Human Lymphocytes Interact Directly with CD47 through a Novel Member of the Signal Regulatory Protein (SIRP) Family¹

The expression of signal regulatory protein α (SIRPα)³ (CD172) in humans and its homologues in other mammals has been observed on myeloid, neuronal, and endothelial cells. SIRPα has been cloned several times in different species leading to a variety of names, including SIRPα, SHPS-1, MyD-1, P84/BIT, and MFR (1, 2, 3, 4, 5, 6). The role of SIRPα is generally assumed to be an inhibitory one, mainly because of its interaction via ITIM motifs in its cytoplasmic tail with the Src homology 2 domain-containing phosphatase 1 (Shp1) and Shp2 protein tyrosine phosphatases (1, 2, 7, 8), and there are functional data supporting this (9). SIRPα can also have negative effects on the expression of inflammatory cytokines, especially TNF-α (10, 11). The extracellular ligand for SIRPα is CD47, an unusual five-pass transmembrane protein with a single Ig-like domain (12, 13). CD47 itself has been ascribed a wide variety of functions and is ubiquitously expressed. It interacts in a cis manner with cell surface integrins and was originally termed integrin-associated protein (14, 15). There is evidence that it affects cell behavior through an interaction with heterotrimeric G proteins (16, 17). CD47 has been shown to have effects on integrins, migration, phagocytosis, IL-12R expression, and T cell activation and conversely on anergy or cell death (18, 19, 20, 21, 22, 23, 24, 25, 26). Surprisingly, considering its ubiquitous expression and the multiple effects that CD47 ligation can involve, CD47^−/− mice are viable and healthy. An obvious phenotype is only apparent when mice are shown to succumb to bacterial infection more quickly than their wild-type relatives. This seems to be due to defects in neutrophil migration (27). CD47 has also been postulated to act as a marker of self, as CD47^−/− RBC are rapidly phagocytosed when injected into wild-type mice (28).

The role of CD47/SIRPα is further complicated by the finding that thrombospondin-1 has been shown to interact with CD47 (29) via the C-terminal region. However, many of the effects seen may also be mediated by thrombospondin-1 adhesion with integrins or simultaneous ligation with integrins and CD47 (30). It is therefore important to discover other protein interactions in this system and their affinities and tissue distribution. In contrast to SIRPα, SIRPβ appears to exert activatory stimuli by virtue of interacting with the DAP12 adapter protein via a charged lysine in the SIRPβ transmembrane region. DAP12 is thought to function via the binding and activation of Src family kinases such as Syk through ITAM motifs (31). However, despite the sequence similarity between SIRPα and SIRPβ, SIRPβ · Fc fusion proteins do not bind to CD47 expressed on the cell surface (32). A third SIRP-related protein has been suggested at the cDNA level (33). In this study we show that SIRPγ arises from a unique gene that, at the amino acid level, is approximately equally conserved to both SIRPα and SIRPβ. However, although the expressed protein has a truncated cytoplasmic tail, it is unlike SIRPβ in that it does not require DAP12 for surface expression and binds to CD47. Specific mAb show that it also has different cellular distributions to SIRPα and -β. Taken together, we suggest that the protein be named SIRPγ to differentiate it from SIRPα and SIRPβ, but still indicate that it belongs to the same closely related protein family. Finally, we also show that the differences in affinity of interaction with CD47 may control the functional outcome of an interaction between different SIRPα and SIRPγ.

