Effects of α5 GABAA receptor modulation on social interaction, memory, and neuroinflammation in a mouse model of Alzheimer's disease

Abstract Aims GABAergic modulation involved in cognitive processing appears to be substantially changed in Alzheimer's disease (AD). In a widely used 5xFAD model of AD, we aimed to assess if negative and positive allosteric modulators of α5 GABAA receptors (NAM and PAM, respectively) would affect social interaction, social, object and spatial memory, and neuroinflammation. Methods After 10‐day treatment with PAM, NAM, or solvent, 6‐month‐old transgenic and non‐transgenic 5xFAD mice underwent testing in a behavioral battery. Gene expressions of IL‐1β, IL‐6, TNF‐α, GFAP, and IBA‐1 were determined in hippocampus and prefrontal cortex by qPCR. Results PAM treatment impaired spatial learning in transgenic females compared to solvent‐treated transgenic females, and social recognition in transgenic and non‐transgenic males. NAM treatment declined social interaction in transgenic and non‐transgenic males, while had beneficial effect on cognitive flexibility in non‐transgenic males compared to solvent‐treated non‐transgenic males. Transgenic animals have not fully displayed cognitive symptoms, but neuroinflammation was confirmed. NAM reduced proinflammatory gene expressions in transgenic females and astrogliosis in transgenic males compared to pathological controls. Conclusion PAM and NAM failed to exert favorable behavioral effects in transgenic animals. Suppression of neuroinflammation obtained with NAM calls for more studies with GABAergic ligands in amyloid beta‐ and/or tau‐dependent models with prominent neuroinflammation.


| INTRODUC TI ON
Alzheimer's disease (AD) is a devastating neurological illness that predominantly affects cognition in elderly people, resulting in disability. 1 In AD pathogenesis, the GABAergic system emerged as an important facet in cognitive impairment. 2 GABA A receptors containing the α5 subunit are common but not ubiquitous in the brain, with roles in neuronal excitability, synaptic plasticity, neurogenesis, and cognition in the hippocampus being mediated by both synaptic and tonic inhibition. 3 Accumulating evidence supports the involvement in AD pathology of α5 GABA A receptors in various brain subregions, most conspicuously the hippocampus and cerebral cortex. 4 Excitatory/inhibitory imbalance, observed in AD patients, is likewise present in mouse transgenic models of AD with amyloid beta (Aβ) accumulation and cognitive decline. 3 One such model is 5xFAD mouse model, characterized by an early onset of amyloid plaques deposition in cortical and hippocampal areas. 5 It was suggested that excessive glutamatergic stimulation in an early stage of disease in 5xFAD model is probably compensated by higher GABA currents. 6 The induced prolonged tonic inhibition could give rise to memory impairment, which was restored by acute administration of a negative allosteric modulator (NAM) of α5 GABA A receptors (L-655,708) or with SNAP-5114 to block astrocytic GABA release via GAT3/4. 6 Apparently contradictory, in another mouse model of AD it was shown that both, activation and inhibition of GABA A receptors (by muscimol and bicuculline, respectively), may improve spatial recognition memory. 7 In various tests other than those modeling AD, both a positive allosteric modulator (PAM) [8][9][10] and NAM 11,12 appeared to have a potential to restore object and spatial memory, with PAM being favored in the context of aging. 8 Activity of mouse parvalbumin-positive (PV) interneurons in medial prefrontal cortex (PFC) elicited by low gamma frequency stimulation, as well as muscimol treatment in the dorsomedial PFC, resulted in a prosocial effect leading to increased social interaction. 13 Gene expression study in PFC demonstrated existence of the α5 subunit in pyramidal neurons (39.7% of α5 GABA A receptor positive cells in humans and 54.14% in mice) and in PV interneurons (20% in humans and 16.33% in mice). 14 Positive allosteric modulation at α5 GABAA receptors showed improved social interaction in a rat autism model. 10 On the contrary, both, pharmacogenetic inhibition of PV neurons 13 and bicuculline stereotaxic administration in the PFC 15 decreased social interaction. Even though a selective α5 GABA A receptor NAM failed to decrease social interaction in healthy rats, 16 FG 7142, as a non-selective inverse agonist acting also at α5 GABA A receptors, 17 did reduce social interaction. 16 Hence, one could posit that α5 GABA A receptor PAM and NAM may mitigate and further deteriorate sociability impairment in AD, respectively.
Although Aβ and tau accumulations seemed to be a hallmark of AD for decades, even the most advanced and promising pharmacological strategies directed to reduce Aβ load failed so far to demonstrate consistent cognitive improvements in AD patients (cf. Ref. [18]). Thus, research focus becomes shifted to other contributing mechanisms, such as neuroinflammation. 19 A predominant proinflammatory phenotype is observed in microglia exposed to Aβ in an in vitro AD model, 20 as well as during the pre-plaque stage in murine AD models. 21 Such a proinflammatory microglial phenotype was also observed in humans with mild cognitive impairment without detected Aβ, 22 which may suggest its protective role at the beginning of disease, with an opposite role when AD progresses. 19 Activated microglia secrete proinflammatory cytokines, and could further induce an astrocytic shift to a proinflammatory phenotype. 23 Neurotoxic reactive astrocytes were detected in post-mortem brain tissue, and around 30%-60% of astrocytes in hippocampus and PFC in AD patients had a proinflammatory phenotype. 24 Furthermore, proinflammatory astrocytes could cause excessive GABA release affecting memory in AD models, 25 and also lead to neuronal and oligodendrocyte death. 19 Based on the listed evidence, we aimed to assess the influence of protracted bidirectional α5 GABA A receptor modulation on different cognitive domains and sociability in 6-month-old transgenic and non-transgenic 5xFAD mice of both sexes, as well as on neuroimmune profile in the hippocampus and PFC, characterized by gene expression of proinflammatory cytokines IL-1β, IL-6 and TNFα, as Ligands were dosed at 5 mg/kg once per day during 10 days in transgenic and non-transgenic animals, while control animals obtained SOL.  Figure 1). Hence, all animals were subjected to the same experimental procedure under considerably similar conditions, but divided in three time points, which is relevant for increasing reproducibility of this study. 29 Following behavioral procedures, animals were terminally anesthetized with ketamine (100 mg/kg, i.p.), PFC and hippocampus were collected and snap-frozen in liquid nitrogen. Samples were stored at −80°C until qPCR analysis was conducted.

