Research Article

Restriction of PD-1 function by cis-PD-L1/CD80 interactions is required for optimal T cell responses

See allHide authors and affiliations

Science  10 May 2019:
Vol. 364, Issue 6440, pp. 558-566
DOI: 10.1126/science.aav7062

Sparing T cells from inhibition

Programmed cell death 1 (PD-1) is an inhibitory receptor that normally keeps T cell immune responses in check. Immunotherapy targeting PD-1 has proven successful for certain types of cancer, but it remains unclear how PD-1 is regulated. Sugiura et al. found that a costimulatory molecule, CD80, can restrict PD-1 function during the activation of T lymphocytes. Binding of CD80 to the PD-1 ligand PD-L1 in cis on primary activated dendritic cells interfered with the ability of PD-L1 to access PD-1 on T cells, which would otherwise have inhibited T cell activation. Functional insights into PD-L1–CD80 interactions may explain the outcomes of anti–PD-1 and anti–PD-L1 cancer therapy.

Science, this issue p. 558

Abstract

Targeted blockade of PD-1 with immune checkpoint inhibitors can activate T cells to destroy tumors. PD-1 is believed to function mainly at the effector, but not in the activation, phase of T cell responses, yet how PD-1 function is restricted at the activation stage is currently unknown. Here we demonstrate that CD80 interacts with PD-L1 in cis on antigen-presenting cells (APCs) to disrupt PD-L1/PD-1 binding. Subsequently, PD-L1 cannot engage PD-1 to inhibit T cell activation when APCs express substantial amounts of CD80. In knock-in mice in which cis-PD-L1/CD80 interactions do not occur, tumor immunity and autoimmune responses were greatly attenuated by PD-1. These findings indicate that CD80 on APCs limits the PD-1 coinhibitory signal, while promoting CD28-mediated costimulation, and highlight critical components for induction of optimal immune responses.

Cancer immunotherapies targeting the inhibitory co-receptors PD-1 and CTLA-4 have revolutionized cancer treatment and demonstrated durable clinical benefit for several types of tumors (1, 2). Such immunotherapies aim to destroy cancer by activating tumor-specific T cells that are otherwise nonresponsive. In addition to tumor-specific T cells, the targeted blockade of PD-1 or CTLA-4 can also activate autoreactive T cells to cause destruction to various tissues, which has been collectively termed immune-related adverse events (1, 2). The eradication of tumors and the subsequent development of immune-related adverse events in patients receiving PD-1– or CTLA-4–targeted immunotherapies have highlighted their importance in cancer and autoimmunity (17). When PD-1 interacts with either of two ligands, PD-L1 or PD-L2, it inhibits T cell receptor (TCR) activation signals by recruiting the SHP-2 protein tyrosine phosphatase (3, 6, 7). By contrast, CTLA-4 has been shown to suppress T cell activation upon interacting with either CD80 or CD86 by transducing inhibitory signals and/or hampering engagement of the costimulatory protein CD28 (in part due to the 10- to 20-fold higher affinity for CD80/86 compared with CD28) (4, 5).

Although PD-1 is a potent coinhibitory receptor, PD-1 signaling does not necessarily abolish all immune responses (3, 6, 7). It is generally believed that PD-1 mostly functions in effector T cells, whereas CTLA-4 mainly targets the T cell activation phase (8, 9). The expression of PD-L1 but not CD80/86 on tumor and tissue-parenchymal cells, and the requirement of CD28 for the activation of naïve T cells, have been proposed to account for the key roles of PD-1 and CTLA-4 at effector and activation phases, respectively (8, 9). In addition, delayed PD-1 expression is thought to explain its limited effect at the early stage of T cell activation (8, 9). However, PD-L1 and PD-L2 are also expressed on antigen-presenting cells (APCs), and PD-1 is expressed on T cells within several hours of antigen stimulation (1012). Therefore, the mechanisms underlying the differential regulation of the function of PD-1 at the activation versus effector phase of T cell responses are currently unclear.

Although PD-L1 and CD80 are well recognized as ligands of PD-1 and CD28/CTLA-4, respectively (3, 6, 7), previous studies showed that these two molecules also interact with each other (13, 14). Furthermore, PD-L1 on T cells can interact with CD80 on APCs, and CD80 on T cells engages PD-L1 on APCs (i.e., trans-PD-L1/CD80 and trans-CD80/PD-L1 interactions, respectively) to transduce coinhibitory signals for T cell activation (13, 15). However, the molecular bases of coinhibition and the physiological significance of the trans interactions between PD-L1 and CD80 remain unclear (16, 17). By contrast, other studies reported that PD-L1 and CD80 interact in cis when these molecules are overexpressed on the same cell (18, 19). Because of these controversial reports, the actual mode and the physiological relevance of their interaction are currently unclear.

In this study, we found that CD80 physiologically interacts with PD-L1 in cis on primary activated dendritic cells (DCs), which interferes with PD-L1/PD-1 binding and subsequently abrogates the function of PD-1. By generating knock-in mice harboring PD-L1- and CD80-mutants that cannot form the cis-PD-L1/CD80 duplex, we demonstrate that the function of PD-1 is persistently restricted by cis-PD-L1/CD80 interactions, which underlie the induction of optimal immune responses against foreign-, tumor-associated–, and self-antigens. Our findings reveal how costimulatory and coinhibitory signals are regulated to optimize beneficial immunity while suppressing deleterious immune responses.

