Tacrolimus-Induced BMP/SMAD Signaling Associates With Metabolic Stress–Activated FOXO1 to Trigger β-Cell Failure

The calcineurin inhibitor tacrolimus (also known as FK506) is the most widely used immunosuppressive drug in solid organ transplantation, but 40% of organ recipients on tacrolimus develop posttransplant diabetes mellitus (PTDM) (1). Understandably, prevention of graft rejection prevails over risk of developing PTDM. However, diabetes diminishes quality of life while it increases risk of graft loss, mortality, and health care expenditure (2). Therefore, deciphering the mechanisms underlying diabetogenicity of tacrolimus is essential to develop novel strategies to prevent PTDM while upholding the efficacy of immunosuppression.

Tacrolimus exerts its immunosuppressive action on T-cells through binding FK506-binding protein 12 (FKBP12) and inhibiting calcineurin activity, which prevents nuclear translocation of the nuclear factor of activated T-cells (NFAT), and the subsequent boost in inflammatory cytokines (3). Inhibition of calcineurin/NFAT has also been proposed as a main mechanism for PTDM, since calcineurin signaling has been shown to regulate β-cell function and proliferation (4,5). However, recipients on tacrolimus develop diabetes more often than recipients on cyclosporin A (CsA) (6). Tacrolimus is often described as a more potent calcineurin inhibitor (7), but comparable levels of calcineurin inhibition (and nuclear NFAT) have been found for tacrolimus and CsA both in vitro and in vivo (8,9). Furthermore, we observed that metabolic stress predisposes β-cells to failure upon tacrolimus treatment (8,10). Importantly, these experimental evidences are in line with clinical observations indicating that PTDM is predominantly observed in patients on tacrolimus compared with other immunosuppressive drugs and in patients displaying features of metabolic syndrome (11). Overall, these findings indicate that other pathways beyond calcineurin/NFAT inhibition are involved in tacrolimus-induced diabetes.

Recent studies show that tacrolimus activates bone morphogenetic protein (BMP) signaling in hepatocytes and pulmonary endothelial cells (12,13). FKBP12 prevents phosphorylation of transforming growth factor-β family type 1 receptors in absence of its exogenous ligand, e.g., BMP (14). Therefore, binding of tacrolimus to FKBP12 may competitively release FKBP12 from the receptors, leading to activation of the downstream receptor regulated (R-)SMADs.

Here, we identified, in primary human β-cells, a novel mechanism of action of tacrolimus through activation of BMP/SMAD signaling and a subsequent loss of β-cell maturity and function.

Human islet isolations from cadaveric human organ donors were performed in the Good Manufacturing Practice facility of the Leiden University Medical Center according to the method used in the center for the procurement of clinical-grade material (15). Islets were used for research only if they could not be used for clinical purposes and if research consent had been obtained according to national laws. Data from 13 human islet preparations are shown in this study (see human islet checklist in the Supplementary Data). Given that the availability of human islets is limited, not all read-outs could be used for each batch of islets used. Islets were cultured in CMRL 1066 medium (5.5 mmol/L glucose) containing 10% human serum, 20 mg/mL ciprofloxacin, 50 mg/mL gentamycin, 2 mmol/L l-glutamine, 0.25 mg/mL fungizone, 10 mmol/L HEPES, and 1.2 mg/mL nicotinamide in a humidified atmosphere with 5% CO₂.

Treatments with glucose and palmitate were performed using a clinical-grade glucose stock solution of 2.2 mol/L and a palmitate stock solution of 10 mmol/L prepared following the protocol described by Cousin et al. (16). Tacrolimus (TAC), CsA, and LDN193189 (LDN) were dissolved first in ethanol (25 mg/mL) and then diluted with PBS to prepare the final stock solutions (1 ng/µL TAC, 50 ng/µL CsA, and 100 µmol/L LDN). VIVIT peptide was directly diluted in phosphate buffer saline to a stock solution of 100 µmol/L. Medium and treatments were refreshed daily.

