Alzheimer’s disease (AD) is the leading cause of dementia, characterized by progressive cognitive and functional decline, often accompanied by a long-term need for care and a significant socioeconomic burden. Epidemiological data indicate that over 56 million people worldwide are currently living with dementia, a number expected to continue rising due to population aging. Consequently, the direct and indirect costs associated with dementia have become an unavoidable long-term strain on public health systems (1–3). However, existing mainstream pharmacological treatments, such as cholinesterase inhibitors and N-methyl-d-aspartate receptor antagonists, provide only limited symptomatic relief and are unable to halt or reverse the underlying pathological processes of the disease (1, 4, 5). In recent years, disease-modifying therapies such as monoclonal antibodies targeting amyloid-beta (Aβ) have made breakthroughs, but their heterogeneous efficacy, potential risk of imaging abnormalities, and high costs have limited their widespread application (5, 6). Therefore, the search for disruptive therapeutic strategies capable of multi-target intervention in AD’s complex pathological mechanisms, with the potential to achieve neurorepair, has become a critical challenge in the field.

Beyond cognitive impairment, AD is frequently accompanied by a range of neuropsychiatric symptoms (NPS), such as anxiety, depression, and apathy. Collectively, these manifestations are referred to as the behavioral and psychological symptoms of dementia (BPSD) and profoundly compromise the quality of life of both patients and their caregivers (7, 8). The hippocampus, a key brain region for cognition and emotional regulation, is a common pathological substrate underlying both cognitive deficits and BPSD in patients with AD (9).

In this context, mesenchymal stem cells (MSCs) have gained attention as promising therapeutic candidates due to their self-renewal capacity, multipotent differentiation potential, robust immunomodulatory functions, and ability to secrete neurotrophic factors (10, 11). Among the various sources of MSCs, human umbilical cord-derived mesenchymal stem cells (hUCMSCs) offer distinct advantages for clinical translation. Derived from perinatal medical waste, hUCMSCs are ethically uncontroversial, exhibit superior proliferative ability, lower immunogenicity, and notably potent paracrine activity (12, 13). Preclinical evidence from AD animal models also provides biological feasibility for the use of hUCMSCs. Previous studies in transgenic mouse models have shown correlations between hUCMSCs transplantation and reductions in Aβ load, as well as improvements in behavioral outcomes such as learning, memory, and sensorimotor functions, suggesting that hUCMSCs may exert a comprehensive effect by modulating the pathological microenvironment (13–16). Recently, Murayama (17) published a review that systematically summarized diverse therapeutic strategies targeting Aβ and tau proteins and highlighted the potential value of stem cell therapy in the treatment of AD, providing an important theoretical foundation for the present work. However, existing animal studies differ markedly in the types of models used, and the pathological characteristics and disease timelines vary across AD mouse models. For example, the APP/PS1 model co-expresses human APP carrying the Swedish mutation (K595N/M596L) and human PSEN1 carrying the L166P mutation, leading to increased Aβ42 production and progressive plaque deposition; the 5xFAD model co-expresses APP mutations (Swedish, Florida [I716V], and London [V717I]) together with PSEN1 mutations (M146L/L286V), resulting in extremely rapid pathological progression; whereas the Tg2576 model overexpresses human APP with the Swedish mutation alone, with later onset of Aβ deposition and cognitive deficits. In addition, some studies have used SAMP8 mice, which exhibit accelerated senescence, or models induced by hippocampal injection of Aβ oligomers (18, 19). This diversity in model types, together with substantial variation in the route, dose, and frequency of hUCMSC administration, may not only obscure the true magnitude of treatment effects but also make it difficult to translate seemingly consistent positive into reproducible clinical protocols.

