Mining in Endometrial Cancer Based on the TCGA Database and Constructing Prognostic Models

Summary

This protocol outlines a computational approach for identifying genes associated with nicotinamide metabolism in endometrial cancer, utilizing TCGA and GEO datasets. To facilitate the discovery of biomarkers and enhance risk estimation in uterine corpus endometrial carcinoma, this research describes the analysis of distinct expression, route enhancement, prognostic model design, and evaluation procedures.

Abstract

Endometrial cancer (EC) ranks among the least prevalent gynecological tumors worldwide, with rising incidence due to aging and obesity. Lymph node metastasis remains common in Uterine Corpus Endometrial Carcinoma (UCEC), necessitating new prognostic biomarkers to guide treatment. In this research, UCEC information from the Cancer Genome Atlas (TCGA) was analyzed, and the results were validated using the Gene Expression Omnibus (GEO). Eighteen differentially expressed genetic factors associated with nicotinamide metabolism (NMRDEGs) were identified. Gene Set Enrichment Analysis (GSEA) indicated their role in oxidative stress, hypoxia, glycolysis, and apoptosis processes. Univariate Cox regression identified six key genes (AURKA, CDKN3, FOXM1, CDKN2A, TK1, and CDK1), utilized to progress a hazard prediction framework. Protein-protein interaction (PPI) analysis uncovered additional hub genes such as CDK2, CCNA2, TP53, and FOXM1. The six key genes showed strong prognostic value, and the study's risk model could guide clinical decisions. Nicotinamide metabolism was found to be significantly linked with EC progression. This study offers new perceptions into the role of nicotinamide metabolism in EC and suggests possible avenues for treatment advancements.

Introduction

Endometrial cancer (EC) is common in female patients¹. EC can be classified into the hormone-dependent type and the hormone-independent type². In 2020, the updates of guidelines for managing EC patients included molecular marker detection and molecular typing, significantly influencing postoperative adjuvant treatment and clinical prognosis for EC³. The pathogenesis of EC is multifaceted, involving mutations or loss of several oncogenes and/or tumor suppressor genes, along with abnormalities in multiple signaling pathways that can influence disease progression and outcome. Searching for new biomarkers and identifying effective treatment targets are critical. This pursuit aims to deepen the understanding of EC's pathogenesis and potential prognostic biomarkers, establishing a foundation for early detection and targeted therapy.

Niacinamide resembles niacin and plays a part in fat absorption and glycogen breakdown. It is essential for the human body as an important coenzyme component. Nicotinamide metabolism encompasses the processes through which the human body absorbs, transforms, and utilizes Nicotinamide.

Niacinamide metabolism begins with absorption. It can be acquired through food intake or amino acid conversion within the body and is absorbed by the intestinal mucosal epithelial cells. It then transforms into its more complex forms, crucial in many biochemical processes. Additionally, Nicotinamide is convertible into other compounds, including the metabolite 1-methylnicotinamide via nicotinamide adenine dinucleotide methyltransferase (NMT). These transformations potentially influence the onset and progression of various metabolic diseases. As reported, supplementing with Niacinamide has various beneficial effects, including the prevention and treatment of tumors⁴. Current examination has confirmed that it could downregulate the phosphorylation level of STAT3Y705 in lung cancer cells⁵. A study has found that using Niacinamide alone could inhibit invasive nonmelanoma membrane tumors in high-risk patients⁶. Kourtzidis et al.⁷ found that supplementing with Niacinamide could inhibit weight gain in rats, enhance the activity of transketolase (TK) and red blood cell Na+- K+- ATPase, and affect energy metabolism in the body. A prior study has recognized nicotinamide N-methyltransferase (NNMT), a methyltransferase, as an important issue in the activity of cancer-associated fibroblasts⁸. Furthermore, nicotinamide phosphoribosyl transferase has been shown to contribute to venetoclax resistance in relapsed critical myeloid leukemia stem compartments⁹. Therefore, the objective was to comprehensively identify NAD+ metabolism-related genes (NMRGs) linked to the risk of EC.

Protocol

This study used publicly available, de-identified clinical and transcriptomic data from The Cancer Genome Atlas and the Gene Expression Omnibus. All contributing studies had prior institutional review board approval and informed consent. As only secondary analysis of anonymized data was performed, no additional ethical approval was required. The databases and software used are listed in the Table of Materials.

