In this study, we demonstrate that primary cultures of HBECs express robust basal levels of PD-L2 and increased levels of PD-L1 and PD-L2 in response to pro-inflammatory cytokines. Such PD-1 ligand expression contributes to the capacity of HBECs to reduce the migration and activation of human T cells. Our analysis of post-mortem human brain tissues underlines that PD-L2 is expressed by all brain endothelial cells under normal physiological conditions but that a significant proportion of these cells do not express PD-L2 in MS brain lesions. Finally, PD-L1 although easily observed on other CNS cell types in MS brain lesions is not detected on brain endothelial cells.
PD-L1 and PD-L2 expression by endothelial cells from various origins, but not CNS, has been previously shown. Using primary cultures of HBECs, we observed that under physiological conditions PD-L1 was not detected as assessed by flow cytometry and qPCR. On the other hand, PD-L2 was already highly expressed at basal level (Figure 1). Upon inflammation, both ligands were up-regulated, reaching around 100% of cells positive for these ligands (Figure 1). Previous studies have shown similar observations using human umbilical vein endothelial cells (HUVECs) by qPCR and flow cytometry: HUVECs did not express PD-L1 [15, 22, 23], but did bear considerable PD-L2 levels under basal conditions . Moreover, IFN-γ stimulation increased both PD-L1 and PD-L2 expression while the combination IFN-γ+TNF was even more potent . Also using flow cytometry, human cornea endothelial cells increased PD-L1 and PD-L2 levels following IFN-γ stimulation . In contrast, mouse heart endothelial cells upregulated PD-L1 levels in inflammatory conditions but did not express detectable levels of PD-L2 under basal or activated conditions as assessed by flow cytometry and microscopy [15, 21], suggesting distinct regulation of PD-L1 and PD-L2 by endothelial cells in different species and in different organs.
Massive infiltration of immune cells into the CNS is one of the first steps leading to the formation of new MS lesions and mechanisms controlling such infiltration have not been completely elucidated. Blocking PD-L1 and PD-L2 in EAE, the mouse model of MS, leads to earlier onset and increased severity of the disease, mainly due to elevated number of infiltrating immune cells, especially CD8 T cells [25, 26]. In our study, we demonstrated that blocking PD-L1 and PD-L2 on HBECs leads to elevated number of CD8 and CD4 T cells migrating through an in vitro BBB model (Figure 2), supporting a contributing role for these ligands expressed by the local endothelium in regulating immune cell infiltration into the CNS. In contrast, our group has recently shown that MHC class I blockade does not modify the migration of human CD8 T cells across BBB-endothelial cells . These observations also demonstrate that although CD8 T cells and HBECs were obtained from different human donors, the allo-reactivity did not play a role in CD8 T cell migration in our in vitro BBB model. Furthermore, it has been previously demonstrated that the ligation of PD-1 blocks the β1 and β2 integrin-mediated adhesion by human T cells induced with anti-CD3 . Therefore, based on these published data and our own novel data, we suggest that the binding of PD-1 on T cells by PD-L1/2 on HBECs prevents these T cells from crossing the endothelium potentially via a mechanism implicating integrins. CD8 T cells were shown to be particularly affected by a general PD-L1 and PD-L2 blockade in the EAE model [26, 27]. We can speculate that PD-1 ligand expression by the CNS-endothelium may play a role in regulating the migration of other activated immune cells expressing the cognate receptor, as PD-1 is expressed not only on activated T cells but also on B cells and monocytes .
Endothelial cells from different organs have been shown to display the capacity to modulate T cell responses via the expression of PD-L1 and/or PD-L2 [15, 21–23]. In our studies, blocking PD-L1 and PD-L2 on inflamed HBECs did not affect the proliferation of CD8 T cells. However, it had an impact on the production of IFN-γ and granzyme B (Figure 3). Rodig and colleagues have similarly demonstrated that blocking PD-L1 and/or PD-L2 on HUVECs increased the production of IFN-γ, but did not influence proliferation and IL-2 production by CD8 T cells . This group also reported that blocking PD-L1 on mouse heart endothelial cells increase the killing capacity of CD8 T cells. We believe that the effects seen in our in vitro assays were mainly due to the blocking of PD-L1 and PD-L2 on inflamed HBECs but we cannot rule out that the anti-PD-L1 antibody could bind to PD-L1-expressing activated CD8 T cells. However, as we could not detect PD-L1 on ex vivo CD8 T cells and only low levels of PD-L1 on a small fraction of CD8 T cells (8-20%) after anti-CD3+anti-CD28 activation (data not shown), this would be a less important contribution.
