MicroRNAs in Pancreatic Cancer: Advances in Biomarker Discovery and Therapeutic Implications

Pancreatic cancer remains a formidable malignancy characterized by high mortality rates, primarily attributable to late-stage diagnosis and a dearth of effective therapeutic interventions. The identification of reliable biomarkers holds paramount importance in enhancing early detection, prognostic evaluation, and targeted treatment modalities. Small non-coding RNAs, particularly microRNAs, have emerged as promising candidates for pancreatic cancer biomarkers in recent years. In this review, we delve into the evolving role of cellular and circulating miRNAs, including exosomal miRNAs, in the diagnosis, prognosis, and therapeutic targeting of pancreatic cancer. Drawing upon the latest research advancements in omics data-driven biomarker discovery, we also perform a case study using public datasets and address commonly identified research discrepancies, challenges, and limitations. Lastly, we discuss analytical approaches that integrate multimodal analyses incorporating clinical and molecular features, presenting new insights into identifying robust miRNA-centric biomarkers.


Introduction
Pancreatic cancer (PC) stands as one of the most lethal malignancies, with a mere 9% five-year survival rate.Typically diagnosed in advanced stages, treatment options remain limited, with only 15-20% of cases amenable to resection upon diagnosis, leading to a dismal prognosis [1].The critical necessity for early detection of pancreatic cancer is undeniable yet hindered by the absence of specific symptoms and effective screening methodologies.Current diagnostic methods, exemplified by the best-validated PC biomarker, carbohydrate antigen 19-9 (CA19-9), exhibit insufficient sensitivity and specificity for early screening, underscoring the urgent need for innovative approaches [2].
It is well noted that miRNAs may act as oncogenes, named oncomiRs, in some types of cancers but tumor suppressors in others, depending on their specific expressions and the cellular context.In pancreatic cancer, in addition to those reviewed in [21,22], newly reported tumor suppressor miRNAs include miR-634 [23], miR-1225 [24], miR-623 [25], miR-345-5p [26], and miR-30c [27], while oncomiRs include miR-194 [28] and miR-221 [29].Their altered expression patterns in pancreatic cancer versus controls make them promising diagnostic and prognostic indicators [9,10,[30][31][32][33].Despite ongoing efforts [34][35][36][37], the roles of miRNA in the etiology and progression of pancreatic cancer are often not fully studied.This is partially attributed to the lengthy and challenging process of determining miRNA functions, which requires reliable target identification through computational predictions and extensive experimental validations.Given that a single miRNA can simultaneously target multiple genes, while a single gene can be targeted by multiple miRNAs, which also varies in conditions, modeling such complex miRNA-gene interactions should be supported by a sophisticated systematic network analysis [38,39].
The miRNA expression profiles obtained from various biopsy methods offer valuable insights into pancreatic tumor molecular characteristics, serving as a comprehensive pool of potential biomarkers for clinical use.Aberrant miRNA expression patterns in pancreatic cancer tissues and biofluids, analyzed through techniques like PCR, a microarray, and small RNA sequencing (sRNA-seq), hold promise for distinguishing cancer patients from healthy individuals [9,33,40,41].For instance, elevated expressions of tissue miR-21, miR-155, and miR-451 have demonstrated promising diagnostic value in pancreatic cancer [20,42].
To provide an overview of the most recent research progress on miRNA biomarkers in pancreatic cancer, we screened the titles and abstracts of over 900 articles published in PubMed between 2014 and 2024 that include the keywords "microRNA, biomarker, and pancreatic cancer".We focus on miRNAs identified in human pancreatic tissues and biofluids with diagnostic and prognostic potential by large omics-based studies or with therapeutic implications verified by both in vitro and in vivo experiments.In addition, we conducted a case study utilizing public miRNA expression datasets collected from the Gene Expression Omnibus (GEO) repository (https://www.ncbi.nlm.nih.gov/geo/, 1 August 2023).The discussion focuses on significant achievements, lessons learned, potential limitations and challenges, and possible solutions by identifying reliable combinatory biomarkers.Specifically, with a focus on circulating exosomal miRNAs, the aim is to elucidate their potential for non-invasive liquid biopsy in early PC screening alongside new approaches focusing on multimodal analysis and mechanistic discovery.
Furthermore, incorporating miRNA with other circulating markers has further improved diagnostic sensitivity [40].For instance, Nakamura et al. demonstrated an enhanced diagnostic accuracy, from 74-89% to 91%, when combining pancreatic juice-derived exosomal miRNAs, miR-21, and miR-155 with pancreatic juice cytology (PJC) for differentiating PDAC from CP patients [60].Combining serum miR-21-5p with human satellite II (HSATII) RNA level can also improve the diagnostic performance (with an AUC of 0.90 compared to 0.87) by using one marker alone [56].
Overall, current evidence underscores the potential of exosomal miRNAs as noninvasive, high-sensitivity biomarkers for the early detection and monitoring of pancreatic cancer.

