This study is an umbrella review, defined as a systematic review of systematic reviews and meta-analyses. The objective was to aggregate and evaluate the therapeutic potential, mechanisms, and methodological quality of MSC-EV interventions in preclinical animal models. The review focused on meta-analyses to provide a high-level synthesis of evidence, capturing a broad range of diseases, exosome sources, and outcomes. The protocol was developed a priori and registered with PROSPERO to ensure transparency and reproducibility.

A comprehensive and systematic literature search was conducted to identify relevant meta-analyses. The search strategy was designed to capture studies evaluating MSC-EV therapeutic efficacy in preclinical models, with specific queries tailored to extracellular vesicles, mesenchymal stem cells, and meta-analyses (Table 1). The search was executed across multiple electronic databases, and the strategy was adapted from Table 1 of the provided article. The Scopus and Web of Science databases were searched from inception to July 2025 by two independent reviewers (N.M.M. and K.R.Z.) using standardized search protocols. Search results were exported to EndNote 20 for deduplication, and duplicates were removed using both automated and manual checks (Figure 1). The search strategy was validated by a medical librarian to ensure comprehensiveness and accuracy.

For inclusion in this umbrella review, studies were selected based on predefined inclusion and exclusion criteria to ensure both relevance and methodological quality. Eligible studies were systematic reviews that included meta-analyses of preclinical studies, specifically those investigating MSC-EVs—including exosomes, microvesicles, or other EV subtypes—as the primary therapeutic intervention. Studies combining MSC-EVs with other therapies, such as scaffolds or pharmacological agents, were included provided that MSC-EVs remained the central focus. The target population comprised preclinical animal models used to study a broad range of diseases or conditions. Included studies had to report quantitative outcomes relevant to therapeutic efficacy, such as functional assessments, histological evaluations, molecular biomarkers, or survival rates. Only English-language, peer-reviewed journal articles were considered.

Studies were excluded if they were narrative reviews, systematic reviews without meta-analyses, or primary research articles. Additional exclusion criteria included studies that focused on EVs not derived from MSCs, unless MSC-EVs constituted a major component of the analysis. Clinical trials or studies involving human subjects were excluded, as were meta-analyses limited solely to in vitro data. Non-English publications, conference abstracts, grey literature, preprints, and other non-peer-reviewed materials were also excluded from this review.

The study selection process was conducted in two distinct stages to ensure methodological rigor and transparency. In the first stage, titles and abstracts were independently screened by two reviewers (A.B. and M.A.K.). This initial screening was performed against the predefined eligibility criteria. Any discrepancies between the reviewers were resolved through discussion or, if necessary, by consulting a third reviewer (A.T.). In the second stage, the full texts of studies deemed potentially eligible were retrieved and independently evaluated by two additional reviewers (A.B. and M.A.K.) to determine their final inclusion. At this stage, specific reasons for exclusion were carefully documented. To provide a clear overview of the selection process, a PRISMA flow diagram was generated, outlining the number of records identified, screened, included, and excluded at each phase of the review (Figure 1).

Data extraction was carried out independently by two reviewers (N.M.M. and K.R.Z.) using a standardized form developed in Microsoft Excel. This form was piloted on five studies to ensure consistency, clarity, and completeness in data capture. After extraction, data were cross-verified for accuracy by the reviewers. Any inconsistencies were resolved through consensus or, when necessary, by consulting a senior author (A.T.).

The data extraction encompassed several key elements. For study characteristics, information was collected on the authors, year of publication, journal name, and reference number, along with the total number of studies included in each meta-analysis and the specific disease or condition being investigated. Intervention details included the type of MSC-EVs, the origin of the MSCs, and the method of delivery.

Regarding animal models, data were gathered on the species used, the specific strains, and the experimental disease models employed. Outcomes extracted included both primary outcomes and secondary outcomes. Where available, effect sizes such as standardized mean differences (SMD), weighted mean differences (WMD), hazard ratios (HR), or odds ratios (OR) were recorded, along with their corresponding 95% confidence intervals. Measures of heterogeneity, such as the I² statistic, were also documented.

