Decoding the Impact of Genetic Variants in Gastric Cancer Patients Based on High-Dimensional Copy Number Variation Data Using Next-Generation Knowledge Discovery Methods

Abstract

Objectives: Despite a reduction in the incidence and mortality rates of gastric cancer (GC), it remains the fifth most frequently diagnosed malignancy globally. A better understanding of the regulatory mechanisms involved in the progression and development of GC is important for developing novel targeted approaches for treatment. We aimed to identify a set of differentially regulated pathways and cellular, molecular, and physiological system development and functions in GC patients infected with H. pylori infection based on copy number variation (CNV) data using next-generation knowledge discovery (NGKD) methods. Methods: In this study, we used our previous CNV data derived from tissue samples from GC patients (n = 33) and normal gastric samples (n = 15) by the comparative genome hybridization (CGH) method using Illumina HumanOmni1-Quad v.1.0 BeadChip (Zenodo Accession No: 1346283). The variant effects analysis of genetic gain or loss of function in GC was conducted using Ingenuity Pathway Analysis (IPA) software. In addition, in silico validation was performed with iPathwayGuide software using high-throughput RNA sequencing (RNAseq) data (GSE83088) from GC patients. Results: We observed 213 unique CNVs in the control group, 420 unique CNVs in the GC group, and 225 common variants. We found that cancer, gastrointestinal diseases, and organismal injury and abnormalities were the three diseases or disorders that were most affected in the GC group. We also identified that the programmed cell death ligand 1 (PD-L1) cancer immunotherapy pathway, T-cell apoptosis, T-cell exhaustion, and Type 1 regulatory T-cell (Tr1 cells) specialization were dysregulated in GC patients. RNAseq data from GC patients showed that the PD-1/PD-L1 pathway was significantly upregulated in GC samples compared with controls. Conclusions: In conclusion, in the present study, we decoded differentially impacted GC-specific diseases and biological functions and pathways based on CNV data using NGKD methods that can be adopted to design personalized therapeutic approaches for patients with GC in a typical clinical milieu.

1. Introduction

The prevalence of gastric cancer (GC) has been declining; still, GC is the fifth most commonly diagnosed cancer worldwide [1]. Early diagnosis is crucial for improving prognosis and decreasing the rate of mortality [2]. It is usually diagnosed at an advanced and incurable stage, as clear symptoms are not exhibited in the early stages [3]. Therefore, it is necessary to identify the early stages of GC by identifying regulatory molecules or biomarkers for its diagnosis and prognosis.

Sequence alterations of DNA, such as copy number variations (CNVs), can lead to the oncogenes activation and inactivation of anti-oncogenes. CNVs can affect gene expression and confer differential susceptibility to various diseases. Several studies have shown that genomic duplication or deletion affects gene expression and cancer-related biological processes [4,5]. CNVs appear to have a higher mutation rate than single-nucleotide polymorphisms (SNPs), thereby affecting larger genomic fragments of genomes [6]. In cancer, cells divide and grow in an unregulated manner involving alterations in specific genes [7,8]. Recent studies have shown how CNVs provide key information and their impact on biological processes in humans at the genomic level. It is generally acknowledged that CNV affects gene expression and is associated with carcinogenesis and the progression of different cancers [9,10].

Integration of CNVs data has made it possible to systematically study and understand genetic changes and signaling pathways. However, owing to variations in different tumor types, a complete understanding of the pathogenic mechanisms of cancer is essential to identify potential therapeutic possibilities. Many studies have shown that the correlation between CNV and gene expression has biological implications in the development and progression of cancer [11]. In GC, CNVs are essential indicators of the risk and progression of GC. Our previous study on GC patients from the Saudi population using a comparative genomic hybridization array (CGH) identified high copy number gains and losses of DNA regions [2]. In the present study, CNV data in GC were investigated using CGH profiling, followed by in-depth systems biological analyses to predict prognosis responses and better responses to immunotherapy. We used the genetic loss or gain analysis method to discover unique CNVs, diseases and biofunctions, and canonical pathways in GC patients using NGKD methods to better design personalized therapies in a typical clinical setting.

