Nrf2-Mediated Neuroprotection Against Recurrent Hypoglycemia Is Insufficient to Prevent Cognitive Impairment in a Rodent Model of Type 1 Diabetes

Hypoglycemia is a common adverse side effect of insulin therapy in type 1 diabetes (T1D), largely due to hyperinsulinemia, a diminished counterregulatory response, and impaired awareness of hypoglycemia (1). The brain is especially vulnerable to hypoglycemia due to its high metabolic demand and minimal fuel stores. As such, there is potential for recurrent hypoglycemia to produce long-term neuronal damage. This is a source of major concern and fear for individuals with T1D (2), especially as studies in animals and humans have yielded inconsistent findings.

Profound hypoglycemia, sufficient to cause coma, results in brain damage in humans (3,4), but epidemiological studies yield conflicting data on the impact of reversible severe and nonsevere hypoglycemia on long-term cognitive function (5). For instance, 18 years’ follow-up of T1D individuals in the Diabetes Control and Complications Trial (DCCT) (6) found no evidence of an association between severe hypoglycemia and cognitive decline. In contrast, prospective studies in prepubertal children with T1D have reported that severe hypoglycemia may result in long-term neurological damage and psychomotor retardation (7–11). A limitation of these studies is that the cognitive decline resulting from recurrent hypoglycemia may take place over many decades, and accurate documentation of the frequency of hypoglycemia over such timescales is extremely difficult.

Studies in animal models offer the opportunity to address many of these questions over shorter time frames as well as to examine underlying mechanisms. This literature though is equally conflicting. Again, profound hypoglycemia (sufficient to induce an isoelectric electroencephalogram and/or multiple seizures) in animal models causes brain damage particularly in hippocampus and frontal cortex (12–15). However, significant neuronal damage under such conditions would be anticipated, and the relevance therefore to T1D in which hypoglycemia of such severity is rare is not clear. In contrast, nonsevere hypoglycemia does not appear to induce neuronal cell death (16), and long-term recurrent nonsevere hypoglycemia has even been shown to protect against age-related cognitive decline (17). In addition, recurrent nonsevere hypoglycemia in rodents potentially preconditions the brain, protecting it to some extent from neurological damage resulting from subsequent very severe hypoglycemia (18).

An additional limitation of studies in animals is that many were primarily conducted using nondiabetic models. This is important, because in T1D, intermittent exposure to hypoglycemia will always occur in the context of chronic hyperglycemia of varying degrees. Chronic hyperglycemia (19), severe hypoglycemia (20,21), and glucose recovery from hypoglycemia (13,22) have each independently been shown to stimulate reactive oxygen species (ROS) production. In addition, chronic hyperglycemia may impair antioxidant defense mechanisms (23–25). Therefore, the ability of neurons to respond to nonsevere hypoglycemia may be uniquely impaired in T1D increasing vulnerability of the brain particularly to oxidative stress, but this question has to date not been addressed.

To test this hypothesis, we studied insulin-treated T1D and nondiabetic rodent models that were exposed to intermittent episodes of nonsevere hypoglycemia over 4 weeks and examined the impact of these interventions on cognitive function and markers of oxidative stress and inflammation. Having demonstrated that recurrent hypoglycemia in T1D but not nondiabetic rodents induced defects in cognitive function that were associated with hippocampal inflammation and oxidative damage, we subsequently sought to clarify the role of the oxidative stress response by studying mice lacking the transcription factor nuclear factor–erythroid 2 p45–related factor 2 (Nrf2) because it plays a critical role in regulating basal cellular antioxidant defenses as well as orchestrating responses to oxidative stress (for review, see Ref. 26). In this study, we report that Nrf2^−/− mice are very vulnerable to both T1D and recurrent hypoglycemia and that loss of Nrf2 resulted in strongly enhanced hippocampal inflammatory and oxidative damage responses to these two metabolic stimuli.

