on the migration of E15.5 born neurons is very interesting. It suggests that the migration of the E14.5 born upper layer neurons is independent of BMP signaling and might involve a separate set of molecular players.

One possible explanation for this effect on migration is a delay in neurogenesis that results in a subsequent delay in migration. Indeed, when we examined the effect of inhibition of BMP signaling on cell proliferation at E17.5, we found that a greater proportion of the proliferating cells remained in the cell cycle in the dnBMPR1B-electroporated cortex compared with the control. This suggests that the delay in migration might occur due to a delay in neurogenesis. However, further experiments need to be carried out in order to determine whether inhibition of BMP signaling is likely to affect the later stages of migration of upper layer cortical neurons that occur post-neurogenesis.

Subsequently, when we inhibited canonical BMP signaling by overexpressing Smurf1 or inhibited non-canonical BMP signaling by overexpressing BMPRIIΔLIMK in the mouse cortex, we did observe a delay in the migration of these neurons, but the effect was not as dramatic as that observed with overexpression of dnBMPR1B. This indicates that both canonical and LIMK-mediated non-canonical BMP signaling are necessary for proper migration of E15.5 born upper layer neurons and explains the severity of the migration defect observed upon expression of dnBMPR1B, which inhibits both of these pathways (Fig. 8).

Upon inhibition of BMP signaling by overexpression of dnBMPR1B, we also observed morphological changes such as the bifurcation of the leading process in the neurons that were stuck in the IZ during migration. A bifurcated leading process has previously been reported in neurons in which modulators of RhoA activity have been knocked down (Pacary et al., 2011; Azzarelli et al., 2014; Hand et al., 2005; Pacary et al., 2013). Thus, we speculate that BMP signaling might regulate the migration of E15.5 born upper layer cortical neurons through modulating RhoA activity.

At P6, we observed another effect of inhibition of BMP signaling, namely an alteration in the polarity of the E15.5 born upper layer cortical neurons. This was manifested as an indistinguishable apical dendrite and mislocalization of the Golgi. We observed similar mislocalized Golgi in neurons overexpressing Smurf1, but not in significant numbers in neurons overexpressing BMPRIIΔLIMK. Although the process of establishment of cell polarity may involve cytoskeletal changes, our data seem to indicate that the cytoskeletal changes orchestrated by non-canonical BMP signaling do not play a role in this process. This suggests that canonical BMP signaling is primarily required for polarity of the E15.5 born upper layer neurons (Fig. 8).

At P21 we observed a significant effect of inhibition of BMP signaling on dendritic morphology, which included dendritic branching, length and spine density. The branching and length of dendrites were most significantly decreased in neurons in which LIMK-dependent non-canonical BMP signaling had been inhibited. By contrast, in neurons in which canonical BMP signaling was selectively inhibited by Smurf1 overexpression, there was no significant change in dendritic branching or length compared with the control. In the case of spine density, the maximum decrease was observed with inhibition of LIMK-mediated non-canonical BMP signaling. This suggests that the effect of overexpression of dnBMPR1B on morphogenesis of dendrites is mostly a reflection of the inhibition of LIMK-mediated non-canonical BMP signaling (Fig. 8). Previous in vitro studies of cortical neurons have demonstrated that LIMK-mediated non-canonical BMP signaling is a key regulator of dendritogenesis (Lee-Hoeflich et al., 2004), and our study has provided in vivo data in support of this.

In conclusion, this study has yielded data in strong support of a hypothesis that BMP signaling is involved in regulating several diverse developmental phenomena in the mouse cortex. Inhibition of BMP signaling specifically affects the migration, establishment of polarity and dendritic morphogenesis of E15.5 born upper layer pyramidal neurons. In addition, this study has also dissected out the contribution of canonical and LIMK-dependent non-canonical BMP signaling in regulating each of these phenotypes. Further exploration in future will aim to identify the specific downstream effectors of BMP signaling that regulate migration, polarity and dendritic morphogenesis.