In terms of methodological quality, each study’s risk of bias was assessed using established tools like SYRCLE or CAMARADES. The overall risk of bias was categorized as low, moderate, high, or unclear. Evaluation of publication bias included methods such as Egger’s test and visual inspection of funnel plots. Furthermore, the AMSTAR 2 tool was used to appraise the methodological quality of the included systematic reviews and meta-analyses, with ratings categorized as high, moderate, low, or critically low, and critical flaws explicitly noted. Data were extracted from main texts, tables, and Supplementary Material. When numerical data was missing, attempts were made to contact the original authors for clarification. In cases where no response was obtained, data were estimated from graphical figures.

To evaluate the methodological rigor of the included meta-analyses and the risk of bias in the primary studies they synthesized, two complementary assessment tools were employed. The AMSTAR 2 was used to appraise the overall quality of the included meta-analyses. Two independent reviewers (A.B. and G.A.T.) applied the 16-item checklist, with particular attention to critical domains such as protocol registration (item 2), comprehensiveness of the literature search strategy (item 4), justification for excluded studies (item 7), risk of bias assessment of included studies (item 9), appropriateness of the meta-analytic methods (item 11), and consideration of publication bias (item 15). Based on the number and severity of critical flaws identified, each meta-analysis was rated as having high, moderate, low, or critically low confidence in its findings. Any disagreements between reviewers were resolved through discussion and consensus. AMSTAR-2 ratings were assigned according to the number of critical domains rated ‘No.’ Reviews with ≥1 critical flaw were downgraded to low or critically low confidence.

The risk of bias in the primary studies included within each meta-analysis was assessed using the tools employed by the original meta-analyses themselves. The most commonly used instruments were the SYRCLE risk of bias tool and the CAMARADES checklist. These tools evaluated key domains of bias, including selection bias, performance bias, detection bias, attrition bias, and reporting bias. The overall risk of bias for each meta-analysis—categorized as low, moderate, high, or unclear—was recorded as reported in the studies. If a meta-analysis utilized a custom or non-standard assessment tool, its specific criteria were documented accordingly.

To improve clarity, we distinguished the use of the SYRCLE and CAMARADES tools based on the model type and reporting structure of the original meta-analyses. Specifically, the SYRCLE tool was applied when the included meta-analysis assessed basic animal studies with heterogeneous outcomes such as behavioral scores, histological findings, or inflammatory markers. In contrast, the CAMARADES checklist was used when analyzing more structured preclinical models—particularly in neurological and cardiovascular studies—where endpoints such as infarct volume, mNSS, or neurobehavioral scores were commonly and consistently reported. In instances where both tools were used or a modified version was employed, we recorded that distinction accordingly in Table 4.

Data were synthesized both narratively and quantitatively to comprehensively evaluate the therapeutic efficacy of MSC-EVs across various diseases, exosome sources, and outcome measures. The synthesis was structured to align with the objectives of the umbrella review, with a particular focus on therapeutic effectiveness, underlying mechanisms of action, and the methodological quality of the included meta-analyses.

A narrative synthesis was performed to describe the diversity of conditions addressed in the included studies, the types and tissue sources of MSC-EVs used, the animal models employed, and the administration routes applied. This synthesis also outlined the primary outcomes assessed, their consistency across studies, and the proposed mechanisms of action, such as anti-inflammatory, anti-apoptotic, and regenerative effects. Findings were organized into comprehensive tables and illustrative figures to facilitate interpretation and comparison. For instance, Table 3 presents a detailed summary of exosome-based therapies across different diseases and conditions, while visual aids such as bar graphs and merged heatmaps were used to depict data trends and outcome distributions.

In the quantitative synthesis, effect sizes, heterogeneity measures, and statistical significance were summarized based on the results reported in the included meta-analyses. Key metrics included SMD, WMD, HR, and OR, all accompanied by 95% confidence intervals. These metrics were typically reported for primary outcomes such as functional recovery scores, wound healing rates, or infarct volume reduction. Heterogeneity across studies was assessed using the I² statistic, with values greater than 50% considered indicative of substantial variability. Where available, subgroup analyses or sensitivity analyses were reported to explore sources of heterogeneity. Publication bias was evaluated based on the original meta-analyses.

