Soluble Interleukin (IL)-15Rα Is Generated by Alternative Splicing or Proteolytic Cleavage and Forms Functional Complexes with IL-15*

Interleukin 15 (IL-15) is a pleiotropic cytokine that is hardly detectable in biological fluids. Here, we show that IL-15 forms functional heterocomplexes with soluble high affinity IL-15 receptor α (IL-15Rα) chain in mouse serum and cell-conditioned medium, which prevents IL-15 detection by ELISA. We also demonstrate that two soluble IL-15Rα (sIL-15Rα) sushi domain isoforms are generated through a novel alternative splicing mechanism within the IL-15Rα gene. These isoforms potentiate IL-15 action by promoting the IL-15-mediated proliferation of the CTLL cell line and interferon γ production by murine NK cells, which suggests a role in IL-15 transpresentation. Conversely, a full-length sIL-15Rα ectodomain released by tumor necrosis factor-α-converting enzyme (TACE)-dependent proteolysis inhibits IL-15 activity. Thus, a dual mechanism of sIL-15Rα generation exists in mice, giving rise to polypeptides with distinct properties, which regulate IL-15 function.

Interleukin (IL) 2 -15 is a pleiotropic cytokine that belongs to the four-␣-helix bundle cytokine family and displays many immunomodulatory activities, which partially overlap with those of IL-2 (1). Contrary to IL-2, which is produced mainly by activated T cells, IL-15 has a much broader pattern of expression in both immune and nonimmune cell types and tissues (2). The existing similarities in action between IL-15 and IL-2 are explained in part by sharing of IL-2/IL-15R␤ and IL-2R␥/␥c subunits (3). However, each cytokine has its own unique high affinity receptor ␣ chain, which confers ligand specificity (4). Both IL-15R␣ and IL-2R␣ have common structural features that include rather short cytoplasmic domains and Pro/Thrrich extracellular regions containing a conserved protein binding motif (sushi domain, also known as the glycoprotein-I motif or short consensus repeat) responsible for ligand binding (4). In the absence of the IL-2/IL-15R␤ and ␥c subunits, IL-15R␣ binds IL-15 with high affinity (K a ϳ 10 11 M Ϫ1 ), which is in striking contrast to IL-2R␣ that exhibits low affinity for IL-2 (K a ϳ 10 8 M Ϫ1 ) (4). A number of distinct IL-15R␣ isoforms were described in human and mouse as a result of an alternative splicing within the IL-15R␣ gene. These include deletions of exon 2, exon 3, an alternative usage of exon 7 or 7Ј (5,6), and deletions of exon 4, exons 3 and 4, and exons 3, 4, and 5 (7).
Recently, we demonstrated that spontaneous and inducible shedding of a natural soluble IL-15R␣ (sIL-15R␣) depends on activity of tumor necrosis factor-␣-converting enzyme (TACE/ ADAM17) and could be blocked by inhibitors of this metalloproteinase (8). TACE-deficient murine fibroblasts exhibited a significant reduction in both types of IL-15R␣ shedding, whereas reconstitution of TACE restored sIL-15R␣ release. Notwithstanding, secreted forms of diverse membrane-linked proteins may also appear as a result of an alternative splicing mechanism that gives rise to a polypeptide lacking transmembrane and/or other regions (9). In some species, both modes can be operative, as shown for sIL-4R and sIL-6R (9 -11). In addition, mechanism of soluble receptor generation may undergo evolutionary divergence in different species. Production of growth hormone receptor by proteolytic cleavage in humans and rabbits and by alternative splicing in mice and rats provides a classical example of such a phenomenon (12).
In this study, we demonstrate that an alternative splicing mechanism within the IL-15R␣ gene is responsible for the generation of a natural sIL-15R␣ sushi domain. Alternatively spliced or shed sIL-15R␣ isoforms associate with IL-15, promoting or inhibiting IL-15 activity, respectively. The presence of IL-15⅐sIL-15R␣ in mouse serum and cell-conditioned medium implicates these heterocomplexes in IL-15 transpresentation and regulation of the cytokine activity, suggesting an important role in the generation and maintenance of multiple lymphocyte subsets.
Cell Culture, Stimulation, and Transfection Conditions-COS-7, RENCA, and CTLL-16 cell lines were obtained from ATCC and L929 cells were from European Collection of Cell Cultures. WT and IL-15R␣ Ϫ/Ϫ mouse embryonic fibroblasts were described elsewhere (14). NK cells were purified from mouse spleens using CD49b (DX5) microbeads (Miltenyi Biotech), yielding cell purities of Ͼ93%. Concanavalin A blasts were prepared by incubation of splenocytes with 2 g/ml concanavalin A for 48 h. Bone marrow-derived dendritic cells (DCs) and macrophages (M⌽) were generated as described previously (15,16). Cells were cultured in RPMI 1640 or Dulbecco's modified Eagle's medium supplemented with 10% fetal calf serum, 2 mM L-glutamine, 100 units/ml penicillin, and 100 g/ml streptomycin. Cells were transfected using Lipofectamine 2000 (Invitrogen). Transfection efficiency confirmed by fluorescent microscopy and WB was about 60%. After 48 h, conditioned medium was replaced by a fresh medium without fetal calf serum (500 l), and cells were incubated for another 4 h. Supernatants and cells were harvested and analyzed by WB or ELISA. Cells were infected with 2.4 hemagglutinating units/ml Newcastle disease virus or stimulated with 100 ng/ml LPS.
