Pancreatic @ Cells Express Two Autoantigenic Forms of Glutamic Acid Decarboxylase, a 65-kDa Hydrophilic Form and a 64-kDa Amphiphilic Form Which Can Be Both Membrane-bound and Soluble*

The 64-kDa pancreatic &cell autoantigen, which is a target of autoantibodies associated with early as well as progressive stages of ,&cell destruction, resulting in insulin-dependent diabetes (IDDM) in humans, has been identified as the y-aminobutyric acid-synthesiz-ing enzyme glutamic acid decarboxylase. We have identified two autoantigenic forms of this protein in rat pancreatic @-cells, a M, 66,000 (GADBs) hydrophilic and soluble form of PI 6.9-7.1 and a M. 64,000 (GADe4) component of PI 6.7. GADs4 is more abundant than GADsa and has three distinct forms with regard to cellular compartment and hydrophobicity. A major portion of GADe4 is hydrophobic and firmly mem-brane-anchored and can only be released from membrane fractions by detergent. A second portion is hydrophobic but soluble or of a low membrane avidity, and a third minor portion is soluble and hydrophilic. All the GADs4 forms have identical PI and mobility on sodium dodecyl sulfate-polyacrylamide

a target of autoantibodies associated with early as well as progressive stages of ,&cell destruction, resulting in insulin-dependent diabetes (IDDM) in humans, has been identified as the y-aminobutyric acid-synthesizing enzyme glutamic acid decarboxylase. We have identified two autoantigenic forms of this protein in rat pancreatic @-cells, a M, 66,000 (GADBs) hydrophilic and soluble form of PI 6.9-7.1 and a M. 64,000 (GADe4) component of PI 6.7. GADs4 is more abundant than GADsa and has three distinct forms with regard to cellular compartment and hydrophobicity. A major portion of GADe4 is hydrophobic and firmly membrane-anchored and can only be released from membrane fractions by detergent. A second portion is hydrophobic but soluble or of a low membrane avidity, and a third minor portion is soluble and hydrophilic. All the GADs4 forms have identical PI and mobility on sodium dodecyl sulfate-polyacrylamide gel electrophoresis. Results of pulse-chase labeling with [36S]methionine are consistent with GADe4 being synthesized as a soluble protein that is processed into a firmly membrane-anchored form in a process which involves increases in hydrophobicity but no detectable changes in size or charge. All the GAD64 forms can be resolved into two isoforms, a and B, which differ by approximately 1 kDa in mobility on sodium dodecyl sulfatepolyacrylamide gel electrophoresis but are identical with regard to all other parameters analyzed in this study. GAD65 has a shorter half-life than the GADe4 forms, remains hydrophilic and soluble, and does not resolve into isomers. Comparative analysis of the brain and @-cell forms of GAD show that GADeS and GADe4 in pancreatic &cells correspond to the larger and smaller forms of GAD in brain, respectively. The expression of different forms and the flexibility in subcellular localization of the GAD autoantigen in Bcells may have implications for both its function and autoantigenicity.
* This study was supported by National Institutes of Health Grant 1POlDK418 and by grants from Novo-Nordisk, the Juvenile Diabetes Foundation International, March of Dimes, and by the Foundation Nordisk Insulinlaboratorium, in its initial phases. The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked "advertisethis fact. )I To whom correspondance should be addressed.
Insulin dependent diabetes mellitus (IDDM)' is characterized by a selective loss of the insulin producing P-cells in a process which can span several years, and is characterized by clear indications of an autoimmune response, that includes circulating islet cell antibodies (Castano and Eisenbarth, 1990). We have used the circulating antibodies present in IDDM sera to identify a M, 64,000 (64 kD) target human islet cell autoantigen by immunoprecipitation of [35S]methioninelabeled human islets (Baekkeskov et al., 1982). Antibodies to the human islet cell 64-kD protein and its rat islet counterpart are present in approximately 80% of newly diagnosed IDDM patients and have been detected up to several years before the clinical onset of the disease, concomitantly with a decrease of P-cell function (Sigurdsson and Baekkeskov, 1990 for review). The 64-kD protein was recently identified as the GABAsynthesizing enzyme glutamic acid decarboxylase (GAD) in pancreatic P-cells . The presence of GAD and its product GABA in islet P-cells (Garry et al., 1986) and the presence of GABAA receptors on islet a-and &cells (Rorsman et al., 1989;Reusens-Billen et al., 1984) suggests a role of GABA in paracrine signalling between the P-cell and the other endocrine islet cells. GAD is expressed in high concentrations in GABA-ergic neurons in the central nervous system (Mugnaini and Oertel, 1985) and in the oviduct (Erdo et al., 1989). Two major forms of GAD have been detected in the brain and their molecular masses have been described as 59-67 kDa Legay et al., 1987;Chang and Gottlieb, 1988). The larger brain form has been cloned and sequenced from cat (Kaufman et al., 1986;Kobayashi et al., 1987), rat (Julien et al., 1990Wyborski et al., 1990), and mouse (Katarova et al., 1990). The smaller GAD form in rat brain has recently been cloned, sequenced, and shown to be a product of a different gene (Erlander et al., 1991).

