Long-Term IKK2/NF-κB Signaling in Pancreatic β-Cells Induces Immune-Mediated Diabetes

Male mice were housed under specific pathogen-free conditions at the animal facility of the University of Ulm. Pdx-1.tTA mice (C57BL/6) (6) and (tetO)₇.IKK2-CA (constitutively active mutant of human IKK2) mice (NMRI) (7) were described previously. Pdx-1.tTA mice are knockin animals in which the coding sequence of tetracycline-dependent transactivator (tTA) has replaced the endogenous Pdx-1 gene, thereby rendering this mouse line heterozygous for Pdx-1 (Pdx-1^+/−). Control mice group include wild-type and single-transgenic (tetO)₇.IKK2-CA mice unless otherwise stated. Experiments were performed in accordance with institutional guidelines and German animal protection law.

Blood glucose was measured using the One-Touch Ultra glucometer (LifeScan Inc., Mipitas, CA). Pancreatic insulin was extracted by overnight agitation with cold acid ethanol (0.18 mol/L HCl in 70% ethanol) at 4°C. The supernatant was collected, and the pellet was re-extracted. The pooled supernatant was used for insulin measurement. Insulin was determined in plasma samples containing protease inhibitor and pancreatic extracts using the Ultra-Sensitive Mouse Insulin ELISA Kit (Chrystal Chem Inc.) following the manufacturer’s instructions.

Pancreata were snap-frozen in liquid nitrogen and pulverized, and proteins were extracted for Western immunoblotting and luciferase activity measurement (7). For immunoblotting, antibodies against IKK2 (Abcam) and extracellular signal–related kinase-2 (Santa Cruz Biotechnology) were used. Electrophoretic mobility shift assay was performed as described (7) using whole-cell extract from isolated islets.

For paraffin sections, pancreata were excised, fixed overnight in 3.8% buffered formalin, dehydrated, paraffin embedded, cut in 3-μm sections, and further processed as previously described (8). For cryosections, 10-μm slices from natively frozen pancreata were fixed with 4% paraformaldehyde. Sections were incubated with the primary antibodies: rabbit (Cell Signaling Technology) or guinea pig (Abcam) anti-insulin, goat anti-human IKK2 (Santa Cruz Biotechnology), rabbit anti-RelA/p65 (Laboratory Vision), anti–Pdx-1 and antichromogranin A (Abcam), anti-Ki67 (Thermo Scientific), rat anti-CD4 and anti-CD8 (Abcam), anti–CD11c-phycoerythrin and anti-B220 (BD Biosciences), and anti–CD25-phycoerythrin and anti–major histocompatibility complex (MHC) class II (MHC II)–FITC (eBioscience). Secondary antibodies were coupled with Alexa Fluor (Invitrogen) for immunofluorescence and with horseradish peroxidase in immunohistochemistry that was developed by 3-amino-9-ethylcarbazole (DakoCytomation). Immunofluorescent stainings were visualized as before (8), and other stainings were analyzed on a Leica DM IRB microscope (Leica Microsystems) equipped with ProgRes C14 digital camera (Jenoptik) and Openlab software (Improvision).

In situ detection of DNA strand breaks was performed using the TUNEL labeling method with the FragEL DNA Fragmentation Detection Kit and colorimetric-TdT enzyme (EMD Millipore) according to the manufacturer’s instructions.

Pancreata were inflated in situ with 0.5 mg/mL ice-cold collagenase XI solution (Sigma-Aldrich, St. Louis, MO) dissolved in PBS (with Ca/Mg). Pancreatic tissue was dissected out and digested for 19–24 min at 37°C with subsequent washing and centrifugation. Islets in the sediment were purified with Histopaque 10771 (Sigma-Aldrich), washed, and frozen in liquid N₂.

RNA was extracted with RNeasy kit (Quiagen) and cDNA synthesis was done using the Transcriptor High Fidelity cDNA Synthesis Kit (Roche). Quantitative real-time PCR was performed with the Roche LightCycler 480 (Roche) using gene-specific primers and hydrolysis probes designed by the Roche Universal Probe Library system (Roche). Microarray analysis was performed with the Mouse Gene 1.0 ST array (Affymetrix) and evaluated with the “Genesifter” sofware (Geospiza). The expression data are available at Gene Expression Omnibus super series (accession number GSE47504).

