Expression of the Co-inhibitory Molecules Across Different Cancers
The co-inhibitory molecules play an important role in the tumor microenvironment (TME) by contributing to T cell exhaustion and immune escape. We hypothesized that these molecules might have unique expression landscapes in various immune cells to perform their respective functions. The specific expression of these molecules in immune cells including Granulocytes, Monocytes, T cells, B cells, DCs, NK cells, and Progenitors was shown in the Supplementary Fig. 1A. CTLA4, LAG3, PD-1, TIGIT were mainly expressed on T cells, whereas CD80 was highly expressed on B cells, and VISTA as well as CD155 was primarily expressed on Granulocytes and Monocytes. Noticeably, the expression of HMGB1 in the above immune cells was relatively even and high. These results suggested that the expression of co-inhibitory molecules was not limited in effector cells, so their roles in anti-tumor immunity needed to be further investigated. We then compared the mRNA levels of these molecules in tumor and normal tissues. Fig. 1A illustrates the landscape of the log2(fold change) (FC) of these 18 genes across 24 TCGA cancers with available tumor and normal tissues. CTLA4, LAG3, PD-1, TIGIT, TIM3, CD80, CD86, CD112, CD155, FGL1, PD-L1 and VEGFA tended to have log2 (FC) values greater than zero, while the values of log2 (FC) of VISTA were mostly below zero (P < 0.05).
Then, we searched the TIMER platform which provides the visualization of pan-cancer expression profiling of co-inhibitory molecules. CTLA4 and PD-1 expression levels were upregulated in tumor tissues relative to non-carcinoma tissues in BRCA, CHOL, ESCA, HNSC, KIRP, LIHC, LUAD, STAD, UCEC, whereas both were downregulated in KICH. PD-1 expression also rose in KIRC and decreased in THCA (Fig. 1B, C). The similar changing trends of these two molecules indicated that they might affect each other in the development and even in the treatment of cancer. The expression of TIM3 in BRCA, CESC, CHOL, ESCA, GBM, HNSC, KIRC, KIRP, STAD, THCA and UCEC tumor cells was significantly higher than that in normal cells, while its expression levels in LUAD and LUSC tumor cells were lower than normal cells (Fig. 1D). The specific expression of TIM3 in lung cancer might suggest differences when it plays its role. Besides, there was an increase in LAG3 and TIGIT expression levels in tumor tissues in BRCA, ESCA, HNSC, KIRC, LUAD, LUSC, while a similar decline in COAD and SKCM. LAG3 was also upregulated in GBM, PCPG and downregulated in KICH, LIHC, PRAD, READ, THCA, UCEC, whereas TIGIT was additionally upregulated in KIRP, STAD, UCS (Fig. 1E, F). Not only did the two show relatively high similarity in differential expression, but also in prognostic studies. NRP1 could be more detected in CHOL, ESCA, GBM, HNSC, KIRC, LIHC, STAD tumor cells than in normal cells, while its expression levels in BLCA, BRCA, CESC, COAD, KICH, LUSC, READ, SKCM, UCEC tumor cells were lower (Fig. 1G). The expression of VISTA showed an upward trend in CHOL, KIRC and LIHC tumor cells. In BLCA, BRCA, CESC, COAD, KICH, LUAD, LUSC, PRAD, READ, STAD and UCEC tumor cells, interestingly, the opposite phenomenon was true (Fig. 1H). From the expression perspective alone, VISTA had already shown great differences from other co-inhibitory molecules. More detailed information about the ligands was available in the Supplementary Fig. 1B and Supplementary Table 1.
