Characterization and Favorable in Vivo Properties of Heterodimeric Soluble IL-15·IL-15Rα Cytokine Compared to IL-15 Monomer*

Background: IL-15 acts in vivo in association with IL-15Rα, as a heterodimeric cytokine. Results: Authentically processed and glycosylated IL-15·sIL-15Rα heterodimer was purified from human cells and characterized by sequencing and functional studies. Conclusion: IL-15 heterodimer shows favorable in vivo properties compared with monomeric IL-15. Significance: Characterization and availability of the IL-15 heterodimer is important for functional studies and clinical applications. Interleukin-15 (IL-15), a 114-amino acid cytokine related to IL-2, regulates immune homeostasis and the fate of many lymphocyte subsets. We reported that, in the blood of mice and humans, IL-15 is present as a heterodimer associated with soluble IL-15 receptor α (sIL-15Rα). Here, we show efficient production of this noncovalently linked but stable heterodimer in clonal human HEK293 cells and release of the processed IL-15·sIL-15Rα heterodimer in the medium. Purification of the IL-15 and sIL-15Rα polypeptides allowed identification of the proteolytic cleavage site of IL-15Rα and characterization of multiple glycosylation sites. Administration of the IL-15·sIL-15Rα heterodimer reconstituted from purified subunits resulted in sustained plasma IL-15 levels and in robust expansion of NK and T cells in mice, demonstrating pharmacokinetics and in vivo bioactivity superior to single chain IL-15. These identified properties of heterodimeric IL-15 provide a strong rationale for the evaluation of this molecule for clinical applications.

Interleukin-15 (IL-15), a 114-amino acid cytokine related to IL-2, regulates immune homeostasis and the fate of many lymphocyte subsets. We reported that, in the blood of mice and humans, IL-15 is present as a heterodimer associated with soluble IL-15 receptor ␣ (sIL-15R␣). Here, we show efficient production of this noncovalently linked but stable heterodimer in clonal human HEK293 cells and release of the processed IL-15⅐sIL-15R␣ heterodimer in the medium. Purification of the IL-15 and sIL-15R␣ polypeptides allowed identification of the proteolytic cleavage site of IL-15R␣ and characterization of multiple glycosylation sites. Administration of the IL-15⅐sIL-15R␣ heterodimer reconstituted from purified subunits resulted in sustained plasma IL-15 levels and in robust expansion of NK and T cells in mice, demonstrating pharmacokinetics and in vivo bioactivity superior to single chain IL-15. These identified properties of heterodimeric IL-15 provide a strong rationale for the evaluation of this molecule for clinical applications.
IL-15 acts on the surface of the cell in complex with membrane-embedded IL-15R␣ to engage the IL-2⅐IL-15 receptor ␤/␥ complex on nearby cells, a process termed trans-presentation (16). Genetic and cell transfer experiments suggested that the simultaneous expression of IL-15R␣ in the same cell is necessary for the production and secretion of IL-15 under physiological conditions (17)(18)(19)(20). We demonstrated that co-expression of the two molecules in the same mammalian cell leads to rapid intracellular association of IL-15 and IL-15R␣ in the endoplasmic reticulum, stabilization of both molecules, and efficient transport of the heterodimeric complex to the cell surface, where it is bioactive (21,22). In addition, the surface heterodimer is rapidly cleaved and released in the plasma as bioactive IL-15⅐sIL-15R␣ cytokine (22). These results demonstrated that IL-15 and IL-15R␣ are coproduced and form heterodimers, and that the active form of IL-15 found in the body is the heterodimer in two distinct forms, either cell associated or soluble. Recently, we demonstrated that the circulating form of IL-15 in biological fluids is indeed in complex with soluble IL-15R␣ (sIL-15R␣) 3 in both mouse and human (23). These findings provide additional support to previous works reporting the ability of recombinant sIL-15R␣ to act as potent agonist of IL-15 function in vivo (6,10,22,24,25), and suggest that the IL-15⅐IL-15R␣ heterodimeric form is responsible for IL-15 bioactivity. The molecular mechanism of expression and function of IL-15⅐IL-15R␣ and its in vivo persistence appear unique among the ␥ chain family of cytokines and suggest that IL-15⅐sIL-15R␣ may provide important advantages over monomeric IL-15 for clinical use.
The mechanism responsible for shedding of the heterodimer from the cell surface is poorly understood and the C terminus sequence of naturally cleaved sIL-15R␣ is unknown. For these reasons, the molecular characterization of the bioactive heterodimeric cytokine is important. Isolation of the heterodimer from human tissues is difficult, because of the small quantities of circulating cytokine (26). We therefore developed methods for production and purification of IL-15⅐IL-15R␣ heterodimers synthesized, processed, and secreted by human cells after gene transfer. We developed stable, clonal HEK293-derived human cell lines overproducing naturally processed IL-15⅐sIL-15R␣ heterodimers and an efficient purification procedure to yield biologically active heterodimeric IL-15⅐sIL-15R␣ cytokine. This also allowed the characterization of the IL-15⅐sIL-15R␣ complexes, determination of the amino acid sequence, and the proteolytic cleavage site of the sIL-15R␣, analysis of the glycosylation, and evaluation of pharmacokinetics and in vivo bioactivity.

