The N-terminal Region of Neuregulin Isoforms Determines the Accumulation of Cell Surface and Released Neuregulin Ectodomain*

Two neuregulin-1 isoforms highly expressed in the nervous system are the type III neuregulin III- b 1a and the type I neuregulin I- b 1a. The sequence of these two isoforms differs only in the region that is N-terminal of the bioactive epidermal growth factor-like domain. While the biosynthetic processing of the I- b 1a isoform has been well characterized, the processing of III- b 1a has not been reported. In this study, we compared III- b 1a and I- b 1a processing. Both III- b 1a and I- b 1a were synthesized as transmembrane proproteins that were proteolytically cleaved to produce an N-terminal fragment containing the bioactive epidermal growth factor-like domain. For I- b 1a, this product was released into the medium. However, for III- b 1a, this product was a transmembrane protein. In cultures of cells expressing III- b 1a, the amount of neuregulin at the cell surface was much greater, and the amount in the medium was much less than in cultures expressing I- b 1a. Phorbol ester treatment and truncation of the cytoplasmic tail had markedly different effects on III- b 1a and I- b 1a processing. These results demonstrate an important role for the N-terminal region in determining neuregulin biosynthetic processing and show that a major product of III- b 1a processing is a tethered ligand that may act as a cell surface signaling molecule. Neuregulin-1

Neuregulin-1 gene products are cell-cell signaling proteins that are ligands for receptor tyrosine kinases of the ErbB/HER subfamily (for reviews, see Refs. [1][2][3][4]. Signaling events mediated by neuregulins (NRGs) 1 have been shown to be essential for normal development of the nervous system and heart (5)(6)(7)(8)(9). Roles for NRG-1 proteins in the development of other organs and in the adult have also been suggested.
Gene transcripts encoding at least 14 different NRG isoforms have been identified (2). Based on the structure of their Nterminal region, NRG-1 isoforms can be divided into three types (Refs. 1 and 10; see "Neuregulin Isoforms and Nomenclature" and Fig. 1A). The NRG isoforms originally called neu differentiation factor (11 and 12) and heregulin (13) have a type I N terminus. The N terminus of chick ARIA (14) is most similar to the mammalian type I N terminus. The isoforms originally called glial growth factor (15,16) that include a "kringle domain" have a type II N terminus. The isoforms originally called SMDF, nARIA, or CRD-NRG (17-21) have a type III N terminus. The type I and type II isoforms contain an Ig-like domain in the N-terminal region (Ig-NRGs), whereas the type III isoform contain a cysteine-rich domain (CRD-NRGs). Studies of the temporal and spatial expression patterns of NRG isoforms suggest that type I/II and type III NRGs serve distinct functions (10,18). This inference has been confirmed by "knockouts" that have specifically deleted the Ig-NRGs (6) or the CRD-NRGs (Ref. 8; see also Ref. 10). With respect to the type III NRGs, analysis of the knockout animals indicates that type III NRGs play a critical role in the interactions of peripheral nerve axons with muscle and with Schwann cells. Defects in animals lacking type III NRGs include retraction of nerve terminals from newly formed synapses, absence of Schwann cells from peripheral nerves, and loss of motor and sensory neurons (8).
The cellular processing of type I NRGs has been the subject of several investigations (22)(23)(24)(25). Most type I NRG isoforms include a hydrophobic stretch of amino acids C-terminal of the epidermal growth factor (EGF)-like) domain that serves as a transmembrane (TM) domain. These type I isoforms are synthesized as transmembrane "proproteins" from which a paracrine signal can be produced by proteolytic cleavage in the stalk region (Fig. 1B). The release of the ectodomain into the extracellular space may require transport of the uncleaved proprotein to the cell surface, followed by stalk cleavage and "shedding" of the ectodomain from the surface. Alternatively, after arriving at the surface, the proprotein may be internalized into endocytotic pathway compartments in which proteolytic processing occurs with subsequent secretion of the ectodomain fragment (24). Release of the type I NRG ectodomain into culture medium is accelerated by activation of protein kinase C (12,22,24) and blocked by shortening the cytoplasmic tail to fewer than ϳ90 amino acids (7,26) or by deletion of specific segments within this 90-amino acid stretch (26).
NRG III-␤3 was the first reported NRG isoform with a type III N-terminal region (17). This isoform, originally referred to as SMDF, has a ␤3 type of EGF-like domain, and thus lacks the transmembrane domain C-terminal of the EGF-like domain (Fig. 1A). Subsequent to the isolation of the III-␤3 isoform, a III-␤1a isoform (Fig. 1A) was discovered (Refs. 18 -20; see also Ref. 27). Based on RNA studies (10, 12, 18 -20, 28 -30), it is likely that the III-␤1a isoform is one of the most abundant NRG isoforms in the late embryonic and postnatal nervous system and that the III-␤3 isoform is relatively rare.
The III-␤1a and I-␤1a isoforms differ in their N-terminal regions but are identical in their EGF-like, "stalk," transmem-brane, and cytoplasmic regions (Fig. 1A). Based on the similarity in structure of the III-␤1a and I-␤1a isoforms, schematic diagrams (4,8) have represented the topology of III-␤1a as similar to I-␤1a (Fig. 1, B and C). This model of III-␤1a topology suggests that III-␤1a, like I-␤1a, will be cleaved in the stalk region and that the consequence of this cleavage will be release of the entire N-terminal fragment into the extracellular fluid. To test this model and to determine whether the cellular mechanisms that regulate I-␤1a topology and biosynthetic processing also govern III-␤1a, we have compared III-␤1a processing to I-␤1a processing using biochemical and immunocytochemical techniques.

