Construction of an aging-related risk signature in high-grade serous ovarian cancer for predicting survival outcome and immunogenicity

Abstract

Studies have shown that aging significantly impacts tumorigenesis, survival outcome, and treatment efficacy in various tumors, covering high-grade serous ovarian cancer (HGSOC). Therefore, the objective for this investigation is to construct an aging-relevant risk signature for the first time, which will help evaluate the immunogenicity and survival status for patients with HGSOC. Totaling 1727 patients with HGSOC, along with their mRNA genomic data and clinical survival data, were obtained based on 5 independent cohorts. The Lasso-Cox regression model was utilized to identify the aging genes that had the most significant impact on prognosis. The risk signature was developed by integrating the determined gene expression and accordant model weights. Additionally, immunocytes in the microenvironment, signaling pathways, and immune-relevant signatures were assessed based on distinct risk subgroups. Finally, 2 cohorts that underwent treatment with immune checkpoint inhibitor (ICI) were employed to confirm the effects of identified risk signature on ICI efficacy. An aging signature was constructed from 12 relevant genes, which showed improved survival outcomes in low-risk HGSOC patients across discovery and 4 validation cohorts (all P < .05). The low-risk subgroup showed better immunocyte infiltration and higher enrichment of immune pathways and ICI predictors based on further immunology analysis. Notably, in the immunotherapeutic cohorts, low-risk aging signature was observed to link to better immunotherapeutic outcomes and increased response rates. Together, our constructed signature of aging has the potential to assess not only the prognosis outcome and immunogenicity, but also, importantly, the efficacy of ICI treatment. This signature provides valuable insights for prognosis prediction and immunotherapeutic effect evaluation, ultimately promoting individualized treatment for HGSOC patients.

1. Introduction

High-grade serous ovarian cancer (HGSOC) is a type of epithelial ovarian cancer that is featured by aggressive behavior, genomic instability, and high heterogeneity.^[1] HGSOC accounts for approximately 70% of all cases of epithelial ovarian cancer and always occurs at an advanced clinical stage, despite initial response to standard treatment regimens.^[2] The disease is associated with significant morbidity and mortality, making it a major healthcare issue.^[3] Significant efforts are being made to improve the understanding of the underlying molecular mechanisms driving HGSOC development and progression, as well as to identify novel biomarkers and therapeutic targets for the disease.

Immune checkpoint inhibitors (ICIs) are a class of drugs that modulate the immune system by blocking negative regulatory pathways, thus enhancing the intrinsic capacity of immunological regulators to discern tumor cells.^[4] These drugs work by inhibiting certain proteins, known as immune checkpoints, that help cancer cells evade detection by the immune system.^[5] By blocking these checkpoints, ICIs can unleash better immunity, resulting in improved survival outcomes for patients with various types of cancer. The development of ICIs has revolutionized cancer treatment, and they have become a cornerstone of immunotherapy.^[6] Despite their promising activity, not all patients respond to ICIs, and there is still much to explore about the mechanisms underlying their efficacy and resistance.^[7] Ongoing research efforts are focused on identifying predictive biomarkers, improving treatment strategies, and developing novel combination therapies to further enhance the clinical benefit of ICIs.

The process of aging is widely recognized as a significant regulator for many diseases, including human cancer.^[8] While aging itself is a contributing factor, several findings have revealed that genes with respect to aging, such as APOE^[9] and FOXO3,^[9,10] and genome regions such as 5q33.3, were connected with longevity. However, since the complex and multifaceted interactions between various factors, such as the genome, environment, and aging-related diseases,^[11] investigating the aging process remains a challenging endeavor. To shed light on this problem, Peters et al performed an extensive population-based transcriptomic analysis to identify genes that are implicated in the aging process.^[11]

In this investigation, we meticulously curated a comprehensive dataset comprising 1727 patients of HGSOC derived from 5 independent cohorts obtained from publicly available data sources to develop and corroborate an aging prognosis model. Subsequently, we carried out multi-level immune mechanistic exploration, which provided compelling evidence for the exceptional ability of aging signature to evaluate immunological traits. Moreover, based on analyses of immunotherapeutic transcriptomic data, this investigation observed this aging signature displayed remarkable predictive power with respect to ICI treatment efficacy. The outcomes of our research may offer crucial insights into the clinical outcome assessment and immunotherapeutic effect prediction for individuals afflicted with HGSOC.