Sixty-four adult male C57BL/6J mice (20–25 g; Charles River Laboratories, U.K.) and 30 male Nrf2^−/− (20–25 g) mice were used. Generation and genotyping of Nrf2^−/− mice (provided by Ken Itoh and Masayuki Yamamoto, Centre for Tsukuba Advanced Research Alliance and Institute of Basic Medical Sciences, University of Tsukuba, Tsukuba, Japan) have been described previously (27). Animals were fed ad libitum, on a 12:12-h light/dark schedule. All animal procedures were approved by the University of Dundee Ethical Review Process and performed in accordance with U.K. Home Office regulations (under the auspices of Project License PIL60/4120).

Groups of animals were randomly assigned to receive streptozotocin (STZ) (125 mg/kg i.p.) to induce T1D or control (citrate acid buffer i.p.). Tail vein blood glucose (Accu-Read; Accu-Read, Guangdong, China) was measured 3 and 7 days post-STZ and a reading ≥16.0 mmol/L (288 mg/dL) regarded as diabetic. Animals failing to reach this were given a second injection of STZ and retested as above. Mice were subsequently subdivided into recurrent hypoglycemia (Hypo) or control, giving the following four test groups: 1) Control, 2) Control + Hypo, 3) T1D, and 4) T1D + Hypo (n = 8/group) (Supplementary Fig. 1A). Similarly, for studies of Nrf2-null mice, animals were randomly allocated to the following four test groups: 1) Nrf2^−/−, 2) Nrf2^−/− + Hypo, 3) T1D Nrf2^−/−, and 4) T1D Nrf2^−/− + Hypo (n = 7 to 8/group).

Animals were anesthetized by isoflurane and Linbit insulin implants inserted subcutaneously. Insulin replacement was used in an attempt to replicate more closely human insulin-treated T1D and to ensure the animals remained healthy and showed positive weight trajectory over the 12 weeks of the experiment (Supplementary Fig. 1B and C). However, for the purposes of this study in which the interaction between chronic hyperglycemia and Hypo was being explored, insulin implants at half of the recommended dose (∼0.05 U/kg/day) were used (Fig. 1A and C). Control animals were also anesthetized and subjected to sham surgery.

Mice were subjected to eight episodes of hypoglycemia (two per week for 4 weeks) (Supplementary Fig. 1A). Following a 4-h fast, basal glucose was measured from the tail vein, and insulin was injected (0.75 mU/g Nrf2^−/−, 1 mU/g Control and wild-type [WT] animals and 4 mU/g T1D and T1D Nrf2^−/− i.p.) to induce moderate hypoglycemia (2.5 –3 mmol/L) (Fig. 1B and D). Hypoglycemia was maintained for 2 h and animals allowed to return to euglycemic levels with food. Animals were monitored continuously during the hypoglycemic period. After 2 weeks, insulin doses were reduced to 0.5 mU/g Nrf2^−/−, 0.75 mU/g Control animals, and 2.5 –3 mU/g for T1D and T1D Nrf2^−/− animals, respectively. Any animal showing signs of physical impairment received glucose (1 mg/kg i.p.). Two animals in the Control group, one animal in the T1D group, and one animal from the Nrf2^−/− group required recovery on one occasion. Control animals were fasted and given saline injections. No animals suffered from seizures.

Cognition was assessed by novel object recognition (NOR) and spontaneous alternation tests (28,29), at least 3 days after the last hypoglycemic episode (Supplementary Fig. 1A). Beckman activity box and open field maze tasks were also made to exclude any confounding effects of altered activity or anxiety.

A simple hippocampal-mediated task based on the innate tendency of rodents to seek novelty was used (28). The primary outcome measure was the discrimination index (D3), calculated as the total time spent exploring the familiar objects/total time spent exploring the novel object, for short-term (10-min) and long-term (24-h) memory.

This was used to test spatial working memory as described previously (29). Memory was assessed as a percent 4/5 alternation with an alternation counted when all four arms are visited within a span of five arm entries (29).

