PD-L1 was inversely correlated with the infiltration of Tm cells in CRC
To explore the potential correlation between tumor PD-L1 expression and lymphocytes infiltration in CRC, we detected PBLs and TILs of 64 patients with stage II-III and tumor PD-L1 expression. Tumor PD-L1 expression was scored as 0 (absent), 1 (weak), 2 (moderate) or 3 (strong) in cytoplasm, and membrane expression was scored as 0 (absent) or 1 (present; distinct membrane staining) (Fig. 1A) and a total PD-L1 score ranging from 0 to 4 was calculated. Based on the total PD-L1 score, patients were divided into PD-L1 low (0–2) or high (3–4) expression group. We detected the levels of circulating immune cells by flow cytometry (fig. S1) and analyzed potential correlations with PD-L1 expression. More CD8(+)T cells, Tm cells, including CD4(+)Tm cells, CD8(+)Tm cells were observed in peripheral blood of patients with high PD-L1 expression (Fig. 1B).However, the levels of other kinds of immune cells in peripheral blood had no significant difference between the PD-L1 high and low expression group (fig. S2,A-I).
Patients with high PD-L1 expression had increased infiltrated of CD8(+)T cells in intraepithelial area and increased levels of circulating CD8(+)T cells in blood (Fig. 1C). Interestingly, in terms of Tm cells, high PD-L1 expression accompanied with an increased level of circulating Tm cells, but a decreased level of infiltrated Tm cells in tumor (Fig. 1D). In addition, we analyzed the potential correlation between the infiltration differences and patients survival. By calculating the ratio of P/T(PBLs/TILs), we found the P/T value was significantly higher in patients with postoperative metastasis (M1) than that of patients without metastasis (M0) (Fig. 1E). Patients with PD-L1 high expression had a higher postoperative metastasis rate (Fig. 1F).
Whether the presence of PD-L1 inhibited the infiltration of Tm cells, or the lower infiltration of Tm cells leading to the increased expression of PD-L1, aroused our curiosity. We hypothesized that tumor PD-L1 can inhibit the infiltration of Tm cells.
Tumor PD-L1 inhibits the migration and infiltration of Tm cells
To investigate the effects of supernatant of PD-L1 positive cells on the migration of Tm cells, the supernatant of control/si-PD-L1 HT-29 cells separately was used to stimulate the peripheral blood lymphocytes of healthy people. It was found that the migrated number of Tm cells increased significantly after PD-L1 was knocked down(Fig. 2A).
Recent study have shown that in a mouse ovarian cancer xenograft model, tumor-infiltrating CD4(+)T cells and CD8(+)T cells were significantly increased after PD-L1 were knockout, and the levels of chemokines CXCL9 and CXCL10 in tumor were increased(18)which can promote the migration of T cells by binding to its CXCR3 receptor(19). For Tm cells, the expression of the CCR7 membrane protein is closely related to cell migration, cell activation and differentiation. The combination of CCR7 and its ligand CCL19 can activate downstream signaling pathways and change the migration, differentiation and other behaviors of Tm cells(19).Therefore, we speculate that in CRC, PD-L1 may inhibit the infiltration of Tm cells by affecting chemokines secretion in TME. In order to explore the effects of PD-L1 on these receptors, we compared the expression of surface CCR7/CXCR3 on Tm cells after stimulated by si-PD-L1 /control HT-29 supernatant. After a 24-hours stimulation with the si-PD-L1 supernatant, the proportion of CCR7-positive Tm cells was decreased, while the proportion of CXCR3-positive cells did not change significantly (Fig. 2B), suggesting that the expression of CCR7 was decreased in si-PD-L1 group. This short-term decrease was probably due to the endocytosis after the membrane receptor CCR7 bind to its ligand, which greatly consumed the CCR7 protein. The decrease of CCR7 indicated the ratio of effector Tm cells was up-regulated, consistent with the increase of Tem cells in the xenograft model, that is, the expression of PD-L1 inhibit the differentiation of Tcm cells into its effector type(fig. S3A). Furthermore, we co-cultured human peripheral lymphocytes with HT-29 cells (Fig. 2C) and found the CCR7 and CXCR3 receptors were significantly decreased in si-PD-L1 group (Fig. 2D). In summary, knocking down PD-L1 can reduce the expression of CCR7, CXCR3 on Tm cells, and PD-L1 protein can inhibit the transformation of Tm cells into Tem cells.