1. Data download

Research utilized the EC (TCGA-Uterine Corpus Endometrial Carcinoma (TCGA-UCEC)) dataset¹⁰, which consists of 589 samples, including 554 tumor tissue samples from UCEC patients (UCEC group) and sequencing data from 35 adjacent normal tissues (Normal group). The UCSC Xena database was utilized to recover the corresponding clinical data¹¹, excluding those lacking complete clinical information. Ultimately, 577 samples with clinical data were available for analysis. Detailed baseline information is provided in Table 1.

Additional datasets related to EC, GSE115810, and GSE63678¹², were downloaded using the GEOquery package¹³. The GSE115810 dataset and GSE63678 were merged to create the Combined Datasets for further analysis (Table 2).

NAD+ metabolism-related genes (NMRGs) are situated after the GeneCardsrecord¹⁴ and relevant literature¹⁵. Using "Niacinamide metabolism" as the search term in GeneCards, identified 345 NMRGs with relevance scores above 4. Combining and removing duplicates from the 42 NMRGs found in the literature compiled a total of 371 NMRGs (Supplementary Table 1). Clinical data were acquired as.tsv phenotypic files; data were downloaded in HTSeq-FPKM format. Excluded samples had more than 20% of their clinical data missing. FPKM was log₂-transformed and converted to TPM (transcripts per million). Probe IDs were mapped to gene symbols for GEO datasets, and duplicate probes were averaged.

2. Differentially expressed genes of nicotinamide metabolism

Research began by applying the R setsva¹⁶ to eliminate set possessions after the GSE115810 and GSE63678 datasets, resulting in a mutual dataset containing 31 EC (UCEC) and 8 adjacent normal samples. Next, the utilization of limmaset¹⁷ to conduct discrepancy gene expression examination on the TCGA-UCEC dataset.

The intersection of DEGs from the TCGA-UCEC analysis with 337 NMRGs to pinpoint DEGs connected to nicotinamide metabolism. This produced a list of Niacinamide metabolism-related differentially expressed genes (NMRDEGs), which were pictured by a Venn illustration. The outcomes of the discrepancy appearance examination were illustrated by the ggplot2 R package¹⁸, while a heatmap of the NMRDEGs was generated using the pheatmapset¹⁹. Inter-dataset variance is eliminated through batch correction using ComBat (empirical Bayes). The limma empirical Bayes linear model framework was employed in DEG analysis. Explicitly applied differential expression thresholds:

|log₂FC| ≥ 1

FDR less than 0.05.

The NMRG list only intersected with DEGs that met both requirements. Plots of volcanoes and heatmaps made with ggplot2 and pheatmap.

3. NMRDEGs function (GO), pathway (KEGG) enhancement examination

GO²⁰ and KEGG²¹enhancement examines was completed by the clusterProfiler set²². For both analyses, significance thresholds were recognized at p. adjust< 0.05 and FDR (q value) < 0.25. Research also integrated logFC values into the enrichment analysis, representing the results in circular and chord diagrams. Thresholds for enrichment significance:p-value adjusted < 0.05 FDR < 0.25 (q-value). Cluster Profiler was used for GO and KEGG studies. Gene-directionality is displayed using chord and circle plots that incorporate log₂ fold change data.

4. Gene Set Enrichment Analysis (GSEA)

The type of inheritable factor sets that contributed most to the phenotype can be identified using GSEA²³. For this analysis, the TCGA-UCEC dataset was ranked based on logFC values, and an improvement examination was conducted using the clusterProfiler package. Key restrictions encompassed a seed value of 2022 and 10,000 permutations. The MSigDB gene set "c2.all.v2022.1.Hs.symbols.gmt" was employed²⁴. The most enriched pathways, including Manalo hypoxia-induced genes, oxidative stress-induced senescence, glycolysis, and apoptosis pathways, were visualized using a mountain plot. Prior to GSEA, genes were ordered by log₂ fold change.10,000 permutations were used in the analysis.c2.all.v2022.1.Hs.symbols.gmt is the MSigDB collection that was used. For reproducibility, a fixed random seed (2022) was employed. Significant pathways were those with p < 0.05 and q < 0.25.

