Inhibition of SGLT2 Rescues Bone Marrow Cell Traffic for Vascular Repair: Role of Glucose Control and Ketogenesis

The primary consequence of diabetes is the accelerated development of cardiovascular disease (1), driving a huge burden for affected individuals, health care systems, and society. Unlike microangiopathies (i.e., retinopathy, nephropathy, and neuropathy), cardiovascular complications can marginally be prevented by achieving near-normal glucose levels (2).

The demonstration that treatment with sodium–glucose contransporter 2 inhibitors (SGLT2i) improves cardiovascular outcomes of diabetes (3) represented the major groundbreaking finding in the last 10 years of research in the field. Despite huge efforts, the mechanisms by which SGLT2i protect from the adverse cardiovascular consequences of hyperglycemia have not been elucidated (4). The simultaneous effect of SGLT2i on multiple risk factors (glycemia, blood pressure, and obesity) only accounts for part of the observed cardiovascular benefit (5). Since SGLT2i exert much of their protection against heart failure, the atypical natriuretic action has been hypothesized to be primarily involved (6). Yet, this would not explain the delayed mortality observed in patients with diabetes and cardiovascular disease who received SGLT2i (7). Interestingly, SGLT2i is associated with an increase in hematocrit (8), which appeared to be mediating the delay in cardiovascular death (9,10). Inhibition of the cardiac sodium/potassium exchanger by SGLT2i has been shown in vitro (11) and might protect the heart from ischemic injury, but its clinical relevance remains unclear. Another candidate mediator of cardioprotection by SGLT2i is the switch to ketone body utilization (12), which would improve myocardial energetics (13).

Lastly, it has been hypothesized that SGLT2i may stimulate circulating hematopoietic stem/progenitor cells (HSPCs). Low levels of circulating HSPCs with vascular tropism have been consistently reported in both type 1 diabetes (T1D) and type 2 diabetes (T2D) (14), which is considered to contribute to cardiovascular damage (15). Prior studies indicate that low HSPC levels are due to an impaired mobilization from the bone marrow (BM) to peripheral blood (16). Such stem cell “mobilopathy” (17) is driven by a combination of BM remodeling (microangiopathy and neuropathy) and inflammation (14,18), with functional pathways that can be targeted by specific therapies (19). In patients with T2D, Hess et al. (20) reported that the SGLT2i empagliflozin increased expression of the HSPC marker CD133 on aldehyde dehydrogenase–expressing cells, which are supposed to be provided with vasculotropic properties. On the contrary, we found that effect of SGLT2i on antigenically defined HSPCs and endothelial precursors are mostly related to the improved glucose control (21,22).

In this study, we used mouse models to evaluate if the SGLT2i dapagliflozin can improve traffic of HSPCs from the BM to sites of vascular damage in diabetes. We also explored some systemic pathways that could explain the effects of dapagliflozin on HSPCs and their vascular effect.

C57BL/6J wild-type (Wt) and C57BL/6-Tg(UBC-GFP) mice were purchased from The Jackson Laboratory and established as a colony since 2001 and 2017, respectively. Mice were randomly assigned to treatments or experimental groups. All animal studies were approved by the Animal Care and Use Committee of the Veneto Institute of Molecular Medicine and by the Italian Health Ministry.

Three- to 4-month-old male mice were used. Type 1-like diabetes was induced by a single i.p. injection of 175 mg/kg streptozotocin (STZ) in 100 mmol/L, pH 4, Na-citrate buffer (23). Blood glucose, ketones (β-hydroxybutyrate [BOHB]), and urinary glucose were measured using a point-of-care device (FreeStyle; Abbott, Abbott Park, IL). Diabetes onset was confirmed for blood glucose ≥300 mg/dL 48 h after STZ injection, but only mice with persistent hyperglycemia in the subsequent 2 weeks were used.

Type 2-like diabetes was induced by feeding mice with a high-fat diet (HFD) (60% of the total energy derived from fat) (ssniff EF, acc. D12492 [I] mod.; Spezialdiäten, Soest, Germany) for 6 weeks. To verify dysmetabolism, an i.p. glucose tolerance test was performed by injecting 1 g/kg glucose i.p. and measuring blood glucose at 0, 30, 60, and 120 min.

