Neuronal LRP4 regulates synapse formation in the developing CNS

During the last few decades we have acquired substantial knowledge on how neuromuscular junctions (NMJs) are formed in the peripheral nervous system (PNS; for a review, see Tintignac et al., 2015). Conversely, the molecular determinants orchestrating synapse formation in the central nervous system (CNS) are complex and much less understood. Although the molecular cascades governing synaptogenesis in the CNS and the PNS seem to be fundamentally different, the existence of organizational proteins common to both types of synapses suggests that the development of both structures might share several common pathways.

A key regulator for NMJ development is the low-density lipoprotein receptor-related protein 4 (LRP4; Weatherbee et al., 2006; Wu et al., 2012; Yumoto et al., 2012). At the NMJ, LRP4 forms a complex with the muscle-specific tyrosine kinase MuSK in the muscle fiber plasma membrane; this complex serves as the receptor for the motoneuron-derived extracellular matrix protein agrin (Kim et al., 2008; Zhang et al., 2008). Binding of agrin to the tetrameric LRP4/MuSK complex activates an intracellular signaling cascade in the muscle fiber resulting in the formation of most, if not all, postsynaptic specializations (Tintignac et al., 2015). LRP4 can also act as a muscle-derived retrograde stop signal for motor axons and promote the formation of presynaptic specializations at the NMJ (Wu et al., 2012; Yumoto et al., 2012). Although motoneurons also express LRP4, the functions of neuron-derived LRP4 at the NMJ are not fully understood.

In addition to the PNS, LRP4 plays crucial roles in the adult CNS, including maintenance of excitatory synaptic transmission, hippocampal synaptic plasticity, fear conditioning, associative and spatial learning, and long-term potentiation (Gomez et al., 2014; Pohlkamp et al., 2015; Sun et al., 2016). Accordingly, Lrp4 mRNA is present in the neocortex, hippocampus, cerebellum and olfactory bulb (Lein et al., 2007; Sun et al., 2016; Tian et al., 2006). Moreover, LRP4 protein has been detected in postsynaptic membrane fractions purified from adult rat forebrain where it interacts with the postsynaptic scaffold protein PSD95 (also known as DLG4) (Gomez et al., 2014; Tian et al., 2006). Recently, astrocyte-expressed LRP4 activated by agrin has been shown to modulate glutamatergic synaptic transmission by regulating the release of ATP from astrocytes (Sun et al., 2016). Although these data show a requirement of astroglial LRP4 for normal function of the adult brain, it is still unknown whether LRP4 regulates synaptogenesis in the developing CNS in a similar way as at the NMJ.

In this study, we investigated the effects of increased or decreased expression of LRP4 on cultured CNS neurons as well as on neurons in mouse cortex and hippocampus in vivo. We show a crucial role for neuronal LRP4 activated by agrin in the developing CNS during dendritic development and synapse formation.

To begin to examine LRP4 in CNS neurons, we investigated its expression using dissociated cells from the embryonic cerebral cortex and hippocampus at different days in vitro (DIV3, 6, 8, 10 and 12) with two independently generated antibodies against LRP4 (see Materials and Methods). Cultures were co-labeled with antibodies against microtubule-associated protein 2 (MAP2), which localizes to the soma and the dendrites of neurons (Caceres et al., 1984). We detected LRP4 immunoreactivity in cortical and hippocampal neurons at all stages analyzed (Fig. 1A). In particular, we observed abundant punctate LRP4 staining in neuronal cell bodies (Fig. 1Ba,Ba′) and in the proximal MAP2-positive dendrites (Fig. 1Bb,Bb′). Staining was also evident in axons as well as dendritic and axonal growth cones (Fig. 1Bb,Bb′, insets with one and two asterisks; Fig. 1C). Immunostaining of hippocampal neurons with antibodies for LRP4 and either CamKII (Jones et al., 1994) or GABA (Sloviter and Nilaver, 1987) demonstrated the presence of LRP4 in both excitatory and inhibitory neurons, respectively (Fig. S1). The specificity of the LRP4 antibodies used for immunocytochemistry was confirmed by western blotting, where a single band with an apparent molecular mass of 250 kDa was detected in cultured astrocytes, as previously described (Sun et al., 2016), as well as extracts from cultured embryonic neurons (Fig. 1D). Collectively, these results demonstrate the presence of LRP4 in embryonic neurons.

