Therapeutic potential of mesenchymal stem cell-derived extracellular vesicle in nonalcoholic fatty liver disease: a systematic review and meta-analysis of preclinical evidence

Abstract

Objective

Nonalcoholic fatty liver disease (NAFLD) is a global chronic health challenge, demanding the development of innovative therapeutic strategies. Mesenchymal stromal cell-derived extracellular vesicles (MSC-EVs) have emerged as a promising therapeutic approach for NAFLD; however, current evidence is limited to preclinical studies. This systematic review and meta-analysis assessed the therapeutic efficacy of MSC-EVs in rodent models of NAFLD and its progressive form, nonalcoholic steatohepatitis (NASH). By synthesizing preclinical data, we aim to establish a robust evidence base that can guide future clinical trials and optimize MSC-EV-based therapies.

Methods

Comprehensive searches of the PubMed, Web of Science, Embase, CNKI, Wanfang, and VIP databases identified eligible animal studies. Methodological quality was assessed via the SYRCLE risk-of-bias tool. The meta-analyses were conducted following Cochrane Handbook guidelines via Stata 18.0.

Results

MSC-EVs led to significant reductions in key metabolic parameters, including AST (SMD = -2.79, 95% CI [-3.64, -1.94], p< 0.01), ALT (SMD = -2.47, 95% CI [-3.44, -1.50], p < 0.01), TG (SMD = -1.86, 95% CI [-2.98, -0.73], P < 0.01), liver TG (SMD = -4.02, 95% CI [-5.84, -2.20], p < 0.01), TC (SMD = -2.52, 95% CI [-3.56, -1.48], p < 0.01), liver TC (SMD = -5.28, 95% CI [-7.71, -2.84], p < 0.01), NAS score(SMD = -3.56, 95% CI [-5.04, -2.09], P < 0.01), FBG SMD = -1.89, 95% CI [-2.94, -0.83], p < 0.01), and body weight (SMD = -2.34, 95% CI [-3.94, -0.74], p < 0.01). Additionally, MSC-EVs improved the level of inflammatory cytokines (TNF-α and IL-6) and oxidative stress markers (SOD and MDA). These effects surpass those reported in previous MSC-EVs studies targeting liver disease, particularly regarding unassessed lipid parameters and oxidative stress indicators.

Conclusion

MSC-EVs show promising potential for treating NAFLD/NASH, with substantial evidence supporting their therapeutic and reparative effects. Our findings directly inform clinical trial design by identifying optimal parameters-such as human-derived EVs, treatment durations longer than four weeks, and exosome preparations obtained via differential ultracentrifugation-to maximize therapeutic efficacy. These findings warrant further clinical investigation to facilitate the clinical translation of MSC-EVs as a therapeutic option for NAFLD/NASH.

Supplementary Information

The online version contains supplementary material available at 10.1186/s12944-025-02635-1.

Introduction

Nonalcoholic fatty liver disease (NAFLD) and its progressive form, nonalcoholic steatohepatitis (NASH), represent a spectrum of liver disorders characterized by excessive fat accumulation in hepatocytes in the absence of significant alcohol consumption. These conditions have emerged as major global public health concerns [1]. NAFLD is currently the most prevalent chronic liver disease, with an estimated global prevalence of 30% [2], and its incidence continues to rise in parallel with the increasing prevalence of obesity and metabolic syndrome [3, 4]. As the more severe stage of NAFLD, NASH is characterized by hepatic steatosis accompanied by hepatocellular injury, inflammation, and fibrosis, which can ultimately progress to cirrhosis and even hepatocellular carcinoma, posing a significant threat to public health Given the limitations of current therapeutic strategies and the absence of definitive pharmacological interventions, there is an urgent need for innovative treatment approaches [5, 6].

Mesenchymal stem cells (MSCs), known for their immunomodulatory and regenerative properties, have shown therapeutic promise in diverse diseases [7, 8]. However, safety concerns, such as potential tumorigenicity and immune rejection associated with live-cell therapies, have limited their clinical translation [9]. This has shifted focus to MSC-EVs—nanoscale particles that recapitulate the therapeutic benefits of MSCs while circumventing risks related to cell transplantation.

MSC-EVs exhibit lower immunogenicity, reduced tumorigenic potential, and enhanced stability, making them attractive candidates for NAFLD/NASH therapy [10]. A specialized subclass of EVs, exosomes (EXOs), are nanovesicles (40–150 nm in diameter) released through the fusion of multivesicular bodies with the cell membrane. These vesicles carry a diverse array of bioactive molecules, including miRNAs, proteins, and lipids, enables modulation of inflammation, oxidative stress, and lipid metabolism—key pathways in NAFLD/NASH progression [10].

Despite preclinical enthusiasm, significant heterogeneity exists in reported outcomes of MSC-EV efficacy. Variability in EV isolation methods (e.g., differential centrifugation vs. size-exclusion chromatography), dosing regimens, administration routes, and model systems (e.g., different diet-induced) complicates the interpretation of results [11, 12]. Furthermore, critical questions remain unanswered: Which NAFLD/NASH biomarkers are most consistently improved by MSC-EVs? What factors contribute to interstudy variability in therapeutic outcomes? Addressing these gaps requires a systematic synthesis of preclinical evidence.

This systematic review and meta-analysis aim to evaluate the efficacy of MSC-EVs across key indicators in NAFLD/NASH rodent models, and provide evidence-based recommendations to optimize MSC-EVs therapy for clinical translation.

Materials and methods

The current systematic review and meta-analysis were designed and carried out following the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines. It was registered with PROSPERO (registration number CRD42024568746).

Search strategy

The PubMed, Web of Science, Embase, China National Knowledge Infrastructure (CNKI), Wanfang Data Knowledge Service Platform, and VIP Periodical Service Platform were searched from database inception until November 2024. The language was limited to English or Chinese. Medical subject headings (MeSH) and free words for database searches were as follows: [(“Extracellular Vesicles” OR “Exosomes” OR “Microvesicles” OR “Cell-Derived Microparticles”) AND (“Nonalcoholic Fatty Liver Disease” OR “NAFLD” OR “Nonalcoholic Steatohepatitis” OR “NASH”) AND (“Mesenchymal Stem Cells” OR “MSCs” OR “Mesenchymal Stem Cells”)].

Inclusion criteria

1) Model: studies used rodent models; 2) Intervention: extracellular vesicles with all dosages and durations; 3) Comparison: the control group was untreated-controlled or vehicle-controlled; 4) Outcomes: The outcome parameters included at least one of AST (aspartate transaminase), ALT (alanine aminotransferase), TG (triglyceride), liver TG, TC (total cholesterol), liver TC, BW (body weight), NAS score(NAFLD activity score), FBG (fasting blood glucose), TNF-α (Tumor Necrosis Factor-alpha), IL-6 (Interleukin-6; TNF-α), IL-1β (Interleukin-1 Beta), SOD (Superoxide Dismutase), and MDA (Malondialdehyde).

Exclusion criteria

1) Not published in English or Chinese; 2) in vitro studies or clinical trials; 3) non-extracellular vesicle interventions; 4) combined with other therapies; 5) reviews, crossover studies, and studies without a separate control group; 6) duplicate publications; and 7) data or full texts of the studies were not available.

Data extraction

The retrieved literature was imported into EndNote 21, and duplicate records were removed. Two independent researchers screened the titles and abstracts to exclude irrelevant studies. Subsequently, full-text articles were reviewed independently by the same researchers to identify and exclude those not meeting the inclusion criteria. In cases of disagreement, the reviewers first discussed their rationale for inclusion or exclusion. If consensus could not be reached, a third researcher, who was blinded to the initial decisions, independently reviewed the disputed articles and made a final judgment based on the predefined inclusion and exclusion criteria. All discrepancies and final decisions were documented in a screening log to ensure transparency and reproducibility. The following details were extracted from the selected studies: 1) the first author and publication year; 2) Animal characteristics (species, sex, sample size and weight); 3) Disease induction method (e.g., high-fat diet [HFD], methionine-choline-deficient [MCD] diet); 4) EV/EXO details (source [e.g., HUMSCs, HAMSCs], isolation method, characterization, particle size, dosage, frequency, administration route [intravenous, intraperitoneal], duration); And 5) outcome measures (AST, ALT, TG, TC, liver TG, liver TC, NAS score, FBG, BW, inflammatory cytokines, oxidative stress markers). Data were independently extracted by two researchers using a piloted spreadsheet to ensure consistency.