The sequence representing the three Ig-like domains that comprise the extracellular region of the human SIRPγ was amplified by PCR using human peripheral blood leukocytes cDNA as a template. The oligonucleotides ATGATTCAGCCTGAGAAG (sense) and TCAGGTCTTCTGCTTCCAG (antisense) were designed using the known SIRPγ sequence (EMBL accession no. AB042624). For soluble SIRPγ the antisense primer used was GTAGCATCTGAGCTCTGG. The products were blunt end ligated into pCR2.1 (Invitrogen Life Technologies, Carlsbad, CA). For chimeric soluble proteins, the SIRP sequence was ligated into a pEFBOS-XB expression vector containing rat CD4 Ig-like domains 3 and 4 (henceforth referred to as CD4) and a biotinylation motif (34, 35). For construction of surface-expressed SIRPγ bearing the FLAG epitope, the N terminus of the three extracellular Ig-like domains of SIRPγ was ligated (SalI-BamHI) into a CD4Lflag-pEFBOS construct (using a rat CD4 leader (CD4^L) ending … VVTTQG, followed by the FLAG epitope DYKDDDDKST). Protein expression was detected with anti-FLAG-M2 mAb (Sigma-Aldrich, Poole, U.K.). For chimeric constructs containing additional leader peptides, either human SIRPα (SIRPα^L) or rat CD4^L sequences were ligated onto the 5′ end of SIRPγ. The resulting SIRPα^L chimera had the following sequence at the join site: EEELQMIQP (SIRPα^L sequence underlined). To produce the SIRPα^L/SIRPγ construct, the SIRPα signal peptide was joined to SIRPγ using an endogenous SIRPα PstI site (the PstI site was inserted into the SIRPγ sequence using PCR). Soluble recombinant proteins with the rat CD4 domains were purified by OX68 mAb affinity chromatography (34).

Surface plasmon resonance measurements were obtained using a BIAcore 2000 biosensor instrument (BIAcore, Stevenage, Herts, U.K.) using CM5 research grade chips. The streptavidin was immobilized directly via amine coupling in 10 mM sodium acetate, pH 4.5. Equilibrium affinity and kinetic measurements were conducted in 10 mM HEPES (pH 7.4), 150 mM NaCl, 3 mM EDTA, and 0.005% surfactant P20 at 37°C using short injection times of 3 s (5 μl at 100 μl/min) to minimize the contribution of any aggregated material. For equilibrium binding, immediately before the experiment, purified CD47 · CD4 was size fractionated to exclude any aggregated protein. Increasing and decreasing concentrations of monomeric CD47 · CD4 were passed over SIRPα, SIRPγ, or control CD4 (all coated on chip at 1000 response units (RU)). For off-rate determinations, CD47 · CD4 (40 μM) was passed over immobilized SIRPγ · CD4 (at 1600 and 800 RU), or control CD4 (1600 RU). K_d values were obtained by both nonlinear curve fitting of the Langmuir binding isotherm and Scatchard transformations of the binding data (Origin software, OriginLab, Northampton, MA). k_off values were determined by fitting a first-order exponential decay curve to normalized data after subtraction of the negative control values.

Six-week-old male BALB/c mice were immunized s.c. with 10–20 μg of purified SIRPγ · CD4 in CFA and then with IFA. A mouse generating good immune responses to the immunogen was boosted, and 4 days later the spleen was removed, and hybridomas were generated using standard procedures by fusing with the NS-1 cell line. Hybridoma supernatants were initially screened by ELISA, using SIRPγ · CD4, human SIRPα.Fc (13), human SIRPβ · Fc, or CD4 to eliminate hybridomas with cross-reactivity to rat CD4, human SIRPα, and SIRPβ. The remaining hybridoma supernatants were screened for the ability to stain SIRPγ-transfected 293T cells. Four hybridomas recognizing SIRPγ were named OX116–OX119 and were recloned. All mAb were of the mouse IgG1 isotype.

PBMC were purified from blood of healthy volunteers on Ficoll-Paque density gradient (Amersham Biosciences, Arlington Heights, IL) and cultured in tissue culture medium (RPMI 1640, 10% FCS, 2 mM glutamine, 50 μM 2-ME, 100 U/ml penicillin, and 100 μg/ml streptomycin). 293T cells were transiently transfected with human SIRPγ · pEFBOS, human SIRPβ · pcDNA6 (gift from M. Colonna, Washington University School of Medicine, St. Louis, MO) and human FLAG-tagged DAP12 · pREP10 (gift from J. Sedgwick, DNAX, Palo Alto, CA) using FuGene transfection reagent (Roche, Indianapolis, IN) and the manufacturer’s protocol. Surface expression of transfected cells was analyzed after 48 h in culture.