2. Materials and Methods

2.1. Ethical Statement

The raw CNV data generated from our previous study [2] were used for variant effect analysis, as indicated in the clinical samples section below. Therefore, this study was exempted from institutional review board (IRB) approval.

2.2. Clinical Samples

In this study, we used our previous CNV data derived from tissue samples of 33 GC patients and gastric samples from 15 normal controls by the comparative genome hybridization (CGH) method [2]. Control gastric tissue samples were collected from 6 Saudi males and 9 Saudi females with an age range of 22–53 years. The Saudi GC patients consisted of 6 males and 2 females with an age range of 43–87 years from the early stage and 4 females and 21 males with an age range of 27–82 years from last stage. All GC patients before surgery had no history of chemotherapy or radiotherapy. The CNV data were deposited in Zenodo (Accession No: 1346283) and shared under the Creative Commons Attribution 4.0, international license (CC BY 4.0). Chromosomal alterations were recorded by comparing the normalized intensity of the GC samples to that of the control samples [2].

2.3. Variant Effects Analysis for Genetic Loss or Gain Using Ingenuity Pathway Analysis Software

The genetic gain- or loss-of-function data from the normal healthy control and GC groups were imported into Ingenuity Pathway Analysis (IPA) software (Qiagen, USA). IPA is a web-based software application developed and maintained by Qiagen, USA. It uses a curated backend knowledge base that is regularly updated based on the latest literature and state-of-the-art next-generation backend algorithms to analyze, integrate, and interpret high-throughput data derived from variant effects analysis, RNA sequencing (RNA seq), metabolomics, proteomics, MicroRNA, and SNP microarrays. Downstream effects analysis in IPA predicts cellular and molecular biofunctions, diseases and disorders, and other phenotypes regulated by patterns in control and test data sets. Loss or gain of activity of a gene was assigned on a scale of −2, −1, 0, 1, 2 to indicate how much each gene has lost or gained activity due to mutation in both groups before uploading the datasets into IPA. The core analysis module was selected to identify significant downstream effects on canonical pathways, diseases and biofunctions, tox functions, pathological functions, etc. [12,13], in the control and GC groups separately. Fisher’s exact test with a p-value cut-off ≤ 0.05 and a minimum of p ≤ 0.05 and p ≤ 0.01 to compute the Benjamini–Hochberg (B-H) corrections was used to calculate statistical significance for the canonical pathways and diseases and biofunctions, respectively. Fisher’s exact test is computationally less rigorous and commonly used. In IPA, the Fisher’s exact test was used calculate the p-value in IPA analysis since the computation depends on the random statistical model or null model. The random model correlates a random selection of a group of molecules (N) and tests if this group is enriched significantly in a specific annotation in the functional analysis.

The activation or inhibition of canonical signaling pathways, diseases and disorders, molecular and cellular functions, and the development and function of physiological systems were computed based on the Z-score algorithm of IPA and compared to an idealized activation or inhibition pattern for a signaling pathway, disease/disorder, or biological function. The molecular activity predictor (MAP) tool in IPA [13] was used to evaluate the effects of variants on canonical signaling pathways in the control and GC groups. Heatmaps were constructed based on the Z-score showing the diseases and biofunctions significantly affected by the CNVs in the control and GC groups using the heatmap tool in IPA. In addition, the IPA Comparison Module was used to decipher unique variants in the control and GC groups as well as common variants. Pictorial representation of the underlying pathological processes in GC group was done using BioRender webtool.

2.4. In Silico Validation with iPathwayGuide Software

We analyzed high-throughput RNAseq data of GC samples from Gene Expression Omnibus (GSE83088) using iPathwayGuide (Advaita Bioinformatics, Ann Arbor, MI, USA, as previously described [14,15,16,17].

3. Results

3.1. Copy Number Variations in GC Patients

In all GC samples, copy number gains were higher than losses compared with normal tissue samples from Saudi patients. A high copy number gain and loss was observed in our earlier study [2]. Many common CNVs have been reported along with novel CNVs at 1p36.32, where 11 gains and 2 losses were recorded in GC samples from Saudi patients. These copy number gains were observed from early stages, whereas in late stages, copy number losses were observed in GC samples (Supplementary Figure S1). In other populations, this region is still unreported in the Database of Genomic Variants and may be specific to the Saudi population [2].