For studies requiring data extraction from graphical representations, GetData Graph Digitizer software (version 2.26) was used to obtain numerical values. In accordance with the Cochrane Handbook for Systematic Reviews of Interventions (CHSRI), data extraction prioritized results from the highest-dose cohort in multiarm dose–response studies and endpoint data for longitudinal outcome measures.

Quality assessment

The methodological quality of the included studies was independently assessed by two researchers via the Systematic Review Center for Laboratory Animal Experimentation (SYRCLE) risk-of-bias tool. This assessment consists of ten domains: (1) sequence generation, (2) baseline characteristics, (3) allocation concealment, (4) random housing, (5) blinding (for animal breeders and researchers), (6) random outcome assessment, (7) blinding (for outcome evaluators), (8) incomplete outcome data, (9) selective outcome reporting, and (10) other sources of bias. Each entry was identified as"low risk""high risk"or"unclear risk". Any disagreement that arose from the evaluation was settled by discussion it with a third researcher.

Data synthesis and analysis

STATA software (version 18.0) was used for the statistical analyses. Standardized mean difference (SMDs) with 95% confidence intervals (95% CI) were employed to assess continuous outcomes, with p < 0.05 considered statistically significant. We used random-effects models (DerSimonian-Laird method) to calculate the pooled estimates. To evaluate study heterogeneity, the I-squared (I²) statistic was used, with values < 25% indicating low heterogeneity, 25%–50% indicating moderate heterogeneity, and > 50% indicating high heterogeneity. To investigate potential sources of heterogeneity, subgroup analyses were conducted based on key variables, including animal model, source of EVs, duration, type of particles, and isolation method. These subgroup analyses were pre-specified, as each of these factors is known to influence therapeutic outcomes in preclinical studies. The selection of these variables was guided by prior literature, biological plausibility, and the heterogeneity observed across the included studies. Given the exploratory nature of these analyses, no formal adjustments for multiple comparisons were applied. Sensitivity analysis was performed to assess the robustness of the overall results. Additionally, if at least 10 studies reported on a specific outcome, Egger’s test was conducted to assess potential publication bias.

Results

A total of 135 studies were identified across six databases: PubMed (n = 32), Web of Science (n = 61), Embase (n = 31), CNKI (n = 4), Wanfang (n = 2), and VIP (n = 0). After the removal of duplicates, 85 studies remained. Following title and abstract screening, 24 studies were selected for full-text review. Ultimately, 14 studies met the inclusion criteria and were included in the systematic review and meta-analysis, while 10 studies were excluded. A detailed flowchart of the selection process is presented in Fig. 1.

Fig. 1 .

Flowchart for selection of studies

Study characteristics

This meta-analysis included 14 studies [13–26] with 12 published in English and 2 in Chinese, involving a total of 212 animals. Among these, 12 studies utilized C57BL/6 J mice, whereas 2 studies employed Sprague–Dawley (SD) rats, with the majority of the animals being male. With respect to the disease models, the analysis included 7 NASH models and 7 NAFLD models, as one study adopted both the models. High-fat diet (HFD) induction was the primary method used across all the models. Regarding the source of EVs or exosomes, 10 studies utilized human umbilical mesenchymal stem cells (HUMSCs), 1 study used human adipose-derived mesenchymal stem cells (HAMSCs), and 1 study used human placenta-derived mesenchymal stem cells (HEMSCs). The remaining two studies employed mesenchymal stem cells derived from the bone of Sprague–Dawley rats (RBMSCs) and mesenchymal stem cells derived from the adipose tissue of mice (MADSCs). All studies provided detailed descriptions of the extraction and characterization methods for EVs/EXOs. These methods include transmission electron microscopy (TEM) to examine exosome morphology, nanoparticle tracking analysis (NTA) to measure exosome size and concentration, and Western blotting to identify exosomal surface antigens. The diameters of the isolated EVs/EXOs ranged primarily from 40 to 136.8 nm. In all included studies, the intervention groups received the maximum reported dose of EVs/EXOs, whereas PBS served as the control treatment. In terms of isolation methods, 5 studies used reagents, 8 studies used gradient centrifugation, and only 1 study employed filtration. The treatment duration ranged from 3 −15 weeks. The primary outcome measures included: AST, ALT, TG, and TC, whereas secondary outcomes included fasting blood glucose, body weight, inflammatory factors (TNF-α, IL-6, and IL-1β), and oxidative stress indicators (SOD and MDA). The detailed characteristics of the included studies are listed in Table 1.

Table 1.

Basic characteristics of included studies

Study	Species (sex;n = Intervention/control group), weight	Model	Source of MSC	EV Size	EV characterisation	EV isolation method	Intervention	Duration	Control	Outcome
Cheng et al.,2021 [14]	SD rats (male, 6/6); 240–260 g	NAFLD (HFHF)	HUMSCs (exo)	96 nm	TEM, NTA, WB	reagent	100ug/rat/w (IV)	2 months	PBS	AST, ALT, TG, TC, FBG
Chen et al.,2024 [24]	C57BL/6 J mice (famale, 5/6) 18–20 g	NAFLD (HFD)	HUMSCs (exo)	None	TEM, NTA, WB	differential centrifugation	10 mg/Kg (IV)	4 weeks	PBS	ALT, AST, TG, TC, BW
Du et al.,2022 [21]	C57BL/6 J mice (male, 12/12) 20–23 g	NAFLD(HFD)	HUMSCs (exo)	40–120 nm	TEM, NTA, WB	reagent	120ug/mouse/qw (IV) 4.8 mg/Kg	15 weeks	PBS	ALT, AST, TG, TC,liver TG, liver TC, FBG, BW, NAS score, TNF-α, IL-6, SOD, MDA
El-Derany et al.,2021 [18]	SD rats (male 9/9) 120–150 g	NASH (HFD)	RBMSCs (exo)	None	FCM, TEM	differential centrifugation	120 μg/Kg/biw (IV)	6 weeks	PBS	AST, ALT, TG, TC, liver TG, liver TC
Kang et al.,2022 [23]	C57BL/6 J mice (male, 6/6); 18–22 g	NASH (HFHD, MCD)	HUMSCs (exo)	72.22 nm	TEM, NTA, WB	reagent	100 μg/mouse/q3d (IV)	6 weeks	PBS	ALT, TG, BW, NAS score, liver TG, TNF-α, IL-6
Kim et al.,2023 [25]	C57BL/6 J mice (male, 6/6); 22–25 g	NASH(MCH)	HUMSCs (ev)	131.1 nm	TEM, NTA, FCM	differential centrifugation	50ug/mouse/tiw (IP)	4 weeks	PBS	ALT, AST, liver TG
Li et al.,2023 [17]	C57BL/6 J mice (male, 10/10); none	NAFLD(HFD)	HUMSCs (exo)	100 nm	TEM, NTA, WB	differential centrifugation	20 mg/Kg/q2d (IV)	10 weeks	PBS	FBG
Liang et al.,2023 [26]	C57BL/6 J mice (male,8/820-25 g	NASH (HFMRCD)	HUMSCs (ev)	110 nm	TEM,NTA,WB	reagent	2 mg/Kg/q5d (IV)	1 month	PBS	BW
Niu et al.,2022 [20]	C57BL/6 J mice (famale,10/10) none	NAFLD (HFD)	MADSCs (ev)	95.8 ± 1.2 nm	TEM, NTA, WB	differential centrifugation	100ug/mouse/biw(IV)	7 weeks	PBS	AST, ALT
Shi et al.,2022 [13]	C57BL/6 J mice (male, 6/6) 22–25 g	NASH (MCD)	HUMSCs (exo)	120 nm	TEM, NTA, WB	reagent	20 mg/Kg/biw (IV)	6 weeks	PBS	AST, ALT, BW, NAS score, TNF-α, IL-1β, IL-6
Watanabe et al.,2020 [22]	C57BL/6 J mice (male, 8/8) none	NASH (LPS)	HAMSCs (ev)	100 nm	NTA	differential centrifugation	5ug/mouse (IV)	4 weeks	PBS	ALT, TG, TC, FBG, BW
Yang et al.,2021 [16]	C57BL/6 J mice (male. 3/6) none	NAFLD(HFD)	HUMSCs (exo)	110 nm	TEM, NTA, WB, FCM	differential centrifugation	20ug/Kg/biw (IV)	4 weeks	PBS	ALT, AST, TG, TC, liver TG, liver TC, BW
Yang et al.,2023 [19]	C57BL/6 J mice (male 6/6) 18–20 g	NAFLD (HFD)	HUMSCs (exo)	110 nm	TEM, NTA, WB	differential centrifugation	10 mg/Kg (IV)	4 weeks	PBS	ALT, AST, TG, TC, liver TG, liver TC, FBG, BW
Zhang et al.,2023 [11]	C57BL/6 J mice (famale 8/8) none	NASH(HFD)	HEMSCs (ev)	138.6 nm	NTA, WB	filtration	100ug/mouse/qod (IV)	3 weeks	PBS	ALT, liver TG, BW, NAS score, IL-1β, IL-6

