Exosomal prognostic biomarkers predict metastatic progression and survival in breast cancer patients

Objectives: This study aims to comprehensively evaluate extracellular vesicle (EV)-based biomarkers circulating in body ﬂ uidswith signi ﬁ cant prognostic value in breast cancer (BrCa). Methods: We systematically searched WOS, PubMed, and Scopus databases on 14 February 2023 for studies indicating overall survival(OS), progression/disease/event-free survival(PFS/DFS/EFS), and metastatic progression. We computed univariate(UHR) or multivariate adjusted(AHR) hazard ra-tios, and AUC values for all prognostic EV-based biomarkers of blood-origin using random e ﬀ ect model and Stata 16.0 software. Subgroup analysis was conducted for positive and negative prognostic factors. Results: Twenty-one articles comprising twenty-six studies and 3,423 patients satis ﬁ ed the inclusion criteria. EV-based negative biomarkers indicated lowOS(UHR=2.31, CI=1.77 – 3.03, I 2 =60.12 %, p<0.001); worse DFS/PFS/EFS(UHR=3.91, CI=2.82 – 5.43, I 2 =19.08 %, p=0.24); increased risk for metastasis(pooled AUC=0.91). Out of 56 EV-based biomarkers that have been previously described, we identi ﬁ ed PD-L2, sHLA-G, exo-XIST, and miR4800 as the best predictors of OS of BrCa patients. Expression levels of miR155, Annexin-A2, sHLA-G, PD-L2, miR1246, PSMA and the biomarkers constructing the EV P - panel hold signi ﬁ cant potential to be combined in a prognostic-panel predicting DFS/PFS/EFS of BrCa patients. PD-L2 and sHLA-G standing out as leading biomarkers in both OS and DFS highlights the importance of immune system evasion for patient survival. In addition, we suggest that reinforcement with additional RNA biomarkers could significantly increase the metastatic prediction power of the previously described EV DX -panel. Conclusions: This meta-analysis provides an overview of the liquid biopsy-based EV-biomarkers associated with OS, DFS, and metastatic progression of BrCa for the ﬁ rst time. Prognostic e ﬃ ciency of the proposed panels should be further investigated before transition to clinical use.


Introduction
Breast cancer is the most commonly diagnosed female cancer, and the leading cause of cancer mortality in females worldwide with over 2.25 million new cases reported in 2020 [1].While average recovery rate from breast cancer is relatively high with respect to other cancer types, early detection of the cancer and its recurrence, and patientdependent selection of the optimal therapeutic approaches are key to minimal morbidity and increased survival.Prognostic and predictive biomarkers serve as tools in determining which patients would benefit from chemotherapy/adjuvant therapy, and for the selection of the optimal therapeutic regimen.
In the last few decades, there has been a growing interest in the development of non-invasive diagnostic tools, and many circulating biomarkers identified in serum, plasma, urine, saliva, cerebrospinal fluid, seminal or vaginal fluid, even tear have been suggested to be used for diagnostic and prognostic purposes.Few commercial kits relying on liquid biopsies have even received FDA approval as companion diagnostics for tumour profiling.Recently, attention has been focused on membrane-bound or membrane-enclosed vesicular biomarkers circulating in bodily fluids not only because they are more stable in the circulation, but also because they provide additional information about the tumour origin and contain cues regarding the tumour microenvironment.Extracellular vesicles (EVs) which were initially thought to be formed solely as apoptotic bodies, were later discovered to be produced also through healthy cell-shedding in the form of exosomes and microvesicles (MVs) leading to cell-cell communications between local and distant cells, and cell-microenvironment interactions.Extracellular vesicles ranging from 30-150 nm sized exosomes to 100-1,000 nm sized microvesicles are lipid-bound membranes that encapsulate tissue-specific bioactive molecules such as proteins, mRNAs, miRNAs, small interfering RNAs (siRNAs), long noncoding RNAs (lncRNAs), circular RNAs (circRNAs), metabolites, etc. [2].Exosomes originate from endosomal vesicles; in addition to the endosome-associated proteins such as tetraspanins (i.e.CD9, CD63, CD81) that participate in multivesicular endosome (MVE) biogenesis and be used as markers of exosomal-signature, they inherit bioactive molecules specific to the cell they stem from.MVs on the other hand, are formed by direct budding of the plasma membrane, mainly contain cytosolic and plasma membrane-associated proteins that cluster at plasma membrane; while they are rich in tetraspanins, cytoskeletal proteins and integrins, MV's lack organelle-associated proteins (i.e.proteins of nucleus, ER, Golgi, mitochondria, etc.).EVs may act on the original cell through autocrine signalling, may interact with neighbouring cells in close proximity through paracrine signalling, or may travel to distant tissues through circulation and act on distal tissues.Via the cargo molecules they carry, they may interfere with the signalling pathways of the cells they target; mRNA cargo may be translated into effector proteins; non-coding RNAs may regulate expression profile of the target cell; metabolites and messengers might interfere with the function or fate of the target cell.When the circulating EVs originate from cancer cells, through the messengers they carry they may reprogram a recipient healthy cell into malignant growth, even leading to metastatic progression.
