Efficacy of Chemical Chaperones (TUDCA, 4-PBA) in Restoring Glucose-Stimulated Insulin Secretion and Mitigating Endoplasmic Reticulum Stress in Pancreatic Beta-Cells: A Systematic Review and Bayesian Meta-Analysis

1. Introduction

The relentless global rise of type 2 diabetes mellitus (T2DM) represents one of the most severe public-health and socioeconomic challenges of modern medicine, characterized by peripheral insulin resistance and progressive pancreatic beta-cell functional failure [1]. Initially, the endocrine pancreas compensates for diminished peripheral insulin sensitivity by hyperactivating its synthetic and exocytotic machinery, maintaining normoglycemia at the expense of chronic hyperinsulinemia [2]. However, persistent exposure to elevated blood glucose and circulating non-esterified free fatty acids, a pathological state termed “glucolipotoxicity,” imposes an unsustainable synthetic burden on beta-cells [3]. This burden is associated with ER redox imbalance, delayed ER export, accumulation and misfolding of proinsulin, and beta-cell dysfunction [4].

The ER is the primary site for structural folding, disulfide-bond formation, and quaternary assembly of secretable proteins [5]. Pancreatic beta-cells are highly specialized for insulin production, but chronic glucolipotoxicity increases the misfolded-protein burden in the ER, elevating BiP/GRP78 and other chaperones and robustly activating all three branches of the UPR. Initially, this response aims to reduce protein load and enhance folding; however, with persistent glucolipotoxicity, the UPR shifts toward pro-apoptotic outputs (CHOP, JNK, mitochondrial apoptosis), contributing to beta-cell failure [6,7].

In its physiological phase, the UPR initiates an adaptive feedback loop to restore proteostasis. Following dissociation from BiP, PERK oligomerizes and autophosphorylates, phosphorylating the α-subunit of eukaryotic initiation factor 2 (eIF2α) and inducing a transient global attenuation of translation [1]. Concurrently, activated IRE1α displays endoribonuclease activity, processing Xbp1 mRNA to produce splicing-active XBP-1s, which transcriptionally upregulates ER chaperones and ER-associated degradation components [1]. Simultaneously, ATF6 translocates to the Golgi apparatus for proteolytic cleavage, releasing its active cytosolic domain to act as a pro-survival transcription factor [1].

When ER stress is chronic, severe, and unresolved, the UPR transitions from a pro-survival program to a terminal pro-apoptotic cascade [7]. Prolonged phosphorylation of eIF2α selectively promotes translation of activating transcription factor 4 (ATF4), which upregulates transcription of Ddit3, encoding C/EBP-homologous protein (CHOP) [1]. Nuclear CHOP accumulation suppresses anti-apoptotic Bcl2 and activates pro-apoptotic Bax, Bim, and p53, triggering mitochondrial outer-membrane permeabilization, cytochrome-c release, and caspase-3-dependent apoptosis [5]. Crucially, before cell death is finalized, the loss of proteostasis drives beta-cells into functional silence and epigenetic dedifferentiation, downregulating identity genes such as PDX-1, MAFA, and NKX6.1 and silencing the calcium-dependent exocytotic machinery, thereby abolishing glucose-stimulated insulin secretion (GSIS) [8].

To preserve exocytotic identity, pharmacological strategies utilizing low-molecular-weight chemical chaperones have gained significant interest [9]. TUDCA and 4-PBA act as molecular scaffolds, stabilizing protein conformation, accelerating folding kinetics, and reducing the burden of ER-misfolded proteins. Although preclinical evidence suggests these agents are beneficial, the literature is characterized by small-sample studies and methodological noise, hindering quantitative synthesis. To address these limitations, we conducted a PRISMA-compliant systematic review and a hierarchical Bayesian meta-analysis of preclinical studies assessing the efficacy of TUDCA and 4-PBA in restoring GSIS and mitigating terminal ER stress under metabolic glucolipotoxicity (PROSPERO CRD420261370436).

