Diagnostic acuuracy of cerebrospinal fluid and serum-isolated exosomes for glioblastoma: a systematic review and meta-analysis CURRENT

Glioblastoma (GBM) is the most malignant glioma cancer with a high morbidity and mortality worldwide. Unfortunately, a routine method is not available for screening or preoperative early detection of GBM. However, early detection in a none-invasive or minimally invasive method could be beneficial and increase the survival rate. In this systematic review and meta-analysis, we aimed to examine the diagnostic accuracy of exosomal RNAs that were extracted from patients’ CSF or serum for GBM diagnosis. We searched Web of Science, Scopus, PubMed (including Medline), Embase and ProQuest (as databases for grey literature) up to December 2019; we also performed backward and forward reference checking of included and relevant studies. Finally, included studies were assessed with QUADAS-2 checklist and their data extracted. We carried out a meta-analysis of included study, regarding to the diagnostic meta-analysis guidelines for obtaining pooled accuracy estimates. In addition, sensitivity analysis and meta-regression were also conducted. We retrived 1730 records from databases, nine of them included in systematic review and qualitative synthesis. Six studies were considered to statistical analysis and performed diagnostic meta-analysis. Our results suggested that the pooled sensitivity and specificity of exosomal biomarkers for GBM were 0.76 and 0.80, respectively. In addition, the pooled positive likelihood ratio (PLR), negative likelihood ratio (NLR), and diagnostic odds ratio (DOR) were 3.7, 0.30 and 12, respectively. The overall area under the curve (AUC) of exosomal biomarkers for GBM diagnosis was found to be 0.85. According to our results, the value of 0.85 for AUC, suggesting that exosomal biomarkers might serve as a high potential and non-invasive diagnostic tool for GBM.


Introduction
Glioblastoma or glioblastoma multiforme (GBM) is grade IV and the most malignant type of gliomas [1,2]. For GBM cure, traditional treatment is tumor resection followed by radiotherapy and chemotherapy that is limited in effectiveness due to high rates of relapse, overall resistance to therapy, and serious neurotoxicity and neurological side effects [3,4]. Due to lacking early detection and personalized treatment of GBM, the prognosis is often poor [5].
Detection and biopsy of tumors, especially in brain cancer, is high invasive, expensive and time-consuming. For this reason, developing of non-invasive, affordable and efficient methods for detection and grade prediction matters a great deal. In this regard, identification of biomarkers of brain cancer in the patient's body fluids, especially CSF and serum, including DNAs, miRNAs, mRNAs, lncRNAs, and proteins could be much helpful for diagnosis of GBM and other neurological cancers and diseases [5].
In the past several decades, many studies aimed to introduce new, accurate, and specific biomarkers for different types of cancers [6][7][8]. One of the new type of biomarkers that introduced recently, is the biomarkers that occure inside the exosomes. Exosomes are a type of extracellular nanovesicles with 30-150 nm in diameters [9]. Regarding the origin, exosomes are distinct from other types of extracellular vesicles including microvesicles (with heterogeneous size from 50 to 1000 nm in diameter) and apoptotic bodies [10,11].
However, microvesicles are crucial in cellular communication and has been used as a biomarker source similar to exosomes, but they are not popular as exosomes [12].
However, there is a problem in terminology of exosomes and microvesicles; sometimes, they are mistakenly used interchangeably [13].
Exosomes are the carreres of their origin biomarkers that most of the time are indistinguishable in the corresponding biological fluid as cell-free or naked biomarkers [14]. As a result of the latter property of exosomes, they can provide various sensitive and specific diagnostic biomarkers for different pathophysiological conditions [11]. At the beginning, Valadi et al. for the first time stated that exosomes, in addition to proteins, possess different types of RNAs [15]. Today it is well known that exosomes transport diverse molecular constituents (such as lipids, proteins, and nucleic acids) of their source cells and tissues [10]. Next, over important finding about exosomes biological nature and their contents, scientists made such an idea that exosomal contents as a tool for discriminating cancerous cells from healthy cells may simplify cancer diagnosis through a minimal invasive approach [16]. Interestingly, furthure studies demonstrated that the amount of exosomes in patients with cancer is higher than healthy controls [17]. Exosomal biomarkers compared to other biomarkers enable similar or higher specificity and sensitivity due to their higher stableness compared to unvesicled or cell-free biomarkers.
New technical improvements in exosome isolation facilitate exosome study, and it could make exosomal diagnostics as a new method for diagnosis of diseases, especially cancers [16,18].

