Distinct contribution of PD-L1 suppression by spatial expression of PD-L1 on tumor and non-tumor cells

Abstract

Programmed cell death receptor 1 (PD-1) and its ligand, PD-L1, are important immune checkpoint proteins. Although antibodies that block PD-1/PD-L1 have shown promising clinical efficacy in a subset of cancer patients, the detailed cellular and molecular mechanisms behind anti-PD-1 and anti-PD-L1 immunotherapy are not well defined. Specifically, the way in which PD-L1 contributes to immune suppression on tumor and non-tumor cells remains controversial. By selectively blocking PD-L1 on either tumor or non-tumor cells, we demonstrated that PD-L1 from both sources suppressed the anti-tumor T-cell response. Blocking PD-L1 on either tumor cells or non-tumor cells inhibited tumor growth and enhanced immune cell infiltration, as well as the tumor-specific T-cell response. Further, simultaneously blocking tumor- and non-tumor-derived PD-L1 maximized anti-tumor T-cell responses and demonstrated synergy. In addition, the relative contribution of PD-L1 on tumor and non-tumor cells to immune suppression depended on the PD-L1 expression level. Lastly, we found that the F4/80 receptor was involved in the anti-tumor effect of PD-L1 blockade. Taken together, our data indicate that PD-L1 on both tumor and non-tumor cells is critical for T-cell inhibition, which provides new directions for the optimization of PD-L1-blocking antibodies and the development of clinical biomarker strategies.

Introduction

Tumor cells acquire the characteristic hallmarks of cancer through intrinsic and extrinsic mechanisms.¹ Evasion of the immune system is one such hallmark and this enables cancer cells to escape destruction by immune cells. To accomplish this, cancer cells use a variety of mechanisms, including downregulation of antigen presentation molecules to avoid recognition by T cells² or active upregulation of inhibitory molecules to cause immune cell dysfunction.^3–7 Programmed cell death receptor ligand 1 (PD-L1) is one of these key modulatory molecules. The engagement of PD-L1 with PD-1 transduces an inhibitory signal for T-cell activation. Blockade of this co–inhibitory pathway by either anti-PD1 or anti-PD-L1 antibodies can profoundly enhance the T-cell response, as evidenced by increased effector cytokine production and cytotoxicity.^8,9 According to this simple concept, anti-PD1- and anti-PD-L1-blocking antibodies have achieved promising clinical efficacy in ~ 10–30% of cancer patients.¹⁰ However, the mechanisms that contribute to the efficacy of these blocking antibodies are not fully understood. It has been reported that the efficacy of anti-PD-L1 and anti-PD-1 antibody therapy is correlated with infiltrating T cells, PD-L1 expression, and tumor mutational burden.^9–12 PD-L1 can be expressed on tumor cells and multiple types of non-tumor cells, including macrophages, myeloid-derived suppressor cells (MDSCs), stromal cells, and T cells.¹³ The expression of PD-L1 can be upregulated by cytokines including type I interferons (IFNs), IFN-γ, and tumor necrosis factor through either increased messenger RNA transcription or increased protein stability.^14–16 Initially, tumor cells were considered the dominant source of PD-L1 for T-cell suppression, which was supported by the decreased immunogenicity of PD-L1-overexpressing tumor cells³, and the clinical correlation between PD-L1 expression levels on tumor cells and the efficacy of PD-L1 blockade.^12,17–19 However, recent studies have shown that non-tumor-derived PD-L1 is also correlated with anti-PD-1 antibody efficacy.^12,20,21 These controversial observations suggest that multiple underlying mechanisms may be involved in PD-L1-mediated T-cell suppression. The determination of the contribution of PD-L1 from different cell sources is critical for understanding the anti-tumor mechanism of anti-PD-L1 antibodies and for screening predictive biomarkers for these therapies.

Using novel tumor models, we were able to selectively block tumor- and non-tumor-derived PD-L1 in a naturally developed tumor microenvironment, rather than simply study the absence of PD-L1 on either tumor cells or non-tumor cells. We demonstrated that both tumor- and non-tumor-derived PD-L1 contributed to T-cell inhibition in a non-redundant way and that blocking both sources of PD-L1 achieved synergy and resulted in the maximum anti-tumor effect. Furthermore, we found that F4/80 was critical for anti-PD-L1 antibody-mediated tumor regression. Thus, our findings not only demonstrate the mechanisms involved in the anti-tumor effect of anti-PD-L1 antibodies but also provide new directions for the design of combinational strategies and the optimization of predictive biomarker screening for PD-1/PD-L1-related therapies.