Although the expression of tumor PD-L1 did not affect its own secretion of CXCL9 and CXCL10(fig. S3B), it cannot be ruled out that PD-L1 protein can regulate the secretion of CXCL9, CXCL10 in DC cells. We used different cell supernatants to stimulate healthy human peripheral lymphocytes, and found that the secretion of CXCL10 in DC cells(CD14+) was significantly increased after stimulated by PD-L1 knockdown supernatants(fig. S3,C-D). The co-culture experiments of tumor cells and lymphocytes indicated that after PD-L1 was knocked down, CXCL9 and CXCL10 secreted by CD14(+) cells were significantly increased(Fig. 2E). Therefore, the PD-L1 protein can reduce these chemokines secreted by DC cells.
In vivo, Lenti-Cas-sgPD-L1 was used to knock down the PD-L1 expression in MC-38 cells to establish a C57BL/6 mouse orthotopic xenograft model(Fig. 2F). In sg-PD-L1 group, the infiltration of Tm cells (CD44 + CD62L+/-) was significantly higher than that in control(Fig. 2G), especially with an increase of Tem subtypes(CD44 + CD62L-)(fig. S3A).There was no difference on the expression of GanB and Perf between the sg-PD-L1 group and its control(Fig. 2H).Furthermore, we found the infiltrated number of Tm cells was also increased in metastases after knocking out PD-L1 in a mouse lung metastasis model(Fig. 2I), also with an increase of Tem cells(fig. S3E). In general, PD-L1 expressed by tumor cells can inhibit the infiltration of Tm cells by reducing the secretion of CXCL9,CXCL10 in DCs in colorectal cancer (as the blue cycle in fig. S4).
Exploring the main subsets of Tm cells inhibited by PD-L1 on infiltration
In order to identify the main subsets of inhibited Tm cells, we selected individuals with significant differences in Tm cells infiltration as research subjects. The number of Tm cells in peripheral blood(P) and tumor(T) from 5 CRC patients (P1-P5) were measured by flow cytometry (Fig. 3A). The ratio of Tm cells in peripheral blood to Tm cells in tumor (P/T) were calculated. It was found that P1, P2 has the maximum, minimum P/T ratio and their PD-L1 expression was obviously distinct(Fig. 3B). Therefore, P1 and P2 were selected for further research. We sorted out Tm cells from this two matched samples by flow cytometry and generated single-cell transcriptome profiles. Using GEM (Gel bead in emulsion) to capture single cells through the 10x Genomics platform. A total of 20,900 cells passing the quality control were used for further analysis (4,832-5,578 cells; median: 5,225 cells / sample). To clarify the subsets, we identified 12 clusters by T-distributed stochastic neighbor embedding (t-SNE)(Fig. 3C). There were five CD4(+)Tm cells clusters, including naive Tm cells (C1-blood, C3-tissue, C8-tissue), effector Tm cells (C9-tissue) and regulatory Tm cells (C6-tissue). At the same time, seven CD8(+)Tm clusters were identified, including cytotoxic Tm cells (C2 ,C4 ,C5), exhausted Tm cells (C7, C10, C11) and proliferative Tm cells (C12)(Fig. 3C).
Tm cells in blood are mainly divided into three clusters, C1-CD4 LEF1, C2-CD8 GZMH, C5-CD8 GZMB(Fig. 3D). C1-CD4 LEF1 showed high naive marker expression LEF1, TCF7. C2-CD8 GZMH cells overexpressed cytotoxic gene GZMH, GNLY, with a high expression of FGFBP2 which encoded a serum protein related to fibroblast growth factor. C5-CD8 GZMB was characterized by NKG7 and KLRG1 expression (Table. S1), showing similar characteristics with natural killer cells. In terms of the number of this three clusters, P2 with high PD-L1 expression has more C2-CD8 GZMH cells, less C1-CD4 LEF1 cells and a similar content of C5-CD8 GZMB cells as P1(Fig. 3E).
There are more cytotoxic CD8(+)T cells(Cluster4) and naive T cells(Cluster8) in P1 who had a higher lymphocyte infiltration(Fig. 3F-G). The proportion of exhausted CD8(+) T cells (represented by Cluster7 and Cluster11) is higher in the tumor of P2, patient with low lymphocytic infiltration, while in P1 the exhausted CD8(+)T cells are mainly dominated by Cluster 10 expressing HAVCR2, CXCL13 (Fig. 3H).In addition, there is a group of effector CD4(+)T cells (Cluster9) in P2. C9-CD4-effector displayed the high expression of chemokine related genes CXCR6, tissue-resident marker which can regulate Tm lymphocyte migration(20), CCL20 which repressed the proliferation of myeloid progenitors(21), IL23R which was described as a exclusion chemokine can dampen the tumor infiltration of CD8 + T cells(22). More this CD4 effector cells were observed in P2 than P1 tumor, implying that the lower lymphocyte infiltration in P2.
