Forty-three naked mole-rats (H. glaber; Rüppell, 1842) (25 female, 18 male) encompassing a range of body masses and ages were used in this study. For a preliminary study using a non-continuous maze design, eight animals were used (Table 1). Following improvements to the maze design and incorporation of a continuous maze, thirty-five animals were used. Six of these animals were excluded due to missing data for their dominance ranking. Eleven additional animals had no more than one missing trial data point. For these eleven animals, the missing data point was replaced with linear interpolation. One additional animal was excluded as an outlier due to data points in the first learning trial exceeding ten standard deviations of the mean, while another was excluded due to passing away before the end of data collection. Thus, 27 animals were included in the continuous maze study (Table 1). For the memory probe analysis, two additional animals were excluded, one as an outlier with values that exceed 10 standard deviations of the mean and the other excluded for missing multiple data points. Thus, 25 animals of the initial 27 were included in the memory probe analysis. Body masses of the animals included in the analysis ranged from 23.0 to 101.0 g (Braintree Scientific compact portable scale model CB 1,001 with 0.1 g precision), with ages ranging from 0.39 to 6.68 years at the start of testing. All of the naked mole-rats were members of one of two laboratory breeding colonies (Colony A and Colony B) maintained in-house, with varying status within the colony (queens, breeders, non-breeders). Not all members of each colony were used. Animals were excluded based on involvement in other ongoing studies. The vivarium housing for each colony was maintained at approximately 27.8–30°C and at least 40% relative humidity, without a light/dark cycle due to the subterranean habitat of the animals (Sherman et al., 1991). Animal housing consisted of a series of chambers interconnected with tubing and designed to mimic their natural burrow systems. Free access to food and ample bedding were provided (for further housing details, see Artwohl et al., 2002). All aspects of the research performed for this study complied with our protocol approved by the Institutional Animal Care and Use Committee at Southern Illinois University, Carbondale, IL, United States, and were in accordance with the National Institutes of Health Guide for the Care and Use of Laboratory Animals (NIH Publication Nos. 8023 and 1978).

We designed modified versions of the Hebb-Williams maze, a common paradigm for testing spatial learning and memory in rodents. Modifications to the Hebb-Williams maze were based on studies that used similar tests of spatial navigation in other atypical rodent animal models, including the subterranean tuco-tuco (Antinuchi and Schleich, 2003; Schleich, 2010; Brachetta et al., 2014) and the blind mole-rat (Kimchi and Terkel, 2001, 2004). The mazes were constructed using PVC pipes that were cut in half longitudinally and covered with a sheet of clear acrylic, creating a trough for the animals to navigate through while allowing visibility for video recordings. The tubing had a 2” diameter, equal to the tubing in the housing chamber tunnel systems for the naked mole-rats. This tubing width was chosen based on previous research that showed greater learning and memory performance in the blind mole-rat when using narrower pathways that provide tactile cues in mazes (Kimchi and Terkel, 2004).

The first maze design we tested was non-continuous and included dead-ends (Figure 1A) similar to other Hebb-Williams-based mazes (Kimchi and Terkel, 2001, 2004; Costanzo et al., 2009; du Toit et al., 2012). Preliminary testing using a cohort of eight animals was conducted for this version of the maze (Table 1). Any dead-ends within the maze were sealed and a clear acrylic sheet was placed on top of the maze, allowing visualization of animal behavior and performance in the mazes from above. The reward chamber consisted of a cylindrical housing chamber made of clear cast acrylic with 8” diameter, 10” height, and 1/4” wall thickness. This reward chamber was heated with a heat lamp and filled with bedding from the animal’s home colony. It was connected to the maze via a clear cast acrylic 2” diameter connector tube with 1/4” wall thickness (US Plastics, Item #44530). Performance was measured using latency to the reward chamber, total distance traveled before reaching the reward chamber, and number of entries into the error zones. An entry into an error zone was counted if over 50% of the animal’s body passed into the zone, regardless of the direction of entry.

Naked mole-rats react strongly to blockages in tunnels by persistently digging and biting in an attempt to get past the blockage (Sherman et al., 1991). During initial maze design testing (Figure 1A), we found that this distraction affected completion of the maze (“Fail Rate,” Table 1). Fail rate was calculated as the percentage of learning trials (Trial 1–15) in which animals did not complete the maze in the allotted 5 min (Table 1).

To address this, we redesigned the maze to a continuous configuration that lacked dead-ends (Figure 1B) and tested thirty-five animals in this maze configuration (Table 1). This allowed us to more effectively investigate whether the animals would be more motivated to complete the maze without having attention diverted from seeking the reward chamber due to dead-end distractors.

In order to test the effectiveness of a range of motivators in naked mole-rats, and subsequently use the most effective motivator in the reward chamber, a separate experiment was performed prior to using the continuous maze. The same eight animals used in the testing of the first maze design were utilized to determine appropriate motivators for naked mole-rats. First, as a control, a start chamber was connected to a separate empty chamber using an L-shaped connector tube. Each animal’s behavior was first assessed in this control condition (see Data Analysis below). To test the effectiveness of a range of motivators, a motivator was added to the empty chamber, making it a reward chamber (Figure 1C). For the control test, each animal was placed in the start chamber and observed for one trial of 2 min in duration. Time spent in each chamber and latency to enter the empty chamber were recorded. This was then compared to each animal’s performance with a motivator present in the reward chamber. All animals were tested with one motivator (e.g., heat lamp) for one trial of 2 min in duration before switching to the next motivator, and motivators were tested in the following order: heat lamp, darkened chamber (shielded from outside light with black construction paper), food (sweet potato, banana baby food, and ProNutro—all items placed together in the food chamber, for all animals tested), home colony bedding, and foreign colony bedding (Table 2). The start and empty or reward chambers were wiped with 70% isopropyl ethanol between testing sessions with different animals. The resulting optimized motivator, a combination of home colony bedding and a heat lamp, was used for the continuous maze reward chamber.

