|Home | About | Journals | Submit | Contact Us | Français|
We have identified a group of actin-binding/bundling proteins that are expressed in cerebellar Purkinje cells (PCs) but are not detected in other neurons of the central nervous system. These proteins are novel isoforms of the actin-bundling protein espin that arise through the utilization of a unique site for transcriptional initiation and differential splicing. Light and electron microscopic localization studies demonstrated that these espin isoforms are enriched in the dendritic spines of PCs. They were detected in the head and neck and in association with the postsynaptic density (PSD) of dendritic spines in synaptic contact with parallel or climbing fibers. They were also highly enriched in PSD fractions isolated from cerebellum. The PC espins efficiently bound and bundled actin filaments in vitro, and these activites were not inhibited by Ca2+. When expressed in transfected neuronal cell lines, the PC espins colocalized with actin filaments and elicited the formation of coarse cytoplasmic actin bundles. The insulin receptor substrate p53 (IRSp53), an SH3 adapter protein and regulator of the actin cytoskeleton, was identified as an espin-binding protein in yeast two-hybrid screens. Cotransfection studies and pull-down assays showed that this interaction was direct and required the N-terminal proline-rich peptide of the PC espins. Thus, the PC espins exhibit the properties of modular actin-bundling proteins with the potential to influence the organization and dynamics of the actin cytoskeleton in PC dendritic spines and to participate in multiprotein complexes involving SH3 domain-containing proteins, such as IRSp53.
Dendritic spines are dynamic components of the neuron that undergo changes in density or shape during development, learning and disease and in response to hormones, neurotransmitters and synaptic activity (Nimchinsky et al., 2002). Many of these changes reflect the dynamics of the dendritic spine actin cytoskeleton (Matus, 2000). Studies of live neurons indicate that dendritic spines undergo rapid actin-based movements (Fischer et al., 1998; Dunaevsky et al., 1999; Korkortian and Segal, 2001) and that a major portion of spine-associated actin is dynamic (Star et al., 2002). In cultured hippocampal neurons, glutamate receptor activation or electrical stimulation decreases actin-based spine movements (Fischer et al., 2000) and the fraction of dynamic spine actin (Star et al., 2002), but can also lead to the redistribution (Colicos et al., 2001) or loss (Halpain et al., 1998) of spine F-actin.
To understand the molecular mechanisms that underlie dendritic spine dynamics, it is important to identify the proteins that influence the actin cytoskeleton of dendritic spines and mediate its connection to the postsynaptic density (PSD). Efforts to identify the protein components of spines have focused on hippocampal neurons (Zhang and Benson, 2000; Hering and Sheng, 2001). Fewer such studies have examined the spines of cerebellar Purkinje cells (PCs). Implicated in motor control and cognitive functions (Hansel et al., 2001; Molinari et al., 2002; Houk and Mugnaini, 2002), PCs are rich in dendritic spines that receive excitatory input from two major sources. The spines of PC distal dendrites form synapses with ~150,000–200,000 parallel fibers, whereas the sparser spines of PC proximal dendrites form ~1,000–1,500 synapses with a single climbing fiber, resulting in one of the most powerful synaptic contacts in the brain (Strata et al., 2000; Hansel, 2001). There are indications that the spines and PSDs of PCs differ from those elsewhere in the central nervous system in ultrastructure, protein composition and signaling pathways (Carlin et al., 1980; Araki et al., 1993; Brenman et al., 1996; Capani et al., 2001; Okubo et al., 2001, Miyagi et al., 2002).
We have determined that the dendritic spines of PCs contain novel isoforms of the actin-bundling protein espin. Espins have been found previously in association with parallel actin bundles in Sertoli cell-spermatid junctions (Bartles et al., 1996; Chen et al., 1999), brush border microvilli (Bartles et al., 1998) and hair cell stereocilia (Zheng et al., 2000). [Espin should not be confused with epsin, an endocytic adaptor protein with a similar name (De Camilli et al., 2002).] Encoded by a single gene, espin isoforms share a C-terminal peptide that is necessary and sufficient for potent actin-bundling activity, but their N-terminal peptides contain different protein-protein interaction motifs as a result of differences in transcriptional initiation and splicing (Bartles, 2000). Here we report the localization of the novel PC espin isoforms, elucidate their sequences and highlight their interactions with F-actin and the insulin receptor substrate p53 (IRSp53), an SH3 adapter protein and known regulator of the actin cytoskeleton (Krugmann et al., 2001; Miki and Takenawa, 2002).
