Long-lasting modifications in synaptic transmission depend on de novo gene expression in neurons. The expression of activin, a member of the transforming growth factor β (TGF-β) superfamily, is upregulated during hippocampal long-term potentiation (LTP). Here, we show that activin increased the average number of presynaptic contacts on dendritic spines by increasing the population of spines that were contacted by multiple presynaptic terminals in cultured neurons. Activin also induced spine lengthening, primarily by elongating the neck, resulting in longer mushroom-shaped spines. The number of spines and spine head size were not significantly affected by activin treatment. The effects of activin on spinal filamentous actin (F-actin) morphology were independent of protein and RNA synthesis. Inhibition of cytoskeletal actin dynamics or of the mitogen-activated protein (MAP) kinase pathway blocked not only the activin-induced increase in the number of terminals contacting a spine but also the activin-induced lengthening of spines. These results strongly suggest that activin increases the number of synaptic contacts by modulating actin dynamics in spines, a process that might contribute to the establishment of late-phase LTP.
Activin, a member of the transforming growth factor β family, is a multi-functional ligand protein that regulates the proliferation and differentiation of numerous cell types (Mather et al., 1997; Ying et al., 1997). For example, activin stimulates the release of follicle-stimulating hormone, of an erythroid differentiation factor and of a mesoderm-inducing factor. Activin also mediates the neuroprotective effect of basic fibroblast growth factor during excitotoxic brain injury (Tretter et al., 2000). Activin binds to the Ser/Thr kinase receptor ActRII (Acvr2a), which then phosphorylates the Smad2 and Smad3 transcription factors (Derynck and Zhang, 2003; Massague et al., 2000; Pangas and Woodruff, 2000). Phosphorylated Smad2/Smad3 subsequently interacts with Smad4 and the resulting complex translocates to the nucleus to activate target gene expression. The ActRII-binding PDZ protein, Arip-1 (also known as S-scam and Magi2), localizes at the synaptic site of hippocampal neurons (Shoji et al., 2000). In addition, other Smad-independent pathways are also activated by activin receptors, including MAP kinase signaling routes (ten Dijke et al., 2000; Werner and Alzheimer, 2006).
In the central nervous system the expression of many genes is regulated during long-term potentiation (LTP) (a model form of synaptic plasticity) (Matsuo et al., 2000; Nedivi et al., 1993; Qian et al., 1993; Yamagata et al., 1993; Yamazaki et al., 2001). One of these genes encodes activin β-A (Inhba), a subunit of activin A, and expression of this gene is upregulated during LTP and convulsive seizure as a late effector gene (Andreasson and Worley, 1995; Inokuchi et al., 1996).
Several studies strongly suggest that the long-lasting maintenance of synaptic plasticity relies on structural changes in neuronal cells (Bailey and Kandel, 1993; Bozdagi et al., 2000; Engert and Bonhoeffer, 1999; Fukazawa et al., 2003; Maletic-Savatic et al., 1999; Matsuzaki et al., 2004; Nagerl et al., 2004; Okamoto et al., 2004; Toni et al., 1999; Trachtenberg et al., 2002; Zhou et al., 2004). These structural changes involve alterations in the sizes of dendritic spines and the synaptic cleft, the formation of new spines, and major changes in dendritic and axonal arbors. Indeed, the morphogenesis of dendritic spines, which are the main site of innervation by excitatory glutamatergic presynaptic terminals, plays an important role in synaptic plasticity (Fukazawa et al., 2003; Matsuzaki et al., 2004; Nagerl et al., 2004; Okamoto et al., 2004; Zhou et al., 2004). A remarkable feature of the cytoskeleton of dendritic spines is its enriched level of filamentous actin (F-actin) (Fifkova and Morales, 1992; Matus et al., 1982). The mechanisms that regulate F-actin dynamics in the spine contribute to the induction, extent and maintenance of hippocampal LTP (Fukazawa et al., 2003; Kim and Lisman, 1999; Krucker et al., 2000; Okamoto et al., 2004). These observations suggest that genes that are upregulated during LTP include those that modulate spinal actin dynamics and spine morphology.
