All animal work was performed with the approval of the Garvan Institute and St. Vincent’s Hospital Animal Ethics Committee, in accordance with the National Health and Medical Research Council animal experimentation guidelines and the Australian Code of Practice for the Care and Use of Animals for Scientific Purposes (2013). Hemizygous transgenic (J20) and non-transgenic littermates (wildtype; WT) were used from the B6.Cg-Tg (PDGFB-APPSwInd)20Lms/2J (J20) line. These J20 mice (sometimes referred to as hAPP-J20) (Wright et al., 2013) overexpress mutant APP (Amyloid precursor protein) containing both the Swedish and Indiana mutations, under the control of the PDGF-β chain promoter. Mice were maintained on a C57BL/6J background. Mice were group housed on a 12 h light/dark cycle with ad libitum access to food and water. For all experiments, mice were acclimatised for at least 1 week in holding rooms, before being used in experimental procedures. A total of 40 male mice were used throughout the study.

The apparatus comprised a circular, white platform measuring 920 mm in diameter, positioned at a height of 1 m above the ground. The perimeter of the platform contained 20 identical holes that were equally spaced. A black escape box, measuring 175 mm (D) × 75 mm (W) × 80 mm (H), was hidden beneath one of the holes, while the other holes were blocked. Mice were subjected to negative reinforcements (bright lights and loud noises) and were expected to use the visual cues located in the room to find the hidden escape box. The activity of the animals was recorded using the ANYmaze Video Tracking System 6.33 (Stoelting Co.) with a DMK 22AUC03 camera placed directly above the maze.

44-week-old mice underwent three phases during the Barnes Maze protocol: habituation, acquisition, and the probe (test) trial. During the 2 min habituation phase, mice were placed in the centre of the platform and covered with a glass beaker and moved slowly toward the escape box to learn its location. The acquisition phase consisted of 3 training trials per day with a 35–45 min interval between each trial, for 5 days. During these trials, the mice were placed in a cylindrical chamber measuring 80 mm in diameter and 12.5 mm in height, positioned at the centre of the maze and facing a random direction. The chamber was then lifted, and the mice were given 2 min to locate the hidden escape box. If the animals were unable to locate the escape box within the given time, they were manually guided to it. The location of the escape box was randomly assigned to each animal before the experiment, but it remained constant for each individual mouse throughout the trial. The probe trial was conducted 24 h after the last acquisition trial, during which the hidden escape box was removed, and its hole was undifferentiated from the other holes on the platform. Mice were given 60 s on the platform, during which they were expected to move to the location where the escape box was previously located. The results measured for each mouse included primary latency, primary distance, and primary errors.

A separate cohort of 44-week-old mice were anesthetised with isoflurane and cervically dislocated. Whole brains were extracted and placed on a sterile plate cooled with ice. The hippocampus was rapidly dissected and removed from both the left and right cerebral hemispheres. The extracted hippocampal tissue was placed within sterile Eppendorf tubes and snap-frozen in dry ice. Samples were then stored at −80°C before being processed for protein extraction. For all experiments performed within this study, only the right hippocampus was used.

Hippocampal tissue was suspended in 200 μl of RIPA buffer (Sigma-Aldrich, Cat# MFCD02100484) containing Protease Inhibitor Cocktail (Sigma-Aldrich, Cat#S8830) and Phos-stop dephosphorylation inhibitors (Sigma-Aldrich, Cat#4906845001). Tissue was homogenised by sonication and the resulting cell suspension was centrifuged at 14,000 rpm for 15 min at 4°C and supernatant was collected. If being used immediately, protein extracts were placed on ice. Otherwise, extracts were aliquoted and stored at −80°C until required. Each aliquot was freeze-thawed no more than 3 times to ensure their integrity.

The concentrations of all protein samples were quantified in technical triplicates utilising a Bradford assay. A standard curve was generated with Pre-Diluted Bovine Serum Albumin (ThermoFisher, Cat# 23208) by performing standards in triplicate across a dilution range of 0–2000 μg/ml. The concentrations of unknown samples were interpolated from this curve according to the manufacturer’s instructions. Absorbance was measured using a FLUOstar Omega Microplate reader (BMG-Labtech). Concentrations were determined by interpolating the standard curve.