Prognostic Potential of miRNAs in Pancreatic Cancer
The expression levels of specific miRNAs have been found to correlate with cancer progression, treatment response, and patient survival in pancreatic cancers.Numerous studies have identified miRNA signatures associated with the disease, facilitating the stratification of cancer patients.As summarized in Table 2, elevated levels of serum miR-451a [47], plasma miR-221-3p [29], exosomal miR-21 [64], and cyst fluid EV-derived miR-200b were found in pancreatic cancer patients compared to healthy controls.The increased expression levels of these miRNAs are linked to poorer outcomes, recurrences, and advanced disease stages.For instance, an elevated expression of miR-221-3p was found in PC tissues, cell lines, and plasma, and the expression level of plasma miR-221-3p was correlated with distant metastasis and TNM stages [29].Similarly, an upregulation of plasma miR-370-3p expression correlates with poor prognosis in PDAC patients (hazard ratios (HR) 2.13, p = 0.004) [84].The overexpression of tissue miR-31 and miR-205 was significantly correlated with higher histological grades and decreased survival in TCGA PDAC clinical samples [85].In another study [86], miR-200b is present at a higher level in EVs from mucinous pancreatic cystic neoplasms' (M-PCNs) EVs compared to EVs isolated from other pancreatic cysts (not-M samples).Its expression discriminates patients with these two groups with a higher sensitivity (83.3%) and specificity (90.9%) than other diagnostic methods.Conversely, decreased levels of tissue miR-132 [30], miR-7 [32], and miR-26a-5p [87] have been observed in pancreatic cancer patients compared to healthy controls.The expression of these miRNAs is associated with tumor progression and metastasis.In [30], it shows that the 1-year survival rate of patients with high miR-132 expression, a tumor suppressor, was greater than that of patients with low miR-132 expression, and the lower expression was associated with poorer prognosis in PDAC.Similarly, low miR-7 expression may contribute to tumor progression and poor prognosis in pancreatic cancer [32].The elevated expression of miR-26a-5p, an oncomiR, was found to inhibit PDAC cell proliferation, migration, and invasion by targeting ARNTL2, which is correlated to a favorable prognosis [87].

