Cognitive Trajectories in Longitudinally Trained 3xTg-AD Mice

Abstract

Preclinical studies in Alzheimer’s disease (AD) often rely on cognitively naïve animal models in cross-sectional designs that can fail to reflect the cognitive exposures across the lifespan and heterogeneous neurobehavioral features observed in humans. To determine whether longitudinal cognitive training may affect cognitive capacities in a well-characterized AD mouse model, 3xTg and wild-type mice (n=20) were exposed daily to a training variant of the Go-No-Go (GNG) operant task from 3 to 9 months old. At 3, 6, and 9 months, performance on a testing variant of the GNG task and anxiety-like behaviors were measured, while long-term recognition memory was also assessed at 9 months. In general, GNG training improved performance with increasing age across genotypes. At 3 months old, 3xTg mice showed slight deficits in inhibitory control that were accompanied by minor improvements in signal detection and decreased anxiety-like behavior, but these differences did not persist at 6 and 9 months old. At 9 months old, 3xTg mice displayed minor deficits in signal detection, and long-term recognition memory capacity was comparable with wild-type subjects. Our findings indicate that longitudinal cognitive training can render 3xTg mice with cognitive capacities that are on par with their wild-type counterparts, potentially reflecting functional compensation in subjects harboring AD genetic mutations.

1. Introduction

Transgenic rodent models of Alzheimer’s disease (AD) have revealed important insights, but these models can fail to recapitulate cognitive exposures across the lifespan and the heterogeneous neurobehavioral features observed in human populations, including individuals with life-long cognitive enrichment who do not develop cognitive impairment (cognitive resilience) [1, 2]. As most studies examine cognitively naïve animals in cross-sectional study designs, it remains unclear how longitudinal cognitive training may affect resiliency to cognitive deficits that are otherwise associated with AD neuropathology.

One of the most commonly used AD models, the triple-transgenic AD mouse (3xTg-AD or 3xTg), exhibits both extracellular plaques (approximately beginning at 6 months old) and tangle development (approximately beginning at 12 months old) attributed to three AD-related mutations (the APP_swe and Tau_P301L transgenes on a PS1_M146V knock-in background) [3–5]. While its drawbacks include the presence of a MAPT mutation that does not naturally occur in AD and spatially distinct patterns of amyloid-beta (Aβ) and tau pathology, this model nonetheless displays other key hallmarks of AD, including impaired neuronal viability and reactive astrogliosis [6–8]. In cross-sectional experiments, cognitive deficits in this model typically emerge by 6–9 months old, and include associative learning (passive avoidance), spatial working memory (Morris water maze [MWM], Y-maze), and recognition memory (novel object recognition [NOR]) [9]. Of the few longitudinal studies in 3xTg mice, impairments in NOR and MWM performance have been reported as early as 4 and 6 months old, respectively [5, 10].

Despite these well-characterized phenotypes, it remains unclear how longitudinal cognitive training influences cognitive capacities in 3xTg mice, particularly in a paradigm that translates well to humans. Along with the beneficial effects of general environmental enrichment reported in AD mouse models [11], intermittent cognitive training on the MWM throughout life (i.e., every 3 months from 3 to 18 months old) can enhance MWM, NOR and Barnes maze performance of aged 3xTg mice compared to their cognitively-naïve, transgenic counterparts [12, 13], while similar MWM training during adulthood (i.e., every 2 months from 5 to 9 months old) can mask the cognitive benefits otherwise associated with an Aβ-immunotherapy [14]. However, the stress of swimming and the impacts of other non-cognitive confounds (e.g., visual sensory function) limit the translational relevance of the MWM to humans [15, 16]. Conversely, the Go-No-Go (GNG) is a highly translatable task which measures attention and response inhibition deficits that are evident in the early stages of AD pathogenesis, and relies on cortical networks that are implicated in the development of mild cognitive impairment (MCI) and AD[17–21]. Thus, the aim of the current study was to examine how daily exposure to a training variant of the GNG task differentially impacted performance of 3xTg and wild-type mice in a testing variant of the GNG task in early-adulthood (3 months old), middle-age (6 months old) and late-adulthood (9 months old).

To address this aim, 3xTg and wild-type mice were trained to criterion on a training variant (50:50) of the GNG task at 3 months old, followed by testing on another variant of the task with a different stimulus probability (20:80) across three consecutive testing days. Testing performance was also measured at 6 and 9 months, and daily exposure to the training variant continued in-between these time points (Figure 1). Because trait anxiety can moderate inhibitory control processes in the GNG task [22, 23], we also measured anxiety-like behaviors using an elevated plus maze (EPM) at 3, 6 and 9 months. Finally, long-term recognition memory was assessed with NOR at 9 months old. We hypothesized that exposure to life-long cognitive training would functionally compensate for the cognitive impact of AD-related genetic mutations, as evidenced by similar GNG testing performance between genotypes across ages as well as similar NOR performance.

