The PRY/SPRY/B30.2 Domain of Butyrophilin 1A1 (BTN1A1) Binds to Xanthine Oxidoreductase

Butyrophilin 1A1 (BTN1A1) and xanthine oxidoreductase (XOR) are highly expressed in the lactating mammary gland and are secreted into milk associated with the milk fat globule membrane (MFGM). Ablation of the genes encoding either protein causes severe defects in the secretion of milk lipid droplets, suggesting that the two proteins may function in the same pathway. Therefore, we determined whether BTN1A1 and XOR directly interact using protein binding assays, surface plasmon resonance analysis, and gel filtration. Bovine XOR bound with high affinity in a pH- and salt-sensitive manner (KD = 101 ± 31 nm in 10 mm HEPES, 150 mm NaCl, pH 7.4) to the PRY/SPRY/B30.2 domain in the cytoplasmic region of bovine BTN1A1. Binding was stoichiometric, with one XOR dimer binding to either two BTN1A1 monomers or one dimer. XOR bound to BTN1A1 orthologs from mice, humans, or cows but not to the cytoplasmic domains of the closely related human paralogs, BTN2A1 or BTN3A1, or to the B30.2 domain of human RoRet (TRIM 38), a protein in the TRIM family. Analysis of the protein composition of the MFGM of wild type and BTN1A1 null mice showed that most of the XOR in mice lacking BTN1A1 was released from the MFGM in a soluble form when the milk lipid droplets were disrupted to prepare membrane, compared with wild-type mice, in which most of the XOR remained membrane-bound. Thus BTN1A1 functions in vivo to stabilize the association of XOR with the MFGM by direct interactions through the PRY/SPRY/B30.2 domain. The potential significance of BTN1A1/XOR interactions in the mammary gland and other tissues is discussed.

B30.2 or the PRY/SPRY domain, which comprises two sheets of antiparallel ␤-strands folded into a ␤sandwich, which in some proteins is contiguous at the N terminus with one or two ␣-helices (20 -24) (for a discussion of the relationship between PRY, SPRY, and B30.2 domains, see Ref. (25)). This domain (here abbreviated as B30.2), 4 is postulated to serve as a protein-binding module, by which proteins interact through the extended surface loops that adjoin individual ␤-strands (22).
One protein that may bind to the cytoplasmic region of BTN proteins (and the B30.2 domain) is the redox enzyme, xanthine oxidoreductase (XOR), 5 because it was shown to bind to the cytoplasmic domain of mouse BTN1A1 in an in vitro binding assay (26). Furthermore, one XOR-deficient mouse strain (Xdh ϩ/Ϫ ) (27) displayed a lactation phenotype similar to that of Btn1a1 null mice (1), suggesting that the two proteins may be functionally linked by direct interaction. These conclusions, however, have been challenged, because XOR does not co-localize with BTN1A1 in immunolabeled freeze-fractured replicates of secreted milk lipid droplets (28), and a second mouse strain deficient in XOR does not appear to have an altered lactation phenotype (29).
In this paper, we devise in vivo and in vitro assays to show that the cytoplasmic domain of BTN1A1 binds to XOR via the B30.2 domain and that BTN1A1 is required for the stable association of XOR with the MFGM in vivo. Furthermore, interaction with XOR appears to be limited to BTN1A1 orthologs. These results are discussed in the context of potential functions of BTN1A1 in the mammary gland and other tissues and the relationship of BTN1A1 to other BTN family members.

EXPERIMENTAL PROCEDURES
Materials-Reduced glutathione, vitamins, amino acids, pepstatin A, leupeptin, N ␣ -p-tosyl-L-lysine chloromethyl ketone, N ␣ -p-tosyl-L-phenylalanine chloromethyl ketone, aprotinin, digitonin, and adjuvants were obtained from Sigma. The ESP yeast protein expression and purification system was from Stratagene (La Jolla, CA). Glutathione-coated beads for the purification of glutathione S-transferase (GST) fusion proteins, thrombin, ECL Western blot detection reagents, supplies for surface plasmon resonance (SPR), cyanogen bromide-acti-vated Sepharose 4B, and protein A-coated Sepharose beads were from Amersham Biosciences/GE Healthcare. Rabbits were purchased from Covance (Princeton, NJ). Horseradish peroxidase-conjugated goat anti-rabbit secondary antibody (blotting grade), Triton X-100, SDS, acrylamide, and bisacrylamide were from Bio-Rad, and protease inhibitor tablets were from Roche Applied Science. DNA polymerases, restriction enzymes, dNTPs, and trypsin were from Promega (Madison, WI), and phenylmethylsulfonyl fluoride and keyhole limpet hemocyanin were obtained from Calbiochem. Lipofectamine, rabbit anti-green fluorescent protein (GFP) antibody, and Dulbecco's phosphate-buffered saline (PBS) were from Invitrogen. All other chemicals were from Fisher.
Expression and Purification of Recombinant Proteins-Recombinant proteins were expressed in Schizosaccharomyces pombe, because the genome does not contain the XOR gene (30). Potential interactions between BTN and endogenous XOR were thus avoided. cDNAs were inserted 3Ј to the GST gene in the pESP-1 vector supplied with the ESP-Yeast Protein Expression and Purification System (Stratagene) (see supplemental Table 1S for details, definitions, nomenclature, and abbreviations for the vectors used). Deletion constructs were prepared using PCR (see supplemental Table 2S for primer pairs). SP-Q01 S. pombe cells were transformed with the completed vectors and grown in 1l cultures, as recommended by the manufacturer. In initial experiments, cells were resuspended in a mixture of 140 mM NaCl, 2.7 mM KCl, 10 mM Na 2 HPO 4 , and 1.8 mM KH 2 PO 4 (PBS1) and lysed by vortexing with glass beads for 5-7 min in the presence of proteinase inhibitors (1 mM phenylmethylsulfonyl fluoride, 1 l/ml aprotinin (8.2 IU/ml), 1 M pepstatin A, 100 M leupeptin) and 0.5% (v/v) Triton X-100. In later experiments, improved yields of protein (6 -10 mg of protein/liter) were obtained by resuspending the cells in a modified lysis buffer (150 mM NaCl, 0.5 mM EDTA, 50 mM Tris-HCl, pH 7.4, Roche Applied Science protease inhibitors, 1 mM phenylmethylsulfonyl fluoride, and 1 l/ml aprotinin (8.2 IU/ml)) and breaking the cells with a French press. Lysates were clarified by centrifugation at 17,000 ϫ g av for 30 min, followed by 34,000 ϫ g av for 1 h in a Beckman model XL-90 ultracentrifuge. Recombinant proteins were purified by chromatography on glutathione resins following standard procedures.