2. Materials and methods

2.1. HGSOC samples, immunotherapeutic samples, and aging-related genes

In order to acquire eligible HGSOC cases with both mRNA transcriptomic data and prognosis information, an extensive search was conducted for the Gene Expression Omnibus and the Cancer Genome Atlas (TCGA) databases. Samples lacking essential transcriptome data or follow-up characteristics were excluded from this study, ultimately resulting in a total of 1727 samples deemed suitable for inclusion criteria. These samples were acquired from various datasets, that is TCGA (N = 581), GSE13876 (N = 415), GSE9891 (N = 277), GSE32062 (N = 260), and GSE49997 (N = 194). Given that TCGA possessed the largest sample size of 581 patients among the included HGSOC datasets, it was deemed the discovery cohort and utilized to construct the aging signature. A comprehensive cohort and transcriptomic platform information across included HGSOC cases were presented (see Table S1, http://links.lww.com/MD/J584, Supplemental Digital Content, which shows the summary of TCGA and 4 validation HGSOC cohorts utilized in this work). To understand the immunotherapeutic implications derived from this determined aging signature, publicly available immunogenomic datasets were employed. Specifically, mRNA expression profiles and ICI treatment data collected from 348 samples with advanced urothelial cancer (UC) who were treated with anti-programmed death ligand-1 agents from the IMvigor210 cohort is the first immunotherapeutic validation cohort.^[12] Moreover, the second cohort comprising 121 melanoma cases that were treated with anti-programmed cell death 1/programmed death ligand-1 or integrated agents^[13] was collected to ulteriorly confirm ICI roles of such risk signature. Finally, a total of 1106 aging-related genes were acquired from Peters et al research.^[11] The ethics committee of The First Hospital of Lanzhou University approved the study.

2.2. Development of aging risk signature

Development of this aging risk signature involved a series of rigorous steps. Firstly, univariate Cox regression analyses were conducted to assess correlations between each of those 1106 aging-related genes and survival outcomes in the TCGA-HGSOC dataset. Secondly, we employed the LASSO Cox regression analysis (with R glmnet package^[14]) to lessen aging-related genes by removing those with low predictive power for prognosis. Thirdly, by integrating the remaining gene expression values with their regression weights, risk scores were evaluated using the method below: ${\sum }_i^{n}coefficientofgene\left(i\right)\ast Expressionofgene\left(i\right)$, where i represents the particular aging-related genes. The samples were divided into high- and low-risk groups based on the median risk score of the TCGA-HGSOC cohort to investigate the association with survival outcome. Finally, we corroborated the accuracy and robustness of this aging-related model through additional external verification in multiple datasets.

2.3. Tumor infiltration immunocytes and immune checkpoints

In order to shed light on the varying abundance of immunocyte infiltration between the low- and high-risk subgroups, we conducted an assessment of infiltration levels of 28 different phenotypes of immunocytes according to a recently reported method.^[15] These 28 immunocytes were subsequently classified into 3 discrete groups to facilitate analysis of their specific biological functions: anti-tumor, pro-tumor, and intermediate tumor immunocytes. Characteristic genes associated with these immunocyte subtypes were presented (see Table S2, http://links.lww.com/MD/J585, Supplemental Digital Content, which shows feature genes for 28 immunocyte subtypes). We then estimated the abundance of 22 infiltration immunocytes under the CIBERSORT algorithm,^[16] which is based on the default signature matrix (i.e., LM22 matrix) provided by the CIBERSORT website for deconvolution analysis. In the present investigation, we utilized both approaches in order to achieve a mutually validating outcome.