Two weeks after the onset of diabetes (STZ) or 4 weeks after initiation of HFD, dapagliflozin (AstraZeneca) was administered in the drinking water at a dosage of 5 mg/kg/day for 2 weeks.

Mice were anesthetized with inhaled isoflurane. The left common carotid artery was exposed and injured by applying an electric current with a bipolar tweezer for 4 mm from the branching. Three days later, the extent of residual endothelial damage was evaluated by tail vein injection of 200 µL of 5% Evans blue. Mice were then killed by overdose of inhaled carbon dioxide, perfused with PBS to remove blood and excess staining, and both carotid arteries were excised, placed between glass slides with mounting medium, and photographed with a light microscope. In BM transplantation experiments, at sacrifice, carotid arteries were excised and processed with 0.025% trypsin-EDTA (Merck), 2 mg/mL collagenase II (Worthington Biochemical Corporation), and 60 units/mL of DNase (Merck) for 1 h at 37°C to obtain a cell suspension for flow cytometry.

Recipient mice were treated with a myeloablative total-body irradiation with 10 Gy, split in two doses of 5 Gy 3 h apart, and followed by intravenous injection of BM cells from donor C57BL/6-Tg(UBC-GFP) mice (10⁷/each) isolated by flushing femurs and tibias with sterile ice-cold PBS. Animals were housed with sterile cages, water, food, and bedding for 4 weeks to allow BM reconstitution. Engraftment was assessed by peripheral blood cell count. Complete engraftment and development of the BM-GFP chimera were confirmed when ≥95% of white blood cells (WBC) were GFP⁺.

Treatment with the SGLT2i dapagliflozin was initiated 2 weeks after confirming the onset of STZ diabetes. Dapagliflozin (AstraZeneca) was given in the drinking water at a dosage of 5 mg/kg/day for 2 weeks. HSPC mobilization was induced by injecting animals subcutaneously with 200 mg/kg/day human recombinant granulocyte colony-stimulating factor (G-CSF) daily for 4 consecutive days. CD49d blockade was performed by injecting the animals daily intravenously with 30 μg of an anti-mouse CD49d antibody (clone R1–2; Thermo Fisher Scientific) from day −1 through day 3 after carotid injury. AMD3100 (Tocris Bioscience) was given by daily subcutaneous injections of 1.25 mg/kg for 3 days. In a set of experiments, diabetic mice were treated with 1.5 units/day of insulin glargine (Eli Lilly and Company) for 2 weeks by daily i.p. injection. Ketone supplement (Keto BHB Salts Supplement; Zenwise Health, Wilmington, DE) was added to the drinking water at 5 g/kg/day for 2 weeks.

Total WBC count was performed using the CELL-DYN Emerald Hematology Analyzer (Abbott) on fresh EDTA-treated mouse blood. For measuring circulating erythropoietin (EPO), EDTA-treated blood samples were centrifuged at 2,000g for 10 min to obtain plasma that was stored at −20°C. Plasma EPO was quantified using the Mouse Erythropoietin Quantikine ELISA Kit (R&D Systems, Minneapolis, MN) according to the manufacturer’s instructions.

For colony-forming unit (CFU) assay, BM cells (3 × 10⁴) were plated in 35-mm Petri dishes containing 1 mL methylcellulose-based medium MethoCult supplemented with 1% penicillin/streptomycin. Alternatively, after red blood cell (RBC) lysis, 25 µL of peripheral blood were plated in 24-well plates containing 0.5 mL MethoCult (Voden) supplemented with 1% penicillin/streptomycin. Colony formation was scored after 10 days of culture. In separate experiments, dapagliflozin at different concentrations was added to the MethoCult medium.

Flow cytometry was performed on murine BM cells, EDTA-treated peripheral blood, or cell suspensions, as needed. BM cells were isolated from femurs and tibiae by flushing the bone cavity with ice-cold PBS through a 40 μmol/L cell strainer. A total of 100 µL of BM cell suspension, peripheral blood cells, or cells from carotid arteries was labeled with specific antibodies, as reported in Supplementary Table 1.