CD80 interacts with PD-L1 in cis to disrupt PD-L1/PD-1 binding

To test whether PD-1 is actually engaged by its ligands during T cell activation by physiological APCs, we examined the binding capacities of PD-1 to activated CD8α+ and CD11b+ splenic DCs and thioglycollate-induced peritoneal macrophages (TG-MΦs) using the soluble form of the extracellular region of mouse PD-1 (PD-1-EC) (Fig. 1A and fig. S1). Intriguingly, PD-1-EC–binding capacities of these three cell populations largely differed, despite having similar and high PD-L1 expression levels. TG-MΦs strongly bound PD-1-EC, whereas CD8α+ and CD11b+ DCs weakly bound PD-1-EC. We next examined the possible involvement of CD80, given that its expression levels on CD8α+ and CD11b+ DCs were much higher than those on TG-MΦs (Fig. 1A) and because CD80 reportedly interacts with PD-L1 (1315, 1820). We found that CD8α+ and CD11b+ DCs from CD80-deficient mice (Cd80–/–) bound PD-1-EC as strongly as TG-MΦs, indicating that CD80 is required for the weak association between CD8α+ and CD11b+ DCs and PD-1-EC in wild-type mice (Fig. 1A and fig. S2A). Because DCs from PD-L1–deficient mice (Cd274–/–) scarcely bound PD-1-EC, but were substantially stained with an antibody (Ab) against PD-L2 (anti–PD-L2), PD-1-EC binding was determined to be mostly dependent on PD-L1 (Fig. 1A and fig. S2B). In flow cytometry, staining intensities largely depend on the quality of Abs. Therefore the strong signal obtained from the anti–PD-L2 Ab might have been due to the high detection efficiency, rather than the actual PD-L2 expression level on these DCs. Furthermore, the expression of CD80 on TG-MΦs was not sufficiently increased to substantially interfere with PD-L1/PD-1 binding, and the binding between PD-1-EC and TG-MΦs was not changed by CD80 deficiency (Fig. 1A).

Fig. 1 Disruption of PD-L1/PD-1 binding by cis-PD-L1/CD80 interactions results in impaired PD-1–mediated T cell suppression.

(A) Weak binding of PD-1-EC to splenic mature DCs. Binding intensities of PD-1-EC and anti–PD-L1, anti–PD-L2, or anti-CD80 to LPS-activated splenic CD8α+ and CD11b+ DCs and TG-MΦs from wild-type (WT), Cd80–/–, Cd274–/– mice are shown. Shaded histograms represent isotype control staining. (B and C) Attenuated binding of PD-1-EC to CD80-expressing B cell lines overexpressing PD-L1. Binding intensities of PD-1-EC and indicated Abs to IIA1.6, WEHI-231.5, CH12F3-2A, and BCL1-B20 cells overexpressing PD-L1 are shown. Shaded histograms represent isotype control staining (B). Mean of fluorescent intensities (MFI) of PD-1-EC and anti-CD80 are plotted for B cell lines overexpressing PD-L1. The solid line represents the regression line. The correlation coefficient is indicated (C). (D to F) cis-PD-L1/CD80 interactions interfere with PD-L1/PD-1 binding. Binding intensities of PD-1-EC and indicated Abs to IIA1.6ΔPD-L1 B cells expressing PD-L1 (D) or PD-L2 (E) with CD80 or CD86 are shown. Binding intensities of PD-1-EC and anti–PD-L1 to IIA1.6ΔPD-L1-PD-L1 B cells with or without coculture with IIA1.6ΔPD-L1-mock or IIA1.6ΔPD-L1-CD80 B cells are shown (F). (G) Coimmunoprecipitation of CD80 but not CD86 with PD-L1. Green, red, and yellow arrowheads indicate monomeric PD-L1, CD80 or CD86, and PD-L1/CD80 heterodimer, respectively. WCL: whole-cell lysates without cross-linking. (H to J) IL-2 production from PD-1-sufficient (left) or -deficient (right) DO11.10 T cells upon coculture with indicated APCs pulsed with the indicated amount of antigenic peptide. Percentage of PD-1–mediated inhibition of IL-2 production is indicated. (K) Binding intensities of Abs against PD-L1 (left) and CD80 (right) on IIA1.6ΔPD-L1 B cells expressing these molecules at different levels. (L) Binding intensities of PD-1-EC on IIA1.6ΔPD-L1 B cells expressing PD-L1 at the highest level and CD80 at different levels. (M) Heatmap showing relative binding intensities of PD-1-EC on IIA1.6ΔPD-L1 B cells with 25 differential expression patterns of PD-L1 and CD80. (N) Heatmap showing the percentage of PD-1–mediated inhibition of IL-2 production from DO11.10 T cells elicited by IIA1.6ΔPD-L1 B cells with 25 differential expression patterns of PD-L1 and CD80. Error bars denote SEM. Representative data from more than three independent experiments are shown (A to N).

Next, we overexpressed PD-L1 in various B cell lines that harbored varying CD80 expression levels, and tested the ability to bind PD-1-EC (Fig. 1B). PD-1-EC–binding intensities did not correlate with PD-L1 expression levels. PD-L1–overexpressing CH12F3-2A and BCL1-B20 cells were found to only weakly bind PD-1-EC despite their high PD-L1 expression. Notably, CH12F3-2A and BCL1-B20 cells showed high expression of CD80, and the PD-1-EC–binding ability of PD-L1–overexpressing B cell lines negatively correlated with CD80 expression, suggesting that CD80 interferes with the PD-L1/PD-1 association (Fig. 1C).