Results

Blocking PD-L1 on non-tumor cells reactivates the anti-tumor T-cell response

Anti-PD-L1 antibodies interfere with the binding of PD-L1 to PD-1, which leads to T-cell activation and tumor control. However, the way in which different sources of PD-L1 (tumor-derived vs. non-tumor-derived) contribute to immune suppression remains unclear. To investigate this, we constructed a B16-OVA melanoma cell line deficient in mouse PD-L1 (mPD-L1^null B16-OVA) using the CRISPR/Cas9 gene-editing strategy (Fig. 1a). The growth of the B16-OVA mPD-L1^null cell line in vitro or in immune-compromised mice is similar to that of parental B16-OVA cells (Supplementary Figures S1 and 2). In this B16-OVA mPD-L1^null cell line, OVA is stably expressed and serves as a tumor-specific model antigen. The peptides OT-I and OT-II, generated from OVA, can be presented by major histocompatibility complex (MHC)-I and MHC-II molecules, respectively, and thus they can activate OT-I- and OT-II-specific T-cell responses. By inoculating wild-type (WT) B6 mice, B16-OVA mPD-L1^null cells, we established a tumor model in which PD-L1 is only expressed on non-tumor cells. Interestingly, we found that the administration of an anti-mouse PD-L1 antibody (clone 10F.9G2, no cross-reactivity with human PD-L1 (hPD-L1)) in this model could effectively control tumor growth (Fig. 1b and Supplementary Figure S3). To determine whether the inhibition of non-tumor cell-derived PD-L1 is sufficient to reactivate the T-cell response, we further examined the activation of tumor-specific CD8⁺ and CD4⁺ T cells. Indeed, we observed elevated IFN-γ production in the anti-PD-L1-treated group after OT-I peptide stimulation (Fig. 1c), which suggests restoration of tumor-specific CD8⁺ T-cell responses. However, in this model, tumor-derived PD-L1 is absent during the entire tumor development period, which may lead to a very different tumor microenvironment than that observed in cancer patients. To better simulate the clinical situation, we inoculated hPD-L1 knock-in (KI) mice with B16-OVA tumor cells. With the use of an anti-hPD-L1 antibody (MEDI-4736), which binds to hPD-L1 without cross-reactivity to mouse PD-L1 (mPD-L1)²², we can selectively block non-tumor-derived PD-L1 (Fig. 1d). Although hPD-L1 is known to bind mouse PD-1, as demonstrated by crystal structure analysis,²³ it is not clear whether hPD-L1 could activate mouse PD1 signaling similar to the way this signaling is activated by mPD-L1. To confirm that hPD-L1 can functionally engage mouse PD-1, we examined the impact of an anti-hPD-L1 antibody on T-cell activation in hPD-L1 KI mice. Similar to anti-mouse PD-L1 antibody treatment, we detected increased IFN-γ production in a dose-dependent manner with anti-hPD-L1 treatment (Fig. 1e), which suggests that hPD-L1 can functionally bind to mouse PD-1 and transduce an inhibitory signal in T cells. In this PD-L1-intact model, the selective blocking of PD-L1 on non-tumor cells using an anti-hPD-L1 antibody resulted in tumor regression and increased anti-tumor CD8⁺ T-cell responses (Fig. 1f,g). Therefore, these data suggest that blocking PD-L1 on non-tumor cells can effectively reverse T-cell inhibition and control tumor growth.

Fig. 1.

Blocking PD-L1 on non-tumor cells reactivates the anti-tumor T-cell response. a PD-L1 expression of B16-OVA and B16-OVA mPD-L1^null cells in the presence or absence of IFN-γ was analyzed by flow cytometry. b WT B6 mice (n = 5/group) were injected subcutaneously with 5 × 10⁵ B16-OVA mPD-L1^null cells, which was followed by the administration of 25 μg of anti-mPD-L1 or control IgG on days 9 and 12. Tumor growth was measured and compared twice a week. The mean + SEM are shown. c Seven days after the last treatment dLN cells were collected, and an IFN-γ CBA was performed after restimulation with OT-I and OT-II peptides. d Splenocytes from WT B6 mice or hPD-L1 KI mice were stained with anti-mPD-L1 or anti-hPD-L1 and analyzed by flow cytometry. e Splenocytes from WT B6 mice or hPD-L1 KI mice were stimulated with anti-CD3 and anti-CD28 and different concentrations of anti-mouse or human PD-L1. IFN-γ production was measured by CBA 2 days later. f hPD-L1 KI mice (n = 5/group) were injected subcutaneously with 5×10⁵ B16-OVA cells, which was followed by the administration of 25 μg of anti-hPD-L1 or control IgG on days 9 and 12. Tumor growth was measured and compared twice a week. The mean + SEM are shown. g Seven days after the last treatment, dLN cells were collected, and an IFN-γ CBA was performed after restimulation with OT-I and OT-II peptides. p < 0.05 compared with the control group. One representative experiment of three is depicted

