Male C576BL/6J mice (20–25 g; Charles River, UK) were used. The generation (mice were backcrossed over six generations onto a C57BL/6J background) and genotyping of Nrf2^−/− mice lacking Nrf2 (also known as Nfe2l2), kindly provided by K. Itoh and M. Yamamoto (Centre for Tsukuba Advanced Research Alliance and Institute of Basic Medical Sciences, University of Tsukuba, Tsukuba, Japan), were performed as described previously [21]. Mice were housed four per cage with food and water available ad libitum, on a 12 h light–dark schedule. All animal procedures were approved by the University of Dundee Ethical Review Process and performed according to UK Home Office regulations and the ARRIVE 2.0 guidelines (under the auspices of Project License PIL PE82c1898).

C576BL/6J mice were randomly assigned to receive streptozotocin (STZ; 150 mg/kg i.p.) to chemically induce STZ-diabetes or control (Hanks’ Buffered Salt Solution buffer; Gibco, UK; i.p.). At 72 h and 7 days post-injection, blood glucose was measured from tail-vein samples using a hand-held glucose monitor (Accuread, Roche, UK); blood glucose ≥16.0 mmol/l was regarded as diabetic. Any mouse that failed to reach this criterion was given a second injection of STZ, and blood glucose was re-tested. To maintain body weight and health, Linbit insulin implants (LinShin, Canada; at half of the recommended dose [~0.05 U/kg per day]) were inserted subcutaneously under isoflurane anaesthetic as described [13]. Control mice were also anaesthetised.

After 4 weeks of stable hyperglycaemia (STZ-diabetes) or euglycaemia (wild-type [WT] control and Nrf2^−/−mice), the mice underwent surgery for the insertion of vascular catheters as described previously [22]. Mice were allowed to recover for 5 days (or until they reached pre-surgery weight).

As previously described, a 2 h 4 mU kg⁻¹ min⁻¹ infusion of human short-acting insulin (Actrapid, Novo Nordisk, UK) was initiated in mice fasted for 5 h [23]. Mice were then allocated into groups (see Fig. 1 and the Text box detailing mouse groups). Target glucose levels (5.2 mmol/l [euglycaemia], 2.8 mmol/l [hypoglycaemia] and >16 mmol/l [hyperglycaemia]) were achieved and maintained for at least 30 min using a variable 50% glucose infusion based on frequent plasma glucose determinations. Additional blood samples to measure counterregulatory hormones were taken at the end of the second step of the clamp. At the end of the clamp, the mice were given food and water ad libitum and allowed to recover to their endogenous glucose levels (i.e., hyperglycaemia for STZ-diabetes mice and euglycaemia for WT control and Nrf2^−/− mice). Fig. 1

Experimental design of mouse glycaemic clamps. (a) Control (A, B, C) and Nrf2^−/− (G, H, I) mice were exposed to stable euglycaemia (WT- or Nrf2^−/−-EE), hypoglycaemia (~ 2.8 mmol/l) with recovery to euglycaemia (~5.2 mmol/l) (WT- or Nrf2^−/−-LE), or hypoglycaemia with recovery to hyperglycaemia (>16 mmol/l) (WT- or Nrf2^−/−-LH). (b) STZ-diabetic (D, E, F) mice were exposed to stable hyperglycaemia (STZ-HH), hypoglycaemia with recovery to hyperglycaemia (STZ-LH), or hypoglycaemia with recovery to euglycaemia (STZ-LE)

Plasma glucagon and adrenaline levels were measured using commercially available ELISA kits (adrenaline, DEE5100R, Demeditec, Germany; glucagon, 10-1281-01, Mercodia, Sweden).

Sixteen hours after the clamp, mice were killed humanely, and the brain (hippocampus) was dissected and flash-frozen in liquid nitrogen for subsequent biochemical analysis.

