F₁-Adenosine Triphosphatase Displays Properties Characteristic of an Antigen Presentation Molecule for Vγ9Vδ2 T Cells Free

Human lymphocytes of the Vγ9Vδ2 subset represent the main pool of nonconventional T lymphocytes in adult blood and lymph nodes. They expand early during life, probably in response to bacterial and parasitic infections, and their effector functions are triggered following the recognition of nonpeptidic phosphorylated molecules called phosphoantigens (1, 2). Mature Vγ9Vδ2 cells sampled from blood or expanded in vitro also recognize diverse tumor cell types against which they frequently exert cytotoxicity. This cytolytic activity is regulated by activatory and inhibitory receptors for classical and nonclassical MHC class I (MHC-I) Ags (NKG2D and NKRs) (3, 4) and molecular pattern recognition receptors, such as TLRs, although activation through their TCR is essential (5, 6). The Vγ9Vδ2 TCR is thought to function as a pathogen-associated molecular pattern receptor dedicated to the recognition of small pyrophosphorylated alkyls. Strong agonist phosphoantigens, such as hydroxy-methyl-butenyl pyrophosphate, are produced by bacterial organisms through the deoxyxylulose phosphate pathway (7). However, eukaryotic cells produce weaker Vγ9Vδ2 TCR agonist Ags, such as isopentenyl pyrophosphate (IPP) and dimethylallyl pyrophosphate, through the ubiquitous mevalonate pathway leading to isoprenoid and steroid synthesis. Accumulation of IPP can naturally occur in tumor cells or in bacteria-infected macrophages following dysregulation of this metabolic pathway (8, 9). IPP accumulation can also be induced by cell treatment with aminobisphosphonate drugs that inhibit farnesyl pyrophosphate synthase, which normally consumes IPP in downstream isoprenoid synthesis (10).

In addition to alkyl pyrophosphates, phosphoantigens exist as nucleotidic conjugates coproduced by microorganisms (1). Moreover, triphosphoric acid 1-adenosin-5′-yl ester 3-(3-methylbut-3-enyl) ester (ApppI), an adenylated derivative of IPP, accumulates in cells overproducing IPP (11, 12). We found that it is naturally produced in the Vγ9Vδ2 tumoral target Daudi as well. ApppI requires cleavage into IPP+AMP to stimulate Vγ9Vδ2 cells in the absence of APCs. This can be achieved by providing exogenous nucleotide pyrophosphatase (NPP). However, ApppI can be efficiently captured by diverse cell lines and dendritic cells and confers a strong stimulatory activity for Vγ9Vδ2 cells in the absence of exogenous enzyme (13).

TCR transfer and mutagenesis experiments have demonstrated the TCR dependence of phosphoantigen recognition but direct interaction between the TCR and a putative phosphoantigenic ligand has not yet been formally proven (14, 15). Although not absolutely required, APCs increase Vγ9Vδ2 T cell responses to phosphoantigens (16). Nevertheless, the activation of Vγ9Vδ2 lymphocytes by soluble phosphoantigens requires contact with a neighboring cell, which does not need to be a specialized APC, is not required to express MHC molecules, must be of human origin, and can be a neighboring T cell (17, 18). The nature of the signal provided by the contacting cell has not been reported and could involve Ag presentation, costimulation, or both.

We reported previously that Vγ9Vδ2-sensitive tumors display on their surface a complex similar to mitochondrial ATP synthase, ecto–F₁-adenosine triphosphatase (ATPase). Using soluble forms of the Vγ9Vδ2 TCR and of a soluble form of bovine F₁-ATPase, specific interactions between the TCR of the G115 Vγ9Vδ2 T cell clone and ecto–F₁-ATPase have been demonstrated (19). In agreement, recent experiments using soluble TCR tetramers and target cells loaded with hydroxy-methyl-butenyl pyrophosphate or using a photoactivable derivative of it strongly support the notion that phosphoantigens can be displayed on the cell surface on an uncharacterized trypsin-sensitive structure (20, 21). Recent experiments revealed that the cell-surface F₁-ATPase complex is tightly associated with MHC-I Ags, an interaction that can mask antigenic epitopes and prevent detection of the F₁-ATPase complex on many cell types expressing high levels of MHC-I Ags (22).