3.2. Variant Effects Analysis Using IPA

We identified 213 unique CNVs in the control group, 420 unique CNVs in the GC group, and 225 common variants between the two groups. The total number of unique variants in the control and GC groups and the common variants are shown in a Venn diagram (Figure 1). The list of unique CNVs found in the control and patient groups and the common CNVs between these groups are provided in Supplementary Tables S1–S3.

Variant effects analysis of genetic gain or loss of function using the IPA core analysis module to predict the downstream canonical pathways revealed that the programmed cell death protein-1 (PD-1) and programmed cell death ligand 1 (PD-L1) cancer immunotherapy pathway was the most affected signaling pathway in the GC group (Figure 2). Furthermore, MAP analysis of the PD-1/PD-L1 cancer immunotherapy pathway revealed that T-cell apoptosis, T-cell exhaustion, and Type 1 regulatory T-cell (Tr1 cells) specialization were activated in GC (Figure 3a) but not in the healthy control group. Cellular functions, such as T-cell activation, effector functions of T cells, and T-cell proliferation, were inhibited in the GC group compared to the control group (Figure 3b). The mechanism of PD-1/PD-L1 immune checkpoint inhibition of T-cell activation by the tumor cell and the effectiveness of anti PD-1 antibodies in restoring T-cell activation and subsequent immune attack on the tumor cell, leading to its death, is depicted in Figure 3c.

Using the “Diseases and Functions” tab, a bar graph and a categorically organized heatmap were generated to investigate groups of ontologically related diseases and biofunctions predicted to be increased (orange) or decreased (blue) in control and GC groups (Figure 4). We found that cancer, gastrointestinal diseases, and organismal injury and abnormalities were the three diseases or disorders most affected in the GC group (Figure 4a). The least affected diseases and disorders, including neurological disease, dermatological disease and conditions, endocrine system disorders, etc., are shown in Figure 4a. The heatmap-based activation Z-score algorithm of IPA showed that diseases such as cancer, organismal injury and abnormalities, gene expression, cell death, and survival were significantly inhibited compared to the GC group (Figure 4b). In the GC group, disease categories such as cancer, incidence of cancer, and neoplasia were significantly activated based on the Z-score (Figure 4c). The disease and biofunctions activated by the CNVs unique to the control group included cellular development, cellular maintenance, cellular assembly and organization, cell morphology, and tissue development (Figure 4d). The diseases and biofunctions that were significantly activated by the CNVs common in the control and GC groups included cell development, cell movement, cell growth and development, cellular function, and cellular assembly (Figure 4e). The diseases and biofunctions significantly affected by the CNVs unique to the GC group included cancer, solid tumor, abdominal cancer, cell death, and survival (Figure 4f). The diseases and biofunctions affected by CNVs in the control and GC patients are provided in Supplementary Figures S2–S6 and Supplementary Tables S4–S10.

Additionally, the meta-analysis of high-throughput RNAseq data obtained from GC showed that PD-1/PD-L1 pathway was significantly upregulated compared to normal control (Figures S7–S9).

4. Discussion

CNV is a type of structural variation in the genome that results in an abnormal copy number of genes, such as deletions, losses, gains, and amplifications [2]. CNV is critical for the expression of coding and non-coding genes in the genome, significantly affecting the activity of a number of signaling pathways [18,19] and promoting cancer development. Therefore, targeting the amplified genes and the subsequent activation of oncogenic signaling pathways may be crucial for developing novel personalized therapies for GC. Identification of CNVs in either chromosomal or mitochondrial DNA from clinical samples (blood or tissue) may aid in diagnosis, prognosis, and personalized or targeted therapies for GC [20,21]. IPA analysis of the genetic gain- or loss-of-function datasets obtained from genomic sequencing data from control and GC groups revealed the biological impact downstream of canonical pathways, diseases, disorders, and biological functions. We used “variant loss/gain” to determine whether and how much activity each gene was lost or gained from a mutation. The “variant loss/gain” is determined on a scale of −2, −1, 0, 1, 2, where each number contains information about the loss or gain of activity of that gene and not at the variant level.

