Exosomal miRNA Biomarker Panel for Pancreatic Ductal Adenocarcinoma Detection in Patient Plasma: A Pilot Study

Pancreatic ductal adenocarcinoma (PDAC) is rapidly becoming one of the leading causes of cancer-related deaths in the United States, and with its high mortality rate, there is a pressing need to develop sensitive and robust methods for detection. Exosomal biomarker panels provide a promising avenue for PDAC screening since exosomes are highly stable and easily harvested from body fluids. PDAC-associated miRNAs packaged within these exosomes could be used as diagnostic markers. We analyzed a series of 18 candidate miRNAs via RT-qPCR to identify the differentially expressed miRNAs (p < 0.05, t-test) between plasma exosomes harvested from PDAC patients and control patients. From this analysis, we propose a four-marker panel consisting of miR-93-5p, miR-339-3p, miR-425-5p, and miR-425-3p with an area under the curve (AUC) of the receiver operator characteristic curve (ROC) of 0.885 with a sensitivity of 80% and a specificity of 94.7%, which is comparable to the CA19-9 standard PDAC marker diagnostic.


Introduction
Pancreatic ductal adenocarcinoma (PDAC) is the third leading cause of cancer-related deaths in the United States of America with an estimated 62,000 new diagnoses and an estimated 50,000 deaths expected in the USA this year [1]. Currently, the only FDA-approved diagnostic for PDAC is serum antigen CA19-9. Patients with cancer of the pancreas, stomach, lung, liver, or colon typically show levels of CA19-9 exceeding 37 U/mL [2]. For PDAC detection, the CA19-9 serum antigen diagnostic test has a sensitivity ranging between 79% and 95% and a specificity ranging between 82% and 91% [3]. However, since non-cancerous conditions such as pancreatitis, gallbladder infection, liver disease, and gallstones may also show increased CA19-9 levels [4,5], there is a need to develop new modalities with increased sensitivity and specificity of detection for PDAC compared to non-cancerous conditions. Exosomes, and their contents, may offer a more reliable diagnostic alternative to CA19-9. Exosomes are released by all cells in the body, and it is well established that tumor cells release even greater quantities of exosomes [6]. Exosomes are 30-150 nm sized extracellular vesicles that contain proteins, DNA, RNA, and other cellular constituents [7,8]. They are stable in body fluids, allowing for easy collection from patient blood, plasma, serum, saliva, or urine. Recent research has examined both exosomes and their contents for diagnostic feasibility for many diseases [7][8][9][10][11].
The microRNA transcriptome potentially contains diagnostic biomarkers for PDAC that could exceed the sensitivity and specificity of CA19-9 serum markers. MicroRNAs are 19-25-nucleotide-long sequences that have been shown to regulate about a third of human genes, with half being involved in tumor regulation [12]. In non-cancerous cells,

Diagnostic Value of Plasma Exosomal miRNAs
Receiver operative characteristic (ROC) curve analysis was performed on the four mature miRNAs to assess their combined diagnostic efficacy. Since all four miRNAs (miR-93-5p, miR-339-3p, miR-425-5p, and miR-425-3p) exhibited significantly greater ∆Cq values in PDAC samples compared to the control, two threshold values were used to establish a positive "hit" for diagnosing PDAC. If a sample exhibited an miRNA with a ∆Cq value that was (1) greater than the upper limit of the 99% confidence interval (CI) of the average control sample and (2) greater than the lower limit of the 99% CI of the average PDAC sample ∆Cq value, it was recorded as a positive hit for PDAC. Additionally, if a sample exhibited an miRNA with a ∆Cq value greater than both thresholds, it was recorded as two positive hits for PDAC. The upper limit, or the highest average ∆Cq value for the control, was determined for each miRNA and used as a threshold for establishing a potential diagnostic. These values (99% CI) were >9.60 for miR-93-5p, >14.82 for miR-339-3p, >9.16 for miR-425-5p, and >13.77 for miR-425-3p. Using these thresholds, a panel consisting of these four miRNAs was assessed via ROC analysis. The AUC was 0.865 (p < 1 × 10 −8 , Figure 3a) with a sensitivity of 66.7% and a specificity of 94.7%. Similarly, the lower limit of the average PDAC ∆Cq value (99% CI) for each miRNA as the threshold was >8.92 for miR-93-5p, >14.27 for miR-339-3p, >9.59 for miR-425-5p, and >13.15 for miR-425-3p. If the values exceeded the minimum average ∆Cq for PDAC, it was considered a positive hit for PDAC detection. Using this threshold, the four-miRNA panel yielded an AUC of 0.878 (p < 1 × 10 −9 , Figure 3b), a sensitivity of 80%, and a specificity of 89.5%. Finally, combining both the upper limits of the control thresholds (99% CI) and the lower limits of the PDAC thresholds (99% CI), (the ROC analysis resulted in an AUC of 0.877 (p < 1 × 10 −9 ), and an overall sensitivity of 80% and specificity of 94.7% ( Figure 4). Therefore, this marker allows us to detect PDAC in 12 of the 15 samples, including four out of five early-stage PDAC patients. By comparison, CA19-9 could only detect PDAC in 8 out of the 15 samples and was only able to detect one out of the five early-stage PDAC patients ( Figure 5). Interestingly, control sample 8 exhibited the maximum possible of 8 positive hits for PDAC (2 hits for each of the four miRNA), which marks it as a clear outlier sample compared to the other control samples ( Figure 5). The provided patient history for the control samples was limited, and, therefore, it could not be verified whether or not control patient 8 may have had a known pancreatic cancer diagnosis.  , and compared to control samples (n = 19). Student's t-test was used to establish significance, where *, **, and *** denote p < 0.05, p < 0.01, and p < 0.001, respectively.

Diagnostic Value of Plasma Exosomal miRNAs
Receiver operative characteristic (ROC) curve analysis was performed on the four mature miRNAs to assess their combined diagnostic efficacy. Since all four miRNAs (miR-93-5p, miR-339-3p, miR-425-5p, and miR-425-3p) exhibited significantly greater ΔCq values in PDAC samples compared to the control, two threshold values were used to establish a positive "hit" for diagnosing PDAC. If a sample exhibited an miRNA with a ΔCq value that was (1) greater than the upper limit of the 99% confidence interval (CI) of the average control sample and (2) greater than the lower limit of the 99% CI of the average PDAC sample ΔCq value, it was recorded as a positive hit for PDAC. Additionally, if a sample exhibited an miRNA with a ΔCq value greater than both thresholds, it was recorded as two positive hits for PDAC. The upper limit, or the highest average ΔCq value for the control, was determined for each miRNA and used as a threshold for establishing . Student's t-test was used to establish significance, where *, **, and *** denote p < 0.05, p < 0.01, and p < 0.001, respectively. 8 out of the 15 samples and was only able to detect one out of the five early-stage PDAC patients ( Figure 5). Interestingly, control sample 8 exhibited the maximum possible of 8 positive hits for PDAC (2 hits for each of the four miRNA), which marks it as a clear outlier sample compared to the other control samples ( Figure 5). The provided patient history for the control samples was limited, and, therefore, it could not be verified whether or not control patient 8 may have had a known pancreatic cancer diagnosis.    out of five early-stage PDAC patients. By comparison, CA19-9 could only detect PDAC in 8 out of the 15 samples and was only able to detect one out of the five early-stage PDAC patients ( Figure 5). Interestingly, control sample 8 exhibited the maximum possible of 8 positive hits for PDAC (2 hits for each of the four miRNA), which marks it as a clear outlier sample compared to the other control samples ( Figure 5). The provided patient history for the control samples was limited, and, therefore, it could not be verified whether or not control patient 8 may have had a known pancreatic cancer diagnosis.