Pancreatic cells were isolated and purified as previously described (9) and subsequently stained with monoclonal antibodies using standard procedures. The following antibodies were purchased from eBioscience: anti–MHC II (M5/114.15.2), anti–MHC class I (MHC I; AF6–88.5.5.3), anti-CD11b (M1/70), anti-CD3e (17A2), anti-CD25 (PC61.5), and anti-CD69 (H1.2F3); from BD Biosciences: anti-CD11b (M1/70), anti-CD8a (53–6.7), anti-CD4 (RM4–5), CD44 (IM7), anti-NK1.1 (PK136), and anti-CD11c (HL3). Anti-CD45 (30-F11), anti-CD19 (6D5), anti-F4/80 (BM8), and anti-Ly6G (1A8) were from BioLegend, while Ly6C (1G7.G10) was from Miltenyi Biotec. Fixable viability dye (eBioscience) was used to exclude dead cells. FACS was performed on an FACSCanto II (BD Biosciences), and data were analyzed with FACSDiva 6.2 (BD Biosciences) and FlowJo softwares (Tree Star).

Values are given as mean ± SEM. Statistical analysis was performed with the Prism software (GraphPad) using two-tailed Student t test. Data with P values of ≤0.05 were considered statistically significant.

To directly explore the biological consequences of canonical NF-κB activation in pancreatic β-cells, we generated IKK2-CA^Pdx-1. This was achieved by crossing transgenic mice expressing the tTA under the control of Pdx-1 promoter (Pdx-1^+/−) (6) with mice carrying the luciferase-(tetO)₇-IKK2-CA minigene (7) (Fig. 1A). The transgenic expression system was not activated until weaning in order to avoid any effects of IKK2-CA activity on pancreas development. Measurement of luciferase activity, the coexpressed reporter gene, revealed strong transgene activity in the pancreas of IKK2-CA^Pdx-1 mice after doxycycline withdrawal and only minor activity in the intestine (Fig. 1B). Pancreatic IKK2-CA expression was confirmed by immunoblot and could be switched off by doxycycline (Fig. 1C), thus allowing conditional regulation of IKK2 activity. Immunofluorescence staining demonstrated a mosaic expression of IKK2-CA exclusively in the islets of IKK2-CA^Pdx-1 mice (Fig. 1D), correlating with obvious reduction in insulin immunoreactivity (Fig. 1D).

Electrophoretic mobility shift assays showed strong basal NF-κB activity in IKK2-CA^Pdx-1 mice relative to their Pdx-1^+/− littermates, demonstrating transgene functionality (Fig. 1E). NF-κB activity was also elevated in Pdx-1^+/− mice compared with controls. Furthermore, IKK2 activation led to nuclear translocation of the NF-κB subunit RelA in β-cells of IKK2-CA^Pdx-1 mice, while almost no nuclear RelA was detected in islets of either Pdx-1^+/− or control mice (Fig. 1F).

We then assessed the physiological and histological consequences of IKK2/NF-κB signaling in pancreatic β-cells. Interestingly, at the age of 11 weeks, some animals started to develop hyperglycemia, and at ∼24 weeks of age, all IKK2-CA^Pdx-1 mice showed substantial elevation in fed (538.5 ± 16 vs. 154.9 ± 7 mg/dL in Pdx-1^+/− mice) and fasting (423.3 ± 34 vs. 96.9 ± 4 mg/dL in Pdx-1^+/− mice) blood glucose levels, with several animals >600 mg/dL (Fig. 2A and B). Hyperglycemia was accompanied by a 57% reduction of plasma insulin levels compared with Pdx-1^+/− mice (Fig. 2C), indicating that these mice were overtly diabetic. Animals displayed clinical signs of diabetes, including polyuria, polydipsia, weight loss, and general sickness, and in severe cases, mortality was recorded. Furthermore, immunohistological analyses indicated an extensive loss of insulin-positive β-cells in IKK2-CA^Pdx-1 mice, with the remaining β-cells appearing degranulated with only faint residual insulin immunoreactivity (Fig. 2D) and reduced Pdx-1 expression (Supplementary Fig. 1A). Costaining with IKK2 revealed that most of the IKK2-CA–expressing cells were no longer positive for insulin. Interestingly, few insulin-negative cells still expressed nuclear Pdx-1, while more such cells retained the expression of the endocrine marker chromogranin A (Supplementary Fig. 1A and B). This implicates that IKK2/NF-κB activation in β-cells leads to suppression of insulin expression and thus loss of their functionality in addition (or prior) to their demise. In accordance, pancreatic insulin content and insulin transcription had dropped to ∼3 and 26% of the values of the Pdx-1^+/− littermates, respectively (Fig. 2E and F).

