PD-L2 suppresses T cell signaling via coinhibitory microcluster formation and SHP2 phosphatase recruitment

The coinhibitory receptor, PD-1, is of major importance for the suppression of T cell activation in various types of immune responses. A high-resolution imaging study showed that PD-1 forms a coinhibitory signalosome, “PD-1 microcluster”, with the phosphatase, SHP2, to dephosphorylate the TCR/CD3 complex and its downstream signaling molecules. Such a consecutive reaction entirely depended on PD-1–PD-L1/2 binding. PD-L2 is expressed on professional antigen-presenting cells and also on some tumor cells, which possibly explains the discrepant efficacy of immune checkpoint therapy for PD-L1-negative tumors. Here, we performed precise imaging analysis of PD-L2 forming PD-1–PD-L2 clusters associating with SHP2. PD-L2 could compete with PD-L1 for binding to PD-1, occupying the same space at TCR microclusters. The PD-1 microcluster formation was inhibited by certain mAbs with functional consequences. Thus, PD-1 microcluster formation provides a visible index for the effectiveness of anti-PD-1- or anti-PD-L1/2-mediated T cell suppression. PD-L2 may exert immune suppressive responses cooperatively with PD-L1 on the microcluster scale. Takehara et al performed imaging analysis of microcluster formation between the PD-L1 and PD-L2, which are known to play a role in T cell activation in response to tumour cell signaling. Their analysis showed that the cluster formation inhibited T cell receptor signaling and could serve as a visual index for PD-L1/2-targeted cancer therapies.

T cell activation is generally modulated by two distinct signals, the first is antigen-specific signaling from the T cell receptor (TCR) and the second is antigen-independent signaling from various costimulatory and coinhibitory receptors. Programmed cell death 1 (PD-1, also known as CD279) is currently the most notable coinhibitory receptor to suppress T cell activation 1 and finely tunes various types of immune responses, such as immunotolerance, antiviral immunity, tumor immunity, and antibody maturation 2 . PD-1 is a member of the CD28 superfamily, which also includes CTLA-4 3 , and inducibly expressed on T and B cells, natural killer cells, monocytes and dendritic cells 4 upon antigen or cytokine stimulation 5 . PD-1 contains two tyrosine motifs in its cytoplasmic tail, an immunoreceptor tyrosine-based inhibitory motif (ITIM) and an immunoreceptor tyrosine-based switch motif (ITSM) 6 , that are phosphorylated possibly by a Src kinase, Lck, and/or C-terminal Src kinase upon the engagement of PD-1 in T cells 7 . Phosphorylation of both the ITIM and ITSM is required for the recruitment of Src homology 2 domain-containing tyrosine phosphatase 2 (SHP2), which contributes to the reduction of the overall phosphorylation status of the TCR/CD3-CD3ζ-associated protein of 70 kD (Zap70) axis and the CD28-protein kinase C-θ axis 6 . PD-1 possesses two ligands, programmed death-ligand 1 (PD-L1, B7-H1, or CD274) 8 and PD-L2 (B7-DC or CD273) 9 , which belong to B7 family members. While PD-L1 is ubiquitously expressed throughout an entire body, notably by both immune cells and cancer cells 10 , PD-L2 expression is relatively restricted to professional antigen-presenting cells (APCs) and increased when they are activated via innate receptor signaling 10 .
Another PD-1 ligand, PD-L2 was recently reported to be expressed by cancer-associated fibroblasts 11 and various types of tumor cells 12 . The Cancer Genome Atlas and the results from the analyses of clinical tumor samples demonstrated that the expression of PD-L2 is sometimes more highly correlated with antitumor immune responses than PD-L1 in case of renal cell cancer and lung squamous cell cancer 13 . Practically, immunohistochemical staining of tumor tissues for PD-L2 is reported as a useful method to predict the efficacy of anti-PD-1 treatment in some kinds of malignancies 12 . Surface plasmon resonance analysis revealed that the affinity of PD-L2 for PD-1 is two-fold to six-fold higher than PD-L1 14 . The crystal structure analysis for human PD-1-PD-L2 complex suggested a possibility that this high-affinity binding can be the attractive target for the drug development with small compounds 15 . PD-L2 may possibly exert some influence over the developing tumor microenvironment by a different mechanism than PD-1-PD-L1 binding, therefore PD-L2 might be a promising newer target for antitumor immunotherapy. However, few studies have reported the biological and biochemical functions of PD-L2 for PD-1-mediated inhibition of immune cells and signaling. In particular, the differences between the two PD-1 ligands have remained elusive yet.
