Fischer 344 × Brown Norway F1 (FBN) Hybrid male and female rats from the National Institute on Aging colony at Charles River were used in this study, as these rats are a robust model of non-pathological aging (McQuail and Nicolle, 2015; Hernandez et al., 2020c). A total of 36 total young (4–7 months) and aged (20–23 months) rats were split across control (n = 6 young male, n = 9 aged male, n = 1 young female and n = 2 aged female) and ketogenic (n = 8 young male, n = 8 aged male, n = 1 young female, and n = 1 aged female) diet groups. Additional females were unavailable for inclusion in this study, which precluded the analysis of any potential sex differences. The data from the female rats, however, did not vary from the means of the male rats and were thus included in the analysis. Moreover, similar results were obtained if the female rats were excluded. Because sex alone is not an adequate reason to exclude animals from a sample, the female rats were included. All rats were housed individually and maintained on a 12-h reversed light/dark cycle with all behavioral testing occurring in the dark phase. Note that reported ages correspond to the age of the rats upon arrival at the University of Florida vivarium. All rats were on the diet for 3 months prior to starting shaping on behavioral tasks, which took approximately 3 months. Thus, at the time of tissue collection young rats were 10–13 months and aged rats were 26–29 months. Mortality rates for this strain of rat begin to increase at 30 months (Fernandes et al., 1997). Therefore, to minimize age-related attrition we wanted to complete 3 months of the diet, plus the several months of training prior to 30 months of age. Additionally, it has been reported that Fischer 344 × Brown Norway hybrid rats start to show physical impairments at 28 months of age (McQuail and Nicolle, 2015). Because the WM/BAT task required rats to ambulate around a track, it was critical to begin cognitive testing prior to the emergence of sensorimotor impairments.

All rats were fed either a high-fat, low-carbohydrate KD (75.85% fat, 20.12% protein, 3.85% carbohydrate; Lab Supply; 5722, Fort Worth, TX, USA) or a calorically and micronutrient comparable control diet (CD; 16.35% fat, 18.76% protein, 64.89% carbohydrate; Lab Supply; 1810727, Fort Worth, TX, USA). The specific details on composition of the diets have been reported previously (Hernandez et al., 2018a,b). The primary fat source in the KD was a >95% C8 medium chain triglyceride (MCT; Neobee 895, Stephan, Northfield, IL, USA). Rats were weighed daily and given ∼51 kCal for males and ∼43 kCal for females at the same time each day for the first 12 weeks of the diet. Following a 12-week adjustment period to the diet and confirmation of sustained nutritional ketosis, the amount of food was restricted to reach an ∼15% reduction in body weight to motivate participation during behavioral testing. Access to water was available ad libitum for all animals.

All behavioral data for shaping, continuous alternations, object discrimination testing, and the working memory/bi-conditional association task for the animals in which neurobiology data are reported in the current study were documented in a previous publication (Hernandez et al., 2018a). Briefly, all behavioral testing occurred on a Figure 8-shaped maze (see Figure 1A) that was 67.5 inches long and 25 inches wide. The maze was constructed from wood and sealed with waterproof white paint. The center arm was made of clear acrylic to monitor gait (data not reported). The choice platforms each contained two food wells (2.5 cm in diameter) that were recessed into the maze floor by 1 cm. All arms were 4 inches wide. The choice platforms were contained within 7.5 cm raised walls and the right arm was contained within 20 cm high raised halls, but the center and left arms did not have walls.

During shaping, rats were placed on the maze for 10 min per day for two consecutive days to habituate to the maze. Food rewards (macadamia nuts, Mauna Loa, Keaau, Hawaii) were scattered throughout the maze to encourage exploration. Macadamia nuts, used for rats on both diets, were chosen based on their palatability and macronutrient composition (80% fat, 13.33% carbohydrates, and 6.67% protein), which would not inhibit ketosis. Importantly, the macadamia nuts did not interfere with the consumption of the control diet nor lead to weight gain, nor did they induce ketosis in the CD group. Rats were then trained to continuously alternate between the two outside arms of the maze in a Figure 8 motion, returning through the center between trials until they correctly alternated on at least 16 of 20 trials on two consecutive days.