Methods
We designated a protocol following PRISMA guideline for reporting of a systematic review, and Protocols for Reviews of Diagnostic Test Accuracy [19]. Prior to publication of the study, we were submitted this systematic review on PROSPERO on 26 Jun 2019 and registered on 17/09/2019 with CRD42019132438 ID.

Search strategy designing
To retrieving all possible studies in the area of our purpose, we aimed at developing a full search strategy. We were implemented a comprehensive systematic search which were combined (using the Boolean Operator) text-words and subject headings (MeSH or equivalent) of the following electronic databases: PubMed (including Medline), Embase, Web of Science, Scopus, and ProQuest (as databases for grey literature) through December 2019. The search was performed according to the search strategies stated in the protocol Cochrane Handbook for Systematic Reviews of Diagnostic Test Accuracy [20].
For the search in electronic databases, we used the all possible keywords related to "Exosomes" and "Glioblasoma" that was extracted from MeSH database and Emtree. This strategy of search allowed us to conducting of a comprehensive search by recent published protocol [21]. The databases were searched without any restriction.
The literature search was performed independently by two investigators (AT and DJ). Any possible discordance was compared to that of an additional investigator (YM). In order to retrieve any missed related articles through database searching, in addition to the cited references in the final identified studies, a comprehensive search was carried out for the published reviews adapted to our title. Subsequently similar to systematic screening of online databases results, the references of all these articles screened manually. This step ensures that all studies relevant to assessing the diagnostic accuracy of exosomal RNAs in the diagnosis and early detection of GBM are included in our final library. Full search strategy of PubMed, Scopus, Embase and Web of Science databases was shown in Table S1 in Aditional file 1.

Selection of studies
Two reviewers independently (DJ and FJ) screened the titles and abstracts of all retrieved records and then evaluated to determine whether inclusion criteria were met. Inclusion assessment of full papers was conducted by one author (DJ) and checked by a second investigator (AT). In case of disagreement, a consensus was reached by discussion or referral to a third author (YM). We conducted PRISMA diagram to illustrate the study selection process [19].

Inclusion criteria
We included all original articles that were surveyed diagnostic value of exosomes RNAs in human GBM patients, studies that conducted on body fluid isolated exosomes and studies that were conducted on two types of extracellular vesicles (exosomes, microvesicles).

Exclusion criteria
We excluded duplicate citations, non-peer-reviewed, review papers and book chapters.

Data collection process
The following items were collected from each article by two investigators (DJ and ER): first author, publication year, country, number participants, exosomes source, isolation and purification methods, related identified biomarkers, biomarker extraction method, not provided with studies) calculated using the specificity and sensitivity. The data extracted in fully pair method from individual studies. We sought further information from the study authors if necessary. Any disagreements between data collectors were resolved through either discussion or consultation with a third author (AT).

Assessment of methodological quality and risk of bias
We also assessed the methodological quality of the included studies independently by two reviewer (DJ and AT) by using the revised Quality Assessment of Diagnostic Accuracy Studies (QUADAS-2) tool, recommended by the Cochrane collaboration, for risk of bias and applicability concerns [22]. It assesses the risk of bias by scoring questions in the four domains as follows: 1. Patient Selection (the method of patient selection and the patients included), 2. Index Test (the test being studied and how it was conducted and interpreted), 3. Reference Standard (the reference standard test used and how it was conducted and interpreted), and 4. Flow and Timing (the flow of patient inclusion and exclusion, testing procedure and the interval between tests). The first three domains also assess the applicability concerns considering review question. Each of domains was categorized as high, low, or unclear and disagreements were resolved by discussion with a third reviewer (RJ).