Approximately 50 islet equivalents (IEQ) per well were placed in a 96-well transwell plate. Islets were preincubated for 90 min in low-glucose (1.67 mmol/L) secretion assay buffer at pH 7.4 containing 11.5 mmol/L NaCl, 0.5 mmol/L KCl, 2.4 mmol/L NaHCO₃, 2.2 mmol/L CaCl₂, 1 mmol/L Mgcl₂, 20 mmol/L HEPES, and 0.2% human serum albumin. Islets were subsequently transferred to low (1.67 mmol/L) glucose-containing buffer for 60 min and then to high (20 mmol/L) glucose condition for 60 min. Insulin secretion was assessed using a human insulin ELISA kit (Mercodia).

Human islets were washed in PBS and total RNA was extracted immediately using RNeasy kit (Qiagen) according to the manufacturer’s protocol. Total RNA (1 µg) was reversed transcribed using M-MLV retro-transcriptase (Invitrogen). Quantitative PCR was performed in a CFX system (Bio-Rad). Fold changes were calculated using the ΔCt method with human β-2-microglobulin as reference gene. Primers used are listed in Supplementary Table 3.

Human islets were washed in PBS and fixed in 4% formaldehyde for 20 min. Formaldehyde was removed and islets were included and spun down at 20,000 rcf in 2% melted agar. Agar-containing islet pellets were embedded in paraffin. Paraffin blocks were cut in 4-µm sections. Following rehydration of sections, antigen retrieval was performed by heating slides in citrate buffer (pH 6.0) using a pressure cooker. Tissue was then permeabilized in 0.1% triton X-100 and, blocked with donkey serum, and primary antibodies were incubated overnight at 4°C. Secondary antibodies (Alexa Fluor; Thermo Fisher Scientific) were incubated for 2 h at room temperature and protected from light. DAPI-containing mounting medium (Vector) was used as nuclear counterstaining. All sections were examined using confocal microscopy (SP8 WLL; Leica).

For quantifications of nuclear localization of target proteins FOXO1, MAFA, PDX-1, and SMADs in β-cells, two to four sections were evaluated per target, per condition, and per donor. To avoid the staining of the same islet two times we kept an interval of at least 200 μm between sections. Each section was assessed entirely to prevent potential bias due to selections of a subset of islets. Nuclei of β-cells were identified with anti-NKX6.1 IgG, which was used in combination with the corresponding antibody against FOXO1, MAFA, PDX-1, phosphorylated (phospho)-SMAD1/5, or SMAD4 (Supplementary Table 1). Quantifications were performed by counting at least 500 NKX6.1⁺ cells per donor, condition, and target. Proportion of double positive nuclei (FOXO1⁺NKX6.1⁺, MAFA⁺NKX6.1⁺, PDX-1⁺NKX6.1⁺, and SMADs⁺NKX6.1⁺) is represented relative to total number of NKX6.1⁺ cells. Nuclear staining was only considered positive if there was a clear and uniform signal from the nucleus. Investigators were blinded to experimental conditions during quantification of staining.

Human isolated islets (3,000 IEQ per condition) were washed with PBS and lysed in 3× SDS loading buffer (100 mmol/L Tris-Cl, 4% SDS, 20% glycerol, 200 mmol/L dithiothreitol, and 0.2% bromophenol blue) by passing 10 times through a 25-gauge needle and boiling for 10 min at 95°C. Lysates were centrifuged, and 10 μL of supernatant was loaded into a SDS-PAGE gel and tested for immunoblotting with antibodies against phospho-SMAD1/5, MAFA, and tubulin. Horseradish peroxidase–conjugated anti-rabbit or anti-mouse was used as secondary antibody, and enhanced chemiluminescent substrate was used to develop the signal (Supplementary Table 1).