Blocking PD-L1 on tumor cells effectively restores the T-cell response

The idea that non-tumor-derived PD-L1 is sufficient to inhibit T-cell activation and promote tumor growth diverges from the current understanding that tumor-derived PD-L1 is the major regulator of immune suppression. However, the studies that support this conclusion primarily used PD-L1^–/– mice to exclude the role of non-tumor-derived PD-L1. It is worth noting that in vitro and in vivo CD4⁺ and CD8⁺ T-cell responses are remarkably enhanced in PD-L1^–/– mice compared with WT mice,²⁴ which may not be representative of suppressed T cells in a clinical setting and may potentially exaggerate the treatment effect of PD-L1 inhibition. To determine the contribution of tumor-derived PD-L1 in the inhibition of anti-tumor T-cell responses in a more clinically relevant model, we generated a B16-OVA cell line deficient in mPD-L1 but that expressed hPD-L1 (B16-OVA mPD-L1^null hPD-L1⁺) and inoculated these cells into WT B6 mice. In this model, non-tumor cells express mouse PD-L1 and tumor cells express hPD-L1. The two sources of PD-L1 may be distinguished using anti-mPD-L1 (10F.9G2) and anti-hPD-L1 (MEDI-4736) antibodies (Fig. 2a). We took advantage of this novel mouse model and selectively blocked PD-L1 on tumor cells using an anti-hPD-L1 antibody, whereas the inhibitory signal from PD-L1 on non-tumor cells remained intact. We found that blocking PD-L1 on tumor cells suppressed tumor growth (Fig. 2b and Supplementary Figure S4) and reactivated tumor-specific T-cell responses (Fig. 2c). As hPD-L1 on B16-OVA mPD-L1^null hPD-L1⁺ cells is constitutively expressed under the control of the EF1α promoter, the expression pattern of hPD-L1 on these cells may be different from the inducible expression pattern of endogenous PD-L1. To test whether inducible and constitutively expressed PD-L1 respond differently to anti-PD-L1 antibody treatment, we used the previously established B16-OVA tumor model in hPD-L1 KI mice. In this model, B16-OVA tumor cells express mouse PD-L1 controlled by an endogenous promoter and non-tumor cells express hPD-L1. We selectively blocked tumor-derived mPD-L1 using an anti-mPD-L1 antibody. Similar to our previous observations using the model in which PD-L1 is constitutively expressed, the blockade of inducible PD-L1 on tumor cells could also effectively control tumor growth and induce tumor-specific CD8⁺ T-cell responses (Fig. 2d,e). Taken together, these data indicate that blocking PD-L1 on tumor cells can also effectively reactivate T-cell responses and inhibit tumor growth.

Fig. 2.

Blocking PD-L1 on tumor cells effectively restores the anti-tumor T cell response. a Human and mouse PD-L1 expression on B16-OVA and B16-OVA mPD-L1^null hPD-L1⁺ cells was analyzed by flow cytometry. White, B16-OVA; gray, B16-OVA mPD-L1^null hPD-L1⁺. b WT B6 mice (n = 5/group) were injected subcutaneously with 5 × 10⁵ B16-OVA mPD-L1^null hPD-L1⁺cells, which was followed by the administration of 25 μg of anti-hPD-L1 or control IgG on days 9 and 12. Tumor growth was measured and compared twice a week. The mean + SEM are shown. c Seven days after the last treatment, dLN cells were collected and an IFN-γ CBA was performed after restimulation with OT-I and OT-II peptides. d Human PD-L1 KI mice (n = 5/group) were injected subcutaneously with 5 × 10⁵ B16-OVA cells, which was followed by the administration of 25 μg of anti-mPD-L1 or control IgG on days 9 and 12. Tumor growth was measured and compared twice a week. The mean + SEM are shown. e Seven days after the last treatment, dLN cells were collected and an IFN-γ CBA was performed after restimulation with OT-I and OT-II peptides. *p < 0.05 compared with the control group. The mean + SEM are shown. One representative experiment of three is depicted

PD-L1 on tumor cells and non-tumor cells synergistically suppresses anti-tumor T-cell responses