EXPERIMENTAL PROCEDURES
Neuregulin Isoforms and Nomenclature-The isoform designation "III-␤ 1a" refers to a NRG with a "type III" N-terminal region, a "␤ " type EGF-like domain, a "1" type sequence at the carboxyl terminus of the EGF-like domain, and an "a" type cytoplasmic tail (see Fig. 1A for illustration). NRG isoforms can differ in the sequence of their N-terminal region (type I, II, and III), EGF-like domain (␣ or ␤), the C-terminal end of the EGF-like domain (1, 2, 3, or 4), and cytoplasmic tail (a, b, c, or none). In NRG isoforms with ␤1, ␤2, or ␤4 EGF-like domains, there is a stalk region, transmembrane domain, and cytoplasmic tail Cterminal of the EGF-like domain. (The stalk is the sequence between the EGF-like domain and the transmembrane domain.) These features are not found in NRG isoforms with a ␤3 EGF-like domain. Most NRGs FIG. 1. Structure of the I-␤1a, III-␤1a, and III-␤3 NRG isoforms and constructs used in this study. A, structural regions of the full-length I-␤1a, III-␤1a, and III-␤3 proteins. See "Neuregulin Isoforms and Nomenclature" for discussion of NRG isoform names. III-␤1a and I-␤1a differ only in their N-terminal region; their sequence is identical from the EGF-like domain through the carboxyl terminus. The sequence of III-␤1a and III-␤3 is identical from the N terminus to just beyond the end of the EGF-like domain. The III-␤3 isoform terminates 20 amino acids following the last cysteine of the EGF-like domain and thus lacks the stalk, transmembrane, and cytoplasmic regions shared by III-␤1a and I-␤1a. Based on hydrophobicity analysis, III-␤3 and I-␤1a each have one potential transmembrane domain (black boxes). III-␤1a contains both of these potential transmembrane domains. The amino acids that define the 1 and 3 EGF-like domain subtypes are shown below the diagrams of III-␤1a and III-␤3, respectively. Numbers above each diagram indicate the position of the first or last amino acid of various structural regions within the protein. B, membrane orientation and processing proposed for the I-␤1a isoform. The EGF-like domain in the NRG ectodomain is alone sufficient for high potency activation of the cognate receptors, ErbB2, -3, and -4. I-␤1a is synthesized as an N out /C in transmembrane protein. Proteolytic cleavage in the "stalk" region (arrow) produces an NTF containing the bioactive EGF-like domain and a CTF, also referred to as the "a-tail remnant." The NTF is efficiently released into the medium. C, model for III-␤1a membrane orientation and processing derived from I-␤1a model. This model predicts that the CRD is within the ectodomain and that stalk cleavage creates a soluble ectodomain fragment containing both the EGF-like domain and CRD domain. D, constructs used in this study. An HA epitope tag (*) was introduced immediately N-terminal of the EGF-like domain. The FLAG-III-␤1a construct included, in addition, an N-terminal FLAG tag. For one series of experiments, we created constructs encoding NRGs with cytoplasmic tail truncations. Instead of the 374-amino acid a-tail, these forms had tail lengths of 79 amino acids (III-␤1-T79), 45 amino acids (III-␤1-T45), or 13 amino acids (III-␤1-T13 and I-␤1-T13). In this series of experiments, we also tested I-␤1c, an NRG form with a 157-amino acid tail. The location of the epitope for the a-tail antibody (␣-a-tail) used is indicated by a thick underbar. The number of the first and last amino acid in some regions of the proteins is indicated above the diagrams. in the nervous system have a ␤-type EGF-like domain and a-type tail (12,28). The amino acid sequence of rat NRG I-␤1a can be found in Ref. 12, and the amino acid sequence of the rat type III N terminus can be found in Ref. 19.
Based on the results presented in this study, we propose that the III-␤1a proprotein has a cytoplasmic tail at both its N terminus and C terminus. However, to be consistent with the published literature, the term "cytoplasmic tail" used without qualification refers to the cytoplasmic region C-terminal to the EGF-like domain. For example, III-␤1a has the "a-tail" type of cytoplasmic tail.
We refer to neuregulin proteins with both an ectodomain epitope and cytoplasmic tail epitope as "proproteins," in recognition of their potential to be biosynthetic precursors to cleaved bioactive products. However, this does not exclude the possibility that these "proproteins" are themselves bioactive signaling proteins.
Throughout this paper, "neuregulin" and the abbreviation "NRG" refer only to the proteins encoded by the nrg-1 gene. Three related genes (nrg-2, nrg-3, and nrg-4) have now been identified, but the proteins encoded by these genes are not discussed.
Creation of Neuregulin Constructs-All neuregulin proteins analyzed in this study included an HA epitope tag in the ectodomain (Fig. 1D). The sole exception is untagged I-␤1a (labeled "native") used in the experiment illustrated in Fig. 2. Standard recombinant DNA techniques (31,32) were used to create expression vectors for NRG proteins with an HA epitope tag in the ectodomain, just N-terminal of the EGF-like domain. The sequences of I-␤1a and III-␤1a differ at their N-terminal ends but are identical from a serine residue located 14 amino acids N-terminal of the first cysteine residue of the EGF-like domain through their C termini (Fig. 1A). The PCR-based strategy used to insert the HA tag changes the sequence between this serine and cysteine from STSTSTTGTSHLIKC to STSTSTTGTSIDYPYDVPD-YASLHLIKC (underlined residues represent HA epitope tag). All HAtagged constructs used contained this identical sequence within the ectodomain. A similar polymerase chain reaction-based strategy was used to introduce a FLAG epitope tag at the N terminus of an HAtagged III-␤1a construct for examination of the membrane orientation of the III-␤1a N-terminal fragment. In this FLAG-III-␤1a construct, the N-terminal sequence was changed from MEIYSP . . . to MDYKDDD-KEFGGMEIYSP . . . (underlined sequence represents FLAG tag). The fidelity of all sequences amplified by polymerase chain reaction was verified by DNA sequencing. The vector backbone for all constructs was pcDNA 3.1 (Invitrogen). The tail deletion constructs and NRG I-␤1c construct included a C-terminal Myc epitope tag. The Myc tag was not utilized in this study. Details regarding the primers used and the specific procedures employed to create these constructs are available upon request.
Plasmids were prepared for transfection using Qiagen MaxiPrep kits. Transfection of COS-7 cells using DEAE-dextran was performed as described previously (34) using 0.2 g of DNA/well and 10 g of DNA/ 100-mm dish. Transfection of PC 12 cells was performed using Lipo-fectAMINE 2000 (Life Technologies) according to the manufacturer's instructions. PC12 cells were transfected 1 day after plating (ϳ90% confluent); 1 g of DNA was used per well. Schwann cell cultures were transfected when 50 -80% confluent. Transfection of Schwann cells was performed with FuGene reagent (Roche Molecular Biochemicals) according to the manufacturer's directions. For each well, 0.75 l of FuGene was mixed with 25 l of serum-free DMEM and then incubated with 0.5 g of DNA for 15 min at room temperature prior to adding to the media of the cultured cells. COS-7 and PC 12 cells were processed 72 h after transfection. Schwann cells were analyzed 48 h after transfection.
