How GLP-1 Receptor Agonists May Influence Cell Cleanup in Brain Diseases

This study was conducted in accordance with the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) (Page et al., 2021). A literature search was performed for studies that investigated the effect of GLP-1RAs on autophagy in models of neurodegenerative disorders, specifically AD and PD.

A systematic search was conducted on PubMed, Web of Science, and OVID (including Medline, Embase, and APA PsycInfo databases) from inception to June 17, 2025 for English-language articles using the following search string: ((‘GLP-1’ OR ‘Glucagon-Like Peptide-1’ OR ‘GLP-1 agonist’ OR ‘GLP-1 receptor agonist’ OR ‘Glucagon-Like Peptide-1 agonist’ OR ‘Glucagon-Like Peptide-1 receptor agonist’ OR ‘Glucagon like peptide-1’ OR ‘Semaglutide’ OR ‘Ozempic’ OR ‘Wegovy’ OR ‘Rybelsus’ OR ‘Dulaglutide’ OR ‘Trulicity’ OR ‘Liraglutide’ OR ‘Saxenda’ OR ‘Victoza’ OR ‘Tirzepatide’ OR ‘Mounjaro’ OR ‘Zepbound’ OR ‘Exenatide’ OR ‘Byetta’ OR ‘Bydureon’ OR ‘Bydureon BCise’) AND (Autophagy) AND (‘Alzheimer’s’ OR ‘Alzheimer’s Disease’ OR ‘Neurodegenerative Disorder’ OR ‘Dementia’ OR’ Parkinson’s’ OR ‘Parkinson’s Disease’ OR ‘Neurological Disorder’ OR ‘Memory Disorder’)).

Subsequently, two reviewers (M.C.S. and I.H.) independently screened the studies on Covidence to prevent any influence on decisions from either of the reviewers and to enable transparency with the screening process. After duplicates were removed, articles were assessed based on their title, abstract, and full text according to the eligibility inclusion/exclusion criteria. Conflicts on screening were resolved via a discussion and a consensus decision.

Any experimental studies using in vitro or in vivo models to investigate AD or PD were included. The presence of AD in these models was assessed based on the detection of two core biomarkers: (1) amyloid-beta (aβ) plaques and (2) NFTs of abnormally phosphorylated tau proteins (Jack et al., 2024). Meanwhile, PD models were included based on the presence of key pathological features, including α-syn aggregation and the progressive loss of dopaminergic neurons (Ioghen et al., 2023; Pandey et al., 2024). Furthermore, the presence of key clinical motor symptoms in in vivo samples must be observed: bradykinesia in combination with rigidity or resting tremors, or both (Postuma et al., 2015).

Studies were included if they investigated the effect of FDA-approved GLP-1RAs in comparison to controls who received either a placebo or no intervention. Additionally, studies that reported any changes in autophagy along with improvements in core symptoms of AD or PD were included.

Studies were excluded if they: (1) did not have participants with AD or PD (i.e., had any other neurodegenerative condition); (2) used non-FDA-approved GLP-1RAs or any form of dipeptidyl peptidase-4 inhibitors due to any potential differences in mechanism from GLP-1RAs; (3) did not assess autophagy as an outcome; (4) had any other study design except an experimental/preclinical or randomised controlled trial (RCT).

Data extraction was performed independently by two reviewers (M.C.S. and I.H.) on Google Spreadsheets. The studies were split into two sections: an in vitro section and an in vivo section. Once in their respective sections, the studies were organised and analysed through several components: (1) author; (2) disease model (AD or PD); (3) study design (experimental vs RCT); (4) cell type used (for in vitro studies) or animal model characteristics (for in vivo studies); (5) stressor used; (6) sample size; (7) type of GLP-1RA used and its dosage; (8) groups (control(s) and experimental); (9) main findings pertaining to autophagy. To enable some level of comparability between all the studies, a focus on extracting data on autophagy-specific markers such as beclin-1, ATG7, ATG3, microtubule-associated protein light chain 3 (LC3-II/LC3-I), lysosomal-associated membrane protein 1 (LAMP1), and p62 [also commonly known as sequestosome 1 (SQSTM1)] was done. The extracted data were assessed for inclusion by both reviewers.

Risk of bias (RoB) was assessed by two independent reviewers (M.C.S. and I.H.). The SYRCLE’s RoB tool for animal studies was utilised for in vivo studies (Hooijmans et al., 2014). Meanwhile, the Modified SYRCLE’s RoB tool was employed for in vitro studies (De la Rosa González et al., 2024). All conflicts were resolved by discussion between the two reviewers.

