Mechanisms of Sodium–Glucose Cotransporter 2 Inhibition: Insights From Large-Scale Proteomics

Sodium–glucose cotransporter 2 (SGLT2) is a transmembrane protein encoded by genes of the SLC5A family (1). SGLT2 is a high-capacity, low-affinity transporter almost exclusively expressed on the luminal side of the S1 segment of the proximal renal tubule. Its function is to reabsorb glucose and sodium from the glomerular filtrate into the circulation. Convincing evidence shows that SGLT2 is functionally overactive in patients with type 2 diabetes (T2D), in whom it therefore contributes to the hyperglycemia (2). SGLT2 inhibition results in urinary excretion of glucose, sodium, and water, leading to hemodynamic changes as well as glycemic improvements in patients with T2D (1). SGLT2 inhibitors (SGLT2i) have been developed as glucose-lowering drugs for the treatment of T2D. Somewhat unexpectedly, large cardiovascular (CV) safety trials testing SGLT2i against placebo in patients with T2D at high CV risk have consistently demonstrated clinically relevant benefit for both CV and renal outcomes (3). This has led to changes in clinical recommendations and practice (4). Because the glucose-lowering efficacy of SGLT2i is generally modest (i.e., HbA_1c reductions in the range of 0.6–0.8%), it is widely believed that mechanisms other than (or in addition to) glycemic control must be at work to explain the CV and renal protection of SGLT2i. The recently published Dapagliflozin and Prevention of Adverse Outcomes in Heart Failure (DAPA-HF) results also suggest that SGLT2i lessen risk of heart failure by mechanisms independent of glycemia, since benefits were identical in those with and without diabetes (5). Multiple hypotheses have been advanced, not necessarily exclusive of one another. Thus, SGLT2i-induced natriuresis and osmotic diuresis with blood volume contraction and increased hematocrit may be responsible for the reduction in blood pressure and arterial stiffness, which can improve cardiac function by decreasing both pre- and afterload (6). Also, SGLT2i may optimize tissue energy metabolism and mitochondrial bioenergetics in impaired heart and kidney by supplying an excess of “thrifty” substrates (e.g., ketones) (7). Reduction of uricemia, modulation of the tubuloglomerular feedback, and lowering of inflammation and fibrosis biomarkers (8) are additional putative contributors to reported benefits (9).

From the in vivo mechanistic information so far gathered, it is likely that the primary action of SGLT2i on the proximal renal tubule induces a range of physiological consequences in several domains of bodily functions. Proteomics is a useful approach to explore canonical biological pathways in a hypothesis-free fashion (10). This prompted us to screen the plasma proteome of individuals with T2D or impaired glucose tolerance (IGT) receiving SGLT2i treatment by applying a large-scale version of an aptamer-based multiplex proteomics platform (SOMAscan) to quantify plasma proteins (11).

A total of 72 subjects (25 women and 47 men) were recruited into the study (Supplementary Table 1). Of these, 61 were patients with T2D (29 were drug naive or off any glucose-lowering agent for at least 12 weeks, and 32 were on a stable dose of metformin of ≥1,500 mg/day for at least 12 weeks). Inclusion criteria were either sex, age >18 years, BMI 20–40 kg/m², HbA_1c 6.5–10.5% [48–91 mmol/mol], and estimated glomerular filtration rate (eGFR) >60 mL/min/1.73 m² (by the Chronic Kidney Disease Epidemiology Collaboration equation). Duration of diabetes was <1 year in 3%, <5 years in 18%, <10 years in 26%, and ≥10 years in 19% of the patients. Exclusion criteria were history of malignancy in the past 5 years, significant CV disorder within the past 6 months, pregnancy or women expecting to conceive within the study duration, bariatric surgery within the past 2 years, treatment with antiobesity drugs in the past 3 months, neurogenic bladder disorders, ALT and AST >3.0× the upper limit of normal, changes in thyroid hormone dosage within 6 weeks or any other endocrine disease except T2D, and alcohol or drug abuse. Detailed metabolic analysis of these patients has been published previously (12). Eleven subjects with IGT (by American Diabetes Association criteria) were included; their anthropometric and metabolic characteristics have been previously reported (13).