EXPERIMENTAL PROCEDURES
Generation of Mammalian Cell Lines Overproducing Heterodimeric IL-15⅐IL-15R␣-DNA vectors optimized for the efficient expression of human IL-15 and full-length IL-15R␣ (22,27) were used for the generation of stable clonal cell lines. Highly purified, endotoxin-free DNA plasmid (Qiagen EndoFree Giga kit, Hilden, Germany) was linearized by restriction enzyme digestion and purified using the Nucleotide Removal Kit (Qiagen) and ethanol precipitated under sterile conditions. HEK293 cells (Invitrogen, number 11631017) were stably transfected by the calcium phosphate coprecipitation technique using optimized plasmids. Clones 19.7 and 1.5 were among the highest producers of IL-15⅐sIL-15R␣. Another HEK293-derived human cell line (clone 2.66) producing IL-15⅐sIL-15R␣ heterodimers was generated using DNA vectors expressing IL-15 and the extracellular region of IL-15R␣ (truncated sIL-15R␣, aa 1-175 (22,28) of the mature molecule). The cells were expanded and seeded in serum-free media in a hollow fiber system (FiberCell Systems Inc). Glucose consumption was measured daily and serum-free media was replaced when the glucose concentration dropped below 100 mg/dl. Cell supernatants (20 ml) were harvested daily for up to 5 months and assayed for IL-15 levels by ELISA (R&D Systems).
Reverse Phase-High Performance Liquid Chromatography (RP-HPLC) Separation and Analysis of IL-15 Heterodimers-For analytical RP-HPLC, samples were centrifuged to pellet cellular debris and 100 l of media containing IL-15⅐sIL-15R␣ complexes were separated by RP-HPLC under nonreducing conditions at a flow rate of 0.3 ml/min on 2.1 ϫ 100-mm Poros R2/10 column (ABI, USA), using aqueous acetonitrile/trifluoroacetic acid solvents and a Shimadzu HPLC system equipped with LC-10AD pumps, an SCL-10A system controller, a CTO-10AC oven, an FRC-10A fraction collector, and an SPD-M10AV diode array detector at 55°C. Buffer A was 0.1% trifluoroacetic acid (TFA) in water. The column was equilibrated with 10% buffer B (0.1% TFA in acetonitrile). The gradient of buffer B was: 10 -43%, 9 min; 43-44%, 12 min; 44 -85%, 4 min; 85%, 5 min. Peaks were detected by UV absorption at 206 and 280 nm. Quantitation of purified proteins was performed by amino acid analysis using a Hitachi L-8800 Amino Acid Analyzer. Purified IL-15 and sIL-15R␣ subunits were mixed at a 1:1 molar ratio to allow re-association of complexes. Analysis of reconstituted IL-15⅐sIL-15R␣ complexes was performed in both denaturing and nondenaturing conditions on polyacrylamide gels (12 and 4 -20% gradient, respectively), and visualized by Coomassie Blue staining. Formation of the complexes was confirmed by Western immunoblot analysis, using antihuman IL-15 or anti-human IL-15R␣ antibodies (AF315 and AF247, respectively, R&D Systems).
Preparative RP-HPLC was performed using a Dionex HPLC system equipped with Ultimate 3000 Binary pumps model number HPG-3400A, Ultimate 3000 Photodiode Array Detector model number PDA-3000, Ultimate 3000 Column Compartment model number TCC-3000, Ultimate 3000 Solvent Rack and Degasser model number SRD-3400, Isco Fraction Collector, model number Foxy 200. Typically, 20 ml/run were loaded directly on the column after centrifugation to remove cell debris.
For further purification of IL-15, fractions corresponding to IL-15 were pooled and re-purified first on 16 ϫ 100-mm POROS R2/10 at 26°C at flow rate 5 ml/min. The gradient of buffer B was: 20 -36%, 30 min; 36 -50%, 150 min; 50 -90%, 20 min; 90%, 10 min. Peaks were detected at 206 and 280 nm and analyzed by sequencing using an automated Applied Biosystems Inc. 477 Protein Sequencer, by SDS-polyacrylamide gel electrophoresis (PAGE), and by immunoblot analysis using an Enhanced Chemiluminescence (ECL) procedure (Amersham Biosciences Life Science). Quantitation of total protein in mixtures or purified subunits was done by amino acid analysis using a Hitachi L-8800 Amino Acid Analyzer. The sIL-15R␣ and IL-15 pools were mixed in equimolar quantities and lyophilized.