To investigate the role of LRP4 in embryonic neurons, in particular to determine if its expression influences the establishment of the neuronal network, we examined the effect of LRP4 knockdown and overexpression in cultured CNS neurons. To reduce LRP4 expression we used two different synthetic microRNAs (miR1232 and miR1544), both targeting the open reading frame of Lrp4 mRNA. Both microRNAs were selected based on their efficacy in reducing LRP4 levels (see Materials and Methods; Fig. S2A,B) and were cloned in tandem into a single vector together with the DsRed cDNA under the control of a CAG promoter (pCAG-miRLRP4-DsRed; Fig. S2B). As control, we used the pCAG-DsRed vector lacking both microRNA sequences (Fig. S2B). Primary cortical and hippocampal neurons were co-transfected at DIV3 with a cDNA encoding a synapsin-driven eGFP:actin fusion protein (SYN-GFP:actin) in combination with either the pCAG-miRLRP4-DsRed or with the pCAG-DsRed vector. The pSYN-GFP:actin expression allowed a detailed analysis of neuronal processes and dendritic spines in the transfected cells (Fischer et al., 1998) that were analyzed 7-9 days post-transfection (DPT). The LRP4 expression levels and the morphology of neurons transfected with the control plasmids were indistinguishable from non-transfected neurons (Fig. S2B,C; data not shown), demonstrating that neither the transfection nor the expression of either GFP:actin or DsRed influence the normal phenotype of cultured neurons. In contrast, transfection with the pCAG-miRLRP4-DsRed vector led to a severe reduction of LRP4 immunoreactivity (Fig. S2B), providing additional evidence for the specificity of the anti-LRP4 antibodies. Knockdown of LRP4 resulted in a noticeable increase of the length of the primary dendrites in both cortical and hippocampal neurons (Fig. 2A; Fig. 2B,E, blue bars). Additionally, reduced LRP4 expression led to a decrease in the number of primary dendrites extending from the neuron soma (Fig. S2B; Fig. 2A; Fig. 2C,F, blue bars) and the density of spines (Fig. 2D,G, blue bars). These results demonstrate that reduced levels of LRP4 affect number of dendritic processes, dendritic length and density of spines in cultured cortical and hippocampal neurons.

We hypothesized that increased levels of LRP4 might induce the opposite phenotype compared with the effect of the miRNAs. To investigate this, cortical and hippocampal neurons were co-transfected at DIV3 with a vector encoding LRP4 under the control of the neuron-specific human synapsin promoter (pSYN-LRP4) and the SYN-GFP:actin vector. Recombinant LRP4 was exclusively expressed in neurons as predicted by the synapsin promoter (Gascón et al., 2008; Kügler et al., 2001; Fig. S2C). Morphometric analysis of the dendritic structure revealed a significant decrease in the length of primary dendrites of both cortical (Fig. 2A; Fig. 2B, black bar) and hippocampal (Fig. 2E, black bar) neurons overexpressing LRP4 compared with neurons transfected with the control vector (Fig. 2A; Fig. 2B,E, gray bars). We also observed an increased number of primary dendritic processes and density of spines in hippocampal and cortical neurons transfected with the LRP4-encoding vector (Fig. 2A; Fig. 2C,D,F,G, black bars). Thus, reduced expression of LRP4 in embryonic neurons increased the overall length of primary dendrites and decreased the number of dendrites and spines, whereas increased levels of LRP4 expression had the opposite effect on dendritic number and length, as well as density of dendritic spines.

Previous studies demonstrated that in Lrp4^−/− mice the α-motoneuron growth cone continues to grow along the muscle fiber and fails to develop into a presynaptic terminal (Weatherbee et al., 2006). As a consequence, axons with deficient capacity to establish synapses might become unusually long. This suggests that the dendritic phenotype observed upon altered LRP4 levels might be secondary to differences in synaptic density at the CNS. To test this hypothesis, we analyzed the progressive development of spines and dendrites in cultured neurons with increased and reduced levels of LRP4 at different time points. We first focused on the early stages of development (DIV3-8), when dendrites are almost devoid of mature spines and synapses. Interestingly, upon LRP4 overexpression, both hippocampal and cortical neurons exhibited reduced length and increased number of primary dendritic processes already at DIV6 (Fig. 3A, top; Fig. 3B,C). In contrast, neither hippocampal nor cortical neurons overexpressing LRP4 displayed changes in the density of spines at DIV6 (Fig. 3D). Likewise, the number of dendrites was significantly reduced upon LRP4 knockdown in hippocampal neurons at DIV8, prior to the reduction of spine density (Fig. 3D, left). Thus, these data indicate that the effect of LRP4 on dendritic arborization takes place before the development of synaptic specializations.