21258
Autoantigenic Forms of GAD in Pancreatic /Wells analyzed autoantigenic forms of GAD in both soluble and membrane compartments of rat islets and characterized those with regard to size, charge, hydrophobicity, and half-life. We show that islets of Langerhans express a less abundant 65-kDa soluble hydrophilic form of GAD (GAD,,) in addition to GADM a and P, and provide evidence that only GADs4 becomes membrane bound in a process which involves modification with small noncharged hydrophobic residue(s). We also show that GAD,, and GAD64 are homologous to the larger and smaller brain forms of GAD, respectively.

MATERIALS AND METHODS
Isolation of Islets and Biosynthetic Labeling-Isolation of rat islets, maintenance in culture and radioactive labeling with [35S]methionine for 4 h was carried out as described (Baekkeskov et al., 1989). In pulse-chase labeling experiments, rat islets were starved in methionine-free RPMI 1640 medium at 37 "C for 30 min and then given a pulse of [36S]methionine (specific activity, >lo00 Ci/mmol, Amersham) for 30 or 40 min followed by chase periods in medium containing 5 X the normal content of nonradioactive methionine for 0-72 h before harvesting. Labeled islets were harvested by centrifugation, washed twice in nonradioactive medium and once in 20 mM Hepes, pH 7.4, 150 mM NaCl and 10 mM benzamidine/HCl, and then either immediately processed for homogenization and isolation of soluble and membrane compartments or snap frozen and stored in aliquots at -80 "C.
For quantitative analysis of GAD, the resulting pellet was subjected to repeated extractions and ultracentrifugations. Usually all detergent releasable GAD was recovered in the two first extracts of the WP-100 fraction. The S1-100 and S2-100 fractions and the WP-100 extracts were immunoprecipitated quantitatively with GAD antisera  either directly or after preclearance with control sera as described (Baekkeskov et al., 1987). In brief fractions were incubated with an excess of serum for 16 h at 4 "C to form immunocomplexes, which were then isolated by adsorbtion to excess amounts of protein A-Sepharose (Pharmacia LKB Technology Inc., Uppsala, Sweden). Analysis of supernatants after immunoprecipitation by reimmunoprecipitation with antisera and protein A-Sepharose revealed that GAD was depleted quantitatively by the first immunoprecipitation. Immunoprecipitates were analyzed by SDS-PAGE followed by fluorography. Densitometric scanning of fluorograms was carried out as described . The results of density measurement of GAD in each fraction were expressed as percent of the sum of density values for GAD in all three fractions. Distribution of total protein into the three fractions was expressed as a percentage of trichloroacetic acid-precipitable counts. WP-100 values were expressed as the sum of GAD recovered in repeated extracts of the WP-100 fraction.
For salt and EDTA washing experiments identical aliquots of the WP-100 fraction were taken up in either Hepes buffer A or the same buffer supplemented with one of the following components: 0.5 M NaCl, 0.2 M Na2C03, 0.05 M Na4PZ07, 0.2 M MgC12, 0.05 M EDTA, or 2% TX-100 and incubated at 4 'C for 1 h. The samples were then ultracentrifuged at 100,000 X g and the resulting supernatants immunoprecipitated and subjected to SDS-PAGE and fluorography.
TX-114 Partitioning Assays-TX-114 detergent phase separations of detergent extracts of the WP-100 fraction or of soluble fractions following addition of TX-114 were performed as described (Baekkeskov et al., 1989). For comparative analysis of TX-114 detergent phase partition of GAD in soluble and particulate fractions, SI-100, S2-100, and a WP-100 extract were each diluted with either extraction buffer or homogenization buffer, supplemented with TX-114 to obtain an identical buffer composition in each fraction duringphase separation. Following TX-114 phase separation (Bordier, 1981;Baekkeskov et al., 1987) the aqueous (a) and detergent (d) phases of each fraction were immunoprecipitated and analyzed by SDS-PAGE and fluorography. Distribution of GAD into a and d in each fraction was measured by densitometric scanning of fluorograms and expressed as a percentage of the sum of the values in both fractions.
Gel Electrophoretic Analysis-One-dimensional SDS-PAGE was performed using uniform 8.5 or 10% SDS-polyacrylamide gels and either the buffer system of Laemmli (1970) (0.055 M Tris base, 0.192 M glycine, and 0.1% SDS) or a modified Laemmli buffer system (0.025 M Tris base, 0.192 M glycine, 0.05% SDS) (Fey et al., 1984). The GAD& 01 and ( 3 components usually only resolved in the latter buffer system. Two types of two-dimensional (2D) gel electrophoresis experiments were performed. The first involved NEPHGE (O'Farrell et al., 1977) in the first dimension as described earlier (Baekkeskov et al., 1989). This method has the advantage of applying the sample at the acidic end (anode) which results in less aggregates of GAD at the top of the gel than in the isoelectric focusing gels where proteins are applied at the basic end (cathode). However, this method does not allow determination of true isoelectric points. For this purpose equilibrium IEF was carried out in the first dimension in tube gels (170 X 1.55 mm (id)) as described by O'Farrell (1975) consisting of 5.8% v/v ampholines (pH 5-7) and 6.7% ampholines (pH 3.5-10) (Pharmacia). Following prefocusing at 1200 V for 4 h, the samples were subjected to electrophoresis for 20 h using 0.02 M NaOH at the anode and 0.01 M H3P04 at the cathode. The pH gradient was determined by elution of the ampholines from 1-cm slices of gel and measuring the pH. The second dimension for both the NEPHGE and IEF gels was SDS-PAGE on 10% slab gels using modified Laemmli buffers (Fey et al., 1984).
