B6.129S7-Chrna7^tm1Bay/J ¹ male and C57Bl/6 J ² female mice were purchased (The Jackson Laboratory, Bar Harbor, ME) and bred in house to produce heterozygous α7 nAChR offspring. These offspring were bred together to produce homozygous knockouts (KO), heterozygous partial knockouts (not used in this study), and wildtype (WT) offspring. The B6.129S7-Chrna7^tm1Bay/J strain is a result of deleting exons 8–10 of the Chrna7 gene (Orr-Urtreger et al., 1997). Genotypes were determined and validated from tail clip biopsies obtained at weaning (Transnetyx, Cordova, TN, United States). All animals were housed in same-sex sibling cohorts of at least three per cage. The first cohort of mice generated for stereology at 3 months old, an age where brain growth has reached a plateau, were allowed food and water ad libitum and were not handled or exposed to any behavioral task (Figure 1A). A second cohort of mice began diet restriction (85–90% initial mass) and daily handling 10 days prior to behavioral testing at 4 months old (Figure 1B). This age was chosen based on scheduling and being a comparable age to the stereology cohort. All procedures were approved and performed in compliance with the NIEHS/ NIH Humane Care and Use of Animals Protocols.

At 3 months old, the first cohort of mice were deeply anesthetized using sodium pentobarbital (FatalPlus; 100 mg/kg) and transcardially perfused using ice cold 0.1 M phosphate buffer saline (PBS; pH 7.4 with 0.1% heparin) followed by ice cold 4% paraformaldehyde (PFA) in PBS. The entire skull was left in 4% PFA for 48 h after perfusion to prevent the likelihood of dark neuron artifacts (Jortner, 2006). Dissected brains were then cryoprotected at 4°C in a 30% sucrose and 0.1 M PBS solution. Brains were then transferred to Charles River Laboratories, Inc. (Durham, NC, United States).

Trimmed brains were embedded in paraffin in a coronal/transverse orientation. After a random start, two 3 μm thick disector pairs, a section for hematoxylin and eosin staining, and a reserve section were collected at 270 μm intervals, which produced four slides per interval, with approximately 8–12 intervals per mouse. The disector pair from each interval was stained with Cresyl Violet. Stained disector pairs were scanned at 40X magnification using a Nanozoomer whole slide scanner (Hamamatsu Photonics K.K., JP) and imported into the Autodisector platform (Visiopharm, DK).

For granule cell counts, sections were aligned and regions of interest (ROI) were manually drawn around the dentate gyrus (Figure 2A). For pyramidal cell counting, sections were aligned and ROI were manually drawn around the three fields of the CA1-3 (Figure 3A). For each cell type, three independent samplings were performed using seven fields randomly selected from the 8–12 intervals with an unbiased counting frame of 3,000 μm² (60 μm × 50 μm). Sampled fields were loaded into the Stereology module of the Visiopharm software and the number of granule or pyramidal cells were manually counted with the help of an analysis algorithm for each animal. Cell types were clearly identified by a 5–20 μm diameter elliptical soma with a granular nucleus. Individual nuclei were identified within the unbiased counting frame by the presence of a nuclear membrane and an unobstructed view of a portion of the interior of the nucleus (Figures 2B,C, 3B,C). Cells were counted in both directions, i.e., if the nucleus in a stained cell was present in either the reference or the look-up section and not the other, that cell was counted (Figures 2D,E, 3D,E).

The total counts across all sampling fields for granule and pyramidal cells were summed by the software separately. The weighted counts were calculated by the software by multiplying the raw counts in each sampling field by the inverse of the probability of having a positive count in that field (Gardi et al., 2008). Data was imported into the Visiopharm software calculator. For each independent sampling, the total number of granule or pyramidal cells were calculated according to the equation N = ∑Q/2 × 1/ssf; where ∑Q was the sum of the weighted counts (divided by 2 as counting was performed in both directions of the dissector), and ssf was the section sampling fraction, which is the section thickness divided by the sampling interval (Gardi et al., 2008). The mean and standard deviation of the three samplings was calculated for each animal. The coefficient of error (CE) of the three independent samplings was also calculated for each animal according to the equation CE = STD/(mean × √2) (Gardi et al., 2008). To determine whether the precision of the estimate was adequate, the precision range of an optimally balanced estimator (PROBE) calculation was performed for each estimate (Løkkegaard, 2004). The average CE and coefficient of variation (CV) for each group were determined for each estimate and the following equation was performed: PROBE = CV2/CE2. A PROBE value of greater than 2 indicates that the biological variability, represented by the coefficient of variation, is sufficiently larger than the sampling error, represented by the coefficient of error, that any difference may confidently be attributed to the variability between animals. All estimates were below the a priori PROBE value of 2.

