Genetic reduction of Nrf2 exacerbates cognitive deficits in a mouse model of Alzheimer’s disease

Abstract

Aging is the major risk factor for several neurodegenerative diseases, including Alzheimer’s disease (AD). However, the mechanisms by which aging contributes to neurodegeneration remain elusive. The nuclear factor (erythroid-derived 2)-like 2 (Nrf2) is a transcription factor that regulates expression of a vast number of genes by binding to the antioxidant response element. Nrf2 levels decrease as a function of age, and reduced Nrf2 levels have been reported in postmortem human brains and animal models of AD. Nevertheless, it is still unknown whether Nrf2 plays a role in the cognitive deficits associated with AD. To address this question, we used a genetic approach to remove the Nrf2 gene from APP/PS1 mice, a widely used animal model of AD. We found that the lack of Nrf2 significantly exacerbates cognitive deficits in APP/PS1, without altering gross motor function. Specifically, we found an exacerbation of deficits in spatial learning and memory, as well as in working and associative memory. Different brain regions control these behavioral tests, indicating that the lack of Nrf2 has a global effect on brain function. The changes in cognition were linked to an increase in Aβ and interferon-gamma (IFNγ) levels, and microgliosis. The changes in IFNγ levels are noteworthy as previously published evidence indicates that IFNγ can increase microglia activation and induce Aβ production. Our data suggest a clear link between Nrf2 and AD-mediated cognitive decline and further strengthen the connection between Nrf2 and AD.

Introduction

The nuclear factor (erythroid-derived 2)-like 2 (Nrf2) is a transcription factor that regulates the expression of a vast number of genes by binding to the antioxidant response element (1,2). While Nrf2 is considered a master regulator of the cellular redox status, it also controls the expression of genes involved in the overall metabolic reprogramming during stress and inflammation (3,4). Physiologically, Nrf2 function is regulated by Kelch ECH-associated protein (Keap1; 5,6). Keap1 binds to Nrf2 in the cytoplasm and targets it for proteasomal degradation (7). In contrast, high levels of reactive oxygen species and inflammation reduce Keap1 levels leading to migration of Nrf2 into the nucleus where it induces gene expression (8).

Alzheimer’s disease (AD) is the most common neurodegenerative disorder (9). Neuropathologically, AD is characterized by the accumulation of plaques, mainly made of amyloid-β (Aβ), and tangles, mostly made of hyperphosphorylated tau (10). Aging is the major risk factor for both sporadic and familial AD. The latter is caused by mutations in the amyloid precursor protein, presenilin 1, or presenilin 2 genes, while the former is of unknown causes (10). Therefore, identifying the mechanisms by which aging contributes to the development of AD may unveil new critical insights into the pathogenesis of this insidious disorder. Reduced Nrf2 activity is linked to aging and age-dependent neurodegenerative diseases (11). Evidence indicates that oxidative stress increases as a function of age in many organs, including the brain (12,13). To this end, highly reactive oxygen species are another constant feature of AD (14,15). For example, the levels of several detoxifying enzymes are reduced in AD brains compared with control cases (16–19); these changes are linked to a reduction in nuclear Nrf2 levels in AD brains compared with control cases (20). In contrast, whether the levels of Nrf2 change in animal models of AD remain unclear as contradictory reports exist in the literature (21–24).

Overwhelming data support a primary role of inflammation in AD (25). Neuropathologically, several laboratories have reported an increase in brain inflammation in AD brains. Genetically, genome-wide association data have consistently highlighted a link between AD and genes associated with the inflammatory response (26). Nrf2 regulates brain inflammation by controlling the expression of several inflammatory markers (27). For example, Nrf2 knockout mice (Nrf2^−/−) show an exacerbated hippocampal inflammation, including microgliosis, in response to lipopolysaccharide injection (28). Nrf2^−/− injected with 1-methyl-4-phenyl-1, 2, 3, 6-tetrahydropyridine (MPTP) show a similar oversensitivity to brain inflammation. Notably, overexpression of Nrf2 in glia cells protected against MPTP-mediated toxicity (29). Despite the large body of evidence linking Nrf2 to neurodegenerative diseases, the effect of reducing Nrf2 levels on cognitive decline associated with AD-like pathology remain elusive. In this manuscript, we used a genetic approach to filling this gap in knowledge.

Results

To study the role of Nrf2 in AD pathogenesis, we used a cross-breeding approach and bred Nrf2 knockout mice (Nrf2^−/−) with APP/PS1 mice (APP/PS1^+/0). We obtained mice with the following genotypes: (i) APP/PS1 with both endogenous copies of the Nrf2 gene. (ii) APP/PS1 lacking both copies of the Nrf2 endogenous gene (APP/PS1;Nrf2^−/−). (iii) Wildtype mice with both copies of the Nrf2 gene (WT). (iv) Wild type mice lacking both copies of the Nrf2 gene (Nrf2^−/−). Notably, all four groups of mice were littermates (Fig. 1A). We measured body weight monthly and found that all four groups of mice gained weight at the same pace during the duration of the experiment (Fig. 1B). To confirm the removal of the Nrf2 genes, we measured its mRNA levels by quantitative real-time PCR. Consistent with the genotype of the mice, Nrf2 was clearly detectable only in WT and APP/PS1 mice (APP/PS1 vs. APP/PS1;Nrf2^−/−: P = 0.0013; WT vs. Nrf2^−/−: P = 0.0024). Notably, the presence of the APP transgene did not change Nrf2 mRNA levels (WT vs APP: P = 0.8568, Fig. 1C). To confirm the lack of the Nrf2 gene functionally, we measured heme oxygenase 1 (HO-1) levels by western blot, as HO-1 is an essential Nrf2 target gene (3). As expected, we found that both the Nrf2^−/− and the APP/PS1;Nrf2^−/− mice had significantly lower HO-1 levels compared with the other two groups (APP/PS1 vs. APP/PS1;Nrf2^−/−: P = 0.0234; WT vs. Nrf2^−/−: P = 0.0093; Fig. 1D–E). These data suggest that there were no major compensatory mechanisms during gestation or aging in the Nrf2^−/− mice. Notably, while the levels of HO-1 were higher in APP/PS1 mice compared with WT mice, this difference was not statistically significant. This finding is not consistent with a recent report showing an increase in HO-1 levels in APP/PS1 mice (30). While the cause behind this apparent contradiction is not clear, the lack of quantitative analyses and the low number of animals used by Xing and colleagues might confound the interpretation of their results. Another possible explanation is the different background of the APP/PS1 mice used by Xing and colleagues compared with the ones used here.

