Paracrine and Transpresentation Functions of IL-15 Are Mediated by Diverse Splice Versions of IL-15Rα in Human Monocytes and Dendritic Cells*

Background: IL-15 can either be transpresented by IL-15Rα or be secreted. Results: New N- and C-terminal splice versions of human IL-15Rα determine whether IL-15 is secreted or stays bound to the cell membrane. Conclusion: IL-15Rα isoforms determine the mode of action of IL-15. Significance: IL-15Rα isoforms may modify immune response outcomes in humans. Species-specific differences of post-translational modifications suggested the existence of human IL-15Rα isoforms. We identified eight new isoforms that are predicted to modify the intracellular C termini of IL-15Rα, and another N-terminal exon “Ex2A” that was consistently present in all but one of the C-terminal isoforms. Ex2A encodes a 49-amino acid domain that allowed the transfer of IL-15/IL-15Rα complex to the cell surface but prevented its cleavage from cell membranes and its secretion thus facilitating the transpresentation of IL-15 as part of the immunological synapse. The Ex2A domain also affected the O-glycosylation of IL-15Rα that explained the species-specific differences. The Ex2A domain appeared to be removed from major IL-15Rα species during protein maturation, but both Ex2A and IL-15Rα appeared on the surface of monocytic cells upon activation. The membrane-associated form of the only C-terminal isoform that lacked Ex2A (IC3) was retained inside the cell, but soluble IL-15/IL-15Rα complexes were readily released from cells that expressed IL-15/IL-15Rα-IC3 thus limiting this IL-15/IL-15Rα isoform to act as a secreted molecule. These data suggest that splice versions of IL-15Rα determine the range of IL-15 activities.

IL-15 is a common gamma chain cytokine with a well-defined set of target cell populations, mainly NK cells and CD8 T cell subsets (1,2). The cytokine is necessary for their survival and proliferation and contributes to their activation.
IL-15 and IL-15R␣ appear to form a dimeric cytokine. IL-15 and IL-15R␣ are strictly co-induced on monocytic cells (3)(4)(5). Strong inductions are achieved with a combination of an interferon and either TLR-ligands or CD40 ligand, and the tran-scriptions of both IL-15 and IL-15R␣ depend on interferonresponse factors and on NF-B recognition sites (6 -8).
Adoptive transfer experiments in mice showed a necessity of both IL-15 and IL-15R␣ expression in the same cells for the survival of target cells (3,5). The expression of IL-15R␣ on both CD11b ϩ and CD11c ϩ cells contribute to the survival of IL-15dependent NK and CD8 cells (9).
IL-15/IL-15R␣ complexes function via several distinct modes: In the well-documented "transpresentation" mode, the dimeric cytokine stays anchored on the surface of monocytic cells forcing target cells to accept co-stimulatory signals in addition to the cytokine stimulation (10). Alternatively, IL-15/ IL-15R␣ complexes can be cleaved from the surface of producing cells (11) to act as a soluble heterodimer on neighboring cells in a paracrine fashion. IL-15 also exerts effects on cells that produce the cytokine themselves in that it increases the survival of mature dendritic cells (DCs) 2 and regulates protease activities in mastocytes (4,12). These effects may be accomplished by autocrine activities of IL-15, and similar defects that we observed in DCs that lacked either the cytokine or part of the receptor appear to support this scenario (4). Alternatively, the binding of signaling molecules to the intracellular portion of IL-15R␣ itself may also suggest reverse signaling events (13). The precise basis for the various modes of IL-15 activity is currently unknown.
Directing immune responses to specific targets such as cancerous cells by treatments with IL-15 may represent a valuable approach (14). We and others have shown an additive effect of soluble IL-15R␣ on IL-15 activity both in vivo and in vitro (15)(16)(17). IL-15R␣ had accessory activity on the amplitudes of CD8 responses as well as during the response against NK cell-sensitive tumors in vivo. The IL-15R␣ effect may be partially caused by an increased half-life of IL-15 when bound to IL-15R␣, but direct activity-modulating effects also appear to be mediated by IL-15R␣. A potential clinical use of IL-15R␣ accessory to IL-15 appears to necessitate knowledge about the exact nature of this protein under physiologic conditions.
Here we explore the molecular basis for the various modes of IL-15 action. We report that a number of previously unrecognized isoforms of human IL-15R␣ exists that direct the cytokine to paracrine versus transpresentation activities by facilitating secreted or membrane-bound forms of the cytokine.

EXPERIMENTAL PROCEDURES
Mice and Human Cells-C57BL/6 wildtype mice were purchased from the Jackson Laboratory and were housed and treated within published guidelines of humane animal care, and all procedures were approved and performed according to National Cancer Institute Animal Care and Use Committeeapproved protocols for animal research. The use of human cells was approved by The Institutional Review Board of the National Cancer Institute, NIH, and all samples were obtained with informed consent.
