Identification phase

A systematic search was carried out across four databases (PubMed, Scopus, Google Scholar, and Europe PMC) using these six keyword combinations:

• 1.

  Phytoconstituents AND Diabetes Mellitus AND Molecular Mechanisms

• 2.

  Plant-derived Compounds AND Antidiabetic Activity AND Signal Transduction

• 3.

  Herbal Medicine AND Diabetes Management AND Cellular Pathways

• 4.

  Natural Products AND Insulin Resistance AND Gene Expression

• 5.

  Botanical Extracts AND Glucose Metabolism AND Therapeutic Targets

• 6.

  Phytochemicals AND Diabetes Therapy AND Inflammatory Pathways

The search results are broken down as follows: PubMed (893 identified, 871 after duplicates removal), Scopus (329 identified, 325 after duplicates removal), Google Scholar (1,000 identified, 798 after duplicates removal), and Europe PMC (1,000 identified, 821 after duplicates removal). The overall workflow [13] of study identification, screening, eligibility, and inclusion is summarized in Figure 1. All studies summarized in Table S1 are collectively cited here for numerical continuity [14–190].

Display full size

Figure 1. PRISMA flow diagram showing the selection of studies. Adapted from [13]. © 2019 The Authors. Licensed under a CC BY 4.0.

Screening phase

Eligibility phase

Studies that used experimental models explicitly created to research diabetes were included in this review. In vivo models that met the criteria were genetically diabetic db/db mice, mice induced by a high-fat diet (HFD), and rats induced by streptozotocin (STZ). For in vitro research, only papers that modeled diabetes circumstances utilizing insulin-resistant adipocytes, hepatic cells, or pancreatic β-cells were included. In silico research was considered if it targeted diabetes-related proteins, such as protein tyrosine phosphatase 1B (PTP1B) or peroxisome proliferator-activated receptor gamma (PPARγ), using molecular docking or simulations. Only English-language research papers published after 2015 were considered for this analysis. Articles were excluded due to the following factors: those published before 2015, inaccessible because of paywalls or lack of author response, involving non-diabetic models or unrelated conditions, or review articles or editorials. All retrieved full-text articles were independently screened by two reviewers, and any disagreements were resolved through discussion to ensure uniform application of the eligibility requirements.

Data extraction variables

Data extraction was guided by predefined variables, including plant source, study type, diabetes model, molecular target, and key outcomes (Table 1). Across the 177 included studies, 92 (52%) were in vitro, 66 (37.3%) in vivo, 15 (8.5%) in silico, and 4 (2.3%) others [hybrid: human trial, DIA proteomics, high-content screening (HCS)]. The majority of the investigated phytoconstituents were derived from medicinal plants traditionally associated with antidiabetic activity, notably polyphenols, alkaloids, and flavonoids, which together accounted for over half of all reported compounds. This distribution indicates a prevailing emphasis on in vivo validation and molecular mechanisms involving antioxidant and insulin-sensitizing pathways.

A thematic synthesis approach was used to categorize extracted data by pathway and model type, as shown in Table 1. Data synthesis was conducted qualitatively using a thematic framework approach. The extracted data were first coded into key themes, including pathway targeted, study model, biomarker outcomes, and therapeutic effects. These themes were then compared across studies to identify recurring mechanistic patterns, convergence of therapeutic outcomes, and cross-validation between in vitro, in vivo, and in silico designs. This process enabled an integrative narrative synthesis, which was chosen over meta-analysis due to the methodological and outcome heterogeneity across included studies.

Inclusion phase

A total of 177 studies were incorporated into the qualitative synthesis.

SYRCLE’s risk of bias (ROB) tool for in vivo investigations is used for quality assessment of in vivo investigations. In vitro investigations were assessed using the ToxRTool, which evaluates study quality based on test substance characterization, test system description, study design, and data reporting.

The combined results were arranged according to the molecular pathway (e.g., PI3K/AKT: 44.6%, GLUTs: 19.8%, AMPK: 14.1%, others: 21.5%).