Figure 1.

Removing the Nrf2 genes from APP/PS1 mice does not alter body weight. (A) Schematic representation of the breeding strategy used to delete both copies of the Nrf2 gene from APP/PS1 mice. Four littermate groups were generated and used for the experiments described here: WT mice (APP/PS1^0/0; Nrf2^+/+, n =  5 females and 3 males), APP/PS1 mice (APP/PS1^+/0;Nrf2^+/+, n =  5 females and 8 males), Nrf2^−/− mice (APP/PS1^0/0;Nrf2^−/−, n =  7 females and 5 males), and APP/PS1;Nrf2^−/− mice (APP/PS1^+/0;Nrf2^−/−, n =  8 females and 4 males). (B) The graph shows the body weight of mice, which was taken monthly. All groups gained weight at the same pace, and no statistically significant differences were detected for any of the genotypes. Data are presented as means ± SEM and were analysed by two-way ANOVA. (C) The graph shows Nrf2 mRNA levels in the different genotypes (n =  4/genotype). As expected, Nrf2 was only detected in WT and APP/PS1 mice. (D) Representative western blots of protein extracted from brains of 12-month-old mice probed with the indicated antibodies. (E) A significant reduction of HO-1 protein was detected in the brain of mice lacking the Nrf2 gene. Data were generated by normalizing the levels of the protein of interest to β-actin loading control. Data are presented as box plots and were analysed by unpaired t-test.

We assessed non-cognitive behavior by measuring general motor function and anxiety and stress in 11-month-old mice (n = 5 females and 3 males for WT, n = 5 females and 8 males for APP/PS1, n = 7 females and 5 males for Nrf2^−/−, and n = 8 females and 4 males for APP/PS1;Nrf2^−/−). We found that spontaneous activity and gross motor function were similar among the four groups, as indicated by distance traveled and average speed in the testing area during the open field test (Fig. 2A and B). To determine the effects of removing Nrf2 on cognitive function, we tested the four groups of mice in the Morris water maze (MWM), in the radial arm water maze (RAWM), and the contextual fear conditioning (CFC). These three tasks are routinely used to assess spatial learning and memory (MWM and RAWM), working memory (RAWM), and association learning (CFC; 31). We started with the MWM by training mice to find a hidden platform using extra-maze cues; mice were given four training trials per day per five consecutive days. When we analysed the time to find the hidden platform, we found a significant effect for days [P < 0.0001; F(4, 220) = 30.71] and genotype [P < 0.0001; F(3, 220) = 13.93; Fig. 2C]. The former indicates that all mice learned the task across the 5 days of training. The latter indicates that one or more genotypes had a different pace of learning. Post hoc analyses indicated that APP/PS1 mice performed significantly worse than WT mice at day 4 (P < 0.05) and day 5 (P < 0.05). Notably, removing the Nrf2 gene, exacerbated this difference, as APP/PS1;Nrf2^−/− mice performed significantly worse than WT mice at day 3 (P < 0.05), day 4 (P < 0.001), and day 5 (P < 0.001). We found similar results when we analysed the distance traveled to find the hidden platform: there was a significant effect for days [P < 0.0001, F(4, 220) = 14.30] and genotype [P < 0.0001, F(3, 220) = 12.92; Fig. 2D]. Post hoc analyses indicated that APP/PS1 mice performed significantly worse than WT mice at day 4 (P < 0.05) and day 5 (P < 0.05). However, APP/PS1;Nrf2^−/− mice performed significantly worse than WT mice at day 3 (P < 0.05), day 4 (P < 0.001), and day 5 (P < 0.001).

Figure 2.

Lack of Nrf2 exacerbates cognitive function in APP/PS1 mice. (A,B) The graphs show total distance traveled and speed during the open field test. The data were not statistically different among the four groups (P = 0.8112 and 0.8661, respectively). (C,D) Learning curves of 11-month-old mice trained in the spatial reference version of the MWM. Each day represents the average of four training trials. For the escape latency (time), we found a significant effect for day and genotype [day effect: P < 0.0001; F(4, 220) = 30.71; genotype effect: P < 0.0001; F(3, 220) = 13.93]. APP/PS1 mice were impaired at day 4 and 5, while APP/PS1;Nrf2^−/− were impaired starting from day 3. The same results were obtained when analysing the distance traveled to find the platform [day effect: P < 0.0001, F(4, 220) = 14.30; genotype effect: P < 0.0001, F(3, 220) = 12.92]. Data were analysed by two-way ANOVA. (E) The graph shows the numbers of platform location crosses (frequency) during a single 60-s probe trial. We found a significant difference among groups (P < 0.0001). Post hoc analysis with Bonferroni’s correction showed that the WT mice performed better than all the other groups (P < 0.001 vs. APP/PS1 and APP/PS1;Nrf2^−/−, P < 0.05 vs. Nrf2^−/−). (F,G) The graphs show the time mice spent in the target and the opposite quadrants during the probe trials. In both cases, only the APP/PS1;Nrf2^−/− performed at chance level (15 s). (H) Average swim speed during the probe trials. The data were not statistically different among the four groups (P = 0.3591). (I,J) Mice were evaluated in the RAWM. The graphs show the number of reference errors during the two days of testing (I) and the second day (J). All four groups learned the task but APP/PS1;Nrf2^−/− mice performed significantly worse than WT mice on day 2 (P < 0.01). (K,L) Mice were evaluated in the RAWM. The graphs show the number of working memory errors during the two days of testing (K) and the second day (L). All four groups learned the task but APP/PS1;Nrf2^−/− mice performed significantly worse than WT mice on day 2 (P < 0.01). (M) Mice were tested in the CFC task. The graph shows the percentage freezing on different days. Mice received a mild foot shock on day 0 only. For the other days, they were re-exposed to the environment without receiving further foot shocks. The APP/PS1;Nrf2^−/− was the only group that did not extinguish the memory after repetitive exposure to the same environment [days effect, P < 0.0001, F(5, 210) = 16.93; genotype effect, P = 0.001, F(3, 210) = 5.650]. Post hoc analysis with Bonferroni’s correction showed that on day 5 (P < 0.05) and 6 (P < 0.01) APP/PS1;Nrf2^−/− mice performed worse than the other groups. Data in panels A, B, E-H, J, and L are presented as box plots and were analysed by one-way ANOVA. Data in panels C-D, I, K, and M are presented as means ± SEM and were analysed by two-way ANOVA.

