Hierarchical Costimulator Thresholds for Distinct Immune Responses: Application of a Novel Two-Step Fc Fusion Protein Transfer Method¹

Activation of T cells is dependent upon coordinate signaling through Ag and costimulator receptors on their surfaces (reviewed in Refs. 1, 2, 3). An “activation threshold model” has been proposed to explain various phenomena associated with antigenic signaling through the TCR (4, 5, 6, 7, 8, 9, 10). There are three essential aspects to this model. First, T cell activation is elicited once a threshold number of TCRs have been triggered. Second, a hierarchy of TCR thresholds are operative even at the single-cell level, with differing T cell cytokine outputs and functional capabilities resulting from the triggering of differing numbers of TCRs. Third, TCR activation thresholds can be lowered via costimulation (8) or elevated by at least one anti-inflammatory protein.³

Little is known about quantitative aspects of costimulator signaling, and the possibility of “costimulator activation thresholds,” that might parallel TCR activation thresholds, has not been systematically evaluated. There are two related questions in this regard. First, at any given Ag density, are there a threshold number of costimulator receptors that must be triggered to activate T cells? Second, if costimulator thresholds do indeed exist, is there a hierarchy of such thresholds within single cells, with different T cell outputs pegged to different costimulator inputs? Evidence in favor of such costimulator thresholds comes from recently reported studies by several groups that point to the importance of APC costimulator levels on elicited T cell responses (8, 11, 12).

The quantitative study of costimulation has been limited by technical factors. Gene transfer strategies do not permit fine control of costimulator densities at APC surfaces. As an alternative, cell-free systems have been invoked by some investigators. For example, differing concentrations of costimulators can be immobilized on plastic substrates (13, 14, 15) or on cell-sized latex microspheres (16). Such cell-free systems, however, by definition ignore contextual molecular determinants at the APC surface.

Protein transfer offers an alternative means for titrating costimulator densities at APC surfaces. We (17, 18) and others (19, 20) have described one costimulator protein transfer method that invokes chimeric GPI-modified costimulators as membrane-reincorporable “protein paints.” However, scaling up the production of these protein derivatives is complicated by the need to purify them from transfectant cell membranes. In the present study, we report a unique protein transfer strategy that bypasses the need to work with recombinant membrane proteins. This strategy builds upon work by Kim and Peacock (21) who described a two-step approach for binding Abs to cell surfaces. According to their approach, protein A, after derivatization with palmitate, is first incorporated into cell membranes, and in turn, this membrane-associated palmitated protein A (pal-prot A)⁴ is used as a trap for secondarily added IgG molecules. Building upon this finding, the present study demonstrates that Fc fusion proteins can be similarly delivered to pal-prot A-precoated cell surfaces with preservation of trans-signaling protein function. Moreover, we proceed to utilize costimulator · Fcγ₁ (Fc domain of human IgG1) protein transfer to establish the existence of costimulator activation thresholds within T cells.

Recombinant protein A (Calbiochem, La Jolla, CA) was derivatized with the N-hydroxysuccinimide ester of palmitic acid (Sigma, St. Louis, MO) as described (21). The lipid-derivatized protein A was purified as described (22) on a 30-ml Sephadex G-25 (Sigma) column. The protein product was quantitated using a bicinchoninic acid kit (Bio-Rad, Richmond, CA), filter sterilized, and stored at 4°C until use.

Cells (3–7 × 10⁶/ml) were resuspended in RPMI 1640 medium (BioWittaker, Walkersville, MD) after three washes with this same medium. Varying concentrations of pal-prot A (or nonderivatized protein A as negative control) were added to the cell suspension, and the mixture was incubated at 4°C for 2 h with constant mixing. To assess the incorporation of pal-prot A onto cell surfaces, cells were washed twice in buffer (0.25% BSA/0.01% sodium azide/PBS) and then incubated on ice for 1 h with 100 μl of 100 μg/ml FITC-human IgG (Sigma) diluted with the same buffer. Cells were washed twice in the buffer and analyzed on a FACStar (Becton Dickinson, Mountain View, CA).