No additional meta-analyses were conducted within this umbrella review, as the aim was to synthesize and evaluate existing meta-analytic evidence rather than generate new pooled estimates. However, reported effect sizes were qualitatively summarized to identify therapeutic trends—for example, MSC-EVs demonstrated high efficacy in preclinical models of stroke and moderate effects in kidney transplantation models.

Because umbrella reviews synthesize findings from published meta-analyses without re-analyzing primary studies, we did not exclude individual studies on the basis of heterogeneity. Instead, we applied a rule-based classification: outcomes were labeled as High effectiveness only when SMD >1.5, p < 0.01, and I² < 70% in ≥2 independent meta-analyses. Outcomes with I² ≥ 70% were reclassified as Promising but heterogeneous and interpreted with caution. Sensitivity summaries were added to indicate whether conclusions remained robust after considering only meta-analyses with I² < 70% and without AMSTAR-2 critical flaws.

Because this is an umbrella review, we did not exclude meta-analyses solely on the basis of high heterogeneity. Instead, we applied a rule-based classification: outcomes were labeled as High effectiveness only when SMD >1.5, p < 0.01, and I² < 70% in ≥2 independent reviews. Outcomes with I² ≥ 70% were reclassified as Promising but heterogeneous and interpreted with caution.

Subgroup analyses reported within the included meta-analyses were extracted to identify factors that may influence the therapeutic efficacy of MSC-EVs. These analyses explored variations based on the source of exosomes—such as bone marrow-derived MSCs (BM-MSCs), adipose-derived MSCs (AD-MSCs), and human umbilical cord-derived MSCs (hUC-MSCs)—as well as animal model characteristics, including species and specific strains used in the experiments. Differences in disease models were also considered, such as contusion versus compression injury models for SCI, to evaluate how pathophysiological variations affect outcomes.

Additional subgroup variables included the route of MSC-EV administration and the timing and dosage of EV delivery. These factors were examined to determine their potential role in modulating therapeutic effectiveness across studies.

Sensitivity analyses conducted within the original meta-analyses were also summarized. These included procedures such as excluding studies with a high risk of bias to test the stability of the main findings, as well as statistical methods like trim-and-fill adjustments to evaluate the impact of publication bias. Together, these subgroup and sensitivity analyses provided important insights into the robustness and generalizability of MSC-EV therapy outcomes across different experimental conditions.

As this study involved no primary data collection or animal experimentation, ethical approval was not required. However, the review considered the ethical conduct of included studies, noting compliance with animal welfare regulations as reported by the meta-analyses.

Several tools were employed to facilitate data management and ensure methodological consistency throughout the review process. EndNote 20 was used for reference management and to identify and remove duplicate records prior to screening. For data extraction and the creation of summary tables, Microsoft Excel was utilized, offering a structured format to capture and organize information efficiently. Additionally, RStudio was employed to generate heatmap graphs, enabling visual representation of data patterns and relationships derived from the synthesized findings.

No new statistical analyses were performed in this umbrella review, as its primary goal was to synthesize and interpret results from existing meta-analyses. However, statistical metrics reported in the included studies were carefully reviewed and verified for accuracy to ensure the reliability of the synthesized findings.

MSC-EVs demonstrated high therapeutic efficacy across most evaluated diseases, with consistent improvements in functional, histological, and molecular outcomes (Figure 2). The following summarizes key findings by disease category (Figure 3; Supplementary Table S1). MSC-EVs consistently reduced inflammation and apoptosis, while enhancing functional scores and histological repair. Effectiveness was high across most conditions, with bone marrow-derived MSC-EVs (BMSC-EVs) and preconditioned EVs showing superior results, though heterogeneity was moderate to high and risk of bias varied. The classification of therapeutic effectiveness into “high” and “moderate” was based on reported meta-analytic metrics. “High” effectiveness was assigned to outcomes with standardized mean difference (SMD) > 1.5, p < 0.01, and low-to-moderate heterogeneity (I² < 70%) observed in at least two independent meta-analyses. “Moderate” effectiveness was applied to outcomes with SMD values between 0.8 and 1.5 or when heterogeneity exceeded 70%.

MSC-EVs exert their therapeutic effects through a range of interconnected biological mechanisms. These mechanisms contribute to the regenerative and protective roles of MSC-EVs in various pathological conditions.

