Pigment epithelium-derived factor behaves like a noninhibitory serpin. Neurotrophic activity does not require the serpin reactive loop.

Pigment epithelium-derived factor (PEDF), a neurite-promoting factor, has an amino acid primary structure that is related to members of the serine protease inhibitor (serpin) family. Controlled proteolysis of native PEDF (50 kDa) with either trypsin, chymotrypsin, elastase, or subtilisin yields in each case one major limited product of 46 kDa as analyzed by SDS-polyacrylamide gel electrophoresis. N-terminal sequence analysis of the isolated 46-kDa products indicates a favored cleavage region located toward the C-terminal end of PEDF. A proteolyzed PEDF protein reaction mixture reveals two overlapping sequences: that of the N terminus of intact PEDF and that of an internal region, consistent with cleavage of PEDF about position 382. These data indicate that PEDF protein has a globular conformation with one protease-sensitive exposed loop that contains the homologous serpin-reactive site. Cleavage within the reactive-site loop of PEDF does not cause a conformational change in the molecules (the stressed (S) relaxed (R) transition) and results in heat denaturation identical to its native counterpart. This lack of conformational change is also seen upon cleavage within the reactive-site loop of the noninhibitory serpin ovalbumin. Furthermore, the PEDF neurite-promoting function is not lost with cleavage of the exposed loop. Recombinant PEDF polypeptide fragments with larger truncations from the C-terminal end show neurotrophic activity. Our results clearly indicate that integrity of the PEDF homologous serpin reactive center is dispensable for neurotrophic activity. Thus, the PEDF induction of neurites must be mediated by a mechanism other than serine protease inhibition. Altogether our data indicate that PEDF belongs to the subgroup of noninhibitory serpins and that its N-terminal region confers a neurite-promoting activity to the protein. The neurotrophic active site of PEDF is separated from the serpin reactive-site loop, not only in the primary structure, but also in the folded protein structure.

PEDF 1 was first described as a neurite-promoting factor that is released by human fetal retinal pigment epithelial (RPE) cells. It was reported that PEDF isolated from medium conditioned by human fetal RPE primary cultures promotes neurite outgrowth in cultured human retinoblastoma Y-79 cells (1). Information about the PEDF peptide sequence has permitted the isolation and cloning of a human PEDF cDNA (2). From cDNA clones, expression vectors were constructed (3,4) and, in turn, specific antisera to PEDF were developed from the recombinant PEDF proteins (4,5). Antiserum Ab-rPEDF has been instrumental in the identification of PEDF protein in physiological sources. PEDF is present in bovine eyes in the interphotoreceptor matrix (IPM), i.e. the extracellular matrix located between RPE and the neural retina, and is the sole IPM component responsible for the IPM neurite-promoting activity (5). In addition to the effect on retinoblastoma cells, PEDF has the capacity of promoting neuronal survival of primary cerebellar granule cell neurons (6). These observations support the idea that PEDF is secreted from the RPE and has a neurotrophic effect on retinal cells.
PEDF is a 50-kDa glycoprotein with a sequence homologous to members of the serpin family (2,4,5). Sequence analysis of the 418 amino acids in the human PEDF demonstrates a 27% identity to ␣ 1 -antitrypsin, the serpin prototype. However, no inhibitory function of PEDF against serine proteases has yet been demonstrated. Serpins constitute a group of Ͼ40 proteins thought to share the same overall tertiary structure (7). Analysis of the serpin folded protein structure in solution indicates that the C-terminal region of all serpins has an exposed peptide loop that is highly susceptible to proteolysis. In the case of inhibitory serpins, the serpin-reactive site, P 1 , is located within the exposed loop, and the bond between residues at positions P 1 and P 1 Ј is cleaved by the target protease (where known). Cleavage of an inhibitory serpin by its target protease induces a change in conformation from stressed to relaxed (S 3 R) as revealed by an increase in stability to denaturation. The residue at position P 1 binds at the primary specificity pocket of the target protease, and the protease and serpin form a complex that impairs further proteolytic activity. It is believed that the amino acid in position P 1 helps define the specificity of inhibition for serpins. Thus, the serpin reactive center is a well defined structural-functional characteristic of serpin proteins. Not all serpins have a demonstrable inhibitory activity against a serine protease, e.g. chicken ovalbumin (8), rat angiotensinogen (9), and barley Z-protein (10) have no reported inhibitor activities and sequence comparison reveals that they are also members of the serpin family. As opposed to the inhibitory serpins, ovalbumin and angiotensinogen do not undergo the serpin conformational change. The human PEDF sequence has amino acid Leu in position P 1 , known to be specific for inhibition of chymotrypsin and chymotrypsin-like activities; however, the recombinant PEDF protein (positions 44 -418) does not demonstrate inhibition against these activities (3). As in ovalbumin, angiotensinogen, and some dysfunctional serpins, PEDF has residues on the N-terminal side of the P 1 residue that are considered unfavorable for the insertion of the serpin loop into the A ␤-sheet of the folded serpin protein. Incorporation of the serpin-exposed loop as an additional strand into the A ␤-sheet explains the serpin conformational change S 3 R that establishes the inhibitory status of a serpin.