When a T cell recognizes its cognate antigen peptide in a major histocompatibility complex (pMHC) molecule expressed on an APC, an immunological synapse is formed at the adherent interface between the two cells 16 . By using the combined imaging system of high-resolution microscopy and antigen-presenting supported lipid bilayers (SLBs), we and other groups found that a few hundred clusters of TCRs are imaged at the immunological synapse [17][18][19] . These clusters consist of tens of TCRs and their proximal signaling molecules, many of which are tyrosine-phosphorylated, and they have been identified as a minimal unit for T cell activation and named TCR microclusters 19 . Similar to the TCR, tens of PD-1 molecules cluster together after binding their ligand, PD-L1, and form PD-1 microclusters at the same region of TCRs, accompanied by recruitment of the phosphatase SHP2. These PD-1-SHP2 microclusters contribute to the dephosphorylation of the TCR/CD3 complex and CD28, as well as their downstream signaling molecules 20 .
In this study, we focused on the functional differences between PD-L1 and PD-L2 and examined PD-1-mediated T cell suppression from the viewpoint of PD-1 signalosome formation by using an advanced live cell imaging technique. We newly established SLBs reconstituted with both mouse (m) PD-L1 and mPD-L2 and confirmed the clustering of PD-1 triggered by PD-1-PD-L2 binding, which was blocked by several antibodies against PD-1, PD-L1, or PD-L2 in the specific manner to the antibody-antigen reactions. Similarly, the PD-1-PD-L2 binding transiently recruited SHP2, which induced dephosphorylation of TCR/CD3 complexes and their downstream signaling molecules at the TCR microclusters to suppress T cell responses. Our imaging system revealed the competition between PD-L1 and PD-L2 in binding to PD-1. When the density of PD-1 was less than its ligands, PD-L2 excluded PD-L1 from TCR-PD-1 microclusters, due to the higher affinity of PD-L2 for PD-1, while PD-L2 and PD-L1 colocalized together at the microclusters when there was a low density of PD-L2. Our study highlighted the significance of PD-L2 in the formation of the PD-1-mediated inhibitory signalosome, and also in the induction of immune suppression. These results could lead to an expanded panel of immune checkpoint inhibitors (ICIs) for a variety of clinical settings.