One of the most prominent mechanisms is the anti-inflammatory effect. MSC-EVs were consistently shown to downregulate proinflammatory cytokines such as tumor necrosis factor-alpha (TNF-α), interleukin-1 beta (IL-1β), and interleukin-6 (IL-6), while simultaneously upregulating anti-inflammatory mediators including interleukin-10 (IL-10) and transforming growth factor-beta 1 (TGF-β1). These immunomodulatory effects were observed across multiple disease models, particularly in stroke, SCI (SCI), acute kidney injury (AKI), and asthma.

Anti-apoptotic effects were also widely reported. MSC-EVs reduced markers of apoptosis, such as caspase-3 and Bax, in neurological, renal, and cardiovascular models. By inhibiting apoptotic pathways, MSC-EVs helped preserve tissue integrity and cell viability in damaged organs.

Functional improvements were another key therapeutic outcome, with enhanced performance in disease-specific scoring systems such as the Basso, Beattie, Bresnahan (BBB) score for SCI, the modified Neurological Severity Score (mNSS) for stroke, and the Osteoarthritis Research Society International (OARSI) score for joint degeneration. These improvements were largely attributed to mechanisms such as neuroregeneration, angiogenesis, and overall tissue repair facilitated by MSC-EVs.

Finally, histological improvements supported the regenerative potential of MSC-EVs. Across studies, MSC-EVs were shown to stimulate collagen deposition, promote angiogenesis and neurogenesis, and reduce fibrosis, lesion size, and tissue damage. These histological changes were particularly evident in models of wound healing, liver fibrosis, and kidney disease, underscoring the broad-spectrum therapeutic action of MSC-EVs across organ systems.

Across conditions such as ischemic stroke, diabetic wounds, SCI, and acute kidney injury, MSC-EVs significantly reduced inflammation, apoptosis, and tissue damage while enhancing functional recovery and histological repair (Table 3). BMSC-EVs, adipose-derived MSC-EVs (ADSC-EVs), and preconditioned EVs showed superior efficacy in conditions like ischemic stroke, diabetic wounds, and multiple sclerosis, with notable improvements in neurovascular repair, wound closure, and clinical scores. However, effectiveness was low in kidney transplantation, where MSC-EVs showed no significant benefit. Consistency across studies was moderate (I² = 23–95%) for most conditions, with high heterogeneity in bone injury (I² = 97–98%) and acute kidney injury (I² = 96%), likely due to variability in animal models, exosome sources, and administration methods. For disease areas where heterogeneity was very high (I² ≥ 70%), such as bone injury and acute kidney injury, the results were reclassified as Promising but heterogeneous. While these conditions showed large effect sizes, the variability across studies limits certainty in the pooled estimates. For such disease areas with I² ≥ 70%, outcomes were downgraded to Promising but heterogeneous. While effect sizes were large, the variability across studies limits the certainty of pooled estimates.

Administration routes varied substantially across conditions. Intravenous delivery was the predominant method in most disease models, including renal and hepatic injury. For CNS models such as spinal cord injury and ischemic stroke, intrathecal, intranasal, or intracerebroventricular administration was frequently used and, in some cases, demonstrated greater efficacy by enabling direct delivery across the blood–brain barrier. For local diseases such as diabetic wounds and periodontal regeneration, local injections or hydrogel/scaffold-based delivery systems were commonly applied, supporting tissue retention and enhancing therapeutic benefit. These findings, summarized in Table 4, indicate that administration route is an important factor influencing MSC-EV efficacy and should be tailored to the target organ and disease.

Across the included meta-analyses, MSC-EV doses varied widely depending on disease model, administration route, and MSC source. The reported doses ranged from as low as 2 μg to as high as 700 μg of EV protein per injection, or from 1 × 10⁵ to 1 × 10¹¹ particles per dose. Most studies administered EVs intravenously, although intranasal, intrathecal, subcutaneous, intrauterine, and local delivery via hydrogels or scaffolds were also frequently reported. A new supplementary table (Table 4) was created to summarize these dosing parameters, including dose units, routes, and whether dose-response relationships were investigated. Among the reviewed studies, approximately one-third conducted some form of dose-response assessment, with 100 μg per injection emerging as a commonly effective dose across multiple conditions, including spinal cord injury, ischemic stroke, and diabetic wound healing.