We directly tested whether PD-L1/PD-1 binding was disrupted by CD80. Mouse IIA1.6 B cell lymphoma lines express PD-L1 but not PD-L2. Thus, we knocked out the gene encoding PD-L1 using the CRISPR-Cas9 system in IIA1.6 B cells (hereafter termed IIA1.6ΔPD-L1 B cells) (fig. S3A) (21, 22). Then we evaluated the PD-1–binding capacities of IIA1.6ΔPD-L1 cells overexpressing PD-L1, PD-L2, CD80, and CD86 in various combinations by using PD-1-EC. PD-1-EC binding to IIA1.6ΔPD-L1 B cells overexpressing PD-L1 (i.e., IIA1.6ΔPD-L1-PD-L1 B cells) was completely blocked by the simultaneous overexpression of CD80 but not CD86 (Fig. 1D). By contrast, PD-1-EC binding to IIA1.6ΔPD-L1-PD-L2 B cells was not affected by the simultaneous overexpression of CD80 and CD86 (Fig. 1E). The addition of IIA1.6ΔPD-L1-CD80 B cells to IIA1.6ΔPD-L1-PD-L1 B cells, which permits trans- but not cis-interactions of PD-L1 and CD80, did not affect PD-1-EC binding to IIA1.6ΔPD-L1-PD-L1 B cells, indicating that CD80 needs to be expressed on the same cell to interfere with PD-L1/PD-1 binding (Fig. 1F). Notably, the cis-PD-L1/CD80 interaction did not interfere with CD80/CD28 or CD80/CTLA-4 binding (fig. S4A). To examine the effect of cis-PD-L1/CD80 interactions on the costimulatory and coinhibitory functions of CD28 and CTLA-4, respectively, we generated DO11.10 T cells deficient in PD-L1, PD-1, and CD28 (termed DO11.10ΔPD-L1ΔPD-1ΔCD28 T cells) using CRISPR-Cas9 and compared these cells with or without overexpression of CD28 and CTLA-4 (fig. S3B). We observed comparable costimulatory and coinhibitory effects of CD28 and CTLA-4 respectively, in the presence or absence of PD-L1 on APCs (fig. S4, B to G). These results suggest that cis-PD-L1/CD80 interactions do not attenuate the function of CD28 and CTLA-4. Further confirming the cis interaction between PD-L1 and CD80, we observed that CD80 but not CD86 were coimmunoprecipitated with PD-L1 when neighboring cell-surface proteins were cross-linked with a cell-impermeable cross-linker, bis(sulfosuccinimidyl)suberate (BS3) (Fig. 1G).

cis-PD-L1/CD80 interactions inhibit PD-L1 from eliciting PD-1 functions

To test the functional consequences of the aforementioned interactions, we used IIA1.6ΔPD-L1 B cells overexpressing PD-L1, PD-L2, CD80, or CD86 in various combinations as APCs to stimulate DO11.10 T hybridoma cells that recognize pOVA323-339/I-Ad. Upon antigen stimulation, interleukin-2 (IL-2) production from DO11.10 T cells was strongly inhibited when either PD-L1 or PD-L2 was expressed on APCs (Fig. 1H). PD-L1– or PD-L2–mediated inhibition appeared dependent on PD-1 because the suppression of IL-2 production was not observed in DO11.10ΔPD-1 T cells. As expected, the simultaneous expression of CD80 but not CD86 completely abrogated the PD-1–mediated inhibitory effects caused by IIA1.6ΔPD-L1 B cells overexpressing PD-L1 but not PD-L2 (Fig. 1, I and J). Thus, CD80 interacts with PD-L1 in cis to hinder PD-L1 from triggering PD-1 function during T cell activation.

We also examined the dose-dependent effects of cis-PD-L1/CD80 interactions. CD80 and PD-L1 were expressed independently or simultaneously on IIA1.6ΔPD-L1 B cells at four different expression levels (comprising 25 combinations in total) (Fig. 1K). As expected, PD-1-EC binding to PD-L1 was attenuated when substantial amounts of CD80 were expressed simultaneously (Fig. 1, L and M). APCs with PD-1-EC–binding capacity could elicit PD-1 function and suppress IL-2 production from DO11.10 T cells upon antigen stimulation (Fig. 1N). These results suggest that the relative amount of PD-L1 and CD80 determines the capacity of cells to associate with PD-1 and elicit PD-1 function.

As mentioned previously herein, trans interactions between PD-L1 on T cells and CD80 on APCs, and the association of CD80 on T cells with PD-L1 on APCs, both transduce coinhibitory signals. We therefore tested how trans interactions between PD-L1 and CD80 affect the activation of DO11.10 T cells. A previous study reported that coinhibition by trans PD-L1 and CD80 interactions was most evident in the absence of the CD28 signal (13). Thus, we used DO11.10ΔPD-L1ΔPD-1ΔCD28 T cells (fig. S3B), in which the genes encoding CD28, PD-L1, and PD-1 have been deleted using CRISPR-Cas9. DO11.10ΔPD-L1ΔPD-1ΔCD28 T cells overexpressing CD80 or PD-L1 were stimulated with pOVA-pulsed IIA1.6ΔPD-L1 B cells overexpressing PD-L1 or CD80, respectively (fig. S5, A to D). IL-2 production from DO11.10ΔPD-L1ΔPD-1ΔCD28 T cells overexpressing CD80 or PD-L1 was not suppressed either by trans-CD80/PD-L1 or trans-PD-L1/CD80 interactions, respectively, even though the expression levels of CD80 and PD-L1 on these cells were much higher than those on activated primary T cells (fig. S6). These results suggest that PD-L1 mainly interacts with CD80 in cis rather than in trans.

