Neuroprotective Potential of SGLT2 Inhibitors in Animal Models of Alzheimer’s Disease and Type 2 Diabetes Mellitus: A Systematic Review

1.

The research gap is not clearly mentioned in the introduction section. Clarify the specific gap with which this systematic review aims to address

Rewrite and clarified our research gap.

“Despite the growing interest in the potential of SGLT2i to mitigate cognitive decline, clinical data on AD patients remain scarce. The only randomised controlled trial designed to evaluate the mental effects of SGLT2i in AD patients is trial NCT03801642, conducted by the University of Kansas Medical Centre. Its full results have not yet been published. As such, majority of the mechanistic insights have been gained from preclinical data. However, the potential role of SGLT2is in preventing or treating AD has not been systematically or comprehensively assessed. Hence, there is an important gap in the current body of evidence.”

2.

Elaborate on why SGLT2 inhibitors would mechanistically benefit type 3 DM in relation to AD

Thank you for the comment. After reviewing all of the feedback, we have decided to remove the term “T3DM” throughout the manuscript, as it is not clinically accepted and only a few publications acknowledge its existence. Instead, we have revised the manuscript to consistently use “T2DM.” This change is also in line with comment no. 12 from Reviewer 1.

The proposed mechanism underlying the therapeutic benefits of SGLT2i in AD is discussed in the Discussion section, particularly in Section 4.1.

3.

Give information about search filters and Boolean operators used in the study.

Thank you for the comment. We received a similar comment from the second reviewer. 

The complete search strings for each database have been included in the supplementary materials, and all other details related to the PRISMA 2020–compliant flow diagram are already presented in Figure 1.

4.

PROSPERO registration number (CRD420251061359) appears unusually long; verify its accuracy and confirm whether the protocol is accessible.

We have double checked the accessibility, and it is available under the following link.

https://www.crd.york.ac.uk/PROSPERO/view/CRD420251061359

The screenshot from the website is attached.

5.

Justify the criteria chosen for rodent selection; how these factors influence heterogeneity.

Thank you for this insightful comment. In our review, we did not predefine inclusion criteria based on rodent strain or model; instead, we included all preclinical studies that evaluated SGLT2 inhibitors in combined T2DM/AD or AD-like conditions using recognized mouse or rat models. This approach was chosen to capture the full range of available evidence and avoid an overly narrow synthesis at this early stage of the field.

Rats (mainly Wistar) and mice (C57BL/6, db/db, APP/PS1, Swiss) differ in baseline metabolic profile, genetic background, and susceptibility to cognitive impairment, while the various AD and T2DM induction methods (icv‑STZ, AlCl3, ovariectomy/ D‑galactose, scopolamine, cadmium, high‑fructose diet, STZ-induced diabetes) model distinct aspects and severities of pathology. These differences inevitably 6contribute to clinical and m7ethodological het8erogeneity by influencing the onset, magnitude, and pattern of neurodegeneration, metabolic disturbance, and behavioural outcomes. In the review, this heterogeneity was addressed by:

1)       acknowledged the heterogeneity in animal strain, model type, and induction method in detail in the characteristics table 1.

2)       We also have clarified in the Methods that all rodent strains and AD/T2DM induction protocols were eligible, and in the Discussion we explicitly acknowledge that variation in species, strain, and disease model is a key source of heterogeneity and a limitation that may partly explain differences in effect size across studies.

6.

Provide the Error margin or the calibration approach of WebPlotDigitizer used.

Thank you for the comment.

WebPlotDigitizer does not provide a fixed universal error margin; accuracy depends on image quality and proper axis

calibration. As part of the standard procedure, we calibrated each graph by assigning two reference points on both the x- and y-axes before extraction.

This validated approach (Drevon et al., 2017) typically yields errors <2%, and we confirmed consistency by using high-resolution figures and visually

checking extracted values against the original curves.

We have edited in the Manuscript Section 2.3 “WebPlotDigitizer was used to extract numerical values from graphs. Each figure was calibrated by assigning two reference points on both the x- and y-axes, which is a standard procedure to ensure accurate scaling; this approach

has been validated and typically yields extraction errors of &lt;2% (Drevon et al., 2017). A.H.M.R. and T.M.H. independently performed the data extraction

using a designated Microsoft Excel file.”

Drevon D, Fursa SR, Malcolm AL. Intercoder Reliability and Validity of WebPlotDigitizer in Extracting Graphed Data. Behav Modif. 2017

Mar;41(2):323-339. doi: 10.1177/0145445516673998.

7.

