Apatinib combined with PD-L1 blockade synergistically impedes tumor growth and improves OS in a GC mouse model
To investigate whether combined therapy would be more efficacious than apatinib or PD-L1 blockade monotherapy, we performed a therapeutic study in a cell line-derived xenograft (CDX) murine model. The mouse GC cell line MFC was inoculated subcutaneously into the flanks of immunocompetent mice. Once tumors were palpable, mice were randomized and systemically treated with isotype control antibody, apatinib, anti-PD-L1 antibody or the combination of anti-PD-L1 antibody and apatinib until the study endpoint. The therapy protocol is shown in Fig. 1a. The groups treated with apatinib monotherapy and anti-PD-L1 antibody monotherapy had significantly improved median overall survival (OS) compared with the control group (Fig. 1b). Strikingly, the combined treatment of apatinib and anti-PD-L1 antibody led to a substantial improvement in median OS compared with control or a single agent alone (Fig. 1b). To further elucidate whether the combination of apatinib and anti-PD-L1 antibody exerted synergistic treatment effects, the predicted additive OS curves were generated according to the method in a recently published paper[16]. The observed survival for the combination therapy was significantly longer than the predicted survival for the combination therapy, which indicated synergism between apatinib and PD-L1 blockade (Fig. 1c). Moreover, after 2 weeks of treatment, we found that treatment with apatinib or anti-PD-L1 antibody alone significantly inhibited the growth of MFC tumors in immunocompetent mice, and combination treatment with apatinib and anti-PD-L1 antibody led to a further reduction in tumor volume and tumor weight (Fig. 1d-e). In addition, combined treatment dramatically decreased the number of Ki-67-positive proliferative tumor cells compared with the control or a single agent alone (Fig. 1f). Collectively, these data suggest that treatment of MFC-bearing mice with combined apatinib and PD-L1 blockade therapy synergistically improves OS and inhibits tumor growth compared with apatinib or PD-L1 blockade monotherapy.
Combined apatinib and PD-L1 blockade reprograms the immune microenvironment and promotes antitumor immunity in a GC mouse model
We next examined how apatinib monotherapy, anti-PD-L1 monotherapy and combined apatinib and PD-L1 therapy impact lymphocyte infiltration in MFC tumors. Immunohistochemical analysis showed that apatinib treatment alone increased the number of Foxp3+ Treg cells, while the number of Foxp3+ Treg cells was not significantly altered upon anti-PD-L1 monotherapy and combined apatinib/anti-PD-L1 therapy compared with control treatment (Fig. 2a). Moreover, the number of CD8+ T cells and CD20+ B cells was significantly increased upon anti-PD-L1 and combined apatinib/anti-PD-L1 therapy but was not significantly altered upon apatinib therapy compared with those in the control group (Fig. 2b-c). Furthermore, the combination therapy led to a higher ratio of CD8+ T cells to Foxp3+ Treg cells, indicating a shift towards immune activation rather than an immunosuppressive microenvironment (Fig. 2d). To validate the activation of tumor-infiltrating CD8+ T cells after PD-L1 blockade, we evaluated biomarkers (IFN-γ, TNFα and IL-2) indicative of CD8+ T cell activation in mouse plasma at the study endpoint after treatment. Significantly increased levels of IFN-γ, TNFα and IL-2 were observed when mice were treated with the apatinib/anti-PD-L1 combination treatment, while apatinib or anti-PD-L1 monotherapy only weakly increased activated CD8+ T cells (Fig. 2e). Furthermore, when evaluating the Th1/Th2 cytokine balance by ELISA, we found that the ratio of Th1 to Th2 cytokines (IFN-γ to IL-10) was significantly increased upon combined apatinib and ani-PD-L1 therapy compared with control treatment for monotherapy (Fig. 2f). Collectively, these data suggest that combined apatinib and PD-L1 blockade reprograms the immune microenvironment and promotes antitumor immunity in a GC mouse model.