(a) (b)   The heatmap shows which samples surpassed the two thresholds (ΔCq is greater than the upper limit of the control for each miRNA AND ΔCq is greater than the lower limit of PDAC for each miRNA, 99% CI), indicating a positive hit for PDAC. Red indicates a positive hit and blue indicates a negative hit. Columns labeled as "Cont" represent control plasma samples while columns labeled as "PDAC" represent plasma from patients with pancreatic cancer. PDAC 1, 2, 3, 4, and 5 represent early-stage (I and II) pancreatic cancer; PDAC 6, 7, and 12 represent mid-stage (stage III) pancreatic The heatmap shows which samples surpassed the two thresholds (∆Cq is greater than the upper limit of the control for each miRNA AND ∆Cq is greater than the lower limit of PDAC for each miRNA, 99% CI), indicating a positive hit for PDAC. Red indicates a positive hit and blue indicates a negative hit. Columns labeled as "Cont" represent control plasma samples while columns labeled as "PDAC" represent plasma from patients with pancreatic cancer. PDAC 1, 2, 3, 4, and 5 represent early-stage (I and II) pancreatic cancer; PDAC 6, 7, and 12 represent mid-stage (stage III) pancreatic cancer; and PDAC 8, 9, 10, 11, 13, 14, and 15 represent late-stage (stage IV) pancreatic cancer. CA19-9 values are added for comparison where the threshold is the medically established value of >37 U/mL.

Discussion
Pancreatic cancer continues to be a difficult disease to diagnose and treat. Mortality rates for PDAC are high, though each year has brought marginal improvements in survival rate, with 2022 reporting a survival rate of 11% [1]. The need to develop more reliable methods of detecting pancreatic cancer at earlier stages remains a top priority. Some avenues of research have turned to freely circulating miRNAs or exosomal miRNAs for an array of diseases, including various cancers [28]. Recently, a study by Zou et al. proposed a panel for the early detection of PDAC comprising six circulating miRNAs (let-7b-5p, miR-192-5p, miR-19a-3p, miR-19b-3p, miR-223-3p, and miR-25-3p) that were identified by machine learning and validated in patient samples [29]. Another study by Wang et al. discovered a single serum exosomal marker, miR-1226-3p, that could diagnose and predict pancreatic cancer invasion and metastases [30]. Despite these important contributions, there remain alternative diagnostic biomarkers to be discovered with the potential for earlier and more reliable detection of pancreatic cancer.
Our previous study identified 18 candidate miRNAs from a bioinformatics analysis of publicly available cBioPortal data collected from pancreatic cancer patients [23]. Seven miRNAs were verified to be differentially expressed in exosomes collected from PDAC cell lines compared to an immortalized pancreatic cell line in vitro. Therefore, we sought to validate the expression levels of these candidate miRNAs in plasma exosomes collected from known PDAC patients. The present study identified four miRNAs, miR-93-5p, miR-339-3p, miR-425-5p, and miR-425-3p, with significantly greater ∆Cq values in PDAC plasma samples compared to control plasma samples. A biomarker panel consisting of these four miRNAs and a dual threshold cutoff consisting of the upper limit of the ∆Cq values (99% CI) of the miRNAs from control samples and the lower limit of the ∆Cq values of the miRNAs (99% CI) from PDAC samples resulted in a diagnostic with an AUC of 0.887, a sensitivity of 80%, and specificity of 94.7%. This is comparable to the current FDA-approved CA19-9 diagnostic, which exhibits both a variable sensitivity (70-90%) and specificity (68-91%) [31]. Interestingly, CA19-9 levels were reported as elevated in only 8 of the 15 (53.33%) PDAC patient samples collected for this study. For comparison, by using the optimal cutoff of the four-miRNA two-threshold biomarker panel proposed in this study, our panel positively identified 12 of the 15 (80.00%) PDAC patient samples. This could indicate a potential improvement in PDAC diagnostic sensitivity using the proposed biomarker panel compared to the established CA19-9 serum diagnostic. However, further studies on a larger sample size are needed to fully evaluate this novel finding.