Interestingly, we observed that, whereas the differences in the dendritic length and spine density induced upon LRP4 overexpression and knockdown were maintained or even increased over time (from DIV3 to DIV21; Fig. 3B,D), the early effect on dendritic number (DIV6 to DIV12; Fig. 3C) became milder or disappeared during the maturation period (DIV12 to DIV21; Fig. 3C). Thus, at DIV21 we did not observe statistical differences in the number of dendrites in cortical and hippocampal neurons transfected with either the control or the miRNA-encoding plasmids (Fig. 3C). This suggests that the phenotype induced by altered levels of LRP4 on spine density and dendritic length is persistent, whereas the changes in dendritic numbers are rather transient.

To investigate further the mechanism by which LRP4 regulates dendritic development we followed the dynamics of neurite growth by time-lapse video microscopy (Fig. S3). As expected, control neurons exhibited extensive changes in the dendritic length over the tracking time period. Some dendrites decreased in length; however, the majority of them increased. In contrast, upon LRP4 overexpression neither shrinkage nor growth was observed, indicating that LRP4 expression diminishes growth cones' extension and retraction dynamics, which ultimately results in a reduced dendritic length. This suggests that LRP4 might regulate molecular pathways involved in the elongation or retraction of dendrites, which is in concordance with the expression of LRP4 in dendritic growth cones (Fig. 1C).

Our previous observations suggested that LRP4 expression levels affect the formation of synaptic specializations. To scrutinize this hypothesis further, we first examined whether knocking down LRP4 in CNS neurons was accompanied by a decrease in the density of puncta labeled by antibodies against pre- and postsynaptic marker proteins. In both hippocampal and cortical neurons, the density of puncta labeled by antibodies against the presynaptic proteins bassoon and synaptobrevin-2 (VAMP2) was significantly decreased (Fig. 4A; Fig. 4B,D, blue bars). We also observed a small reduction in the density of puncta labeled by an antibody against the postsynaptic protein PSD95, although this was not statistically significant (Fig. 4A; Fig. 4C, blue bar). Overall, these results demonstrate that reducing the amount of endogenous LRP4 leads to a decrease in the number of synapses in embryonic neurons.

We next investigated whether the number of synaptic puncta was increased upon LRP4 overexpression. In hippocampal and cortical neurons transfected with the LRP4-encoding vector, we observed an increase in the number of bassoon- (Fig. 4A; Fig. 4B, black bar) and synaptobrevin-2- (Fig. 4D, black bar) positive puncta. Likewise, the number of PSD95-positive puncta was also increased in hippocampal neurons (Fig. 4A; Fig. 4C, black bar). Taken together, these results demonstrate that an increase of LRP4 levels affected the density of synapse-like specializations in CNS neurons in vitro. Overall, our results demonstrate that both dendritic morphology and density of synapse-like specializations on dendrites are modulated by expression levels of LRP4 in cultured cortical and hippocampal neurons.

Our results so far demonstrated that LRP4 knockdown significantly reduced the number of dendritic spines and synapse-like specializations in hippocampal and cortical neurons (Figs 2-4). To investigate the extent to which these alterations affect functional synaptic connectivity, we next analyzed the spontaneous miniature excitatory postsynaptic currents (mEPSCs) in neurons transfected with the plasmid encoding the LRP4 miRNAs. Although we observed a reduction in the frequency and amplitude of mEPSCs upon LRP4 knockdown (Fig. S4), the differences were not statistically significant, probably owing to the high variability among neurons in the control cultures. Furthermore, changes in frequency of mEPSCs are not always proportional to the number of synapses, as changes in the stochastic release of presynaptic neurotransmitters or density of postsynaptic receptors could affect the mEPSC values as well. Thus, to further examine whether LRP4 expression levels affect synaptic connectivity, we used the rabies virus-mediated monosynaptic tracing technique, a method that allows an efficient analysis of functional presynaptic partners (Wickersham et al., 2007a). For that, cortical neurons were transfected with a retroviral vector encoding the EnvA-receptor TVA, the rabies virus glycoprotein G (which is responsible for retrograde transport of the virus across synapses), and the fluorescent reporter DsRedExpress2 (Fig. 5A). Neurons transfected with this vector are susceptible to subsequent infection by the glycoprotein G-deficient, eGFP-encoding EnvA-pseudotyped rabies virus (RABV) and capable of retrograde transfer of the virus to all immediate presynaptic partners (Deshpande et al., 2013; Wickersham et al., 2007a) (Fig. 5B). The schematics in Fig. 5B,C illustrate the strategy and the timeline for monosynaptic tracing of presynaptic partners of neurons in vitro. Transfection of the G- and TVA-encoding plasmid together with either pCAG-DsRed or pCAG-miRLRP4-DsRed followed by RABV infection 7 days later resulted in the appearance of double reporter-positive neurons (starter cells) at low density, indicating that these cells had received both the plasmids and the virus (yellow cells; Fig. 5D, white arrows). Green cells expressing only eGFP represent cells with transsynaptic transmission of RABV from the starter neurons (Fig. 5D). Because transsynaptic transmission of the RABV requires synaptic contacts, the green cells are presynaptic to the starter cells (DsRed⁺/GFP⁺, yellow). Analysis of the eGFP-labeled cells demonstrated that the number of direct presynaptic partners of LRP4 knockdown neurons was significantly lower compared with control neurons (Fig. 5E). This indicates that the LRP4 knockdown-mediated reduction of synapse-like specializations and spines is accompanied by a reduced number of synapses.