Given this context, the present study aims to provide a more accurate estimate of the overall effect size of hUCMSCs on cognitive, behavioral, and core pathological outcomes through a rigorous and comprehensive systematic review and meta-analysis within a unified cell source framework. Subgroup analyses will be conducted to identify key variables that influence efficacy, such as administration routes, dose/frequency, and outcome measurement systems. The goal is to identify potential “optimal windows” for therapeutic intervention and, through systematic bias risk and reporting quality assessments, to provide a higher level of evidence for optimizing future animal study designs and clinical translation pathways.

This study provides robust preclinical evidence for the use of hUCMSCs in treating AD mouse models through a rigorous systematic review and meta-analysis of 13 randomized controlled trials. Overall, our analysis not only confirms that hUCMSCs transplantation significantly reverses spatial learning and memory deficits in AD model mice but also reveals its dual role in alleviating core pathological features (Aβ load) and remodeling the neural microenvironment (upregulating BDNF and inhibiting apoptosis). Importantly, by exploring the sources of heterogeneity, we identified high-dose intravenous infusion as a potential optimal clinical translation window for achieving the best therapeutic effects, providing critical guidance for the design of future clinical trials.

Notably, Chen et al. (21) recently reported the first clinical observational study of intravenous infusion of human umbilical cord blood mononuclear cells (hUCBMCs) in patients with mild to moderate AD (21). In that study, 11 patients with AD were enrolled, and three infusions of hUCBMCs were found to transiently improve cognitive function (MMSE and ADAS-Cog scores), activities of daily living (ADL), and sleep quality (PSQI), with a favorable safety profile. Although hUCBMCs and hUCMSCs differ in cellular composition and biological properties (hUCBMCs comprise multiple cell populations, including hematopoietic and endothelial progenitors, whereas hUCMSCs are purified mesenchymal stem cells), both are umbilical cord-derived and share comparable immunomodulatory and neurotrophic potential. The clinical observations by Chen et al. (21) provide preliminary evidence for the feasibility and safety of umbilical cord-derived cell therapy in patients with AD, and are consistent with our meta-analytic findings, jointly supporting the therapeutic potential of umbilical cord-derived stem cells/cells for AD. Our systematic review further delineates the multi-target of hUCMSCs in animal models and suggests that medium-to-high dose intravenous infusion may be critical for optimizing clinical efficacy. These findings offer important guidance on dosing and administration routes for future clinical trials based on hUCMSCs.

Delving into the biological mechanisms, the cognitive benefits of hUCMSCs appear to stem from a multi-targeted, synergistic disease-modifying effect rather than a singular symptomatic treatment. Our meta-analysis results show a high degree of consistency between improvements in cognitive behavior and significant reductions in brain Aβ levels, along with a marked increase in BDNF protein expression. This suggests that hUCMSCs may exert neuroprotective and reparative effects through their paracrine activity. On one hand, the various bioactive factors they secrete can modulate the immune microenvironment in the brain, promoting the transformation of microglia into an anti-inflammatory phenotype (M2 type) with phagocytic capabilities, thereby enhancing the clearance of Aβ plaques (22, 23). On the other hand, hUCMSCs can continuously release neurotrophic factors such as BDNF, which not only support neuronal survival and synaptic plasticity but also directly counteract synaptic toxicity induced by Aβ oligomers and inhibit the resultant neuronal apoptosis (23–25). Compared to current clinical monoclonal antibody therapies that solely target Aβ clearance, this comprehensive intervention strategy, which both clears pathological products and actively promotes neural repair, clearly offers a more holistic advantage in addressing the complex pathological network of AD.