Twenty-four hours after the last training trial, we removed the platform from the maze and conducted probe trials to assess spatial memory during a single 60-s probe trial. We found that the number of platform location crosses was significantly different among the four genotypes [P < 0.0001, F(3, 43) = 22.69; Fig. 2E]. Post hoc analyses indicated that WT mice performed significantly better than the other three groups; however, the difference between APP/PS1 and APP/PS1;Nrf2^−/− mice was not statistically significant, which is likely due to a flooring effect. We then analysed the time mice spent in the target and the opposite quadrants during the probe trials. We found that the time spent in the target and opposite quadrants was significantly different among the four groups [P < 0.0001, F(3, 43) = 9.51; and P = 0.0017, F(3, 43) = 6.06, respectively; Fig. 2F and G]. Notably, while the WT mice spent significantly more time in the target quadrant compared with the APP/PS1 and APP/PS1;Nrf2^−/− mice (P < 0.05 and P < 0.001, respectively), only the APP/PS1;Nrf2^−/− performed at a chance level, with a group average of 13.08 ± 1.40 s [P = 0.1970, t(11) = 1.373]. This performance was significantly worse than APP/PS1 mice, which spent significantly more time in the target quadrant, with a group average of 25.78 ± 2.04 s [P = 0.0209, t(11) = 2.695]. We obtained similar results when we analysed the time spent in the opposite target [for APP/PS1: average time = 9.90 ± 1.51, P = 0.0061, t(11) = 3.380; for APP/PS1;Nrf2^−/−: average time = 16.10 ± 1.26, P = 0.4023, t(11) = 0.8712]. Notably, swim speed was not statistically significant among the four groups (Fig. 2H), suggesting that the performance deficits are not due to motor issues.

To further assess cognitive function in mice with and without Nrf2, we tested all groups in the RAWM, which measures reference and working memory, by counting the number of errors made between trials and within the same trial, respectively. For reference memory, we found that all groups learned the task as indicated by the significantly lower reference errors made during the probe trials (day 2) compared with the learning trials (day 1). Specifically, we found an effect for days [P < 0.0001, F(1, 78) = 59.13] and genotype [P = 0.001, F(3, 78) = 5.664; Fig. 2I]. The latter indicates that there was a difference in the pace at which one or more of the groups learned. To this end, post hoc analyses on day 2 indicated that APP/PS1 mice were not significantly different than WT or Nrf2^−/− mice. In contrast, APP/PS1;Nrf2^−/− mice performed significantly worse than WT and Nrf2^−/− mice (P < 0.01 and P < 0.05, respectively; Fig. 2J). For working memory, we also found an effect for days [P < 0.0001, F(1, 78) = 41.11] and genotype [P = 0.0005, F(3, 78) = 6.603; Fig. 2K]. Notably, post hoc analyses indicated that on day 2, APP/PS1;Nrf2^−/− mice performed significantly worse than WT and Nrf2^−/− mice (P < 0.01 and P < 0.05, respectively; Fig. 2L).

We next assessed changes in contextual fear conditioning and extinction among the four groups. To this end, we placed mice in a conditioning box where they received a mild foot shock. Subsequently, mice were replaced in the same conditioning box once daily for six consecutive days, without receiving additional foot shock. For all groups, the freezing time was significantly higher on day one, 24 h after the initial shock, compared with day 0, before the shock. There were no differences among the four groups [days effect, P < 0.0001, F(1, 78) = 87.47; genotype effect, P = 0.5162, F(3, 78) = 0.7674; Fig. 2M]. These data indicate that all groups make a strong association between the environment and the aversive stimulus. However, under physiological condition, if mice are replaced in the context several times without receiving new foot shocks, they will relearn a new association, a process known as extinction (32). Indeed, the freezing time of WT mice went from 40.00 ± 9.53% on day 1 to 13.14 ± 2.93% on day 6, Fig. 2M). We observed a similar reduction in freezing time for Nrf2^−/− and APP/PS1 mice. The only group that did not show any significant extinction was the APP/PS1;Nrf2^−/− mice, which showed a freezing time of 48.08 ± 7.28% and 47.58 ± 7.62% on day 1 and day 6, respectively [days effect, P < 0.0001, F(5, 210) = 16.93; genotype effect, P = 0.001, F(3, 210) = 5.650]. Post hoc analysis with Bonferroni’s correction showed that on day 4 (P < 0.05) and 6 (P < 0.01) APP/PS1;Nrf2^−/− mice performed worse than the other groups (Fig. 2M). Taken together, the behavioral data indicate that removing the gene encoding Nrf2 exacerbates cognitive function in APP/PS1 mice.