To integrate the evidence across sources and disease categories, we created a Bubble chart (Figure 4) mapping MSC-EV sources against disease models. This visualization includes only meta-analyses with AMSTAR-2 high or moderate confidence and I² < 70%. Cells indicate the number of supporting meta-analyses, with darker shading representing stronger evidence. Hollow dots mark disease–source pairs where evidence exists but heterogeneity was high (I² ≥ 70%). This figure highlights consistent support for BM-MSC-EVs in neurological diseases (stroke, SCI), AD-MSC-EVs in diabetic wound healing, and UC-MSC-EVs in musculoskeletal and periodontal regeneration.

The therapeutic efficacy of MSC-EVs varied notably depending on their cellular source (Table 5). Among the sources, BM-MSCs were the most extensively studied, with approximately 308 studies. These EVs demonstrated high effectiveness across multiple conditions, including ischemic stroke, SCI, acute kidney injury, and cardiovascular diseases. BMSC-EVs were particularly effective in reducing infarct size, improving neurological function scores, and promoting neuroregeneration.

AD-MSCs, represented in about 154 studies, showed the highest efficacy in the treatment of diabetic wounds. These EVs promoted angiogenesis and accelerated wound closure, and also demonstrated consistent therapeutic benefits in models of liver fibrosis and chronic kidney disease.

hUC-MSCs, reported in around 119 studies, were most effective in models of knee osteoarthritis, periodontal tissue regeneration, and skin wound healing. hUC-MSC-EVs consistently reduced inflammation and improved functional outcomes across various disease models.

EVs derived from other MSC sources, such as menstrual blood, synovial tissue, and dental pulp, were less frequently studied but showed high therapeutic potential in specific conditions. For example, EVs from menstrual blood and synovial MSCs were effective in intrauterine adhesion and osteoarthritis, respectively, while periodontal ligament-derived EVs showed strong efficacy in models of multiple sclerosis and periodontal regeneration.

Notably, modified or engineered EVs—such as those loaded with specific microRNAs or preconditioned under hypoxic conditions—often outperformed their native counterparts. These engineered vesicles showed enhanced efficacy in models of stroke, SCI, and diabetic wounds. The method of EV delivery also influenced outcomes to some extent; while hydrogels and scaffold-based approaches were used in several studies, no delivery method demonstrated consistent superiority over direct injection.

The methodological rigor of the included meta-analyses and their underlying primary studies revealed several key challenges (Table 6). Most reviews reported a moderate to high risk of bias, assessed using tools such as SYRCLE and CAMARADES. Common methodological shortcomings included unclear random sequence generation, lack of blinding of personnel and outcome assessors, and insufficient details regarding allocation concealment. Furthermore, publication bias was detected in several high-interest disease models—including stroke, SCI, and diabetic wounds—although many findings remained robust after trim-and-fill adjustments. Across the included reviews, the most frequent biases were inadequate or unclear random sequence generation, lack of blinding of investigators and outcome assessors, and insufficient allocation concealment. These issues were consistently reported in the majority of meta-analyses and represent systemic weaknesses in preclinical MSC-EV research.

In terms of methodological quality, all included meta-analyses received a moderate AMSTAR 2 rating (Figure 5; Supplementary Table S2). This was primarily due to high heterogeneity (with I² values ranging from 35% to 99%) and limited reporting of essential methodological components such as randomization procedures and blinding. Important methodological shortcomings were identified in several studies, particularly incomplete or unclear risk of bias assessments and lack of consideration for publication bias. Where AMSTAR-2 critical domains were rated ‘No,’ these reviews were classified as low or critically low confidence.

Heterogeneity was a significant concern across the dataset, with I² values often exceeding 70%. This variability was largely attributed to differences in animal models, MSC sources, EV dosages, and delivery routes. Despite this, sensitivity and subgroup analyses frequently confirmed the robustness of results, suggesting that the therapeutic effects of MSC-EVs were consistent across different experimental conditions.

