Diabetes Exacerbates Infection via Hyperinflammation by Signaling through TLR4 and RAGE

ABSTRACT For more than a century, diabetic patients have been considered immunosuppressed due to defects in phagocytosis and microbial killing. We confirmed that diabetic mice were hypersusceptible to bacteremia caused by Gram-negative bacteria (GNB), dying at inocula nonlethal to nondiabetic mice. Contrary to the pervasive paradigm that diabetes impedes phagocytic function, the bacterial burden was no greater in diabetic mice despite excess mortality. However, diabetic mice did exhibit dramatically increased levels of proinflammatory cytokines in response to GNB infections, and immunosuppressing these cytokines with dexamethasone restored their resistance to infection, both of which are consistent with excess inflammation. Furthermore, disruption of the receptor for advanced glycation end products (RAGE), which is stimulated by heightened levels of AGEs in diabetic hosts, protected diabetic but not nondiabetic mice from GNB infection. Thus, rather than immunosuppression, diabetes drives lethal hyperinflammation in response to GNB by signaling through RAGE. As such, interventions to improve the outcomes from GNB infections should seek to suppress the immune response in diabetic hosts.

T wenty-six million people in the United States suffer from diabetes, a disease projected to afflict one-fifth of the population by 2050 (1). People with diabetes are three times more likely to develop bacterial sepsis and suffer far worse outcomes (2)(3)(4)(5). In particular, diabetes is a risk factor for significantly increased mortality from infections caused by Gram-negative bacteria, especially Acinetobacter baumannii (6)(7)(8)(9). Over a century of literature has indicated that defective phagocytes in diabetic hosts are responsible for worse outcomes from bacterial infections, leading to the characterization of diabetes as an immunosuppressive state. This immunosuppression has long been ascribed to deficits in neutrophil phagocytosis and microbial killing found ex vivo in cells taken from diabetic hosts (4,5,(10)(11)(12)(13)(14)(15)(16)(17)(18)(19)(20)(21)(22), yet neutropenia is not a substantial risk factor for worse outcomes from A. baumannii infections (23,24). Indeed, even in those with cancer, the majority of patients who develop A. baumannii infections are not neutropenic (23), and neutropenia is not an independent predictor of outcome from these infections (24). We therefore suspected that suppression of phagocytic activity was unlikely to explain why diabetic hosts would be more susceptible to infections caused by A. baumannii and other Gram-negative bacteria.
While studying A. baumannii pathogenesis, we found that the principal cause of lethality in nondiabetic mice was inflammation driven by stimulation of Toll-like receptor 4 (TLR4) by lipopolysaccharide (LPS) (25). It has also been reported that high levels of inflammatory mediators in mice correlated with increased susceptibility to bacterial infection (26). Thus, we investigated a novel hypothesis: rather than immunosuppression, worse outcomes in diabetic hosts with Gram-negative bacterial infections are due to overactivation of the innate immune system. As such, we sought to determine the etiology for increased mortality from infections caused by Gramnegative bacteria and to define the source of excess inflammation in the diabetic host.