Hippocampal samples (single lobe from each mouse) from groups A (WT-EE) and E (STZ-LH) were analysed using SILAC-based proteomic analysis [24]. Samples were randomised, and the analysts were blinded during data acquisition. Samples were homogenised, and the protein extracted in sodium dodecyl sulfate (SDS, 4% wt/vol.) dithiothreitol (DTT, 0.1% wt/vol.), 100 mmol/l Tris-HCl (pH 7.6). After centrifugation, protein lysates from each experimental sample were spiked with an equivalent amount of SILAC protein lysate. After heating to 60°C for 30 min, the samples were alkylated by adding an equal volume of 150 mmol/l iodoacetamide (in 100 mmol/l Tris-HCl buffer [pH 7.6]). The protein was precipitated using the MeOH–chloroform method [25], and the protein concentration was measured using Protein 660 nm reagent (Pierce, UK). Samples were reduced/alkylated, digested with LysC (Pierce, 1:100), and then fractionated using a strong anion exchanger [26].

The top 6 ms/ms programs (collision-induced dissociation [CID] or pulse-Q dissociation [PQD]) on LTQ-Orbitrap (Thermo Scientific, Germany) peptide identification and protein quantification were assessed using Maxquant (Ver 1.5.0.30; www.maxquant.org) or PEAKS 7.0 (Bioinformatics Solution Inc, Canada; database =Uniprot mouse 2017-02-29). Quantification is based on the methods described [24]. Missing data points were replaced (with 0) only for principal component analysis (PCA). A false detection rate (FDR) was set to 1% at the identified peptide spectrum match level. N-terminal acetylation, cysteine carbamidomethylation, and phosphorylation at S/T/Y were the only permitted post-translational modifications. Normalisation was performed using z-score normalisation in Perseus (Version 1.5.4.0, www.maxquant.org).

The second hippocampal lobe from all groups was powdered on liquid nitrogen using a pestle and mortar. A portion of the powder was homogenised in lysis buffer containing protease inhibitors and prepared for western blotting or proteomics analysis as described [27]. The remaining powder was frozen at −80°C for subsequent biochemical analysis. Membranes were probed for the following target proteins identified from SILAC (all from Cell Signalling Technology, UK): heat shock protein 90-β (HSP90B); proteasome subunit α type-2 (PSMA2); proteasome subunit α type-3 (PSMA3); proteasome subunit β type-7 (PSMB7); 6-phosphogluconate dehydrogenase (6PGD); long-chain acyl Co-A dehydrogenase (ACADL); and HSP90 co-chaperone (CDC37). All blots were normalised to the housekeeping protein GAPDH.

Malondialdehyde concentration was determined in all hippocampal samples (A–I) by the thiobarbituric acid-reactive substances assay [28] using a 96-well plate format. The amount of malondialdehyde was determined spectrophotometrically at 532 nm, and concentrations were determined by standard curve. All samples were assayed in duplicate.

The level of carbonylated protein within the hippocampus of all groups (A–I) was measured by ELISA (Caymen Chemicals, US). Protein carbonyl concentration was calculated using the following equation: protein carbonyl (nmol/ml) = CA/(× 0.011 [mmol/l]⁻¹) (500 ml/200 ml), where CA is the corrected absorbance (mean absorbance of controls – mean absorbance of samples).

Total RNA was extracted from hippocampal tissue from all groups (A–I) using TRIzol reagent (Invitrogen, UK). 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 (encoding for nuclear factor erythroid 2-related factor 2 [NRF2]); Nqo1 (encoding for NAD(P)H: quinone oxidoreductase 1); Hmox-1 (encoding for haem oxygenase 1); and Sod2 (encoding for superoxide dismutase 2). All samples were performed in triplicate and normalised to the housekeeping genes Actb and Ppia. Values are expressed as a fold-change relative to group A (WT-EE) for STZ-diabetic mice and group G (Nrf2^−/−-EE) for Nrf2^−/− mice.