To calculate hippocampal volume, sections were aligned, and ROI were manually drawn around the CA1, CA2, CA3, dentate gyrus, and the subiculum of each section (Figures 4A,B). Three-dimensional tissue shrinkage during processing was estimated by the following formula: 3D global shrinkage = 1 −(post-processing weight/pre-processing weight) (Gundersen et al., 2013). The number of points hitting the hippocampus (tagged H) were summed by the software and imported into the calculator module for total volume estimation (Figure 4B). The volume of each sampling was calculated according to the Cavalieri method using the following equation: V = ΣP × A(p) × T; where ΣP was the number of intersecting points, A(p) was the area per point (0.269992 mm²), and T was the sectioning interval (Howard and Reed, 2005). The volume for each animal was corrected for shrinkage using the equation V_corrected = V_estimated/(1–3D shrinkage) (Gundersen et al., 2013). Noise was calculated as: Noise = 0.0724 * (b/√a) * √(n * ΣP); where n is the number of sections and b/√a is the average profile shape, where b is the perimeter, and a is the profile area. The profile shape was determined to be 4. Noise was combined with the estimator variance of Σ_area to produce a total variance, which was used to calculate a coefficient of error (CE) for each estimate: CE = √(total variance)/ΣP (Gundersen and Jensen, 1987).

Bussey-Saksida touchscreen operant chambers (Loughborough Campden Instruments, United Kingdom) housed within sound attenuating cubicles (MED-OFA-017, Med Associates Inc., Fairfax, VT, United States) were utilized to test pattern separation. This touchscreen task was generated to mimic the Cambridge Neuropsychological Test Automated Battery (CANTAB), a highly selective test for dementia in humans (Sahgal et al., 1991; Égerházi et al., 2007) in which the ability to discriminate choices has been found to be dependent on hippocampal neurogenesis (Clelland et al., 2009). Both tests require the subject to touch visual cues in a pattern to precisely test spatial memory combined with immediate reward. A 5-Choice Mask (4 × 4 cm each with 1 cm spacing) was utilized for both initial training and position discrimination tasks. The operant chambers (25 cm × 20 cm × 14 cm) contained a touch screen (25 cm × 15 cm) on one wall and a reward-port on the opposing wall. The reward stimulus (evaporated milk) was presented using a peristaltic pump. Training began with an “Initial Touch” phase where one of the five squares was illuminated (at random) for up to 10 s. A 21 μL reward was presented if the mouse touched the illuminated square during the 10 s. If no touch occurred, the illuminated square was turned off and a 7 μL reward pulse was presented. The next phase of training was a “Must Touch” similar to the previous phase except no reward was presented in the event the illuminated square was not touched. All mice were trained to a criterion of performance requiring greater than 85% correct performance over 2 consecutive days before moving to the next phase. After successful completion of the training, the mice were then tested for 10 days on the Large Separation task (Figure 1C). In this task, either the far right or far left illuminated squares, separated by three dark squares, were designated as correct. The contingency reversed after seven correct touches in eight attempts. The mice were then given 10 days of the Small Separation task (Figure 1D) where either the left or right illuminated squares, separated by a single dark square, were designated as correct. Like the Large Separation task, the contingency reversed after seven correct touches in eight attempts. For both tasks, the next trial started 2 s after reward was consumed (when the correct poke was made) and 10 s timeout was imposed when an incorrect poke was made (Supplementary Video 1).

For stereology data, two-way ANOVAs with genotype and sex as factors were employed to determine differences in hippocampal cell counts and volume with α set at 0.05. For the touchscreen data, repeated measures ANOVAs with genotype and sex as the primary factors to determine differences in the average percentage of correct attempts were employed, with α set at 0.05. Tukey post hoc tests were used in the case of significant effects. After initial data analysis, additional analyses were performed to determine differences in motivational variables: reward latency and selection latency with α set at 0.05. To characterize behavioral strategies, a generalized linear model (GLM) logit link function (logistic regression) adapted from Chen et al. (2021) was used to predict left vs. right choice based on which side was chosen previously and whether it was rewarded based on the following equation (Chen et al., 2021). Genotypes were blinded to experimenters throughout the experimental phase but not during analyses.