Results
PD-L2 induces PD-1 clustering at TCR microclusters in a manner similar to PD-L1. PD-1 was previously reported to accumulate at an immunological synapse and also at TCR microclusters upon a T cell-APC conjugation 20 . To analyze the localization of PD-1 and its ligands more precisely, we utilized the SLB system, which contains glycosylphosphatidylinositolanchored I-E k (I-E k -GPI) and mICAM-1 as its basic components, to which mPD-1 ligands, mPD-L1 and/or mPD-L2, can be added. The molecular densities of mPD-L1-GPI or mPD-L2-GPI were calculated by flow cytometry, so that they could be adjusted to the same level expressed on a primary antigen presenting cell ( Supplementary Fig. 1). Splenic CD4 + T cells were prepared from AND-TCR [specific for moth cytochrome c 88-103 (MCC 88-103 ) on I-E k ] transgenic (Tg) Rag2-deficient (Rag2 −/− ) PD-1-deficient (Pdcd1 −/− ) mice, stimulated and then retrovirally transduced by enhanced green fluorescent protein-tagged mPD-1 (mPD-1-EGFP). The cells were allowed to settle on the SLB and imaged by confocal microscopy or total internal reflection fluorescence microcopy. On an SLB, PD-1 formed clusters at the nascent T cell-bilayer contact regions in the presence of PD-L1 or PD-L2 and the PD-1 microclusters migrated toward the center of an immunological synapse, forming a central-supramolecular activation clusters (c-SMAC) (Fig. 1a, b, and Supplementary Movie 1). To examine the colocalization of PD-1 with TCRs at microclusters, these AND-Tg CD4 + T cells expressing mPD-1-EGFP were prestained with DyLight 650-labeled anti-TCRβ (H57) Fab and imaged on the same SLB as in Fig. 1a. In the presence of PD-L1 or PD-L2, PD-1 initially accumulated at the same clusters with TCRs (PD-1-TCR microclusters) (Fig. 1c, left panel) and later these PD-1 clusters eventually translocated into a c-SMAC (Fig. 1c, right panel and Fig. 1e, left). To compare the behaviors of PD-1 in cytotoxic T cells with those in helper ones, we imaged the splenic CD8 + T cells from OT-I-TCR [specific for Ovalbumin 257-264 (OVA 257-264 ) on H2-K b ] Tg Rag2 −/− mice on the same SLB except for H-2K b -GPI, and confirmed no difference in the clustering and kinetics of PD-1 between the two T cell subsets (Fig. 1d, e, right).
To examine the localization of PD-1 plus PD-L1 or PD-L2 at a T cell-bilayer interface, we conjugated a T cell expressing mPD-1-EGFP to an I-E k -positive APC cell line, DC-1, introduced with mPD-L1 and/or mPD-L2 tagged by HaloTag. DC-1 cells were sorted based on the levels of staining for HaloTag to set PD-L1 and PD-L2 to all same density ( Supplementary Fig. 2a-c). The accumulation of PD-1 could be imaged at an T cell-APC interface even if only PD-L2 was expressed (Fig. 1f). From these data, we confirmed that not only PD-L1 but also PD-L2 induces both PD-1 microcluster formation, and its translocation to the c-SMAC associating with TCRs.
PD-1-PD-L2 binding induces PD-1 microcluster formation colocalizing with both TCRs and CD28. As shown in Fig. 1 with primary T cells, we confirmed PD-1 microcluster formation by using a T cell hybridoma expressing AND TCRαβ-chains (2D12) further introduced by mPD-1-EGFP. We observed PD-1 clusters at the T cell-bilayer interface when mPD-L1-GPI or mPD-L2-GPI was reconstituted into the SLB ( Supplementary Fig. 3a), and almost all of the T cells developed PD-1 clustering colocalized with TCRs ( Supplementary Fig. 3b). Hui et al. reported that CD28 could be a primary target of PD-1-mediated inhibitory signaling 21 . CD28 possesses two ligands, CD80 and CD86, and augments both phosphatidylinositol-3 kinase-Akt and nuclear factor-kappa B signaling collaborating with TCRs after CD28-CD80/CD86 binding 22 . We showed the colocalization of PD-1 with both TCR and CD28 at microclusters by using the AND-TCR T cell hybridoma expressing mPD-1-HaloTag and mCD28-EGFP ( Supplementary Fig. 3c). Although most of the PD-1, TCR, and CD28 microclusters were merged together, statistical analysis demonstrated that, in the presence of mPD-L1-GPI, there was a lower correlation efficiency between PD-1 and TCRs than those between PD-1 and CD28 or CD28 and TCRs ( Supplementary Fig. 3d, left). In contrast, the correlation efficiency was almost identical between PD-1 and TCR, PD-1 and CD28, or CD28 and TCR in the presence of PD-L2-GPI (Supplementary Fig. 3d, right). These data indicated that PD-1 forms microclusters colocalized with both TCR and CD28 in the presence of CD80 and PD-L1 or PD-L2 with a slightly less tendency of PD-1-TCR co-clustering.