Membrane Association Analysis-COS-7 cells transfected with the III-␤1a construct were washed twice with PBS (0.1 M phosphate buffer, pH 7.4, 150 mM NaCl). One ml of hypotonic lysis buffer (20 mM Hepes, 2 mM MgCl 2 , pH 7.4) was added per 100-mm dish of transfected cells. The cells were collected in this buffer using a rubber policeman and passed through a 21-gauge needle 30 times. Nuclei and unbroken cells were removed by centrifugation at 200 ϫ g for 10 min at 4°C. The supernatant from this low speed spin (Input in Fig. 5A) was centrifuged at 125,000 ϫ g for 1 h at 4°C using a Beckman TLA 100.3 rotor. To analyze the proportion of cellular III-␤1a proprotein and its products bound to membranes, the supernatant (S2) and membrane pellet (P2) from this high speed spin were prepared in SDS sample buffer. For membrane extraction experiments, the membrane pellet was extracted on ice for 30 min in hypotonic lysis buffer with 1 M KCl or with 50 mM Na 2 CO 3 , pH 12. Following extraction, the samples were centrifuged at 125,000 ϫ g for 1 h at 4°C to separate the soluble (S; supernatant) and insoluble (I; pellet) fractions. An equal percentage of the starting material for supernatant and pellet fractions was analyzed by Western blotting.
Triton X-114 partitioning experiments were performed using a modification of Bordier's original protocol (35). The membrane pellet (see above) was resuspended in 960 l of hypotonic lysis buffer and mixed with 240 l of 8% (v/v) Triton X-114 in 10 mM Tris, pH 7.5, 150 mM NaCl (final volume 1.2 ml; final Triton X-114 concentration of 1.6%). This mixture was centrifuged at 125,000 ϫ g for 1 h at 4°C using a Beckman TLA 100.3 rotor to produce a soluble and an insoluble (pellet) fraction. To induce phase separation, 200 l of the soluble fraction was warmed to 28°C for 3 min. The warmed mixture was then layered onto a 6% sucrose cushion (300 l) in a microcentrifuge tube and centrifuged at 200 ϫ g for 5 min. This resulted in a three-phase solution: aqueous, top; sucrose cushion, middle; detergent, bottom. The aqueous phase was mixed with an equal volume of 2ϫ sample buffer. The detergent phase was adjusted to 200 l with lysis buffer and then mixed with an equal volume of 2ϫ sample buffer.
Lysate Preparation and Western Blot Analysis-To prepare lysates for Western blot analysis, cells in 100-mm dishes were rinsed twice with HBS (20 mM Hepes, pH 7.4, 150 mM NaCl, 2 mM CaCl 2 ) and lysed by scraping in 400 l of lysis buffer (1% SDS, 20 mM Tris-HCl, pH 7.5, 5 mM EDTA, 150 mM NaCl). The lysate was passed through a 27-gauge needle four times to shear DNA and cleared of insoluble material by centrifugation at 15,000 ϫ g for 20 min at 4°C. The samples analyzed in Fig. 2 were total cell lysates, not cleared lysates. These samples were prepared by harvesting the cells directly in 400 l of 2ϫ SDS sample buffer/dish.
Western blot analysis was performed as described elsewhere (34). Samples were heated for 5 min at 95°C prior to being loaded on a gel for analysis. The antibody dilutions used were as follows: ␣-HA monoclonal antibody 16B12 (raw ascites fluid; Berkeley Antibody Company, Richmond, CA), 1:3000; ␣-NRG a-tail rabbit polyclonal antibody SC348 (Santa Cruz Biotechnology, Inc., Santa Cruz, CA), 0.1 g/ml; horseradish peroxidase-conjugated goat anti-rabbit (Pierce), 1:50,000; horseradish peroxidase-conjugated donkey anti-mouse (Jackson ImmunoResearch Laboratories), 1:50,000. Blots were developed using Renaissance chemiluminescence substrate (PerkinElmer Life Sciences). Prestained standards used on Western blots (BenchMark protein ladder; Life Technologies) were calibrated against unstained molecular weight markers (Bio-Rad). A standard curve (distance migrated versus log M r ) was created based on the migration of the marker proteins, and molecular weights were assigned to each band of interest according to the position of the center of the band.
Experiments illustrated in the figures were repeated three times (Figs. 2-4, 6, and 7) or twice (Figs. 5 and 8). Each repetition gave results similar to those illustrated. For quantitation of cell surface or released NRG, samples to be compared were analyzed on a single blot along with multiple dilutions of a standard sample. Films were scanned, and band "volume" (summed pixel density) was measured using the NIH Image software. A standard curve was created by plotting the pixel density for each dilution of the standard sample as a function of the relative amount loaded. The relative amount of NRG in each test sample was calculated based on this standard curve.
Cell Surface Biotinylation-For each 100-mm dish of transfected COS-7 cells, the cells were washed twice with 5 ml of ice-cold HBS and then incubated in 3 ml of HBS containing 0.5 mg/ml sulfo-NHS-LCbiotin (Pierce) for 45 min at 4°C. Following removal of the sulfo-NHS-LC-biotin solution, residual sulfo-NHS-LC-biotin was quenched by incubating with 10 mM glycine in HBS for 5 min. Cleared lysate (prepared as described above) was mixed with an equal volume of TENT buffer (20 mM Tris HCl, pH 7.4, 150 mM NaCl, 5 mM EDTA, 1% Triton X-100) and incubated with 50 l of streptavidin-conjugated agarose beads (Pierce) for 1 h at 4°C with end-over-end mixing. The agarose beads were washed three times with lysis buffer and eluted in 2ϫ SDS sample buffer. A gel sample was also prepared from the cleared lysate to assess the total amount of NRG proteins (cell surface plus internal) expressed in each dish.
