Extracellular vesicle biomarkers for pancreatic cancer diagnosis: a systematic review and meta-analysis

Extracellular vesicle (EV) biomarkers have promising diagnosis and screening capacity for several cancers, but the diagnostic value for pancreatic cancer (PC) is controversial. The aim of our study was to review the diagnostic performance of EV biomarkers for PC. We performed a systematic review of PubMed, Medline, and Web Of Science databases from inception to 18 Feb 2022. We identified studies reporting the diagnostic performance of EV biomarkers for PC and summarized the information of sensitivity, specificity, area under the curve (AUC), or receiver operator characteristic (ROC) curve) in according to a pre-designed data collection form. Pooled sensitivity and specificity was calculated using a random-effect model. We identified 39 studies, including 2037 PC patients and 1632 noncancerous, seven of which were conducted independent validation tests. Seventeen studies emphasized on EV RNAs, sixteen on EV proteins, and sixteen on biomarker panels. MiR-10b, miR-21, and GPC1 were the most frequently reported RNA and protein for PC diagnosis. For individual RNAs and proteins, the pooled sensitivity and specificity were 79% (95% CI: 77–81%) and 87% (95% CI: 85–89%), 72% (95% CI: 69–74%) and 77% (95% CI: 74–80%), respectively. the pooled sensitivity and specificity of EV RNA combined with protein panels were 84% (95% CI: 81–86%) and 89% (95% CI: 86–91%), respectively. Surprisingly, for early stage (stage I and II) PC EV biomarkers showed excellent diagnostic performance with the sensitivity of 90% (95% CI: 87–93%) and the specificity of 94% (95% CI: 92–95%). Both in sensitivity and subgroup analyses, we did not observe notable difference in pooled sensitivity and specificity. Studies might be limited by the isolation and detection techniques of EVs to a certain extent. EV biomarkers showed appealing diagnostic preference for PC, especially for early stage PC. Solving the deficiency of technologies of isolation and detection EVs has important implications for application these novel noninvasive biomarkers in clinical practice.

distant metastasis even is 2.9% [2,4]. Surgical intervention, as the only curative strategy for patients with PC so far, is generally estimated to improve the 5-year survival rate to 30 ~ 40% for PC patients diagnosed at an early stage [5]. However, due to most of PC patients diagnosed at an advance stage, less than 20% of the tumors are eligible for surgical resection [6][7][8]. Imaging-based methods, such as compute tomography, magnetic resonance imaging, and echo-guided ultrasound, have been studied as screening tools only for populations at high risk of PC, which are not only expensive, radiation exposure, or less tolerant, but also have a high rate of false-positive results [9][10][11]. Current conventional serological biomarkers widely used for PC diagnosis and recurrence, such as carbohydrate antigen 199 (CA199), are neither sensitive nor specific enough to act as accurate early diagnostic strategies [12][13][14][15][16] and also are overexpressed in benign pancreatic diseases [17]. Therefore, novel circulation-based noninvasive biomarkers that can specifically and accurately diagnose PC at early stage in the general population have attracted great interest worldwide.
Extracellular vesicles (EVs), mainly classified exosomes and microvesicles, are as membrane-bound vesicles with biologically active molecules including proteins and nucleic acids and can be released from tumor cells to transport these molecules from parental cells to recipient cells to mediate tumor initiation, progression, and metastasis [18][19][20][21]. EVs stably existing in the circulation can protect these functional molecules from impaired by hydrolysis [21]. circulation EV proteins and nucleic acids beneficially classify tumor type for a diagnosis in patients with cancer of unknown primary tumor origin and the concentration of these molecules increases as cancer stage and tumor size [22][23][24][25]. Circulation EV proteins and nucleic acids as potential biomarkers for early cancers diagnosis and continuous monitoring have been investigated for a decade [26,27]. The potential of circulation EV proteins and nucleic acids as biomarkers for PC indicates increasing application and attention. Accumulating evidences have suggested that EV proteins and nucleic acids can be as promising biomarkers for PC diagnosis [28][29][30][31][32]. Two recent prospective studies identified that circulation EV proteins discerned PC patients from individuals with chronic pancreatitis (CP) and healthy individuals, which showed absolute sensitivity and specificity [33,34]. However, circulation EV biomarkers were also reported as invasive markers for benign pancreatic diseases, such as intraductal papillary mucinous neoplasms (IPMNs) [35] and CP [36,37]. Yang et al. demonstrated that circulating EV proteins and miRNAs can predict the presence of invasive carcinoma within IPMN [35,38]. Circulating EV miR-579-3p was identified significant lower expression in CP patients compared to healthy controls [36]. A bioinformatics analysis resulted that exsome miRNAs may be promising markers for early diagnosis and treatment of CP [39]. Therefore, the purpose of this study was to systematically review the characteristics of circulation EV biomarkers for diagnosing PC and further to evaluate the diagnosis value of these biomarkers for distinguishing PC from noncancerous.

