In situ and in silico kinetic analyses of programmed cell death-1 (PD-1) receptor, programmed cell death ligands, and B7-1 protein interaction network

Programmed cell death-1 (PD-1) is an inhibitory receptor with an essential role in maintaining peripheral tolerance and is among the most promising immunotherapeutic targets for treating cancer, autoimmunity, and infectious diseases. A complete understanding of the consequences of PD-1 engagement by its ligands, PD-L1 and PD-L2, and of PD-L1 binding to B7-1 requires quantitative analysis of their interactions at the cell surface. We present here the first complete in situ kinetic analysis of the PD-1/PD-ligands/B7-1 system. Consistent with previous solution measurements, we observed higher in situ affinities for human (h) than murine (m) PD-1 interactions, stronger binding of hPD-1 to hPD-L2 than hPD-L1, and comparable binding of mPD-1 to both ligands. However, in contrast to the relatively weak solution affinities, the in situ affinities of PD-1 are as high as those of the T cell receptor for agonist pMHC and of LFA-1 (lymphocyte function-associated antigen 1) for ICAM-1 (intercellular adhesion molecule 1) but significantly lower than that of the B7-1/CTLA-4 interaction, suggesting a distinct basis for PD-1- versus CTLA-4-mediated inhibition. Notably, the in situ interactions of PD-1 are much stronger than that of B7-1 with PD-L1. Overall, the in situ affinity ranking greatly depends on the on-rate instead of the off-rate. In silico simulations predict that PD-1/PD-L1 interactions dominate at interfaces between activated T cells and mature dendritic cells and that these interactions will be highly sensitive to the dynamics of PD-L1 and PD-L2 expression. Our results provide a kinetic framework for better understanding inhibitory PD-1 activity in health and disease.