Cells were labeled by indirect immunofluorescence at 4°C in the presence of 10 mM sodium azide according to standard procedures and were analyzed by flow cytometry on a FACScan (BD Biosciences, Mountain View, CA). All mAb, unless otherwise stated, were obtained from BD Pharmingen (San Diego, CA) and were directly conjugated to FITC or PE. Annexin-FITC (Roche)/propidium iodide (Sigma-Aldrich) staining was performed according to the manufacturer’s protocol. Two-color staining was performed using biotinylated SIRPγ mAb OX116 or OX119 with streptavidin-PE and a second, directly FITC-conjugated mAb (CD3, CD4, CD8, CD19, and CD25). Isotype-matched biotinylated or fluorochrome-labeled mAb were used as controls. Generation of multivalent SIRP reagents using avidin-coated fluorescent beads (Sphero beads; Biotechnologie, www.kisker-biotech.com) and subsequent binding assays were previously described (36, 37). For lymphocyte activation, cells were resuspended in 200 μl of RPMI 1640, 10% FCS, and 50 μM 2-ME with antibiotics in flat-bottom, 96-well tissue culture plates at 10⁶/ml. At the indicated time points, the cells were frozen in 10% DMSO/90% FCS at −80°C. All samples were thawed and resuspended in PBS with azide before analysis as described above.

Jurkat or U937 cells at 4 × 10⁵/well in 96-well plates were cultured for 3 h with the indicated mAb or fusion proteins bound to avidin-coated fluorescent beads (Sphero) in tissue culture medium. Ten microliters of beads (streptavidin-coated, preincubated with 20 μl of biotinylated fusion protein) or mAb at 5 μg/ml (final concentration) were added per well. The mAb used were control mAb (OX2), CD3 (mAb OKT3), CD47 (mAb 1796; Cymbus Biotechnology (Southampton, Hants, U.K.) or mAb MCA911 (Serotec, Oxford, U.K.)), and CD51/61 (BD Pharmingen). For blocking assays soluble CD47 · CD4 or CD4 alone as a control was added at 100 μg/ml before addition of beads. Cells were pelleted and washed in binding buffer (10 mM HEPES/NaOH (pH 7.4), 150 mM NaCl, 5 mM KCl, 1 mM MgCl₂, and 1.8 mM CaCl₂). Cells were then incubated with annexin-PE/7-aminoactinomycin D (7-AAD) according to the manufacturer’s protocol (BD Pharmingen) and analyzed on a FACScan (BD Biosciences, Mountain View, CA).

PBMC were isolated by Ficoll density gradient, surface-biotinylated, and lysed at 2 × 10⁷/ml in lysis buffer (150 mM NaCl, 50 mM Tris-HCl (pH 7.4), 1 mM EDTA, 1% Nonidet P-40, 1 mM PMSF, 50 mM benzamidine, 1 mg/ml leupeptin, 1 mg/ml aprotinin, and 1 mg/ml pepstatin) for 30 min, and insoluble material was spun down (13 krpm, 10 min). For all preclearing and immunoprecipitate steps, 10⁷ goat anti-mouse conjugated Dynal beads (Dynal Biotech, Great Neck, NJ) preincubated with 3 μg of mAb/2 × 10⁷ cells were used. After immunoprecipitation, Dynal beads were resuspended in reducing sample buffer by boiling for 5 min and loaded onto a 4–12% gradient polyacrylamide gel (NOVEX, San Diego, CA). After electrophoresis, the gel was blotted onto polyvinylidene difluoride, which was then blocked in 1% BSA before incubation with extra-avidin conjugated to peroxidase and developed with ECL reagents (Amersham Biosciences).

Analysis of the genomic sequence using the University of California Santa Cruz human genome browser (http://genome.ucsc.edu) showed that SIRPγ is present on chromosome 20p13 and has six possible exons (Table I). This also shows a predicted signal peptide that shows a high degree of homology with SIRPβ peptide (Fig. 1,a). There are at least two transcript variants of SIRPγ shown by established sequence tags (ESTs). One form has exons 1–6 (variant 1), whereas the other has exons 1, 2, 5, and 6 (variant 2). Both variants 1 and 2 are predicted to have the same transmembrane domain. Variant 1 consists of an N-terminal, Ig-like V domain and two Ig-like C domains, whereas variant 2 only contains the N-terminal, Ig-like V domain (Fig. 1,b). The existence of both forms (at least at the mRNA level) was confirmed using PCR on cDNA from PBLs (data not shown). However, transient expression of constructs containing variant 2 failed to produce surface protein, and immunoprecipitation data with available mAb (see below) did not indicate the presence of a protein corresponding to the predicted m.w. of variant 2. Therefore, surface expression of this protein by leukocytes seems unlikely. SIRPγ variant 1 is highly related to SIRPα and SIRPβ, as shown by the comparison of amino acid sequences in Fig. 1 a. Over the Ig-like regions, there is an equal level of conservation (79%). Very low levels of conservation were seen in the transmembrane regions and cytoplasmic tail with either SIRPα or SIRPβ.