Young adult Sprague-Dawley rats and CD-1 or CBA/CaJ mice were purchased from Harlan (Indianapolis, IN) or Jackson Labs (Bar Harbor, ME). All experiments conformed to protocols approved by the Northwestern University Institutional Animal Care and Use Committee and Center for Comparative Medicine, an AAALAC-accredited facility, and followed guidelines issued by the National Institutes of Health and the Society for Neuroscience.
Espin antibodies were produced in rabbits and affinity purified on columns of recombinant PC espin 1 or its N- or C-terminal fragments (Bartles et al., 1996; Chen et al., 1999). Rabbit polyclonal antibody to rat PSD-93/Chapsyn-110 and mouse monoclonal actin antibody C4 were from Chemicon (Temecula, CA). Mouse monoclonal tubulin antibody TuJ1 was from Dr. Anthony Frankfurter (Department of Biology, University of Virginia, Charlottesville, VA).
For immunoperoxidase histochemistry, rodents were perfused transcardially with 4% formaldehyde in 0.12 M phosphate buffer, pH 7.4, and brains were infiltrated with 30% sucrose. Frozen sections (30 µm thick) were treated with 0.6% H202 /10% methanol and labeled with espin antibody or preimmune IgG. Bound antibody was visualized by the ABC method (Vector Labs, Burlingame, CA). Some labeled frozen sections were dehydrated, flat-embedded in Epon and examined as 1 µm-thick sections. Images were obtained with a Spot RT CCD camera and Nikon Eclipse 800 microscope. Some frozen sections were labeled with espin antibody and tubulin antibody followed by Alexa488- and Alexa594-labeled goat secondary antibodies or phalloidin (Molecular Probes, Eugene, OR), and optical z-sections (1.5 µm thick) were obtained using a Nikon microscope with a PCM 2000 confocal laser scanning system. For labeling at the electron microscopic level, rats were perfused with 4% formaldehyde (with or without 0.1% glutaraldehyde) in 0.12 M phosphate buffer, pH 7.4. Brain sections (60 µm thick) were cut on a Vibratome, cryoprotected with glycerol-dimethylsulfoxide mixtures, frozen and thawed four times, treated with 1% sodium borohyride followed by 0.6% H202 /10% methanol and labeled with espin antibody or preimmune IgG. Bound antibody was visualized by the peroxidase-antiperoxidase method (Sternberger Immunochemicals, Lutherville, MD). Sections were treated with 2% OsO4, 1% uranyl actetate, dehydrated and flat-embedded in Epon. Ultrathin sections were counterstained with uranyl acetate and lead citrate and examined on a Zeiss EM10 electron microscope at 80 kV.
PSDs were isolated from rat cerebellum by subcellular fractionation in the presence of 1 mM phenylmethylsulfonyl fluoride, 1 µg/ml antipain and 1 µg/ml leupeptin using the modified method of Carlin et al. (1980) followed by Dosemeci and Reese (1993). Proteins were detected on western blots using the ECL method (Amersham Biosciences). In some experiments, a modified PSD fraction, prepared using only a single extraction with 0.5% Triton X-100, was subjected to additional extraction (Cho et al., 1992).
The sequences of the PC espin isoforms were inferred by DNA sequence analysis of overlapping PCR products resulting from RT-PCR and 5’ RACE-PCR reactions conducted using RNA isolated from rat and mouse cerebellum and selected espin primers in conjunction with kits and reagents purchased from Invitrogen (Carlsbad, CA). Full-length cDNAs for the rat PC espins and rat PC espin 1 deletion constructs missing the N-terminal (96-SSLPPPPPPSFPPPPPPGTQLPPPPGTPAPNPPVGL-132 = δ1) or C-terminal proline-rich peptide (256-PPPPPPPPLPEALSSPPPAPPLPIEG-281 = δ2) were prepared by ligating selected restriction fragments. A near full-length cDNA encoding the short form of mouse IRSp53 (GenBank accession no. BC016411) was obtained in yeast two-hybrid screens (see below). A cDNA encoding the 186-amino acid C-terminal peptide (M367-R552) of the long form (transcript variant 2) of human IRSp53 (GenBank accession no. NP_059345) was obtained from HeLa cell RNA by RT-PCR.
Screening was carried out using the Clontech Matchmaker GAL4 System 2 (BD Biosciences Clontech). A cDNA encoding the N-terminal portion of rat PC espin 1 (D13–Q421), which included the two proline-rich peptides, was fused in-frame to the GAL4 binding domain of the pAS2-1 vector and used as “bait” to screen Clontech Matchmaker pACT2 cDNA libraries from mouse brain and testis in Y190 yeast cells. His+ transformants were confirmed through a second round of growth on selection plates and a β-galactosidase filter assay.