Proteolytic Digestion to Identify C Terminus of sIL-15R␣-60 g of naturally cleaved sIL-15R␣ were dissolved in 0.02 M Tris-HCl, pH 8.5, and Lys-C endoproteinase (Roche Applied Science Gmbh, Mannheim, Germany) (20:1 (w/w) protein-protease) was added and incubated for 22 h at 37°C. After digestion, reverse-phase HPLC separation of Lys-C fragments were performed under nonreducing conditions on a 2.1 ϫ 100-mm Vydac C18 column at 0.3 ml/min, using a Shimadzu HPLC system. The gradient of buffer B was: 7-30%, 25 min; 30 -70%, 5min; 70%, 5 min at 55°C. Peaks were detected at 206 and 280 nm and analyzed by sequencing using an automated Applied Biosystems Inc. 477 A protein sequencer. MALDI-TOF MS-Aliquots of HPLC fractions were mixed with equal volume of matrix solution (␣-cyano-4-hydroxycinnamic acid), and 1.5 l of mixture was spotted on a target plate and analyzed by MALDI-TOF MS (matrix-assisted laser desorption ionization time-of-flight mass spectrometry) on Ultra-flex III TOF/TOF (Bruker Daltonics, Billerica, MA). Spectra were externally calibrated in reflector mode using Bruker Peptide Calibration Standard II. Monoisotopic masses were determined using FlexAnalysis 3.3 (Bruker Daltonics) with the SNAP peak picking algorithm. The spectra were analyzed using BioTools 3.2 software (Bruker Daltonics). Mass spectrometric sequencing of particular peptides was performed by MALDI-TOF MS/MS analysis in the "LIFT" mode with manual selection of precursor ions. Fragment ion spectra were used for SwissProt protein database search by the Mascot MS/MS ion search program on the NIH Mascot server. The obtained identifications were confirmed by comparing the BioTools generated fragment ions series with experimental MS/MS data. Confidence in the identification was assessed based on the Mascot Protein or Ion score, which is Ϫ10 ϫ log(P), where P is the probability that the observed match is a random event.
In-Source Decay MALDI-TOF analysis of intact HPLC-purified sIL-15Ra was done using 1,5-diaminonaphthalene as matrix as suggested by in Bruker's technical note number TN-36 (automated acquisition of in-source decay MALDI spectra for the N-and C-terminal sequence determination of intact proteins). MS/MS tolerance in analysis of In-Source Decay spectrum using BioTools was 1.5 Da. Mass spectrometry of the sIL-15R␣ preparation was performed on an Applied Biosystems Voyager-DE Pro time-of-flight mass spectrometer operated in linear mode under positive ion conditions. Typical voltages were 25 kV accelerating, guide wire 0.15%, and grid voltage 91.5%. A nitrogen laser was used at 337 nm with 250 laser shots averaged per spectrum. A CovalX HM-1 high mass detector was used with HV-1 set to 2.5 kV and HV-2 at 20 kV. Bovine serum albumin and apo-myoglobin were used as standards for external calibration. Sinapic acid was used as matrix. Data analysis was carried out using "Data Explorer" software resident on the Voyager mass spectrometer.
IL-15 Treatment in Mice-Six-week-old female C57BL/6 mice were obtained from Charles River Laboratories, Inc. (Frederick, MD). Escherichia coli-derived monomeric IL-15 (29) and purified IL-15 heterodimers were injected at a dose of 3 g of IL-15 eq/mouse intraperitoneally. Mice were bled at different time points after protein injection, and the serum IL-15 levels were measured using human IL-15 chemiluminescent immunoassay (Quanti-Glo, R & D Systems).

CFSE Labeling of Cells and Cell Adoptive Transfer in Mice-
To make single cell suspensions, spleens from 6-week-old female C57BL/6 mice were gently squeezed through a 100-m Cell Strainer (Thomas) and washed in RPMI 1640 medium (Invitrogen) to remove any remaining organ stroma. Cells were incubated for 10 min at 37°C with CFSE (2 M; Molecular Probes) and washed twice. CFSE-labeled cells (20 ϫ 10 6 ) were resuspended in PBS and injected intravenously into congenic mice. IL-15 treatment was performed the day after cell injection. At day 4, mice were sacrificed and splenocytes were analyzed by multiparameter flow cytometry to evaluate the bioactivity of IL-15. Briefly, the cells were washed in FACS buffer containing 0.2% fetal calf serum and stained with the following panel of conjugated rat anti-mouse antibodies: CD3-APCCy7, CD4-PerCP, CD8-Pacific Blue, and NK1.1-PeCy7 or -APC (BD Biosciences). The percentage of cells of the original population that had divided in response to IL-15 treatment was calculated based on CFSE intensity. Samples were acquired using a LSRII flow cytometer (BD Biosciences), and the data were analyzed by FlowJo software (Tree Star, San Carlos, CA). In some experiments, surface staining of cells was followed by intracellular staining with Ki-67 antibody (BD Biosciences) for the detection of proliferating cells.

