Survival and Enrichment Analysis of Epithelial–Mesenchymal Transition Genes in Bladder Urothelial Carcinoma

Abstract

The escalating prevalence of bladder cancer, particularly urothelial carcinoma, necessitates innovative approaches for prognosis and therapy. This study delves into the significance of genes related to epithelial–mesenchymal transition (EMT), a process inherently linked to carcinogenesis and comparatively better studied in other cancers. We examined 1184 EMT-related gene expression levels in bladder urothelial cancer cases through the TCGA dataset. Genes shown to be differentially expressed in relation to survival underwent further network and enrichment analysis to uncover how they might shape disease outcomes. Our in silico analysis revealed a subset of 32 genes, including those significantly represented in biological pathways such as VEGF signaling and bacterium response. In addition, these genes interact with genes involved in the JAK-STAT signaling pathway. Additionally, some of those 32 genes have been linked to immunomodulators such as chemokines CCL15 and CCL18, as well as to various immune cell infiltrates. Our findings highlight the prognostic utility of various EMT-related genes and identify possible modulators of their effect on survival, allowing for further targeted wet lab research and possible therapeutic intervention.

1. Introduction

Bladder cancer is the 10th most common malignancy worldwide with 573,278 new cases and 212,536 deaths in 2020 [1]. Urothelial carcinoma accounts for over 90% of bladder cancers, which costs the U.S. alone $4 billion annually [2]. The prevalence is predicted to continue to rise due to the increasing industrialization and urbanization in developing countries, and the aging population [3,4]. Well-studied risk factors include cigarette smoking and occupational exposures especially in metal workers, painters, and chemical process workers [5,6]. Various altered genes have been implicated in amplifying the effect of these environmental exposures including carcinogen detoxification genes like UDP Glucuronosyltransferase Family 1 Member A Complex Locus (UGT1A) and N-acetyltransferase 2 (NAT2) [7,8,9], fibroblast growth factor receptor 3 (FGFR 3) [10], p16, p53, retinoblastoma (RB), matrix metalloproteinases, genes involved in folate metabolism, and high activity metabolic activators like high activity P450 cytochrome enzymes [11,12,13,14,15,16,17]. Within the TNM staging used, T1 represents tumor invasion up to the muscular layer of the bladder, with further stages T2,3,4 representing further progression into the muscle, perivesicular layer, and adjacent structures and organs, respectively. Currently, treatment options include transurethral resection of bladder tumor (TURBT) and intravesical therapy for non-muscle invasive, while options for metastatic disease include radial cystectomy, neoadjuvant chemotherapy, and newer immunotherapies [2,18,19,20,21,22].

Several genes have been implicated in the development of urothelial cancer and metastases and used as targets for therapy. Some of these include LRP1B [23], ERRC2, FANCC, ATM, RB1 [24], p53 [25], and SLC14A1 [26]. There are currently no widely accepted bladder cancer screening programs, though biannual cystoscopies have been found to be efficacious in vulnerable subpopulations [27,28,29]. Currently follow-up is time-consuming and expensive, consisting of cystoscopy, imaging, and surgery; urine biomarkers are being studied to supplement those options [18,19].

Cells involved in the invasion of bladder cancer alter their surroundings and can also become transiently and reversibly plastic, turning into mesenchymal stem cells. This is the epithelial–mesenchymal transition (EMT), which is among the most relevant paradigm shifts in how we view cancer progression and can combat its growth. During EMT, epithelial cells lose their polarized, adhesive characteristics and gain a mesenchymal phenotype, enabling them to migrate and invade surrounding tissues [30,31]. Transcription factors including Snail, Zeb, and Twist aid in this process by repressing E-cadherin, an epithelial transmembrane protein [32]. In contrast to epithelial cells, mesenchymal carcinoma cells exhibit specific metabolic needs. As they undergo EMT, cancer cells finely regulate multiple metabolic pathways to support the demands of rapid cell proliferation [33]. The molecular pathways shown to be associated with EMT include Snail/Slug, Twist, Six1, Cripto, TGF-β, and Wnt/β-catenin [34]. The literature shows how genes such as CDH1, ZEB1, TGFB, CDH2, VIM, and TIMP1 have been linked to inducing the EMT phenotype, driving cell migration, and adapting to changing demands on the primary tumor [33,35,36].