Antibodies that block PD-1-PD-L1/2 binding interfere with PD-1 microcluster formation. Anti-PD-1 and also anti-PD-L1 antibodies can have tremendous clinical benefits in treatment of human malignancies. To examine the effects of anti-mPD-1, anti-mPD-L1 and anti-mPD-L2 blocking antibodies on mPD-1 microcluster formation, we imaged the clustering of PD-1-EGFP on the AND-TCR T cell hybridoma settled onto the SLB in the presence or absence of antibodies against mPD-1 or mPD-L1 and/or mPD-L2. In the presence of mPD-L1-GPI on the SLB, PD-1 microclusters were disrupted by adding anti-mPD-1, anti-mPD-L1 or anti-mPD-L1 + anti-mPD-L2, but not by anti-mPD-L2 (Fig. 2a, left). If instead mPD-L2-GPI was on the SLB, anti-mPD-1, anti-mPD-L2 or anti-mPD-L1 + anti-mPD-L2, but not anti-mPD-L1, disrupted the PD-1 microclusters (Fig. 2a, right). The percentage of T cells forming PD-1 microclusters paralleled the blocking ability of each antibody (Fig. 2b). Various anti-mPD-1 clones have been already examined for their blocking efficiencies both in vitro and in vivo, therefore we verified available three typical clones of anti-mPD-1, J43 23 , RMP1-14 24 , and 29F.1A12 25 , in our experimental settings. All clones worked properly as a staining antibody but they possessed their own optimal concentrations (Supplementary Fig. 4). We next compared the blocking efficiencies among these clones against the clustering of PD-1 in imaging. While J43 and RMP1-14 could not disrupted the clustering of PD-1 at a concentration of 10 µg/ml, they could at the higher concentration, 50 µg/ml. On the other hand, 29F.1A12 at a concentration of 10 µg/ml was enough to interfere the clustering of PD-1 (Fig. 2c). Collectively, these data showed that antibodies known as a blocking one can generally inhibit the clustering of PD-1 depending on their optimal concentrations.
The phosphatase SHP2, but not SHP1, transiently associates with PD-1 microclusters triggered by PD-1-PD-L2 binding. We next examined the association of the phosphatases, SHP1 and SHP2, to PD-1 microclusters generated by the PD-1-PD-L2 binding. By using AND-TCR T cell hybridomas co-expressing mPD-1-HaloTag and EGFP-mSHP1 or EGFP-mSHP2, we confirmed that SHP2, but not SHP1, colocalized at the PD-1 microclusters formed in the presence of either PD-L1 or PD-L2 at the very early phase after T cell-bilayer contact (20 s, Fig. 3a), and that SHP2 clustering could not detected at later time points. The translocation of SHP2 into the immunological synapse was further imaged at the interface between a T cell and an APC expressing PD-L1 or PD-L2 (Fig. 3b).
SHP2 is a responsible molecule in the PD-1-mediated T cell suppression and it could be a promising target to enhance the clinical effectiveness in ICI therapy. Meanwhile, SHP2 inhibitor was recently reported to suppress the in vivo growth of some kinds of cancers bearing ras mutation 26,27 . We therefore examined the formation of PD-1-mediated coinhibitory microclusters in the presence of a SHP2 inhibitor, RMC4550. However, we could find any changes in neither clustering of PD-1 nor translocation of SHP2 ( Supplementary Fig. 5), suggesting that physical association between PD-1 and SHP2 forming microclusters might not rely on the catalytic function of SHP2.