Optimal antibody concentrations and linear dynamic ranges for each target protein were determined prior to conducting expression analyses. To quantify target proteins in J20 mice compared with WT littermates, samples were run on a standard 24-well WES operating plate, as per the manufacturer’s instructions. Reagents were obtained from 12 to 230 kDa separation modules (ProteinSimple Cat#SM-W004) and total protein detection modules (ProteinSimple Cat#DM-TP01). For experiments investigating the expression of target proteins, samples were pipetted in two separate wells within the plate. An internal control sample (derived from a WT littermate) was run in technical duplicates for both the primary antibody targets and total protein concentration so that data could be standardised to this internal calibrator across multiple plates. Several wells were used to run controls on each plate including controls for background biotinylation in sample diluent in the absence of sample, background biotinylation in the sample in the absence of the biotinylation label and background antibody signal within the sample diluent in the absence of sample.

Data was analysed using the WES instrument software (ProteinSimple, Compass for SW 4.1 Windows 7/8/10 64 bit). Peak analysis settings were performed as follows: Range: (1–250); Baseline: Threshold (0.1), Window (400), Stiffness (0.1); Peak Find: Threshold (10), Width (9), Area Calculation (Dropped lines). Baseline adjustments were made to fit relative background chemiluminescence signals with all samples measured at identical conditions. Dropped line analysis was preferred over a Gaussian fit model to adjust for interfering additional peaks and for better control of the relative peak signal. Supplementary Table 1 describes antibodies measured in WES analysis and peaks recorded for antibodies used in the procedure.

Statistical analyses were performed using GraphPad Prism 8.2.1 (441). Outliers were identified and excluded using Grubbs’ test for outliers (α = 0.05). The D’Agostino-Pearson test was employed to assess normality and lognormality, in order to determine whether the data were parametric or non-parametric. For parametric groups, an F-test of variance was conducted to ascertain whether standard deviations were equal between individual data sets. In cases where standard deviations were equal, a student’s t-test was utilised to determine significance between the two groups. Conversely, for groups with unequal standard deviations, Welch’s t-test was applied. For non-parametric groups, the Mann-Whitney test was used to determine the significance between the two groups. Statistical significance was defined as p < 0.05 and denoted as *p < 0.05, **p < 0.01, ***p < 0.001, ****p < 0.0001. Data are presented as mean ± SD unless otherwise specified. For the quantification of whole hippocampal GluA1-4 protein expression, as well as ADAR1-3, FXR1P, WWP2, PIN1, and CREB1 protein expression, a sample size of n = 12 per group was utilised. Due to the novelty of these experiments, the expected means, standard deviations, and effect sizes were unknown. The maximum sample size available for these initial investigations was n = 12. The data supporting the findings of this study are provided in a Supplementary document.

Prior to analysing protein expression in the hippocampus of J20 mice, we first conducted a series of experiments to optimise our automated capillary western protocols. Identical to traditional western blotting, protein quantification can be determined by normalising target protein expression levels to loading controls. In traditional western blotting, these loading controls have predominantly been “housekeeping” proteins (e.g., β-Tubulin, GAPDH, and β-Actin) (Li and Shen, 2013). However, although housekeeping proteins are thought to be expressed consistently within tissue under disease conditions, research now suggests this may not always be the case (Ferguson et al., 2005; Lin and Redies, 2012). In the context of neurological diseases, several studies have demonstrated that commonly used housekeeping proteins do in fact fluctuate in disease conditions (Li and Shen, 2013; Vigelsø et al., 2015). To overcome these issues, it has become common to normalise target protein expression to the total protein expression of a given sample (Aldridge et al., 2008; Kirshner and Gibbs, 2018).