At the neuromuscular junction, LRP4 is the receptor for the motoneuron-derived extracellular matrix proteoglycan agrin (Kim et al., 2008; Zhang et al., 2008). We hypothesized that agrin might also regulate the effect of LRP4 on dendritic morphology in cultured CNS neurons. As a first step, we examined whether LRP4 influences the distribution of agrin in cultured neurons. To this end, we co-transfected cultured cortical neurons with pCAG-DsRed and pSYN-LRP4 on DIV2 and subsequently with pSYN-GFP:actin and a vector in which the transmembrane form of agrin was expressed under the control of the synapsin promoter (pSYN-TMagrin) on DIV3. Analysis of the sites of axon-dendrite contact by confocal microscopy revealed that agrin was highly concentrated at contact sites between dendrites of neurons overexpressing LRP4 and axons overexpressing TM-agrin (Fig. 6A). In contrast, neurons overexpressing TM-agrin did not concentrate agrin immunoreactivity at the sites of contact to neurons in which LRP4 had been knocked down (Fig. 6B). These results suggest that LRP4 affects the distribution of TM-agrin in neurons and are consistent with the hypothesis that LRP4 is involved in aggregating TM-agrin at CNS synapses.

We next analyzed the effect of LRP4 overexpression on cortical neurons in the presence or absence of functional blocking antibodies for agrin (Fig. 6C). These polyclonal antibodies were generated against a 95 kDa C-terminal fragment of mouse agrin (Eusebio et al., 2003), which has been shown to be responsible for LRP4 binding and acetylcholine receptor aggregation at the NMJ. Neurons overexpressing LRP4 had reduced numbers of spines (Fig. 6D) and primary dendrites (Fig. 6E), and increased dendritic length (Fig. 6F) in the presence of anti-agrin antibodies, compared with neurons overexpressing LRP4 in the presence of the preimmune serum. Thus, addition of anti-agrin antibodies reversed the effect of LRP4 overexpression (for control values see white bars in Fig. 6D-F), suggesting that agrin is involved in the LRP4 overexpression-mediated phenotype of CNS neurons in vitro.

The effect of anti-agrin antibodies on dendritic growth suggested that an interaction of LRP4 with TM-agrin might be necessary for the formation of a normal dendritic arbor. To test this further, we investigated whether a soluble C-terminal fragment of an agrin isoform (A4B8), which is able to bind to LRP4 but is not anchored to the cell membrane or the extracellular matrix, interferes with the effect of LRP4 expression in cortical neurons. To this end, we analyzed the spine density, number of dendrites and dendritic length in embryonic neurons upon LRP4 overexpression in the presence or absence of the C-125 fragment of chick agrin (Pevzner et al., 2012; Tsim et al., 1992). Cultures were fixed 3-4 days after transfection and agrin addition, and morphometric analysis was performed as described above. The presence of the soluble agrin fragment led to a decrease in spine density (Fig. 6D) and number of primary dendrites (Fig. 6E), as well as an increase in dendritic length (Fig. 6F) in cortical neurons overexpressing LRP4. Thus, soluble agrin reversed the effect of LRP4 overexpression on dendritic growth and spine formation (for control values see white bars in Fig. 6D-F) mimicking the effect of the anti-agrin antibodies. This further supports that an interaction of LRP4 with TM-agrin is required for normal dendritic and spine development in cortical neurons. Furthermore, the similar effects of the anti-agrin antibodies and the soluble agrin fragment demonstrate that a membrane association of agrin is required for the LRP4-mediated changes in dendritic morphology and spine density in CNS neurons.