As the number of EV-biomarkers discovered in breast cancer progression, therapeutic response and prognosis is progressively increasing; there has been a growing need for overall assessment of the association between EV-based biomarkers and survival outcomes of breast cancer.This study fills a gap in the literature by bridging the observational data acquired from individual cohorts with limited sample size regarding prognostic values of EV-based biomarkers on therapy-response, metastatic progression, overall survival, and disease/progression/ event-free survival of breast cancer patients.For clarity of the current text, the prognostic biomarkers of extracellular vesicle-origin that are retrieved from liquid biopsies of breast cancer patients will be mentioned as "liquid biopsy-based EV-biomarkers" here onwards.

Search strategy
We have followed Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) 2020 Statement for identifying, selecting, appraising and synthesizing the studies included in this meta-analysis [3].Web of Science, PubMed and Scopus were systematically searched on 14 February 2023 for studies reporting prognostic potential of circulating exosomal biomarkers indicating overall survival (OS), progression-free survival (PFS) and disease-free survival (DFS) in breast cancer patients.No date restriction was applied.The MeSH terms and additional keywords used for each database is provided as supplementary information (Supplementary Table S1).

Screening and selection of studies
Documents shortlisted by initial keywords search were downloaded as bibtex documents from Web of Science, Pubmed and Scopus databases individually and bibtex files were uploaded to Mendeley reference manager.Duplications were initially removed by Mendeley upon approval, remaining publications were then aligned according to document name and additional duplications were removed manually.Final document list for the initial screening was exported from Mendeley as an Excel sheet, and abstract of each article on the list was screened by both authors (CCS and ŞO) according to PECOS criteria listed on (Supplementary Table S2).Descriptive information including article title, publication type/year/journal, author name, study type, biomarker information, and prognostic information type were extracted from each abstract, entered into individual excel sheets named Initial Screening Results (ISR) separately by CCS and ŞO, and used for the initial elimination process.Book chapters, case reports, conference proceedings, comments, dissertations, editorials, guidelines, meeting abstracts, metaanalyses, reviews, and technical reports were excluded.Articles studying model organisms (i.e.mice, rat) and cell-culture studies that lack human patients were excluded.Studies that examine cancer types other than breast cancer patients were also excluded.Two investigators crosschecked each other's ISR-sheet; discrepancies regarding article selection were resolved by discussion; and documents to be included in the detailed evaluation were shortlisted as a Detailed Screening (DS) list.

Data extraction
During secondary selection and data extraction process, full-text contents of each article shortlisted in the DS-list and their supplementary materials were further examined and comprehensive data from the eligible studies were extracted as detailed on Table 1, Supplementary Tables S3 and S4, Table 2, Supplementary Tables S5 and S6.Respective data from the human studies examining biomarkers of extracellular origin (extracellular vesicle, exosome, microvesicle, etc.) reporting prognostic value on breast cancer patients which are compatible with the PECOS criteria (Supplementary Table S2) were extracted as Eligible Data (ED) by both authors (CCS and ŞO) separately.Two investigators crosschecked each other's ED-sheets; discrepancies were double checked until consensus was reached; and compiled data were further analysed by statistical evaluation.The impact factor (IF) of journals at the time of publication of the studies were extracted from https://www.scijournal.org/.In addition to reporting hazard ratios (HRs) and odd ratios (ORs), some studies also investigated the receiver operating characteristics (ROC) of the examined liquid biopsy-based EV-biomarkers in terms of their prognostic potential on metastatic progression, OS and PFS.Area under the curve (AUC) values of these biomarkers for these prognostic criteria were pooled and analysed.
Where multiple biomarkers were reported in a single study, only the biomarkers with AUC≥0.7 were accepted as effective markers and included in the final analyses.

Results
Initial keyword search (Supplementary Table S1) on Web of Science, PubMed, and Scopus databases revealed 492 articles accepted by the publishers as of 14 February 2023 (Figure 1).At the initial screening step, abstracts (and fulltext documents if accurate elimination was jeopardized by the lack of sufficient information in the abstract) of these articles were thoroughly reviewed and 148 articles were shortlisted for in-depth full-text screening.Detailed evaluation of 148 articles retrieved from scientific literature indicated that some of the reported prognostic biomarkers were actually not of extracellular vesicle origin; instead, they were serum biomarkers, freely circulating in the blood.These studies were excluded from this metaanalysis.As the aim of this study was to evaluate extracellular vesicle-based biomarkers circulating in body fluids, studies that report biomarkers from extracellular vesicles isolated directly from the tumour mass or other tissues were also excluded.In some studies, the prognostic roles of circulating exosomal biomarkers have been determined solely as a result of bioinformatics analysis of publicly available data present in databases such as TCGA or GEO.In most of these cases, the analysed biomarkers were mentioned as biomarkers obtained from liquid biopsy, yet it was not clear whether they were specifically isolated from extracellular vesicles, or they were solely circulating free biomarkers.Therefore, bioinformaticsbased studies were also excluded unless they indicated the utilization of databases specific for extracellular vesicles.Only the studies which clearly pointed out the examination of biomarkers present in the isolated extracellular vesicles, exosomes, or microvesicles recovered from liquid biopsies (Tables 1 and 2) were included in the final analysis.