In bladder cancer, various microRNAs (miRNAs) have been found to regulate proteins such as Smad7 or Twist1, either promoting or disrupting EMT and metastasis [37]. Understanding the mechanisms underlying EMT is crucial for developing targeted therapies to control cancer metastasis and may prove useful in treatment options going forward. Comparatively, there has been less work in this field in bladder cancer than in other cancers. This paper examines a multitude of EMT-related genes in relation to not only outcomes, but also the biologic networks and pathways which allow these genes to influence carcinogenesis and affect these outcomes.

2. Materials and Methods

2.1. Selection of Genes

To have a comprehensive overview of genes involved in the epithelial–mesenchymal transition, dbEMT 2.0 (http://dbemt.bioinfo-minzhao.org/ (accessed on 1 September 2022)), a database curated for focus on EMT-related genes, was utilized. A spreadsheet was generated with 1184 genes listed on the database, obtained from an initial PubMed abstract query for “Epithelial Mesenchymal Transition Genes” with the results mined for unique genes linked to EMT (see Supplemental Materials).

2.2. Survival Analysis

Publicly available cases from the NIH-funded “The Cancer Genome Atlas” (TCGA) project were utilized to examine gene expression pertaining to survival in bladder urothelial cancer (data portal: https://portal.gdc.cancer.gov/projects/TCGA-BLCA (accessed on 1 September 2022)). Kaplan–Meier plots were generated through the R2 platform (https://hgserver1.amc.nl/cgi-bin/r2/main.cgi (accessed on 1 September 2022)) using the TCGA dataset for “Bladder Urothelial Carcinoma”, n = 407. The built-in “KaplanScan” algorithm was used to divide mRNA gene expression into “high” versus “low” categories (n values for each based on KaplanScan groupings of expression). Overall survival was compared to follow-up time in months being analyzed. For multiple hypothesis testing, p-values were adjusted to a false discovery rate (FDR) of 0.05.

2.3. Expression Analysis

To compare normal versus tumor levels of those EMT genes which showed differential expression regarding survival, mRNA levels for the “Bladder Urothelial Carcinoma” TCGA dataset were analyzed with a Welch’s t-test through the UCSC Xena platform. This platform is a genome browser and visualization tool of genomic and phenotypic data for both public and private datasets (https://xena.ucsc.edu/ (accessed on 1 September 2022)). An FDR cutoff of 0.05 was used for significance. Violin plots were generated to visualize expression values.

Through the aforementioned R2 platform, the expression levels of the previously mentioned genes which showed differential expression in regard to survival were analyzed. mRNA gene levels were compared in respect to pathologic staging and, given the small n associated with cases of stage 1 bladder urothelial cancer in the TCGA dataset (n = 2), Kruskal–Wallis analysis with corresponding pairwise Welch’s t-tests were used. Again, p-values were adjusted and an FDR cutoff of 0.05 was used.

2.4. Network & Enrichment Analysis

GeneMANIA (http://www.genemania.org (accessed on 1 September 2022)) serves as a platform for visualizing diverse biological interactions encompassing co-expression, co-localization, and domain similarity. In this study, GeneMANIA, R (https://www.r-project.org/ (accessed on 1 September 2022)), and the Cytoscape platform (https://cytoscape.org/ (accessed on 1 September 2022)) were utilized to construct a gene–gene interaction network focusing on EMT-related genes exhibiting significant differential expression based on survival data.

Subsequently, the network analysis highlighted certain genes alongside the aforementioned EMT genes. These genes underwent enrichment analysis to shed light on their potential involvement in specific biological processes, using annotations from the Gene Ontology (GO), Kyoto Encyclopedia of Genes, and Genomes (KEGG) databases. It was carried out through the Metascape platform (http://metascape.org (accessed on 1 October 2022)). The criteria for this analysis included a minimum overlap of 3 genes and an enrichment threshold of 1.5. Statistical significance was set at p < 0.05.