In this study we have investigated the overall conformation of PEDF protein and its inhibitory/substrate status. We have used native PEDF protein purified from bovine eyes (5) and recombinant PEDF polypeptides derived from a human PEDF expression vector, pRC-BH (3). Limited proteolysis of native PEDF with a range of proteases showed that it is consistently vulnerable to cleavage at sites around the homologous P 1 site (position 382) 2 leaving most of the molecule resistant to proteolysis. We show that, unlike inhibitory serpins, the thermal stability of PEDF did not increase upon cleavage at its exposed loop. PEDF without its exposed loop and even without larger segments from its C-terminal end retained its neurite outgrowth-inducing activity. From our results we conclude that PEDF indeed has a folded protein structure in solution typical of serpins and belongs to the subgroup of noninhibitory serpins. The N-terminal region of PEDF contains a neurotrophic active site which is distal from the serpin reactive loop. The structurefunction relationships of the PEDF protein are discussed.
Preparation of PEDF Polypeptides-Native PEDF protein was purified from the IPM of fresh bovine eyes as described before (5). Briefly, IPM protein extract was subjected to 45-80% ammonium sulfate fractionation. The precipitated proteins were resuspended and dialyzed against 100 mM NaCl, 50 mM sodium phosphate, pH 6.5, 1 mM dithiothreitol, 10% glycerol, and layered onto a Mono-S column (HR 5/5) on an FPLC system. PEDF was eluted with a linear gradient of 100 -500 mM NaCl in the above buffer with a flow rate of 0.8 ml/min. Finally, the PEDF-containing fractions were pooled and subjected to gel filtration chromatography attached to an HPLC system (Gold Beckmann) using a 7.5 mm ϫ 60-cm TSK-Gel G3000SW HPLC column (TosoHaas, Inc.), equilibrated with phosphate-buffered saline, pH 7.4, at a flow rate of 1 ml/min. PEDF protein eluted as a single peak with a retention time immediately behind that for ovalbumin (M r 43,000). The final PEDF sample was stored at Ϫ80°C. Sequence analysis on the purified protein indicated that the N terminus of the mature bovine PEDF protein started at position 23 of the human PEDF precursor sequence.
Recombinant PEDF polypeptides were produced in Escherichia coli from expression vectors containing human PEDF cDNA fragments. Preparation of DNA fragments, ligation reactions, and bacterial transformations were performed as described before (11). Deletion mutant plasmids were derived from pRC-BH, an expression vector with PEDF coding sequences Asp 44 -Pro 418 (3). Plasmid pRC-BH (5.8 kb) was digested with PvuII and the ends of the 3.2-kb DNA fragment self-ligated to obtain pRC-BP. Plasmid pRC-BH was digested with XhoI and Hin-dIII, their ends filled in with Klenow fragment for a blunt-end ligation to produce pRC-BX (5.2 kb). Plasmid pRC-BH was digested with Asp718 and HindIII, their ends filled in with dNTPs and Klenow fragment followed by blunt-end ligation to obtain pRC-BA (4.8 kb) DNA. PEDF amino acid positions in the expression plasmids were as follows: BH, 44 -418; BP, 44 -267; BX, 44 -228; BA, 44 -121. Expression of PEDF genes in bacteria, cell lysis, and protein isolation from inclusion bodies were as described before for pRC-BH (3). Polypeptide fragments BP and BX were further purified by S-Sepharose chromatography in 4 M urea in 50 mM phosphate buffer, pH 6.5, and eluted with a NaCl linear gradient. Fractions containing BP and BX were pooled and stored at Ϫ80°C.
