The researchers conducted a systematic literature search using 3 databases, including MEDLINE, EMBASE, and Web of Science, to screen for targeted studies on the efficacy of BMSCs in treating sepsis. The detailed search strategy is shown in Additional file 1: Table S1 . The last search was updated on August 31, 2021. The publication language was limited to English. We also searched the reference lists of eligible studies.

According to the PICOS scheme (population, intervention, control, outcome, and study design), the studies’ eligibility criteria were formulated (40). This meta-analysis included studies that met the following criteria: (i) They tested the efficacy of MSC(M) treatment in animal models of sepsis (all types of animals and both genders). (ii) The research centered on animal models of sepsis or endotoxemia. (iii) The study included mortality as part of its evaluation index. When two or more articles contained overlapping data, the most recent or informative article was used. (iv) Present experimental studies in original research papers. (v) The study was published in English.

The exclusion criteria were as follows: (i) Researches that only evaluated the efficacy of transfected or modified cell transplantation. (ii) Studies that only tested stem cells other than MSC(M). (iii) MSC(M) administered before sepsis model.

All published articles were independently reviewed by two investigators after the removal of duplicates. After the two investigators reached an agreement, irrelevant studies were excluded. For a comprehensive review, all relevant articles were retrieved, and two researchers independently evaluated each article using the above selection criteria. Any disagreements or uncertainties were resolved by consensus, and if necessary by a third investigator.

The following information were abstracted by two investigators independently and entered electronically: author, publishing year, study country, animal characteristics (species, gender, sample size, and model), intervention characteristics (origin, dose, route, and timing of the MSC(M) treatment), follow-up (the longest observation time of outcomes after MSC(M) administration), and our primary measures related to secondary outcomes. If only graphs were available, values were calculated from images using GetData Graph Digitizer software. Analyzing the data involved averaging the two researchers’ readings. In the absence of the standard deviation, the standard error was converted into a standard deviation by multiplying it by the square root of the group size. When multiple experimental groups that were distinguished by a variety of factors such as cell dose or delivery route and timing were contrasted with the control group, these groups were considered separate and independent studies. When results were evaluated over a range of follow-up periods, only the longest interval was selected.

Study authors assessed independently the risk of bias for each study included using the method developed by the Systematic Review Center for Laboratory Animal Experimentatio(SYRCLE) (41). This tool evaluates selection bias, performance bias, detection bias, attribution bias, and reporting bias and grades them according to “Yes, No, or Unclear.” Arguments were resolved through discussions with additional authors.

The primary outcome of this meta-analysis was mortality. OR was calculated between the MSC(M) treated group and the control group to determine the combined effect size. All statistical analyses and graphs were performed by R (version 4.0.2), using a random-effects model and the Hedges calculation (42). Typically, an effect size of 0.2 represents a small effect, 0.5 represents a medium effect, and 0.8 represents a large effect (43). A P-value <0.05 was considered statistically significant. The I ² statistic was used to analyze heterogeneity, and it was divided into three categories, low (25–50%), moderate (50–75%), and high (>75%) (44).

We used six clinical characteristics to group the effect size of outcome: Gender (male, female); sepsis model (CLP, non-CLP Roteneone); MSC(M) species (Allogeneic, Syngeneic, or Xenogeneic); MSC(M) dose (<1×10⁶, ≥1×10⁶ and <1×10⁷, ≥1×10⁷); follow-up duration (≤6hours, >6 hours); delivery route (intravenous or intraperitoneal). In order to examine the possible associations between the outcomes and the above clinical characteristics, subgroup analyses and meta-regression analyses were performed (45). Sensitivity analyses were performed by omitting one study at a time to evaluate whether the results were affected by a single study. Publication bias was evaluated using funnel plots (46), and the symmetry of funnel plots was performed with Egger regression (47). If necessary, any non-negligible bias would be corrected using the trim-and-fill approach (48).

A total of 1551 records were identified following the search procedure shown in Figure 1 , including 184 articles in PubMed, 538 articles in EMBASE, and 784 articles in Web of Science. After excluding duplicates, 432 studies were screened by title and abstract and 364 studies were subsequently excluded. Full-text screening for eligibility was performed on 68 studies and, based on the inclusion and exclusion criteria, 24 studies were included in this systematic review, No new records met our inclusion criteria, and thus, all were excluded.

The characteristics of the 24 studies are summarized in Table 1 . All studies were carried out in rodents (rats and mice). Intervention included MSCs obtained from mice, rat, or human bone marrow. The most common sepsis models were the cecal ligation and puncture (CLP) and intraperitoneal injection of lipopolysaccharide (LPS). Following induction of sepsis, MSC(M) were injected within a few hours of the induction of the sepsis animal models. The mean follow-up ranged from 8 days to 20 weeks. The most common delivery routes used for MSC(M) were intravenous route or intraperitoneal route. The total dose of MSC(M) administered ranged from 1×10⁵ to 1×10⁷ cells.