To investigate whether the altered dendritic morphology and synapse number observed in cultured embryonic neurons after knockdown of LRP4 could also be observed in vivo, we performed in utero electroporation of the LRP4 miRNAs. To this end, the SYN-miRLRP4-DsRed plasmid was co-electroporated with the SYN-GFP:actin vector into the hippocampus and cortex of embryonic day (E) 14.5 embryos. Electroporation of the pSYN-DsRed plasmid together with the pSYN-GFP:actin vector was used as control. This protocol allowed us to target groups of neurons in layers II-IV of the cortex and CA1 area of the hippocampus (see Fig. 7A for a schematic of the vectors and Fig. 7B for the injection sites and the timeline of the experiments). Dendritic morphology and spines were analyzed by confocal microscopy of brain sections from postnatal day (P) 21 animals (Fig. 7C,E). To investigate the effect of LRP4 knockdown on spines we analyzed their density on dendrites of electroporated neurons. Quantifications revealed that expression of miRLRP4 in embryonic neurons in vivo resulted in an approximately 40% decrease in the density of dendritic spines in cortical (Fig. 7D) and hippocampal (Fig. 7F) neurons. These results demonstrate a role for LRP4 in establishing or maintaining synapses in neurons in the cortex and CA1 region of the hippocampus.

Finally, to investigate the effect of LRP4 knockdown on dendritic development in vivo, we performed a morphometric analysis of single electroporated cells. Within cortical areas, the compact distribution of electroporated neurons did not allow us to identify single cells (Fig. 7C). However, only a few cells within the hippocampus were electroporated and could be unambiguously identified in coronal sections (Fig. 7E). Fig. 7G shows examples of individual CA1 hippocampal neurons electroporated with either the control plasmids (Fig. 7G, left) or the miRLRP4 plasmid (Fig. 7G, right) and traced with the Simple Neurite Tracer (Longair et al., 2011). Sholl analysis revealed that control hippocampal neurons had a greater number of intersections in a radius between 10 and 500 μm from the cell body compared with neurons in which LRP4 expression was reduced (Fig. 7H). Moreover, morphometric analysis of transfected cells revealed that reduced LRP4 expression significantly decreased the number of primary dendrites in embryonic hippocampal neurons (Fig. 7I) whereas their average dendritic length was not noticeably altered (data not shown). Because the electroporation procedure was not ideal for targeting neurons located in the CA3 area of the hippocampus, we were not able to determine whether this region would also be affected by LRP4 knockdown. However, we observed that cultured neurons from CA3 and CA1/dentate gyrus areas, which can be identified as CTIP2 (BCL11B)⁻ and CTIP2⁺ cells, respectively, are all immunoreactive for LRP4 (Fig. S5), suggesting that neuronal LRP4 might play an important role in all the areas of the hippocampus.

LRP4 has a fundamental role during formation, maintenance and regeneration of the NMJ (for a review, see Tintignac et al., 2015). Moreover, cumulative evidence indicates that LRP4 plays crucial roles in synaptic function in the CNS (Kim et al., 2008; Wu et al., 2012; Zhang et al., 2008). However, the molecular mechanisms regulating these functions are still mostly unknown. A recent study indicated that, in Drosophila, presynaptic expression of LRP4 regulates synapse number (Mosca et al., 2017). Studies in LRP4-deficient mice rescued from perinatal death by re-expression of LRP4 in the neuromuscular system, as well as mice with a selective loss of the intracellular and the transmembrane domains showed a significantly decreased density of dendritic spines in the CA3-hippocampal region, but had no gross anatomical abnormalities in the prenatal and adult hippocampus, cortex and cerebellum (Pohlkamp et al., 2015; Gomez et al., 2014). In addition, Lrp4 mRNA is present in the neocortex, hippocampus, cerebellum and olfactory bulb (Sun et al., 2016; Lein et al., 2007; Tian et al., 2006). A recent study (Sun et al., 2016) reported that in adult mice β-galactosidase expression driven by the endogenous Lrp4 promoter is mostly detectable in astrocytes, which is in concordance with our data showing that LRP4 levels, detected by western blotting, are higher in astrocytes than in neurons (Fig. 1D). Moreover, it was demonstrated that the activation of astroglial LRP4 by neuronal agrin induced the release of ATP, which modulates the synaptic activity of hippocampal pyramidal neurons (Sun et al., 2016). Consistently, mice with a conditional knockout for astroglial Lrp4 mRNA exhibited reduced glutamatergic synaptic transmission in neurons from the CA3 region. However, depletion of LRP4 expression in astrocytes did not lead to a decreased density of dendritic spines in the hippocampus (Sun et al., 2016), suggesting that other cell sources of LRP4 (i.e. neurons, oligodendrocytes or microglia) play additional roles in the CNS. In our study, we revealed that reduced LRP4 expression specifically in cortical and hippocampal neurons in vitro and in vivo led to a reduced density of spines and dendrite-associated synapse-like specializations, including bassoon- and synaptobrevin2-positive puncta. Accordingly, increased LRP4 expression in cultured CNS neurons resulted in an increase in the number of spines. As the experiments reported in our study affected LRP4 expression only in neurons (due to the neuron-specific synapsin promoter), the altered spine density that we observed must be caused by changes in neuronally expressed LRP4. Moreover, using a rabies virus-mediated monosynaptic tracing technique (Wickersham et al., 2007a), we demonstrated that the decreased number of synapse-like specializations and dendritic spines mediated by the expression of LRP4 miRNAs is paralleled by a decrease in the number of synaptic inputs as well, and therefore functional synapses.