PD-1 2 (CD279) is an immune checkpoint receptor expressed mainly on activated T cells and B cells. Its primary function is to maintain peripheral tolerance within the adaptive immune system. PD-1 deficiency results in spontaneous development of autoimmunity in mouse models (1,2), and polymorphism of its gene PDCD1 in humans is associated with various autoimmune diseases including systemic lupus erythematosus, rheumatoid arthritis, type I diabetes, etc., among different populations (3). High surface expression of PD-1 has also been found to be a hallmark of T cell exhaustion, where antigen-specific CD8 ϩ T cells lose their ability to combat tumor cells or virus-infected cells (4,5). Antibody blockade of the PD-1 pathway is able to restore the effector functions of exhausted CD8 ϩ T cells for tumor or viral clearance, and this approach is emerging as a promising immunotherapeutic strategy to treat a wide range of cancer and infectious diseases, e.g. nivolumab and pembrolizumab among others in the pipeline (3,6). In addition, the more complex functions of PD-1 are evidenced by its important role in the generation and activity of induced regulatory T cells (7)(8)(9), its high expression on follicular helper CD4 ϩ T cells (10), the improved cognitive performance in an Alzheimer's mouse model with PD-1 blockade (11), and its expression and growthpromoting effect on certain tumor cells (12).
PD-1 is a type I transmembrane glycoprotein with a single IgV domain in the extracellular region and two tyrosine-based signaling motifs in the cytoplasmic tail: an immunoreceptor tyrosine-based inhibitory motif and an immunoreceptor tyrosine-based switch motif. The potent inhibitory effect of PD-1 relies on the phosphorylation of the immunoreceptor tyrosinebased switch motif and subsequent recruitment of SHP-2, which attenuates TCR or B cell receptor (BCR) proximal signaling (13). The two ligands, PD-L1 (CD274, B7-H1) and PD-L2 (CD273, B7-DC), are both type I transmembrane glycoproteins, each consisting of an IgV and an IgC domain with high similarities to other B7 family proteins (14 -17). Although both ligands inhibit T cell function in vitro upon binding to PD-1, their in vivo effects are largely governed by their distinct expression patterns, with PD-L1 universally expressed, whereas PD-L2 is restricted to activated antigen-presenting cells (13). Interest-ingly, the structures of PD-1⅐PD-ligand complexes manifest the interactions of variable domains from antigen receptors (V H /V L for B cell receptor and V ␣ /V ␤ for TCR) by using equivalent ␤-sheets to interact with each other while leaving the loops exposed (18 -20). Moreover, a structural comparison of the apo and complexed forms of hPD-1 and mPD-1 reveals significant differences, suggesting potentially distinct ligand binding properties (20,21). In addition, PD-L1 has been shown to bind B7-1 and induce bidirectional inhibitory signaling in the absence of CD28 and CTLA-4, the previously identified receptors for B7-1 (22).
The great therapeutic potential of PD-1 and its critical role in lymphocyte biology call for a better understanding of the interactions within the complex PD-1/PD-ligands/B7-1 system. Extensive efforts have been made to interrogate these interactions using structural, mutagenesis, and surface plasmon resonance (SPR)-based approaches, establishing our current understanding of the binding interfaces involved, with hotspot regions and potentially distinct binding modes identified and solution binding properties measured (18 -20, 23-27). However, most of the kinetic studies used different dimeric protein constructs and reported kinetic parameters with large discrepancies. Recently, Cheng et al. (20) characterized in detail the monomeric-binding properties of purified PD-1, PD-ligands, and B7-1 proteins alongside a structural investigation of human PD-1. They reported relatively weak affinities of PD-1/PD-ligand interactions and even lower values for the affinity of the B7-1/PD-L1-binding interactions. We and others have shown that kinetic parameters determined in situ with molecules expressed on the native cell membrane display distinct characteristics. In contrast to solution measurements, in situ parameters reflect the effective binding properties integrating both the physical and chemical determinants of the binding interface and the modulation of molecular organization by the cellular environment (28). In addition, the in situ kinetics of TCR-pMHC interactions have been shown to correlate with antigen recognition and discrimination, signaling, effector function, and developmental fate of T cells better than their counterpart solution parameters (29 -34). In situ kinetic studies of the PD-1/PD-ligands/B7-1 system will provide us not only a direct description of how they interact on the cell membrane, a key step to delineating the interaction network, but also mechanistic insights as to how in situ and solution kinetics would correlate or differ for various molecular interactions.
Here we report a systematic in situ kinetic analysis of the PD-1/PD-ligands/B7-1 system in both human and murine species. The in situ parameters, although generally exhibiting the same trends as the solution measurements using the same protein constructs, categorize PD-1/PD-ligand binding as intermediate-to-strong interactions based on comparisons with the in situ affinities of TCR interacting with agonist pMHC or of activated lymphocyte function-associated antigen 1 (LFA-1) binding to intercellular adhesion molecule 1 (ICAM-1). In addition, the human and murine B7-1/PD-L1 interaction was found to be much weaker than the PD-1/PD-L1 interaction. Finally, simulations based on the kinetic analysis allowed us to delineate the contribution of individual receptor-ligand species to the interaction network, offering a better understanding of the immunological role of this system of interactions.

