Implications of prognosis-associated genes in pancreatic tumor metastasis: lessons from global studies in bioinformatics

Pancreatic cancer (PC) is a highly lethal malignancy with a 5-year survival rate of 10%. The occurrence of metastasis, among other hallmarks, is the main contributor to its poor prognosis. Consequently, the elucidation of metastatic genes involved in the aggressive nature of the disease and its poor prognosis will result in the development of new treatment modalities for improved management of PC. There is a deep interest in understanding underlying disease pathology, identifying key prognostic genes, and genes associated with metastasis. Computational approaches, which have become increasingly relevant over the last decade, are commonly used to explore such interests. This review aims to address global studies that have employed global approaches to identify prognostic and metastatic genes, while highlighting their methods and limitations. A panel of 48 prognostic genes were identified across these studies, but only five, including ANLN, ARNTL2, PLAU, TOP2A, and VCAN, were validated in multiple studies and associated with metastasis. Their association with metastasis has been further explored here, and the implications of these genes in the metastatic cascade have been interpreted.


Introduction
With an incidence rate that has increased approximately 1% each year for the past 20 years, PC is currently ranked the 11 th most common cancer worldwide [1]. This year alone, the numbers of diagnoses and deaths attributed to this disease in the USA are projected to exceed 60,000 and 48,000, respectively. In fact, due to the consistently increasing incidence rate of PC, it is expected to surpass breast cancer as the third leading cause of cancer deaths by 2025 [2]. Currently, the stage-wide 5-year survival rate for PC is 10%, which is one of the lowest among all major cancers, demonstrating the need for novel detection and treatment modalities [3].
While the causes of PC are not completely understood, genetic factors, such as DNA mutations, have emerged as key factors in PC etiology. Genetic alterations can be hereditary, leading to disorders such as Lynch syndrome and Peutz-Jeghers syndrome. Further, lifestyle or environment, such as smoking, could cause mutations leading to the development of various malignancies. Additionally, the presence of certain benign conditions increases the risk for the development of pancreatic cancer. Diabetes mellitus and pancreatitis are two diseases that are commonly seen in patients prior to the development of PC [1,4]. Interestingly, late-onset diabetes mellitus is also observed in individuals who have previously been diagnosed with PC, indicating a bi-directional association between these diseases [5]. However, understanding beyond PC etiology is urgently needed to develop effective interventions for improved diagnosis and treatment.
Over the last decade, treatment strategies have been actively exploring pathways associated with chemoresistance, metastasis, hypoxia, and immunosuppression for PC targeting. Further efforts are being made to explore the underlying heterogeneity of pancreatic tumors and target pathways associated with distinct subtypes [6][7][8]. Each of these components contributes to the recalcitrant nature of all major cancers, including PC, and explored therapies have been designed to take advantage of cell surface markers, metabolism, stroma, the immune system, and important signaling pathways, among other factors. However, we have yet to develop effective therapies which target the major attributes, specifically metastasis, which are responsible for the aggressive nature of PC.

Gaps in the understanding of pancreatic tumor metastasis
Tumor metastasis is one of the main characteristics attributed to poor prognosis in PC. Due to its asymptomatic nature, PC is commonly diagnosed at late stage, when the cancer is locally advanced or metastasized, and treatment options are limited. In these aggressive tumors, excessive desmoplasia is often observed, and the metastases are primarily found in the liver, lung, and peritoneum [9]. Unfortunately, metastatic cases account for approximately 50% of new PC diagnoses, and their 5-year survival rate is a dismal 3% [3,10]. Despite these facts, there is not yet a standard targeted therapy for metastasis, partly due to our limited understanding of systemic progression models and the molecular events that underlie the metastatic cascade. The metastatic cascade consists of several fundamental processes, including invasion of the basement membrane, intravasation, survival in the circulation, extravasation, and colonization at secondary sites [11]. As cells progress through each of these steps, they undergo selective pressures, ensuring that only the fittest cells colonize the secondary tumor sites [12]. It begins when epithelial cells undergo a series of mutations that give them the metastatic capacity to invade the basement membrane, move across the extracellular matrix (ECM), and intravasate into the bloodstream. Once in the bloodstream, these metastatic tumor cells, or circulating tumor cells (CTCs), must acclimatize to the new environment and evade immune surveillance [13][14][15]. CTCs that have survived in the bloodstream and reached a distant organ, then referred to as disseminated tumor cells (DTCs), arrest and receive cues from the distant organ for the formation of micrometastases [16,17]. While only a fraction of these CTCs and DTCs survive the selection process to form distant secondary tumors, their speculated role in recurrent cancer and drug resistance further implicates them in disease lethality [18,19]. Moreover, this tumor cell motility, and consequently their metastasis, is a consequence of the dysregulation of cell adhesion molecules. This is highlighted by the long-standing hallmark of cancer that loss of E-cadherin expression is correlated with the migratory and invasive potential of the tumor cells both under in vitro and in vivo conditions [20].