Preparation of Thrombin-cleaved Proteins-GST was removed from the recombinant proteins by digestion with thrombin (GE Healthcare). ϳ5-mg batches of purified fusion protein were rebound to glutathione-Sepharose beads and incubated overnight with gentle agitation at 22°C in 5.0 ml of 20 mM Tris-HCl buffer, pH 8.4, containing 150 mM NaCl, 2.5 mM CaCl 2 , and thrombin (10 NIH enzyme units/mg of fusion protein). The recombinant protein was recovered from the supernatant, and the thrombin was removed by gel filtration on Superdex 200.
Preparation of Vectors Encoding Fluorescent Fusion Proteins of mBTN1A1 and mXOR-Total RNA was purified from lactating mouse mammary gland using TRIzol reagent (Invitrogen) and XOR and BTN1A1 cDNAs prepared by reverse transcription-PCR using the SuperScript TM III One-Step RT-PCR kit from Invitrogen and specific primers complementary to the 5Ј and 3Ј mouse Xdh and Btn1a1 open reading frames, respec- 4 The B30.2 domain was first defined as a region of similarity in linear sequence encoded by a single exon in BTN and TRIM protein genes (12). Recent structural determinations of PRY-SPRY-19q13.4.1, GUSTAVUS, TRIM21, and SOCS box protein 2 (SSB-2) (20,21,23,24) have shown that this exon encodes the ␤-sandwich. However, in some cases, additional N-terminal ␣-helices, encoded by adjacent 5Ј exons are an integral part of the entire domain and may play a role in the domain's stability. Secondary structure predictions using NPSA and Jpred programs suggest that the equivalent BTN1A1 domain will also contain N-terminal ␣helices. Therefore, the term "B30.2" is used throughout this paper to include both the predicted N-terminal ␣-helices (outside of the historical B30.2 domain) and the ␤sandwich (for hBTN1A1, residues 285-476; see Fig. 2 of Ref. 21 for sequence alignments and further details). 5 The enzyme is expressed in most tissues as xanthine dehydrogenase. Conversion to the oxidase form occurs in the mammary gland during or following secretion into milk (see Ref. 1 for discussion). For convenience, the protein is denoted throughout as xanthine oxidoreductase (XOR), regardless of whether it is in the dehydrogenase or the oxidase form. According to international convention, the genes encoding human and bovine BTN1A1 and XOR are denoted BTN1A1 and XDH, respectively, and the mouse genes are denoted Btn1a1 and Xdh, respectively.
tively. The region of the cDNA encoding the processed form of mBTN1A1 with the N-terminal signal sequence removed was amplified with a second set of primers (supplemental Table 2S), which introduced XhoI and XmaI restriction enzyme sites into the 5Ј-and 3Ј-ends, respectively, and the amplified product was cloned into the pECFP-C1 vector (Clontech Living Color vectors), which had been mutated so as to encode monomerized ECFP (enhanced cyan fluorescent protein). The signal peptide of mBTN1A1 was then inserted 5Ј of the ECFP gene using primers incorporating NheI and AgeI restriction enzyme sites at the 5Ј-and 3Ј-ends, respectively. The completed vector thus encoded the mBTN1A1 signal sequence, followed by the ECFP gene fused to the mature form of mBTN1A1. Deletion constructs (supplemental Table 1S) were prepared by amplifying the appropriate regions of the open reading frame of Btn1a1 using the primers listed in supplemental Table 2S. A vector encoding mXOR with monomerized EYFP (enhanced yellow fluorescent protein) at the C terminus was constructed using primers encoding NheI and XmaI restriction enzyme sites at the 5Ј-and 3Ј-ends, respectively (supplemental Table 2S), and the amplified product cloned into the pEYFP-N1 vector (Clontech). All PCR products were verified by sequencing. Affinity Chromatography of Tissue Extracts on a Glutathione Matrix-C3H mice were killed on day 10 of lactation by asphyxiation in CO 2 , and all of the mammary glands were excised and weighed. Connective and adipose tissue were removed, and three volumes of Tris-buffered saline (TBS; 10 mM Tris-HCl, 140 mM NaCl, pH 7.2) containing proteinase inhibitors (0.1 mM phenylmethylsulfonyl fluoride, 0.1 mM N ␣ -p-tosyl-L-lysine chloromethyl ketone, 0.1 mM N ␣ -p-tosyl-Lphenylalanine chloromethyl ketone, 1.0 ml/100 ml (8.2 IU/ml) aprotinin, and 1 mM ⑀-aminocaproic acid) were added per g of tissue. The tissue was minced finely and homogenized on ice using a Polytron homogenizer by giving five pulses of 10 s each at a speed setting of eight. Homogenates were filtered through two layers of cheesecloth premoistened with TBS, and the filtrates were centrifuged at 1,000 ϫ g av for 10 min at 4°C to obtain sediments, operationally called the nuclear pellet. The postnuclear supernatants were further centrifuged at 100,000 ϫ g av at 4°C for 1 h, and the resulting postmicrosomal supernatants were retained for analysis. Postmicrosomal supernatants (50 mg of protein) were fractionated on 2-ml glutathione affinity columns, which had been saturated previously with either GST-mBTN1A1cyto (test column) or GST (control column). The columns were washed with PBS1, and bound protein was eluted with 10 ml of 10 mM reduced glutathione. Fractions containing protein were combined and analyzed by SDS-PAGE. The entire procedure was repeated a total of three times with similar results.