The phosphorylation of CD3ζ and SLP-76 in primary CD4 + T cells were also reduced at the TCR microclusters upon PD-1-PD-L1 or PD-1-PD-L2 binding (Fig. 4c, d). Phosphorylation of the molecules in the further downstream of TCR/CD3 complex, PLCγ1, Akt, and Erk, was also attenuated ( Fig. 4e and Supplementary Fig. 6). We further confirmed the reduction of IL-2 production by mPD-1-expressing T cell hybridomas stimulated by DC-1 cells expressing, or not expressing mPD-L1 or mPD-L2 at any concentration of antigen peptides (Fig. 4f). Using these T cells, DC-1 cells and blocking antibodies as in Fig. 2a, we showed that each antibody restored IL-2 production by T cells, except for anti-mPD-L2 in the case of DC-1 cells expressing mPD-L1 and anti-mPD-L1 with DC-1 cells expressing mPD-L2 (Fig. 4g). Primary OT-I Tg CD8 + T cells transduced with mPD-1 ( Supplementary Fig. 7a) were further tested for the reduction of IFNγ production after stimulation with H-2K b -expressing tumor cell line, EL-4 cells, with or without mPD-L1 or mPD-L2 expression ( Supplementary Fig. 7b). Both cytotoxicity and IFNγ production by OT-I Tg CD8 + T cells were suppressed if these cells were stimulated by EL-4 cells expressing mPD-L1 or mPD-L2 and the suppression was reversed by addition of anti-mPD-1 (29F.1A12) (Fig. 4h, i). Related to the results in Fig. 2c evaluating the function of different anti-mPD-1 antibodies in clustering of PD-1, functional difference in each clone in vitro was evaluated from the viewpoint of recovery of IL-2 production. 29F.1A12 fully restored IL-2 production by T cells, whereas J43 and RMP1-14 could partially restored it (Fig. 4j). From these results, PD-L2 suppresses T cell function through dephosphorylation of TCR/ CD3 and its downstream molecules in a similar fashion as does another PD-1 ligand, PD-L1.
PD-L2 takes an advantage of binding to PD-1 against PD-L1. A recent clinical report described the dual-expression of PD-L1 and PD-L2 by some of malignant tumors 12 . We analyzed the endogenous expression of human (h) PD-L1 and hPD-L2 by different human lung cancer cell lines before or after the culture with IFNγ, and found different expression patterns of PD-1 ligands in each lung cancer cell line (Supplementary Fig. 8). To mimic this expression pattern of PD-l ligand, we reconstituted both mPD-L1-GPI and mPD-L2-GPI into an SLB, and confirmed the predominance of PD-L2 in binding to PD-1 with various densities of PD-L1 and PD-L2. As expected, PD-1 microclusters were still observed even at a low density of PD-L2, but not PD-L1 ( Supplementary Fig. 9a, b).
We next examined the competition between PD-L1 and PD-L2 in the binding to PD-1. To analyze the distribution of PD-L1 and PD-L2 on the same SLB, the ligands were labeled with different fluorescent dyes (PD-L1, AlexaFluor 647; PD-L2, AlexaFluor 488) and HaloTag-tagged PD-1 was stained by TMR-labeled HaloTag ligand after retroviral introduction into T cells. PD-L2 at a high density (more than 150 molecules/μm 2 in Fig. 5a, b, left, c) interfered with the translocation of PD-L1 within PD-1 microclusters, while PD-L2 still remained within PD-1 microclusters if PD-L1 was reconstituted even at a high density (300 molecules/ μm 2 in Fig. 5a, b, right, c).
We next attempted to evaluate T cell responses after stimulation by APCs with the same densities of PD-L1 and PD-L2 as those on the SLBs in Fig. 5a. We coated 5 μm silica beads with the same lipid-bilayers as in Fig. 5a to mimic live APCs expressing these ligands. IL-2 production from T cells was suppressed, when the cells were stimulated by these silica beads with a fixed low-density mPD-L1-GPI (75 molecules/μm 2 ) and mPD-L2-GPI with from high-density (300 molecules/μm 2 ) to low-density. However, it was completely suppressed when the cells were stimulated by the beads with a fixed low-density mPD-L2-GPI (75 molecules/μm 2 ) and mPD-L1-GPI from high to low (Fig. 5d).