Analysis of Conditioned Medium and Phorbol 12-Myristate 13-Acetate (PMA) Stimulation-48 h after transfection, COS-7 cells were washed once with PBS, and the standard growth medium was replaced with Opti-MEM I (Life Technologies). 12 h later, this medium was replaced with 5 ml of fresh Opti-MEM I per 100-mm dish with or without 100 nM PMA, and the cells were returned to the incubator. Following a 30-min (Fig. 6) or 2-h (Fig. 8) incubation at 37°C, the conditioned medium was collected and filtered through a 0.22-m filter (Millipore Corp.) to remove floating cells. The medium was then concentrated 40 fold by ultrafiltration using Centricon-10 units (Millipore). To assess the effect of PMA on cell surface NRG, the same treatment protocol was followed. At the end of the 30-min incubation with PMA, the cells were rinsed twice with ice-cold HBS, and biotinylation with sulfo-NHS-LC-biotin was performed as described above.
Immunocytochemistry-To assess cell surface NRG, cells were washed twice with ice-cold serum-free DMEM and then incubated in primary antibody solutions for 30 min on ice, washed again with ice-cold serum-free DMEM, and incubated in secondary antibody solution for 30 min on ice. Cells were then washed and fixed with 2% paraformaldehyde containing 0.1% Triton X-100 in PBS for 20 min on ice. Fixed cells were blocked in 10% normal donkey serum diluted in PBS (blocking buffer) for 30 min at room temperature. For labeling of intracellular epitopes, these fixed and permeabilized cells were incubated in primary antibody solution for 1 h at room temperature and then washed in PBS (3 times for 5 min each). Cells were then incubated in the appropriate secondary antibody solution for 1 h at room temperature. Cells were washed with PBS and mounted with Vectashield Mounting Medium (Vector Laboratories, Burlingame, CA). As a negative control to show that cell surface label was not due to antibody internalization, cells were incubated in the ␣-a-tail and secondary antibodies prior to permeabilization, as described above. When ␣-a-tail was applied prior to permeabilization, no labeling of cells was detected (not shown). Primary and secondary antibodies for cell surface label were diluted in serumfree DMEM. Primary and secondary antibodies for intracellular staining were diluted in blocking buffer. Antibody dilutions were as follows: ␣-HA monoclonal 16B12, 1:500; ␣-FLAG monoclonal M2, 17 g/ml; ␣-NRG a-tail rabbit polyclonal antibody SC348, 0.5 g/ml; fluorescein isothiocyanate-conjugated donkey anti-mouse, 1:100; lissamine-rhodamine sulfonyl chloride-conjugated donkey anti-rabbit, 1:1000. Secondary antibodies were purchased from Jackson ImmunoResearch Laboratories, Inc.
Images were collected using a Bio-Rad 1024 laser-scanning confocal system (Bio-Rad) coupled to a Zeiss Axioskop microscope. Images shown in the figures were obtained using a 20ϫ Plan-Neofluar (NA 0.5) or 40ϫ Plan-Neofluar (NA 0.75) lens. A Z-series of images was collected for each preparation at a step of 0.5 m. The images shown were compiled from a stack of five individual Z-series images using the Bio-Rad Lasersharp software. Images were further processed using Adobe Photoshop. For each cell type, all images of the same fluorophore were collected and processed identically.

RESULTS
To compare the processing of NRG III-␤1a with that of I-␤1a, we constructed cDNAs encoding NRG isoforms with an HA epitope tag inserted immediately N-terminal of the EGF-like domain (Fig. 1D). The use of a shared tag allowed us to directly infer the relative amount of III-␤1a and I-␤1a ectodomain from signal strength in immunoassays. The position chosen for the HA tag corresponded to the position used in a previous study of transforming growth factor-alpha (TGF␣) processing (36). By using antibodies to the HA tag and to an epitope located in the cytoplasmic a-tail, we were able to detect both the N-terminal fragment (NTF) and the C-terminal fragment (CTF) of NRG proproteins proteolytically cleaved in the stalk region. We have also tested a similarly constructed HA-tagged III-␤3. Our results with the HA-tagged I-␤1a and III-␤3 proteins are consistent with previous studies of native I-␤1a and III-␤3 isoforms and studies of these isoforms with epitope tags located at the extreme N or C terminus (12,22,24,25,37). Therefore, the processing of the HA-tagged NRG proteins does not seem to differ substantially from processing of the native forms of these proteins.
The III-␤1a Proprotein Is Proteolytically Processed to a 76-kDa N-terminal and a 60-kDa C-terminal Fragment-To determine the size of full-length III-␤1a and of fragments that might be derived from this, we transfected COS-7 cells with a III-␤1a expression vector and analyzed cell lysates by Western blot. Bands of 140 and 110 kDa were labeled by both the ␣-HA and ␣-a-tail antibodies ( Fig. 2A, lanes 2 and 2Ј; Fig. 2B, lane 2). Since the 140-and 110-kDa proteins have both an ectodomain (HA) and a-tail epitope, these are III-␤1a proprotein forms. The predicted size of the peptide encoded by the III-␤1a cDNA is 78 kDa (including 1.5 kDa of mass contributed by the HA tag), which is considerably smaller than the observed size of the III-␤1a proproteins. The larger than predicted mass of these proprotein forms might be due to posttranslational modifications such as glycosylation.
We also observed a band of 76 kDa that labeled only with ␣-HA and a broad 60-kDa band that labeled only with the cytoplasmic domain antibody. These results suggest that the 76-kDa protein is an NTF and the 60-kDa protein is a CTF derived from the III-␤1a proprotein by proteolytic cleavage. Consistent with this interpretation, the sum of these fragment sizes is 136 kDa, approximately equal to the size of the 140-kDa protein labeled by both antibodies. The III-␤1a CTF is similar in size to the I-␤1a CTF (Fig. 2B). Since the cleavage site of I-␤1a has been shown to be in the ectodomain stalk region (23,25), the similar size of the III-␤1a and I-␤1a CTFs indicates that III-␤1a is also processed by cleavage in the stalk region. The 76-kDa NTF band was much more intense than the 140and 110-kDa proprotein bands, indicating that the NTF, not the proprotein, is the predominant ectodomain containing NRG form in cells expressing III-␤1a. It is noteworthy that proteins similar in size to the III-␤1a NTF (76 kDa) and larger proprotein form (140 kDa) have been detected in Western blot analyses of brain and spinal cord samples developed with pan-NRG antibodies (20,38,39).