Locomotor activity was assessed using a Beckman activity box. Animals, habituated to the box for 4 days, were placed into the box and allowed to explore freely for 15 min. Mobile, active, and static counts were recorded.

Activity was recorded on film and later analyzed by two independent scorers for total time spent within the inner and outer zones of the maze. Activity is presented as percent total time spent in each zone.

On completion of behavioral testing, animals were killed humanely, brain tissue dissected, and flash frozen in liquid nitrogen for subsequent biochemical analyses.

The concentration of malondialdehyde was determined in hippocampus using the thiobarbituric acid–reactive substances assay as described in Mihara and Uchiyama (30) and adapted for a 96-well plate format. The amount of malondialdehyde in the samples was determined spectrophotometrically at 532 nm and concentration determined from a standard curve. All samples were assayed in duplicate.

Levels of carbonylated protein within the hippocampus were measured by ELISA (Cayman Chemical). Protein carbonyl concentration was calculated using the following equation: protein carbonyl (nmol/mL) = ([CA]/[0.011 μmol/L⁻¹])(500 μL/200 μL), where CA is equal to the corrected absorbance (average absorbance of controls − average absorbance of samples).

The total protein levels of a panel of inflammatory cytokines (interferon-γ, interleukin [IL]-10, IL-12p70, IL-1β IL-2, IL-4, IL-5, IL-6, keratinocyte chemoattractant/human growth-related oncogene, and tumor necrosis factor-α [TNF-α]) were measured using the V-Plex Proinflammatory Panel (mouse) kit (Meso Scale Discovery). Hippocampal homogenates (50 µg) were assayed in duplicate and values presented relative to their appropriate Control group (Control or Nrf2^−/−, respectively).

Total RNA was extracted from hippocampal tissue using TRIzol reagent (Invitrogen). Reverse transcription was performed with 1 ng RNA using SuperScript III First Strand Synthesis system for RT (Invitrogen). Real-time PCR was performed using TaqMan gene expression assays for the following genes: Nrf2 and Nrf2 target genes [Hmox1, Nqo1, Gsta1, Txn1, Txrnd1, and Srxn1 (31)], the inflammatory genes IL-1β, IL-6, Tnfα, and Nos2, and housekeeping genes (Applied Biosystems) (see Supplementary Table 1 for full details of primer/probe sequences). All samples were performed in triplicate and normalized to actin. Values are expressed as a fold-change relative to Control.

Data were analyzed using SPSS version 18 (SPSS, Chicago, IL). Multivariate ANOVA was used to compare groups with treatment (TID) or genotype (Nrf2^−/−) and Hypo as between subject variables. Post hoc analysis was performed using the Tukey multiple-comparisons test. Data are expressed as mean values ± SEM. Statistical significance was set at P < 0.05.

As expected, blood glucose levels were higher in T1D compared with WT animals over the 12-week study (Fig. 1A) (T1D × week F[3.74,93.48] = 35.89; P < 0.05). However, the level of hypoglycemia achieved was comparable between T1D and Control groups (Fig. 1B) (average glucose, Control = 2.9 ± 0.3 vs. TID = 3.0 ± 0.4 mmol/L; t = 0.088; P = 0.97) and did not change over the duration of the experiment (T1D, F[1,26] = 0.621; P = 0.45).

Similarly, T1D Nrf2^−/− mice had significantly higher fasting glucose levels when compared with nondiabetic Nrf2^−/− animals for the duration of the study (Fig. 1C) (T1D × week, F[3.66,36.66] = 21.53; P < 0.01) but the degree of hypoglycemia achieved between T1D Nrf2^−/− and Nrf2^−/− animals was comparable (Fig. 1D) (average glucose, Nrf2^−/− = 2.4 ± 0.3 vs. TID Nrf2^−/− = 3.1 ± 0.6 mmol/L; t = 0.062, P = 0.95).