Apatinib increases PD-L1 expression in gastric cancer cells via hypoxia and IFN-γ expression
We next investigated the impact of apatinib monotherapy, anti-PD-L1 monotherapy and combined apatinib/anti-PD-L1 treatment on PD-L1 expression in GC tumors. Immunohistochemical analysis showed that the expression of PD-L1 was dramatically increased in tumor samples upon apatinib monotherapy and combined apatinib/anti-PD-L1 therapy compared with anti-PD-L1 monotherapy (Fig. 3a). Flow cytometry further confirmed a substantial upregulation of PD-L1 on tumor cells isolated from MFC xenografts upon apatinib monotherapy and combined apatinib/anti-PD-L1 therapy, whereas the expression of PD-L1 on anti-PD-L1-treated tumor cells remained low, comparable to that in control tumors (Fig. 3b). PD-L2 expression was not significantly altered upon apatinib, anti-PD-L1 or combined anti-VEGF/anti-PD-L1 targeted therapy compared to control treatment (Fig. 3b). Hypoxia, which can be caused by antiangiogenic therapy, has been reported to increase the expression of PD-L1 on tumor cells via HIF-1α[17]. Therefore, we investigated whether apatinib-induced hypoxia increases PD-L1 expression in MFC tumors. We observed increased nuclear expression of HIF-1α in apatinib- and combined apatinib/anti-PD-L1-treated MFC-bearing mouse models compared with that in control-treated mice, accompanied by increased PD-L1 expression (Fig. 3c). We also observed that during apatinib treatment, PD-L1 was strongly correlated with HIF-1α expression (Fig. 3d), indicating PD-L1 expression by GC cells, particularly in hypoxic regions. To confirm the direct effect of hypoxia on PD-L1 expression in GC cells, we examined the expression of PD-L1 in GC cells under hypoxic vs normoxic conditions and observed a significant increase in PD-L1 expression on GC cells upon exposure to hypoxia (Fig. 3e-f). We further examined whether apatinib exerted a direct effect on PD-L1 expression in GC cells. We found that apatinib did not result in an appreciable change in PD-L1 expression in GC cells under hypoxic and normoxic conditions (Additional file 1: Figure S1A). Given that IFN-γ has been reported to mediate PD-L1 expression in tumor cells[18], we next investigated whether IFN-γ was involved in the PD-L1 expression induced by apatinib treatment in GC cells. As shown in Fig. 3g-h, IFN-γ dramatically induced PD-L1 expression in GC cells. By ELISA, we found that IFN-γ was significantly increased in the plasma of MFC-bearing mice upon apatinib and apatinib/anti-PD-L1 combination treatment (Fig. 2e). The expression of IFN-γ-mediated genes, CXCL10, MX1 and IFIT3[19], was also higher in MFC cells isolated from tumors treated with apatinib and apatinib/anti-PD-L1 combined therapy compared with that in the control or anti-PD-L1 group (Fig. 3i). These results suggest that apatinib enhances the expression of PD-L1 partially by inducing IFN-γ expression as well as by inducing a hypoxic microenvironment in GC.