PD-1–mediated inhibitory effects are abrogated by cis-CD80/PD-L1 interactions on mature primary DCs

Primary CD8α+ and CD11b+ DCs and TG-MΦs were used as APCs to stimulate T lymphoma cells that respond to a major histocompatibility complex class I (MHC I)–restricted peptide (H2-Kb-pOVA257-264, BW-OT-I cells) or an MHC II–restricted peptide (I-Ab-pOVA323-339, BW-OT-II cells) (Fig. 2, A to D). We detected substantial PD-1–mediated inhibitory effects on the activation of BW-OT-I and BW-OT-II cells when TG-MΦs were used as APCs. In contrast, PD-1 did not inhibit the activation of BW-OT-I and BW-OT-II cells when CD8α+ and CD11b+ DCs were used as APCs, which is in agreement with the strong association between PD-1-EC and TG-MΦs and the weak association between PD-1-EC and CD8α+ or CD11b+ DCs. Collectively, the ability of PD-1 to inactivate T cells is restricted upon T cell-DC interactions, likely because CD80 on DCs interacts with PD-L1 in cis to prevent PD-L1 from engaging PD-1.

Fig. 2 cis-PD-L1/CD80 interactions block PD-1 from inhibiting T cell activation in mature primary DCs.

(A to D) Activated DCs are unable to elicit PD-1–mediated inhibition. IL-2 production from BW-OT-I (A) and BW-OT-II (C) cells stimulated by LPS-activated TG-MΦs and splenic CD8α+ and CD11b+ DCs pulsed with the indicated amount of antigenic peptides is shown. Blocking Abs against PD-L1 and PD-L2, and isotype control IgG, were added as indicated. Extended analyses of (B) (10 pM) and (D) (1 μM) show the percentage inhibition of IL-2 production mediated by PD-L1, PD-L2, and PD-L1/2 upon stimulation with the indicated APCs (B and D). One-way analysis of variance (ANOVA) with Dunnett’s post hoc test (B and D). *p < 0.05; **p < 0.01; ns, not significant. Error bars denote SEM. Representative data from more than three independent experiments are shown (A to D).

PD-L1-Y56A and CD80-L107E mutants fail to interact with CD80 and PD-L1 in cis respectively

We then isolated mutants of PD-L1 and CD80 that cannot bind CD80 and PD-L1, respectively (yet retain the other functions). The binding characteristics of chimeric molecules in which immunoglobulin V (IgV) and IgC domains of PD-L1, PD-L2, CD80, or CD86 were swapped revealed that IgV domains of PD-L1 and CD80 are involved in their interaction (fig. S7, A to D). We therefore introduced random mutations in the IgV domain of PD-L1 by using error-prone polymerase chain reaction and overexpressed these mutants in IIA1.6ΔPD-L1-CD80 B cells, which was followed by sorting the cells on the basis of their acquisition of PD-1-EC–binding capacity (fig. S5E). PD-L1 mutants isolated from cells with PD-1-EC-binding ability had high mutation rates at V54, Y56, and E58 (fig. S7F). The predicted three-dimensional (3D) structure of mouse PD-L1 indicated that these amino acid residues were located in the C strand of PD-L1, which is close to the PD-1–interacting surface (Fig. 3A) (23). These results suggest that the PD-L1/CD80 interaction surface partially overlaps with that of PD-L1/PD-1. By examining a series of point mutants at Y56, we elected to use PD-L1-Y56A for subsequent analyses because this mutant bound to PD-1-EC comparably in the presence or absence of CD80 (Fig. 3C and fig. S7G) and was expressed in IIA1.6ΔPD-L1 B cells at similar levels to wild-type PD-L1 (fig. S7H).

Fig. 3 PD-L1-Y56A and CD80-L107E mutants abrogate cis-PD-L1/CD80 interactions.

(A and B) Predicted 3D structures of mouse PD-L1 (A) and mouse CD80 (B). Amino acid residues that affected cis-PD-L1/CD80 interactions are indicated (E, Glu; L, Leu; V, Val; and Y, Tyr). (C) Binding of PD-1-EC to PD-L1-Y56A in the presence of CD80. Binding intensities of PD-1-EC to IIA1.6ΔPD-L1 B cells expressing the indicated molecules are shown. (D and E) Binding of PD-1-EC to PD-L1 in the presence of CD80-L107E. Binding intensities of PD-1-EC (D), anti-CD80, CD28-EC, and CTLA-4-EC (E) to IIA1.6ΔPD-L1 B cells expressing indicated molecules are shown. (F) Defective coimmunoprecipitation of PD-L1-Y56A/CD80 and PD-L1/CD80-L107E. Green, red, and yellow arrowheads indicate monomeric PD-L1, CD80, and PD-L1/CD80 heterodimer, respectively. WCL: whole-cell lysates without cross-linking. (G to J) IL-2 production from PD-1–sufficient (left) or –deficient (right) DO11.10 T cells upon coculture with APCs expressing the indicated molecules. Indicated amounts of antigenic peptide were used. (K) Extended analysis of data presented in (G). Percentage of PD-1–dependent inhibition mediated by PD-L1 and PD-L1-Y56A is shown. (L) Extended analysis of data presented in (H to J). Percentage of PD-1–dependent inhibition mediated by APCs expressing indicated molecules pulsed with 0.3 μM antigenic peptide is shown. One-way ANOVA with Dunnett’s post hoc test (L). ***p < 0.001. Error bars denote SEM. Representative data from more than three independent experiments are shown (C to L).