Proteolytic Reactions of PEDF-Trypsin and ␣-chymotrypsin reaction mixtures contained 20 mM Tris-Cl, pH 7.5, 150 mM NaCl, and 1 mM EDTA. Subtilisin, endoproteinase Glu-C, endoproteinase Lys-C, and elastase reaction mixtures contained 20 mM Tris-Cl, pH 8.8, and 10% glycerol. Protein amounts, incubation times, and temperatures were as indicated. The reactions were stopped by freezing in dry ice and the addition of SDS-PAGE sample buffer. Before use, the activity of each protease was confirmed by assaying cleavage of a polypeptide substrate of 39 kDa used previously in limited proteolysis (12), followed by SDS-PAGE analysis of products.
Thrombin Inhibition Solution Assays-Thrombin activity was assayed against tosyl-glycyl-prolyl-arginine-4-nitroanilide acetate (Chromozym TH; Boehringer Mannheim) following the manufacturer's protocol. Reactions were conducted at 25°C in 50 mM Tris-Cl, pH 8.3, and 200 mM NaCl, with 35 ng/ml thrombin and a substrate concentration of 0.12 mg/ml. Reaction mixtures were in a volume of 1 ml. The increase in absorbance was monitored on a Beckman DU-30 spectrophotometer at 405 nm. The rates of hydrolysis of substrate were found to be linear, at about 0.044 OD 405 nm /min. For inhibition assays, thrombin was preincubated with PEDF in PBS (20 l) at 37°C for 1 h, in the absence or presence of heparin. Preincubation periods longer than 3 h inactivated thrombin by more than 50%. The final concentration of heparin in the reaction mixture was 7 units/ml, and that of PEDF varied between 17.5 and 70 ng/ml.
Preparation of Cleaved PEDF-PEDF was treated with subtilisin at 1:60 (w/w) for 30 min at 25°C. PEDF cleavage was confirmed by SDS-PAGE. The reaction was stopped by addition of 9 mM phenylmethylsulfonyl fluoride (final concentration) when used in heat stability assays and by heating at 75°C for 20 min when used in neurite outgrowth assays. Trypsin treatment of PEDF was at a ratio of 1:100 (w/w) for 120 min at 25°C. PEDF cleavage was confirmed by SDS-PAGE analysis and N-terminal sequencing of the products. Cleaved PEDF was separated from trypsin and other products by gel filtration on a TSK-3000 column attached to an HPLC system. Cleaved PEDF was then concentrated by ultrafiltration using Centricon-30 (Amicon) concentrators.
Heat Stability Assays-Aliquots of PEDF and cleaved-PEDF protein were each incubated at 15 g/ml for the subtilisin-treated and at 150 g/ml for the trypsin-treated samples, in 75 mM Tris, 75 mM glycine, 75 mM sodium phosphate, pH 7.6, at increasing temperatures between 30 and 100°C for 2 h as described before (13). Proteins were fractionated by centrifugation at 14,000 ϫ g for 30 min at 4°C. Residual soluble PEDF protein was quantified by densitometry of immunoblot analysis with Ab-rPEDF using 4-chloro-1-naphthol as the color development reagent (5).
Sequence Analysis-N-terminal sequence determination of peptides was performed using an Applied Biosystems model 477 sequencer following the manufacturer's protocols. Amino acid sequencing of PEDF proteins was carried out on 7-10 g of protein in solution. Sequence determination of PEDF proteolytic fragments was performed on protein that had been resolved by SDS-PAGE, transferred to Immobilon poly-(vinylidene difluoride) membrane (Millipore), and stained briefly with Coomassie Blue (14). The membrane pieces containing stained bands were subjected to amino acid sequencing.
Neurite Outgrowth Analysis-Human Y-79 retinoblastoma cells were grown and treated for neurite outgrowth as described in Becerra et al. (3). Addition of effectors to 2-ml cultures was as indicated and at a final PEDF concentration as indicated. The differentiation state of the cultures was monitored by light microscopy at intervals after attachment.
Other Methods-Protein concentration was determined by the Bradford (15) assay using the Bio-Rad protein assay kit. SDS-PAGE of proteins was performed according to Laemmli (16). All protein markers were from Bio-Rad.