Cell Expression Assay for Protein Binding-HEK 293T cells were grown to 80 -90% confluence in 20 ϫ 10 ϫ 10-mm LAB-Tek culture chambers and co-transfected, using Lipofectamine 2000 (Invitrogen), with pmXOR-EYFP and a pECFP-mBTN1A1 vector (0.15 g, each), encoding either full-length ECFP-mBTN1A1 or deletion constructs (supplemental Table  1S). Cultures were incubated overnight, and the cells were then washed three times with KHM buffer (20 mM HEPES, pH 6.5, 110 mM potassium acetate, and 2 mM MgCl 2 ). Micrographs of selected cells were recorded using a Leica DMIRE2 fluorescence light microscope employing excitation wavelengths of 438 nm for ECFP and 500 nm for EYFP and separately visualizing the fluorophores between 426 and 450 nm and 488 and 512 nm, respectively. KHM buffer containing 20 M digitonin was then added to permeabilize the plasma membrane. Soluble fluorescent fusion protein was allowed to diffuse out of the cells, and a second set of micrographs was recorded after there was no further change in the fluorescence signal (3-5 min). Transfection efficiencies were in the 50 -70% range for each experiment, and ϳ90% of the doubly transfected cells in any one microscope field displayed the fluorescence patterns described.
Cell Fractionation and Detection of ECFP and EYFP Fusion Proteins-Control and transfected cells, as above, were suspended in 3 mM imidazole buffer, pH 7.4, containing 250 mM sucrose, 0.5 mM EDTA, and proteinase inhibitors, and mechanically disrupted by passage through a 27-gauge needle (10 times). The postnuclear supernatants were fractionated into membrane and cytosol fractions by centrifugation at 100,000 ϫ g av for 1 h and analyzed by Western blotting. Fusion proteins on the blots were detected with cross-reactive antibody to GFP (Invitrogen) diluted 1-2,500-fold, followed with a 1-10,000fold dilution of goat anti-(rabbit IgG) horseradish peroxidase conjugate.
Immunoprecipitation-For immunoprecipitation from HEK 293T cells, cultures were grown in 10-cm dishes and transfected with a total of 12 g of DNA using Lipofectamine 2000. Control and transfected cells were harvested in Dulbecco's PBS (Invitrogen) and the above proteinase inhibitors (PBS2) and mechanically disrupted by passage through a 27-gauge needle a total of 10 times. The nuclei were removed by centrifugation at 1,000 ϫ g av for 10 min at 4°C, and the protein concentration of the postnuclear supernatants was adjusted to 2.4 mg/ml. RIPA buffer (2ϫ) was then added so that the final concentrations of the constituents were protein 1.2 mg/ml, 50 mM Tris-HCl, pH 7.4, 40 mM NaCl, 1.0% (v/v) Nonidet P-40, 0.5% (w/v) sodium deoxycholate, and 0.1% (w/v) SDS. The mixtures were incubated at 4°C with end-over-end stirring for 10 min, the detergent-soluble extracts were recovered by centrifugation at 16,000 ϫ g av for 5 min and then precleared by incubation with protein A-coated beads at 4°C for 1 h (20 l of beads per 1.2 ml of RIPA buffer extract). To ensure optimal precipitation of potential XOR-BTN1A1 complexes, protein A-Sepharose beads (GE Healthcare) were titrated with rabbit polyclonal antibody to mXOR to determine the minimum amount of antibody required to precipitate all of the XOR from the RIPA buffer extracts. Immunoprecipitates were collected on the beads by incubating 20 l of optimally coated XOR-antibody/protein A beads with 1.0 ml of each RIPA buffer extract, overnight, at 4°C with gentle agitation. The beads were then washed three times with ice-cold PBS2, and the bound protein was dissolved in SDS-PAGE buffer at 95°C for 3 min and analyzed by SDS-PAGE and Western blotting. For immunoprecipitations from lactating mammary tissue, total membrane fractions were collected from the postnuclear supernatants of tissue homogenates as described under "Affinity Chromatography of Tissue Extracts on a Glutathione Matrix." After washing once with TBS, the microsomal membranes were extracted with RIPA buffer, and the immunoprecipitations were conducted on the clarified RIPA buffer extracts essentially as described above for HEK 293T cells. The immunoprecipitations were repeated with total membrane fractions from three wild-type (Btn1a1 ϩ/ϩ ) and three Btn1a1 Ϫ/Ϫ C57/Bl6 mice, all at day 10 of lactation.
In Vitro Glutathione Bead Binding Assay for Protein-Protein Interactions-GST fusion proteins at final concentrations of 0.01-2 M were bound to 0.2 ml (packed volume) of glutathione-coated beads in 0.5-1.0 ml of 50 mM Tris-HCl buffer, pH 7.4, for 3 h at 4°C, and the beads were then washed three times with 1.0-ml aliquots of PBS. For the binding assays, the beads were divided into 20-l portions and incubated overnight with XOR purified from bovine milk (bXO) (0.1-2 M, monomeric M r ) in a total volume of 1.0 ml at 4°C, with end-over-end stirring, in 1.5-ml Eppendorf tubes. The beads were then briefly washed three times with 1.0-ml aliquots of ice-cold PBS and heated to 95°C in SDS-PAGE sample buffer. Aliquots of the centrifuged supernatants were analyzed by SDS-PAGE, together with known amounts of fusion protein and bXOR as standards (0.5-8 g). Gels were stained with Coomassie Blue and destained in 10% (v/v) acetic acid, and amounts of fusion protein and bXOR in the samples were determined using Quantity One 1-D Analysis software (Bio-Rad). Binding was assayed either with constant amounts of fusion protein on the beads and increasing amounts of bXOR or with increasing amounts of fusion protein on the beads and saturating amounts of bXOR. Association and dissociation constants (K A and K D ) were determined from the binding data by non-linear regression curve fit (hyperbola) using Prism version 5.0 software (Graphpad Software Inc.).
Electrophoresis and Western Blot-Techniques for electrophoresis and immunoblotting were essentially as described (31)(32)(33), using the ECL blotting kit according to the manufacturer's instructions (Amersham Biosciences/GE Healthcare). Antibodies to mouse (1) and bovine BTN1A1 (34) and bovine XOR (35) have been described previously. Rabbit anti-peptide antibodies to the C terminus of mouse XOR were prepared as described by Kreis (36) using the 21-mer peptide QFTTLCAT-GTPENCKSWSVRI (American Peptide, Sunnyvale, CA) as immunogen.