Cell Culture-293HEK and U2OS cells (ATCC) were propagated in DMEM supplemented with 10% FBS. For transfections, cells were trypsinized, washed, and seeded into 6-well plates. Transfections were done 18 h later using a total amount of 3 g of plasmid DNA per well containing various amounts of specific plasmid DNA and empty pcDNA3.1 (Invitrogen) and 4 l of Lipofectamine 2000 (Invitrogen), resulting in greater than 90% transfection efficiency for 293 cells. The culture medium was exchanged 8 h later with RPMI containing 8% human AB serum (Cellgro) to generate supernatants for human NK cell proliferation, or otherwise with DMEM-10% FBS. Supernatants were collected 48 h later; otherwise cells were used 24 h after transfection. Where indicated, the O-glycosylation inhibitor Benzyl-GalNAc was added into the culture medium to 4 mM 1 h after transfections. Stable U2OS clones were generated by the presence of 400 g/ml G418 (Invitrogen) and limited dilution.
To generate membrane fractions containing human CD40 ligand, 293HEK cells that had been transfected with a CD40 ligand-encoding plasmid 24 h prior were washed, resuspended in 210 mM D-mannitol, 70 mM sucrose, 1 mM EDTA, 10 mM Hepes, pH 7.2, mechanically disrupted by 30 passages through 27-gauge needles and centrifuged at 14,000 ϫ g. The resulting pellet from ϳ5 ϫ 10 6 transfected cells was used to mature ϳ5 ϫ 10 6 human DCs. As a control we used membrane fractions prepared from mock-transfected 293HEK cells that did not affect DC maturation or their expression of IL-15 or IL-15R␣ (not shown).
Human DCs were derived from elutriated monocytes (Blood Bank, NIH, Bethesda) and cultured for 5 days in RPMI supplemented with 8% human serum and 50 ng/ml recombinant human GM-CSF and IL-4 (Peprotech). Maturations were done by overnight exposures to combinations of 20 ng/ml human IFN-␥ (Peprotech) and 100 ng/ml LPS (Escherichia coli 055:B5, Sigma), the membrane fraction of CD40 ligand-transfected 293HEK cells, 10 g/ml poly I:C (Sigma), exponentially growing listeria monocytogenes bacteria at a multiplicity of infection of 0.5 or vaccinia virus (Western Reserve, multiplicity of infection of 10). Maturations were verified by FACS analyses of CD80, CD86, and MHC class II expression (not shown). Murine bone marrow-derived DCs were prepared by growing C57BL/6 bone marrows in RPMI supplemented with 10% FBS and 40 ng/ml recombinant murine GM-CSF (Peprotech) for 5 days. To generate murine monocyte-derived DCs, murine PBMCs were isolated via Ficoll-centrifugation of blood samples and allowed to adhere to tissue culture plates for 2 h. Plates were rinsed repeatedly, and the remaining adhering cells were incubated for 5 days in RPMI containing 10% FBS, 40 ng/ml murine GM-CSF, and 50 ng/ml murine IL-4 (Peprotech). DCs were matured overnight in RPMI containing 10% FBS, 50 ng/ml LPS (E. coli 055:B5, Sigma), and 20 ng/ml murine IFN-␥ (Peprotech).
To derive human NK cells, blood samples from healthy donors were depleted of erythrocytes via Ficoll-centrifugation and sorted with the negative NK cell isolation kit (Miltenyi). NK cells were expanded in culture in RPMI containing 8% human AB serum and 1 nM human IL-2 (Peprotech) for 7 days prior to use in proliferation assays. PBMCs were also used directly after labeling with CFSE (500 nM, 10 min at 37°C, Invitrogen).
Generation of Antibodies-The entire Ex2A domain was expressed as a GST fusion protein (pGEX-2T, Pharmacia) in E. coli, isolated and used to immunize rabbits and mice. Rabbit antiserum was produced by Prosci, Poway, CA. Mouse mAbs were generated using standard procedures (18). The screening of mAbs was performed by ELISA using supernatants from 293HEK cells that had been transfected with plasmids encoding IL-15 and a fusion construct of the extracellular portion of human IL-15R␣ containing Ex2A and the two C-terminal constant regions of human IgG1. Ascites were produced from positive clones by Harlan Laboratories, Indianapolis, IN. For use in cytometry, mAbs were biotinylated using the EZ-Link Sulfo-NHS-LC-Biotinylation Kit (Pierce). Point mutation analyses showed binding of ␣Ex2A1/2/3 mAbs to be affected by amino acids C27, C36, and H41 of the mature Ex2A domain in cytometry analyses, while binding of ␣Ex2A4 in immunoblots involved amino acids E15, F17, H19, and E20 (not shown).