Therapeutic Implications of miRNAs in Pancreatic Cancers
In addition to their diagnostic and prognostic roles, miRNAs show promise for targeted therapies in pancreatic cancer.For example, the upregulation of miR-103 in PDAC leads to increased tumor metastasis and a poor prognosis through regulating target USP10 (ubiquitin-specific peptidase 10) [90].MiRNA-based therapies involve the restoration or inhibition of specific miRNAs to modulate oncogenic or tumor-suppressive pathways.Clinical trials evaluating miRNA-targeted therapies are underway, providing a novel avenue for personalized treatment.
In a review by Tesfaye et al. [22], the aberrant expression of miRNAs and their target genes and pathways in PDAC, such as oncogene KRAS and tumor suppressor genes like p53, p16, and SMAD4, were examined.Key observations included the loss of tumor suppressive miRNAs leading to an overexpression of oncogenes (let-7, miR-34, and miR-200), miRNAs conferring resistance to chemotherapy (miR-21 and miR-221/222), and the sensitization of cells to gemcitabine upon restoration of tumor suppressor miRNAs (miR-34).Similarly, Amartya et al. reviewed the potential therapeutic implication of miRNA with respect to hallmarks in PDAC [21].In Table 3, we summarize a list of miRNAs dysregulated in PDAC whose therapeutic potential have been verified by both in vitro and in vivo experiments.
Several miRNAs were overexpressed in PDAC samples and cells compared with normal controls, including miR-21, miR-27, miR-194-5p, and miR-29.They play pivotal roles in regulating key signaling and cellular pathways in PDAC.An elevated level of miR-21 was shown to promote drug resistance to 5-FU and enhance the proliferation of pancreatic cancer cells by downregulating the expression of its target genes phosphatase and tensin homolog (PTEN) and programmed cell death factor 4 (PDCD4) [91].It also promotes EGF-induced cell proliferation by targeting Sprouty RTK Signaling Antagonist 2 (Spry2) [92].Exosomal miRNA-27a derived from pancreatic cancer cells promotes the angiogenesis of human microvascular endothelial cells in pancreatic cancer via regulating BTG2 [93].MiR-194-5p downregulates tumor cell PD-L1 expression and promotes antitumor immunity in pancreatic cancer [28].The upregulation of miR-29a was found to be positively correlated with metastasis, and its ectopic expression increased the expression of pro-inflammatory factors, epithelial-mesenchymal transition (EMT) markers, and cell viability and invasion by downregulating Tristetraprolin (TTP).
In contrast, the downregulation of a few tumor-suppressing miRNAs was observed in PDAC cells.The ectopic expression of miR-15a was also found to cause cell cycle arrest and inhibit PDAC cell proliferation and EMT via the downregulation of Bmi-1 expression [94].MiR-145 can act as a suppressor to slow pancreatic cancer progression by suppressing cancer cell proliferation, migration, and invasion through regulating different targets, including MUC13, the TGF-β receptor, and SMAD2 [95,96].MiR-34a inhibits the migration and invasion of pancreatic cancer cell lines through regulating EMT and Notch signaling via targets Snail1 and Notch1 [97].
Epigenetic mechanisms were also noted to regulate miRNA expression [22].More recently, Zheng et al. demonstrated the oncogenic role of a circular RNA derived from exons of myoferlin (MYOF), named circMYOF, in PDAC [98].CircMYOF was upregulated in PDAC tissues and cell lines, acted as a sponge of miR-4739, and regulated its target vascular endothelial growth factor A (VEGFA), leading to the activation of PI3K/AKT signaling and increased aerobic glycolysis (Warburg effect).The circMYOF/miR-4739/VEGFA axis facilitates PDAC progression and may serve as a prognostic biomarker and therapeutic target.A similar observation was made regarding the circ-membrane-bound O-acyltransferase domain containing two (circ-MBOAT2)/miR-433-3p/glutamic-oxaloacetic transaminase 1 (GOT1) axis [99].In pancreatic cancer tissues and cells, circ-MBOAT2 and GOT1 expression were significantly upregulated, while miR-433-3p expression was downregulated.Circ-MBOAT2 acted as a sponge of miR-433-3p, which was associated with GOT1 and repressed cell proliferation, migration, invasion, and glutamine catabolism in pancreatic cancer.

Challenges, Solutions, and Future Directions
While miRNAs from various sources show immense promise as biomarkers in pancreatic cancers, several challenges persist.The standardization of sample collection, isolation methods, and data analysis remains critical for these endeavors.Previous research, as reviewed in the preceding sections, has predominantly focused on exploring aberrant miRNA expression using microarray or sequencing profiling alongside bioinformatics analysis to identify key potential biomarkers.However, several issues were revealed.

Discrepancies in Reported Biomarkers
As evidenced in Tables 1 and 2, different studies using distinct cohorts have reported divergent biomarkers, even based on similar statistical analysis.To further assess this issue, we conducted a comparative analysis using three datasets of tissue miRNA expression profiles and three datasets of serum miRNA expression profiles from TCGA and GEO, as summarized in Table 4.Our differential expression analysis using the Limma package revealed several key observations, as follows.

•
In tissue, there are 250, 93, and 83 differentially expressed (DE) miRNAs identified in each of the three datasets, respectively, with a fold change >2 and p-value < 0.05 (detailed lists in Supplementary Materials (Table S1)).
No single common DE miRNAs were found among all three datasets.Two, nine, and twenty-two DE miRNAs were common among 2/3 datasets.