At the end of the behavioral testing, we sacrificed all mice, which at this point were 12 months of age, and used their brains for neuropathological assessment. To determine whether removing the gene encoding Nrf2 influenced Aβ pathology, we first immunostained sections from APP/PS1 and APP/PS1;Nrf2^−/− with an Aβ42-specific antibody (Fig. 3A and B). We found that the total plaque count in the cortex and the hippocampus was not statistically different between the two groups (Fig. 3C). We then measured soluble and insoluble Aβ levels by sandwich ELISA. In the soluble fraction, we found that Aβ40 and Aβ42 levels were significantly higher in APP/PS1;Nrf2^−/− mice compared with APP/PS1 mice [for Aβ40: P = 0.01; t(13) = 2.85. For Aβ42: P = 0.01; t(13) = 2.70; Fig. 3D]. In the insoluble fraction, only Aβ42 levels were significantly different between the two groups [P = 0.01; t(13) = 2.77; Fig. 3E]. To probe for possible mechanisms that could account for the changes in Aβ levels after removing the Nrf2 gene, we measured the steady-state levels of full-length APP and its two major C-terminal fragments, C99 and C83. For all of these measurements, we found that removing the Nrf2 gene in the WT or the APP/PS1 mice has no effects (Fig. 3F–I). These data suggest that the increase in Aβ levels is not due to an increase in Aβ production.

Figure 3.

Lack of Nrf2 increases Aβ levels but not plaque number in APP/PS1 mice. (A,B) Representative microphotographs of brain sections immunostained with an Aβ₄₂ specific antibody. (C) Quantitative analysis of the Aβ₄₂ immunoreactivity in the cortex and hippocampus showed no differences among groups (P = 0.6568 for the cortex, P = 0.4785 for the hippocampus, n =  6/genotype). (D,E) Aβ₄₀ and Aβ₄₂ levels measured by ELISA (n = 8/genotype). In the soluble fraction, both Aβ₄₀ and Aβ₄₂ levels were significantly different between APP/PS1 and APP/PS1;Nrf2^−/− mice (P = 0.0143 and 0.0182, respectively). In the insoluble fraction, Aβ₄₂ levels were significantly different between APP/PS1 and APP/PS1;Nrf2^−/− mice, while no differences were detected for Aβ₄₀ (P = 0.0158 and 0.4484, respectively). (F) Representative western blots of protein extracted from brains of 12-month-old mice probed with the indicated antibodies. (G–I) Quantitative analyses of the blots. In APP/PS1 and APP/PS1;Nrf2^−/− mice the levels of APP, C83, and C99 were significantly higher than the two WT groups. These changes were independent of Nrf2. Data are presented as box plots and were analysed by unpaired t-test (D,E) or one-way ANOVA (G–I).

The steady-state levels of Aβ levels are regulated by its production and its degradation. Multiple laboratories have reported that Aβ can be degraded by both the proteasome and autophagy (33–35). We, therefore, measured the trypsin, chymotrypsin, and caspase-like activities of the proteasome using selective fluorogenic substrates. While the caspase-like activity was not statistically different among the four groups (Fig. 4A), we found that the trypsin and chymotrypsin activities were [P < 0.0001 and F(3, 47) = 10.90, P < 0.0001 and F(3, 47) = 29.89, respectively; Fig. 4B and C]. However, this difference was driven by WT mice, which had the highest levels of proteasomal function compared with the other three groups. These data suggest that in WT mice, removal of Nrf2 decreases two of the three major catalytic activities of the proteasome. However, in APP/PS1 mice, which already have significantly lower levels of trypsin and chymotrypsin activities compared with WT mice, removal of Nrf2 had no effects, suggesting a flooring effect for Nrf2-mediated changes on proteasomal function.

Figure 4.

Lack of Nrf2 does not change proteasome or autophagy activity in APP/PS1 mice. (A–C) Proteasomal activity was evaluated using selective fluorogenic substrates. No differences among genotypes were detected for the caspase-like activity. Trypsin- or chymotrypsin- activities were different among groups (P < 0.0001 for both). Post hoc analysis with Bonferroni’s correction highlighted that these differences were driven by the WT group (P < 0.001 for both). (D) Representative western blots of protein extracted from brains of 12-month-old mice probed with the indicated antibodies. (E–I) Quantitative analyses of the blots show no differences among genotypes for any of the proteins analysed (P = 0.8690 for p-mTOR, P = 0.4928 for p-p70S6K, P = 0.7004 for Atg3, P = 0.2273 for Atg5, and P = 0.4092 for Atg7). Data are presented as box plots and were analysed by one-way ANOVA.

The second major intracellular turnover mechanism is autophagy. mTOR is one of the major negative regulators of autophagy, and its activity is increased in AD (25,36,37). Routinely, mTOR activity is assessed by measuring the steady-state levels of p70S6K1, a kinase directly phosphorylated by mTOR. We found that the levels of phosphorylated mTOR and p70S6K1 were not statistically significant among the four groups (Fig. 4D–F). Consistent with these data, the levels of Atg3, Atg5, and Atg7, three key autophagy-related proteins involved in autophagy induction, were also not statistically different among the four groups (Fig. 4D, G–I). Together, these data indicate that a decrease in protein turnover is most likely not associated with the increase in Aβ levels.

Nrf2 plays a key role in oxidative stress response (11,24). To evaluate the oxidative stress status of the different genotypes, we employed two distinct approaches. First, we assessed the levels of 4-Hydroxynonenal (4-HNE) by western blot, an index routinely used for lipid peroxidation. Notably, we found that 4-HNE levels were higher in transgenic mice compared with wildtype mice, independent of the Nrf2 genotype [APP transgene effect: P = 0.0278 and F(1, 24) = 5.518. Nrf2 effect: P = 0.6554 and F(1, 24) = 0.2044. Fig. 5A and B]. We then evaluated protein carbonyl levels, which are routinely used as a marker of protein oxidation. We detected that there was a difference in protein oxidation driven by the genotype [APP transgene effect: P = 0.0039 and F(1, 24) = 9.127. Nrf2 effect: P = 0.0141 and F(1, 24) = 6.451. Fig. 5C]. Post hoc analyses indicated that the WT mice had lower protein carbonyl levels than Nrf2^−/− mice, in which protein carbonyl levels were similar to APP/PS1 mice. Moreover, removing Nrf2 from APP/PS1 did not further increase protein oxidation, suggesting a possible ceiling effect.

Figure 5.