Abbreviation: ALT Alanine aminotransferase, AST Aspartate transaminase, biw twice a week. BW Body weight, EV Extracellular vesicle, EXO Exosome, FBG Fasting blood glucose, FCM Flow cytometry, HAMSCs Human amniotic mesenchymal stem cells, HEMSCs Human embryonic derived mesenchymal stem cells, HFD High-fat diet, HFHD High-fat high-sugar diet feeding, HFMRCD High-fat methionine-choline deficient diet feeding, HUMSCs Human umbilical cord mesenchymal stem cells, IL-1β Interleukin-1 Beta, IL-6 Interleukin-6, IV Intravenous injection, IP Intraperitoneal injection, MADSCs Mouse adipose-derived stem cells, MCH Methionine-choline deficient, MDA Malondialdehyde, NAS score NAFLD activity score, NTA Nanoparticle tracking analysis, q3d every third day, q5d every five day, qod every other day, RBMSCs Rat bone marrow mesenchymal stem cells, SOD Superoxide Dismutase, TC Yotal cholesterol, TEM Transmission electron microscope, TG Triglyceride, TNF-α Tumor Necrosis Factor-alpha, WB Western blot

Risk of bias and quality of included studies

The risk of bias scores ranged from 3 to 4, with five studies scoring 3 points and nine studies scoring 4 points. None of the 14 studies provided details on allocation sequence generation, and two studies did not specify their randomization methods. Only one study reported baseline similarity between groups. Allocation concealment was unclear in all studies, and the adequacy of masking for group assignments was not specified. Eight studies implemented randomized housing to ensure consistent environments for the animals. However, none of the studies provided sufficient information on caregiver or investigator blinding, nor was it possible to confirm whether blinding or randomization occurred during outcome assessments. Primary outcome data were complete in all studies, with no evidence of selective reporting or other significant deviations. Further details are provided in Table 2.

Table 2.

The methodological quality of included studies

Study	A	B	C	D	E	F	G	H	I	J	Total
Cheng et al.,2021 [14]	?	?	?	?	?	?	?	+	+	+	3
Chen et al.,2024 [24]	?	?	?	?	?	?	?	+	+	+	3
Du et al.,2022 [21]	?	?	?	+	?	?	?	+	+	+	4
El-Derany et al.,2021 [18]	?	?	?	+	?	?	?	+	+	+	4
Kang et al. ,2022 [23]	?	?	?	?	?	?	?	+	+	+	3
Kim et al.,2023 [25]	-	?	?	+	?	?	?	+	+	+	4
Li et al.,2023 [17]	?	?	?	+	?	?	?	+	+	+	4
Liang et al.,2023 [26]	?	?	?	+	?	?	?	+	+	+	4
Niu et al.,2022 [20]	-	?	?	?	?	?	?	+	+	+	3
Shi et al.,2022 [13]	?	?	?	?	?	?	?	+	+	+	3
Watanabe et al.,2020 [22]	?	?	?	+	?	?	?	+	+	+	4
Yang et al.,2021 [16]	?	?	?	+	?	?	?	+	+	+	4
Yang et al.,2023 [19]	?	?	?	+	?	?	?	+	+	+	4
Zhang et al.,2023 [11]	?	+	?	?	?	?	?	+	+	+	4

Selection bias: A, Sequence generation; B, Baseline characteristics; C, Allocation concealment. Performance bias: D, Random housing; E, Blinding. Detection bias: F, Random outcome assessment; G, Blinding. Attrition bias: H, Incomplete outcome data. Reporting bias: I, Selective outcome reporting. Other: J, Other sources of bias.?, unclear; +, low risk; -, high risk

Effects on AST levels

Nine studies assessed the impact of EVs on AST levels. The pooled results indicated a significant reduction in AST levels in the treatment group compared with the control group (SMD = −2.79, 95% CI [−3.64, −1.94], p < 0.01). However, heterogeneity analysis revealed substantial variability among the studies (I² = 67.57%, p < 0.01) (Fig. 2).

Fig. 2.

Effects on AST levels

Effects on ALT levels

Eleven studies provided pairwise comparisons of ALT levels. The meta-analysis revealed a significant reduction in ALT levels in the treatment group (SMD = −2.47, 95% CI [−3.44, −1.50], p < 0.01). However, high of heterogeneity was detected (I² = 84.00%, p < 0.01) (Fig. 3).

Fig. 3.

Effects on ALT levels

Effects on TG levels

Eight studies examined the effect of EVs on TG levels, and revealed significant reduction in TG levels in the treatment group compared to the control group (SMD = −1.86, 95% CI [−2.98, −0.73], P < 0.01). Despite this, substantial heterogeneity was present (I² = 84.12%, p < 0.01) (Fig. 4).

Fig. 4 .

Effects on TG levels

Effects on liver TG levels

Eight studies assessed liver TG levels and found a significant reduction in the treatment group (SMD = −4.02, 95% CI [−5.84, −2.20], p < 0.01), accompanied by considerable heterogeneity (I² = 89.97%, p < 0.01) (Fig. 5).

Fig. 5 .

Effects on liver TG levels

Effects on TC levels

Seven studies reported the effect of EVs on TC levels, with a significant decrease in the treatment group (SMD = −2.52, 95% CI [−3.56, −1.48], p < 0.01). However, significant heterogeneity was observed (I² = 74.18%, p < 0.01) (Fig. 6).

Fig. 6.

 Effects on TC levels

Effects on liver TC levels

According to five studies, liver TC levels were significantly lower in the treatment group than in the control group (SMD = −5.28, 95% CI [−7.71, −2.84], p < 0.01), with substantial heterogeneity being observed (I² = 85.04%, p < 0.01) (Fig. 7).

Fig. 7 .

Effects on liver TC levels

Effects on NAS score

Four studies investigated the effect of EVs on the NAS score and found a significant reduction in the treatment group compared to the control group (SMD = −3.56, 95% CI [−5.04, −2.09], p < 0.01). Despite this, substantial heterogeneity was present (I² = 70.80%, p = 0.01) (Fig. 8).

Fig. 8 .

Effects on NAS score levels

Effects on body weight

Nine studies examined the effects of EVs on BW. The combined results demonstrated a significant decrease in BW in the treatment group compared with the control group (SMD = −2.34, 95% CI [−3.94, −0.74], p < 0.01). Heterogeneity analysis indicated significant variability among the studies (I² = 91.38%, p < 0.01) (Fig. 9).

Fig. 9 .

Effects on BW levels

Effects on FBG levels

Five studies assessed the effects of EVs on FBG levels. The pooled results indicated a significant reduction in FBG levels in the treatment group compared with the control group (SMD = −1.89, 95% CI [−2.94, −0.83], p < 0.01). The heterogeneity analysis revealed significant variability (I² = 76.27%, p < 0.01) (Fig. 10).