Figure 1.

Overview of study design. After initial autoshaping and pretraining, 3xTg and wild-type mice were trained to criterion on the 50:50 variant of the Go-No-Go (GNG) task at 3 months old, followed by testing on the 80:20 variant of the task across three consecutive testing days. Testing performance was also measured at 6 and 9 months old, while daily training on the 50:50 variant continued in-between these time points. After GNG testing at each time point, behavior in an elevated plus maze (EPM) was collected, and long-term memory (24 hr.) was also measured in a Novel Object Recognition (NOR) task at 9 months. ^A The 50:50 variant of the Go-No-Go task presented equal distributions of “Go” and “No-Go” trials. ^B The 20:80 variant of the Go-No-Go task presented 20% and 80% of trials as “Go” and “No-Go”, respectively. ^C Immunoblotting on a subset of brain tissues (hippocampus) was used to confirm the presence and absence of Aβ oligomers in 3xTg and wild type mice, respectively. Lysates from 3xTg subjects show the presence of a 64kDa band that corresponds to high molecular weight Aβ oligomers.

2. Methods

2.1. Subjects

Wild-type (n = 10; Female = 5, Male = 5) and 3xTg mice (n = 10; Female = 5, Male = 5) were utilized in the current study. Breeding pairs were acquired from Jackson Laboratories (Bar Harbor, Maine) and a colony for each strain was subsequently established. Wild-type mice exhibited a B6129SF2/J background (hybrid of C57BL/6J and 129 Sv strains), and 3xTg mice (B6;129-Psen1^tm1Mpm Tg(APPSwe,tauP301L) 1Lfa/Mmjax strain) exhibited mutations in three genes associated with AD (APP_swe, MAPT_P301L and PSEN1_M146V). The expression of these transgenes is primarily restricted to brain regions afflicted in AD, such as the hippocampus and cerebral cortex, resulting in the pathological hallmarks of AD at approximately 6 months of age [3–5]. One 3xTg subject (n = 1) expired at 9 months of age prior to the completion of NOR testing. Mice were individually housed in temperature/humidity-controlled standard housing conditions, with a 12h light/dark cycle (lights on 07:00 local time), and progressively water-restricted to 5 min of water access per day prior to the commencement of behavioral training. Mice were individually housed to ensure equal access to water during daily watering sessions without potential interference from conspecifics. All behavioral paradigms were conducted 7 days/week between 09:00 and 16:00. Food pellets (LabDiet, St. Louis, MO) were available ad libitum during the experiment. All experimental procedures were authorized by the Institutional Care and Use Committee (IACUC) of Temple University and complied with National Institute of Health regulations.

2.2. Go-No-Go (GNG) decision-making task

Apparatus

Mouse operant conditioning chambers (MED Associates; St Albans, VT, USA) were equipped with a standard grid floor and house light (28 V, 100 mA), a central reward port attached to a fluid dipper, and two ultra-sensitive retractable levers positioned on the left and right sides of the chamber with large cue lights (2.5 cm; 28 V, 100 mV) above each lever. All events, including light presentation, lever operations, and reward delivery, were controlled by a SmrtCtrl interface running MED-PC IV software on a Dell PC (Optiplex 960). Operant boxes were equipped with a video tracking system consisting of HD cameras (960H/700TVL) connected to an 8-channel HD Analog DVR system (Q-See, Anaheim, CA).

Behavioral Training and Testing

Subjects were trained and tested in an operant Go-No-Go (GNG) visual discrimination task, which is designed to emulate response conflict with the added component of impulse control by introducing the need to actively withhold from responding on specific trials, as described in our previous study [24].

Animals (6–7 weeks old at the beginning of the study) were initially autoshaped on a FR-2 reinforcement schedule to acquire lever press responding and subsequent reward reinforcement (10ul of 0.066% saccharine water). After attaining criterion (i.e., 30 lever presses/session), subjects were advanced to a pretraining phase. Each session of the pretraining phase began with the illumination of the house-light, located in the rear of the chamber. Following an intertrial interval (ITI) of 9 ± 3s, a trial commenced when the subject was presented with an illumination of a panel light in conjunction with a single, corresponding lever. In order to obtain a reward, the animal was required to press the active lever (as denoted by a light cue above the lever) within 10s of presentation. Lever presentations were randomized, and to control for novelty effects associated with the visual stimulus, trials were randomly assigned with unpredictably occurring visual cues. Once subjects attained criterion on pretraining (30 rewards and ≤10% omissions for three consecutive days), they were advanced to the GNG task.