The 10-mm–thick kidney and femur cryosections were obtained with a Leica CM1950 cryostat (Leica Biosystems, Milan, Italy). Sections were incubated with anti-rabbit anti–SGLT2 polyclonal antibody and after with goat anti-rabbit Cy3 (catalog number 111-165-003; Jackson ImmunoResearch Laboratories, West Grove, PA). Nuclei were counterstained with Hoechst 33352. Images were taken with a Leica DM6 B microscope and then processed with Fiji/ImageJ 1.50 software (National Institutes of Health, Bethesda, MD).

Human umbilical vein endothelial cells (HUVECs) were seeded in six-well plates with complete Endothelial Cell Basal Medium (catalog number C-22210; PromoCell) supplemented with 10% FBS and 1% penicillin/streptomycin and glutamine (Corning). Upon reaching 80% confluence, cells were synchronized by serum starvation for 24 h. After that, cells were incubated with 0.5 μmol/L AMD3100 (catalog number 155148-31-5; Tocris Bioscience) for 2.5 h before stimulation for 30 min with 20 ng/mL human CXCL12 (catalog number 350-NS; R&D Systems). Cells were lysed on ice with modified radioimmunoprecipitation assay lysis buffer (50 mmol/L Tris-HCl, 150 mmol/L NaCl, 10 mmol/L MgCl₂, 0.5 mmol/L DTT, 1 mmol/L EDTA, and 10% glycerol) supplemented with 1% Triton X-100, 1% SDS, cOmplete Protease Inhibitor Cocktail (Roche), and Phosphatase Inhibitor Cocktails 2 and 3 (Merck). Lysates were heated at 70°C for 10 min, and, after centrifugation at 13,000g for 15 min, protein was quantified in supernatants using the BCA assay (Thermo Fisher Scientific). Proteins (10 µg) were separated on 10% SDS-PAGE and transferred onto nitrocellulose membranes. Membranes were then blocked and probed using the following primary antibodies: rabbit anti–phospho-AktSer473 antibody (1:1,000; #9271; Cell Signaling Technology, Danvers, MA), rabbit anti-Akt (1:1,000; #9272; Cell Signaling Technology), and mouse anti-GAPDH (1:10,000; ab8245; Abcam, Cambridge, U.K.). After washing, membranes were incubated with appropriate secondary horseradish peroxidase–conjugated antibodies (all from Jackson ImmunoResearch Laboratories). Bands were detected by chemiluminescence using the WesternBright Quantum HRP substrate (Advansta, Inc., Menlo Park, CA). Images were acquired with an ImageQuant LAS 4000 (GE Healthcare, Chicago, IL). Densitometric analysis was performed with ImageJ 1.47v (National Institutes of Health).

Fibroblasts were isolated from skin explants of Wt mice and cultivated with DMEM 5 mmol/L glucose medium (Merck) supplemented with 10% FBS (Corning) and 1% l-glutamine/penicillin-streptomycin (Corning). Cells were seeded in 12-well plates until 100% confluent. Total BM was isolated from C57BL/6-Tg(UBC-GFP) mice by flushing femurs and tibias with sterile ice-cold PBS. GFP⁺ Ly6G/Cmid/low Siglec-F⁺ cells were sorted and incubated with 2 μg of anti-mouse CD49d antibody (clone R1–2) or isotype IgG2bk antibody for 45 min at 37°C. Cells were then plated on fibroblasts for 2 h. Nonadherent cells were removed by a gentle wash with PBS followed by fixation with 4% paraformaldehyde before imaging.

RNA was isolated from flushed BM or cells by use of QIAzol or with a Total RNA Purification Micro Kit (Norgen Biotek Corp.) and quantified with a NanoDrop 2000 Spectrophotometer (Thermo Fisher Scientific). cDNA was synthesized from 500 ng RNA using a SensiFAST cDNA Synthesis Kit (Bioline, London, U.K.). Quantitative PCR was performed using SensiFAST SYBR Lo-ROX Kit (Bioline) via a QuantStudio 5 Real-Time PCR System (Thermo Fisher Scientific). The list of primers is given in Supplementary Table 2.