RESULTS
Controlled Proteolysis of PEDF-First we examined the effect of trypsin and ␣-chymotrypsin on PEDF protein purified from bovine eyes. By incubation with trypsin or ␣-chymotrypsin at a protease:substrate ratio of 1:100 (w/w), native PEDF of 50 kDa was digested in a one-step fashion into a product of 46 kDa, i.e. a decrease of 4 kDa. Complete substrate utilization was achieved by 120 min (Fig. 1, lanes 4 and 7). A band corresponding to the released low molecular weight peptide was not readily identified by Coomassie Blue staining of the gel.
To determine the position(s) of the cleavage sites, isolated tryptic and chymotryptic 46-kDa fragments were subjected to automated amino acid sequencing from their N termini. Both fragments shared the same sequence with the undigested PEDF protein (Table I). Sequence analysis on the intact protein indicated that the N terminus of bovine PEDF starts at position 23 2 of the human PEDF precursor sequence. Note that there is a basic amino acid, Arg, at position 48 that is a potential site for trypsin cleavage, and two aromatic amino acids, Phe-Phe, at positions 46 and 47 that are potential sites for ␣-chymotrypsin cleavage. To determine the sequence of the low molecular weight fragment, analysis was performed on total reaction mixtures of PEDF treated with trypsin or ␣-chymotrypsin (at a ratio of 1:100 for 120 min). Two overlapping sequences were obtained, one corresponding to the N terminus of intact PEDF starting at Asp 23 and the other corresponding to the internal C-terminal region starting at positions 382 and 383 for the tryptic and chymotryptic fragment, respectively ( Table I). Note that the internal sequence (20 amino acid residues) shares 95% identity with the human PEDF (see Fig. 7). This sequence has potential sites for chymotrypsin cleavage, e.g. amino acids Phe (positions 384, 395, and 397), Leu (positions 386 and 390), and for trypsin cleavage, e.g. Arg (position 399). These results clearly indicate that the 46-kDa fragments represent the N-terminal domain of the protein and that limited tryptic and chymotryptic cleavage sites are located between positions 381/382 and 382/383 of the PEDF molecule. Position 382 maps to the homologous P 1 site and is occupied by amino acid Leu.
Proteolytic digestion of PEDF with subtilisin at a protease: substrate ratio of 1:120 generated a 46-kDa fragment, and within 60 min of proteolysis, most of the PEDF substrate molecules were converted to product (Fig. 2, lanes 2 and 3). Sequence analysis of the isolated 46-kDa fragment indicated that it started at Thr 32 (Table I). Treatment with elastase, at a protease:PEDF ratio of 1:1, also generated a 46-kDa polypeptide fragment (data not shown), and sequence determination of the 46-kDa elastase fragment revealed that two-thirds of the molecules began at Asp 23 and one-third at Ser 30 (Table I). The loss of nine or seven amino acid residues cannot account for an overall size reduction of ϳ4 kDa. This implies that PEDF has limited cleavage sites for subtilisin and elastase located about position 380, i.e. around the P 1 position. However, treatment of PEDF with endoproteinase Glu-C or endoproteinase Lys-C, at a higher protease:substrate ratio (1:24 and 1:57, respectively) than with subtilisin, did not generate a size reduction of the 50-kDa PEDF fragment (Fig. 2, lanes 4 -7). The N-terminal sequence of the 50-kDa endoproteinase Glu-C and 50-kDa endoproteinase Lys-C fragments started at Ala 28 and at Asp 23 , respectively ( Table I

DAGQEAG
a Estimated from SDS-PAGE. b Amino acids are given in single letter code. Numbering corresponds to the derived sequence from a human PEDF cDNA where residue 1 is the first position derived for the precursor PEDF polypeptide (2) and position 23 corresponds to the first residue of the mature bovine protein (5).
c Two N-terminal sequences were obtained, parentheses indicate the molar ratio of their first amino acid. and 44, that are potential sites for endoproteinase Glu-C cleavage. Thus, sequence and size analysis of these fragments revealed that the protease-sensitive region of bovine PEDF does not contain Asp, Glu, or Lys. The limited tryptic site and the lack of an endoproteinase Lys-C site suggests that bovine PEDF has Arg around its P 1 position. In contrast, the known human sequence has Asp and lacks Arg around P 1 .