SPR-The binding of bXOR to GST fusion proteins in real time was evaluated by SPR using a Biacore 3000 Biosensor (GE Healthcare, Biacore, Uppsala, Sweden). Carboxymethyldextran sensor chips (CM5) in flow cells 1 and 2 were coated with 14000 -15000 resonance units (RU) of monoclonal anti-GST antibody by standard amine coupling techniques according to the manufacturer's instructions (Biacore). GST fusion proteins were captured on the chips at a level of 50 RU by injecting an aliquot of GST alone (0.5 g/ml) into flow cell 1, and GST fusion proteins (1.0 g/ml) into flow cell 2, respectively, in HBS-EP buffer (10 mM HEPES, 150 mM NaCl, 3 mM EDTA, and 0.005% (v/v) surfactant p20, pH 7.4, filtered through a 0.2-m filter and degassed before use). The kinetics of binding of bXOR to the immobilized GST fusion proteins were evaluated at a flow rate of 30 l/min at 25°C to minimize mass transport effects. bXOR at concentrations from 0 to 500 nM, in 60-l aliquots, was injected into flow cells 1 and 2, and the association and dissociation rates were recorded for up to 6 min. To assess the effect of pH on binding kinetics, bXOR was diluted in HBS-EP buffer adjusted to pH 6.0, 6.5, 7.0, 7.4, or 8.0 with HCl or NaOH, and the association and dissociation analysis were performed as described above. The effect of salt on binding was similarly determined by using HBS-EP buffer containing variable amounts of NaCl. Chip surfaces were regenerated by removing GST proteins with 50-l injections of 10 mM glycine-HCl buffer, pH 2.2. Sensorgrams were analyzed with BIAevaluation version 3.2 software (Biacore). Signals from the reference channel (GST) were subtracted from the channel containing GST fusion proteins to correct for refractive index changes, injection noise, and nonspecific binding to the reference surface. The signal obtained with a blank injection of HBS-EP buffer alone was then subtracted from the resulting data. Data were globally fitted to the Langmuir model for 1:1 binding and analyzed for significance by Student's t test.
CD Spectroscopy-Purified protein samples (0.5 mg/ml) were analyzed in a 0.2-mm cuvette using a -180 spectrophotometer. Secondary structure was calculated using CDNN software (37), and extinction coefficients of proteins in native and denatured states were calculated, as described (38).
Mass Spectrometry-Proteins were digested in polyacrylamide gels with trypsin, and the peptide products were recovered following the procedure of Shevchenko et al. (39). Solubilized tryptic peptides from each band were analyzed by infusion electrospray ionization on a QStar Pulsar quadrupole-time-offlight tandem mass spectrometer (Applied Biosciences, Foster City, CA). Molecular masses and tandem mass spectra of 1-5 peptides were used to search the NIHPD data base using the MASCOT search program (Matrix Science, London, UK). Peptide identification required probability scores exceeding 95%. The peptides identified from each band were used to identify the corresponding protein.
Assays-XOR for all of the binding studies was purified from bovine milk by the method of Sullivan et al. (35), omitting Triton X-100 from the electrofocusing step to ensure that variable traces of detergent did not compromise the binding assays. XOR in mouse milk lipid (MFGM and 100,000 ϫ g av soluble fractions) was assayed aerobically, in duplicate, in 100 mM sodium phosphate buffer, pH 7.2, containing 10% (v/v) dimethyl sulfoxide and 150 M xanthine at 37°C. The conversion of xanthine into uric acid was followed at 293 nm using a PerkinElmer Life Sciences Lambda 25 spectrophotometer. Enzyme activity (IU) was calculated using a molar extinction coefficient for uric acid under the assay conditions of 12.5 ϫ 10 3 M Ϫ1 cm Ϫ1 . Protein was assayed by the bicinchoninic acid method of Smith et al. (40), using bovine serum albumin as a standard.

RESULTS
To directly explore potential interactions between BTN1A1 and XOR, we initially prepared the cytoplasmic domain of mouse BTN1A1 (mBTN1A1cyto) from yeast as a C-terminal fusion protein with GST (GST-mBTN1A1cyto) (supplemental Table 1S). Analysis by SDS-PAGE and two-dimensional gel electrophoresis showed that the purified protein was of the expected size (56,000) (supplemental Fig. 1Sa) and focused as a series of isoelectric variants with pI values between 5.9 and 6.6 (supplemental Fig. 1Sb) (the theoretical pI for GST-mBTN1A1cyto based on amino acid composition is 5.9). The CD spectrum of mBTN1A1cyto, obtained by subtracting the spectrum of GST from that of GST-mBTN1A1cyto, was consistent with a structure rich in ␤-sheets with no overt structural perturbations (supplemental Fig. 1Sc) (16% ␣-helix, 35% antiparallel sheet, 8% parallel sheet, 17% ␤-turn, and 25% random coil). These values are consistent with a protein in which the B30.2 domain comprises about 75% of the total structure and is predicted to contain one or two ␣-helices and 13-15 ␤-strands (20,21,23).
Using the purified and characterized GST-mBTN1A1cyto, we first asked which soluble proteins from lactating mouse mammary gland will bind to mBTN1A1cyto in vitro. Postmicrosomal supernatants obtained from the mammary glands of day 10 lactating mice (Fig. 1, lane 1) were fractionated on either GST or GST-mBTN1A1cyto immobilized on glutathione-agarose, and the bound proteins were eluted with glutathione. The control GST column failed to reproducibly pull down proteins from extracts, and hence only GST was detected after the glutathione wash (Fig. 1, lane 2; protein 6, white asterisk). Specifically bound proteins from the test column with approximate M r of Ͼ200,000, 150,000, 66,000, and 50,000 were among the most prominent detected in the eluates by SDS-PAGE (Fig. 1, lane 3; proteins 1-3 and 5) together with GST-mBTN1A1cyto (Fig. 1, lane 3; protein 4, white asterisk). These proteins were identified as fatty acid synthase, XOR, serum albumin, and eukaryotic elongation factor ␥1, respectively, based on peptide analysis by tandem mass spectrometry ( Table 1). The 150-kDa protein, identified as XOR, was the most abundant. That GST-mBTN1A1cyto binds to XOR was confirmed by repeating the affinity chromatography with XOR purified from bovine milk (see further data below).
