Introduction
Hepatocellular carcinoma (HCC), the main histology type (70–90%) of liver cancer, ranks sixth in cancer incidence and fourth in death. In recent years, the incidence of HCC has increased in most regions of the world and decreased in some countries in Asia [
1,
2]. Currently, the main treatment for HCC patients in early stages is surgery, combination with transarterial chemoembolization, ablation and liver transplantation [
3]. For others in advanced stages, the effective approaches involve molecular targeting agents (tyrosine kinase inhibitors: sorafenib, lenvatinib and regorafenib) [
4], and huaier granule, a traditional Chinese medicine [
5]. Although these methods have improved the prognosis of HCC patients, the overall survival of HCC remains challenging for the heterogeneity of HCC [
1,
6].
Immunotherapy has been a growing focus because of its effectiveness in many tumors [
7,
8], Particularly, targeted therapies for immune checkpoints such as anti-CTLA4 and anti-PD1 have benefited a part of patients with solid tumors [
9‐
12], although not all patients show the response to immunotherapy [
13]. Even though clinical trials are under way, the future of immunotherapy in HCC is uncertain. In chronic hepatitis caused by viral infection (HBV, HCV), alcoholism, metabolic diseases (non-fatty liver disease), and drug damage (aflatoxin, aristolochic acid), changes in the liver microenvironment and imbalance in the proportion of immune cells eventually lead to immune escape and the promotion of HCC [
14,
15]. The role of different components (including tumor-associated macrophages (TAM), myeloid-derived suppressor cells (MDSC), regulatory T cells (Tregs), CD8+ cytotoxic T lymphocytes, fibroblasts) of tumor microenvironment (TME) in hepatocellular carcinoma has also been discussed in many studies [
16]. Determining TME of patients with cancers before treatment can demonstrate the immune status to predict the prognosis of patients and the response to chemotherapy and immunotherapy drugs [
17,
18].
To understand the immune microenvironment of HCC, some researches have done to investigate the immune subtypes of HCC [
19]. However, there are lack of multi-omics (mRNA, miRNA, long non-coding RNA (LncRNA), somatic mutation, DNA methylation, copy number variations and reverse phase protein array (RPPA)) studying focusing on immune microenvironment characteristics and immunotherapy strategies of HCC.
In this study, we used two computational algorithms to estimate the abundance of 26 TME cells of 1000 HCC samples and performed three clustering methods to confirm 3 clusters, the optimal number of clusters. Then we recognized the differences of immunes cell abundance, immune gene expression, genomic characteristics, molecular and biological function, and clinical outcomes among the three subtypes of HCC. Finally, we developed a support vector machine (SVM) classifier using multi-omics signatures to identify patients with significant prognostic differences and different responses to immunotherapy in HCC, and preliminarily demonstrated that the expression of MMP9 could predict the immune characteristics of HCC.
Discussion
Immunotherapy has turned into one of the most promising treatments in cancer [
7,
8,
13]. However, the future of immune checkpoint therapy in liver cancer still remains unclear, with the failure of phase III clinical trial [
52] (anti-PD-1) despite a small proportion (15%) of response to PD-1 inhibitor in phase II clinical trial [
53]. One of the keys of immunotherapy to HCC is deeply understanding the immunological characteristics of liver cancer. In this study, we recognized the TME in HCC including differences of immune microenvironment between carcinoma and adjacent tissues, clinical, molecular and genomic characteristics in HCC immune subtypes, and signatures helping identify patients with high T cell infiltration but T cell dysfunction and higher response to immune checkpoint therapy from TCGA-LIHC and other independent cohorts.
Our work showed that most TME cells varied between tumor and adjacent non-tumor tissues. However, few researches focused on the infiltration alterations of TME cells, and we inferred that changes of immune cell composition might initiate the occurrence of tumors during the process of immune surveillance, especially in HBV and HCV related HCC. And also, the roles of TME cells in the diagnosis and prediction of liver cancer are rarely reported. This result could provide more new sights to seek the mechanisms of HCC initiation. And this research was dedicated to exploring the heterogeneity of immune subtypes of HCC based on large public datasets. There existed 3 clusters in HCC by unsupervised learning, with distinct immune cell abundance and different from the discovery of Yutaka et al. [
54] that HCC patients could be purely classified as high, middle and low immune cell infiltration. The main reasons for the deviations in analysis may be the differences in immune cell estimation and classification methods, which were realized based on unsupervised clustering of machine learning in this research. Despite the distribution diversities of TME cells among the three clusters, we found cluster1 had more mature adaptive immune cells such as CD8 T cells and cytotoxic lymphocytes, which used to be considered an indicator of improved survival for cancer patients [
30,
55]. Also, this conclusion was also validated in our study that high infiltration of CD8 T cells and cytotoxic lymphocytes predicted better outcome both in the whole sample and in each subtype of HCC by univariate Cox regression (Fig.