PC espin isoforms were expressed in E. coli BL-21 Star (DE3) with a short (~1.5-kDa) N-terminal 6 × His tag using the ProEXHTa vector (Invitrogen), affinity purified from soluble bacterial extracts under nondenaturing conditions on Ni-NTA agarose (Qiagen, Valencia, CA), dialyzed against assay buffer and clarified by ultracentrifugation prior to use (Bartles et al., 1998, Chen et al., 1999). Cosedimentation F-actin-bundling assays were carried out as described (Bartles et al., 1998, Chen et al., 1999). For pull-down assays, the 186-amino acid C-terminal peptide of human IRSp53, which includes its SH3 domain, was expressed as a glutathione S-transferase (GST) fusion protein using the pGEX4T-3 vector (Amersham Biosciences). Recombinant rat PC espin proteins were incubated with glutathione-Sepharose 4B beads preloaded with GST-IRSp53 SH3 domain fusion protein or GST alone in 0.1 M KCl, 20 mM imidazole-HCl, 1 mM DTT, 1.5 mM NaN3, pH 7.4, for 1 h at 4 °C. After washing 4 times at 13,000 × g for 30 s, the bound proteins were released by heating in SDS and resolved in SDS gels.
Cells of the mouse Neuro-2a neuroblastoma line (American Type Culture Collection, Manassas, VA) were cultured in DMEM containing 10% fetal bovine serum and differentiated by culturing in DMEM containing 2% fetal bovine serum and 0.5 mM dibutyryl-cAMP (Sigma, St. Louis, MO) for 1–11 d (Wu et al., 1998). 24 h prior to transfection, cells were trypsinized and replated on coverslips coated with poly-L-lysine (Sigma) followed by laminin. Transient transfection with Lipofectamine (Invitrogen), fixation, immunolabeling and examination by fluorescence microscopy were carried out as described previously (Chen et al., 1999) using a Zeiss Axioplan 2 Imaging microscope system equipped with an Axiocam digital camera. The PC espins and the mouse IRSp53 construct were expressed using pEGFP-C vectors (BD Biosciences Clontech) and detected as green fluorescent protein (GFP) fusions. For cells examined 8 h after transfection, the signal was boosted using GFP monoclonal antibody (Roche, Indianapolis, IN) followed by Alexa488-secondary antibody (Molecular Probes). For cotransfections, untagged PC espins were expressed using the pcDNA3 vector (Invitrogen). Multiple labeling for F-actin and DNA were carried out with using Texas Red-phalloidin or 4’,6-diamidino-2-phenylindole (Sigma), respectively.
When frozen sections of rat or mouse brain were labeled with affinity purified rabbit polyclonal espin antibodies, intense specific immunostaining was observed in the cerebellar cortex (Fig. 1A and B), but was absent in cerebral cortex, hippocampus, striatum and brainstem. The dark diaminobenzidine (DAB) reaction product occupied the PC and molecular layers of the cerebellar cortex, while the granular layer and white matter were free of staining (Fig. 1A and B). When immunoperoxidase-labeled frozen sections of rat cerebellum were embedded in plastic and analyzed as 1 µm-thick sections, the DAB reaction product was detected as dark puncta, ~1 µm in diameter, distributed densely over the molecular layer and as a weaker diffuse staining in the cell bodies and stem dendrites of PCs, but not in their axons (Fig. 1B and C). The density of puncta over the molecular layer was estimated to be ~2.5- to 3-fold higher than over the cell bodies of PCs. No other cell types were labeled, including the granule and Golgi cells of the granular layer and the stellate and basket cells of the molecular layer. Therefore, these results suggested that PCs were the only cerebellar cell type containing espins and that the PC espins were enriched in PC dendritic spines.
A dense punctate staining of the molecular layer was also observed when espins were localized by confocal immunofluorescence (Fig. 2B). Consistent with a localization to PC dendritic spines, the espin-positive puncta showed extensive overlap with F-actin-rich puncta as revealed by double labeling with fluorescent phalloidin (Fig. 2A–C). Moreover, double labeling for espin and tubulin showed an enrichment of espin in tiny (~1 µm), gemmule-shaped projections studding the tubulin-positive shafts of PC spiny branchlets (Fig. 2D and E). The level of espin antibody staining appeared consistently intense in the heads of the spines.