Proximity ligation assay (PLA) experiments were carried out using Duolink by Sigma-Aldrich according to the manufacturer’s instructions. Islets were processed as described for immunofluorescence staining with an additional permeabilization step in cold methanol. Antibodies anti-SMADs and FOXO1 (Supplementary Table 1) were incubated overnight at 4°C. Incubation with probes as well as ligation procedures was performed following instructions by the manufacturer, and an amplification step was carried out overnight at 37°C. A method for quenching lipofuscin autofluorescence was used at the end of the procedure (17). PLA signal was detected using confocal microscopy (SP8 WLL; Leica).

Gene expression counts were used that were generated in a previous study from our group by Muraro et al. (18) and deposited on the NCBI Gene Expression Omnibus database with accession number GSE85241. The transcript counts of donors 29, 30, and 31 were filtered and normalized using R. Cells containing at least 1,500 genes and 5,000 transcripts expressed in at least 5% of cells were included, leaving a total of 1,974 cells. The transcript counts of remaining cells were normalized by down-sampling to 5,000 transcripts per cell. Dimensionality reduction was applied (t-SNE) using the Rtsne package, and cell-type clusters were identified using typical marker genes. More specifically, cluster-wide expression on the t-SNE of insulin, glucagon, somatostatin, pancreatic polypeptide, CD24, and type 1 collagen α-1 was used to mark β-, α-, δ-, PP-, exocrine, and mesenchymal cells, respectively.

Transplantation experiments were performed in 8- to 10-week-old male C57BL6 Rag2^−/− immunodeficient mice (Charles River). Mice were cared according to institutional guidelines. Transplantation was performed under isofluorane anesthesia; 3,000 human IEQ were transplanted under the kidney capsule. Animals were then separated in two groups, six animals per group, and after recovery from surgery were fed ad libitum with high-fat diet (HFD) pellets (rodent diet with 60 kcal% fat; Research Diets) for 13 weeks. Animals on tacrolimus received an intraperitoneal dose from week 10 of 0.3 mg/kg × day based on previous studies (10). Random and fasting blood glucose was measured weekly during the first 10 weeks and then twice a week during treatment (tail vein). Intraperitoneal glucose tolerance tests (IPGTTs) were performed in overnight-fasted animals. Blood samples were drawn from the tail vein before injecting 2 g/kg glucose and after 15, 30, 60, and 120 min in the animals. Blood glucose concentrations were measured using a glucose meter. Human insulin concentrations were measured by ELISA (Mercodia). At the end of treatments, grafts were removed, fixed, and embedded in paraffin for immunostaining.

Normality for all variables was tested using Kolmogorov-Smirnov test. When variables were distributed normally statistical significance was determined by Student t test or ANOVA followed by Tukey multiple comparison tests, as appropriate. If the variable did not pass the normality test, then Kruskal-Wallis and Dunn’s multiple comparisons tests were used. Data from every donor are represented as single points. When data are depicted in boxplot diagrams, the box represents the median and the interquartile range and the plus symbol the mean. Whiskers represent minimum and maximum values. In all histogram representations, values are expressed as mean ± SEM. A P value lower than 0.05 was considered significant.

All experimental procedures involving animals were approved by the ethics committee of the Leiden University Medical Center.

Data sets generated during and/or analyzed during the current study are available from the corresponding author upon reasonable request.

Clinical and preclinical data indicate that diabetogenicity of tacrolimus is more pronounced in individuals presenting an insulin resistance condition. To explore mechanisms underlying the effect of tacrolimus in a metabolic milieu, we first established a culture model to emulate mild metabolic stress in primary human islets. Islets were initially cultured in media containing increasing levels of glucose (5.5, 11, and 22 mmol/L) combined with palmitate (0, 100, 200, and 500 µmol/L) for 72 h. We selected 11 mmol/L glucose and 200 µmol/L palmitate for 72 h as optimal stress condition based on the following criteria: 1) no increased apoptosis (Supplementary Fig. 1A), 2) an increased frequency in nuclear localization of FOXO1 as marker of β-cell stress (Supplementary Fig. 1B), and 3) no changes in nuclear localization patterns of key β-cell markers MAFA and PDX-1 (Supplementary Fig. 1B).