We further tested the antigen-specific blocking and T cell suppression by immune checkpoint antibodies, anti-mPD-1, anti-mPD-L1, and/or anti-mPD-L2, in imaging. When a PD-1expressing T cell settled on an SLB reconstituted with both mPD-L1-GPI and mPD-L2-GPI, PD-1 clustering was collapsed by adding anti-mPD-1 or anti-mPD-L1 together with anti-mPD-L2, but not by anti-mPD-L1 or anti-mPD-L2 alone (Fig. 5e). When these blocking antibodies were added into the conjugation between the T cells expressing PD-1 in Fig. 5e Fig. 10). Similarly, anti-mPD-1 or anti-mPD-L1 plus anti-mPD-L2 could restore, at least partially, the IL-2 production, whereas anti-mPD-L1 or anti-mPD-L2 alone could not (Fig. 5f). From these data, PD-L2 was shown to have relatively higher affinity than PD-L1 for PD-1 and was more efficient capacity for PD-1mediated T cell inhibition.

Discussion
Here, we have provided a comprehensive analysis of the inhibitory capacity of the less well-studied PD-1 ligand, PD-L2, as well as side-by-side analysis of activity of PD-L1 and PD-L2. PD-1-PD-L2 binding triggered the clustering of PD-1 with TCRs together forming a TCR-PD-1-PD-L2 signalosome in a fashion similar to PD-1-PD-L1 binding. Our studies, particularly the competition assays between PD-L1 and PD-L2 in the binding to PD-1, have resolved several issues of interest, scientifically and also clinically. First, upon binding to PD-L2, PD-1 forms coinhibitory microclusters colocalized with TCRs and transiently recruits the phosphatase SHP2. Second, we evaluated the blocking effects of several monoclonal antibodies by visualizing the collapse of PD-1 microclusters and confirmed the epitope-specific interference of each antibody. Third, PD-L2 occupies the space where PD-1 forms microclusters, preventing PD-L1 entry, when PD-1 was present at a relatively lower number than its ligands.
A TCR microcluster, the minimal unit for T cell activation, contains TCRs and their proximal signaling molecules such as Zap70, SLP-76, Vav1, and PLCγ1. We found that like PD-L1, PD-L2 induces clustering of PD-1 colocalized at a TCR microcluster in both CD4 + and CD8 + T cells. Immediate but transient recruitment of SHP2 to PD-1 microclusters was also the same when PD-1 was crosslinked by PD-L2. Since SHP2 translocates at a TCR-PD-1 signalosome, the substrates of PD-1-mediated dephosphorylation could be the downstream signaling molecules of TCR and also CD28. Based on the kinetics of the recruitment of SHP2 and the components of the signaling molecules within TCR-PD-1-SHP2 microclusters, PD-L1 and PD-L2 may convey similar inhibitory signals.
Surface plasmon resonance analysis has shown that PD-L2 can bind 2-6 times more potently to PD-1 than PD-L1 14 . Affinity between two molecules is generally measured as the interaction between an analyte and an immobilized ligand in surface plasmon resonance analysis, therefore it is not being evaluated as in a physiological situation, particularly in the case of receptor-ligand pairs both expressed on a cell surface. Our imaging analysis using SLBs thus has advantages for evaluating the affinity between two molecules on a cell surface correlating with their clustering. Each ligand on an SLB is movable and easily controlled in a density in a vast excess of lipids, therefore one can analyze the behavior of a single molecule in a spatiotemporal fashion. When clustering of PD-1 was imaged on an SLB reconstituted by PD-L1 or PD-L2 in a different density from high to low, PD-1 was tightly accumulated forming microclusters even in the presence of PD-L2 at a low density, indicative of the higher affinity of PD-L2 for PD-1. To make the density of PD-L1 or PD-L2 on SLBs close to the actual density on tumor cells, we adjusted their numbers and performed several near physiological, biological and biochemical assays. PD-1 signaling triggered by either PD-L1 or PD-L2 inhibits to almost the same extent the amount of IL-2 and IFNγ cytokine produced, and the cytotoxic function of CD8 + T cells. This suggests that final T cell output is regulated by the PD-1 expression level, not by ligand expression level or its affinity for the receptor. The biological response of T cells triggered by two PD-1 ligands might be less affected, if the T cells expressed a smaller amount of PD-1 than those of PD-1 ligands on APCs.