There are currently two models that are commonly used to describe the metastatic process: the linear progression model and the parallel progression model [21]. The linear progression model, which is more widely recognized, proposes that the dissemination of tumor cells and the development of metastases occurs during late-stage primary tumor progression. The seeding cells (clones) from a late-stage tumor will harbor advanced mutations and growth potential, thereby allowing minimal genetic deviations between the primary tumor and metastases [22,23]. Alternatively, the parallel progression model proposes early dissemination of tumor cells during the first stages of tumor progression. These early-stage cells have had limited genetic progression, and so the clones from this population are less genetically advanced compared to their linear counterpart. Moreover, early dissemination allows the primary tumor and metastases to evolve separately from a less advanced stage, leading to a higher level of genetic divergence between them [22,23]. Although no direct evidence exists in support of either model, sequencing data and animal models of various cancer metastases, namely those of the liver, lung, and peritoneum, have provided indirect corroboration for both models, indicating that metastasis may not only be a cancerspecific but also a case-specific phenomenon [23].
The primary driver gene mutations of PC, including KRAS, TP53, p16/CDKN2A, and SMAD4, have a crucial role in early pancreatic lesions, local and advanced tumors, metastasis, and the hypovascular and hypoxic nature of the stroma, which contributes to the evasion of the immune response and alterations in cellular metabolism [24][25][26]. Tiwari et al. demonstrated that in PC, hypoxia-inducible factor 1 alpha (HIF1α) acts as a tumor suppressor by suppressing the expression of protein phosphatase 1 regulatory subunit 1B (PP1R1B), leading to the degradation of p53 protein in pancreatic cancer cells and an increase in the invasive and metastatic activity of tumors cells [27]. Another study found that methyltransferase-like 14 (METTL14) upregulation decreases p53 apoptosis effector related to PMP-22 (PERP) expression mediated through m 6 A modification and promotes pancreatic cancer metastasis [28]. Exosomes, which have become invaluable in cancer research, play an essential role in tumor initiation and the formation of extracellular signalosomes, which influence tumor microenvironment remodeling [29]. PC exosomes have also been shown to transport nucleic acids, proteins, or lipids from parental to recipient cells; produce pro-inflammatory cues; and facilitate immunosuppression, anti-apoptosis leading to angiogenesis, proliferation, and tumor metastasis. More specifically, exosomes produced by cancerassociated fibroblasts (CAFs), tumor-associated macrophages (TAMs), cancer-initiating cells (CICs), and pancreatic stem cells (PSCs) have a diverse potential of cellular functionalities such as growth, proliferation, drug resistance, epithelial-mesenchymal transition (EMT), migration, invasion, and metastasis [30].
Intriguingly, many investigators believe that pancreatic cancer metastasis is one of the primary causes of death. Yet, few studies envisage the molecular mechanism of the tumor cell journey to distant organs. While many questions regarding the metastatic process remain elusive, our knowledge base thus far has demonstrated the importance of metastasis in disease lethality and the critical nature of developing targeted metastasis therapies. In other words, recent evidence indicates that metastasis is strongly associated with poor prognosis, and the targeting of these metastatic pathways may lead to improved patient outcomes [31]. Indeed, studies in breast, prostate, esophageal, and liver cancer have established a correlation between prognosis and the occurrence and progression of metastasis [32][33][34][35][36][37][38]. Specifically, the cell functions which are characteristic of metastasis, which include proliferation, angiogenesis, migration, and invasion, among others, correlate with prognostic gene signatures [32]. While there is no direct evidence of this relationship in PC, the aforementioned studies support the supposition that there may be an association between the stages of metastatic progression or the metastatic cascade and poor prognosis in PC. For these reasons, there has been extensive research on the mechanism of metastasis and prospective drug targets. Current targeted therapies, including the epidermal growth factor receptor (EGFR) inhibitor erlotinib and the neurotrophic tyrosine receptor kinase (NTRK) inhibitors larotrectinib and entrectinib, are given to advancedstage PC patients in combination with standard of care drugs to slow the progression of the tumor. This may stall local invasion, which occurs prior to the dissemination of cancer cells. However, these metastasis-inhibiting drugs are not always effective, and they do not serve in a neoadjuvant capacity [39].
Alongside our efforts to develop therapies that target the quintessential characteristics of these tumors, there has been an extensive search for prognostic markers, markers of metastasis, and associated biological pathways of PC. The expectation is that establishing a panel of prognostic genes for PC, specifically those associated with progression and metastasis, could lead to the identification of novel pathways for therapeutic targeting. In recent years, there has been a consensus that computational approaches in bioinformatics may be employed to better understand metastatic cascades and genes implicated in the aggressive nature of PC. In this review, we aim to address such studies and assess their findings while highlighting the merits and disadvantages of these approaches.

Computational tools employed for identification of metastasisand prognosis-associated genes
In recent years, computational approaches used to analyze gene data have become invaluable tools in cancer research.
Identifying key genes involved in tumorigenesis, tumor pathogenesis, or the hallmarks of cancer, among other classifications, is crucial to our developing knowledge of the molecular mechanisms of cancer and our search for effective treatment targets. For particularly lethal cancers, such as PC, these computational methodologies are especially indispensable, as they may shed light on the genes which may be used for early diagnosis or play pivotal roles in the progression of the disease [40,41].
There are numerous methods employed to determine differentially expressed genes between the phenotypes included in microarray datasets. The use of statistical software packages, such as Linear Models for Microarray Analysis (limma), GEO2R, and Weighted Gene Coexpression Network Analysis (WGCNA) in R programming language, is among the most common techniques used for this purpose. While identifying similar gene networks, limma determines the genes which can be attributed to each phenotype using a linear model, and WGCNA defines the differential co-expression networks between the phenotypes to identify differentially expressed genes (DEGs) [42]. GEO2R is a limma-based method also used for the analyses of differentially expressed pathways. These statistical packages are advantageous because they do not require prior command-line proficiency, making them a widely accessible resource.