Consistent with previously published results (24), Western analysis of lysates from cells expressing I-␤1a revealed prominent bands of 110 and 95 kDa representing the I-␤1a proprotein and a broad band of 45-kDa protein representing the I-␤1a NTF (Fig. 2, A and B, lanes 4 and 5). We also routinely observed a series of fainter bands extending from ϳ43 to 30 kDa on blots developed with ␣-HA but not on blots developed with ␣-a-tail. These bands may represent proteolytic degradation products of the I-␤1a proprotein or NTF.
Western analysis of lysates from cells expressing III-␤3 (SMDF) was also consistent with published results (37). In these lysates, a band of 83 kDa, representing full-length III-␤3, was labeled by the ␣-HA antibody (Fig. 2A, lane 3). The sequence of the III-␤1a NTF, which is created by cleavage of the III-␤1a proprotein in the stalk region (see above), should be identical to the sequence of III-␤3 except in the region Cterminal of the EGF-like domain (Fig. 1A). The slightly larger size of the full-length III-␤3 relative to the III-␤1a NTF (83 versus 76 kDa) might be due to differences in glycosylation. Consistent with this possibility, the ectodomain sequence unique to the ␤3 type of EGF-like domain is rich in serine.
In summary, Western blot analysis indicates that the III-␤1a proprotein is proteolytically processed in the stalk region and that the predominant cellular form that includes the ectodomain differs between III-␤1aand I-␤1a-expressing cells. In III-␤1a-expressing cells, the major form that includes the ectodomain is the NTF resulting from stalk cleavage, but in I-␤1a-expressing cells, it is the proprotein. The bands seen in lysates of III-␤1a-, I-␤1a-, and III-␤3-expressing cells are listed in Table I. Table I also summarizes information from experiments described below that define NRG protein fragments at the cell surface and released into medium.
The NTF Produced by Cleavage of the III-␤1a Proprotein Accumulates at the Cell Surface-To determine whether the N-terminal region of NRG isoforms affects the amount of NRG ectodomain exposed at the cell surface, we first compared the cellular distribution of III-␤1a and I-␤1a using immunocytochemical methods. Since neurons (10,(17)(18)(19)30) and Schwann cells (20,21) are known to express NRGs, we examined NRG distribution in NGF-differentiated PC12 cells (cells that have neuronal characteristics) and primary cultures of Schwann cells as well as in COS-7 cells. Transfected PC12 cells (Fig. 3, A  and B), Schwann cells (Fig. 3, C and D), and COS-7 cells (Fig.  3, E and F) gave similar results. Nonpermeabilized cells (Ϫ) expressing III-␤1a were strongly labeled with ␣-HA, whereas no specific surface labeling of I-␤1a-expressing cells was detected. Since NRG activity has been demonstrated in axonal membranes (40,41), it is noteworthy that the surface of neurites as well as the cell body was strongly labeled in PC12 cells expressing III-␤1a (Fig. 3A). Permeabilized cells (ϩ) expressing III-␤1a and I-␤1a were labeled by ␣-a-tail (Fig. 3, A-F) or ␣-HA (not shown) with similar intensity, suggesting that the total amount of NRG is roughly similar in cells expressing III-␤1a and I-␤1a. The larger cell surface accumulation of III-␤1a than I-␤1a for all three cell types examined suggests that cell biological mechanisms common to many cell types govern these differences.
Cell surface biotinylation (Fig. 4) was used to determine more quantitatively the relative amounts of the NRG proproteins and their products exposed at the cell surface. In III-␤1aexpressing cells, most of the NRG that was biotinylated was the 76-kDa NTF (Fig. 4A). In long exposures of blots developed with the ␣-a-tail antibody, faint bands representing cell surface 140-kDa proprotein and 60-kDa CTF were also detected (not shown), but the 110-kDa proprotein band was not observed.
Although for I-␤1a-expressing cells, cell surface NRG was not observed by immunocytochemistry (Fig. 3), it was detected in the biotinylation experiments (Fig. 4A). This is consistent with the results of others, who have detected low levels of type I NRGs at the cell surface by immunohistochemistry or cell surface biotinylation (22,24). The most abundant NRG forms at the surface of the I-␤1a-expressing cells were the 45-kDa NTF and the 110-kDa proprotein (Fig. 4A). In the experiment shown in Fig. 4, the predominate form was the 45-kDa form, but in other experiments, the amount of 110-kDa proprotein at the surface exceeded the amount of the 45-kDa form (see, for example, Fig. 7). The I-␤1a NTF might accumulate at the cell surface due to interaction between the Ig-like domain of the type I NRG ectodomain and cell surface heparin sulfate proteoglycans (42). We did not detect the 95-kDa I-␤1a proprotein at the cell surface.