Analysis of the human genomic sequence also showed that SIRPγ, SIRPβ, and SIRPα are situated in close proximity to each other, over a combined distance of ∼377 kb. There are three other putative genes in the SIRP cluster (Fig. 2,a). A comparison of domain structure of all SIRP genes is shown in Fig. 2 b. Two of these genes show a relationship with the SIRP genes and therefore appear to be other members of the SIRP gene family, although more distantly related. Accession number AAH33502 represents the first and is reported as an expressed cDNA. This has a deduced protein sequence of 197 aa, which includes a leader sequence and one predicted Ig-like domain. It has 48% amino acid identity with SIRPγ in the Ig-like domain. The second is represented by GeneScan prediction NT_011387.32. This is not well supported by EST data, but does consist of an open reading frame predicted to encode a peptide of 651 aa. This has a putative leader sequence and five predicted Ig-like domains. Most of the conservation with SIRPs resides in the most C-terminal, two Ig-like domains, which both have ∼60% amino acid identity with the first N-terminal, Ig-like domain of SIRPα. There are EST data for mouse and rat homologues (SWISSPROT: Q8BJ958 and XP_230596.1) that show closest similarity to NT_011387.32 and whose predicted transmembranes are similar to SIRPβ, including a single lysine residue that may indicate DAP10 or DAP12 association. The amino acid identities of the other NT_011387.32 Ig-like domains with SIRPα are much lower (∼30%). The third putative gene is represented by five expressed sequences, namely, BX096358, AA398753, AA292852, AA398753, and AA292852, and does not share any sequence conservation with the other SIRP genes.

Human SIRPγ showed a high degree of conservation (78% amino acid identity) with a partial cDNA from another large mammal, cattle (Fig. 1 b). Although encoding a stop codon, the cDNA does not encode a transmembrane region and may encode a soluble protein. Database searches for genomic and EST data have shown no evidence yet of SIRPγ in mice or rats.

The published SIRPγ sequence appeared truncated at the N terminus (amino acid sequence commenced MIQP..) (33). SIRPγ was amplified by PCR from leukocyte cDNA using primers based on this sequence, but protein could not be expressed (data not shown). A leader sequence from SIRPα or CD4 was inserted, and SIRPγ was expressed both at the cell surface, as detected by the FLAG label, and with SIRP mAb B1D5 (gift from H. G. Buhring, University of Tubingen, Tubingen, Germany), and as a soluble chimeric construct with rat CD4 domains 3 and 4 (data not shown). Recently, a full-length protein sequence for SIRPγ has been described (SWISSPROT accession no. Q9P1W8) that confirmed that the original cDNA sequence lacked a leader peptide. This sequence also confirmed that the constructs we produced with the SIRPα signal peptide did not confer any SIRPα-specific amino acids to the predicted mature SIRPγ protein that could have affected the data.

Given the sequence similarity between SIRPα and SIRPγ, we wanted to establish whether there is also a direct interaction between SIRPγ and CD47. The binding of purified soluble recombinant CD47 · CD4 to immobilized SIRPγ was analyzed using surface plasmon resonance with a BIAcore. Similar amounts of biotinylated SIRPα · CD4, SIRPβ · CD4, SIRPγ · CD4, and control CD4 were immobilized on separate flow cells via binding to covalently attached streptavidin. Fig. 3 shows that SIRPγ, like SIRPα, does bind to CD47, but the lower equilibrium binding indicates that it interacts with a lower affinity than SIRPα. SIRPβ gave no signal above the control CD4 background level, indicating that it does not react with CD47, in agreement with cell binding data (32) (E. Vernon Wilson, D. Voulgaraki, M. H. Brown, A. N. Barclay, and G. Brooke, unpublished observations).