Meta-analysis approaches utilized to analyze differentially expressed genes (DEGs), or subsets of genes identified from microarrays, are also highly diverse. A few of the most common resources include Gene Ontology (GO), Kyoto Encyclopedia of Genes and Genomes (KEGG), Gene Set Enrichment Analysis (GSEA), and the Database for Annotation, Visualization and Integrated Discovery (DAVID). These knowledge bases contain integrated analysis methods which utilize pathway databases such as KEGG, Reactome, BioCarta, and PANTHER to perform functional enrichment analysis. This identifies which biological processes, signaling pathways, or molecular functions are enriched in the genes of interest and demonstrates how they affect other genes and pathways to infer downstream effects. However, the best method for pathway analysis is highly dependent on the needs of the study and the statistical sensitivity required. For example, while DAVID is a highly efficient and comprehensive tool for gene set analysis, it is prone to false positives in pathway analysis. Alternatively, GSEA is arguably the least biased method for determining which pathways most significantly encompass the input genes, though it may also have limited sensitivity in gene set analysis [43]. Another interesting aspect of pathway analysis is that most methods integrate multiple resources and databases in order to have the most inclusive analysis; only a select few, such as KEGG pathway analysis and GSEA, have a single primary data source.
Protein-protein interaction networks and hub genes are also commonly explored in studies that utilize bioinformatics. Tools such as Cytoscape, Metascape, cBioPortal, and the Search Tool for the Retrieval of Interacting Genes/Proteins (STRING) allow for the analysis of genomics data, visualization of interactome networks of input genes, and identification of highly interconnected hub genes across given gene sets. Multiple databases comprising network and annotation data are integrated to support these tools, which helps elucidate genes or biological processes that may play a vital role in the phenotype of interest in a given dataset. Unlike the other tools, however, the STRING database incorporates both known and predicted interactions, and the type of interaction between genes is annotated in the analysis.

Prognostic and metastatic markers as identified by computational approaches
The poor prognosis of PC is often associated with the presence of metastases in the lymph nodes and distant secondary sites. With that in mind, it is unsurprising that a number of the prognostic genes recognized by recent studies have been associated with the stages of the metastatic cascade in PC and other cancer types. Specifically, our review of all the literature has led to the collection of 48 prognostic genes, many of which are associated with various aspects of cancer progression. Among these genes, a set of five genes were differentially altered across multiple studies and implicated in the metastatic progression of PC. These include anillin actin-binding protein (ANLN), DNA topoisomerase II alpha (TOP2A), urokinase-type plasminogen activator (PLAU), versican (VCAN), and aryl hydrocarbon receptor nuclear translocator-like 2 (ARNTL2). Nearly a dozen studies in the last 5 years have utilized bioinformatics analyses to identify genes associated with PC prognosis [44][45][46][47][48][49][50][51][52][53][54]. A recent study conducted by Luo et al. aimed to determine the molecular signatures for PDAC progression and a survival score to predict PDAC prognosis. Using PDAC data retrieved from GEO datasets GSE28735, GSE62452, and GSE57495, and DEGs, key genes associated with PDAC tumors, their association with prognosis, and their clinical significance were determined by assessment of WGCNA and miRNA profile survival analysis in R programming language. Kaplan-Meier survival analysis of metastasis data obtained from the International Cancer Genome Consortium (ICGC) demonstrated an association between several DEGs and metastasis. Further GO enrichment analysis, KEGG pathway analysis, and GSEA presented the biological processes, molecular functions, cellular components, and pathways that are enriched in these DEGs. Several genes were associated with poor prognosis, leading to a 7-gene signature which could accurately predict PDAC prognosis and metastasis: ARNTL2, desmoglein 3 (DSG3), protein tyrosine phosphatase receptor type R (PTPRR), ANLN, S100 calcium-binding protein A14 (S100A14), ankyrin repeat domain 22 (ANKRD22), and tetraspanin 7 (TSPAN7) [44]. The detailed annotation of these genes in PC progression and metastasis is included in Table 1).
Intriguingly, ANLN has been shown to play a role in the promotion of EMT in lung adenocarcinoma and cell-cell adhesion, migration, and invasion in PC [55,56], indicating a potential role of ANLN in various stages of the metastatic cascade, including invasion of the basement membrane, intravasation, extravasation, and colonization at secondary sites ( Table 2). While EMT is commonly recognized as a central aspect of cancer metastasis, it is the loss, rather than promotion, of cell adhesions that are associated with cancer development. However, the collective migration of cancer cells, a process where two or more cells whose cell-cell junctions are intact move together into nearby tissues and vasculature, could be one explanation for ANLN-mediated cell-cell adhesion [57,58]. As this type of migration would require remodeling of the extracellular matrix (ECM) to accommodate the movement of these cell groups, and these changes would incite cell motility through integrins, the enriched pathways support this supposition for ECM disassembly, ECM organization, collagen catabolic process, integrin binding, and cell migration, which have been observed in PC DEGs.