To determine the appropriate loading controls for our study, we used linear regression to assess the relationship between signal intensity and concentration for a range of traditional housekeeping proteins and compared these to the total protein expression of each sample. For all housekeeping proteins, we used saturating concentrations of antibodies for detection (0.5 μg/ml). We observed a strong linear relationship between signal intensity and protein concentration for all housekeeping proteins and the total protein assay (Figures 1A–D: β-Tubulin r² = 0.9882, GAPDH r² = 0.9724, β-Actin r² = 0.9983, total protein r² = 0.9960). Another way to illustrate this is to quantify the fold change in signal after each twofold dilution, which should theoretically halve each time. This approach allowed us to visually compare traditional housekeeping proteins to the total protein assay (Figure 1E). Using this approach, it was noticeable that the signals from the housekeeping proteins were not following a twofold decreasing trend in signal intensity at each twofold dilution. In fact, we observed that both GAPDH and β-Actin signals did not fall within the twofold range until the sample protein concentration decreased to 0.032 μg/ml. Furthermore, β-Tubulin exhibited signal values that were outside the twofold range at all dilutions, suggesting a non-linear relationship within these concentration ranges. This poses a challenge when comparing the signal of these proteins to weakly expressed proteins that may exhibit a linear decrease in signal at higher protein concentrations. In contrast, our findings revealed that the total protein assay demonstrated a linear decrease of twofold across a wider range of concentrations (0.032–0.250 μg/ml), suggesting the possibility of normalising data to total protein even for weakly expressed proteins.

To confirm that the total protein assay provided a strong linear relationship between signal and protein concentration, we repeated the total protein assay over a larger number of replicates (n = 6), finding that the assay was consistently linear within the ranges 0.032–0.250 μg/ml (Figures 1F, G). These findings illustrate that the total protein assay is suitable for normalisation over a wide range of protein concentrations. These concentrations were subsequently used as the starting point when designing assays to test the linear range of target proteins so that both our target proteins and the total protein assay were assessed within their linear range.

Finally, we determined the reproducibility of the total protein assay across multiple technical and biological replicates. One downside of non-fluorescent capillary western blotting is that the total protein assay cannot be multiplexed with other antibodies, as the chemiluminescence signal from the total protein assay would obscure the signal from other antibodies. Therefore, to compare the expression levels of our proteins of interest to the total protein expression within a sample, it was necessary to determine the signal variance when a single sample is pipetted multiple times.

We conducted a total protein assay on hippocampal tissue from 3 WT mice, each of which was analysed across 8 technical replicates at 0.064 μg/ml. There was no significant difference found in the signal between the 3 WT samples (Figure 1H). Furthermore, the coefficient of variance was <10% for each biological replicate (4.50, 4.98, and 6.76%), a result similar to those found in another study investigating the efficiency of capillary western blotting (Rustandi et al., 2012) and less than the reported variability found in traditional western blotting of >35% (Koller and Wätzig, 2005). This provided confidence we could accurately normalise our target protein signals to the total protein signal.

We next assessed the linear range of our target proteins to determine the optimal protein concentration for each target. To do this, we conducted a six-point, twofold dilution of protein derived from a calibrator sample. For all target proteins, the antibody concentration was run at a saturating dilution of 0.5 μg/ml. Similar to the housekeeping proteins, a strong linear relationship was displayed for the ADAR proteins (Figure 2A: ADAR1 r² = 0.9994, ADAR2 r² = 0.9994, ADAR3 r² = 0.9991) and GluA subunits (Figure 2B: GluA1 r² = 0.9990, GluA2 r² = 0.9996, GluA3 r² = 0.9917, GluA4 r² = 0.9841) between signal intensity and protein concentration. Specifically, ADAR1-3 were linear between 0.125 and 0.250 μg/ml (Figure 2C). This was also the case for the AMPA subunits GluA3 and GluA4 (Figure 2D). GluA3 and GluA4 are relatively lowly expressed in comparison to GluA1 and GluA2, likely explaining why GluA1 and GluA2 were still linear at the lowest tested concentrations (between 0.016 and 0.063 μg/ml; Figure 2D). In subsequent quantification experiments, we adopted these ranges when assessing each protein to ensure we were working within the linear range of each protein.

Following this, we assessed the saturation point for each antibody to ensure antibody binding was not a rate-limiting step in the assay. For each target protein, the protein concentration was within the linear range of the target, based on our previous experiments (i.e., ADAR1-3 and GluA3-4: 0.250 μg/ml; GluA1-2: 0.063 μg/ml). ADAR1-3 antibodies all reached saturation at >0.1 μg/ml (Figure 2E), whereas the GluA1-4 antibodies reached a plateau at >0.05 μg/ml (Figure 2F). Antibody concentrations above these saturation points were subsequently used for each assay. Optimisation of protein concentrations and antibody saturation points were similarly conducted for all other target proteins PIN1, WWP2, FXR1P, and CREB (Supplementary Figures 1, 2).