Here, we used low-density cultures at mature stages from the rat hippocampus to investigate changes in spine morphology by focusing on both presynapses and postsynapses after the addition of purified activin from bovine follicular fluid. The low-density culture method developed by Banker (Goslin et al., 1998) offers a useful model system for cell biological studies of central nervous system neurons. Because neurons spread in monolayers and both dendrites and axons extend in the same focal plane, immunostaining of specific proteins provides a convenient means to visualize and quantify spine morphology by allowing presynapses and postsynapses to be discriminated. We found that activin modulates actin dynamics in the spine, thereby altering spine morphology and the synaptic connectivity of hippocampal neurons.
Activin enhances the number of pre- and post-synaptic contacts
Because activin is known as a differentiation factor that modulates cell morphogenesis, we examined whether it regulates synaptic morphology in 18-day-old hippocampal neurons in dissociated culture. The shape of spinal F-actin directly reflects the morphology of the dendritic spine (Fischer et al., 1998). We hereafter refer to the shape of phalloidin-stained F-actin within the spine as spine morphology, because rhodamine-phalloidin is a specific probe for F-actin. Antibodies to presynaptic proteins such as synaptophysin or synaptotagmin offer a convenient means to visualize the distribution of presynaptic specializations on neurons. Fig. 1 shows the number of presynaptic contacts per dendritic spine in cultured hippocampal neurons with or without activin treatment. In the control hippocampal culture, most of the phalloidin-positive puncta made contact with only one synaptophysin-positive punctum (Fig. 1A). However, 6 hours after the introduction of activin, there was a significant increase in the number of phalloidin-positive puncta interacting with two or three synaptophysin-positive puncta (Fig. 1B). Additionally, the majority of phalloidin-positive puncta that interacted with two or more synaptophysin-positive puncta were mushroom-shaped spines with an elongated neck. Quantitative analysis clearly demonstrated that activin significantly enhanced the number of contacts made by presynaptic terminals on each spine (Fig. 1C). Activin increased the population of spines that were contacted by multiple presynaptic terminals from 17.8% in control cultures to 28.7% in activin-treated cultures (Fig. 1D).
A similar increase in the number of presynaptic contacts on each postsynaptic spine was observed after activin treatment for either 24 hours or for 4 days (Fig. 1E). This increase was blocked when anti-activin A antibody was added to the culture medium 2 hours after the start of activin treatment. Therefore, continuous activin signaling is required for altering synaptic contacts.
Activin modulates spinal F-actin morphology
We next examined whether activin regulates the morphology of spines. A particularly attractive hypothesis is that an increase in average spine length elevates the number of synaptic contacts made by each neuron, because this increases the chance that a spine makes contact with an axonal terminal (Stepanyants et al., 2002). We first visualized the morphology of 21-day-old hippocampal neurons from a high-density culture previously transfected with EGFP. A live image shows that the spines of activin-treated neurons were more mobile than those of control neurons. Some spines were unaltered, but others underwent changes in length (Fig. 2A,B). A comparison of spine lengths showed that some populations of spines underwent elongation following activin treatment for 6 hours (Fig. 2C,D).
To investigate changes in spine morphology in more detail, we treated 18-day-old hippocampal neurons from a low-density culture with activin and then stained them with rhodamine-phalloidin and an anti-Map2 (anti-Mtap2) antibody to visualize spine heads and synaptic dendrites, respectively (Fig. 3). In our culture system, most spines (∼95%) from untreated hippocampal neurons consisted of a spine head with a thin neck (mushroom-shaped), and headless spines (filopodia and other protrusions) were rare (∼5%) (Fig. 3A,B). The presence of activin did not alter the ratio of mushroom-shaped to headless spines (data not shown). Thus, the main effect of activin treatment was to elongate the neck, resulting in longer mushroom-shaped spines. The lengthened spines formed synapses, as shown by their colocalization with the presynaptic marker protein synaptotagmin I (Fig. 3A,B). Morphological changes were first detected 2 hours after the addition of activin to the culture system, and these increased with longer incubation times (Fig. 3C). The effect of activin was dose-dependent and was completely blocked by follistatin, an activin-binding protein that inhibits activin function (Nakamura et al., 1990) (Fig. 3D,E). The number of spines and the size of spine heads were not significantly affected by activin treatment (Fig. 3F,G). Blocking presynaptic activity by tetrodotoxin did not inhibit the response to activin (Fig. 3H).