The expression plasmid pCDM8/B7Ig, encoding the complete human B7-1 extracellular domain linked in-frame to the Fcγ₁, was obtained from the American Type Culture Collection (Manassas, VA). The sequence encoding B7-1 · Fcγ₁ was mobilized from pCDM8/B7Ig by digesting with XbaI, filling-in with Klenow fragment, and subsequently digesting with HindIII. The mobilized fragment was subcloned into the EBV episomal expression vector pREP7β (Invitrogen, San Diego, CA) with HindIII and filled-in BamHI sites. The plasmid was transfected into 293 cells (human kidney cell line; American Type Culture Collection), and hygromycin B-resistant colonies were selected in serum-free UltraCulture medium (BioWhattaker) supplemented with 10 mM glutamine, penicillin/streptomycin, and 200 μg/ml hygromycin B. Secreted B7-1 · Fcγ₁ was purified from conditioned medium by protein A-agarose (Life Technologies, Germantown, MD) affinity chromatography and verified by SDS-PAGE.

Cells precoated with pal-prot A were washed once and resuspended in RPMI 1640 medium (3–7 × 10⁶ cells/ml). The mixture was incubated at 4°C for 1 h with constant mixing. To monitor protein delivery, 10⁶ cells were washed twice with the same buffer as above, incubated on ice for 1 h with 1 μg of human B7-specific mAb BB-1 (PharMingen, San Diego, CA) in 100 μl of buffer. Cells were washed once and immunostained (on ice for 1 h) with 100 μl of 1:100 diluted FITC-conjugated goat F(ab′)₂ anti-mouse Ig (Boehringer Mannheim, Indianapolis, IN) as secondary Ab. Cells were washed once, resuspended in PBS, and analyzed on a FACStar.

Human B7-1 · Fcγ₁ was iodinated using Iodo-beads (Pierce, Rockford, IL) according to the manufacturer’s protocol, and the labeled protein was purified on a Sephadex G-25 column (Pharmacia, Piscataway, NJ). The specificity was adjusted to 2.1 × 10⁶cpm/μg by addition of unlabeled B7-1 · Fcγ₁. Protein transfer was performed as described earlier, substituting the labeled protein. All experiments were performed in duplicate. To control for nonspecific binding, excess amounts of unlabeled human IgG (Sigma) were added to specifically block the binding of B7-1 · Fcγ₁ to protein A. After repeated washing, counts in cell pellets were determined using a gamma counter (1272 Clinigamma; LKB Instruments, Gaithersburg, MD). Counts resulting from specific binding of B7-1 · Fcγ₁ were calculated by subtracting nonspecific counts obtained with human IgG. The average number of molecules on a single cell was calculated according to the formula A × B⁻¹ × C⁻¹ × N_A, where A is the determined radioactivity (cpm) in the cell pellet, B is the specific activity of the labeled protein expressed as cpm/mol, C is the number of cells in the cell pellet, and N_A is Avogadro’s constant.

PBMC were isolated from fresh whole blood by Ficoll density centrifugation. T cells were purified by two rounds of treatment with Lympho-kwik (One Lambda, Canoga Park, CA). T cell purity was verified by lack of a proliferative response to PHA or PMA in the absence of accessory cells. The human CD3-specific mAb HIT3a (PharMingen) was bound to 96-well plates at the indicated concentrations and used in this form to provide a first activating signal to T cells. Alternatively, PHA was used in soluble form as a source of a first signal. K562 cells transfected with the negative control vector pREP7β (K562/pREP7β) were precoated with pal-prot A and secondarily coated with B7-1 · Fcγ₁. For each proliferation assay, 1 × 10⁵ T cells were incubated with 4 × 10⁴ B7-1 · Fcγ₁-coated and mitomycin C-treated K562/REP7β cells for 60 h at 37°C. Wells were pulsed with 1 μCi [³H]thymidine for the last 16 h of the incubation period. Cells were harvested and counted on a Betaplate liquid scintillation counter (Wallac Oy, Turku, Finland).