| Novel object recognition test
In a Plexiglas box (40 × 30 × 30 cm), test animals were exposed during 10 min to a pair of two identical objects located 10 cm far from each other (plastic Rubik's cube or cylinder of glass, randomly distributed across groups), according to the protocol adopted and modified from Colié et al. 30 After 45 min, one of the objects was replaced by the remaining new object and test animal was allowed to freely explore them for 10 min (Novel object recognition test (NORT) for shortterm memory assessment; NORT short-term). 24 h later, animals were again introduced to the same arena for 10 min, but the old object was replaced by the new one (plastic pyramid) (NORT for longterm memory assessment; NORT long-term).

| Three-chamber test (3ct)
Non-transparent Plexiglas apparatus consisted of three parts: the center chamber and two identical side chambers, as previously described. 31 The test comprised of three consecutive phases: habituation, social interaction test (SIT), and social recognition test (SRT). The duration of the first phase was 5 min, and test mouse was allowed only to explore the center zone, with entries to side chambers closed. In the SIT phase, the test mouse was placed into the center chamber, with the passages blocked and a single unfamiliar mouse was placed into either of the side chambers under the wire cage, while the cage in another chamber remained empty. The sliding doors were opened, and the test mouse was allowed to move freely within the apparatus for 10 min. The test mouse was then returned to the center chamber, the doors were closed, and a new unfamiliar mouse was enclosed into the empty cage in the other side chamber. Next, the SRT phase began with opening the sliding doors and the test mouse was allowed to move freely within the apparatus for another 10 min. Test animal exploration time in the narrow zone of cage with or without stimulus animal was used in statistical analysis.

| Morris water maze
Animals were tested in a large tank (170 cm diameter), and the proto- With the exception of the probe swims, trial duration in all phases of MWM was 2 min, and an animal was allowed to swim in 4 consecutive trials per day with pseudorandom departure points.
If the animal was unable to reach the platform, it was guided to the platform and allowed to remain there for 25 s. The average of 4 consecutive trials in the regular training phase was calculated and used for statistical analysis. In reversal cued, latencies to locate the new platform in the first two trials per test day, labeled as trial 1 and trial 2, respectively, were analyzed as modified from Hu et al. 32 The latency to find the platform and number of entries in the platform zone, and old and new platform and number of entries in these platform zones in regular probe and reversal probe, respectively, were used for statistical analysis.

| Tissue preparation and qPCR
Total RNA was isolated by Trizol protocol and RNA concentration was measured. After reverse transcription reaction, the converted cDNA underwent real-time polymerase chain reaction (qPCR) in order to determine IL-1β, IL-6, TNFα, GFAP, and IBA-1 expression levels. Protocol details are provided in the Appendix S1. on the data without normal distribution. 33 Moreover, as there is no nonparametric equivalent for three-way ANOVA, we used this test for the analysis, but for the groups that failed to follow normal distribution, the non-parametric Mann-Whitney U test was utilized to confirm those differences obtained after Sidak post hoc test between groups where at least one of groups did not follow normal distribution (cf. Ref. [34]). Out of 19 differences for the analyzed parameters revealed between two groups where at least one of them did not follow normal distribution, the non-parametric testing confirmed significant differences in 14 cases, in 4 cases the difference was at a non-significant statistical trend level (0.05 < p < 0.1), and only in 1 case, the significant difference disappeared;