The average spine length was the same for spines with single synaptic contacts, whether cells were cultured in the presence or absence of activin (Fig. 3I). Similarly, spines with two synaptic contacts had the same length irrespective of culture conditions. There were, however, significant differences in spine length between spines having a single synaptic contact and those with two contacts (single versus double in control, P<0.05; single versus double in activin-treated, P<0.001). Thus, activin modulates not only the physical patterning of pre- and post-synaptic contact sites but also overall spine morphology.
The effect of activin on spine morphology does not depend on protein or RNA synthesis
Cycloheximide and emetine are protein-synthesis inhibitors that operate via distinct inhibitory mechanisms. Neither drug blocked the effect of activin (Fig. 4A,B). Moreover, inhibition of RNA synthesis by actinomycin D did not block the activity of activin (Fig. 4B). Therefore, the effect of activin on spine morphology is independent of protein and RNA synthesis.
F-actin-destabilizing reagents block the effects of activin
To test for an influence of activin on pre- and post-synaptic contact patterns, the effects of F-actin-destabilizing reagents (ADRs) were examined. The ADR cytochalasin D (Cyt D, 0.1 μM) specifically blocked the increase in the average number of presynaptic terminals contacted per spine (Fig. 5A,B). Cyt D alone at this concentration had no effect on the number of contacts. Similarly, Cyt D inhibited the activin-induced lengthening of F-actin in spines (Fig. 5C,D). Another ADR, mycalolide B, also blocked the effect of activin on spine length, and the average spine lengths in neurons treated with activin alone versus neurons treated with activin and 50 nM mycalolide B were 1.59±0.054 μm versus 1.50±0.049 μm, respectively.
The MAP kinase pathway is involved in the effects of activin
Because MAP kinase signaling modulates dendritic spine morphology (Wu et al., 2001) and LTP maintenance (Kelleher, 3rd et al., 2004; Patterson et al., 2001), we asked whether the MAP kinase pathway also mediates the activin signaling that alters spinal morphology. When hippocampal primary cultures were treated with activin, phosphorylation of ERK1/2 increased by twofold within 0.5-1 hour (Fig. 6A-C). This increase was specifically observed in neurons and was not detected in astroglial-enriched cultures treated with activin (Fig. 6C). U0126, a specific inhibitor of MEK, completely inhibited the activin-induced increase in the number of synaptic contacts and the lengthening of dendritic spines (Fig. 6D-F). By contrast, activin did not cause a marked increase in phosphorylation of JNK (supplementary material Fig. S1). Phosphorylation of p38 MAP kinase increased slightly in neurons treated with activin (supplementary material Fig. S1).
We have described here a novel function of activin in modulating actin dynamics in dendritic spines. Activin increased the population of spines with multiple synaptic contacts in cultured hippocampal neurons (Fig. 1D), which resulted in a significant increase in the average number of presynaptic contacts per spine (Fig. 1C). At the same time activin also lengthened dendritic spines (Figs 2, 3). The average length of spines with multiple contacts was greater than that of spines with single contacts (Fig. 3I). Cytochalasin D as well as U0126 inhibited the effects of activin on both spine length and synaptic contact number (Fig. 6). Thus, there is a strong correlation between spine length and the number of presynaptic contacts. A theoretical study has suggested that an increase in spine length leads to an increase in the number of potential synapses by enhancing the chance of making contact with an axonal terminal, which then alters synaptic strength (Stepanyants et al., 2002). Our data suggest that activin increases the number of contacts by lengthening spines. This interpretation is supported by the recent finding that spines grow towards presynaptic buttons to make new synapses in the adult neocortex (Knott et al., 2006).