One of these articles was found to be retracted, therefore this article was excluded.Three studies were further excluded from the final analysis as they lacked detailed statistical data such as standard error (SE), standard deviation (SD) or 95 % confidence interval (95 % CI).At the end of the detailed full-text screening, twenty-one articles comprising 26 studies and 3,423 patients were found to be eligible for this meta-analysis, which had reported prognostic values (OS, PFS, DFS, EFS, metastatic progression) of EV-based biomarkers recovered from liquid biopsies accompanied by the accurate statistical data (Table 1) .
Article information, study characteristics and biomarker information (Table 1), sample information (Supplementary Table S3) and patient data (Supplementary Table S4) were extracted from each article and summarized in Tables 2 and 3. Eligible studies were classified according to the type of results they provided (either AUC values or hazard ratio) and respective survival data (including statistical data) were extracted (Supplementary Tables S5 and S6) and analysed.For cases when the analysis type was not specified as univariate or multivariate, it was accepted to be univariate, and the respective data were included as univariate HR or univariate OR values in the final analysis.
EV is an umbrella term covering various membranous structures released by cells such as exosomes, microvesicles, apoptotic bodies, shedding vesicles, etc.While 13 of the eligible studies (61.9 %) specified the examined EV-type as exosome; two of the studies (9.5 %) specifically investigated microvesicles, and seven studies (33.3 %) mentioned the membraneous origin only as EV.Due to their exceptionally small size and peculiar biochemical properties, special techniques are applied for the isolation of extracellular membranous structures, such as ultracentrifugation-based techniques (differential ultracentrifugation, density gradient centrifugation), sizebased techniques (ultrafiltration, microfiltration, sequential filtration, size exclusion chromatography, flow field-flow Ceran Serdar and Osmanlıoğlu: Exosomal prognostic biomarkers in breast cancer fractionation, hydrostatic filtration dialysis), immunoaffinitybased techniques (enzyme-linked immunosorbent assay, magneto-immunoprecipitation), precipitation-based techniques (polyethylene glycol precipitation, lectin induced agglutination, immunoprecipitation), microfluidic techniques (acoustic nanofilter, immuno-based microfluidic isolation), chromatographic techniques (ion exchange chromatography, fast protein/high performance liquid chromatography), fluorescence-activated sorting, etc. [26,27].In addition to these in-house methods, various companies have developed commercial kits to aid researchers in their EV-oriented studies.Each isolation technique has its own strengths and impediments, having different tendencies towards higher recovery or higher specificity.As there is no gold standard for EV-isolation, the International Society for Extracellular Vesicles has advised that the optimal separation method is chosen according to the downstream requirements [26].Ultimate care must be taken during the isolation procedures, as misconduct leading to cross-contamination with cell-debris or large lipoproteins was associated with an increased risk of spurious results [26].Currently, especially for biomarker studies, it is strongly recommended and even obligated by an increasing number of authorities that, the results obtained through a particular separation method are verified by complementary techniques [28].Extracellular vesicles investigated in each of the twenty-one studies had been isolated by only a single technique (Supplementary Tables S1 and S3), calling the interference of potential non-vesicular or non-EV contamination into question.While 19.1 % of the studies used centrifugation or ultracentrifugation for EV-isolation, 71.4 % preferred to use commercial kits of various origins (Table 1).
Characterization of the EV-isolates is strongly necessitated in order to resolve the reservations regarding possible contamination with co-isolated materials, even when complementary techniques are used for EV-isolation.For characterization; 1) source and quantity of the EVs; 2) size; 3) presence of a lipid bilayer (presence of a transmembrane or GPI-anchored extracellular protein); 4) intactness of the isolated membranous vesicles (presence of cytosolic/periplasmic proteins with membrane protein-or lipid-binding abilities); 5) purity (absence of negative markers such as commonly expected contaminants); 6) subcellular origin (presence of proteins specific to certain subcellular compartments other than endosomes or plasma membrane); 7) association with extracellular proteins if present (such as growth factors, cytokines, extracellular matrix components) should be demonstrated [28].Characterization of single vesicles also has to be confirmed by at least two complementary techniques; 1) techniques providing images at high resolution such as, transmission electron microscopy (TEM), atomic-force microscopy (AFM), scanning-probe microscopy (SPM); 2) techniques estimating biophysical features such as nanoparticle tracking analysis (NTA), Raman spectroscopy, high-resolution flow cytometry, etc. [28].When twenty-one articles were inspected in this perspective; seven of them (33.3 %) were found to have characterised the EV-isolates by more than one technique, one of which was either NTA or TEM; six studies (28.6 %) have utilized only one method, four of which was NTA-based; and eight studies (38.1 %) have not reported any characterization method at all.61.9 % of the eligible studies have investigated the presence of at least one exosome-marker protein; most preferred marker was CD63 (52.4 %), followed by CD9 and CD81 (33.3 % each), TSG101 (14.3 %), and EpCAM (9.5 %).Nine studies (38.1 %) had not declared utilization of any exosomal-markers (Table 2).While some of the explored liquid biopsy-based EV-biomarkers had been associated with increased prognostic risk (will be mentioned as "negative biomarkers"), others were associated with decreased prognostic risk (will be mentioned as "positive biomarkers").As positive biomarkers and negative biomarkers have hazard ratios in opposite directions, they tend to cancel the effect of one another in a pooled analysis.Therefore, were first analysed positive biomarkers and negative biomarkers in separate subgroups, then the prognostic values of all biomarkers were combined in an overall score.