PEDF Is Not a Substrate or an Inhibitor of Thrombin-
The PEDF and heparin cofactor II (HCII) sequences have Leu at position P 1 . HCII forms a complex with thrombin and the rate of inhibition of thrombin by HCII increases Ͼ1000-fold when heparin or dermatan sulfate is present (17). At the same time, thrombin is the physiological target for the neurotrophic serpin protease nexin-1/glia-derived nexin (PN-1/GDN) that has Arg at position P 1 (18). Therefore, it was of interest to examine whether PEDF would form an inhibitory complex with thrombin. Incubation of PEDF with thrombin in the absence or presence of heparin did not generate a complex or a proteolytic product (Fig. 3). In solution assays of thrombin inhibition, preincubation of thrombin with PEDF did not decrease the rate of hydrolysis of the thrombin substrate, not even when heparin was present (data not shown). Thus, PEDF did not behave as an inhibitor of thrombin.
S 3 R Transition Assay-From the information above, the preferred cleavage site on PEDF for several proteases falls within the serpin-exposed loop and specifically for trypsin next to the homologous P 1 site. Inhibitory and noninhibitory serpins can be classified a priori by the thermal stability properties of their unmodified and cleaved species. The stability to heat of intact PEDF protein and trypsin-PEDF was assayed and is shown in Fig. 4. PEDF protein precipitated with increased temperature in a similar fashion regardless of cleavage at its exposed loop. Cleaved PEDF with or without the N-terminal cleavage (subtilisin-versus trypsin-cleaved PEDF) showed the same matching stability curves (not shown).
Biological Activity of Cleaved PEDF-To investigate the role for the PEDF serpin-exposed loop in the neurite-promoting function, we tested PEDF cleaved at its exposed loop for induction of neurite outgrowth in human retinoblastoma Y-79 cells in culture. Samples of cleaved PEDF were prepared by limited proteolysis with subtilisin or trypsin as above. The subtilisin reaction mixture containing the cleaved PEDF 46-kDa fragment (positions 32-ϳ380) was incubated at 75°C for 20 min to heat-inactivate the protease and used for analysis. The results showed that cleaved PEDF induced morphological differentiation in Y-79 cell cultures, as did the unmodified PEDF protein (Fig. 5). Induction of neurite outgrowth from the Y-79 cell cultures was also observed with the 46-kDa tryptic fragment FIG. 4. S 3 R transition assay. Aliquots of intact PEDF and PEDF cleaved by trypsin at its homologous serpin reactive loop were incubated between 30°C and 100°C for 2 h, as described before (13). Soluble proteins were then fractionated by centrifugation and analyzed by SDS-polyacrylamide gel electrophoresis (Panel A). PEDF proteins were followed by immunoblot against antiserum Ab-rPEDF and photographs of immunoblots stained with 4-chloro-1-naphthol are shown. Assay with ovalbumin stained with Coomassie Brilliant Blue is shown as control. (positions 23-381) (data not shown). Incubation at 75°C did not inactivate the PEDF protein from differentiating the retinoblastoma cells. These observations showed that integrity of the serpin-exposed loop is dispensable for neurotrophic activity. Thus, action as a serpin inhibitor is not what mediates the PEDF-induced neurotrophic activity. The resulting expression vectors produce polypeptides of 223, 185, and 77 PEDF residues, respectively. Purification of BP, BX, and BA was carried out from the inclusion bodies of their hosts, followed by S-Sepharose column chromatography for BP and BX in 4 M urea to maintain the peptides in solution. BP and BX were eluted with a linear gradient, and peak fractions containing pure BP and BX, as well as BA in inclusion body extracts, were used for analysis. The results showed that BP, BX, and BA exhibit neurite outgrowth-inducing activity (Fig.  6). Thus, even when large segments from the C-terminal end of PEDF were deleted, the neurotrophic activity is retained. DISCUSSION Analyses of the primary structure of the human PEDF sequence predict that the PEDF protein shares the common FIG. 5. Biological activity of PEDF protein cleaved at its serpin exposed loop by subtilisin. Neurite-outgrowth assay was