Notably, the ADAR1 antibody used in this study binds both p110 and p150 isoforms. Due to the significantly higher expression of p110 in the brain and the larger peak observed in the electropherogram traces obscuring p150 at lower concentrations, optimisations were performed solely on the p110 peak. This may impact the accuracy of p150 quantification and should be noted in regard to the conclusion obtained from its analysis. Representative examples of original, uncropped capillary western blot traces for all proteins analysed in this paper in WT and J20 tissue are included in Supplementary Figures 3–8.

The J20 mouse is a widespread model of AD with well-characterised behavioural and histopathological alterations (Palop et al., 2003; Esposito et al., 2006; Galvan et al., 2006; Cheng et al., 2007; Harris et al., 2010; Cissé et al., 2011; Verret et al., 2012). In our own previous studies using J20 mice at similar, or younger ages than those used in this study, we have identified impaired devaluation performance in goal-directed action tests, spatial reference/working memory deficits in the radial arm maze, short-term memory deficits in a Y Maze and hyperactivity in an open field test. Despite this, no changes in anxiety on an elevated plus test, fear memory in a context fear conditioning paradigm, or novel object recognition in an object recognition task have been observed (Wright et al., 2013, 2023; Dhungana et al., 2023). To both expand upon the characterisation of the J20 mouse model and confirm our cohort had not experienced genetic drift, which has been shown to occur in AD mouse models (Cho et al., 2022), a separate cohort of J20 and WT mice underwent behavioural testing in a Barnes maze task, to identify if memory deficits were still present in our colony of J20 mice. As expected, both WT and J20 mice were able to learn the location of the escape box within the 5-day acquisition phase (Figures 3A, C, E). During the probe trial, J20 mice took significantly more time (Figure 3B: 7.800 ± 9.367 vs. 20.00 ± 10.03, p = 0.0276), travelled a significantly longer distance (Figure 3D: 0.7872 ± 0.7974 vs. 2.087 ± 0.8970, p = 0.0092), and made significantly more errors (Figure 3F: 9.100 ± 8.504 vs. 23.17 ± 6.735, p = 0.0040) to find the escape box location. These results confirmed that spatial memory deficits were still present in our J20 mice indicating a preserved AD phenotype.

To measure the expression of both AMPAR subunits and RNA-editing-related proteins in J20 mice, we extracted protein from the hippocampus of twelve 44-week-old J20 and 12 age-matched WT littermates. We observed no significant difference in the expression of GluA1 (Figure 4A: 100.0 ± 6.556% vs. 101.7 ± 5.827%, p = 0.8154), GluA2 (Figure 4B: 100.0 ± 3.773% vs. 95.36 ± 1.597%, p = 0.0899), or GluA4 (Figure 4D: 100.0 ± 4.806% vs. 95.36 ± 4.373%, p = 0.6776) in WT vs. J20 mice. However, there was a significant increase in the expression of the GluA3 subunit (Figure 4C: 100.0 ± 4.245% vs. 95.36 ± 3.061%, p = 0.0150) within the J20 cohort when compared to the WTs.

ADAR1 has two isoforms, ADAR1 p110 and ADAR1 p150, which have predicted molecular weights of ∼110 and ∼150 kDa, respectively. Two peaks corresponding to these molecular weights were observed following capillary western blotting with the ADAR1 antibody, allowing us to quantify the isoforms separately. Due to the relatively low expression of p150 compared to p110, it was necessary to verify the antibody was accurately detecting p150. Treatment of SH-SY5Y cells with type I and II interferons led to a moderate increase of p110 (≈20–50%) and a significant upregulation of p150 (≈300–500%) which was not mirrored following treatment with IL1-B, a cytokine not significantly linked to ADAR1 p150 (Supplementary Figure 9). Considering the p150 isoform is highly interferon-inducible (George et al., 2016; Li et al., 2021) this provided confidence that the p150 peak observed was accurate and could be used to quantify changes in mice.