| Statistical analysis
the p values are provided in Table S2 in Appendix S1. Accordingly, we decided to present and discuss all changes revealed by ANOVA testing, with the single exception of the case where non-parametric analysis failed to demonstrate even a trend level of difference.
Statistical analysis was run in IBM SPSS Statistics 25 software and graphs were plotted in GraphPad Prism 9. The statistical significances are shown on graphs as * for 0.01 < p < 0.05, ** for 0.001 < p < 0.01, *** for p < 0.001. If significant, overall effects were reported in each graph.

| Control transgenic females had impaired object memory and α5 GABA A ligands were devoid of any effects on object memory in AD model
Results and significant overall effects of factors genotype, treatment, sex or object or their interaction in NORT were shown in Figure 2.
In NORT for short-term memory estimation, solvent-treated nontransgenic, but not transgenic females spent more time with a new compared to an old object (p = 0.045 and p = 0.985, respectively, Figure 2A). In males, neither solvent-treated non-transgenic nor transgenic animals did discriminate between the new and the old object (p = 0.186 and p = 0.251, respectively, Figure 2A). However, after PWZ-029 administration, non-transgenic males preferentially explored the new as compared to the old object (p = 0.022, Figure 2A). Neither PAM nor NAM treatment did have any effect in transgenic animals in NORT (p = 0.325 for males and p = 0.560 for females, and p = 0.416 for males and p = 0.379 for females, respectively, Figure 2A). In NORT for longterm memory assessment, only non-transgenic female controls showed a trend in favor of spending more time with the new as compared to the old object (p = 0.053, Figure S1 in Appendix S1).

| MP-III-022 and PWZ-029 impaired social recognition and social interaction, respectively, in 5xFAD mice
For 3ct, results and overall effects of factors genotype, treatment, sex or chamber or their interaction were included in Figure 2.
In SIT and SRT, both transgenic and non-transgenic males treated with solvent spent more time in the chamber that contained a single unfamiliar animal or new animal, respectively, compared to chamber with empty cage or old animal, respectively (p = 0.036 and p = 0.028, respectively, Figure 2B). MP-III-022-treated transgenic and nontransgenic males explored more cage with conspecific compared to empty cage in SIT (p = 0.013 and p = 0.012, respectively, Figure 2B).
On the other hand, PWZ-029-treated transgenic and non-transgenic males explored more chamber with new conspecific, as compared to old in SRT (p = 0.005 and p = 0.015, respectively, Figure 2c). In SRT, PWZ-029-treated non-transgenic females investigated more chamber with new conspecific compared to chamber with old conspecific (p = 0.046, Figure 2c).

| MP-III-022 treatment impaired procedural learning in transgenic females, and transgenic females showed cognitive inflexibility regardless of treatment
Results obtained in MWM, together with overall effects of factors genotype, treatment, sex or day (where applicable) or their interactions, were illustrated in Table 1 and Figure 3.
Transgenic females treated with MP-III-022 had higher latency to find platform compared to solvent-treated transgenic females (p = 0.024, Table 1 Table 1). No significant difference for the latency to find the platform zone was found between groups in the probe day ( Figure S2).
In probe trial, transgenic female controls spent less time in the periphery compared to non-transgenic female controls (p = 0.016, Figure 3A). MP-III-022-treated transgenic females spent more time in the peripheral zone compared to transgenic females treated with solvent (p = 0.007, Figure 3A).
In the reversal cued day, transgenic female animals treated with MP-III-022 had lower latencies to find the new platform in trial 2 compared to trial 1 (p = 0.031, Figure 3B). Further, transgenic male controls showed lower latency to find the new platform in trial 1 compared to control non-transgenic male mice (p = 0.013, Figure 3B).
Non-transgenic male mice treated with PWZ-029 had lower latency to find a new platform in trial 1 compared to control non-transgenic males (p = 0.006, Figure 3B).

| Neuroinflammation in transgenic animals and protective effects of PWZ-029 treatment
Results from qPCR experiment and overall effects of factors genotype, treatment, and sex or their interactions are shown in Figure 4.
GFAP levels in HC in transgenic male controls were higher compared to non-transgenic male control (p = 0.008, Figure 4G), while transgenic female controls showed a trend effect toward increasing GFAP levels in HC compared to female non-transgenic controls (p = 0.058, Figure 4G). Additionally, transgenic male controls had higher GFAP levels compared to transgenic female control in HC (p = 0.021, Figure 4G). PWZ-029-treated transgenic males and non-transgenic females showed lower and higher GFAP levels compared to solvent-treated transgenic males (p = 0.026, Figure 4G) and non-transgenic females (p = 0.011, Figure 4G), respectively, in HC. In PFC, both male and female transgenic animals treated with solvent had higher GFAP levels compared to non-transgenic control males (p = 0.003, Figure 4H) and females (p = 0.027, Figure 4H), respectively. IBA-1 levels were higher in male and female transgenic controls compared to male (p = 0.004, Figure 4i) and female nontransgenic controls (p < 0.001, Figure 4i), respectively, in HC. No statistical difference for IBA-1 levels in PFC was found between groups ( Figure 4J).