Fig. 10.

 Effects on FBG levels

Effects on inflammatory markers

Two studies evaluated the effect of EVs on IL-1β levels and revealed a significant reduction in the treatment group (SMD = −1.53, 95% CI [−2.42, −0.63], p < 0.01), with low heterogeneity (I² = 11.38%, p = 0.29) (Fig. 11).

Fig. 11 .

Effects on IL-1β levels

Four studies assessed IL-6 levels, and reported a notable decrease in the treatment group (SMD = −2.59, 95% CI [−3.99, −1.18], p < 0.01). However, substantial heterogeneity was present (I² = 79.95%, p < 0.01) (Fig. 12).

Fig. 12.

 Effects on IL-6 levels

Similarly, three studies analyzed TNF-α levels, and demonstrated a significant reduction in the treatment group (SMD = −2.76, 95% CI [−4.12, −1.32], p < 0.01), with high heterogeneity (I² = 81.87%, p < 0.01) (Fig. 13).

Fig. 13 .

Effects on TNF-α levels

Effects on oxidative stress markers

Two studies measured the levels of SOD and MDA. SOD activity was greater in indicated that the treatment group than in the control group (SMD = 1.63, 95% CI [0.89, 2.38], p < 0.01). Similarly, MDA levels were significantly lower in the treatment group than in the control group (SMD = −1.85, 95% CI [−2.55, −1.55], p < 0.01). Heterogeneity tests revealed low variability among these studies, but the results were not statistically significant (SOD: I² = 15.91%, p = 0.30; MDA: I² = 0%, p = 0.52) (Figs. 14 and 15). Subgroup analyses were not feasible for these two outcomes because of the limited number of studies.

Fig. 14 .

Effects on SOD levels

Fig. 15 .

Effects on MDA levels.

Subgroup analysis

To explore potential sources of heterogeneity, subgroup analyses were conducted base on the basis of the animal model, treatment duration, source of EVs/EXOs, and particle type (Table 3, Figs. 16, 17, 18, 19, 20, 21, 22, 23 and 24).

Table 3.

Subgroup analyses in AST, ALT, TG, TC, liver TG, liver TC, FBG, NAS score and BW with the extended parameters

Indicators	Comparison	Subgroup	Number of studies	SMD	P for meta-analysis	I2	P for heterogeneity
AST
	Animal model	NAFLD	6	−3.20 [−4.57, −1.83]	< 0.05	78.92%	< 0.01
		NASH	3	−2.30 [−3.06 −1.55]	< 0.05	0%	0.83
	Source	Animal	2	−5.33 [−11.54, 0.96]	0.10	94.16%	< 0.01
		Human	7	−2.38 [−2.90, −1.86]	< 0.05	0%	0.43
	Type of particles	EVs	2	−5.54 [−11.33, −0.26]	0.06	92.50%	< 0.01
		EXOs	7	−2.32 [−2.82, −1.82]	< 0.05	0%	0.45
	Duration	> 4 weeks	6	−3.54 [−5.04, −2.04]	< 0.05	80.01%	< 0.01
		≤ 4 weeks	3	−2.07 [−2.77, −1.36],	< 0.05	0%	0.65
	Isolation method	Centrifugation	6	−2.84 [−4.11, −1.58]	< 0.05	76.77%	< 0.01
		Reagent	3	−2.80 [−3.71, −1.89],	< 0.05	22.48%	0.28
ALT
	Animal model	NAFLD	6	−3.02 [−4.41, −1.62]	< 0.05	81.03%	< 0.01
		NASH	7	−2.03 [−3.42, 0.63]	< 0.05	86.70%	< 0.01
	Source	Animal	2	−4.32 [−8.05, −0.58]	< 0.05	89.28%	< 0.05
		Human	11	−2.20 [−3.16, −1.17]	< 0.05	82.62%	< 0.01
	Type of particles	EVs	4	−2.40 [−5.39, 0.59]	0.16	94.47%	< 0.01
		EXOs	9	−2.43 [−3.05, −1.80]	< 0.05	44.06%	0.74
	Duration	> 4 weeks	7	−3.38 [−4.35, −2.40]	< 0.05	63.96%	< 0.05
		≤ 4 weeks	6	−1.36 [−2.71, −0.01]	< 0.05	85.09%	< 0.01
	Isolation method	Centrifugation	6	−2.48 [−3.40, −1.56]	< 0.05	57.53%	< 0.01
		Reagent	5	−2.80 [−3.71, −1.89]	< 0.05	39.47%	< 0.05
		filtration	1	−1.47 [−0.41, 2.53]	< 0.05	NA	NA
Serum TG
	Animal model	NAFLD	5	−2.49 [−3.90, −1.06]	< 0.05	78.96%	< 0.01
		NASH	3	−0.91 [−2.46, 0.65]	0.25	84.40%	< 0.01
	Source	Animal	1	−2.66 [−3.90, −1.42]	< 0.05	NA	N
		Human	7	−1.74 [−2.99, −0.50]	< 0.05	84.95%	< 0.01
	Type of particles	EVs	2	−0.08 [−0.76, 0.61]	0.82	0%	0.74
		EXOs	6	−2.50 [−3.63, −1.36]	< 0.05	74.12%	< 0.01
	Duration	> 4 weeks	3	−2.83 [−3.85, −2.01]	< 0.05	0%	0.90
		≤ 4 weeks	5	−1.23 [−2.59, 0.13]	0.08	82.58%	< 0.01
	Isolation method	Centrifugation	6	−1.48 [−2.77, −0.20]	< 0.05	83.92%	0.77
		Reagent	2	−2.93 [−3.85, −2.01]	< 0.05	0%	< 0.01
Liver TG
	Animal model	NAFLD	4	−3.88 [−5.29, −2.47]	< 0.05	54.94%	0.08
		NASH	5	−4.23 [−7.28, −1.76]	< 0.05	93.43%	< 0.01
	Source	Animal	1	−6.05 [−8.21, −3.89]	< 0.05	NA	NA
		Human	8	−3.71 [−5.58, −1.85]	< 0.05	89.62%	< 0.01
	Type of particles	EVs	2	−7.68 [−24.56, 9.21]	0.37	95.95%	< 0.01
		EXOs	7	−3.77 [−4.78, −2.77]	< 0.05	55.61%	< 0.05
	Duration	> 4 weeks	4	−3.40 [−4.62, −2.18]	< 0.05	60.76%	0.54
		≤ 4 weeks	5	−5.10 [−8.81, −1.40]	< 0.05	93.38%	< 0.01
	Isolation method	Centrifugation	5	−6.08 [−8.51, −3.65]	< 0.05	76.47%	< 0.01
		Reagent	3	−2.81 [−3.58, −2.03]	< 0.05	89.97%	< 0.01
		filtration	1	0.61 [−0.34, 1.56]	0.21	NA	NA
Serum TC
	Animal model	NAFLD	5	−2.76 [−4.38, −1.13]	< 0.05	82.73%	< 0.01
		NASH	2	−2.21 [−3.03, −1.38]	< 0.05	0%	0.83
	Source	Animal	1	−2.29 [−3.45, −1.14]	< 0.05	NA	NA
		Human	6	−2.60 [−3.87, −1.33]	< 0.05	78.48%	< 0.01
	Type of particles	EVs	1	−2.12 [−3.30, −0.93]	< 0.05	NA	/
		EXOs	6	−2.63 [−3.89, −1.37]	< 0.05	78.41%	< 0.01
	Duration	> 4 weeks	3	−3.27 [−4.59, −1.95]	< 0.05	59.25%	0.09
		≤ 4 weeks	4	−1.93 [−3.30, −0.55]	< 0.05	74.84%	< 0.01
	Isolation method	Centrifugation	5	−1.98 [−3.04, −0.93]	< 0.05	68.42%	< 0.05
		Reagent	2	−3.93 [−5.65, −2.21]	< 0.05	48.01%	0.17
Liver TC
	Animal model	NAFLD	4	−4.21 [−6.32, −2.10]	< 0.05	77.75%	< 0.01
		NASH	1	−8.91 [−11.95, −5.87]	< 0.05	NA	NA
	Source	Animal	1	−8.91 [−11.95, −5.87]	< 0.05	NA	NA
		Human	4	−4.21 [−6.32, −2.10]	< 0.05	77.75%	< 0.01
	Duration	> 4 weeks	2	−5.45 [−11.92, −1.00]	0.98	93.84%	< 0.01
		≤ 4 weeks	3	−5.28 [−8.21, −2.35]	< 0.05	74.28%	< 0.05
	Isolation method	Centrifugation	4	−6.28 [−9.24, −3.32]	< 0.05	79.99%	< 0.01
		Reagent	1	−2.32 [−3.33, −1.31]	< 0.05	NA	NA
FBG
	Animal model	NAFLD	4	−2.19 [−3.43, −0.95]	< 0.05	76.50%	< 0.01
		NASH	1	−0.83 [−1.80, 0.14]	0.09	NA	NA
	Type of particles	EVs	1	−0.83 [−1.80, 0.14]	0.09	NA	NA
		EXOs	4	−2.19 [−3.43, −0.95]	< 0.05	76.50%	< 0.01
	Duration	> 4 weeks	3	−2.57 [−4.16, −0.98]	< 0.05	79.82%	< 0.01
		≤ 4 weeks	2	−0.97 [−1.71, −0.23]	< 0.05	0%	0.67
	Isolation method	Centrifugation	3	−2.05 [−3.98, −0.13]	< 0.05	86.69%	< 0.01
		Reagent	2	−1.81 [−2.77, −085]	< 0.05	36.15%	0.21
BW
	Animal model	NAFLD	5	−4.03 [−5.90 −2.17]	< 0.05	82.35%	< 0.01
		NASH	4	−3.02 [−2.56, 1.95]	0.79	92.55%	< 0.01
	Type of particles	EVs	3	−1.34 [−3.51, −0.83]	0.23	90.73%	< 0.01
		EXOs	6	−3.05 [−5.50, −0.60]	< 0.05	92.29%	< 0.01
	Duration	> 4 weeks	4	−1.41 [−3.92 −1.11]	0.27	93.27%	< 0.01
		≤ 4 weeks	5	−3.29 [−5.73, −0.86]	< 0.05	91.31%	< 0.01
	Isolation method	Centrifugation	5	−4.00 [−6.13, −1.87]	< 0.05	86.26%	< 0.01
		Reagent	3	−0.86 [−4.33, 2.61]	0.63	94.38%	< 0.01
		filtration	1	0.55 [−0.39, 1.50]	0.25	NA	NA
NAS score
	Animal model	NAFLD	1	−3.64 [−5.58 −1.69]	< 0.05	NA	NA
		NASH	4	−3.60 [−5.44, −1.76]	< 0.05	77.11%	< 0.01
	Type of particles	EVs	1	−1.61 [−2.69, −0.53]	< 0.05	NA	NA
		EXOs	4	−4.07 [−5.08, −3.05]	< 0.05	2.10%	0.38
	Duration	> 4 weeks	4	−4.07 [−5.08, −3.05]	< 0.05	2.10%	0.38
		≤ 4 weeks	1	−1.61 [−2.69, −0.53]	< 0.05	NA	NA
	Isolation method	Reagent	4	−1.61 [−2.69, −0.53]	< 0.05	NA	NA
		Filtraion	1	−4.07 [−5.08, −3.05]	< 0.05	2.10%	0.38