Animal experiments were performed according to the protocol sanctioned by the Institute Animal Ethics Committee, registered with CPCSEA (no. 810/03/ac/CPCSEA, dated 15/10/2003). Wild-type mice (CByB6F1/J, Stock no. 100009, The Jackson Laboratory) were crossed to generate timed pregnant females for the experiments.

Timed pregnant females were anesthetized using an isoflurane vaporizer (Patterson Veterinary). Uterine horns of the pregnant female were exposed and construct(s) DNA injected at 1 µg/µl total concentration along with 1% Fast Green dye into the forebrain of the embryos. Electroporation was performed with tweezer-disc electrodes (LF650S10, BEX) providing an electric pulse of 35 V for 50 ms, five times at an interval of 950 ms, using an electroporator (ECM 830, BTX, Harvard Apparatus).

The following constructs were used: (1) pCAG-GFP (a gift from Prof. Constance Cepko, Harvard Medical School, Boston, USA), (2) pCAG-dnBMPR1B, (3) pCAG-BMPRIIΔLIMK, (4) pCAG-Smurf1. For constructs 2-4, see the supplementary Materials and Methods.

Mouse forebrain sections of the desired stage were subjected to immunohistochemistry using the following primary antibodies: anti-pSMAD1/5/8 (1:100; 8828, Cell Signaling Technology), anti-Tbr1 (1:2000; Ab31940, Abcam), anti-Tbr2 (1:1000; Ab15894, Abcam), anti-Ctip2 (1:300; Ab28448, Abcam), anti-Brn2 (1:4000; SC6029, Santa Cruz Biotechnology), anti p-Cofilin (1:300; SC21867R, Santa Cruz Biotechnology), anti-GM130 (1:300; 610822, BD Biosciences), anti-GFP (1:500; A-6455, ThermoFisher Scientific) and anti-Ki67 (1:500; Ab39260, Abcam). See the supplementary Materials and Methods for details.

The timed pregnant females were injected with EdU via intraperitoneal injection at 20 mg/kg body weight 2 h before in utero electroporation at E15.5, and embryos were harvested at E17.5 for analysis. EdU labeling in the electroporated cells in cortical sections was detected using the Click-iT EdU Imaging Kit (C10338, ThermoFisher Scientific). EdU⁺ GFP⁺ double-positive cells were further analyzed for the presence of Ki67, a marker for proliferating cells by performing immunohistochemistry in the cortical section after the EdU detection assay. The fraction of GFP⁺ cells that remained in the cell cycle was estimated by counting the number of GFP⁺ EdU⁺ Ki67⁺ triple-positive cells among GFP⁺ EdU⁺ double-positive cells.

For each quantification, GFP⁺ cells from three consecutive sections of an embryo were counted, averaged and considered as n=1. At least three such embryos (n=3) were analyzed for each experiment. For quantification of dendritic branches, the average number of primary, secondary and tertiary branches was calculated for 50 GFP⁺ neurons per animal and this was considered as n=1. Counting of GFP⁺ cells and dendritic branches was performed using ImageJ software. See the supplementary Materials and Methods for details.

Sections of the mouse cortex electroporated with the various constructs were imaged and analyzed using a multiphoton laser scanning confocal microscope (LSM780) and ZEN 2011 software (Carl Zeiss). Neurite length was estimated using NeuronJ software (https://imagescience.org/meijering/software/neuronj), an ImageJ plug-in for neurite tracing, following the instructions provided in the manual.

Statistical analysis was performed for all data sets using Excel (Microsoft), and data are represented as mean±s.d. An unpaired Student's t-test was performed to analyze statistical significance. One-way analysis of variance (ANOVA) was performed to compare the means of more than two groups, and post-hoc analysis was performed using a two-tailed Student's t-test (Figs 4-6).

We are grateful to Dr Amitabha Bandyopadhyay and Dr Sandeep Gupta for critical comments on the manuscript; Mr Ali Abbas Zaidi for help in subcloning of dnBMPR1B; Mr Tathagat Biswas for help in preparation of movie files of GM130 immunostaining; and Ms Neetu Dey for help with confocal imaging.