Distinct PD-L1 and PD-L2 expression has been reported in different human organs. Several groups demonstrated that PD-L1 and/or PD-L2 are detected in immuno-privileged organs under physiological conditions. PD-L1 is elevated in human placenta, while PD-L2 is highly expressed on the endothelium of placenta blood vessels . Although PD-L1 is constitutively expressed in testis, another immuno-privileged organ, no PD-L2 is observed . PD-L1 is also constitutively expressed at high levels by corneal epithelial cells. However, these cells bear significantly reduced PD-L1 levels during dry eye disease, a T-cell mediated inflammation , paralleling our observations for PD-L2 on human CNS endothelium in controls vs. MS. Using an endothelial cell specific marker (caveolin-1), we easily detected PD-L2 expression by all blood vessels (caveolin-1+) in post-mortem CNS tissues obtained from normal controls, but only on about 50% of blood vessels in MS lesions (Figure 5). We observed non-endothelial cells around blood vessels expressing PD-L2 in MS lesions. According to the shape and the localization of these cells, we hypothesize that these are infiltrating immune cells. Experiments performed in EAE documented PD-L1 and PD-L2 detection on a fraction of infiltrating immune cells such as macrophages, dendritic cells and microglia [26, 40]. We could not detect PD-L1 on CNS brain endothelium although this ligand was easily observed on other CNS cells in MS lesions (Figure 4) and has been observed on malignant gliomas . We have previously shown that PD-L1 is significantly elevated in MS brain lesions especially on astrocytes and microglia/macrophages , while this ligand is barely detectable in normal controls. These observations correlate with our previous in vitro data obtained with primary cultures of glial cells; we detected low levels of PD-L1 on microglia and astrocytes under basal conditions but a significant increase of PD-L1 levels on these cells upon pro-inflammatory stimulation . In contrast to our observations in human CNS, PD-L2 was not detected on CNS cells of control and EAE animals, although PD-L1 was observed on resident brain cells, including the endothelium, in EAE mice [26, 42]. These results support the notion that PD-L1 and PD-L2 expression is differently regulated in human and murine CNS .
Recent work suggests that the expression of PD-1 ligands is regulated by different promoters in distinct cell types . Indeed, whereas murine PD-L2 expression has been shown to be controlled by NF-κB and STAT6, PD-L1 expression is not . Moreover, platinum-based chemotherapeutics have been shown to downregulate PD-L2 expression in human dendritic and tumor cells  via a STAT6-mediated mechanism. Therefore, a more detailed dissection of the mechanisms regulating PD-L1 and PD-L2 expression under physiological and disease conditions is warranted and could result in new therapeutic tools.
Our in vitro data showed that HBECs express low or no PD-L1 but high PD-L2 levels under basal conditions; similarly in normal control brain tissues we did not detect PD-L1 but observed robust PD-L2 expression by the brain endothelium. Although inflammatory cytokines increased PD-L1 and PD-L2 levels in vitro, these ligands were not upregulated in MS lesions compared to controls. In contrast to glial cells, endothelial cells are sitting at the boundary between the periphery and the CNS. We can hypothesize that factors, others than pro-inflammatory cytokines, present in the periphery on the lumen side, or other CNS cells closely interacting with the endothelium, may impact on the in vivo PD-L1 and PD-L2 expression by the CNS endothelium. Finally, we can speculate that under physiological conditions, the elevated PD-L2 basal levels contribute to inhibit the activation and migration of T cells across the BBB, but given the reduced levels of PD-L2 on MS brain endothelium, this function is impaired. CD8 T cells have been reported to be localized more frequently in the parenchyma of MS brain [8, 9, 45]. Furthermore, we observed an important PD-L1 upregulation  in MS lesions in perivascular and parenchymal areas, correlating with the absence of PD-1 on infiltrating CD8 T cells. Therefore, we speculate that the BBB capacity to control cell entry into the CNS is impaired in MS patients, leading to the entry of T cells regardless whether they express PD-1 or not, but that PD-1-negative CD8 T cells will be favored for progressing into the inflamed parenchyma which abundantly expresses PD-L1.