Our results indicate that activin exerts its effects through the MAP kinase pathway. Thus, hippocampal neurons use a novel signaling pathway that differs from the conventional Smad2/Smad3 cascade (Derynck and Zhang, 2003; Massague et al., 2000); this novel pathway is initiated by the binding of activin to ActRII, leading to gene expression. Activin activates the MAP kinase pathway, specifically ERK1/2, in striatal cells (Bao et al., 2005). The ERK group regulates multiple targets in response to growth factors, but JNK and p38 MAP kinase are activated in response to pro-inflammatory cytokines and environmental stress (Raingeaud et al., 1995). MAP kinase activation is correlated with morphological changes in dendritic spines (Wu et al., 2001). Electrophysiological examination of excitatory synapses from transgenic mice expressing a dominant-negative activin receptor IB mutant in forebrain neurons showed a reduced NMDA current response (Muller et al., 2006). NMDA receptor activation regulates actin-based structural changes in dendritic spines (Fischer et al., 2000). Thus, activin activates NMDA receptors indirectly and it also activates the MAP kinase pathway, so that, overall, it regulates actin dynamics in spines to change their morphology and synaptic connections.
Neurotrophins participate in activity-dependent synaptic plasticity, linking synaptic activity with long-term functional and structural modification of synaptic connections (Poo, 2001). Activin exerts a neurotrophic effect on cultured hippocampal neurons (Iwahori et al., 1997). Protein synthesis is necessary for activin to perform its neuroprotective role. On the other hand, we show here that the morphological change in spines caused by activin is independent of protein and RNA synthesis. Therefore, it is an interesting possibility that, as a neurotrophin, activin uses the Smad pathway and, as a synaptic modulator, it promotes the Smad-independent MAP kinase pathway to regulate spine morphology.
Inhibition of activin function blocks the establishment of late-phase LTP (L-LTP) without affecting early LTP (E-LTP) in the dentate gyrus of the hippocampus (I.K., K.T., H.S. et al., unpublished). We show here that activin signaling modulates dendritic spine morphogenesis and thereby alters the pattern of synaptic contacts. This in turn would contribute to prolonged maintenance of the potentiation of synaptic transmission (L-LTP). The finding that L-LTP is accompanied by an increased number of synaptic puncta that are identified by synaptophysin and N-cadherin (Bozdagi et al., 2000) supports this idea.
Materials and Methods
All procedures involving the use of animals complied with the guidelines of the National Institutes of Health and were approved by the Animal Care and Use Committee of Mitsubishi Kagaku Institute of Life Sciences (MITILS).
Follistatin and activin
Follistatin and activin were prepared from bovine follicular fluid as previously described (Nakamura et al., 1992).
Dissociated primary hippocampal neurons were cultured as described (Goslin et al., 1998). Briefly, rat hippocampi were dissected from 18-day embryos and dissociated with papain. The neurons were then plated at a density of approximately 1.0×104 cells/cm2 on cover slips coated with poly-L-lysine (1 mg/ml in 0.1 M borate buffer) in Dulbecco's modified Eagle's medium (DMEM) (Invitrogen, Carlsbad, CA) with 5% horse serum and 5% fetal calf serum. After the neurons attached to the substrate, they were transferred face down to a dish containing a monolayer of astrocytes and maintained for 18 days in serum-free DMEM with N2 supplements, 1 mM sodium pyruvate and 0.1% ovalbumin. Activin, follistatin, cycloheximide (Sigma), emetine (Alexis, San Diego, CA), actinomycin D (Sigma), cytochalasin D (Calbiochem, La Jolla, CA), mycalolide B (Calbiochem), U0126 (Sigma) and anti-activin A antibody (R&D systems, MN) were directly added to the culture medium from concentrated stocks. In Fig. 6C, astroglial-enriched cultures were prepared as described (Kato et al., 1981).