Abbreviations: 95% CI 95% confidence intervals, I² I-squared, NA Not available, NAFLD Nonalcoholic fatty liver disease, NASH Nonalcoholic steatohepatitis, EV Extracellular vesicle, EXO Exosome, SMD Standardized mean difference

Fig. 16.

Subgroup of AST

Fig. 17.

 Subgroup of ALT

Fig. 18 .

Subgroup of TG

Fig. 19 .

Subgroup of liver TG

Fig. 20 .

Subgroup of TC

Fig. 21 .

Subgroup of liver TC

Fig. 22 .

Subgroup of NAS score

Fig. 23 .

Subgroup of BW

Fig. 24 .

Subgroup of FBG

Animal model

The animal models were categorized into two subgroups: the NAFLD model and the NASH model. Subgroup analysis revealed that EVs derived from the NAFLD model exhibited greater therapeutic effects on AST (SMD −3.20 vs. SMD −2.30, p = 0.26), ALT (SMD −3.02 vs. SMD −2.03, p = 0.33), TG (SMD −2.49 vs. SMD −0.91, p = 0.14),TC (SMD −2.76 vs. SMD −2.21, p = 0.55), FBG (SMD −2.19 vs. SMD −0.83, p = 0.09), and BW (SMD −4.03 vs. SMD −3.02, p = 0.01), compared with EVs derived from the NASH model. Among these indicators, only the difference in BW reached statistical significance (p = 0.01), while the others did not. In contrast, EVs derived from the NASH model showed slightly better therapeutic effects on liver TG (SMD −4.23 vs. SMD −3.88, p = 0.84) and liver TC (SMD −8.91 vs. SMD −4.21, p = 0.13). Regarding the NAS score, both subgroups demonstrated comparable improvements (SMD −3.64 vs. SMD −3.60, p = 0.98).

Within the NASH model, AST (I² = 0%, p = 0.84) and TC (I² = 0%, p = 0.83) showed low heterogeneity, suggesting that differences in the animal models might have contributed to the observed heterogeneity.

Duration

Studies were stratified into two subgroups on the basis of the treatment duration: > 4 weeks and ≤ 4 weeks. The > 4 weeks subgroup presented significantly greater reductions in AST (SMD −3.54 vs. SMD −2.07, p = 0.08), ALT (SMD −3.38 vs. SMD −1.36, p = 0.018), TG (SMD −2.83 vs. SMD −1.23, p = 0.043), TC (SMD −3.27 vs. SMD −1.93, p = 0.17), liver TC (SMD 5.45 vs. SMD −5.28, p = 0.96), NAS score (SMD −4.07 vs. SMD −1.61, p = 0.001), and FBG (SMD −2.57 vs. SMD −0.97, p = 0.074) levels.

The ≤ 4 weeks subgroup presented a greater reduction in BW (SMD −3.29 vs. SMD −1.41, p = 0.29). No statistically significant differences were observed in AST, liver TC, TC, or BW levels between the two subgroups, except for ALT, TG, and NAS score, which showed significant differences.

Notably, heterogeneity decreased markedly in the > 4 weeks subgroup for TG (I² = 0%, p = 0.09), and NAS score (I² = 2.10%, p = 0.38). In contrast, heterogeneity was low in the ≤ 4 weeks subgroup for AST (I² = 0%, p = 0.65) and FBG (I² = 0%, p = 0.67), suggesting that treatment duration may be a key source of heterogeneity.

Sources of EVs

The sources of EVs were categorized into two subgroups: human-derived and animal-derived. Subgroup analysis indicated that, compared with human-derived EVs, animal-derived EVs had superior therapeutic effects ALT (SMD −4.32 vs. SMD −2.20, p = 0.28), AST (SMD −5.33 vs. SMD −2.38, p = 0.36), TG (SMD −2.66 vs. SMD −1.74, p = 0.31), liver TG (SMD −6.05vs. SMD −3.71, p = 0.11), and liver TC (SMD −8.91 vs. SMD −4.21, p = 0.013) levels. However, for TC (SMD −2.29 vs. SMD −2.20, p = 0.72) levels, human-derived EVs exhibited marginally better therapeutic effects.

Despite these differences, inter-group comparisons revealed no statistically significant differences between the subgroups, except for liver TC levels. Due to insufficient data in the subgroups for FBG, BW, and NAS score, analyses of these indicators were not performed. Subgroup analysis revealed significantly reduced heterogeneity (I² = 0%, p = 0.83) in AST outcomes within the human-derived subgroup, suggesting that the origin of EVs/EXOs may play a role in modulating heterogeneity.