In the present work, we have used ApppI and its nonnucleotidic analog IPP to explore the contribution of nucleotidic Ags and F₁-ATPase to phosphoantigen responses.

Cell-surface F₁-ATPase was detected by standard indirect immunofluorescence using anti-ATP synthase α-subunit (7H10) or β-subunit (3D5), both from Molecular Probes (Eugene, OR). Anti–TCRVδ2-FITC (IMMU360) was from Beckman-Coulter France (Villepinte, France) and anti–CD107a-PE (H4A3) from BD Biosciences (San Jose, CA). Bovine heart mitochondrial F₁/Fo-ATPase complex (bF₁), solubilized in undecyl-β-D-maltoside, was used throughout this study and was kindly provided by Prof. J. E. Walker (Medical Research Council, Cambridge, U.K.). IPP was from Isoprenoids (Tampa, FL). Crotalus adamanteus venom NPP and all other reagents were from Sigma-Aldrich (St. Louis, MO). The procedure for ApppI synthesis is described elsewhere (22).

Cell culture reagents were from Invitrogen except human serum (PAA Laboratories, Pasching, Austria) and IL-2 (Sanofi-Synthelabo, Toulouse, France). Daudi (β₂-microglobulin deficient), K562 (from American Type Culture Collection, Manassas, VA), and 721.221 (TAP-deficient) (23) are MHC-I–deficient cell lines; Raji (Burkitt lymphoma from American Type Culture Collection) and 721 (conventional EBV-B cell line) are MHC-I positive (23). All cell lines were grown in complete medium (CM): RPMI 1640 supplemented with glutamax, 10% heat-inactivated FCS, sodium pyruvate, and penicillin/streptomycin. For pulsing with ApppI, cells were washed once in serum-free hybridoma-SFM medium (SFM), incubated overnight in SFM containing 20 μM ApppI, and extensively washed before coincubation with T cells. Overnight treatment with pamidronate (Sigma-Aldrich) and zoledronate (Novartis AG, Basel, Switzerland) was performed in standard CM at the indicated concentration.

For polyclonal Vγ9Vδ2 T cell line establishment, PBLs isolated from buffy coats from healthy donors (Etablissement Français du Sang, Toulouse, France) were stimulated with 5 μM IPP in CM containing 10% human serum instead of FCS. Twenty-four hours poststimulation, IL-2 was added at a final concentration of 200 U/ml. Alternatively, a γδ line (PHA-γδ) was raised by magnetic sorting of pan-γδ cells (Miltenyi Biotec Kit, Miltenyi Biotec, Auburn, CA) and expansion with PHA (1 μg/ml), irradiated feeder cells (10⁷ PBL and 2 × 10⁵ EBV-B cells for 2 × 10⁶ γδ cells), and IL-2. On day 5, IL-2 concentration was raised to 400 U/ml. Cells were passed every 2 to 3 d, maintained at a concentration of 6 × 10⁵/ml, and frozen after 22–24 d of culture for further use. Using this protocol, the cell lines were >90% Vδ2⁺ (IPP-γδ) and 65% Vδ2⁺ (PHA-γδ). Control αβ T cell lines were obtained with the same protocol using staphylococcal enterotoxin B (200 ng/ml) instead of IPP. Upon thawing, lymphocytes were left to recover overnight in CM (without IL-2) before the assessment of Ag responses.

CD107 expression was monitored for assessing cytolytic activity as previously described (24). After antigenic stimulation in the presence of the anti–CD107a-PE Ab, cells were washed twice in PBS containing 1% BSA and stained for TCRVδ2. Data were acquired on an FACScan cytometer (BD Biosciences). Results are expressed as the mean of triplicate microwell cultures ± SD.

For cytokine release assay, T cells (5 × 10⁵) were cocultured with Ag- or Ab-coated beads (10⁶) in 100 μl/well in round-bottom 96-well plates in CM supplemented with a substimulatory dose of PMA (0.6 ng/ml) for 24 h. Supernatants were recovered, and IFN-γ was measured using a standard ELISA assay (OptEIA, BD Biosciences).