We analysed the prognostic values of EV-based nineteen mRNA, nine miRNA, three lncRNA, seventeen proteins and eight metabolites in total (Table 3).In addition, we included the prognostic values of the biomarker panels that had been reported in the eligible literature in this metaanalysis [5,15,[19][20][21][23][24][25].Although liquid biopsies include urine, saliva, semen, eyedrops, etc. in addition to serum and plasma; EV-based biomarkers investigated in all eligible studies were found to be only of blood origin (mentioned either as blood, serum or plasma in the respective articles).Therefore, although we have screened the literature for EV-based biomarkers present in all liquid biopsies, biomarkers from no other liquid biopsy origin were used in the pooled analyses.
Of the nineteen articles that reported hazard or odd ratios of the serum/plasma-based EV-biomarkers in breast cancer prognosis, ten have analysed the hazard/odd ratios via univariate analysis, five articles have utilized multivariate analysis, and four have used both univariate and multivariate analysis (Supplementary Table S6).Since odd ratios were presented only in one of these studies, pooled analyses were performed only from univariate or multivariate hazard ratios.
Only six of the eligible articles reported AUC values of the ROC curves analysing the prognostic potential of serum/ plasma-based EV-biomarkers [8, 19-21, 24, 25].One of these articles [19] and six others had reported AUC values for the diagnostic values of serum/plasma-based EV-biomarkers in addition to their hazard ratios [5,6,10,12,16,23] (Supplementary Table S5).Researchers have also investigated the distinctive role of some of these biomarkers in breast cancer diagnosis in general, or in differentiating between BC-subtypes such as luminal A, luminal B, Her2+ or TNBC.While subtype discrimination is also a feature related with prognostic potential, since the initial MeSH words did not include diagnosis-related words, we did not perform a pooled analysis in this aspect to avoid bias that would have otherwise been caused due to immature sampling.

The association between liquid biopsy-based EV-biomarkers and overall survival of breast cancer patients
We analysed fourteen EV-based biomarkers indicating worse overall survival (negative biomarkers), and five EV-based biomarkers indicating better overall survival (positive biomarkers) that were detected in liquid biopsy of breast cancer patients.Positive and negative biomarkers were initially analysed as separate subgroups.One of the fourteen EV-based negative biomarkers that were investigated in two separate studies, was "total EV concentration" ([total EV]) instead of a specific protein/DNA/RNA/metabolite encaptured in an EV.Highest hazard ratios were demonstrated for PD-L2 EV , total EV concentration ([total EV]), and sHLA-G EV (UHR=8.98,UHR=8.11,UHR=6.21 respectively) [8,9,21].Apart from the two biomarkers of metabolite origin, all of the examined negative-biomarkers had indicated significantly worse overall survival.The pooled univariate analysis we performed computing the hazard ratios of the fourteen EV-based negative biomarkers present in liquid biopsy indicated a high risk of mortality (UHR=2.31;95 % CI=1.77-3.03;I 2 =60.12 %; p<0.001; Figure 2A and B) [5-10, 13, 15-17, 21].Studies reporting the prognostic value of PD-L2 EV and [total EV] in terms of overall survival had also performed ROC analysis for these conditions [9,21]; and our pooled AUC analyses revealed a profound prognostic efficiency for their combination (pooled AUC=0.74 (95 % CI=0.64-0.83;I 2 =0.00 %; p=0.97; Figure 2E).Only two of the studies had reported positive EV-based biomarkers [17,18]; one of which had revealed prognostic values for four different EV-based miRNAs separately [17,18].Our subgroup analysis of these five EV-based positive biomarkers revealed a pooled univariate hazard ratio of UHR=0.32 (95 % CI=0.15-0.68;I 2 =82.44 %; p<0.001; Figure 2A); with miR-4800 being the best indicator for better overall survival (UHR=0.19;95 % CI=0.16-0.60 [18]).We found that the overall univariate hazard ratio of all the reported EV-based biomarkers indicative of mortality risk that can be recovered from liquid biopsy was 1.51 (95 % CI=1.08-2.12;I 2 =86.51 %; p<0.001; Figure 2A).