Study selection and data extraction process

Four reviewers participated in this systematic review, each with a clearly defined role throughout the process. The study selection process was conducted collaboratively by two reviewers (Reviewers A and B) who jointly performed the systematic search across all four databases (each person working on two databases), downloaded articles, removed duplicates, and conducted title and abstract screening. All screening decisions during this phase were made by consensus among these two reviewers to ensure consistent application of inclusion and exclusion criteria. Following the completion of study selection, data extraction was performed independently by the remaining two reviewers (Reviewers C and D) using the standardized extraction form to ensure consistency in captured variables. Reviewer C extracted data from 78 studies while Reviewer D extracted data from 99 studies, totaling the 177 included studies. Each reviewer was responsible for removing all relevant variables from their assigned studies, including plant/phytoconstituent information, study type, diabetes model used, molecular targets, and key outcomes.

Methodological rigor

To ensure methodological rigor and compliance with the PRISMA 2020 guidelines, this review was prospectively registered in PROSPERO (CRD420251073083) before the screening process began. The search strategy, inclusion and exclusion criteria, and data extraction framework were defined before registration, while data analysis and synthesis were conducted afterward. The studies were carefully examined and chosen by four reviewers. To verify the reproducibility of the screening process, a sensitivity check was performed by randomly selecting 20% of the full-text articles for independent screening by two reviewers. Inter-rater reliability was then quantified using Cohen’s kappa (κ) to measure the level of agreement between reviewers. Agreement was substantial across key domains, with κ = 0.81 for study type classification, 0.78 for molecular target identification, and 0.85 for outcome classification. Disagreements were resolved by consensus through discussion between the two reviewers, and a third senior reviewer adjudicated unresolved cases. This approach ensured consistent and reliable data capture while maintaining the efficiency of the collaborative process. Table S1 contains the complete search strategy for replication, and a tabular summary of exclusion grounds is provided for transparency. Furthermore, 85% of the included studies used mammalian models, enhancing the clinical relevance and translational validity of the findings. Consistent dosage reporting (in mg/kg) across in vivo studies enabled insightful comparisons across different experimental setups.

For data management and reference handling, we used Publish or Perish, Microsoft Excel, and EndNote. Publish or Perish was used for bibliometric retrieval and citation analysis; EndNote for structured referencing and citation management; and Excel for organizing extracted variables, tabulating study characteristics, and generating descriptive statistics.

Molecular pathways and targets

The studies reviewed targeted key molecular mechanisms involved in diabetes pathogenesis collectively; the most molecular pathways and targets are PI3K/AKT signaling (44.6%) of the studies, GLUTs (19.8%), AMPK activation (14.1%), and other pathways (α-glucosidase inhibition, PPAR modulation, antioxidant/ROS regulation, nuclear factor kappa-light-chain-enhancer of activated B cells (NF-κB), dipeptidyl peptidase (DPP)-4 inhibition, insulin receptor modulation) were reported across 38 studies, which cover about 21.5% of the studies.

Table 2 presents representative biomarkers modulated within each pathway. PI3K/AKT had the highest count (79 studies), followed by GLUT transporters (35 studies) and AMPK (25 studies). Biomarker-level findings highlight mechanistic plausibility, including insulin receptor substrate 1 (IRS-1) and GLUT4 for PI3K/AKT, acetyl-CoA carboxylase/PPARγ coactivator 1-alpha (ACC/PGC-1α) for AMPK, and GLUT2/GLUT4/SGLT2 for GLUT transporters.

The summarized molecular targets and biomarkers (Table 2) highlight the predominance of PI3K/AKT, GLUT, and AMPK signaling in phytoconstituent research on diabetes. These pathways (AMPK, PI3K/AKT, and GLUT) were prioritized because they represent critical molecular nodes in glucose homeostasis and insulin signaling. AMPK is a master regulator of cellular energy metabolism, enhancing glucose uptake and fatty acid oxidation. The PI3K/AKT pathway is the canonical insulin signaling cascade, essential for GLUT4 translocation and pancreatic β-cell survival. GLUT transporters, particularly GLUT2 and GLUT4, directly mediate cellular glucose uptake. Dysregulation of these three mechanisms is central to the pathophysiology of diabetes, making them highly relevant therapeutic targets. Their combined modulation offers strong mechanistic plausibility for phytoconstituents as multi-target antidiabetic agents.