Lack of Nrf2 affects oxidative stress in APP/PS1 mice. (A) Representative western blots of proteins extracted from brains of 12-month-old mice probed with the indicated antibodies (n =  7/genotype). (B) Quantitative analyses of the blots show higher 4-HNE levels in presence of the APP transgene, independent of the Nrf2 genotype [APP transgene effect: P = 0.0278 and F(1, 24) = 5.518. Nrf2 effect: P = 0.6554 and F(1, 24) = 0.2044]. (C) Protein carbonyl levels were measured by ELISA (n =  7/genotype) in the soluble fraction. Both APP transgene and Nrf2 genotype affected protein carbonyl levels [APP transgene effect: P = 0.0039 and F(1, 24) = 9.127. Nrf2 effect: P = 0.0141 and F(1, 24) = 6.451. Fig. 5C]. No differences were detected between APP/PS1 and APP/PS1;Nrf2^−/− mice. Data are presented as box plots, and were analysed by two-way ANOVA.

Extensive evidence supports a role for Nrf2 in brain inflammation (e.g. 27,28). In turn, brain inflammation can alter Aβ deposition (38,39). To assess astrogliosis, we immunostained sections from all four genotypes with an antibody that recognizes glial fibrillary acidic protein (GFAP; Fig. 6A–C). We found that GFAP immunoreactivity was similar among the four groups in the dentate gyrus and CA1 (Fig. 6D and E). In contrast, in the cortex we found a significant effect for genotype [APP transgene effect: P < 0.0001 and F(1, 16) = 472.9. Nrf2 effect: P = 0.0044 and F(1, 16) = 10.94. Interaction: P = 0.0005 and F(1, 16) = 18.61; Fig. 6F]. Post hoc analysis with Bonferroni’s correction indicated an effect for Nrf2 only in wildtype mice, probably due to a ceiling effect (P < 0.001). To assess microglial activation, we immunostained sections from all four genotypes with an antibody raised against ionized calcium binding adapter molecule 1 (Iba1; Fig 6A–C), and that recognizes activated microglia. We found that in dentate gyrus and CA1, Iba1 immunoreactivity was higher in transgenic mice compared with wildtype mice, independent of the Nrf2 genotype [P = 0.0001 and F(1, 16) = 26.01 for dentate gyrus; P = 0.0008 and F(1, 16) = 17.06 for CA1, Fig. 6G and H]. In contrast, in the cortex, we found that there was a difference in Iba1 immunoreactivity driven by the genotype [APP transgene effect: P < 0.0001 and F(1, 16) = 52.39. Nrf2 effect: P = 0.0119 and F(1, 16) = 8.04. Interaction: P = 0.0057 and F(1, 16) = 10.18; Fig. 6I]. Post hoc analysis with Bonferroni’s correction indicated that Iba1 immunoreactivity was significantly higher in APP/PS1;Nrf2^−/− mice compared with APP/PS1 mice (P < 0.001, Fig. 6I). To further dissect the relationship between Nrf2 and neuroinflammation, we measured the levels of 25 cytokines by multiplex ELISA (Table 1). We analysed the data by two-way ANOVA and found that the levels of 21 cytokines were significantly different between transgenic and wildtype mice, independently of the Nrf2 genotype (Table 1). For two cytokines, IFNγ and MIP1β, there was also a Nrf2 genotype effect. Specifically, we found that the levels of IFNγ were significantly higher in the APP/PS1;Nrf2^−/− mice compared with the other three groups (Table 1). Also, the MIP1β levels were significantly higher in the two transgenic groups compared with the wild type, and they were higher in APP/PS1;Nrf2^−/− mice compared with APP/PS1 mice (Table 1). Together, these data show that reducing Nrf2 levels in APP/PS1 increases the levels of specific cytokines and microglia activation.

Table 1.

Brain homogenates (n = 8/genotype) were analysed with Bioplex. Data were normalized for protein concentration (pg of analyte/mg of protein), are presented as mean ± SD, and were analysed by two-way ANOVA. *P < 0.05, **P <0.01, and ***P < 0.001

	WT	Nrf2−/−	APP/PS1	APP/PS1; Nrf2−/−	APP transgene effect	Nrf2 effect
	Average ± SD	Average ± SD	Average ± SD	Average ± SD
G-CSF	0.621 ± 1.097	1.310 ± 2.349	4.651 ± 3.885	2.418 ± 2.577	*	ns
GM-CSF	0.285 ± 0.033	0.345 ± 0.068	0.529 ± 0.123	0.423 ± 0.110	***	ns
IFNγ	0.218 ± 0.040	0.203 ± 0.031	0.224 ± 0.110	0.394 ± 0.254	*	**
IL-1α	2.625 ± 1.878	2.189 ± 1.126	3.399 ± 0.838	3.791 ± 0.808	*	ns
IL-1β	0.283 ± 0.040	0.282 ± 0.042	0.326 ± 0.055	0.333 ± 0.142	ns	ns
IL-2	0.243 ± 0.033	0.220 ± 0.020	0.212 ± 0.033	0.225 ± 0.044	ns	ns
IL-4	0.151 ± 0.022	0.151 ± 0.023	0.136 ± 0.009	0.136 ± 0.011	*	ns
IL-5	0.143 ± 0.046	0.194 ± 0.061	0.472 ± 0.303	0.332 ± 0.121	***	ns
IL-6	0.063 ± 0.008	0.104 ± 0.047	0.238 ± 0.066	0.784 ± 1.694	***	ns
IL-7	0.079 ± 0.017	0.094 ± 0.016	0.132 ± 0.014	0.121 ± 0.015	***	ns
IL-9	21.598 ± 15.269	19.223 ± 11.358	30.415 ± 7.136	34.283 ± 6.497	**	ns
IL-10	0.229 ± 0.266	0.184 ± 0.201	0.659 ± 0.488	0.478 ± 0.376	**	ns
IL-12 (p40)	0.297 ± 0.042	0.296 ± 0.045	0.266 ± 0.017	0.267 ± 0.021	*	ns
IL-12 (p70)	0.288 ± 0.041	0.287 ± 0.043	0.258 ± 0.016	0.259 ± 0.020	*	ns
IL-13	1.120 ± 0.159	1.115 ± 0.168	1.004 ± 0.064	1.008 ± 0.079	*	ns
IL-15	0.939 ± 0.692	1.812 ± 0.694	3.451 ± 0.590	3.207 ± 0.609	***	ns
IL-17	0.069 ± 0.010	0.068 ± 0.010	0.081 ± 0.014	0.073 ± 0.026	ns	ns
IP-10	4.851 ± 1.672	4.162 ± 1.009	9.969 ± 1.675	11.598 ± 1.404	***	ns
KC	1.491 ± 0.222	1.499 ± 0.331	2.565 ± 0.657	2.110 ± 0.626	***	ns
MCP-1	0.251 ± 0.034	0.286 ± 0.076	0.440 ± 0.104	0.425 ± 0.125	***	ns
MIP1α	1.835 ± 0.926	2.215 ± 0.722	3.025 ± 0.454	2.994 ± 0.447	***	ns
MIP1β	0.322 ± 0.090	0.715 ± 0.441	4.230 ± 0.642	5.078 ± 0.579	***	**
MIP2	0.542 ± 0.279	0.660 ± 0.302	1.352 ± 0.249	1.471 ± 0.314	***	ns
RANTES	0.144 ± 0.022	0.160 ± 0.026	0.185 ± 0.022	0.202 ± 0.036	***	ns
TNFα	0.364 ± 0.052	0.362 ± 0.055	0.326 ± 0.021	0.327 ± 0.026	*	ns