In situ ligand interactions of human PD-1
The solution binding between monomeric hPD-1 and hPD ligands displays weak solution affinities (M level K d ) despite the potent inhibitory effect they elicit (20,24). We performed micropipette adhesion frequency assays (see "Experimental procedures") using CHO cells expressing hPD-1 and RBCs coated with hPD-L1 or hPD-L2 to study their in situ interactions (Fig. 1). This assay has previously been confirmed to have single-bond sensitivity yet remains highly specific (35). For CHO cells expressing 128 hPD-1 molecules per m 2 , RBCs coated with hPD-L1 (16.4/m 2 ) or hPD-L2 (5.3/m 2 ) generated P a of 0.62 or 0.6, respectively, whereas RBCs coated with SA alone (ligand-free) gave a P a of Ͻ0.1 (Fig. 1C). The P a was also reduced to background level when hPD-1 was blocked using a monoclonal antibody (clone EH12.2H7) or using WT CHO cells without PD-1 expression, further confirming that detected bond formation is specific for the receptor-ligand pair being analyzed.
Both P a versus t c curves for hPD-L1 and hPD-L2 binding to hPD-1 fitted well to Equation 1 with R 2 of 0.9831 and 0.9987, respectively (see "Experimental procedures") at three combinations of hPD-1 and hPD-ligand densities (Fig. 2, A and B), indicating that the previously reported monomeric-binding model in solution also applies to binding on the cell surface. The effective in situ affinity (A c K a ) and off-rate (k off ) were extracted from the curve-fitting and used to derive the effective on-rate using Equation 2 (see "Experimental procedures"). Comparing the in situ kinetic parameters (Fig. 2, E-G, open columns, left ordinate) with their counterparts measured using SPR (Fig. 2, E-G, gray columns, right ordinate) (20), the effective in situ affinities follow the same rank order as the solution affinities but show a greater difference between the two ligands, with the A c K a for hPD-1/hPD-L2 3.75-fold higher than that of the hPD-1/ hPD-L1 interaction (Fig. 2E). The A c K a values, 4.74 Ϯ 0.30 ϫ 10 Ϫ4 and 2.12 Ϯ 0.56 ϫ 10 Ϫ3 m 4 for hPD-L1 and hPD-L2, respectively, are similar or higher than that of TCR interacting with potent cognate pMHC (2.4 Ϯ 0.  (20). The in situ on-rate is comparable with those of PSGL-1 interacting with P-selectin or L-selectin, where the fast on-rates facilitate the capture and rolling of trafficking leukocytes on endothelial cells (37). These in situ kinetic parameters indicate stronger interactions between PD-1 and PD-ligands on the cell surface with slower off-rate and faster on-rates than previously appreciated based on solution measurements with cell-free systems. Our results may account, at least in part, for PD-1's potent inhibition of TCR signaling.

In situ interactions of PD-L1 with B7-1
PD-L1 has been reported to interact with B7-1 and deliver inhibitory signals bidirectionally (22,26). However, it remains controversial as to how strong this interaction is compared with PD-1/PD-L1 binding (20,26). To compare their in situ kinetics, we expressed hB7-1 and mB7-1 in CHO cells and tested their binding to PD-L1-coated RBCs. B7-1 has been shown to form non-covalent dimers on the cell surface (38,39). For both human and murine cases, however, a monomeric binding model (Equation 1) with the same kinetic parameters could be fitted simultaneously to two adhesion frequency curves generated using two sets of molecular densities of B7-1 and PD-L1 (Fig. 3, A and B), suggesting that dimeric binding did not occur under our experimental conditions (40). The in situ affinities are 1.21 Ϯ 0.16 ϫ 10 Ϫ5 m 4 for the human interaction and 3.47 Ϯ 1.25 ϫ 10 Ϫ6 m 4 for the murine following the same trend as the solution affinities ( Fig. 3C, Table 1). The higher A c K a for hB7-1/hPD-L1 is largely because of its 8.3-fold faster on-rate, although its dissociation is more rapid as well (Fig.  3, D and E). Interestingly, B7-1/PD-L1 interactions are much weaker than PD-1/PD-L1 interactions, with 37-and 49-fold A c K a differences for human and murine, respectively. The differences in the in situ parameters are larger than those previously estimated by SPR experiments (20), suggesting that potential restrictions on B7-1/PD-L1 interactions are imposed by the cellular environment. Consistent with this, mPD-L1-coated RBCs generated an adhesion frequency of Ͼ50% when tested against activated CD8 ϩ T cells from WT P14 mice, whereas negligible binding was observed in cells from PD-1 Ϫ/Ϫ P14 mice despite the significant levels of B7-1 expressed by both cells (data not shown).