2.5. Tumor Immune Microenvironment Analysis

The relationship between the EMT-related genes differentially expressed in connection with survival and the immune system in cases of bladder urothelial cancer was examined through the TISIDB and TIMER platform. Spearman’s correlations between mRNA gene expression of the EMT genes and clinically relevant immunoinhibitor and cytokine gene expressions were calculated and visualized. Those with a rho of >|0.40| were considered meaningful with a p-value of >0.01. Devolution methods were used to estimate the immune infiltration of a wide variety of immune cells based on gene expression on the TIMER platform. The Spearman’s correlation was adjusted based on tumor purity, with a rho of >|0.30| visualized.

3. Results

3.1. Survival Analysis

A myriad of EMT-related genes showed differential expression in normal versus cancer tissues. Kaplan–Meier plots (Figure 1) were generated using TCGA data on “Bladder Urothelial Carcinoma” for each of the 1184 genes mined from the EMTdb, with only 32 meeting significant cutoff following FDR correction (

Table 1. Tabulated data from Kaplan–Meier plots based on mRNA expression of EMT-related genes which showed to be differentially expressed in regard to overall survival.

Gene	p-Value	Expression in Worse Prognosis
ADAM17	8.65 × 10−6	low
AGER	4.13 × 10−7	high
ANXA1	1.80 × 10−6	low
ARMC8	3.99 × 10−8	low
ART1	1.36 × 10−3	high
BBC3	5.28 × 10−6	high
CEMIP	1.83 × 10−5	low
ELSPBP1	1.65 × 10−3	high
FBP1	3.43 × 10−5	high
FN1	1.11 × 10−5	low
FOXA1	8.51 × 10−5	high
HOOK1	4.71 × 10−12	high
HTN1	1.02 × 10−3	high
IL22	4.18 × 10−4	high
INPP4B	2.37 × 10−5	high
LAMC2	1.98 × 10−9	low
LYPD3	3.33 × 10−5	low
MAP2K1	3.73 × 10−7	low
NES	2.91 × 10−9	low
NR2F2	1.34 × 10−6	low
NRP2	5.85 × 10−5	low
PDCD6IP	1.31 × 10−5	low
PEBP4	2.16 × 10−4	high
PRKCI	2.65 × 10−4	low
PTPN6	2.03 × 10−7	high
RUNX2	7.19 × 10−8	low
SCEL	2.27 × 10−7	low
SLC9A3R1	3.78 × 10−4	low
SOX3	3.92 × 10−4	high
SPRR2A	1.17 × 10−5	low
STIM2	2.11 × 10−11	high
TBX3	2.50 × 10−4	high

).

Genes with “high” mRNA expression leading to worse prognosis include those in

Table 2. List of genes and p-values for which high mRNA expression is correlated with worse prognosis.

Gene	p-Value
ADAM17	0.0182
ANXA1	0.0433
ARMC8	0.0009
CEMIP	0.0300
FN1	0.0443
LAMC2	0.0050
LYPD3	0.0461
MAP2K1	0.0419
NES	0.0413
NR2F2	0.0495
NRP2	0.0481
PDCD6IP	0.0311
PRKCI	0.0457
RUNX2	0.0007
SCEL	0.0018
SLC9A3R1	0.0467
SPRR2A	0.0342

.

On the other hand, genes which showed “low” mRNA expression leading to worse prognosis include those in

Table 3. List of genes and p-values for which low mRNA expression is correlated with worse prognosis.

Gene	p-Value
AGER	0.0441
ART1	0.0446
BBC3	0.0317
ELSPBP1	0.0486
FBP1	0.0359
FOXA1	0.0356
HOOK1	0.0011
HTN1	0.0478
IL22	0.0389
INPP4B	0.0314
PEBP4	0.0322
PTPN6	0.0137
SOX3	0.0324
STIM2	0.0500
TBX3	0.0482

.

3.2. Expression Analysis

Out of the 32 genes highlighted in the survival analysis, some of them also showed statistically significant expression levels when comparing normal versus tumor samples. This can be seen in the density plot (Figure 2) or in more detail in the sample violin plots (Figure 3); the rest of the genes and p-values from the violin plots in Supplemental Materials can be seen in

Table 4. Violin plot genes and p-values.