Combined apatinib and PD-L1 blockade induces HEV formation that is associated with LTβR signaling activation mediated by dendritic cells
Reprogramming the tumor vasculature by normalization could lead to a shift towards an immune-activated microenvironment[6]; therefore, we hypothesized that combined apatinib and PD-L1 blockade therapy may affect the tumor vasculature in GC. Immunohistochemical analysis showed that the vessel density, as indicated by the number of CD-31-positive vessels per area, was dramatically reduced in tumor samples upon apatinib monotherapy and combined apatinib/anti-PD-L1 therapy but not in anti-PD-L1 monotherapy-treated tumors (Fig. 4a). HEVs, which are structurally distinct blood vessels, are thought to facilitate the generation of tissue-destroying lymphocytes inside chronically inflamed tissues and cancers[10]. Using immunohistochemical staining for the HEV-specific marker MECA-79 [20], we found that naive MFC tumors were devoid of MECA-79+ vessels, whereas MECA-79+ vessels were markedly induced by combined apatinib/anti-PD-L1 therapy (Fig. 4b). Previous studies suggested that continuous stimulation of lymphotoxin‑β receptor (LTβR) on HEV endothelial cells by CD11c+ DCs is essential for the induction and maintenance of HEVs in lymphoid tissues[21, 22]. Indeed, the number of CD11c+ DCs was significantly increased in MFC tumor tissues upon combined apatinib/anti-PD-L1 treatment (Fig. 4c). We next isolated CD11c+ DCs from MFC tumors upon apatinib monotherapy, anti-PD-L1 monotherapy and combined apatinib/anti-PD-L1 therapy. We found that the activity of CD11c+ DCs, indicated by upregulation of the CCR7, CD40, IL-10 and IL-12 genes[23], was greatly enhanced upon anti-PD-L1 monotherapy and combined apatinib/anti-PD-L1 treatment (Fig. 4d). Additionally, the expression of the LTβR ligands LTβ and LIGHT, but not LTα, was also upregulated in CD11c+ DCs isolated from tumors upon anti-PD-L1 monotherapy and combined apatinib/anti-PD-L1 treatment, whereas no significant change in the expression of LTα, LTβ and LIGHT was observed in MFC cells isolated from tumors upon apatinib monotherapy (Fig. 4e). We next examined the role of LTβR signaling in HEV formation during combined apatinib/anti-PD-L1 treatment. We found that an LTβR antagonist exerted little effect on the proliferation of GC cell lines in vitro (Additional file 1: Figure S1B). We next tested the effect of combined treatment with an LTβR signaling antagonistic antibody and combined apatinib/anti-PD-L1 therapy in an immunocompetent mouse GC model. As shown in Fig. 4f-h, no significant tumor suppression was observed in the LTβR antagonist-only-treated group compared with the control group. Meanwhile, combined apatinib/anti-PD-L1 treatment significantly decreased the tumor burden of MFC tumors by approximately 80%. Notably, the addition of an LTβR antagonist during combined apatinib/anti-PD-L1 treatment substantially weakened the antitumor effect (Fig. 4f-h). Immunohistochemistry was used to evaluate the density of MECA-79+ HEVs, CD8+ T cells and CD20+ B cells in the xenograft tumors. We found that the HEV formation, and the CD8+ T cell and CD20+ B cell infiltration driven by combined apatinib/anti-PD-L1 treatment were also blocked by treatment with an LTβR antagonist (Fig. 4i-j). These data support the idea that LTβR signaling activated by DCs contributes to HEV formation during combined apatinib/anti-PD-L1 treatment.
Presence of HEVs correlates with clinical outcome in GC patients
To identify the presence of HEVs in human GC, tumor tissues from 192 surgically resected gastric adenocarcinoma patients were analyzed for expression of the HEV-specific marker MECA-79. Vessels expressing MECA-79 were detected in the majority of GC tumors analyzed (118/192). A portion of MECA-79-positive vessels displayed a cuboidal endothelial appearance, typical of lymph node HEVs (Fig. 5a). MECA-79 was also expressed on endothelial cells with a flat morphology, typical of the tumor vasculature (Fig. 5a). The density of MECA-79-positive HEVs varied in patients, ranging from 0 to 6.2 HEVs per mm3 (median: 0.8 HEVs/mm3, mean: 1.1 HEVs/mm3) (Fig. 5a). Moreover, immunostaining of the pan-vascular endothelial cell markers CD31 and MECA-79 in GC samples demonstrated that the percentage of MECA-79+ vessels ranged from 1–11% of the CD31+ vessels (Fig. 5b), and no significant correlation between intratumoral MECA-79+ and CD31+ vessels was observed (Fig. 5c), suggesting that the differences in angiogenesis were not responsible for the differences in the density of HEVs in gastric tumors. Furthermore, a strong association between DC-LAMP+ (LAMP3+) mature dendritic cells and the presence of MECA79+ high endothelial venules was identified (Fig. 5d). We further tested the correlations between the density of MECA-79-positive HEVs and the clinicopathologic status and prognosis of patients with GC. We found that the density of HEVs in gastric adenocarcinoma samples of early stage (TNM stage I–II, n = 81) was significantly elevated compared to that in gastric adenocarcinoma samples of advanced stage (TNM stage III–IV, n = 111) (Fig. 5e; Additional file 2: Table S7). Further analyses showed that a low density of HEVs in gastric adenocarcinoma was positively associated with relatively higher tumor size, vascular invasion and lymph node metastasis than a high density of HEVs (Fig. 5f-h; Additional file 2: Table S7). Kaplan-Meier analysis revealed that a low density of HEVs predicted a poor prognosis for overall survival in gastric adenocarcinoma patients after surgery (Fig. 5i). Cox regression multivariate analysis also revealed that HEV density, tumor differentiation, lymph node metastasis and TNM stage were independent prognostic factors for the survival of gastric adenocarcinoma patients (Fig. 5j; Additional file 2: Table S8).