5. Cox model construction and related prognosis examination

To determine the predictive value of Nicotinamide metabolism-linked differentially expressed genes (NMRDEGs) in endometrial carcinoma (UCEC), researchers used univariate Cox regression analysis to first classify applicant inheritable factors; those with a danger proportion (HR) > 1 and p-value < 0.1 were determined to be appropriate for the multivariate Cox relative risks framework.

Criteria for univariate Cox selection: p < 0.10 and HR > 1.Log₂-TPM-normalized expression values were employed in the multivariate Cox model. A linear combination of Cox coefficients × gene expression is used to determine the risk score.1-, 3-, and 5-year OS probabilities were employed in nomogram calibration.1-, 3-, and 5-year AUC values were employed in time-dependent ROC. The surv_cutpoint max-statistic approach was used to find survival cut-off values. Both KM and ROC analyses used the same thresholds.

A nomograph was constructed from the multivariate Cox model to assess its accuracy or predictive capacity and to calculate the chances of 1-, 3-, and 5-year total existence. Standardization arcs are employed for assessing the steadiness between projected prospects and real consequences, and decision curve analysis (DCA) was utilized to measure the medical helpfulness of the structure²⁵.

mRNA expression levels were determined as normalized log₂-transformed records per million (TPM) values by means of the DESeq2 package. TPMs accounted for sequencing complexity and gene measurement to provide robust and unbiased estimates of expression levels between samples.

Using the multivariate Cox model's coefficients, each patient's prognostic hazard rating was determined as follows:

riskScore = Σ_i Coefficient (gene_i) *mRNA Expression (gene_i)    (1)

Kaplan-Meier (KM) existence arcs were prepared to assess the general endurance of tall- and low-hazard clusters created on determined hazard ratings. Time-dependent receiver operating characteristic (ROC) arcs were produced in evaluating frameworks' routine at 1-, 3-, and 5-year periodideas^26,27.

To categorize gene expression by high- and low-expression collections for survival stratification, the utilization of the surv_cut point role after the survminer R package is used. This function determines the greatest cut-off value by maximizing the standardized log-rank statistic, giving an unbiased, statistically optimal cut-off point.

The cut-off values obtained for each prognostic gene are indicated in the ROC curves as dashed lines. Research applied the same thresholds for all survival and ROC analyses.

TCGA RNA-seq was downloaded in HTSeq-FPKM format; clinical data was imported as TSV phenotype files; FPKM was converted to TPM and log₂-transformed; GEO datasets were mapped from probe IDs to gene symbols using platform annotations; duplicate probes were averaged for a single gene value; samples with more than 20% missing clinical information were excluded; ComBat (empirical Bayes) was used for batch correction for GSE datasets; PCA and boxplots were used to verify that the batch correction was successful. TPM normalization using standard expression transformation techniques; batch correction using ComBat with dataset origin as the batch variable; differential expression calculated using limma linear modeling (tumor vs. normal design matrix); ranked gene lists generated from log₂ fold changes for GSEA input; and univariate and multivariate Cox regressions carried out using survival analysis tools

6. Gene set variation analysis (GSVA)

GSVA²⁸ was employed to measure the way enhancement among the clusters. In the TCGA-UCEC dataset, 50 hallmark pathways were enriched, with 41 showing important alterations among the binary assemblies. GSVA was used with hallmark gene sets to obtain pathway activity per sample; STRING protein interaction data were imported into Cytoscape; the MCC algorithm was used to identify hub genes; risk scores were computed as the sum of gene expression values multiplied by their Cox coefficients; time-dependent ROC curves were generated using survival-time ROC routines. For each sample, GSVA computed pathway-level enrichment scores. The Wilcoxon rank-sum test is used to assess differences in hallmark pathway activity. Of the fifty signature pathways, forty-one were significantly different (adjusted p < 0.05).

7. Protein-protein interaction (PPI) system

A PPI system containing the important genes (AURKA, CDKN3, FOXM1, CDKN2A, TK1, and CDK1) has been created using the STRING file²⁹ and a communication value threshold of 0.70, indicating high confidence. This network has been created using Cytoscape³⁰, highlighting interactions that may play crucial roles in UCEC pathogenesis. The Maximal Clique Centrality (MCC ) method³¹ was useful to rank the inheritable factor created on their interaction scores within the net. The upper 10 protein sequence with the uppermost interface scores was recognized, including CDK2, CDK4, CCNA2, CCNB1, CCNE1, CDK1, TP53, and FOXM1. These genes were further analyzed for their involvement in critical biological processes. The GeneMANIA platform³² was also used to predict additional protein interactions and provide a broader context for the key genes' roles in UCEC progression. Threshold for STRING confidence score: >0.70 (high confidence). Cytoscape displays the network. The Maximal Clique Centrality (MCC) technique is used to rank hub genes. The MCC ranking was used to identify the top interacting genes (CDK2, CCNA2, TP53, etc.). Additional interaction predictions are made using GeneMANIA.