Antisera-The antisera used in this study included three IDDM patient sera which were strongly positive for GAD autoantibodies and three control sera from healthy individuals; a polyclonal rabbit antiserum (1266) generated against a synthetic peptide prepared according to described methods (Atar et al., 1989) and containing the C-terminal sequence of the larger rat brain form of GAD (Cys-Trp-

Ser-Ser-Arg-Thr-Gln-Leu-Leu-His-Ser-Pro-Ile-Leu-Thr-Ser-Ser-
Ser-Arg-Arg) (Julien et al., 1990;Wyborski et al., 1990) coupled to keyhole limpet hemocyanin (a gift from Dr. J. S. Petersen, Hagedorn Research Laboratory, Gentofte Denmark); GAD6, a mouse monoclonal antibody (ascites), which was raised against immunoaffinity purified rat brain GAD and which specifically recognizes the smaller GAD form in brain on Western blots (Chang and Gottlieb, 1988) (a gift from Dr. D. Gottlieb, Univ. of Washington, St. Louis), and K2, a rabbit antiserum raised to the larger form of cat GAD produced in a bacterial expression system and which almost exclusively reacts with the larger brain form on Western blots ) (a gift from Dr. Allan Tobin, UCLA).
Transient Expression of the Larger Form of Brain GAD in COS Cells-The eukaryotic expression vector 91023B (Wong et al., 1985) (a gift by Dr. Randall Kaufman, Genetics Institute, Boston) was modified at the EcoRI cloning site with synthetic linkers to produce a SpeI site. This plasmid is called pEXP130. A plasmid containing a full-length cDNA clone encoding the larger rat brain form of GAD (a gift from Dr. A. Tobin, UCLA) was digested with XbaI to isolate a 3kb fragment corresponding to nucleotides -250 to 2750. The 5' 200 base pairs are intronic (an artifact in cDNA synthesis) but contain no translational start sequence. The XbaI fragment was ligated to the SpeI site of pEXP130 and transformed using standard methods. Sense and antisense conformations were screened and named pGADl7 and pGAD16, respectively.
COS7 cells (American Type Culture Collection) were transfected using a lipofectin reagent (BRL) according to the manufacturer's protocol. 60 pg of lipofectin and 30 pg of either pGAD16 or pGAD17 were added to 100-mm dishes of COS7 cells at 80% confluency in GIBCO optimum medium supplemented with 20% FCS. Following 5h incubation at 37 "C, an equal volume of Dulbecco's modified Eagle's medium with 20% fetal bovine serum was added. Cells were harvested after 48 h, snap frozen, and stored in aliquots at -80°C.

RESULTS
Analysis of Amphiphilic GAD, a and @ in Membrane Fractions of Rat Islets-We have previously identified an amphiphilic 64-kD autoantigenic form of GAD (GAD,) in human islets and showed that this form can resolve into two isomers CY and @ which have an identical PI of 6.7 and differ in molecular mass by about 1 kDa. The frequent inavailability of fresh human material has hampered further analyses and motivated the identification of alternative sources of islets for the characterization of this protein in pancreatic @-cells. Rat islets have been shown to express amphiphilic GAD, in particulate fractions , Baekkeskov et al., 1989. To analyze this form in more detail we immunoprecipitated GAD64 from TX-114 detergent phase-purified extracts of crude particulate fraction (P-100) of a large number (n = 88) of rat islet preparations using GAD, antibodypositive IDDM sera. GAD, in particulate rat islet fractions was also detected as two isoforms, a and @, which differ in molecular mass by about 1 kDa (Fig. IA) and have a PI of 6.7 ( Fig. lB), suggesting a significant homology between the human and rat GAD, autoantigen. The a/@ ratio differed among the preparations (Fig. 1C) and the @ isoform was practically absent in approximately 25% of the preparations. The @-isoform did not differ from the CY isoform in any discernible properties, except mobility on SDS-PAGE. Thus the a and @ isoforms behaved identically with regard to subcellular distribution, hydrophobicity, membrane anchoring, and turnover time (see below). The @ isoform may be derived from the CY isoform or vice versa, either de mvo or during isolation of GAD. For this reason we refer to the 64-kD PI 6.7 CY/@ doublet as one entity, GAD,, unless otherwise indicated.