Methods
The systematic review and meta-analysis followed a preferred protocol and is reported according with the PRISMA guidelines [40].

Data sources and searches
We performed an electronic search of PubMed, Medline, and Web of Science databases to identify relevant studies assessing circulation EV biomarkers for PC detection up to 18

Study selection
We initially screened all titles and abstracts and studies matched any of the following criteria were excluded: (1) non-English articles; (2) non-original articles; (3) not pancreatic cancer articles; (4) non-human studies; (5) not relevant to the topic. Two authors (Erna Jia and Na Ren) reviewed all potentially relevant full texts, the following studies were included: (1) studies that investigated EV biomarkers in plasma, serum, blood, or peripheral blood in PC patients; (2) PC patients were diagnosed as pancreatic ductal adenocarcinoma depending on the cytological or histological examination; (3) studies that reported the diagnostic performance of EV biomarkers for PC had relevant data (such as sensitivity, specificity, area under the curve (AUC), or receiver operator characteristic (ROC) curve); (4) control groups containing healthy people or benign disease. Any discrepancies were resolved by discussion.

Data extraction and quality assessment
The two reviewers independently extracted available information from eligibility studies in according to a pre-designed data collection form and resolved any disagreements by discussion again. We extracted key information on first author, year of publication, country, study design, population characteristics (including sample size, mean age, and gender distribution), type of blood-based specimen, PC stage, population composition of control groups, names or panels of target biomarkers, detection methods of target biomarkers, preparation approaches of EVs, sensitivity, specificity, AUC, and P Value. Risk of bias and application for each study was assessed using the Quality Assessment of Diagnostic Accuracy Studies 2 (QUADAS-2) checklist [41]. Publication bias was analyzed and represented by a funnel plot, and funnel plot symmetry was assessed quantitatively with egger's test through R software (version 3.5.3, R Foundation, Vienna, Austria) [42].

Data synthesis and statistical analysis
We calculated mean age and sex distribution of eligibility studies using statistical software R (version 3.5.3, R Foundation, Vienna, Austria) if relevant data was not obtained but raw data was available. If the values of sensitivity, specificity or AUC were not reported, we estimated these diagnostic indicators based on ROC curve using OriginPro software (version 9.0) according to maximum Youden's index.
We combined the sensitivity and specificity of EV biomarkers in studies reporting the relevant data to obtain a pooled diagnostic performance for PC using Meta-DiSc version 1.4 by the random-effect model (DerSimonian-Laird method). If noncancerous groups were consisted of healthy controls and/ or benign diseases, we studied them as a whole if the relevant data was reported, or we studied the healthy controls if not. We investigated heterogeneity between studies using Cocharan's Q test and the inconsistency index (I 2 value), with P < 0.05 or I 2 > 50% as statistically significant heterogeneity. We performed sensitivity analysis for studies considered at low risk of bias and low concern for applicability to give convincing diagnostic performance of EV biomarkers for PC based on the assessment of QUADAS 2 using Review Manager 5.3. We also applied subgroup analysis to pool the sensitivity and specificity of individual EV miRNAs and individual EV RNAs detected using qPCR for PC diagnosis.