When performing our mouse experiments we found the expression of ADAR1 p110 was significantly increased by 21.8% in J20 mice compared to WTs (Figure 5A: 100.0 ± 4.9% vs. 121.8 ± 6.2%, p = 0.0121), but the expression of ADAR1 p150 was unchanged (Figure 5B: 100.0 ± 11.4% vs. 102.6 ± 8.3%, p = 0.8541). Notably, the expression of ADAR2 was found to be significantly decreased in J20 mice compared to WT controls (Figure 5C: 100.0 ± 4.669% vs. 85.08 ± 4.335%, p = 0.0296). Finally, the expression of ADAR3 was not significantly altered in J20 mice compared to WT mice (Figure 5D: means 100.0 ± 4.901% vs. 95.36 ± 4.366%, p = 0.4871). Collectively, this data suggests there are changes in the expression of key enzymes involved in A-to-I RNA-editing and AMPA receptor subunits in the hippocampus of J20 mice.

Despite finding no significant changes in the overall expression of proteins PIN1, WWP2, FXR1, or CREB in the hippocampus of J20 mice (Supplementary Figure 10), a benefit of capillary western blotting is that little sample is used for each assay, allowing protein homogenates to be used for several assays. Considering we analysed the expression of multiple protein targets from the same hippocampal homogenates, it was feasible to conduct a post hoc regression analysis to assess the relationship between expression changes in these proteins across different samples (Supplementary Figure 11). We therefore performed a correlation matrix on the relative expression of the analysed proteins. Pearson correlation coefficients were determined for each protein pairing within both WT and J20 groups, with both genotypes exhibiting a combined total of 11 significantly correlated proteins (Figure 6A). Only four protein-protein correlations were shared between the two groups, indicating specific correlation patterns between WT and J20 mice. Of particular interest were the significant correlations within the J20s that were not observed within the WTs. This was distinctly seen in the AMPA subunits wherein GluA1-GluA3 (r = 0.82, p < 0.001), GluA1-GluA4 (r = 0.6, p = 0.039), GluA1-ADAR3 (r = −0.63, p = 0.025), GluA3-ADAR2 (r = 0.65, p = 0.023), GluA3-ADAR3 (r = −0.59 p = 0.042), GluA4-PIN1 (r = −0.62, p = 0.03), and GluA4-WWP2 (r = 0.70, p = 0.012) were all significantly correlated.

Furthermore, using the Pearson correlation coefficients obtained from the initial heat map a further analysis was performed to examine the protein-protein correlations that exhibited the largest differences between WT and J20 R-values (Figure 6B). Specific focus was placed on results wherein the J20 mice displayed larger variance in absolute correlation coefficient values relative to the WT controls. From this PIN1, CREB1, and the AMPAR subunits (specifically GluA3) showed the most consistent association with other RNA-editing related proteins within the J20s. Interestingly, the most significant variation between genotypes was found in the positive association of GluA3-ADAR2. This was unexpected as these proteins were found to be up and downregulated in the J20 mice, respectively. These observations suggest potential disparities in protein interaction networks between the two genotypes beyond the surface-level expression profile and the differences found in these correlations.

Our findings in the J20 AD mice generally parallel what has previously been seen in human AD tissue. Specifically, our observation of an upregulation of ADAR1 p110 protein in the hippocampus parallels that observed in one study of mRNA expression in the frontal lobe of AD patients. Meanwhile, the aforementioned study reported a downregulation of ADAR1 p110 mRNA in the temporal lobes but a significant increase in the ADAR1 p150 isoform RNA in the hippocampus (Khermesh et al., 2016).

Similarly, our observation of downregulation in ADAR2 protein in the hippocampus is consistent with decreased ADAR2 mRNA expression in the prefrontal cortex (Ma et al., 2021) and caudate nucleus (Gaisler-Salomon et al., 2014) of AD patients. It is also consistent with our recent findings of ADAR2 protein downregulation in both the J20 and 5xFAD models of AD using traditional and capillary western blotting, respectively (Wright et al., 2023). However, there are differences between what we observed in J20 mice and some human studies. Our observation of ADAR2 protein downregulation in the hippocampus is in contrast to evidence indicating an upregulation of ADAR2 mRNA in the temporal lobe in AD (Khermesh et al., 2016). Furthermore, while ADAR3 mRNA has been reported to be upregulated in both the hippocampus (Annese et al., 2018) and prefrontal cortex (Ma et al., 2021), we observed no alterations in ADAR3 protein expression in the hippocampus of our J20 mice.