Espin localization at the electron microscopic level definitively confirmed immunostaining of PC dendritic spines (Fig. 3). Dendritic spines containing DAB reaction product were involved in synapses with either parallel fibers (Fig. 3B and C) or climbing fibers (Fig. 3D). After fixation with 4% formaldehyde and moderate permeabilization of the Vibratome sections by repeated freezing and thawing prior to immunolabeling, the majority of PC spines stained positively for espin (Fig. 3A). In these sections, the PSD, subsynaptic globules and tubules of the endoplasmic reticulum in the head and neck were stained components of the spines (Fig. 3A and B). Little staining was detected in the dendritic shaft. Addition of even small amounts (0.1%) of glutaraldehyde to the fixative resulted in a sharp decline in immunostaining, causing it to appear more restricted to the PSD (Fig. 3C and D). The conspicuous labeling of the PSD was remarkable, although it is well known that the DAB reaction product can diffuse locally within the ~1 µm-thick spine. Attempts to localize PC espins by postembedment immunogold labeling of ultrathin sections prepared using LR White (Bartles et al., 1998) or the freeze-substitution, Lowicryl embedding method (Landsend et al., 1997; Roche et al., 1999) were not successful, presumably due to epitope masking or inactivation. We were, however, able to obtain independent biochemical evidence in support of the localization of PC espins to the PSD using subcellular fractionation (see below).
Western blot analysis using affinity purified espin antibodies showed the presence of multiple immunoreactive polypeptides in cerebellum. Intense specific labeling of a poorly resolved doublet at 60–65 kDa was observed in homogenate of rat cerebellum (Fig. 4A). A minor band at ~53 kDa was also labeled specifically. The pattern of specific labeling for mouse cerebellar homogenate showed a major band at ~68 kDa and a poorly resolved multiplet at 55–60 kDa (Fig. 4A). No specific labeling was detected in homogenate of cerebral cortex (Fig. 4A).
The multiplicity of bands observed on western blots appeared to be due to the presence of multiple espin isoforms in PCs. Sequence analysis of cDNAs obtained by RT-PCR and 5’-RACE PCR using espin primers revealed the presence of transcripts for four novel espin isoforms in RNA isolated from cerebellum (GenBank accession numbers AF540942–AF540945 for mouse and AF540946–AF540949 for rat). The stick-figure diagram in Fig. 4B summarizes the structural relationships between these four espin isoforms, which we have designated PC espins 1, 1+, 2 and 2+. It also includes an updated map of the mouse espin gene showing the relative positions and sizes of known exons. Mouse and rat counterparts were 92–93% identical at the amino acid sequence level. The C-terminal peptides of all four PC espins (encoded by exons w–z) were identical to those of the other espin isoforms characterized to date (Chen et al., 1999), but their N termini were different. Overall, their N termini more closely resembled that of the ~110-kDa Sertoli cell espin isoform (Bartles et al., 1996) than the ~30-kDa small espin isoform of enterocytes and renal proximal tubular epithelial cells (Bartles et al., 1998). For example, all four PC espin isoforms contained the two proline-rich peptides (exons o and r) and the additional F-actin-binding peptide (exon o) just downstream of the N-terminal proline-rich peptide. However, in place of the ankyrin-like repeats found at the N terminus of Sertoli cell espin (Bartles et al., 1996), the PC espins contained a unique 3-amino acid peptide (initiator methionine plus two additional amino acids) encoded, along with the 5’-UTR, by exon m. In addition, the PC espins were missing the short peptide encoded by exon p. Two of the isoforms, PC espin 2 and 2+, were missing the peptide between the two proline-rich peptides encoded by exon q (asterisk in Fig. 4B), thereby bringing the two proline-rich peptides closer together in primary sequence. Finally, two of the isoforms – one with and one without the peptide encoded by exon q – contained an additional 9-amino acid peptide rich in positively charged amino acids (KVRVLRHRK in mouse, KVRILRHRK in rat) encoded by small exon s (+ in Fig. 4B) and hence were designated PC espins 1+ and 2+, respectively.
To clarify the relationship between cDNA sequence and apparent molecular mass, rat versions of all four PC espin isoforms were expressed in untagged form in transiently transfected cells of the mouse Neuro-2a neuroblastoma line and also in E. coli with a short (~1.5-kDa) N-terminal 6 × His-tag (Fig. 4A). Regardless of expression system, the rat PC espin proteins migrated in SDS gels in the same general region where the immunoreactive bands were detected on western blots of cerebellum (Fig. 4A). The minor bands detected at lower positions on the western blot of the transfected Neuro-2a cells represent proteolytic breakdown products. Ignoring the effects of post-translational modification, these comparisons suggested that the ~60-kDa PC espin 1 and, to a lesser extent, the ~65-kDa PC espin 1+ were the two major espin isoforms present in rat cerebellum. Only small amounts of proteins comigrating with the expressed rat PC espins 2 or 2+ were detected on the western blots of rat cerebellum. A different outcome was obtained for the mouse, whose PC espin isoforms are predicted to be ~2-kDa larger than their rat counterparts due to minor differences in amino acid sequence. More than half of the label detected on the western blots of mouse cerebellum migrated at the lower molecular mass (55–60 kDa) expected for mouse PC espins 2 and 2+. In agreement with this trend toward increased expression of PC espin 2 and 2+ proteins in the mouse, RT-PCR analysis of cerebellar RNA showed a greater relative abundance of transcripts encoding PC espins 2 and 2+ in mouse than in rat (not shown).