Proliferation Assay-Human NK cells that had been cultured for 7 days in 1 nM IL-2 were washed three times and plated into 96-well plates at 5 ϫ 10 4 cells per well. Cells were incubated for 48 h with various amounts of supernatants that had been collected from transfected 293HEK cells. [ 3 H]Thymidine (1 mCi, Perkin-Elmer) was present during the final 12 h of the assay.
RT-PCR and Sequencing-Total RNA was prepared with Trizol (Invitrogen) according to the manufacturer's instructions. 1 g of RNA was subsequently transcribed into cDNA using an oligo-dT primer and 200 units of Superscript III reverse transcriptase (Invitrogen). Primers used for PCR are given in the supplemental Table S1. Parameters for all PCR amplifications were 3 min at 94°C, followed by 35 cycles of 95°C for 15 s, 60°C for 15 s, and 72°C for 90 s, with a final extension of 72°C for 5 min. To amplify the entire coding region of human IL-15R␣, a nested step was necessary in which 1 l of the first PCR was used as template for the nested step. PCR reactions were sizefractionated, bands were isolated with GenElute Spin Columns (Sigma), and sequenced directly. The new IL-15R␣ sequences that are described in this report have been scanned against the database, and no sequence with significant relatedness to the new sequences other than genomic sequence has been identified. The nucleotide sequences reported in this paper have been submitted to GenBank TM with accession numbers HQ401283, JX987304, JX987305, JX987306, JX987307, JX987308, JX987309, JX987310, JX987311.
Plasmid Constructs-Coding sequences for human IL-15, IL-15R␣, CD25, and CD40 ligand were amplified by RT-PCR, cloned into pCR2.1-TOPO (Invitrogen), verified by sequencing and recloned into the mammalian expression vector pcDNA3.1. The generation of chimeric expression constructs containing the extracellular portion of IL-15R␣ and the two C-terminal constant regions of human IgG1 has been described previously (15). Oligonucleotides that were used to generate other chimeric, deletion, and tagged constructs are given in the supplemental Table S1, and cloning steps should be self-explanatory. Point mutations were introduced using the QuikChange II XL Kit (Agilent).
Immunoprecipitation and Western Blot-All antibodies used for immunoprecipitation were covalently bound to protein G-agarose (Pierce). As controls for goat anti-human and goat anti-mouse IL-15R␣ antibodies (AF247 and AF551 from R&D Systems) we used goat IgG (Sigma), as control for mouse antihuman IL-15R␣ (7A4), mouse anti-human IL-15 (M111) and anti-Ex2A we used mouse IgG (Pierce). Cell lysates were used from ϳ5 ϫ 10 6 DCs for the detection of endogenous proteins, and from 10 6 transfected 293HEK cells for the detection of overexpressed proteins per sample. Lysis was done in 150 mM NaCl, 1% Triton-X-100, 25 mM Hepes, pH 7.3, 10% glycerol, supplemented with protease inhibitors (Roche). Cell lysates were cleared by centrifugation and incubated with agarosebound immunoprecipitating antibodies for 2 h at 4°C with constant rotation, washed four times with lysis buffer and subjected to SDS-PAGE (Tris-Glycine Gels, Invitrogen) that was followed by immunoblotting with antibodies against human IL-15R␣ (SC12378 from Santa Cruz Biotechnology and SAB1100259 and 1100260 from Sigma), murine IL-15R␣ (AF551, R&D Systems), CD25 (SC665, Santa Cruz Biotechnology), human IL-15 (SC7889 from Santa Cruz Biotechnology), the Flag tag (M2 from Sigma), or the Ex2A domain (described above).
ELISA-Nunc Maxisorp plates were coated with antibody (5 g/ml anti-IL15 M111 or goat anti-human IgG from Pierce in PBS at 4°C overnight) and blocked with 1% w/v bovine serum albumin at 37°C for 1 h. Samples were added and incubated at 37°C for 1 h that was followed by the second antibody (antihuman IL-15R␣ AF247, R&D Systems, anti-IL-15 M111, or anti-Ex2A1, all biotinylated and at 1 g/ml), alkaline phosphatase-conjugated streptavidin (Pierce, 1 g/ml, 37°C for 1 h) and substrate.
Microscopy-For immunofluorescence microscopy, U2OS clones were grown on coverslips, fixed with methanol (Ϫ20°C for 5 min), blocked with 1% BSA and 5% goat serum (37°C for 1 h), incubated with primary antibodies (rabbit anti-GM130 or anti-PD1, Abcam, 37°C for 1 h), incubated with secondary antibody (goat anti-rabbit-Cy3, Invitrogen, 37°C for 1 h), incubated with DAPI (100 nM, room temperature, 20 min) and mounted. Confocal fluorescence images were acquired using a Zeiss LSM510 META laser-scanning microscope equipped with a 63ϫ Plan-aprochromat (N.A. 1.4) oil immersion objective lens, with an optical slice thickness of 0.9 m, a x-y pixel sampling of 0.07 m, and a z-step size of 0.35 m. The resultant z-stack image series were background subtracted and an intensity threshold mask applied representative of the respective cell labels used to calculate colocalization coefficients through the volume of the sample using the colocalization analysis module of Imaris software (v 7.0).