Continuous data are expressed as mean ± SE unless otherwise specified. Normality was checked using the Kolmogorov-Smirnov test, and nonnormal data were log-transformed before analysis. Comparison between two or more groups was performed using the Student t test and ANOVA for normal variables or the Mann-Whitney U test and Kruskal-Wallis test for nonnormal variables that could not be log-transformed (e.g., because of frequent zero values). To analyze data from experiments with two groups and two treatments, two-way ANOVA was used, and the effect of group and treatment was analyzed. All tests were two-tailed. Bonferroni adjustment was used to account for multiple testing. Biological replicates (individual mice) are shown as individual data points superimposed on bar charts. Significance was conventionally accepted at P < 0.05.

Data and resources used for this study are available from the corresponding author upon reasonable request.

We first evaluated gene expression of SGLT2 (Slc5A2) in various mouse organs and tissues relative to the ubiquitin housekeeping gene. BM cells constituted the extrarenal tissue with the highest gene expression of Slc5A2, followed by skeletal muscle and the heart (Supplementary Fig. 1A). Slc5A2 expression was about three times higher in BM than in skeletal muscle, although it was four orders of magnitude lower (10⁻4) than in the kidney (Supplementary Fig. 1B). We examined Slc5A2 expression in hematopoietic versus mesenchymal precursor cells in vitro. Neither cultured hematopoietic CFUs nor isolated mesenchymal stem cells exhibited the same Slc5A2 expression as BM cells freshly eluted from the bones (Supplementary Fig. 1C), suggesting that mature cells in the BM contributed most to Slc5A2 expression. In fact, both mononuclear cells and polymorphonuclear cells had levels of Slc5A2 expression similar to those of BM cells (Supplementary Fig. 1C). In view of the low Slc5A2 gene expression, we were unable to detect SGLT2 protein expression in the BM by either flow cytometry (data not shown) or section immune staining (Supplementary Fig. 1D).

We then assessed the effects of inhibiting SGLT2 with dapagliflozin in STZ-induced diabetic mice (Fig. 1A). At the dose used, dapagliflozin increased urinary glucose more than fivefold in nondiabetic mice, confirming the pharmacologic effect (Fig. 1B). In STZ diabetic mice, urinary glucose was markedly elevated in untreated conditions and, at the end of the treatment period, was not further increased by dapagliflozin. Consistently, body weight was significantly lower in diabetic versus nondiabetic mice and was not affected by dapagliflozin (Fig. 1C). While dapagliflozin did not significantly reduce blood glucose levels in nondiabetic mice, it decreased blood glucose in diabetic mice on average from 437 to 349 mg/dL (equal to −20%; P = 0.008) (Fig. 1D). These results indicate that the extent of SGLT2 inhibition achieved was sufficient to elicit a pharmacologic response, but not to normalize glycemia in this model.

As expected, diabetes increased RBC count and hematocrit, an effect of hemoconcentration (Fig. 1E and F). As noted before, diabetes significantly increased the granulocyte/lymphocyte (G/L) ratio (Fig. 1I), a reflection of enhanced myelopoiesis induced by hyperglycemia (18,24). Dapagliflozin had no effect on RBCs, hematocrit, and platelets, while it increased WBC count in both diabetic and nondiabetic mice, but did not modify the G/L ratio (Fig. 1E–I). Within the BM, dapagliflozin modestly increased mature leukocytes (especially in nondiabetic mice) (Fig. 2A), while not affecting the generation of hematopoietic CFU in vivo (Fig. 2B and Supplementary Fig. 2A) or in vitro (Fig. 2C and Supplementary Fig. 2B). No change was also observed in primitive HSPCs expressing signaling lymphocyte activation molecule CD150 (SLAM) and various myeloid progenitors (Fig. 2D and E).

We previously reported that BM macrophages, especially those expressing the sialoadhesin CD169, retain HSPCs in the diabetic BM niche and prevent them from being mobilized in response to G-CSF (25). While not affecting the total number of macrophages (Fig. 3A and B), STZ diabetes increased CD169⁺ BM macrophages up to 70%, an effect that was mostly abolished by dapagliflozin (Fig. 3C and D).