Clarification on the rationale of the two different doses used or pooled during meta-analysis

Thank you for pointing this out. To clarify, the 2 doses are specifically from Borikar et al., and Samman et al.,. that investigated 2 different doses of SGLT2i. By splitting them into 2 different group is to allow dose‑specific effects if clinically relevant.

In particular, it is expected different effects for low and high doses, and we want to estimate them separately, we treated each dose as its own comparison (low vs control, high vs control).

8.

The document states there are 6 figures, but the figure legends list 7. There is an inconsistency in the numbering. Verify it.

Thank you for the comment. We have double checked that there are only 6 figures and 6 figure legends.

9.

Revise figures, axis labels and units for uniform presentation.

Thank you for your comment. We have revised the figures and ensured that the labels and units are presented uniformly.

10.

Authors should explain the limitations observed during CAMARADES scoring and how it impacted overall conclusion.

Thank you for the comment. We have now explain the limitations observed during CAMARADES scoring and clarify how this might influence the overall conclusion.

“The methodological quality of the included animal studies was assessed using a risk of bias checklist adapted from The Collaborative Approach to Meta-Analysis and Review of Animal Data from Experimental Studies (CAMARADES) [39]. Each item in the tool was rated as Yes (Y) or No (N), with Yes scored as 1 and No scored as 0. The quality score for each study was calculated as the sum of Y responses across all items, and the mean quality score was then derived from the scores of all included studies

The papers demonstrated a moderate quality of methodology, with a mean quality score of 6.83 across all 12 studies (Table 2). All were published in peer‑reviewed journals and reported their adherence to relevant ethical guidelines. The studies avoided using anaesthetics that display intrinsic pharmacological activity, used appropriate animal models, and provided declarations of conflicts of interest.

Two studies did not report any temperature control as part of the husbandry conditions, a factor that may confound the assessment of animal behaviour [20,29]. All studies reported the absence of blinding during allocation, and only one study implemented blinded outcome assessment.[29] The lack of blinding in outcome assessment can in-troduce detection bias, which could have influenced the semi‑quantitative scoring of animal parameters. Furthermore, we rated almost all studies as ‘N’ for sample size calculation. This point indicates that the experiments could have been underpowered or overpowered, which could have led to imprecise estimates of effect sizes..”

11.

Does the author include a trim and fill analysis of the Eagers test as it indicated publication bias in multiple outcomes

We thank the reviewer for this comment. We have conducted a trim and fill analysis. The raw results is presented in the supplementary material. We have added the following text in section 3.7.

“Across the four behavioural outcomes, trim-and-fill analysis indicated varying degrees of small-study or publication bias. For escape latency T2DM (Figure 6A) and time spent in target quadrant T2DM (Figure 6B), the procedure either imputed very few studies or none at all, implying a minimal impact of publication bias. For escape latency AD (Figure 6C) and time spent in target quadrant AD (Figure 6D), the method imputed a few missing studies in the opposite direction of the observed effects, which means the magnitude of benefit in those domains could be slightly overestimated. Importantly, in all cases, the direction of the pooled effect remained unchanged after adjustment. This result indicates that some asymmetry is present, probably due to the small number of animal studies. Our overall conclusions regarding the beneficial effects of SGLT2 inhibitors were robust. The corresponding results are presented in Supplementary Material”

And in the Limitations section 4.6:

“Our Eagers test suggested the presence of publication bias. In other words, studies reporting significant positive outcomes may have been more likely to be published than negative outcomes. This bias could potentially compromise the overall validity and generalisability of the findings presented in this review. Finally, although trim-and-fill analysis was performed, the small number of studies means that the interpretability is limited.”

We have also updated the methods Section 2.5 with the following text: “Additionally, trim-and-fill analyses were performed in R to explore the potential impact of small study effects.

”

12.

The discussion sections should be strongly supported by existing literature and not speculations.

Thank you for the comments. We have revised the discussion section and changed the sentences that showed a certain level of ambiguity and added necessary references.

13.

References are duplicated, revise it

Thank you for the comment. We realized that the reference list for Borikar et al 2024 is duplicated and has now rectified this.

14.

Several grammatical errors and formatting issues need to be corrected

Thank you for your valuable comment. We have submitted the manuscript to a professional English editing service to improve the language quality and formatting. The certificate of proofreading is attached.

No

Comment from Reviewer

Reply from Authors

1

The meta-analysis for AD models reports very high heterogeneity (I² > 85%) for both escape latency and target quadrant time. Pooling these studies without deeper investigation may compromise interpretability.