For immobilization on beads, solubilized bF₁ was diluted to 1/3000 (5 μg/ml) in 1 ml PBS and incubated with 10⁷, 6-μm-diameter polystyrene beads (Biovalley, Marne-la-Vallée, France) for 4 h at room temperature on a rotating wheel. Beads were then pelleted by centrifugation (2 min, 12,000 × g), washed, and incubated in 1 ml 1% BSA in PBS overnight, washed once with PBS, and kept up to 10 d at 4°C in PBS containing 1% BSA. Beads were used in stimulation assays as cultured cell lines. Positive and negative control beads were prepared similarly except that bF₁ was omitted or replaced by anti-CD3ζ (MEM-57, Exbio Praha, Vestec, Czech Republic), or control Abs (anti-NKG2D, W6/32, 10 μg/ml) followed by saturation with BSA.

In some experiments, the beads were loaded with phosphoantigens: to this end, after washing, a suitable aliquot of beads was incubated with ApppI or IPP (50 μM) for 2 h at 37°C in complete SFM medium, washed extensively, and used for Vγ9Vδ2 T cell stimulation. In phosphatase sensitivity assays, NPP or CM was added to the beads during 15–60 min before or after pulsing the beads. These were then washed extensively and cocultured with T cells in the presence or absence of NPP. At either step, NPP was used at the concentration of 0.02 U/ml.

bF₁ was assayed by a bioluminescence measurement method. A total of 100 ng purified bF₁ was incubated for 1 min in the presence of 1 μM ATP and the indicated ApppI concentration in 50 μl ATPase buffer (10 mM HEPES, 150 mM NaCl, 5 mM KCl, 2 mM MgCl₂ [pH 7.5]). A total of 25 μl reconstituted luciferase solution (ATP CLS II kit, Roche, Basel, Switzerland) were then added, and luminescence was acquired during 2 s using an Orion luminometer (Berthold Detection Systems, Huntsville, AL). For each dose of ApppI, the luminescence was recorded. The absence of interference of ApppI with luciferase activity was checked. Data are expressed as the mean of triplicate experiments ± SD.

Vγ9Vδ2 T cells were loaded with fura 2-AM (Molecular Probes; 5 μM, 45 min) in 1 ml RPMI and washed, and 10⁵ cells were left to adhere for 10 min to the bottom of polylysin-d–treated (5 μg/ml) eight-well Lab-Tek chamber slides (Thermo Scientific, Roskilde, Denmark) in 100 μl RPMI containing 5% FCS and 10 mM HEPES. Fluorescence measurements were performed on a Zeiss Axiovert 200M inverted microscope equipped with a CCD camera and a 37°C chamber (Zeiss, Jena, Germany). Cells were consecutively excited at 340 and 380 nm at intervals of 5 or 10 s, and emission at 510 nm was collected with the CCD camera in turn with transmitted light images. Analysis was performed on individual cells that did not make contacts with surrounding T cells during the time of recording. After initial focusing, acquisition was paused, and Ags and beads were added in a volume of 20 μl deposited on the top of the cell medium. It took 2 to 3 min for the beads to come in contact with the adherent T cells. The camera output was analyzed with the Metafluor software. Data were transferred into the Excel software (Microsoft, Redmond, WA) for conversion of 340 nm/380 nm fluorescence ratios to approximate intracellular calcium concentrations (intracellular calcium indexes). These were calculated as described (25) by recording fluorescence ratios during 5 min in a separate experiment after ionomycin stimulation (5 nM) in the presence or absence of 4 mM EDTA to calculate the minimum and maximum ratios, respectively. At each time point, the ratios from 8–15 cells were averaged and data were expressed as means ± SD.

K562 or 721.221 cells were either pulsed with ApppI (20 μM) or kept untreated for 16–18 h, washed twice, and stained with hexidium iodide (Molecular Probes; 5 μM) and used as APCs. Alternatively, bF₁-coated beads were pulsed with 50 μM ApppI (see above) and used for interaction with T cells. Vγ9Vδ2 T cells were mixed at the ratio of 1:1 with APCs or with beads at the ratio of 1:5 in presence or absence of nucleotide pyrophosphatase in round-bottom 5-ml tubes. Tubes were centrifuged at 1200 rpm for 1 min and incubated at 37°C for 30 min (APCs) or 50 min (beads). The cells were then resuspended gently and layered onto poly-l-lysine–coated eight-well slides. After 10 min, 0.02% NaN₃ was added, and slides were kept at room temperature for a further 15 min. The wells were washed once with PBS containing 5% FBS and 0.02% NaN₃. Finally, cells were stained with anti-δ2–FITC, and the slides were mounted and observed under a Zeiss LSM 510 (Zeiss) confocal microscope with a 63 Plan-Apochromat objective (1.4 oil), electronic zoom 3, as described (26).