The metastatic significance of ARNTL2 expression has also been explored in lung adenocarcinoma and PC. In lung adenocarcinoma, ARNTL2 expression is an important factor in the survival of DTCs and CTCs and metastatic seeding [81]. In a similar manner, its expression in PC has been shown to positively regulate the TGF-β signaling pathway, which has been observed to promote metastasis when expressed in tumors and is known to promote tumor progression in PC [82,85]. ANTL2 is also observed to promote cell focal adhesion in PC, which contributes to cell dissociation from the primary tumor and reattachment to the ECM for invasion and intravasation [82,86]. These findings indicate that ARNTL2 may be expressed throughout the entire metastatic cascade, including invasion of the basement membrane, intravasation, survival in the circulation, extravasation to secondary sites, and colonization at secondary sites ( Table 2). The enriched pathways for integrin binding, laminin-binding, ECM disassembly, and PI3K-AKT signaling pathway in PC further emphasize the metastatic involvement of ARNTL2, as they are associated with invasion and metastasis, and there is a link between TGF-β and PI3K-AKT signaling pathways, demonstrated by crosstalk in cancer [85,87].
Studies have also aimed to delineate the genes associated with PC initiation, progression, and prognosis. Jin Table 1 Recent studies that utilized computational methods to identify key genes in PC   [81][82][83][84] et al. designed a study to identify genes responsible for the molecular mechanism of PC tumorigenesis and proliferation. DEGs were determined from GEO datasets GSE32676, GSE15471, and GSE71989 using GEO2R. DAVID and the STRING database were used to classify the enriched biological processes, molecular functions, cellular components, and pathways, while the PPI network for the DEGs was built using Cytoscape. Hub genes and their mode of regulation were also identified using Cytoscape, and their association with poor survival was determined by constructing Kaplan-Meier curves via cBioPortal. Further, expression changes throughout disease progression were explored for each DEG using Oncomine, which demonstrated increased expression of cyclin-dependent kinase 1 (CDK1) and centrosomal protein 55 (CEP55) as PC progresses. They identified 10 hub genes that were associated with decreased survival: ANLN, assembly factor for spindle microtubules (ASPM), CDK1, CEP55, denticleless E3 ubiquitin protein ligase homolog (DTL), epithelial cell transforming 2 (ECT2), NIMA-related kinase 2 (NEK2), TOP2A, and protein regulator of cytokinesis 1 (PRC1) [45] (Table 1). Like ANLN, TOP2A positively regulates invasion and migration and promotes EMT in lung adenocarcinoma, and in PC, TOP2A expression gives rise to enhanced cell proliferation, migration, and EMT [63,64]. These downstream effects of TOP2A expression, along with the enhanced expression of TOP2A observed in metastatic luminal breast cancer and prostate cancer, indicate a potential role in all five stages of the metastatic cascade [88][89][90][91] (Table 2). This is further reflected in the upregulated pathways observed in PC DEGs, comprising integrin binding, laminin-binding, collagen catabolic process, and cell migration, as integrins are key molecules for migration, the proteolytic breakdown of collagen is important in an invasion, and laminin and collagen-binding are significant in intravasation and extravasation.
Xu et al. recently investigated potential prognostic genes and the molecular mechanisms of PC metastasis. A total of 109 DEGs in metastatic PC were determined from GEO datasets GSE19279, GSE42952, and GSE71729 using the limma package in the R programming language. Annotation of function, pathway analysis, PPI network analysis, and prognostic analysis of these genes were performed using DAVID, the STRING database, Cytoscape, and the GEPIA analysis tool. Decreased survival was significantly associated with the low expression of four DEGs, including secretogranin V (SCG5), crystallin beta A2 (CRYBA2), carboxypeptidase E (CPE), and chromogranin B (CHGB), indicating that the abnormal regulation of these genes in cancer may impact overall survival and the metastatic properties of PC tumors [46] (Table 1).
In recent years, seven studies explored potential gene targets of PC and genes associated with prognosis and progression of the disease using similar techniques. The interest of a study by Chen et al. lay in the use of integrated bioinformatics to identify genes involved in PC tumorigenesis. DEGs were screened from GEO datasets GSE15471, GSE16515, and GSE6245 using GEO2R. The functional significance, enriched pathways, and PPI network for the common DEGs from the three datasets were determined using DAVID, the STRING database, and Cytoscape. Cytoscape also allowed for the identification of 24 main hub genes, and their association with poor survival was determined through cBioPortal Kaplan-Meier curves. The authors identified four genes with differential survival: TOP2A, periostin (POSTN), PLAU, and VCAN ( Table 1). The ROC curves of these genes identified a significant area under the curve (AUC), indicating that they may also have diagnostic potential. VCAN was determined as a relatively novel marker for PC progression and further exploration using Oncomine and the R2 Genomics Analysis and Visualization Platform, an online genomics analysis tool, suggested that VCAN expression may play an important role in PC response to chemotherapy treatment [47].
Overexpression of VCAN has been observed in numerous cancers and their respective metastases but has only been associated with the invasion and motility of breast cancer, prostate cancer, PC, and melanoma, and invasion and migration in renal cell carcinoma [75][76][77][78][79]. As a major component of the ECM, its influence on cell adhesion, cell migration, and cell invasion is logical. Further, as invasion is a key aspect of the metastatic process, it can be surmised from these findings that VCAN may be involved in the invasion of the basement membrane, intravasation, extravasation to secondary sites, and colonization at secondary sites, which are all stages of the metastatic cascade that depend on the invasive nature of cancer cells (Table 2).