The relative amounts of NRG ectodomain exposed at the cell surface of III-␤1aand I-␤1a-expressing cells were estimated by quantitating the biotinylated protein bands (Fig. 4C). Assuming the two proteins were equally biotinylated, there was FIG. 2. The III-␤1a proprotein is proteolytically processed to produce a 76-kDa NTF and a 60-kDa CTF. COS-7 cells were transiently transfected with expression vectors for indicated NRG isoforms. Lysates prepared from these cells were analyzed by Western blot. All NRG proteins included an HA-epitope tag in the ectodomain except for the native I-␤1a (lane 5). The blot was developed with ␣-HA (A) and then stripped and reprobed with ␣-a-tail (B). Lane 2Ј is a longer exposure of A, lane 2. It is included to demonstrate the 140-and 110-kDa III-␤1a bands. Bands labeled by each antibody are summarized in Table  I. The NRG proproteins and some of their products are represented by multiple bands or very broad bands. At least in part, this is due to heterogeneous glycosylation (14,22). The band pattern is similar for HA-tagged and native I-␤1a (B, lanes 4 and 5), which supports the working assumption of this study that the HA epitope tag does not substantially alter the processing of the NRG isoforms examined. 1% of the total cell lysate prepared from a 100-mm dish of transfected or mock-transfected COS-7 cells was loaded in each lane. ϳ25 times more III-␤1a ectodomain at the cell surface than I-␤1a ectodomain. Furthermore, not only was there much more NRG ectodomain at the surface of III-␤1a-expressing cells, but also a larger proportion of the cellular NRG was at the surface, since III-␤1a lysates contained less NRG than the I-␤1a lysates (Fig. 4B). Results of experiments in which cell surface proteins were digested with trypsin are consistent with this interpretation. 2 In these experiments, a large percentage of the cellular III-␤1a was degraded, but only a tiny percentage of the total cellular I-␤1a 110-and 45-kDa forms were degraded In summary, both immunocytochemical and biochemical experiments indicate that cells expressing III-␤1a have much more NRG ectodomain exposed at their cell surface than cells expressing I-␤1a, and this difference appears to be largely due 2 J. Wang and D. Falls, unpublished data. F) were transfected with the III-␤1a or I-␤1a expression vector, as indicated. The cells were incubated with ␣-HA prior to permeabilization (Ϫ) to reveal NRG ectodomain exposed at the cell surface and with ␣-a-tail following permeabilization (ϩ). The surface of cells expressing III-␤1a was brightly labeled by ␣-HA (rows A, C, and E), but no specific cell surface label of I-␤1a-expressing cells was detected (rows B, D, and F). The intensity of label with ␣-a-tail was similar in the III-␤1a-and I-␤1a-expressing cells.  With heavier loadings and long exposures, a weak 140-kDa band was also detected, and this band was also detected with ␣-a-tail (not shown). 110-and 45-kDa bands are present in the biotinylated I-␤1a sample (lane 4). These I-␤1a bands are much less intense than the III-␤1a NTF band. Control samples (lanes 3 and 5) were prepared identically except that the biotinylation reagent was omitted (Ϫ). The absence of signal in these no biotin controls demonstrates that NRG precipitation by streptavidin was dependent on protein biotinylation. An identically prepared sample of biotinylated proteins from mock-transfected cells was loaded in lane 1. Each lane was loaded with 10% of the fraction material derived from a 100-mm dish of transfected COS-7 cells. B, Western blot analysis with ␣-HA of lysates from which biotinylated proteins (A) were isolated. The band intensities are similar for the lysates prepared from dishes subjected to biotinylation (ϩ) and to mock biotinylation (Ϫ). Each lane was loaded with 1% of the total cell lysate prepared from each dish. C, relative amounts of biotinylated NRG proteins. Bands were quantitated by densitometry as described under "Experimental Procedures." The amount of the III-␤1a NTF (76 kDa) was assigned a value of 1. Mean Ϯ S.E. (n ϭ 3) is shown.

FIG. 3. The NRG ectodomain accumulates at the cell surface of III-␤1a-expressing PC12 cells, Schwann cells, and COS-7 cells. PC12 cells (A and B), Schwann cells (C and D), and COS-7 cells (E and
to the accumulation of the NTF of III-␤1a in the plasma membrane. The N-terminal Product Produced by Cleavage of the III-␤1a Proprotein Is a Transmembrane Protein-The accumulation of the III-␤1a NTF at the cell surface (Figs. 3 and 4) and the presence of a potential transmembrane domain within the CRD (Fig. 1A) suggested the possibility that the III-␤1a NTF is a transmembrane protein. To test this hypothesis, we first assessed the proportion of cellular III-␤1a associated with membranes. After high speed centrifugation, none of the NTF in a cleared cell lysate was found in the supernatant, the fraction that contains the soluble, cytoplasmic proteins (Fig. 5A). Next we determined whether the NTF behaves as an integral or a peripheral membrane protein. We incubated a membrane fraction prepared from cells expressing III-␤1a in 1 M potassium chloride or sodium bicarbonate, pH 12, and then separated the membrane-associated proteins from soluble proteins by centrifugation. With both of these treatments, the III-␤1a NTF remained associated with the membranes (Fig. 5B), the behavior expected for an integral membrane protein. The NTF also largely partitioned into the detergent phase in Triton X-114 phase partitioning experiments (Fig. 5B), further evidence that the NTF is an integral membrane protein. Finally, we examined whether the NTF has intracellular as well as extracellular epitopes. When a III-␤1a protein with an N-terminal FLAG epitope tag was expressed in COS-7 cells, the FLAG tag could only be detected in permeabilized cells (Fig. 5C), indicating that this epitope is cytoplasmic, whereas an HA epitope tag located just N-terminal of the EGF-like domain is exposed on the cell surface (Fig. 3). Together, these data indicate that the III-␤1a NTF is an N in /C out transmembrane protein.
The III-␤1a Ectodomain Is Inefficiently Shed into the Medium Compared with the I-␤1a Ectodomain-Previous studies have demonstrated that the ectodomain of I-␤1a is efficiently released from cells (24) and that this release is accelerated by FIG. 6. The concentration of NRG in medium conditioned by COS-7 cells expressing III-␤1a is low relative to the level in medium conditioned by COS-7 cells expressing I-␤1a. A, Western analysis of medium conditioned for 30 min by COS-7 cells expressing III-␤1a or I-␤1a and either untreated (Ϫ) or treated (ϩ) with 100 nM PMA for 30 min. Two separate exposures of the same blot are shown, because the amount of NRG in medium conditioned by III-␤1a-expressing cells is much less than in medium conditioned by I-␤1a-expressing cells. Treatment with PMA dramatically increased the amount of NRG released by I-␤1a-expressing cells but had little effect on release by III-␤1a-expressing cells. 15% of the concentrated medium conditioned by a 100-mm dish of transfected COS-7 cells was loaded in each lane. B, Western analysis of the lysates from the cultures analyzed in A. The results show that the amount of total cell-associated NRG is similar for untreated (Ϫ) and PMA-treated (ϩ) cultures. 1% of the total cell lysate derived from a 100-mm dish of transfected COS-7 cells was loaded in each lane. C, relative amount of NRG in conditioned medium. Bands were quantitated by densitometry as described under "Experimental Procedures." The amount of HA immunoreactivity in the medium conditioned by PMA-treated cells expressing I-␤1a was assigned a value of 1. Mean Ϯ S.E. (n ϭ 3) is shown.

FIG. 5. The 76-kDa N-terminal fragment of III-␤1a is a transmembrane protein.