For the generation of the exon 2 deletion mutant, a 65-amino acid sequence corresponding to the cytokine-binding domain ( 33 GTTCPPPVSIEHADIRVKNYSVNSRERYVCNSGFKRKA-GTSTLIECVINKNTNVAHWTTPSLKCI 97 ) was deleted using inverse PCR strategy. The pair of primers (sense, 5Ј-AGAGA-CCCCTCCCTAGCTCAC-3Ј; antisense, 5Ј-CGGCGTCACC-CTCAGCGGGAG-3Ј) in the IL-15R␣ coding sequence was designed in such a way that, after PCR amplification, the complete plasmid pEGFP-N1 was obtained again, lacking only the bases located between the two primers. The obtained PCR products were subsequently phosphorylated and ligated. The identity of the deletion-containing construct was verified by a standard DNA sequencing.
Mice-IL-15R␣ ϩ/ϩ and IL-15R␣ Ϫ/Ϫ C57BL/6 mice were bred in Research Center Borstel under specific pathogen-free conditions. Mice were bled from the tail vein, and sera were analyzed by ELISA.
Immunoprecipitation and WB-Supernatants were concentrated 10-fold using 10 kDa cut-off filtration units (Vivaspin; Vivascience). Nonidet P-40 (0.5% final concentration) and a mixture of protease inhibitors were added to supernatants, and immunoprecipitation was performed for 2 h at 4°C. Immunocomplexes were captured on protein A/G-agarose (Pierce). To analyze glycosylation, the samples were treated with 250 milliunits of N-glycosidase F (Roche Applied Science) for 3 h at 37°C according to the manufacturer's instructions. WB analysis of precipitates and protein lysates was performed as described elsewhere (7,8). ELISA-Concentration of mouse sIL-15R␣, IFN␥, and human IL-15 in cell supernatants, lysates, and mouse sera was evaluated by DuoSet ELISA kits according to the manufacturer's recommendations. IL-15R␣⅐IL-15 heterocomplexes were detected by a two-site ELISA. Plates were coated with monoclonal anti-IL-15R␣ Abs, and IL-15R␣⅐IL-15 heterocomplexes in samples were detected using polyclonal biotinylated goat anti-human IL-15 Abs followed by incubation with streptavidin-peroxidase. Chromogenic substrate (R&D Systems) was used for visualization, and the reaction was stopped after 20 min of incubation by the addition of 1 N H 2 SO 4 . Optical density was determined at 450 nm using an ELISA reader (Dynatech). Anti-IL-15R␣ or anti-IL-15 Abs did not show cross-reactivity for IL-15 or IL-15R␣, respectively.
The reference standard used for the two-site ELISA was a dilution of recombinant IL-15 and recombinant soluble IL-15R␣, consisting of a full extracellular domain of this receptor subunit. These two molecules were mixed at a fixed molar ratio of 1:1 and incubated for 2 h at room temperature with periodic gentle stirring to allow complex formation before addition in 2-fold serial dilutions to the ELISA plate. In addition, a standard curve for IL-15 or IL-15R␣ was also established using specific ELISA for IL-15 or IL-15R␣, respectively, and correlated well with the standard curve using IL-15⅐IL-15R␣ complexes. Additional control experiments demonstrated that the two-site ELISA detects only IL-15⅐IL-15R␣ complexes and neither component alone (IL-15 or IL-15R␣, respectively).
Data Analysis-All experiments were performed in at least three independent assays, which yielded highly comparable results. Semiquantitative PCR data were quantified using ImageQuant TL software (Amersham Biosciences). Data are summarized as mean Ϯ S.D. Statistical analysis of the results was performed by Student's t test for unpaired samples. A p value of Ͻ0.05 was considered statistically significant.

sIL-15R␣ Forms Heterocomplexes with IL-15-
The ability of recombinant sIL-15R␣ to bind exogenous and endogenous IL-15, preventing the IL-15-specific interaction with IL-15R complex and the IL-15-mediated downstream signaling, is well documented (13,19,20). Given that endogenous sIL-15R␣ is were detected in mouse serum by specific and two-site ELISA, respectively. Serum from IL-15R␣ Ϫ/Ϫ mice was used as a control. Sera from at least 10 animals per strain were tested. The figure depicts actual measurements of representative serum samples from three IL-15R␣ ϩ/ϩ mice and a mean value from 10 IL-15R␣ Ϫ/Ϫ mice. B, detection of IL-15R␣ and IL-15 in cell lysates from L929, M⌽, mouse embryonic fibroblasts (MEFs) and DCs by specific ELISA and detection of IL-15⅐IL-15R␣ complexes by two-site ELISA. C, detection of IL-15R␣, IL-15, or IL-15⅐IL-15R␣ complexes in cell-conditioned medium from L929 cells, M⌽, and DCs. The supernatants were concentrated 10-fold in order to increase the sensitivity of the assay. The mean concentration of sIL-15R␣ or IL-15 was detected by specific ELISA for IL-15R␣ or IL-15, respectively (white or gray bars). The mean concentration of IL-15R␣-associated IL-15 was assessed by the two-site ELISA (black bars). D, IL-15R␣ was precipitated from cell lysates or mouse serum using specific Abs. Immunocomplexes were subjected to WB and tested using anti-IL-15 Abs. Detection of IL-15R␣ on the same blot served as a loading control. MAY (8) and the high affinity of IL-15R␣ toward IL-15, we questioned whether this soluble receptor molecule associates with IL-15 in biological fluids and cell supernatants, affecting ligand half-life and functional properties and causing difficulties in IL-15 detection. To test this hypothesis, mouse serum samples were assessed by a specific ELISA for IL-15 and sIL-15R␣ concentration and by a two-site ELISA for IL-15/sIL-15R␣ interactions, respectively. In a direct two-site ELISA, the 96-well plates were covered with monoclonal coating Abs that recognize the extracellular domain of IL-15R␣. After the incubation with serum samples, polyclonal detection Abs targeting IL-15 were added. In inverted setup, monoclonal anti-IL-15 Abs served as coating Abs, and polyclonal anti-IL-15R␣ served as detection Abs. Direct two-site ELISA showed that varying amounts of endogenous IL-15 are constitutively associated with sIL-15R␣ in mouse serum, ranging from 100 to 900 pg/ml (Fig. 1A), whereas specific ELISA for IL-15 or inverted ELISA did not generate any signal (data not shown), presumably due to the inability of anti-IL-15 monoclonal coating Abs to bind sIL-15R␣-associated IL-15. This suggests that binding sites on IL-15 are inaccessible to these Abs and indicates that most if not all IL-15 molecules in mouse serum are presumably bound to sIL-15R␣. Thus, sIL-15R␣ in mouse serum forms heterocomplexes with IL-15, and sIL-15R␣ prevents IL-15 recognition by monoclonal Abs.