as described (3). Human retinoblastoma Y-79 cells exponentially growing in serum containing medium were washed twice with PBS, and plated (2.5 ϫ 10 5 cell per ml) in serum-free minimal essential medium supplemented with a mix of insulin, transferrin, and selenium. Effectors were then added to the cultures. After 7 days at 37°C in 5% CO 2 , the cells were attached to poly-Dlysine coated plates with fresh serum-free medium. The differentiation state of the cultures was monitored at different intervals after attachment. Morphological characteristics of 9-day post-attachment cultures are shown. Addition of effectors were as follows: Panel A, subtilisintreated PEDF reaction mixture heated at 75°C for 20 min added at 100 ng of PEDF protein per ml of culture; Panel B, reaction mixture as in Panel A minus PEDF. serpin tertiary structure. By automated assembly of protein blocks for data base searching (19) the human PEDF protein sequence shows five high scoring peptide blocks that align with more than 50 serpin sequences (data not shown). This represents a strong relationship for PEDF to the serpin family. In addition, its linear sequence (418 residues) contains small insertions and deletions that are compatible with the threedimensional serpin model, i.e. ␣ 1 -antitrypsin, and the highly conserved residues considered critical for maintaining the serpin spatial structure (7). However, the actual folded protein conformation of native PEDF has not yet been investigated. In the studies presented here, we have used native bovine PEDF protein, which based on its strong immunoreactivity with antiserum against human PEDF is presumed to share sequence homology with the human protein (5). Our data confirm that the bovine and human sequences are highly homologous within an internal region (Fig. 7). Controlled proteolysis has proven successful in analyzing the structural conformation of proteins in solution and, in particular, the common overall structure of serpins (20). Analysis of the PEDF products from several proteases indicates that the overall native conformation of PEDF protein includes a highly protease-resistant core consisting of most of the residues from the N-terminal region, and a protease-sensitive area located around the homologous serpinreactive site P 1 (Table I). P 1 is occupied by Leu at position 382 in the bovine and human sequences. Our limited proteolysis data demonstrate that native PEDF has the common serpin structure, i.e. a globular conformation with an exposed loop located at the C-terminal end of the molecule. A circular dichroism spectrum confirms that native bovine PEDF protein contains 35% ␤ structures as shown for the folded structure of serpins (data not shown). Thus, not only the linear but also the folded protein conformation of PEDF in solution is homologous to the serpin family of proteins.

Truncation of the C-terminal Region of PEDF Does Not Affect
There are now two serpins that are known to be neurotrophic factors, PN-1/GDN and PEDF. PN-1/GDN promotes neurite outgrowth from different neuronal cell types, including neuroblastoma as well as primary hippocampal and sympathetic neurons (21,22), and rescues motor neurons from naturally occurring and axotomy-induced cell death (23). The physiological target for PN-1/GDN has been identified as thrombin (18,24). Sequence comparison between PEDF (2) and PN-1/GDN (25) reveals 23% identity and 48% homology. 3 Human and rat PN-1/GDN have Arg in position P 1 , that is specific for thrombin inhibition; while both human and bovine PEDF have Leu in P 1 (Fig. 7), specific for chymotrypsin. However, PEDF (native or recombinant) is not an inhibitor of ␣-chymotrypsin 3 (3). The effect of PEDF on thrombin was investigated since HCII, a serpin with Leu in P 1 , changes its target specificity to thrombin when in the presence of sulfated polysaccharides (17). In contrast to PN-1/GDN, we have found that PEDF does not decrease the hydrolysis of thrombin substrate or form serpin: proteinase complexes with thrombin, even in the presence of sulfated polysaccharides (Fig. 3). Thus, PEDF is not an inhibitor of thrombin and is structurally and biochemically distinct from the well established neurotrophic serpin PN-1/GDN.