Type of particles

Studies have been categorized into two subgroups on the basis of particle size: EVs and EXOs. Subgroup analysis revealed that the EXOs subgroup demonstrated greater efficacy in reducing ALT (SMD −2.43 vs. SMD −2.40, p = 0.99), TG (SMD −2.50 vs. SMD −0.08, p < 0.001), TC (SMD −2.63 vs. SMD −2.12, p = 0.56), FBG (SMD −2.19 vs. SMD −0.83, p = 0.09), NAS score (SMD −4.07 vs. SMD −1.61, p < 0.01) and BW (SMD −3.05 vs. SMD −1.34, p = 0.31) levels compared to the EVs subgroup. Conversely, the EVs subgroup showed greater reductions in AST (SMD −5.54 vs. SMD −2.32, p = 0.28) and liver TG (SMD −7.68 vs. SMD −3.77, p = 0.65) levels. Despite these trends, statistically significant differences were only observed for NAS score and TG.

Importantly, the EXOs subgroup exhibited a significant reduction in heterogeneity for AST outcomes (I² = 0%, p = 0.45) and a trend toward decreased heterogeneity for ALT (I² = 44.06%, p = 0.74). In the EVs subgroup, heterogeneity was notably reduced for NAS score (I² = 2.10%, p = 0.38) and TG (I² = 0%, p = 0.74), suggesting that particle type may influence heterogeneity outcomes.

Isolation method

Studies were divided into three subgroups based on EVs/EXOs extraction methods: centrifugation, reagent, and filtration. Subgroup analysis revealed that the centrifugation subgroup demonstrated greater efficacy in reducing AST (SMD −2.84 vs. SMD −2.80, p = 0.95), liver TG (SMD −6.08 vs. SMD −2.81 vs. SMD 0.61, p < 0.001), liver TC (SMD −6.28 vs. SMD −2.32, p = 0.013), FBG (SMD −2.05 vs. SMD −1.81, p = 0.82) and BW (SMD −4.00 vs. SMD −0.86 vs. SMD 0.55, p = 0.001) compared to the other subgroups.

The reagent-based subgroup showed superior efficacy in reducing ALT (SMD −2.80 vs. SMD −2.48 vs. SMD −1.47, p < 0.001), TC (SMD −3.93 vs. SMD −1.98, p = 0.06) and TG (SMD −2.93 vs. SMD −1.48, p = 0.07) Conversely, the filtration subgroup showed the greatest reduction in NAS score (SMD −4.07 vs. SMD −1.61, p = 0.001). However, statistically significant differences were only observed for liver TG, liver TC, BW, ALT, and NAS score.

In the reagent subgroup, reduced heterogeneity was observed for AST (I² = 22.48%, p = 0.28), ALT (I² = 39.47%, p < 0.05), TG (I² = 0%, p < 0.01), TG (I² = 48.01%, p = 0.17) and FBG (I² = 36.15%, p = 0.21), suggesting that the method of EVs/EXOs isolation may be an important source of heterogeneity.

Sensitivity analysis

A comprehensive sensitivity analysis was conducted via leave-one-out method to evaluate the robustness of the effect sizes for AST, ALT, TC, liver TC, TG, liver TG, NAS score, FBG, BW, TNF-α, IL-6, IL-1β, SOD, and MDA. The results indicated that the effect sizes remained consistent, with no single study significantly altering the overall findings, confirming the reliability and stability of the results across the dataset.

Significant heterogeneity was observed in AST outcomes. The study conducted by Niu et al. presented results that deviated markedly from those of other studies [20], which may account for the observed heterogeneity. After excluding this study, EV treatment still significantly reduced AST levels (SMD = −2.36, 95% CI [−2.83, −1.88], p < 0.01), and no residual heterogeneity was detected (I² = 0, p = 0.54) (Supplementary Figs. 1 and 2).

Publication bias

For ALT, where at least 10 studies were available, potential publication bias was assessed. Visual inspection of the funnel plot suggested substantial publication bias, which was further supported by Egger’s test (p < 0.001) (Fig. 25). To evaluate the robustness of the results, we applied the trim-and-fill method to estimate the potential impact of unpublished studies. The adjusted pooled estimate based on a random-effects model remained comparable to the original results, indicating that the presence of publication bias did not substantially affect the overall findings (Fig. 26).

Fig. 25 .

Publication bias plot

Fig. 26 .

Trim-and-fill method of ALT

Discussion

Summary of the meta-analysis

This systematic review and meta-analysis comprehensively assessed the therapeutic potential of MSCs-EVs in preclinical rodent models of NAFLD and NASH. A total of fourteen studies involving 212 animals were included in the analysis. The pooled results revealed that MSCs-EVs significantly improved NAFLD/NASH-related biomarkers, demonstrated by reductions in AST, ALT, TC, liver TC, TG, liver TG, NAS score, BW, FBG, TNF-α, IL-1β, and IL-6 levels. Additionally, MSCs-EVs exerted beneficial effects on the levels of oxidative stress markers, including SOD and MDA.

Despite the promising results, substantial heterogeneity was observed across all outcomes. To investigate its potential sources, subgroup analyses were conducted based on animal model, source of EVs, duration, type of particles, and isolation method. The results indicated that these factors contributed substantially to the variability in effect sizes, identifying them as key sources of heterogeneity. Sensitivity analyses confirmed the robustness of the pooled estimates. Notably, publication bias was detected for ALT levels; however, trim-and-fill analysis showed that the overall findings remained consistent, suggesting that the publication bias did not materially impact the conclusions.

Findings of the subgroups

Animal model

This meta-analysis revealed that MSC-EVs demonstrate superior therapeutic efficacy in NAFLD models compared with NASH models. Specifically, MSCs-EVs led to more pronounced improvements in parameters such as ALT, AST, TG, TC, FBG, and BW, although statistical significance was not achieved for some indices. Mechanistically, NAFLD is characterized by hepatocellular lipid accumulation without significant inflammation or fibrosis, presents a microenvironment where EVs can preferentially modulate lipid metabolism via miRNA-mediated pathways (e.g., miR-24-3p targeting Keap1 [27].In contrast, NASH is characterized by a more complex histopathological profile, including steatosis, inflammatory cell infiltration, hepatocyte ballooning, and collagen deposition. These may impair EVs bioavailability or overwhelm their reparative capacity [28].More evidence also suggests that MSCs-EVs have differential dose-dependent effects. Low-dose MSC-EVs may provide optimal efficacy in early-stage steatosis (NAFLD), whereas higher doses are more effective in addressing late-stage fibrotic conditions (NASH), as shown in studies on liver fibrosis [29]. Notably, treatment with EVs induced only marginal weight fluctuations in NASH models, in stark contrast to the pronounced weight reduction observed in NAFLD models. This discrepancy may be attributed to systemic metabolic disturbances in NASH models (commonly induced by methionine-choline-deficient diets), which lead to caloric restriction, hypoglycemia, rapid weight loss, and hepatic atrophy [30]. We hypothesize that, compared with NAFLD models, EVs help alleviate energy metabolism dysregulation in NASH by enhancing hepatic parenchymal repair and regeneration, thereby moderating weight loss.

Duration

The analysis revealed that MSCs-EVs administered for more than 4 weeks demonstrated greater efficacy in reducing serum hepatic enzymes (ALT, AST) and lipid parameters (TG, TC) than shorter treatment durations (≤ 4 weeks). These findings suggest that extending the duration of EV therapy may optimize therapeutic outcomes in both NAFLD and NASH. The observed duration-dependent effect could be due to the extended treatment period, which allows more comprehensive regulation of lipid metabolism and sustained attenuation of inflammatory pathways.