Statistical Analysis
We used the metandi and midas modules in the STATA 11.2 (Stata Corporation, College Station, TX, USA) statistical software to perform all analyses [23,24]. TP, FP, FN and TN data were used to calculate sensitivity, specificity, positive and negative likelihood ratio and diagnostic odds ratio. Their pooled estimates and their corresponding CI 95% for exosomes were calculated by using the bivariate and hierarchical meta-analysis that included bivariate mixed-effects regression model and the hierarchical summary ROC (HSROC) modeling [25,26]. Results were displayed graphically on forest-plots and HSROC curve. Heterogeneity between included studies was assessed using Cochran's Q test and the inconsistency index (I 2 ) describing the percentage of total variation across studies due to heterogeneity rather than chance [27]. A p-value ≤ 0.05 and an I 2 value ≥ 50% would indicate substantial heterogeneity. The threshold effect was checked using Spearman's rho and potential sources of heterogeneity were explored by meta-regression. Sensitivity analysis was performed by omission of outlier studies indicated by Cook's distance and standardized predicted random effects to investigate the influence of this study on the pooled estimates [24]. We assessed publication bias using Deek's funnel plot, and a pvalue < 0.1 in the Deek's asymmetry test was considered to indicate publication bias [28].

Literature search and study selection
Initially, 1730 articles (279 from PubMed, 438 from Scopus, 400 from Web of Science and 613 from Embase) were retrieved based on the search strategies from the online databases. From wich, 900 duplicate publications were identified and removed. The remaining 830 articles were screened by titles and abstracts, led to the exclusion of 802 items, including 259 reviews and 196 conference abstracts. The full texts of the 28 remaining records studied, and 19 articles were excluded according to inclusion and exclusion criteria. Finally, nine eligible studies were included in the present systematic review from 2013 to 2019 and from four countries of Australia, China, Italy and USA (Table   1). Figure 1 shows the process of literature search and study selection as a flow diagram based on the PRISMA.

Quality assessment
According to the results of quality assessment using QUADAS-2 check list in Figure 2 and Table S2 (in Aditional file 1), in the section of risk of bias, the most elements lowering the quality of studies are reference standard decription as well as the index test. In this regard, some studies did not reported ratios of diagnostic value. In addition two studies reported diagnostic potential of an exosomal biomarker for GBM from other types of disease or cancers. In the patient selection, flow and timing section totall six studies were unclear or weak. This is due to undefined patients pathophysiological condition, sex, and age as well as some technical analysis and experimental efforts.
In the case of applicability, consequently as expected, the index test is still suffering from low quality in four studies due to low applicability of developed index test for diagnosis of GBM. However, the rest of the studies have developed their index test compliance with our review question and inclusion criterias.
Alltogether, we have found that only the study of Manterola et al. [30] was perfectly meet our inclusion criterias and review questions in all parts of risk of bias and applicability assessment sections. Finally, we identified six studies that filled the thresholds to be included in meta-analysis and three studies were excluded ( Figure 1).

Figure 2
Quality assessment of the included studies performed using the QUADAS-2 checklist.

The characteristics of studies included in systematic review and metaanalysis
Altogether, 487 patients histopathological and MRI diagnosed with GBM cencer were investigated for biomarker development in nine included studies. The patient population is ranged from 13-107 and the control population consisted of healthy controls, non-brain tomur patients [33] and other brain tomur patients [37].
Six studies used the serum, two studies used CSF and one study used both serum and CSF as biomarker source. In addition, one study used CSF from two different anatomical loacation (cisternal and lumbral) ( Table 2).
In order to analysis of exosomal biomarkers, exosomes must first be isolated and purified from the other vesicles and molecules of the serum or CSF samples. As shown in Table 2 three studies used ultracentrifugation (UC), five studies used commercial kits and one study used size exclusion chromatography for isolation and purification of exosomes. In this regard three methods of exosome isolation was used in this studies (Table 2).
For RNA extraction from isolated exosomes, entirely all of the studies used RNA extraction commercial kits (Table 2). Besides, for profiling and analyzing the expression pattern of selected RNAs or between patients and controls, q-RT-PCR and RNA sequencing methods were used. From wich, eight studies used q-RT-PCR and only one study used sequencing.