The IgV domain of CD80 consists of nine β strands that are termed A, B, C, C’, C”, D, E, F, and G strands. A, G, F, C, C’, and C” strands are assembled into a β sheet that forms one side surface of the molecule termed AGFCC’C” face, whereas D, E, and B strands are assembled into another β sheet that forms the other side of the molecule termed DEB face (24). Because CD28 binds to the AGFCC’C” face of CD80 (2426) and CD80 can bind CD28 in the presence of PD-L1 (fig. S4A), we focused on the other surface (i.e., DEB face). By comparing the hydrophobicities of CD80 (24) and CD86 (27), we found a distinctive hydrophobic patch in the DEB face of CD80 (fig. S8, A and B). We next mutated amino acid residues in this hydrophobic area and found that CD80 with either the L96E or L107E mutation failed to interfere with PD-L1/PD-1 binding (Fig. 3D and fig. S8C). We therefore used CD80-L107E for subsequent analyses because (i) this mutant did not interfere with PD-L1/PD-1 binding, (ii) it bound CD28 and CTLA-4 in a manner comparable to that of wild-type CD80, and (iii) it showed expression levels in IIA1.6ΔPD-L1 B cells similar to those of wild-type CD80 (Fig. 3, D and E). We also observed that coimmunoprecipitations of CD80 with PD-L1-Y56A and of CD80-L107E with PD-L1, respectively, were severely attenuated and nearly abolished (Fig. 3F).

The function of these mutants was next tested (Fig. 3, G to L). When IIA1.6ΔPD-L1 B cells expressing PD-L1-Y56A were used as APCs to stimulate DO11.10 T cells, PD-L1-Y56A produced PD-1–mediated inhibitory effects at levels comparable to those of wild-type PD-L1 in the absence of CD80 (Fig. 3, G and K). In addition, PD-1 function occurred with PD-L1-Y56A in the presence of CD80 (Fig. 3, H and L), indicating that it can evade the inhibitory effect of CD80 and elicit PD-1 activity. When CD80-L107E was expressed on IIA1.6ΔPD-L1 B cells, CD80-L107E enhanced IL-2 production from DO11.10 T cells after antigen stimulation, similar to that seen with wild-type CD80. In addition, PD-L1 was found to trigger PD-1 functions in the presence of CD80-L107E, indicating that CD80-L107E cannot interfere with the function of PD-L1, similar to what was observed for CD86 (Fig. 3I, J, and L).

Abrogation of PD-1 function by cis-PD-L1/CD80 interactions also occurs in human orthologs

To examine whether the abrogation of PD-1–mediated inhibitory effects by cis-PD-L1/CD80 interactions occurs in human orthologs (as well as mouse orthologs), we carried out similar experiments using human orthologs. Indeed, human CD80 but not CD86 bound to human PD-L1 in cis but not in trans and attenuated the binding of human PD-L1 to human PD-1 (fig. S9, A to C). Human CD80 but not CD86 could be coimmunoprecipitated with human PD-L1 (fig. S9D). In addition, cis interaction between human CD80 and human PD-L1 precluded human PD-L1 from mediating the inhibitory function of PD-1 (fig. S9, E to G).

We then isolated the human PD-L1 mutant, PD-L1-N63D/G119S, which was unable to undergo cis-PD-L1/CD80 interactions, by employing the same strategy that we used to obtain mouse PD-L1-Y56A (fig. S10A). In the absence of human CD80, human PD-L1-N63D/G119S was able to bind human PD-1 at levels comparable to those of wild-type PD-L1. In addition, human PD-L1-N63D/G119S bound human PD-1 comparably in the presence or absence of human CD80 (fig. S10B). Next, we examined human CD80-I92E and CD80-L104E, which correspond to mouse CD80-L96E and CD80-L107E, respectively, and found that these mutants completely lacked cis-PD-L1/CD80 interactions and failed to interfere with the binding between human PD-L1 and human PD-1 (fig. S10C).

Functional analysis revealed that human PD-L1-N63D/G119S produced human PD-1–mediated inhibitory effects comparably to wild-type PD-L1 in the absence of human CD80 (fig. S10D). In addition, human PD-L1-N63D/G119S triggered PD-1–mediated inhibitory effects in the presence of human CD80 (fig. S10E), indicating that human PD-L1-N63D/G119S can evade the effect of CD80, resulting in PD-1 activity. Likewise, CD80-L104E augmented IL-2 production from DO11.10 T cells upon antigen stimulation comparably to wild-type CD80. We also observed that human PD-L1 resulted in PD-1 function in the presence of human CD80-L104E, indicating that human CD80-L104E cannot hinder PD-L1 from eliciting PD-1 function similar to that observed for human CD86 (fig. S10, F and G). Our data suggest that the abrogation of PD-1 function by cis-PD-L1/CD80 associations could be a universal strategy to mitigate the inhibitory effect of PD-1 during the activation of T cells upon T cell–DC interactions.

DCs from Cd274Y56A and Cd80L107E mice have substantial PD-1–binding capacities

To explore the biological significance of cis-PD-L1/CD80 interactions in vivo, we generated knock-in mice expressing PD-L1-Y56A or CD80-L107E by using CRISPR-Cas9 (28) (hereafter termed Cd274Y56A and Cd80L107E mice) (figs. S11 and S12). Cd274Y56A and Cd80L107E mice were born normally and did not show overt spontaneous phenotypes. We stimulated CD8α+ and CD11b+ splenic DCs and granulocyte-macrophage colony-stimulating factor–induced bone marrow–derived dendritic cells (BM-DCs) from knock-in mice with lipopolysaccharide (LPS), the major component of the outer membrane of Gram-negative bacteria and the agonist of Toll-like receptor 4 (TLR4) (29). As expected, LPS-activated CD8α+ and CD11b+ splenic DCs and BM-DCs from knock-in mice showed much stronger PD-1-EC–binding ability compared to those from wild-type mice despite similar expression profiles for CD80, PD-L1, and PD-L2 (Fig. 4A and fig. S13A). The expression levels of CD86 and MHC II on these cells were also comparable (fig. S13A). Moreover, the binding intensities of CD28-EC and CTLA-4-EC on LPS-activated CD8α+ and CD11b+ splenic DCs from Cd274Y56A and Cd80L107E mice were comparable to those from wild-type mice, as expected, on the basis of experimental results with cultured cells (Fig. 3E and figs. S4A and S13B).