We thank the reviewer for this suggestion. Due to the limited number of studies

available for each subgroup, performing subgroup analyses by model type, species, SGLT2 inhibitor type, dosage, or study duration was not feasible. However, we have applied a random-effects model throughout the meta-analysis to account for expected between-study heterogeneity, which is appropriate given the variability in

experimental models and study designs.

We have revised the Statistical Analysis

section (section 2.5) to the following:

“A meta-analysis was conducted using R (version 4.1.1) using the meta and metafor packages. Outcomes were treated as continuous data. Given the variability in experimental models and study designs, a random effects model was applied to pool

effect sizes across studies.”

2

The search method lacks full reproducibility. Include complete search strings for each database in the main text or Supplementary Materials. Provide dates of searches, number of duplicates removed, and full PRISMA 2020-compliant flow diagram.

Thank you for your comment. The complete search strings for each database have been included in the supplementary materials, and all other details related to the PRISMA 2020–compliant flow diagram are already presented in Figure 1.

3

Sample sizes, animal characteristics, and model induction methods are inconsistently presented (e.g., “3–8 animals” range). This weakens transparency for meta-analysis. 

Standardize Table 1 to include exact group sizes, sex, and strain. Clarify missing data or state “not reported.”

Thank you for your comment. We have double checked and corrected the number of animals. We have standardized the table to include exact group sizes, sex and strain of the animals tested. We also have clarified the missing data as not reported.

4

Forest plots and funnel plots are currently low-resolution and lack essential details. Provide high-resolution figures with all standard graphical elements (study weight, effect size models, heterogeneity statistics). Add clear axis labels and legends –

Thank you for your comment. We have now regenerated all forest and funnel plots in high resolution with publication-quality formatting. The updated figures include all standard graphical elements, including study weights, effect size estimates with 95% confidence intervals, and heterogeneity statistics. Clear axis labels, legends, and ‘Favours Control / Favours Intervention’ annotations have been included to improve readability. These updated figures replace the previous low-resolution versions in the revised manuscript (Figures 2–5). 

5

Egger’s test suggests bias, but the manuscript stops short of explaining its implications. Discuss sources of potential publication bias (small animal studies, positive-result bias, selective reporting). Explain how this may inflate effect estimates. –

We thank the reviewer for this important point. As a similar concern was also raised by Reviewer 1, we have addressed it comprehensively in the revised manuscript.

Specifically, we have now conducted trim-and-fill analyses for all four outcomes, summarised the findings in Section 3.7, and added the full results to the Supplementary Materials. We have also incorporated an explicit note in Section 4.6 acknowledging the limitations of interpreting trim-and-fill results in the context of a small evidence base. These revisions collectively clarify the extent and implications of potential publication bias in our analysis.

6

Values taken from figures using WebPlotDigitizer may introduce error. Discuss limitations associated with digitizing data. If possible, conduct sensitivity analyses excluding digitized studies.-

Thank you for the comment. As mentioned to the first reviewer, the use of WebPlotDigitizer indeed has limitations.

WebPlotDigitizer does not provide a fixed universal error margin; accuracy depends on image quality and proper axis calibration. As part of the standard procedure, we calibrated each graph by assigning two reference points on both the x- and y-axes before extraction.This validated approach (Drevon et al., 2017) typically yields errors <2%, and we confirmed consistency by using high-resolution figures and visually checking extracted values against the original curves.

We have edited in the Manuscript Section 2.3 “WebPlotDigitizer was used to extract numerical values from graphs. Each figure was calibrated by assigning two reference points on both the x- and y-axes, which is a standard procedure to ensure accurate scaling; this approach

has been validated and typically yields extraction errors of &lt;2% (Drevon et al., 2017). A.H.M.R. and T.M.H. independently performed the data extraction

using a designated Microsoft Excel file.”

7.

Several DOIs appear repeated three times, and some references are duplicated.

Thank you for the comment. We realized that reference list for Borikar et al 2024 is duplicated and has now rectify this.

8.

Inconsistent Terminology e.g “intracerebrovascular STZ” vs “icv-STZ”, “nootropic model” vs “memory model”,“Alzheimer’s like conditions” → “Alzheimer’s-like conditions, Standardize terminology throughout.

Thank you for the comment. We have revised the manuscript to consistently using the suggested words.

9.

Table 1 contains missing abbreviations e.g., “8OHDG” only defined later

It was defined at the bottom of the table.

10.

The PROSPERO ID appears unusual (CRD420251061359). Verify accuracy and ensure the registration is publicly accessible.

Thaml you for the comment. We received a similar comment from reviewer 1.

We have double checked the accessibility, and it is available under the following link.

https://www.crd.york.ac.uk/PROSPERO/view/CRD420251061359

The screenshot from the website is attached.