Consistent with the presence of the C-terminal peptide shared among known espin isoforms and the additional F-actin-binding site encoded by exon o (Chen et al., 1999), all four PC espin isoforms efficiently bound and bundled preformed actin filaments in vitro under physiological buffer conditions. These results are illustrated in Fig. 5A using a standard cosedimentation actin-bundling assay. When preformed actin filaments were incubated in the absence of espins and then centrifuged at medium speed, >95% of the preformed actin filaments remained in the supernatant. In contrast, when even very low amounts of the PC espin isoforms were mixed with the preformed actin filaments (in this case only ~1 espin for every 15–20 actin monomers), most of the actin filaments were recovered in the pellet. This shift of F-actin from the supernatant to the pellet in this assay is indicative of actin bundle formation (Bartles et al., 1998; Chen et al., 1999). From this assay, it was also apparent that all four PC espin isoforms bound efficiently to actin filaments. In the presence of F-actin, the PC espins were quantitatively recovered in the pellet, but in the absence of F-actin they remained soluble and were not detected in the pellet (Fig. 5A). Actin bundle formation was also noted in these mixtures by increases in solution turbidity and by negative staining electron microscopy (not shown), which confirmed the presence of copious actin bundles resembling the partially ordered parallel actin bundles elicited by other espin isoforms (Bartles et al., 1998; Chen et al., 1999). Unlike other proteins with actin-bundling activity, such as villin and fimbrin (Glenney et al., 1981; Alicea and Mooseker, 1988; Namba et al., 1992; Lin et al., 1994), the espins appear not to be inhibited by Ca2+. This result is illustrated in Fig. 5B for rat PC espin 1, which shows no difference in the ability to bind and bundle F-actin in cosedimentation assays conducted in the presence of EGTA (to chelate any trace Ca2+ that might be present in the buffer) or 20 µM CaCl2. Similar results were obtained with using all four PC espin isoforms and for CaCl2 concentrations as high as 100 µM (not shown).
To examine the localization and effects of the PC espins in vivo, we expressed GFP-tagged versions of the four rat PC espin isoforms by transient transfection in cells of the mouse Neuro-2a neuroblastoma line. We did not detect endogenous espins in these cells using espin antibody. The results were similar for the four rat PC isoforms, so only examples of rat PC espin 1 are shown in Fig. 6A–F. When examined at times after transfection ranging from 8 h to 48 h, the GFP-PC espins were colocalized with F-actin. At early times after transfection (8 h), when levels of the GFP-PC espins were still relatively low, the GFP-PC espins were localized to spiky, F-actin-rich projections (filopodia) and arcs beneath some concave cell margins (Fig. 6A and B). As the time after transfection and level of GFP-PC espin increased, the GFP-PC espins and F-actin remained colocalized, but the labeling of filopodia decreased as espin-rich coarse cytoplasmic F-actin bundles became the dominant labeled structure (Fig. 6C and D). These coarse cytoplasmic F-actin bundles were not detected in untransfected control cells (not shown). A similar effect was noted previously upon expression of other espin isoforms in cells of the NRK and BHK fibroblastic lines (Bartles et al., 1996, 1998; Chen et al., 1999), but was more evident in the Neuro-2a cells because they contain fewer actin stress fibers than NRK or BHK cells. In cells displaying a neuronal morphology, the coarse cytoplasmic espin/F-actin bundles were frequently present in large neurites (Fig. 6E and F). None of these results depended on the presence of the GFP tag, because identical results were obtained when the PC espin isoforms were expressed in untagged form using the pcDNA3 vector and detected with espin antibody (not shown).