By using the data from a recent study,^[17] the comprehensive immune checkpoint genes were assembled. However, due to the utilization of diverse sequencing platforms, we did not observe the gene VISTA in the discovery cohort.

2.4. Tumor immunogenicity-related signatures

Several recently published molecular signatures have demonstrated associations with immunogenicity and response to ICI treatments. Consequently, we have curated 3 representative signatures, including: T cell-inflamed signature, comprised of 18 genes that are linked with establishing a sensitized immune microenvironment and have been connected to the effect of immunotherapy^[18]; cytolytic activity^[19]; and IFN gamma signature,^[18] which is linked with the T cell-inflamed signature. Feature genes of each immunogenicity signature were also collected (see Table S3, http://links.lww.com/MD/J586, Supplemental Digital Content, which shows specific feature genes for 3 representative immune-related signatures).

2.5. GSEA and ssGSEA

In order to ascertain the potential biological circuits behind 2 risk subpopulations, we have utilized gene set enrichment analysis (GSEA). The fold change values that were obtained from differential analyses between 2 subgroups, performed using the limma package,^[20] along with their corresponding genes, were employed as input variables for the GSEA function that was implemented using the R clusterProfiler package.^[21] Signaling pathways that were sourced from KEGG and GO BP databases were employed as the annotation pathways. Single sample GSEA (single sample gene set enrichment analysis) algorithm implemented by the GSVA package^[22] was utilized to assess enrichment levels of each immune signature and immunocyte subtype based on the characteristic genes.

3. Results

3.1. Construction of aging risk signature

Following the exclusion of samples lacking necessary transcriptomic expression profiles or prognosis data, a total of 1727 patients diagnosed with HGSOC were deemed suitable for subsequent analysis. This cohort consisted of datasets TCGA (N = 581), GSE13876 (N = 415), GSE9891 (N = 277), GSE32062 (N = 260), and GSE49997 (N = 194). Of these cohorts, the TCGA HGSOC subset, which contained the largest population size, was chosen as the discovery cohort. By employing univariate Cox regression analysis on collected 1106 aging genes based on mRNA expression data from the TCGA cohort, we found that 152 genes were prognostic (all P < .05; see Table S4, http://links.lww.com/MD/J587, Supplemental Digital Content, which shows the results of univariate Cox regression analysis of 1106 aging-related genes in the TCGA HGSOC cohort). Next, we utilized the Lasso-Cox regression model with 10-fold cross-validation to pinpoint the aging genes that contributed most significantly to HGSOC survival. Lasso coefficient profiles of the log (λ) with included gene counts were illustrated in Figure 1A. While the minimum partial likelihood deviance could be obtained at a number of 61, we opted instead to utilize the 1 - standard error criteria to choose a more concise model, resulting in the selection of 12 aging-related genes for the construction of a risk signature aimed at evaluating the prognosis of HGSOC patients (Fig. 1B).

Figure 1.

Construction of the aging risk signature and exploration of its survival prediction capacity. (A) Lasso-Cox model was employed to identify 153 prognostic aging genes, as evidenced by their coefficient profiles in the TCGA cohort. (B) Partial likelihood deviance analysis was conducted to determine distinct panels of these genes using the Lasso-Cox model, wherein red dots represent detailed partial likelihood of deviance values while gray lines indicate standard error. (C) HGSOC patients were partitioned into low- and high-risk subgroups based on the median risk score serving as the cutoff value. Following this, distinct survival status and time were compared between low- and high-risk subgroups. Moreover, a heatmap was utilized to represent the different expression patterns of the 12 identified aging-related genes in 2 groups. (D) Kaplan–Meier survival analysis of the determined 2 HGSOC subgroups. (E) The association between the identified risk signature and HGSOC prognosis was assessed in a multivariate Cox regression model, taking clinical confounding factors into account. HGSOC = high-grade serous ovarian cancer, TCGA = the Cancer Genome Atlas.