Pdx-1 is an important transcription factor for β-cell function, and its heterozygosity, present in the Pdx-1^+/− knockin mouse, was shown to be associated with some β-cell defects (6). Indeed, Pdx-1^+/− mice showed slight elevation in fed blood glucose level (Fig. 2A) and reduction in insulin transcription and pancreatic content compared with control littermates (Fig. 2E and F); however, in contrast to IKK2-CA^Pdx-1 mice, they were not diabetic.

Invasion of pancreatic islets by leukocytes is the hallmark of immune-mediated diabetes. Unlike Pdx-1^+/− mice, islets of IKK2-CA^Pdx-1 showed marked peri-insulitis (perivascular, periductal, and peri-islet infiltrates) and insulitis (Fig. 3A), reminiscent of the insulitic process in humans. Of note, similar to human histopathology, not all islets showed mononuclear infiltration (Supplementary Fig. 2).

Flow cytometry revealed that hematopoietic (CD45⁺) cells in IKK2-CA^Pdx-1 pancreata increased fourfold (Fig. 3B) and that infiltrating leukocytes were activated CD4⁺ and CD8⁺ T, B, and dendritic cells as well as macrophages and few natural killer cells (Fig. 3B–D and Supplementary Fig. 3A). This was further confirmed on the histological level (Fig. 3E and F and Supplementary Figs. 3B and C and 4). The majority of infiltrating T cells was of the CD4⁺ type (Fig. 3B and F). Furthermore, we detected hyperexpression of MHC I and II by β-cells of IKK2-CA^Pdx-1 mice (Fig. 3G, Supplementary Fig. 5A–C), as previously reported in diabetic patients (10). This suggests increased antigen presentation capacity of β-cells in these mice, which might be critical for phenotype development. In accordance, there was transcriptional upregulation of the MHC I and II components, H2-Q4 and H2-Aa (Fig. 3H), respectively.

The insulitic process in IKK2-CA^Pdx-1 mice was associated with elevated expression of inflammatory cytokines and chemokines like tumor necrosis factor (Tnf), Ccl5, Ccl2, and Cxcl10 and the adhesion molecule intracellular adhesion molecule-1 (Fig. 3I). This inflammatory profile is reminiscent of those found in islets from patients, in animals with type 1 diabetes (11,12), and in normal human islets subjected to cytokines or enteroviruses (13–15).

As apoptosis is thought to be the major cause of β-cell death in diabetes (16), we performed TUNEL assays that showed the presence of apoptotic cells only in IKK2-CA^Pdx-1 islets (Fig. 4A and B). This indicates that apoptosis may account at least in part for the reduced β-cell mass in IKK2-CA^Pdx-1 mice. β-Cells are susceptible to endoplasmic reticulum (ER) stress, and ER stress-mediated apoptosis in β-cells has been implicated in the pathogenesis of diabetes (17). Moreover, islets from prediabetic NOD mice (18) and patients with type 1 diabetes (19) exhibit signs of ER stress. Consistent with that, the ER stress-related factors Ddit3/Chop and Atf3 were upregulated in IKK2-CA^Pdx-1 mice (Fig. 4C). In addition, other stress-associated apoptosis-inducing factors like Nos2 (iNOS) and Myc were also elevated in IKK2-CA^Pdx-1 pancreata (Fig. 4D and E).

To examine the possibility of reverse remodeling the diabetic phenotype, doxycycline was readministered for 30 days to diabetic IKK2-CA^Pdx-1 mice (Fig. 5A). Diabetes in IKK2-CA^Pdx-1 mice was confirmed by elevated fed and fasted blood glucose values and reduced plasma insulin levels (Fig. 5B and C). We observed a clear reduction in fed blood glucose values already within the first 10 days, which virtually completely normalized in all animals by 30 days (Fig. 5B). Consistently, fasting blood glucose and fed plasma insulin levels were also restored (Fig. 5C). Doxycycline-dependent transgene inactivation was confirmed by immunofluorescence staining (Fig. 5D) and Western blot (Fig. 1C). Diabetic IKK2-CA^Pdx-1 mice also regained normal structured islets showing virtual absence of infiltrating cells upon doxycycline treatment (Fig. 5D and E). Insulin immunostaining demonstrated the reappearance of β-cell–rich islets and prominent β-cell regranulation, which strongly stained for insulin (Fig. 5D). Importantly, the reversion of the diabetes status was accompanied by increased levels of Ki67 immunoreactivity in islets of IKK2-CA^Pdx-1 mice (Fig. 5F and Supplementary Fig. 6), indicating that proliferation is involved in β-cell regeneration in this model. Furthermore, transcription of insulin, MHC I/II molecules, and various inflammatory markers was effectively normalized (Fig. 5G–I).