W I T H D R A W N W I T H D R A W N
We have earlier demonstrated that mouse sera contain varying amounts of sIL-15R␣ using an ELISA newly developed in our laboratory, since an ELISA for sIL-15R␣ was not commercially available at that time (8). The results from commercial ELISA for sIL-15R␣ showed that C57BL/6 mice display different serum levels of circulating sIL-15R␣ (ϳ0.3-2 ng/ml), which are lower than those obtained by our custom ELISA and correlate with IL-15 concentration detected by two-site ELISA (Fig. 1A). It should, however, be mentioned that some animals exhibited very low, sometimes undetectable serum levels of sIL-15R␣ and a corresponding decrease in IL-15 amount (data not shown). The physiologic reason for such heterogeneity remains unclear. Not only was sIL-15R␣ undetectable by specific ELISA in serum from IL-15R␣ Ϫ/Ϫ animals, which were used here for control and comparison, but also IL-15 was absent, suggesting that IL-15R␣ is required for IL-15 liberation and might play an important role in cellular mechanism(s) controlling this process.
Next, we tested whether IL-15⅐IL-15R␣ heterocomplexes exist in primary cells from normal and IL-15R␣ Ϫ/Ϫ animals or cells and cell lines of various origin. These experiments revealed that murine M⌽, mouse embryonic fibroblasts, DCs, and L929 cells express IL-15R␣ and, to a lesser extent, IL-15, whereas IL-15R␣ was undetectable in the cell lysates of mouse embryonic fibroblasts and DCs from IL-15R␣ Ϫ/Ϫ animals (Fig. 1B). Interestingly, the IL-15 level was also rather low in these cells, as detected by specific ELISA. The concentration of IL-15 in the cell lysates was always higher according to twosite versus specific ELISA, indicating that at least some IL-15 molecules associate with IL-15R␣ already within the cell. Furthermore, sIL-15R␣-associated IL-15 (ϳ100 pg/ml) was also detected by two-site ELISA in cell-conditioned medium from M⌽ and DCs but not L929 fibroblasts, whereas specific ELISA for IL-15 did not generate any signal (Fig. 1C). Remarkably, all of these cells abundantly release sIL-15R␣, as evaluated by specific ELISA for this receptor chain (ϳ320 -550 pg/ml). The ability of sIL-15R␣ to form heterocomplexes with IL-15 intracellularly was also confirmed by WB analysis after immunoprecipitation from the cell lysates using specific anti-IL-15R␣ Abs, whereas these Abs were not able to precipitate IL-15R␣ and IL-15 from lysates of IL-15R␣ Ϫ/Ϫ DCs (Fig. 1D). Furthermore, the association between IL-15R␣ and IL-15 was also found in serum from WT but not IL-15R␣ Ϫ/Ϫ mice (Fig. 1D). Isotype-matched control Abs did not precipitate these proteins (data not shown). These results show that IL-15R␣ can associate with IL-15 in the cytoplasm and forms soluble heterocomplexes with IL-15 in mouse serum and cellconditioned medium from distinct primary murine cells.