RESULTS
We confirmed that C57BL/6 mice with diabetes induced by streptozotocin (STZ) (27) were hypersusceptible to lethal A. baumannii bacteremia at an inoculum nonlethal to nondiabetic mice (Fig. 1A). Even mice with diet-induced-obesity (DIO) diabetes, which causes mild hyperglycemia consistent with clinical type 2 diabetes (28)(29)(30), had significantly increased mortality compared to that in nondiabetic mice (Fig. 1B). One possible explanation for greater mortality in diabetic mice is that hyperglycemia creates preferable growth conditions for the bacteria. However, we found no difference in bacterial growth in DIO diabetic versus nondiabetic mouse serum, indicating that simple growth enhancement caused by hyperglycemia was not operative (Fig. 1C). Moreover, despite clear survival differences between DIO diabetic and nondiabetic mice, there was no disparity in blood bacterial density (Fig. 1D), indicating no deficiency in phagocytic clearance of bacteria in DIO diabetic mice.
Despite similar bacterial densities, diabetic but not nondiabetic mice experienced hypothermia, severe metabolic acidemia (low blood pH, low serum bicarbonate, and high base deficit), and markedly higher serum levels of interleukin-1␤ (IL-1␤), IL-10, and tumor necrosis factor (TNF) than nondiabetic mice (P Ͻ 0.05 for all comparisons), plus a strong trend toward higher levels of IL-6 (P ϭ 0.05) ( Fig. 2A and B). Finally, diabetic mice developed higher ratios of IL-10 to TNF (Fig. 2C), consistent with prior clinical findings that predict worse survival in bacteremic patients (31).
The worse survival rates and increased levels of inflammatory cytokines in diabetic mice with no correlation between bacterial burden and survival suggested that hyperinflammation was driving the increased lethality. To test this hypothesis, we immunosuppressed diabetic mice by treating them with a potent corticosteroid (dexamethasone at 4 mg/kg of body weight). Diabetic mice treated once with dexamethasone 20 min postinfection had markedly improved survival compared to diabetic control mice treated with a placebo (Fig. 3A). Thus, diabetes did not exacerbate lethality by inhibiting clearance of A. baumannii but by amplifying the inflammatory response to the organism.
We have previously shown that A. baumannii infections result in profound hyper-inflammation and sepsis in nondiabetic mice via TLR4 (25). We therefore sought to determine whether TLR4 disruption also mitigated mortality in diabetic mice. While all wild-type mice (C3H/Fe) with streptozotocin-induced diabetes died from A. baumannii infection, diabetic TLR4 mutant mice (C3H/He) had markedly improved survival (Fig. 3B). Similarly, a proprietary small-molecule competitive TLR4 antagonist (E5564, 10 mg/kg/day intravenously [i.v.] for 4 days postinfection) completely protected mice from otherwise lethal A. baumannii infection (Fig. 3C). Furthermore, knocking out TLR4 in C57BL/6 mice did not affect the hyperglycemia in DIO mice ( Fig. 3D) but did completely protect them from lethal infection (Fig. 3E). These data confirm that diabetic mice are protected against A. baumannii infection in the absence of TLR4 signaling, as was previously found in nondiabetic mice. Despite worse survival, DIO diabetic mice did not have greater bacterial burdens or higher serum endotoxin levels than nondiabetic mice, whether wild type or TLR4 knockout (TLR4 KO) ( Fig. 4A and B). Tellingly, DIO diabetic TLR4 KO mice had marked improvements in sepsis biomarkers and near-baseline (32-34) levels of inflammatory cytokines compared to DIO diabetic wild-type mice ( Fig. 4C and D). These data further underscore the lack of correlation between outcomes of infection and innate immune clearance of bacteria, instead attributing mortality to hyperinflammation in diabetic hosts.
To explore the generalizability of these findings, we established models of infection in DIO diabetic mice caused by a clinical bloodstream isolate of Escherichia coli expressing an extended-spectrum ␤-lactamase (ESBL) (strain type 131/H30 with the bla CTX-M gene) and a strain of pan-drug-resistant Klebsiella pneumoniae expressing a Klebsiella pneumoniae carbapenemase (KPC). DIO diabetic mice were significantly more susceptible than nondiabetic mice to both Gram-negative Enterobacteriaceae species  5B). Just as we found with A. baumannii, DIO diabetic TLR4 KO mice were more resistant to K. pneumoniae and E. coli infection than DIO diabetic wild-type mice (Fig. 5C).
Since both diabetic and nondiabetic mice express TLR4 but diabetic mice die from inocula nonlethal to nondiabetic mice despite similar bacterial burdens, we hypothesized that a second inflammatory signaling pathway unique to diabetic mice must drive sepsis, in addition to TLR4. One well-studied mechanism by which diabetes induces endovascular inflammation proceeds via stimulation of the receptor for advanced glycation end products (RAGE), due to elevated levels of advanced glycation end products (AGEs) (35,36). Recently, TLR4 and RAGE were shown to share the downstream proinflammatory signaling transducer MyD88 (37). We therefore tested the ability of two soluble RAGE antagonists to affect the outcome from infection, FPS-ZM1 (small molecule) and RAP (peptide antagonist) (38,39). Antagonizing RAGE during A. baumannii infection was protective in mice with DIO diabetes but not in nondiabetic mice (Fig. 6A), demonstrating that the effect was related to diabetes. Despite improving the survival of DIO diabetic mice, RAGE antagonism did not alter the bacterial burdens ( Fig. 6B) but did significantly ameliorate the metabolic acidemia and cytokine storms seen in the DIO diabetic mice ( Fig. 6C and D). DIO diabetic mice also experienced significantly higher ratios of IL-10 to TNF than mice treated with the small-molecule RAGE antagonist (FPS-ZM1), similar to the results for the nondiabetic wild-type mice in the experiment whose results are shown in Fig. 2C (Fig. 6E). Additionally, antagonizing RAGE in DIO diabetic mice infected with E. coli (Fig. 6F) or K. pneumoniae (Fig. 6G) significantly improved survival without affecting bacterial burdens, demonstrating that the protective effect was generalizable across Gram-negative species.
To ensure that off-target effects of the two RAGE antagonists were not responsible for the observed phenotype, we sought to corroborate the RAGE antagonist results in knockout mice. We first demonstrated that DIO wild-type and DIO RAGE KO mice had similarly elevated levels of fasting blood glucose (Fig. 7A). Like the outcome of treatment with soluble RAGE antagonists, knocking out RAGE in diabetic mice also made them highly resistant to lethal infection compared to the susceptibility of diabetic wild-type control mice (Fig. 7B). Indeed, knocking out RAGE in diabetic mice reverted their susceptibility to infection to the level of nondiabetic wild-type mice (compare Fig. 1B and 7B). Severing RAGE signaling also resulted in marked reductions of inflammatory cytokine levels in diabetic mice (Fig. 7C), to levels comparable to those in nondiabetic wild-type mice (compare Fig. 2B and 7C). Diabetic wild-type mice again developed higher ratios of IL-10 to TNF than diabetic RAGE KO mice (Fig. 7D), resembling the ratios in nondiabetic wild-type mice (compare Fig. 2C and 7D).
As previously mentioned, a recent study found a common signaling pathway between RAGE and TLR4, whose unique cytoplasmic effectors both bind to TIRAP and signal through MyD88, which transduces downstream signals that activate the transcription of inflammatory cytokines (37). We therefore sought to determine whether MyD88 signaling was a common pathway driving the increased susceptibility of diabetic mice to infections caused by Gram-negative bacteria. DIO wild-type and DIO MyD88 KO mice had similar levels of fasting blood glucose, which were significantly higher than the levels in nondiabetic wild-type mice (Fig. 8A). Despite intact TLR4 and RAGE, diabetic MyD88 KO mice were completely resistant to lethal infection compared to the susceptibility of diabetic wild-type control mice (Fig. 8B).