After attaining criterion on the pretraining phase, animals progressed to the final stage of the GNG task, during which they were required to discriminate between the pseudo-randomized presentation of response cues (Go Trials) and withholding cues (No-Go Trials) in 48-trial sessions (Figure 1). A Go trial began with the continuous illumination of a 7s visual cue (either the left or right light), followed by the presentation of both levers 2s later. Two lever press responses (FR-2) on the cued lever were scored as a hit and followed by reward delivery. A response to the incorrect lever was not rewarded and resulted in a time-out phase characterized by a 10s extinguishing of the house light. If no response was recorded after 5s of lever presentation, levers co-terminated with the stimulus light and the trial was scored as an omission, and the subject received neither a reward nor a time-out period. A No-Go trial began with the flashing (2Hz) of the visual cue for 7s (either left or right). As with Go trials, levers where presented 2s later and co-terminated with the flashing cue after 5s. During a No-Go trial, lever press responses (FR-2) were not rewarded, resulted in a timeout phase, and were labeled as false alarms. To obtain a reward during a No-Go trial, animals were therefore required to completely refrain from responding to lever presentations, which was considered a correct rejection. A single response on either lever during all trial types was not scored, allowing the animal an opportunity to withhold or continue lever pressing. Again, an ITI of 9 ± 3s was employed throughout. Criteria performance was defined as 70% overall correct responses on Go and No-Go trials (i.e., > 70% of Hits and <30% of False Alarms, respectively), with ≤20% omissions over three consecutive days.

Animals were trained daily on a 50:50 variant of the task, which randomly presented 50% response cues (Go trials) and 50% of withholding cues (No-Go trials). At 3, 6 and 9 months old, animals were tested on a 20:80 variant of the task over three consecutive testing days, which randomly presented 20% response cues (Go trials) and 80% withhold cues (No-Go trials). This strategy was adopted to assess decision-making and impulse control under conditions of uncertainty, where the decision threshold for choice accuracy and reaction times are suggested to increase under stringent inhibitory control conditions [25]. Testing days occurred 24-hr after the final 50:50 GNG training session at each time point. In-between testing at 3, 6 and 9 months, animals were maintained daily on the 50:50 variant of the GNG task (Figure 1). Depending on individual variation in response times, GNG sessions (i.e., a training session on the 50:50 variant or a testing session on the 20:80 variant) were completed in approximately 30 minutes per animal.

Behavioral Measures

Training performance on 50:50 GNG was assessed as an average of the three days on which subjects attained the criterion at 3 months old, and the three days preceding 20:80 GNG testing sessions at 6 and 9 months old. Testing performance on 20:80 GNG was assessed as an average of the three testing days in primary analyses and separately across each testing day in secondary analyses (e.g., testing day 1 performance at 3 months vs 6 months vs 9 months). Measures included sessions to criterion, percentage of hits (h), percentage of false alarms (fa), hit latencies (2^nd lever press), false alarm latencies (2^nd lever press), sensitivity index scores and efficiency scores. The sensitivity index is a composite measure of attention and discriminability, and was calculated using the formula [p(h)−p(fa)]/[2(p(h)+p(fa))]−[p(h)+p(fa)]², as described previously [24, 26]. Efficiency scores were calculated as the ratio of number of rewards earned to the total number of responses in Go trials [27].

2.3. Elevated plus maze (EPM)

Subjects were assessed in the EPM at 3, 6 and 9 months. Prior to EPM, animals were habituated to the testing environment for 10 min. The apparatus was opaque in color, composed of polyvinyl chloride plastic, and consisted of two open arms (10cmx50cmx30cm) as well as two closed arms (10cmx50cmx30cm). Subjects were placed in the center of the maze and allowed to explore the environment for 5 min. At the conclusion of testing, a subject was removed and placed back in standard housing conditions. Sessions were video recorded and later measured for time spent in the open and closed arms by an observer blind to testing conditions. Performance was defined as percent time in open arms relative to time spent in closed arms, and percent time in open arms relative to total time. EPM performance was measured 24-hours after animals were tested on the 20:80 variant of the GNG task and 24-hours before animals resumed daily training on the 50:50 variant of the GNG task.