In situ interactions of human B7-1 with CD28 and CTLA-4
To better orient the in situ kinetics of the PD-1-ligand interactions, we analyzed the in situ interactions of hB7-1 with Upon separation, the test was scored 1 when stretch of the RBC membrane by molecular bond(s) formed during contact was observed (upper right) or 0 when no membrane stretch was observed (lower right). The same test was repeated 30 -50 cycles per cell pair yielding an averaged adhesion frequency, P a . B, adhesion frequency versus contact duration curves for hPD-1-ligand interactions. Each point represents the mean Ϯ S.E. of 3-6 cell pairs tested for the corresponding contact duration. In situ effective affinity (A c K a ) and off-rate (k off ) were extracted by fitting each curve to Equation 1 in conjugation with measured PD-1 (m r , #/m 2 ) and ligand (m l , #/m 2 ) densities by flow cytometry. C, representative specificity controls of adhesion events for hPD-1-ligand interactions including ligand-free RBCs, receptor-free WT CHO cells, or PD-1 CHO cells with anti-PD-1 blocking.

PD-1 in situ interaction
CD28 and CTLA-4, the best-studied interactions in the B7-CD28 family (Fig. 4). CTLA-4 binds to the same ligands as CD28 and antagonizes its costimulatory signaling via multiple mechanisms (41). The kinetic basis manifests as a 10-fold higher solution affinity for B7-1/CTLA-4 (42), which is further enhanced by bivalent binding on the cell membrane (43,44). The in situ affinities follow the same trend but display a 60-fold difference (2.68 Ϯ 0.05 ϫ 10 Ϫ4 versus 1.63 Ϯ 0.21 ϫ 10 Ϫ2 m 4 ; Fig. 4C). Given that low densities of B7-1 were used to reduce dimerization on the membrane, the much larger difference in in situ affinity versus solution affinity suggests that the in situ interactions of these proteins are highly differentially modulated by the cellular environment, i.e. native B7-1 is much less favored for binding to CD28 than CTLA-4, potentially due to the membrane constraints (e.g. orientation on the membrane) missed in solution measurement. Although the difference in solution affinity was largely attributed to the smaller k off of B7-1/CTLA-4, the in situ off-rates are different by only 2-fold (Fig. 4D). Instead, the in situ on-rate of B7-1/CTLA-4 binding is 50-fold higher than that of that of B7-1/CD28 binding (1.13 Ϯ 0.001 ϫ 10 Ϫ2 versus 3.78 Ϯ 0.53 ϫ 10 Ϫ4 m 4 s Ϫ1 ; Fig. 4E), accounting for the 60-fold higher in situ affinity of CTLA-4.

Simulation of complex formation at the T cell/DC interface
Due to the complex nature of the PD-1 and B7-1 network of interactions, the net outcome will depend on the types of complexes formed by each receptor against a background of competition for ligands. To gain quantitative insights into the behavior of this system, we simulated molecular complex formation between an activated human T cell and a mature dendritic cell (DC) by combining receptor and ligand expression levels (20) with the in situ binding kinetics measured in this study. Using ordinary differential equations, the model described the process of complex formation involving PD-1 and B7-1 on T cells with PD-L1 and PD-L2 on DC coupled with molecular diffusion of proteins between contacting and noncontacting membrane compartments. The equations and parameter are the same as previously reported (20) except that the kinetic parameters only were used in the present study. Examining the fraction of individual molecular complexes formed reveals the fast establishment (within 30 s) of a steady state over time wherein PD-1/PD-L1 becomes the dominant interaction species (79%) despite its ϳ3-fold lower in situ affinity than PD-1 for PD-L2 (Fig. 5A). The dominance of PD-L1 is largely explained by its 15-fold higher (80,372 versus 5,243 molecules/cell) expression level versus PD-L2 on mature DC (20). Steady-state analysis wherein the number of PD-L1 or PD-L2 on the DC is varied further reveals a range between 2,000 and 200,000 molecules/cell where the fractions of PD-1⅐PD-L1 and PD-1⅐PD-L2 complexes changes dramatically (Fig. 5, B and C). This range is physiologically feasible particularly for PD-L1 (20,45), suggesting that the contributions of PD-L1 versus PD-L2 to complex formation is highly regulated by their expression dynamics. In contrast, increasing B7-1 expression on the T cell did not significantly affect the fraction of PD-1⅐PD-L1 bonds until reaching a level of Ͼ20,000 molecules/cell (Fig. 5D), which is much higher than the estimated level of Ͻ4,000 molecules/ cell (46).