Data were analysed using SPSS version 18 (IBM, UK). One-way ANOVA was used to compare clamp groups within each genotype (groups A–F for control [WT] and STZ-diabetes mice; groups G–I for Nrf2^−/− mice). Post hoc analysis was performed using Tukey’s multiple comparisons test. For data that were not normally distributed, Kruskal–Wallis, followed by Dunn’s multiple comparisons test, was performed. Data are expressed as mean values ± SEM. Statistical significance was set at p<0.05.

Stable hypoglycaemic (groups B, C, E and F) and hyperglycaemic plateaus (groups C, D, E and F) were achieved during the clamp procedures (Table 1; p<0.05 for each group vs WT control [group A]). In groups B and C, hypoglycaemia from a euglycaemic baseline resulted in significantly elevated glucagon and adrenaline plasma levels compared with group A. In contrast, in the STZ-diabetic mice (groups E and F), consistent with human type 1 diabetes, the glucagon response to a hypoglycaemic challenge was impaired (Table 1) [29] and the adrenaline response was severely blunted. The hormonal counterregulatory response to hypoglycaemia in Nrf2^−/− mice was comparable with that in C57Bl6/J control mice (Table 1). Table 1

Mean plasma glucose levels during each phase of the hyperinsulinaemic glucose clamps along with counterregulatory hormone levels (glucagon and adrenaline) measured at the end of the eu/hypoglycaemia period

To examine the impact of hypoglycaemia on Nrf2 and NRF2 target genes Nqo1, Sod2 and Hmox-1, their expression levels were measured in the hippocampus of all control and STZ-diabetes groups (electronic supplementary material [ESM] Table 1). Transcript levels of Nqo1 and Sod2 were significantly elevated in STZ-diabetic mice following acute hypoglycaemia (STZ-LH vs WT-EE; p<0.05 for both genes), and the levels of Sod2 were further increased (>fivefold) in chronic hyperglycaemia. In WT non-diabetic mice, Sod2 and Hmox-1 transcript levels were significantly elevated by hypoglycaemia (WT-LE vs WT-EE; p<0.05). As anticipated, RNA levels of these NRF2 target genes were unaltered in Nrf2-knockout mice (ESM Table 2), demonstrating NRF2 dependence.

In non-diabetic WT control mice, acute hypoglycaemia did not significantly increase lipid peroxidation irrespective of the glucose level at which the clamp finished (Fig. 2b; WT-EE vs WT-LE, p>0.05; WT-EE vs WT-LH, p>0.05). In contrast, hippocampal lipid peroxidation was significantly increased in all STZ-diabetic models, with the most significant effect seen where there was post-hypoglycaemic hyperglycaemia (Fig. 2a; STZ-LH vs WT-EE, p<0.01). In STZ-diabetes, maintaining post-hypoglycaemic euglycaemia ameliorated this effect (STZ-LE vs STZ-LH, p<0.05). The levels of lipid peroxidation in Nrf2^−/− mice were elevated in all conditions when compared with control (WT-EE) mice (Fig. 2c; main effect of genotype, p<0.01). Fig. 2

Chronic hyperglycaemia is associated with hippocampal oxidative damage. (a) Levels of hippocampal lipid peroxidation were increased in STZ-diabetic mice (white bars) compared with control (WT) mice (black bars) maintained at euglycaemia. (b) Euglycaemic control mice exposed to an acute episode of hypoglycaemia exhibited no change in hippocampal lipid peroxidation. (c) Euglycaemic Nrf2^−/− mice (grey bars) displayed increased levels of hippocampal lipid peroxidation irrespective of hypoglycaemic challenge. (d) Protein carbonylation levels were elevated in STZ-diabetic mice exposed to hyperglycaemia compared with control mice at euglycaemia. (e) Control WT mice exposed to an acute hypoglycaemic episode showed a rise in protein carbonylation only when recovered to a hyperglycaemic state. (f) Nrf2^−/− mice displayed increased levels of protein carbonylation irrespective of glycaemic variability. n=4–7/group. Results represent mean values ± SEM. *p<0.05, **p<0.01 vs WT-EE; ^¶p<0.05 vs STZ-diabetes (one-way ANOVA followed by Tukey post hoc test). E, euglycaemia; H, high, hyperglycaemia; L, low, hypoglycaemia; MDA, malondialdehyde