•
In serum, 238, 122, and 203 DE miRNAs are identified in each of the three datasets, respectively, using less stringent criteria (Table S2).
Two common DE miRNAs were found among all three datasets, namely, miR-1246 and miR-1290.Eleven, fourteen, and twenty-six DE miRNAs were common among 2/3 datasets.
• Tissue vs. serum groups: 117 miRNAs were differentially expressed in one or more datasets of each group (Table S3).
No single miRNA is common in all six datasets.MiR-1246 is common in all serum datasets and 2/3 tissue datasets, while miR-205-5p is common in 2/3 of each group.However, the dysregulation trend is not consistent.
It is well noted that the discrepancies in methodology, patient cohorts, and clinical settings across studies likely contribute to these inconsistencies.The inherent heterogeneity of pancreatic cancer further complicates the identification of universal biomarkers, as tumors can exhibit diverse molecular subtypes, and patient responses to treatment can vary significantly.The inconsistency of the DE miRNAs could also suggest that the miRNAs' role in pancreatic cancer may be more complex than initially thought, possibly influenced by various biological mechanisms and clinical factors.Thus, a single or small set of miRNAs may not adequately capture this heterogeneity, necessitating the exploration of composite biomarkers.
Moving forward, efforts should focus on the following.

2.
Exploring associated clinical metadata to understand how variables such as disease subtypes, stages, previous treatments, or comorbidities influence miRNA expression patterns.
Integrating functional analysis to predict gene targets and associated pathways, enhancing the relevance of miRNAs to pancreatic cancer.

Discrepancies in Analytical Methods
Advancements in computational and statistical methods have enabled the analysis of large-scale miRNA expression data for cancer detection.Logistic regression, LASSO regression, and several machine learning models, such as Support Vector Machine and Decision Tree Classifier, have been applied to identify potential miRNA signatures [42,59,74].However, given the complexity and heterogeneity of pancreatic cancer, the high dimensionality and noise inherent in miRNA expression data demand careful consideration.
One of the foremost challenges is selecting the most relevant miRNAs (features) from the vast pool.Feature selection techniques are critical for identifying informative miRNAs that discriminate between cancer and normal samples.However, striking a balance between identifying the most discriminatory miRNAs while avoiding overfitting and ensuring generalization is a delicate task.Developing robust feature selection algorithms that effectively filter out noise and irrelevant miRNAs while capturing critical ones contributing to the disease phenotype is imperative [102].
Another challenge arises from the need for more well-annotated and sufficiently large datasets.While machine learning algorithms, particularly deep learning models, thrive on big data to learn complex patterns effectively, obtaining large and well-curated miRNA expression datasets for pancreatic cancer remains challenging due to the rarity of the disease and associated difficulties in sample collection and data sharing.Limited data can lead to model overfitting and reduced generalizability.
Therefore, it is crucial to develop machine learning models that can effectively handle small datasets and extract meaningful information.These models must be designed to account for variability and be robust enough to accommodate the diverse miRNA expression patterns across different patient groups.Another challenge lies in interpreting the biological significance of selected miRNA biomarkers.Machine learning models often operate as 'black boxes', making it elusive to understand why specific miRNAs were chosen as biomarkers.Developing methods that predict biomarkers and provide insights into their functional roles and underlying biological mechanisms can enhance the credibility and clinical relevance of the identified biomarkers [54,98].
To illustrate a standard approach to identify miRNA signatures, we conducted a case study using a public dataset.This study involved the following.

•
The utilization of example datasets from GSE59856, comprising serum miRNA expression data of 100 pancreatic cancer patients and 150 healthy controls.

•
The use of Principal Component Analysis (PCA) for dimensionality reduction and visualization to understand the distribution of normal healthy and pancreatic cancer samples.

•
The employment of the Limma package for differential gene expression analysis, aiding in the identification of top circulating miRNA biomarkers differentiating cancer patients from healthy controls.