2.4. Novel object recognition (NOR)

At 9 months old, long-term recognition memory was assessed in the NOR task, as described previously [28]. The apparatus was rectangular in shape (40cmx30cmx30cm) and composed of polyvinyl chloride plastic. NOR objects (e.g., Lego blocks, Erlenmeyer flask, conical tube etc.,) were chosen based on their relative size to an animal’s body and their distinguishing stimuli. NOR testing occurred 24-hours after animals were tested on EPM. In brief, the NOR task consisted of a habituation phase, a sample phase, and a testing phase. On Day 1, subjects were individually submitted to the apparatus and allowed to explore the arena for 10 min. On Day 2, with two identical objects in a symmetric position from the center of the arena, subjects were again introduced to the testing arena and allowed to explore for 10 min. On Day 3, subjects were reintroduced to the testing arena with a familiar object used during the sample phase and a novel object placed in the same location as the sample stimuli. NOR testing sessions were video recorded and later assessed for the time spent exploring each object by an observer blind to testing conditions. Exploration was defined as an animal being within 2cm of the object and oriented towards it. Performance was defined as absolute preference (time spent exploring the novel object minus the time exploring familiar object) as well as relative preference (time exploring novel object relative to the total time exploring both objects).

2.5. Immunoblotting

Immunoblotting with an Aβ oligomer antibody was conducted for genotype validation in a subset of randomly selected wild type and 3xTg mice (n = 5/genotype). After behavioral testing at 9 months old concluded, cervical dislocation was applied, followed by rapid decapitation and dissection of bilateral hippocampi. Samples were homogenized in cold (4C) 0.05M HEPES-NaOH buffer (pH 7.4) containing 0.32M sucrose, 5mM MgCl₂, 0.2mM EDTA, 0.2mM EGTA, 1% Triton X-100 and a protease inhibitor cocktail. Protein concentrations were estimated with a modified Lowry assay (Pierce, Rockford, IL). Lysates were subjected to SDS-PAGE electrophoresis for protein separation and blots were transferred onto PVDF membranes to detect protein bands for oligomeric Aβ using a rabbit anti-oligomer AR-11 polyclonal antibody (ThermoFisher Scientific, #AHB0052). Beta-actin (β-actin) was used as a gel-loading control. Samples from 3xTg, but not wild-type mice, showed the presence of high molecular weight Aβ oligomers (Figure 1C), consistent with a previous report [29].

2.6. Statistical analyses

Differences in performance across repeated measures (e.g., age [3, 6, 9 months old] or testing days [1, 2, 3]), as a function of between-subjects factors (e.g., genotype [3xTg, wild-type] or sex [female, male]), were assessed with mixed-model analyses of variance (ANOVA). One-way ANOVAs were conducted for direct comparison of between-subjects factors and post-hoc analyses. Associations between performance at 9 months old (i.e., the time point at which animals were euthanized for immunoblotting) and hippocampal Aβ oligomer levels (normalized to β-actin) in 3xTg mice were assessed using Spearman’s correlations. Analyses were performed using SPSS version 28.0.1 (IBM-SPSS, Armonk, NY) and figures were generated using R version 4.1.2 (R Foundation, Vienna, Austria). Outliers were defined by 3SDs, and a two-sided p < .05 defined statistical significance.

3. Results

3.1. Improved longitudinal performance in cognitive training

The number of sessions to attain criterion on the training variant (50:50) of the GNG task at 3 months old did not differ between genotypes (F_{1, 18} = 0.19, p = .672) (Figure 2A) or sex (F_{1, 18} = 0.03, p = .869). To examine if genotype influenced GNG 50:50 performance across ages, we compared performance across 3, 6, and 9 months as a function of genotype. With age and genotype as within-and between-subject factors, respectively, we did not find changes in hit percentages (F_{2, 36} = 1.08, p = .350) (Figure 2B) with increasing age, but we did observe significantly decreased percentages of false alarms (F_{2, 36} = 6.30, p = .005) (Figure 2C), shortened hit (F_{2, 36} = 8.31, p = .001) and false alarm (F_{2, 36} = 8.99, p < .001) latencies, as well as elevated efficiency (F_{2, 36} = 4.06, p = .026) and sensitivity index scores (F_{2, 36} = 4.54, p = .017) (Figure 2D). Such main effects of age were not modified by genotype, and using sex as a between-subjects factor yielded a similar pattern of results. In the subset of 3xTg mice used for genotype validation, Aβ oligomer levels were not associated with GNG 50:50 performance at 9 months old (all measures rho < 0.80, p >.05). These findings indicate that subjects learned the training variant (50:50) of GNG task at comparable rates, and that further training on this cognitively-demanding task throughout adulthood leads to general improvements in performance across multiple measures that are not dependent on a subject’s genotype nor sex.

Figure 2.

Performance on the training variant (50:50) of the Go-No-Go (GNG) task. A) The number of training sessions to attain criterion on the GNG task did not differ between genotypes. B) Improvements in hit percentages were not statistically significant across ages or across genotype. C) Decreased false alarm percentages were significantly reduced across ages independent of genotype. D) Improvements in sensitivity index scores were statistically significant across ages independent of genotype. Differences in performance across ages as a function of genotype were assessed with mixed-model analyses of variance, and one-way analyses of variance were used to directly compare genotypes at each age. *p < .05; **p < .01.