Discussion
The critical role of PD-1 in maintaining peripheral T-cell tolerance and its key suppressive effect on exhausted T cells have made it a promising therapeutic target for restoring T cell functions in a wide range of cancers and infectious diseases. To better understand the biophysicochemical basis of PD-1's function, a spatiotemporal map of ligand binding of PD-1 and

PD-1 in situ interaction
related molecules with specified kinetic properties at the intercellular junction is needed, which have not been reported previously. Here we performed a detailed in situ characterization of these kinetics, which combined with the simulation of molecular complex formation at the interface of an activated T cell and a mature DC, provide several insights as discussed below.
Molecular recognition on the T cell membrane is largely governed by the intrinsic properties of the binding sites within a receptor-ligand pair, the net effects on binding of which are usually characterized by SPR analysis or thermodynamic approaches using soluble recombinant polypeptides truncated before the membrane anchor. These measurements in conjunction with structural studies have been invaluable for understanding the binding step of molecular recognition. However, applying the solution-phase results directly to the cellular environment is not necessarily straightforward as at the cell membrane, the orientation, organization, diffusion, and even structures of the proteins are likely to be subject to modulation by the cellular environment. To mimic such physiological situations, we probed the binding to native receptors expressed on the CHO cell membrane with recombinant ligands anchored on the RBC membrane, a configuration better reflecting the representation of the native receptors and a close approximation of the native ligands. We and others have shown in multiple molecular systems that binding kinetics measured in situ often differ dramatically from the equivalent solution measurements. For example, the in situ kinetics of TCR/pMHC extend over a much wider dynamic range than solution kinetics and correlate better with the functional potency of pMHC ligands (29 -34). They are also sensitive to perturbations of the cellular environment such as inhibition of actin polymerization with latrunculin A or disruption of membrane microdomains with cholesterol oxidase (28,29). Therefore, instead of the intrinsic physicochemical properties of the binding site only, the in situ kinetics measured by our assays represent a potentially tunable property of the molecular recognition in the cellular context. The context dependence of such interactions at the very least is illustrated in the molecular interactions measured in this study by the expanded dynamic range of in situ affinities compared with their solution counterparts. The solution affinities from the weakest interaction (mB7-1/PD-L1) to the strongest (hB7-1/CTLA-4) spans 200-fold, whereas the in situ affinities have a range of almost 4 orders of magnitude (Fig. 6A). Interestingly,