Qualitative synthesis of diagnostic value of exosomal RNAs for GBM
In Table 3, extracted data related to diagnostic value of different biomarkers including, AUC, CI95%, TP, TN, FP and FN as well as specificity, sensitivity and their cut-offs, has been shown.
Through nine studies, 24 single RNA biomarker and there miR-panels developed for diagnosis of GBM. Among the single biomarkers, miR-21 was investigated in four individual study [29,31,32,34]. In addition miR-222 , miR-29b, miR-320, miR-574-3p, miR-124-3p, RNU6 and HOTAIR biomarkers were also individually evaluated for the diagnosis of GBM in different studies. According to the studies provoded, or calculated specificity and sensitivity of these single biomarkers, they shows high diagnostic potential for GBM (Table  3). From wich, miR-21 with 85 and 100, 87 and 93, 84 and 77 for sensitivity and specificity respectively in three different GBM patient populations shows a high diagnosis value for GBM. Interestingly, in the study conducted by Lan et al. [33], they assessed the expression level of this miR-301 in healthy controls, but they did not performed any diagnostic analysis. However, they only provided diagnostic values for this miR for discriminating GBM from none-brain tomur patients.
Finally another miR panel with miR-21, miR-222 and miR-124-3p showed 84% and 77% for sensitivity and specificity [34]. In addition to these panels, Ebrahimkhani et al. [35] developed six miR panels that predict GBM with perfect accuracy. Their panels including although have perfect accuracy but they did not reported any sensitivity and specificity for these panels.

Diagnostic accuracy
Meta-analysis was carried out on the data from six studies including 592 participants (325 GBM patients and 267 healthy controls) on 16 exosomal biomarkers. We excluded three articles from meta-analysis due to unreported TP, FP, FN and TN as well as working on patient controls [31,33,35].
Across all six included studies, the pooled sensitivity and specificity of exosomal biomarkers for detecting the presence of GBM were 0.76 (95% confidence interval [CI] = 0.68-0.82) and 0.80 (95% CI = 0.72-0.86), respectively. A forest plot of the included studies along with sensitivities, specificities and pooled estimates is provided in Figure 3.
The hierarchical summary receiver operating characteristic (HSROC) curve for diagnostic accuracy of exosomal biomarkers is provided in Figure 4. The area under the curve (AUC), as an overall measure for test performance, was 0.85 (95% CI: 0.81-0.87), indicating that exosomal biomarkers have good diagnostic accuracy for GBM.

Publication bias
The potential publication bias was examined by the Deeks' funnel plot. As displayed in Figure 5, a P-value of 0.70 indicated no evidence of publication bias in the meta-analysis.

Heterogeneity
Due to no threshold effect existence and given the presence of between-study heterogeneity, we conducted a meta-regression to finding source of heterogeneity. The multiple meta-regression of sensitivity and specificity indicated that source and isolation method of exosomes, RNA extraction methods, sample size and quality of study acted as the potential source of heterogeneity in the pooled estimates (Aditional file 1, Figure S3). Table 5.
When these outliers were omitted, the pooled sensitivity and specificity were 0.79 (95% CI 0.73-0.84) and 0.76 (95% CI 0.69-0.81). The omission of these studies not substantially influenced pooled estimates but reduced the extent of between-studies heterogeneity in the pooled sensitivity and specificity from I 2 73.44% to 49.25% and 59.24% to 40.68%, respectively (Aditional file 1, Figure S5).