When stimulated with G-CSF, antigenically defined HSPCs (lineage [Lin]⁻c-kit⁺Sca-1⁺ [LKS] cells) increased in nondiabetic mice by 7.3 ± 1.7-fold (Fig. 3E), paralleled by a marked increase in the generation of functionally defined HSPCs (CFU in Fig. 3F). As demonstrated before, diabetic mice were completely unresponsive to the HSPC-mobilizing effect of G-CSF, with no change in HSPCs (1.4 ± 0.5-fold) (Fig. 3G) and CFUs (Fig. 3H). Dapagliflozin did not significantly modify response to G-CSF in nondiabetic mice (Fig. 3E and F), while it partially rescued HSPC mobilization in STZ diabetic mice: in dapagliflozin-treated STZ diabetic mice, HSPC increased by 3.2-fold and generation of CFU increased by 4.5-fold (Fig. 3G and H). Thus, despite a modest effect on blood glucose, dapagliflozin prevented one pathologic feature of inflammatory BM alterations in STZ diabetes, namely the surge in CD169⁺ macrophages, and partially rescued HSPC mobilization.

Next, we examined if the partial rescue of HSPC mobilization translated into an improved traffic of BM-derived cells to sites of vascular damage using the carotid damage model. Endothelial injury was induced experimentally by a standardized low-voltage electrical current yielding a 4-mm denudation, which was stained with Evans blue. Re-endothelization in healthy mice normally reaches 50% at 3 days after injury and is evidenced by reduced Evans blue staining. As shown in Fig. 4A, spontaneous endothelial healing at day 3 after injury was dramatically reduced in STZ diabetic versus nondiabetic mice (5.5 ± 1.1% vs. 40.0 ± 1.3%; P < 0.001), while it was improved in diabetic mice treated with dapagliflozin (16.9 ± 4.0%; P = 0.02 vs. diabetic mice receiving vehicle). Yet, the extent of endothelial healing in diabetic mice treated with dapagliflozin was still significantly lower than in nondiabetic mice (P < 0.01).

In type 2-like diabetic mice induced by HFD, despite fasting and postload hyperglycemia (Supplementary Fig. 3A and B), HSPC mobilization was preserved and enhanced compared with control mice (Supplementary Fig. 3D and E), and there was no defect in re-endothelization (Supplementary Fig. 3F). Dapagliflozin normalized glucose levels in HFD, induced ketogenesis, and dampened HSPC mobilization (Supplementary Fig. 3A–E). However, the lack of mobilopathy, which has been shown before in models of type 2 diabetes (18,26), along with unaffected endothelial healing (Supplementary Fig. 3F), makes the HFD model unsuitable to evaluate the role of dapagliflozin on HSPC traffic and vascular repair.

Having shown that dapagliflozin improved vascular healing in STZ diabetes, we then evaluated the traffic of BM-derived cells. To this end, we generated BM-GFP⁺ chimeric mice by transplanting BM cells from mice ubiquitously expressing GFP into myeloablated Wt mice. After reconstitution, chimerism was always >95%, and there was no overt hematologic disturbance (Supplementary Fig. 4). Chimeric mice were rendered diabetic or kept nondiabetic by injecting them with STZ or vehicle, respectively. After 4 weeks, we performed carotid endothelial damage under treatment with dapagliflozin or vehicle. At day 3 after injury, we evaluated the accumulation and phenotype of BM-derived GFP⁺ cells homed to the damaged carotids (Fig. 4B). Upon immunofluorescence staining of whole mount carotids, there was clear evidence of abundant GFP⁺ cells (Fig. 4B). Using flow cytometry, we found that, in nondiabetic mice, GFP⁺ cells were more than doubled at day 3 postinjury compared with the uninjured carotid, whereas this effect was completely absent in vehicle-treated diabetic mice. Dapagliflozin treatment rescued homing of GFP⁺ cells to the injured carotids toward normal levels (Fig. 4C). Injury increased HSPCs in the carotid wall of nondiabetic mice, whereas it did not in diabetic mice, even under dapagliflozin treatment (Fig. 4D). However, antigenically defined HSPCs (LKS cells) accounted for 0.1% of GFP⁺ cells homed to the damage site, whereas most GFP⁺ cells were mature cells, especially granulocytes and monocytes, with a minority assuming the macrophage phenotype (Fig. 4E). GFP⁺ macrophages were increased at sites of endothelial injury by treating diabetic mice with dapagliflozin (Fig. 4E), although they still represented a minority of total GFP⁺ cells. Among mature cells, we then focused on CD49d⁺ granulocytes, which we have recently found to be a provascular cell phenotype impaired in human and murine diabetes and rescued by dapagliflozin (27). After carotid endothelial injury, CD49d⁺ granulocytes (mostly Gr-1⁺) represented 4–5% of homed cells in nondiabetic mice, were significantly reduced in STZ diabetic mice to <2% (P < 0.001), and were rescued by dapagliflozin (Fig. 4F). To explore mechanistic salience of this finding, we performed additional experiments on CD49d⁺ cells. G-CSF induced a significant mobilization of CD49d⁺ neutrophils but not eosinophils, without any modification by STZ diabetes (Fig. 5A). We then examined if defective homing of CD49d⁺ cells was causally linked to impaired re-endothelization. We isolated Ly6G^lowSiglec-F⁺ granulocytes from GFP-transgenic mice, as a population enriched in CD49d⁺ cells, and examined their adhesion capacity (Fig. 5B). Blocking CD49d with a neutralizing monoclonal antibody (mAb) impaired adhesion of these cells to fibroblasts in vitro (Fig. 5C). Then, we blocked CD49d in vivo by injecting the mAb in nondiabetic mice undergoing carotid endothelial injury. Compared with vehicle-treated mice, those receiving anti-CD49d mAb displayed a significantly lower re-endothelization (Fig. 5D), suggesting that homing of CD49d⁺ cells is required for normal endothelial healing.