Hippocampal and cortical neurons in culture also responded to reduced LRP4 expression by changes in dendritic morphology. Fewer but longer dendritic processes were generated when the density of synapse-like specializations was decreased. Overexpression of LRP4 in these neurons had the opposite effect, i.e. more but shorter dendrites extended from the soma with a concomitant increase in synaptic density. Moreover, we observed that knockdown of neuronal LRP4 also led to a decrease in the number of dendritic branches in the CA1 area of the hippocampus in vivo. Interestingly, the effect of LRP4 knockdown and overexpression on dendrite numbers disappears in cultured neurons over time (Fig. 3). This could explain why changes in dendritic morphology were not previously detected in adult LRP4-deficient brains (Gomez et al., 2014; Pohlkamp et al., 2015; Sun et al., 2016). However, changes in dendrite length remain evident after 21 days of maturation in vitro, suggesting that LRP4 levels would still have a long-term impact on neuronal arbor.

Previous studies have shown that dendritogenesis is a highly dynamic process that is influenced by synaptic activity as well as by signals from the immediate environment (for reviews, see Cheng and Poo, 2012; Dong et al., 2015). The opposite effects of LRP4 overexpression and knockdown on the number of primary dendrites, their length and on spine/synapse density suggest that all these effects might be mechanistically linked. One possibility is that the increase in the dendritic length observed in neurons with reduced levels of LRP4 is a consequence of the decrease in synaptic density. However, this is unlikely to be the case as we observed that the effect of LRP4 on dendritic arborization takes place before the development of synaptic specializations (Fig. 3). Another possibility is that LRP4 regulates the dynamics of cytoskeleton assembly, as rearrangements of cytoskeletal architecture is crucial for both dendrite morphogenesis (Flynn, 2013) and development of presynaptic terminals (Dent et al., 2011). Supporting this hypothesis, we observed that LRP4 overexpression results in a decreased mobility of dendrite terminals, blocking both extension and retraction of growing dendrites (Fig. S3). Accordingly, we detected immunoreactivity for LRP4 in dendritic growth cones. However, further studies are needed to determine the molecular signals regulating these processes.

At the NMJ, the interaction of LRP4 with agrin is crucial for synaptogenesis (Tintignac et al., 2015). In CNS neurons, we show that anti-agrin antibodies or addition of a soluble C-terminal agrin fragment reversed the effects of LRP4 overexpression. As both treatments would interfere with the TM-agrin-LRP4 interaction either by blocking agrin binding to LRP4 (antibodies) or by competing with the ligand-binding site of LRP4 (soluble agrin fragment), these results implicate TM-agrin as one potential binding partner for LRP4 in the CNS. Thus, TM-agrin-LRP4 interactions might shape dendritic morphology and induce the formation of synapses in cultured neurons of the CNS. This hypothesis is further strengthened by our observation that agrin accumulated at contact sites between dendrites of neurons with increased LRP4 expression and axons overexpressing agrin, whereas agrin did not aggregate at contact sites with neurons in which LRP4 expression was reduced. Interestingly, neurons in the brain of agrin^−/− mice, in which the perinatal lethality has been rescued by motoneuron-specific expression of agrin, also develop fewer synapses and shorter dendrites (Ksiazek et al., 2007).

Although our study focused on cortical and hippocampal neurons, LRP4 has a widespread expression pattern in the CNS, including astrocytes and neuronal populations outside of the forebrain, such as the cerebellum and the olfactory bulb (Tian et al., 2006; data not shown). It is therefore conceivable that LRP4 generally regulates dendritogenesis and synapse formation in many regions of the brain beyond the cerebral cortex and hippocampus. Our study provides the first indications of a function for LRP4 in dendrite and synapse formation in developing forebrain neurons, opening up avenues for elucidating the underlying molecular mechanisms in these and other regions of the CNS. Our results also suggest that at the molecular level, synaptogenesis at the NMJ and in the CNS are more similar than previously anticipated.

Use and care of animals was approved by German authorities and according to national law (TierSchG). C57BL/6J wild-type mice were bred in the animal facility of the Institute of Physiology of LMU Munich. The day of the vaginal plug was considered E0. Mice with a targeted deletion of the Musk gene have been described previously (DeChiara et al., 1996). For in utero electroporation, animals were operated upon as approved by the Regionspräsidium Karlsruhe (Baden Württemberg, Germany) under license number 35-9185.81/G-28/16. All experimental procedures were performed in accordance with German and European Union guidelines.