Fig. 4 Defective T cell responses in Cd274Y56A and Cd80L107E mice.

(A) Contour plots showing binding intensities of PD-1-EC and anti-CD80 on LPS-activated BM-DCs and splenic CD8α+ and CD11b+ DCs from Cd274Y56A, Cd80L107E, and wild-type mice. Frequencies of PD-1-EC+ cells are indicated. (B and C) PD-1–mediated inhibition of IL-2 production from T cells upon activation with BM-DCs from Cd274Y56A and Cd80L107E but not wild-type mice. IL-2 production from BW-OT-I (left) and BW-OT-II (right) cells upon stimulation with BM-DCs from Cd274Y56A, Cd80L107E, and wild-type mice are shown (B). PD-1–mediated inhibition of IL-2 production compared with BM-DCs from wild-type mice in the presence of the indicated Abs is shown for BW-OT-I (10 pM) and BW-OT-II (0.1 μM) cells (C). (D) PD-1–mediated inhibition of IL-2 production from BW-OT-I (left, 10 pM) and BW-OT-II (right, 0.1 μM) cells upon stimulation with splenic CD8α+ and CD11b+ DCs from mice with the indicated genotype. (E and F) Defective induction of OVA-specific T cell responses upon immunization in the absence of cis-PD-L1/CD80 interactions. Draining LN cells from Cd274Y56A, Cd80L107E, and wild-type mice immunized with OVA were restimulated with the indicated antigens (n = 9 each). Concentration of IFN-γ (E) and IL-2 (F) is shown. One-way ANOVA with Dunnett’s post hoc test (C to F). *p < 0.05; **p < 0.01; ***p < 0.001. Error bars denote SEM. Representative data from more than three independent experiments are shown (A to D).

Splenic DCs from Cd80L107E and Cd80–/– mice were next stimulated with the additional agonists of TLRs, namely poly(I:C) (TLR3), CpG (TLR9), R848 (TLR7), and zymosan (TLR2/6) (fig. S14, A and B). All TLR agonists tested induced the expression of PD-L1, PD-L2, and CD80, but to variable degrees on CD8α+ and CD11b+ DCs. Accordingly, DCs from Cd80L107E and Cd80–/– mice strongly bound PD-1-EC upon stimulation with all TLR agonists tested, whereas the binding of PD-1-EC was attenuated in wild-type mice to variable degrees. Because the PD-1-EC–binding capacities were comparable between DCs from Cd80L107E and Cd80–/– mice, CD80-L107E appeared to almost completely lack cis-PD-L1/CD80 interactions (fig. S14, A and B). The expression levels of PD-L1 showed a strong correlation with the PD-1-EC–binding intensities in activated DCs from Cd80L107E and Cd80–/– mice, whereas this correlation was much weaker in those from wild-type mice (fig. S14C). By contrast, the ratio of PD-L1 to CD80 strongly correlated with the PD-1-EC–binding capacities in activated DCs from wild-type but not Cd80L107E mice (fig. S14D). These results revealed the PD-1-EC–binding properties of DCs in response to stimulation with TLR agonists and further confirmed that PD-1–binding ability is dependent upon the relative amount of PD-L1 and CD80.

T cell responses to immunogens are severely attenuated in Cd274Y56A and Cd80L107E mice that lack cis-PD-L1/CD80 interactions

In accordance with the binding capacities of DCs to PD-1-EC, activated DCs from Cd274Y56A and Cd80L107E mice, but not those from wild-type mice, elicited substantial PD-1–mediated inhibitory effects on the activation of BW-OT-I and BW-OT-II cells (Fig. 4, B to D). PD-1–mediated inhibition was mostly dependent on PD-L1, and the contribution of PD-L2 was marginal (Fig. 4, B and C), as was the case for PD-1-EC binding (Fig. 1A).

Next, we examined the effects of cis-PD-L1/CD80 interactions on immune responses to foreign antigens in vivo. We immunized Cd274Y56A and Cd80L107E mice with OVA protein emulsified in complete Freund’s adjuvant and restimulated T cells in the draining lymph nodes 1 week later (Fig. 4, E and F). Interferon-γ (IFN-γ) and IL-2 production from T cells, upon stimulation with OVA protein and MHC I– and MHC II–restricted OVA peptides, was significantly reduced in both knock-in mice compared to those in wild-type mice, indicating that PD-1 function is restricted in wild-type mice by cis-PD-L1/CD80 interactions during the induction of immune responses against foreign antigens.

Antitumor immune responses are severely attenuated in Cd274Y56A and Cd80L107E mice lacking cis-PD-L1/CD80 interactions

Next, we tested the involvement of cis-PD-L1/CD80 interactions in antitumor immune responses. Mice were inoculated with E.G7 T lymphoma cells that express OVA (30) and vaccinated with OVA protein together with poly(I:C) (Fig. 5, A to C). Tumor growth was substantially suppressed by vaccination in wild-type mice, whereas significant curative effects were not observed in Cd274Y56A or Cd80L107E mice (Fig. 5, A to C). To directly examine the role of cis-PD-L1/CD80 interactions in APCs, we tested DC-based vaccination. Wild-type mice inoculated with E.G7 T lymphoma cells were immunized with OVA-pulsed BM-DCs from wild-type and knock-in mice (Fig. 5, D to F). Consistent with the effects of protein vaccination, immunization with BM-DCs from wild-type but not Cd274Y56A or Cd80L107E mice induced substantial antitumor immune responses (Fig. 5, D to F).