Identification of Two Alternatively Spliced sIL-15R␣ Sushi Domain Isoforms-By screening for specific IL-15R␣ transcripts that may encode polypeptide chains lacking transmembrane and/or other regions, thereby giving rise to alternatively spliced sIL-15R␣, two mRNA products potentially representing such sIL-15R␣ isoforms were detected in L929 fibroblasts by reverse transcription-PCR. These transcripts were also identified in kidney epithelial cell line RENCA, excluding the possibility of a cloning-associated artifact ( Fig. 2A). The mRNA products were cloned and sequenced and predicted to encode two slightly different IL-15R␣ isoforms (referred to hereafter as S1 and S2, respectively). The S1 isoform is 300 bp in length, whereas S2 is encoded by an mRNA that is 321 bp in length. Both isoforms contain regions coding only for the signal pep-FIGURE 2. Cloning of novel soluble IL-15R␣ isoforms. A, reverse transcription-PCR amplification of different IL-15R␣ isoforms from RENCA and L929 cells. A mock PCR (without cDNA) was included to exclude contamination. B, comparative sequence analysis of S1 and S2 IL-15R␣ isoforms. The sequence corresponding to exon 1 is in the gray box, and the sequence corresponding to exon 2 (sushi domain) is in boldface type. The start codon is underlined. The glycosylation site is indicated by an asterisk. C, cell lysates were analyzed by WB using anti-IL-15R␣ Abs and anti-S2 antiserum. Detection of ␤-actin on the same blot was used as a loading control. D, detection of S2 isoform in 50-fold concentrated conditioned medium from L929 cells by WB using S2 antiserum. The supernatants from COS-7 cells transfected with a construct coding for S2 isoform or mock-transfected (empty vector) served as a positive or negative control, respectively. E, expression of IL-15R␣ WT and S2 isoforms in different mouse tissues was analyzed by semiquantitative PCR. Expression levels are plotted as a ratio between specific cDNA amplification and amplification of control gene (␤-actin). F and G, regulation of expression of IL-15R␣ WT and S2 isoforms in different cells. DCs and M⌽ were stimulated with LPS (100 ng/ml) for 4 or 8 h (E). L929 fibroblasts were stimulated for 3 or 6 h with LPS (100 ng/ml) or Newcastle disease virus (NDV) (2.4 hemagglutinating units) (F). Total RNA was extracted from cells, reverse-transcribed, and subjected to semiquantitative PCR amplification using specific primers for IL-15R␣ WT and S2 as described under "Experimental Procedures." The image shows the amplified bands after 35 cycles. The amount of cDNA was equalized by PCR amplification of ␤-actin. A mock PCR (no cDNA) was included as a negative control. H, the S2 isoform is present in murine sera in vivo, as detected by specific ELISA for sIL-15R␣, in which monoclonal anti-IL-15R␣ Abs targeting the extracellular domain of multiple IL-15R␣ isoforms served as capture Abs, whereas anti-S2 Abs served as specific detection Abs for S2 protein. Detection of IL-15R␣ WT is shown as a positive control. Detection of both isoforms in IL-15R␣ Ϫ/Ϫ mice represents negative controls. MAY (Fig. 2B). These isoforms are slightly different in length (the S2 isoform is longer by 7 amino acids) and exhibit minor variations at the 3Ј end of mRNA, whereby serine and lysine in the S1 isoform are replaced by arginine and proline in the S2 isoform, respectively. The open reading frame in the S1 and S2 isoform shares with the extracellular region of IL-15R␣ (518 bp) the first 293 bp until isoleucine at position 97, flanking in S1 what is presumably an unspliced adjacent intron 2 that terminates the open reading frame following the lysine residue. The S2 isoform also seems to an end in unspliced intron 2, although the corresponding coding sequence is located more distantly (data not shown). The predicted molecular mass of S1 and S2 is about 12-15 kDa. L929 and RENCA cells were also found to express IL-15R␣ WT as well as IL-15R␣⌬4, IL-15R␣⌬3,4, and IL-15R␣⌬3,4,5 isoforms ( Fig. 2A) recently identified and characterized in murine mast cells (7).

W I T H D R A W N W I T H D R
To confirm the existence of novel sIL-15R␣ isoforms at the protein level, polyclonal Abs of rabbit origin directed against the unique last 9 amino acids of the S2 isoform were generated. As shown in Fig. 2C, varying amounts of S2 protein were detected in L929 fibroblasts, mouse embryonic fibroblasts, DCs, and M⌽ but not in concanavalin A-activated T cell blasts. IL-15R␣ WT protein was also detected in these cells. Remarkably, the expression of S2 isoform was considerably higher in DCs, whereas no S2 isoform was detected in DCs from IL-15R␣ Ϫ/Ϫ animals (Fig. 2C). Moreover, S2 protein was also detected in the concentrated culture medium (CM) from L929 cells (Fig. 2D). Given that the S1 isoform differs from IL-15R␣ WT and S2 only by 2 amino acid residues, specific Abs to the S1 isoform could not be generated, thereby excluding an experimental confirmation of S1 protein expression. Taken together, these experiments demonstrated that a novel alternative splicing mechanism is responsible for the generation of two natural sIL-15R␣ sushi domain isoforms.
To gain information about the source of circulating sIL-15R␣ and to analyze the pattern of expression of the novel S2 isoform, we surveyed a wide variety of cell types by semiquantitative reverse transcription-PCR for the expression for S2 isoform mRNA, using lower primer targeting the unique sequence at 3Ј end of S2. Concomitantly, the expression of IL-15R␣ WT mRNA was assessed for control and comparison (Fig. 2E). Both IL-15R␣ WT and S2 isoform were highly expressed in lymph nodes and testis, whereas the highest level of IL-15R␣ WT expression was found in the heart. Relatively low levels of expression of both proteins were detected in muscle and brain. Interestingly, the level of the S2 isoform was relatively high in thymus as compared with IL-15R␣ WT. Furthermore, stimuli, such as LPS and Newcastle disease virus, differentially regulated expression of the alternatively spliced IL-15R␣ S2 isoform and IL-15R␣ WT at the mRNA level. Both stimuli up-regulated IL-15R␣ WT expression in DCs, M⌽, and L929 cells, whereas the S2 isoform was down-regulated upon LPS stimulation (Fig.  2, F and G). This fact suggests that cells can reciprocally regulate the production of distinct IL-15R␣ isoforms according to particular environmental challenges.