One hundred forty-four fasting EDTA plasma samples were obtained from the subjects at baseline and after 4 weeks of treatment with 25 mg/day empagliflozin. The modified aptamer binding reagents and SOMAscan assay and its performance characteristics have previously been described (14–16). In brief, each of the individual proteins measured has its binding reagent made of chemically modified DNA, referred to as a modified aptamer. Each plasma sample was incubated with the mixture of modified aptamers to generate modified aptamer-protein complexes under equilibrium conditions. Unbound modified aptamers and unbound or nonspecifically bound proteins were eliminated by two bead-based immobilization steps. After elution of the modified aptamers from the target protein, the fluorescently labeled modified aptamers were directly quantified on hybridization array (Agilent Technologies). Calibrators were included so that the degree of fluorescence was a consistent reflection of protein concentration. Protein concentration was expressed as relative fluorescence units. All samples were placed randomly on 96-well plates, run in a single batch, and normalized against protein calibrator samples included on each plate. Technicians were blinded to the before and after status of samples. Details of scaling, normalization, and control of batch effects are given in Williams et al. (8). Median intra- and interassay coefficients of variation are ∼5% (16). A total of 4,005 proteins were measured; 292 of them failed to pass SomaLogic’s quality control metrics, leaving 3,713 proteins for analysis. Differences between the baseline and 4-week samples were expressed as log₂ ratios for each of the proteins measured.

Plasma glucose and creatinine concentrations were measured by standard laboratory methods. Creatinine clearance was measured on carefully collected urine samples as the ratio of urinary creatinine excretion and plasma creatinine levels.

Differences between on-treatment and baseline values were analyzed by the Wilcoxon signed rank test for paired comparisons in 3,713 protein measurements. P values were corrected for multiple comparisons using both the Bonferroni familywise error rate correction and the Benjamini-Hochberg false discovery rate (FDR) correction. Proteins with an FDR-adjusted value of P < 0.05 were considered statistically significant for inclusion in the pathway analysis. Ingenuity Pathway Analysis (IPA) was used to cluster differentially expressed proteins after 4 weeks of treatment compared with baseline into pathways and functional groups. For those modified aptamers that had multiple UniProt identifications associated with one result, only the first listed UniProt identification was used in pathway analysis (8). The Fisher right-tailed exact test was used to calculate a P value to determine the probability that the association of the differently expressed proteins in the measured data set and the pathway is explained by chance alone. Correction for changes in creatinine clearance was carried out using a mixed model with patient (b0i) as a random effect: log₁₀(p)_ij = b0_i + β₁eGFR_ij + β₂week_ij + ε_ij, where p is the protein signal, i is the ith subject, j is the jth observation, β₁ is the fixed-effect parameter for eGFR, and β₂ is the fixed-effect parameter for time (week 4 and week 0).

As expected, in the whole group of subjects, 4 weeks of treatment resulted in significant decreases of body weight, HbA_1c, fasting glycemia, and creatinine clearance (Supplementary Table 1). Of the 3,713 proteins measured, 21 were not recognized by IPA, and the remainder were used as background reference. No statistically significant differences in the within-subject responses were seen between baseline and treatment in the three groups of participants (drug-naive T2D, metformin-treated T2D, and IGT) (r² = 0.035, P = 0.22). There were also no discernible differences on an individual protein target level, as evidenced by univariate Kruskal-Wallis tests on the within-subject (baseline − treatment) differences (e.g., Fig. 1). Therefore, in all subsequent analyses, the three groups were treated as one.

On an initial screen using an FDR P ≤ 0.05 corrected for 3,713 dimensions, 44 proteins were associated with a significant change between baseline and treatment (16 reaching a Bonferroni P < 0.05) (Table 1). Pathway analysis grouped these proteins into biological pathways ranging from iron handling and cardiomyocyte contractility and relaxation to lipid and glucose metabolism (Fig. 2). The target proteins—and their genes—that showed the largest change with treatment (i.e., a percent change lower than −15% or higher than +19%) with an FDR P < 0.05 (model 1) are listed in Table 2.

Creatinine clearance was mildly, but significantly reduced after 4 weeks of empagliflozin treatment (Supplementary Table 1). This change may bias the assessment of treatment effects because of accumulation of waste products in the blood. The analysis was therefore repeated after adjustment for individual changes in creatinine clearance (model 2). Several proteins that were significant (at the FDR P < 0.05 level) in model 1 lost significance. In particular, seven protein targets with P < 0.05 moved to P > 0.20 after the correction for creatinine clearance: cystatin-SN (CST1.5459.33.3), guanylate cyclase activator 2B (uroguanylin, GUCA2B.6223.5.3), trefoil factor 3 (TFF3.4721.54.2), C-C motif chemokine 15 (CCL15.3509.1.1), thrombospondin 4 (THBS4.3340.53.1), renin (REN.3396.54.2), and neutral ceramidase (ASAH2.3212.30.3). The results of model 2 are given in Supplementary Table 2 in descending order of FDR-corrected P values, down to P < 0.10.