3.2. Subtle effects of genotype on testing performance in 20:80 GNG task

After averaging performance across all three testing days for each time point (3, 6, 9 months old) in the 20:80 variant of GNG task, our primary analyses showed no interaction effects of genotype across ages, including hits (F_{1, 18} = 0.55, p = .588), false alarms (F_{1, 18} = 1.98, p = .170) and sensitivity index scores (F_{1, 18} = 1.51, p = .249), indicating that longitudinal cognitive training renders 3xTg mice with average testing performance that is on par with their wild-type counterparts. Analyses with sex as a between-subjects factor similarly indicated no significant differences between males and females (all measures F < 4.41, p >.05). Aβ oligomer levels were not associated with average GNG 20:80 performance in the subset of 3xTg mice used for genotype validation (all measures rho < 0.50, p >.05). However, with testing day and genotype as within-and between-subject factors, respectively, we detected interactions across multiple measures in early-adulthood (3 months old), including hit (F_{2, 36} = 4.42, p = .019) and false alarm percentages (F_{2, 36} = 10.42, p < .001), as well as sensitivity index (F_{2, 36} = 9.24, p < .001) and efficiency (F_{2, 36} = 5.10, p = .011) scores; false alarm (F_{2, 36} = 1.75, p = .188) and hit latencies (F_{2, 36} = 0.91, p = .411) were not modified by genotype at 3 months old. To better understand the sources and implications of these interactions across ages, we conducted secondary analyses of individual testing days.

Using testing day 1 performance, we found that percentages of false alarms across ages (3, 6 and 9 months old) were significantly dependent on genotype (F_{2, 36} = 8.36, p = .001). Such effects were primarily driven by higher false alarm rates among 3xTg mice at 3 months old (F_{1, 18} = 8.08, p = .011), whereas false alarm rates at 6 (F_{1, 18} = 0.14, p = .709) and 9 (F_{1, 18} = 0.00, p = .986) months old were similar between genotypes (Figure 3A). We did not find genotype interactions on other testing day 1 measures across ages, including hit percentages (F_{2, 36} = 0.55, p = .580), sensitivity index (F_{2, 36} = 2.31, p = .114) and efficiency (F_{2, 36} = 2.00, p = .150) scores; however, we did detect a significant decrease in hit percentages (F_{1, 18} = 5.69, p = .028) and sensitivity index scores (F_{1, 18} = 5.85, p = .026) among 3xTg mice at 9 months old (Figure 3B–C).

Figure 3.

Performance on the testing variant (20:80) of the Go-No-Go (GNG) task. A) Percentages of testing day 1 false alarms across ages were significantly modified by genotype, primarily driven by significantly higher rates among 3xTg mice at 3 months old. Genotype interactions were not detected across other testing day 1 measures, although we did detect a significant decrease in B) hit percentages and C) sensitivity index scores among 3xTg mice at 9 months old. D) Genotype interactions were not detected on testing day 2 false alarm percentages, but E) hit percentages and F) sensitivity index scores across ages were significantly modified by genotype, primarily driven by significantly higher values among 3xTg mice at 3 months old. G) Genotype interactions were not detected on testing day 3 false alarm percentages. H) Percentages of testing day 3 hits were significantly modified by genotype, primarily driven by significantly higher and lower rates among 3xTg mice at 3 and 9 months old, respectively. I) Testing day 3 sensitivity scores were significantly modified by genotype, primarily driven by significantly lower values at 9 months old. Differences in performance across ages as a function of genotype were assessed with mixed-model analyses of variance, and one-way analyses of variance were used to directly compare genotypes at each age. *p < .05; **p < .01.

On testing day 2, we did not observe an interaction of genotype on false alarm rates across ages (F_{2, 36} = 2.03, p = .146) (Figure 3D); however, hit percentages (F_{2, 36} = 5.03, p = .012), sensitivity index (F_{2, 36} = 5.08, p = .011) and efficiency (F_{2, 36} = 3.46, p = .042) scores across ages were significantly modified by genotype. Such effects were attributed to higher hit percentages (F_{1, 18} = 6.64, p = .019) and sensitivity index scores (F_{1, 18} = 6.83, p = .018) among 3xTg animals at 3 months old, whereas performance at 6 (hits: F_{1, 18} = 0.00, p = .987; sensitivity index: F_{1, 18} = 0.01, p = .931) and 9 (hits: F_{1, 18} = 0.57, p = .460; sensitivity index: F_{1, 18} = 0.37, p = .549) months old was similar between genotypes (Figure 3E–F). Efficiency scores were not significantly modified by genotype in post hoc analyses.

