An Erythropoietin Fusion Protein Comprised of Identical Repeating Domains Exhibits Enhanced Biological Properties*

The hematopoietic growth factor erythropoietin (Epo) initiates its intracellular signaling cascade by binding to and inducing the homodimerization of two identical receptor molecules. We have now constructed and expressed in COS cells a cDNA encoding a fusion protein consisting of two complete human Epo domains linked in tandem by a 17-amino acid flexible peptide. On SDS-polyacrylamide gel electrophoresis, the Epo-Epo fusion protein migrated as a broad band with an average apparent molecular mass of 76 kDa, slightly more than twice the average apparent molecular mass of Epo, 37 kDa. EnzymaticN-deglycosylation resulted in an Epo-Epo species that migrated on SDS-polyacrylamide gel electrophoresis as a narrow band with an average apparent molecular mass of 39 kDa. The specific activity of the Epo-Epo fusion protein in vitro (1,007 IU/μg; 76 IU/pmol) was significantly greater than that of Epo (352 IU/μg; 13 IU/pmol). Moreover, secretion of Epo-Epo by COS cells was 8-fold greater than that of Epo. Subcutaneous administration of a single dose of Epo-Epo to mice resulted in a significant increase in red blood cell production within 7 days. In contrast, administration of an equivalent dose of conventional recombinant Epo was without effect. The pharmacokinetic behavior of Epo-Epo differed significantly from that of Epo. The results suggest that Epo-Epo may have important biological and therapeutic advantages.

Recombinantly produced proteins are gaining wide use as injectable pharmaceuticals to treat a variety of deficiencies and diseases. A problem encountered in their use is the frequency with which injections must be made in order to maintain a therapeutic level in the circulation. One means of increasing the plasma half-life of injected proteins is chemical conjugation with polyethylene glycol ("pegylation") (1). Although apparent success has been achieved with some proteins (2), this method can sometimes alter protein structure, reduce biological activity, or cause unanticipated changes in specificity and function (3)(4)(5). We hypothesized that an alternative approach would be the production of a multivalent molecule consisting of two or more biologically active units of the same protein. We speculated that these molecules would exhibit an increased plasma half-life and would also possess enhanced activity due to facilitated binding of the repeating units to their cognate receptors and to amplification of the intracellular signaling pathways. We and others have shown previously that such molecules could be produced by chemical cross-linking (6,7). However, a fusion protein with two human erythropoietin (Epo) 1 domains linked by three to seven amino acids exhibited reduced in vitro activity compared with the wild-type monomer (8).
We now report the production of a recombinant fusion protein consisting of two complete human Epo molecules in tandem separated by a 17-amino acid linker. Both Epo domains of the fusion protein are equally biologically active. Importantly, the protein has substantially enhanced potency and efficacy over conventional recombinant Epo in vitro and in vivo and is efficacious after a single subcutaneous injection.
Epo A/blunt was digested with ScaI and XhoI, whereas Epo B/blunt was digested with ScaI and BamHI. Gel-purified Epo A/blunt and Epo B/blunt digests were ligated in a 1:1 molar ratio. The resulting 1.2-and 1.4-kilobase bands were gel-purified and then ligated into expression vector pCDNA3.1(Ϫ) (InVitrogen) that had been digested previously with BamHI and XhoI and gel-purified. Ampicillin-resistant colonies (10 g/ml) were plucked, and positive clones were identified by restriction digest analysis using NgoM.
Ligation and Transformation Reactions-Ligation reactions for pCRblunt contained 25 ng of vector and a 10-fold molar excess of either Epo A3 or Epo B. 10-l reactions contained 6 mM Tris-HCl, pH 7.5, 6 mM MgCl 2 , 5 mM NaCl, 0.1 mg/ml bovine serum albumin, 7 mM mercaptoethanol, 0.1 mM ATP, 2 mM dithiothreitol, 1 mM spermidine, and 4 Weiss units of T4 DNA ligase (InVitrogen). Incubations were carried out for 1 h at 16°C. pcDNA 3.1(Ϫ) ligations contained a vector:insert ratio of 5:1, and the reaction conditions described above were used. 50 l of One Shot TOP10 competent cells (InVitrogen) were transformed with 2 l of ligation reaction mixture according to the manufacturer's procedure.