We previously reported a role for ecto–F₁-ATPase (also called membrane ATP synthase by others) in the Vγ9Vδ2 T cell response to tumor cells. Ecto–F₁-ATPase can be detected on many cell lines by surface immunostaining with the Abs 7H10 (α-subunit) and 3D5 (β-subunit), although high MHC-I expression strongly interferes with its detection by masking target epitopes (22). However, it is barely detected on the 721.221 cell line as opposed to Daudi and K562, which are similarly MHC-I deficient (Fig. 1A).

ApppI is a nucleotidic analog of IPP that is endogenously produced in cells treated with aminobisphosphonates. We thus pulsed MHC-I–deficient cells with aminobisphosphonates or ApppI to assess their ability to promote phosphoantigenic responses after endogenous or exogenous Ag loading. Incubation with ApppI increased the stimulatory potential of Daudi cells. K562 cells became strong stimulators postincubation with ApppI or aminobisphosphonates. In contrast, the 721.221 lymphoblastoid cell line promoted no or very weak responses to ApppI and pamidronate (Fig. 1). This was not due to the inhability of B cells to present phosphoantigens as Raji Burkitt lymphoma cells as well as the EBV-B cell line 721 could be pulsed with ApppI or pamidronate (Supplemental Fig. 1). Although ineffective at low doses, pulsing 721.221 cells with zoledronate at 40 μM or over made cells stimulatory (Fig. 1C). These doses, however, were toxic to cells and promoted their morphological change, largely due to apoptosis as they became positive for Annexin V. Moreover, 4–24 h of this treatment led to increased staining with anti–F₁-ATPase Abs in up to ∼40% of cells, as well as increased stimulatory activity, suggesting a direct link between F₁-ATPase expression and phosphoantigen presentation (Fig. 2).

To examine the involvement of the TCR in these anti-tumor responses, TCR aggregation was studied by confocal microscopy. Following 30 min of contact with ApppI-pulsed K562 cells, the TCR on most Vγ9Vδ2 cells underwent a typical redistribution in patches or caps. This was observed neither with unpulsed K562 cells nor with 721.221 cells, pulsed or unpulsed (Fig. 1D).

We then examined the ability of K562 or 721.221 cells to support the response to the pyrophosphate Ag IPP. As this Ag does not require APCs and may induce Vδ2 responses following T cell-T cell contact, we used calcium flux monitoring in individual Vδ2 cells by videomicroscopy. Live Vγ9Vδ2 T cells were loaded with the fluorescent calcium probe fura 2 and immobilized in culture medium on the bottom of Lab-Tek chamber slides in conditions in which they did not make contacts with each other. IPP was provided in solution. Cell contact was provided by adding unlabeled tumor cells on top of the culture (Fig. 3). In the absence of cell-cell contact, IPP did not induce intracellular Ca²⁺ flux in individual cells. Addition of K562 cells in the presence of IPP produced a calcium flux that was maximal a few minutes after initial contacts with K562 cells and decreased gradually to reach the baseline after ∼30 min. Calcium flux was not observed in the absence of IPP. As opposed to K562, 721.221 cells failed to promote an IPP-mediated calcium response.