Risk of bias assessment

The overall distribution of ROB ratings (low, moderate, high) is presented in Table 3 and visualized in Figure 2.

Display full size

Figure 2. ROB chart. ROB: risk of bias.

Figure 2 presents the distribution of ROB ratings (low, moderate, high), showing that 68.93% of the studies carried a moderate risk, 29.94% a low risk, and 1.13% a high risk.

Trends in publication year

From this research, we observed a progressive increase in publications on plant-based therapies for diabetes over 11 years. The statistics are as follows: 2015–2019 (47 studies), 2020–2022 (53 studies), 2023–2025 (77 studies). However, 2024 had the highest number of publications (40 studies), indicating a recent increase in interest in plant-based therapies for diabetes (Table 4).

Study selection and characteristics

Articles published from 2015 to 2025 were used for the research, of which a total of 177 articles met the inclusion criteria for the systematic review. These articles were chosen for the systematic review using systematic screening procedures during the study identification and selection process. The included studies have different types of research designs and approaches (experimental methodology), the study types include in vivo found in about 66 studies covering 37.3% of the studies, in vitro (n = 92, 52%), in silico (n = 15, 8.5%), and other hybrid methods, which include human trial, computational, DIA proteomics, HCS (n = 4, 2.3%). Due to rounding, the sum of various percentages may not equal 100% (Table 5). These studies have their origins in multiple research groups that use various models and analytical techniques to investigate and analyze the effects of phytoconstituents on diabetes (Figure 1). Table S1 lists the studies’ authors, year, title, plant/phytochemical, study type, diabetes model used, molecular target pathway, main findings, outcome, notes, diabetes model category, ROB, and tools used.

Relationship between study type and pathway focus

Cross-tabulation suggests that, focusing on the relationship between study type and molecular pathway, the PI3K/AKT and AMPK pathways were most investigated using combined in vitro and in vivo designs. In contrast, modulation of GLUT activity was more evenly distributed across in vitro and in vivo studies. Additionally, enzymatic inhibition (e.g., α-glucosidase, DPP-4) was predominantly explored in vitro or in silico. In contrast, computational models were primarily used for molecular docking, absorption, distribution, metabolism, excretion, and toxicity (ADMET) profiling, and virtual screening of phytoconstituents. This relationship complements the molecular evidence summarized in Table 2, confirming that hybrid models primarily investigated PI3K/AKT and AMPK mechanisms.

An experimental diabetes model was used

Various study models were used in the studies to investigate the phytoconstituents activity, the study models used includes; rodent models only (n = 85, 48.02%), cell line models only (n = 26, 14.69%), combined rodent and cell line models (n = 28, 15.82%), in silico computational models (n = 14, 7.91%), and others (e.g., zebrafish, organ-specific ex vivo system) (n = 23, 13.00%) (Table 6). This distribution demonstrates a firm reliance on whole-animal testing, which is supported by in vitro and in silico mechanistic studies.

Therapeutic outcomes observed

From the articles reviewed, therapeutic outcomes were observed. The outcomes shown were heterogeneous. The most frequently observed therapeutic effects involve reduced glucose levels (n = 49), anti-inflammatory effects (n = 20), antioxidant activity (n = 10), improved insulin sensitivity (n = 7), improved lipid profile (n = 5), and other outcomes (renal function, pancreatic protection, cognitive improvements, hormonal regulation, etc.) were observed in 86 studies.

Synthesis of molecular-based therapeutic trends

An integrative review of the data revealed a strong alignment between pathway targeting and therapeutic outcomes, consistent improvements in glucose homeostasis across multiple models and compounds, and a high rate of insulin signaling modulation, suggesting a potential for mechanistic synergy and an emerging preference for hybrid study designs that support deeper validation of bioactivity.