Detection of Amphiphilic and Hydrophilic GAD, but Only Hydrophilic GAD65 in Soluble Fractions of Rat Islets-To analyze the distribution of the GAD autoantigen into soluble and membrane compartments, islets were homogenized and subjected to ultracentrifugation to prepare a primary cytosol (Sl-100) fraction. The crude particulate fraction (P-100) was resuspended in an isotonic Hepes buffer to remove cytosolic proteins and proteins loosely associated to membranes, followed by ultracentrifugation to prepare a secondary cytosol fraction (S2-100) and a washed particulate fraction (WP-100). GAD was immunoprecipitated from the different fractions using GAD antibody-positive IDDM sera and analyzed by SDS-PAGE and fluorography.
Analyses of the soluble fractions of S1-100 and S2-100 showed that the expression of GAD in rat islets was not restricted to membrane bound compartments. Thus SDS-PAGE (Fig. 2, A and B ) and 2D gel electrophoresis (Fig. 2C) showed the presence of the GAD, form of PI 6.7 in both the S1-100 and S2-100 cytosol fractions in addition to the particulate fractions. No difference was detected in the mobility of GADs4 in the different compartments ( Fig. 1C; Fig. 2). The analysis of the soluble and particulate fractions furthermore revealed an additional form of GAD of M , 65,000 and PI 6.9- analysis of particulate amphiphilic GADM immunoprecipitated from rat islets by GAD antibody-positive sera from three IDDM patients (lanes 4-6). Immunoprecipitation with sera from three healthy control individuals is shown in lanes 1-3.
[3sS]Methionine-labeled rat islets were homogenized in an isotonic Hepes/sucrose buffer and the homogenate centrifuged at 100,000 X g for 1 h. The pellet (P-100) was extracted in Triton X-114 and the lysate subjected to temperature-induced phase separation. The detergent-enriched phase was precleared with normal human serum and then immunoprecipitated.
In addition to the positions of GADM alp, the positions of actin (a) and tubulin ( t ) , which were present as background in all immunoprecipitates, are indicated. The sera used in lanes 4 and 5 were used in all subsequent analyses of GAD and sera in lanes 1 and 2 as control sera. B, NEPHGE/SDS-PAGE analysis of particulate amphiphilic GADw prepared as in A. The PI of the a and components was determined by a separate analysis using IEF/SDS-PAGE (not shown). C, SDS-PAGE analysis of amphiphilic GAD-immunoprecipitated from either soluble (S2-100) or particulate (WP-100) fractions of three different rat islet cell preparations. The crude particulate fraction (P-100) from each islet preparation was washed by resuspension in Hepes buffer A, followed by centrifugation at 100,000 X g for 1 h to prepare a supernatant (S2-100) and a washed particulate fraction (WP-100). TX-114 detergent phase-purified material from each fraction was immunoprecipitated with either a control serum (C) or a GAD antibody-positive IDDM serum (I). The GADM a/@ ratio varies between the three preparations shown in lanes 14,543, and 9-12, respectively, but is similar for S2-100 and WP-100 in each preparation. Note that samples from S2-100 (lanes 5 and 6 ) and WP-100 (lanes 7 and 8) of one islet preparation were analyzed on two separate gels. 7.1 (GAD=) in the SI-100 and S2-100 fractions (Fig. 2A, lane  2; Fig. 2C, a and c ) . GADs5 was absent in the WP-100 fraction (Fig. 2C, d). Electron microscopic analysis of the S1-100 and S2-100 fractions did not reveal the presence of small membrane vesicles or fragments (data not shown), suggesting that the GAD, and GADs5 forms detected in these fractions were indeed soluble.
To analyze the hydrophobicity of GADs5 and GAD, in the different fractions, the soluble fractions S1-100 and S2-100 and the particulate fractions were subjected to TX-114 phase separation. The distribution of GAD65 and GAD, into the detergent and aqueous phases was analyzed by immunoprecipitation. This analysis showed that GADss was only present in the aqueous phase and never partitioned into the detergent phase demonstrating that this form is hydrophilic (Fig. 2, A  and B). However, whereas GAD, in both the S2-100, P-100, and WP-100 fractions partitioned into the TX-114 detergent phase (Fig. 2, A and B ) , GADs4 in the S1-100 fraction was Autoantigenic Forms of GAD in Pancreatic B-Cells detected predominantly in the aqueous phase (see below), suggesting heterogeneity in hydrophilic/hydrophobic properties of GAD, in the different compartments.
We considered that the difference in mobility between the a and B components of GAD,, might represent the membrane anchor, and therefore that the distribution of those components might differ between the soluble and particulate compartments. The a/@ ratio was however identical in each compartment ( Fig. 1C; Fig. 2B and results not shown). In the preparations where the @ isoform was not evident, the a isoform was still found in the different compartments (Fig. 2, A and C). Thus the a and B isoforms of GAD, show identical behavior with regard to compartmentalization. As described earlier (Baekkeskov et al., ,1989) TX-114 detergent phase purification of the amphiphilic form of GAD, efficiently eliminates background proteins in immunoprecipitates ( Figs. 1 and 2). In contrast, several background proteins can be detected in immunoprecipitates of crude cellular fractions and in particular in the aqueous phases using either IDDM or control sera (Fig. 2 A , lanes 25; Fig. 2B, lanes 3,4,  6,7). As shown elsewhere (Baekkeskov et al., 1989) the background proteins represent a minor fraction of abundant cellular proteins carried nonspecifically through the immunoprecipitation procedure. Since GAD is a very rare protein in the crude extracts, a small fraction of an abundant cellular protein may be detected on a fluorogram similarly to the total amount of GAD present in this fraction, which is being specifically immunoprecipitated by IDDM sera.