We identified the position of each cluster in cell differentiation and the dynamic changes of key genes by cell trajectory analysis (Fig. 3I). In pseudotime analysis, we found that the C2-CD8 GZMH and C5-CD8 GZMB cells were at the beginning of the trajectory, C4-CD8-cytotoxic cells as an intermediate cytotoxic state characterized by expressing EOMES, whereas the C7, C10 cells and most CD11 cells at a terminal state. Additionally, most C12-CD8-MKI67 cells were at the terminal state.(fig. S5 A-B)
Single-cell regulatory network inference and clustering (SCENIC) analysis of Cluster 2
In order to clarify the identity of each cluster, we investigated the differences in transcription factors and downstream target genes as previously described(23). Single-cell regulatory network inference and clustering (SCENIC) analysis was performed for the blood and tumor samples (Fig. 4A) and 12 clusters (Fig. 4B),respectively (23). The connection specificity index (CSI) analysis of 4 samples was performed among 28 regulons within nine distinct regulon modules (Fig. 4C). Obviously, module 5 displayed the highest regulation activity in both blood and tumor samples, especially in P2P (Fig. 4A). Consistently, the analysis for each cluster revealed that the module 5 has the highest regulation activity among the nine regulon modules, especially in Cluster 5 (Fig. 4B). There is no difference in the proportion of Cluster 5 between P1P and P2P. In P2P, only the cell number of Cluster 2 is much higher than that in P1, so the high activity of module 5 in P2P may be attributed to Cluster2. Therefore, Cluster 2 should be the focus group for further research. We obtained the specific correspondence between regulons and clusters by calculating the regulon specificity score (RSS). The RSS ranking showed that in Cluster 2, TBX21, EOMES and RUNX3 scored the highest, implying that these transcription factors may be specifically associated with this cluster(Fig. 4D).TBX21 regulated the expression of IFN-γ in Th1 and NK cells, suggesting a role for this gene in initiating Th1 lineage development from naive cells, as well as EOMES and RUNX3, were IFN-γ regulating transcription factors(24).EOMES was necessary for the differentiation of effector CD8(+)T cells(25). RUNX3 can increase the aggregation of CD8(+)T cells in tumor(26). It seems that RUNX3 transcriptional activation may promote tissue infiltration, but in fact, the infiltration of Cluster2 was not increased. We wondered what inhibited the infiltration of such cells.
Gene expression characteristics of FGFBP2(+)Tm cells
C2-CD8 GZMH showed high expression levels of FGFBP2, GZMH, TRBV7-2(Fig. 5A). Differential gene analysis (P2P vs P2T) also found that the expression of the gene FGFBP2, GZMH, TRBV7-2 were significantly increased in P2P, confirming that C2-CD8 GZMH is the main differential subset in blood(Fig. 5B). It was reported that nearly all fgfbp2-expressing CD4(+) and CD8(+) T cells showed the ability to produce IFN-γ but not IL-2 (12), which is consistent with our results(Fig. 5C). It has confirmed IL-2 can increase the infiltration of CD8(+)T cells in tumor(27). Therefore, higher numbers of fgfbp2-expressing CD8(+)T cells means more secretion of IFN-γ (the main cytokine responsible for the upregulation of PD-L1 in tumors), while less IL-2 means less CD8(+)T cells infiltration. This is consistent with the distribution of Tm cells in P2 that a high number of fgfbp2-expressing CD8(+) Tm cells in blood, low infiltration of CD8(+)Tm cells in tumor with high expression of PD-L1 in tumor(as the red cycle in fig. S4). The heatmap displayed FGFBP2 genes, as well as the cytotoxic gene GZMH, GNLY, NKG7 were downregulated along the cell trajectory(fig. S6).These FGFBP2(+)Tm cells can promote tumor angiogenesis by secreting FGFBP2 protein, which can bind to FGFs in the extracellular matrix and regulate tumor growth and metastasis(15,16). The density of vessels shapes blood flow of the tumor, and the abundance of blood supply is critical for immune cells infiltration. Therefore, FGFBP2(+)Tm cells may be closely related to tumor angiogenesis and progression.