Gene	p-Value	Higher Expression
ADAM17	8.31 × 10−4	Primary tumor
AGER	2.77 × 10−2	Primary tumor
ANXA1	1.41 × 10−2	Normal tissue
ARMC8	2.35 × 10−4	Normal tissue
ART1	1.73 × 10−1	Normal tissue
BBC3	5.39 × 10−5	Primary tumor
CEMIP/KIAA1199	1.93 × 10−3	Primary tumor
ELSPBP1	2.52 × 10−1	Normal tissue
FBP1	7.76 × 10−1	Normal tissue
FN1	3.88 × 10−1	Primary tumor
FOXA1	1.64 × 10−1	Primary tumor
HOOK1	5.13 × 10−3	Primary tumor
IL22	9.30 × 10−2	Normal tissue
INPP4B	2.06 × 10−2	Primary tumor
LAMC2	1.05 × 10−4	Primary tumor
LYPD3	3.03 × 10−1	Primary tumor
MAP2K1	2.88 × 10−3	Primary tumor
NES	8.36 × 10−8	Normal tissue
NR2F2	3.04 × 10−1	Normal tissue
NRP2	1.24 × 10−4	Normal tissue
PDCD6IP	3.67 × 10−1	Primary tumor
PEBP4	2.17 × 10−4	Normal tissue
PRKCI	4.31 × 10−3	Primary tumor
PTPN6	3.14 × 10−3	Primary tumor
RUNX2	1.90 × 10−2	Primary tumor
SCEL	2.05 × 10−1	Primary tumor
SLC9A3R1	8.95 × 10−3	Primary tumor
SOX3	3.35 × 10−2	Normal tissue
SPRR2A	2.77 × 10−1	Primary tumor
STIM2	8.06 × 10−1	Normal tissue
TBX3	7.13 × 10−1	Primary tumor

. Some genes also showed significant differential expression based on tumor stage (Figure 4).

From the XENA browser, 19 normal tissues were compared with 407 TCGA cases. Following FDR correction (cut-off p-value is 0.003143), 10 genes were shown to be differentially expressed when considering overall survival (

Table 5. List of genes differentially expressed regarding overall survival when comparing normal versus tumor samples with p-values.

Gene	p-Value
ADAM17	0.0008
ARMC8	0.0002
BBC3	0.0001
CEMIP	0.0019
LAMC2	0.0001
MAP2K1	0.0029
NES	<0.0001
NRP2	0.0001
PEBP4	0.0002
PTPN6	0.0031

).

Several genes approach significance post-correction (

Table 6. List of genes that approach significance after FDR correction with p-values (cutoff p value is 0.003143).

Gene	p-Value
HOOK1	0.0051
HTN1	0.0069
SLC9A3R1	0.0090

).

Out of the same set of EMT-genes regarding survival, 10 out of the 32 showed differential expression when considering pathologic staging (

Table 7. List of stages that showed significantly different expression of the relevant gene based off TCGA datasets.

Gene	Different Stages
FN1	2 vs. 3 (p = 1.22 × 10−8), 2 vs. 4 (4.20 × 10−12)
NRP2	2 vs. 4 (p = 2.10 × 10−8)
FOXA1	1 vs. 3 (p = 5.08 × 10−3)
NES	2 vs. 3 (p = 1.71 × 10−5)
AGER	2 vs. 3 (p = 2.00 × 10−3), 2 vs. 4 (p = 2.39 × 10−5)
RUNX2	2 vs. 4 (p = 2.99 × 10−4)
PTPN6	2 vs. 4 (p = 7.24 × 10−3)
FBP1	2 vs. 3 (p = 5.65 × 10−3)
TBX3	2 vs. 3 (p = 3.56 × 10−3)
STIM2	2 vs. 4 (p = 3.97 × 10−5)

).

3.3. Identification of Further Gene Interactions and Enriched Biological Processes

3.3.1. Network Analysis

Network analysis was employed to identify genes with interconnected relationships, utilizing factors such as physical interactions, co-localization, and co-expression data (Figure 5). The genes chosen for constructing each network analysis were the previously mentioned EMT-associated genes that displayed distinct expression patterns in survival outcomes.