Figure 6.

Increased microglia immunoreactivity in the cortex of APP/PS1 mice lacking the Nrf2 gene. (A–C) Representative microphotographs of different brain regions (DG, dentate gyrus; CA1, Cornu Ammonis 1; CX, cortex) from sections immunostained with GFAP (red) and Iba1 (green). i, WT; ii, Nrf2^−/−; iii, APP/PS1; iv, APP/PS1;Nrf2^−/−. (D–F) Semiquantitative analysis of GFAP immunoreactivity showed no differences among groups in DG and CA1. In contrast, there was a significant difference in the CX [APP transgene effect: P < 0.0001 and F(1, 16) = 472.9; Nrf2 effect: P = 0.0044 and F(1, 16) = 10.94; interaction: P = 0.0005 and F(1, 16) = 18.61]. (G–I) Semiquantitative analysis of Iba1 immunoreactivity showed that the presence of the APP transgene changed Iba1 immunoreactivity in DG and CA1 [P = 0.0001 and F(1, 16) = 26.01 for DG; P = 0.0008 and F(1, 16) = 17.06 for CA1]. Moreover, in the CX Iba1 immunoreactivity was different among the four groups [APP transgene effect: P < 0.0001 and F(1, 16) = 52.39; Nrf2 effect: P = 0.0119 and F(1, 16) = 8.04; interaction: P = 0.0057 and F(1, 16) = 10.18]. Post hoc analysis with Bonferroni’s correction indicated that Iba1 immunoreactivity was significantly higher in APP/PS1;Nrf2^−/− mice compared with APP/PS1 mice (P < 0.001). Data are presented as box plots and were analysed by two-way ANOVA.

Discussion

Nrf2 plays a definite role in brain aging and several neurodegenerative disorders (1,4,20,27). Here, we report the novel finding that decreasing Nrf2 levels exacerbates cognitive deficits in a mouse model of AD. While aging is the major risk factor for AD, even in cases in which AD is caused by mutations in APP, PS1, or PS2, the penetrance of the mutation changes as a function of age (40). Regardless of this indisputable evidence, the mechanisms by which aging contributes to the disease pathogenesis are unknown. Although further studies are needed to fully dissect the relationship between the age-dependent decrease in Nrf2 and AD pathogenesis, given the data presented here, it is tempting to speculate that Nrf2 may act as a molecular link between brain aging and AD. This hypothesis is supported by the findings from several laboratories reporting that Nrf2 activity decreases as a function of age (32,35). Also, Nrf2 activity is directly linked to tau pathology further strengthened the connection between Nrf2 and AD (41,42).

We report that lack of Nrf2 exacerbates spatial learning and memory, as well as working and associative memory deficits in APP/PS1 mice. The results of the contextual fear conditioning show that under the conditions used here, the APP/PS1 mice perform as well as WT mice. These results suggest that associated memory function related to the amygdala-hippocampus axis is not altered in APP/PS1 mice. It is important to note that the outcomes of cognitive tests are highly dependent on the genetic background of mice. Along these lines, even the neuropathological phenotype of many AD mouse models is linked to the genetic background of the mice (43–47). Thus, given that there are several colonies of APP/PS1 mice on different genetic backgrounds, it is not surprising that contradicting reports have been published. For example, many have shown that APP/PS1 mice show impairment in contextual fear conditioning (48,49), while others have found no differences (50). To minimize the effects of genetic background on the cognitive outcome (and all the other experiments reported here), all the four groups of mice used here were littermates.

We investigated changes in Aβ pathology and found that the lack of Nrf2 gene does not modify plaque number but does increase soluble Aβ40 and Aβ42, therefore highlighting a dissociation between Aβ plaques and cognitive deficits. Consistent with this observation, we and others have repeatedly reported on such dissociation. For example, several studies in animal models have reported a lack of association between cognitive deficits and Aβ load (51,52). Moreover, clinical evidence showed that the amyloid burden does not correlate with the severity of cognitive deficits in AD patients (36,37,53). The effects on plaque number reported here are not consistent with a previous report showing that the lack of Nrf2 exacerbates Aβ deposition in a mouse model of AD (21). While it is always hard to understand this type of discrepancies, Joshi and colleagues used 7–8 months of age mice (21), whereas the mice used here were 12 months of age. It is possible that as the mice aged, the number of Aβ plaques in APP/PS1;Nrf2^−/− mice increased faster than APP/PS1 mice. In addition, the difference in genetic background between the mice used by Joshi and colleagues and those used here could account for some discrepancies in the results.