To gain insight into molecular events induced by IKK2/NF-κB activation and possibly involved in diabetes development, we performed gene expression profiling of pancreatic islets isolated from 11-week-old IKK2-CA^Pdx-1 animals. This age represents a critical checkpoint in phenotype development at which some animals already showed hyperglycemia, while others were still normoglycemic (Fig. 6A). In this analysis, including samples from both types, 288 transcripts were found to be upregulated, 46 of which are known NF-κB targets (Supplementary Table 2) and 28 to be downregulated (threshold, 1.5-fold; P < 0.05 in t test with Benjamini-Hochberg correction) (Supplementary Table 1, selected genes, and Table 1). Several of those genes were verified by quantitative PCR (qPCR) as depicted in Fig. 6B. The increased expression of various cytokines, chemokines, adhesion molecules, and antigen presentation molecules clearly demonstrated an activation of innate and adaptive immune responses in islets of IKK2-CA^Pdx-1 mice (Table 1, Supplementary Table 1, and Fig. 6B). However, aside from these factors typically detected in other type 1 diabetes models, other less-characterized genes were highly upregulated (Table 1, Supplementary Table 1, and Fig. 6B). The T-cell–directed chemokine Ccl17 (also known as thymus and activation-regulated chemokine) was the most prominent one. Ccl17 together with Ccl22 (also known as macrophage-derived chemokine), another elevated chemokine, are known ligands for the Ccr4 receptor, which has been linked to the development of autoimmune diabetes (20). Also, a cluster of inflammation-related serine proteases called serine protease inhibitor clade A member 1 (Serpina1) and different members of the cationic protein products of eosinophils, the eosinophil-associated ribonucleases (Ears), were upregulated. Kyoto Encyclopedia of Genes and Genomes analysis clearly revealed a signature reflecting antiviral responses and reactions to infections. This was represented by the upregulation of different interferon (IFN)-regulated genes, especially involved in the type 1 IFN response including signal transducer and activator of transcription 1 (Stat1), IFN regulatory factor (Irf)7, Irf8, and Irf5 as well as IFN-inducible genes like Ifih1 and Ifit1. Furthermore, genes involved in oxidative stress (Nox1) and tissue remodeling (Mmp12) were also elevated (Table 1 and Supplementary Table 1). Taken together, the observed gene expression profile in the IKK2-CA^Pdx-1 model mimics to a great extent that detected in virus-induced immune-mediated diabetes and, importantly, points to other potential novel candidate genes for diabetes.

In an attempt to identify candidate genes involved in phenotype initiation, we followed up the temporal expression pattern of distinct genes. Interestingly, Ccl17 was massively upregulated in islets of 8-week-old IKK2-CA^Pdx-1, when animals were asymptomatic, while there was only mild upregulation of Ear2, Serpina1a, Irf7, and Madcam1 and no elevation of MHC I/II molecules (Fig. 7A). These data support the hypothesis that Ccl17 might be a critical initial effector of IKK2/NF-κB–mediated diabetes development. In addition, qPCR analysis of highly diseased IKK2-CA^Pdx-1 mice (∼24 weeks old) revealed sustained upregulation of Ccl17, Ear2, and Serpina1a that was completely normalized in the reverse-remodeling experiment (Fig. 7B).