In order to identify the interactive region of mouse BTN1A1 that binds to XOR, we devised two assays, one in which fluorescent fusion proteins of mouse BTN1A1 and mouse XOR were expressed in HEK 293T cells and a second in which mutated GST-mBTN1A1cyto fusion proteins were bound to glutathione-coated beads and used in pull-down assays with bXOR.
For the in vivo cell expression assay, vectors were constructed encoding mouse BTN1A1 fused at the N terminus (after the signal sequence) to ECFP (pECFP-mBTN1A1) and mXOR fused at the C terminus to EYFP (pmXOR-EYFP) (supplemental Table 1S). When HEK 293T cells were separately transfected with these vectors, ECFP-mBTN1A1 was expressed and sorted to the plasma membrane ( Fig. 2A, a) and a variable fraction to an intracellular site. This intracellular site appeared to be associated with or close to the trans-Golgi/trans-Golgi network, because co-expression of mBTN1A1-EYFP and the Golgi marker galactosyltransferase-ECFP showed close association of the two expressed proteins (supplemental Fig. 2SA). In contrast, mXOR-EYFP was expressed in the cytoplasm (Fig. 2A,  c). Both expressed mBTN1A1 and mXOR fluorescent fusion proteins were of the expected size, with little evidence of degradation (shown by immunoblots of cell extracts probed with an antibody to GFP; supplemental Fig. 3S). That mXOR-EYFP  was primarily expressed in the cytoplasm was confirmed by treating the transfected cells with digitonin to permeabilize the cells and remove cytoplasmic proteins (41). Virtually all of the detectable mXOR-EYFP was washed out of the cells (Fig. 2A, compare c and d). As expected for a membrane protein, the distribution of ECFP-mBTN1A1 was unaffected by digitonin treatment (Fig. 2A, compare a and b).
To demonstrate binding between the two proteins, HEK 293T cells were co-transfected with both vectors. A fraction of the expressed mXOR-EYFP co-localized with ECFP-mBTN1A1 at the plasma membrane and in the Golgi region, even after treatment with digitonin ( Fig. 2B and supplemental  Fig. 2SB). That the two proteins were physically bound to each other was confirmed by immunoprecipitating clarified detergent extracts of transfected cells with antibody to mXOR (Fig.  3a). A specific band of protein, which reacted with antibody to mBTN1A1, was detected on blots of immunoprecipitates from doubly transfected cells (Fig. 3a, lane 4) but not from nontransfected control cells (Fig. 3a, lane 1), cells transfected only with mXOR-EYFP (Fig. 3a, lane 2), or cells doubly transfected with mXOR-EYFP and a vector encoding GFP (Fig. 3a, lane 3).
Mutated forms of ECFP-mBTN1A1 were then co-expressed with mXOR-EYFP, to determine which region of mBTN1A1cyto is required for localization of both proteins to cellular membranes. The cytoplasmic domain of BTN1A1 comprises three domains: a stem region closest to the membrane anchor, followed by a B30.2 domain and a C-terminal tail region (schematic in Fig. 4). Progressive deletion of the C-terminal region of ECFP-mBTN1A1cyto (supplemental Table 1S) showed that the entire tail domain could be deleted with no effect on binding to mXOR-EYFP (results summarized in Fig.  4). However, the removal of the most C-terminal 10 amino acids of the B30.2 domain resulted in relocation of the ECFP fusion protein to intracellular vesicles and abrogated binding to mXOR-EYFP (Figs. 2C and 4).
Binding between XOR and the cytoplasmic domain of BTN1A1 was confirmed by glutathione/GST pull-down assays. Fusion proteins of GST joined at the C terminus, either to mBTN1A1cyto or mutated forms of mBTN1A1cyto, were bound to glutathione-coated agarose beads and then incubated with a range of concentrations of bXOR. Binding of bXOR to GST-mBTN1A1cyto was saturable with apparent K D values in the 50 nM range (Fig. 5a and Table 2). Deletion of the tail domain of mBTN1A1cyto had no influence on the binding of bXOR at saturating concentrations (Figs. 4 and 6a, compare   3 and 4). GST-coated beads or uncoated beads served as controls (Fig. 6a, lanes 1 and 2). However, deletion of 10 amino acids from the C-terminal region of the B30.2 domain or deletion of 25 amino acids from the N-terminal region, abrogated binding (Fig. 6a, lanes 5 and 6). Most significantly, the B30.2 domain alone was sufficient for binding to bXOR (Figs. 4 and  6a, lane 7).
To investigate binding specificity within and across species, a series of GST fusion proteins were prepared that incorporated, at the C terminus, the entire cytoplasmic domains or the B30.2 domains of bovine and human BTN1A1 or the entire cytoplasmic domains of human BTN2A1 and BTN3A1. In addition, a protein comprising GST fused to the predicted B30.2 domain of human RoRet (TRIM 38) (42,43) was prepared to test potential binding to a B30.2 domain in a protein outside of the BTN protein family. bXOR bound to the entire cytoplasmic domain or the B30.2 domain of BTN1A1 from mice (see above), cows, or humans (Fig. 6b, lanes 1-3) (data not shown). However, binding was limited to BTN1A1 subfamily members, since no interaction was detected between bXOR and BTN2A1, or BTN3A1 (Fig. 6b, lanes  4 and 5). Furthermore, bXOR did not bind to the B30.2 domain of human RoRet (TRIM 38) (Fig. 6b,  lane 6; control in lane 7). These data demonstrate that XOR does not promiscuously bind to the B30.2 domains of even closely related paralogs of BTN1A1.