The most significant cognitive defect found was in long-term memory as assessed by the NOR task in T1D + Hypo mice (Fig. 2A) (T1D × Hypo, F[1,30] = 5.936; P < 0.05). A less marked cognitive defect at this time point was also found in T1D mice (Fig. 2A) (T1D, F[1,30] = 52.40; P < 0.05), but not in Control or Control + Hypo animals. These differences were not due to a reduction in time spent exploring the objects (effect of T1D and Hypo, both P = not significant). Performance on the NOR task after a 10-min interval was intact in all groups (main effects T1D and Hypo on D3 indices and Exploration time; all P = not significant), indicating that both the ability to learn the task and short-term memory for objects were intact.

On the spontaneous alternation task, T1D + Hypo animals were also most affected, demonstrating a reduction in percentage alternations ([total alternations/total entries − 4] * 100) (Fig. 2B) (T1D × Hypo, F[1,15] = 5.71; P < 0.01) and fewer total alternations (T1D × Hypo, F[1,15] = 7.734; P < 0.01; average total alternations; Control 46.29 ± 3.66, Control + Hypo 48.36 ± 3.63, T1D 47.10 ± 4.67, and T1D + Hypo 41.67 ± 3.80). In contrast, Control animals receiving Hypo showed a trend toward enhanced percentage of alternation (Fig. 2B) (F[1,15] = 3.04; P = 0.10). There was no effect of T1D alone on the total number of entries (F[1,30] = 2.92; P = not significant).

As with wild types, all Nrf2^−/− animals performed the short-term NOR task above chance levels (D3 index of ≥0.2; F < 1; P = not significant), and as before, there were no differences between groups in discrimination indices and time spent exploring the objects (D3 indices and Exploration time; all F < 1; P = not significant). However, after 24 h, only Control Nrf2^−/− animals were able to perform the task with both Hypo, T1D, and Hypo + T1D markedly impairing long-term memory (Fig. 2C) (Hypo, F[1,25] = 12.09, P < 0.01; T1D, F[1,25] = 6.95, P < 0.05; and T1D × Hypo, F[1,25] = 6.24, P < 0.05). These differences were not due to a reduction in time spent exploring the objects (Exploration time; all F < 1; P = not significant).

In both nondiabetic and T1D Nrf2^−/− animals, Hypo resulted in impaired working memory ([total alternations/total entries − 4] * 100) (Fig. 2D) (Hypo F[1,25] = 18.18; P < 0.01; F < 1 for the effect of T1D and T1D × Hypo). This impairment was accompanied by a significant reduction in the total number of alternations (Hypo F[1,25] = 9.22, P < 0.01; T1D, F[1,25] = 17.28, P < 0.01; F < 1, P = not significant for effect of T1D × Hypo; average total alternations: Nrf2^−/− 28.35 ± 2.78, Nrf2^−/− + Hypo 21.00 ± 2.80, T1D Nrf2^−/− 24.86 ± 4.11, and T1D Nrf2^−/− + Hypo 10.14 ± 1.83). As with wild-type mice, there was no difference in total number of entries (T1D, F[1,25] = 4.12; P = 0.776; F < 1, P = not significant for effect of Hypo and T1D × Hypo).

Assessments of locomotor activity and anxiety on the open field maze (data not shown) did not differ between any groups (active, mobile, or static counts [all F < 1]; time spent in central inner zone or the outer peripheral zone [all F < 1]).

To explore the mechanisms underpinning the cognitive impairments demonstrated in working and long-term memory, transcript abundance of transcription factor Nrf2 as well as that of its target genes Hmox1, Nqo1, Gsta1, Txn1, Txnrd1, and Srxn1 were measured in the hippocampus of each animal model. Significant changes in gene expression were seen in response to T1D and Hypo in both wild-type and Nrf2^−/− animals (Table 1). Both T1D and T1D + Hypo increased mRNA for Hmox1 and Nqo1 in all animals, whereas a marked increase in Nrf2 expression was seen in the hippocampus of T1D + Hypo wild-type mice. As expected, the expression of other Nrf2-regulated genes (Gsta1, Txn1, Txnrd1, and Srxn1) appeared to be significantly diminished in all Nrf2^−/− mice irrespective of the metabolic model applied. Interestingly, mRNA for Gsta1 was significantly decreased in response to Hypo in T1D and nondiabetic WT mice, and this effect was lost in Nrf2^−/− mice (Table 1).