IPA has more than 700 pre-defined canonical pathways or well-understood signaling pathways for the interpretation of loss or gain functions. In IPA, pathway overlap of significant dataset molecules of the control and GC groups with canonical pathways that are likely activated or inhibited can be obtained. IPA uses an algorithm that relates the effect on the gene from the mutation in the dataset to the pattern anticipated for the specific canonical pathway if it was activated. To understand the variant effects on these well-known pathways, canonical pathways were investigated using IPA to predict the involvement of signaling pathways in the control and disease groups. In the GC group, the PD-1/PD-L1 immunotherapy pathway had the highest impact on the canonical pathway. In line with our findings based on IPA, a high-throughput knowledge discovery platform, several studies have shown that PD-1 and PD-L1 are overexpressed in GC tissues [22,23,24]. Increased PD-1/PD-L1 expression augments the growth of GC cells [25]. Furthermore, our in silico validation study using iPathwayGuide, another NGKD platform, based on RNAseq data derived from GC patients (14), proved that the genes implicated in the PD-1/PD-L1 pathway was differentially regulated and reduced T-cell activity and increased T-cell apoptosis and effector T-cell development.

Cancer immunotherapy has been a success story in antitumor therapy over the past few years [26,27]. PD-1 is one of the major factors that inhibits tumor-specific immune responses and induces immune tolerance or self-tolerance by promoting apoptosis of antigen-specific T cells, T-cell activity, and inhibiting apoptosis of Tr1 cells [28,29]. On the other hand, the PD-L1, a transmembrane protein, binds with PD-1 to co-inhibit immune responses against tumor cells, decrease the proliferation of PD-1-positive immune cells, decrease cytokine secretion, and augment apoptosis of T cells [30]. PD-L1 reduces host immunity against tumor cells in various malignancies [31,32]. Hence, PD-1/PD-L1 signaling is essential for immune escape in various cancers and has a substantial impact on cancer therapy (Figure 3c). In our study, the PD-1/PD-L1 cancer immunotherapy pathway was significantly affected by CNVs in GC group. Further in-depth MAP analysis revealed that cellular functions such as T-cell apoptosis, T-cell exhaustion, and type 1 regulatory T-cell (Tr1 cells) specialization were activated in GC but not in the healthy control group. Cellular functions, such as T-cell activation, effector functions of T cells, and T-cell proliferation, were inhibited in the GC group compared to the control group.

The anti-PD-1 monoclonal antibody pembrolizumab and nivolumab are the first group of checkpoint inhibitors to be fast-tracked by the U.S. Food and Drug Administration (FDA) for the treatment of ipilimumab-refractory melanoma [33,34]. Clinical trials with antibodies against PD-1, such as pembrolizumab and nivolumab, have shown good efficacy and reasonable safety in gastrointestinal cancer [35,36]. Therefore, it is important to assess new monoclonal antibodies (mAbs) against PD-1 and PD-L1 in gastrointestinal cancer. The major strength of our study is that we have demonstrated the use of NGKD methods such as IPA and iPathwayGuide to rapidly describe the underlying pathological pathways in GC patients using CNV and RNAseq data, respectively. However, major limitations include the small sample size and lack of functional validation of CNVs and their impact on cellular and molecular functions and cancer immunotherapy pathways in GC patients.

5. Conclusions

Our study adopted NGKD methods to decode how CNVs impact the PD-1/PD-L1 cancer immunotherapy pathway and the specific activation of T-cell apoptosis, T-cell exhaustion, and type 1 regulatory T-cell (Tr1 cells) specialization in GC patients. Furthermore, cellular functions, such as T-cell activation, effector functions of T cells, and T-cell proliferation, were inhibited in GC patients. In the future, we aim to add more data and integration analysis to define the relationship between CNVs, cellular functions, and cancer immunotherapy pathways using NGKD methods to design personalized treatment regimens to effectively control GC in the Saudi population.