RNA Extraction and Northern Blot Analysis-Total RNA was prepared using TRIZOL reagent (Life Technologies, Inc.). The manufacturer's protocol was followed, and the RNA obtained was separated on 1.2% agarose containing 5.5% formaldehyde and transferred to Gene-Screen Plus. The DIG-labeled probe was generated according to the procedure described by the manufacturer.
Transient Expression in COS1 Cells-COS1 cells (American Type Culture Collection) were grown to 70% confluence, washed twice with phosphate-buffered saline and 6 mM glucose, and resuspended in phosphate-buffered saline/glucose to a concentration of 2 ϫ 10 6 cells/ml. 10 g of DNA were added to 0.5 ml of cells, and the cells were electroporated at 0.3 kV, 250 farads (Bio-Rad Electroporator). The cells were plated into 10 ml of Dulbecco's modified Eagle's medium/high glucose and 10% fetal bovine serum and incubated for 72 h at 37°C, 5% CO 2 . Supernatant medium was harvested, centrifuged at 10,000 ϫ g at 4°C, and then dialyzed against ␣minimum Eagle's medium. These samples were assayed for Epo protein and biological activity.
Epo Protein Determination-Epo and Epo-Epo fusion proteins were measured by Western blot and by enzyme-linked immunosorbent assay (ELISA). Samples were subjected to SDS-PAGE and transferred electrophoretically to 0.45 m nitrocellulose membranes in 25 mM Tris-HCl, 192 mM glycine, and 10% methanol. Membranes were then rinsed twice with distilled water and incubated overnight at 4°C in TBST and 10% nonfat dry milk, pH 7.5. The membranes were rinsed twice with TBST, washed once with TBST for 15 min and washed twice with TBST for 5 min each. The membranes were then incubated with anti-erythropoietin monoclonal antibody AE-7A5 (11) (Genzyme); 0.7 g/ml TBST, and 5% nonfat dry milk for 1 h at 23°C. Rinsing and washing were carried out as described above, followed by incubation with a horseradish peroxidase-conjugated goat anti-mouse IgG (Cappel) diluted 1:1000 in TBST and 5% nonfat dry milk for 1 h at 23°C. Rinsing and washing were again carried out as described above, plus two additional washes for 5 min each. Protein bands recognized by the primary antibody were detected using a luminescence kit (ECL; Amersham Pharmacia Biotech). The ELISA (Genzyme) was calibrated with recombinant human Epo.
Deglycosylation of the Epo-Epo Fusion Protein-SDS (0.5%) and 5% ␤-mercaptoethanol were added to a COS1 cell culture supernatant containing the Epo-Epo fusion protein, and the solution was heated at 100°C for 2 min. After cooling, 0.5 volume of 5% Nonidet P-40 (to counteract the inhibitory effects of SDS on N-glycanase) and 1 volume of incubation buffer (20 mM sodium phosphate and 50 mM EDTA, pH 7.5) were added to the supernatant. Specified concentrations of Nglycanase (peptide-N 4 -(N-acetyl-␤-glucosaminyl) asparagine amidase; Genzyme) were added to 100-l aliquots of the prepared supernatants. All samples were incubated for 30 min at 37°C and then prepared for SDS-PAGE. Electrophoretic transfer and Western blotting were performed essentially as described above.
Bioassay-The bioactivity of samples (in IU) was determined in vitro essentially as described by Krystal (12). The laboratory standard of recombinant Epo used to generate the standard curve was calibrated against the World Health Organization Second International Reference Preparation. Each sample was diluted with bioassay medium containing 78% ␣-minimum Eagle's medium, 20% heat-inactivated fetal bovine serum, 1% ␤-mercaptoethanol, and 1% penicillin/streptomycin/fungizone (Life Technologies, Inc.).
In Vivo Biological Activity-Three groups of C57BL/6J mice (8 -10week-old females) were used. Before administration of Epo or Epo-Epo, each animal was anesthetized, and its hematocrit was determined using blood obtained by filling two heparinized microhematocrit tubes from the retro-orbital venous plexus. An identifying mark was placed on each animal to permit the determination of its pre-and post-treatment hematocrit individually. The animals were weighed, and each received an identical amount (in IU) of Epo or Epo-Epo fusion protein by subcutaneous injection. The biological activities of the treatment samples were verified in triplicate by in vitro bioassay. The frequency of treatment was once only (day 1). On day 8, the post-treatment hematocrit of each animal was determined. Student's t test was used to determine whether the mean changes between the pre-and post-treatment hematocrits of each treatment group were different. An ␣ level of 0.05 (two-sided test) was used.