To obtain independent biochemical verification for the presence of espins in PC PSDs, PSDs were isolated from rat cerebellar homogenate by centrifugation in discontinuous sucrose density gradients using standard procedures. When western blots containing 1 µg of protein from cerebellar homogenate, synaptosomes and PSD fraction were labeled with espin antibody, a dramatic enrichment of PC espins was observed in the PSD fraction (Fig. 7A). The extent of enrichment observed for the PC espins was similar to that noted for a known protein of the PC PSD (Brenman et al., 1996), the membrane-associated guanylate kinase PSD-93/chapsyn-110 (Fig. 7A). The weak band positioned immediately beneath the espin band in the synaptosome lane was due to nonspecific binding and was not evident on the western blot containing 30 µg of protein (Fig. 7A). We next checked to see whether PC espins fractionated like “core” proteins of the PSD, as defined operationally by resistance to extraction with 3% n-laurylsarcosine (Cho et al., 1992). PSDs isolated from rat cerebellum by a single extraction with 0.5% Triton X-100 were subjected to a second extraction step involving buffer alone, 0.5% Triton X-100 or 3% N-laurylsarcosine. The PC espins remained associated with the PSD fraction following a second extraction in Triton X-100 (Fig. 7B), suggesting that they were tightly associated with the PSD. However, they were mostly solubilized by 3% n-laurylsarcosine (Fig. 7B), suggesting that they do not represent core proteins of the PSD. In contrast, PSD-93/chapsyn-110 better resisted extraction with 3% n-laurylsarcosine (Fig. 7B). The 3% n-laurylsarcosine also extracted the majority of the actin from the PSD fraction (Fig. 7B).
We obtained evidence for a direct binding interaction between PC espins and IRSp53, an SH3 adapter and known regulator of the actin cytoskeleton (Krugmann et al., 2001; Miki and Takenawa, 2002) that has previously been localized PC PSDs (Abbott et al., 1999). The N-terminal portion of PC espin 1, including its two proline-rich peptides, was used as bait in a yeast two-hybrid screen to identify candidate espin-binding proteins in commercial mouse brain and testis cDNA libraries. Out of the 35 positive clones obtained in two independent screens, 7 were overlapping clones of the short form of mouse IRSp53 (GenBank accession number BC016411). The inserts ranged in size from 1.8 to 2.5 kb. All encoded the C-terminal half of IRSp53, which contained its SH3 domain (Yeh et al., 1998; Krugmann et al., 2001; Miki and Takenawa, 2002) and included the 3’-UTR. The longest clone was nearly full length, starting at A49. To confirm this binding interaction in mammalian cells, we expressed this longest IRSp53 construct as a GFP fusion in the presence of untagged versions of the rat PC espins in transiently cotransfected mouse Neuro-2a neuroblastoma cells. Regardless of PC espin isoform, the GFP-IRSp53 colocalized with the espin-rich coarse cytoplasmic F-actin bundles shown previously (Fig. 6C and D) to be elicted by expression of the PC espins. This result is illustrated for rat PC espin 1 in Fig. 6G and H. This interaction required the N-terminal proline-rich peptide of the PC espins. Deletion of either proline-rich peptide from the PC espins had no effect on the ability to elicit the formation of coarse cytoplasmic F-actin bundles (cf. Fig. 6G, I and K with control in Fig. 6M) or to become colocalized with the F-actin bundles (data not shown). When expressed in the presence of PC espins that were missing the C-terminal proline-rich peptide (δ2), the GFP-IRSp53 maintained its colocalization with the coarse cytoplasmic espin/F-actin bundles (Fig. 6I and J). In contrast, deletion of the N-terminal proline-rich peptide of the PC espins (δ1) eliminated the binding of GFP-IRSp53 to the espin/F-actin bundles. Instead, the GFP-IRSp53 showed lower levels of accumulation and localization to a dense aggregate, possibly an aggresome (Kopito, 2000), in the perinuclear region (Fig. 6K and L). This was the same pattern of localization observed for the GFP-IRSp53 when it was expressed in the absence of the PC espins (Fig. 6M and N). These results suggested that the N-terminal proline-rich peptide of the PC espins bound IRSp53, presumably through its SH3 domain.
The direct nature of this interaction between the PC espins and the SH3 domain-containing C-terminal peptide of IRSp53 was confirmed in-vitro using pull-down assays. When glutathione-Sepharose beads loaded with GST-IRSp53 SH3 domain fusion protein or GST alone were incubated with the PC espin isoforms, relatively large amounts of the espins bound to the GST-IRSp53 beads, but not to the GST control beads (Fig. 7C). All bead pellet samples contained minor amounts of a bacterial protein that comigrated with PC espins 1 and 1+ and was contributed by the beads loaded with GST or GST-IRSp53 SH3 domain, not the PC espins. Additional pull-down assays comparing equivalent amounts of PC espin 1 and PC espin 1 missing either its N-terminal or C-terminal proline-rich peptide confirmed the involvement of the N-terminal proline-rich peptide. PC espin 1 and PC espin 1 missing its C-terminal proline-rich peptide (δ2) bound to the beads at high levels (Fig. 7D). However, deletion of the N-terminal proline-rich peptide of PC espin 1 (δ1) decreased the level of binding to the low background levels noted for beads loaded with GST alone (Fig. 7B).