Correlation Analysis of Co-inhibitory Molecules
As suggested by the above analysis, there is a relationship between the expression of co-inhibitory genes, such as CTLA4 and PD-1, in different types of tumors, which prompted us to think about the correlation between them. In fact, co-inhibitory molecules are not insulated from each other. Evidence has suggested that the blockade of one co-inhibitory molecule probably alters the sensitivity of another to its antibody, thus influencing the effect of treatment [13, 95-97]. Hence, these published studies also indicate that it is meaningful to study the correlation between co-inhibitory molecules. As we can see from Fig. 2A, CTLA4 was significantly correlated with TIGIT (R = 0.84), PD-1 (R = 0.71), CD80 (R = 0.70), and LAG3 (R = 0.69). PD-1 was significantly correlated with TIGIT (R = 0.81), and LAG3 (R = 0.77). TIGIT had a clear correlation with LAG3 (R = 0.76), CD80 (R = 0.72), and PD-L2 (R = 0.66). TIM3 was clearly correlated with CD86 (R = 0.84), and PD-L2 (R = 0.61). CD80 was obviously correlated with CD86 (R = 0.70), and PD-L2 (R = 0.68). CD86 had a dramatically positive correlation with PD-L2 (R = 0.72), LGALS9 (R = 0.61). PD-L1 was clearly correlated with PD-L2 (R = 0.69) (all P < 0.05). These correlations are consistent with the results of the previously mentioned analyses and the available literature. However, the above analysis is based on samples from the whole TCGA data, and according to the literature, sometimes different types of tumors may exhibit different expression profile characteristics. To explore their correlations in individual tumor types, we further investigated the 10 combinations with the highest correlation coefficient. They showed a high correlation in most cancer types (Fig. 2B - K). Nonetheless, interestingly, these combinations of PD-1 and CTLA4 (Fig. 2B), PD-1 and LAG3 (Fig. 2C), PD-1 and TIGIT (Fig. 2F), PD-L2 and CD86 (Fig. 2K) all showed an exceptionally low correlation in THYM (close to or even below zero), with the correlation of TIGIT and LAG3 as well as CD80 and CD86 showing a similar phenomenon in LGG and AML, respectively (Fig. 2E, J). Notably, the combination of anti-LAG3 and anti-PD-1 was recently approved by the FDA for the treatment of patients with unresectable or metastatic melanoma, as also significantly noted in our analysis (Fig. 2C) [98].
Survival Analysis of Co-inhibitory Molecules
Since the result of the effect of coinhibitory molecules on various signaling pathways is in the patients’ survival time, the relationship between the expression levels of some co-inhibitory molecules and the survival of tumor patients has been widely and intensively studied to help find reliable targets. However, not every gene has been studied in depth, and at the same time, there is a lack of systematic survival analysis for all co-inhibitory molecule genes. To better understand their prognostic value, we next investigated the relationships between co-inhibitory molecules expression and cancer prognosis. Based on COX analysis, we first explored the overall implications of OS (Fig. 3A) and PFI (Fig. 3B) using the data from TCGA. Although somewhat complicated, we could obtain prognostic indicators for each gene from this analysis and make comparative observations between them. CTLA4 was indicated as a prognostic factor both for OS and PFI in 8 cancers (5 positive vs. 3 negative), LAG3 (2 positive vs. 4 negative), NRP1 (0 positive vs. 6 negative), PD-1 (5 positive vs. 4 negative), TIGIT (4 positive vs. 4 negative), TIM3 (2 positive vs. 2 negative), VISTA (6 positive vs. 1 negative), CD80 (2 positive vs. 4 negative), CD86 (2 positive vs. 2 negative), CD112 (1 positive vs. 3 negative), CD155 (0 positive vs. 5 negative), FGL1 (1 positive vs. 2 negative), LGALS9 (4 positive vs. 3 negative), PD-L1 (2 positive vs. 2 negative), PD-L2 (1 positive vs. 1 negative), SEMA4A (7 positive vs. 1 negative), VEGFA (1 positive vs. 4 negative). Notably, the majority of co-inhibitory molecules were threatening to the prognosis of KIRC, KIRP, LGG, THYM, and UVM patient. All data can be found in Supplementary Tables 2, 3.