In this study, we describe a novel mouse model of immune-mediated diabetes, the IKK2-CA^Pdx-1 model that is based on genetic activation of proinflammatory IKK2/NF-κB signaling, specifically in β-cells. IKK2-CA^Pdx-1 animals develop severe hyperglycemia and hypoinsulinemia mirroring substantial β-cell loss that is associated with marked islet inflammation. Similar to the human immunopathology, infiltrates are not seen in all pancreatic islets and include different kinds of immune cells (21). This model is the first animal model to activate IKK2/NF-κB in β-cells and to prove directly its capability to trigger diabetes development on its own. IKK2-CA^Pdx-1 islets show a gene expression signature reflecting the activation of innate immunity and type I IFN response, which could generate a proinflammatory microenvironment sufficient to recruit different immune cells. These infiltrating cells are able to contribute to and/or amplify the inflammatory insult, thereby promoting β-cell destruction as observed in diabetic subjects and NOD mice (1,2). Furthermore, increased expression of MHC I and II in transgenic β-cells indicates that IKK2/NF-κB signaling is capable of increasing islet antigen presentation to infiltrating T cells, which could participate in the autoimmune insult. In addition, MHC II expression by β-cells can promote their death, as seen by MHC II ligation in antigen-presenting cells (22). This may allow for an intensive dialogue between β-cells and immune cells that finally promotes β-cell destruction in a T-cell–dependent manner. The main T-cell subtype in our model is the CD4⁺ type, in contrast to the concept that CD8⁺ T cells predominate in humans (21), However, there is a considerable heterogeneity within diabetic patients (e.g., even no autoreactive T cells were found in a subset of recent-onset patients) (23). In addition, CD4⁺ T cells can mediate β-cell death in transgenic NOD mice (24), and diabetes development was shown to require the presence of both CD4⁺ and CD8⁺ T cells (25).

The IKK2-CA^Pdx-1 phenotype is in line with the reported role of NF-κB as a mediator of cytokine-induced β-cell destruction (4,5) and its activation in islets of prediabetic NOD mice (18). Consistent with this notion, resistance to streptozotocin (STZ)-induced diabetes was achieved by β-cell–specific inhibition of NF-κB (26,27), and various natural products were found to inhibit STZ-induced diabetes and protect against β-cell damage through NF-κB inhibition (4). However, β-cell–specific repression of NF-κB in normal mice elicits hyperglycemia and defective glucose-stimulated insulin secretion and accelerates diabetes development in NOD mice (4,5), which may implicate that the extent/level, context, and timing of NF-κB activation dictate the overall outcome of diabetes pathogenesis.

Our gene expression data suggest that activation of IKK2/NF-κB in β-cells is sufficient to initiate different cellular mechanisms formerly shown to affect β-cell function/survival and induce diabetes in humans and animals (16,17,28–32). These diverse factors (TNF-α, Atf3, C/EBP homologous protein, Nos2, c-myc, Stat1, and Nox-1) are known to induce inflammation, oxidative stress, ER stress, and nitric oxide production (Fig. 7C) that together may finally funnel in β-cell dysfunction and/or apoptosis in our model. However, we cannot exclude other unknown effects of IKK2/NF-κB signaling that might interfere with β-cell function and promote disease development (Fig. 7C).

The expression profile in our model is further pointing to the involvement of innate immunity and type I IFN response (Fig. 7C), characteristic for viral infection, and is similarly detected in pancreata and islets from patients at clinical onset and long-standing diabetes (12). Furthermore, the gene-expression program induced by enteroviral infection of human islets (14,15) is also prominently mimicked by the IKK2-CA^Pdx-1 model, suggesting that IKK/NF-κB is a critical downstream effector in this context. Viral infections have been proposed as a triggering factor of type 1 diabetes in humans and animals (33), and importantly, they activate Toll-like and nucleotide-binding oligomerization domain–like receptors, which are well-known inducers of NF-κB. Indeed, NF-κB signaling is activated in β-cells by enterovirus infection (15) and double-stranded RNA treatment (34) and mediates at least partially the deleterious effects of these insults. Recently, Irf7, the master regulator of type I IFN–dependent immune responses and upregulated in our model, has been implicated in type 1 diabetes pathogenesis (35). Similarly, Ifih, another candidate gene for type 1 diabetes that is expressed in human islets and can be induced by inflammatory cytokines (13) and by double-stranded RNA in rat β-cells (36), is elevated in IKK2-CA^Pdx-1 mice. Therefore, it is well conceivable that NF-κB activation in β-cells on its own could create an innate immune response similar to an antiviral response, which finally culminates in diabetes development. However, we cannot exclude a possible role of Pdx-1 haploinsufficiency on promoting β-cell susceptibility to IKK2/NF-κB–induced inflammation and phenotype development.