On the other hand, if PD-1 was more highly expressed on T cells compared to its ligands on APCs, the difference in affinity between PD-L1 and PD-L2 for PD-1 could totally contribute to PD-1-mediated inhibition of T cells. It might mean a delicate balance of the costimulatory and coinhibitory receptor network, which is harmoniously regulated by the affinity and also the spatiotemporal expression pattern of each receptor and ligand.
TCRs and costimulatory and coinhibitory receptors cooperatively manage T cell reactivity and various downstream signaling are considered to be the substrates primarily targeted by PD-1.
Our imaging system with SLBs precisely elucidated the Fig. 4 PD-1-PD-L1 or PD-1-PD-L2 binding attenuates the phosphorylation of both TCR/CD3 complex and its downstream signaling molecules. a 2D12 transduced with mPD-1-HaloTag were preincubated with the HaloTag ligand-TMR (cyan) and DyLight 488-labeled H57 Fab (green), plated onto the SLB as in Fig. 1a without (top) or with mPD-L1-GPI (150/μm 2 , middle) or mPD-L2-GPI (150/μm 2 , bottom), fixed at 2 min after contact, stained with Alexa Fluor 647-labeled anti-phospho (p) CD3ζ (red) and imaged by confocal microscopy. Histograms show fold fluorescent intensities of TCRβ (green), PD-1 (cyan) and pCD3ζ (red) on the diagonal yellow lines in the DIC images. b The graph shows pCD3ζ/TCRβ fluorescent intensity ratio at the T cell-bilayer interface in a in the absence (left) or presence of mPD-L1-GPI (middle) or mPD-L2-GPI (right) (n = 20). Horizontal bars, average. c Primary CD4 + T cells in Fig. 1c transduced with mPD-1-GFP (green) were prestained with the DyLight 549-labeled H57 Fab (cyan), plated onto the SLB, fixed, stained for pCD3ζ (red, left panel) or pSLP-76 (red, right panel) and imaged as in a. Histograms show fold fluorescent intensities of TCRβ (cyan), PD-1 (green) and pCD3ζ or pSLP-76 (red) on the diagonal yellow lines in the middle column.  colocalization of PD-1 with both TCR and CD28 at microclusters. By statistical analyses, the correlation coefficient between PD-1 and TCR clustering was lower than that of PD-1-CD28, when PD-1 was crosslinked by PD-L1. In contrast, that of PD-1-TCR was almost same as that of PD-1-CD28 on the crosslinking of PD-1 by PD-L2. Recently, two papers reported that a cis-interaction of CD80 to PD-L1 inhibits PD-1-PD-L1 binding but still preserves the CD80-CD28 binding 28,29 . Such differences in the correlation coefficient among PD-1, TCR, and CD28 clustering crosslinked by PD-L2 could be partly due to the cis-interaction between PD-L1 and CD80. One of the mechanisms for antitumor effects of ICI treatment is the blocking of the inhibitory signaling via PD-1 resulting from PD-L1-or PD-L2-crosslinking. While anti-hPD-1 can block both PD-1-PD-L1 and PD-1-PD-L2 binding, the PD-1-PD-L2 binding still occurred even in the presence of anti-PD-L1. It is a critical point for the potential clinical efficacy of an ICI to evaluate whether the monoclonal antibody possesses a blocking effect, which results in interference between a receptor-ligand pair, and not a simple affinity against a receptor. This difference results from the location of the cognate epitope within an antigen. Anti-mPD-1 monoclonal antibodies, J43, RMP1-14, and 29F.1A12, can equally detect the surface expression of mPD-1 and they differently blocked the PD-1-PD-L1/L2 binding in our in vitro experiments (Fig. 2). Their function was almost consistent with the results of the imaging analysis for PD-1 clustering that J43 and RMP1-14 showed lower recovery ratio even with higher concentration of these antibodies compared with that of 29F.1A12. Although it is true that each clone functions differently in vitro and/or in vivo due to its isoform type or a real concentration in the microenvironment, our imaging technique can be effective to evaluate the actual function of the clones and to screen the most suitable one for ICIs. Anti-mPD-L2, which we examined in our in vitro and ex vivo experiments, fortunately restored both cytokine production and cytotoxicity against tumor cells expressing PD-L2. PD-1 is translocated into TCR microclusters together forming clusters upon crosslinking by PD-1 ligands and pMHC, and an antibody that blocks PD-1-PD-L1/2 binding collapsed PD-1 clustering but still maintained the TCR microclusters, which produce activation and restoration for T cell signaling. We found a tight correlation between the collapse of PD-1 microclusters and recovery of T cell functions, indicating the benefit of our imaging technique as a powerful screening tool for blocking antibodies likely to recover T cell functions more effectively.