2. Results

2.1. Study Selection

The systematic search of PubMed, Scopus, and Web of Science (January 2016–May 2026) returned 1,663 records. After removal of 8 duplicates, 1,655 records were screened by title and abstract, of which 1,650 were excluded against the PICOS criteria (1,353 for wrong intervention, 185 for wrong population or tissue, 70 for wrong comparator or stressor, and 42 for wrong study design or outcomes). Five full-text articles were assessed for eligibility; one was excluded because its primary outcome concerned GLP-1-receptor signaling rather than the GSIS stimulation index or CHOP mitigation. Consequently, four preclinical experimental studies were included in the qualitative and quantitative syntheses [5,10,13,14]. The selection workflow is summarized in the PRISMA flow diagram (Figure 1).

2.2. Qualitative Synthesis and Risk of Bias

Application of the SYRCLE tool indicated a low-to-moderate risk of bias across the four included studies. Selection bias was judged low because the in vivo studies reported random allocation to diet and treatment arms. Attrition and reporting biases were low, with complete datasets reported for the primary variables. Detection bias was rated unclear because explicit blinding of outcome assessors during GSIS assays and Western-blot densitometry was not comprehensively documented. The study characteristics and risk-of-bias judgments are summarized in Table 1. The models ranged from in vitro immortalized cell lines (INS-1, βTC-6) to ex vivo isolated islets from in vivo rodents fed high-fat or high-fat/high-fructose diets [10,13].

2.3. Primary Outcome: Quantitative Restoration of the GSIS Stimulation Index

Chronic overnutrition and lipotoxic insults impaired or dysregulated GSIS across the dietary and lipotoxic models. In rodents fed an HFD [10] or an HFHFD [13], islets exhibited severe impairment of exocytotic capacity in response to high-glucose stimulation (16.7 mM). This defect was mirrored in vitro after cholesterol loading of INS-1 and βTC-6 cells [5]. By contrast, in mice subjected to protein restriction followed by HFD, the islets developed glucose-stimulated insulin hypersecretion, which TUDCA normalized [14]. The per-study dosing regimens, high-glucose GSIS values, and resulting effect sizes are summarized in Table 2.

Dosing regimens, group sizes (n), glucose concentrations, and stressor protocols are text-reported in the primary publications and are verifiable. GSIS means ± SEM (†) were extracted from the source figures (Binayi Figure 5B; Izadi Figure 8A; Kong Figure 4K; dos Reis Araujo Figure 3K) and are digitized approximations. GSIS was assessed at high-glucose stimulation (16.7 mM for Binayi and Izadi; 25 mM for Kong; 11.1 mM for dos Reis Araujo). Units: ng/10 islets/90 min (Binayi, Izadi); ng/mg protein (Kong); ng/islet/h (dos Reis Araujo). ‡ In the malnutrition-plus-HFD model, the stressor-induced hypersecretion; the restoration ratio is oriented toward the healthy-control value so that normalization is positive on the same scale as the suppression models. The pooled estimate is the posterior mean (95% CrI) from the hierarchical Bayesian random-effects model.

At the level of individual studies, the rescue was substantial and directionally consistent (Table 2; Figure 2). In the chronic HFD model, islet insulin output under 16.7 mM glucose was approximately 120 ± 14 ng/10 islets/90 min under the stressor and rose to approximately 270 ± 18 ng/10 islets/90 min after 4-PBA, a 2.25-fold restoration (95% CI 1.73 to 2.93) [10]. In the HFHFD model, secretion increased from approximately 115 ± 11 to 190 ± 13 ng/10 islets/90 min with 4-PBA (1.65-fold; 95% CI 1.31 to 2.08) [13]. In cholesterol-loaded βTC-6 and INS-1 cells, glucose-stimulated secretion recovered from approximately 95 ± 12 to 175 ± 15 ng/mg protein after 4-PBA pre-treatment (1.84-fold; 95% CI 1.37 to 2.48) [5]. In the malnutrition-plus-HFD model, where the stressor instead produced hypersecretion, TUDCA normalized output at 11.1 mM glucose from approximately 0.90 ± 0.09 to 0.52 ± 0.06 ng/islet/h, a 1.73-fold movement toward the healthy-control value (95% CI 1.28 to 2.33) [14]. These per-study GSIS values are approximate, having been digitized from the source figures, whereas the chaperone doses, routes, schedules, and group sizes listed in Table 2 are taken directly from the primary reports.