The association between liquid biopsy-based EV-biomarkers and progression-free survival of breast cancer patients
Of the twenty-one eligible studies, four had investigated the prognostic potential of the EV-based biomarkers in terms of progression-free survival [8,9,20,21], two in terms of disease-free survival [6,11], and one study had scrutinized the examined biomarkers for their potential in providing insight on all of the three (progression-free survival, diseasefree progression, event-free survival and pathological complete response) [19].Since disease-free and event-free survival may also be regarded as progression-free survival [27], we have pooled all the hazard ratios for PFS, DFS and EFS for a holistic progression-free point of view.Yet, as the majority of these studies preferably reported PFS, we analysed pooled hazard ratios for PFS also as an individual peer-group.Since odd ratios had only been reported by a single study [19] while hazard ratios were a common denominator of all studies, we performed pooled analyses only with the reported hazard ratios.Pooled univariate hazard ratios of negative EV-based biomarkers indicated significant association with worse PFS (UHR=3.76;95 % CI=2.48-5.68;I 2 =30.29 %; p=0.13; Figure 4A and B), and worse PFS/DFS/EFS combined (UHR=3.91;95 % CI=2.82-5.43;I 2 =19.16 %; p=0.24; Figure 3A and B).While the highest hazard ratios had been reported for [Total EV], miR-155 [20], sHLA-G EV [9], PD-L2 EV / ERBB3+CTC [21], and miR-1246 [20]; the most significant indicators were miR-155 and miR-1246 [20].As one would expect, reported multivariate analyses had revealed hazard ratios of higher significance.In parallel, pooled multivariate hazard ratios of the negative EV-based biomarkers revealed profound prognostic efficiency for predicting PFS-alone (Pooled AHR=4.73;95 % CI=2.96-7.56;I 2 =0.00 %; p=0.91; Figure 4C and D) and in combination with DFS/EFS (Pooled AHR=4.65;95 % CI=3.14-6.89;I 2 =12.74 %; p=0.33; Figure 3C  and D).The EV-based biomarker indicating the highest risk for worse PFS/DFS/EFS was by far Del-1 [11], which was followed by EV P -panel [19], miR-155 [20], and PD-L2 EV [21].miR-155 and miR-1246 were suggested as efficient indicators for both EFS and PFS [20].Three of the studies reporting the prognostic value in terms of PFS had also performed ROC analysis for these conditions [8,19,21]; and their pooled AUC value indicated profound prognostic potential (pooled AUC=0.80 (95 % CI=0.73-0.86I 2 =0.00 %; p=0.57; Figure 4E and  F).Positive EV-based biomarkers were not significant enough in predicting PFS either in a univariate setting (UHR=0.66;95 % CI=0.34-1.32;I 2 =0.00 %; p=0.98; Figure 4C  and D) or a multivariate setting (pooled analysis was not possible since multivariate HR value had been reported only for a single biomarker), as 95 % CI of their individual and pooled subgroup hazard ratios inclosed the value 1.00.Pooled-UHR (I 2 =19.16 %; p=0.24; Figure 3A and B) and AHR (I 2 =12.74 %; p=0.33; Figure 3C and D) results of the biomarkers indicating worse PFS/DFS/EFS both had limited heterogeneity.On the contrary, when only the studies disclosing prognostic values of biomarkers indicating worse PFS were pooled, the results were relatively homogeneous (I 2 =0.00 %; p=0.91; Figure 4C and D).

Liquid biopsy-based EV-biomarkers indicative of metastatic progression in breast cancer
Three of the eligible articles had reported AUC values regarding the prognostic potential of the liquid biopsy-Ceran Serdar and Osmanlıoğlu: Exosomal prognostic biomarkers in breast cancer

556
Ceran Serdar and Osmanlıoğlu: Exosomal prognostic biomarkers in breast cancer based EV-biomarkers for predicting metastatic progression [19,24,25].The pooled analysis we performed computing fourteen RNA-based, two miRNA-based, three proteinbased biomarkers, and a protein-based biomarker panel (EV DX ) indicated that, detection of these liquid biopsy-based EV-biomarkers reveals high risk for metastatic progression (Pooled AUC=0.91;95 % CI=0.86-0.95;I 2 =76.78 %; p<0.001; Figure 5A and B).While the discovery of nine biomarkers with absolute prognostic accuracy of AUC=1.00 in a single article is somewhat controversial; EV DX -panel and CA 125 appear to be strong predictors of metastatic progression in breast cancer.This is also reflected in the relatively high heterogeneity index (I 2 =76.78 %; p<0.001; Figure 5A), and the Funnel-graph (Figure 5B).

Discussion
To the best of our knowledge, the current study is the first meta-analysis evaluating the prognostic potential of EV-based biomarkers which are detected in liquid biopsies of breast cancer patients, and are suggested to predict overall survival, disease-free survival, and metastatic progression.
Extracellular vesicles (EVs) circulating in body fluids serve as messengers of the cells they originate from.Regarded as indicators of tumour shedding, they have been used in the diagnosis of different types of cancer.In addition to the information they carry from the tumour cells, they also contain cues from the tumour microenvironment.