It is plausible that PLAU may function as a promoter of metastasis at the beginning and end of the metastatic cascade. The primary function of its protein product, uPA is involved in remodeling ECM, which in breast and cervical cancer enhances cancer cell motility and ability to invade the basement membrane and migrate, thereby promoting metastasis through proteolytic destruction of ECM [92,93]. Intravasation and extravasation are similarly associated with EMT and ECM degradation, while colonization at secondary sites is associated with the motility and invasion of cancer cells (Table 2). ECM disassembly and protein activation cascade, which were enriched pathways of PC DEGs, highlight the function of PLAU and its potential role in the metastatic cascade.
Wu et al. further explored genes that may be associated with PC prognosis. Unlike many other studies, only a single GEO dataset, GSE62165, was used to identify DEGs with the limma package in R programming language. While this limits the sample number, the use of one large size dataset is beneficial in that normalization is not needed, and the available clinical study data is uniform. Clusterprofiler in R programming language classified the enriched biological processes, molecular functions, cellular components, and pathways for DEGs, and their PPI network was constructed using the STRING database. Cytoscape and UAL-CAN, an online tool for analyzing omics data, were used to identify 18 core genes and their association with survival. They found two genes, PLAU and collagen type XVII alpha 1 chain (COL17A1), associated with poor prognosis [48] ( Table 1).
In a similar manner, Zhou et al. investigated biomarkers and prognostic targets for PDAC. As microRNAs (miRNAs) have been previously shown to have diagnostic and prognostic potential in PC, DEGs and differentially expressed miRNAs were determined from GEO datasets GSE41368, GSE43795, GSE55643, and GSE41369, which were analyzed using GEO2R. Commonly identified DEGs from all four GEO datasets were further analyzed using Metascape, GO analysis, KEGG pathway analysis, the STRING database, Cytoscape, and FunRich, a tool for analyzing omics data, for functional enrichment. The functional analysis of these DEGs helped determine their associated biological processes, molecular functions, cellular components, enriched pathways, and PPI networks. Hub genes were identified using Cytoscape, and survival analysis of these genes was performed using the KM plotter. miRNAs were similarly analyzed, and the overlap between the miRNA gene targets and DEGs revealed seven hub genes, including proto-oncogene c-Myc (MYC), solute carrier family 2 member 1 (SLC2A1), pyruvate kinase M1/2 (PKM), PLAU, peroxisome proliferator-activated receptor gamma (PPARG), proto-oncogene c-Met (MET), and integrin subunit alpha 3 (ITGA3), which may be associated with poor prognosis, and whose miRNA regulators are associated with EMT and the PI3K-AKT and MAPK/ERK signaling pathways, among other metastatic processes [49] (Table 1).
In a study by Lu et al., hub genes that could be used as targets in PC diagnosis and treatment strategies were elucidated. The limma package in R programming language was used to determine DEGs from GEO datasets GSE15471, GSE19650, GSE32676, and GSE71989. DAVID, KEGG Orthology Based Annotation System (KOBAS) enrichment, the STRING database, and Cytoscape were used to classify the enriched biological processes, molecular functions, cellular components, and pathways for the DEGs, as well as their PPI network. Hub genes and their association with survival were identified and explored using Cytoscape and UALCAN analysis. Five hub genes were found to be associated with decreased survival: MET, maternal embryonic leucine zipper kinase (MELK), syndecan 1 (SDC1), thrombospondin 1 (THBS1), and TOP2A [50] (Table 1).
Unlike prior studies, Wu et al. developed a prognostic signature and nomogram, a statistical model used for risk prediction, which could be used to predict overall survival in PC. The limma package in R programming language was used to analyze GEO datasets GSE71729, GSE62165, GSE62452, GSE28735, GSE15471, GSE16515, and GSE32676 to identify DEGs. They further explored enriched biological processes through GO enrichment and KEGG pathway analysis. Signaling pathways and the PPI network of the DEGs were identified with DAVID and the STRING database. Cytoscape was used to identify hub genes of the PPI network, and their prognostic potential was explored and tested through Cox regression analysis, Kaplan-Meier analysis, ROC curves, and Harrell's concordance index. Nine prognostic genes, including MET, kallikrein-related peptidase 10 (KLK10), COL17A1, CEP55, ANKRD22, integrin subunit beta 6 (ITGB6), ARNTL2, mucolipin TRP cation channel 3 (MCOLN3), and solute carrier family 25 member 45 (SLC25A45), were identified from these analyses [51] (Table 1).
Li et al. used computational approaches to identify therapeutic targets for PC and gain insight into the underlying molecular mechanisms of PDAC using a bioinformatics approach. To determine DEGs, GEO dataset GSE28735 was analyzed using GEO2R. The biological processes, molecular functions, cellular components, and pathway enrichment were classified using DAVID. The mRNA expression of the top DEGs and their association with survival were determined using boxplot analysis and survival analysis from the GEPIA web tool, and identified 20 key hub genes from the PPI network of DEGs using the STRING database. Four genes were associated with the tumor stage, including solute carrier family 6 member 14 (SLC6A14), polypeptide N-acetylgalactosaminyltransferase 5 (GALNT5), tetraspanin 1 (TSPAN1), and islet amyloid polypeptide (IAPP). Excluding IAPP, which was associated with a favorable prognosis, these genes were also associated with poor prognosis [53] ( Table 1).