In addition to GADs5 and GADs4, a component of M, 55,000 was sometimes specifically detected in immunoprecipitates with IDDM sera (Fig. 2A, lanes 2 and 4; Fig. 2B, lanes 3 and  7). This component was only detected in the aqueous phases following phase separation. Mild proteolytic digestion of both GADss and GADs4 with trypsin and other proteases results in a hydrophilic fragment of this size .' Furthermore we have occasionally detected a 55-kDa GAD fragment in islet cell extracts which have not been treated with proteases. Thus GAD seems to be unusually susceptible to degradation resulting in a 55-kDa fragment? It is therefore likely that the 55-kDa component detected in some immunoprecipitates with IDDM sera in this study represents a hydrophilic fragment of either GADss or GADs4 generated by proteolytic cleavage either in vivo or during preparation of immunoprecipitates.
GADs4 and GADfis were quantitatively immunoprecipitated from the S1-100, S2-100, and WP-100 fractions in six independent experiments, and the relative distribution estimated by densitometric scanning of fluorograms and compared to the distribution of total proteins (Fig. 3). The fraction of GADs4 detected in S1-100 (-10%) was 4.5-fold lower than the fraction of total [35S]methionine-labeled proteins in this fraction. In contrast, the fraction of GADM detected in the S2-100 fraction (-30%) was about 1.7-fold higher than the fraction of total [35S]methionine-labeled proteins. This difference in fractionation suggests that GADrn in the S2-100 fraction consisted not only of a truly soluble protein entrapped in membrane vesicles, but also of protein directly released from the membranes. The washed particulate (WP-100) fraction contained approximately 60% of the ["Slmethionine-labeled GADe4, consistent with a membrane anchoring. Quantitative determination of GADs5 was more difficult due to an interfering background band in this area in some preparations of the particulate fraction. The values presented in Fig. 3, -55% in S1-100, -40 in S2-100, and 5% in WP-100 are therefore a * K. Hejnaes and S. Baekkeskov

FIG. 2. One-and two-dimensional gel electrophoretic analysis of GAD in membrane bound and soluble compartments.
A, SDS-PAGE analysis of GAD immunoprecipitated from SI-100 and P-100 fractions. Primary cytosol (SI-100) and a crude particulate (P-100) fraction were prepared and detergent phase ( d ) and aqueous phase ( a ) derived from the P-100 extract as described in Fig. 1, followed by immunoprecipitation with control (C) and IDDM (I) sera, SDS-PAGE analysis using modified Laemmli buffer, and fluorography. The immunoprecipitates in lanes 2 and 3 represent S1-100 prepared from 3000 rat islets and the immunoprecipitates in lanes 4-7 represent P-100 prepared from 1000 islets to obtain comparable signal intensities between the soluble S1-100 and particulate P-100 fractions. In addition to GADM a, the IDDM serum specifically immunoprecipitates a 65-kDa band (GAD,) from S1-100 (lane 2). This band is also present in the aqueous phase of the P-100 fraction (lane 4 ) , although it is not clearly distinguished from the background. Only GADM a is present in the detergent phase of the particulate fraction (lane 6). M, markers at the positions indicated are shown in lane 1. The M, marker with mobility between the GADS and GAD, a (indicated with an arrowhead in lane I) and the background band indicated with an arrowhead in lane 3, which has been shown to have mobility in between the GADM a and @ components, were used together with the TX-114 phase distribution pattern to identify the two bands as GAD65 and GADM a as opposed to GADM a and @.2D analyses confirmed that this preparation of islets lacked the GADM @ (not shown). The background bands in immunoprecipitates of SI-100 and the aqueous phases represent a small fraction of major cellular proteins carried nonspecifically through the immunoprecipitation procedure. IDDM sera may, however, specifically recognize a weak 55-kDa band in lanes 2 and 4 (see legend to Fig. 2B); B, SDS-PAGE analysis of GAD immunoprecipitated from aqueous and detergent phases prepared from S2-100 and WP-100 fractions as described in the legend to Fig. 1B (lanes 3 and 7). This protein may represent a analysis of immunoprecipitates of S1-100 with IDDM serum ( a ) and a control serum ( b ) and of S2-100 (c) and WP-100 ( d ) with IDDM serum. The exact PIS of the GADS and GAD% a/P components were methionine-labeled rat islets were subjected to homogenization followed by ultracentrifugation a t 100,000 X g to fractionate S1-100 (300 p l ) and P-100 fractions. The crude P-100 was resuspended in 300 pl of HEPES buffer A and ultracentrifuged to yield S2-100 and WP-100 fractions. WP-100 was extracted twice in 150 pl of extraction buffer containing 2% TX-100. S1-100, S2-100, and P-100 extracts were precleared twice by immunoprecipitation with normal serum and then divided in two aliquots and immunoprecipitated with GAD antibody-positive IDDM serum and a negative control serum, followed by resolution on SDS-gels in modified Laemmli buffers and fluorography. The distribution of GAD,, (dark-shaded bars), GAD64 (shaded bars), and total cellular proteins (open bars) into the three fractions was assessed as described under "Materials and Methods" and mean ?S.D. calculated from six experiments. rough estimate. Nevertheless they demonstrate that GADs5 is soluble and not membrane-anchored. Thus rat islets express two forms of GAD recognized by autoantibodies in IDDM, a soluble and hydrophilic form, GADs5, of PI 6.9-7.1 and a second form, GADs4, of PI 6.7 which is heterogeneous with regard to both subcellular localization and hydrophobicity.