Primary cultures of embryonic cortical and hippocampal neurons were prepared as described (Brewer et al., 1993). Transfections with Lipofectamine 2000 (Invitrogen) were performed on DIV2-3 according to the manufacturer's instructions. The number of transfected cells was in the range of 2% and the efficiency was similar for each of the vectors used. One day after transfection, neurons were treated with 16 U chick agrin (C-terminal 125 kDa fragment isoform A4B8; Pevzner et al., 2012) or with 0.6 μl rabbit polyclonal anti-agrin antibody (serum 204; Eusebio et al., 2003) in 0.5 ml culture medium. We observed no effect of the antibodies on cell death and number of neurons and non-neuronal cells in the cultures. Fixation and staining of the neuronal cultures was performed as described (Zhang et al., 2015; see supplementary Materials and Methods for details on antibodies).

Time-lapse imaging of dissociated cortical neurons was performed as previously described (Costa et al., 2011). For further details, see the supplementary Materials and Methods.

The construct encoding for DsRedExpress2, the RABV glycoprotein (G) and the TVA800 (the GPI-anchored form of the TVA receptor), designed as CAG-DsRedExpress2-2A-G-IRES2-TVA (i.e. G-TVA construct), as well as the construction of the G gene-deleted and EnvA-pseudotyped GFP-expressing RABV (SADΔG-GFP) have been described previously (Wickersham et al., 2007a,b). Cells that express the G-TVA can be infected with EnvA-pseudotyped SADΔG-GFP and allow for transsynaptic transmission into all presynaptic neurons synaptically innervating the transfected and infected neuron. The connectivity with presynaptic neurons was determined by counting the double-fluorescent cells (GFP⁺/DsRED⁺) and the RABV-only single-positive cells (GFP⁺). Results were expressed as connectivity ratio, representing the number of GFP-positive cells per GFP- and DsRed-double positive cells as described (Wickersham et al., 2007a).

For overexpression of LRP4 in cortical and hippocampal neurons, the full-length coding sequence of mouse Lrp4 from pCMV-Lrp4 (commercially obtained from the German Center of Genome Research, Source BioScience, Nottingham, UK) was cloned into the pcDNA3.1(−) vector, in which the cytomegalovirus (CMV) promoter was replaced by the neuron-specific synapsin (SYN) promoter (Kügler et al., 2001). For knockdown of LRP4 by RNA interference, four different synthetic microRNA (miRNA) sequences (BlockIT Invitrogen) targeting the open reading frame and 3′ UTR of Lrp4 mRNA were individually cloned into the pcDNA6.2-GW vector. The targeted cDNA sequences for miRNAs were: LRP4 miR_1 (miR1232): 5′-TGAGGAGAACTGCAATGTTAA-3′; LRP4 miR_2 (miR1544): 5′-CGAGTACACGCTGCTACTGAA-3′; LRP4 miR_3 (miR6854): 5′-CTTTCAAAGACTGGCTCAACA-3′; LRP4 miR_4 (miR7072): 5′-CATTAGCTTCATGCTCCTTTA-3′.

Western blotting (see supplementary Materials and Methods) and quantitative analysis of LRP4 protein levels in HEK293 cells transiently transfected with full-length mouse LRP4 cDNA and subsequent transfection with the miRNAs revealed that miR1232 and miR1544 were the most efficient in reducing LRP4 protein levels (Fig. S2A). For knockdown of LRP4 by RNA interference in vitro and in utero, the miR1232 and miR1544 microRNAs were expressed as tandems with the DsRed cDNA sequence and cloned in an expression vector under the control of either the CAG or the neuron-specific synapsin (SYN) promoter (CAG/SYN-miRLRP4-DsRed), respectively. This plasmid was co-electroporated with the SYN-GFP:actin vector, in order to visualize spines and small protrusions. As control for the electroporation we used the pSYN-DsRed plasmid together with the pSYN-GFP:actin vector.

For overexpression of TM-agrin in neurons, the coding sequence of mouse TM-agrin was cloned into a pMES plasmid in which the CMV promoter was replaced by the SYN promoter. As control for the transfections we used the pSYN-GFP:actin vector (generously provided by Fransisco G. Scholl; Gascón et al., 2008) and the pCAG-DsRed (Heinrich et al., 2014). All constructs were verified by DNA sequencing and were produced under endotoxin-free conditions (Invitrogen EndoFree Plasmid Maxi Kit).

Whole-cell patch-clamp analysis of hippocampal neurons in culture after transfection with control vector or with miRLRP4 cDNA were used to record spontaneous mEPSCs. For further details, see supplementary Materials and Methods.