RESULTS
Cloning of IL-15R␣ Isoforms-Activated monocytes and DCs are major sources of IL-15. As previously reported (4,19,20) and as shown in Fig. 1A, endogenous IL-15R␣ from activated human DCs differs from endogenous murine IL-15R␣ and from human IL-15R␣ generated by overexpression in that the majority migrates at a lower molecular weight. We had previously shown that O-and N-glycosylation events modify IL-15R␣ resulting in migration at a higher molecular weight (19). The glycosylation discrepancies as shown in Fig. 1A could have been caused by at least two different mechanisms: (A) human DCs could possess a specific glycosylation-inhibiting mechanism, or (B) sequence of human IL-15R␣ may not be fully known, and additional domains may affect glycosylation patterns. We investigated the latter possibility. We generated cDNAs from human DCs that had been matured with combinations of IFN-␥ and TLR ligands or CD40 ligand and applied several PCR-based cloning strategies (supplemental Fig. S1). To search for additional internal exons of fully spliced message, a first PCR step used oligonucleotides that annealed inside the first and last exon of IL-15R␣ mRNA that was followed by a nested step to amplify a sequence that comprised one or two exon borders. We observed that all IL-15R␣ mRNAs from variously treated monocytes and DCs contained an additional 153-bp sequence inserted between the previously dedicated exons 1 and 2 that we will refer to as "exon 2A" or "Ex2A" (Fig. 1B, the predicted amino acid sequence is shown below the panels). It is important to note that amplification of Ex2A largely depended on the first PCR step that targeted full-length and fully spliced IL-15R␣ mRNA in that most PCR products lacked Ex2A when the exon 1/2 border was amplified from cDNA directly. In contrast to human IL-15R␣, GenBank TM searches failed to reveal any sequence predicted to encode a homologous domain in the murine intron 1. In addition, we were unable to detect a corresponding exon by PCR in cDNAs that had been derived from mature murine bone marrow-or monocyte-derived DCs. These data suggest that human DC-derived IL-15R␣ mRNA contains an additional exon not found in mice.
A second cloning series targeted alternative C-terminal isoforms of IL-15R␣. Using 3Ј-RACE on cDNA samples as above we identified seven new splice versions that modified the 3Ј-end (Fig. 1C) in addition to the previously known IL-15R␣ message that we refer to as "wildtype" in this manuscript. Alignments with genomic sequences revealed that all exons were located within ϳ45 kbp of exon 1 (Fig. 1E). One isoform (EM2) is predicted to encode an additional extracellular domain and an alternative membrane domain. All other isoforms retained the wildtype membrane domain but altered the intracellular amino acid sequence. The two isoforms IC2 and IC6 resulted from alternative splice locations of the same C-terminal exon similar to what has been described for wildtype IL-15R␣ (21).
We also investigated whether the inclusion of Ex2A is linked to C-terminal isoforms. We performed initial PCR steps with oligonucleotides annealing in the first and last exons of the respective isoforms that was followed by a nested step to amplify the exon 1/2 border. Fig. 1D shows the consistent presence of Ex2A in all C-terminal isoforms with the exception of IC3 for which species with or without Ex2A were detected. Together these data show the existence of additional coding exons in the human IL-15R␣ gene.
IL-15/IL-15R␣ Complex with the Ex2A Domain Is Functional-Ex2A is predicted to encode an amino acid domain immediately N-terminal of the IL-15-binding sushi domain. We investigated whether this constellation would render the resulting IL-15/IL-15R␣ complex inactive. We generated chimeric constructs in which the membrane and intracellular domains of IL-15R␣ were replaced by immunoglobulin constant domains of human IgG1 thus inducing their secretion (15). We analyzed supernatants from transfected 293 cells by ELISA and observed that antibodies to both IL-15 and IL-15R␣ recognized the IL-15/IL-15R␣-IgFc complex that had been captured by anti-Fc antibodies ( Fig. 2A) indicating that binding of IL-15 to IL-15R␣ Analyses were done by immunoprecipitations/immunoblots. While most murine IL-15R␣ from DCs and human IL-15R␣ from transfected 293 cells migrated at ϳ65 kDa with minor species at 35 kDa, the majority of IL-15R␣ from human DCs was detected at 35 kDa with a minor species at 65 kDa. Human IL-15R␣ from 293 cells and DCs were detected with the same antibody combinations, and two alternative antibodies that were used for both immunoprecipitations and immunoblots gave similar results. B, cDNAs were derived from variably activated human monocytes and DCs and subjected to RT-PCR. A first step with oligonucleotides annealing in the first and last exons of IL-15R␣ was followed by a nested step that used oligonucleotides in exons 1 and 3. All resulting PCR products that contained the sushi domainencoding and IL-15-binding exon 2 also contained an additional 153-bp sequence between exons 1 and 2 that we termed "exon 2A". The predicted amino acid sequence is shown below the panels. C, further analyses revealed the existence of alternative IL-15R␣ splice products at its 3Ј-end in cDNAs that had been derived from activated DCs. Specific PCR products that contained IL-15R␣ sequences are indicated by asterisks left of the bands, double-asterisks denote wild-type IL-15R␣. D, targeting the various C-terminal isoforms in the first PCR step revealed IC3 to be the only isoform that may lack the Ex2A sequence. All PCR products were verified by direct sequencing. E, genomic organization of newly identified exons of human IL-15R␣ on chromosome 10. New exons are shown in black and are named below the horizontal bar. Previously described exons are gray and are numbered above the horizontal bar. Lines below the bar depict splicing (black for new and gray for previously known). Exon structures are depicted above the bar with coding regions in black and noncoding regions in white. EM2 encodes an alternative membrane domain that is shown in gray. NOVEMBER 23, 2012 • VOLUME 287 • NUMBER 48 remained unaffected by Ex2A although its presence appeared to decrease the levels of secretion. The same supernatants also induced the proliferation of native human NK cells (Fig. 2B) with differences that corresponded to the levels of secretion. Therefore, there appeared to be no interference by Ex2A with the binding or activity of IL-15.