We found an increase in microgliosis, IFNγ, and MIP1β in APP/PS1;Nrf2^−/− mice compared with APP/PS1 mice. This finding is consistent with previous reports showing that Nrf2^−/− mice are more prone to neuroinflammation (29,54). For example, increasing α-synuclein expression in Nrf2^−/− mice alters microglial morphology and changes cytokine production (54). Along these lines, lack of Nrf2 exacerbates the microglia activation, including the release of toxic cytokines, in mice injected with 1-methyl-4-phenyl-1, 2, 3, 6-tetrahydropyridine (55). Consistent with these results, others have shown that activation of Nrf2 attenuates microglia activation (56,57). Brain inflammation and alterations in cytokine levels, including IFNγ, have been linked to aging and neurodegenerative diseases (58,59). IFNγ is expressed in neurons, glia, and T cells; it is related to activation of microglia and is considered a critical cytokine in the regulation of brain inflammation (60). The connection between IFNγ and AD is strengthened by data showing that IFNγ levels are higher in AD brains compared with control cases and in mouse models of AD (61,62). Notably, IFNγ contributes to the generation of Aβ in neurons and astrocytes (63,64), while chronic hippocampal gene delivery of IFNγ exacerbates AD-like pathology in mice (65). These data strongly suggest that higher IFNγ can increase Aβ levels and could explain the Nrf2-mediated increase in Aβ shown here. MIP1β is a chemoattractant for natural killer cells, monocytes, and other immune cells. Others have reported that MIP1β levels are also higher in the brains of APP/PS1 mice (66), thus supporting our results.

We found that APP/PS1;Nrf2^−/− mice have higher soluble Aβ levels compared with APP/PS1 mice. These Aβ changes most likely contribute to the increase in inflammation seen in APP/PS1;Nrf2^−/−. To this end, Selkoe and colleagues have reported that intracerebral injections of Aβ oligomers into WT mice increase microgliosis and levels of several cytokines (67). Given the complex interaction between inflammation and Aβ (68) and Nrf2 and inflammation (28), the data presented here are consistent with a crosstalk among Aβ, microgliosis, and Nrf2. It is plausible that the presence of Nrf2 in APP/PS1 mice mitigates the inflammatory response (to a certain extent) due to high soluble and insoluble Aβ levels. In contrast, removing Nrf2 from APP/PS1 mice, lowers such mitigation leading to an increase in microgliosis and the associated cytokine levels. In turn, the higher microglia activation and IFN1γ levels can further contribute to increasing Aβ levels.

In conclusion, we report the novel finding that reducing Nrf2 levels exacerbates behavioral deficits in APP/PS1 mice. Specifically, the cognitive changes were generalized across different domains as we found an exacerbation of deficits in spatial learning and memory, as well as in working and associative memory. These behavioral tests are controlled by different brain regions (31), indicating that the lack of Nrf2 has a global effect on brain function.

Materials and Methods

Mice

APP/PS1 and Nrf2^−/− mice were purchased from The Jackson Laboratory (stock number 017009 and 034832, respectively). Using the crossbreeding strategy reported in Figure 1, we generated the four groups utilized for this study. Specifically: WT (APP/PS1^0/0;Nrf2^+/+); Nrf2^−/− (APP/PS1^0/0;Nrf2^−/−); APP/PS1 (APP/PS1^+/0;Nrf2^+/+); and APP/PS1;Nrf2^−/− (APP/PS1^+/0;Nrf2^−/−). Notably, all mice used in this study were littermates, and we used both male and female mice. Mice were housed 4–5 per cage, were kept on a 12-h light/dark cycle with ad libitum access to food and water. Animal care and treatments were in accordance with the applicable regulations in the vivarium (The Institutional Animal Care and Use Committee of the Arizona State University).

Behavioral analyses

Open field. We performed the open field as previously described (69). It is routinely used to investigate the spontaneous explorative activity of rodents, as well as the general locomotor activity.

Morris water maze (MWM). The MWM is commonly used to evaluate spatial learning and memory, and it was performed as previously described (70). Briefly, to find a hidden platform, mice received four training trials per day for five consecutive days. Escape latency and distance traveled were scored as an indication of learning. On day 6, we removed the platform and allowed mice to swim for 60 s freely. During this time, several dependent variables were scored: number of platform location crosses (frequency), and time spent in the target and opposite quadrants. Both open field and MWM were analysed using EthoVision XT tracking system (Noldus Information Technology, VA, USA).

Radial arm water maze (RAWM). RAWM is used to evaluate spatial reference and working memory. We performed the task as previously described (71,72), and measured two different kinds of errors: incorrect arm entries (reference memory errors) and re-entries (working memory errors) for day one and day two. The scoring was done manually by an experimenter blind to the genotype.

Contextual fear conditioning (CFC). The contextual fear conditioning (CFC) test was performed using an apparatus from San Diego Instruments (CA, USA) as previously described (73). Briefly, on day 1, mice were placed in the conditioning chamber and allowed to freely explore for 148 s at the end of which they received a mild foot shock (0.3 mA, 2 s). After the shock, mice remained in the chamber for 30 additional seconds. On all the other days (2, 3, 4, and 6), mice were placed back in the chamber for 3 min without any shock administration, to evaluate formation and extinction of the adverse memory. Freezing, defined as the complete lack of motion except breathing for a minimum of 2 s, was recorded and analysed by Freeze Monitor software (San Diego Instruments). We reported the percent of freezing during each day.

Protein extraction

Mice were anesthetized with isoflurane and perfused with cold phosphate buffer saline buffer (PBS) for 5 min with a perfusion pump (flux rate 25 ml/min). After perfusion, their brains were removed and bisected sagittally. We processed the tissue as previously described (74). Briefly, we fixed half of the brain in 4% paraformaldehyde, and flash frozen the other half in dry ice and stored at −80 °C until use. We used the fixed tissue for the immunohistochemical experiments. The frozen brains were homogenized in a solution of tissue protein extraction reagent (T-PER, ThermoFisher Scientific) supplemented with protease (Roche Applied Science, IN, USA) and phosphatase inhibitors (Millipore, MA, USA). The homogenized tissues were centrifuged at 4 °C for 1 h at 100 000 g. We stored the supernatant (soluble fraction) at −80 °C until use. We then homogenized the pellet in 70% formic acid and centrifuged it (as mentioned previously). We stored the supernatant of this centrifugation as the insoluble fraction.