We have demonstrated that PD-L1 on tumor cells and non-tumor cells contributes to T-cell inhibition. However, whether PD-L1 on tumor and non-tumor cells is able to synergistically inhibit T-cell responses remains unclear. Using both of our novel tumor mouse models (B16-OVA mPD-L1^null hPD-L1⁺cells into WT B6 mice and B16-OVA into hPD-L1 KI mice), we blocked PD-L1 on both tumor and non-tumor cells simultaneously using a combination of anti-hPD-L1 and anti-mPD-L1 antibodies. Consistent with previous observations, we found that blocking either tumor- or non-tumor-derived PD-L1 could partially inhibit tumor growth (Fig. 3a,b and Supplementary Figure S5). Importantly, when we simultaneously blocked both tumor- and non-tumor-derived PD-L1, we observed a dramatic reduction in tumor size compared with the blocking of either tumor- or non-tumor-derived PD-L1 alone (Fig. 3a,b). Blocking the PD-L1-mediated inhibition from both tumor and non-tumor cells resulted in maximum anti-tumor responses, which was indicated by increased infiltration of CD3⁺ cells (Fig. 3c) and increased IFN-γ production after OT-I peptide stimulation (Fig. 3d,e). In particular, the increase in tumor-specific IFN-γ production showed greater than additive effects when PD-L1 on both tumor and non-tumor cells was blocked. These data suggest that PD-L1 on both tumor cells and non-tumor cells functions synergistically to suppress T-cell function and reshape the immune-suppressive tumor microenvironment.

Fig. 3.

PD-L1 on tumor cells and non-tumor cells synergistically suppresses anti-tumor T-cell responses. a WT B6 mice (n = 5/group) were injected subcutaneously with 5 × 10⁵ B16-OVA mPD-L1^null hPD-L1⁺cells, which was followed by the administration of 25 μg of the indicated control IgG, anti-hPD-L1, and anti-mPD-L1 on days 9 and 12. Tumor growth was measured and compared twice a week. The mean + SEM are shown. b Human PD-L1 KI mice (n = 5/group) were injected subcutaneously with 5 × 10⁵ B16-OVA cells, which was followed by the administration of 25 μg of the indicated control IgG, anti-mPD-L1, or anti-hPD-L1 on days 9 and 12. Tumor growth was measured and compared twice a week. The mean + SEM are shown. c Same as in a, 7 days after the last treatment, tumor tissue was collected for anti-CD3 staining Scale bar, 50 μm. d Same as in a, 7 days after the last treatment, dLN cells were collected and an IFN-γ CBA was performed after restimulation with OT-I and OT-II peptides. e Same as in b, 7 days after the last treatment, LN cells were collected and an IFN-γ CBA was performed after restimulation with OT-I and OT-II peptide. The mean + SEM are shown. One representative experiment of three is depicted

Expression level determines T-cell inhibition ability of different sources of PD-L1

Based on our data and previous publications, it appears that in different tumor models or experimental settings, tumor- and non-tumor-derived PD-L1 can contribute to the suppression of T-cell responses and the promotion of tumor growth in three different ways: (1) tumor-derived PD-L1 is dominant¹⁹; (2) non-tumor-derived PD-L1 is dominant²⁰; or (3) both tumor- and non-tumor-derived PD-L1 are important.¹⁸ We hypothesized that the relative expression level of PD-L1 on tumor cells compared with non-tumor cells determines its relative contribution to immune suppression. To test our hypothesis, we constructed a series of tumor cell lines to simulate different PD-L1 expression levels: B16-OVA mPD-L1^null hPD-L1^low, B16-OVA mPD-L1^null hPD-L1^medium, and B16-OVA mPD-L1^null hPD-L1^high. These cell lines mimic a low, medium, and high ratio of PD-L1^tumor/PD-L1^non-tumor expression, respectively (Fig. 4a). Interestingly, we found that when the ratio of PD-L1^tumor/PD-L1^non-tumor changed from low to medium or high, the anti-tumor efficacy of blocking non-tumor-derived PD-L1 changed from strong to weak (Fig. 4b,c,d). When tumor cells expressed medium or high levels of PD-L1 relative to non-tumor PD-L1, the anti-tumor efficacy of blocking tumor-derived PD-L1 was superior than blocking non-tumor-derived PD-L1 (Fig. 4c,d). On the contrary, when tumor cells expressed low levels of PD-L1 relative to non-tumor PD-L1, PD-L1 on non-tumor cells seemed to contribute equally to immune suppression (Fig. 4b). Consequently, our data support the hypothesis that both tumor- and non-tumor-derived PD-L1 are important for the suppression of T-cell responses. Moreover, their relative contribution to immune suppression is determined by the relative expression level of PD-L1 from both sources. From a clinical perspective, our findings indicate the importance of evaluating PD-L1 expression on both tumor and non-tumor cells as a biomarker for patient selection and the prediction of treatment efficacy.

Fig. 4.