Protein carbonylation is commonly used as a biomarker of oxidative damage for many proteins. Levels increase with age, and this increase has been linked to changes in specific enzymes, such as members of the tyrosine kinase family [30], GLUT4 [31] and the 19s and 20s proteasomal subunits [32], and to diseases such as diabetes [33–35]. In non-diabetic WT mice, there was no impact of a single acute hypoglycaemic challenge on levels of carbonylated proteins when returned to euglycaemic levels (Fig. 2e; WT-EE vs WT-LE, p>0.05). In contrast, in STZ-diabetic mice, hypoglycaemia followed by recovery to hyperglycaemia resulted in a marked increase in protein carbonylation (Fig. 2d; WT-EE vs STZ-LH, p<0.01). There were also small but significant increases in carbonylated protein levels in STZ-diabetic mice that had not been exposed to hypoglycaemia (Fig. 2d; WT-EE vs STZ-HH, p<0.05), as well as non-diabetic mice who were exposed to post-hypoglycaemic hyperglycaemia (Fig. 2e; WT-LH vs WT-EE, p<0.05). Interestingly, recovery of STZ-diabetic mice to euglycaemia largely reversed the increase in protein carbonylation (Fig. 2d; WT-EE vs STZ-LE, p>0.05). Notably, levels of carbonylated proteins were significantly elevated in the hippocampus of all Nrf2^−/− mice compared with non-diabetic WT mice (Fig. 2f; main effect of genotype, p<0.01).

SILAC is a method of accurately quantifying changes in protein expression [24]. In vivo SILAC with label-free proteomics was used to assess changes in hippocampal protein expression in STZ-diabetic mice exposed to post-hypoglycaemic hyperglycaemia (STZ-LH, group E) compared with control mice (WT-EE, group A). This procedure identified 71 proteins that were differentially expressed between groups (ESM Table 3 [upregulated proteins] and ESM Table 4 [downregulated proteins]). Pathway analysis identified significant upregulation of proteins involved in long-chain fatty acid metabolism (predominantly β-oxidation) and components of the proteasome, suggesting an enhanced capacity for long-chain fatty acid oxidation and the degradation of damaged proteins (ESM Table 3). Conversely, significant downregulation of proteins involved in mediating the stress response, including several heat shock proteins, was observed (ESM Table 4).

We then examined candidate proteins from the key pathways identified in the SILAC analysis (fatty acid metabolism, proteasomal degradation and chaperone/stress response) across all study groups. ACADL, a mitochondrial protein involved in the initial step of fatty acid β-oxidation, was increased following hypoglycaemia in STZ-diabetic mice, an effect that was not seen when glucose was recovered to euglycaemia (Fig. 3a). In addition, we considered upregulation of 6PGD of interest in relation to the oxidative damage associated with the post-hypoglycaemic hyperglycaemic phase (ESM Table 3). 6PGD is a key enzyme of the oxidative arm of the pentose phosphate pathway (PPP) and the largest contributor to cytosolic NADPH, an important component of cellular antioxidant defences. 6PGD was enhanced in control and STZ-diabetic mice exposed to an acute hypoglycaemic challenge compared with control mice, although the impact of hypoglycaemia was less pronounced in STZ-diabetic mice (Fig. 3e, i; WT-EE vs STZ-LH, p<0.05; WT-EE vs WT-LE, p<0.01). Fig. 3