IL-15R␣ Isoforms
Ex2A Affects the Glycosylation of IL-15R␣-We addressed the initial question of the causes of glycosylation differences. When Ex2A domain was present, IL-15R␣ indeed migrated at a lower molecular weight on SDS-PAGE (Fig. 3A, left panels). To investigate causality we expressed the Ex2A domain at the N-terminal end of mature CD25 and observed similar mobility changes (Fig. 3A, right panels). Pulse-chase experiments (Fig.  3B) showed that for IL-15R␣ lacking Ex2A, the higher molecular weight species started to appear after a 30-min chase and remained stable while unmodified IL-15R␣ disappeared. The presence of Ex2A appeared to prevent such a modification to result in unstable protein only. These data suggest that the observed protein modification confers stability. Ex2A appeared to be acting in cis only since co-expressions of Ex2A-containing IL-15R␣ or CD25 failed to affect migration patterns of IL-15R␣ or CD25 lacking Ex2A (supplemental Fig. S2). We failed to identify amino acid motifs within the Ex2A domain that caused the changes in the migratory patterns (supplemental Fig. S3) suggesting that Ex2A changes the conformation of IL-15R␣. Together these data show that the Ex2A domain affects glycosylation of IL-15R␣ in cis.
IL-15R␣ undergoes multiple glycosylation events (19). We studied the nature of the protein modifications that were asso-ciated with the presence of Ex2A. The early N-glycosylation appeared to be unaffected since both digestions with EndoH as well as amino acid substitutions of the targeted asparagines shifted IL-15R␣ regardless of the presence of Ex2A (Fig. 3, C  and D). In contrast, O-sialoglycoprotease digestion was prevented by the presence of Ex2A (Fig. 3, C and D). This suggests that Ex2A affected O-glycosylation since O-sialoglycoprotease requires this protein modification for its activity (22). O-Glycosylation modifications of IL-15R␣ may be atypical since digestions with neuraminidase/O-glycosidase failed to affect the high molecular weight-species of IL-15R␣ (Fig. 3C). The existence of such atypical modifications may further be supported by the inability of the O-glycosylation inhibitor Benzyl-GalNAc to affect mobility (see below). Digestions with chondroitinase, heparinases, or combinations thereof also failed to affect IL-15R␣ mobility (not shown). In conclusion, the presence of the Ex2A domain appears to inhibit the O-glycosylation of IL-15R␣ in cis.
The Ex2A Domain Prevents Release of IL-15/IL-15R␣ into the Soluble Phase-We investigated the functional consequences of Ex2A presence. As mentioned above, the IL-15/IL-15R␣ complex acts as either a membrane-bound cytokine or after its release into the soluble phase. To study the former we generated IL-15R␣ Ϯ Ex2A with C-terminal GFP tags that were expressed together with IL-15 in 293 cells and analyzed by cytometry. Fig. 4A shows that for IL-15R␣ containing Ex2A, increases of intracellular GFP detection were accompanied by smaller increases of surface IL-15R␣ suggesting that Ex2A inhibited but did not prevent the surface export of IL-15R␣. Ex2A also slightly inhibited the transport of IL-15R␣ from the endoplasmatic reticulum to the Golgi apparatus (supplemental Fig. S4). When concentrations of IL-15/IL-15R␣ heterodimer in supernatants were determined by ELISA using plate-bound anti-IL-15 and soluble anti-IL-15R␣ antibodies as a measure of cytokine release into the soluble phase (Fig. 4B), Ex2A completely abolished the ability of 293 cells to release the IL-15/IL-15R␣ complex into the soluble phase. It therefore appears that Ex2A inhibits surface export and prevents release of the soluble IL-15/IL-15R␣ heterodimer.