Generation of Human Cell Lines Producing High Levels of IL-15⅐sIL-15R␣-
We previously reported the generation of optimized combination vectors for the coordinate expression of the two chains of the human heterodimeric cytokine IL-15⅐IL-15R␣ (21,22). Yields of secreted bioactive IL-15 achieved by these plasmids were Ͼ1,000-fold higher compared with wt IL-15 cDNAs (22,27). These plasmids were used to develop stable clonal IL-15⅐sIL-15R␣-producing human HEK293 cell lines. We used intact genes of IL-15 and IL-15R␣ for the generation of these cell lines. Membrane-bound IL-15R␣ is composed of 5 domains, the sushi domain responsible for the binding to IL-15, a linker region, the Pro/Thr-rich domain, the transmembrane domain and cytoplasmic tail (30, 31) (Fig. 1). Upon stable introduction of the genes into HEK293 cell lines, the heterodimer was obtained in the culture supernatant in a soluble form, after transport of the IL-15⅐IL-15R␣ to the plasma membrane and proteolytic cleavage of the extracel- lular part of IL-15R␣ by cellular enzymes. The mature secreted IL-15R␣ molecule is depicted in Fig. 1. Clones 19.7 and 1.5 were selected as the highest producers of IL-15⅐sIL-15R␣ heterodimers. Both stable clones were grown in continuous culture in serum-free medium, using a hollow fiber culture system, and the supernatants were harvested daily. Clone 19.7 produced 70 mg of IL-15/liter (calculated as monomer IL-15 by ELISA) for up to 5 months (supplemental Fig. S1A). HPLC analysis under nonreducing conditions of weekly samples of supernatants produced by clone 19.7 collected from day 29 through day 137 in the bioreactor demonstrated stable production of IL-15⅐sIL-15R␣ heterodimer over this interval (supplemental Fig. S1B). Similar results were obtained for clone 1.5. These results demonstrated that high levels of IL-15⅐sIL-15R␣ heterodimeric cytokine production were achieved in stable human HEK293-derived cell lines. We have previously reported the generation of a DNA vector encoding the extracellular portion of IL-15R␣ (22) (truncated sIL-15R␣, encoding the signal peptide and 175 amino acids of the mature IL-15R␣, Fig. 1). We established an additional cell line (clone 2.66) producing the engineered IL-15⅐sIL-15R␣ complexes after transfer of genes encoding IL-15 and the truncated sIL-15R␣ form. The known C terminus sequence of truncated sIL-15R␣ generated from clone 2.66 ( Fig. 1) was used as control for the identification of the natural cleavage site on IL-15R␣ after expression of the full-length molecule on the cell membrane (see below).
Purification IL-15⅐sIL-15R␣ by HPLC Under Nonreducing Conditions-We employed a purification strategy using reverse-phase HPLC (RP-HPLC) under nonreducing conditions, which keeps intact the disulfide bonds of the noncovalently associated IL-15 and sIL-15R␣ subunits but dissociates the subunits from each other. The chains were purified and characterized separately, and they were subsequently re-associated in vitro at a 1:1 molar ratio to regenerate the intact heterodimeric cytokine. The first purification step was performed using Waters RCM 25 ϫ 100-mm Bondapak C18 column, with the IL-15 and sIL-15R␣ subunits dissociating in the acetonitrile containing buffer used for separation. Fig. 2A shows a typical RP-HPLC elution profile of heterodimer produced from clone 19.7. The content of sIL-15R␣ and IL-15 peaks were analyzed by SDS-PAGE (Fig. 2B) and the identities of both proteins were confirmed by immunoblot analysis (not shown). sIL-15R␣ and IL-15 were eluted as broad peaks in several fractions), fractions 1-3 for IL-15R␣ and fractions 4 -13 for IL-15. Comparisons of the different fractions revealed small differences in size among eluted proteins (Fig. 2), likely due to differences in posttranslational modifications (e.g. glycosylation). To achieve Ͼ95% purity for sIL-15R␣, a pool of fractions 1-3 was subjected to chromatography on a 16 ϫ 100-mm POROS R2/10 column (supplemental Fig. S2, A and B). IL-15 containing fractions 4 -13 were also re-purified on the same column (supplemental Fig. S2, C and D). All purifications were performed under nonreducing conditions to maintain natural disulfide bonds in both proteins. The final pools of sIL-15R␣ (supplemental Fig.  S2, A-B, Fx 1-3) and IL-15 (supplemental Fig. S2, C and D,  were analyzed by N-terminal Edman sequencing and the amount of each protein was determined by quantitative amino acid analysis. The molar ratio of the purified proteins recovered from the HPLC separation was ϳ1:1. sIL-15R␣ and IL-15 were mixed at equivalent molar amounts in PBS, allowed to re-associate, then analyzed by native PAGE. Under native conditions, all three molecular species of sIL-15R␣, IL-15, and heterodimer were detected as diffuse bands, likely due to heterogeneity of glycosylation in human HEK293 cells. Determination of C-terminal Sequence of sIL-15R␣-To determine the C-terminal sequence of naturally cleaved sIL-15R␣, purified sIL-15R␣ (from clone 19.7) was digested with Lys-C endoproteinase, generating peptides suitable for N-terminal sequencing and mass spectrometry (MS) analyses. Truncated sIL-15R␣ produced from clone 2.66 was also purified as described above for clone 19.7 and was used as reference because its C terminus sequence was known. The expected peptides after Lys-C endoproteinase digestion from truncated sIL-15R␣_2.66 are shown in supplemental Table S1. The peptides generated after Lys-C digestion from the naturally cleaved sIL-15R␣_19.7 were separated by RP-HPLC under nonreducing conditions, generating a smaller number of peptides for further analysis. The C-terminal peptide produced a broad peak that eluted early in RP-HPLC and was collected in several fractions (Fig. 3, Fx 21-23). Analysis by N-terminal protein sequencing of fraction 22 obtained after Lys-C proteolysis and RP-HPLC (Fig. 3) revealed 23 amino acid residues, XIRDPALVHQR-PAPPS(T)VXXAGV, corresponding to residues 63-85 of the mature IL-15R␣ (residues numbered according to the mature amino acid sequence) along with the 19-mer NWELXAXAS-HQPPGVYPQG, which corresponds to the sequence after