3.3.2. Enrichment Analysis

Enrichment analysis was executed on two distinct gene sets: first, on the genes pinpointed in the network analysis, and additionally, on the genes situated at the periphery (“outer rim”) of the network analysis, apart from the initial EMT genes (Figure 6). Enrichment analysis of the initial 32 highlighted EMT genes showed numerous pathways significantly overrepresented in the cohort of genes, such as endothelial cell migration and regulation of cell shape, in line with the known physiologic processes involved with the epithelial–mesenchymal transition. However, other less canonically associated ones such as defense response to bacterium and carbohydrate response were also uncovered through the analysis. Regarding the genes listed in the network analysis, similar biologic processes such as the VEGFA signaling pathway was highlighted alongside other related ones such as HIF-1 survival signaling.

3.4. Correlation to Inflammation Mediators

3.4.1. Immunomodulator, Cytokine

Through examining for correlation between various immunomodulators and chemokines (see Appendix A,

Table A1. Complete list of genes included as “immunomodulators”.

Immunomodulators	Chemokines	Chemokines
ADORA2A	CCL1	CX3CL1
BTLA	CCL2	CXCL1
CD160	CCL3	CXCL2
CD244	CCL4	CXCL3
CD274	CCL5	CXCL5
CD96	CCL7	CXCL6
CSF1R	CCL8	CXCL8
CTLA4	CCL11	CXCL9
HAVCR2	CCL13	CXCL10
IDO1	CCL14	CXCL11
IL10	CCL15	CXCL12
IL10RB	CCL16	CXCL13
KDR	CCL17	CXCL14
KIR2DL1	CCL18	CXCL16
KIR2DL3	CCL19	CXCL17
LAG3	CCL20	XCL1
LGALS9	CCL21	XCL2
PDCD1	CCL22
PDCD1LG2	CCL23
PVRL2	CCL24
TGFB1	CCL25
TGFBR1	CCL26
TIGIT	CCL27
VTCN1	CCL28

for list), the following plots were generated (Figure 7), with the statistically and clinically significant ones (p < 0.05 post correction, |rho| > 0.4) being shown. Out of the 32 genes, 12 were shown to have at least significant correlation with an immunomodulator or chemokines (TBX3, NRP2, FN1, FOXA1, FBP1, ANXA1, LAMC2, HOOK1, NES, PTPN6, RUNX2), with the first four having at least 25 different immunomodulators or cytokines to be significantly correlated with (

Table 8. Summary of Spearman’s correlation data between EMT-related gene and various immunomodulators, grouped by EMT-gene of interest.

Gene	Immunomodulator (Positive Correlation)	Immunomodulator (Negative Correlation)
TBX3		CXCL16, CXCL13, CXCL11, CXCL10, CXCL9, CXCL5, CXCL3, CXCL2, CXCL1, CCL26, CCL23, CCL18, CCL13, CCL8, CCL7, CCL5, CCL4, CCL3, TIGIT, TGFBR1, PDCD1LG2, PDCD1, LAG3, IL10, IDO1, HAVCR2, CTLA4, CSF1R, CD274
NRP2	CXCL13, CXCL12, CXCL11, CXCL10, CXCL9, CXCL2, CCL26, CCL23, CCL21, CCL19, CCL18, CCL13, CCL11, CCL8, CCL7, CCL5, CCL4, CCL3, CCL2, LAG3, TIGIT, TGFBR1, PDCD1LG2, PDCD1, IL10, HAVCR2, CTLA4, CSF1R, BTLA, ADORA2A
FN1	TGFBR1, TGFB1, PDCD1LG2, LAG3, IL10, HAVCR2, CSF1R, CD274, CXCL13, CXCL12, CXCL11, CXCL10, CXCL9, CXCL5, CXCL2, CCL26, CCL23, CCL21, CCL18, CCL13, CCL11, CCL7, CCL5, CCL4, CCL3, CCL2
FOXA1	CCL15	CXCL12, CXCL11, CXCL10, CXCL9, CXCL5, CXCL3, CXCL2, CCL26, CCL23, CCL21, CCL18, CCL13, CCL8, CCL7, CCL5, CCL4, CCL3, CCL2, TGFBR1, TGFB1, PDCD1LG2, LAG3, IL10, HAVCR2, CTLA4, CSF1R, CD274
FBP1	CCL15	CCL4, TGFBR1, PDCD1LG2, CD274
ANXA1	PDCD1LG2, CD274, CCL7
SPRR2A	CXCL8, TGFB1
LAMC2	CXCL8, CXCL1, TGFB1
HOOK1	TGFB1, CSF1R, CCL23
NES	CXCL12, KDR
PTPN6	LGAGLS9
RUNX2	PDCD1LG2

).