Endothelial healing occurs through two different mechanisms: 1) migration of resident endothelial cells; and 2) recruitment of circulating BM-derived cells. As CXCR4 is involved in these processes (28,29), we investigated whether CXCR4 was implicated in the effects of dapagliflozin on BM cell homing and endothelial repair. To this end, we blocked CXCR4 signaling with AMD3100 at a dose (1.25 mg/kg for 3 days) that did not stimulate HSPC mobilization and release of CD49d⁺ neutrophils (Supplementary Fig. 3). We first verified that in vitro AMD3100, at a concentration (?0.5 μmol/L) expected to be achieved in plasma after the in vivo dose we used (30), inhibited a typical intracellular signal CXCR4 elicited by the ligand CXCL12 (i.e., AKT phosphorylation on serine-473) (Fig. 6A and B) As expected, AMD3100 impaired in vivo endothelial healing at day 3 after carotid injury in nondiabetic mice. However, AMD3100 did not modify the effects of dapagliflozin on the recovery of endothelial healing in STZ diabetic mice (Fig. 6C and D). Notably, using BM-GFP⁺ chimeric mice, we found that AMD3100 impaired re-endothelization without affecting the homing of BM-derived cells in nondiabetic mice (Fig. 6E).

In view of the very low gene expression of SGLT2 on extrarenal tissues without protein detection and on the limited effects of dapagliflozin on hematopoiesis in baseline conditions, we hypothesized that the effects exerted by dapagliflozin on BM-derived cell traffic may be mediated at a systemic level. EPO regulates the mobilization of hematopoietic cells with vascular repair capacity (31,32), and red cell mass has been implicated in cardiovascular protection by SGLT2i (9,10). Thus, we first examined if dapagliflozin increased EPO gene expression and release. There was a trend toward increased EPO gene expression in the kidney of STZ diabetic mice and of increased plasma EPO concentrations. However, dapagliflozin exerted no significant effect on EPO gene expression (Supplementary Fig. 6A). Plasma EPO concentrations were significantly increased by dapagliflozin in nondiabetic mice but not in STZ diabetic mice (Supplementary Fig. 6B).

Since any systemic effect of SGLT2i cannot be separated from the improvement in blood glucose, we tested if insulin therapy mimicked the effects of dapagliflozin. We treated STZ diabetic mice with low-dose basal insulin to achieve a degree of glucose control that was similar to that obtained with dapagliflozin. At the end of treatment, blood glucose was reduced by 25% in insulin-treated mice compared with vehicle-treated mice but was still markedly elevated (on average 437 vs. 579 mg/dL) (Fig. 7A). Insulin also induced a recovery of body weight (Fig. 5B). HSPC mobilization after G-CSF was partially rescued in insulin-treated diabetic mice to an extent similar to that obtained with dapagliflozin (4.5 ± 1.7-fold increase) (Fig. 7C and D). Interestingly, insulin treatment also improved re-endothelization after carotid injury (Fig. 7E). These results suggest that the effects observed with dapagliflozin may be mediated, at least in part, by an improved glucose control.