Novel hub genes and pathways which can be utilized to diagnose, predict the prognosis of, or treat PDAC were recently explored in a study by Lu et al. For determining hub genes, GEO datasets GSE62452, GSE15471, GSE102238, GSE16515, and GSE62165 were analyzed using GEO2R. A total of 21 core upregulated hub genes were consistently present in all five GEO datasets. They were further explored using DAVID, the STRING database, Cytoscape, and OncoLnc, a tool for analyzing survival data and correlated RNA expression, to elucidate their biological function, enriched pathways, PPI network, and survival association. Six core hub genes were associated with decreased survival and may be potential clinical markers: integrin subunit alpha 2 (ITGA2), matrix metallopeptidase 7 (MMP7), integrin subunit beta 4 (ITGB4), ITGA3, VCAN, and PLAU [54] ( Table 1).

Significant pathways in PC associated with newly identified prognostic signature and metastasis
Several significantly enriched pathways were identified among all the studies aimed to identify prognostic, therapeutic, or metastatic genes. For the studies which performed GO enrichment analysis for combined upregulated and downregulated DEGs, enrichment was commonly depicted in the biological processes and molecular functions associated with ECM maintenance and cell adhesion and migration [47,[49][50][51]. Interestingly, one study performed clustering analysis on DEGs prior to enrichment analysis, highlighting additional biological pathways, including blood vessel development, vasculature development, smooth muscle development, and cell junction assembly [51]. Likewise, the studies with combined DEGs had similar terms returned from KEGG pathway analysis, which described the most significant enrichment in ECM-receptor interaction, focal adhesion, and the PI3K-AKT signaling pathway [47,[49][50][51].
Receptor tyrosine kinases (RTKs) are membrane-bound receptors that initiate signaling events upon binding to growth factors, hormones, cytokines, neurotrophic factors, and, pertinently, extracellular signaling molecules and ECM components. RTKs act in conserved pathways that involve signaling events of cellular proliferation, differentiation, survival, and migration in cancer [94]. RTK signaling is tightly regulated under normal conditions but can be aberrantly dysregulated upon oncogenic insults. For example, the enhanced signaling of epidermal growth factor receptor (EGFR), one of the members of the ERBB family of RTKs comprising ERBB1 (EGFR), ERBB2, ERBB3, and ERBB4, is implicated in the development of many solid tumors [94]. RTKs exert their signaling by auto-and transphosphorylation of their intracellular C-terminal region. This can activate many important signaling pathways, including PI3K-AKT, RAS/ RAF/MAPK, JAK-STAT, and PLC-γ1, which govern cancer cell events and functionalities such as proliferation, metabolism, angiogenesis, progression, and survival [95][96][97].
Numerous studies have elucidated the role of EGFR in the early and late stages of pancreatic cancer progression. Though KRAS activation is a prerequisite for the initial stages of the progression and is found to be mutated in 90% of PC patients, EGFR activity is also important for inducing PC progression through MEK/ERK activity [98]. The challenges in combating PC are majorly due to recurrence, metastatic events, and drug resistance attributed to cancer stem cell populations residing in the tumor bulk. It was recently shown that afatinib, a pan-EGFR inhibitor, reduces SOX9, a key molecule in maintaining stem cell populations. As the EGFR/ERK/FOXA2/SOX9 axis regulates pancreatic cancer stem cells (PCSCs), inhibition of EGFR hampers the growth and motility of PCSCs mediated through this axis [99]. Provided that EGFR signaling activates many pathways associated with carcinogenesis and metastasis, such as MAPK, PI3K, and JAK-STAT, it is no surprise that it has made an attractive target for cancer therapy. Various studies have described small molecule EGFR inhibitors, including erlotinib, gefitinib, afatinib, and osimertinib (AZD9291) [100,101]. In addition, specific monoclonal antibodies against EGFR such as cetuximab and panitumumab are FDA approved for various cancer indications [102,103]. Among small molecule inhibitors, erlotinib is FDA approved and is used in clinics in combination with gemcitabine for treating PC patients with local, advanced, unresectable, or metastatic tumors [104].