The above results were obtained from downloaded data for local COX analyses. To verify the accuracy and to further validate these results, we performed KM analyses based on the widely recognized online tool Xena. Based on hazard ratio (HR) values above, we show the KM diagram with significant logRank P values, indicating that our COX and KM analysis are consistent. Taking VISTA as an example, we were unexpected to find that VISTA usually played a positive role in tumor cells, such as ACC (OS: P-value = 0.051; DSS: P-value = 0.025; PFI: P-value < 0.001), CESC (OS: P-value = 0.046; DSS: P-value = 0.044; PFI: P-value = 0.046), MESO (OS: P-value < 0.001; DSS: P-value = 0.009; PFI: P-value = 0.089), SARC (OS: P-value = 0.002; DSS: P-value = 0.007; PFI: P-value = 0.026), SKCM (OS: P-value < 0.001; DSS: P-value <0.001; PFI: P-value = 0.001), which was consistent with the literature [99]. Nonetheless, raised VISTA expression had negative impacts on survival in THYM (OS: P-value = 0.008; DSS: P-value = 0.210; PFI: P-value = 0.143), and UVM (OS: P-value < 0.001; DSS: P-value < 0.001; PFI: P-value = 0.014) (Fig. 3C). The result suggested that the role of VISTA was heterogeneous in prognosis across different cancer types and it might not be suitable for a therapeutic target in some tumors.
Relationship Between Co-inhibitory Molecules and Immune Cell Infiltration
The intensive research on immunotherapy in recent years has made us realize the importance of TME, especially the infiltration of immune cells. Among these, the CD8+ T cells are gaining increasingly popularity due to their anti-tumor effect and potential to function as a favorable prognostic biomarker in solid tumors [100-103]. If immune cells can recognize and cluster around tumor cells, why can’t they effectively stop the development and progression of cancer? Co-inhibitory molecules must be important participants in this process. Therefore, we hypothesized that effective blocking points could be selected based on the strength of the correlation between expression level and infiltration. Overview of the relationship between co-inhibitory molecules and CD8+ T cells infiltration was shown in Fig. 4A (As for other immune cells, the corresponding information can be seen in Supplementary Fig. 2). As we can see in most cancer types, IRs were almost positively correlated with CD8+ T cell infiltration. On the other hand, in terms of ligands, PD-L1, PD-L2, CD80, CD86 and LGALS9 were positively associated with CD8+ T cell infiltration, whereas PVR, PVRL2, VEGFA generally had negative correlation with it. What caught our attention was that in THCA as well as THYM, the expression of co-inhibitory molecules was usually associated with poor CD8+ T cell infiltration. To further validate the above results, we used different algorithms, such as TIMER, CIBERSORT, QUANTISEQ and MCPCOUNTER, to assess T-cell infiltration and the analyses showed relatively high agreement between these algorithms (Fig. 4B).
Thus, a new question was raised. Since the infiltration of CD8+ T cells is related to co-inhibitory molecules, how about the contribution of other immune cells? Although T cells are the most important immune infiltrating cells in tumor immunity, other cells, such as DCs, are also thought to be involved in the immune regulation and tumor clearance of TME. We moved on to investigate the association between these molecules and B cell, CD4+ T cell, Neutrophil, Macrophage, Myeloid dendritic cell infiltration (Supplementary Tables 4 and 5 displayed the correlation of 18 co-inhibitory molecules with various immune cells from TIMER and CIBERSORT, respectively). Here, we presented the correlation plot between VISTA and six types of immune cells (Fig. 4C). Overall, VISTA showed a positive correlation with immune cell infiltration in various cancers, while THYM was the obvious exception. Given all this, the role of VISTA is more elusive and worth further exploring.