Discussion
Cancer is a result of perturbation in cell cycle checkpoints molecules that usually occurs by mutations or other genetic changes in a single cell [38]. This molecular event of the transformation will be exist in all of a tomur cells [39]. Identifying and tracking of the molecular information of cancerous transformation could help to develop new therapies and diagnostics [40].
Malignant gliomas including GBM is a high mortality brain tomur with poor prognosis.
There are no preoperative routine diagnosis method for GBM. However GBM most of the times, grew locally and rarely metastasized outside of CNS and if resected in the early stages even could completely cured [41].
In the recent few decades, plasma circulating cancer proteins [42], cell-free DNAs [43] and RNAs [44] were raised as diagnostic tools and significantly increased the prediction power of various cancers. Exosomal biomarkers possess multiple advantages compared to cell-free biomarkers. Cell-free biomarkers continuously oppose to blood degrading enzymes while exosomal biomarkers are encapsulated inside the vesicles and are more stable outside of the body after isolation so that they could be reliable than cell-free biomarkers like RNAs, that are most prone to degradation [13].
Studies showed that exosomal biomarkers, specially RNAs (miRNAs, LncRNAs and mRNAs) could be used for prediction of the cancers and their stages [45]. For instance, exosomal miR-21 with high sensitivity and specificity for discriminating multiple cancers from healthy patients has been widely studied. This exosomal biomarker has different expression in many cancers [46][47][48][49][50].
The present study, with aim to find the diagnostic accuracy of exosomes, is the first systematic review and meta-analysis study for aluation of overall exosomes diagnostic value for GBM. The main goal of this study was to assess how could exosomes be used as a non-invasive method for diagnosis of GBM. Regarding, we conducted a comprehensive search strategy for searching the databases for all possible exosomal biomarkers for GBM.
All of the included articles to our systematic review were evaluated the exosomal RNA biomarkers and we did not find any exosomal protein, DNA, lipid or carbohydrates biomarkers that met our inclusion and exclusion criterias. It is worth mentioning that we found only exosomal protein and DNA putative biomarkers for GBM diagnosis that were not suitable for this study.
We first assessed that how these nine included studies conducted the diagnostic experiment using exosomal biomarkers. According to previous studies we found that, the method of exosome extraction, RNA profiling, and analyzing could affect the results of diagnostic evaluation [51][52][53][54][55]. Therefore, the workflow from the extraction of exosomes to extraction and analyzing the biomarker assessed in included studies. Recently ultracentrifugation introduced as the gold standard for exosome isolation [56]. Other methods including, affinity chromatography method [57] and specially the exosome isolation kits are mostly used for exosome isolation [58]. As discussed in detail in our previous review [13], exosomes contents are heterogeneous and dependent on the methods of isolation and purification [59]. In our included studies three methods of exosome isolation (UC, Kit and SEC) and mostly UC and exoquic were used. As downstream analysis of exosomes also are determining factors in the results of diagnostic studies [54,60]. We found that different RNA isolation kits, as well as two main methods were used for extraction and profiling of exosomal RNAs (q-RT-PCR and Sequencing).
According to the results of extracted data and quality assessment using QUADAS-2 checklist, we excluded three study and conducted a meta-analysis with six primary research. We calculated pooled estimates of AUC, sensiticity, specificity, DOR, PLR and NLR for overall diagnostic value of 16 exosomal biomarker for GBM diagnosis.
Our results suggest that the pooled sensitivity and specificity of exosomal biomarkers for GBM are 0.76 and 0.80, respectively. In addition, the pooled PLR, NLR, and DOR were 3.7, 0.30 and 12, respectively. The 3.7 calculated PLR shows that using this index test GBM patients had nearby 3.7-fold higher chance of a positive result than healthy population. Furthermore, 0.30 NLR, suggests that from the negative outcome of the index test results, 30 % could be GBM positive. Similarly, DOR is an index that correlates with the diagnostic performance of our index test. The higher DOR the better diagnostic performance [61]. In addition, AUC is the most important index to show the overall diagnostic power of an index test. The AUC value higher than 0.75 is considered as acceptable diagnostic performance [62]. According to our results, the value of DOR and AUC were 12 and 0.85, respectively, suggesting that exosomes might serve as a high potential diagnostic tool for GBM.