Fig. 5 Vaccine-induced antitumor immune responses are limited in the absence of cis-PD-L1/CD80 interactions.

(A to F) Defective induction of antitumor immune responses in the absence of cis-PD-L1/CD80 associations. Experimental design of the vaccination with OVA and poly(I:C) (A). The absolute (B) and relative (at day 15) (C) volumes of E.G7 tumors in Cd274Y56A, Cd80L107E, and wild-type mice immunized with phosphate-buffered saline (Ctrl) or OVA and poly(I:C) (Vac) are shown (number of mice ≥ 8 each). Experimental design of the vaccination with OVA-pulsed BM-DCs (D). The absolute (E) and relative (at day 14) (F) volumes of E.G7 tumors in wild-type mice immunized with BM-DCs from Cd274Y56A, Cd80L107E, and wild-type mice with or without OVA are shown (number of mice = 7 each). (G) PD-1-EC–binding intensities and PD-L1-, PD-L2-, CD80-, and CD86-expression levels of CD11cintMHCIIhighXCR1+OVA+ and CD11cintMHCIIhighCD11b+OVA+ migratory DC subsets in skin dLNs from Cd274Y56A, Cd80L107E, and wild-type mice immunized with Alexa488-OVA and poly(I:C). (H to J) Defective induction of tumor antigen–specific CD8+ T cells in the absence of cis-PD-L1/CD80 interactions. Skin dLNs cells from Cd274Y56A, Cd80L107E, and wild-type mice were stained with H-2Kb-OVA-tetramer 7 days after immunization with OVA and poly(I:C). Representative contour plots (H), frequency (I), and the absolute number (J) of H-2Kb-OVA-tetramer+ cells are shown (number of mice = 9 each). Unpaired two-tailed Student’s t test (B and E), one-way ANOVA with Dunnett’s post hoc test (C, F, I, J). *p < 0.05; **p < 0.01; ***p < 0.001. Error bars denote SEM.

We then characterized APCs that present antigens to tumor-specific T cells. We immunized mice subcutaneously with Alexa488-labeled OVA protein and poly(I:C) and analyzed DCs that took up antigen and migrated to the draining lymph nodes (dLNs) (Fig. 5G). Binding of PD-1-EC to XCR1+ and CD11b+ migratory DCs, which preferentially activate CD8+ and CD4+ T cells, respectively (31, 32), was substantially reduced in wild-type mice compared to that in Cd274Y56A and Cd80L107E mice. Notably, the expression levels of PD-L1, PD-L2, CD80, CD86, CD40, MHC I, and the other costimulatory and coinhibitory ligands were comparable among wild-type and knock-in mice, with less than a 1.3-fold difference (Fig. 5G and fig. S15). These results suggest that the engagement of PD-1 on tumor-specific T cells with PD-L1 on APCs is restricted by cis-PD-L1/CD80 interactions in vivo.

We then evaluated the generation of antigen-specific CD8+ T cells upon immunization by using MHC I-tetramers presenting OVA peptide (H-2Kb-pOVA-tetramer). The frequency and number of OVA-specific CD8+ T cells upon immunization were considerably lower in Cd274Y56A and Cd80L107E mice compared to those in wild-type mice (Fig. 5, H to J). By contrast, the tetramer staining intensity and the expression levels of other activation markers were not changed (fig. S16). Collectively, these data indicate that PD-1 function is persistently restricted by cis-PD-L1/CD80 interactions during the initiation of antitumor immune responses.

cis-PD-L1/CD80 interactions reduce PD-1 function to exacerbate autoimmunity

We then explored the role of cis-PD-L1/CD80 interactions in autoimmunity by inducing experimental autoimmune encephalomyelitis (EAE), which is a mouse model of multiple sclerosis (Fig. 6A). Cd274Y56A and Cd80L107E mice showed significantly reduced signs of EAE upon immunization with the myelin oligodendrocyte glycoprotein (MOG) peptide compared to those in wild-type mice (Fig. 6B). Further, migratory DCs in skin dLNs from immunized Cd274Y56A and Cd80L107E mice showed higher PD-1-EC–binding ability compared to those from wild-type mice. We consistently observed a substantial reduction in IL-17 production by splenocytes from immunized Cd274Y56A and Cd80L107E mice after restimulation with the MOG peptide as compared to that in wild-type mice (Fig. 6D). These findings indicate that the suppressive effects of PD-1 are prevented during the development of EAE by cis-PD-L1/CD80 interactions in wild-type mice and that the liberation of PD-L1 from CD80 can restore PD-1–mediated inhibitory effects and alleviate the signs of EAE in Cd274Y56A and Cd80L107E animals.

Fig. 6 Autoimmunity is alleviated in the absence of cis-PD-L1/CD80 interactions.

(A) Experimental design of EAE. (B) Milder autoimmune signs in the absence of cis-PD-L1/CD80 interactions. The clinical score of EAE in Cd274Y56A, Cd80L107E, and wild-type mice is shown (number of mice = 12 each). (C) PD-1-EC–binding intensities and PD-L1-, PD-L2-, CD80-, and CD86-expression levels of CD11cintMHCIIhighXCR1+ and CD11cintMHCIIhighCD11b+ migratory DC subsets in skin dLNs from Cd274Y56A, Cd80L107E, and wild-type mice on day 1. (D) Defective induction of IL-17–producing cells upon induction of EAE in the absence of cis-PD-L1/CD80 interactions (number of mice = 8 each). Two-way repeated measure ANOVA with Tukey HSD post hoc test (B), one-way ANOVA with Dunnett’s post hoc test (D). *p < 0.05; **p < 0.01; ***p < 0.001. Error bars denote SEM.