Lys-151, the last Lys residue before the IL-15R␣ transmembrane domain (supplemental  (Fig. 4). Upon protein sequence analysis of fraction 22 (NWELXAXASHQPPGVYPQG), we did not detect the expected Thr-5(in the protein:Thr-156) and Ser-7(in the protein:Ser-158), whereas we were able to detect Ser-9(in the protein:Ser-160). This suggested that Thr-5 and Ser-7 were most likely modified, probably through O-glycosylation. Analysis of m/z 2020.927, 2056.934, 2308.082, 2365.108, 2455.138, and 2770.224 in MS/MS mode showed very similar fragmentation spectra in the low mass region (up to 1225 Da), suggesting that these peptides were most likely derived from the same peptide sequence but present different post-translational modifications. The fragmentation spectra of m/z 2020.927, 2056.943, 2365.108, and 2770.224 and their analysis are shown in supplemental Fig. S4. Because O-glycosylation was likely to be on Thr-5 and Ser-7, the formation of unmodified N-terminal b-ions and C-terminal y-ions preceding these residues could be expected, potentially allowing the identification of the corre-  sponding peptide using Mascot search of the protein sequence database. In all cases (m/z 2020.927, 2056.934, 2308.082, 2365.108, 2455.138, and 2770.224), the Mascot MS/MS ion search resulted in the identification of the same sequence of IL-15R␣ that was obtained by protein sequencing, confirming that the NWELTASASHQPPGVYPQG sequence corresponds to the whole C-terminal peptide of naturally cleaved sIL-15R␣_19.7. A similar analysis was performed on naturally cleaved sIL-15R␣ purified from clone 1.5, leading to the same conclusion. Taken together, these data demonstrate that the proteolytic cleavage of membrane-bound IL-15R␣ takes place between Gly-170 and His-171 in two different cell clones. Truncated engineered sIL-15R␣ produced from clone 2.66 includes 5 additional amino acids at the C terminus (Fig. 1).
Identification of Post-translational Modifications of sIL-15R␣-Analysis of the naturally cleaved sIL-15R␣ expressed from stable human cell lines by MALDI-TOF MS revealed the presence of numerous post-translational modifications. The expected molecular mass of the polypeptide chain of sIL-15R␣ (170 amino acids) is 17,839.86 Da (Fig. 1). MALDI-TOF MS analysis of the purified sIL-15R␣ revealed a broad peak with a center at 34,910 Da (supplemental Fig. S3), implying that almost half of its molecular mass is due to post-translational modifications, most likely O-and N-linked glycosylation. Detailed analysis characterized the post-translational modification in the C-and N-terminal region of naturally cleaved sIL-15R␣. The C-terminal peptide with m/z 2020.927 (Fig. 4) was 18 Da less than the predicted NWELTASASHQPPGVYPQG peptide with m/z 2038.962, indicating loss of water. Because O-glycosylations were the likely modifications in the C-terminal region of sIL-15R␣, it is possible that 2020.927 is the product of the ␤-elimination reaction during proteolysis of the purified sIL-15R␣ by Lys-C digestion. A ␤-elimination reaction at O-glycosylated Ser and Thr residues would lead to loss of water with formation of dehydroalanine and dehydroaminobutyric acid, correspondingly. Peptides with dehydroamino acid residues containing a reactive double bond are not stable and quickly react with available nucleophilic compounds. The peptide with m/z 2020.927 appeared to be stable and did not react with 2-aminoethanethiol suggesting that its double bond probably reacted intramolecularly with the hydroxyl group of a neighboring Ser or Thr residue. The ion fragment spectrum of parent ion 2020.927 is shown in supplemental Fig. S4A. The peptide with m/z 2056.934 differs from 2020.927 by 36 Da and represented the m/z of the same NWELTASASHQPPGVYPQG peptide, which first lost water due to ␤-elimination and then added HCl to the double bond of a dehydroamino acid. The ion fragment spectrum of parent ion 2056.934 is shown in supplemental Fig. S4B. Analysis of this spectrum in BioTools with modification (chlorine addition) on Thr-5 and Ser-7 suggested that O-glycosylation took place at these sites with equal probability. Analysis of ion fragment spectra of parent ions 2308.082, 2365.108, 2455.138, and 2770.224 suggested that they have O-linked oligosaccharides attached to the peptide with m/z 2038.962 (supplemental Fig. S4, C and D).
We also performed a detailed characterization of the N-terminal region of sIL-15R␣. MALDI-TOF MS/MS of m/z 1922.963 ( Fig. 3; fraction 40) identified the N-terminal peptide ITCPPPMSVEHADIWVK of sIL-15R␣. However, m/z 2126.150 of fraction 39 gave a similar fragment ion spectrum, suggesting that both ions correspond to the same peptide. Mascot MS/MS ion search using fragment ions derived from 2126.150 when the parent ion was set as 1922.950 (corresponding to the theoretical m/z of unmodified peptide) confidently identified the same sequence ITCPPPMSVEHADIWVK (supplemental Fig. S5A). The mass difference between these m/z is 203.187, which is close to the mass of N-acetylhexosamine (HexNAc; N-acetylgalactosamine or N-acetylglucosamine). Analysis of the spectra with the modification set on Ser-8 (supplemental Fig. S5B) and Thr-2 (supplemental Fig. S5C) suggested that the modified residue was most likely Ser-8 because no additional expected modified b-ions were detected when modification was set on Thr-2. There are several Ser residues in the N-terminal sequence of IL-15R␣ beyond Ser-8. To determine whether some of these residues are also modified, purified sIL-15R␣ was analyzed by In-Source Decay MALDI-TOF MS. The spectrum was analyzed using Top-Down Mascot Search of the Sprot human database, which allowed the identification of the mature N-terminal sequence of IL-15R␣ (supplemental Fig.  S6A). Analysis of the spectrum confirmed that Ser-8 is partially modified by N-acetylhexosamine (HexNAc) along with Ser-18, -20, -23, and -31 (supplemental Fig. S6, B-F). These data showed that naturally cleaved sIL-15R␣ produced from human cells is heavily glycosylated with N-and O-linked glycosylation both in the N-and C-terminal regions of the protein (Fig. 1).