3.4.2. Immune Cell Infiltrate

Through the TIMER platform, deconvolution methods were used to estimate the amount of various immune cells (T cells CD4+, Tregs, B cells, Neutrophils, Monocytes, Macrophages, DCs, NK and Mast Cells; see Figure 8). Out of the 32 initial genes, 14 (ADAM17, AGER, ANXA1, ARMC8, FBP1, FN1, FOXA1, LAMC2, MAP2K1, NRP2, PTPN6, RUNX2, STIM2, TBX3) showed to have at least significant correlation (Spearman’s correlation > 0.05 following adjustment based on tumor purity, with a rho of >|0.30|; see

Table 9. Summary of Spearman’s correlation data between EMT-related gene and various immune cell types, grouped by EMT-gene of interest.

Gene	Immune Infiltrates (Positive Correlation)	Immune Infiltrates (Negative Correlation)
ADAM17	Macrophage, Neutrophil
AGER		T cell CD8+
ANX1	T cell CD8+, Neutrophil, Myeloid dendritic cell
ARMC8	Macrophage
FBP1		T cell CD8+, Neutrophil, Myeloid dendritic cell
FN1	T cell CD4+, T cell CD8+, Macrophage, Myeloid dendritic cell
FOXA1	T cell CD4+	T cell CD8+, Myeloid dendritic cell
LAMC2	T cell CD8+, Neutrophil, Myeloid dendritic cell
MAP2K1	T cell CD8+, Neutrophil
NRP2	T cell CD8+, Macrophage, Myeloid dendritic cell
PTPN6		B cell
RUNX2	T cell CD8+, Myeloid dendritic cell
STIM2	T cell CD4+
TBX3		T cell CD8+, T cell CD4+, Neutrophil, Myeloid dendritic cell

).

4. Discussion

Overall, we see a robust cohort of EMT-related genes that are differentially expressed as pertaining to survival in bladder urothelial carcinoma. RUNX2, a gene most associated with cartilage production, has been linked to pancreatic cancer and to breast cancer progression through modulation of MicroRNAs and the metastasis-associated 1 (MTA1)/NuRD complex [38,39]. By activating the Wnt signaling pathway, ARMC8 has been linked to increased invasion in cutaneous squamous cell carcinoma and lung cancer [40,41,42]. Also implicated in the Wnt pathway, as well as the VEGF pathway, CEMIP (formerly known as KIA1199) has emerged via immunohistochemical studies as a possible biomarker for a variety of cancers [43,44,45]. The phosphatase INPP4B is an inhibitor of the Wnt pathway; its knockout has been linked to increased proliferation [46]. Knockout or downregulation of the transcription factor gene FOXA1 has also been linked to worse prognosis, altering the carcinogenic activity of the Snail/Twist1 axis in breast cancer as well as prostate cancer [47,48]. TBX3 has been linked to breast and cervical cancer proliferation, but it also inhibits the activity of the YAP/TAZ signaling pathway involved with cellular regeneration and growth [49,50,51].

Of note, PTPN6, a tyrosine phosphatase, affects cell growth and carcinogenesis in both bladder and colon cancer. Interestingly, overexpression of PTPN6 has been associated with worse prognosis and increased metastasis, while the opposite was seen in other studies [52,53]. Additionally, there were cases where our findings were incongruent with other cancers. PEBP4 expression was correlated with metastasis in colorectal and breast cancer but is significantly associated with better outcomes in our analysis. AGER (advanced glycosylation end-product-specific receptor) was shown to be associated with increased cell migration in cervical cancer, but the opposite as per our analysis [54]. The pro-carcinogenic effect that ART1 has in colorectal cancer was not seen in our Bladder TCGA analysis [55].