A total of 10⁶ T cells was incubated with 5 × 10⁵ processed K562/REP7β cells (B7-1 · Fcγ₁ positive or negative) in 48-well plates using either plate-bound HIT3a or PHA as a source of first signal. Supernatants were collected after 48 h, and ELISAs for human IFN-γ and IL-2 were performed using a commercial ELISA kit according to manufacturer’s protocol (Genzyme, Cambridge, MA).

A total of 10⁶ T cells was incubated with 5 × 10⁵ B7-1 · Fcγ₁-coated K562/REP7β cells in 48-well plates for 48 h. Again, either plate-bound HIT3a or PHA was used as a source of first signal. Monensin (Sigma) was added to a final concentration of 3 μM, and the mixture was incubated for an additional 6 h to accumulate cytokine within the cells. Cells were then collected, fixed by incubating them in 100 μl of fixation solution (4% paraformaldehyde/PBS (pH 7.4)) on ice for 20 min, and then washed twice with staining buffer (0.1% saponin/1% heat-inactivated FCS/0.1% sodium azide/Dulbecco’s PBS). Immunostaining for intracellular cytokines was performed by incubating the cells on ice for 1 h with 100 μl of the staining buffer containing 0.5 μg of FITC-anti-IFN-γ and 0.5 μg of PE-anti-IL-2 Abs (PharMingen). Cells were subsequently washed once with staining buffer without saponin. T cells were gated using forward light scatter/side light scatter parameters, and 2–5 × 10⁴ cells were analyzed in each run.

A two-step strategy was developed for quantitatively “painting” costimulators onto cell surfaces. This approach was adapted from a method originally designed by Kim and Peacock (21) for delivering intact IgG to cell surfaces. According to our modification, Fc fusion proteins are substituted for intact IgG. The first step involves precoating cells with a chemically palmitated prot A derivative (pal-prot A; see Materials and Methods). In a first set of optimization experiments, efficient incorporation of palmitated prot A was documented in four cell lines (Fig. 1,A) as detected with FITC-conjugated human IgG. As a negative control, nonderivatized protein A lacked the capacity to bind to the same cells. Membrane incorporation was dose dependent and started to plateau at 33 μg/ml pal-prot A (Fig. 1,B). Pal-prot A incorporation was rapid, appearing immediately after addition to the cells and reaching a plateau at ∼1 h (Fig. 1 C).

The second step involves adding the Fc fusion protein to cells precoated with pal-prot A. A B7-1 · Fcγ₁ fusion protein, incorporating the Fc region of human IgG1, was produced (see Materials and Methods) and visualized by SDS-PAGE as a dominant single band of ∼80 kDa under both reducing and nonreducing conditions (Fig. 2 A). Its identity was confirmed by ELISA, with the recombinant protein binding strongly to the human B7-1-specific mAb, BB-1, but not to control Ab (data not shown).