Following alternation training, rats were trained to perform simple object discriminations both with and without alternations. These tasks did not have the bi-conditional object-in-place rule and were used to train the animals in the procedural aspects of moving an object to obtain a reward. Rats performed 20 trials/day of this task until a criterion of at least 80% correct was reached on two consecutive days. Rats were then tested on the cognitive dual task which combined spatial alternations with a biconditional association task (WM/BAT) that required the animal to continue to alternate, while also associating a particular object of a pair with the different testing platforms (see Figure 1A). Specifically, rats were presented with the same two objects in both arms with each object being the correct choice in opposing arms. This task requires working memory to complete the continuous alternation while also imposing an interrupting bi-conditional association task between alternations to increase the cognitive load. Should a rat mistakenly enter the wrong arm (for example, entering into the right arm following a right arm trial), the trial was recorded as a working memory error and rats were not presented with the object discrimination problem. All rats completed 20 trials/day until they reached a criterion performance of better than 80% correct on two consecutive days and then a final shorter session was run, and tissue was collected. Rats were trained to a common criterion to ensure that differences in the expression of Arc and/or Homer1a were due to age and/or diet rather than differences in performance or the number of rewards obtained.

On the final day of behavior, the testing session was comprized of two behavioral epochs (approximately 5 min each in duration), counterbalanced across rats. Rats completed yoked trials so that young and aged rats performed a comparable number of trials. To do so, an aged rat performed as many trials as possible within 5 min. The next young rat completed the same number of trials, or 5 min total, whichever came first. Following a 20-min rest in their home cage, rats performed the second epoch of behavior. Rats were then immediately placed into a bell jar containing isoflurane-saturated cotton (Abbott Laboratories, Chicago, IL, USA) separated from the animal by a wire mesh shield. Animals rapidly lost the righting reflex (<30 s), after which they were immediately euthanized via rapid decapitation. Tissue was extracted and flash frozen in 2-methylbutane (Acros Organics, NJ, USA) chilled in a bath of dry ice with 100% ethanol (∼−70°C). Tissue was stored at −80°C until cryosectioning. Prior to cryosectioning, one hemisphere from each experimental group was blocked so that tissue from different groups would be sliced together and so each slide would contain all experimental groups to control for potential variability is staining across slides. Sectioning was performed at 20 μm on a cryostat (Microm HM550) and thaw mounted on Superfrost Plus slides (Fisher Scientific). Sliced tissue was stored at −80°C until in situ hybridization.

In situ hybridization was performed to label Arc and Homer1a mRNA for catFISH (Vazdarjanova et al., 2002; Marrone et al., 2008) using commercial transcription kits and RNA labeling mix (REF #: 11277073910, Lot #: 26591021 and REF #11426346910:, Lot # 24158720; Roche Applied Science) to generate digoxigenin-labeled and fluorescein-labeled riboprobes using a plasmid template containing cDNA. Tissue was incubated in the generated probes overnight and labeled with an anti–digoxigenin-HRP conjugate or anti-fluorescein- (Ref#: 11207733910, Lot # 43500600 and REF#11426346910; LOT#45220020; Roche Applied Science). Fluorescein (Fluorescein Direct FISH; PerkinElmer, REF# 14921915, Lot #: 2587974) and Cy3 (TSA Cyanine 3 Tyramide; PerkinElmer, REF#: 10197077, Lot #: 2629833) were used to visualize labeled cells, and nuclei were counterstained with DAPI (Thermo Scientific).