Determination of in Vivo Half-life of Different Forms of IL-15-IL-15⅐sIL-15R␣ heterodimer produced from human cell lines retains all the naturally occurring post-translational modifications, such as native disulfide bonds and both N-and O-linked glycosylation (Fig. 1). These post-translational modifications are absent in E. coli-derived monomeric IL-15 (29) and, along with the lack of the IL-15R␣ subunit, could affect the stability, pharmacokinetics, and bioactivity of IL-15 cytokine.
The purified IL-15 and sIL-15R␣ chains were reconstituted as described above (Fig. 2C) and used in experiments in vivo to evaluate their pharmacokinetic and pharmacodynamic properties in comparison to both single chain nonglycosylated IL-15 produced in E. coli and purified by conventional chromatography (29) and glycosylated single chain IL-15 produced by human HEK293 cells and purified as described above (Fig. 2). To evaluate the in vivo half-life of IL-15⅐sIL-15R␣ heterodimers in comparison to single chain IL-15, mice (5/group) were injected intraperitoneally with 3 g of human E. coli-derived single chain IL-15, 3 g of human HEK293 cell-derived single chain IL-15, or an equimolar amount of purified human IL-15⅐sIL-15R␣ (corresponding to 3 g of IL-15 monomer, clone 1.5 lot 1). Serum was collected at various times after injection and IL-15 levels were measured by ELISA (Fig. 5). Both single chain IL-15 preparations reached peak plasma levels at 30 min after protein administration (ϳ70 and ϳ25 ng/ml for E. coli-derived and human HEK293 cell-derived IL-15, respectively). The difference in the IL-15 plasma levels between these two preparations is most likely a consequence of the different capability of the ELISA antibodies to detect unglycosylated and glycosylated IL-15 (data not shown). In contrast, administration of the IL-15⅐sIL-15R␣ heterodimer resulted in higher peak IL-15 levels 2 h after injection (ϳ120 ng/ml). The half-life of both single chain IL-15 preparations was similar and less than 30 min, whereas the IL-15⅐sIL-15R␣ heterodimer had a significantly extended half-life (ϳ4 h) (Fig. 5). The area under the curve for the 24-h period was 20 times higher for the heterodimer, compared with both single chain IL-15. Similar results were obtained after intravenous or subcutaneous administration (data not shown). Human cell-produced IL-15⅐sIL-15R␣ heterodimer thus has a favorable pharmacokinetic profile in mice in comparison to single chain IL-15.
Bioactivity of Heterodimeric IL-15⅐sIL-15R␣ in Vivo-We also compared the biological activity of the different forms of IL-15 in mice. CFSE-labeled splenocytes were transferred into C57BL/6 mice. The mice were subsequently treated with PBS, with 3 g of E. coli-derived single chain IL-15, or with an equimolar amount of IL-15⅐sIL-15R␣. Proliferation of transferred cells was evaluated 4 days after treatment. IL-15⅐sIL-15R␣ heterodimer induced a greater proliferation of donor CD8 ϩ T cells (Fig. 6A, top panels) and NK cells (Fig. 6A, bottom panels) in comparison to single chain IL-15 with a higher frequency of proliferating cells, and more rounds of cell division. Upon single chain IL-15 injection, few CD8 ϩ T cells divided even once, whereas the IL-15⅐sIL-15R␣ heterodimer induced multiple rounds of division, resulting in a significant increase in the frequency of CD8 ϩ T cells in spleen. Thus, the IL-15⅐sIL-15R␣ heterodimer was more stable in vivo, had a prolonged serum half-life, and was more bioactive on a molar basis, compared with single chain IL-15. The serum levels of IL-15 correlated with the biological activity, as measured by CFSE dilution of transferred cells (Fig. 6A). Although the overall proliferation of CD4 ϩ T cells did not change upon IL-15 administration, analysis of different subsets of memory CD4 ϩ T cells showed an expansion of effector memory cells (data not shown), as previously reported (32)(33)(34). We also confirmed the increased pro-liferation of CD8 ϩ T and NK cells upon IL-15 administration by measuring the frequency of cells expressing Ki-67. Mice (5/group) were injected intraperitoneally with 3 g of human E. coli-derived single chain IL-15, 3 g of human HEK293 cellderived single chain IL-15, or an equimolar amount of purified human IL-15⅐sIL-15R␣. At day 4 after protein administration, all IL-15-treated mice showed an increased frequency of Ki-67 ϩ CD8 ϩ T and NK cells in comparison to PBS-treated mice (Fig. 6B). IL-15 heterodimer induced a greater proliferation of these lymphocyte subsets in comparison to both single chain IL-15 preparations, which were characterized by a similar bioactivity in agreement with their similar pharmacokinetic profile.