On testing day 3, we did not observe an interaction of genotype on false alarm rates (F_2, 36 = 1.50, p = .237) (Figure 3G), but hit percentages (F_{2, 36} = 7.75, p = .002), sensitivity index (F_{2, 36} = 5.27, p = .010) and efficiency (F_{2, 36} = 4.31, p = .021) scores across ages were significantly modified by genotype. Such effects on hit percentages were driven by higher rates at 3 months old (F_{1, 18} = 9.61, p = .006) and lower rates at 9 months old (F_{1, 18} = 4.56, p = .047) among 3xTg animals (Figure 3H). Regarding the sensitivity index, although scores were not different between genotypes at 3 (F_{1, 18} = 2.97, p = .103) or 6 (F_{1, 18} = 0.58, p = .457) months old, we detected lower scores among 3xTg subject at 9 months old (F_{1, 18} = 7.10, p = .016) (Figure 3I). Efficiency scores were not significantly modified by genotype in post hoc analyses. In the subset of 3xTg mice used for genotype validation, Aβ oligomer levels were related to improved GNG testing performance at 9 months old on testing day 1 (sensitivity index scores: rho = 0.93, p < .001; efficacy scores: rho = 0.78, p < .001), testing day 2 (hits: rho = 0.51, p = .021; efficacy scores: rho = 0.59, p = .006) and testing day 3 (hits: rho = 0.71, p < .001; sensitivity index scores: rho = 0.69, p < .001; efficacy scores: rho = 0.71, p < .001). Despite general improvements in average testing performance with increasing age across genotypes, these findings suggest that 3xTg subjects display subtle inhibitory control deficits and improved signal detection at 3 months old (i.e., increased testing day 1 false alarms, increased testing day 2 hit percentages and sensitivity index scores, increased testing day 3 hit percentages), and by 9 months old, such animals show signs of impaired signal detection (i.e., reduced hits and sensitivity index scores on testing days 1 and 3).

3.3. Effects of genotype on anxiety-related behavior

With age and genotype as within-and between-subject factors, we next examined EPM performance. Analyses showed that the percent of time spent in open arms across ages was significantly modified by genotype, both relative to time in closed arms (F_{2, 36} = 4.83, p = .014) and total time (F_{2, 36} = 7.04, p = .003). Such effects were attributed to greater time in open arms at 3 months old among 3xTg animals (% rel. to time in closed arms: F_{1, 18} = 4.43, p = .050; % rel. to time total time: F_{1, 18} = 4.78, p = .042), whereas time in open arms at 6 month old (% rel. to time in closed arms: F_{1, 18} = 0.00, p = .967; % rel. to time total time: F_{1, 18} = 0.42, p = .524) and 9 months old (% rel. to time in closed arms: F_{1, 18} = 0.14, p = .715; % rel. to time total time: F_{1, 18} = 0.34, p = .569) was similar across genotypes (Figure 4A–B). EPM measures with sex as a between-subjects factor revealed no significant differences between males and females (% rel. to time in closed arms: F_{2, 36} = 0.21, p = .812; % rel. to time total time: F_{2, 36} = 0.34, p = .716). Among the subset of 3xTg mice used for genotype validation, Aβ oligomer levels were not associated with EPM performance at 9 months old (% rel. to time in closed arms and % rel. to time total time: rho = −0.40, p = .510). These findings suggest that 3xTg subjects display a lack of inhibition and/or decreased anxiety at 3 months old, but such differences are attenuated by later adulthood.

Figure 4.

Performance in the elevated plus maze (EPM). A) Time in open arms (% relative to time in closed arms) and B) time in open arms (% relative to total time) across ages were significantly modified genotype, with significantly higher times in open arms at 3 months old among 3xTg animals. Differences in performance across ages as a function of genotype were assessed with mixed-model analyses of variance, and one-way analyses of variance were used to directly compare genotypes at each age. *p < .05; **p < .01.