To measure the affinity of the interaction, purified human CD47 · CD4 was fractionated by gel filtration to ensure that it was monomeric. As shown in Fig. 3, CD47 · CD4 bound specifically to SIRPγ in a dose-dependent manner. Nonlinear curve fitting of the data collected at 37°C produced a K_d of 23 μM (Fig. 3). This is within the range of normal leukocyte cell surface interactions between proteins, which range from 0.1–100 μM (38). Scatchard analysis gave a similar result (Fig. 3). For comparison, CD47 · CD4 was also passed over SIRPα · CD4 bound to the chip, which produced a K_d of 2 μM, also shown on the same Scatchard plot (Fig. 3). Kinetic analysis of the dissociation rate with two different levels of bound SIRPγ · CD4 at 37°C showed a rate constant (k_off) of 3.1 s⁻¹ for the higher level of bound SIRPγ · CD4 (1600 RU) and a k_off of 5.3 s⁻¹ for the lower level (800 RU; Fig. 3). This increase in the apparent k_off most likely reflects the existence of rebinding effects after dissociation, which result in underestimation of the true k_off. Thus, the true k_off is ≥5.3 s⁻¹. This fast off-rate is comparable to other low affinity interactions, such as the CD2-CD58 interaction (39), and explains why the affinity is relatively low for SIRPγ. From the dissociation rate constant k_off and the K_d, the association rate constant was calculated to be ∼2 × 10⁵ M⁻¹s⁻¹, and from the k_off, the half-life was estimated to be 0.1 s. In comparison, measurements made at the same time show that the affinity for human SIRPα-CD47 interaction was K_d = 2 μM (data not shown) and the k_off = 1.6 s⁻¹ (0.4 s half-life; Fig. 3). The k_off is independent of concentration and is the same as that determined previously (13). However, the K_d value for human SIRPα-CD47 was lower (8 μM) than that reported in this study (2 μM), probably due to a higher proportion of active CD47 in this preparation. The comparison shown above is valid and is the best estimate, because the same CD47 preparation was used for both SIRPα and SIRPγ; thus, SIRPγ has ∼10-fold lower affinity for CD47 than SIRPα.

Recombinant soluble SIRPγ · CD4 protein was used to raise mAb in mice. The resulting mAb were screened against the three known SIRP proteins by ELISA. Four mAb (mAb OX116–OX119) were picked for further study by flow cytometry using transfected cells. None of the mAb stained untransfected 293T cells (data not shown). From the mAb obtained, only one, mAb OX119, appeared to be largely specific for SIRPγ (Fig. 4), although some marginal binding to SIRPβ transfectants was observed. The specificity of OX119 was confirmed by immunoprecipitation, because this showed a single immunoprecipitated band of ∼55 kDa (Fig. 4). The positive control mAb SE5A5 bound with a band of the expected size for SIRPα and also a particularly prominent band corresponding to SIRPβ. Studies with transfectants confirmed that this mAb binds SIRPα and SIRPβ, but not SIRPγ (Table II), although the prominence of the SIRPβ band was unexpected. Data produced from transfected cells with some of the mAb may not reflect the specificity of staining on ex vivo cells, because although mAb OX116 apparently stained SIRPα-transfected 293T cells, it did not cross-react with SIRPα by ELISA and did not stain the U937 cell line (that expresses SIRPα and SIRPβ) or PBLs. It may be that alternative glycosylation by some cell lines (e.g., 293T) can alter the specificity of some mAb. Of the other mAb, OX117 cross-reacted with SIRPβ (SIRPβ/DAP12-transfected RBL cells were gifts from E. Tomasello and E. Vivier, Centre d’Immunologie de Marseille-Luminy, Marseille, France), and OX118 cross-reacted with SIRPα (Fig. 4). A summary and comparison of SIRP mAb specificity is shown in Table II (other mAb were SE5A5 and B1D5 (32) and ILA24 (3)). The expression of SIRPγ was equally expressed in the presence or the absence of DAP12, in marked contrast to SIRPβ, which requires coexpression of DAP12 for surface expression (Fig. 4).