A Major Portion of GADs4 Is Firmly Membrane-anchred-
The results described above demonstrate that a subpopulation of GAD, behaves as a soluble protein during traditional cell fractionation, and suggested that membrane-bound GADs4 might not be an integral membrane protein, but rather associated with the periphery of the membrane, such that it can be released during washing of the particulate fraction. To address this possibility, compounds which release peripheral membrane proteins were tested for their ability to release GADs4. Aliquots of WP-100 fractions were incubated in Hepes buffer A supplemented with one of the following compounds: 0.5 M NaCl; 0.2 M Na2C03 (pH 10.9); 0.5 M NaP207; 0.2 M MgC12; or 50 mM EDTA. In addition, the detergent TX-100 was used to release integral membrane proteins. After ultracentrifugation to remove insoluble material, the supernatants were analyzed for released proteins by immunoprecipitation and SDS-PAGE (Fig. 4). GAD, in the WP-100 fraction was only released from the membranes in a significant manner by detergent, but not by any of the agents known to release peripheral membrane proteins. This result indicates that the hydrophobic GADs4 exists in two different forms with regard to cellular compartments. One form, found in the S2-100 fraction, is soluble or has a low membrane avidity. The other form, found in the WP-100 fraction, is firmly membraneanchored and, in the presence of enzyme inhibitors (Hepes buffer A), can only be released from the membrane by detergent. Time-dependent spontaneous release of membrane anchored GADs4 from the WP-100 fraction can, however, take place in buffer compositions without enzyme inhibitors and may signify an endogenous enzyme capable of removing the determined by IEF/SDS-PAGE analysis in separate experiments (not shown). Background spots are labeled with small numerals to enable comparison between different panels. rat islets were homogenized in homogenization buffer and separated into S1-100 and a crude P-100. The crude P-100 fraction was resuspended in Hepes washing buffer, aliquoted in seven equal portions and ultracentrifuged. Each WP-100 aliquot was extracted in Hepes washing buffer supplemented as indicated with salts, detergent, EDTA, or nothing, and the extracts were immunoprecipitated with GAD antibody-positive IDDM serum (I) and a negative control serum ( C ) followed by analysis of GADI4 in immunoprecipitates by SDS-PAGE in standard Laemmli buffers and fluorography. membrane anchor of GADs4 and resulting in the amphiphilic but soluble S2-100 form.3 Differential Hydrophobicity of the Membrane-bound and Soluble Forms of GADs4-A portion of GADs4 consistently remained in the aqueous phase even after repeated detergent extractions and phase separations. This distribution pattern might either be due to its size precluding quantitative partitioning into the detergent phase (Bjerrum et al., 1983), or to a heterogeneity in the protein with regard to hydrophobic properties. As shown above, a large proportion of GADs4 in both the S2-100 and WP-100 fractions partitioned into the TX-114 detergent phase. To investigate whether the soluble and membrane-bound fractions of GAD, differed in their amphiphilic properties, we analyzed their distribution between the two phases. The S1-100, S2-100, and WP-100 fraction were each subjected to a phase separation using identical buffer compositions in the three fractions. The distribution of GADs4 in the aqueous phase and detergent phase was analyzed by quantitative immunoprecipitation and SDS-PAGE. The relative quantity of GAD, in the different phases was estimated by densitometric scanning of fluorograms from nine experiments and compared to the distribution of total cell proteins (Fig. 5 ) . In all the experiments, no differences in the ratio between the (Y and p forms of GADe4 were detected between the aqueous and detergent phases. In the nine independent experiments, total cellular proteins in the S1-100 and S2-100 fractions behaved similarly, with -10% and -11% distributing into the detergent phase, respectively. As expected, a much higher proportion, -55% of the total cellular proteins in the WP-100 fraction separated into the detergent phase. The corresponding figures for GADs4 in the detergent phases, as assessed after immunoprecipitation, were -8% in the S1-100, and -60% in the WP-100 fractions, respectively. The results show that the membrane-anchored and S2-100 of GAD in Pancreatic @-Cells forms of GADe4 are more hydrophobic than GAD,, found in the S1-100 cytosol fraction. Furthermore, the membraneanchored GAD,., and the S1-100 GAD,, each follow the general pattern of TX-114 partition characteristics in their respective compartments. The hydrophobic characteristics of GAD,, in the S2-100 fraction were, however, anomalous for that fraction, in that -50% partitioned into the detergent phase. GAD,, residing in this fraction is either cytosolic or has been released from the membrane, and yet displays a TX-114 binding pattern resembling that of the protein in the particulate fraction. The data are compatible with the existence of three populations of GADM which differ with regard to compartment and hydrophobicity. One form is localized to the S1-100 fraction and is mainly hydrophilic. The second form, which is most concentrated in S2-100, has a significantly increased hydrophobicity compared to the S1-100 form and seems to have a low membrane avidity. The third form, which predominates in the WP-100 fraction, has similar hydrophobicity as the S2-100 form but