Effect of glycaemic variability on SILAC-outcome selected protein levels in the hippocampus of WT control and STZ-diabetic mice. (a–g) Hippocampal protein levels (ratio of signal intensities to control euglycaemia [WT-EE] data) in mice exposed to an acute hypoglycaemic episode from a euglycaemic (WT control mice, black bars) or hyperglycaemic (STZ-diabetic mice, white bars) baseline and returned to euglycaemia or hyperglycaemia: ACADL (a); PSMA2 (b); PSMA3 (c); PSMB7 (d); 6PGD (e); HSP90B (f); and CDC37 (g). (h–j) Representative immunoblots of ACADL, PSMB7 and PSMA2 (h), HSP90B, 6PGD and PSMA3 (i), and CDC37 (j) and their respective GAPDH loading controls. Results represent mean values ± SEM. *p<0.05, **p<0.01, ***p<0.001 (Kruskal–Wallis one-way ANOVA followed by Dunn’s multiple comparisons test). E, euglycaemia; H, high, hyperglycaemia; L, low, hypoglycaemia

PSMA2 (Fig. 3b, h), PSMA3 (Fig. 3c, i) and PSMB7 (Fig. 3d, h), which form part of the 20S core structure, were all significantly increased following exposure to hypoglycaemia in both non-diabetic and STZ-diabetic mice (all p<0.05). HSP90, a chaperone protein that assists in correct protein folding and aids degradation of damaged proteins [36], was reduced following hypoglycaemia in STZ-diabetic mice (Fig. 3f, i; WT-EE vs STZ-LH, p<0.05). This contrasts with non-diabetic mice where acute hypoglycaemic challenge induced an increase in expression of HSP90B (Fig. 3f, i; WT-EE vs WT-LE, p<0.05). Similarly, hypoglycaemia in STZ-diabetic but not non-diabetic mice downregulated CDC37, an HSP90B co-chaperone protein (Fig. 3g). Interestingly, this effect was lost when STZ-diabetic mice were recovered to euglycaemia; however, CDC37 was also suppressed in non-diabetic mice recovered from hypoglycaemia to hyperglycaemia, suggesting that post-hypoglycaemic hyperglycaemia suppresses CDC37.

To further examine the role of NRF2 in mediating protection against the oxidative stress associated with both hyper- and hypoglycaemia, we also assessed the impact of acute changes in glycaemia on hippocampal levels of these proteins. Protein abundance of the mitochondrial protein ACADL was significantly elevated, whereas 6PGD did not increase in Nrf2^−/− mice (Fig. 4a, e). The increase in 6PGD was also seen in STZ-diabetic mice that had been exposed to acute hypoglycaemia euglycaemia (Fig. 3e). Exposure to hypoglycaemia increased the expression of PSMA3 and PSMB7 (Fig. 4c, i; p<0.05 vs WT-EE; and Fig. 4d, h; p<0.01 vs WT-EE) in Nrf2^−/− mouse hippocampus, with a non-statistically significant increase in PSMA2 (Fig. 4b, h; p=0.07). Similarly, the pattern of change in HSP90B after hypoglycaemia in Nrf2^−/− mice was also seen in STZ-diabetic mice but not non-diabetic WT mice who experienced post-hypoglycaemia (Fig. 4f). This suggests roles for NRF2 particularly in mediating the increase in reductive biosynthesis and chaperone/stress responses, which appear key pathways in the cellular response to hypoglycaemia. Fig. 4

Effect of glycaemic variability on SILAC-outcome selected protein levels in the hippocampus of Nrf2^{− /−} mice. (a–g). Hippocampal protein levels (ratio of signal intensities to Nrf2^-/- mice at euglycaemia [Nrf2^-/--EE] data) in Nrf2^−/− mice exposed to an acute hypoglycaemic episode from a euglycaemic baseline and returned to euglycaemia or hyperglycaemia: ACADL (a); PSMA2 (b); PSMA3 (c); PSMB7 (d); 6PGD (e); HSP90B (f); and CDC37 (g). (h–j) Representative immunoblots of ACADL, PSMB7 and PSMA2 (h), HSP90B, 6PGD and PSMA3 (i), and CDC37 (j), and their respective GAPDH loading controls. Results represent mean values ± SEM. *p<0.05, **p<0.01 (Kruskal–Wallis one-way ANOVA followed by Dunn’s multiple comparisons test). E, euglycaemia; H, high, hyperglycaemia; L, low, hypoglycaemia