Pharmacological administration of 4-PBA or TUDCA systematically restored exocytotic responsiveness. The hierarchical Bayesian model yielded a pooled posterior mean restoration ratio of μ = 1.87 (95% credible interval [CrI]: 1.39 to 2.46), indicating a substantial positive effect of chemical chaperones on the GSIS stimulation index across all experimental platforms (Table 2; Figure 2 and Figure 3). The entire credible interval lay above the no-effect value of 1, corresponding to a posterior probability of benefit exceeding 0.99. Between-study heterogeneity was modest (posterior median τ² = 0.02; posterior mean 0.07; 95% CrI 0.00 to 0.45), indicating that, despite divergent biological models (rodent islets versus immortalized cell lines) and stress protocols, the functional exocytotic rescue was reproducible in both magnitude and direction (Figure 3). The width and lower bound of the τ² interval reflect the limited precision attainable with only four studies, as discussed in Section 4.4.

2.4. Secondary Outcome: Expression and Mitigation of BiP and CHOP

The principal mechanism underlying preservation of GSIS was direct suppression of terminal ER stress signaling [10]. Western blot analyses demonstrated that glucolipotoxicity induced a marked upregulation of BiP/GRP78, ATF4, and the terminal pro-apoptotic transcription factor CHOP [5]. Co-treatment with 4-PBA or TUDCA significantly blunted this response [10]. In the pancreas of HFD-fed rats, 4-PBA decreased BiP and CHOP levels and restored the ER-membrane glycoprotein Wolfram syndrome 1 (WFS1) [10]. In the malnutrition-plus-HFD model, BiP and CHOP were not directly quantified; instead, TUDCA normalized the glutamate-dehydrogenase-driven islet insulin hypersecretion and improved peripheral insulin sensitivity [14].

2.5. Molecular Mechanism of 4-PBA: Structural Scaffolding and WFS1 Stabilization

4-PBA acts predominantly by binding hydrophobic regions of unfolded or partially folded proteins in the ER lumen, preventing nonspecific aggregation and facilitating structural maturation and transit through the secretory pathway [5]. Under overnutrition, the rapid influx of nascent proinsulin exceeds BiP’s folding capacity [13]; 4-PBA reduces the thermodynamic energy required for proinsulin folding, alleviating ER congestion [5]. Crucially, this reduction in luminal stress stabilizes WFS1 [10]. Under chronic glucolipotoxic stress, WFS1 is targeted for rapid degradation by the E3 ubiquitin ligase Smurf1 [18]; by preventing UPR hyperactivation, 4-PBA suppresses Smurf1-mediated WFS1 degradation [18]. Because WFS1 maintains ER calcium homeostasis, its stabilization prevents excessive calcium leak into the cytosol and preserves the calcium-regulated exocytotic machinery [10].

2.6. Molecular Mechanism of TUDCA: Receptor-Mediated Signaling

In contrast to 4-PBA, TUDCA is a hydrophilic bile-acid conjugate with a dual mechanism, combining physical chaperone activity in the ER with cell-surface receptor-mediated signaling [15]. TUDCA acts as a high-affinity agonist of the G protein-coupled bile acid receptor TGR5 on the beta-cell membrane, coupling to Gαs, activating adenylyl cyclase, and triggering a localized rise in cyclic AMP [15]. Elevated cAMP activates protein kinase A (PKA), which phosphorylates CREB [15] and potentiates the late, Ca²⁺-dependent steps of insulin-granule exocytosis (docking, priming, and SNARE-mediated fusion) [15,19]. This potentiation is glucose-dependent: at basal glucose (2.8 mM), TUDCA does not stimulate secretion, thereby avoiding hypoglycemia; at stimulatory glucose (11.1–22.2 mM), it dose-dependently potentiates secretion and shifts the glucose-response curve leftward [15].