To examine for evidence of ROS-induced cellular damage, levels of lipid peroxidation and protein carbonylation were determined in hippocampal homogenates. Hypo in both nondiabetic and T1D mice increased lipid peroxidation within the hippocampus (Fig. 3A) (Hypo, F[1,30] = 8.36; P < 0.05). Moreover, there was a T1D × Hypo interaction, indicating an additional stimulus to lipid peroxidation in T1D mice exposed to repeated hypoglycemia (T1D × Hypo, F[1,28] = 6.24; P < 0.05). Protein carbonylation, an irreversible process resulting from exposure to ROS and an indicator of severe oxidative damage (32), was significantly elevated in the hippocampus only in T1D mice exposed to Hypo (Fig. 3B) (T1D × Hypo, F[1,28] = 4.35; P < 0.05).

In Nrf2^−/− mice, hippocampal lipid peroxidation (Fig. 3C) was increased by Hypo (F[1,25] = 76.86; P < 0.01), and greatest in T1D Nrf2^−/− animals following Hypo (T1D × Hypo, F[1,25] = 5.07; P < 0.05). Protein carbonylation was increased with Hypo (Fig. 3D) (Hypo, F[1,25] = 38.08; P < 0.01), but not T1D (F < 1; P = not significant for T1D and T1D × Hypo).

To determine whether Hypo in T1D elicited an inflammatory response, cytokine protein levels were measured within hippocampal homogenates. In wild-type mice, IL-1β, IL-2, IL-4, IL-6, and TNF-α were significantly enhanced in T1D animals, whereas exposure to Hypo significantly increased levels of IL-12p70 and IL-5 (Table 2) (all P < 0.05). The combination of Hypo + T1D resulted in a marked stimulus to the production of IL-1β, IL-2, TNF-α, IL-4, IL-6, IL-12p70, and IL-5 (Table 2) (all P < 0.01).

A similar, but more marked pattern in cytokine release was seen in Nrf2^−/− animals. IL-1β, IL-2, TNF-α, IL-4, IL-6, IL-12p70, and IL-5 levels were increased in T1D, whereas IL-6, IL-12p70, and IL-5 were significantly increased following Hypo (Table 2) (all P < 0.05). Again, in Nrf2^−/− mice the combination of Hypo + T1D resulted in a marked stimulus to the production of IL-1β, IL-2, TNF-α, IL-4, IL-6, IL-12p70, and IL-5 (Table 2) (all P < 0.001). Interestingly, levels of all cytokines measured were increased even in untreated Nrf2^−/− mice and amplified following each metabolic stimulus, although the pattern of change was nearly identical to that seen in wild-type mice.

The transcript abundance of the inflammatory genes IL-1β, IL-6, Tnfα, and Nos2 were also checked and showed a similar pattern of change to protein levels of each cytokine, although these increases only reached significance in T1D and T1D + Hypo Nrf2^−/− animals (Supplementary Table 2).

In this study, we show for the first time in a healthy insulin-treated animal model of T1D that recurrent nonsevere hypoglycemia may impact on neuronal integrity and function. We show that eight episodes of nonsevere hypoglycemia in T1D over a 4-week period induces defects in memory consolidation and working memory, which is associated with persisting biochemical evidence of oxidative stress and inflammation in the hippocampus. Interestingly, protein synthesis, which is considered to play an integral role in memory consolidation, is disrupted at many levels by oxidative stress (33). In contrast and consistent with the work of others, recurrent hypoglycemia in nondiabetic rodents did not impair cognitive function (34). A key role for oxidative stress in inducing cognitive impairment was further illustrated by demonstrating a marked amplification of the hippocampal oxidative stress and inflammatory response in mice lacking the Nrf2 transcription factor (following recurrent hypoglycemia).