Source of EVs

This meta-analysis revealed that, compared with human-derived EVs, animal-derived EVs or exosomes exhibit superior efficacy in reducing hepatic enzymes (ALT and AST) and improving TG levels in NAFLD/NASH models. This disparity may be explained by three principal factors. First, interspecies immunological compatibility plays a critical role. Murine-derived EVs tend to exhibit lower immunogenicity when administered in murine models, thereby reducing the risk of immune rejection and enhancing therapeutic efficacy. In contrast, human-derived EVs may be recognized as xenogeneic, potentially triggering immune responses that diminish their therapeutic performance [27].Second, murine EVs carry species-specific bioactive molecules—such as growth factors, cytokines, and mRNAs-that are more effectively utilized within the same species to promote hepatic repair and regulate metabolic pathways. When administered across species, however, these bioactive cargos may fail to exert their full therapeutic potential due to immune clearance mechanisms, including activation of the mononuclear phagocyte system and MHC mismatch–induced adaptive immune responses [31, 32]. These findings highlight the importance of considering both immunogenicity and species compatibility when translating EV-based therapies into clinical use. Third, murine EVs demonstrate greater hepatic accumulation in murine models, likely facilitated by species-specific receptor–ligand interactions with hepatocytes. Conversely, human-derived EVs show reduced hepatic uptake in mice, presumably due to interspecies differences in receptor expression profiles [33]. These observations further underscore the therapeutic relevance of species matching in EV delivery and suggest that species-matched EVs may enhance biodistribution and functional outcomes [34]. Nonetheless, rigorous clinical investigations are still required to confirm the efficacy and safety of this approach in humans with NAFLD/NASH.

Type of particles

Exosomes (EXOs,30–150 nm), a specific subset of EVs, exhibit a more uniform size distribution and greater enrichment of bioactive molecules (e.g., miRNAs, proteins) than do heterogeneous EVs (100–1000 nm), which encompass a range of vesicle subtypes with diverse cargos. EXOs have shown superior therapeutic efficacy in NAFLD, which is attributed to their enrichment in lipid metabolism-regulating miRNAs (e.g., miR-122 and miR-34a) and surface markers (such as CD63 and CD81), which facilitate hepatocyte-specific uptake [35, 36]. In contrast, larger EVs have reduced targeting efficiency due to size-dependent limitations in biodistribution.

Isolation method

Subgroup analysis based on isolation methods revealed that EVs obtained via centrifugation demonstrated superior therapeutic efficacy in reducing AST, liver TG, liver TC, FBG, and BW levels compared with those isolated by other methods. EVs from the reagent-based group exhibited greater effects on ALT, TG, and TC levels, while the filtration group showed the most pronounced effect on NAS score. However, only liver TG, liver TC, BW, ALT, and NAS score reached statistical significance. Overall, centrifugation appeared to be the most effective method for isolating therapeutically potent EVs, likely due to its ability to yield a higher quantity and better-preserved vesicles. Nevertheless, its primary limitation lies in the time-consuming and labor-intensive procedure. In contrast, reagent- and filtration-based methods are more convenient and time-efficient but often result in lower EV yields.

Mechanistically, isolation methods can influence the composition and functionality of EVs. Centrifugation may preserve a broader spectrum of vesicles, including larger EVs enriched in pro-regenerative growth factors. Filtration techniques tend to favor smaller exosomes, which may be rich in anti-inflammatory microRNAs. However, reagent-based kits may compromise EV integrity due to the use of chemical additives, potentially altering surface receptor expression and biodistribution. These methodological differences highlight the critical need for standardizing EV isolation protocols in accordance with therapeutic objectives.

Future studies should emphasize transparent reporting of isolation parameters—such as centrifugal speed, duration, and specific reagent brands—to enhance reproducibility and cross-study comparability. Additionally, the adoption of emerging isolation techniques, including size-exclusion chromatography and microfluidic-based sorting, holds promise for improving EV purity and functional consistency. Aligning isolation strategies with disease-specific therapeutic mechanisms (e.g., exosome-mediated miRNA delivery in NASH-associated fibrosis) may further enhance the clinical translation of MSC-EV therapies.

Mechanisms of MSCs-EVs in ameliorating NAFLD/NASH

Bioactive molecule delivery

MSCs-EVs are rich in bioactive molecules, including proteins, lipids, and nucleic acids, which play crucial roles in intercellular communication and contribute to their therapeutic effects [37]. These bioactive molecules can modulate various biological processes, such as inflammation, cell proliferation, and tissue repair, enhancing the overall efficacy of MSCs-EVs in treating liver diseases such as NAFLD and NASH.

Lipid metabolism regulation

EVs regulate lipid metabolism through the modulation of various metabolic pathways. These mechanisms are closely aligned with our meta-analysis findings, which demonstrated significant reductions in TG, TC, ALT, AST, and FBG levels following EV treatment. Yang et al. demonstrated that EVs deliver CAMKK1, which activates the AMPK-mediated PPAR-α/CPT-1A and SREBP-1c/FAS signaling pathways, promoting fatty acid oxidation while suppressing lipogenesis [19]. Kang et al. reported that MSC-EVs downregulate hepatic mRNA expression of key lipogenic markers (SREBP-1c, FAS), oxidative markers (PPAR-α, CPT1-α, ACOX), and lipid transport markers (FABp5) [23]. Kim et al. reported that AMPK phosphorylation enhances metabolic homeostasis by upregulating the expression of mitochondrial oxidative genes (PGC1-α, MCAD, ACSL1, CPT1-α and UCP-2) and downregulating the expression of genes (ACC-1, FABP-1, SREBP-1 and FATP5), thereby alleviating hepatic steatosis [25].

On a deeper mechanistic level, EVs modulate lipid metabolism by delivering specific miRNAs that target key regulatory genes. One such miRNA, miR-627-5p, is highly enriched in MSC-EVs [38]. Fat mass and obesity-associated gene (FTO), which plays a pivotal role in obesity and NAFLD, exacerbates hepatic insulin resistance, oxidative stress, and lipid accumulation [39, 40]. Cheng et al. demonstrated that MSC-EVs suppress FTO expression through miR-627-5p delivery, thereby improving glucolipid metabolism [41]. El-Derany et al. reported that MSC-EXOs upregulate miR-627-5p to reverse SREBP1 and ACC protein levels in NASH mice, inhibiting lipogenesis and activating fatty acid oxidation [18]. Similarly, Niu et al. identified miR-223-3p in MSC-EVs as a negative regulator of E2F1 through 3'UTR binding. Suppression of E2F1 reduces lipid accumulation, which aligns with findings of Denechaud that E2F1 depletion inhibits glycolysis and lipogenesis in hepatocytes [42]. E2F1 is a validated miR-223 target and has been identified as a potential therapeutic target for inflammatory diseases [43]. Furthermore, Du et al. demonstrated that exosomal miR-24-3p improves NAFLD by targeting Keap1 mRNA in hepatocytes [21]. Liang et al. revealed that EVs enhance autophagy via the AMPK/mTOR pathway by activating ULK1 complex dephosphorylation, which promotes lipid phagocytosis and degradation [26]. Collectively, these studies underscore the diverse and multifaceted role of EVs in modulating lipid metabolism, providing diverse therapeutic avenues for treating liver diseases such as NAFLD and NASH.

As biomarkers of hepatocellular injury, the reductions in ALT and AST levels indicate that MSC-derived EVs may protect liver function by alleviating hepatic lipid accumulation and inflammatory responses. Moreover, the decrease in FBG levels suggests an improvement in insulin sensitivity and glucose metabolism mediated by MSC-EVs [41].

Anti-inflammatory and antioxidative effects

Dysregulated lipid metabolism triggers macrophage foam cell formation and the excessive release of inflammatory cytokines, which play pivotal roles in sustaining NASH-related inflammation [44, 45].Macrophage polarization toward the M1 phenotype promotes proinflammatory cytokine release, accelerating NASH progression, whereas polarization toward the M2 phenotype enhances anti-inflammatory factor secretion, mitigating disease development [46]. Kang et al. demonstrated that exosome therapy significantly reduced M1 macrophage populations in the livers of NASH mice while increasing M2 macrophage proportions, suggesting that MSC-EXOs may alleviate NASH by modulating macrophage polarization, thereby reducing pro-inflammatory cytokine levels such as TNF-α, IL-1β, and IL-6. Similarly, Nie et al. reported that exosome intervention significantly reduced NF-κB (p65) phosphorylation in the livers of MCD diet-induced NASH mice, effectively suppressing inflammation. This intervention also promoted M2 macrophage polarization and enhanced the production of anti-inflammatory factors, collectively mitigating NASH-related liver damage and improving disease outcomes in the MCD-induced model [47].