Differential expression of ecto–F₁-ATPase by these cell lines could explain their different ability to support phosphoantigen responses. However, other Ags, such as ligands targeting the activatory receptor NKG2D, could be involved. Attempts to block responses induced by cell–cell interactions with anti–F₁-ATPase Abs failed, although the same Abs were previously shown to affect interactions between soluble F₁ and TCR complexes (19). This, however, did not rule out a possible involvement of F₁-ATPase. To assess its possible implication in the calcium response to phosphoantigen, calcium flux analysis was performed as above except that contacting tumor cells were replaced by polystyrene beads coated with purified bF₁ and saturated with BSA (Fig. 4). Again, no significant calcium flux could be detected in the presence of IPP in cells making no contact with surrounding partners when beads devoid of bF₁ were provided for contact. On the contrary, the contact of bF₁-coated beads promoted a delayed increase of intracellular calcium when IPP was present in the culture medium. Unlike with whole APCs, this calcium flux started asynchronously in T cells 20–40 min after the contact with beads. The pattern of intracellular calcium pulse also varied in individual cells. In most of them, the peak of maximum intracellular Ca²⁺ concentration was preceded by a smaller peak that lasted 10–20 min, with or without a return to the baseline (Fig. 4B). A similar flux was not induced by bF₁ in the absence of phosphoantigen or with control beads devoid of bF₁. Moreover, the calcium flux induced by the combination of IPP and bF₁ was severely depressed by the addition of a combination of anti-α and anti–β-F₁ Abs but not by soluble control isotype-matched Abs. This shows that a cognate interaction with F₁-ATPase promotes the response to IPP (Fig. 4C). As controls, we analyzed the γδ-T cell responses to beads carrying anti-CD3 and anti-NKG2D Abs instead of F₁-ATPase in the absence of phosphoantigen. Although more efficient than the stimulation with the bF₁+IPP combination, anti-CD3 stimulation was more delayed than with cells (Fig. 3), suggesting that the delay is inherent to T cell stimulation in the absence of accessory molecules and costimulation. Anti–MHC-I (W6/32 Ab) and anti-NKG2D alone did not promote significant responses (Fig. 4D).

Mitochondrial ATP synthase is known to bind di- and triphosphonucleotides in its catalytic sites. Additional nonhydrolytic sites can accommodate multiple nucleotide analogs and present allosteric regulatory properties on the enzyme activity (27, 28). We thus wanted to see if ApppI could also bind to bF₁. Surface plasmon resonance was used unsuccessfully due to unfavorable size ratios between the two ligands (∼1:1000). Binding was thus tested by two alternative approaches.

In a first set of experiments, bF₁ immobilized on polystyrene beads was incubated with IPP or ApppI, extensively washed, and used to stimulate Vγ9Vδ2 T cells. ApppI is not active in the absence of APCs and requires cleavage of the β-γ phosphoester bond by a nucleotide pyrophosphatase (13). The stimulatory activity of Ag-pulsed beads was thus evaluated in the presence and absence of exogenous crotale NPP (Fig. 5A). Vγ9Vδ2 T cells were activated to produce cytokines when stimulated with ApppI-pulsed bF₁, in the presence of NPP, but not with IPP-pulsed beads. This indicates that ApppI can be stably loaded on bF₁ and that IPP can subsequently be released from the complex by exogenous NPP. As no activity could be retained on IPP-pulsed beads, binding requires the Ag to be in nucleotidic form. Nonspecific binding of nucleotidic Ag on beads is excluded because no activity is retained on control beads devoid of bF₁ and BSA saturated. Finally, the effect is specific for Vγ9Vδ2 cells as a control αβ T cell line was unresponsive to ApppI-loaded bF₁, although it responded well to anti-CD3 stimulation (Fig. 5B). Monitoring CD107 expression poststimulation gave qualitatively identical results (data not shown).

The stimulatory activity of bF₁ loaded with ApppI on polystyrene beads was subsequently evaluated in the single-cell calcium mobilization assay described above. Beads coated with bF₁ and loaded with ApppI did not promote Vγ9Vδ2 T cell activation upon contact. However, activation was induced upon contact with beads as soon as NPP was added into the culture medium (Fig. 5C). To confirm the involvement of the TCR in γδ T cell activation with immobilized F₁-ATPase, TCR aggregation was analyzed using confocal microscopy (Fig. 5D). Capping of the TCR could be observed following a 50-min contact with F1-coated and ApppI-loaded beads but not with control beads or with beads devoid of ApppI. Strikingly, capping occurred even in the absence of exogenous NPP, indicating that IPP release is not mandatory for TCR aggregation.

In the second set of experiments, we reasoned that a putative interaction of ApppI with F₁-ATPase might be revealed by a modulation of its enzymatic activity in an ATP hydrolysis acellular assay (Fig. 6). When ApppI was added to ATP in the presence of F₁/Fo ATPase, this increased the rate of ATP hydrolysis by F₁ at low ATP concentration (1 μM). This indicates that ApppI can bind to the enzyme complex in a noncompetitive fashion, resulting in an allosteric stimulatory effect on the enzymatic activity of the complex. Altogether, these experiments indicate that F₁-ATPase can bind and present ApppI to Vγ9Vδ2 cells. However, in the absence of APCs, IPP must be released from complexes by addition of exogenous NPP to allow activation of Vγ9Vδ2 T cells.

