Identification of immune microenvironment subtypes that predicted the prognosis of patients with ovarian cancer

Abstract Ovarian cancer (OC) is associated with high mortality rate. However, the correlation between immune microenvironment and prognosis of OC remains unclear. This study aimed to explore prognostic significance of OC tumour microenvironment. The OC data set was selected from the cancer genome atlas (TCGA), and 307 samples were collected. Hierarchical clustering was performed according to the expression of 756 genes. The immune and matrix scores of all immune subtypes were determined, and Kruskal‐Wallis test was used to analyse the differences in the immune and matrix scores between OC samples with different immune subtypes. The model for predicting prognosis was constructed based on the expression of immune‐related genes. TIDE platform was applied to predict the effect of immunotherapy on patients with OC of different immune subtypes. The 307 OC samples were classified into three immune subtypes A‐C. Patients in subtype B had poorer prognosis and lower survival rate. The infiltration of helper T cells and macrophages in microenvironment indicated significant differences between immune subtypes. Enrichment analyses of immune cell molecular pathways showed that JAK–STAT3 pathway changed significantly in subtype B. Furthermore, predictive response to immunotherapy in subtype B was significantly higher than that in subtype A and C. Immune subtyping can be used as an independent predictor of the prognosis of OC patients, which may be related to the infiltration patterns of immune cells in tumour microenvironment. In addition, patients in immune subtype B have superior response to immunotherapy, suggesting that patients in subtype B are suitable for immunotherapy.


cells. Immune cells release immunoregulatory factors and interact
with each other to influence the progression, prognosis and treatment responsiveness of patients with OC. [6][7][8] Studies have reported that OC cells can regulate the differentiation of macrophages into M2 tumour-associated macrophages in the microenvironment. 9 Tumour-associated macrophages can further mediate an imbalance in the T-cell differentiation ratio in the microenvironment. 10 M2 tumour-associated macrophages can promote peritoneal metastasis of OC by accelerating angiogenesis. 11 However, the correlation between immune microenvironment and prognosis of OC remains unclear. Therefore, this study aimed to explore prognostic significance of OC tumour microenvironment.
We divided OC into three immune subtypes on the basis of clinical features according to the expression of immune-related genes.
We analysed immune subtypes of 307 samples from the cancer genome atlas (TCGA) database and 93 samples from the international cancer genome consortium (ICGC) database, OC patients with high immune and matrix scores in subtype B exhibited poor prognosis.
Additionally, the infiltration of M1-type macrophages and helper T cells decreased, whereas the infiltration of Treg-type macrophages and M2-type macrophages increased. The enrichment fraction of the IL-6-JAK-STAT3 molecular pathway increased in the immune subtype samples of subtype B. However, the enrichment fraction of the KRAS-down-regulation pathway increased in A and C immune subtype samples with superior prognosis.

| Data acquisition
OC data set was selected from the TCGA, and the samples lacking survival data were excluded. Finally, the expression of 20 188 genes in 307 ovarian serous adenocarcinoma samples was analysed.
The ICGC data set selected in the present study was transformed according to the GeneSymbol. The expression of the same GeneSymbol was combined, whereas the expression of 43 638 genes was obtained from 93 OC samples (http://DCC.icgc.org/relea ses/curre nt/Proje cts/OV-AU).

| Hierarchical clustering
The distance between samples was calculated using hierarchical clustering. The nearest points were merged into the same class each time. Then, the distance between classes was calculated, and the nearest classes were merged into one large class until a single class was synthesized. The distance between classes was calculated using methods such as single linkage, complete linkage, average linkage and unweighted pair group method with arithmetic mean (UPGMA).
The complete linkage was chosen as the calculation method, the calculated distance was Chebyshev distance, and the clustering algorithm constructed in the pheatmap package in R was used.

| Survival analyses
Survival analyses were performed using survival and survminer packages in R and further visualized using the Kaplan-Meier curve.

| Immune score and matrix score
The immune and matrix scores of each OC sample were calculated by ESTIMATE (R package: estimate).

| Evaluation of immune cell infiltration
CIBERSORT was used to normalize gene expression data to infer the infiltration ratio of 22 types of immune cells. To evaluate the reliability of deconvolution, CIBERSORT was used to calculate the P value for each sample, and the correlation analysis was performed after screening based on a P value of < 0.05.

| Gene set variation analysis
Gene set variation analysis (GSVA) is a non-parametric and unsupervised algorithm used to calculate the enrichment scores of specific gene sets in each sample without grouping samples in advance.
GSVA transforms gene expression from an expression matrix characterized by a single gene to an expression matrix characterized by a specific gene set. GSVA was used to calculate the enrichment scores of different gene sets in each sample of TCGA, and correlation analyses were performed.

| OC immunotyping
A total of 756 genes were filtered from 770 immune-related genes in the TCGA database. Based on the expressions of 756 genes, hierarchical clustering analyses were performed on 307 genes in the TCGA database (complete, longest distance method, and Maximum and Chebyshev distance). Three immune subtypes of OC (clusters A, B and C) were obtained, and the expression heat map ( Figure 1A) was constructed according to clinical phenotype.
The estimate in R package was used to calculate the immune and matrix scores of each sample in the TCGA data set, and the immune score ( Figure 1B) and matrix score ( Figure 1C) were compared between the three subtypes of OC samples with different immune subtypes. Comparison of immune scores (Figure S1A-C) and matrix scores ( Figure S1D-F) in pairwise groups indicated that the immune and matrix scores of OC samples with immune subtypes in subtype B were significantly higher than those in subtype A and C; however, no significant difference was observed in the scores between subtype A and C. Therefore, subtype A + C was compared with subtype B ( Figure 1D,E), and subtype B (n = 117) had significantly higher immune and matrix scores than subtype A + C (n = 191).