As described above, Ex2A affected O-glycosylation and release of IL-15R␣. We studied whether both processes are linked. For unrelated reasons we had generated a number of IL-15R␣ mutants where serines had been replaced by alanines and that did not include Ex2A. The analysis of one of these clones (S158/160A) by Western blot revealed a lack of O-glycosylation (Fig. 4C). Contrary to the presence of Ex2A however, these amino acid replacements did not decrease the IL-15/IL-15R␣ heterodimer concentration in supernatants suggesting that the effects of Ex2A on O-glycosylation and heterodimer release are not linked.
We also studied the contributions of O-and N-glycosylation to IL-15/IL-15R␣ heterodimer release. Fig. 4D (graph) shows that the presence of the O-glycosylation inhibitor Benzyl-Gal-NAc increased the release of the IL-15/IL-15R␣ complex that lacked Ex2A without an obvious effect on O-glycosylation of IL-15R␣ itself as analyzed by the migration on Western blots (Fig. 4D, panel) suggesting that O-glycosylation of alternative proteins affected the IL-15/IL-15R␣ complex release. In con- trast, N-glycosylation of IL-15R␣ itself appeared to support the secretion since substituting the target asparagine (N107Q) in IL-15R␣ decreased IL-15/IL-15R␣ complex concentration in the supernatant (Fig. 4D). Together these data reveal a complex picture in which independent effects of N-and O-glycosylation as well as the presence of the Ex2A domain determine IL-15/ IL-15R␣ heterodimer release.
Mature IL-15R␣ Lacks the Ex2A Domain-To detect and study the Ex2A-encoded domain, we generated several mAbs in mice that could be used for cytometry and immunoblotting (supplemental Fig. S5 and not shown). FACS analyses revealed the induction of Ex2A on the surface of DCs upon activation with LPS/IFN-␥ (Fig. 5A). Ex2A was co-expressed together with IL-15 and IL-15R␣ on monocytic (CD14 ϩ , MHC class II ϩ ) but not on lymphocytic (CD3 ϩ , CD56 ϩ , CD20 ϩ ) cells among PBMCs from healthy donors (Fig. 5B). Despite co-expression, immunoprecipitation/immunoblot analyses revealed that the major species of IL-15R␣ in activated DCs did not comprise the Ex2A domain (Fig. 6). Analyses of monocytes and DCs that had been activated under various conditions also revealed that Ex2A was consistently expressed on cells that were positive for IL-15 (supplemental Fig. S6) or IL-15R␣ (not shown). These data suggest that the Ex2A domain is removed from the major IL-15R␣ species during protein maturation and that both proteins appear on the surface separately.
Alternative C-terminal Isoforms of IL-15R␣-We investigated whether the newly identified C-terminal isoforms of human IL-15R␣ similarly affected secretion and/or surface expression. Fig. 7A shows that the expression of two alternative isoforms was associated with an increased secretion of IL-15/ IL-15R␣ heterodimer into the supernatants when compared with wildtype. When surface expressions were analyzed (Fig.  7B), most isoforms showed behaviors similar to wildtype IL-15R␣ in that the surface expressions of IL-15 and IL-15R␣ increased proportionally. However, the C-terminal isoforms with the highest secretion levels differed: Levels of IL-15R␣ increased less when compared with IL-15 for the isoforms EM2 and IC5 that we interpret as conformational changes that were induced by the alternative C termini and that affected the detection by the anti-IL-15R␣ antibody. In addition, the iso-form with the highest secretion level (IC3) was nearly absent from the cell surface. These data suggest that C-terminal isoforms of IL-15R␣ determine secretion and surface expression.