The underlying causes for these differences are not clear. A limitation of the prior published studies is that they have exclusively evaluated the expression of mRNA, rather than protein. It is important to note that changes in mRNA expression are often not reflected at the protein level (Nie et al., 2006; Guo et al., 2008). This is especially relevant given that RNA editing is a post-transcriptional modification and because ADAR2 can negatively regulate the efficiency of its own protein translation through self-editing (Fu et al., 2016). Thus, the RNA may be highly expressed but the self-editing may prevent protein translation. The fact we have assessed protein rather than mRNA may therefore account for some of the differences observed between our work and prior investigations in human tissue. Alternatively, it is widely acknowledged that mouse models do not perfectly replicate human AD. Specifically, all prior investigations of ADARs in AD have investigated changes in late-onset AD (LOAD) brains, which is not associated with familial AD mutations (Götz et al., 2009; Morris et al., 2014; Cho et al., 2022). Our mouse model, in contrast, is engineered to express mutations associated with familial AD and therefore may not necessarily be an optimal representation of the molecular events transpiring in LOAD.

It remains unclear if the previously published regional variability of ADAR2 mRNA within the brain also reflects a regional variation in ADAR protein levels and activity, given the complex regulation of ADARs. Nonetheless, collectively our observations and those from previous studies suggest the dysregulation of ADAR expression is a clear pathological feature of AD.

ADAR1 has been identified as the principal factor responsible for the majority of editing within non-coding regions and repeat elements in mRNAs. It does not exhibit the specificity that ADAR2 has for protein-recoding genes, particularly in regards to A-I editing events that alter codons such as the GluA2 Q/R site (Tan et al., 2017; Chalk et al., 2019; Costa Cruz et al., 2020). Nevertheless, both proteins possess the capability to edit numerous identical sites and are essential for the preservation of A-to-I editing in the epitranscriptome (Wong et al., 2001). This is supported by the finding that ADAR2 KO mice retain ∼10% editing at the GluA2 Q/R site, suggesting ADAR1 may be responsible for this residual editing (Higuchi et al., 2000). Additionally, this residual editing is also present in varying degrees at other Q/R sites within GluK2, Gabra3, and Cyfip2 (Higuchi et al., 2000; Horsch et al., 2011).

Accordingly and perhaps unsurprisingly, ADAR1 and ADAR2 proteins generally show different expression profiles, are regulated differently, generally edit different targets and are differentially altered in disease (Palop et al., 2003; Esposito et al., 2006; Galvan et al., 2006; Cheng et al., 2007; Wright et al., 2013; Ameen-Ali et al., 2019). ADAR2 and its associated editing is downregulated in several conditions, and this tends to be associated with neurodegeneration (Maas et al., 2001; Peng et al., 2006; Hwang et al., 2016; Srivastava et al., 2017; Moore et al., 2019, 2019; Tran et al., 2019). In contrast, the upregulation of ADAR1 appears to have both ameliorative and pathological effects (Yang et al., 2003; Hartner et al., 2009; Sagredo et al., 2020; de Santiago et al., 2021; Cai et al., 2023). While beyond the scope of this study, future research will be needed to identify if ADARs are dysregulated at earlier time points in mouse models of AD, and if this dysregulation may precede amyloid pathologies and/or whether it plays a role in initiating other AD-associated pathologies beyond loss of dendritic spines and neurodegeneration (Konen et al., 2020; Wright et al., 2023).

It is noteworthy that there is a significant increase in expression of the GluA3 AMPA receptor subunit at 44-weeks in our J20 AD model mice. The result is consistent with the recent report of a significant increase in GluA3 in the CSF of AD patients when compared to cognitively normal patients with subjective cognitive decline (SCD) (Enache et al., 2020). An increase in GluA3 has also been observed in ALS, where it appears that the co-occurrence of ADAR2 downregulation and upregulation of GluA3-containing AMPARs in motor neurons is associated with excitotoxicity and neuronal death (Rembach et al., 2004; Kwak et al., 2010; Gregory et al., 2020). Our findings therefore appear consistent with prior literature.