Correlation Analysis of VISTA and TMB, MSI in Pan-Cancer
In addition, we wondered what factors may affect the expression of co-inhibitory molecules and the efficacy of immunotherapy. Although there is no comprehensive evidence that TMB or MSI is associated with every co-inhibitory molecule, they are both important factors in tumor immunotherapy [78-82] and can even serve as emerging biomarkers associated with the sensitivity to ICIs such as PD-1/PD-L1 mAbs [81, 104, 105]. Hence, we were very interested in whether the expression of these genes was associated with TMB or MSI. Since VISTA was observed to have important effects on the tumor prognosis and TME, here we focused on the correlation between VISTA and TMB as well as MSI to further explore the role of VISTA in tumors (Relevant information about other genes was presented in the Supplementary Table 6). Our results indicated that the higher the expression of VISTA, the lower TMB in ACC (r = -0.225), DLBC (r = -0.499), LIHC (r = -0.160), PCPG (r = -0.151), PRAD (r = -0.179), READ (r = -0.203), STAD (r = -0.206), THCA (r = -0.172), while the higher TMB in COAD (r = 0.132), KIRC (r = 0.145) (Fig. 5A). With respect to MSI, VISTA had a positive association with it in COAD (r = 0.233), whereas a negative correlation could be seen in CHOL (r = -0.344) LUSC (r = -0.195), MESO (r = -0.295) STAD (r = -0.236), TGCT (r = -0.235) (Fig. 5B). Considering the difference of physiological indexes of different patients, the levels of TMB and MSI can be detected before choosing inhibitors to predict the feasibility of immunotherapy.
Correlation Analysis of VISTA and Cancer Cell Stemness
The progression of cancer is accompanied by a gradual loss of differentiation phenotypes as well as the increasingly pronounced stem cell properties which can be used as a predictive biomarker for tumor prognosis [83, 84, 106]. One of the major difficulties in cancer treatment lies in its heterogeneity, and even with the successful removal of numerous cancer cells, a small number of cancer stem cells are sufficient to form new tumors. Because the heterogeneity is mostly induced by abnormal cell differentiation and stem cell signaling, understanding the impact of cancer cell stemness will greatly improve the clinical treatment and help to predict the tumor prognosis. The direct assessment of tumor stemness is a little complicated, so we evaluated this feature indirectly by scoring each patient based on the literature. In short, two independent stemness indexes, mDNAsi which indicates epigenetic features and mRNAsi which reflected gene expression, were obtained by a multiplatform analysis of the stem cell transcriptome, methylome, and transcription factor binding sites. The closer the mDNAsi or mRNAsi index is to 1, the stronger the degree of cancer cell stemness, which also means that the tumor cells are less differentiated. As we can see, the expression of VISTA was positively correlated with the mDNAsi in ACC (r = 0.248), PRAD (r = 0.216), THCA (r = 0.133), THYM (r = 0.363), and UVM (r = 0.272), while it had negative correlation with the mDNAsi in 17 cancers (Fig. 6A). We also found that the expression of VISTA was dramatically negatively correlated with mRNAsi in 27 cancers. However, there was no significantly positive association between VISTA and mRNAsi in any type of cancer (Fig. 6B). We therefore speculated that this result might serve as one of the reasons why VISTA is associated with a good prognosis and also increase the reliability of VISTA as a tumor biomarker.
Correlations Between VISTA and DNA/RNA Methylation Regulatory Genes
DNA and RNA methylation are strongly implicated in tumorigenesis and have guiding implications for the development of powerful diagnostic [87, 107]. Unlike TMB, stemness, etc., there is no scoring algorithm for methylation and in fact it is difficult to quantify the degree of methylation scored for each patient. However, the genes responsible for methylation in cells are well defined, such as DNA methylation regulatory genes (typically 5mC: DNMT1, DNMT3A, DNMT3B, DNMT3L) [86] or RNA methylation (typically m6A: ALKBH5, FTO, METTL3, METTL14, WTAP) [87], and we can roughly estimate the degree of methylation in tumors by assessing the expression of these genes. The relationships between co-inhibitory molecule genes and DNA/RNA methylation regulatory genes were presented in the Supplementary Tables 7-8. Here, again, we covered VISTA in detail to deepen our understanding of it. Our results indicated that VISTA and DNA/RNA methylation genes were closely related. VISTA had a significant association with DNMT1 (2 positive vs. 16 negative), DNMT3A (2 positive vs. 18 negative), DNMT3B (2 positive vs. 26 negative), DNMT3L (2 positive vs. 8 negative), ALKBH5 (7 positive vs. 6 negative), FTO (6 positive vs. 7 negative), METTL3 (0 positive vs. 19 negative), METTL4 (0 positive vs. 23 negative), and WTAP (13 positive vs. 7 negative). The specific correlation values were shown in Fig. 5C, D, respectively. Our observation that VISTA had a general negative correlation with these genes contributed to the hypothesis that it could influence tumor prognosis by altering the methylation levels.