In Vivo Pharmacokinetics-Two groups of female CD-1 mice (8 -10 weeks old) were used. Before the administration of Epo or Epo-Epo, 40 -50 l of blood were collected from the tail vein into heparinized microhematocrit tubes. The tubes were centrifuged, and the plasma was collected and frozen at Ϫ80°C. Each animal received 6 IU of Epo or Epo-Epo by intravenous injection, and blood samples were obtained at 5 min, 1 h, 2 h, 4 h, 8 h, and 24 h thereafter. The injections were performed by a technical specialist staff member of the Beth Israel Deaconess Medical Center Animal Research Facility. The integrity of each injection and the absence of extravasation were confirmed by an independent observer.

RESULTS
Construction and Expression of Epo-Epo cDNA-We constructed an Epo-Epo fusion protein cDNA ( Fig. 1) beginning at the 5Ј end with the complete coding region of the 193-amino acid human Epo pre-protein (13,14) with no stop codon (Domain A). This is followed by 51 nucleotides encoding a flexible peptide linker of the sequence A-[G-G-G-G-S] 3 -T. Downstream of the linker is a domain encoding the 166-amino acid mature Epo protein followed by a stop codon (Domain B). COS1 cells were transfected with either conventional Epo cDNA (8,12) or Epo-Epo cDNA, and Northern blot analyses were carried out (Fig. 2). Cells transfected with Epo cDNA expressed a 2.0kilobase transcript. An approximately equal amount of a transcript of ϳ2.8 kilobases was expressed by cells transfected with Epo-Epo cDNA.
Efficient translation of the fusion protein message and secretion of Epo-Epo by the transfected COS1 cells were demonstrated by SDS-PAGE and Western blot analysis using an anti-Epo monoclonal antibody (10) (Fig. 3). Epo migrated as a broad band with an average apparent molecular mass of 37 kDa, consistent with the glycosylation of the 18.4-kDa polypeptide (15) at each of its four sugar attachment sites. In contrast, Epo-Epo migrated with an average apparent molecular mass of 76 kDa. This is what would be predicted from efficient glycosylation of the four sugar attachment sites in each domain (37 kDa ϩ 37 kDa) with the addition of the 17-amino acid peptide linker sequence (1.8 kDa). This result implies little or no steric hindrance of the co-translational glycosylation and oligosaccharide processing of the fusion protein.
The results of N-deglycosylation support this conclusion. Treatment of Epo-Epo with an optimal concentration of Nglycanase (16) yielded a species that migrated on SDS-PAGE as a narrow band with an average apparent molecular mass of 42 kDa, consistent with the removal of six highly branched complex oligosaccharides (Fig. 4). The use of sub-optimal amounts of the enzyme resulted in a typical ladder pattern, consistent with stepwise removal of multiple N-linked oligosaccharides. The magnitude of the reduction in average apparent molecular mass (ϳ34 kDa) achieved by complete N-deglycosylation of Epo-Epo is twice that reported for N-deglycosylation of the Epo monomer (15.5-18 kDa) (15,17,18). In some experiments, the N-deglycosylated Epo-Epo migrated on SDS-PAGE as a closely spaced doublet, suggesting that some molecules lack the Olinked sugar. This has also been reported for monomeric Epo (15). Taken together, our results are consistent with an Epo-Epo polypeptide backbone to which six N-linked complex oligosaccharides similar in molecular mass to those reported for monomeric Epo are attached. Additionally, as with monomeric Epo, the degree of O-linked glycosylation of Epo-Epo exhibits some variability.