ROS-mediated hepatic inflammation and lipid accumulation are known contributors to NAFLD progression [48]. Nrf2, a key transcription factor in the antioxidant response, protects cells from ROS-induced oxidative damage [49]. Keap-1 promotes the continuous ubiquitination and degradation of Nrf2 in the cytoplasm. Upon dissociation from Keap-1, Nrf2 translocates to the nucleus, where it activates the expression of antioxidant genes such as HO-1, SOD, and GSH, thereby counteracting oxidative stress and maintaining redox homeostasis [50]. Du et al. revealed that MSC-EXOs transfer miR-24-3p to hepatocytes, where it targets Keap-1 to activate Nrf2 while concurrently inhibiting the NF-κB signaling pathway, leading to reduced ROS production and alleviation of inflammation, thereby decreasing pro-inflammatory cytokines such as TNF-α, IL-1β, and IL-6 [21].

Kang et al. further discovered that in HepG2 cells treated with ML385, a specific Nrf2 inhibitor, MSC-EXOs upregulated Nrf2/NQO-1 protein level, thereby activating downstream antioxidant enzymes (SOD, GSH, and NQO-1). Moreover, MSC-EXOs downregulated CYP2E1 expression in HepG2 cells. Since CYP2E1 has been shown to mediate ROS production and exacerbate oxidative stress, these findings suggest that MSC-EXOs may help mitigate oxidative damage by reducing ROS generation [23].

In addition, Watanabe et al. reported that EVs downregulate fibrosis-associated genes (Mmp2, Mmp12, and Mmp13) while increasing anti-inflammatory macrophage populations in the liver, ultimately mitigating hepatic fibrosis [22]. Similarly, Niu et al. observed that MSC-EVs suppress the expression of key fibrotic biomarkers, including α-SMA, COL1A1, and TGF-β1, whose downregulation is indicative of fibrosis regression [20].

Inflammation and oxidative stress are central to the progression of NAFLD and NASH [51]. MSCs-EVs can downregulate inflammatory cytokines and ameliorate inflammation and oxidative stress, thereby alleviating liver inflammation and fibrosis, restoring hepatocellular function, and ultimately reducing ALT and AST levels.

Mitochondrial and endoplasmic reticulum function repair

Mitochondrial autophagy dysfunction is a key player in the development and progression of NASH [52]. Malva reported that MSC-EXOs upregulated the expression of mitochondrial autophagy-related genes (Parkin, PINK1, ULK1, BNIP3L) and Bcl-2 in HFD-induced NASH mice, while inhibiting the expression of caspase-2, which has been shown to promote mitochondrial autophagy [53]. Therefore, MSCs-EXOs can improve the progression of apoptosis in NAFLD/NASH by modulating these molecules [18].

Nie et al. found that exosomes increased the transcriptional levels of mitochondrial markers (COXIV and VDAC1) and essential cofactors for mitochondrial biogenesis (PGC1α), thereby promoting mitochondrial fission and biogenesis [47].

Endoplasmic reticulum (ER) dysfunction is also a key factor in the pathogenesis of NASH, as prolonged ER stress can lead to cell apoptosis [54]. Kim et al. discovered that MSC-EVs significantly reduced the mRNA expression of ER stress-related genes (Xbp1, Grp78, and Chop) in primary hepatocytes treated with thapsigargin (an ER stress inducer), alleviating ER dysfunction and the resulting hepatocyte apoptosis in NASH [25].

The restoration of mitochondrial and endoplasmic reticulum function contributes to the recovery of normal hepatocyte metabolic activity, thereby reducing lipid accumulation in the liver. This leads to decreased levels of TG and TC, alleviates hepatic inflammation, and further lowers ALT, AST, and pro-inflammatory cytokine levels.

Liver regeneration

MSC-EVs stimulate hepatocyte proliferation and promote liver tissue repair. Song et al. demonstrated that miRNA-124, derived from human umbilical cord MSC-EVs, enhances liver regeneration in partial hepatectomy rats by downregulating Foxg1 [55]. Additionally, MSC-EVs induce angiogenesis, improving hepatic perfusion and oxygenation, which supports tissue recovery.

Emerging evidence also underscores the critical role of the gut microbiota in the pathogenesis of NAFLD/NASH.

MSC-EVs alleviate NASH by modulating the gut-liver axis, improving gut microbial composition and metabolite profiles, which in turn reduces hepatic inflammation and metabolic dysregulation. This leads to decreased levels of inflammatory cytokines and improved lipid and glucose metabolism, ultimately influencing biomarkers such as AST and ALT.

Bo et al. revealed that MSC-EXOs partially restore dysfunctional gut microbiota and enhance microbial metabolic pathways in rat hepatic injury models, promoting hepatic tissue regeneration [56].

Limitations and future directions

The results of this study should be interpreted in light of the following potential confounding factors. Variations in the design of animal experiments may affect the robustness of the meta-analysis outcomes. Although we have attempted to collect information on animal strains, sex, and weight, most of the included studies did not report detailed animal model characteristics such as diet and age, which limited our ability to fully control for these confounders. Future primary research should aim to comprehensively report these fundamental variables and strive for standardization in experimental design to enhance the reliability of systematic reviews. Nevertheless, through sensitivity analyses and the application of random-effects models, we have minimized the impact of heterogeneity, ensuring that the results remain robust.

Despite the promising results observed, this study has several limitations that should be addressed. First, the significant heterogeneity among the included studies highlights the need for further research to identify potential sources of variation and confirm the consistency of the therapeutic effects. Additionally, since the studies included in this meta-analysis were preclinical, the generalizability of the findings to clinical applications remains limited. Therefore, future clinical trials are essential to evaluate the safety and efficacy of MSC-EVs in human patients with NAFLD/NASH. Finally, considering the potential for long-term administration, additional studies are needed to assess the safety and durability of the therapeutic effects of MSC-EVs in human subjects.

Conclusion

This meta-analysis revealed that MSC-EVs are an effective biotherapy for combating NAFLD/NASH, primarily by modulating lipid metabolism and improving inflammation and oxidative stress, thereby reversing the progression of NAFLD/NASH. However, the overall methodological quality of these studies is relatively low, and further large-scale, long-term, high-quality animal model studies and human clinical trials are needed for further validation.

Abbreviations

NAFLD

Nonalcoholic fatty liver disease

MCH

Methionine-choline deficient

HFD

High-fat diet

EXO

Exosome

EV

Extracellular vesicle

HUMSCs

Human umbilical cord mesenchymal stem cells

HEMSCs

Human embryonic derived mesenchymal stem cells

RBMSCs

Rat bone marrow mesenchymal stem cells

HAMSCs

Human amniotic mesenchymal stem cells

MADSCs

Mouse adipose-derived stem cells

AST

Aspartate transaminase

ALT

Alanine aminotransferase

TG

Triglyceride

TC

Total cholesterol

FBG

Fasting blood glucose

BW

Body weight

TNF-α

Factor-alpha

IL-6

Interleukin-6

IL- IL-1β

Interleukin-1β

SOD

Superoxide dismutase

MDA

Malondialdehyde

CAMKK1

Calmodulin-dependent protein kinase kinase1

Authors’ contributions

Qiangqiang Dai :D, Di Zhu:Z; Xiaoming Du:D, Hao Tan:T, Qiu Chen:C. D. and C.conceived, initiated, and supervised the project. D.,Z.,D., and T. collected and analyzed the data. D.and Z. wrote a draft of the manuscript. All authors critically reviewed and revised the manuscript, and agreed with the published version of the manuscript.

Funding

This work was supported by grants from the Science and Technology Research Special project of Science and Technology Department of Chengdu (2023-yf09-00052-sn) and the Noncommunicable Chronic Diseases-National Science and Technology Major Project (No. 2023ZD0509400).

.

Data availability

No datasets were generated or analysed during the current study.

Declarations

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.