We also confirmed that the S2 isoform is present in murine sera in vivo using specific ELISA for sIL-15R␣, in which mono-clonal anti-IL-15R␣ Abs targeting the extracellular domain of multiple IL-15R␣ isoforms served as capture Abs, whereas anti-S2 Abs served as specific detection Abs for S2 protein. This approach has an advantage of enhanced sensitivity as compared with immunoprecipitation and Western blotting analysis. It should be mentioned that the concentration of S2 protein in serum was rather low (below 100 pg/ml) (Fig. 2H). However, the fact that anti-S2 Abs target the last 9 unique amino acids of the S2 isoform essentially limits the sensitivity of the assay, making more precise detection of the S2 isoform a technically daunting task.
Next, we set forth to investigate whether IL-15R␣ WT, IL-15R␣⌬4, IL-15R␣⌬3,4, and IL-15R␣⌬3,4,5 as well as novel S1 and S2 isoforms are released as soluble proteins to the CM and form complexes with IL-15. To this end, COS-7 and HeLa cells were transfected with constructs encoding these IL-15R␣ chains. Indeed, these proteins were effectively translated in COS-7 and HeLa cells and released to the cell supernatants, as detected by WB of the cell lysates and CM, with the exception of the IL-15R␣⌬3,4,5 isoform, which was absent in the CM (Fig. 3, A and B; data not shown). Next, supernatants from these cells were tested by specific ELISA for IL-15R␣. Interestingly, all isoforms except IL-15R␣⌬3,4,5 were present in rather high concentration (about 6 -8 ng/ml) in both CM and cell lysates after transfection with respective expression vectors, whereas IL-15R␣⌬3,4,5 was predominantly found in the cell lysates (Fig.  3C). Expression of IL-15R␣ on the cell surface of transfected COS-7 cells was confirmed by immunostaining and fluorescence-activated cell sorting analysis. Phorbol 12-myristate 13-acetate treatment for 2 h induced shedding of IL-15R␣, resulting in the decrease of surface IL-15R␣ WT, IL-15R␣⌬4, and IL-15R␣⌬3,4 isoform expression, whereas expression of the IL-15R␣⌬3,4,5 isoform remained intact (Fig. 3D).
The Novel IL-15R␣ Isoforms Show Usage of Their N-Glycosylation Site-IL-15R␣ has single N-glycosylation and multiple O-glycosylation sites in the extracellular domain (5). Reportedly, a number of IL-15R␣ isoforms are N-glycosylated, whereas no O-glycosylation was observed (7). To study the role of N-glycosylation in the post-translational processing of S1 and S2 isoforms, these were transiently transfected into COS-7 cells. The whole cell lysates or concentrated CM from transfected cells were treated with N-glycosidase, and the expression products were analyzed by WB. Fig. 3E demonstrates that indeed both novel IL-15R␣ isoforms are N-glycosylated in the cell lysates and CM, since treatment with N-glycosidase shifts corresponding protein bands to a lower molecular mass position.
IL-15R␣ Is Essential for IL-15 Release-Reportedly, both IL-15 and IL-15R␣ must be co-expressed by the same cells to transpresent IL-15, indicating a requirement of IL-15R␣ at a cellular level for IL-15 elaboration (22). Thus, we sought to investigate whether IL-15R␣ could affect IL-15 secretion using co-transfection experiments. For this purpose, COS-7 cells were co-transfected with vectors coding for IL-15 and IL-15R␣ WT as well as novel S1 and S2 isoforms. Interestingly, these experiments showed that IL-15R␣ expression was indeed essential for IL-15 release into the CM (Fig. 4A)  centrations of sIL-15R␣ WT, S1, and S2 isoforms are present in the CM. Notwithstanding, the release of IL-15 to the CM was observed only upon concomitant transfection with a respective IL-15R␣ construct. Given that this effect was also seen in cells transfected with S1 or S2 isoforms, this indi-cates that distinct cellular mechanism(s) orchestrate and control the coordinated release of both IL-15R␣ and IL-15. Notably, the amount of IL-15 was much lower compared with its high affinity chain (about 100 -170 pg/ml versus 2.7-8.6 ng/ml, respectively).

. Analysis of expression and shedding/secretion of different IL-15R␣ isoforms in transfected COS-7 and HeLa cells.
Cells were transfected with different constructs as indicated. Empty vector (mock)-transfected cells were used as a negative control, transfected with IL-15R␣ WT construct as a positive control. Expression of membrane-bound IL-15R␣ isoforms (A) and sIL-15R␣ (B) isoforms was analyzed in cell lysates (L) and culture medium (CM), respectively, from COS-7 by WB using anti-GFP Abs. C, expression and shedding/secretion of IL-15R␣ isoforms in L and CM from transfected COS-7 and HeLa cells were analyzed by specific ELISA. *, p Ͻ 0.05 versus IL-15R␣ WT. D, effect of phorbol 12-myristate 13-acetate (PMA) on the surface expression of IL-15R␣ isoforms in transfected COS-7 cells. Cells were treated with 200 ng/ml phorbol 12-myristate 13-acetate for 2 h, and expression of IL-15R␣ was analyzed by fluorescenceactivated cell sorting analysis. Untreated cells were used as a control. E, glycosylation pattern of S1 and S2 IL-15R␣ isoforms. Samples were treated with N-glycosidase or left untreated as described under "Experimental Procedures." After treatment, protein lysates were analyzed by WB using anti-GFP Abs. MAY

W I T H D R A W N W I T H D R A W N
Next, we tested whether blocking of IL-15R␣ shedding by a broad spectrum metalloproteinase inhibitor, GM6001, affects IL-15 secretion. To this end, COS-7 cells transfected as above were treated with this chemical compound, and the supernatants were tested by ELISA for IL-15 or sIL-15R␣. In fact, GM6001 considerably reduced IL-15R␣ WT shedding (an about 4-fold decrease), which was accompanied by a comparable decline in IL-15 release (Fig. 4B). Conversely, GM6001 did not affect the release of S1 and S2 isoforms. Importantly, the release of IL-15 to the supernatants from IL-15R␣ S1-or S2-transfected cells also was not affected by GM6001 (Fig. 4B). This is in accord with the fact that these isoforms lack membrane-proximal and transmembrane regions and thus cannot undergo proteolytic cleavage.