Biological Activity of Epo-Epo in Vitro-Epo-Epo is highly biologically active in vitro. We transfected COS1 cells with either conventional Epo wt cDNA or Epo-Epo (Epo wt /Epo wt ) cDNA. After 3 days, the supernatants were harvested and subjected to an in vitro bioassay. Protein was measured by ELISA calibrated against recombinant human Epo (Table I). Epo wt was expressed at modest levels (3.5-9.2 IU/ml; 0.010 -0.026 g/ml) and exhibited a mean specific activity of 352 IU/g (13 IU/pmol). This specific activity is slightly higher than that obtained by us previously for wild-type Epo expressed in COS1 cells (201-284 IU/g) (10). In marked contrast, we found a much higher level of expression of Epo wt /Epo wt (110 -195 IU/ ml; 0.109 -0.194 g/ml), which exhibited a substantially higher mean specific activity (1007 IU/g; 76 IU/pmol).
Both Epo domains of Epo wt /Epo wt are biologically active. We showed this by mutation of arginine 103 (or the arginine 103 equivalent in Domain B, which is arginine 223 of the fusion protein) to alanine (R103A). This mutation results in complete inactivation of the Epo wt protein (19,20). We prepared three different constructs that incorporated this R103A mutation: (a) Epo R103A /Epo wt , (b) Epo wt /Epo R103A , and (c) Epo R103A /Epo R103A . We reasoned that if each of the two domains were active, then mutating only one domain would result in a protein that still retained biological activity, but with some decrease in specific activity. Only the double mutant Epo R103A /Epo R103A should be inactive. As seen in Table I, both Epo R103A /Epo wt and Epo wt / Epo R103A were expressed efficiently, and both exhibited biological activity. Interestingly, the mean specific activities of the two mutants were 480 (36 IU/pmol) and 516 IU/g (39 IU/ pmol), respectively, essentially one-half that of the non-mutated Epo wt /Epo wt (1007 IU/g; 76 IU/pmol). This result strongly suggests that introduction of the R103A mutation into each Epo domain resulted in inactivation of that domain, allowing the other, wild-type domain to activate the receptor. This indicates that both Epo wt domains of Epo wt /Epo wt are biologically active, i.e. independently capable of receptor activation, and nearly equally so. The double mutant Epo R103A / Epo R103A was expressed at the mRNA level (data not shown). However, no activity or protein was detected in the COS1 supernatant.
In Vivo Biological Activity and Pharmacokinetics-Epo-Epo exhibited enhanced activity in vivo compared with monomeric Epo (Fig. 5). We obtained individual pretreatment hematocrits from three groups of four mice. Each animal in the first group received 300 IU Epo-Epo/kg subcutaneously in the form of COS1 cell supernatant, whereas each animal in the second group received a single injection of 300 IU Epo/kg subcutaneously. Supernatant medium from COS1 cells transfected with vector alone was administered to the third group as a control. Seven days later, the post-treatment hematocrit of each animal  was determined. A substantial increase in hematocrit was observed in all four animals treated with Epo-Epo. In contrast, none of the Epo-treated animals exhibited a significant increase in hematocrit, an expected result in view of the relatively short half-life of Epo. The hematocrit of the Epo-Epotreated animals increased an average of 2.5% compared with a mean decrease of Ϫ0.2% in the Epo-treated group and a mean decrease of Ϫ0.8% in the control group. The mean change in hematocrit of the Epo-Epo-treated group was significantly different from that of the Epo-treated group (p ϭ 0.015) and that of the control group (p ϭ 0.008). We showed that the pharmacokinetic behavior of Epo-Epo is markedly different from that of monomeric Epo (Fig. 6). We injected two groups of mice intravenously either with 6 IU of monomeric Epo or with 6 IU Epo-Epo. At specified times, blood samples were obtained, and the plasma concentration of Epo or Epo-Epo was determined by ELISA. The plasma levels of mice injected with monomeric Epo decreased rapidly (Fig. 6A). In three of the four animals, the Epo concentration decreased to less than half of the peak (5 min) level within 1 h. Epo was detected in the plasma of only one animal after 4 h. In contrast, the plasma levels of all four animals injected with Epo-Epo remained high for many hours after injection (Fig. 6B). Epo-Epo levels remained detectable after 8 h in all four animals and were above 50% of the peak in two of the four animals. Epo-Epo was detectable in the plasma of two of the four animals even after 24 h. Epo-Epo also exhibited another, unexpected pharmacokinetic property. In three of the four animals, Epo-Epo levels continued to rise after the 5-min time point, peaking at 2 h after injection. This unanticipated finding was not due to variations in injection technique. All intravenous injections were performed by skilled technical personnel and monitored independently.