Research results
We identified 1641 potentially relevant studies resulted from the above search strategy, with 688 from PubMed, 323 from Medline and 630 from Web of Science (Fig. 1). After deduplication and review of titles and abstracts, we narrowed to 97 studies for full text screening and then a further 56 studies were excluded: the specimens of 13 studies were not peripheral blood; 41 studies had no related information data, such as sensitivity, specificity, AUC value, or ROC curve; the control group of one study contained malignant tumors and the case group of one study contained post-treatment cases. Consequently, we identified 41 studies that were met the inclusion criteria [25,30,31,33,34,. Among these 41 studies, the related information data of four studies was respectively from the same two population cohort, we selected the recently published two studies for analyses [48,52]. Ultimately, 39 studies identifying EV biomarkers in serum or plasma for diagnosing PC were included in the systematic review [30,31,34,[43][44][45][46][47][48][49][50][51][52][53][54][55][56][57], and 34 of which were included in the meta-analysis [24, 25, 30, 31, 33, 34, 44-47, 53, 59-62, 64, 67-84].

Quality assessment
We judged the risk of bias and applicability concerns of the included studies using QUADAS-2 evaluation tool, which were grouped four domains: participant selection, index test, reference standard, flow and timing. The                        outcomes of our assessment of methodological quality were summarized in Fig. 2. Twenty-seven studies were high quality studies with low risk of bias for all four domains. The risk of bias in the "index test" was high and the applicability concerns of the "index test" was unclear in Melo et al. 2015, as the index test might not be reproduced [30]; the index test of one study had unclear risk due to the unspecified of EVs isolation and detection method [80]; the risk of bias and applicability concerns of the "patient selection" were unclear in 6 studies [52,56,68,71,74], as the age and gender distribution was significant differences both in PC cases and control groups; The risk of bias and applicability concerns of the "reference standard" were unclear in Goto 2018, as the reference standard in this study was not clarified [84].
The results of egger's test (Z = 0.28, P = 0.78) did not provided any evidence of publication bias. The funnel plot showed reasonably symmetrical, which also supports the results of egger's test (Fig. 3).
A total of 177 EV proteins reported in 13 studies were statistically significant for PC diagnosis, 19 EV proteins were included in panels. GPC1 was the most frequently reported EV protein with median sensitivity and specificity of 72% (from 43 to 100%) and 86% (from 52 to 100%), respectively, which was reported in nine times. And the following were EphA2, EGFR, and EPCAM (Additional file 3). For individual EV proteins, the median reported sensitivity and specificity were 64% (36-100%) and 85% (52-100%), respectively. Both sensitivity and specificity of 20% EV proteins exceeded 80%.
172 EV biomarker panels in 16 studies were reported and 12 panels were verified by independent validation among six studies (Table 3). Most panels showed powerful diagnostic accuracy with median sensitivity and specificity of 82% (from 38 to 100%) and 90% (from 63 to 100%), respectively. Both the sensitivity and specificity of 63% panels exceeded 80%. Nine studies reported diagnosis performance of EV biomarkers (17 individual biomarkers and 2 panels) for early stage (stage I-II) PC, two of individual biomarkers were conducted independent validation tests. And across these two validation studies, both the sensitivity and specificity of EV biomarkers for early stage (stage I-II) PC was more than 85% [34]. The median reported sensitivity and specificity for early stage PC were 94% (from 67 to 100%) and 86% (from 73 to 100%), respectively. Both sensitivity and specificity of 68% EV biomarkers exceeded 80%.