2.7. Downstream Systemic and Peripheral Metabolic Repercussions

The benefits of chemical-chaperone therapy extend beyond the pancreas to systemic lipid and glucose metabolism [2]. In streptozotocin-induced diabetic models, TUDCA (300 mg/kg i.p. daily) reduced blood glucose, HbA1c, and HOMA-IR, increased serum insulin and GLP-1, and decreased the serum levels and activity of ceramide synthase, thereby limiting the accumulation of lipotoxic ceramides [16]. In prediabetic-aged mice, TUDCA attenuated hyperinsulinemia by restoring hepatic insulin clearance: aging downregulates hepatic insulin-degrading enzyme (IDE), and TUDCA upregulated hepatic IDE through S1PR2/Akt signaling, normalizing circulating insulin and restoring peripheral insulin sensitivity [2]. This normalization was accompanied by reduced visceral adiposity, lower hepatocyte triglyceride content, increased whole-body energy expenditure, and reversal of age-related cognitive decline [2]. In offspring exposed to an HFHFD from birth to young adulthood, 4-PBA improved insulin resistance (lower HOMA-IR, higher QUICKI), reduced pancreatic and hypothalamic malondialdehyde, downregulated BiP and CHOP, and lowered plasma leptin, although the diet-induced fall in HDL and rise in LDL persisted [13].

3. Discussion

The integrated preclinical evidence indicates that chemical chaperones are a powerful tool to arrest beta-cell functional decline in T2DM. Rather than merely delaying cell death, TUDCA and 4-PBA preserve exocytotic identity by maintaining the ER folding environment during chronic glucolipotoxic stress.

The principal finding of this synthesis is quantitative: across four mechanistically distinct preclinical models, co-administration of TUDCA or 4-PBA produced a large and directionally consistent restoration of the GSIS stimulation index, corresponding to a pooled Bayesian posterior mean of 1.87 (95% CrI: 1.39 to 2.46) on the restoration-ratio scale (Figure 2 and Figure 3). Because the entire credible interval lies above the null, the posterior probability that chemical chaperones improve secretory function under glucolipotoxic stress exceeds 0.99. Equally informative is the modest between-study heterogeneity (posterior median τ² = 0.02): despite divergent stressors (chronic high-fat feeding, high-fat/high-fructose feeding, cholesterol loading, and protein restriction followed by high-fat feeding) and platforms (in vivo and ex vivo islets and immortalized beta-cell lines), the secretory rescue was reproducible. This convergence suggests that the benefit is driven by a shared upstream mechanism—relief of ER proteostatic stress—rather than by model-specific idiosyncrasies, reinforcing the biological plausibility of the pooled estimate despite the small evidence base [10,13].

Two features distinguish the present work. First, rather than cataloguing chaperone effects qualitatively, it integrates them within a single hierarchical Bayesian framework that propagates both within- and between-study uncertainty, yielding an interpretable probabilistic estimate well suited to the small, noisy datasets characteristic of basic beta-cell research. Second, it reconciles two mechanistically separable modes of action—the purely physical scaffolding of 4-PBA, which prevents proinsulin aggregation and stabilizes WFS1 against Smurf1-mediated degradation [18], and the dual chaperone-plus-receptor activity of TUDCA, which additionally engages TGR5/cAMP/PKA signaling to potentiate the SNARE-dependent late steps of granule exocytosis [15,19]—within a single account of how ER proteostasis is coupled to secretory identity. The translational relevance is heightened by the clinical maturity of both agents: 4-PBA (sodium phenylbutyrate) is an approved chemical chaperone for urea-cycle disorders with a well-characterized human safety profile [21], and TUDCA has been administered safely to insulin-resistant adults at a dose (1.75 g/day) closely matching the allometrically modeled effective dose [17].