We further investigated IC3. We generated and analyzed three deletion mutants (supplemental Fig. S7). Deleting parts of the N-terminal or C-terminal ends of the intracellular portion of IC3 was accompanied by both higher surface expression levels and by lower secretion levels suggesting that secretion and intracellular retention were linked. Membrane-bound IC3-IL-15R␣ appeared to be prevented from entering the Golgi apparatus as most protein co-localized with the endoplasmatic reticulum in microscopic analyses (supplemental Fig. S8). IC3-IL-15R␣ retained IL-15 inside the cell (Fig. 8A) since increased levels of the C-terminal and therefore intracellular GFP tag were accompanied by surface appearance of IL-15 for wildtype but not for IC3-IL-15R␣. Replacing the intracellular portion of CD25 with IC3 also caused a reduced surface appearance of CD25 suggesting causality between IC3 and intracellular retention (Fig. 8B). We also analyzed the cell-associated and secreted Increased GFP expression was accompanied by a smaller increase in the expression of extracellular IL-15R␣ when Ex2A was present suggesting that Ex2A inhibited the surface export of IL-15/IL-15R␣ heterodimer. B, no IL-15/IL-15R␣ heterodimer was detected by ELISA in the supernatant when Ex2A was present. The panel on the right shows the levels of cell-associated IL-15R␣ for this experiment. C, wildtype IL-15R␣ without Ex2A as well as a construct with two amino acid substitutions (S158/160A) that lacked IL-15R␣ modified by O-glycosylation similar to when Ex2A is present were coexpressed with IL-15 and analyzed by ELISA (graph) and Western blotting (panel). Secreted IL-15/IL-15R␣ heterodimers were detected for both conditions suggesting that Ex2A-induced changes in the O-glycosylation of IL-15R␣ and its secretion are not linked. D, both O-and N-glycosylation affect IL-15/IL-15R␣ heterodimer secretion. The presence of the O-glycosylation inhibitor Benzyl-GalNAc increased levels of IL-15/IL-15R␣ heterodimer secretion (graph) despite a lack of discernible effects on IL-15R␣ migration on Western blots suggesting that an O-glycosylation of alternative proteins is involved in determining secretion levels. Expressions of an asparagine mutant (N107Q) that prevented N-glycosylation decreased IL-15/IL-15R␣ heterodimer secretion suggesting a positive effect of IL-15R␣ N-glycosylation on its secretion. forms of wildtype and IC3-IL-15R␣ by immunoprecipitation/ immunoblot and made several observations (Fig. 8C): 1) The intracellular presence of IC3 prevented O-glycosylation similar to the presence of the Ex2A domain (see above). 2) The molecular weight of both isoforms differed for both surface and secreted proteins. As expected, the intracellular portion of IL-15R␣ was present only in cell-associated but not in secreted protein (not shown). The molecular weight difference of the secreted protein suggests alternative cleavage sites for wildtype and IC3-IL-15R␣.
3) The molecular weight of secreted IC3-IL-15R␣ showed an unexpected increase when compared with cell-associated protein. 4) Cleavage of both IL-15R␣ isoforms appeared to occur inside or close to the membrane domain since the secreted proteins were recognized by antibodies directed against the 15 amino acids immediately N-terminal of the membrane domain. A cleavage inside the membrane domain is further supported by the finding that IL-15/IL-15R␣ heterodimer secretions were detected for all natural IL-15R␣ variants that lacked various parts of the extracellular domain encoded by exons three through five (supplemental Fig. S9). We also determined the proliferation-inducing capacity of secreted IL-15/IC3-IL-15R␣ complex. Equimolar amounts of IC3-IL-15R␣/IL-15 complex as determined by ELISA appeared to be superior to wildtype IL-15R␣/IL-15 in inducing the proliferation of CD8 and NK cells among PBMCs (Fig. 8D). Together these data suggest that the intracellular isoform IC3 prevented surface expression of the IL-15R␣/IL-15 complex but caused high levels of secretion to induce the proliferation of target cells in a paracrine fashion.

DISCUSSION
The pro-inflammatory cytokine IL-15 has several modes of action. Paracrine activities involving secreted IL-15/IL-15R␣ complex assure the survival of responding NK and CD8 T cells without the requirement of cell-to-cell interactions. Cell surface IL-15 transpresentation by IL-15R␣ gives the potential for additional signaling from the IL-15-expressing to the IL-15-dependent cells. It has been shown that IL-15 transpresentation rather than soluble IL-15/IL-15R␣ heterodimer serves as a coactivation signal for NK cells (23,24), and similar mechanisms probably also apply to CD8 T cells where IL-15 signaling has distinct effects during various immune response phases (25,26). Here we describe a number of previously unrecognized isoforms of human IL-15R␣ that determine the cytokine mode of action. We observed that wildtype IL-15R␣ appears to act exclusively via transpresentation due to the new N-terminal Ex2A domain that prevented its cleavage and secretion into the soluble phase while another isoform, IC3 caused the appearance of soluble but not cell membrane-anchored IL-15/IL-15R␣ complex due to the retention of membrane-bound IL-15R␣ in the endoplasmatic reticulum.
An interesting finding is the apparent absence of similar IL-15R␣-regulating mechanisms in non-primates. GenBank TM searches identified sequence homologous to Ex2A in chimpanzees only but neither murine, rat, nor bovine genomes appear to contain such an exon. In addition, cDNAs from various mouse monocyte and DC preparations lacked any additional sequence between the first two exons when analyzed by RT-PCR (not FIGURE 6. Ex2A is absent from the major species of mature IL-15R␣ in DCs. Human DCs with or without LPS/IFN-␥-or CD40L/IFN-␥-induced maturations as well as Ex2A-IL-15R␣-transfected 293 cells were subjected to immunoprecipitation with antibodies directed against extracellular IL-15R␣ that was followed by immunoblotting with antibodies against the intracellular portion of wildtype IL-15R␣ or against Ex2A. The major 35-kDa IL-15R␣ species that was induced by DC maturation was detected with immunoblotting antibodies against IL-15R␣ (upper) but not against Ex2A (lower) suggesting that both epitopes were not part of the same protein. A 42-kDa protein that was present in immature DCs and after CD40L/IFN-␥ treatment appeared to contain both Ex2A and extracellular but not intracellular IL-15R␣ sequence (lower panel) since an antibody against extracellular IL-15R␣ was used to precipitate and anti-Ex2A was used to blot. The exact nature of this protein remains unclear. shown). This isoform-dependent IL-15/IL-15R␣ regulation may therefore be a recent event in phylogenesis, and organisms such as mice may employ alternative mechanisms to achieve similar outcomes.