PD-1 in situ interaction
the enhanced dynamic range does not undermine the general agreement between in situ and solution measurements, as the ranking of in situ affinities correlates almost perfectly with that of the solution measurements (Fig. 6A). Neither does the enhancement come from a simple transformation uniformly applied to all interactions. For example, the conversion from solution to in situ affinities according to the method of Bell (47,48) gives variable confinement lengths, a characteristic length reflecting the volume of the search space for molecular binding on the cell surface (Fig. 6B). Such variation suggests differential perturbation of in situ affinities by the cellular environment. The mechanisms, which are likely complex, may work upon differences in effective molecular length and orientation (48 -50). Moreover, it is likely that the in situ on-rate, not off-rate, will be subject to the largest effects. The on-rate governs bond formation and hence depends on processes affecting the encounter rate of the interacting molecules at the intercellular junction, which accounts for the major differences in in situ versus in-solution binding. The off-rate, on the other hand, is determined by the durability of a bond after it forms and, hence, depends to a greater extent on the physicochemical property at the interface of the molecular complex. This line of reasoning is strongly supported by the remarkable correlation between the in situ on-rate and the in situ affinity (Fig. 6C). The in situ on-rate also spans a similar dynamic range as the in situ affinity, whereas the in situ off-rate changes within a 4-fold range (Fig. 6,  C and D). Overall, our data emphasize the significance and advantages of in situ kinetic studies of membrane receptorligand interactions, which in contrast to the off-rate dominated solution affinities (20,42) are more highly dependent on onrate effects that, as suggested by our data, are a potentially tunable property of the in situ interactions.
An important question that has not been fully addressed previously is how strong PD-1-ligand binding as well as PD-L1/ B7-1 interactions are compared with the interactions of antigen receptors and other co-stimulatory and co-inhibitory receptors. With enhanced resolution the in situ kinetics now defines a more precise ranking of such interactions of interest (Fig. 6A). Using the interaction between LFA-1 and ICAM-1 as a reference, the solution K d is 185 nM when the I-domain of LFA-1 is open, corresponding to the high affinity interaction (51). The solution affinity for the human B7-1/CTLA-4 interaction is moderately lower (K d ϭ 420 nM). The SPR measurements of bivalent PD-1/PD-ligand interactions (i.e. Fc fusion proteins) yielded K d values ranging from 0.01 M to 0.8 M (22,24,26,27), whereas the monomeric interactions are much weaker (K d ϭ 2-8 M) (20). In contrast, the in situ affinity of B7-1/ CTLA-4 is 3-fold higher than that of LFA-1/ICAM-1, and PD-1/ PD-ligand in situ affinities span the range from intermediate to strong, with the strongest (hPD-1/hPD-L2) comparable to the high affinity LFA-1/ICAM-1 interaction. Similar kinetic advantage for in situ binding of these inhibitory receptors is also evident for comparisons based on another co-stimulatory receptor, CD28, interacting with B7-1; i.e. interactions that are weaker in solution are similar or even stronger on the cell membrane, whereas the differences of stronger interactions become further amplified, giving rise to 2-5-fold lower confinement lengths than CD28 (Fig. 6B). Therefore, the relatively high in situ affinities of PD-1/PD-ligand interactions in comparison to the low solution affinities may provide a better understanding of PD-1's potent inhibitory function to counter activating signals by the TCR and co-stimulatory receptors such as CD28. On the other hand, the in situ affinities of PD-1-ligand binding are considerably lower than that of CTLA-4, suggesting that distinct mechanisms for their inhibitory effects are at play. This notion is also supported by the fact that CTLA-4 shares the same ligands (B7-1 and B7-2) with the co-stimulatory receptor CD28 and is also evidenced by the different signaling pathways that are thought to act downstream of CTLA-4 and PD-1 (52). Notably, our in situ measurements indicate that both human and mouse B7-1/PD-L1 interactions are much weaker than interactions with their canonical receptors; the A c K a of B7-1/ PD-L1 binding is ϳ37-49-fold, 20-fold, or 3-log lower than that of PD-1/PD-L1 binding, B7-1/CD28 binding, or B7-1/ CTLA-4 binding, respectively. The differences are much larger than the previous solution measurements, manifesting Ͼ10fold higher confinement lengths (Fig. 6B). Again, the differences seem to have their source from the in situ on-rate instead of off-rate (Fig. 6, C and D), suggesting that these interactions occur less favorably in the cellular context. Accordingly, in silico simulations using these kinetic rates imply that B7-1 will only ever have a negligible effect on the fraction of PD-1⅐PD-L1 bonds formed at physiological levels of B7-1 expression. These results suggest that these interactions would only be functional when very abundant B7-1 and PD-L1 are present, which then would restrict the significance of this interaction to some particular physiological or pathological conditions in contrast to the broader and more potent effect of PD-1 in maintaining peripheral tolerance. Overall, understanding how these molecules interact in situ provides a basis for delineating this network of interactions under physiological conditions, where the interplay is likely to be affected by expression and affinity-based competition and signaling-based cellular regulation. The in situ kinetic parameters obtained here would also be a useful source for mathematical simulations of membrane receptor-ligand interactions in pharmacodynamics and thus facilitate drug development targeting this network of molecules.
We found distinct binding kinetics for human versus murine PD-1 interactions as well as for PD-L1 versus PD-L2 of both species, as also revealed by solution kinetic studies (20,24,27). These differential binding properties may be related to the structural differences reported earlier. Compared with mPD-1, hPD-1 has different positioning of the FG loop and also replaces the CЉ stand with a flexible loop. Both these regions contribute to ligand binding as shown by an NMR structure (20). The recent structure of a hPD-1⅐hPD-L1 complex further shows significant plasticity associated with the ligand binding of hPD-1; the CCЈ loop that adopts an open conformation in apo-hPD-1 closes upon binding to hPD-L1, whereas the CCЈ loop of apo-mPD-1 already displays a closed conformation (21). Also, the two PD-1-ligands may bind to PD-1 via structurally distinct mechanisms. Thermodynamic analysis reveals an entropically driven process for hPD-1/hPD-L1 binding, whereas a large enthalpic term is found for the hPD-1/hPD-L2 interaction (20). All these kinetic and structural differences raise the possibility of differential signaling capacities by PD-L1 versus PD-L2. On the other hand, their functions could be highly regulated by their distinct expression patterns; PD-L1 is universally expressed, whereas PD-L2 expression is restricted to professional antigen-presenting cells. Combining the kinetic and expression profiles, our model simulations demonstrate that the contribution of individual ligands on DCs in forming molecular complexes with PD-1 on T cell is likely to be wholly dominated by their expression dynamics. This result seems to argue against the non-redundant role of PD-1/PD-L2 interaction unless it triggers distinct signaling outcomes by virtue of its enhanced on-rate and capacity for rebinding and its somewhat slower off-rate. Together these structural and kinetic differences between human and murine PD-1 and their interactions with their two ligands imply that PD-1 is subject to continuous selection pressure during evolution, the optimization of which might be critical for balancing immune tolerance with the challenge of dealing with fast-evolving pathogens.