3.1. Concordance with the Broader Preclinical Literature

Although only four studies met the registered analytic window and PICOS criteria for quantitative pooling, the direction and magnitude of the pooled effect are consistent with a broader preclinical literature that was not eligible for meta-analysis. 4-PBA restored palmitate-impaired GSIS in beta-cells [23], and TUDCA improved islet function by reducing ER stress [24]; chaperone treatment also ameliorated beta-cell dysfunction in islet-amyloid-polypeptide overexpression models [25] and preserved the UPR to protect against autoimmune (type 1) diabetes in NOD mice [26]. Early human data are likewise supportive: oral sodium phenylbutyrate partially alleviated lipid-induced beta-cell dysfunction in adults during clamp studies [27]. These reports were excluded from the quantitative synthesis because they fell outside the January 2016 analytic window or employed stressors (amyloid proteotoxicity, autoimmune cytokine attack) or designs (human clamp studies) that did not meet the registered PICOS criteria; they therefore inform the biological context and external consistency of the pooled estimate rather than the estimate itself.

3.2. Cellular Proteostasis and Systemic Metabolic Crosstalk

Organized by physiological domain, Table 3 shows that the two chaperones share a common glucose-normalizing action but differ in breadth. For glycemic control, TUDCA reversed fasting hyperglycemia, lowered HbA1c, and restored HOMA-IR and QUICKI toward control values [16], whereas 4-PBA lowered fasting glucose, reduced HOMA-IR, and raised QUICKI in the HFHFD-offspring model [13]. Both improved islet function, TUDCA restored glucose-dependent secretion and raised GLP-1 [16], and 4-PBA upregulated insulin mRNA and intracellular insulin content [13], consistent with the secretory rescue quantified in Section 3.3. Their lipid effects diverged in magnitude: TUDCA markedly lowered total cholesterol, triglycerides, LDL-C, and VLDL-C while raising HDL-C [16], whereas under 4-PBA, the diet-induced dyslipidemia persisted and was only partially corrected [13]. Both restored redox balance and dampened inflammatory and apoptotic signaling, TUDCA suppressing TNF-α, IL-6, IL-1β, and PGE2 and blocking the intrinsic apoptotic cascade [16], and 4-PBA reducing CHOP-mediated apoptosis and preventing Smurf1-dependent WFS1 degradation [13,18], directly linking the systemic readouts to the ER-proteostatic mechanism of Section 3.4 and Section 3.5. Collectively, these data position 4-PBA as a predominantly proteostatic, β-oxidation-promoting agent and TUDCA as a broader metabolic and neuroprotective modulator, while both converge on improved glycemic control.

4.3. Allometric Calibration and Clinical Translation

3.3. Allometric Calibration and Clinical Translation

To translate preclinical efficacy into clinical therapy, allometric dose calibration is required. In the included rodent trials, the standard intraperitoneal TUDCA dose was 300 mg/kg/day [2,14]. Using the body-surface-area conversion HED = animal dose × (Kₘ,animal / Kₘ,human), with Kₘ = 3 for the mouse and 37 for the adult human [2], a 300 mg/kg mouse dose corresponds to a human-equivalent dose of approximately 24.3 mg/kg/day, or about 1.7 g/day for a 70-kg adult. This estimate is closely consistent with the dose already shown to be safe and metabolically active in humans: obese, insulin-resistant adults received 1.75 g/day of TUDCA for four weeks without adverse events and with improved hepatic and muscle insulin sensitivity [17]. This convergence supports the translational plausibility of the modeled dose, although it modestly exceeds the 10–20 mg/kg/day range licensed for cholestatic liver disease [2]. Controlled clinical trials are required to confirm whether structural beta-cell preservation and restored insulin clearance observed in animals translate to patients with T2DM.