The consensus of GO enrichment analysis terms for upregulated DEGs among studies that performed separate analyses for each set of DEGs included pathways for biological process concerning ECM maintenance, collagen catabolic process, cell migration, and cell adhesion. Similarly, enrichment in the molecular function category for upregulated DEGs included ECM structure, collagen binding, integrin binding, cell adhesion molecule binding, and cadherin binding involved in cell-cell adhesion. In support of this, cellular component enrichment for upregulated DEGs was found in the ECM and cell junctions [44-46, 48, 52-54]. Further, these features play key roles in migration and EMT, which are predominantly associated with metastasis. Fig. 1 Schematic of metastatic cascade in PC. Multiple in silico studies have identified prognostic and metastatic gene sets in PC. By reviewing all the literature, we found a total of 48 prognostic genes, and further identified a set of five genes which were observed in multiple studies and found to be implicated in the metastatic progression of PC. These includes anillin actin-binding protein (ANLN), DNA topoisomerase II alpha (TOP2A), urokinase-type plasminogen activator (PLAU), versican (VCAN), and aryl hydrocarbon receptor nuclear translocator-like 2 (ARNTL2). The invasion, metastasis, and migration of tumor cells are common characteristics influenced by these genes, implicating them in the early and late stages of the metastatic cascade. ANLN is exclusively associated with the collective migration of tumor cells to secondary sites through the promotion of cellcell adhesions, and ANTL2 and TOP2A are uniquely associated with tumor cell survival while in circulation PC metastasis is mediated through a culmination of environmentally derived cell-intrinsic and cell-extrinsic cues, which give tumor cells the ability to migrate from the primary tumor to distant organs. Extrinsic cues, including paracrine and autocrine mechanisms, enable these tumor cells to reach the destined organ for colonization [105]. Extrinsic cues can also cause phenotypic changes in neoplastic cancer cells, which must be attained in order to gain motility and the ability to cross the hurdles of physical restrictions and vasculature. EMT is the trans-differentiation cellular process where epithelial cells acquire a mesenchymal phenotype via a series of biochemical changes induced by several growth factors, including TGF-β, HGF, EGF, IGF, and FGF [106,107]. In the prerequisite phenotype, cells lose their epithelial characteristics (markers like E-cadherin, occludin, claudin, and laminin-1) and switch to mesenchymal traits (N-cadherin, vimentin, and fibronectin), causing the dysregulation of cell-cell contacts and the dissociation of cells from the epithelial layer [108]. Many transcriptional regulators can regulate this mesenchymal phenotype. TWIST, SNAIL1, SNAIL2, ZEB1, and ZEB2 repress E-cadherin expression and activate mesenchymal differentiation markers such as N-cadherin and vimentin, cellular matrix and focal adhesion proteins, and matrix metalloproteinases involved in promoting motility [109][110][111]. Additionally, calreticulin, a calciumbinding endoplasmic reticulum protein known to have various cellular roles, including that of a chaperone, has been shown to promote EMT through the Integrin/EGFR-ERK/ MAPK axis in PC [112][113][114]. Longping Go-Ichi-Ni-San 2, an oncogene, is known to have a role in clinically advanced stages of PC through activation of the ERK/MAPK signaling pathway, which might mediate EMT. The reverse process of EMT, a concept called mesenchymal to epithelial transition (MET), occurs once metastatic tumor cells reach distant organs and colonize, and utilize many of the same pathways [24,[115][116][117]. This stresses the importance of genes and pathways involved in the process of EMT and points to their potential as therapeutic targets for metastasis.
Integrins belong to the cell adhesions family and comprise 24 αβ heterodimers formed from different α and β subunits. Integrin-β1, for example, associates with multiple α subunits to create 12 receptors for ECM components like collagen, laminin, and fibronectin rich in arginine-glycineaspartic acid [118,119]. They are the integral receptors that mediate cell adhesion and function as a mechanotransmitter for oncogenic and metastatic signals. Their interaction with the ECM involves the organization of the cytoskeleton and relaying intracellular signals from the ECM to regulate survival, proliferation, migration, and EMT, among other cell fate transitions [120,121]. Integrin-facilitated cell adhesion to the extracellular matrix is highly controlled and, upon dysregulation, causes pathogenesis. In the case of pancreatic cancer, this elicits phenotypes and signaling pathways conducive to tumor growth and migration [122]. Further, it has been shown that integrin-β1 is involved in ERK signaling, and its inhibition decreases KRAS signaling in PC Fig. 2 Meta-analysis of prognostic genes identified across multiple studies. Among the 48 computationally-derived prognostic genes, 21 were found to play a role in processes involved in metastasis, including adhesion, invasion, ECM regulation, and migration in PC. Interestingly, several of these genes, including integrin subunit alpha 2 (ITGA2), integrin subunit alpha 3 (ITGA3), thrombospondin 1 (THBS1), syndecan 1 (SDC1), matrix metallopeptidase 7 (MMP7), periostin (POSTN), proto-oncogene c-Met (MET), versican (VCAN), NIMA related kinase 2 (NEK2), and peroxisome proliferator activated receptor gamma (PPARG), were observed to play a role in a number of these processes, indicating that they may also be important in the metastatic cascade cells lacking ECM attachment [123,124]. Integrin-β1 has also been implicated in the activation of PI3K signaling in PC [122]. The significance of integrins in cancer is further implicated by integrin-β8, which plays a vital role in PC cell radiochemoresistance, intracellular vesicle trafficking, and autophagy upon irradiation [125].
For downregulated DEGs, GO pathway enrichment for the biological process category included those related to digestion and proteolysis, organismal homeostasis, collagen maintenance, and ECM maintenance. The molecular function category for downregulated DEGs mainly encompassed peptidase and lipase activity, and cell adhesion. Enrichment of downregulated DEGs in the cellular component category was found in the ECM, vesicles, endoplasmic reticulum lumen, and, uniquely, platelet alpha granules [44,46,48,[52][53][54].
KEGG pathway analysis of upregulated and downregulated DEGs revealed differential enrichment of upregulated DEGs in the interleukin-17 signaling pathway and the PPAR signaling pathway. Pathways differentially enriched in downregulated DEGs, as determined by KEGG pathway analysis, included pancreatic secretion, and complement and coagulation cascade pathways [44-46, 48, 52-54]. The most significantly enriched pathway for upregulated DEGs, and most common across studies, was the PI3K-AKT signaling pathway.