There were many significant univariate associations between some top “hit” proteins and physiological parameters (e.g., insulin-like binding factor protein-1 [IGFBP-1] and age [Supplementary Fig. 1] or growth differentiation factor 15 [GDF15] and age [r = 0.55, P < 0.0001] or IGFBP-1 or GDF15 and fasting glycemia [r = 0.24 [P = 0.042] and r = 0.25 [P = 0.033], respectively; data not shown]). More relevant, we also found a few associations between changes in top hit proteins and changes in plasma glucose concentrations. For example, the treatment-induced changes in IGFBP-1 were reciprocally related to the corresponding changes in fasting glycemia (Supplementary Fig. 1).

Empagliflozin treatment for 4 weeks altered the proteomic profile in the plasma of individuals with T2D. Among the proteins that changed in response to treatment, pathway analysis indicated that their biological roles pertained to lipid and glucose metabolism, iron handling, cardiomyocyte function (contraction and relaxation), cytokines, and molecular transport. In accord with previous metabolic results in these subjects (17), there was no major difference across background antihyperglycemic treatment (metformin or none) or glucose tolerance status (T2D vs. IGT). This made it possible to analyze all available data as one set.

Circulating protein concentrations are the net balance of release into the bloodstream (resulting from changes in gene expression/transcription and/or cellular secretion/shedding) and removal from the plasma (by degradation or excretion). Empagliflozin, like other SGLT2i, causes a reduction in eGFR, which persists as long as treatment continues and is reversed on stopping it (3,5,12). eGFR adjustment of our patients’ proteomic profile was therefore necessary to protect against bias introduced by hemodynamic and filtration changes. It is of interest that with eGFR adjustment, seven proteins lost strength of association with treatment. Among them are cystatin, a clinical marker of GFR, and three other proteins (REN, GUCA2B, and TFF3) that also are influenced by changes in GFR. This finding, along with the use of nonparametric testing and stringent multicomparison P value restriction, adds confidence in the reliability of the remaining observed treatment-induced protein changes. However, adjustment for eGFR changes also subtracts any pharmacodynamics that coincidentally have the same time course and magnitude as the change in GFR.

Among the plasma protein changes achieving the strongest statistical significance, IGFBP-1, transgelin-2, fatty acid–binding protein 3 and 4 (FABPA), GDF15, and sulphydryl oxidase 2 precursor (QSOX2) were increased by empagliflozin, while ferritin, THBS3, and rearranged during transfection (RET) were decreased. Interpretation of the biological significance of such changes in the context of the biochemical modifications observed in vivo must be done with caution. In fact, IPA almost invariably suggests that any given protein plots onto several different pathways. Thus, IGFBP-1 (produced mainly by the liver in response to nutritional stimuli) acts as a transport protein of IGF to prolong its half-life, guide a tissue-specific localization, and modulate its actions (18) (Supplementary Fig. 1). A strong, consistent association between circulating IGFBP-1 and insulin sensitivity has been reported in diverse populations (19). Furthermore, plasma IGFBP-1 levels are inversely correlated with leptin levels (20). In patients with T2D, SGLT2i treatment reduces visceral fat (21) and ameliorates insulin resistance (22). This may suggest a causal path whereby empagliflozin improves insulin resistance through upregulation of IGFBP-1. However, IGFBP-1 is also involved in retinoid X receptor activation and estrogen receptor signaling (19), vascular function (23), cell migration through integrin (24), and pancreatic β-cell regeneration (25). Therefore, additional links to IGFBP-1 may emerge as the range of physiological consequences of SGLT2i therapy in patients with diabetes is expanded and physiologically explained.

Serum ferritin, a key parameter of iron homeostasis (Supplementary Fig. 2), also is a well-known marker reflecting insulin resistance and liver fat accumulation (26). The reduced ferritin levels after empagliflozin (−34%) resonate with the alleviation of hepatic steatosis reported in several studies of SGLT2i (27). On the other hand, the reduction in ferritin may be linked with the observed (mild and transient) increase in erythropoietin (28), reflecting a boost in erythropoiesis through suppression of hepcidin (29).