We also identified other novel diabetes-associated candidate genes. Ccl17, the most upregulated one, is a chemokine involved in immune and allergic inflammatory responses (20,37–42). The early marked upregulation of Ccl17 at normoglycemia supports the idea of being a foremost effector of IKK2/NF-κB signaling that contributes to diabetes development. Consistently, viral infection of human B cells induces NF-κB–dependent expression of Ccl17 and Ccl22 (also elevated in our model) (37), and IKK/NF-κB inhibition prevents cytokine-induced Ccl17 production in keratinocytes (38). Ccl17 is mainly produced by dendritic cells (39) and plays an important role in T cell development, trafficking, and activation. Furthermore, it was highlighted as a novel biomarker for allergic inflammatory diseases like asthma (40) and atopic dermatitis (41), in which it is strongly coexpressed with Ccl22 (42).

Both Ccl17 and Ccl22 preferentially attract CD4⁺ T cells via the Ccr4 receptor. Cells expressing Ccl17 were detected within infiltrated islets from prediabetic NOD mice, and Ccr4-positive T cells were shown to be critically involved in autoimmune diabetes development (20). Additionally, neutralizing Ccl22 antibodies inhibited insulitis and diabetes, whereas transgenic Ccl22 expression accelerated disease development (20). However, autoimmune diabetes was prevented by Ccl22-mediated regulatory T-cell recruitment in another study (43). Furthermore, no significant difference in plasma Ccl17 was detected in type 1 diabetes patients (44). Yet, this study was conducted in a very small cohort of diabetic Japanese subjects, which certainly does not exclude a different pattern in other patient subsets.

Ears, also prominently induced in our model, represent a subgroup of the RNase A family secreted by rodent eosinophils (45) and also expressed by macrophages. Ears participate in host defense by their antibacterial and antiviral activity together with chemotaxis to dendritic cells (45). The Serpina1 family of genes encodes for inhibitors of serine proteases. In humans, it is represented by a single gene, called α1-antitrypsin (AAT) for which expression was enhanced by proinflammatory cytokines in islet cells (46). AAT is an acute-phase protein with anti-inflammatory, tissue-protective, and antiapoptotic properties that is able to prolong islet allograft survival and to inhibit cytokine and STZ-induced β-cell apoptosis (47,48) as well as diabetes development in NOD mice (49). Although decreased functional activity of serum AAT was shown in diabetes, serum levels are variable, probably reflecting changes in the inflammatory status of the disease (50). To what extent these newly identified genes contribute to human disease pathology, however, needs further investigation.

One striking aspect of the IKK2-CA^Pdx-1 model is the enormous recovery potential from the diabetes status including the resolution of insulitis and the efficient regeneration of insulin-positive β-cells after transgene inactivation. A possible rationale is that a portion of β-cells is not destroyed but, rather, has lost its functional and differentiated state and therefore can regain its functionality upon IKK2-CA inactivation with the consequent resolution of proinflammatory mediators and disappearance of immune cells. Furthermore, the increase of Ki67-positive β-cells during the regeneration phase suggests that proliferation is also involved in the restoration of β-cell mass in this model as found in other models; however, other mechanisms cannot be excluded. This finding implies that the IKK/NF-κB system might be a potential target for clinical intervention. Furthermore, the IKK2-CA^Pdx-1 mouse model could be valuable for assessing mechanisms and identifying genes involved in β-cell regeneration as well. Since diabetes passes through relapsing-remitting onsets, our model is further potentially useful for evaluating clinical therapeutics at different stages of the human disease presentation through switching the system off and on.

In summary, the IKK2-CA^Pdx-1 model represents a novel conditional mouse model of immune-mediated diabetes with very distinct checkpoints of β-cell destruction and regeneration. This model phenocopies major aspects of the human disease and may represent a valuable tool for improving preclinical drug assessment for diabetes treatment.

Acknowledgments. The authors thank Ute Leschik, Melanie Gerstenlauer, and Bianca Ries (University of Ulm) for excellent technical assistance and Karlheinz Holzmann (Genomics-Core Facility, University Hospital Ulm) for performing the microarray analysis.

Funding. This work was supported by grants GRK-1041-P3 (Deutsche Forschungsgemeinschaft) and BIU-C6 (Boehringer Ingelheim Ulm University BioCenter) to B.B.

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

Author Contributions. H.H.S. and B.B. designed and performed the experiments, analyzed data, and wrote the manuscript. B.T. and K.F. performed research and analyzed data. H.J.M. contributed to discussion and reviewed and edited the manuscript. R.S. and M.W. contributed to discussion, performed the experiments, and reviewed and edited the manuscript. B.O.B. and T.W. designed research and reviewed and edited the manuscript. H.H.S. and B.B. 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.