Fluorescence microscopy on a Keyence BZX-810 digital microscope (Keyence Corporation of America, Itasca, IL) was utilized to obtain z-stacks at 1 μm increments from 3 sections per region of interest for each rat. A total of 4 regions were imaged: deep and superficial prelimbic cortex (PL) and distal (CA3b) and proximal CA3 (CA3c), for a total of 12 images per rat (see Figure 1B). Because of the potential for conflating CA3a for CA2, this subregion of CA3 was not analyzed. Deep and superficial layers of PL were analyzed separately due to their distinct patterns of connectivity. Specifically, superficial PL layers are more reciprocally connected to other cortical areas, such as the perirhinal cortex (Burwell, 2000; Furtak et al., 2007; Agster and Burwell, 2009), while deep layers of PL are more connected to subcortical structures such as the nucleus reuniens (Vertes, 2002; Vertes et al., 2006; Jayachandran et al., 2019). A custom ImageJ plugin (available upon request) was utilized to segment nuclei and to classify cells as having no mRNA present, being positive for Arc, Homer1, or both, as well as the subcellular location of the mRNA. All experimenters analyzing cellular imaging data were blind to age, diet, and behavioral task order (WM/BAT vs. alternation first). As in previous studies (e.g., Maurer et al., 2017; Hernandez et al., 2018c), only fully visible cells within the median 20% of the optical planes were considered.

Figure 1C shows examples of all possible cell classifications and the temporal activity profile inferred from the cellular location of Arc and Homer1a mRNA. To summarize, cells were classified as one of the following: (1) negative for any staining, indicating no activity 60 min prior to sacrifice, (2) Arc cytoplasm + Homer1a foci, indicating activity during epoch 1 only, (3) Arc foci, indicating activity during epoch 2 only, (4) Arc cytoplasm + Homer1a foci + Arc foci, indicating activity during both epochs, (5) Homer1a cytoplasm, indicating home cage activity only, (6) Homer1a cytoplasm + Arc cytoplasm + Homer1a foci, indicating home cage and epoch 1 activity, (7) Homer1a cytoplasm + Arc foci, indicating home cage and epoch 2 activity, and (8) Homer1a cytoplasm + Arc cytoplasm + Homer1 foci + Arc foci, indicating activity in the home cage and during both epochs.

To control for differences in activity between regions similarity scores (Vazdarjanova and Guzowski, 2004) were calculated across the two behavioral epochs as follows: B-(E⁢1×E⁢2)(E⁢1+E⁢2)2-(E⁢1×E⁢2) where B is the proportion of cells active during both epochs of behavior and E1 and E2 are the proportion of cells active during the first or second behavioral epoch, respectively. If identical populations of cells were active during the two behavioral epochs, the subject would receive a value of 1. Conversely, a value of 0 indicates two entirely statistically independent populations of cells were activated during the two behavioral epochs. These values allow for the comparison across brain regions with different neural activity patterns by reducing each region to a single value, rather than information from multiple epochs of behavior.

Differences in neuronal activation during the behavioral epochs in relation to age, cortical layers and diet group were statistically evaluated by calculating the mean percentage of Arc positive cells per rat as done previously (Hartzell et al., 2013; Hernandez et al., 2018c,2020b) to avoid inflating statistical power by including multiple measures from one animal and ensure data do not violate the assumption of independent observations (Aarts et al., 2014).

Potential effects of age, layers/subregion and task were examined using repeated measures ANOVAs (ANOVA-RM) with the within-subject factors of behavioral task, cortical layers, pre-active status, and the between-subjects factors of age and diet. All analyses were performed with the Statistical Package for the Social Sciences v25 (IBM, Armonk, NY, USA) or GraphPad Prism version 9.1.0 for Windows (GraphPad Software, San Diego, CA, USA).¹ Statistical significance was considered at p-values less than 0.05.

Figure 2 shows the percent of neurons that were positive for Homer1a in the cytoplasm, indicating activation prior to behavior during baseline for PL (Figures 2A, B) and CA3 (Figures 2C, D). The proportion of cells active at baseline was significantly different between the PL cortex and CA3 subregions [F_(3,96) = 44.15; p < 0.001], with more baseline activity within both subregions of CA3 compared to PL. This was not significantly affected by age [F_(1,32) = 1.43; p = 0.24] nor diet [F_(1,32) = 1.79; p = 0.19], and none of these factors significantly interacted (p > 0.24 for all comparisons).