In PC, nearly 59% of patients have elevated PI3K-AKT signaling. This pathway is normally regulated by phosphatase and tensin homolog (PTEN), a natural antagonist of PI3K, but this tumor suppressor is often lost in cancer [126,127]. Similarly, about 60% of PC patients have increased expression of AKT2, which acts as an oncogene that presides over many cellular processes, including survival. AKT2 is the major downstream effector for the PI3K and RTK pathways [126,128,129]. PI3K/AKT/mTOR elicits signaling events responsible for regulating many essential cellular processes, including cell growth, metabolism, survival, metastasis, and resistance to chemotherapy [130]. It has also been established that this pathway plays a vital role in angiogenesis, macrophage transcriptional reprogramming, T cell differentiation, tumor cell homeostasis, and fibroblastsupported chemoresistance, apoptosis, invasion, tumorigenesis, and EMT [128,[131][132][133][134]. Further, PI3K signaling in stromal cells modulates the surrounding microenvironment, creating a space conducive for metastatic events. Its dysregulation leads to oncogenic signals, which involve changes in proliferation, migration, and immune modulations [135][136][137][138]. Unsurprisingly, the frequent amplification, mutation, or loss of key PI3K/AKT/mTOR regulators in many solid cancers has made it an attractive therapeutic target [139,140]. Everolimus, an mTOR inhibitor, has been studied in numerous PC clinical trials. However, poor efficacy has been observed for everolimus alone and in combination with various standard of care therapies. It is postulated that the identification of drugs that are synergistic with everolimus may result in more successful responses [141].
Remarkably, many of the enriched pathways for the upregulated genes were associated with invasion, metastasis, tumorigenesis, and angiogenesis. In contrast, enriched pathways for downregulated genes were concerned with cellular maintenance, homeostasis, and inhibition of tumor progression. Inflammation and immune response pathways, tumor cell adhesion and motility pathways, lipid uptake and processing pathways, and the PI3K-AKT signaling pathway were enriched for both upregulated and downregulated genes. The upregulation of genes involved in invasion, metastasis, and angiogenesis falls in line with the expected aggressive nature of PC. The presence of shared immune pathways, cell movement, and energy pathways between these upregulated and downregulated genes may point to the regulatory role of the tumor microenvironment or indicate that they are complexly regulated throughout pancreatic tumorigenesis and tumor progression.

Similarities and limitations of study design
Due to the limited availability of publicly available microarray datasets, many studies utilized at least one common dataset to define a prognostic signature. The data availability is even less for the studies interested in metastatic genes. Interestingly, while the use of these datasets led to the identification of similar upregulated and downregulated pathways across studies, the pipeline employed by each group was unique, and therefore, very few genes overlapped across each group's prognostic gene set. However, multiple members belonging to the same gene families were identified, and DEGs had a moderate degree of similarity between studies. It is important to note that despite the distinct pipeline of each study, the methodologies across them are comparable. While R programming language was commonly used to identify DEGs in many of these studies, DAVID, the STRING database, and Cytoscape were heavily relied on to discover the biological functions, enriched pathways, and the PPI network of DEGs. Remarkably, GSEA was used to identify the enriched pathways of DEGs in only one study, possibly due to the low sensitivity of this analysis. Moreover, methods like STRING and GO enrichment analysis rely on experimental annotations and those derived electronically. While this increases the number of annotations available for analysis, it can lead to the assignment of incorrect ones, as electronic annotations are not as accurate as their experimental counterparts [142]. Unfortunately, there is currently no way to limit analyses to experimental annotations, and unlike STRING, GO enrichment analysis does not distinguish between the two in the output. Further verification of these annotations is often needed to determine their true functional significance.
The key difference among these studies is how DEGs were grouped for analysis. Four of these studies performed pathway analyses on DEGs as one group, while seven studies divided DEGs into upregulated and downregulated categories. Though a holistic look at the pathway enrichment of all DEGs can be insightful and elucidate which pathways are altered in pancreatic tumors as compared to the normal pancreas, the scope of this analysis is limited. When considering all DEGs, we are unable to determine whether increased or decreased expression is associated with the altered pathways. The separate analysis of upregulated and downregulated genes allows us to better understand whether the pathway changes are due to increased or decreased expression of specific genes. In other words, with differentiation of upregulated and downregulated DEGs, we can recognize with greater clarity the molecular basis of important pathway changes between normal pancreas and pancreatic tumors.

Conclusion
The use of integrated bioinformatics has allowed us to identify key genes common in many malignancies. In PC, these methodologies have aided in determining genes associated with prognosis, tumorigenesis, and, as recently shown, metastasis. The knowledge acquired from these studies will help develop biomarkers and drugs specifically targeting these processes and lead to better disease management. Further, PC prognosis is often dependent on the presence of metastases. Metastasis is a multi-stage process; we must understand which stage has the best potential for targeting. Bioinformatics analysis provides special insight into the pathobiology and the stage-specific expression of metastatic genes, but further elucidation is needed to take advantage of the metastatic cascade in a clinical setting. Future studies will likely incorporate these computational approaches to accomplish this in PC and other cancers.
Funding The authors of this work were supported, in parts, by the NIH grants (P01 CA217798, R01 CA183459, CA210637, R01 CA228524, U01 CA200466, U01 CA210240, and R44CA224619), LB506 award to SK from the Nebraska State Department of Health & Human Services, NRI collaboration initiative grant to SK from University of Nebraska)

Declarations
Conflict of interest SKB is one of the co-founders of Sanguine Diagnostics and Therapeutics, Inc. The other authors declare no competing interests.
Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article's Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http:// creat iveco mmons. org/ licen ses/ by/4. 0/.