Interestingly, in contrast, the 3xTg-AD mouse model of AD shows an increase in the expression of the GluA3 subunit at 3 months of age, transitioning to a profound decrease at 12 months of age (Cantanelli et al., 2014). It is possible the 3xTg-AD model may have a different temporal profile of GluA3 expression compared with the J20 model, due to a different pathological profile, e.g., the added presence of neurofibrillary tau tangles (Oddo et al., 2003). Alternatively, the decrease in GluA3 expression may occur at a later time point in J20 mice. It would be useful in future work to determine if the expression of GluA3 declines in J20 mice beyond 44-weeks of age.

In order to consider why GluA3 is upregulated in AD, it is instructive to consider the role of GluA3 at the synapse. GluA3-containing AMPARs are enriched at the synapse when compared to other subunit combinations (Jacob and Weinberg, 2015) and despite their strong presence and preferential placement at synaptic junctions, their contribution to synaptic currents is minimal (Shi et al., 2001, 2001). However, the GluA3 subunit is required for beta-amyloid (Aβ)-mediated synaptic and cognitive deficits (Reinders et al., 2016). Remarkably, hippocampal neurons that do not express GluA3 are protected against Aβ-mediated synaptic depression, spine loss, and degeneration.

Based on the above, we speculate that the initial upregulation in the expression of GluA3 might serve to reduce calcium influx and/or neuronal excitability as a neuroprotective mechanism since AMPA receptors (AMPARs) composed of GluA2/3 subunits show reduced conductance compared to those containing GluA1/2 subunits (Jacob and Weinberg, 2015; Renner et al., 2017). It may then be the case that these neurons become more vulnerable to Aβ-induced toxicity at a later stage, given, as above, GluA3 neurons are more susceptible to Aβ mediated toxicity.

We observed a positive correlation between GluA3 and ADAR2 expression across individual mice. If this were to hold up to further investigation, this finding appears at odds with our observation that, on average, GluA3 was increased and ADAR2 was decreased in J20 mice compared to controls. This provides a hint that the regulation of the expression of these two proteins may be interrelated in a complex manner.

The expression of several proteins known to bind to and regulate ADARs (PIN1, WWP2, FXR1P, and CREB1) were unaltered in the hippocampus of J20 mice. It nevertheless remains possible that these proteins influence the expression and activity of ADARs, especially since protein expression is only one of many determinants of a protein’s activity.

Interestingly, PIN1 and CREB1 exhibited inverse relationships with specific AMPAR subunits in specific mice. Specifically, in the J20 mice model, PIN1 expression was consistently negatively correlated with GluA1, GluA3, and GluA4. Although the direct role of PIN1 in moderating AMPAR expression and function is not documented, one study found that PIN1 ablation in mice induced GluA1 phosphorylation and epileptic seizures (Hou et al., 2021). Moreover, PIN1’s ability to regulate ADAR2 nuclear localisation and facilitate editing of GluA2 offers a plausible theoretical pathway linking these proteins (Marcucci et al., 2011; Behm et al., 2017). In view of that, it is interesting that GluA2 expression was not correlated with PIN1, in either genotype. As above, other changes in PIN1, such as changes in phosphorylation, may modify PIN1 interactions and regulation of GluA2.

Perhaps less surprisingly, the CREB1 protein displayed a positive correlation with the expression of GluA1, GluA2, and GluA3. A substantial body of literature underscores the role of CREB1 in regulating AMPAR subunits. Consistent with our findings, GluA1 appears to have a significant relationship with CREB signalling. Genomic analysis has revealed the potential of CREB1 to bind to at least four identified CRE sequences within the GluA1 promoter (Borges and Dingledine, 2001; Carlezon et al., 2005). Further investigations into the role of CREB1 in synaptic plasticity and learning have shown that inhibition of CREB1 function not only led to a significant reduction in the GluA1 subunit but also potentially has a downregulatory effect on GluA2 and GluA3 subunits as well (Middei et al., 2013). Moreover, it has been hypothesised that CREB1 may act to induce GluA2 expression through the well-established FosB pathway, which has been strongly implicated in AD pathology in both mice and humans (Carlezon et al., 1998; Kelz et al., 1999; McClung and Nestler, 2003). These findings suggest intriguing potential signalling relationships that may play a role in regulating both RNA-editing and AMPAR function in the J20 mice.