The bioactivity of IL-15 heterodimer formulations was similar in different purified lots. Mice were injected intraperitoneally with 3 g of human IL-15⅐sIL-15R␣ from two separate purification lots and sacrificed at day 3 after injection (Fig. 6C). Administration of human IL-15⅐sIL-15R␣ in mice resulted in an increased frequency of proliferating CD8 ϩ T cells (defined as Ki-67 ϩ ) in comparison to untreated mice after both intraperitoneal (Fig. 6C) and intravenous delivery (data not shown). No difference in bioactivity (measured as CD8 ϩ T cell proliferation) was observed comparing IL-15⅐sIL-15R␣ obtained from different production/purification lots (Fig. 6C).

DISCUSSION
IL-15 is a important cytokine with potential clinical applications as a lymphocyte growth and activation factor (15). Recombinant human IL-15 generated in E. coli has been produced as a nonglycosylated monomer of ϳ12 kDa (29). Preclinical studies to evaluate safety, toxicity, pharmacokinetics, and pharmacodynamics of monomeric human IL-15 have been conducted in rhesus macaques, and showed an increase in the absolute number and proliferation of NK and CD8 ϩ T cells (33)(34)(35)(36). Although monomeric E. coli-produced IL-15 is in the initial stages of clinical testing, this form of the molecule poses multiple challenges for clinical use due to its instability and rapid plasma clearance (35,36). IL-15 expression is tightly regulated at the transcriptional level, as well as at several posttranscriptional and post-translational steps such as mRNA stability, generation of alternative spliced isoforms, intracellular trafficking, interaction with IL-15R␣, and secretion (21,22,24,(37)(38)(39)(40)(41)(42). We have employed a systematic approach to reproduce in engineered human cells the natural steps of production and processing of IL-15⅐sIL-15R␣ heterodimers, resulting in efficient production and purification of the bioactive IL-15 heterodimeric cytokine, which appears to have important potential advantages over the monomeric form of the molecule. Taking advantage of the stabilization of IL-15 by co-expression with IL-15R␣ (22), we produced combination vectors expressing the heterodimeric cytokine IL-15⅐IL-15R␣, providing strong improvements in yield (21,22). These vectors were used to develop stable clonal human HEK293 cells that grow in serum-free medium and express and secrete high levels of human IL-15⅐sIL-15R␣ complexes (up to 70 mg of IL-15/liter). We developed an efficient procedure for the purification of biologically active IL-15⅐sIL-15R␣ heterodimers based on nonre- ducing RP-HPLC. The IL-15 and sIL-15R␣ chains of the IL-15 heterodimeric cytokine are noncovalently linked and can be separated under certain conditions, such as pH Ͻ 3.5 (16). This allowed us to produce pure preparations of single chain IL-15 and sIL-15R␣. Importantly, RP-HPLC was performed under nonreducing conditions, avoiding the need for protein refolding after purification. Because the K d of the two chains is ϳ10 Ϫ11 M (30,43), the purified IL-15 and sIL-15R␣ subunits can be combined in vitro in a 1:1 molar ratio in PBS allowing rapid formation of the bioactive heterodimeric cytokine. IL-15 is stabilized in the presence of IL-15R␣, and the generation of IL-15 as a heterodimeric cytokine has the additional benefit to be a more stable structure, avoiding denaturation and inactivation and decreasing the possibility of creating immunogenic forms.
HEK293 human cells produce correctly folded, processed, and glycosylated human IL-15⅐sIL-15R␣ heterodimeric cytokine. Both IL-15 and sIL-15R␣ subunits of the heterodimeric cytokine contain intramolecular disulfide bonds and are heavily glycosylated (Fig. 1). IL-15 has three potential N-linked glycosylation sites (44). In agreement with a previous publication (31), the data presented in this study show that sIL-15Ra contains both N-and O-linked carbohydrates both in the N-and C-portion of the molecule. Several studies have demonstrated the contribution of glycosylation to the effect of cytokines and growth factors (for reviews, see Refs. 45 and 46). Glycosylated interferon-␤ (47), erythropoietin (48,49), granulocyte colony stimulating factor (50), and IL-7 (51) are more stable and bioactive in comparison to the nonglycosylated forms. Glycosylation was also reported to affect the interaction with specific receptors, as IL-7 was able to bind glycosylated IL-7R␣ 300-fold more tightly than unglycosylated IL-7R␣ (52). Additionally, production of factors for clinical use in human cells may reduce the risk of immunogenicity. Administration of recombinant human granulocyte macrophage colony stimulating factor was associated with the development of antibodies against the recombinant protein. These antibodies were found to react against sites on the protein that are normally protected by O-linked carbohydrates (49). Similarly, administration of E. coli-derived human IL-7 in humans induced antibodies against the recombinant protein (53), whereas no anti-IL-7 antibodies were found using the glycosylated cytokine (CYT107) (54). Immunogenicity of E. coli-derived single chain human IL-15 has been reported in macaques where the development of anti-IL-15 antibodies was observed upon subcutaneous administration (35).