We identified 12 genes, namely CCR7, HLA-DOB, PRPF19, SHMT2, EPHA1, CTPS2, RCL1, MATN2, FCGBP, ANXA4, VSIG4, and GALNT10, and determined their corresponding prognostic weights (see Table S5, http://links.lww.com/MD/J588, Supplemental Digital Content, which shows the identified 12 aging-related genes and their regression coefficients in the TCGA HGSOC cohort). To evaluate risk scores for each patient with HGSOC (Fig. 1C), we established a risk signature by linearly integrating the expression values of these 12 genes, along with their regression weights obtained via the Lasso-Cox. Furthermore, distinct expression distribution for these 12 genes between 2 subpopulations was also presented (Fig. 1C).

In the discovery cohort, low- and high-risk subgroups respectively contained 290 and 291 HGSOC samples were identified by using the median risk score. Our findings indicated that patients in the low-risk group exhibited significantly better prognoses relative to those in the high-risk group (Log-rank test P < .001; Fig. 1D). Furthermore, this connection remained statistically significant adjusting for confounders such as age, stage, and grade under a multivariate Cox regression model (HR: 0.53, 95% CI: 0.43–0.66, P < .001; Fig. 1E).

3.2. Corroboration of the aging risk signature

To establish the prognostic significance of this identified aging signature, we conducted further analyses on an additional 4 cohorts of HGSOC patients. Our findings corroborated that low-risk patients harbored markedly improved prognoses when compared to their high-risk counterparts (Log-rank test P = .026 for GSE13876, P = .008 for GSE9891, P = .006 for GSE32062, and P = .021 for GSE49997; Fig. 2A–D). These results provide compelling evidence for the robustness and generalizability of the aging signature as a prognostic indicator in HGSOC.

Figure 2.

Corroboration of the survival prediction capacity of the aging signature. Survival plots of 4 HGSOC validation cohorts of (A) GSE13876, (C) GSE9891, (E) GSE32062, and (G) GSE49997 were obtained based on the 2 risk subpopulations. HGSOC = high-grade serous ovarian cancer.

3.3. Aging signature versus immunocyte infiltration and immunogenicity

Associations between aging, aging-related molecular features, and cancer immune regulation were recently reported.^[23,24] In light of this, we constructed multiple box plots to exhibit the distinct infiltration levels of 28 lymphocytes between 2 risk populations (Fig. 3A). Our findings indicate that anti-tumor immunocytes, represented by activated CD8 T cells, were significantly more abundant in low-risk samples (all P < .05). Additionally, we observed lower infiltration of immune-suppressive cells, such as CD56 dim natural killer cells, in low-risk samples. CIBERSORT algorithm also found a similar pattern of lymphocyte infiltration, with preferable immunocyte infiltration and the immune microenvironment in low-risk samples (see Fig. S1, http://links.lww.com/MD/J589, Supplemental Digital Content, which shows the distinct immunocyte infiltration between 2 risk subgroups based on the CIBERSORT algorithm).

Figure 3.

Immunocyte infiltration and signaling pathways linked with aging risk signature. (A) Totaling 28 immunocyte subtypes were categorized into 3 clusters according to their biological functions, with their distinct infiltration abundance in 2 subgroups exhibiting. Immunocytes with increased infiltration levels or decreased infiltration levels in the low-risk group were labeled with blue and purple, respectively. Significantly enriched signaling pathways of low-risk patients were obtained based on the pathways from (B) KEGG and (C) GO BP databases. Immunogenicity pathways were labeled with the green. (D) T cell-inflamed signature, (E) IFN gamma signature, and (F) cytolytic activity signature enrichment scores in low- and high-risk HGSOC patients. *P < .05, **P < .01, ***P < .001. HGSOC = high-grade serous ovarian cancer.

We performed GSEA analysis on the KEGG/GO BP databases and HGSOC transcriptomic data to investigate the pathways connected with the aging risk signature. Our findings indicated that pathways related to immune response, such as T cell receptor signaling pathway and antigen processing and presentation in the KEGG database (Fig. 3B), as well as pathways like activation of immune response in the GO BP database (Fig. 3C), were markedly enriched in low-risk samples.