Next, we evaluated the effect of tacrolimus on human β-cells in this culture system. We first assessed β-cell apoptosis using concentrations of tacrolimus ranging from 0.2 to 25 ng/mL. We selected a tacrolimus concentration of 2 ng/mL for further experiments based on levels of β-cell apoptosis, which were similar to those in controls (Supplementary Fig. 1C). This dose is also comparable to free plasma levels in patients, which is ∼20% of commonly used target whole blood trough levels after transplantation as the remainder is bound to erythrocytes (19).

Tacrolimus treatment alone did not affect the insulin secretory capacity of treated β-cells, whereas glucose and palmitate (GP) induced a significant reduction in glucose-stimulated insulin secretion compared with control (Supplementary Fig. 1D). Simultaneous treatment with glucose, palmitate, and tacrolimus (GP-TAC) induced a complete abrogation of glucose-stimulated insulin secretion response in human islets (Fig. 1A). To investigate whether deleterious effects of tacrolimus on insulin secretion were not only due to inhibition of calcineurin/NFAT pathway in human β-cells, we tested another calcineurin inhibitor, CsA (100 ng/mL GP-CsA), and the NFAT inhibitor, VIVIT peptide (1 µmol/L GP-VIVIT) (20). None of these treatments affected β-cell function to a similar extent (Fig. 1A).

We subsequently investigated the effects of tacrolimus with glucose and palmitate on nuclear localization of key β-cell transcription factors. Using NKX6.1 as a marker of β-cell nuclei, we observed a reduction of ∼50% in β-cell nuclei (NKX6.1⁺) positive for MAFA (85.8 ± 5.2% vs. 41.1 ± 19.0%, P < 0.001, for nontreated control [NTC] vs. GP-TAC) (Fig. 1B and C). Additionally, we observed a slight but significant decrease in frequency of NKX6.1⁺ nuclei showing also positivity for PDX-1 in the GP-TAC condition compared with untreated controls or the GP condition (Fig. 1D and E). Furthermore, reduction in nuclear staining for MAFA and PDX-1 was accompanied by a reduction in their gene expression (Fig. 1F and G). Importantly, we also showed that tacrolimus treatment in a normal medium (TAC) did not cause alterations in nuclear staining or gene expression for MAFA and PDX-1 (Fig. 1B–G). Altogether these data indicate that deleterious effects of tacrolimus in human β-cells are dependent upon coexisting mild metabolic stress conditions and that pathways beyond calcineurin/NFAT inhibition are involved.

To validate the effect of tacrolimus on human β-cells under metabolic stress in vivo, we transplanted primary human islets (3,000 IEQ) in immunodeficient mice. Mice were subsequently subjected to an HFD, which was maintained for 90 days (Fig. 2A). Body weight increased significantly (47.1 ± 1.5 g at day 70 vs. 29.5 ± 1.6 g at day 0; P < 0.0001) (Supplementary Fig. 2A), while blood glucose levels, determined by both fasting and random measurements, remained in normal range (Supplementary Fig. 2B). At day 70, a group of mice was treated with tacrolimus (HFD + TAC) for 20 days. Untreated mice were used as control (HFD) (Fig. 2A). At the end of the treatment, all animals in both groups showed nuclear FOXO1 staining, indicating the presence of metabolic stress in human β-cells (Supplementary Fig. 2C). To evaluate glycemic control in the two groups, an IPGTT was performed at the end of the 90-day period. HFD mice recovered basal levels of glycemia 2 h after a glucose challenge, whereas HFD + TAC animals displayed blood glucose levels above 20 mmol/L for 2 h after glucose injection, indicating an inability to manage the glucose challenge (Fig. 2B and C). The early insulin secretory response during the IPGTT was increased by 1.7× in the HFD group, whereas no increase in human insulin was observed in HFD + TAC mice (Fig. 2D and E). Importantly, human β-cell dysfunction observed in HFD + TAC animals was associated with reduced frequency in β-cells with a positive nuclear staining for the transcription factor MAFA (Fig. 2F and G). Insulin secretion by murine pancreata was similar between treatments and with no manifest peak upon a glucose challenge (Supplementary Fig. 2D). This might be due to an earlier response of human islets to glucose increase because of their lower glucose threshold for insulin secretion as compared with murine islets (21). These results show that tacrolimus adversely affects human β-cell maturity and function also in an in vivo model of metabolic stress.