3.4. Sex-differences in long-term recognition memory

Preliminary analysis confirmed NOR procedures were effective, given that the overall time spent at the familiar object (M = 15.93, SD = 13.13) was significantly less than the time spent at the novel object (M = 27.98, SD = 15.64; F_{1, 17} = 3.49, p = .002). Neither absolute (F_{1, 17} = 0.26, p = .617) nor relative (F_{1, 17} = 0.00, p = .959) preferences were significantly associated with genotype. Aβ oligomer levels were not associated with NOR performance in the subset of 3xTg mice used for genotype validation (absolute and relative preferences: rho = 0.80, p = .104). Relative preference was not dependent on sex (F_{1, 17} = 2.68, p = .120), but we found that absolute preference was significantly higher among female subjects compared to male subjects (F_{1, 17} = 5.99, p = .026) (Figure 5A–B). Analyses examining the interaction of sex and genotype, although limited by statistical power, suggested no effect moderation for absolute (F_{3, 15} = 0.02, p = .899) or relative (F_{3, 15} = 0.18, p = .677) preferences. These results indicate that among highly trained, cognitively enriched subjects, memory capacities in the NOR task at 9 months old are not modified by genotype, although females seem to display improved capacities compared to males, at least in terms of absolute preference.

Figure 5.

Performance in the Novel Object Recognition (NOR) task. A) Absolute preference (time spent exploring the novel object minus the time exploring familiar object) was significantly higher among female mice. B) Relative preference (time exploring novel object relative to the total time exploring both objects) was not different between male and female subjects. One-way analyses of variance were used assess sex-differences in performance. *p < .05; **p < .01.

4. Discussion

While cognitive impairments have been well characterized in 3xTg mice [9], it remains unclear whether longitudinal training provides resiliency to such deficits, particularly in a task of executive function that translates well to humans. Therefore, the aim of the current study was to examine how daily exposure to a training variant (50:50) of the GNG task differentially impacted performance of 3xTg and wild-type mice in a testing variant (20:80) of the GNG task at early-adulthood (3 months old), middle-age (6 months old), and late-adulthood (9 months old). We found that all animals learned the training variant at comparable rates, and that further exposure to this cognitively-demanding task throughout adulthood leads to general improvements in training and testing performance. Neither genotype illustrated deficits in long-term recognition memory at 9 months, but female mice showed evidence of improved performance, suggesting that the benefits of cognitive training may not fully extend to memory capacities of male subjects. Although differences on select measures were detected across certain GNG testing days in secondary analyses, these results overall suggest that longitudinal cognitive training endows 3xTg mice with cognitive capacities that are on par with their wild-type counterparts.

Despite similar improvements in average testing performance with increasing age across genotypes, we also found that 3xTg subjects displayed slight deficits in inhibitory control that were accompanied by minor signal detection improvements at 3 months old (i.e., increased testing day 1 false alarms, increased testing day 2 hit percentages and sensitivity index scores, increased testing day 3 hit percentages), but by 9 months old, such animals showed signs of impaired signal detection (i.e., reduced hits and sensitivity index scores on testing days 1 and 3). This temporal pattern was particularly interesting because impulsivity and impaired inhibitory control during mid-life are among the earliest signs of late-life cognitive impairment [30], with GNG deficits having been frequently documented among individuals with MCI and early-stage dementia [31, 32]. Although hyperactivity to uncertainty can be an early feature of cogntive decline (e.g., subjective cogntive impairment [33]), the similar hit and false alarm latencies between genotypes suggested that these differences were not driven by hyperactivity or motor impulsivity stemming from uncertaintly in the environment (i.e., the change in the proportion of No-Go trials from 50% to 80%). Because bottom-up cognitive control mechanisms are typically engaged during the proactive phase of the GNG [34], variation in these cognitive processes during novel or uncertain conditions in 3xTg mice at an age when histopathological features of the AD pathology are not typically visible (i.e., 3 months old) may represent an early cognitive endophenotype of AD. Additionally, while these results suggest that exposure to daily cognitive training may not be able to completely compensate for the deleterious effects associated with AD gene expression, the current analyses also support the use of the GNG paradigm as a highly sensitive task that is capable of detecting subtle differences in cognitive performance in AD mouse models.

As evidenced by EPM performance, 3xTg mice exhibited a lack of inhibition/decreased anxiety at 3 months old that was attenuated by 6 and 9 months old. Because decreased anxiety is suggested to exert positive effects on inhibitory control processes [23], such EPM findings are consistent with the slight improvements in signal detection observed among 3xTg mice at 3-months old, but inconsistent with the minor inhibitory control deficits that accompanied these improvements. Although EPM findings in the 3xTg model have been conflicting [35–39], our results align with evidence of decreased EPM anxiety that can be detected as early as 3 months old [40], as well as other data suggesting that these patterns may not persist after 6 months old [41]. However, we did not find evidence of sex differences in the EPM, which have been previously reported in this model [42]. It is possible that extensive handling and habituation during training and testing could have moderated anxiety-like traits related to genotype and sex [43].