The review suggests that MSC-EVs exhibit high efficacy across multiple disease categories, including neurological, renal, wound healing, liver, musculoskeletal, respiratory, and reproductive disorders. Notably, MSC-EVs consistently reduced inflammation and apoptosis while promoting tissue regeneration, angiogenesis, and functional recovery. For instance, in ischemic stroke, MSC-EVs reduced cerebral infarct volume (SMD -3.76) and improved neurological scores (mNSS; SMD -2.11), with BMSC-EVs showing superior efficacy (Zhao et al., 2023). Similarly, in diabetic wounds, adipose-derived EVs (ADSC-EVs) accelerated wound closure (SMD 4.22) and enhanced angiogenesis (SMD 9.27), highlighting their potential in regenerative medicine (Soltani et al., 2024).

These findings align with the broader literature on MSC-EVs, which emphasizes their role as bioactive mediators carrying microRNAs, proteins, and lipids that modulate cellular processes. The high efficacy observed in conditions like SCI and traumatic brain injury, where MSC-EVs improved locomotor scores (BBB; WMD 3.47) and cognitive outcomes (mNSS; SMD -4.48), underscores their neuroprotective and regenerative capabilities (Chen et al., 2024; Ye et al., 2024). The ability of MSC-EVs to outperform conditioned medium in acute kidney injury (Liu C. et al., 2020; Zhang G. et al., 2016) and to match or exceed MSC-based therapies in subarachnoid hemorrhage (He et al., 2022) further supports their therapeutic advantage, likely due to their stability, low immunogenicity, and ability to cross biological barriers.

The clinical implications are significant. MSC-EVs offer a cell-free therapeutic approach that circumvents challenges associated with MSC transplantation, such as immune rejection and tumorigenic risks. Their efficacy in diverse preclinical models suggests potential for broad clinical applications, particularly in conditions with high unmet needs, such as stroke, SCI, and diabetic complications. However, the variability in efficacy across diseases highlights the need for disease-specific optimization of EV sources, dosing, and delivery methods.

The review reveals that exosome source significantly influences therapeutic outcomes. AD-MSC-EVs excelled in wound healing, particularly diabetic wounds, where they promoted angiogenesis and collagen deposition, while BM-MSC-EVs demonstrated superior effects in neurological models. hUC-MSCs showed superior efficacy in knee osteoarthritis and periodontal regeneration, possibly due to their high proliferative capacity and immunomodulatory properties.

Emerging sources, such as periodontal ligament (PDLSCs) for multiple sclerosis (Xun et al., 2022) and menstrual blood (MenSCs) for intrauterine adhesion (Chen et al., 2023), demonstrated high efficacy despite fewer studies, suggesting untapped potential. Modified EVs, such as miRNA-loaded or hypoxia-pretreated EVs, consistently outperformed native EVs, as seen in SCI (Hu et al., 2021; Liu W. et al., 2020; Yang et al., 2024) and stroke (Li et al., 2023; Song et al., 2024), where engineered EVs enhanced functional recovery by targeting specific pathways. These findings align with recent studies emphasizing the role of EV cargo engineering in enhancing therapeutic specificity.

Delivery methods also influenced outcomes. Intravenous and intrathecal routes were most common, with intrathecal administration showing superior efficacy in SCI. Hydrogels and scaffolds improved outcomes in some contexts, but their benefit was not universal, as seen in diabetic wounds where non-hydrogel methods were equally effective (Bailey et al., 2022; Chen et al., 2025). These observations underscore the need for tailored delivery strategies based on disease pathophysiology and target tissue.

The administration route is another determinant of therapeutic outcomes. While intravenous delivery remains the most frequently used method, it may not be optimal for all disease contexts. For CNS conditions, intrathecal and intranasal delivery were more effective in bypassing the blood–brain barrier and enhancing neuroprotective outcomes. For local pathologies, such as wounds and periodontal disease, local injection and hydrogel-mediated delivery improved retention and tissue-specific effects. These observations underscore the need for future preclinical and clinical studies to systematically evaluate route-dependent biodistribution and efficacy of MSC-EVs.