The initial literature search yielded 142 studies, of which 64 duplicates were identified on Covidence and 5 were identified manually and removed. Based on title and abstract screening, 42 studies were deemed irrelevant, while the remaining 31 studies progressed onto full-text screening. Seventeen studies were excluded for having either the wrong intervention [e.g., a dual agonist or non-FDA-approved GLP-1RAs (n = 10)], incorrect patient population [e.g., samples were not models of AD or PD but of other conditions (n = 4)], or wrong outcomes [i.e., did not look at autophagy as an outcome (n = 3)]. As a result, 14 articles were included in the review herein and underwent data extraction (Figure 1).

Of the total 14 articles included, two studies (Bu et al., 2021; Zhang et al., 2021 a) incorporated both in vivo and in vitro experiments; therefore, each component was treated as distinct and assessed separately during data extraction. Accordingly, 8 studies were considered in vivo RCTs, while the other 8 were in vitro experimental studies. Of the in vitro cell studies, only 1 used Lund Human Mesencephalic (LUHMES) dopaminergic-like neurons (Panagaki et al., 2023) while the rest used the SH-SY5Y human neuroblastoma cell line as their model. For their intervention, a total of 3 studies examined liraglutide (Lira) (Panagaki et al., 2017; Kong et al., 2020; Panagaki et al., 2023), 2 studies investigated exendin-4 (Ex-4) (Bu et al., 2021; Zhang et al., 2021 a), and 1 study explored semaglutide (Sema) (Chang et al., 2020); furthermore, 2 studies investigated more than 1 GLP-1RA: 1 study examined Lira and Ex-4 (Jalewa et al., 2016), and 1 study looked at Lira and Sema (Liu et al., 2022).

From the included in vivo studies, only 3 used transgenic models (Bu et al., 2021; Zhang et al., 2023; Elbadawy et al., 2025), while the others primarily utilised environmental stressors such as 1-methyl-4-phenyl-1,2,3,6-tetrahydropyridine (MPTP), 6-hydroxydopamine (6-OHDA), or thapsigargin. For their intervention, a total of 4 studies examined Lira (Zhang et al., 2020; Lin et al., 2021; Wu et al., 2022; Zhang et al., 2023), 2 studies investigated Ex-4 (Bu et al., 2021; Zhang et al., 2021 a), 1 study explored Sema (Elbadawy et al., 2025), and 1 study looked at both Lira and Sema (Zhang et al., 2018).

Overall, 5 articles also included dual agonists as another comparison intervention; however, only results pertaining to FDA-approved single GLP-1RAs were included from these studies (Jalewa et al., 2016; Zhang et al., 2020; Zhang et al., 2021 a; Panagaki et al., 2023; Zhang et al., 2023). A summary of the study characteristics is exhibited in Table 1.

Details of the RoB assessment are presented in Tables 2 and 3 for in vivo and in vitro studies, respectively. In the majority of the in vivo studies, a high risk of selection and performance bias was due to a lack of randomisation and blinding of investigators to intervention groups. Additionally, an unclear RoB was mainly given in circumstances where the criteria of the domain were partially met but not enough to ensure a low RoB. It must be noted that, although all the studies except one (Elbadawy et al., 2025) did not have blinded outcome assessors (D7), this was not deemed a serious concern in terms of bias, as the animals were euthanised to measure the outcome related to autophagy. Furthermore, regarding whether the assessment of incomplete data was adequately addressed (D8), only sections related to autophagy were examined for this domain, since only data related to autophagy were extracted.

As for the in vitro studies, a high risk of detection bias (D7) was due to a lack of blinding of the outcome assessor; moreover, an unclear RoB for other sources of bias (D10) was given to studies that had potentially concerning conflicts of interest, among other reasons.

Among some of its other functions, beclin-1 is a central regulator of autophagy. Beclin-1 serves as a scaffold protein that forms an autophagy-specific complex with multiple other proteins at the ER to create a phagophore, the precursor of an autophagosome, ultimately initiating autophagy (Tran et al., 2021). Deficiency of beclin-1 in response to neurodegenerative-induced stress was consistently observed among the included studies and is supported by other literature (Pickford et al., 2008). Moreover, an elevation of beclin-1 after GLP-1RA administration was a common finding, indicating restored or elevated autophagy.

LC3-I is a cytosolic protein that is conjugated to phosphatidyl-ethanolamine (PE) to form LC3-II. The lipidation of LC3-I to LC3-II enables its insertion into the autophagosome inner and outer membranes, serving as a crucial marker of autophagosomes (Ye et al., 2023; Runwal et al., 2019). The included studies reported a notable reduction in the ratio of LC3-II/LC3-I after AD and PD-like stress was induced, indicating impaired autophagy. Following GLP-1RA treatment, LC3-II levels were rescued, implying restored autophagic flux.