Protein coating on RBC surface
According to a protocol approved by the Institutional Review Board of the Georgia Institute of Technology, human RBCs were isolated from healthy donors, biotinylated with various concentration of Biotin-X-NHS, functionalized with saturating amount of streptavidin (SA), and washed (29). SA-coated RBCs were then incubated with biotinylated recombinant proteins and washed before adhesion frequency assay or flow cytometric analysis.

Adhesion frequency assay
The theoretical framework and detailed procedures have been reported previously (29,35,55). As shown in Fig. 1, a CHO-expressing PD-1 and a RBC coated with PD-ligand were repetitively brought in contact for a well defined duration (t c ) with a constant contact area (A c ). Adhesion frequency (P a ) was calculated over scoring 50 contact cycles, with each giving 1 for adhesion or 0 for no adhesion based on the deflection of the RBC membrane upon separation (Fig. 1A). The adhesion frequency curve P a versus t c monotonically increased with contact duration then plateaus, and the P a versus t c curve changed shape with the PD-1-ligand and level with the receptor (m r ) and ligand (m l ) densities (Fig. 1B). This curve reflects the kinetic nature of the molecular bond formation and dissociation at the cellular interface and can be well fitted to the following equation assuming a single step second-order forward, first-order reverse reaction between monomeric receptor and ligand (35).
The effective in situ affinity (A c K a ) and off-rate (k off ) were then determined from least-mean-square fitting in conjunction with measurement of receptor and ligand densities using flow cytometry. In situ on-rate was further calculated as A c k on ϭ A c K a ϫ k off (Eq. 2)

Simulations of complex formation in the PD-1/PD-ligands/B7-1 system
The mathematical model for simulating the interactions of PD-1, PD-1-ligands, and B7-1 has been published previously (20). The same set of equations and parameters was used except for the in situ binding kinetics being the numbers measured in this study. Briefly, the model describes the case where an activated T cell expressing PD-1 and B7-1 makes contact with a mature DC with PD-L1 and PD-L2. The molecular interactions at the cell-cell interface and the diffusion across different membrane compartments were integrated into coupled ordinary differential equations. The fraction for each complex in the interface was simulated over time to obtain steady-state values. Analysis of complex fractions at steady-state were examined against varying numbers of PD-L1, PD-L2, or B7-1.
Author contributions-K. L., S. J. D., and C. Z. designed the experiments. K. L. performed most of the experiments. K. L. and C. Z. analyzed the data. X. C. made the recombinant proteins. A. T. performed the in silico simulation.