
Background Hepatic insulin resistance is a key driver of metabolic dysfunction-associated steatohepatitis (MASH) and is mechanistically linked to intracellular calcium (Ca2+) dysregulation. Sodium-glucose cotransporter 2 (SGLT2) inhibitors provide hepatoprotection, but their direct mechanisms have been debated because hepatocyte expression of SGLT2 is limited. We hypothesized that dapagliflozin restores hepatic insulin signaling by remodeling hepatocellular Ca2+ homeostasis. Methods HepG2 cells with induced lipotoxicity and high-fat diet (HFD)-fed C57BL/6N mice were treated with dapagliflozin. Their insulin signaling, intracellular Ca2+ dynamics, and hepatic transcriptomic profiles were assessed using confocal microscopy, immunoblotting, and RNA sequencing. Results Dapagliflozin was more effective than the other SGLT2 inhibitors in restoring insulin-stimulated protein kinase B (AKT) phosphorylation in HepG2 cells. Dapagliflozin attenuated cytosolic Ca2+ overload by restoring sarcoplasmic/endoplasmic reticulum Ca2+ ATPase (SERCA2) and plasma membrane Ca2+ ATPase (PMCA2) expression and modulating Ca2+ influx-related proteins. It also significantly ameliorated hepatic steatosis, fibrosis, and insulin resistance in the HFD-fed mice without affecting their body weight. Transcriptomic profiling confirmed a decisive shift away from fibrotic and stress-related programs toward metabolic competence, as evidenced by the stable upregulation of Atp2a2 (SERCA2). Dapagliflozin also restored physiological postprandial insulin signaling and reactivated the AKT-glycogen synthase kinase-3 beta (GSK3β)-forkhead box O3a (FOXO3A) axis. Conclusion Dapagliflozin has direct hepatoprotective effects that involve reprogramming the cellular Ca2+ influx-clearance balance. This pump-channel rebalancing alleviates the Ca2+-dependent biophysical barrier to insulin signaling and provides a molecular rationale for the weight-independent therapeutic efficacy of dapagliflozin in MASH.
Background We aimed to investigate factors contributing to discordance between estimated glycosylated hemoglobin (eHbA1c) and measured HbA1c (mHbA1c) from continuous glucose monitoring (CGM) using both 14- and 90-day data in a large Asian cohort. Methods We retrospectively analyzed 3,724 14-day and 2,664 90-day glycemic profiles with ≥70% active CGM time (AT) from 1,028 individuals with type 1 diabetes mellitus (T1DM). eHbA1c was calculated using a regression model derived from our study population. Discordance was defined as eHbA1c-mHbA1c (overestimation [eHbA1c ≥ mHbA1c] or underestimation [eHbA1c < mHbA1c]). Similarity was defined as a difference within ±0.1%. Results The 14-day eHbA1c model explained 74.7% of the variance in mHbA1c, compared with 88.4% for the 90-day eHbA1c model (conditional R2). Absolute discordance was significantly lower in real-time CGM (rtCGM) than in intermittently scanned CGM (isCGM) (P=0.004), and in those with higher time in range (TIR) and AT (all P<0.001). Overestimation was associated with rtCGM (P=0.016) and lower HbA1c (P<0.001). Underestimation was greater in isCGM users and in those with higher HbA1c and lower body mass index (all P≤0.001). In 14-day profiles, ≥78% AT was required for eHbA1c-mHbA1c similarity, whereas 90-day profiles maintained similarity regardless of AT. A prediction model incorporating AT, TIR, and sensor type improved accuracy (R2: 14-day, 0.774; 90-day, 0.888), particularly for 14-day data. Conclusion Factors influencing eHbA1c-mHbA1c concordance differ by CGM duration. While 14-day data require higher AT (≥78%) for accuracy, 90-day data provide more stable estimates. Clinical interpretation must account for mHbA1c, TIR, and sensor type, especially when using shorter-term CGM profiles in Asian individuals with T1DM.
Early dysglycemia, including prediabetes, represents heterogeneous and dynamic states that are incompletely captured by fasting plasma glucose, glycosylated hemoglobin, or a single oral glucose tolerance test. Continuous glucose monitoring (CGM) provides a higher-density assessment of glucose behavior under free-living conditions, enabling characterization of fasting, postprandial, nocturnal and activity-related glucose patterns before the onset of diabetes. This narrative review summarizes current evidence on CGM in normoglycemia towards early dysglycemia, with emphasis on conventional CGM metrics, higher-order features derived from raw glucose time series, and emerging artificial intelligence approaches. Existing studies show that individuals with prediabetes have higher mean glucose, greater time above tight glucose ranges, lower time in tight range and greater glycemic variability than normoglycemic comparators. However, absolute differences are small, and distributions overlap substantially. More recent approaches, including the concept of glucotypes, multi-feature classifiers, personalized postprandial response models, glucodensity methods and glucose foundation models, suggest that CGM may capture richer temporal and individualized phenotypes than consensus summary metrics alone, supporting future risk stratification, mechanistic subphenotyping and targeted prevention. However, model validation remains limited, and outcome-linked CGM thresholds are not yet established. At present, CGM should still be viewed as a promising research and phenotyping tool in early dysglycemia.