Within the PL, there was significantly more baseline activity within the superficial layers than there was in the deep layers [F_(1,32) = 7.87; p = 0.008]. This did not significantly vary by age [F_(1,32) = 0.59; p = 0.45] or diet group [F_(1,32) = 1.09; p = 0.31] (Figures 2A, B). Furthermore, there were no significant interaction effects between age, diet, and cortical layers (p > 0.18 for all comparisons). Within the deep layers, however, there were significantly fewer neurons that had Homer1a in the cytoplasm in aged compared to young rats [F_(1,32) = 5.72; p = 0.02]. There was not a significant effect of age on baseline activity within the superficial layers [F_(1,32) = 0.03; p = 0.87].

The effect of subregion within CA3 on baseline activity did not reach statistical significance, though there was a trend toward greater proportion of active cells within proximal CA3 relative to distal CA3 [F_(1,32) = 3.68; p = 0.06; Figures 2B–D]. There were no differences across age [F_(1,32) = 2.79; p = 0.10] or diet [F_(1,32) = 1.15; p = 0.29] groups. Furthermore, age and diet did not significantly interact with each other [F_(1,32) = 0.23; p = 0.63], nor with cortical layer (p > 0.72 for both comparisons).

Blood was collected during sacrifice to measure peripheral β-hydroxybutyrate (BHB; a major circulating ketone body) immediately following behavioral testing. Rats fed a ketogenic diet had significantly more BHB than control-fed rats [F_(1,34) = 17.08; p = 0.0002; Figure 3A]. There was no significant effect of age [F_(1,34) = 0.16; p = 0.69] nor did age significantly interact with diet [F_(1,34) = 005; p = 0.95], suggesting that young and aged rats had reached similar levels of ketosis.

As reported in a previous study (Hernandez et al., 2018a), both young and aged rats show improved performance on a spatial alternation and the working memory/bi-conditional association task (WM/BAT) when on a long-term ketogenic diet compared to animals on a calorie-matched control diet. On the final day of behavioral testing day, in which IEG expression was quantified, all rats performed two 5 min epochs of behavior separated by 20 min. In counterbalanced order, rats performed the WM/BAT task and continuous spatial alternations on the Figure 8 maze. While there were initial age-related differences and performance improvements related to the ketogenic diet (Hernandez et al., 2018a), all rats were previously trained to a criterion performance. Thus, on the final day of testing, there were no significant differences across age group [F_{(1, 32)} = 1.17; p = 0.29] or diet condition [F_{(1, 32)} = 0.04; p = 0.85] on WM/BAT performance accuracy, nor did these factors significantly interact [F_{(1, 32)} = 0.19; p = 0.67; Figure 3B]. Similarly, performance accuracy for continuously alternating on the maze did not significantly differ between age [F_{(1, 32)} = 0.24; p = 0.63] or diet [F_{(1, 32)} = 0.18; p = 0.68] groups. Moreover, there was not a significant interaction between age and diet on alternation performance [F_{(1, 32)} = 0.86; p = 0.36; Figure 3C].

Figure 4A shows representative images of Arc and Homer1a expression in superficial layers of PL for aged rats on a control (left panel) or ketogenic diet (right panel). The percent of cells active during the two behavioral epochs did not significantly differ between WM/BAT and the continuous alternation tasks within either the deep [F_(1,32) = 0.59; p = 0.45; Figures 4B, C] or superficial [F_(1,32) = 1.81; p = 0.19; Figures 4D, E] layers of PL. Both age [F_(1,32) = 13.39; p = 0.001] and the ketogenic diet [F_(1,32) = 17.20; p < 0.001] significantly increased activity within the superficial layers, but neither age [F_(1,32) = 0.52; p = 0.48] nor diet [F_(1,32) = 1.70; p = 0.20] significantly altered activity within the deep layers. However, post-hoc analyses indicated that there was a significantly greater degree of activation in aged KD-fed rats than young KD-fed rats [t₍₃₂₎ = 3.91; p = 0.003], but this was not true for young vs. aged control-fed rats [t₍₃₂₎ = 1.31; p = 0.73]. There were no significant interactions between task, age, or diet for the deep or superficial layers of PL (p > 0.14 for all comparisons). The observation that a KD affected activation patterns in superficial but not deep cortical layers of PL, suggests that neurons that project to other cortical structures are more likely to be affected by interventions that elevate ketone bodies in the brain than are neurons that project to subcortical areas.