Next, the feasibility of painting B7-1 · Fcγ₁ onto pal-prot A precoated cells was established. When K562 cells were precoated with pal-prot A, secondarily applied B7-1 · Fcγ₁ attached to the cell surface, as detected by immunostaining of the cells with anti-B7-1 BB-1 mAb and FITC-conjugated goat anti-mouse IgG (Fig. 2,B). When a control Fc fusion protein (CD28 · Fcγ₁) was substituted for B7-1 · Fcγ₁, no BB-1 binding was observed, substantiating BB-1 mAb’s B7-1 specificity (Fig. 2,B). When nonderivatized protein A was substituted for pal-prot A, no BB-1 binding was observed, indicating the dependence of Fc protein attachment on protein A lipid anchoring (Fig. 2,B). Next, the feasibility of quantitatively painting B7-1 · Fcγ₁ onto pal-prot A precoated cells was established. When K562 cells were precoated with excess amounts of pal-prot A (33 μg/ml), surface levels of B7-1 · Fcγ₁ were dependent on the concentrations of applied B7-1 · Fcγ₁ (Fig. 2,C). Surface B7-1 epitope levels started to plateau at 33 μg/ml, and the epitope density was similar to that on B7-1-transfected K562 cells (data not shown). The average number of B7-1 · Fcγ₁ painted per cell was determined using ¹²⁵I-labeled B7-1 · Fcγ₁. Again, K562 cells incorporated increasing amounts of B7-1 · Fcγ₁ as the reagent concentration was increased during the painting process (Table I). At the lowest concentration used (0.033 μg/ml), ∼460 molecules became anchored onto each K562 cell. At the highest concentration used (33 μg/ml), about 460,000 B7-1 · Fcγ₁ molecules became incorporated. Taken together, these data establish that B7-1 · Fcγ₁ can be applied to pal-prot A-coated cells in a quantitative fashion.

In vitro proliferation assays were performed to determine whether membrane-tethered B7-1 · Fcγ₁ can costimulate T cells. In these assays, PHA and B7-1 · Fcγ₁-coated K562/REP7β cells (i.e., K562 cells stably transfected with the pREP7β EBV episomal expression vector) were used to provide first and second signals, respectively, to T cells. K562/REP7β cells lack detectable B7-1 (data not shown) and provide a suitable negative control for experiments with K562/B7-1 cells (i.e., K562 cells stably transfected with a pREP7β vector containing human B7-1 cDNA sequence). Surface B7-1 levels on K562/B7-1-transfected cells and B7-1 · Fcγ₁-coated K562/REP7β cells were determined by immunostaining, and the mean fluorescence intensities were 550 and 450, respectively. As shown in Fig. 3 A, in the presence of suboptimal PHA concentrations (<0.5 μg/ml), B7-1 · Fcγ₁-coated K562/REP7β cells, but not K562/REP7β cells, significantly enhance T cell proliferation. The costimulatory effect was comparable to that achieved with K562/B7-1-transfected cells. The B7-1 · Fcγ₁/pal-prot A-dependence of the observed costimulation was verified by showing that cells treated with a combination of (nonderivatized) protein A and B7-1 · Fcγ₁, or with a combination of pal-prot A and control CD8 · Fcγ₁, did not enhance T cell proliferation. In the presence of higher PHA concentrations (> 1 μg/ml), K562/REP7 cells also costimulate T cell proliferation, although to a lesser extent than the B7-1-positive cells.

To further confirm the costimulatory function of cell-associated B7-1 · Fcγ₁, proliferation assays were performed in which plate-bound anti-human CD3 mAb was substituted for PHA as a more physiological first signal. In this setting, in the presence of suboptimal concentrations of anti-CD3 mAb (< 10 μg/ml), cell-associated B7-1 · Fcγ₁ costimulated even more effectively than native B7-1 expressed at equivalent levels on transfected cells (Fig. 3 B). Again, CD8 · Fcγ₁, as a negative control Fc fusion protein, did not costimulate under the same conditions. Taken together, these results establish that B7-1 · Fcγ₁, tethered to membranes via pal-prot A, effectively costimulates T cell proliferation.

With an effective costimulator protein transfer method in hand, quantitative aspects of B7-1 costimulation were evaluated. To this end, T cell proliferation assays were performed using K562/REP7β cells painted with variable concentrations of B7-1 · Fcγ₁. The concentration dependence of B7-1 · Fcγ₁-mediated costimulation could be readily demonstrated when a fixed suboptimal concentration of PHA (0.25 or 0.5 μg/ml) was used as a source of first signal (Fig. 4 A). For example, in the presence of 0.5 μg/ml PHA, T cell proliferation was observed once a threshold B7-1 · Fcγ₁ concentration (0.1 μg/ml) was reached, and the level of proliferation continued to rise with increasing B7-1 · Fcγ₁ concentrations until reaching a plateau at ∼3.3 μg/ml. In the presence of a lower concentration of PHA (0.25 μg/ml), T cell proliferation was observed when a higher threshold B7-1 concentration (1 μg/ml) was reached, indicating that costimulator thresholds can be modulated by the strength of the first signal.