To validate further the ability of IL-15R␣ to form complexes with IL-15 and affect IL-15 secretion, cell-conditioned medium from transfected COS-7 cells was tested by specific or two-site ELISAs for the presence of sIL-15R␣, IL-15, or sIL-15R␣⅐IL-15 complexes, respectively. Specific ELISA for IL-15R␣ showed that this protein is present (ϳ5.5-6.8 ng/ml) in the CM of all transfected cells (Fig. 4C). Direct two-site ELISA for IL-15 confirmed the release of IL-15 in rather high concentrations (ϳ3.5-4.2 ng/ml) to the CM. IL-15R␣ S1 and S2 isoforms were almost equipotent in sustaining IL-15 release as compared with IL-15R␣ WT (Fig. 4C). Conversely, specific ELISA for IL-15 detected considerably lower cytokine concentrations (ϳ100 -200 pg/ml). Thus, overexpression studies demonstrate that IL-15 associates with sIL-15R␣ in the CM and show that IL-15R␣ is essential for IL-15 release.

Exon 2-encoding Sushi Domain of IL-15R␣ Is Required for IL-15
Secretion-Next, we tested whether the cytokine-binding sushi domain of IL-15R␣ is important for IL-15 secretion. To this end, sushi domain (exon 2) deletion mutant was generated (IL-15R␣⌬2) and co-transfected with IL-15 in COS-7 cells. As shown in Fig. 4D, the deletion of exon 2 indeed dramatically affected IL-15 secretion to the CM, as detected by specific ELISA. Remarkably, IL-15 was almost undetectable in the supernatants of IL-15R␣⌬2-transfected cells, whereas cells expressing IL-15R␣ WT released this cytokine (ϳ100 pg/ml). The protein expression of both IL-15R␣ and IL-15 constructs was at a comparable level (lower panels). These results suggest that the cytokine-binding sushi domain of IL-15R␣ is required for IL-15 secretion.

Soluble Sushi Domain of IL-15R␣ and IL-15 Form Agonistic Heterocomplexes-The recombinant sushi domain of IL-15R␣
was recently shown to act as a potent IL-15 agonist, suggesting that, if naturally produced, such sIL-15R␣ sushi domains might be involved in the IL-15 transpresentation mechanism (23). Conversely, it was also reported to inhibit IL-15 action (20). We and others demonstrated that proteolytically generated sIL-15R␣ ectodomain inhibits IL-15 action (8,24). Thus, we investigated the influence of IL-15⅐sIL-15R␣ complexes consisting of S1 or S2 isoforms in combination with IL-15 on CTLL cell proliferation and cytokine production by NK cells. COS-7 cells were transiently transfected with IL-15 expression vector together with constructs coding for sIL-15R␣ S1 and S2, as well as IL-15R␣⌬4, IL-15R␣⌬3,4, and IL-15R␣⌬3,4,5 isoforms and IL-15R␣ WT for control and comparison. The cell-conditioned media were collected, and concentration of IL-15R␣ and IL-15 was evaluated by direct and two-site ELISA (data not shown) and added to CTLL cells. Final concentration of sIL-15R␣complexed IL-15 after the addition to the cells was equilibrated to be 250 pg/ml. In this experimental setup, IL-15R␣ was expected to associate with IL-15 and control IL-15 secretion, whereby resulting IL-15⅐IL-15R␣ complexes would subsequently be released to the CM by protease-dependent or -independent mechanism(s). In fact, only supernatants from COS-7 cells co-transfected with IL-15 and sIL-15R␣ S1 or S2 and containing respective IL-15⅐sIL-15R␣ sushi domain heterocomplexes were able to stimulate the proliferation of CTLL cells, whereas other combinations were ineffective (Fig. 5A).
Remarkably, the addition of recombinant sIL-15R␣ did not affect the proliferation of these cells.
Next, we isolated murine NK cells and tested using the same setup whether IL-15⅐sIL-15R␣ sushi domain complexes induce IFN␥ production by these cells. Remarkably, NK cells produced considerable amounts of IFN␥ in response to stimulation with these complexes, whereas supernatant from COS-7 cells transfected with IL-15R␣ WT was without effect. The production of IFN␥ was higher in NK cells stimulated with IL-15⅐sIL-15R␣ S1 and S2, as compared with IL-15 alone (100 ng/ml), lying in the range of 80 -100 versus 50 -60 pg/ml, respectively (Fig. 5B). Remarkably, recombinant IL-15 in a lower concentration (0.1 ng/ml) did not show any stimulating capacity. Again, the addition of recombinant sIL-15R␣ had no effect, although it efficiently prevented stimulating action of IL-15 (data not shown). These results demonstrate that naturally produced and elaborated IL-15⅐sIL-15R␣ S1 or S2 heterocomplexes stimulate CTLL cell proliferation and IFN␥ production by NK cells, thereby implicating them in IL-15 transpresentation.