The RoB assessment results of included studies is summarized in Table 2 . There was no study considered to have a low RoB. There was no significant difference in the baseline characteristics between the experimental and control groups, reducing the chance of selection bias based on animal characteristics. No study has explicitly described how random sequences are generated despite the random allocation of experimental and control subjects. Accordingly, the RoB was deemed to be “unclear” in the sequence generation domain of all studies included. However, no study attempted to describe the method of concealed allocation. An evaluation of randomized outcomes was reported, the intervention received by each animal was blinded to researchers and, in one study (22), the evaluator’s blindness was noted. Based on the provided signal questions, all of the included studies showed a low risk of attrition bias and reporting bias. In one study that had a high risk of bias, the decrease in animal numbers was not accounted for between methods and results (15). In one study (16), there were other problems that caused a high risk of bias, such as pollution, experimental design, etc. Furthermore, we did not identify any additional sources of bias or bias tools in relation to systematic risks that were not included.

We conducted a sensitivity analysis to evaluate the stability of the results by sequential omission of each study if heterogeneity between the studies existed. The rate of mortality outcome was not significantly affected by any study ( Figure 2 ).

A total of 24 animal studies involving 1278 animals were used in this meta-analysis and reported animal mortality rates. Heterogeneity test results showed I ² = 9% and P = 0.34, indicating that the heterogeneity between the studies was low; thus, a fixed effects model was used. As shown in Figure 3 , the pooled results demonstrated that the mortality of the animals after MSC(M) treatment was significantly reduced (OR 0.34, 95% CI 0.27-0.44, P<0.0001).

Subgroup analysis was performed based on the animal species and model, as well as the model gender, species, Sepsis model, and MSC(M) dose, injection time, source, and injection route. In general, significant efficacy of MSC(M) transplantation was observed in most subgroups. In animal models with female participants, MSC(M) for females (OR=0.26, 95% CI 0.13-0.49, P <0.001) had a greater effect than MSC(M) for male participants (OR=0.27, 95% CI 0.19-0.30, P<0.001) and not reported group (OR=0.69, 95% CI 0.42-1.13, P<0.001) ( Figure S1 ). Furthermore, MSC(M) administered to non-CLP-induced sepsis animals (OR=0.26, 95% CI 0.16-0.41, P<0.001) were more beneficial than those administered to CLP-induced sepsis animals (OR=0.39, 95% CI 0.28-0.52, P<0.001) ( Figure S2 ), and MSC(M) administered to rats (OR=0.23, 95% CI 0.10-0.55, P<0.001) were more beneficial than those administered to mice (OR= 0.36, 95% CI 0.27-0.45, P<0.001) ( Figure S3 ). A significant decrease in mortality rate is observed in both doses of ≥1×10⁶ and <1×10⁷ MSC(M) (OR=0.21, 95% CI 0.14-0.32, P<0.001) and ≥1×10⁷ MSC(M) (OR=0.32, 95% CI 0.06-1.85, P<0.001), as compared to doses of <1×10⁶ MSC(M) (OR=0.50, 95% CI 0.35-0.70, P<0.001) ( Figure S4 ). Additionally, injection times of ≤6h (OR=0.26, 95% CI 0.19-0.37, P<0.001) and better than >6h (OR=0.48, 95% CI 0.33-0.70, P<0.001) after sepsis induction were noted ( Figure S5 ). Further, xenogenic MSC(M) (OR=0.30, 95% CI 0.19-0.47, P<0.001) significantly reduced mortality compared to syngeneic MSC(M) (OR=0.33, 95% CI 0.21-0.51, P < 0.001) and allogenic MSC(M) (OR=0.40, 95% CI 0.20-0.62, P<0.001) ( Figure S6 ). As well, intraperitoneal administration (OR=0.04, 95% CI 0.00-0.37, P<0.001) is superior to intravenous administration (OR=0.36, 95% CI 0.28-0.46, P<0.001) after sepsis induction in animal models ( Figure S7 ).

We assessed publication bias by funnel plots ( Figure 4 ). No evident publication bias was observed by visual inspection. The Egger’s test presented publication bias for mortality rates (P = 0.004). After adopting trim-and-fill correction for mortality rates outcomes, the estimated value remained unchanged.