A, a cleared lysate ("Input") prepared from III-␤1a-expressing COS-7 cells was separated into soluble (S2) and membrane pellet (P2) fractions. The blot was developed with ␣-HA. 1% of the fraction material derived from a 100-mm dish of transfected COS-7 cells was loaded in each lane. The 76-kDa NTF was detected only in the membrane fraction. B, membranes prepared from COS-7 cells expressing III-␤1a were incubated with 1 M potassium chloride or 50 mM sodium bicarbonate, pH 12. The samples were separated into soluble (S, supernatant) or insoluble fractions (I, pellet). Proteins in the membrane preparation were also fractionated in a Triton X-114 partitioning experiment as described under "Experimental Procedures." Detergentsoluble (S, supernatant) and -insoluble fractions (I, pellet) were prepared. The Triton X-114-soluble material (S) was partitioned into detergent (D) and aqueous (AQ) phases. 0.5% of the fraction material derived from a 100-mm dish of transfected COS-7 cells was loaded in each lane. Only the 76-kDa NTF band can be seen with this exposure. C, COS-7 cells were transfected with an expression vector for FLAG-III-␤1a, which has a FLAG epitope tag at the N terminus. The cells were incubated with ␣-FLAG antibody either before (Ϫ) or after (ϩ) permeabilization. Cells were also stained with ␣-a-tail antibody after permeabilization. Labeling with ␣-FLAG was detected only after permeabilization, indicating that the III-␤1a N terminus is cytoplasmic. Scale bar, 50 m. treatment with phorbol esters (22,24,25), compounds that activate protein kinase C. To compare the release of the III-␤1a and I-␤1a ectodomain, medium conditioned by I-␤1a-and III-␤1a-expressing cells was assayed for NRG protein by Western blot. In medium conditioned by I-␤1a-expressing cells, an HAimmunoreactive peptide of 45-kDa was observed (Fig. 6A, lane  1). In medium conditioned by III-␤1a-expressing cells, a faint 63-kDa band was detected (Fig. 6A, lane 3). A soluble NRG ectodomain fragment of this size could be produced by proteolytic cleavage of the 76-kDa III-␤1a NTF between the HA tag and the TM domain within the CRD (see Fig. 1). Strikingly, the concentration of NRG in the medium conditioned by III-␤1aexpressing cells for 30 min was only ϳ10% of the NRG concentration in medium conditioned by I-␤1a-expressing cells (Fig.  6C). Thus, while there is much more III-␤1a ectodomain at the cell surface (Figs. 3 and 4), there is much less III-␤1a ectodomain in conditioned medium.
The effects of PMA treatment on NRG release (Fig. 6) and cell surface NRG (Fig. 7) were very different for cells expressing III-␤1a and I-␤1a. Treatment of I-␤1a-expressing cells with PMA resulted in a dramatic increase in the amount of NRG in conditioned medium (Fig. 6A, lane 1 versus lane 2; Fig. 6C) and a dramatic decrease in the amount of NRG at the cell surface (Fig. 7, lane 6 versus lane 5), but similar treatment of III-␤1aexpressing cells with PMA had little effect on either the release of NRG into the medium (Fig. 6A, lane 3 versus lane 4; Fig. 6C) or the amount of cell surface NRG (Fig. 7, lane 3 versus lane 2). These results demonstrate that the mechanisms regulating NRG release and cell surface accumulation are different for III-␤1a and I-␤1a.
Truncation of the Cytoplasmic Tail Has Little Effect on the Cell Surface Accumulation and Proteolytic Processing of III-␤1 NRGs-For type I TM-NRGs, truncation of the cytoplasmic tail to less than ϳ90 amino acids has been shown to block release of the ectodomain into the medium (7,26). Since III-␤1a and I-␤1a have identical sequence in the EGF-like, stalk, TM, and cytoplasmic tail regions (Fig. 1A), we investigated whether  shortening of the cytoplasmic tail would similarly affect III-␤1a and I-␤1a cell surface accumulation and release.
The amount of NRG at the cell surface and released into the medium was similar for COS-7 cells expressing III-␤1a (374amino acid tail) and III-␤1 proteins with a truncated tail (Fig.  8A, lanes 3-6, and C, lanes 2-5). However, the amount of cell surface and released NRG was markedly less in cultures of cells expressing type I NRG with a truncated tail (I-␤1-T13) than cultures of cells expressing longer, naturally occurring tail forms (I-␤1a and I-␤1c) (Fig. 8A, lanes 8 -10, and C, lanes 6 -8). Thus, while the cytoplasmic tail is essential for the cell surface localization and release of NRGs with a type I N terminus, it does not serve a similar function for NRGs with a type III N terminus. DISCUSSION Based on the identity of the III-␤1a and I-␤1a sequence in the regions known to be important for processing of I-␤1a, it has been assumed that the topology and processing of III-␤1a would be similar to I-␤1a (see Fig. 1, B and C). Our results show that this is not the case. Cultures expressing III-␤1a had ϳ25 times more NRG at the cell surface and ϳ10 times less NRG in the medium than parallel cultures expressing I-␤1a. Truncation of the cytoplasmic tail had little effect on the accumulation of NRG at the surface of III-␤1a-expressing cells, and treatment with phorbol esters had little effect on the release of soluble NRG from these cells.
Taken together, our data support the alternative model for the membrane topology and processing of III-␤1a illustrated in Fig. 9. Proteolytic cleavage in the stalk region (Fig. 9, site 1) creates an NTF that is an N in /C out transmembrane protein.
Since in immunohistochemical experiments epitopes located both at the N and C terminus of the III-␤1a proprotein are only detected following cell permeabilization (Figs. 3 and 5), we propose that the proprotein passes through the membrane twice.
In cells expressing III-␤1a, the amount of 76-kDa N-terminal fragment is large relative to the amount of proprotein (Fig. 2), indicating that most of the NRG in these cells has been cleaved at site 1 but not site 2. Cleavage of III-␤1a at both site 1 and site 2 is required for release of the 63-kDa ectodomain fragment into culture medium. Since the only cell-associated III-␤1a cleavage product observed was the 76-kDa NTF, the model depicts cleavage at site 1 as preceding cleavage at site 2. A candidate site for cleavage 2 is the peptide bond between Gly 117 and Leu 118 , since when a NRG III-␤3-immunoglobulin chimera was expressed in 293 cells, the released ϳ63-kDa protein had the corresponding leucine as its N terminus (43).