Notwithstanding, a significant heterogeneity was observed in the pooled sensitivity, specificity, DOR and NLR. In the case of publication bias, a P-value of 0.70 showed no evidence for publication bias in the meta-analysis.
Alltogether, as expected, the multivariate meta-regression of sensitivity and specificity indicated that source and isolation method of exosomes, RNA extraction methods, sample size and quality of study acted as the potential source of heterogeneity in the pooled estimates. We omitted two studies as outlier and performed a sensitivity analysis. The result of sensitivity analysis showed that, the pooled sensitivity and specificity are 0.79 and 0.76.
Regarding that exosomal diagnosis is a new emerging field in biomedical sciences, little research has conducted on every single cancer, as well as insufficient studies could find on a particular exosomal biomarker. For this reason, systematic review and meta-analyses studies conducted in this field will be associated with some inherent limitations. However, this yung field, need the secondary studies to lead and clarify the future primary studies.
The aim and the results of the present systematic review and meta-analysis and a recent meta analysis, for assessment of diagnostic value of exosomes for lung cancer are much consistent. Similar to us, they also aimed for evaluating the averall diagnostic value of exosomal biomarkers. Their results showed that the pooled sensitivity, specificity, PLR, NLR, DOR and AUC were 0.82, 0.84, 5.27,, 0.21, 25.14, and 0.90, respectively [63]. In addition, a systematic review and meta analysis of exosomal miR-21 for overall cancer detection showed that the sensitivity and specificity of pooled studies were 75% and 85%, and AUC was 0.93.
Finally, despite our complete protocols of systematic review and meta-analysis according to the latest guidelines for diagnostic studies, our study faced to some ineludable challenges and limitations. Firstly and the most important limitation of our study is heterogeneity in exosomal biomarkers. Generally, we claimed that we estimated the pooled diagnostic accuracy for 16 different biomarkers and we did not estimated that for a single particular exosomal biomarker. The latter limitation, ass discussed, is related to small number of studies on the each biomarkers. Further studies on each of these biomarker could lead a meta-analysis on single biomarker. Secondly, the number of included studies and participants were small. More studies could further validate the diagnostic performance of exosomes for GBM. Thirdly, most of the included studies were conducted in china and USA. Fourthly, the high heterogeneity should be taken into consideration. In this regard, we carried out a meta-regression, and a sensitivity analyses to detect the potential heterogeneity sources. After removing the outlier studies, sensitivity analyses were performed. Considering that the diagnostic accuracy did not changed significantly, the heterogeneity was lowered. The meta-regression analyses found some study characteristics including exosome isolation method, exosome source and RNA profiling and sample size as the source of heterogeneity.
Ultimately, despite described limitation, our study strongly suggest that exosomes have high prediction power for GBM and our results and few conducted meta-analysis show that exosomal biomarkers are a new emerging tool for cancer diagnosis that will attract much attention in the near future.

Conclusions
In this study, we screened the 1730 primary research article in the exosome and GBM field and finally nine primary studies remained for further analysis. We found 24 exosomal biomarker for GBM diagnosis. Six studies consit of 16 biomarker considered for metaanalysis. Our results suggest that the pooled sensitivity and specificity of 16 exosomal biomarkers for GBM are 0.76 and 0.80, respectively. In addition, the pooled PLR, NLR, and DOR were 3.7, 0.30 and 12, respectively. Figure 1 The literature search and study selection process for systematic review according to PRISMA   Deek's funnel plot and asymmetry test for evaluating potential publication bias.

Declarations
ESS: the effective sample size.

Supplementary Files
This is a list of supplementary files associated with the primary manuscript. Click to download.