Hippocampal neurons dissociated as described above were plated at high density (3×105 cells/cm2) on cover slips with silicone rubber walls (Flexiperm). The cultures were transfected with pEGFP (Clontech) at 9 days in vitro (div) with Lipofectamine 2000 (Invitrogen). After images of EGFP-labeled neurons were acquired at 21 div using a fluorescent microscope (Axiovert 200M, Carl Zeiss, Jena, Germany) equipped with a confocal scanner unit (CSU22, Yokokawa, Japan) and a digital camera (ORCA-ER, Hamamatsu, Japan) as a z-series projection, the observed cultures were treated with activin (200 ng/ml) or BSA (1 μg/ml, control) and returned to a CO2 incubator. The images of the pre-recorded neurons were acquired again after incubation for 6 hours. The final images were 3D reconstructions of the stack of z-series projections. Because the boundaries of EGFP-labeled neurons were obscure, the length of individual spines was measured from the center of the spine head to the center of the dendritic stalk.
Immunofluorescence labeling was carried out as previously described (Fukazawa et al., 2003). Briefly, the cells were fixed with 4% paraformaldehyde and treated with PBS containing 5% BSA, 5% goat serum and 0.1% Triton X-100 for permeabilization. Cells were then incubated with the following primary antibodies: anti-Map2 (Chemicon, Temecula, CA), anti-synaptotagmin I (PAb Stg1N) (Shoji-Kasai et al., 1992), anti-synaptophysin (171B5) (Fujita, 1989), or anti-phospho-ERK1/2 (Cell Signaling Technology, Beverly, MA); they were then incubated with secondary antibodies labeled with FITC or Cy5 (Chemicon). Actin was labeled with phalloidin-TRITC (Sigma).
Analysis of dendritic spine morphology
Images of immunostained neurons were acquired with an LSM5 PASCAL confocal microscope (Carl Zeiss). Dendritic spine lengths and areas were quantified using MetaMorph image analysis software (Universal Imaging, Downingtown, PA). Spines were defined as protrusions from the dendritic stalk that contained a rounded head region or that resembled a filopodium without a head. The length of individual spines was measured from the tip of the spine head to the interface with the dendritic stalk using the `region' tool in MetaMorph. The size of a spine head was taken as the spine area that was enclosed by the `region' tool. The pixel numbers of the `region' (distance or area) were imported into Excel for data analysis. To calculate the number of presynaptic contacts per dendritic spine, the number of synaptophysin-positive puncta that were associated with a phalloidin-positive punctum were counted using MetaMorph. The data are expressed as means±standard error and were analyzed by Student's t-test.
Cells were harvested from primary hippocampal cultures (1.25×105 cells/cm2) and homogenized in SDS sample buffer (2% sodium dodecyl sulfate, 10% sucrose, 8 mM EDTA, 0.02% bromophenol blue, 20 mM Tris-HCl, pH 6.8). The homogenates were then subjected to SDS-polyacrylamide gel electrophoresis and transferred onto a membrane filter. The levels of ERK1/2, JNK, p38 MAP kinase and phosphorylated ERK1/2, JNK, p38 MAP kinase were determined with the polyclonal ERK1/2, JNK, p38 MAP kinase antibodies and the monoclonal phospho-ERK1/2 (p44/p42) antibody, polyclonal phospho-p38 MAP kinase and phospho-JNK antibodies (Cell Signaling Technology), respectively, and secondary antibodies and analysis with a chemiluminescence detection system (Super Signal West Dura, Pierce, Rockford, IL). A luminescence image analyzer (LAS-1000, Fuji Photo Film, Tokyo, Japan) was used to quantify immunoreactive bands. Total protein concentrations were measured by the method of Bradford (Protein Assay solution, Pierce) using BSA as a standard. The data are expressed as means±standard error.
The authors thank M. Sekiguchi, K. Hirai and F. Ozawa for the preparation of primary neuron cultures, and Y. Saitoh for critical reading of the manuscript. We also thank T. Shirao and H. Takahashi of Gunma University for valuable advice on culture. This work was supported by Special Coordinate Funds for Promoting Science and Technology and in part by grants for Scientific Research on Priority Areas (A)-Neural Circuit Project and (C)-Advanced Brain Science Project from the Ministry of Education, Culture, Sports, Science and Technology of the Japanese Government to K.I.