Correlation Analysis Between VISTA Expression and Drug Sensitivity
Current strategies targeting co-inhibitory molecules, such as PD1, relies mainly on monoclonal antibodies, with some studies attempting to develop small molecule inhibitors. Since tumor cells can interact with their environment in complicated manners and contribute to an immune suppressive environment that fights against anticancer drugs. ICIs combined with chemotherapy have become one of the research hotpots in the field of tumor immunotherapy to effectively improve response rates of cancer therapies [108]. Given that new use of old drugs is a very important development strategy and that the use of drugs in combination with ICIs is currently a crucial research direction, we sought to analyze the relationship between drug sensitivity and co-inhibitory molecules in the NCI60 database which had more than 20,000 drugs (Supplementary Table 9 and Supplementary Fig. 3A-R). Here, Fig. 8 illustrated that the expression of VISTA had a negative correlation with the sensitivity to Lxazomib citrate (r = -0.49), Vincristine (r = -0.48), Tepotinib (r = -0.47), MG−132 (r = -0.46), Tamoxifen (r = -0.45), SB−590885 (r = -0.43), Bortezomib (r = -0.43), Geldanamycin analog (r = -0.42). The higher the expression of VISTA, the stronger the drug sensitivity to EGF−816 (r = 0.38), Methylprednisolone (r = 0.39), BMS−690514 (r = 0.40), Sapitinib (r = 0.41), Erlotinib (r = 0.43), SW−044248 (r = 0.45), Dexamethasone (r = 0.46), Gefitinib (r = 0.59) (all P < 0.01). Among the drugs mentioned above, several have already been under clinical trials in combination therapy with ICIs (Table 1). Notably, Erlotinib, Dexamethasone and Gefitinib had shown prospective value in combination with PD-1/PD-L1 inhibitors in cancer treatment [109-111]. As a result of this, it’s hopeful that the combination of VISTA and these drugs would also be significantly effective. However, there was no reports linking VISTA to these drugs by the time we finished our study. Relevant clinical research can be carried out based on our analysis.
Clinical Trials on VISTA
Immunotherapies have enriched the types of available cancer treatments, with the current clinically approved ICIs focusing on PD-1, CTLA4 and PD-L1. Although they offer hope for a cure for many cancer patients, they have many limitations, such as restricted indications for tumor types and only a fraction of patients responding effectively to these agents [50]. Novel strategies targeting alternative co-inhibitory molecules including LAG3, TIM3, etc. are therefore proposed [52]. Among these, VISTA has been one of the most recently raised inhibitory molecules as a potential target. A growing number of preclinical trials have shown that blocking VISTA enhances the fraction, proliferation, and function of tumor-infiltrating T cells, thus rescuing the TME from the inhibitory state [39]. From the results of our analysis and the available clinical studies, it is clear that VISTA has important implications, but there is no comprehensive summary of clinical studies for it. To better understand the potential application of VISTA, we summarized currently undergoing clinical trials in Table 2. In a Phase I clinical trial, the safety and efficacy of JNJ-61610588, a fully human IgG1 anti-VISTA mAb, was evaluated in patients with advanced cancer. CA-170, an oral inhibitor targeting both PD-L1/L2 and VISTA, has shown clinical efficacy in phase I and II clinical trials in different advanced solid tumor types [112]. Another candidate, CI-8993, is under dose study based on its safety. Since VISTA and PD-1/PD-L1 pathways were proved to have different mechanisms in controlling T cell activation, co-blocking these two signaling pathways shows great prospects in anti-tumor therapy [113], the HMBD-002 [114], a novel anti-VISTA mAb, is being evaluated combined with Pembrolizumab.