Discussion

The identification of trans-PD-L1/CD80 interactions highlights the complexity of signaling within the B7/CD28 family and has been proposed to explain the difference between anti–PD-1 and anti–PD-L1 in cancer immunotherapy (3335). For example, it has been proposed that anti–PD-L1 might provide better clinical efficacy than anti–PD-1, because anti–PD-L1 can block inhibitory signals through PD-L1 and CD80 in addition to PD-1 (33). However, appreciable differences have not yet been recognized in clinical efficacies between anti–PD-1 and anti–PD-L1, and the actual significance of trans interactions of PD-L1 and CD80 between T cells and APCs has been elusive. In this study, we demonstrate that CD80 interacts with PD-L1 in cis on the same APC and inhibits the ability of PD-L1 to bind PD-1 and inhibit T cell activation. Nonetheless, our results do not exclude the possible interactions and functions of PD-L1 and CD80 in trans in certain conditions or model systems.

Ostrand-Rosenberg and colleagues reported that the overexpression of CD80 on PD-L1–expressing tumor cells attenuates the binding of PD-1 to tumors (19, 20). On the basis of this finding, they proposed the application of the soluble protein of CD80 (CD80-Ig) for cancer immunotherapy, as CD80-Ig might block inhibitory signals through PD-L1/CD80 as well as PD-L1/PD-1 (20). However, the mode of interaction between PD-L1 and CD80 was not clarified in these studies, and the interaction between CD80 and PD-L1 under physiological conditions was not demonstrated. Another recent study reported that the co-overexpression of CD80 and PD-L1 on the same cell resulted in their interaction in cis (18), and competition between PD-1 and CD80 for PD-L1 binding at the protein level was also observed. However, the functional consequence of the competition was not explored, and their interactions under physiological conditions were not shown. Furthermore, the significance of cis-PD-L1/CD80 interactions in the induction and/or regulation of immune responses was not known. We report here that in primary DCs, CD80 bound PD-L1 in cis to strongly interfere with PD-L1/PD-1 binding. In addition, we observed that cis-PD-L1/CD80 interactions strongly abrogate PD-1 function. We generated two lines of knock-in mice in which cis-PD-L1/CD80 interactions were specifically attenuated and found that PD-1 function is persistently restricted by cis-PD-L1/CD80 associations during the immune responses to foreign, tumor-associated-, and self-antigens. Intriguingly, the PD-1–binding capacities of DCs were regulated differently depending on the TLR agonist used, suggesting that PD-1 might mitigate T cell responses to pathogens to varying degrees depending on the pathogen-associated molecular patterns.

Although PD-1 is a potent coinhibitory receptor and can strongly inactivate T cells, especially at the effector phase during autoimmune disorders or response to cancer (8, 9), PD-1 does not necessarily abolish all T cell responses, suggesting that there might be a mechanism by which PD-1 function is restricted. To date, the limitation of PD-1 function at the immune activation phase is generally explained by the expression of PD-1 on effector T cells (8, 9). However, PD-1 has been shown to appear on T cells within several hours after antigen stimulation, making the delayed expression of PD-1 less likely to be the core mechanism (10, 11). Our findings revealed the rational and elaborated mechanisms by which CD80 augments T cell activation not only by mediating the CD28 costimulatory signaling but also by attenuating PD-1–driven coinhibitory signals. In vivo studies using Cd274Y56A and Cd80L107E mice indicated that the restriction of PD-1 function by cis-PD-L1/CD80 interactions is critical for the induction of optimal immune responses against foreign, tumor-associated-, and self-antigens. Owing to the success of cancer immunotherapies targeting PD-1, the number of clinical trials is expanding. Targeted manipulation of cis-PD-L1/CD80 interactions might therefore lead to new therapeutic opportunities for the treatment of cancer, autoimmune diseases, and chronic inflammation.

Supplementary Materials

science.sciencemag.org/content/364/6440/558/suppl/DC1

Materials and Methods

Figs. S1 to S16

Table S1

References (3640)

References and Notes

Acknowledgments: We thank L. Ignatowicz, T. Kurosaki, and T. Kitamura for providing BW-1100.129.237, IIA1.6, and Plat-E cells, respectively; T. Honjo for providing DO11.10 T and WEHI-231.5 cells; H. Tsuduki, Y. Okamoto, M. Aoki, A. Otsuka, and R. Matsumura for technical and secretarial assistance; and the other members of our laboratory for helpful discussions. Funding: This work was supported in part by the Core Research for Evolutional Science and Technology Program of the Japan Science and Technology Agency, Basic Science and Platform Technology Program for Innovative Biological Medicine of the Japan Agency for Medical Research and Development (JP18am0301007), and Grant-in-Aid by the Japan Society for the Promotion of Science (JP18H05417 and JP19H01029). Author contributions: D.S. and T.O. designed the experiments. D.S., T.M., I.m.-O., K.S., and T.K.M. established experimental systems and performed the in vivo and in vitro experiments using cultured cells and mice. D.S. and T.T. generated gene-targeted cells and mice. D.S. and T.O. wrote the manuscript with all authors contributing to writing. T.O. supervised the project. Competing interests: D.S., T.K.M., and T.O. are inventors on provisional patent JP2018-229774, submitted by Tokushima University and Ono Pharmaceutical Co., Ltd., that covers the use of antibodies against immune checkpoint molecules for immunotherapy. The other authors declare no competing financial interests. Data and materials availability: The data presented in this manuscript are tabulated in the main paper and the supplementary materials.
View Abstract

Stay Connected to Science

Navigate This Article