HEVs predict lymphocyte infiltration into the tumor in GC patients
Because HEVs are involved in the large-scale migration of lymphocytes from the blood into lymph nodes, we next investigated the relationship between HEVs and immune infiltration in GC. Subsequently, we quantified the number of tumor-infiltrating CD3+ T cells, CD8+ T cells and CD20+ B cells by IHC staining in resected tumors from 192 gastric adenocarcinoma patients. We found infiltration of CD20+ B cells in 39% of patients; 27% of the tumors had CD20+ B cell localized in clusters, and 34% were devoid of CD20+ B cells (Fig. 6a). Notably, CD20+ B cell clusters were in all cases colocalized with MECA79+ high endothelial venules, as well as CD3+ and CD8+ T cells, which indicates the formation of tertiary lymphoid structures (TLSs) (Fig. 6a). In contrast, the infiltration of CD8+ T cells was found in 49% of cases, and 27% of tumors had CD8+ T cells localized outside of such TLSs (Fig. 6a). Furthermore, we observed a strong correlation between the number of HEVs and intratumoral CD8+ T cells (r = 0.669; P < 0.001) and CD20+ B cells (r = 0.751; P < 0.001) (Fig. 6b-c). The mRNA expression levels of genes associated with naive and central memory T and B lymphocyte migration, including CCL19, CCL21, CXCL13 and CCR7[24], were significantly elevated in the HEV-high versus HEV-low tumors (Fig. 6d).
Microsatellite instability (MSI) is a tumor molecular phenotype caused by a defect in the DNA mismatch repair (MMR) system machinery, and MSI-H-induced immunogenic neoantigens are thought to trigger a potent antitumor response in the presence of immune checkpoint blockade[25]. We therefore explored the relationship between MSI and the immune phenotype in the GC microenvironment and whether HEVs were correlated with immune infiltration in MSI GCs. By comparing 21 MSI (MSI-H = 10, MSI-L = 11) against 12 MSS gastric adenocarcinoma tissue samples, we observed that the abundance of intratumoral CD8+ T cells and CD20+ B cells in MSI-H gastric adenocarcinoma tissues was higher than that in the MSI-L or MSS GC samples (Fig. 6e). Strikingly, the numbers of CD8+ T cells and CD20+ B cells in MSI-L tumors were comparable to those in MSS samples (Fig. 6e). Nest, we tested whether HEVs were correlated with the abundant immune infiltration in MSI GCs. As shown in Fig. 6f, the density of HEVs was higher in MSI-H than MSI-L or MSS GC samples. However, the density of HEVs was comparable between MSI-L and MSS GC samples.
Survival analysis revealed that the presence of TLSs was associated with improved patient outcome (Fig. 6g). Gastric adenocarcinoma patients in the CD8+ T cell-high or CD20+ B cell-high group exhibited significantly longer OS than those in the CD8 + T cell-low or CD20+ B cell-low group (Fig. 6h-i), and when HEVs, CD8+ T cells and CD20+ B cells were considered together, we observed that patients with low levels of all HEVs, CD8+ T cells and CD20+ B cells exhibited the shortest OS, whereas those with high levels of all HEVs, CD8+ T cells and CD20+ B cells displayed the best prognosis (Fig. 6j).