Core study findings

The core study findings from the studies include that the most common study design found from the studies was in vitro having found in 52% of the studies, the most targeted pathway was the PI3K/AKT found in 44.6% of the studies, the top therapeutic outcome was glucose reduction in about 27.7% of the studies, the leading study year was 2024 having a total number of 40 articles written in the year. The most used model type was the rodent model (e.g., rats, mice), accounting for 48.02% of the studies.

Limitations of the review

Despite using a comprehensive search approach and adhering to PRISMA principles, this review has limitations. The restriction on English-language publications may have led to language prejudice. The process and scope of article inclusion are visually summarized in Figure 1 to ensure transparency. Narrative synthesis is inherently more susceptible to interpretive bias than meta-analysis, due to significant variation in experimental paradigms, outcomes, and substances. Furthermore, although we assessed the ROB in the included studies, selection bias may still be present because, due to resource constraints, our own screening and data extraction process did not employ full, dual-independent screening at every stage, despite being conducted with cross-checking and consensus.

Future perspectives

It is necessary to close the indicated translational gap to utilize phytoconstituents for the treatment of diabetes. Future research should concentrate on:

• 1.

  The implementation of proven protocols for the extraction, characterization, and standardization of plant extracts is necessary to ensure repeatability and precise dosing. Finding isolated active principles or producing standardized extracts (with known flag molecules) is a significant scientific and regulatory conundrum.

• 2.

  Mechanistic rigour: Examining causal linkages using specific pharmacological inhibitors or genetic knockout models to confirm the involvement of proposed pathways, going beyond correlational observations.

• 3.

  Clinical translation is the process of developing closely watched early-stage clinical studies that verify pre-clinical findings in humans using mechanistic biomarkers (e.g., assessing pathway activation in patient samples). The successes and failures of earlier clinical studies of better-known diabetic herbs should serve as guidance for these investigations.

• 4.

  Solutions for bioavailability: New delivery strategies (such as nanoparticles and phospholipid complexes) are being researched in an effort to solve the limited bioavailability that plagues many otherwise promising phytoconstituents.

• 5.

  Integrated methods: In vitro and in vivo models are being utilized in conjunction with in silico predictions to efficiently find and evaluate the most promising lead drugs with good ADMET profiles.

This comprehensive review, which focuses on AMPK, PI3K/AKT, and GLUT as the primary mechanisms of action, concludes by combining compelling preclinical evidence demonstrating that phytoconstituents significantly modify key pathways associated with diabetes. Their pleiotropic effects align with the complex nature of diabetes. However, the transition from promising pre-clinical data to clinical application is hampered by methodological errors, bioavailability issues, and a lack of human studies. Addressing these problems through systematic, rigorous, and translational research is necessary to fully realize the medicinal potential of the plant kingdom in the global fight against diabetes.

In conclusion, this systematic review integrates a wealth of pre-clinical evidence demonstrating that phytoconstituents effectively ameliorate diabetes symptoms by targeted modification of key biochemical pathways, including PI3K/AKT, GLUT, and AMPK signaling. Mechanistically coherent and consistently documented are improvements in insulin sensitivity and glucose homeostasis, and reductions in oxidative stress and inflammation across various experimental paradigms. Unfortunately, the translation of this high pre-clinical promise into clinical practice is significantly limited by a severe shortage of human trials, significant methodological heterogeneity, and a modest ROB in current investigations. We are now at a pivotal moment in the field. Future studies should focus on standardizing phytoconstituent extraction and characterization, developing mechanistic evidence, and conducting meticulously organized clinical trials to validate these preclinical mechanisms in human subjects. These problems can be addressed by carefully evaluating the substantial therapeutic potential of plant-derived chemicals and applying them to develop novel, multi-targeted strategies for the worldwide management of diabetes mellitus. This conclusion is based on integrated results across experimental models and pathways (Tables 1–3, Figures 1 and 2).