Recently scientists are investigating the potential of extracellular vesicle-based cargos as prognostic biomarkers.In an effort to facilitate development of new treatment strategies, prognostic potentials of EV-based biomarkers reported in literature had previously been examined in various oncogenic backgrounds such as esophageal adenocarcinomas [29], pancreatic ductal adenocarcinoma [27], thyroid cancer [30], endometrial cancer [31], solid tumors [32], etc.
Detailed screening of the literature revealed twentyone articles eligible according to the selection criteria explained in Supplementary Tables S1 and S2 33].Albeit exosomal biomarkers dominated the pool having been investigated in more than two-thirds (65.2 %) of these studies; some of the documented biomarkers were of microvesicular (8.7 % of the studies) or of EV (30.4 %) origin (one of the studies examined biomarkers from two vesicular origins [13]).While no gold-standard technique is yet available for ultrasensitive and specific EV-isolation and detection free from non-vesicular contaminations; there are various EV-isolation methods available whose advantages and disadvantages are reported elsewhere [34], compensating for the shortcomings of one another.International Society for Extracellular Vesicles (ISEV) has advised that, especially for studies investigating diagnostic/ prognostic biomarkers, results of the studies should be confirmed by parallel analyses where extracellular vesicles are isolated via complementary techniques [26].However, all of the twenty-one articles eligible for this study had reported results for EV-based biomarkers which were obtained only by a single technique (Supplementary Tables S2  and S3), leading to reservations about the interference of potential non-vesicular or non-EV contamination.While most (83.6 %) of these studies utilized commercial kits for EV-isolation; to resolve presumptive disputes for an extracellular vesicular origin, one-third of them (34.8 %) had used more than one technique for EV-characterization, and almost two-thirds (60.9 %) had demonstrated presence of at least one exosome-marker proteins.Biomarkers that are associated with increased prognostic risk are referred in this manuscript as "negative biomarkers", while those that are associated with decreased prognostic risk are referred as "positive biomarkers".In order to prevent the counterbalancing between positive biomarkers and negative biomarkers that have hazard ratios in opposite directions, prognostic capacities of positive biomarkers and negative biomarkers were analysed in separate pools.
While multiple variables are investigated simultaneously in most studies aiming to identify diagnostic/prognostic biomarkers; univariate analyses are most frequently carried out instead of multivariate analyses, due to the ease of their statistical analyses and interpretation.As a simpler model, univariate analysis is easier to build, test and understand; yet occasionally leads to overestimations as it disregards inter-relation and inter-dependence among covariates.There is a common misguided assumption that, the parameters identified as discriminants in multivariate analyses will be present as a subset in the list of parameters identified as discriminants in univariate analyses.However, independent variables which do not appear significant in univariate analyses, occasionally complement each other and stand out as biomarkers associated with the investigated condition only in multivariate analyses, which will be missed unless univariate analyses are complemented with multivariate analyses [35].Among the twenty-one articles that were included in this meta-analysis, nineteen had reported hazard or odd ratios of the liquid biopsy-based EV-biomarkers in breast cancer prognosis.Of these nineteen studies, only nine (47 %) had utilized multivariate analysis to identify discriminating biomarkers (five studies only used multivariate analyses, and four had used both univariate and multivariate).Biomarkers identified through univariate analysis and multivariate analysis were pooled separately.In their detailed examination, Tian et al. have demonstrated that through sophisticated statistical evaluation, individual algorithms may be developed that will enable the utilization of the same set of biomarkers as distinct prognostic panels for detection of treatment efficiency, progression-free survival, or metastatic progression [19].Multivariate analysis of the data revealed in the other ten studies (53 %) that had reported univariate hazard ratios only might be useful for the identification of better prognostic biomarkers with higher sensitivity and specificity.
We identified liquid-biopsy based seventeen negative EV-biomarkers (twelve via univariate analysis; three via multivariate analysis; two via both), and five positive EV-biomarkers (all via univariate analysis) for indication of overall survival [5-10, 13, 15-18, 21].While pooled analysis of the negative biomarkers indicated significant high mortality risk (UHR=2.31;95 % CI=1.77-3.03;I 2 =60.12 %; p<0.001; Figure 2A and B and AHR=2.79;95 % CI=2.08-3.74;I 2 =0.00 %; p=0.51; Figure 2C and D); pooled analysis of the positive biomarkers indicated significantly better overall survival (UHR=0.32;95 % CI=0.15-0.68;I 2 =82.44 %; p<0.001; Figure 2A).Our results indicate that, the highest prognostic values were obtained by EV-based PD-L2, sHLA-GEV, exo-XIST, and miR4800, and the combination of these four EV-based biomarkers hold significant potential to be used as a new prognostic-panel for predicting OS of BrCa patients with better clinical performance.We suggest that a prospective cohort study evaluating the potential of a prognostic panel consisting of liquid biopsy-based fourteen EV-biomarkers having the highest significance ([Total EV], PD-L2 EV , sHLA-G EV , exo-XIST, EpCAM, MUC1, Exo-NGF-beta, lnc DANCR, miR148a, miR-4800, miR-4446, miR-2392, miR-2467 and CXCL13) in a multivariate setting might be useful for constructing the optimal algorithm that will enable the prediction of the mortality risk of breast cancer patients with higher accuracy.It should be noted that, while sHLA-G EV , exo-XIST, lnc DANCR and miR148a are among the biomarkers of the highest hazard ratios with better significance (95 % CI farther from HR=1), the studies that report these results were among the ones that failed to provide a detailed explanation for EV-characterization.Therefore, these biomarkers should be considered with reservation (Table 2).