| DISCUSS ION
The current study in 5xFAD mice aimed to model early AD-like changes in different cognitive domains and sociability. It assessed the consequences of protracted α5GABA A receptor positive or negative modulation, elicited by selective ligands MP-III-022 27 and PWZ-029, 26 respectively.
Performance in the novel object recognition task for shortterm object memory was impaired in transgenic compared to nontransgenic 6-month-old female animals, providing the evidence that the model was successful to mimic object memory decline, as previously reported. 35 Not only that MP-III-022 failed to demonstrate improvement in object memory in transgenic female animals, but also suppressed this type of memory in non-transgenic females.
These data are aligned with results obtained in rats, with MP-III-022 suppressing long-term object memory when dosed at 10 mg/kg 24 h before the test. 9 On the other hand, not only control transgenic, but also nontransgenic males did not discriminate between new and old objects, Another contributing factor may be the fact that female rodents tend to explore environment more than males, which may result in less interaction. 41 Control transgenic and non-transgenic animals showed comparable memory during training sessions in MWM as well as on the probe day, as previously reported. 42 The finding could stem from suboptimal learning abilities of the tested mice due to retinal degeneration gene inherited from SJL background, as previously described. 43 MP-III-022 treatment in transgenic females appeared to be detrimental on learning in the training phase of MWM, which is expected for positive modulation of α5 GABA A receptors. 44 In the first day of cued reversal learning phase, MP-III-022 treated transgenic females were enabled to find the new platform across trials.
F I G U R E 4 QPCR results from transgenic and non-transgenic 5xFAD animals of both sexes treated with PWZ-029, MP-III-022, or solvent. Gene expression for IL-6, IL-1β, TNFα as proinflammatory cytokines in hippocampus (HC) and prefrontal cortex (PFC) are shown (a-f, respectively). The gene expression of GFAP and IBA-1 as markers of astrogliosis and microgliosis, respectively, were determined in HC and PFC (g-j, respectively). The statistical significances were revealed with the three-way ANOVA followed by Sidak post hoc test and are shown on graphs as * for 0.01 < p < 0.05, ** for 0.001 < p < 0.01, *** for p < 0.001. The overall effects if significant are given in the upper right corner. The abbreviations used are: g, s, and t for genotype, sex, and treatment, respectively Non-transgenic male animals treated with PWZ-029 had the shortest latency for finding the new platform in reversal training day 1 compared to solvent-treated non-transgenic males, which points to a successful previous learning as a beneficial consequence of negative modulation of α5 GABA A receptors.
In reversal probe, transgenic females, independently of treatment, showed higher latencies to find the new platform compared to old one, which could indicate a sex-dependent impaired cognitive flexibility as reported in different protocols for the reversal probe In 5xFAD transgenic model, pathology-related microglial phenotype is found, and results are translated to human microglia in AD. 49 In our study, microgliosis was demonstrated by increased IBA-1 expression in HC of solvent-treated transgenic animals, as compared to non-transgenic animals regardless of sex influence; in PFC, such a difference was not detected despite an overall genotype effect.
Interleukin-1 (IL-1), interleukin-6 (IL-6), and tumor necrosis factor α (TNFα) are produced mainly by immune cells, but could also originate from neural cells. 50,51 In our study, all proinflammatory cytokine transcripts were upregulated in transgenic animals compared to non-transgenic animals regardless of sex. IL-1 could contribute to synaptic loss and TNFα could induce cell death, 19 suggesting their significant interplay in chronic inflammatory processes such as AD.
Administration of IL-1 receptor antagonist antibody in an AD mouse model restored cognitive deficits observed in Morris water maze and reduced neuroinflammation. 52 IL-6 was (or at least tended to be) upregulated in PFC and HC of AD patients. 53,54 In a rat AD model, IL-1β, IL-6, and TNFα were increased in the hippocampus. 52  We acknowledge the statistical expertise of Ms. Danijela Milenković and expert technical assistance of Dr. Aleksandar Obradović.

CO N FLI C T O F I NTE R E S T
The authors have no conflict of interest to report.

DATA AVA I L A B I L I T Y S TAT E M E N T
The authors confirm that the data supporting the findings of this study are available within the article and its supplementary materi-