L t − 1 = 1 and R t − 1 = 0 if the left side was chosen at the previous trial.

L t − 1 = 0 and R t − 1 = 1 if the right side was chosen at the previous trial.

O t − 1 = 1 if the previous trial’s side choice was correct, otherwise O t − 1 = 0 .

β₀ to represent an overall side bias (left versus right).

β₁ to represent the tendency to stay or switch sides regardless of the previous reward outcome.

β₂ to represent the tendency to stay or switch sides based on the previous reward outcome.

Trials were concatenated across sessions and a single GLM was fit to each animal’s cumulative set of trials. The final beta coefficient estimates values incorporating all trials were analyzed utilizing two-way ANOVA to assess overall strategy differences as a function of sex and genotype. R (Version 4.2.2) was used to fit the models. GraphPad Prism (Version 8.4.1, Graphpad Software Inc., La Jolla, CA, United States) was used to perform remaining statistics and generate figures. Individual data and statistics are displayed in Supplementary Table 1.

To determine if hippocampal structure is affected by the absence of α7 nAChRs, we performed unbiased stereology to quantify hippocampal granule cells (Figure 2), pyramidal cells (Figure 3), and total hippocampal volume (Figure 4) in female and male α7 nAChR KO and WT mice. For granule cells (Figure 2F), females had on average 17% less total cells than males [F(1, 33) = 5.534, p = 0.0248], KOs had on average 22% less than WTs [F(1, 33) = 10.76, p = 0.0025], and an interaction was not significant [F(1, 33) = 0.005702, p = 0.9403]. For pyramidal cells (Figure 3F), females had on average 18% less total cells than males [F(1, 33) = 11.12, p = 0.0021], KOs had on average 20% less than WTs [F(1, 33) = 14.24, p = 0.0006], and an interaction was not significant [F(1, 33) = 0.7789, p = 0.3839]. For hippocampal volume (Figure 4C), there was no significant effect of sex [F(1, 33) = 0.09706, p = 0.7574], genotype [F(1, 33) = 0.6165, p = 0.4380], or an interaction [F(1, 33) = 0.02050, p = 0.8870]. For body mass, females were 23% lighter than males [F(1, 34) = 62.30, p < 0.0001] and there was no effect of genotype [F(1, 34) = 1.408, p = 0.2436] or an interaction [F(1, 34) = 0.4750, p = 0.4954].

To determine if the global KO of α7 nAChRs impairs hippocampal function, we utilized a touchscreen task that was generated to mimic the Cambridge Neuropsychological Test Automated Battery (CANTAB), a highly selective test for dementia in humans (Sahgal et al., 1991; Égerházi et al., 2007). Both tests require the subject to touch visual cues in a pattern which precisely tests spatial memory combined with immediate reward. In mice, the small pattern separation task, and not the large, has been found to require juvenile and adult neurogenesis within the dentate gyrus (Clelland et al., 2009). α7 nAChR KO and WT mice were tested on large and small pattern separation tasks at 4 months old. During the training phase, 50% of mice reached competence (at least 85% correct touches) by the fourth session, 82% by the fifth session, 95% by the sixth session, and 100% by the seventhe. As expected, there were no effects of genotype in the large separation task for females [F(1, 16) = 1.192, p = 0.2912; Figure 5A] or males [F(1, 18) = 0.1809, p = 0.6757; Figure 5B]. There were no effects of genotype in the small separation task for female mice [F(1, 16) = 0.03355, p = 0.8570; Figure 5D]. However, KO male mice performed on average 7% worse than WT male mice in the small separation task [F(1, 18) = 6.047, p = 0.0243; Figure 5C], indicating a male-specific behavioral impairment.

To rule out differences in motivation, we calculated the average reward latency (how long it takes the mice to retrieve the reward after making a correct choice) and the average selection latency (how long it takes to make a response at the beginning of a trial) across trials for the small separation task. We found no differences in reward latency [Genotype: F(1, 34) = 0.0009959, p = 0.9750, Sex: F(1, 34) = 1.408, p = 0.2436; Interaction: F(1, 34) = 0.4567, p = 0.5037] or choice latency [Genotype: F(1, 34) = 0.8712, p = 0.3572, Sex: F(1, 34) = 0.5026, p = 0.4832; Interaction: F(1, 34) = 0.8008, p = 0.3771] indicating that motivational differences cannot explain the sex-dependent deficit in the KO mice.