The increased FGFBP2(+) Tm cells promotes tumor progression
To further clarify the distribution of FGFBP2(+)Tm cells, we collected peripheral blood, lymph nodes, colon polyps, primary tumor, and liver metastases from a colon cancer patient with synchronous liver metastases and sorted out Tm cells from these 5 samples by flow cytometry and generated single-cell transcriptome profiles(Fig. 6A). We found that the distribution of Tm cells in different tissues can accurately reflect the pathological stages of tissues. First, as the tumor progresses, the number of Cluster6 (C6) characterized by high expression CXCR5 is gradually reduced(Fig. 6.A-C, fig. S7). C6 accounted for about 26% of the total Tm cells in polyp, and accounted for about 5% in primary tumor and almost 0 in metastases. Second, from polyp to primary tumor and then to metastases, the ratio of Cluster2(C2) and Cluster4(C4) changed significantly. In polyp, the proportions of C2 and C4 were approximately the same(Fig. 6B), however, the proportions of C2 and C4 in primary tumor and metastases were “reversed”—the proportion of C4 was significantly higher than C2 in primary tumor, while the proportion of C4 was significantly lower than C2 in metastases(Fig. 6.A,C). Third, Cluster3 (C3) with high expression of LEF1 and CCR7 mainly existed in blood (Fig. 6D), and its content decreases sequentially from blood to lymph nodes and then to solid tumor (Fig. 6A-E). Fourth, more importantly, we found C11, C12 highly expressed FGFBP2 gene(Fig. 6F). Most of C11 appeared in blood, and C12 was specifically accumulated in metastases(Fig. 6C). The infiltrated number of FGFBP2(+)Tm cells in tissue increased with tumor progression. Specifically, FGFBP2(+) Tm cells rarely appeared in polyp, with an increased number in primary tumor and exhibited a specific accumulation in metastases.C12 is characterized by high expression of FGFBP2 and GNLY, while the expression of CX3CR1 is significantly lower than that of C11(Fig. 6F-G).Therefore, C12 can be clearly distinguished from the other high FGFBP2 expression cluster in blood and exhibited aggregation in metastases.
We explored the tissue distribution characteristics of FGFBP2(+) Tm cells in CRCs. First, we found the cell migration ability of FGFBP2(+) Tm cells was lower than that of FGFBP2(-)Tm cells (Fig. 7A-B). As a specific cluster of Tm cell, the infiltration of FGFBP2(+)Tm cells is affected by the expression of tumor PD-L1. The number of FGFBP2(+)Tm cells in tumor was reduced in patients with high PD-L1 expression(Fig. 7C). In vivo, in the mouse lung metastases model, the FGFBP2(+)Tm cells in lung metastases of PD-L1 knockout mice were significantly higher than those of the control(Fig. 7D-E).
It was found that FGFBP2(+)Tm had the strongest angiogenic ability in all subsets(Fig. 8A). We observed the distribution of FGFBP2 protein in mouse lung metastases was concentrated around small vessels (Fig. 8B). Similarly, the infiltrating FGFBP2(+)Tm cells were mainly distributed at the margin of the liver metastases in CRC (Fig. 8C), and their distribution was also closely related to the formation of small vessels in tumor interstitium (Fig. 7C, 8C).In mouse lung metastases model(Fig. 7E), PD-L1 knockout mice with high FGFBP2(+)Tm cells infiltration had a higher vessel density in metastases(Fig. 8D) and more lung metastatic nodules(Fig. 8E).It appears that infiltrating FGFBP2(+)Tm cells in metastases are closely related to angiogenesis and involved in the progression of tumor metastases. In addition, dual immunofluorescence showed that FGFBP2(+)Tm cells were significantly higher in metastatic lymph nodes than in non-metastatic lymph nodes in CRC patients(Fig. 8F-G). Overall, the aggregation of FGFBP2(+)Tm cells in CRC indicates the occurrence of tumor metastasis. Furthermore, we collected 28 fresh CRC specimens and performed transcriptome sequencing analysis of 57 lymph nodes, including 24 metastatic lymph nodes and 33 non-metastatic lymph nodes. There were significant differences in gene expression between positive and negative lymph nodes (Fig. 8H), and the expressions of FGFBP2 and FGF19 in metastatic lymph nodes were significantly higher than those in non-metastatic nodes (Fig. 8I). Mechanistically, we previously found FGF19 exerts an immunomodulatory function that creates an environment conducive for metastasis in CRC. FGF19-induced inflammatory cancer-associated fibroblasts promote liver metastasis by inducing neutrophil extracellular trap formation. Overall, FGFBP2 secreted by FGFBP2(+)Tm cells is likely to promote the aggregation of FGF19 by binding to FGF19 secreted by tumor cells, inducing the above signaling pathways to promote metastasis. The promotion effect of the FGF-FGFBP system on tumor metastasis has been further verified.
In conclusion, we found that PD-L1 highly expressed in primary tumor can inhibit the infiltration of Tm cells, especially FGFBP2(+)Tm cells, and the increased FGFBP2(+)Tm cells can promote tumor metastasis though FGF-FGFBP system.