| Immunotyping was related to the prognosis of patients with OC
Kaplan-Meier analysis was used to determine the correlation of immunophenotyping with survival and prognosis of patients with OC.
The overall survival of patients in subtype B was significantly shorter than that of patients in subtype A and C (

| Immunotyping and immune cell infiltration
CIBERSORT was adopted to compare the correlation between lymphoid and myeloid infiltration. The infiltration ratio of 22 immune cells was calculated, and the differences in different immune subtypes were compared ( Figure 4A). Then, a cluster diagram was plot-

| Analyses of immunosubtype-related molecular pathways and immunotherapy reactivity
To further analyse the phenotype related to poor prognosis of patients with immune subtype B, GSVA was used to calculate the average gene set enrichment score of the gene set in hallmark, and a cluster map was constructed according to the grouping ( Figure 6A). subtype A + C. The missense mutation was the main type of mutation ( Figure S2A,B). P53 exhibited the highest mutation rate of 88% ( Figure S2C).
The immunotherapy effect was predicted using TIDE (http://tide. dfci.harva rd.edu/) ( Figure 7C), and the predicted immunotherapy effect in patients with immune subtypes in subtype B was superior.  15 Some studies have suggested that helper T cells F I G U R E 5 Immune cells infiltration was associated with prognosis. According to infiltration value of different immune cells, they were divided into two groups (high and low group), and the correlation between the infiltration of different immune cells and the survival and prognosis of patients was analysed F I G U R E 6 Immune clusters and phenotypes in ovarian cancer. (A) Using GSVA, the average gene set enrichment score was calculated in hallmark, and the score heat map was drawn based on different immune subtypes. (B) A generalized linear model was established, and a forest map was drawn. (C) Based on the infiltration of immune cells, the correlation between the enrichment fraction of gene set and the infiltration of immune cells was analysed have significant anti-tumour immunity, and the activation of helper T cells may be a novel method for improving tumour immunotherapy. 16 In this study, we found that the infiltration ratio of M2 macrophages and Treg cells in B immune subtype samples exhibited an increasing trend. However, the difference was not significant, which may be due to the limited sample size. The tumour-promoting function of M2 macrophages has been widely reported. 17 Monocytes or macrophages in OC microenvironment can polarize to M2 macrophages, and M2 macrophages can further shape immune microenvironment that enhances OC progression. 10,18 Therefore, OC immunotherapy focuses on facilitating the transformation of M2type macrophages to M1-type macrophages. 19 Treg cells are considered to be immunosuppressive, and increased infiltration of Treg cells has been found to be generally associated with poor prognosis of patients with OC. 20,21 The enrichment analysis of immune subtype-related molecular pathway indicated an increase in the enrichment fraction of IL-6-JAK-STAT3 molecular pathway in B subtype, which was highly correlated with M2 macrophage infiltration. IL-6-JAK-STAT pathway plays a significant role in the regulation of tumour immune microenvironment, 22 and its activation is related to drug resistance of OC. 23 A recent single-cell sequencing study on high-grade OC indicated that JAK-STAT pathway was abnormally activated in OC cells and small molecule inhibitors of this pathway are promising candidates for clinical application. 24 Pathway analyses indicated that the enrichment scores of KRAS pathway increased in the immune subtype samples of subtype A and C with better prognosis. KRAS gene mutation and the activation of downstream pathway are significant molecular events in OC progression, 25 which are significantly related to poor prognosis in patients with OC. 26 Finally, we analysed the correlation between immune subtypes and immunotherapy. TIDE was used to predict the response of OC samples with different immune subtypes to immunotherapy. The results indicated that patients with immune subtype B with poor prognosis had superior response to immunotherapy. Since immunotherapy has not been widely developed in ovarian cancer, the F I G U R E 7 Gene mutations and immunotherapy effect of immune clusters. (A) According to the grouping of immune subtypes, CNV mutations of two groups of samples were analysed using GenePattern GISTIC_2.0. (B) The mutation information of OV data set was downloaded from GDC to screen 307 samples, the top 20 genes in overall mutation rate were selected, and the mutation information of samples was plotted by using R package-maftools. (C) Using TIDE, the immunotherapy effect of patients with various immune subtypes was predicted patients' response to immunotherapy was predicted by TIDE analysis, which is a limitation of our research. It needs further study to confirm the correlation between immune subtypes and the response to immunotherapy in ovarian cancer patients received immunotherapy.

| D ISCUSS I ON
In conclusion, our study revealed that tumour immune subtypes can be used as an independent predictor for OC prognosis. Special tumour immune subtype might guide clinical treatment decision and improve the immunotherapy responsiveness of OC patients, thus providing a novel direction for the development of effective immunotherapy strategies.

CO N FLI C T O F I NTE R E S T
The authors confirm that there are no conflicts of interest.