This crosswalk illustrates the concentration of higher-quality evidence, showing clear clusters of BM-MSC-EVs with neurological models, AD-MSC-EVs with wound healing, and UC-MSC-EVs with musculoskeletal and periodontal regeneration. These patterns emphasize the importance of tailoring MSC-EV source selection to disease context.

One critical but under-addressed variable in MSC-EV therapy is dosing strategy. Our umbrella review found substantial variability in reported doses, with most studies using a fixed dose (often 100 μg) without justification or titration. While several studies—such as those on SCI, stroke, and reproductive models—performed subgroup or network meta-analyses to examine dose-response relationships, the overall evidence remains fragmented and underpowered. In some cases, 100–200 μg was reported as optimal for neuroprotection or tissue regeneration, yet other studies used much higher doses (up to 700 μg) or particle-based quantifications (1 × 10⁹ to 10¹¹ particles).

The lack of standardized dosing metrics (mass vs. particle count), inconsistent reporting of EV characterization, and variable injection regimens further complicate cross-study comparisons. Notably, some studies administered EVs via specialized delivery systems, which could enhance local bioavailability and reduce systemic loss. However, head-to-head comparisons across these delivery platforms remain limited.

To support clinical translation, future preclinical trials should incorporate formal dose-response analyses, adopt standardized reporting in line with MISEV2023 guidelines, and evaluate pharmacokinetics and tissue distribution in parallel with efficacy outcomes (Su et al., 2025).

The therapeutic effects of MSC-EVs are mediated through multiple mechanisms, including anti-inflammatory, anti-apoptotic, and regenerative pathways (Liao et al., 2022). The consistent reduction in proinflammatory cytokines and upregulation of IL-10 across diseases like asthma, sepsis, and liver fibrosis highlight their immunomodulatory role. In neurological disorders, MSC-EVs reduced neuronal apoptosis and promoted neurogenesis and axonal regeneration, contributing to functional recovery (Dabrowska et al., 2020). In wound healing, enhanced angiogenesis and collagen deposition were driven by EV-mediated delivery of growth factors and microRNAs (Pulido-Escribano et al., 2023).

These mechanisms are consistent with the literature, which attributes MSC-EV efficacy to their cargo of bioactive molecules, including miRNAs, proteins, and lipids. The ability of MSC-EVs to modulate multiple pathways simultaneously explains their broad efficacy but also complicates efforts to pinpoint specific mechanisms for each disease (Tsuji et al., 2020). Future studies should leverage omics technologies to elucidate disease-specific EV cargos and their targets, facilitating precision medicine approaches.

A major limitation across the evidence base is the prevalence of randomization bias, lack of blinding, and inadequate allocation concealment, as summarized in Table 6. These issues undermine internal validity and may inflate reported effect sizes. The review identified significant methodological challenges that temper the interpretation of findings. Most meta-analyses reported moderate to high risk of bias, primarily due to unclear randomization, lack of blinding, and inadequate allocation concealment in primary studies. The SYRCLE and CAMARADES tools highlighted these issues, with only a few studies achieving low risk across all domains. High heterogeneity (I² often >70%) was another concern, driven by variations in animal models, EV sources, doses, and administration protocols. While sensitivity analyses and trim-and-fill adjustments often confirmed robust findings, publication bias was evident in conditions like stroke and SCI, suggesting a potential overestimation of effect sizes. Although some outcomes showed very large effect sizes, they were accompanied by high heterogeneity (I² ≥ 70%). In this umbrella review, we did not exclude these results but reclassified them as Promising but heterogeneous to preserve comprehensiveness while reflecting their limited certainty.

The AMSTAR 2 assessments rated all meta-analyses as moderate quality, reflecting limitations in reporting randomization, blinding, and publication bias assessments. The lack of standardized EV characterization further complicates comparisons across studies. These methodological issues align with broader challenges in preclinical research, where poor reporting and experimental design can undermine reproducibility (Simon-Tillaux et al., 2022).

The umbrella review itself has limitations. The restriction to English-language studies may have excluded relevant non-English meta-analyses (Wang et al., 2015). The reliance on reported data from included meta-analyses meant that incomplete or inconsistent reporting could affect our synthesis. Additionally, the diversity of diseases and outcomes precluded a formal meta-analysis of effect sizes, limiting our ability to quantify overall efficacy.