We reported previously Vγ9Vδ2 T cell activation with bovine F₁-ATPase in the absence of added phosphoantigen (19). As this was not observed with the IPP-expanded lines used in the current study, we hypothesized that expansion with IPP could induce a selection for strictly IPP-dependent clones. To test for this possibility, a γδ line was produced after sorting γδ cells and polyclonal expansion with PHA and irradiated feeder cells. When activated with immobilized bovine F₁-ATPase, a significant intracellular IFN-γ production (∼30% of PHA-expanded Vδ2⁺ cells) was observed when phosphoantigen was omitted, whereas no significant activation occurred in the IPP-expanded line. Thus, the Vγ9Vδ2 T cell population comprises a subpopulation of cells that can be activated, at least in vitro, by bovine F₁-ATPase (Supplemental Fig. 2).

Tumor cells remain stimulatory following treatment with phosphatases, including NPP (13). We thus wondered whether NPP treatment of ApppI/F₁ complexes prior to their exposure to T cells would prevent activation. Beads carrying ApppI/F₁ complexes were thus treated with NPP during variable times and washed. As a control for NPP activity, ApppI was similarly treated prior to its incubation with F₁-coated beads. After the loading step, the beads were washed and tested for stimulatory activity in the presence of NPP by monitoring CD107a expression (Fig. 7) or intracellular cytokine accumulation (not shown). As expected, ApppI hydrolysis by a 30-min enzyme treatment preloading on beads completely abrogated subsequent responses due to deficient loading on bF₁. Nevertheless, when the NPP treatment was performed after loading on beads, a significant antigenic activity was still present on beads after 1 h of treatment, indicating that ApppI is somewhat protected from hydrolysis when loaded on bF₁ (Fig. 7). A model for Vγ9Vδ2 T cell activation is provided in Fig. 8.

The recognition of phosphoantigens by Vγ9Vδ2 cells requires T cell contact with neighboring cells, but the nature of the signal provided by this contact may be multiple as costimulation through CD28, NKR, or TLR families of receptors as well as adhesion molecules might be involved. However, in the case of tumor cell recognition, experimental data support the need for some Ag presentation molecule. As a possible TCR ligand, F₁-ATPase could play this role. The 721.221 cell line that does not express F1-ATPase does not support phosphoantigen responses as opposed to Daudi and K562, which are similarly MHC-I deficient. Strikingly, however, 721.221 cells undergo apoptosis and express F₁-ATPase upon treatment with zoledronate at high doses. This also promotes their stimulatory potential for Vγ9Vδ2 cells, likely due to the concomitant expression of phosphoantigens and F₁-ATPase. A direct role of F₁-ATPase is further substantiated by the fact that the F₁-ATPase complex immobilized on beads can provide at least part of the signal provided in physiological conditions by membrane contact. Provision of immobilized F₁-ATPase promotes TCR aggregation, calcium mobilization, CD107 expression, and cytokine secretion in Vγ9Vδ2 cells when phosphoantigen is present. Nevertheless, the delayed calcium response as well as the incomplete TCR relocalization following interaction with F₁-ATPase on beads indicates that additional signals are required for a full T cell response.

Vγ9Vδ2 T cells expanded with IPP are not activated by immobilized bovine F₁-ATPase in the absence of exogenous phosphoantigen. This contrasts with our previous report on Vγ9Vδ2 cells from fresh PBL samples, PHA-expanded Vγ9Vδ2 T cell clones derived from PBMCs (19), and, in the present work, for ∼30% of PHA-expanded polyclonal γδ cells. This reactivity could be due to autoreactivity or xenoreactivity and appears to be counterselected after IPP stimulation. Although revealing a basal affinity of the TCR for F₁-ATPase, this reactivity may be of little significance in the context of cell–cell interactions in which costimulatory and regulatory molecules are involved.

Our experiments indicate that the ATP-conjugated analog of IPP can stably bind to F₁-ATPase, whereas IPP does not, and the stimulatory effect of ApppI on ATP hydrolysis suggests binding on noncatalytic nucleotide binding sites (29, 30). Once bound, the Ag is somehow protected from degradation by exogenous NPP. A similar situation is observed with tumors, although there is no direct proof that phosphoantigens are displayed in an unmodified form on their surface. Indeed, we have observed that the stimulatory activity of Daudi cells is unchanged after tumor treatment with NPP (13). Thus, although pyrophosphorylated Ags are produced in tumors as well as in cells infected with bacteria-producing phosphoantigens, they may actually be presented as nucleotide derivatives.