differs in being tightly membrane-bound. Pulse-chase Analysis of GADfis and GAD,,-Islets were labeled for short periods and then subjected to chase periods of 1-48 h to assess the half-life of GADfis and GAD,, in soluble and membrane-bound compartments. The results of one such experiment are shown in Fig. 6. Based on densitometric scanning of the autoradiogram shown in Fig. 6 and of three other experiments it was concluded that a and P of GAD,, had the same rate of turnover in each individual fraction, with the maximum incorporation a t 4 h of chase. In contrast, GADfis had a maximum incorporation a t 0 h of chase, and had a shorter half-life (54 h) than the GAD,, a/@ doublet. GADfis was hydrophilic throughout the time course in that it did not separate into the detergent phase (data not shown). The halflife of GADfi, in the three different fractions S1-100, S2-100, and WP-100 was estimated to be about 6-10 h in the S1-100 Portions of 2000 islets were starved for 30 min in 2-ml methionine-free medium, the medium was decreased to 600 pI and the islets were labeled for 40 min with 1 mCi of [:"S]methionine. Islets were chased in medium containing five times the normal methionine content for the indicated number of hours and then washed and fractionated. S-100, S2-100, and an extract of WP-100 were immunoprecipitated and then analyzed by SDS-PAGE in modified Laemmli buffers and fluorography. Lane 1 shows an immunoprecipitate with control serum, lanes 2-16 with IDDM serum. The gel was fluorographed for 1 month to obtain an exposure suitable for densitometric analysis of GAD in the WP-100 fraction (toppanel) and then refluorographed for 3.6 months to obtain an appropriate exposure of GAD in the S1-100 and S2-100 fractions. The WP-100 fraction in the long exposure is shown in the lowerpanel for comparison. fraction, about 22-28 h in the S2-100 fraction, and about 20-30 h in the WP-100 fraction. TX-114 phase separation of pulse-chase-labeled material showed that GAD,, in the S1-100 and S2-100 fractions displayed the same amphilicity pattern as demonstrated in the 4-h-labeled islets, i.e. the form predominating in SI-100 being hydrophilic and the S2-100 and WP-100 forms being amphiphilic (about 50% distributing into the detergent phase, data not shown). These results suggest that the GAD,, autoantigen in the P-cells is synthesized as a hydrophilic soluble form which predominates in the S1-100 fraction and then is processed into the hydrophobic forms seen in the S2-100 and WP-100 fractions by a maturation process that results in membrane anchoring for the WP-100 form.

Comparative Analysis of Brain and @-Cell F o r m of GAD-
We have shown that brain and @-cell forms of GAD have identical mobility by SDS-PAGE and identical patterns on two-dimensional gels (NEPHGE/SDS-PAGE) , suggesting that islets and brain express identical forms of GAD and that GADfis and GAD,, in islets correspond to the larger and smaller forms of GAD in brain (Chang and Gottlieb, 1988), respectively. The smaller brain form has been cloned and sequenced recently and shown to be encoded by a different gene than the larger form (Erlander et al., 1991). Thus in brain the two forms are clearly different and do not have a precursor-product relationship. To further assess the possible identity between the GAD forms in the two tissues, the brain and P-cell forms of GAD, as well as the larger brain form transfected and expressed in COS cells, were analyzed in parallel by immunoblotting using a set of distinctive antibodies: 1) An antibody (1266) raised against a C-terminal peptide in the larger brain form, which recognizes both forms of GAD in brain (Fig. 7, lune 5 ) , in agreement with the homology between the two forms at the C terminus (Erlander et al., 1991); 2) an antibody which preferably recognizes the larger brain form (the K2 antibody, ; or 3) an antibody which is specific for the smaller brain form (the GAD6 antibody, Chang and Gottlieb, 1988). As shown previously ) the mobility of GADM and GADfi4 in islet cells on SDS-PAGE was identical to that of the larger and smaller brain forms of GAD respectively (Fig. 7 ) . The 1266 antibody recognized both GADG and GAD64 in P-cells (Fig. 7 , lune 6). The K2 antibody specifically stained GADfis in islets (Fig. 7 , lane 3 ) (lanes 2,5,8, and 11, 25 pg of protein per lane), and from rat islets (lanes 3,6,9, and 12, 25 pg of protein per lane). Lunes 1-3 were probed with the K2 antiserum, which recognizes almost exclusively the larger form of GAD in brain (Kaufman et al., 1991, lanes I and 2). This antibody specifically recognizes GADGS and does not show detectable staining of GAD,, in islets (lane 3). Lunes 4-6 were probed with the 1266 antiserum, which recognizes both forms of GAD in brain (lanes 4 and 5 ) and in islets (lane 6 ) and furthermore stains some background bands. Lanes 7-9 were probed with the GAD6 ascites, which only recognizes the smaller form of GAD in brain (Chang and Gottlieb, 1988, lane 8). This antibody specifically binds GAD,, and does not react with GAD65 in islets (lane 9). the larger brain form of GAD in both the COS expression system and brain tissue (Fig. 7, lanes 1 and 2). In contrast GADs4 and the smaller brain form were specifically recognized by the GAD6 antibody which did not stain GADs5 or the larger brain form (Fig. 7, lams 7-9). Those results are consistent with GADss and GADs4 representing two distinct forms of GAD in pancreatic @-cells which are identical in size and antigenicity to the larger and smaller brain forms of GAD, respectively.