Because umbrella reviews rely on published meta-analyses, we cannot exclude or re-pool individual primary studies. Instead, we downgraded evidence strength for outcomes with I² ≥ 70% to Promising but heterogeneous. This ensures transparency while retaining the comprehensive scope of the umbrella review.

Several limitations must be considered when interpreting the findings of this umbrella review. Study quality was a notable concern, as poor reporting of critical methodological aspects such as randomization, blinding, and allocation concealment limited the reliability of some conclusions. Many primary studies scored between 3 and 7 on the SYRCLE scale, reflecting low to moderate methodological quality.

Several included reviews were of low or critically low confidence according to AMSTAR-2, and while retained for completeness, sensitivity summaries excluding these reviews are presented to indicate robustness of conclusions.

Future preclinical MSC-EV studies should implement rigorous randomization and blinding, with transparent allocation concealment, in line with ARRIVE reporting standards, to improve the reliability of pooled evidence.

Publication bias was evident in numerous conditions, including stroke, SCI, and post-operative ileus, as indicated by asymmetrical funnel plots and significant Egger’s or Begg’s test results. However, subsequent trim-and-fill analyses often confirmed the stability of the observed effects, lending credibility to the synthesized outcomes.

Another issue was the variability in exosome characterization. Some studies did not include essential quality control data, such as electron microscopy images or expression analysis of EV surface markers, which may affect the comparability and reproducibility of MSC-EV therapies.

Lastly, translational challenges remain. While MSC-EVs demonstrated high efficacy across a range of preclinical disease models, differences in dosing regimens, timing of administration, and delivery strategies must be standardized to advance these findings toward clinical application.

Several key priorities have emerged to guide future research on MSC-EVs, with the goal of enhancing scientific rigor and accelerating clinical translation. First and foremost, there is a critical need for standardization. Uniform protocols for EV isolation, characterization, and dosing must be developed and widely adopted to ensure reproducibility and comparability across studies. In this context, strict adherence to the MISEV2023 (Minimal Information for Studies of Extracellular Vesicles) guidelines should be considered essential (Welsh et al., 2024).

In addition, mechanistic studies should be expanded using advanced omics technologies—such as proteomics, transcriptomics, and metabolomics—alongside bioinformatics tools, to elucidate disease-specific EV cargos and their molecular targets. Such insights will support the development of more tailored and effective therapeutic strategies. Optimization of MSC-EV therapies is another important area of focus. This includes exploring novel and less-studied EV sources, such as PDLSCs and MenSCs, as well as employing bioengineering strategies like microRNA loading or surface modification to enhance therapeutic potency.

Across the included meta-analyses, the most commonly reported methods were hypoxic preconditioning, miRNA-engineering, cytokine/growth factor priming, and scaffold-based conditioning. These preconditioning approaches were consistently associated with improved therapeutic efficacy, including enhanced angiogenesis, neuroprotection, and anti-inflammatory effects. For example, hypoxia-enhanced EVs showed superior functional outcomes in spinal cord injury models, while miRNA-modified EVs demonstrated targeted regulation of inflammatory and regenerative pathways (Jiang et al., 2025). Scaffold incorporation also supported sustained EV release and localized tissue repair (Leung et al., 2022). These findings suggest that preconditioning may be a key determinant of EV potency, and future research should prioritize standardized evaluation of these strategies (Liu et al., 2025).

For clinical translation, the field must now progress toward conducting early-phase clinical trials (Phase I/II) to assess the safety, tolerability, and efficacy of MSC-EVs in human subjects. Priority should be given to high-impact conditions where preclinical data already show strong therapeutic potential, such as ischemic stroke and diabetic wounds. Alongside these translational efforts, improving methodological rigor in preclinical studies is crucial. This involves proper implementation of randomization, blinding, and allocation concealment, with transparent reporting practices aligned with the ARRIVE (Animal Research: Reporting of In Vivo Experiments) guidelines.

Finally, addressing publication bias remains a vital consideration. The use of prospective study registration and open-access data platforms can help ensure that both positive and negative results are reported, thereby strengthening the integrity of the evidence base. By tackling these research priorities, the field can move toward more reliable, effective, and clinically applicable MSC-EV therapies.