Similar results were obtained when anti-human CD3 mAb was used as a source of first signal instead of PHA (Fig. 4 B). Again, in the presence of a fixed suboptimal concentration of plate-bound anti-CD3 mAb (0.37 or 1.1 μg/ml), costimulation was observed only after a threshold B7-1 · Fcγ₁ concentration was reached, and a further dose-dependent increase in proliferation was also seen. Hence, in the presence of a suboptimal first signal (whether PHA or anti-CD3 mAb), a threshold B7 level is required for T cells to proliferate and the extent of T cell proliferation is dictated by the costimulator level.

Having documented that B7-1 levels can modulate the extent of T cell proliferative responses, we next determined whether B7-1 levels can also dictate the quality of immune responses by altering the ratios of cytokines produced by activated T cells. ELISA was used to measure T cell cytokine secretion in response to varying painted B7-1 · Fcγ₁ concentrations and fixed suboptimal primary stimulus concentrations. At a fixed PHA dose, the B7-1 · Fcγ₁ concentrations eliciting minimal and maximal cytokine responses differed for IFN-γ and IL-2 with the general hierarchy being IFN-γ < IL-2 (Fig. 5,A). A similar hierarchy for the cytokine responses was observed when anti-CD3 mAb (3.3 μg/ml) was used as a source of first signal (Fig. 5,B). For instance, at a B7-1 · Fcγ₁ concentration of 0.33 μg/ml, IFN-γ output was 60% of the maximal response, whereas IL-2 output showed no increase above basal levels (Fig. 5 B). This observed IFN-γ > IL-2 hierarchy for B7-1 costimulator thresholds matches the order described for TCR activation thresholds.

To substantiate the ELISA findings with bulk T cell populations, multiparameter flow cytometric analyses were performed to assess intracellular IFN-γ and IL-2 levels within individual cells. At low B7-1 · Fcγ₁ concentrations, the T cell response is dominated by IFN-γ-only producers; however, at higher B7-1 · Fcγ₁ concentrations, substantial numbers of IFN-γ and IL-2 double producers emerge (Fig. 6). Relatively few IL-2-only producers were observed, even at the highest B7-1 · Fcγ₁ concentrations. These findings are consistent with the bulk T cell cytokine response data, showing that an IFN-γ response requires less B7-1 costimulation than does an IL-2 response.

In the present study, we have introduced a novel approach for quantitatively delivering costimulators to cell surfaces, and, in turn, this method has been used to look for B7-1 costimulation thresholds. Specifically, a B7-1 · Fcγ₁ fusion protein was “painted” onto tumor cells that had been precoated with chemically lipidated protein A, and the influence of varying levels of painted protein on T cell activation was monitored. Key findings included the following: 1) B7-1 · Fcγ₁ can be quantitatively titrated onto protein A-precoated cell surfaces. 2) Painted B7-1 · Fcγ₁ retains costimulator function. 3) At a specified level of TCR triggering, there is a threshold level of costimulator trans-signaling required for T cell activation. 4) There are distinct costimulator thresholds at the single-cell level that are each coupled to different T cell outputs, and the hierarchy of observed T cell cytokine responses parallels that previously documented for TCR activation thresholds.