T cell activation by F₁-ATPase/ApppI complexes in the absence of APCs requires addition of NPP. It seems likely that tumor cells or APCs can provide an NPP activity that is missing or poorly expressed on T cells. Although candidate ecto-enzymes have been reported on tumoral and nontumoral tissues (31, 32), the persistence of phosphoantigenic activity on the cell surface implies that the postulated NPP activity is active only after encounter with T cells. This could be achieved through Ag release from the presenting molecule or through a local activation of the NPP activity concomitantly with TCR binding.

Our TCR aggregation analyses are in agreement with studies using pamidronate-treated tumor cells showing clear TCR modulation (33) or aggregation in caps (34). More conflicting results were described when phosphoantigens were used in solution. Sireci et al. (35) have shown that weak agonist phosphoantigens like IPP-induced TCR modulation, whereas strong agonists did not (35), and in another study, IPP stimulation did not promote TCR modulation (36). This suggests that aggregation and modulation are more readily induced when phosphoantigens are encountered in membrane-bound form, as is probably the case with pamidronate or ApppI-treated cells. Strikingly, we also observed dissociation between aggregation and activation as, in the absence of exogenous NPP, ApppI on bF₁ induces aggregation but no activation.

Finally, assuming 100% binding of properly conformed F₁-ATPase to polystyrene beads and hypothesizing six nucleotide binding sites on the complex, a rough estimate of the phosphoantigen (ApppI) concentration in stimulation assays with F₁-coated beads and NPP indicates that it cannot exceed the nanomolar range. This is several orders of magnitude below the IPP or ApppI concentrations that give detectable activation when supplied in solution (micromolar range). Thus, presentation on F₁-ATPase increases the efficiency of phosphoantigen recognition, possibly through concentrating the Ag in the proximity of the TCR or by providing protection from unwanted surrounding phosphatases. Based on these observations, a model for Ag recognition by Vγ9Vδ2 cells is suggested (Fig. 8).

In our experimental conditions, responses to IPP require T cell recognition of F₁-ATPase, although IPP does not bind to this complex. This strongly supports the notion that F₁-ATPase is generally involved in all responses induced by phosphoantigens. Our explanation for this requirement is that the Vγ9Vδ2 TCR is dedicated to the recognition of Ags on cell surfaces, as is that of other T cell subsets. As activation apparently requires a hydrolytic step, our view is that this is short-circuited when phosphoantigens are encountered in pyrophosphate form. Nevertheless, recognition of the presentation molecule remains requested. This must hold true also for the self-presentation of phosphoantigens by Vγ9Vδ2 T cells. Indeed, the presence of F₁-ATPase on Vγ9Vδ2 cells can be revealed after disruption of MHC-I Ags (P. Vantourout, unpublished observations) and may allow autopresentation.

Finally, although they efficiently inhibit T cell activation by purified F₁-ATPase, anti-α/β ATPase Abs do not alter Vγ9Vδ2 T cell responses to stimulatory cells. F₁-ATPase may thus be one of several presentation structures for phosphoantigens. However, an alternative hypothesis is that ecto–F₁-ATPase is only in part accessible to Abs due to tight interactions with surrounding surface proteins. Indeed, high expression of classical MHC-I molecules hampers detection of F₁-ATPase epitopes. Further studies are required to investigate whether F₁-ATPase preferentially interacts with MHC-I related structures that could tightly regulate the potential autoreactivity of the Vγ9Vδ2 TCR.

We thank Prof. John Walker (Cambridge, U.K.) for generously providing F₁/Fo-ATPase preparations. We also thank F. L’Faqihi and V. Duplan-Eche (Institut Fédératif de Recherche Bio-Médicale de Toulouse, Technical Platform of Cytometry, Toulouse, France) for technical assistance in cytometry and Sophie Allart (Institut Fédératif de Recherche Bio-Médicale de Toulouse, Technical Platform Cell Imaging) for help in confocal and wide-field microscopy.

Disclosures The authors have no financial conflicts of interest.