Regarding the choice of administration route, our study found that systemic administration (intravenous injection) demonstrated efficacy that was superior to or equivalent to that of local administration (hippocampal or ventricular injection) under specific conditions. Intuitively, direct injection into the brain should provide a higher local concentration of cells; however, our subgroup analysis revealed that bilateral hippocampal injection at low doses did not significantly improve cognitive behavior or reduce Aβ levels, while the same dose via tail vein injection showed significant efficacy. This seemingly paradoxical result may be attributed to local neuroinflammation caused by surgical trauma, which could offset the modest benefits of stem cells. It also suggests that hUCMSCs may possess a unique peripheral-central immune regulatory mechanism (26). At present, direct comparative evidence on different administration routes of hUCMSCs in AD models remains scarce, and some mechanistic insights are extrapolated from other disease models. For example, in ischemic stroke models, intravenously administered MSCs can confer neuroprotection by modulating peripheral immune cells (26). Building on these indirect data, we propose the following hypothesis for future testing: although most intravenously cells are initially trapped in the lungs, they may act by regulating peripheral immune organs such as the spleen and suppressing systemic inflammation, thereby improving the brain microenvironment through immune crosstalk along the gut–brain axis or across the blood–brain barrier (27–30). This raises the possibility that, for a disease such as AD, characterized by diffuse brain involvement and systemic metabolic disturbances, broad immunomodulation via the peripheral circulation may be more effective and safer than focal, target-specific injections. Nonetheless, this hypothesis requires validation in AD animal models through head-to-head comparisons of different administration routes and mechanistic experiments. Moreover, peripheral administration is also the most acceptable and readily translatable route for cell-based therapies in current clinical practice.

In addition to the administration route, dose dependence is another critical factor influencing therapeutic efficacy. Our data clearly demonstrate that under the tail vein administration route, the high-dose group exhibited significantly better Aβ clearance and BDNF enhancement compared to the low-dose group, indicating a pronounced “threshold effect” for hUCMSCs treatment. This suggests that a sufficient number of cells must be administered to generate a high enough concentration of cytokines to effectively reverse established pathological changes. This finding poses a challenge for human clinical translation, as simple allometric scaling based on body weight may result in a massive required cell quantity. Therefore, future translational research should focus on exploring how to maintain effective drug concentrations through multiple high-dose administration regimens in primate models while ensuring safety, thereby overcoming the efficacy threshold.

In addition to improving cognitive function, hUCMSCs transplantation may also exert beneficial effects on BPSD. Murayama et al. reported that neural stem cell transplantation significantly ameliorated anxiety- and depression-like behaviors in AD model mice, and that these improvements were associated with enhanced hippocampal neurogenesis and modulation of neurotransmitter systems such as the cholinergic and GABAergic pathways (31). Although the studies included in this meta-analysis primarily focused on cognitive performance and Aβ pathology, hUCMSCs may indirectly alleviate BPSD by reshaping the neural microenvironment, promoting the release of neurotrophic factors, and suppressing neuroinflammation. Future work should more systematically evaluate the impact of hUCMSCs on BPSD and incorporate these outcomes as an integral component of overall efficacy assessment.

Despite the encouraging evidence provided by this study, we must acknowledge that the included primary studies have notable methodological limitations. The risk-of-bias assessment showed that nearly all studies were judged as having an “unclear risk of bias” in key domains, including random sequence generation, allocation concealment, blinding of caretakers and investigators, and blinding of outcome assessment, suggesting a substantial likelihood of performance and detection bias at the level of the overall evidence base. Thus, although the meta-analysis indicates that hUCMSCs exert significant therapeutic effects in AD mouse models, the credibility of these findings is constrained by the generally low quality of the underlying studies. In addition, the effect sizes observed for outcomes such as Aβ reduction (SMD = −5.34) and BDNF elevation (SMD = 4.25) were extraordinarily large. Such inflated effect sizes are a common feature of preclinical meta-analyses and are typically closely linked to limitations in the original studies. As highlighted by the risk-of-bias assessment, blinding procedures were seldom clearly reported, sample sizes were generally small, and publication bias favoring positive findings likely contributed to systematic overestimation of treatment effects. For these pronounced pooled effect sizes, we therefore place greater emphasis on their direction rather than their absolute magnitude—that is, hUCMSCs treatment consistently favors improvement—rather than overinterpreting their possibly exaggerated numerical values. This interpretation is consistent with the observation that, after adjustment for publication bias, effect sizes decreased but remained significant, further supporting the robustness of the beneficial direction of hUCMSCs effects, while underscoring that the precise magnitude of benefit be confirmed in future large-scale, rigorously blinded, high-quality studies.