The OX119 mAb stained the majority (70%) of human PBMC (Fig. 4), and this was repeated in several different unrelated individuals (data not shown). Staining on peripheral blood myeloid cells was generally negative/very low (Fig. 4). The marginal level of staining seen in some cases (data not shown) was assumed to be cross-reactivity due to high levels of SIRPβ on these cells, which were negative for SIRPγ cDNA by PCR (data not shown). SIRPα/SIRPβ staining (mAb SE5A5) is shown for comparison. Within the lymphocyte population, the majority of T cells expressed SIRPγ (85% of CD3 cells), with 92% of CD4 and 71% of CD8^high T cells (excluding CD8^low NK cells) staining. The expression level of SIRPγ on CD25⁺ T cells is equivalent to that on CD25⁻ T cells. There was no change in relative SIRPγ proportions based on CD45Ra/Rb expression (results not shown). There was, however, a reduction in the overall percentage of SIRPγ-expressing cells after 2-day activation of PBMC with Con A, although the expression levels remained approximately the same on the remaining positive cells (Fig. 4). Approximately 10–20% of CD19 B cells were labeled, with some variation between individuals (Fig. 4).

To show that SIRPγ can influence the behavior of CD47 on T cells, an apoptosis assay was used in a similar manner to that described by Manna and Frazier (40). To overcome the low affinity of the SIRP-CD47 interaction and achieve binding of recombinant proteins to cells, a high avidity system of biotinylated SIRPγ fusion protein bound to avidin-Sphero beads was used. These, or mAb (also bound to beads as a control), were incubated with Jurkat or U937 cells in tissue culture medium for 2–3 h at 37°C before levels of apoptosis were measured with annexin. As expected, the CD47 mAb (clone 1796; Cymbus Biotechnology) induced high levels of apoptosis. The results show that SIRPγ and SIRPα induced apoptosis at much higher levels than with control beads in both Jurkat and U937 cells (Fig. 5,a). The slight reduction in apoptosis induction by SIRPγ vs SIRPα is consistent with lower affinity binding to CD47. Levels of apoptosis induction were comparable to the CD47 mAb itself (Fig. 5, b and c). The CD47 mAb MCA911-coated beads show that not all CD47 mAb are capable of inducing apoptosis as has been previously described (40). Abs that cross-linked the CD47-associated integrin CD51/61 had no effect on apoptosis, indicating a lack of direct involvement by CD47-associating integrins. The effect could also be specifically blocked by either excess soluble CD47 protein (CD4 chimera) or the OX119 mAb (Fig. 5 c).

Human PBMC were stimulated with Con A for up to 4 days. To ensure consistency in the results, cells were frozen at each time point, and all staining was conducted at the same time and with the same preparations of reagents. To measure binding of SIRP proteins with CD47, the high avidity system of avidin-coated, fluorescently labeled beads saturated with biotinylated SIRPα · CD4, SIRPγ · CD4, or CD4 control was used. SIRPγ binding to PBMC was mediated solely by CD47, as a CD47 mAb blocked binding of the beads to the cells (Fig. 6,a). The results presented in Fig. 6 show that PBMC at time zero had weak binding to SIRPα or SIRPγ fluorescent beads despite expression of CD47. Upon Con A activation, PBMC expression of CD47 increased ∼2-fold until day 2 before falling to lower levels by day 4. Despite this modest increase in CD47 expression, binding with both SIRPα and SIRPγ beads increased dramatically. However, SIRPγ binding decreased more quickly once CD47 levels started falling. Thus, the reduction in affinity between SIRPγ and SIRPα does not alter the overall amount bound at optimum levels of CD47, but, instead, changes the threshold of binding at intermediate concentrations of CD47. Thus, a threshold of CD47 expression is required before SIRPγ beads can remain bound. Around the threshold concentration, small changes in CD47 lead to dramatic differences in SIRPγ binding, as shown in Fig. 6. The higher affinity SIRPα interaction allows a lower threshold, leading to smaller changes in binding relative to the initial binding seen on day 0. These findings are in agreement with other low affinity cell surface molecular interactions, such as that between CD2 and CD48 (36).