Discussion
A systematic understanding of diagnosis performance of tumor-related moleculars in circulation EVs is critical for PC screening and earliest detection. The present systematic review and meta-analysis has assembled the diagnostic performance of 183 EV RNAs, 177 EV proteins, and 172 EV biomarker panels based on serum/plasma for diagnosing PC in 39 studies from 2015 to 2022. The control groups selected noncancerous population containing healthy population, benign pancreatic disease, IPMN, serious cystadenoma, and nonpancreatic benign diseases. MiR-21 and miR-10b were the highest frequently reported EV RNA for PC diagnosis. GPC1 and EphA2 were mostly reported EV proteins for PC diagnosis. Both the EV RNAs and the EV proteins in high quality studies showed moderate diagnostic values and the EV panels revealed good diagnostic values for PC. Especially, the EV RNA combined protein panels had great diagnostic performances for PC with the pooled sensitivity of 84% and specificity of 89%. Surprisingly, for early stage PC the EV biomarkers showed excellent diagnostic performance, the sensitivity and specificity were 90% and 94%. Respectively. Overall, available studies highlighted the diagnostic performance of EV biomarkers to differentiate PC from noncancerous and further researches will be needed to validate the diagnostic value of EV biomarkers as noninvasive, effective, specific screening tools for PC diagnosis in general population.
EV GPC1 and EV EphA2 were the mostly frequent reported diagnostic proteins for PC in present study. GPC1 was discovered in two prospective studies [47,75], one of which was conducted a blind validation test [47]. In 2015 Nature, EV GPC1 was demonstrated extremely surprising diagnostic accuracy for diagnosing PC, the AUC value reached 1.00 for all stage PC, and the validation test got consistent results that both the sensitivity and specificity of GPC1 for PC diagnosis was 100% [30]. This conclusion attracted highly attention worldwide at a time, but other researchers found inconsistent results at later [30]. A blind prospective study by Yang et al. reported that EV GPC1 had a good sensitivity for PC diagnosis, but the specificity poorly was 52% [47]. In 2018, Li et al. used exosome-based GPC1 to diagnose PC by an ultrasensitive immunoassay method. Although exosome GPC1 could distinguish PC patients from healthy controls, the diagnosis ability was limited and the sensitivity and specificity was 59% and 58%, respectively [82]. Reccently, Xiao et al. established a simple reproducible analysis method for detection exosome GPC1 for PC screening in chinese cohort, the results showed a excellent diagnostic performance with 100% sepcificity and 100% sensitivity [79]. We speculated that the inconsistent diagnosis performance of EV GPC1 for PC might be related to the non-standardized isolation and detection methods, nonuniform experimental procedure, different sample collection protocol, and various data analysis methods. Although the diagnostic performance of EV GPC1 for PC is varied, EV GPC1 indeed is a promising noninvasive biomarker for early diagnosis PC. Almost all of studies reported the diagnosis performance of EV EphA2 revealed great diagnosis value with AUC value equal or beyond 0.85, except one study with AUC value of 0.64 [76]. The result of a large cohort validated study demonstrated that for both I-II stage PC and all stage PC EV EphA2 showed strongly diagnostic efficiency. Especially, EV EphA2 could accurately distinguish PC patients with I-II stage from healthy controls (AUC = 0.96) or pancreatitis (AUC = 0.93) [34].
EV miRNAs recently have raised more interest as novel noninvasive biomarkers for malignant tumors. In our subgroup analysis EV miRNAs appeared powerful diagnosis capacity for PC with both sensitivity and specificity exceeded 80%, which was consistent with that of the whole EV RNAs. EV miR-21 was the most frequency reported miRNA and was significantly high expressed in PC with consistent direction in all five studies. EV miR-21 performed outstanding specificity for PC diagnosis, excepting in study by Goto et al. the specificity was 81%, the specificity value was all greater than 90% in the remaining studies and even two studies showed a specificity of 100% [25,45,61,62,84]. Aberrant expressed miR-21 related to the development and metastatic of malignant tumors and impacted signal transducer and activator of transcription 3 (STAT3), epidermal growth factor receptor (EGFR), transforming growth factor-β, and the p53 pathway in malignant tumor progression [87][88][89][90]. MiR-10b high expressed in PC and played important role in various malignant cancers, including PC. For example, PC patients had significantly higher expression level of miR-10b than healthy controls or pancreatitis and increased miR-10b expression in PC patients indicated tumor aggressiveness [91][92][93][94]; Abnormal expressed miR-10b promoted hepatocellular carcinoma cell proliferation and invasion and high expression level of EV miR-10b was related to advanced tumor [95]. However, there are number of challenges in using EV miRNAs as diagnostic biomarkers for PC, such as racial differences [96], unstandardized sample preparation procession [97], and nonuniform miRNA extraction kits [98]. Continued refinement of EV miRNAs techniques may help identify specific EV miRNAs for screening PC, which offers as promising non-invasive biomarkers for PC diagnosis in clinical practical.