Western blot, ELISA, and Bioplex

Proteins from soluble fractions were resolved by 10% BisTris SDS-polyacrylamide gel electrophoresis (ThermoFisher Scientific) under reducing conditions and transferred to a nitrocellulose membrane. We developed the membranes as described previously (75). We measured Aβ40 and Aβ42 levels in the soluble and the insoluble fractions by ELISA (ThermoFisher Scientific) using the manufacturer’s instructions. We measured protein carbonyl levels in the soluble fraction using a commercially available ELISA kit (CellBiolabs Inc) and following the manufacturer’s instructions. The cytokine profile was obtained with the mouse cytokine panel (MCYTOMAG) 25-plex (Millipore). The analysis was conducted in the soluble fraction following the manufacturer’s instructions.

Immunohistochemistry and immunofluorescence

Brain free-floating slices were obtained using a vibratome from the hemibrains fixed in 4% paraformaldehyde. We performed the immunostaining as previously described (76). Briefly, for immunohistochemistry, 50 µm-thick sections were washed twice with TBS (100 mM Tris pH 7.4, 150 mM NaCl), and the endogenous peroxidase activity was quenched in 3% H₂O₂ in methanol. Sections were then transferred for 15 min into TBS-A (100 mM Tris pH 7.4, 150 mM NaCl, 0.1% Triton X-100) and for 30 min in TBS-B (100 mM Tris pH 7.4, 150 mM NaCl, 0.1% Triton X-100, 2% bovine serum albumin). Finally, sections were incubated overnight at 4 °C with a primary antibody, which was diluted in TBS-B. To remove the excess of primary antibody, sections were washed and then incubated with the proper secondary antibody for 1 h at room temperature. Signal was enhanced by incubating sections first in the avidin-biotin complex (Vector Labs) for 1 h, and then with the avidin-biotin horseradish peroxidase system (Vector Labs, CA, USA). Images were obtained with a digital Zeiss camera and analysed using ImageJ. The same protocol was followed for immunofluorescence staining, but the quenching and the avidin-biotin steps were skipped. For the quantification of GFAP and Iba1, we used one section per mouse (6 mice/genotype). Within each section, we took five different images per brain region (i.e. CA1, dentate gyrus, and cortex) with a Leica Confocal microscope. The experimenter was blind to genotype and the selection of the images and data acquisition were made randomly. Images were analysed using ImageJ.

Proteasome activity assay

To conduct this test, we used the proteasomal substrates, Suc-LLVY-AMC, Bz-VGR-AMC, and Z-LLE-AMC (Enzo Life Sciences, PA, USA), as previously described (77). Briefly, we incubated ten μl of brain homogenate with the fluorogenic substrate of interest. The reactions were conducted in black 96-well plates with assay buffer (25 mM HEPES, pH 7.5, 0.5 mM EDTA, 0.05% NP-40). Kinetic readings were taken at 37 °C every 1.5 min for 60 min (excitation 360 nm, emission 460 nm) using the Synergy HT multimode microplate reader using the Gen5 software (BioTek). Readings were normalized to protein concentration.

Antibodies

From Abcam (MA, USA): 4HNE (Cat. # ab46545, 1: 2000); from Cell Signaling Technology (MA, USA): β-actin (Cat. #3700, 1: 10 000); Atg3 (Cat. #3415, 1: 1000); Atg5 (Cat. #12994, 1: 1000); Atg7 (Cat. #8558, 1: 1000); GFAP (Cat. #3670, 1: 300); p-mTOR(S2448) (Cat. #5536, 1: 1000); p-p70S6K(T389) (Cat. #9234, 1: 1, 000). From Covance (NJ, USA), APP (Aβ amino acids 1–16) monoclonal antibody, 6E10 (Cat. #MAB1560, 1: 3, 000). From Millipore, Aβ42 (Cat. #AB5078P, 1: 200). From Santa Cruz Biotechnology (TX, USA): Heme Oxygenase 1 (Cat. #sc-136960, 1: 500). From Sigma-Aldrich (MO, USA), amyloid precursor protein (APP) C-terminal (Cat. #A8717, 1: 1, 000). From Wako (VA, USA), Iba1 (Cat. #019–19741, 1: 300).

Transcript level determinations

Total RNA was extracted from the cerebellum (n = 4 mice/genotype) using the RNeasy Mini Kit (Qiagen). RNA (1 µg/sample) was reverse-transcribed with the Quantitect® Reverse Transcription Kit (Qiagen), accordingly to the manufacturer’s instructions. After cDNA synthesis, quantitative real time PCR was performed using the following primers (0.3 pmol/μL): Nrf2, Fw 5’- AGGACATGGAGCAAGTTTGG -3’ and Rev 5’- TCTGTCAGTGTGGCTTCTGG-3’; β-actin, Fw 5’-GGCTCTTTTCCAGCCTTCCT-3’ and Rev 5’- ATGCCTGGGTACATGGTGGT-3’ primers. ΔCt values were obtained by subtracting the Ct (cycle threshold) of Nrf2 to the Ct value of β-actin for every sample. ΔΔCt values were obtained by subtracting the ΔCt values in Nrf2^−/−, APP/PS1, and APP/PS1;Nrf2^−/− to the ΔCt values of WT mice. Fold change represents 2 power (−ΔΔCt).

Statistical analyses

All data were analysed using GraphPad Prism (GraphPad Software, CA, USA, www.graphpad.com). Data were analysed by one- or two-way ANOVA followed by Bonferroni’s post hoc analysis, when applicable, as detailed in the figure legend.

Acknowledgements

The authors thank Mr. Nikhil Dave for his technical contribution and Dr. John P Konhilas for his help with the Bioplex data.

Conflict of Interest statement. None declared.

Funding

This work was supported by grants from the Arizona Alzheimer’s Consortium and the National Institutes of Health (R01 AG037637) to S.O.