GDF15, a member of the transforming growth factor-β superfamily, is a myomitokine whose induction in mice protects against the onset of obesity and insulin resistance by promoting oxidative function and lipolysis in liver and adipose tissue (30). GDF15 has recently received interest as a potential therapeutic target in obesity as well as cancer-associated cachexia (31). Recently, glial cell–derived neurotrophic factor receptor α-like (GFRAL) was identified as a specific receptor of GDF15, requiring the coreceptor RET to execute intracellular signaling (32,33). In the current data, the joint increase in GDF15 and decrease in RET might stand for a coordinated stress-like response to empagliflozin-induced weight loss (34), and/or such changes may contribute to the weight loss associated with SGLT2i therapy. As a general tissue stress and injury signal, GDF15 has been reported to be a predictive biomarker for CV events and death in individuals with dysglycemia (35). It is intriguing that in the large CV outcomes trial Outcome Reduction With Initial Glargine Intervention (ORIGIN) trial, GDF15 was found to be increased in patients with T2D on treatment with metformin (36).

FABP4 is primarily located in adipocytes and macrophages (37). It has been implicated in the pathogenesis of insulin resistance and atherosclerosis (38). FABP4 is released from adipocytes in a nonclassical pathway associated with lipolysis. In the current study patients, empagliflozin stimulated lipolysis through a reduction in plasma insulin concentrations (12). The observed increase in plasma FABP4 can therefore be plausibly connected with increased lipolysis.

The thrombospondin family consists of adhesive glycoproteins that mediate cell-to-cell and cell-to-matrix interactions (39). THBS3, which is abundant in muscle and kidney, is linked to the endoplasmic reticulum stress response (40). Presumably, abatement of endoplasmic reticulum stress by SGLT2 inhibition may relate to the 16% decrement in THBS3 levels found in the current study.

This exploratory study has limitations. The sample size was modest, with no placebo control, and treatment only lasted 4 weeks. Furthermore, there was no active comparator that would control for the decrease in HbA_1c, which averaged 0.4% (12). Finally, simultaneous urine proteomics was not done. We therefore opted for a conservative analysis of the proteomic data to down rank random variation between baseline and treatment. That noted, we did see the expected changes in glycemia, weight, and eGFR and in surrogates of liver fat, lending external validity to the findings. We can neither exclude the possibility that the proteome might change further with prolonged treatment nor extrapolate the findings in our cohort of well-controlled patients free of macro- and microvascular complications to the participants of the BI 10773 (Empagliflozin) Cardiovascular Outcome Event Trial in Type 2 Diabetes Mellitus Patients (EMPA-REG OUTCOME) trial (4), who had established CV disease. While vascular and/or renal complications are likely to add their specific proteomic signatures, the protein panel here described may reflect early effects of SGLT2 inhibition.

In conclusion, treatment with empagliflozin in patients with T2D was associated with a shift in the plasma proteome. The most significant differences were detected in protein species potentially related to the clinical and metabolic changes that were actually measured in the same patients (20) (e.g., Supplementary Fig. 1). Novel results on GDF15, ferritin, IGFBP-1, and FABP should stimulate further studies using targeted proteomics and a prospective design. At present, our findings represent the first hypothesis-free evidence that SGLT2 inhibition affects the plasma proteome in a biologically plausible fashion.

Acknowledgments. The authors thank the SomaLogic assay team and Darryl Perry (SomaLogic, Inc., Boulder, CO) for the bioinformatics.

Duality of Interest. SOMAscan assays and the Covance study were funded by SomaLogic, Inc. E.F. has consulted for AstraZeneca, Boehringer Ingelheim, and Sanofi and has received grant support from Boehringer Ingelheim. S.W., R.M.O., and S.A.W. are employees of SomaLogic, Inc. N.S. has consulted for Amgen, AstraZeneca, Boehringer Ingelheim, Eli Lilly, Novo Nordisk, and Sanofi and has received grant support from Boehringer Ingelheim. P.G. serves on a medical advisory board to SomaLogic, Inc., for which he accepts no salary, honoraria, or any other financial incentives. No other potential conflicts of interest relevant to this article were reported.

Author Contributions. E.F. conceived the study and wrote the manuscript. A.C.M., Y.L., and N.S. contributed to the discussion and reviewed/edited the manuscript. E.M., S.W., and R.M.O. researched data. S.A.W. and P.G. reviewed/edited the manuscript. E.F. is the guarantor of this work and, as such, had full access to all the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.