Next, we investigated the molecular mechanisms underlying tacrolimus-induced β-cell dysfunction. Previous studies showed that tacrolimus can activate BMP signaling in other cellular models. Single-cell transcriptomics data show that human β-cells express all necessary components for BMP signaling: BMP type 1 receptor genes ACVR1 (ALK2) and BMPR1 (ALK3); BMP type 2 receptor BMPR2; transcription factors SMAD1, SMAD5, and SMAD8; and canonical BMP target gene ID1 (Supplementary Fig. 3A). In contrast, we did not detect transcripts for type 1 receptor ACVRL1 (ALK1) (data not shown).

We investigated whether tacrolimus activates BMP signaling in primary human islets, and more specifically in β-cells, by examining expression level of phospho-SMAD1/5, intracellular effectors of the BMP pathway (22). We detected the presence of phospho-SMAD1/5 in NTC, probably due to endogenous expression of BMPs or their presence in the medium; however, levels were not enough to induce their nuclear accumulation in β-cells (Fig. 3A). Treatment with tacrolimus for 72 h increased the frequency of human β-cells (NKX6.1⁺) with positive nuclear staining for phospho-SMAD1/5 in both normal medium (TAC, 72.6 ± 2.63%) and medium containing glucose and palmitate (GP-TAC, 74.0 ± 13.1%) as compared with both untreated control (NTC, 16.4 ± 3.19%) and glucose and palmitate alone (GP, 18.73 ± 1.99%) (Fig. 3A and B). Tacrolimus-treated islets also showed an increase in ID1 gene expression, a direct target gene of BMP/SMAD signaling (23), in both normal medium (TAC) and glucose and palmitate condition (GP-TAC) (Fig. 3C). Importantly, treatment with cyclosporin (CsA and GP-CsA) neither induced expression of the ID1 gene nor resulted in nuclear accumulation of phospho-SMAD1/5 (Fig. 3A–C). Additionally, we observed that nuclear staining for phospho-SMAD1/5 was strongly increased in transplanted human β-cells (NKX6.1⁺) in mice from the HFD + TAC group compared with the HFD control group, thereby validating activation of the BMP/SMAD pathway by tacrolimus in vivo (Supplementary Fig. 3B).

Next, to determine whether tacrolimus-induced BMP activation is responsible for the deleterious effects of tacrolimus on β-cells, we tested the BMP type 1 receptor kinase inhibitor LDN (120 nmol/L) (24). As expected, LDN treatment prevented nuclear accumulation of phospho-SMAD1/5 produced by tacrolimus (GP-LDN-TAC). In this condition, we observed no increase in nuclear staining for phospho-SMAD1/5 in β-cells (Fig. 3A and B) or phospho-SMAD1/5 levels in whole islets (Fig. 3E and Supplementary Fig. 4B and E). The increased ID1 islet gene expression was also prevented (Fig. 3C). Remarkably, LDN maintained insulin secretory capacity of human islets exposed to GP-TAC to a level similar to that of islets exposed to GP only (Fig. 3D). Notably, LDN protected β-cells from loss of MAFA protein levels (Fig. 3F and Supplementary Fig. 4A and D), nuclear localization (Fig. 3G and H), and gene expression (Fig. 3I). Additionally, PDX1 nuclear staining and gene expression were also maintained (Fig. 3J–L). These results indicate that tacrolimus-induced β-cell dysfunction is dependent upon metabolic stress and is mediated by activation of the BMP/SMAD signaling pathway.