We have localized espins to PCs in the cerebellar cortex of mouse and rat and identified four novel espin isoforms in these neurons. These espins are enriched in PC dendritic spines and are not detected in other spiny neurons of the central nervous system. The PC espins exhibit the properties of modular actin-bundling proteins that could efficiently cross-link the actin cytoskeleton in PC dendritic spines and/or mediate its connection to the PSD by binding SH3 adapter proteins. The PC espins show no obvious relationship to other modular actin-binding proteins, such as spinophilin/neurabin-II, synaptopodin or α-actinin-2, that are present in the dendritic spines of other brain regions but are either absent or expressed at low levels in PCs (Mundel et al., 1997; Allen et al., 1997; Wyszynski et al., 1998; Satoh et al., 1998; Deller et al., 2000).
Previously, espins have been detected only in epithelial cells (Bartles et al., 1996, 1998; Zheng et al., 2000). Thus, this first report of espins in neurons indicates that the espin family of actin-binding/bundling proteins has a wider cell-type distribution than originally appreciated. Although novel in sequence, the PC espin isoforms fit the established pattern for espin isoform structure (Bartles, 2000): they contain the C-terminal peptide shared by all known isoforms, but they differ from other espin isoforms in the N-terminal peptides that are connected to the shared C-terminal peptide. A PC-specific site for transcriptional initiation specifies espin isoforms that resemble superficially a truncated version of the ~110-kDa Sertoli cell espin missing its N-terminal ankyrin repeats (Bartles et al., 1996). However, the situation is further complicated by differential splicing, which through the differential utilization of exons q and s, accounts for a total of four sequence permutations. The significance of these differences in splicing with regard to isoform localization or function remains to be established.
All four PC espin isoforms contain the espin C-terminal peptide, which is believed to include at least two F-actin-binding sites and is both necessary and sufficient for espin-mediated actin-bundling activity in vitro (Bartles et al., 1998). All four PC espin isoforms also contain a third F-actin-binding site, the additional F-actin-binding site encoded by exon o (Chen et al., 1999). Accordingly, all four PC espins appear as efficient as Sertoli cell espin constructs at binding and bundling actin filaments in vitro (Chen et al., 1999). In view of the high activity they demonstrate in actin-bundling assays and their structural similarity to the ΔN338 truncated version of Sertoli cell espin characterized previously in quantitative binding assays (Chen et al., 1999), it is likely that PC espins also bind F-actin with Kds in the 50–100 nM range. This affinity is 1–2 orders of magnitude greater than that observed for many other actin-binding and actin-cross-linking proteins (Bartles, 2000). These observations lead naturally to the hypothesis that the PC espins function in part to cross-link actin filaments in PC spines in situ.
Light and electron microscopic immunocytochemistry demonstrates that the PC espins are enriched in dendritic spines, although they are distributed to some extent throughout the somatodendritic compartment of the PC. Compared to the parallel actin bundles where other espin isoforms have been localized (Bartles, 2000; Zheng et al., 2000), the actin filaments of dendritic spines appear sparse, less organized and more dynamic (Landis and Reese, 1983; Hirokawa, 1989; Matus, 2000; Fischer et al., 1998, 2000; Star et al., 2002; Colicos et al., 2002). Nevertheless, the PC espins could still perform an actin cross-linking function and, thereby, stabilize the actin cytoskeleton of PC dendritic spines. This could help explain why PC cell spines appear relatively similar in size and shape (Strata et al., 2000) and uniformly rich in F-actin (Capani et al., 2001). A stabilizing effect on the actin cytoskeleton might also help explain what appears to be a unique attribute of PC spines: the ability to grow and/or to be retained in the absence of afferents (Hirano et al., 1977; Takács et al., 1997; Bravin et al., 1999).
When expressed in transfected cells, the PC espins consistently elicit the formation of coarse cytoplasmic F-actin bundles and appear to increase the levels of F-actin. These effects were noted previously in transfected fibroblastic cells expressing espins (Bartles et al., 1996, 1998; Chen et al., 1999). It is unclear whether the increase in the level of F-actin reflects an up-regulation of actin synthesis in response to the creation of an espin cross-linked actin bundle “sink” for actin monomer (Lyubimova et al., 1999) or whether espins might also somehow stimulate actin polymerization.