DISCUSSION
In this study, we show for the first time that an alternative splicing mechanism within the IL-15R␣ gene is responsible for the generation of natural sIL-15R␣ sushi domain isoforms. Thus, sIL-15R␣ in mice is produced through a dual mechanism that includes both proteolytic processing of the membrane-tethered receptor and alternative splicing. We  MAY

W I T H D R A W N W I T H D R A W N
also demonstrate that heterocomplexes consisting of IL-15 and sIL-15R␣ are present in mouse serum and cell culture supernatants, which suggests that maintenance of a dynamic equilibrium between serum levels of sIL-15R␣ and IL-15 is important for IL-15 biology. Given that natural sIL-15R␣ sushi domain promotes IL-15 activity, inducing the IL-15-mediated proliferation of CTLL cells and IFN␥ production by NK cells, transpresentation of IL-15 in such soluble agonistic complexes may constitute a heretofore unappreciated novel mode of action for this pleiotropic cytokine. Finally, IL-15R␣ is required for a coordinated secretion of IL-15, suggesting that distinct control mechanism(s) orchestrate this process.
Thus, although murine sIL-15R␣ is predominantly generated through TACE-dependent proteolysis of the membrane IL-15R␣, both proteolytic cleavage and alternative splicing coexist in the mouse system, adding to our understanding of this process (Fig. 6). It remains to be elucidated whether such a dual mechanism of sIL-15R␣ generation also exists in humans. Whereas proteolysis might primarily be responsible for the generation of sIL-15R␣ capable of inhibiting IL-15 action (8,24), the secreted sIL-15R␣ sushi domain exhibits agonistic qualities. This fact is in agreement with the reported ability of recombinant sIL-15R␣ sushi domain to act as a potent and selective agonist of IL-15 function through the intermediate affinity IL-2/IL-15R␤␥ heterodimer (23). It has been suggested that, if naturally produced, such sIL-15R␣ sushi domains might be involved in the IL-15 transpresentation mechanism.
IL-15⅐sIL-15R␣ complexes might play a particular role in distinct physiologic and/or pathologic conditions and alter in the first place IL-15 action on the cells expressing membranebound IL-15R complex. Depending on the particular sIL-15R␣ subtype (i.e. sushi domain or full-length ectodomain), these soluble molecules may compete for IL-15 with the cellular receptors and inhibit its activity or, contrariwise, promote IL-15 action, especially in cells expressing intermediate affinity IL-15R␤␥ complex. Thus, in addition to the ability of membrane-anchored IL-15R␣ to present IL-15 in trans to neighboring cells during direct cell-to-cell contact (15), IL-15⅐sIL-15R␣ sushi domain heterocomplexes might also perform a similar function of IL-15 transpresentation yet in soluble form. This fact must be particularly relevant to the complex biology of IL-15, with a specific impact on the generation and maintenance of critical immune effector cells, such as memory CD8ϩ T cells, NKT cells, NK cells, and ␥␦ T cells. In fact, injection of recombinant IL-15⅐sIL-15R␣ complexes in vivo induces strong and selective expansion of memory CD8ϩ T cells and NK cells (25). Therefore, transpresentation of IL-15 by natural sIL-15R␣ sushi domain in such agonistic complexes may represent an important mechanism of IL-15 action.
Although shed sIL-15R␣ ectodomain acts as an inhibitory molecule, it might also provide increased molecular stability to IL-15, leading to the reduced activity decay, and expand IL-15 action from autocrine or juxtacrine to paracrine or endocrine modes, resulting in local or systemic effects of the cytokine and influencing the nature and/or duration of the signaling event. Given that sIL-15R␣ ectodomain could extend the bioavailability of IL-15 by prolonging cytokine half-life, whereas disassociation of IL-15 from the complex under conditions that favor this process may serve to provide physiological concentrations of the cytokine in tissues, the precise role of proteolytically generated sIL-15R␣ awaits further studies.
Despite widespread expression of IL-15 mRNA, detection of significant amounts of IL-15 in cell culture supernatants has proven to be extremely difficult (2). Because sIL-15R␣ prevents recognition of IL-15 by monoclonal coating Abs in inverted ELISA, it appears likely that sIL-15R␣ is present in excess over IL-15, and the majority of IL-15 molecules in mouse serum are presumably associated with sIL-15R␣, suggesting that these two molecules exist in a dynamic equilibrium. Considering that human IL-15 has at least two binding sites for recombinant sIL-15R␣ (26) and that IL-15R␣ can, at least partially, oligomerize (27), it is theoretically conceivable that at a low molar ratio, sIL-15R␣ forms complexes with IL-15 where IL-15 could still be detected by distinct anti-IL-15 Abs, whereas at higher molar ratios, most commercially available anti-IL-15 Abs would not detect IL-15 in a specific or two-site ELISA setup. Our results provide the first conclusive evidence that detection of free IL-15 is aggravated by the ability of sIL-15R␣ to form receptorligand complexes with IL-15, concealing certain antigenic determinants from recognizing Abs and resulting in false-negative assessments. It remains to be elucidated which molar ratio(s) of sIL-15R␣ are saturating for IL-15 to prevent its detection by respective Abs. Given that sIL-15R␣ might hide certain IL-15 antigenic epitope(s) also from polyclonal detection Abs, the results from direct two-site ELISA should be interpreted with caution in regard to actual endogenous IL-15 concentration in mouse serum. Also of importance is the fact that antagonistic effects of a particular soluble cytokine receptor are directly related to its concentration and inversely related to the concentration of cytokine (28). Thus, the molar ratio between IL-15 and a distinct sII-15R␣ subtype may define agonistic or antagonistic properties of such complexes.