We considered simultaneous activation of FOXO1 and SMAD transcription factors in the GP-TAC condition as a potential mechanism to explain the synergy between tacrolimus and metabolic stress. Receptor-activated SMAD1/5 as well as the common SMAD mediator, SMAD4, was detected, indicating the presence of all canonical effectors for BMP signaling. First, we confirmed nuclear presence in β-cells of both FOXO1 and SMADs only in the GP-TAC condition (Supplementary Fig. 5A and B). Next, we performed a PLA with endogenous FOXO1 and SMAD1/5 transcription factors as targets. These data indicate the formation of a complex between these proteins only in the GP-TAC condition, and not in the case of sole FOXO1 (GP) or SMADs (TAC) activation (Fig. 4). Importantly, formation of this complex was prevented by LDN treatment. Therefore, synergistic effects of tacrolimus and metabolic stress on primary human β-cells concur with a direct interaction between active SMADs and FOXO1.

Finally, to validate the clinical significance of our findings, we examined pancreatic protocol biopsies from recipients who had undergone pancreas transplantation. Protocol biopsies of pancreas recipients on CsA are, however, uncommon. Nevertheless, we obtained five biopsies from pancreas recipients on tacrolimus and two from patients on CsA. Moreover, we used pancreata from five nondiabetic deceased donors as control samples (Supplementary Table 2). We observed that frequency of β-cells with nuclear phospho-SMAD1/5 was higher in tacrolimus-treated patients compared with CsA-treated patients and control subjects (88.3% ± 12.4% vs. 47.3% ± 21.4% vs. 13.9% ± 14.7, respectively) (Fig. 5). Of note, despite the fact that patients on tacrolimus had a higher frequency of nuclear phospho-SMAD1/5–positive cells, all of them presented with normoglycemia. This was in line with normal (i.e., cytoplasmic) localization of FOXO1, as well as normal nuclear frequency of MAFA in β-cells observed in all protocol biopsies (Supplementary Fig. 6).

Calcineurin inhibitors have boosted the success of organ transplantation, with tacrolimus showing better rejection rates than with CsA (2) but also more drug-induced diabetes (6,25). Here, we present a mechanism underlying tacrolimus-induced β-cell failure in primary human islets.

Tacrolimus binds FKBP12 to inhibit calcineurin (26), whereas FKBP12 prevents type 1 BMP receptors constitutive phosphorylation by type 2 receptors (12,13). Through this mechanism, tacrolimus is a strong BMP activator (13,27). Primary human β-cells express type 1 receptors ACVR1 (ALK2) and BMPR1A (ALK3), as well as type 2 receptor BMPR2 and SMADs. However, there is a controversy about whether active BMP signaling is beneficial or detrimental to β-cells. Goulley et al. (28) showed that BMP4 improved insulin secretion in murine β-cells. In contrast, Bruun et al. (29) showed that in vitro administration of BMPs to rat and human islets had negative effects on proliferation and insulin secretion. Therefore, the role of BMP signaling in β-cells appears to be context-dependent.

Our study shows that tacrolimus activates BMP/SMAD signaling in β-cells, which is deleterious when acting on top of metabolic stress. Combined effects of tacrolimus and metabolic stress seem to converge in a (dys)functional synergy between SMADs and FOXO1 that interact in the nucleus. The formation of this complex is associated with reduced expression of the master regulator MAFA and loss of insulin secretory capacity (Fig. 6). This is supported by the use of a highly selective kinase inhibitor of BMP type 1 receptor (LDN), which prevented SMAD-FOXO1 interaction and protected human β-cells from deleterious effects of tacrolimus.