We next determined the affinity and kinetics of binding between XOR and BTN1A1 by SPR using GST fusion proteins encoding either bovine or mouse BTN1A1cyto domains as ligands and bXOR as analyte (Table 2 and Fig. 7a, example for GST-bBTN1A1cyto). Dissociation constants (K D values) for GST-BTN1A1cyto from either species, determined at 25°C, were not significantly different (average of 112 and 113 nM, for cow and mouse, respectively) and consistent with values in the 40 -50 nM range for K D values determined by the bead binding assay at 4°C (Table 2). bXOR bound to GST-bB30.2 with statistically similar affinity ( Table 2), confirming that the B30.2 domain is necessary and sufficient for maximal interaction between the two proteins. Binding was pH-dependent (determined either by SPR or the bead binding assay), with no binding detected below pH 6.0 (Fig. 7b), and binding affinities increased 5-7-fold when the salt concentration was lowered from 150 to 20 mM NaCl (Fig. 7c).
In order to determine the stoichiometry of binding between BTN1A1 and XOR, we titrated GST fusion proteins bound to glutathione-coated beads with bXOR, either keeping the amount of GST fusion protein constant and varying the amount of bXOR (binding curve) or varying the amount of GST fusion protein on the beads and adding saturating amounts of bXOR (linear titration) and then assaying for the amount of bXOR bound to the beads by quantitative densitometry (Fig. 5, a and b, and Table 2). Either approach gave BTN1A1/bXOR ratios consistent with equimolar binding (1.0:1.0), with the exception of the linear titration of GST-mBTN1A1cyto, which for unexplained reasons gave a ratio of 1.0:0.36. Binding was equimolar for both full-length mouse and bovine BTN1A1cyto or just the bB30.2 domain. Since bXOR is a homodimer (44), these data indicate that one XOR dimer binds to either two BTN1A1 monomers or one BTN1A1 dimer.
The oligomeric state of BTN1A1 and further estimates for stoichiometric ratios were investigated by gel filtration on Superdex 200 (FPLC) to determine the apparent M r values of the purified proteins, separately, and together as BTN1A1-XOR complexes ( Fig.  8 and Table 3). Both GST-mBTN1A1cyto and GST-bBTN1-A1cyto eluted as major peaks of protein with apparent M r values in the range expected for tetramers (ϳ190,000 and 215,000, respectively) ( Table 3; see Fig. 8, a and b, dotted blue lines, for examples of GST-bBTN1A1cyto). Since GST is a dimer (45), we removed GST from GST-bBTN1A1cyto by digestion with thrombin to determine if the ability to oligomerize is an intrinsic property of BTN1A1cyto. The entire cytoplasmic domain of bBTN1A1, eluted with an apparent M r of 48,000, ϳ80% of the mass expected for a dimer (61,020) (Fig. 8b, dotted red line, and Table 3). However, the purified bB30.2 domain eluted with an estimated M r of 19,500, close to that expected for a monomer (23,437), with a lower amount of potential dimer partially separating as a shoulder on the major peak of protein (Fig. 8b, solid red line, and Table 3). These data are consistent with oligomerization of the fusion proteins to tetramers through separate dimeric interactions between the GST domains and dimeric interactions between the cytoplasmic domains of BTN1A1. Furthermore, interactions between the BTN molecules appear to be predominantly due to interactions within either the stem or tail regions, since the B30.2 domain was largely monomeric.
Complexes of the GST fusion proteins and bXOR, which had been prepared by binding and elution from glutathione-coated beads, eluted as broad peaks of protein with apparent M r of ϳ500,000 with significant peak shoulders of ϳ800,000 (Table 3;

TABLE 2 Summary of binding characteristics for interaction between BTN fusion proteins and bXOR
K D values within each assay are not significantly different (p Ͼ 0.05; MIXED procedure, version 9.1, SAS Institute (Cary, NC)). The number of determinations is shown in parentheses.

Parameter
Binding between bXOR and GST-mBTN1A1cyto GST-bBTN1A1cyto GST-bB30.  Fig. 8a, solid black line, example for GST-bBTN1A1). These data are consistent with stoichiometries of one tetrameric GST-fusion protein (M r ϳ225,000) binding to either one or two dimers of XOR (expected M r values for the complex of 520,000 and 815,000, respectively). To eliminate the confounding effects of dimeric GST, complexes of GST-bB30.2 and bXOR were prepared on glutathione-coated beads, and the GST was removed by digestion with thrombin. Such bB30.2-bXOR complexes eluted with an apparent M r of 344 Ϯ 13 ( Fig. 8c and Table  3) (i.e. with a predicted monomeric bB30.2/XOR ratio of 1:0.85). Taken together, these data confirm that the cytoplasmic domain of BTN1A1 binds to bXOR in equimolar amounts via the B30.2 domain. Finally, to assess the potential function of BTN1A1 and XOR in lactation, we first showed that the two proteins bind to each other in vivo. Detergent extracts of membrane fractions, prepared from the lactating mammary tissue of either wild type (Btn1a1 ϩ/ϩ ) or "knock-out" (Btn1a1 Ϫ/Ϫ ) mice, were immunoprecipitated with antibody to mXOR. A band of protein that reacted with antibody to mBTN1A1 was detected when the immunoprecipitates from wild-type mice were analyzed by Western blot using antibody to mBTN1A1 (Fig. 3b, lane 1), whereas there was no reaction with immunoprecipitates from Btn1a1 Ϫ/Ϫ mice (Fig. 3b, lane 2). ϳ37.5 Ϯ 6.2% of BTN1A1 (mean and S.D. of three determinations) was bound to XOR, as determined by densitometric analysis of the detergent extracts before and after incubation with the antibody-coated protein A beads (Fig. 3b, lanes 3 and 5). Interestingly, XOR was recovered in the immunoprecipitates from knock-out mice, indicating that XOR binds to membranes in the absence of BTN1A1 (Fig.  3b, lane 2).