In utero electroporation of E14.5 timed pregnant mice (C57/Bl6; commercially obtained by Janvier Labs, Le Genest-Saint-Isle, France) was performed as previously described (Pacary et al., 2012). Specifically, pregnant mice were anesthetized with isoflurane (induction, 3%; surgery, 1.5%). The analgesic was applied directly after the anesthesia by subcutaneously injecting Carprofen (5 mg/kg). The uterine horns were exposed by Cesarean section and sterile, pre-warmed saline was repeatedly applied during the operation to keep the intestines moist. Animals were kept on a heating pad during the whole process of operation. Plasmids were mixed with Fast Green (2.5 mg/μl, Sigma) and 2 μl (at a concentration of 1 μg/μl) were injected into the lateral ventricle by use of a glass capillary under illumination of the uterine horns by a halogen cold light. DNA was electroporated into the hippocampus with five electrical pulses (amplitude, 40 V; duration, 50 ms; intervals, 1 s). During the pulsing the electrodes were swept from 0° to 45° for targeting the CA1 and CA3 regions, respectively. After electroporation, the uterine horns were carefully repositioned into the abdominal cavity, which was then filled with pre-warmed saline. The abdominal muscle and skin incisions were closed with surgical sutures with medical sewing equipment. Animals were left to recover in a clean cage and embryos allowed to continue their development. At P21, mice were anesthetized with ketamine and xylazine, and then transcardially perfused with PBS (pH 7.4) and fixative containing 4% paraformaldehyde in PBS. Brains were dissected and post-fixed overnight at 4°C. Brains were then sectioned at 100 μm thickness using a VT1200s vibratome (Leica). Sections were treated with 0.3% Triton X-100 in PBS (PBT) and non-specific staining was blocked with 1% bovine serum albumin and 2% normal donkey serum in PBT. All specimens were analyzed using a Zeiss LSM710 or a LSM780 laser-scanning microscope as detailed above. Digital processing of entire images, including adjustment of brightness and contrast, was performed using Photoshop CS3 (Adobe).

The number of primary dendrites emerging from the soma was manually counted. Axons were excluded from the analysis, as they can be distinguished morphologically from the dendrites (axons are thinner, longer, without spines). In addition, to facilitate the identification of dendrites, in most cases we performed immunostaining for MAP2 (labeling dendrites) and for Tau (MAPT; labeling axons). The dendritic length in microns was determined using the Zen2009 software (Zeiss) and the ImageJ Simple Neurite Tracer plugin (Longair et al., 2011) in cortical and hippocampal neurons using the images obtained with a LSM710 laser-scanning confocal or Axio Observer Z1 epifluorescence microscope (Carl Zeiss). The dendritic length in the graphs represents the average length of all the primary dendrites of each neuron (e.g. in a neuron with five primary dendrites the length of all of them was measured starting from the neuron soma and ending at the most distal point of the dendrite and we then calculated the average).

For the in utero electroporation, dendritic branching was analyzed using the Simple Neurite Tracer and the ImageJ plugin for Sholl analysis (Ferreira et al., 2014).

High-resolution z-stacks of dendrites from cortical and hippocampal neurons were collected with a 40× lens and a 2× digital zoom factor (thickness of optical sections: 0.4-1 μm). Spines were morphologically identified in the transfected neurons as short GFP⁺ protrusions emerging from the dendrites. Spine density was manually determined along dendritic segments. The density of bassoon-, synaptobrevin-2- and PSD95-positive puncta was also manually determined.

In all the aforementioned quantifications, experimenter was blind to the conditions analyzed.

Results are presented as mean±s.d. Significance was calculated with GraphPad Prism 5 (GraphPad Software) using a one-way ANOVA test with Dunnett's post-hoc test. Analysis of transfected neurons treated with soluble agrin and anti-agrin antibodies was performed using a one-way ANOVA test with Tukey's post-hoc test. For the in utero electroporation experiments and the experiments with the rabies virus, we used an unpaired t-test with Welch's correction. P<0.05 was set as the level of statistical significance.

We would like to thank Katja Peters and Martina Bürkle for expert technical assistance; Aditi Desphande, Antigona Ulndreaj, John Bixby, Fritz Rathjen and Elisa Murenu for insightful comments and for critically reading and improving the manuscript; Fritz Rathjen for the Lrp4^−/− mouse muscle tissue; Magdalena Götz for constant support and encouragement; and Wilko Altrock for generously providing anti-bassoon antibodies. Parts of the results shown in Figs 1-6 in this paper are reproduced from the PhD thesis of Andromachi Karakatsani (Karakatsani, 2015).