Interestingly enough, some of the genes shown to be differentially expressed in regard to survival were not differentially expressed when comparing normal tissue to tumor; nor did their expression correlate to pathologic staging. For the pathologic staging, only FN1, NRP2, FOXA1, NES, AGER, RUNX2, PTPN6, FBP1, TBX3, and STIM2 were shown to be significantly different.

While there may be differences in Stage I expression versus the other stages, our sample size (n = 2) was too small to reliably detect any except for in the expression levels of FOXA1. Instead, many of the differences were seen between Stage 2 versus 3 and 4. Histopathologically, this correlates with the expansion of the cancer through the muscle with possible lymph node and systemic metastasis. During this process, the pro-mobility changes which accompany the EMT would be fully evident. Our findings were mostly consistent with the survival analysis results, such as with RUNX2 which showed worse prognosis with higher levels, increasing expression from Stages 1 to 4.

Through the network analysis, we were able to see more genes through which the EMT-related genes can modulate and affect carcinogenesis. For example, VEGFA was indicated, a well-known member of a family of growth factors which has proliferative and anti-apoptotic effects [56]. It has been found to be highly expressed in hepatocellular carcinoma (HCC) and triple-negative breast cancer (TNBC). It is associated with worse prognosis in HCC [57] and significantly lower progression-free survival in TNBC treated with chemotherapy [58]. Additionally, SFRP2 was implicated, and recent immunohistochemistry work has linked this gene as a possible early biomarker of pancreatic and colon cancer [59,60]. Another gene highlighted in our analysis, CAV1, codes for a scaffolding protein and often predicts poor prognosis [61].

Enrichment analysis of the EMT-related genes further highlights the multiple modalities through which the EMT process itself encourages cancer growth and metastasis. Included among the highlighted cellular processes in our analysis are those regarding the establishment or maintenance of epithelial cell apical/basal polarity and positive regulation of cell migration. The VEGFA-VEGFR2 signaling pathway and associated angiogenesis, among the seminal hallmarks of cancer, were highlighted as the most overly enriched biologic process (p < 5.0 × 10⁻⁷). Other biological processes less typically associated with EMT such as “defense response to bacterium” and “signaling by interleukins” were picked up by the analysis. However, the same cytokine and inflammatory-mediated response to bacteria can help explain another way EMT-related genes play a role in cancer progression. The increased angiogenic and metabolic changes associated with certain proinflammatory states, which would help fend off a bacterial infection, can serve as fertile ground for initiation of cancer metastasis.

Among the most salient advances in our understanding of cancer progression is the complex role the immune system plays. Notable mediators of the immune system repeatedly occurred in our analysis. TGFBR1, a receptor for the TBFB growth factor, and CSF1R, a cytokine receptor, have been linked to pro-survival activity [62,63,64,65]. In other cancers, the chemokine CCL15 has been linked to tumor-associated macrophage recruitment and CCL18 has been linked to an immunosuppressive tumor microenvironment, allowing for evasion of the host’s immune system [66,67,68,69]. The most common immune cell type with a significant correlation was CD4+ T cells. While classically thought of as having an anti-tumor role, the varied subtypes not completely covered in our current deconvolution methods may play a paradoxical pro-tumor role. It is possible they do so by decreasing activation of other immune cells such as macrophages, warranting further study into the specific gene to immune cell interaction [70].

While this research represents a step forward in investigating the expression patterns of EMT-related genes, there exist certain limitations to these discoveries. These assessments rely on the levels of mRNA in a stable state as a proxy for protein levels. To partially address this, proteomic analysis of the highlighted genes in our study were examined in bladder cancer models through the depmap.org portal (https://depmap.org/portal/ (accessed on 12 September 2023)) and Cancer Cell Line Encyclopedia (CCLE), verifying expression of highlighted genes. Moreover, the functional behavior of these mRNA molecules could be subject to additional regulation by post-translational elements. Additionally, the techniques used to decipher immune cell infiltrations might be influenced by the tumor’s purity, contingent upon the specific formula employed. Ongoing efforts are dedicated to refining methods for accurately estimating cellular infiltrations based on mRNA levels. Overall, this study potentially serves as a starting point for future investigations aimed at directly scrutinizing the highlighted relationship between gene expression and prognosis, extending to the protein or enzymatic realm.