Finally, we examined if ketogenesis induced by dapagliflozin played any role on HSPC traffic. To this end, we administered mice with a ketone drink to achieve the same blood levels of ketone bodies observed during dapagliflozin. With both dapagliflozin and the ketone drink, BOHB concentrations were increased about threefold compared with those observed in untreated diabetic mice (Fig. 7F), yet at much lower levels (?0.3 mmol/L) than those typically occurring during human diabetic ketoacidosis. The ketone drink exerted no effect on glucose levels (Fig. 7G). In this condition, HSPC mobilization after stimulation with G-CSF was rescued to 4.5 ± 0.4-fold in diabetic mice (Fig. 7H and I). Furthermore, artificial elevation of BOHB with the ketone drink led to an improvement in the extent of carotid endothelial healing in diabetic mice (Fig. 7J) that was similar to that observed with dapagliflozin.

The mechanisms by which SGLT2i improves cardiovascular outcomes remains largely unknown (4). Alterations in BM-derived cell kinetics, resulting in the pauperization of circulating HSPCs, contribute to micro- and macrovascular diabetic complications. In this study, we show that therapy with the SGLT2i dapagliflozin was able to reverse, at least in part, mobilization and homing of BM-derived cells to sites of vascular damage, thereby improving endothelial repair.

We used the T1D-like mouse model because it is associated with a complete defect in HSPC mobilization and a severe impairment in vascular healing, whereas the same could not be shown for a T2D-like model. Although initially developed for the treatment of T2D, dapagliflozin can also be used for T1D because its glucose-lowering action is independent from insulin secretion (33). Furthermore, there is solid demonstration that cardiorenal benefits of SGLT2i, which were initially shown in patients with T2D (3,34), also apply to individuals without diabetes (35,36). To minimize the confounding role of glucose control, we chose a lower dapagliflozin dose than that needed to normalize glucose levels in mice. Indeed, most extraglycemic effects of SGLT2i are independent from the degree of glucose control (35,37).

In STZ-induced diabetes, the complete unresponsiveness to G-CSF is ideal to test pharmacologic strategies to rescue HSPC mobilization and improve complications. The observation that SGLT2 is expressed in the BM, though at low levels, provides a rationale for evaluating the effects of dapagliflozin on BM cell traffic. In agreement with the notion that CD169⁺ macrophages retain HSPCs in the BM niche (38), dapagliflozin prevented the increase in CD169⁺ macrophages seen in diabetic mice and recovered the surge in HSPCs after G-CSF toward normal levels. CD169⁺ macrophages display proinflammatory M1 features (25), which rely on glycolytic metabolism (39). We speculate that attenuation of diabetes-induced raise in CD169⁺ macrophages by dapagliflozin resulted from blunting myelopoiesis though specific metabolic effects, including induction of ketogenesis, which may counter M1 polarization (40).

We then evaluated if improving the traffic of BM-derived cells would translate into vascular benefits. We used an endothelial injury model in which healing relies on two processes: 1) activation of local endothelial cells; and 2) recruitment of circulating BM-derived cells (28,41) that we tracked in BM-GFP⁺ chimeric mice. Diabetes strongly reduced homing of BM-derived cells to the injured endothelium and delayed healing. Treatment of diabetic mice with dapagliflozin rescued homing of BM-derived cells and improved endothelial healing. Interestingly, among mature cells homed to the vasculature, dapagliflozin was particularly effective in increasing the amount of CD49d⁺ granulocytes. We have recently shown that CD49d⁺ granulocytes exert proangiogenic activities and are reduced in the blood of murine and human diabetes, possibly due to impaired generation from BM precursors (27). Of note, we already observed that dapagliflozin increased the levels of circulating CD49d⁺ granulocytes in patients with diabetes (27), and we now show that it can improve their homing to sites of vascular damage. Remarkably, artificially inhibiting CD49d cell adhesion by blocking antibodies prevented normal endothelial healing. Therefore, the ability of dapagliflozin to rescue recruitment of CD49d⁺ granulocytes can be mechanistically linked to the improved endothelial repair.