Previous studies bear upon the question of costimulator thresholds. Itoh and Germain (8) reported that at a given Ag density, APC expressing high levels of B7-1 and ICAM-1 stimulated more IL-2-producing cells than did APC with low levels of the costimulators. In another study, increasing LFA-3 density on melanoma cells was shown to enhance cytokine production by melanoma-specific CTL clones (11). Recently, Murtaza et al. (12) reported that the level of costimulation has a significant effect on responses to an Ag, and that a strong costimulatory signal can convert a weak agonist into a full agonist and an agonist into a superagonist. Although these studies point to the importance of APC costimulator levels on elicited T cell responses, and thus support the conclusions of the present study, they were limited in two ways by the experimental systems employed. First, all of these studies used transfected and/or supertransfected sublines that were selected on the basis of low vs high costimulator expression. Although the sublines were carefully examined to ensure comparable expression of several immunoregulatory molecules, such as I-E^K (8), HLA-A2, and ICAM-1 (11), it remains possible that observed functional differences might stem from variable expression of yet other immunoregulatory molecules. Second, the costimulator levels on the transfected sublines could not be tightly controlled, and, consequently, the functional impact of low vs high, but not graded, expression of costimulators was examined in those studies. In contrast, the protein transfer approach of the present study enabled titration of costimulators at the cell surface, while avoiding the confounding effects of subline selection and consequent molecular heterogeneity.

Protein transfer offers a number of advantages over gene transfer for engineering APC. These include the suitability of the former to poorly transfectable cells (for example, biopsy-derived tumor cells), the simplicity of expressing multiple proteins on the same cell surface, and the relative ease and rapidity of the procedure (reviewed in Ref. 23). Previously, we (17, 18, 24, 25) and others (19, 20) have reported the successful application of protein transfer to costimulator and MHC proteins using their recombinant GPI-modified derivatives. However, one limitation of the GPI protein transfer strategy is in scaling up the purification of GPI proteins from membranes of the transfected host cells. The protein transfer approach presented here circumvents this problem, since Fc fusion proteins are soluble, are secreted at reasonably high levels from transfectants (∼1–3 mg/l spent culture medium for B7-1 · Fcγ₁), and are amenable to single-step purification by protein A/G chromatography.

The Fc fusion protein transfer procedure reported here builds upon the Ab protein transfer method first reported by Kim and Peacock (21). However, in contrast to their method, which invoked an 18-h dialysis step for precoating the cells with lipidated protein A, our method reduces this step to 1 h by directly incubating the lipidated protein A with the cells. The extension of Ab protein transfer to anti-CD28 mAbs, and hence costimulation, was reported recently (26). In that report, an alternative multistep approach was employed in which cells were first chemically biotinylated, protein G-biotin was linked to the biotinylated cells using avidin as a bridge, and, finally, mAb was attached to the anchored protein G. The limitation of this latter protein transfer procedure, as compared with the original one advocated by Kim and Peacock and our own, is in its dependence on chemical biotinylation of cells and the associated perturbation of resident cell surface molecules.

The principle finding of the present study is that levels of costimulators can not only tune the magnitude of an evoked immune response, but can also dictate the quality of the immune response. Taken together with the TCR activation threshold model (4, 5, 6, 7, 8, 9, 10), this argues that T cell responses at the single-cell level are quite plastic and can be driven in qualitatively different ways by altering the levels of both antigenic and costimulator inputs. The interplay between these antigenic and costimulatory inputs is likely to be of considerable interest and physiological significance, especially once insights into costimulator cooperativity emerge.

One class of tumor cell vaccines consists of immunogenic tumor cells engineered via gene or protein transfer to express lymphoid costimulators (27). Given its simplicity and effectiveness, our protein transfer method offers an attractive new option for tumor vaccine engineering. Flexible use of costimulator · Fc proteins, invoking diverse costimulator combinations that are tailored to individual tumor types, may provide significant therapeutic advantages.

We thank Joseph Manupello for his assistance with labeling of the protein with ¹²⁵I, Pan Tao and Drs. John Fayen, Jacob Rachmilewitz, and J.-H. Huang for helpful discussions, and Susan Brill for her assistance.