DISCUSSION
In the present study, we have shown that a fusion protein consisting of two complete human Epo domains separated by a flexible peptide linker has significantly enhanced in vitro and in vivo biological activity compared with the monomeric form of recombinant Epo. This enhanced activity appears to be due to an increased specific activity (Table I) coupled with a different pharmacokinetic profile. The unusual geometry of the Epo-Epo pharmacokinetics observed in three of the four animals injected with Epo-Epo (Fig. 6) remains unexplained. We carried out the ELISA analysis of the plasma samples twice at several dilutions and ruled out the possibility of the "high dose hook effect," that is, the paradoxical increase in signal with an increasing dilution of highly concentrated samples (21). The pos-sibility exists that Epo-Epo may associate with itself or with one or more other proteins in the plasma, especially at higher Epo-Epo concentrations. Such an association could interfere with antibody binding in the assay, resulting in a spuriously low reading. The unusual pharmacokinetic behavior of Epo-Epo might also be explained by a rapid and reversible interaction with receptors or other binding sites on cells in close apposition to the plasma, viz., vascular endothelial cells. Both the presence of erythropoietin receptors on these cells and a cellular response to its binding have been documented (22)(23)(24)(25)(26).
Previously, we produced chemically linked Epo dimers and showed that we could achieve an increase in potency and a decreased frequency of injection, all leading to enhanced in vivo action (6). These chemically linked dimers are, in all likelihood, a mixed population of molecules, presumably consisting of more highly active isoforms and, possibly, less active ones. The use of the fusion protein strategy in the present study provided reasonable assurance of structural homogeneity. Although the construction of the Epo-Epo cDNA was relatively straightforward, there were several uncertainties about the ultimate success of the design that could only be answered through experimentation. Specifically, they related to (a) biological activity, (b) potency, (c) the possibility of intramolecular steric hindrance, (d) post-translational processing, (e) stability, and (f) in vivo action.
Based upon our earlier studies, we had concluded that the amino terminus of erythropoietin is not involved in its biological activity because antibodies directed to the first 26 amino acids are not neutralizing (27). However, others have shown that arginine 14 may be important in the action of Epo (19). Therefore, it was unclear whether the second Epo domain (Domain B), that which is tethered by its amino terminus to the

. Pharmacokinetics of Epo (A) and Epo-Epo (B) in mice.
Groups of four mice were injected intravenously with 6 IU of Epo or Epo-Epo. Each symbol represents one animal. Note the prolonged plasma survival of Epo-Epo. See "Experimental Procedures" and "Results." peptide linker, would exhibit biological activity. An even greater uncertainty existed regarding the first Epo domain (Domain A) with a tethered carboxyl terminus, because Fibi et al. (28) demonstrated that antibodies to the carboxyl terminus of erythropoietin are neutralizing. Furthermore, the proximity of cysteine 161 of Domain A to the linker peptide raised questions regarding the fidelity with which the disulfide bond between cysteine 161 and cysteine 7, which is essential for biological activity, would be formed. However, the data show that both Epo domains are active. Presumably, utilization of the flexible linker sequence similar to that used for an interleukin-3/granulocyte macrophage colony-stimulating factor fusion protein (29) permitted proper folding of both Epo domains into their native conformations.
The recent publication of the crystal structure of Epo bound to two extracellular binding domains of the EpoR (30) shows that the amino acids Ser 9 , Arg 10 , Glu 13 , Leu 16 , and Leu 17 of Epo have minor interactions with EpoR in forming the site 1 intermolecular contact area, whereas Leu 5 , Asp 8 , and Arg 10 of Epo have minor interactions in forming site 2. In contrast, Val 11 and Arg 14 of Epo have major interactions at site 2. Near the carboxyl terminus of Epo, residues Asn 147 , Arg 150 , and Gly 151 have major interactions in forming site 1. These observations would suggest that the length of the peptide linker separating the two Epo domains may be critical in allowing sufficient steric freedom so that each Epo domain of Epo wt /Epo wt can bind to and induce dimerization of two EpoR.