Our data also include the identification of a number of C-terminal isoforms in addition to IC3 that change the intracellular amino acid sequence of IL-15R␣. Most of these isoforms functioned in a fashion similar to wildtype IL-15R␣ in that their cleavage from cell membranes was prevented by the presence of the Ex2A domain, but the surface export of membrane-anchored IL-15/IL-15R␣ complex remained intact. This existence of alternative intracellular domains suggests reverse signaling in which various intracellular domains of IL-15R␣ affect signaling events inside the IL-15/IL-15R␣-producing cells. Effects of IL-15 absence on IL-15-producing cells have been reported in the mouse for DCs and mast cells (4,12). In addition, the binding of signaling molecules to IL-15R␣ has been reported (13). Therefore, it appears possible that the expressions of various IL-15R␣ isoforms affect cell differentiation states of IL-15-producing cells.
The new C-terminal isoforms were detected by RT-PCR at levels similar to the wildtype isoform of IL-15R␣ (Fig. 1C). Most of these new isoforms encode short intracellular amino acid motifs. We attempted to generate rabbit antisera directed against these intracellular motifs. We were successful to generate one such antiserum that detected IC4 in immature DCs and after activation with CD40L/IFN-␥ (supplemental Fig. S10). We observed another isoform in Fig. 6 that was immunoprecipitated with antibody against extracellular IL-15R␣ and was recognized by an immunoblotting antibody against Ex2A but not against the intracellular portion of wildtype IL-15R␣. This band may represent the version of IC3 that contains Ex2A (Fig.  1D), and the prevention of membrane-anchored IC3 from exiting the endoplasmatic reticulum may also prevent the cleavage of Ex2A from IL-15R␣. The generation of additional antibodies is necessary to study various IL-15R␣ isoforms on the protein level.
Our data appear to show that IL-15R␣ and possibly CD25 proteins are modified by an unusual O-glycosylation. The proteolytic activity of O-sialoglycoprotease indicates the presence of O-linked oligosaccharides whose attachment appeared to be prevented by both Ex2A and IC3 (Fig. 3, C and D). IL-15R␣ contains a typical mucin-type domain (27) encoded by exons three and four that are often modified by O-glycosylation. Expressions of IL-15R␣ in 293 cells usually generate two species that do or do not contain the protein modification. The distances between these bands in analyses of natural splicing products that lack parts of exons three and four (supplemental Fig.  S9) support the notion of multiple attachment sites in that region that is also typical for O-glycosylation. On the other hand, digestions with O-glycosidase/neuraminidase did not remove sugar moieties from IL-15R␣ (Fig. 3C). O-Glycosidase is specific for cleaving only one O-linked disaccharide, Gal␤1-3GalNAc␣-Ser/Thr (22). In addition, the presence of an inhibitor of mucin-like O-glycosylation failed to affect the mobility of IL-15R␣ while affecting its secretion (Fig. 4). The most likely explanation for these data is that IL-15R␣ is modified by an atypical O-glycosylation that uses sugars other than GalNAc for its core attachment. Such atypical O-glycosylation events have been described for other proteins (28 -30).
Glycosylation patterns in the immune system are highly regulated (31)(32)(33)(34)(35). Such patterns are altered during development, trafficking, activation, and apoptosis, and changes of glycosylation are linked to disease mechanisms such as auto-immunity. The function of a wide variety of receptors is affected by glycosylation (31)(32)(33)(34)(35). Changes of glycosylation may but do not have to affect receptor stability, surface expression, ligand binding, host-pathogen interactions, lectin binding, allosteric conformation, and intracellular signaling. Glycosylation steps also appear to affect IL-15 pathways. Our data show that N-glycosylation of IL-15R␣ supports its secretion while O-glycosylations of other proteins are inhibitory.
In summary, our data reveal the existence of multiple IL-15R␣ isoforms in humans that appear to determine the mode of action of the pro-inflammatory cytokine IL-15. The isoforms containing the Ex2A domain prevent the release of IL-15/IL-15R␣ heterodimers from cell membranes thereby facilitating its transpresentation function. The IC3 isoform that does not contain the Ex2A domain functions as a soluble secreted cytokine.