•
The construction of binary cancer classification models using several machine learning algorithms, including SVM, Decision Tree Classifier, Linear Discriminant Analysis (LDA), Quadratic Discriminant Analysis (QDA), and Gaussian Naive Bayes.The models' performance was evaluated using k-fold cross-validation and standardized evaluation metrics such as accuracy, recall, precision, and F1 score.
As detailed in the Supplementary Materials (Table S4 and Figures S1 and S2), the distinct sample distributions between pancreatic cancer patients and healthy controls, as visualized through PCA, underscore the potential of these miRNAs to distinguish between these two groups.Regarding feature selection, SVM outperformed other models in the classification task and obtained the best model.After examining signatures of different sizes, a combinatory signature comprising miR-125a-3p, miR-6893-5p, miR-125b-1-3p, miR-6075, and miR-4294 was identified, demonstrating the best diagnostic accuracy of 97.59% and an AUC of 99.70%.These findings again highlight the potential of miRNAs as diagnostic biomarkers for pancreatic cancer.
In conclusion, identifying robust and efficient combinatory miRNA biomarkers in pancreatic cancer using feature selection and machine learning is a multifaceted challenge, which requires addressing feature selection, heterogeneity, dataset size, and biological interpretation issues.Addressing these challenges will advance the field towards more accurate diagnostic and prognostic tools, paving the way for personalized approaches to managing pancreatic cancer.

Alternative Approaches
While circulating exosomal miRNAs hold the promises, there are still challenges in handling exosome extraction, purification, and qualification in terms of their small RNA content, making it difficult to profile exosomal miRNA expression in pancreatic cancer tissues and biofluids.Studying the molecular properties of exosomal miRNAs may aid in overcoming these challenges and lead to discoveries in this field [103].For instance, by performing motif analysis using MDS 2 [104] among 880 exosomal miRNA sequences extracted from the ExoCarta database [105], we identified a few conserved sequence motifs for exosomal miRNAs (details in the Supplementary Materials (Table S5)).The motifs' coverage, information content, and p-value were determined, revealing specific motifs highly enriched in miRNAs associated with human exosomes.These motifs may play a crucial role in packaging miRNAs into human exosomes, guided by RNA-binding transporter proteins [106].Considering exosomal miRNA packaging may be a cell-specific process [103], future analysis can focus on exosomal miRNAs detected in specific pancreatic cancer tissue, cells, or biofluids.Understanding these sequence motifs could provide insights into exosomal miRNA packaging mechanisms and guide more target biomarker searches.MiRNAs associated with such motifs are more likely present in the exosomes, even in a low abundance in the early stage of the disease.However, further research is warranted to validate these findings and explore the mechanisms underlying these miRNAs' roles in pancreatic cancer pathogenesis.

Conclusions
MiRNAs and exosomal miRNAs have emerged as promising biomarkers for pancreatic cancer diagnosis, prognosis, and targeted therapy.Their unique features, such as stability and accessibility through liquid biopsies, highlight their potential to transform the landscape of pancreatic cancer management.The results of prior research have provided compelling evidence for the potential of circulating miRNAs as non-invasion biomarkers for pancreatic cancer detection.Continued research and validation efforts are essential to unlock the full clinical potential of these biomarkers.
Despite the promising findings, several challenges must be addressed before miRNAs can be routinely used in clinical practice for pancreatic cancer management.Standardization of sample collection, RNA isolation, and miRNA profiling methods are essential to ensure reproducibility and comparability of the results between different studies.Additionally, large-scale prospective studies are needed to validate serum miRNAs' clinical utility and incorporation into existing diagnostic and prognostic models.New paradigms to explore combinatory signatures by integrating miRNAs with other diagnostics modalities, including clinical information or mechanistic features, enable the identification of markers with enhanced sensitivity and specificity of pancreatic cancer detection, ultimately improving patient prognosis and survival rates.Moreover, specific miRNAs' functional roles and regulatory mechanisms in pancreatic cancer pathogenesis need further exploration.
In conclusion, circulating miRNAs hold great promise as non-invasive biomarkers for pancreatic cancer, offering potential improvements in early detection, prognosis, and treatment monitoring.As research in this field advances, circulating miRNAs may become valuable tools in the fight against this deadly disease.

Table 1 .
MiRNAs as diagnostic biomarkers in pancreatic cancer tissue, fluids, and exosomes.

Table 2 .
MiRNAs as prognostic biomarkers in pancreatic cancer tissue, fluids, and exosomes.

Table 3 .
MiRNAs as potential targets for therapies in PDAC.

Table 4 .
Overview of major research discrepancies, challenges, and solutions illustrated by the case study.