Expression level determines T-cell inhibition ability of different sources of PD-L1. a Human and mouse PD-L1 expression in B16-OVA, B16-OVA mPD-L1^null hPD-L1^low, B16-OVA mPD-L1^null hPD-L1^medium, and B16-OVA mPD-L1^null hPD-L1^high cells was analyzed by flow cytometry. White, B16-OVA; light gray, B16-OVA mPD-L1^null hPD-L1^low; dark gray, mPD-L1^null hPD-L1^medium; black, B16-OVA mPD-L1^null hPD-L1^high. b B16-OVA mPD-L1^null hPD-L1^low, c B16-OVA mPD-L1^null hPD-L1^medium, and d B16-OVA mPD-L1^null hPD-L1^high cells were injected subcutaneously into B6 mice, which was followed by the administration of 25 μg of the indicated control IgG, anti-hPD-L1, or anti-mPD-L1 on days 9 and 12. Tumor growth was measured and compared twice a week. The mean + SEM are shown. One representative experiment of three is depicted

F4/80 is required for the anti-tumor effect of anti-PD-L1 antibody

We have shown that PD-L1 derived from non-tumor cells contributes to T-cell suppression. To further identify the cell populations involved, we used depletion or regulating antibodies to screen the essential cell types for anti-PD-L1 antibody efficacy. As expected, we found that CD8⁺ T cells were important for the therapeutic effect of PD-L1 blockade (Fig. 5a), whereas CD4⁺ T cells and B cells were not required for this anti-tumor effect (Fig. 5a,b). To our surprise, we found that anti-F4/80 antibody treatment remarkably decreased the ability of PD-L1 blockade to mediate tumor control (Fig. 5c). We also found that the CD3⁺ cell infiltration induced by anti-PD-L1 antibody treatment was decreased by anti-F4/80 antibody treatment (Fig. 5d). F4/80 is a transmembrane glycoprotein and, although it is known as a marker of macrophages, its function is not fully understood. The anti-F4/80 antibody we used was not a depletion antibody (Supplementary Figure S6), which indicates that the observed phenotype is not due to the absence of F4/80 ⁺ cells but may be due to alterations in the F4/80 signaling pathway.

Fig. 5.

F4/80 is required for the anti-tumor effect of anti-PD-L1 antibody. WT B6 mice (n = 5/group) were injected subcutaneously with 5 × 10⁵ B16-OVA cells, which was followed by the administration of 25 μg of the indicated control IgG or anti-mPD-L1 on days 9 and 12. a Two hundred micrograms of anti-CD4 or anti-CD8 b anti-CD19, or c anti-F4/80 antibodies were administered on days 9 and 16. Tumor growth was measured and compared twice a week. d Same as in c, 7 days after the last treatment, tumor tissue was collected for anti-CD3 staining Scale bar, 50 μm. e Same as in c, 7 days after the last treatment, tumor tissue was digested, and infiltrated immune cells were analyzed by flow cytometry

It has been reported that anti-PD-L1 treatment can change the immune cell composition in the tumor microenvironment. In agreement with previous data, we observed a significant increase in CD4⁺ and CD8⁺ T cells, and a decrease in immune-suppressive MDSCs among tumor-infiltrating leucocytes with anti-PD-L1 antibody treatment. The administration of an anti-F4/80 antibody abrogated these anti-PD-L1 antibody-induced changes (Fig. 5e).

F4/80 signal affects T-cell activation through modulation of macrophage activation

To test whether the impaired tumor control from anti-F4/80 antibody treatment was due to decreased cytotoxic T-lymphocyte responses, lymphocytes from the draining lymph node of B6 mice inoculated with B16-OVA cells were isolated and stimulated with OT-I or OT-II peptides. Indeed, anti-F4/80 antibody treatment decreased tumor-specific CD8⁺ T-cell responses, which was indicated by lower IFN-γ production after OT-I peptide stimulation (Fig. 6a). To explore this mechanism in greater detail, we stimulated naive B6 splenocytes with anti-CD3 and anti-CD28 in the presence or absence of anti-PD-L1 and anti-F4/80 antibodies in vitro. Consistent with the in vivo data, anti-F4/80 antibody treatment decreased anti-PD-L1 antibody-mediated IFN-γ elevation (Fig. 6b). Initially, we questioned whether anti-F4/80 could affect the viability of macrophages in vitro. By Annexin V staining, we did not observe any difference in apoptosis of bone marrow-derived macrophages (BMMs) after anti-F4/80 treatment in vitro (Supplementary Figure S7). As T-cell activation requires the MHC–peptide complex-mediated first signal, as well as the co-stimulatory second signal, we wondered whether anti-F4/80 antibody treatment affects these two signals. To test this, we used CpG to stimulate BMM in the presence or absence of anti-F4/80 antibody and then analyzed the surface expression of MHC-I, MHC-II, CD80, and CD86 molecules. We observed decreased surface expression of MHC-I, MHC-II, CD80, and CD86 after anti-F4/80 antibody treatment (Fig. 6c and data not shown), which suggests that the anti-F4/80 antibody used may affect T-cell activation through impairment of both antigen presentation and co-stimulation. In addition, the anti-F4/80 antibody can function as either an agonist or antagonist. To distinguish between these two possibilities, we used the extracellular domain of F4/80 to generate a F4/80-hIg fusion protein (F4/80-Fc), which can serve as an antagonist by competing with membrane-bound F4/80 for binding to its ligand(s). Interestingly, treatment of BMM with CpG in the presence of F4/80-Fc resulted in a reduction in MHC-I/II expression similar to that observed in anti-F4/80 antibody treatment (Fig. 6d), which suggests that the anti-F4/80 antibody may function as an antagonist. Taken together, our data indicate that the F4/80 signal may have an important role in anti-PD-L1 antibody-mediated T-cell activation. However, this mechanism needs to be investigated in further detail.