Our study also identified that both mature forms of miR-425 (miR-425-5p and miR-425-3p) had significantly greater ∆Cq values in stage I and II PDAC compared to controls. Consistently, when we applied the two-threshold diagnostic cutoff solely to miR-425-5p, we were able to positively identify five out of five (100% sensitivity) early-stage pancreatic cancer samples. By comparison, CA19-9 was only able to detect one out of five (20%) early-stage pancreatic cancer samples. Therefore, miR-425-5p may offer a more sensitive diagnostic for early-stage PDAC compared to the FDA-approved CA19-9.
Four miRNAs had significantly lower levels of expression in plasma exosomes extracted from PDAC patients compared to control patients. Some of these miRNAs have been previously associated with a variety of cancers and have been implicated in tumor development. For example, miR-93-5p appears to play a role in tumor suppression in ovarian [32] and breast cancers [33,34] by targeting the PD-L1/CCND1 pathway, which is involved in regulating the cell cycle. This is in contrast to the MIR-93 pre-transcript and miR-93-3p, which exhibit oncogenicity. MIR-93 is associated with poor prognosis in PDAC [35], while miR-93-3p predicts poorer outcomes in patients with triple-negative breast cancer [36]. Similarly, overexpression of miR-93-5p appears to promote chemoresistance in PDAC by targeting the PTEN/PI3K/Akt signaling pathway, which is typically involved in tumor suppression [37]. Additional studies confirm the pro-tumorigenic activities of miR-93-5p in endometrial and PDAC tumors [37][38][39][40]. Interestingly, our data are consistent with the previous studies for miR-93-3p, which show a decrease in ∆Cq values, suggesting an increased expression as PDAC transitions from early to late stages. MicroRNA-425-3p is known to be upregulated in response to cisplatin treatment in nonsmall cell lung carcinoma (NSCLC) tissue and exosomes [41]. MicroRNA-425-5p has been shown to be expressed at higher levels in PDAC tissue compared to adjacent healthy tissue [42] as well as in NSCLC [43] and in the serum of gastric cancer patients [44]. Exosomal miR-425-5p from MDA-MB-23 breast cancer cells was shown to convert normal fibroblasts to cancer-associated fibroblasts upon uptake by suppressing the expression of TGFβRII, a TGFβ receptor [45]. Additional research has reported that miR-425-5p promotes tumorigenesis in colorectal cancer by inhibiting the PTEN-p53/TGFβ axis [46] and by activating the CTNND1-mediated β-catenin pathway [47]. MicroRNA-339-5p has been shown to suppress colorectal cancer progression by targeting PRL-1 [48], and while miR-339-3p was observed to interfere with CRC progression, its mechanism is unknown [49]. MiR-339-5p has also been implicated in suppressing melanoma by targeting MCL1, which promotes chemoresistance [50]. Little is known about the role miR-339-3p plays in PDAC, though one paper reports that it is downregulated in the PDAC cell line MIA PaCa-2 [51]. Another work implicates miR-339-3p in inhibiting caerulin-induced acute pancreatitis by targeting TRAF3, which promotes inflammation in pancreatitis cells [52]. Although these previous studies indicate potential roles for miR-93-5p, miR-339-3p, miR-425-5p, and miR-425-3p in cancers, to our knowledge, the present study is the first to propose a diagnostic panel for PDAC using all four biomarkers in plasma exosomes.
Although the present findings indicate a potentially promising novel diagnostic panel to detect pancreatic cancer from plasma exosomes, we cannot rule out the risk of a type I error and the chance that the results are a false positive due to a small sample size. Additionally, a history of cancer treatment regimens for each of the PDAC patients was not available. It is possible that some of the observed miRNA expression changes are due to responses to cancer treatment therapies rather than due to cancer progression. Nevertheless, the proposed four-miRNA two-threshold diagnostic biomarker panel was found to perform comparably to the established CA19-9 diagnostic marker. Further studies on a larger sample size would need to be conducted to conclusively evaluate the diagnostic performance of the proposed panel. To rule out the possibility of sex-, age-, and ethnicity-specific differences, chi-squared tests were performed between the control and PDAC patient information. Chi-squared analyses revealed no significant differences for sex (p > 0.05) and ethnicity (p > 0.05); however, there was a significant difference in the age (p < 0.05) of the PDAC group compared to the control group, when all ages are factored in. Our control group comprised 9 age-matched individuals and 11 individuals who were under the age of 40. There is no significant difference in age variation (p > 0.05), however, between the nine agematched control samples (Cont 2 and Cont 9-16, Figure 4) and their PDAC counterparts. In fact, our data show a clear difference in the expression patterns of the miRNAs (miR-93-5p, miR-339-3p, miR-425-5p, and miR-425-3p) between our age-matched control and PDAC samples. However, the sample set remains small, and further studies will require a larger sample size with more closely matched sample demographics.