One characteristic property of the espins is that their actin-bundling activity is not inhibited by Ca2+. To account for the degeneration of the parallel actin bundles in the hair cell stereocilia of espin-deficient homozygous jerker mice, we have postulated that espin cross-links stabilize the actin bundles against the transient local increases in Ca2+ that accompany mechanoelectrical signal transduction in hair cell stereocilia (Zheng et al., 2000). Dendritic spines are known to experience transient local increases in Ca2+ concentration during synaptic transmission (Nimchinsky et al., 2002; Segal, 2002; Holthoff et al., 2002). Depending upon assay system and conditions, increases in Ca2+ concentration have been reported to reduce spine motility (Fischer et al., 2000), decrease the fraction of dynamic spine actin (Star et al., 2002) and, at higher concentrations, elicit the selective loss of F-actin from spines (Halpain et al., 1998). PC dendritic spines experience large activity-dependent increases in the concentration of Ca2+ through voltage-dependent entry and inositol–1,4,5-trisphosphate-mediated release from intracellular stores (Finch and Augustine, 1998; Wang et al., 2000; Okubo et al., 2001). In fact, these increases in Ca2+ concentration are correlated with cerebellar long-term depression and motor learning (Wang et al., 2000; Hansel et al., 2001). Thus, it is possible that espins may also help protect the actin cytoskeleton of PC spines from the excitotoxic effects associated with these large increases in Ca2+ concentration. We are currently examining the structure and function of PCs in jerker mice to determine whether they show defects in morphology, electrophysiology or dynamics.
Our immunocytochemical evidence suggests a close association between the PC espins and the PSD and explains why PC espins are so highly enriched in cerebellar PSD fractions. It is unclear whether this association is mediated by direct binding between the PC espins and PSD-associated actin filaments (Landis and Reese, 1983; Matus et al., 1982; Hirokawa, 1989) or whether the PC espins bind PSD proteins other than actin. Representatives of >70 families of proteins have been identified in dendritic spines and PSDs (Kennedy, 2000; Scannevin and Huganir, 2000; Zhang and Benson, 2000; Husi et al., 2000; Hering and Sheng, 2001). Among these are proteins that interact with actin filaments indirectly as members of signaling cascades or multiprotein adapter complexes. It is possible that the PC espins bind such proteins via domains believed not to participate directly in binding F-actin. Promising candidates include the two proline-rich peptides of the PC espins. We identified IRSp53 as an SH3 adapter protein that binds to the N-terminal proline-rich peptide of the PC espins in vivo and in vitro. This interaction may be physiologically relevant, because IRSp53 has been localized previously to PC PSDs. IRSp53 mRNA and protein are moderately abundant in rat cerebellum, and, among cerebellar cell types, high levels of transcript are detected only in PCs (Abbott et al., 1999; Thomas et al., 2001). Moreover, the IRSp53 protein has been shown to be highly enriched in rat cerebellar PSD fractions and, when localized by immunofluorescence, to be concentrated at synapses distributed densely throughout the molecular layer of the cerebellar cortex (Abbott et al., 1999). IRSp53 has been identified independently as a substrate for the insulin receptor tyrosine kinase (Yeh et al., 1996) and as a protein ligand for brain angiogenesis inhibitor 1 (Oda et al., 1999) and for atrophin-1, the product of the dentatorubral-pallidoluysian atrophy gene (Okamura-Oho et al., 1999). It is unclear how IRSp53 is tied to signaling by insulin or insulin-like growth factors in the brain, but recent discoveries suggest that IRSp53 can regulate the actin cytoskeleton. Upon binding Cdc42 to its partial Cdc42- and Rac-interactive binding (CRIB) domain, IRSp53 avails its SH3 domain for binding the Ena/VASP family member Mena, and the two proteins appear to act synergistically to promote the formation of F-actin-rich filopodia or neurites (Krugmann et al., 2001; Govind et al., 2001). Conversely, WAVE2, a nucleation-promoting factor for Arp2/3-mediated actin polymerization, can bind the SH3 domain of IRSp53 and thereby enhance the binding of Rac to its CRIB domain (Miki and Takenawa, 2002). Rac is known to regulate the numbers and sizes of PC dendritic spines (Luo et al., 1996; Nakayama et al., 2000; Tashiro et al., 2000). Thus, in addition to affecting the organization and dynamics of PC dendritic spines directly, the PC espins may influence these parameters indirectly through their participation in a multiprotein scaffold that contains IRSp53 or other SH3 adapter proteins. The existence of an alternate multiprotein scaffold in the PC PSD may explain why PSD-93 knockout mice fail to show structural or functional abnormalities (McGee et al., 2001).
This work was supported by NIH grant DC04314 and Independent Scientist Award HD01210 awarded to J.R.B. We thank Drs. Anthony Frankfurter providing tubulin antibody.