Fig. 6.

F4/80 is required for the anti-tumor effect of anti-PD-L1 antibody. a WT B6 mice (n = 5/group) were injected subcutaneously with 5 × 10⁵ B16-OVA cells, which was followed by the administration of 25 μg of the indicated control IgG or anti-mPD-L1 on days 9 and 12. Anti-F4/80 antibodies were administered on days 9 and 16. Seven days after the last treatment, dLN cells were collected and an IFN-γ CBA was performed after restimulation with OT-I and OT-II peptide. b Splenocytes from WT B6 mice were stimulated with anti-CD3 and anti-CD28, as well as with different concentration of anti-mPD-L1 and anti-F4/80 antibodies. IFN-γ production was measured by CBA 2 days later. c BMMs were stimulated with CpG and the indicated concentrations of anti-F4/80 for 24 h. The expression of MHC-I and MHC-II was analyzed by flow cytometry. d BMMs were stimulated with CpG and the indicated concentrations of F4/80-Fc for 24 h. MHC-I and MHC-II expression was analyzed by flow cytometry. One representative experiment of two is depicted

Discussion

PD-L1 is an important immune checkpoint in certain types of cancers and delivers an inhibitory signal to T cells through engagement of PD-1. PD-L1 can be expressed by both tumor cells and non-tumor cells. How these different sources of PD-L1 contribute to T-cell inhibition has recently attracted much attention; however, the contribution of PD-L1 expression on tumor cells compared with that on non-tumor cells has remained unclear, which is partly due to the use of different experimental models in individual studies.^18–21 Using a novel tumor model, we have now revealed the following: (1) both tumor- and non-tumor-derived PD-L1 can support T-cell inhibition in the tumor microenvironment; (2) simultaneously blocking tumor- and non-tumor-derived PD-L1 can synergistically reduce T-cell inhibition and promote anti-tumor T-cell responses; (3) the expression ratio of PD-L1^tumor/PD-L1^non-tumor determines their relative contribution to immune suppression; and (4) the F4/80 receptor is important for anti-PD-L1-mediated T-cell reactivation.

To dissect the role of tumor- and non-tumor-derived PD-L1 in the suppression of the anti-tumor T-cell response, PD-L1^–/– mice and PD-L1^–/– tumor cells are often used. In recent studies,^18,19 PD-L1^–/– mice were used to exclude the contribution of PD-L1 from non-tumor cells. However, genetic disruption of PD-L1 in mice may cause developmental abnormalities. For example, it has been reported that both CD4⁺ and CD8⁺ T cells are overactivated in PD-L1^–/– mice, which is demonstrated by delayed MC38 cell-derived tumor growth in these mice.^18,19,24 Therefore, the immune-activated environment in these mice is very different from the immune suppressive environment that is observed in cancer patients. To understand the role of tumor-derived PD-L1, PD-L1^–/– tumor cell lines are usually used to exclude the contribution of tumor-derived PD-L1.^18–20 However, these tumor cell lines are deficient in PD-L1 during the entire tumor development period, which may result in a different tumor microenvironment. Therefore, the conclusions drawn from these models may not apply to naturally occurring tumors, as most tumor cells can express PD-L1 to some degree and expression is further enhanced by certain cytokines. To overcome the lack of suitable experimental models, we established tumor models using mPD-L1-expressing tumors in hPD-L1 KI mice or hPD-L1-expressing tumors in WT mice. In our models, the immune suppressive role of PD-L1 on both tumor and non-tumor cells persists during the entire tumor development and immunotherapy period, which better resembles the natural setting. We also took advantage of anti-mPD-L1 and anti-hPD-L1 antibodies, which exhibit no cross-reactivity, to selectively block tumor- or non-tumor-derived PD-L1.