2a, b). However, cluster1 had poor survival, which may be caused by high expression of some classic or newly discovered immune checkpoints (PD1, CTLA4 and TIM-3), more infiltration of immunosuppressive cells (TAMs, Tregs and Th17 cells), and genomic alterations (TP53 mutation and deletion). Meanwhile, there are some anti-tumor characteristics in cluster1, such as high lymphocyte infiltration of CD8 T cell and lymphocyte, high IFN-γ response, and high expression of immune co-stimulators, which indicated that there may be more immunotherapeutic responses [
40,
56]. Furthermore, we also found that cluster1 might be more suitable for immunotherapy, which is consistent with the high expression of some immunocheckpoints and makes up for the lack of differences in the expression of classic checkpoints such as PD-L1. However, this conclusion will need to be confirmed by clinical trials in the future.
What’s more, it is worth reminding that our results did not contradict previous findings that high infiltration of CD8 T cells indicated beneficial prognosis, but extended and enriched this conclusion. We demonstrated that there is a group of HCC patients with higher CD8 T cell infiltration, but T cell dysfunction and increased immune escape, resulting in a poor prognosis, which was consistent with discoveries in other tumors [
43,
57]. These results, to some extent, explain the unsatisfactory situation of immunotherapeutic response in HCC [
52,
53,
58], which are in concordance with our results. It further indicates that some patients with increased T cell infiltration are more likely to receive immunotherapy, but show not very high responsiveness for the increased immune escape and T cell dysfunction.
To recognize and immune subtype and predict the response of immunotherapy in HCC, multi-omics signatures were obtained. Using SVM classifier, we regrouped HCC patients into two groups and Type A showed similar characteristics (clinical outcomes, CD8 T cell, T cell dysfunction and response to immunotherapy) with cluster1 in both TCGA cohort and external validation cohorts. Due to the limited responsiveness of targeting immune checkpoint therapy, only a small part of patients show response. The construction of multi-omics SVM model in our study could maximally predict the benefit of immunotherapy in patients with HCC. Additionally, the SVM model could also predict prognosis of several other cancers, suggesting that these tumors might have similar or opposite immune mechanisms to HCC, which was consistent with previous studies that READ, LUSC and BRCA belonged to C1 and C2, while LIHC, PAAD, ACC, LGG belonged to C3,C4 and C5 type [
59‐
61]. These results imply the multi-omics signatures could provide new clues to investigate the TME of HCC or other tumors in future.
Furthermore, we preliminarily verified the immune role of MMP9, a secreted protein produced by TAMs [
47], in HCC through immunohistochemical experiments. High expression of MMP9 indicated higher levels of PD1, CTLA4 and CD8A and poor survival in partial HCC patients, which was in line with our above analysis that some HCC patients with high CD8 T cell infiltration but dysfunction were immunosuppressed. MMP9 can affect the immune state through a variety of ways, such as releasing VEGF to promote angiogenesis [
62] and binding CD44 to release TGFβ [
63]. Furthermore, MMP9 can be used as a biomarker of chemotherapy response, with high expression of MMP9 meaning better responsiveness to chemotherapy [
64]. Perhaps, MMP9 could be a good indicator of T cell dysfunction and immunotherapy responsiveness. In addition to anti-PD1/CTLA4 immune checkpoint therapy, our study suggests that the combination of anti-checkpoint with anti-MMP9 [
51] or anti-TAMs [
65] may be more beneficial to patients with T cell dysfunction in HCC. However, due to the lack of large sequenced HCC cohort and prospective clinical trials that have received immunotherapy, the effect of MMP9 expression on the efficacy of immunotherapy in HCC patients remains concerned.
The advantage of our study is that we used a large number of publicly available independent data sets (from TCGA, ICGC and GEO) and our own cohort, applying different research methods (genomics, transcriptomics, IHC and so on) to detect problems and verify them, which makes the conclusions consistent and reproducible. However, our study is still limited in that the data sources we used were all retrospective. In addition, most of our conclusions are based on bioinformatics analysis. Although multiple datasets from different sources show feasibility, it still needs further experimental verification and application. And the intriguing perspectives and conjectures on our multi-omics SVM model and the immunological role of MMP9 in this study need to be further verified in future prospective clinical trials and molecular biology researches.
Overall, in this research, comprehensive analysis and assessment of TME patterns based on multi-omics in HCC can provide some new strategies about response to immunotherapy, and the combination of targeting drugs.
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.