8. Technology roadmap

The overall workflow and methods employed in this study are summarized in the technology roadmap displayed in Figure 1. This roadmap outlines the steps from dataset acquisition and differential expression analysis to constructing prognostic models and enrichment analyses.

9. Statistical analysis

Data processing and statistical estimation were done using the R program (v4.3.0). The Mann-Whitney U test or Independent Student's t-test was used for two-group comparisons; the Kruskal-Wallis test has been employed for three or more assemblies. The descriptive data were assessed using chi-square or Fisher's exact test. Furthermore, Spearman correlation and Kaplan-Meier survival analysis were performed; p < 0.05 was considered significant.

Applying statistical tests according to the distribution of data: Normal data using the student's t-test. Mann-Whitney U test for data that is not normal. Kruskal-Wallis test for more than three groups. Fisher's exact and chi-square for categorical data. Statistical significance is defined as p < 0.05.

Data dependability is maintained by the pretreatment checkpoints, where boxplots should show consistent expression variance across samples, and PCA plots should demonstrate the absence of batch-specific clusters following ComBat adjustment. Heatmaps that demonstrate tumor-normal grouping and volcano plots that clearly illustrate gene up-/down-regulation are necessary for DEG validation. For the projection Cox model, calibration plots should match predicted and actual survival, ROC AUC values should be greater than 0.65, and KM arcs must show a substantial survival difference. Different route activities between risk groups should be shown by GSVA analysis, in line with established mechanisms like expansion or cell-cycle routes. In order to verify network resilience, highly coupled nodes in the PPI network must appear centrally, and hub genes determined by MCC should match physiologically significant regulators.

Results

Merging the correction of the GEO dataset

Research used the SVA R set to precise for batch possessions in the combined GSE115810 and GSE63678 datasets, producing a unified dataset termed Combined-Datasets. The results of group result elimination were verified using circulation boxplots (Figure 2A,B) and PCA plots (Figure 2C,D), which demonstrated that the batch effects were largely el...

Discussion

The role of energy metabolism in the microenvironment of cancer cells has garnered considerable attention recently. Research has shown that genes involved in nicotinamide metabolism are closely linked to the progression of various malignancies. Once absorbed by the body, Nicotinamide is converted into NAD and NADP, both of which are pivotal in numerous cancer-related processes. Consequently, NAD metabolism is emerging as a promising therapeutic goal for tumor action. It is critical in various cellular activities, includi...

Materials

List of materials used in this article

Name	Company	Catalog Number	Comments
clusterProfiler package	Bioconductor	v4.8.0	GO and KEGG enrichment analysis
Cytoscape	Cytoscape Consortium	v3.9.1	PPI network visualization
DESeq2 package	Bioconductor	v1.40.0	Normalization and expression analysis
GeneCards database	GeneCards Suite	N/A	Source of nicotinamide metabolism-related genes
GeneMANIA platform	genemania.org	N/A	Gene function and interaction prediction
GEO datasets (GSE115810, GSE63678)	Gene Expression Omnibus (GEO)	N/A	Public gene expression datasets for EC
ggplot2 package	CRAN	v3.4.2	Data visualization
limma package	Bioconductor	v3.54.0 (example)	Differential expression analysis
pheatmap package	CRAN	v1.0.12	Heatmap plotting
R software (v4.3.0)	R Foundation for Statistical Computing	Version 4.3.0	Data analysis platform
STRING database	STRING (string-db.org)	v11.5	Protein-protein interaction analysis
survminer package	CRAN	v0.4.9	Survival analysis
sva package	Bioconductor	v3.46.0	Batch effect removal
TCGA-UCEC dataset	The Cancer Genome Atlas (TCGA)	N/A	Transcriptome and clinical data of UCEC patients
UCSC Xena	University of California, Santa Cruz	N/A	Platform for clinical/genomic data download