These results show evidence for the expression and function of a novel member of the SIRP gene family in humans. The increased number of SIRP genes expressed at the cell surface and the finding that most mAb cross-react on different SIRP gene products means that previous mAb binding data need to be reassessed to determine whether functional effects were unique to a particular SIRP gene. A summary of the current known specificity of various SIRP mAb is shown in Table II. We have shown that the SIRP gene family contains three highly related genes, all of which can be expressed. There were two additional genes with some sequence identity to the SIRPs, although these have not been proven to exist at the protein level.

Of the known expressed SIRP genes, all have very different intracellular signaling potential. SIRPα seems to signal via interaction of phosphotyrosines in cytoplasmic ITIM motifs with the Shp-1 and Shp-2 protein tyrosine phosphatases (2), whereas SIRPβ can impart signals via interaction with the DAP12 adapter (41), possibly through subsequent activation of Src family type tyrosine kinases. SIRPγ is unlikely to transmit a signal to the cytoplasm, because it has a negligible cytoplasmic tail with no tyrosines capable of being phosphorylated and no charged transmembrane amino acids capable of interacting with other transmembrane signaling proteins, such as DAP12.

SIRPγ was originally described as SIRP-b2 (33), which suggests that it is a subtype of the SIRPβ gene. We have termed it SIRPγ, because it is a separate gene product distinct from SIRPα or SIRPβ. Analysis of the sequence from the mature expressed protein shows the same level of sequence identity between the genes, although the signal sequence of SIRPβ is more similar to that of SIRPγ. As SIRPα is well conserved between mice and humans, it seems likely that duplication of an ancestral SIRPα gene gave rise to a common CD47 binding precursor of SIRPβ and SIRPγ, followed by further duplication and divergence of SIRPβ into a non-CD47-binding protein, possibly driven by a viral pathogen. However, the low identity (∼57%) of the putative SIRPβ orthologue in mice (EMBL accession no. XM_283831) may indicate that it duplicated independently in this species, and the apparent lack of an SIRPγ gene indicates that mouse SIRPβ evolved directly from SIRPα. The identity of a SIRPβ ligand remains unknown, but it is possible that it binds a pathogen, from the analogy with Ly49 in mice. Most mouse strains have an inhibitory Ly49 form, but some, in addition, have a closely related activatory form that can bind a CMV protein and confer resistance to this virus (42, 43).

The finding of two related proteins that bind the same ligand with differing affinities has also been described for other cell surface receptors. CD80 binds to both CD28 and CTLA-4 with K_d of 4 and 0.4 μM, respectively, and CD80 favors CTLA-4 engagement over CD28 (44, 45). It is also notable that the higher affinity interaction (CD80-CTLA4) is inhibitory, and this parallels the higher affinity SIRPα-CD47 inhibitory interaction. The threshold effect observed with multivalent binding of SIRPγ to CD47 on activation of T cells implies that in vivo, resting T cells would not signal to each other via CD47 due to the low affinity/avidity of SIRPγ, even if cell-cell contact were taking place. However, activation, leading to increased CD47 expression, would allow for the SIRPγ-CD47 interaction to take place. This would be transient, and the interaction would be lost on chronically activated T cells when their CD47 levels decrease. This means that freshly activated CD47^high T cells may be on a “knife edge,” where any extra signals through CD47 could lead to their destruction and uptake by neighboring myeloid cells (46, 47). Indeed, ligation of CD47 has been associated with apoptosis and uptake of cells in some experimental systems (48). This functional consequence of CD47 ligation was reproduced in this study using SIRPγ fusion protein on Jurkat cells. The existence of SIRPγ on T cells means that they have another mechanism for inducing signaling through CD47 with possible multiple effects on integrin function and T cell behavior. Thus, the ability of T cells to directly send signals via CD47 has important implications for T cell biology.

We are grateful to Michael J. Puklavec and Steve Simmonds for help with mAb production, to H. J. Buhring for providing mAb SE5A5 and B1D5, to E. Tomasello and E. Vivier for SIRPβ/DAP12 transfectants and M. Colonna for SIRPβ cDNA, and to J. Sedgwick for DAP12 vector.