IL-15 and IL-15R␣ are critical for the generation, development, homeostasis, and function of memory CD8ϩ T cells, ␥␦ T cells, NKT cells, NK cells, and certain subsets of intestinal intraepithelial lymphocytes, playing important roles in both innate and adaptive immunity (1,29,30). Given that IL-15 is associated with a wide range of immunopathological reactions, it has been suggested that this cytokine may be at the apex of a cytokine cascade that includes downstream production of IL-1, IL-6, granulocyte-macrophage colony-stimulating factor, and other biologically active substances (2). The fine tuned balance between antagonistic and agonistic IL-15⅐sIL-15R␣ may be altered in distinct pathological conditions and can probably serve as a prognostic marker in certain types of diseases, including allergy and inflammation. Diverse neoplastic and inflammatory diseases, including adult T cell leukemia and certain auto-immune disorders like rheumatoid arthritis, inflammatory bowel disease, multiple sclerosis, and pulmonary sarcoidosis, are characterized by abnormalities in IL-15 expression (31). Given that production of IL-15 is burdened by multiple complex regulatory mechanisms and is tightly governed at multiple levels (18), inhibitory sIL-15R␣ may serve as an additional important protective control point to limit excessive and/or undesired IL-15 activity.
Experiments using IL-15-and IL-15R␣-deficient mixed chimera showed that IL-15 and IL-15R␣ must be expressed by the same cells to present IL-15 in trans to neighboring cells, which indicates that IL-15R␣ is required on a cellular level for the elaboration of IL-15 (22). Our data conclusively demonstrate a direct requirement of IL-15R␣, in particular its sushi domain, for IL-15 secretion. Such orchestrated expression suggests that, in addition to multiple control checks regulating release of IL-15, there may exist yet another IL-15R␣-dependent control mechanism. This ensures that IL-15 liberation occurs essentially in the presence of, and perhaps as a result of direct or indirect regulation by, IL-15R␣. It remains to be elucidated which molecular mechanism(s) is responsible for this coupling. The reported ability of IL-15 bound to IL-15R␣ to recycle between early endosomes and the cell surface, maintaining the pool of biologically active IL-15 in trans, provides an explanation as to how such heterocomplexes may be transported to the cell membrane (32). Remarkably, the IL-15R␣ cytoplasmic domain plays an essential role in the recycling process (32). Alternatively, IL-15⅐IL-15R␣ complexes may have intracrine effects on the cell that generates them without ever leaving the cell. In fact, a nonsecretable isoform of IL-15 that shows nuclear co-localization with IL-15R␣ can down-regulate IL-15 gene transcription (21).
Upon the cell surface, IL-15 coupled to IL-15R␣ acts in trans on neighboring cells expressing intermediate affinity IL-15R␤␥ complex (32). However, it is theoretically conceivable that after some persistence on the cell surface, IL-15R␣ may undergo proteolytic processing and enter the circulation either alone or together with IL-15. Conversely, alternatively spliced soluble IL-15R␣ sushi domain may similarly associate with IL-15 in the cell and follow a secretion route whereby resulting complexes act as highly specific and potent agonists of IL-15 function. Such dual properties of sIL-15R␣ isoforms might play an important role in the complex biology of IL-15, negatively or positively affecting the IL-15-mediated biological responses (Fig. 7).
It remains unclear why recombinant sIL-15R␣ is not able to inhibit the ability of IL-15⅐sIL-15R␣ sushi domain complexes to induce CTLL cell proliferation and IFN␥ production by NK  6) in the human system; ***, isoforms cloned from mouse mast cells (7). #, soluble IL-15R␣ generated by proteolytic cleavage (8,24). MAY (23,25). Remarkably, extracellular stimuli, such as LPS and Newcastle disease virus, can affect IL-15R␣ expression, modulating production of certain IL-15R␣ subtypes, including novel alternatively spliced isoforms at the mRNA level. Cells may favor production of distinct IL-15R␣ isoforms that could also be selectively distributed among various intracellular compartments, switching between them according to particular existing requirements and thereby adjusting to rapid environmental changes and demands (Fig. 7). This may have a dramatic impact upon the biologic responses mediated by IL-15 in diverse physiologic or pathologic conditions. In summary, our data demonstrate that sIL-15R␣ in mouse is generated by both alternative splicing and proteolytic cleavage. Qualitatively distinct sIL-15R␣ molecules form functional soluble heterocomplexes with IL-15 capable of inhibiting or potentiating IL-15 action. Thus, IL-15⅐sIL-15R␣ sushi domain heterocomplexes might play an important role in transpresentation of IL-15 and the generation and maintenance of multiple lymphocyte subsets, including memory CD8ϩ T cells and NK cells. This highlights new dimensions in our understanding of how this pivotal cytokine exerts its pleiotropic biological functions and identifies novel targets for the therapeutic manipulation of the pathologic conditions linked with abnormal IL-15 activity.