To establish the substrate/inhibitor status of PEDF, our approach was based on the fact that cleavage of inhibitory serpins at or near the reactive site is followed by a dramatic transformation of structure from a stressed (S), labile conformation to a more ordered, heat stable, relaxed (R) form (26). For example, cleavage of the reactive center loop of PN-1/GDN, ␣ 1 -antitrypsin and antithrombin III gives the typical stressed to relaxed (S 3 R) change in thermal stability (27,28). As opposed to inhibitory serpins, the noninhibitory serpins do not undergo the S 3 R conformational change, as detected by heat stability, transverse urea gels, and even more quantitative 1 H and 31 P NMR spectroscopic data on native and cleaved ovalbumin and angiotensinogen species (13,29,30). Comparison of x-ray structures of cleaved forms of ␣ 1 -antitrypsin and ovalbumin supports the conformational change specific for inhibitory members. The two residues that constitute the reactive center, P 1 -P 1 Ј, in the native ␣ 1 -antitrypsin are located 67 Å apart in the cleaved species while they remain close in the cleaved ovalbumin (31). However, this lack of S 3 R transition for noninhibitory serpins has an exception. The hormone carriers, cortisol-and thyroxine-binding globulins, are reported to undergo serpin conformational change upon cleavage (32). It has been proposed that, in these two noninhibitory serpins, the S 3 R change is utilized to modulate a different property of the protein, such as transport in the hormone carrier globulins. Thermal stability curves for unmodified and cleaved PEDF reveal that PEDF lacks the S 3 R conformational change upon cleavage at its exposed loop (Fig. 4). Interestingly, PEDF becomes thermodynamically unstable at lower temperatures than ovalbumin with denaturation temperatures differing by 20°C. The thermal stability curves for PEDF resemble more the ones for angiotensinogen than the ones for ovalbumin as reported by Stein et al. (13) and presume that PEDF and angiotensinogen (29) have a greater tendency to denature or unfold than ovalbumin. The lack of S 3 R transition for PEDF may be explained by the presence of unfavorable residues in its P 12 region (3). Recently, Pemberton et al. (33) have reported that a newly identified serpin, maspin, is not a protease inhibitory serpin. Similar to the study presented here, the authors show that the tumor suppressor maspin does not undergo the S 3 R transition or inhibit trypsin-like serine proteases. In PEDF, as in angiotensinogen, ovalbumin, and maspin, the S 3 R transition may serve no useful purpose and therefore has been lost by evolutionary change. Thus, PEDF behaves like a typical noninhibitory serpin.
Our data demonstrate that while the C-terminal exposed loop is dispensable, the N-terminal region of PEDF confers a neurotrophic function to the protein (Figs. 5 and 6). Consequently, the neurotrophic induction must be mediated by other than PEDF serpin inhibition. Experiments with cerebellar granule cells have also shown that PEDF polypeptide fragments lacking the exposed loop have a neurotrophic survival effect (6,34). From these observations two distinct regions are identified on the PEDF primary structure: 1) a proximal region (BA) with a neurotrophic domain located within residues 44 -121 and 2) a distal region with an exposed loop around position P 1 (Leu 382 ) (Fig. 7). The sequence of the proximal region is apparently unique, with the highest degree of homology to members of the serpin family (20 -30%). Spatial models for PEDF based on the three-dimensional structures of ␣ 1 -anti-3 S. P. Becerra, unpublished observations. FIG. 7. Organization of PEDF polypeptide sequence. The human sequence corresponds to the derived polypeptide from PEDF cDNA clones (2,4). The bovine sequence was determined by Edman degradation of limited proteolytic fragments of native PEDF (Table I). Position P 1 is the reactive site in inhibitory serpins (36) which in the case, Leu 382 , corresponds to the position homologous to the P 1 from the ␣ 1 -antitrypsin sequence. Two distinct regions are distinguished, a proximal region with a neurotrophic domain located within positions 44 -121, and a distal region, vulnerable to proteolytic cleavage, that constitutes the homologous reactive-site loop of serpins, the P 1 /P 1 Ј region. The sequence of the proximal region has no particular homology with sequences of other serpins and is apparently unique among them. Sequence analysis of the proximal region indicates that this region aligns with helices A, B, C and part of D in the three-dimensional structure of the serpin prototype (7). Sequence alignment of the P 1 /P 1 Ј region of serpins indicates no discernable sequence conservation; however, the homology is increased at the flanking regions, the P 12 -and P 11 Ј regions (3). A signature pattern developed for the serpin family by automated assembly of protein blocks for data base (19) is present in the PEDF sequence (positions 388 -398). trypsin and ovalbumin (7,35) show that the proximal BA region, composed of putative ␣-helices A, B, C, and part of D, is located to the opposite pole from the exposed loop. This model is in agreement with the fact that the PEDF neurotrophic activity is not lost with cleavage of the exposed loop. Thus, in the folded protein structure of PEDF the neurotrophic domain is also separated from the exposed loop.
Altogether, the results presented here demonstrate that PEDF has the protein conformation of a serpin, but inhibition of serine proteases by PEDF cannot account for its neurotrophic activity. The initial biochemical step(s) for the complex biological effect of neurite outgrowth is still unknown. However, the relationships between structure of PEDF and its biological function suggest that the mode of action would include the interaction(s) between peptide residues from its proximal BA region and molecules associated to the membrane of target cells that would then trigger subsequent events of signal transduction.