DISCUSSION
The increased susceptibility of diabetic patients to acute bacterial infectionsparticularly those caused by Gram-negative bacteria-was noted in the 19th century (40). As early as 1907, the mechanism driving the increased severity of infections in diabetic patients was attributed to suppressed phagocytic clearance of microbes (15-18, 20-22, 41-43). Notwithstanding these conclusions, which were primarily drawn from in vitro and ex vivo assays of phagocyte function, we now report that diabetes causes the host to become hypersusceptible to infections caused by Gram-negative bacteria through immune paralysis, independent of bacterial density/clearance. Rather, the hypersusceptibility of diabetic hosts to Gram-negative bacterial infections is primarily due to hyperinflammation-not immunosuppression-inherent to the diabetic host.
TLR4 signaling occurs during infection for both diabetic and nondiabetic mice, but   bacteremia, diabetic wild-type mice experienced significantly higher ratios of IL-10 to TNF than diabetic RAGE KO mice, similar to the nondiabetic wild-type mice in the experiment whose results are shown in Fig. 2C (n ϭ 10 mice per group). *, P Ͻ 0.05.

Diabetes and Hyperinflammation
® receptors, whereas nondiabetic mice activate MyD88 through TLR4 alone. Our findings thus indicate that diabetic mice had increased mortality from infections caused by Gram-negative bacteria not because of immunosuppression but because of hyperinflammation driven by TLR4 and compounded by RAGE, both signaling through a common pathway via MyD88. In 2003, Casadevall and Pirofski proposed the damage response framework model (Fig. 8C) (44). In short, the framework underscores that outcomes of infection depend on contributions from both the microbe and the host. Clinical symptoms and outcomes from infection can therefore result from excessive host inflammatory responses to infection. Diabetes has traditionally been viewed as an immunocompromising condition, causing suppression of phagocytic host defense mechanisms that lead to poor outcomes from infection, an interpretation that pins diabetic hosts to the weak-response side of the damage response framework. Instead, our results indicate that diabetes causes excessive host damage through hyperactive innate immune inflammation in response to Gram-negative bacteria, an interpretation that pins diabetic hosts to the (opposite) strong-response side of the damage response framework.
These results imply that efforts to stimulate the immune system will imperil rather than protect diabetic hosts from acute, Gram-negative bacterial infections. On the contrary, dampening the innate immune response to such infections offers a promising therapeutic approach to such infections in diabetic hosts. Specifically, antagonism of TLR4, RAGE, and MyD88 were each highly effective in protecting mice from lethal infections in our model. Such a strategy has tremendous potential as adjunctive therapy by bolstering antibacterial agents, which alone are often inadequate against Gram-negative bacterial infections in patients with diabetes. Similar pathogenesis and potential therapies may also be relevant to other infections that are more common and more severe in those with diabetes.