ATG7 and ATG3 act as E1-like and E2-like enzymes, respectively. The aforementioned proteins are integral to the process of catalysing the conjugation of LC3-I to PE to form LC3-II after autophagy has been induced (Frudd et al., 2018). To our knowledge, Zhang et al. (2018) was the only in vivo study to explore the impact of neurodegenerative stress and GLP-1RA on ATG7, while several other in vitro studies investigated both ATG7 and ATG3 levels. The collective finding was a suppression of both ATG7 and ATG3 levels after disease induction, with an increase of these markers after GLP-1RA treatment.

In addition to its other roles, p62 functions as a receptor that binds specific ubiquitin-tagged proteins and damaged organelles, delivering them to the autophagosome for degradation. Since p62 is degraded along with the cargo, accumulated levels of p62 are thought to be indicative of impaired autophagy (Liu et al., 2016; Kumar et al., 2022). There were conflicting findings within in vitro and in vivo studies regarding the influence of neurodegenerative-induced stress and GLP-1RAs on p62 levels. Some studies reported elevated p62 in response to disease induction, with GLP-1RAs reducing it. Other articles reported the opposite, where AD and PD lowered p62 levels, and GLP-1RA treatment restored them. Despite this discrepancy, the majority of the included studies, as well as other literature, support the former, where neurodegenerative stress raised p62 levels, with subsequent normalisation of p62 levels following GLP-1RA administration (Runwal et al., 2019; Bjørkøy et al., 2009).

LAMP1 is a glycosylated transmembrane protein found in lysosomes (Cheng et al., 2018). The literature delineates that disruption of LAMP1 is associated with an accumulation of autophagy vesicles as a result of ineffective clearance inside lysosomes (Eskelinen, 2006). Bu et al. (2021) was the only study to investigate this lysosomal marker and found a notable increase in LAMP1 levels following Ex-4 treatment.

While these findings yield valuable insights into GLP-1RAs’ modulation of autophagic processes, there are some limitations to consider, such as the heterogeneity of the included studies. The data reported in this review was collected from preclinical studies that utilised a variety of different cell and animal models, GLP-1RAs (types and dosages), and end points, which made direct comparisons difficult. Moreover, a non-standardised tool was employed to assess the risk of bias for the in vitro studies. None of the studies that we were reviewing were clinical studies, which represents a very significant limitation, and the extent to which the preclinical findings extend to human results is uncertain.

Furthermore, relatively few studies assessed dynamic autophagic flux. As such, deciphering marker levels in association with improved or impaired autophagy can be misconstrued when the timing and context aren’t considered. This limitation could explain the incongruence in p62 levels after disease induction and GLP-1RA administration, as other articles have reported that p62 levels are not always inversely associated with autophagic flux during cellular stress (Liu et al., 2016). Another instance is the interpretation of LC3-II/LC3-I levels, which can also be somewhat problematic, as LC3-II levels measured at a single time point do not necessarily reflect overall autophagic flux (Mizushima and Yoshimori, 2007). For instance, elevations in LC3-II could indicate enhanced autophagosome formation through increased conjugation of LC3-I with PE; however, it could also imply ineffective and substandard lysosomal degradation (leading to an accumulation of autophagosomes with LC3-II) (Sarkar et al., 2014; Sharifi et al., 2015). It remains a possibility that the discrepancies and/or inconsistencies that we have identified in the markers of autophagy that we have evaluated may be related to the inadequate assessment of dynamic autophagy processes as much of the datapoints could be considered static, it would be the case that a more comprehensive and dynamic autophagy approach may result in findings that are discrepant from what we have observed.

An overarching limitation which emanates and surrounds the entire literature review that we have conducted relates to the heterogenous findings. It would be accurate to state that our findings represent a disparate assortment of molecular readouts, most of which are in need of replication especially in human subjects. The degree to which the effects we are observing are specific or pseudospecific cannot be ascertained from the methodologic approach we have taken. Notwithstanding this overarching limitation, a theme throughout the results that is relatively consistent is the observation that molecular changes that are observed relate to processes integral to autophagy.

Future research could assess the dynamic process of autophagy, as it may provide a clearer and more integrative understanding of what is occurring, as opposed to static measurements. Additionally, future studies should broaden their investigation to examine the influence of GLP-1RAs on other neurodegenerative conditions and mental disorders. Performing clinical trials would also prove valuable, as they could evaluate whether these findings translate to human populations.

Recent research has investigated microbiota-based therapeutics in ameliorating autophagy through modulating endogenous GLP-1 as a potential method to mediate AD and PD symptomology (Bonfili et al., 2017; Qi et al., 2025). Further investigation of this approach could reinforce understanding of these relationships. Another avenue of exploration is the influence of dual agonists on autophagy as opposed to single GLP-1RAs, as the dual agonists have also been reported to have a potentially more robust effect on markers of autophagy (Jalewa et al., 2017; Panagaki et al., 2018; Cai et al., 2021).