To determine whether the binding of XOR to BTN1A1 has relevance to the potential function of these proteins in milk lipid secretion, we compared the association of XOR with lipid droplets in the presence and absence of BTN1A1 using wildtype, heterozygous (Btn1a1 ϩ/Ϫ ), or Btn1a1 Ϫ/Ϫ mice. No significant differences in the amount of XOR relative to total protein (determined by immunoblot) or activity (determined as xanthine oxidase activity) were detected in lipid droplets from the milks of wild-type, Btn1a1 ϩ/Ϫ , or Btn1a1 Ϫ/Ϫ mice (Fig. 9). However, although the levels of XOR were comparable between wild-type and Btn1a1 Ϫ/Ϫ mice, there was a highly significant difference in the amount of XOR that remained bound to the MFGM compared with the amount of XOR that was released in a soluble form when membrane was stripped from the core lipid by freeze-thawing the cream samples. Approximately four times as much XOR remained bound to membrane compared with the soluble fraction in wild-type mice, whereas a majority of the XOR in both Btn1a1 ϩ/Ϫ and Btn1a1 Ϫ/Ϫ mice was released into the soluble fraction (Fig. 9, b and c). Furthermore, a fraction of the membrane-bound XOR in wild-type mice separated as aggregates when samples were separated by SDS-PAGE, compared with fractions from Btn1a1 Ϫ/Ϫ mice (bracket in Fig. 9a, i). At least some of these aggregates may consist of complexes of XOR and BTN1A1, because both proteins were detected in the same protein bands by stripping the immunoblots and reprobing them with an antibody to BTN1A1; filled circle in Fig. 9a, ii). Thus, XOR is incorporated into secreted FIGURE 7. Determination of equilibrium and kinetic rate constants for interaction between BTN1A1 proteins and bXOR by SPR. a, representative example of the biosensor analysis of binding of bXOR to GST-bBTN1A1cyto (see Table 2  lipid droplets in the absence of BTN1A1 but is more loosely bound to the MFGM. These data are consistent with the possibility that association of XOR with the MFGM is stabilized by binding to the cytoplasmic domain of BTN1A1, but XOR is initially incorporated into lipid droplets by another mechanism.

DISCUSSION
The data described firmly establish that the cytoplasmic domain of BTN1A1 binds to XOR in vivo and in vitro with relatively high affinity and in a salt-and pH-dependent manner via the B30.2 domain. Binding is stoichiometric with predicted ratios of one XOR homodimer bound to two BTN monomers or one BTN1A1 dimer. Furthermore, binding is independent of species; bovine or mouse XOR binds to the B30.2 domain of mouse, cow, or human BTN1A1. No binding was detected between XOR and even the closely related paralogs, BTN2A1 and BTN3A1, or to the B30.2 domain of the otherwise unrelated RoRet (TRIM 38) protein. Analysis of Btn1a1 Ϫ/Ϫ mice showed that expression of BTN1A1 is required for the stable association of XOR with the MFGM in vivo. These data provide novel insights into the potential roles of BTN1A1 and XOR in milk lipid secretion, the relationship between BTN1A1 and other family members, and the possible functions of BTN1A1 in other cellular contexts.
Based on available structural data, there is an emerging consensus that the B30.2 domain serves as a universal protein binding module (20 -24, 46). The B30.2 domain is widely distributed throughout the proteome (25,47,48), occurring within and outside of the BTN protein family and the immune system. The most conserved region of the domain comprises a core ␤-sandwich, with variable loops linking adjoining ␤-strands. In all B30.2 domain structures that have been characterized with bound ligand (21)(22)(23), the interacting partner binds, via hydrogen bonding and hydrophobic interactions, to one or more discrete pockets formed from the variable loops. Comparison of the linear sequences of these B30.2 domains with those of BTN1A1 and other family members (BTN2A1, BTN2A2, BTN2A3, BTN3A1, BTN3A2, and BTN3A3) reveals a similar predicted pattern of conserved ␤-strands and variable loops (21,23). XOR is therefore likely to interact with BTN1A1 through analogous binding motifs, which are not conserved in BTN2A1 or BTN3A1 (Fig. 6b). Similar binding characteristics are also to be expected because of similarities in the dissociation constants (K D values) for the binding of ligand to BTN1A1 (27 nM in 20 mM NaCl), GUSTAVUS (40 nM in 150 mM NaCl) (21), and TRIM 21 (37 nM in 50 mM NaCl) (23) and the lower K D values (higher affinities) determined at lower salt concentrations (Fig. 7c) (23). However, because of the presumed discontinuous nature of binding motifs in the linear sequence, it is not possible to predict which residues in BTN1A1 function in binding. Resolution of these structural issues will require analysis by x-ray crystallography.
In vitro bead binding assays showed that bXOR binds to the B30.2 domain of BTN1A1 in a monomeric ratio of 1:1 (Fig. 5 and Table 2), which is in agreement with a global best fit of the SPR data using the 1:1 Langmuir model for binding (Fig. 7a,  black lines). These data are also consistent with ligand binding ratios for other B30.2 domain-containing proteins, including GUSTAVUS and TRIM21 (21,23). Since bXOR is a homodimer (44), a maximum of two B30.2 domains are predicted to bind to one native XOR molecule. In vitro analysis by FPLC showed that bBTN1A1cyto can form dimers. Dimerization is most likely through the stem region, which is predicted to have  ␣-helical character, since the B30.2 domain was largely monomeric (Fig. 8b), and the cytoplasmic tail domain is probably disordered because of intermittent P residues (GlobPlot TM prediction). Therefore, XOR has the potential to bind either one or two BTN1A1 monomers or one dimer in a pH-dependent manner.
These binding data are relevant to several potential functions of BTN1A1 in vivo. In the context of milk lipid secretion, fat droplets are coated with apical membrane as they are expelled from mammary cells, and the resulting MFGM in many species contains large amounts of BTN1A1 and XOR (2,49,50). The observation that XOR only binds to BTN1A1, the BTN family member that is highly expressed in the lactating mammary gland, and not to other BTN paralogs provides further evidence that these two proteins function together in lactation. At least two models have been proposed to explain how BTN1A1 and XOR may function in lipid secretion (2,28,51).