Preclinical and clinical studies suggest that marked glycemic variability maybe detrimental to humans (22,35–37). In the current study, glucose variability is represented by three principal metabolic states, namely chronic hyperglycemia, acute hypoglycemia, and recovery from acute hypoglycemia. Hypoglycemia can cause oxidative stress (20) and inflammation (21), and hypoglycemia-mediated ROS production can induce apoptosis (38,39). However, the fact that no impairment in cognitive performance was seen following Hypo in wild-type mice implies that the intrinsic antioxidant capacity, coupled with the oxidative stress response orchestrated by Nrf2, is ordinarily sufficient to protect the neuron from the consequences of moderate glucose deprivation. Consistent with this, recurrent hypoglycemia despite recovery to normal glucose levels resulted in cognitive impairment in Nrf2^−/− mice. More recent studies in hippocampal slice preparations have shown that oxidative stress and neuronal death occur primarily in the recovery period from hypoglycemia during glucose reperfusion and that the extent of oxidative stress correlates with the rise in glucose during recovery (13). In our model, glucose levels post-hypoglycemia were >16 mmol/L, and therefore, glucose reperfusion into neurons that have experienced prolonged energy deprivation (as in ischemia reperfusion injury) may be the major contributor to oxidative stress. But Nrf2 protein levels and function are decreased in humans and rodents with diabetes (40,41); therefore, the Nrf2-mediated defense mechanism may also be insufficient to prevent oxidative damage resulting from nonsevere hypoglycemia in T1D. Neuronal damage secondary to severe hypoglycemia is exacerbated in T1D rats compared with nondiabetic Control subjects (14). Taken together, our data provide robust evidence that recurrent nonsevere hypoglycemia in T1D can provoke sufficient oxidative stress to induce a local inflammatory response and result in neuronal dysfunction, but cannot differentiate between potential effects of hypoglycemia per se or recovery to hyperglycemic levels. Future studies will be required to address this clinically important question.

Chronic hyperglycemia increases the production of ROS through mechanisms such as glucose auto-oxidation and nonenzymatic protein glycation (23) and lowers antioxidant defense mechanisms (24,25) as well as serum free-radical trapping capacity (23). In the current study, both T1D wild-type and T1D Nrf2^−/− mice demonstrated impairments in memory consolidation. Neither model had tissue evidence of increased lipid peroxidation or protein carbonylation, suggesting no significant oxidative damage, but both models had significantly increased levels of Nrf2, Hmox-1, Nqo-1, and the inflammatory cytokines IL-1β, IL-2, IL-4, IL-6, and Tnfα. Intriguingly, loss of Nrf2 amplified both the inflammatory response and cognitive defect. It is recognized that inflammatory responses are exacerbated in Nrf2^−/− mice (42,43), and such an outcome is consistent in humans and rodents with diabetes in which Nrf2 levels and activity are diminished (40,41). This is consistent with a recognized hierarchical response to oxidative stressors by which modest levels of ROS activate an Nrf2-orchestrated adaptation, whereas higher levels of ROS stimulate NF-κB and AP-1 to provide an additional defense mechanism (44). In this case, Nrf2^−/− mice would be anticipated to have higher levels of ROS under both hypo- and also hyperglycemia conditions, leading to a proinflammatory state. Recent studies indicating a role for Nrf2 agonists in the treatment of diabetic nephropathy (43) and cardiomyopathy (45) that are associated with chronic hyperglycemia would be consistent with this possibility. The results of the current study would suggest that chronic hyperglycemia induces a proinflammatory condition through Nrf2-dependent and -independent mechanisms.