In summary, the proposed four-miRNA biomarker panel is able to accurately diagnose PDAC (sensitivity = 80%) compared to control (specificity = 94.7%) samples, performs comparably, and is potentially superior to the established CA19-9 diagnostic method. Additionally, miR-425-5p was identified as a potential marker for the early detection of pancreatic cancer at stages I and II. Further investigation is required to fully evaluate this panel for PDAC diagnosis and monitoring.

Sample Collection
Plasma samples from 15 patients (male: 13, mean age: 66.54; female: 2, mean age: 70) diagnosed with pancreatic cancer were provided by Baptist Health South Florida. The pancreatic cancer samples comprised 10 previously banked plasma specimens collected between 2018 and 2020 and 5 plasma specimens collected in December 2021 by Baptist Health South Florida Hospital. Sex, age, race, staging, and CA19-9 levels are reported in Table 1. Whole blood samples from 19 control patients (male: 13, mean age: 42.62; female: 6, mean age: 44.17) were provided by Continental Blood Bank, Ft. Lauderdale, FL, between 2020 and 2021. The control samples were tested to ensure they were clean of standard blood-borne pathogens prior to release by the blood bank. Patient age, sex, and race are reported in Table 1. In-depth medical history was not recorded by the provider. Plasma was separated from whole blood by centrifugation for 10 min at 2000 g. All samples were deidentified prior to acquisition and stored at −80 • C until exosome isolation steps.

Exosomal miRNA Isolation
Exosomes were isolated from plasma samples using the PEG-based Total Exosome Isolation Kit (from plasma) (Invitrogen, Waltham, MA, USA) as per manufacturer instructions. The Total RNA and Protein Isolation kit (Invitrogen, Waltham, MA, USA) was utilized for the extraction of miRNA from the exosomes isolated from plasma samples. MicroRNA was extracted as per manufacturer instructions, with the addition of an exogenous spike-in control of 1.5 pg of miR-cel-2-3p (Applied Biosystems, Foster City, CA, USA) to monitor RNA extraction efficiency.
RT-qPCR was performed using the AriaMX Thermocycler (Agilent, Santa Clara, CA, USA). The PCR settings are described in TaqMan™ Fast Advanced miRNA cDNA Synthesis Kitprotocols. To normalize sample Cq values, the exogenous spiked-in cel-miR-2-3p control and endogenous hsa-miR-16-5p control Cq values were averaged for each sample. All samples were run in duplicate. MicroRNA levels were calculated and expressed as ∆Cq between the control miRNA Cq value (average of cel-miR-2-3p and hsa-miR-16-5p) and each of the 11 candidate miRNA Cq values. Due to variability in the initial volume of the plasma samples provided, a Cq adjustment was performed to normalize all samples to 750 uL by using the following formula: Cq norm = Cq raw − log 2 750 uL X where X = Initial sample volume in uL.
To minimize PCR background effects, miRNA with Cq values over 35 or not detected after 40 cycles was adjusted to a Cq value of 36 to test for differential expression between PDAC samples versus control samples. The full protocol has been described elsewhere [53].

Statistical Validation
Student's two-tailed t-test was used to test for statistically significant differences in exosomal miRNA expression levels between control and PDAC samples. The chi-squared test was used to predict the associations between age, sex, or ethnicity in the control vs. PDAC groups.