3.4. Limitations and Future Outlook

Several limitations temper these conclusions. First, the quantitative synthesis rests on only four eligible studies; estimating a between-study variance (τ²) from so few studies yields a heterogeneity estimate of limited precision whose credible interval is sensitive to the choice of prior, so the pooled Bayesian estimate should be interpreted as exploratory rather than definitive. A second, related concern is directional heterogeneity among the included models. The four eligible studies do not represent a single secretory phenotype: three reproduced glucolipotoxic suppression of secretion, in which TUDCA or 4-PBA raised a depressed stimulation index [5,10,13], whereas the model of protein-restriction–programmed obesity reproduced insulin hypersecretion, in which TUDCA lowered an exaggerated secretory output toward the control level [14]. Expressing all four on a common restoration-ratio scale oriented toward the healthy-control phenotype (Section 2.4) therefore aggregates physiologically opposite secretory adjustments, and a confirmatory leave-one-out sensitivity analysis—omitting the hypersecretion model—is recommended to verify that the pooled estimate is not driven by this directional asymmetry. This use of a between-arm restoration ratio in place of the within-arm GSIS stimulation index named in the registered protocol, together with the inclusion of BiP alongside the protocol’s pre-specified CHOP as a secondary endoplasmic-reticulum-stress marker, constitute deviations from the registration that should be read together with the operational definition given in Section 2.4. Second, the primary effect sizes were derived from figure digitization because the GSIS values were reported graphically rather than numerically in the source papers. Although digitization was performed with validated software, residual extraction error cannot be excluded, and no GSIS values were fabricated where source data were unavailable. Third, the a priori restriction of the search to January 2016–May 2026 excluded concordant earlier studies (e.g., [23,24]) and studies employing non-glucolipotoxic stressors; these were retained only as qualitative context (Section 4.1) and did not contribute to the pooled estimate, but their exclusion narrows the evidence base relative to the full historical literature. Fourth, as identified by the SYRCLE tool, incomplete reporting of outcome-assessor blinding in some in vivo assays leaves detection bias unclear. Finally, with fewer than ten studies, formal small-study and publication-bias diagnostics were not informative. The principal evidence gap remains the lack of controlled human trials that translate these preclinical findings into human islet physiology; future work should prioritize trials in prediabetes or early-stage T2DM to determine whether structural beta-cell preservation yields durable glycemic control in vivo.

4. Materials and Methods

4.1. Protocol Registration and Search Strategy

This systematic review was conducted in accordance with the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA 2020) guidelines [11] and drew on the methodological standards of the Cochrane Collaboration. The protocol was registered prospectively in PROSPERO (CRD420261370436).

A systematic search was conducted in PubMed (MEDLINE), Scopus, and the Web of Science Core Collection. Although the registered protocol additionally listed EBSCO as a candidate database, EBSCO did not provide indexing coverage beyond the three core databases for the target literature and was therefore not searched; this minor deviation from the protocol is acknowledged. The Boolean search syntax was: (“pancreatic beta cell” OR “beta-cell” OR “islet of Langerhans”) AND (“endoplasmic reticulum stress” OR “ER stress” OR “glucolipotoxicity” OR “lipotoxicity”) AND (“chemical chaperone” OR “TUDCA” OR “4-PBA”) AND (“glucose-stimulated insulin secretion” OR “GSIS” OR “CHOP”). The registered protocol specified an end date (search current through 2026) without a fixed start date; an a priori decision was made to restrict the analytic window to January 2016–May 2026, focusing the quantitative synthesis on contemporary studies that employed standardized glucolipotoxic protocols and reported the quantitative GSIS stimulation index. The implications of this restriction are addressed in the Limitations (Section 4.4).

4.2. Eligibility Criteria (PICOS Framework)

Predefined PICOS criteria were applied for study selection:

• Population (models): preclinical in vivo rodent models, primary isolated islets ex vivo (human or murine), or validated immortalized pancreatic beta-cell lines (e.g., INS-1, INS-1E, MIN6, βTC-6).
• Intervention: pharmacological administration of low-molecular-weight chemical chaperones (TUDCA and/or 4-PBA) under standardized metabolic (lipotoxic or glucolipotoxic) stress.
• Comparator: vehicle-treated controls exposed to the identical metabolic stress (the “stressor-only” arm).
• Outcomes: the primary outcome was the GSIS stimulation index (the ratio of insulin secreted under stimulatory high-glucose vs. basal low-glucose conditions; the corresponding effect-size metric used for quantitative pooling, the between-arm GSIS restoration ratio at stimulatory glucose, is defined in Section 2.4); the secondary outcome was the relative expression of the terminal pro-apoptotic marker CHOP and the ER chaperone BiP/GRP78 (quantified by Western-blot densitometry or qRT-PCR).
• Study design: original experimental basic-science research articles published in peer-reviewed indexed journals in English. Narrative reviews, observational studies, and trials using non-physiological chemical-only ER stressors (e.g., thapsigargin or tunicamycin without a lipid/glucose metabolic component) were excluded.