It is noteworthy that, in contrast to the model shown in Fig.  1C, the model suggested by our data (Fig. 9) predicts that most of the sequence of the CRD is within the membrane or cytoplasm and is not a component of the released ectodomain. We expect the model of Fig. 9 to hold also for the processing of III-␤2a, the other major type III NRG known to be expressed in the nervous system, since this isoform is identical to III-␤1a except that it lacks the eight ␤1-specific amino acids at the C-terminal end of the EGF-like domain.
The amino acid sequence of the III-␤1a NTF is identical to the sequence of the full-length III-␤3 except very near the carboxyl terminus (Fig. 1). A previous study of III-␤3 (37) and our own data 2 demonstrate that this isoform has many characteristics similar to those that we have observed for the III-␤1a NTF. When expressed in fibroblastic cells, full-length III-␤3 is an N in /C out transmembrane protein that accumulates at the cell surface. Cleavage of III-␤3 at a site similarly located to site 2 ( Fig. 9) results in the release of a III-␤3 ectodomain fragment, but the levels of this fragment in medium are low compared with the levels of released I-␤1a ectodomain fragment in medium from parallel cultures. The similarity in behavior of III-␤3 and the III-␤1a NTF suggests that if the biological activities of III-␤3 differ from the activities of III-␤1a, these differences will be attributable to activities of the intact III-␤1a proprotein or C-terminal fragment.
Limitations imposed by the available NRG antibodies have prevented detailed study of NRG processing in cells natively FIG. 9. Proposed model for membrane topology and processing of the III-␤1a isoform. Our data indicate that the III-␤1a proprotein passes through the membrane twice. Proteolytic cleavage in the stalk region (1) produces a transmembrane NTF containing the EGF-like domain. Cleavage of the III-␤1a NTF near the membrane (2) could release a fragment containing the EGF-like domain into the medium. The data presented here suggest that cleavage 2 is rate-limiting for release of soluble III-␤1a ectodomain and that this cleavage is not accelerated by treatment with phorbol esters. expressing NRG proteins. Thus, previous studies of NRG processing have employed COS-7, CHO, or 293 cells transfected with NRG expression constructs (22,24,25,37). Similarly, in this study we compared the processing of NRG III-␤1a with I-␤1a in COS-7 cells transfected with NRG expression vectors. While such expression studies have the potential disadvantage that some aspects of NRG processing might differ in natively expressing cells, the use of transfected fibroblastic cells for biochemical studies has proven to be a very fruitful approach in examining the biosynthetic processing of TGF␣ and other transmembrane ligands (44). Further, cell surface NRG was greater for III-␤1a-than I-␤1a-expressing cells, regardless of whether these NRGs were expressed in COS-7 cells, PC12 cells, or Schwann cells (Fig. 3). That this central result of our study holds true for each cell type tested suggests that the processing of III-␤1a is similar in COS-7 cells, neuronal cells, and Schwann cells.
Several characteristics of transmembrane type I NRG topology and processing fit a model previously established for TGF␣ and other EGF receptor ligands that are synthesized as transmembrane proproteins (44,45). For both TGF␣ and NRG I-␤1a, stalk cleavage results in release (or "shedding") of a soluble ectodomain product, and this release is activated by protein kinase C (Figs. 6 and 7; Refs. 22,24,25,46,47). Also for both, transport of the protein to the cell surface and subsequent ectodomain release are regulated by the cytoplasmic tail (Fig. 8;Refs. 7,26,36,48,and 49). Since the sequence of the NRG III-␤1a and I-␤1a EGF-like domain, stalk, transmembrane domain, and cytoplasmic tail are identical (Fig. 1), it might have been expected that III-␤1a would fit the same pattern, but it does not. Activation of protein kinase C by treatment with phorbol esters caused little change in the amount of III-␤1a at the cell surface or in the medium. This indicates that cleavage at site 2 ( Fig. 9), unlike stalk cleavage (site 1), is insensitive to protein kinase C. Therefore, release of the III-␤1a ectodomain is not governed by the same proteolytic machinery as release of I-␤1a, TGF␣ and a number of other transmembrane proteins. The failure of tail truncation to substantially reduce cell surface III-␤1 indicates that the mechanisms responsible for trafficking III-␤1a and I-␤1a to the cell surface also differ.
One known biochemical difference between type I and type III NRGs that is likely to have significant cell biological consequences is that type I NRGs bind strongly to heparin sulfate proteoglycans, whereas type III NRGs do not (42,50). Consistent with this, evidence indicates that the released ectodomain of type I TM-NRGs is deposited in the basal lamina of neuromuscular synapses and the extracellular matrix of brain (42,50,51). A second reported difference between these classes of NRGs is that soluble, recombinant type III NRG is more potent than type I NRG in inducing nicotinic AChR subunit gene expression in cultured sympathetic neurons (18). The results presented here add new evidence that the cell biological properties of type III NRGs differ substantially from the properties of type I NRGs.
The differences between cells expressing III-␤1a and I-␤1a with respect to cell surface and released NRG suggest the hypothesis that III-␤1a is preferentially employed in vivo for cell-cell communication requiring a juxtacrine (direct contact) mode of interaction, whereas I-␤1a is preferentially employed for paracrine signaling. In fact, the small amount of III-␤1a ectodomain in culture medium (Fig. 6) raises the question of whether release of the type III ectodomain has any physiological role. To date, two cell-cell interactions mediated by membrane-associated NRG activity have been identified. One is the induction by differentiating neuroblasts of glial commitment in neural crest stem cells (52). The second is the induction of Schwann cell proliferation by the axonal membranes of sensory neurons (40,(53)(54)(55). The N-terminal fragment of III-␤1a (or the closely related III-␤2a) is a strong candidate for being the NRG form mediating these interactions, since both neural crestderived neuroblasts and sensory neurons express type III NRGs (19,52) and since Schwann cells along peripheral nerves are markedly depleted in the type III NRG knockout (8). Future studies must elucidate whether type III NRGs are active only in direct contact cell-cell interactions in vivo and determine the novel mechanisms regulating type III NRG biogenesis.