CTLA-4 possesses two ligands, CD80 (also known as B7-1) and CD86 (also known as B7-2), whose expression pattern and affinity are different 30 . In a similar fashion as CTLA-4, while PD-L1 is known to be expressed on somatic cells constitutively and ubiquitously, PD-L2 is generally an inducible ligand. The purpose of having two ligands can be surmised as follows: PD-L1 is the main player maintaining peripheral tolerance by engaging PD-1 and that PD-L2 compensates for PD-L1 function, when the environment requires stronger inhibitory signaling through PD-1 due to, for example, excessive inflammation. Although PD-L1 is considered to be more predominantly expressed in various tumors, we are unfortunately aware of a biased usage of PD-L1 staining as a biomarker for a prospective ICI response. Recent papers have noted a much higher frequency of PD-L2-positive tumor cells than we expected. The expression of each PD-1 ligand on tumors differs with each organ and malignancy 31 . When we examined the expression of PD-L1 or PD-L2 on several human lung cancer cell lines before or after stimulation with IFNγ, those expression levels were shown to vary among the cell lines. A quite lower expression of PD-L2 was enough to form PD-1 microclusters and to introduce inhibitory signaling through PD-1, so anti-PD-1 should be prioritized. Anti-PD-1 is sometimes effective in patients with non-small cell lung cancer, regardless of low PD-L1 expression, in which case it is a strong possibility that PD-L2 was highly expressed in the tumor mass.
In this paper, we have examined in detail the behavior of another PD-1 ligand, PD-L2, and elucidated similarities to the better-studied PD-1 ligand, PD-L1, its unique kinetics and their competition in binding to PD-1. Our results provide basic knowledge for a useful proposal of a new biomarker to indicate tumor immunotherapies. Anti-PD-L2 treatment is expected to be used in the future for some types of tumors.
Methods 2 h. The phosphorylation of CD3ζ and SLP-76 were detected by fluorescent-labeled anti-pCD3ζ and pSLP-76, respectively 20 . A confocal laser scanning microscope (TCS SP8, Leica Microsystems) comprising a 63× oil-immersion objective lens, high sensitivity HyD detectors and 488, 561, and 633 nm laser lines was used. LAS X software (Leica, Germany) was used for image acquisition. A TIRF analysis system was set up on a conventional inverted microscope (Ti-LAPP, Nikon, Tokyo, Japan) equipped with a TIRF objective lens (Nikon), a scientific CMOS camera (ORCA flash 4.0, Hamamatsu photonics) and fiber-coupled 488 nm lasers. The exposure time was set 100 ms with 2.5 s-interval between time points. NISelements software (Nikon) was used for image acquisition and ImageJ software (NIH, Bethesda, MD, USA, RRID:SCR_003070) was used for image processing and final figure preparation.

Data availability
All data supporting the conclusions included in the manuscript are available within the paper and its supplementary information. Source data for figures and graphs in the main text can be found in Supplementary Data. Full immunoprecipitation and western blot images are included in Supplementary Figs. 11 and 12.