Interaction between FOXOs and SMADs has been reported in keratinocytes and glioblastoma cells (30,31). Both families of transcription factors act in multiprotein complexes to integrate different cellular pathways. In β-cells, nuclear shuttling of FOXO1 is a well-known response to metabolic challenges (32), albeit it is unclear whether this constitutes a beneficial or detrimental response for β-cells. FOXO1 nuclear shuttling has been shown to protect murine β-cells by increasing expression of NeuroD and MAFA in response to oxidative stress (33). On the other hand, Kitamura et al. (34) showed that reduced levels of FOXO1 reversed β-cell failure in an IRS2 knockout mouse model, whereas nuclear FOXO1 preceded loss of nuclear MAFA in the db/db mouse (35), and its chemical inhibition improves glycemia (36). This apparent discrepancy on the role of FOXO1 in β-cells is likely to be explained by its interactions with other signaling pathways, such as tacrolimus-activated BMP/SMAD as we show in this study. Moreover, it also underscores the importance of the use of primary human material for evaluation of these mechanisms, even when it implies limited and heterogeneous tissue availability. Additionally, activation of BMP/SMAD signaling by tacrolimus in β-cells may be clinically relevant, since it can be detected in protocol pancreas biopsies from patients who underwent pancreas transplantation and received a tacrolimus-based immunosuppressive regimen. The fact that these recipients were normoglycemic and displayed normal MAFA expression and cytoplasmic FOXO1 is in line with the hypothesis that an interaction between tacrolimus and metabolic stress is needed to produce β-cell failure. However, some of the deleterious effects of tacrolimus in β-cells may be due to alterations produced on mitochondrial calcium uptake, which directly affects oxygen consumption and insulin secretion (37), and we cannot rule out potential negative effects of tacrolimus alone on β-cells with long-term exposures or higher doses.

This study has potential limitations. We have not elucidated the precise mechanism by which interactions between activated SMAD and FOXO1 trigger the loss of MAFA and subsequent β-cell failure. In addition, a potential source of bias arises from the islet preparations used. We used high-purity islet preparations from nondiabetic donors, but we noticed that 69% of them had a BMI >25 kg/m² while 38% had BMI >30 kg/m². Although this was not intentional, it might have some relevance. Finally, systemic BMP inhibition is unlikely to be applicable in a clinical setting due to its broad implications, and more studies are required to determine the potential impact of BMP inhibition on immunosuppression efficacy of tacrolimus.

In conclusion, under conditions of metabolic stress, tacrolimus induces loss of MAFA, a key transcription factor, loss of which has been directly associated with loss of β-cell function in diabetes (38–41). Accordingly, PTDM is regarded by clinicians as a drug-accelerated form of type 2 diabetes (1). Both type 1 and type 2 diabetes are characterized by low-grade inflammation (42,43), and proinflammatory conditions have been associated with activation of BMP pathway (44). The mechanism we are presenting here may, therefore, lead to the identification of novel targets for maintenance of β-cell maturity in all forms of diabetes. Yet, considering that both FOXO1 and BMP/SMAD pathways are implicated in a great variety of cellular processes, the challenge in the near future will consist in designing interventions that will be either cell type or context specific.

Acknowledgments. The authors thank the pancreas donors and their families. The authors also thank Annemieke Tons (Leiden University Medical Center) for helping with staining and Natascha de Graaf (Leiden University Medical Center) for helping with gene expression analysis.

Funding. This work was financially supported by the Dutch Kidney Foundation (DKF 15O130). J.T. received travel grants from the European Foundation for the Study of Diabetes and the European Society for Organ Transplantation. F.C. received grants from JDRF, the Dutch Diabetes Research Foundation, and the DON foundation.

Duality of Interest. No potential conflicts of interest relevant to this article were reported.

Author Contributions. J.T. and M.H. performed the experiments. J.T., F.C., and A.P.J.d.V. conceived the project and drafted the manuscript. P.t.D. provided expertise on BMP signaling. N.G. performed the single-cell transcriptomics analysis. E.P. and A.E.R.-R. participated in the conception of the idea. C.D. provided human pancreatic biopsies. T.J.R. provided the infrastructure for human pancreatic tissue isolation. E.d.K. gave conceptual advice. F.C. and A.P.J.d.V. are the guarantors of this work and, as such, had full access to all the data in the study and take responsibility for the integrity of the data and the accuracy of the data analysis.

Prior Presentation. Parts of this study were presented at the annual conference of the European Pancreas and Islet Transplantation Association, Igls, Austria, 27–29 January 2019, and at the annual conference of the International Pancreas and Islet Transplantation Association, Lyon, France, 2–5 July 2019.