Of particular note, the impact of publication bias on the core outcomes cannot be ignored. Trim-and-fill adjustment for escape latency reduced the pooled effect size from −2.55 to −1.85, a decrease of approximately 27%. This substantial attenuation strongly suggests that, owing to missing negative or neutral studies, the observed pooled effect may be markedly overestimated and the true treatment effect is likely considerably smaller than the current pooled estimate. This further underscores that, when interpreting our findings, greater emphasis should be placed on the direction and robustness of the treatment effect rather than on its absolute magnitude. In addition, this study did not include several key AD-related outcomes such as tau pathology, synaptic markers, and neuroinflammation. These measures are central to AD pathogenesis and may represent important targets through which hUCMSCs exert their therapeutic actions. Because of the lack of available data in the primary studies, we were unable to quantify these effects, which limits a comprehensive understanding of the mechanisms underlying hUCMSCs therapy. Future animal experiments should systematically incorporate these endpoints to more fully evaluate the multi-target therapeutic potential of hUCMSCs.

Another critical limitation lies in the sex bias of the included studies. The nine studies that explicitly reported animal sex, all used male mice only. This pronounced sex bias substantially limits the generalizability of the findings. Given the well-established sexual dimorphism in AD—encompassing incidence, pathological features (such as patterns of Aβ deposition and tau phosphorylation), and immune–inflammatory responses—efficacy data derived solely from male animal models may not be readily extrapolated to females (32, 33). Of particular concern, the immunomodulatory properties of mesenchymal stem cells themselves may be influenced by sex-related hormonal and gene expression differences. Consequently, the current evidence base cannot address the pivotal question of whether hUCMSCs treatment confers comparable benefits in female AD models. Future preclinical studies must include both sexes to evaluate sex-specific treatment responses and to provide more broadly applicable evidence for dose selection and therapeutic regimens in subsequent clinical trials.

Second, a central limitation of this meta-analysis is the substantial heterogeneity observed in some pooled outcomes, particularly escape latency and Aβ levels. Although our prespecified subgroup analyses based on route and dose of administration successfully accounted for most of the heterogeneity and reduced within-subgroup heterogeneity to an acceptable range (with most I² values < 50%), we cannot fully exclude the influence of additional confounders. These may include inherent differences in pathological mechanisms and disease trajectories across AD models (e.g., APP/PS1 vs. 5xFAD), as well as variation in mouse sex and age, length of post-transplant follow-up, and the specific analytical methods used. Such biological and methodological diversity is an inherent challenge in preclinical meta-analyses and may lead to overestimation or imprecise estimation of effect sizes. Consequently, while our subgroup findings provide important clues for identifying a potential optimal therapeutic window, they require confirmation in more standardized animal study designs. Based on the current evidence, we cannot assert the existence of a definitive “optimal clinical window,” but can only propose that medium-to-high doses administered intravenously represent a promising direction for further optimization. Future studies adhere closely to established guidelines for animal study design and reporting (such as the ARRIVE guidelines) to generate higher-quality evidence and provide a more robust foundation for clinical translation.

This study shows that hUCMSCs transplantation significantly improves spatial learning and memory deficits in AD mouse models, reduces brain β-amyloid burden, increases neurotrophic factor levels, and attenuates neuronal apoptosis. Subgroup analyses further suggest that the intravenous route and medium-to-high dosing may be key variables associated with greater efficacy. Although the low methodological quality of the primary studies (with a risk of effect inflation), the likelihood of publication bias, the pronounced sex bias limiting generalizability, and the residual heterogeneity in some outcomes all warrant caution, our findings nonetheless provide directional, systematic evidence supporting hUCMSCs as a promising disease-modifying therapy for AD and offer important guidance for optimizing treatment regimens in future clinical translation.