In one model, BTN1A1 as a transmembrane protein binds to XOR, which serves as a linker protein to other proteins on the lipid droplet surface, including adipophilin (2). Concerted interactions between such BTN-XOR complexes and homophilic adhesive interactions between BTN1A1 molecules are proposed to proceed as a wave around the droplet and lead to the uniform coating of the droplet with membrane and to the eventual expulsion of the lipid droplet from the cell. At some point during or after secretion, associations between BTN1A1 and XOR culminate in formation of a characteristic protein coat complex that can be identified by freeze-etch electron microscopy (52,53). Knock-out of BTN1A1 expression in Btn1a1 Ϫ/Ϫ mice (1) or reduction of the amount of XOR in one strain of Xdh ϩ/Ϫ mice (27) eliminates or reduces the possibility of interactions between BTN1A1 and XOR, leading to the secretion of abnormally large and unstable lipid droplets. Thus, BTN1A1 is seen to function as a transmembrane scaffold, which binds to XOR and other proteins on the droplet surface to ensure the coordinated and regulated secretion of lipid droplets into milk.
The binding data from this current study confirm that BTN1A1, through interactions with the B30.2 domain, binds with high affinity to XOR and that BTN1A1 molecules can bind to each other. However, the observation that XOR is secreted with lipid droplets in the absence of BTN1A1 (Fig. 9) is not compatible with the proposal that BTN1A1 directly recruits XOR into the lipid droplet (2) and suggests that XOR is initially recruited into the MFGM by other means (e.g. by being carried to the apical plasma membrane on the surface of intracellular lipid droplets independently of BTN1A1 or by binding to other proteins as well as BTN1A1) (see Refs. 51 and 54 for discussion).  (a and b) and an enzyme assay (c). a (i), membrane-bound (m) and soluble (s) XOR detected by Western blot. a (ii), the same immunoblots from a (i), stripped and reprobed with an antibody to mouse BTN1A1. Aggregates of XOR and BTN1A1 are indicated by brackets in i and ii, respectively, and potential complexes of XOR and BTN1A1 are shown by filled circles. b, left, comparative amounts of total XOR determined by densitometry of stained immunoblots; right, ratio of membrane-bound to soluble XOR. c, left, comparative amounts of total XOR activity; right, ratio of membrane-bound to soluble XOR activity. For each genotype, milk lipid samples from three mice on day 10 of lactation were analyzed in duplicate. Data within each panel with different letters are significantly different from each other (p Ͻ 0.05) using Student-Newman-Keul's multiple comparison test.
Interestingly, XOR was recovered in crude membrane fractions (Fig. 3b, lane 4) and the MFGM (Fig. 9a, i) in the absence of BTN1A1, indicating that XOR can bind to other membrane components. However, such associations may be artifactually induced in vitro, because XOR has the potential to bind to glycoaminoglycans on exoplasmic membrane surfaces (55) when it is released into the soluble phase from total cell lysates or homogenized milk lipid droplets. Whatever the case, at some point during the secretion process, association of XOR with the MFGM becomes dependent on the presence of BTN1A1, because its elimination in null mice leads to the release of a substantial amount of soluble XOR from isolated milk lipid globules (Fig. 9). A full description of how the MFGM is assembled will require the application of live cell imaging techniques, more refined analysis of the distribution of XOR and BTN1A1 in secretory epithelial cells, and identification of additional protein binding partners and regulatory molecules. Such additional proteins may include fatty acid synthetase, the ␥-subunit of elongation factor 1 ( Table 1), and other GTP-binding proteins (56).
In a second model, based on freeze-fracture labeling techniques, Robenek et al. (28) proposed that secretion of lipid droplets is solely mediated by adhesive interactions between BTN1A1 molecules in the apical plasma membrane and BTN1A1 molecules on the lipid droplet surface and that XOR plays no role at any stage, because it does not colocalize with BTN1A1 in the freeze-fractured faces of secreted lipid droplets. This model is not consistent with the data in this paper showing that BTN1A1 binds tightly to XOR and that the association of XOR with the MFGM is dependent upon the presence of BTN1A1 (Fig. 9). Furthermore, it is inconsistent with the data of McManaman et al. (54) showing that at least a fraction of BTN1A1 may bind to XOR by disulfide bonds in isolated MFGM. Also, the model assumes that BTN1A1, which is an integral membrane protein, binds with unconventional topology to the phospholipid monolayer surrounding cytoplasmic lipid droplets.
Historically, BTN1A1 has been considered to be a mammary-specific protein (15), and its functions have been thought to be limited to potential roles in milk lipid secretion, a process unique to the mammary gland (2). However, extensive tissue surveys by more sensitive histochemical (BTN1A1 on the Human Protein Atlas site on the World Wide Web) and reverse transcription-PCR techniques 6 indicate that BTN1A1 is expressed in other cells and tissues, including spleen, thymus, esophagus, and lymphoid cells in the tonsil. Since none of these other tissues secrete membrane-coated lipid droplets, BTN1A1 must function in other capacities. Two possibilities deserve special consideration. First, BTN1A1 and its binding partner, XOR, possess all of the hallmarks of a novel signaling system; BTN1A1 is an integral protein with IgG folds in the exoplasmic domain, which could bind ligand, and a cytoplasmic B30.2 domain, which binds XOR in a pH-dependent manner (this study). In the appropriate physiological contexts, XOR, either as a dehydrogenase or an oxidase may generate short lived reac-tive species, including H 2 O 2 , superoxide radical, and NO, which may act on downstream targets and regulate gene expression (57). Second, BTN1A1 and XOR (58, 59) may function as components of the innate immune system by generating antibacterial reactive oxygen and nitrogen species following engagement of BTN1A1 with immune receptors or bacterial surfaces. In this context, milk contains large numbers of exosomes, small immunoregulatory 30 -100-nm vesicles that contain significant amounts of both BTN1A1 and XOR (60), which may serve to protect the lactating mammary gland from bacterial pathogens.
The SPRY and B30.2 domains have been identified in at least 86 human genes (25). In other organisms, well over 2,000 sequences encoding B30.2, SPRY, and PRY domains have been cataloged, including several hundred NOD-like immune receptors in teleost fish (48). Although XOR only bound to the B30.2 domain of BTN1A1 in our limited survey of potential binding partners, a more extensive screen is warranted, to determine whether XOR binds to the B30.2 domain in other proteins within and outside of the immune system.