Figure 5A shows representative images of Arc and Homer1a expression in distal CA3 for aged rats on a control (left panel) or ketogenic diet (right panel). In contrast to the PL, the percent of cells active during the two behavioral epochs was significantly higher during WM/BAT compared to continuous alternations within both distal [F_(1,32) = 39.28; p < 0.001; Figures 5B, C] and proximal [F_(1,32) = 28.72; p < 0.001; Figures 5D, E] subregions of CA3. There was a trend for aged rats to have a higher proportion of active neurons during behavior compared to the young rats [F_(1,64) = 2.71, p = 0.1]. However, there was no significant effect of diet within distal or proximal CA3 during either task (p ≥ 0.18 for all comparisons). These observations suggest that more CA3 neurons are active when the cognitive load of the task increases and that aged rats tended to have higher levels of CA3 activity regardless of diet condition.

A similarity score was calculated for each layer of the prelimbic cortex separately for each rat as done previously (e.g., Vazdarjanova and Guzowski, 2004; Burke et al., 2005; Hernandez et al., 2018c). There was no difference in similarity score across the superficial and deep layers of PL [F_(1,32) = 0.93; p = 0.34; Figures 6A, B]. There was, however, a significant decrease in similarity scores in aged rats relative to the young group [F_(1,32) = 7.19; p = 0.01]. Diet did not significantly impact similarity score [F_(1,32) = 0.02; p = 0.90], and there were no significant interactions effects between diet, age, and PL layers (p > 0.11 for all comparisons). These data replicate a previous observation that aged rats have reduced population overlap across two different tasks in the same spatial context compared to young animals (Hernandez et al., 2018c). Because PL neurons has been observed to fire in association with common features across different episodes (Morrissey et al., 2017), these data suggest that PL neural ensembles in older animals are less able to bridge common elements across distinct but overlapping episodes and this is not reversed by diet. In CA3, there was no difference in similarity score between distal and proximal subregions [F_(1,32) = 2.30; p = 0.14; Figures 6C, D]. Furthermore, the similarity score for CA3 was not significantly affected by age [F_(1,32) = 0.22; p = 0.65] or diet [F_(1,32) = 0.46; p = 0.50]. Finally, there were no significant interactions between subregion, age, or diet on similarity score (p > 0.26 for all comparisons).

Previous studies have reported that neuronal populations with higher firing rates are more likely to be active across different behavioral states and contexts than cells with lower firing rates (Mizuseki and Buzsáki, 2013; Buzsáki and Mizuseki, 2014; Grosmark and Buzsaki, 2016; Witharana et al., 2016). This is hypothesized to reflect a skewed or lognormal excitability distribution of neural activity (Buzsáki and Mizuseki, 2014; Grosmark and Buzsaki, 2016). To examine the extent that age and diet may alter the skewed excitability distribution of brain organization, the behavior-related activity profiles of neurons that were active or inactive during baseline in the home cage were compared. In other words, the proportion of cells active at baseline (that is, “pre-active” cells) that were also active during one or more behavioral epochs were compared to the proportion of cells inactive at baseline (that is, “pre-inactive” cells) that were active during one or more behavioral epochs. Within the PL, pre-active cells were significantly more likely to have activity during behavior than pre-inactive cells [F_(1,52) = 123.72; p < 0.001; Figures 7A, B] and this behavioral-related activity bias of pre-active cells did not significantly interact with the cortical layers [F_(1,52) = 1.62; p = 0.21], age groups [F_(1,52) = 0.004; p = 0.95], or diet conditions [F_(1,52) = 0.20; p = 0.66]. This observation suggests that both the deep and superficial layers of the PL follow a skewed excitability distribution that is not altered by age or diet. Finally, there was a significant interaction between diet and PL layer [F_(1,52) = 5.82; p < 0.02], such that the ketogenic diet was associated with greater activation in the superficial [F_(1,28) = 4.63; p < 0.05], but not deep cortical [F_(1,24) = 1.60; p = 0.22] layers of PL. The observation that the ketogenic diet increases activity in superficial but not deep layers of the PL is consistent with what was observed when baseline activity was not accounted for (Figure 4).