Pharmacokinetic and pharmacodynamic properties of the purified glycosylated human IL-15 heterodimers were investigated in mice. In comparison to single chain IL-15 produced both in E. coli and in human HEK293 cells, IL-15⅐sIL-15R␣ complexes showed a more prolonged serum half-life and were more bioactive on a molar basis. The superior bioactivity of IL-15 in the heterodimeric formulation is mainly the result of the presence of IL-15R␣ contributing to increased stability of the protein in vivo. These properties offer the potential to allow lower and less frequent dosing and simpler delivery methods, with increased convenience for both patients and caregivers. The crystal structures of the heterodimer IL-15⅐IL-15Ra and the quaternary IL-15⅐IL-15R␣⅐IL-2R␤-␥c complex have been reported (55,56). These reports identify the amino acids and domains involved in the binding between chains. Importantly, IL-15 has two distinct binding sites, site I for the binding to IL-2R␤ and site II for the binding to IL-2R␥c. In contrast, IL-15R␣ does not contact IL-2R␤, with a distance of Ͼ15 Å separating the subunits at their closest point. It is also reported that the IL-15⅐IL-15R␣ heterodimer binds to IL-2R␤ with an affinity ϳ150-fold greater than that of single chain IL-15, suggesting that IL-15 stabilization is a major function of IL-15R␣ (56).
The production and purification of IL-15 associated with the sushi domain of IL-15R␣ linked to the Fc region of IgG1 has been previously reported (57). This molecule was also reported to show favorable pharmacokinetics and increased biological activity in comparison to single chain IL-15 both in vitro and in vivo. However, the authentically processed and glycosylated IL-15⅐sIL-15R␣, as produced and purified in this study, has the advantage to be the closest to the IL-15 produced in the human body and circulating in plasma (23) and may be the least immunogenic form.
The availability of naturally cleaved purified sIL-15R␣ has allowed us to investigate processing and shedding of IL-15R␣ from the cell surface. MALDI-TOF MS analysis, protein sequencing, and Mascot searches of protein sequence databases confidently identified the proteolytic cleavage site of membrane-bound IL-15R␣ between Gly-170 and His-171 of the mature membrane-associated form of IL-15R␣ (Fig. 1). The determination of the sequence corresponding to the cleavage site in IL-15R␣ and the amino acid sequence of the mature sIL-15R␣ represent an important finding in support of further studies, aiming to identify how this process is regulated. Dysregulated shedding of IL-15⅐IL-15R␣ heterodimers from the cell surface may be one mechanism leading to the altered levels of circulating IL-15 upon certain conditions, i.e. lymphodepleting treatments (23,26) and autoimmune diseases, such as celiac disease (58), rheumatoid arthritis (59), and multiple sclerosis (60). Dissecting the molecular mechanism and regulation of the IL-15⅐IL-15R␣ shedding process may offer opportunities for therapeutic targeting in IL-15-associated pathological conditions and may help the design of better therapeutic and/or prognostic strategies.
In summary, our study demonstrates that high-level production of authentically processed and glycosylated human IL-15⅐sIL-15R␣ heterodimers is achievable in human cells and that RP-HPLC under nonreducing conditions allows the purification of biologically active heterodimers, avoiding protein refolding. Purified glycosylated IL-15 heterodimers have the FIGURE 6. IL-15⅐sIL-15Ra heterodimer is bioactive in vivo. A, IL-15⅐sIL-15R␣ heterodimer is more potent than single chain IL-15, in vivo lymphocyte proliferation. 12 h after transfer of CFSE-labeled splenocytes (20 ϫ 10 6 ), mice were treated intraperitoneally with PBS (3 mice), with 3 g of IL-15 (3 mice), or with equimolar quantity of IL-15⅐sIL-15R␣ (3 mice, 3 g in single chain IL-15). CD8 ϩ T cells (top panels) and NK cells (bottom panels) were analyzed on day 4 by flow cytometry for CFSE fluorescence. One representative mouse per group is shown. B, IL-15R␣ contributes to the superior activity of IL-15 heterodimers. Mice were treated intraperitoneally with PBS (3 mice), with 3 g of E. coli-derived single chain IL-15 (5 mice), with 3 g of human HEK293 cell-derived single chain IL-15 (5 mice), or with an equimolar quantity of IL-15⅐sIL-15R␣ (5 mice, 3 g in single chain IL-15). The frequency of CD8 ϩ T cells (left panels) and NK cells (right panel) expressing the proliferative marker Ki-67 is shown. Individual animals and average are shown for each group. **, p Ͻ 0.01; ***, p Ͻ 0.001; ns, nonsignificant. C, different lots of purified IL-15⅐sIL-15R␣ heterodimer from clone 1.5 induced similar levels of proliferation in splenic CD8 ϩ T cells. Mice were treated intraperitoneally with PBS or with 3 g of IL-15⅐sIL-15R␣ (clone 1.5 lot A) or 3 g of IL-15⅐sIL-15R␣ (clone 1.5 lot B). Isolated splenocytes were evaluated on day 3 by flow cytometry for the presence of the proliferation marker Ki-67. Percentage of Ki-67 ϩ cells within the CD8 ϩ T cell population is shown. One representative mouse per group (3 mice/group) is shown.