According to the results of this meta-analysis: (1) MSC(M) therapy significantly reduced the mortality rate of sepsis animal models, supporting the possibility of using MSC(M) therapy for preclinical studies of sepsis. (2) The administration time was correlated with mortality. MSC(M) therapy initiated within 6 hours of the onset of sepsis exhibited the greatest efficacy. (3) For the mortality rate, intraperitoneal administration injection seems to show the greatest efficacy, followed by intravenous administration, however, this finding should be interpreted cautiously. (4) Xenogenic MSC(M) significantly reduced the mortality rate of sepsis in the animal models, compared to Syngeneic and allogeneic MSC(M). (5) It is unclear from the included studies what is the appropriate dosage of MSC(M). In contrast to the previously published meta-analysis (38), which indicated that less than or equal to 1.0 × 10⁶ MSCs was the ideal dose, the conclusion of our analysis was that more than or equal to 1.0 ×10⁶ MSC(M) is the ideal dose. Therefore, it will be necessary to conduct additional research to determine the ideal dose of MSC(M) for the treatment of sepsis. As a general rule, the above various subgroup analyses are only able to generate hypotheses rather than confirm them. As a result of the subgroup analysis, the results are speculative because they are based on the reanalysis of published data rather than the results of a well-constructed randomized controlled trial. Thus, despite the fact that the subgroup analysis provided updated evidence, the results of this analysis should be considered cautiously.

Numerous preclinical studies have demonstrated the benefit of MSC(M) in animal models of sepsis. The precise mechanism by which MSC(M) may exert beneficial effects in sepsis is still being elucidated, but it appears that multiple mechanisms may contribute. Besides mesenchymal stem cells’ homing properties, progress has also been made with understanding the mechanisms of benefit, including paracrine factors, mitochondrial transfer, and the development of extracellular vesicles and microvesicles. ( Figure 6 ).

Firstly, MSC(M) may be most helpful in alleviating sepsis due to their anti-inflammatory properties. The MSC(M) can promote the repolarization of monocytes and/or macrophages from a type 1 to a type 2 phenotype, characterized by high levels of intracellular IL-10 secretion, increased phagocytosis, low levels of TNF‐α and interferon‐γ production (49). Reprogrammed type 2 monocytes produced by MSC(M) produce large amounts of IL-10, which inhibits neutrophil influx into injured tissue and prevents further damage. Additionally, MSC(M) can also interfere with the differentiation, maturation, and function of dendritic cells, causing them to adopt a regulatory phenotype (50, 51). The MSC(M) can modulate natural killer cells, which are involved in eradicating virus-infected and damaged cells, as well as releasing a range of proinflammatory cytokines, such as interferon‐γ. It has been demonstrated in a number of studies that MSC(M), in co-culture with natural killer cells, impair their cytotoxic activity, their cytokine production, and their ability to release granzyme B (52). Moreover, MSC(M) have also been shown to have antimicrobial effects in animals. MSC(M) contributed to improved bacterial clearance by secreting antibacterial proteins/peptides such as LL-37 (53),lipocalin-2 (54), and so on. Secondly, several studies have demonstrated that MSC(M) have beneficial effects on metabolomics through mitochondrial transfer using extracellular vesicles. In a model of lipopolysaccharide-induced ALI, mitochondrial transfer through connexin-43 was partially responsible for restoring ATP levels (55). The therapeutic effects of MSC(M) in LPS-induced ALI were abrogated when mutant MSC(M) with defective gap junctions or MSC(M) with dysfunctional mitochondria were used. In addition, a model of E.coli pneumonia has shown that mitochondria transfer from MSC(M) to macrophages is necessary for enhancing macrophage bioenergetics and phagocytosis, as well as the antimicrobial effects of MSC(M) in vivo ( 56). Finally, extracellular vesicles (EVs) derived from MSC(M) may contribute to the paracrine effect. These MSC(M)-EVs contribute molecules (such as proteins, peptides, mRNA, microRNA, and lipids) with immunoregulatory properties. MSC(M)-EVs have been found to mimic MSC(M) in alleviating sepsis in healthy people and may serve as an alternative to whole-cell therapy. In contrast to MSC(M), MSC(M)-EVs may offer specific advantages due to their low immunogenicity and high safety profile. Even though several studies have demonstrated the benefits of MSC(M)-EVs in sepsis, the underlying mechanisms of MSC(M)-EVs remain unclear. In order to further understand the mechanism by which beneficial effects are mediated, we need to increase our understanding of paracrine factors and the transfer of mitochondria and microcapsules.

We have identified several limitations to our meta-analysis. Firstly, our approach can include only those studies that have been traditionally published in English. Unpublished data may influence our results. Furthermore, none of the included studies examined the safety of MSC(M) injection in sepsis models in animals. We are unable to evaluate the clinical safety of MSC(M). Finally, a good study requires a sample size calculation that is based on formal procedures. However, no study in the meta-analysis had conducted a sample calculation, indicating insufficient statistical power to determine the treatment effect.