Our data showed that simultaneously blocking tumor- and non-tumor-derived PD-L1 could maximize anti-tumor T-cell responses. Further, by adjusting the expression ratios of PD-L1^tumor/PD-L1^non-tumor, we simulated different clinical situations and concluded that the inhibitory contribution of tumor- and non-tumor-derived PD-L1 was dependent on its relative expression level on tumor and non-tumor cells. Our finding explains why previous studies that used PD-L1^–/– mice determined that tumor-derived PD-L1 was important, whereas other studies that used PD-L1^–/– tumors determined that non-tumor-derived PD-L1 was important. As for which source of PD-L1 has a dominant role in T-cell inhibition, we believe that this is dependent on different tumor models and clinical situations. If tumor cells express low levels of PD-L1, the blockade of non-tumor-derived PD-L1 may be sufficient to re-activate anti-tumor T-cell responses. If tumor cells express high levels of PD-L1, the blockade of tumor-derived PD-L1 is critical for anti-tumor T-cell activation. In addition, it is worthwhile to explore how PD-L1 coordinates with other molecules from tumor and non-tumor cells to inhibit T-cell function. For example, tumor cells express low levels of MHC-I, which could limit T-cell activation. In contrast, non-tumor cells, including dendritic cells (DCs) and macrophages, express high levels of MHC-I. Similarly, other inhibitory molecules may be differentially expressed on tumor and non-tumor cells to enhance the inhibitory signal. These factors should also be considered when the role of PD-L1 in immune suppression is evaluated. Overall, our study emphasized the importance of blocking both tumor- and non-tumor-derived PD-L1 to achieve better therapeutic effects. Therefore, PD-L1 expression on both sources should be considered for patient selection and prediction of therapeutic efficacy.

Although it has been reported that F4/80 may be involved in the development of regulatory T cells,²² the role of F4/80 in anti-tumor responses has not been previously studied. Our data showed that the blockade of F4/80 decreased anti-PD-L1-mediated tumor control and anti-tumor CD8⁺ T-cell responses. Furthermore, blocking F4/80 in BMM in vitro decreased the expression level of MHC-I, MHC-II, CD80, and CD86 in response to CpG stimulation, which suggests that the F4/80 pathway may be critical for antigen presentation and co-stimulation of T cells. Whether inactivation of F4/80 is an anti-PD-L1 resistance mechanism is an interesting question to be addressed in the future. In addition, the activation of F4/80 may present an attractive combinational strategy that may enhance the anti-tumor effect of anti-PD-L1 antibodies. Recent research has shown that an anti-PD-1 antibody can increase macrophage phagocytosis and reduce tumor burden in a macrophage-dependent manner.²⁵ It will be interesting to investigate whether the PD-1 and F4/80 signaling pathways have crosstalk in the regulation of macrophage function in the tumor microenvironment. A detailed understanding of these mechanisms will help to improve the anti-tumor efficacy of anti-PD-L1 antibodies.

Overall, this study has several important implications for cancer immunotherapy by PD-L1 blockade. First, we emphasize that both tumor- and non-tumor-derived PD-L1 can similarly suppress T-cell responses, which suggests that both sources of PD-L1 should be included as predictive diagnostic screening markers. Second, we found that the relative contribution of different sources of PD-L1 is determined by their relative expression level, which can be used to classify different tumor models and cancer patients. Third, we identified that F4/80 is critical for the anti-PD-L1-mediated therapeutic effect, which provides a new direction for the investigation of the mechanism of anti-PD-L1 resistance and provides a potential cancer immunotherapy target. Collectively, our study offers new insights into the cellular mechanisms of immune checkpoint blockade, which may have a significant impact on cancer immunotherapy strategy development and clinical biomarker screening.

Materials and Methods

Mice

C57BL/6J mice were purchased from Beijing Vital River Laboratory Animal Technology Co., Ltd (Beijing, China). hPD-L1 KI mice were purchased from Shanghai Model Organisms Center, Inc. (Shanghai, China). hPD-L1 cDNA was inserted into exon 2 of the mouse PD-L1 locus by CRISPR/Cas9-mediated homologous recombination, which was confirmed by DNA sequencing. All mice were maintained under specific pathogen-free conditions. Animal care and use were in accordance with institutional and NIH protocols and guidelines, and all studies were approved by the Animal Care and Use Committee of Shanghai Jiao Tong University.