Curtis et al. (29) constructed a fusion protein (designated PIXY321) consisting of granulocyte macrophage colony-stimulating factor and interleukin-3 linked by an 11-amino acid linker of the sequence (G) 4 -S-(G) 5 -S (Ϸ40 Å) and another fusion protein consisting of interleukin-3 and granulocyte macrophage colony-stimulating factor linked by a 15-amino acid linker of the sequence (G-G-G-G-S) 3 (Ϸ55 Å). Both of these fusion proteins could bind to cell surface receptors through either cytokine domain, and both exhibited biological activity in vitro consistent with unimpeded function of both cytokine domains. In contrast, a recent study by Qiu et al. (8) used shorter linkers of the sequence (G) 3-7 (Ϸ10 -25 Å) to form fusion proteins of the structure Epo R103A /Epo R103A or Epo wt / Epo wt . Interestingly, these Epo R103A /Epo R103A fusion proteins were biologically active (although less so than Epo wt ), apparently because the site 1 intermolecular contact area of each Epo R103A domain bound one EpoR, and because the linkers were short and flexible enough to allow receptor dimerization by the (theoretically) inactive fusion protein. However, in contrast to our results, Qiu et al. (8) observed no increase but rather a moderate decrease in the activity of their Epo wt /Epo wt fusion proteins compared with Epo wt . Taking into consideration the results of Curtis et al. (29) and the results of the present study, along with the structural data (23), the short linkers used by Qiu et al. (8) probably allowed each Epo wt domain of their Epo wt /Epo wt fusion proteins to bind just one EpoR, resulting in an activity similar to that observed by them for Epo wt and for their Epo R103A /Epo R103A . Apparently, the 17-amino acid linker (A-[G-G-G-G-S] 3 -T) (Ϸ63 Å) used in this study of Epo-Epo allowed both Epo domains to function unimpeded and resulted in enhanced activity.
The potency of our Epo-Epo in vitro is significantly higher than that of conventional recombinant Epo. It is 3-fold higher per microgram and 6-fold higher on a per mole basis. There are several possible explanations for this observation. It may be due to an increased stability of Epo-Epo or to a difference in endocytotic efficiency. Alternatively, this increase in potency may reflect an increase in receptor affinity. Although speculative, the possibility exists that the binding of one domain of Epo-Epo facilitates the binding of the second domain, i.e. positive cooperativity. In this regard, we have shown previously that the induction of EpoR clustering by dimethyl sulfoxide treatment of Rauscher murine erythroleukemia cells resulted in positive cooperativity and an enhanced biological response (31,32). The sequential or simultaneous binding of both Epo domains of the fusion protein might also induce such receptor clusters, thereby leading to increased local concentrations of signal transduction mediators that could result in enhanced signaling.
It is notable that the secretion of Epo wt /Epo wt in this study was 8-fold that of Epo wt (Table I). However, the transcript levels seen on Northern blots were approximately equal (Fig.  2), indicating that an increase in transcription was not responsible. Other potential causes for the difference in secretion include more efficient translation, an increased intracellular trafficking, and enhanced stability during biosynthesis and secretion. Interestingly, both Epo R103A /Epo wt and Epo wt / Epo R103A were secreted at levels greater than that of Epo wt / Epo wt itself. This finding supports the hypothesis that protein stability may play a role in the enhanced secretion of the fusion protein. Our previous work showed that mutations at arginine 103 of Epo could lead to a molecule with increased stability (10,33). Thus, because single R103A mutations in the fusion protein (which presumably enhance its stability) lead to increased rates of secretion, it seems plausible that the enhanced stability of Epo wt /Epo wt over Epo wt plays a role in the higher levels of secretion of the fusion protein.
Although our Epo R103A /Epo R103A mutant was expressed at the mRNA level, we detected no protein by either bioassay or ELISA. This result is in contrast to that reported by Qiu et al. (8) for their dimeric R103A fusion proteins with relatively short Gly 3 , Gly 5 , or Gly 7 linkers that were secreted efficiently by COS7 cells. Although speculative, we believe that these differing results suggest that the length and composition of the linker may play a role. Substitutions at R103 affect the activity and thermal stability of the monomer. They may also influence potential interactions between and among the Epo domains of the fusion protein and the linker itself, thereby altering the protein's interaction with chaperonins and its folding and secretion efficiency.