We curated and assessed the distinct enrichment of 3 immunogenicity-related signatures in different risk subgroups. Our analysis revealed that low-risk patients exhibited markedly higher enrichment of T cell-inflamed signature (Fig. 3D), IFN gamma signature (Fig. 3E), and cytolytic activity signature (Fig. 3F) as compared to high-risk patients (all P < .01).

Our analysis also revealed that low-risk patients harbored enhanced expression of most immune checkpoint genes (all P < .05; see Fig. S2, http://links.lww.com/MD/J590, Supplemental Digital Content, which shows the distinct expression of immune checkpoints between 2 risk groups). These findings suggest a potential correlation between the identified aging risk signature and immune checkpoints, which could have implications for predicting immune infiltration and ICI treatment effects.

3.4. Aging signature versus ICI treatment efficacy

The invention of ICI agents has resulted in a significant enhancement of survival outcomes for several cancers. To scrutinize the correlation of this aging signature with ICI response, we utilized a UC dataset with ICI treatment information. Our analysis demonstrated that low-risk patients presented a noteworthy extension of ICI prognosis (Log-rank test P = .039; Fig. 4A). Moreover, this association persisted even after taking into account multiple clinical factors in a multivariate Cox model (HR: 0.59, 95% CI: 0.49–0.88, P = .024; Fig. 4B). This low-risk subgroup also exhibited significantly elevated objective response rate and disease control rates of ICI treatment (32.4% vs 15.6%, Fisher exact test, P = .021, Fig. 4C; 51.3% vs 37.9%, Fisher exact test, P = .009, Fig. 4D). Furthermore, results revealed low-risk patients harbored remarkably enhanced mutational burden, which is predictive of the ICI efficacy (Wilcoxon rank-sum test P = .002 and .007, respectively; Fig. 4E, F).

Figure 4.

The aging signature for ICI therapy roles under UC patients. (A) Kaplan–Meier ICI survival analysis of 2 risk populations identified based on the genes and weights from discovery cohort. (B) A multivariate Cox regression model was conducted to obtain a more precise association. (C) Objective response rates and (D) disease control rates of ICI treatment in 2 HGSOC groups. Associations of (E) tumor mutation burden and (F) neoantigen burden with determined aging risk signature in the ICI-treated UC cohort. HGSOC = high-grade serous ovarian cancer, ICI = immune checkpoint inhibitors, UC = urothelial cancer.

For the purpose of validating the aging risk signature predictive roles in immunotherapy, another dataset with ICI-treated melanoma samples was also utilized. Our findings revealed that low-risk samples harbored markedly prolonged overall survival and progression-free survival (Log-rank test P = .032 and .091, respectively; Fig. 5A, B). These associations remained significant even after multivariate-adjusted analysis (P = .006 and .069, respectively; Fig. 5C, D). Moreover, our study revealed higher proportions of ICI-responsive and non-progressive melanoma patients in the low-risk subgroup (52.3% vs 32.4%, Fisher exact test P = .031, Fig. 5E; 58.5% vs 42.6%, Fisher exact test P = .039, Fig. 5F). Consequently, these results suggest that this novel aging risk signature holds the potential as a valuable predictor for assessing immunotherapy efficacy.

Figure 5.

The aging signature for ICI therapy roles under melanoma patients. (A) Overall survival and (B) progression-free survival analysis of 2 risk populations identified based on the genes and weights from discovery cohort. (C, D) Multivariate Cox regression models were conducted to obtain more precise associations. (E) Objective response rates and (F) disease control rates of ICI treatment in 2 HGSOC groups. HGSOC = high-grade serous ovarian cancer, ICI = immune checkpoint inhibitors.