According to the general aim of each study, researchers investigate healing processes either as freeing from progression (PFS), from any type of disease-related event (EFS), or from disease in general (DFS).Although each term seems to have slight variations in meaning from the other two, researchers generally tend to use these terms interchangeably.It is expected that the concentration of a biomarker indicative of a healing process will tend to change in the same direction at a comparable rate for each of these prognostic measures (either PFS, DFS or EFS).Illustration of this assumption comes from a study by Zhang et al., where miR-155 and miR-1246 were shown to have statistically indistinguishable hazard ratios for PFS and DFS [20].As anticipated, pooled-UHR values for liquid biopsy-based negative EV-biomarkers indicative of PFS-only or PFS/DFS/ EFS-combined were found to be similar for the studies used in this meta-analysis; pooled-PFS-UHR= 3.76; 95 % CI=2.48-5.68;(Figure 4A and B), pooled-PFS/DFS/EFS-UHR= 3.91; 95 % CI=2.82-5.43(Figure 3A and B).The study suggesting a prognostic potential to Del-1 for disease-free survival prediction, with an exceptionally high hazard ratio (UHR=32.80;95 % CI=6.40-168.90),had not reported detailed technical information for EV-isolation and characterization; therefore should be regarded with reservation until the results are verified by an additional study [11].Consequently, we suggest that a prognostic EV-panel consisting of biomarkers indicating worse prognosis ([total EV], PD-L2 EV , annexin-A2, miR-155, miR-1246, PSMA, sHLA-G EV ) might prove useful in predicting PFS, DFS, EFS.A prospective cohort study with a sufficient sample size would be required to determine the best algorithm for such a prognostic panel.
PD-L2 and sHLA-G stand out as significant prognostic biomarkers of breast cancer patient survival both in terms of overall survival and PFS/DFS/EFS.PD-L2 which is one of the two ligands of programmed cell death receptor PD-1, has been shown as a predictor of worse clinical outcome in numerous solid cancer types [36], including ER+ breast cancer [37].HLA-G which is a non-classical MHC class I molecule, performs significant immune-suppressive roles including prevention of maternal rejection of fetal tissues through maternal immune system.In addition to creating immunecompromised zones in the adult cornea, erythrocytes, and pancreatic islets, HLA-G has been shown to be upregulated in cancer.Expressed on extracellular vesicles, sHLA-G significantly engages in modulation of immune response during tumour progression [38].Through its interaction with immunoglobulin-like transcript 2 (ILT2), sHLA-G was shown to stimulate upregulation of immune checkpoint molecules such as PD-1 on CD8+ T-lymphocytes [33].HLA-G which was demonstrated to be upregulated together with PD-L1/L2 in various cancer types (i.e.pancreatic cancer, papillary thyroid cancer, various oral cancer types and adenoid cystic carcinomas of salivary glands) [38], attracts attention as a prominent biomarker of breast cancer patient survival [9,21,22].As sHLA-G EV and PD-L2 EV are both EV-biomarkers of protein origin, they can easily be combined in a prognostic-panel, offering significant potential to increase predictive power of breast cancer patient survival.In addition to their prognostic role, HLA-G and PD-L2 are also considered as immune-targets for cancer therapy.In parallel with the speculation which suggests that inhibition of HLA-G and PD-1/PD-L1/L2 axis might restore cytotoxic activities of T cells against tumour cells, clinical trials have been initiated which combine anti-HLA-G antibodies with agents blocking PD-1 /PD-L1/L2 axis for treatment of breast cancer patients (Clinical Trial ID: NCT04485013).
Tian et al. further investigated the eight EV-based protein biomarkers (CA 125, CA 15-3, CEA, EGFR, EpCAM, HER2, PSMA, and VEGF) and the prognostic biomarker panel composed by the combination of these eight biomarkers that they had proposed to predict progression-free survival (EV P ), for their potential to predict metastatic progression as well [19].While the same biomarkers participated in prediction of both prognostic events, the algorithm used for the prediction was different for each purpose; EV P for the prediction of progression-free survival; EV DX for the prediction of metastatic progression.Two other studies had investigated the potential of liquid biopsy-based EV-biomarkers to predict metastatic progression [24,25] by analysing the ROC curves of the biomarkers; the AUC values obtained from these studies were pooled with the results obtained by the EV DX panel.
While pooled AUC values of the liquid biopsy-based EV-biomarkers obtained from three studies [19,24,25] were very high (Pooled AUC=0.91;95 % CI=0.86-0.95;I 2 =76.78 %; p<0.001), the discovery of nine biomarkers with absolute prognostic accuracy of AUC=1.00 (with same 95 % CI) in a single article [24] has to be regarded with precaution and has to be verified in a larger cohort.Until then, EV DX appears to be the most optimal prognostic panel for the detection of metastatic progression of breast cancer.