To explore differences in performance strategies and to gain insight into the lack of a pattern separation deficit in female KOs, we performed ad hoc analyses on the strategy of groups using raw trial data from the small separation trials with genotype and sex as main factors. We focused on three different approaches: (1) comparing performance before and after the first reversal in order to analyze whether the performance differs between the initial reversal contingency and subsequent performance (Swan et al., 2014), (2) fitting a GLM on the behavioral strategy to determine side bias, tendency to stay or shift sides, and a tendency to stay or shift sides based on the outcome of the previous choice (Chen et al., 2021), and (3) categorizing the primary strategies used by each group.

Successful performance of this task requires the animal to alternate their response to the right or left based on the current contingency. In each session the left side was always designated as initially correct. To determine how this initial contingency shift affects subsequent performance, we analyzed percent correct during the initial reversal contingency versus all subsequent contingencies. This analysis (Figure 6A) showed males performed worse after the first reversal compared to the females [F(1, 34) = 6.903, p = 0.0128] with no effect of genotype [F(1, 34) = 1.353, p = 0.2528], and no interaction [F(1, 34) = 3.309, p = 0.0777]. This intriguing finding argues that males are using a unique behavioral strategy affected by the initial reversal contingency, which may be exacerbated by the KO.

We next modeled the strategies using a GLM with three components: β₀ to represent an overall side bias (left versus right), β₁ to represent the tendency to stay or switch sides regardless of the previous reward outcome, and β₂ to represent the tendency to stay or switch sides based on the previous reward outcome (Supplementary Figures 1–4). The β₀ coefficients (Figure 6B) were small, but significantly greater than zero for WT females (p = 0.0333) and knockout males (p = 0.0093) indicating a slight left-side bias for these groups. This small effect is likely due to the first trial always being on the left side. Between groups, there was no effect of genotype [F(1, 34) = 0.003864, p = 0.9508] or sex [F(1, 34) = 0.001513, p = 0.9692], but a significant interaction was present [F(1, 34) = 8.514, p = 0.0062]. However, no Tukey post-hoc comparison was significant for the interaction. The β₁ coefficient (Figure 6C) was not significantly different than zero for any group and did not have a significant effect of genotype [F(1, 34) = 0.5405, p = 0.4673], sex [F(1, 34) = 0.4826, p = 0.4920], or an interaction [F(1, 34) = 0.1178, p = 0.7336] indicating no group had an overall tendency to shift or stay regardless of the previous trial. Analysis of the β₂ coefficient (Figure 6D) was significantly greater than zero for each group (p < 0.0001 for each comparison) indicating a dominant strategy to stay when previously correct, and to shift when previously incorrect. Between groups, there was no significant effect of genotype [F(1, 34) = 0.5405, p = 0.4673], sex [F(1, 34) = 0.4826, p = 0.4920], or an interaction [F(1, 34) = 0.1178, p = 0.7336].

We next conducted a trial-by-trial analysis on the small separation task to categorize each trial as Win→Stay, Win→Shift, Lose→Stay, or Lose→Shift, based on the current trial’s choice and previous trial’s outcome (Figures 6E–H). The number of trials for each strategy was then divided by the total trials to determine a proportion. Matching the GLM results, Win→Stay was the dominant strategy (44.5%) followed by Lose→Shift and Lose→Stay (both at 19.4%) and finally Win→Shift (15.6%). Males were 1.2% more likely to Lose→Shift [F(1, 34) = 7.401, p = 0.0102] and 1.7% more likely to Win→Shift [F(1, 34) = 4.788, p = 0.0356] than females. There were no additional effects of genotype, sex, or interactions for any other comparison.

Together, the additional three analysis methods indicate males were more biased by the initial reversal contingency relative to females. While the categorization of trial strategies indicates males had a slightly greater tendency to shift sides regardless of the previous trial’s outcome, the GLM results suggest no difference. In conjunction, this phenomenon is unlikely to be an important contributing factor to differences in performance between sexes. Overall, the difficulty performing the first reversal is the strongest lead in understanding the specific behavioral deficit from the global α7 nAChR KO in males.