4. Discussion

In this research endeavor, we amalgamated transcriptomic and clinical data from numerous independent HGSOC cohorts to formulate and confirm an aging risk signature for purposes of prognosticating survival rates, immunogenicity, and efficiency evaluations of immunotherapy. The study robustness stems from its derivation from a more substantial HGSOC sample and validation across diverse dimensions. While the further prospective study is imperative, our research findings imply potential predictive significance for the aging risk signature concerning HGSOC prognosis, tumor immunogenicity, and immunotherapeutic effect.

Mutational signatures are distinct patterns of mutations that arise from endogenous and exogenous factors.^[25] Recent investigations have revealed that mutational signature 1, which is associated with aging, can lead to weakened tumor immunogenicity and inferior prognoses in prostate cancer and triple-negative breast cancer, indicating its potential relevance for ICI efficacy.^[26] Furthermore, a study conducted by Chong et al reported that patients who exhibited this aging mutational signature had poorer ICI survival outcomes in melanoma and NSCLC.^[27] In our research, we created an aging-related risk signature utilizing transcriptomic analysis instead of mutation-related indicators. Our findings indicate that this signature has predictive value for both survival status and immune infiltration.

CCR7 encodes a chemokine receptor that facilitates the migration of T cells to lymph nodes. Studies have demonstrated that expression is associated with increased infiltration of T cells into tumors, indicating a potentially positive impact on tumor immune response.^[28] HLA-DOB is involved in presenting antigens via major histocompatibility complex (MHC) class II molecules. Higher expression of HLA-DOB has been linked to greater immune cell infiltration in cancer, highlighting a positive correlation with tumor immunogenicity.^[29] SHMT2 encodes an enzyme in mitochondria involved in 1-carbon metabolism. Emerging research suggests that SHMT2 may modulate tumor immune response by regulating T cell activation.^[30] ANXA4 encodes a calcium-dependent phospholipid-binding protein that regulates various cellular processes. Some studies have suggested that ANXA4 could play a role in modulating immune cell function and potentially impact tumor immune response.^[31] VSIG4 encodes a transmembrane protein that regulates macrophage function. Recent research indicates that VSIG4 might be involved in modulating tumor immune response by regulating immune cell activity.^[32] These findings provide additional evidence supporting the predictive power of our developed risk signature in relation to both immune infiltration and immunotherapy effect.

As a result of the paucity of HGSOC cohorts that possess both transcriptomic expression and immunotherapy data, we elected to employ UC and melanoma cohort treated with ICI agents. These 2 cohorts currently represent the largest publicly available immunogenomic cohort. Our aim was to explore the correlation between the aging risk signature and the ICI effect. Our findings demonstrated that low-risk patients exhibit considerably improved ICI survival and enhanced response rate. Given the homogeneity of tumors in certain specific settings such as treatment efficacy evaluation,^[27] it is our conjecture that the aging-related risk signature could prove useful in evaluating not only immunotherapy response and outcome in UC and melanoma but also in HGSOC and other cancer types.

In aggregate, utilizing aging-related transcriptomic profiles sourced from a broader HGSOC cohort, we constructed a risk signature that permits appraisal of survival outcome, immune infiltration, and ICI effect. This determined aging signature represents a promising candidate indicator for the assessment of both HGSOC prognosis and prediction of ICI response.

Author contributions

Conceptualization: Suxia Liu.

Data curation: Suxia Liu, Yuexia Liu, Jianhong Ma.

Formal analysis: Yuexia Liu, Jianhong Ma.

Methodology: Rou Lv.

Software: Rou Lv, Fang Wang.

Writing – original draft: Suxia Liu, Fang Wang.

Supplementary Material

medi-102-e34851-s001.xlsx ^{(9.7KB, xlsx)}

medi-102-e34851-s002.xlsx ^{(17.1KB, xlsx)}

medi-102-e34851-s003.xlsx ^{(10.3KB, xlsx)}

medi-102-e34851-s004.xlsx ^{(37.4KB, xlsx)}

medi-102-e34851-s005.xlsx ^{(9.7KB, xlsx)}

medi-102-e34851-s006.pdf ^{(313.2KB, pdf)}

medi-102-e34851-s007.pdf ^{(313.6KB, pdf)}