As isolation, characterization and analysis of extracellular vesicles are still in their infancy; the investigation of liquid biopsy-based EV-biomarkers is relatively new in breast cancer diagnosis and prognosis.As of today, data available in the literature is far from being sufficient to enable statistically accurate evaluation of the prognostic potential of individual biomarkers.Therefore, in this metaanalysis we have compiled studies investigating liquid biopsy-based candidate EV-biomarkers that are suggested to predict overall survival, disease-free survival, and metastatic progression; aiming to unveil promising biomarkers and biomarker-panels for prognostic prediction of breast cancer.Although the prognostic potential of two of the biomarkers had been investigated by more than one study; the majority of the biomarkers were reported only in single studies.An increased number of systematic studies with large sample sizes in retrospective cohort design are required to reveal additional EV-biomarkers, report additional roles for already identified markers, and verify the current data already available in the literature.
Medical societies such as European Society for Medical Oncology (ESMO) and National Comprehensive Cancer Network (NCCN) periodically establish study groups to evaluate existing prognostic tumour markers and publish guidelines leading the clinicians to the most effective therapeutic options readily available for a particular molecular background [39,40].Once a molecular alteration is paired with an effective therapy type, and there is established association for improved outcome in clinical trials, the alteration/therapy pair is accepted for routine use (ESMO evidence tier -I) [39].The alteration/therapy pairs that are associated with antitumour activity with yet unknown magnitude of benefit (ESMO evidence tier -II) are accepted as "investigational".Currently there are only seven biomarkers accepted as Class-I [estrogen receptor (ER), progesterone receptor (PgR), HER2, programmed death-ligand 1 (PD-L1), germline BRCA1/2 mutation (gBRCA1/2m), phosphatidylinositol-4,5-bisphosphate 3-kinase catalytic subunit alpha (PIK3CA), neurotrophic tyrosine receptor kinase (NTRK)], and four additional biomarkers accepted as Class-II [estrogen receptor 1 (ESR1), partner and localiser of BRCA2 (PALB2), AKT1, androgen receptor (AR)] by ESMO [39].NCCN included RET-fusion as a Category-II biomarker [40].In addition, 21-gene panel Oncotype Dx, and 70-gene panel MammaPrint are classified as Category-I; 50-gene panel Prosigna, and 12-gene panel EndoPredict are classified as Category-II by NCCN for consideration of adjuvant systemic therapy [40].While some of these biomarkers are detected from tissue blocks, some can be analysed relatively noninvasively from blood biopsies.Yet, more scientific evidence is still required before the first EV-based biomarker could make it into the guidelines and be recommended for clinical use for breast cancer either for diagnostic or prognostic purposes.We believe that as additional significant clinical data is accumulated on behalf of EV-based biomarkers, researchers will further be interested in combining conventional markers and new EV-based biomarkers in diagnostic and prognostic panels to increase clinical sensitivity and specificity.
The current study, being the first meta-analysis evaluating the EV-based biomarkers associated with breast cancer prognosis, provides an overview of the liquid biopsy-based EV-biomarkers associated with overall survival, disease-free survival, and metastatic progression of breast cancer.As the prognostic potential of individual biomarkers is verified by an increasing number of research groups in multiple settings, it will be possible to evaluate each biomarker in separate meta-analyses and device much more accurate prognostic panels leading to facilitated follow-up and treatment of breast cancer patients, increasing the overall survival rates.

Figure 2 :
Figure 2: The association between liquid biopsy-based EV-biomarkers and overall survival of breast cancer patients.Pooled univariate hazard ratios (A, B) and pooled multivariate hazard ratios (C, D) of the liquid biopsy-based EV-biomarkers regarding overall survival are seen as forest graphs (A, C) and funnel graphs together with the Egger-test scores (B, D).Forest graph (E) and funnel graph (F) are also presented for pooled AUC values of the liquid biopsy-based EV-biomarkers for predicting the overall survival of breast cancer patients.

Figure 3 :Figure 4 :
Figure 3: The association between liquid biopsy-based EV-biomarkers and PFS/DFS/EFS of breast cancer patients.Pooled univariate hazard ratios (A, B) and pooled multivariate hazard ratios (C, D) of the liquid biopsy-based EV-biomarkers regarding PFS/DFS/EFS-combined (A, B, C, D) are seen as forest graphs (A, C) and funnel graphs together with the Egger-test scores (B, D).

Figure 5 :
Figure 5: The association of liquid biopsy-based EV-biomarkers with metastatic progression of breast cancer patients.Pooled AUC values of the liquid biopsy-based EV-biomarkers regarding the metastatic progression of breast cancer are seen as forest graphs A) and funnel graphs together with the Egger-test scores B).The EV M panel, and the individual biomarkers in that panel had been tested in three separate cohorts; TC, training cohort; VC, validation cohort; PC, prospective cohort.