In present study, we found EV biomarker panels showed high diagnosis value for PC, the pooled sensitivity and specificity was 80% and 86%, respectively, which showed a small advantage diagnosis performance than the whole individual EV RNAs and individual EV proteins. In previous studies, compared with the corresponding individual markers the diagnostic accuracy of either protein panels or miRNA panels was not to show any advantage [25,61,62,79,82]. However, combining EV RNAs with EV proteins, the diagnosis performance for PC was significant greater than that of the corresponding individual biomarkers. For example, both the sensitivity and specificity of EV panel composed by EphA2, miR-451a, miR-21, and miR-10b reached 100% [62]; Another panel consisted by EV miRNAs and EV proteins exhibited great diagnosis efficiency for PC with sensitivity of 100% and specificity of 93% [43]. In present study, we also found the diagnostic value of the EV RNA combined with protein panels showed good diagnostic performance, which was similar to the EV protein panels and a slight greater advantage than the EV RNA panels. Moreover, combined EV RNAs and EV proteins with conventional CA199 also significantly improved the diagnostic accuracy for PC. EV miR-200b and 200c and EpCAM combined with CA199 entailed a diagnostic accuracy of 97%, yielding sensitivity of 92% and specificity of 100% [31]; Therefore, the combine of novel EV biomarkers with conventional CA199 might be explored to increase diagnostic accuracy for PC and can be as noninvasive and low-cost diagnosis methods for PC screening in the future.
Due to high stability and providing specific information from original tumor of RNAs and proteins in EVs, EV RNAs and proteins have attracted considerable attention as noninvasive biomarkers for cancer diagnosis. By far, the analysis of EVs was separated into isolation and detection two steps. Multistep ultracentrifugation is the recommended standard method for EVs isolation, which increased the complexity of the operational process and was time-consuming and expensive, and the centrifugation time and force was nonuniform inducing major difference of EV purity and isolation rate [99]. Most included studies in this systematic review underwent the centrifugation step, but the centrifugation time and force are variable, parts of studies [85] even used once time centrifugation, which greatly affected the isolation concentration of EVs and further impacted the expression level of molecules contained in EVs. For decades, significant progression of EVs isolation and detection techniques has been made and two kinds of microfluidic-based techniques, immunoaffinity-based and size-based platforms, have been reported as most practical solutions to isolate and detect EV biomarkers [34,44,47,62]. Microfluidics-based techniques provide an effective method to integrate the EV isolation and detection into a single chip, which overcome the abovementioned disadvantage [100]. In current study, six studies used microfluidic-based techniques to analysis the expression levels of EV biomarkers, such as 3D microfluidic chip, LPHN-CHDC biochip, ACE integrated biomarker assay, chip-exosome-PEARL SERS immunosensor, surface-Enhanced Raman Scattering, NPS chip, EphA2-EV-nPES assay, SSEC, AbMB-bioChol platform, nanoparticle-and dyebased fluorescent immunoassay, and immuno-capture using magnetic beads. Although microfluidic-based techniques show great advances in EVs isolation and detection, it need sufficient clinical samples to validate the application, accuracy, stability, and reproduction.
The heterogeneity of the present study ranged from mild to severe, neither subgroup analysis nor sensitivity analyses failed to reduce the degree of heterogeneity. Furthermore, we analyzed the pooled diagnostic value of EV biomarkers for early stage PC, the results also did not have impact on heterogeneity. We hypothesized that the sources of heterogeneity might be related to the various methods of EVs isolation and detection, sample selection protocol, and demographic or geographic different of study populations. In addition, the number of samples of case and control groups widely ranged from 6 to 284, which also contributed to the heterogeneity. Although the heterogeneity might limit the reliability of the pooled results in present study, the results of our systematic review and meta-analyses have a certain guide to pick up noninvasive and effective early diagnosis biomarkers for PC.

Conclusions
In this systematic review and meta-analysis, we found circulation based EV biomarkers exhibited considerable diagnosis performance for PC. EV RNAs combined with EV proteins showed appealing higher diagnosis efficiency. Especially, for early stage PC the EV biomarkers revealed excellent diagnostic performance. However, the deficiency of technologies that can effectively isolation and detection EV biomarkers limit the application of EV biomarkers in clinical practice to some extent, highlighting the need for high-quality reproduction researches in this area as well a need for promising accuracy EV biomarkers for PC diagnosis and screening in larger sample prospective cohort.