Although endothelial repair was significantly improved by dapagliflozin, it was still lower than in nondiabetic mice, suggesting that only part of the defect was countered. We thus wanted to understand the relative contribution of resident and circulating cells to the degree of endothelial repair promoted by dapagliflozin. We found that blocking CXCR4 with low-dose AMD3100 reduced endothelial healing in nondiabetic mice without affecting BM-derived cell mobilization and homing, thereby arguably acting on resident endothelial cells. Of note, AMD3100 did not abolish the beneficial effect of dapagliflozin on endothelial repair in diabetic mice. Thus, the partial improvement in re-endothelization by dapagliflozin can be explained by a recovery of BM-derived cell homing without an effect on resident endothelial cells, which, otherwise, would be prevented by concomitant treatment with AMD3100.

Although the SGLT2 gene was expressed in the BM, its levels were 10⁻⁴ times lower than in the kidney, and the protein could not be detected, implying that a strong effect of dapagliflozin on the BM independently from the effect on the kidney seems unlikely. We thus explored whether systemic changes that are typically induced by SGLT2i through its action on the kidney could mimic the effects observed with dapagliflozin. After ruling out a substantial effect on EPO, we focused on glucose control to evaluate if the mild glycemic improvement induced by low-dose dapagliflozin was at least in part responsible for the improved BM-derived cell traffic and vascular repair. Using a protocol of insulin therapy that reproduced the same degree of glucose reduction obtained with dapagliflozin, we observed a similar improvement in HSPC mobilization and endothelial repair. These data suggest that the mild improvement in glucose control during dapagliflozin might have contributed to the observed effects on BM cell traffic and vascular repair. Yet, glucose levels were far from normal, and a direct effect of insulin on the BM and vascular repair cannot be ruled out (42). Furthermore, body weight effects were strikingly different between dapagliflozin and insulin therapy. In fact, STZ-induced diabetes in mice leads to weight loss, a feature of starvation that can be affected by SGLT2i (43) and insulin in opposite ways. Of note, the same results seen with dapagliflozin and insulin were mimicked by treating mice with a ketone drink to achieve similar levels of ketone body elevation as observed during dapagliflozin, without lowering glucose levels. Lactate has been recently shown to regulate BM-cell traffic (44), so that similar effects may be carried out by BOHB. The molecular and metabolic pathways involved in these findings are particularly intriguing and deserve further investigation. Overall, our data are consistent with the view that multiple metabolic effects, carried out at a systemic level, are likely involved.

In summary, we uncover a new mechanism by which dapagliflozin could exert beneficial effects on the macro- and microvasculature. By rescuing mobilization of immature cells from BM to the bloodstream and by improving homing of mature cells to sites of vascular damage, dapagliflozin promoted endothelial healing. Such recovery of endogenous repair is likely to be carried out through multiple mechanisms, including the improvement of glucose control and the induction of ketogenesis.

Funding. This study was supported by grants from: University of Padova (DOR), Ministry of University and Education (PRIN projects 2015ZTT5KB and 201793XZ5A) and AstraZeneca (NCR-17-12732).

The external sponsor had no role in study design, conduction, data analysis, interpretation, and decision to publish.

Duality of Interest. A.A. and G.P.F. received honoraria or lecture fees from AstraZeneca, Boehringer Ingelheim, Mundipharma, and MSD, the manufacturers of SGLT2i. No other potential conflicts of interest relevant to this article were reported.

Author Contributions. M.A. performed experiments, researched and analyzed data, and wrote the manuscript. S.T. performed experiments, researched and analyzed data, and revised the manuscript. F.I.A. performed experiments, researched and analyzed data, and revised the manuscript. M.D’.A. performed experiments, researched and analyzed data, and revised the manuscript. L.M. performed experiments, researched and analyzed data, and revised the manuscript. G.Z. performed experiments, researched and analyzed data, and revised the manuscript. A.R. designed experiments, contributed to discussion, and revised the manuscript. R.C. performed experiments, researched and analyzed data, and revised the manuscript. A.A. provided supervision and funding, contributed to discussion, and revised the manuscript. G.P.F. researched and analyzed data, provided supervision and funding, and wrote the manuscript. M.A. and G.P.F. are the guarantors of this work and, as such, had full access to all of the data in the study and take responsibility for the integrity of the data and the accuracy of the data analysis.