Long-term recognition memory, as measured via NOR, was comparable across genotypes in late adulthood (9 months old), indicating that the benefits of cognitive training may extend to multiple cognitive domains in a mouse model of AD. Our finding of sex-specific effects on NOR performance (at least in regard to absolute preference) adds to contradictory evidence of sex differences in this task. For example, in C57BL/6 mice, some evidence suggests that females show impaired object recognition following a 24 hr. retention interval, yet other data using the same testing parameters have failed to replicate such effects [44, 45]. In experiments that have reported sex differences in NOR performance, such effects are significantly dependent on the duration of retention intervals [46], indicating that our results may be influenced by the 24 hr. retention interval we applied. Although a limited sample size prevented us from reliably assessing a sex by genotype interaction on NOR measures, the improved memory capacities of females was unlikely driven by differences in genotype, as several studies have demonstrated that male and female 3xTg mice perform similarly in NOR [47, 48].

Our findings indicate that life-long cognitive training may functionally compensate for cognitive capacities in subjects harboring AD genetic mutations, given that 3xTg and wild-type mice showed similar improvements in GNG testing performance with increasing age, similar scores in averaged performance across testing days and comparable memory capacities in NOR. We speculate that several mechanisms may account for these results. Consistent with the cognitive reserve hypothesis, GNG training may have enabled 3xTg subjects to utilize pre-existing neural networks more efficiently, or to engage alternative cortical networks, in order to offset deficiencies otherwise associated with abnormal expression of AD-related genes [49]. For example, resiliency to cognitive deficits among APP/PS1 mice induced by daily cognitive training has been linked to increased levels of cell proliferation and immature neurons in the dentate gyrus, suggesting that augmented neurogenesis and plasticity may play a key role in mediating the effects of cognitive exposures on AD-related cognitive impairments [50]. As such compensatory capacities induced by cognitive training are related to reductions in Aβ and phosphorylated tau in the 3xTg mouse [13] but not in other AD-mouse models [51], it remains unclear whether the beneficial effects of longitudinal GNG training involve specific attenuation of AD-related neuropathology. Although we found evidence suggesting that higher hippocampal Aβ oligomer levels were associated with improved GNG testing performance at 9 months old among a subset of 3xTg mice, these findings should be interpreted with caution due to the limited sample size. The benefits of daily cognitive training in our study might also be attributed to blunted neuroinflammation, given that cognitive enrichment can lead to a reduction in pro-inflammatory cytokine expression (e.g., IL-1β, TNF-α) and the homeostatic maintenance of CNS immune cells (e.g., microglia, astrocytes) [52]. For instance, we previously found that cognitive engagement can ameliorate age-related increases in expression of innate immune and neuroinflammatory pathways [53]. Several other biological mechanisms implicated in AD and cognitive resilience may have also contributed to our findings, including variation in mitochondrial dynamics, ion regulation (especially Ca²⁺) and the generation of reactive oxygen/nitrogen species [54].

Although our study had several strengths, including the examination of a widely used AD mouse model, the implementation of a longitudinal study design and the use of a GNG task that translates well to humans, our results should be interpreted with caution due to several limitations. First, because of the financial burdens associated with longitudinal study designs, we were unable to maintain a group of sedentary controls or a group of subjects exposed to a less cognitively demanding paradigm, nor were we able to extend our study beyond 9 months. Therefore, despite the reported consequences of sedentary conditions on cognitive functioning in 3xTg mice [11–14], we cannot conclude that the effects observed in our study were caused solely by exposure to cognitive training or that such effects persist past 9 months of age (i.e., when greater levels of AD pathology are present). Although overall GNG performance (with the exception of minor deficits on certain testing days) was similar across genotypes (i.e., suggesting that the subjects in our study maintained similar capacities for visually discriminating response cue trials [Go trials] from withhold cue trials [No-Go trials]), we were also unable to comprehensively characterize the vision of wild type and 3xTg animals in our study. While reports of vision deficits in the 3xTg model are conflicting [55, 56], we cannot rule out the possibility that impaired visual capacities in these mice related to transgene introduction may have obscured our results. Second, our study lacked a thorough investigation of neurobiological mechanisms that may underlie the benefits of life-long cognitive training in 3xTg mice; future studies are warranted to delineate such mechanisms. Lastly, given the known limitations of AD mouse models [1], it remains unclear whether longitudinal GNG training in human subjects may offer beneficial effects in the context of AD and other forms of cognitive impairment. Despite these considerations, the current analyses suggest that longitudinal cognitive enrichment renders 3xTg mice with cognitive capacities that are on par with their wild-type counterparts, potentially reflecting functional compensation in subjects harboring AD genetic mutations.

Highlights:

• Longitudinal cognitive training improved performance of 3xTg and wild-type mice

• Long-term recognition memory at 9 months old was comparable between genotypes

• Cognitive training may compensate for, or prevent, cognitive deficits in 3xTg mice