4.3. Data Extraction and Risk-of-Bias Assessment

Two reviewers independently extracted data and resolved discrepancies by consensus. Extracted variables included study design, biological model, metabolic stressor, chaperone dosing, primary GSIS outcomes, secondary CHOP and BiP expression, and systemic metabolic parameters. For studies presenting outcomes graphically, exact numerical values were extracted from the source figures using WebPlotDigitizer (version 4.6). Dosing regimens, group sizes, glucose concentrations, and stressor protocols were taken directly from the text and tables of the primary reports and are fully verifiable; only the per-study GSIS values used in the effect-size calculation were obtained via figure digitization and are therefore reported as approximations. The methodological quality and risk of bias of the included studies were assessed using the SYRCLE risk-of-bias tool for animal studies [12], which covers selection, performance, detection, attrition, and reporting biases.

4.4. Hierarchical Bayesian Random-Effects Meta-Analysis

To model the true biological variance (τ²) and to mitigate the small-sample bias common in basic science [22], a hierarchical Bayesian random-effects meta-analysis was performed on the primary outcome (the log-transformed GSIS restoration ratio). For each study, this restoration ratio was operationalized as the between-arm ratio of glucose-stimulated insulin secretion at stimulatory glucose, computed as the treated-to-stressor ratio in the three models in which the metabolic stressor suppressed secretion and as the stressor-to-treated ratio in the single model in which it produced hypersecretion, so that values above unity uniformly denote normalization of secretory output toward the healthy-control phenotype. This metric operationalizes the registered primary outcome (the within-arm GSIS stimulation index) using the high-glucose secretory values reported across all included studies; the deviation is acknowledged in Section 4.4. Let yᵢ denote the estimated log effect size in study i and sᵢ² the within-study sampling variance. The model was specified as.

yᵢ ~ Normal(θᵢ, sᵢ²); θᵢ ~ Normal(μ, τ²)

Weakly informative priors were used: a normal prior on the global mean effect, μ ~ Normal(0, 10²), and a half-Cauchy prior on the between-study standard deviation, τ ~ Half-Cauchy(0, 1) with τ > 0. The joint posterior was estimated by Markov-chain Monte Carlo using the No-U-Turn Sampler (NUTS) implemented in PyMC, with a non-centered parameterization to ensure stable geometry. Four independent chains were run with 5,000 tuning iterations and 20,000 retained draws each (80,000 posterior samples in total), a target acceptance probability of 0.99, and a fixed random seed for reproducibility. Convergence was confirmed by the Gelman–Rubin diagnostic (R̂ < 1.01) and large effective sample sizes (>12,000 for all reported parameters). Because only four studies met the eligibility criteria, formal assessment of small-study effects and publication bias (e.g., funnel-plot asymmetry or Egger’s regression) was not undertaken, as such tests are underpowered and unreliable when fewer than approximately ten studies are available [20.

5. Conclusions

Pharmacological administration of the chemical chaperones TUDCA and 4-PBA is an effective preclinical strategy to preserve pancreatic beta-cell functional identity under severe glucolipotoxic stress. By acting as structural ER scaffolds, these low-molecular-weight compounds prevent terminal UPR activation, downregulate pro-apoptotic CHOP and BiP, protect the ER-membrane glycoprotein WFS1, and restore the GSIS stimulation index, with a pooled Bayesian restoration ratio of 1.87 (95% CrI 1.39 to 2.46) [10]. Combined with TUDCA’s capacity to restore hepatic IDE-mediated insulin clearance, reduce systemic lipotoxicity, and improve cognition, these findings position chemical chaperones as promising disease-modifying candidates to halt the cellular progression of T2DM, pending confirmation in controlled human trials [2].