DISCUSSION
The 64-kDa autoantigen, which is a major target of autoantibodies associated with insulin dependent diabetes has been identified as the GABA-synthesizing enzyme glutamic acid decarboxylase in pancreatic @-cells (Baekkeskov et d., 1990). In the present report we have used human autoantibodies to identify and characterize the different autoantigenic forms of this enzyme in membrane-bound and soluble fractions of pancreatic @-cells in steady state and pulse-chaselabeling experiments. Although GAD has been detected in both membrane bound and soluble compartments of brain (Chang and Gottlieb, 1988;Baekkeskov et ~l . , 1990) and @cells , this study represents the first detailed characterization of membrane bound and soluble forms of this enzyme.
The results show that pancreatic @-cells express two distinct forms of GAD, a larger form of M,, approximately 65,000 (GADss), which is hydrophilic and soluble and has a PI of 6.9-7.1, and a smaller form of M,, approximately 64,000, and PI 6.7 (GADs4), which partitions between soluble and membrane bound compartments and is heterogeneous with regard to amphiphilicity. The pulse-chase analysis suggests that both GADss and GADs4 are synthesized as hydrophilic soluble molecules, which are predominantly found in the S1-100 fraction, and that only GADs4 is posttranslationally modified t o become an amphiphilic molecule which can either be soluble or firmly membrane anchored. It is conceivable that GAD, is modified by hydrophobic residues in a two-step process, which results in first a hydrophobic form which is either soluble or of a low membrane avidity and which is predominantly found in the S2-100 fraction and second, a firmly membrane-anchored form found in the WP-100 fraction. The second step may be reversible." The modification of GAD,, is not accompanied by detectable changes in size or charge, suggesting that the modification is mediated by small hydrophobic noncharged residues. Based on those results we propose that the membrane anchoring is mediated by a small lipid or fatty acid(s).
2D gel electrophoretic analysis of the GAD forms in rat brain and islets showed that both tissues have a larger and a smaller GAD form of similar size and PI (Baekkeskov et d., 1990). Based on the 2D gel electrophoretic analysis (Baekkeskov et d . , 1990) and the size and immunochemical comparisons of the brain and @-cell forms presented here, we conclude that the 65-and the 64-kDa P-cell form correspond to the larger and smaller brain forms of GAD, respectively. We have recently isolated cDNA spanning the entire amino acid coding region for GADw in rat islet cells. Sequencing of the cDNA' confirmed that the sequence of GADGs in islets is identical to that of the larger GAD form in rat brain (Julien et d . , Wyborski et ~l . , 1990). The amino acid sequence of the GAD65 form does not contain membrane anchoring domains (Julien et d., 1990;Wyborski et d., 1990) in agreement with the soluble hydrophilic properties of GAD,, demonstrated in the present study. Similarly the amino acid sequence of the smaller rat brain form of GAD does not contain stretches of hydrophobic amino acids (Erlander et ~l . , 1991), in agreement with our results that GADs4 in islets is also being synthesized as a hydrophilic soluble molecule. The amphiphilic properties of the more mature GADs4 and the ability of this form to become membrane anchored, however, clearly distinguishes it from GADes, which remains hydrophilic and soluble throughout its lifetime.
What is the subcellular localization of GAD? In brain, electron microscopic studies suggest that GAD is present in proximity to or associated with the membrane of synaptic vesicles which contain GABA (Wood et ~l . , 1976). Immunogold electron microscopic analysis of pancreatic sections suggest that GAD is localized to the membrane of small vesicles in P-cells, which contain GABA and stain for the synaptic vesicle protein synaptophysin (Aanstoot et ~l . , 1991;Reetz et d., 1991). It is thus conceivable that membrane bound GAD may become visible at the surface following fusion of GABAcontaining vesicles with the plasma membrane during secretion. However the exact subcellular distribution of the two forms of GAD in brain and @-cells remains to be elucidated. The two forms of mammalian GAD are encoded by two distinct and unlinked genes (Erlander et ~l . , 1991). The reason why two different forms of GAD have evolved is unknown. It is conceivable that the differences in compartmentalization of the two GAD forms demonstrated in this study may affect their enzymatic functions and influence the intracellular routes of transport and secretion of their product, GABA. Furthermore the ability of the smaller @-cell form to be either membrane bound or soluble may reflect the possibility to control its amount in membrane compartments, a characteristic which may influence the visibility of the protein to the immune system.