MATERIALS AND METHODS
Bacterial and mouse strains. Strain HUMC1 is an A. baumannii clinical blood and lung isolate (from a patient with bacteremic ventilator-associated pneumonia) that is resistant to all antibiotics except for colistin and is highly virulent in animal models (25,45,46). Strain KPC KP1 is a panresistant (resistant to all antibiotics, including tigecycline and colistin) clinical bloodstream isolate of K. pneumoniae. Strain JJ1886 is a sequence type 131, ESBL-expressing, H30 strain-derived clinical isolate of E. coli (courtesy of James Johnson).
Infection models. We used our well-established tail vein infection method as the model for bacteremia (25,45,46). In this model, a relatively low inoculum (2 ϫ 10 7 to 4 ϫ 10 7 CFU) of A. baumannii results in lethal infection in susceptible mice, and the model mimics the second most common clinical A. baumannii syndrome (bacteremia), typically initiated by a catheter. Streptozotocin was administered as previously described (46). Some mice were treated with a parenteral competitive TLR4 antagonist, E5564 (supplied by Eisai, Inc.), starting on the day of infection. Based on pilot studies, the drug was dosed at 10 mg/kg, administered i.v. once daily for 4 days when treating i.v. infections.
Bacteria were grown overnight in tryptic soy broth (TSB) at 37°C with shaking. The bacteria were passaged to mid-log growth phase in TSB at 37°C with shaking. Cells were washed twice with phosphate-buffered saline (PBS) and resuspended at the appropriate concentration for infection. Bacteria were administered i.v. via the tail vein. All animal experiments were approved by the Institutional Committees on the Use and Care of Animals (IACUC) at the Los Angeles Biomedical Research Institute and the Keck School of Medicine at the University of Southern California, following the National Institutes of Health guidelines for animal housing and care (48).
Growth curve in serum. Mice were sedated with 100 mg/kg ketamine and 10 mg/kg xylazine. Serum was isolated from blood obtained by cardiac puncture. A. baumannii HUMC1 was added to 300 l serum at 1 ϫ 10 4 CFU/ml and placed on a rotator in a humidified 37°C incubator. Aliquots were taken at 0, 1, 3, and 24 h postinoculation and then diluted in PBS and plated on tryptic soy agar (TSA) to count CFU.
Cytokine and sepsis biomarkers. Serum cytokines were quantified by using the MSD Multi-Spot assay (Meso Scale Diagnostics, LLC) according to the manufacturer's instructions. Blood sepsis biomarkers (pH, base deficit, and bicarbonate) were analyzed using the i-STAT system. For experiments in which i-STAT measurements were made, it was necessary to treat mice with intraperitoneal heparin (25 U; Sigma-Aldrich, St. Louis, MO) with simultaneous sedation by intraperitoneal ketamine (100 mg/kg) and xylazine (10 mg/kg) prior to cardiac puncture to prevent clotting in the i-STAT cartridges. Whole blood was drawn into a 25-gauge syringe and aliquoted into i-STAT cartridges. Values were read on an i-STAT portable clinical analyzer.
Statistics. Survival was compared by the nonparametric log rank test. Categorical variables were compared with the Wilcoxon rank sum test for unpaired comparisons.

ACKNOWLEDGMENTS
We dedicate this paper to Fred Duncanson, respected friend and colleague, who passed away during its drafting. This work was supported by NIAID R01s AI081719 and AI117211, R21 AI127954, and R42 AI106375 and a grant from Eisai, Inc. (B.S.), by Veterans Affairs merit review grant number 1I01BX001974, funding from Geriatric Research Education and Clinical Center VISN, and NIAID R01s AI063517 and AI100560 to R.A.B., and by NIAID UM1AI104681 for both B.S. and R.A.B.