Cell lines and reagents

The expression vector pCDH-EF1-MCS was purchased from System Biosciences (Palo Alto, CA, USA). The gene-editing vector lentiCRISPR V2 puro was purchased from Addgene. The B16-OVA cell line was kindly provided by Hans Schreiber (The University of Chicago). Mouse PD-L1 KO B16-OVA cells were generated using the CRISPR/Cas9 system. Briefly, sgRNA (5′-CAAAACATGAGGATATTTGC-3′) against mPD-L1 was cloned into the lentiCRISPR V2 puro vector and lentivirus was produced. B16-OVA cells were infected with lentivirus that expressed Cas9 and sgRNA against mPD-L1. After selection with puromycin, resistant cells were subcloned and PD-L1 KO cell lines were identified by anti-PD-L1 flow cytometry analysis. To establish hPD-L1-expressing mouse PD-L1^null B16-OVA cells, previous mouse PD-L1^null B16-OVA cells were infected with human-PD-L1 that expressed lentivirus (pCDH-EF1-hPD-L1). hPD-L1-positive clones were then sorted and subcloned. Representative clones with different expression levels of human-PD-L1 were selected for further experiments. The B16-OVA cell line and its derivatives were cultured in 5% CO₂ and maintained in vitro in Dulbecco's modified Eagle's medium supplemented with 10% heat-inactivated fetal bovine serum (Gibco), 2 mmol/L l-glutamine, 100 units/mL penicillin, and 100 μg/mL streptomycin. Anti-CD8 (YTS 169.4.2) and anti-CD4 (GK1.5) antibodies were produced in-house. Anti-mouse PD-L1 (10F.9G2) antibodies were purchased from BioXcell and anti-F4/80 antibodies were obtained from the Frank W. Fitch Monoclonal Antibody Facility (The University of Chicago).²⁶

Production of anti-hPD-L1 antibody and F4/80-Fc fusion protein

For the anti-hPD-L1 antibody, the VL and VH regions of anti-hPD-L1 (MEDI-4736) were obtained from the published patent. The anti-hPD-L1 scFv was synthesized by Genewiz (Suzhou, China). The scFv was cloned into the pCDH-EF1 vector with a C terminal human IgG1 Fc fusion. For F4/80-Fc, the extracellular domain of mouse F4/80 was cloned into the pCDH-EF1 vector with a C-terminal human IgG1 Fc fusion. The plasmids containing scFv-Fc or F4/80-Fc were transfected into Lenti-X 293T cells and supernatants were collected and purified by a Protein A column according to the manufacturer’s instructions (Repligen Corporation).

Tumor growth and treatments

Approximately 0.5–1 × 10⁶ B16-OVA cells or their derivatives were injected subcutaneously into the right flank into 5–12-week-old mice. Tumor volumes were measured along three orthogonal axes (a, b, and c) and calculated as tumor volume = abc/2. After the tumor was established (~ 9–12 days), the mice were treated with 3 intratumoral injections of 25 μg of anti-hPD-L1, anti-mouse PD-L1, or control antibody every 4 days. For cell-depleting experiments, 200 μg of anti-CD8, anti-CD4, and anti-F4/80 antibodies were injected intraperitoneally at the same time as the anti-PD-L1 treatment.

Measurement of IFN-γ secretion by cytometric bead array

OT-I or OT-II peptide-reactive T cells were measured by cytometric bead array (CBA). Spleen or lymph node cells were resuspended in RPMI 1640 supplemented with 10% fetal calf serum, 2 mmol/L l-glutamine, 100 units/mL penicillin, and 100 μg/mL streptomycin. In all, 1–4 × 10⁵ spleen or lymph node cells were used for the assay. OT-I or OT-II peptide was added at a concentration of 10 μg/mL. After 48 h of incubation, IFN-γ production was determined by an IFN-γ CBA assay (BD Biosciences).

Detection of endotoxin in mAb preparations

Endotoxin was measured by the limulus amebocyte lysate assay (Cambrex, Inc., MD). For all monoclonal antibody (mAb) preparations, the amount of endotoxin was determined to be < 0.2 E.U./mg mAb.

Flow cytometric analysis

Single-cell suspensions of cell lines were incubated with anti-CD16/32 (anti-FcγIII/II receptor, clone 2.4G2) for 10 min and were then subsequently stained with conjugated antibodies. All fluorescently labeled monoclonal antibodies were purchased from Biolegend or eBioscience. Samples were analyzed on a Cytoflex (Beckman Coulter) or Sony flow cytometer (Sony Biotech), and the data were analyzed with FlowJo software (TreeStar, Inc.).

Statistical analysis

Mean values were compared using an unpaired Student’s two-tailed t-test. Error bars represent SD or SEM. Statistically significant differences are indicated by *p < 0.05 and **p < 0.01, respectively.

Competing interests

The authors declare no competing interests.

Electronic supplementary material

The online version of this article (10.1038/s41423-018-0021-3) contains supplementary material.