A critical and novel finding in this study is the important role of the transcription factor Nrf2 in initiating the oxidative stress response to hypoglycemia. Nrf2 is a transcription factor that dictates the intrinsic antioxidant capacity of cells under normal physiological conditions and also directs adaptation to oxidative stress. The activity of Nrf2 is itself regulated through a complex transcriptional/epigenetic and posttranslational network in a manner that ensures its function increases during redox perturbation, inflammation, and nutrient/energy fluxes, thereby enabling the factor to orchestrate adaptive responses to diverse forms of stress (for a review, see Ref. 26). Previously, Kraft et al. (46) have reported that Nrf2^−/− mice are significantly more sensitive to kainate neuronal toxicity than their wild-type counterparts and used microarray analysis to show this was associated with markedly reduced expression in the hippocampus of aldehyde oxidase 1, Gstm1, Gstm3, peroxiredoxin 1 (Prdx1), and Prdx2. In our study, the responses of Nrf2^+/+ and Nrf2^−/− mice to T1D and Hypo were examined in separate experiments, and we therefore cannot strictly compare gene expression profiles between wild-type and knockout mice. Nevertheless, it was apparent from our gene expression analyses that large differences exist between the expression of Gsta1, Gstm1, and Srxn1in the hippocampus of in Nrf2^+/+ and Nrf2^−/− mice. In the future, it will be desirable to examine simultaneously the influence that loss of Nrf2 and genetic upregulation of Nrf2 (caused by diminished expression of Kelch-like ECH-associated protein 1 [Keap1], a negative regulator of Nrf2) has on gene expression and cognitive function following T1D and Hypo. It is notable that we found the levels of mRNA for Nrf2 were significantly upregulated by T1D and very dramatically (by 12-fold) by T1D + Hypo. These results are consistent with a recent report that the protein levels of Nrf2 are increased in the diabetic wounds of humans and mice and that pharmacological activation of Nrf2 promotes wound healing in T1D mice (47). Together with the heightened inflammation in T1D and T1D Nrf2^−/− + Hypo mice relative to their wild-type counterparts, these findings suggest that a function of Nrf2 is to act as a brake to control inflammation. Our current results illustrate the critical importance of this function of Nrf2: indeed, inflammation is controlled, and cognitive function is largely preserved in wild-type animals subjected to hypoglycemia, whereas inflammation is enhanced and cognitive function is severely impaired in their Nrf2^−/− counterparts.

In conclusion, our study supports the hypothesis that recurrent moderate hypoglycemia in T1D may have long-term consequences on cognitive function. Our findings suggest that chronic hyperglycemia, recurrent hypoglycemia, and glucose reperfusion following hypoglycemia in T1D interact synergistically to induce pathological oxidative stress in vulnerable brain regions such as the hippocampus. Whether hypoglycemia per se or glucose recovery from hypoglycemia is the major contributor to oxidative damage in T1D cannot be determined from our studies although preclinical research suggests the recovery period is key. This has implications for clinical practice in which treatment of hypoglycemia often leads to marked rebound hyperglycemia. Moreover, we provide evidence that the transcription factor Nrf2 may be integral to neuronal protection against oxidative stress during and following hypoglycemia and in response to chronic hyperglycemia. This raises the possibility of targeting Nrf2 in developing therapies designed to prevent cellular damage in diabetes induced by recurrent hypoglycemia.

Funding. This work was supported by an award from the University of Dundee/Wellcome Trust Translational Medical Research Fund (TMRF2013 to R.J.M.), Diabetes UK (12/0004531), and JDRF (5-2011-464).

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

Author Contributions. A.D.M. designed and performed experiments and wrote the manuscript. J.R.G. performed experiments. A.T.D.-K., J.D.H., and M.L.J.A. contributed to the discussion and reviewed and edited the manuscript. J.S. contributed to the design of experiments and discussion and reviewed and edited the manuscript. R.J.M. designed the experiments and wrote the manuscript. R.J.M. is the guarantor of this work and, as such, had full access to all the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.

Prior Presentation. Parts of this study were presented in abstract form at the 73rd Scientific Sessions of the American Diabetes Association, Chicago, IL, 21–25 June 2013.