Within distal and proximal CA3, pre-active cells were significantly more likely to be active during behavior than pre-inactive cells [F_(1,64) = 360.59; p < 0.001; Figures 7C, D]. The increased probability for pre-active cells to also show activity during behavior compared to neurons without baseline activity significantly differed between distal and proximal CA3 [F_(1,64) = 9.27; p < 0.003], with proximal CA3 having a larger difference in behavioral-related activity between baseline active and inactive cells than distal CA3. Different patterns of anatomical input to distal vs. proximal CA3 could contribute to this observed difference in activity patterns (Witter, 2007; Lee et al., 2015, 2021, 2022). For example, distal CA3 receives more direct input from the entorhinal cortex while proximal CA3 receives mossy fiber input from both the supra- and infrapyramidal blades of the dentate gyrus (Witter, 2007). The increased probability for pre-active cells to also show activity during behavior compared to neurons without baseline activity also significantly differed between age groups [F_(1,64) = 5.75; p < 0.02], with older rats having a greater difference between cells that showed activity prior to behavior vs. those cells that were quiescent in the home cage. In other words, aged CA3 pyramidal cells that had activity in the home cage before behavior (i.e., pre-active) were more likely to fire during behavior than were young pre-active CA3 cells. This observation is consistent with previous reports of elevated activity within aged CA3 neurons relative to young animals (Wilson et al., 2005; Thomé et al., 2016; Maurer et al., 2017; Lee et al., 2021). The behavioral-related activity bias of pre-active neurons did not significantly interact with diet [F_(1,64) = 0.60; p = 0.42]. This suggests that aged CA3 neurons have activity dynamics that are more likely to not change across different behavioral states compared to young CA3 neurons, which is consistent with what has been observed from electrophysiological recordings (Wilson et al., 2005, 2006; Lee et al., 2022). Importantly, diet condition did not have any significant effects on the pre-active vs. pre-inactive bias, nor did it interact with age or CA3 subregion (p > 0.22 for all comparisons). This observation is consistent with a previous finding that the ketogenic diet did not alter the expression of genes that were related to synaptic transmission within CA3 (Hernandez et al., 2019b).

Simple linear regression was used to test whether cellular activity before or during behavioral performance correlates with task acquisition across diet groups. Activity within the superficial prelimbic cortex was investigated, as this is the region in which differences were observed across diet groups. Behavioral performance on days 12–15 of training was used to investigate this potential relationship, as this was the time frame in which rats on a ketogenic diet consistently outperformed rats on the control diet (Hernandez et al., 2018a).

For control-fed rats, task performance significantly correlated with the percent of pre-active cells within the superficial PL [R² = 0.42; F_(1,16) = 11.64; p < 0.01 when adjusted for multiple comparisons] (Figure 8A). However, task performance did not significantly correlate with the percent of pre-active cells within the superficial [R² = 0.03; F_(1,16) = 0.53; p = 0.48] layers of the PL in the ketogenic diet-fed group. Together these data indicate that higher levels of baseline activity in the superficial PL are associated with better cognitive performance under normal conditions. However, with elevated baseline PL activity following a ketogenic diet this relationship is no longer detectable, as all animals show higher activation and better cognitive performance. Cellular activity during task performance, for both WM/BAT and spatial alternations, did not significantly correlate with task performance during this time (p > 0.50 for both comparisons; Figures 8B, C).