Social hierarchy rank was determined using a dominance testing paradigm modeled after similar testing in previous studies in naked mole-rats (Clarke and Faulkes, 1997, 1998). All animals used in the continuous version of the maze were marked with their ID numbers on their backs using a permanent marker to allow for visual tracking of behaviors during dominance testing. Two researchers placed one naked mole-rat each in separate, circular chambers. Both chambers were connected via a clear polycarbonate tube. The animals were oriented at the open ends of the tube such that they met face to face when they moved in a forward direction (Figure 1D). Encounters were only rated if one animal passed over the top of another while in the tube. The test session ended after three instances of face-to-face pass-over behaviors were recorded. After each session, the chambers and tubing were cleaned with 70% isopropyl alcohol. Animal pairings were randomly generated with no pair undergoing more than one testing session together. Each animal underwent at least three testing sessions, each with a novel paired animal.

Dominance ranking was determined for the continuous maze animals using previously described methods with naked mole-rats (Clarke and Faulkes, 1997; Gilbert et al., 2020). Using a package extension for use with R Studio (R Core Team, 2021) to calculate the Elo rating (Neumann and Kulik, 2020), the results of each testing session across multiple days were entered to yield the rank of each individual animal tested. The higher the ranking, the more dominant the animal.

All spatial learning trials occurred within the same 4 h time block across every day of testing (8 a.m.–12 p.m.). All testing was performed in a brightly lit behavior room, maintained at 18.5–21°C and at least 40% relative humidity. Animals were brought into the room in transfer chambers and allowed to acclimate in the reward chamber of the maze for 5 min. Animals were then removed from the reward chamber and placed at the predetermined entry point of the maze, facing forward (Figures 1A,B). The naked mole-rat was allowed to explore the maze until it reached the reward chamber. The trial was considered complete when 5 min elapsed or when the animal successfully navigated to the reward chamber, whichever came first. If the animal was unable to reach the reward chamber before 5 min elapsed, it was gently guided to the chamber by the experimenter following cessation of the trial and the trial counted as a failed trial. Upon entry into the reward chamber, a block was placed at the entrance to prevent reentry into the maze. The animal was allowed 5 min to rest in and explore the chamber, after which time the next trial began. Performance was recorded with a camcorder (FHD 1080P, YEHOOVJIANGER) and analyzed in real-time using ANY-maze software (Stoelting Co., Wood Dale, IL, United States).

Animals underwent spatial learning/training for three consecutive days with five trials per day for a total of fifteen trials (Figure 2). Between each trial, the maze was cleaned with 70% isopropyl alcohol to reduce the effect of olfactory cues on performance in subsequent trials. After each day of testing, the maze was deconstructed and washed with non-scented soap and water. Spatial memory retention was tested 1 day and 1 week following the last day of training for the non-continuous maze and 1 day, 1 week, and 1 month for the continuous maze (Figure 2). On memory probe days, the animal freely explored the maze for one trial of 5 min or until it reached the reward chamber, whichever occurred first. In memory probe trials, the reward chamber contained the same motivators used in learning trials. If the animal was unable to complete the maze within the 5 min allotted, it was returned to the colony without being guided to the reward chamber.

Paired sample t-tests were first used to examine the effectiveness of different motivators for use in the reward chamber. P-values were first calculated by using raw values to compare latency to the reward chamber vs. latency to the empty chamber (Figure 3A) and dwell time in the reward chamber vs. dwell time in the empty chamber (Figure 3B). The dwell time ratio for reward (DTR_R) and control conditions with an empty chamber (DTR_E) were calculated as follows, where DT_S represents the dwell time spent in the start chamber under reward (DT_SR) vs. control conditions with an empty chamber (DT_SE), and DT represents the dwell time in the finish chamber between control (DT_E, dwell time spent in the empty chamber) vs. reward conditions (DT_R, dwell time spent in the reward chamber):

A paired sample t-test was also conducted for DTR_R vs. DTR_E (Figure 3C). Bonferroni corrected p-values were used to determine significance. A summary of results from the paired sample t-tests can be found in Table 2.

For the continuous maze, latency to the reward chamber and distance traveled were analyzed to investigate spatial learning and memory performance (analyses adapted from Vallianatou et al., 2020). For animals with one trial’s worth of missing data points, the missing points were linearly interpolated by averaging the values for the trial preceding the missing point and the trial following the missing point (e.g., a Trial 4 data point was interpolated by averaging the values of the animal on Trial 3 and Trial 5). To investigate differences in the learning phase of the continuous maze, we used two different multivariate repeated measures factorial MANOVAs. These models incorporated body mass and age as covariates; sex and colony as fixed factors; and latency and distance as dependent measures. These models were used to determine any main effects or interactions on performance in a manner similar to previous studies (Costanzo et al., 2009; Oosthuizen et al., 2013; Oosthuizen, 2020). The first repeated measures MANOVA used both Day and Trial as within-subjects factors, wherein each day was subdivided into its respective 5 trials. The second model was used for ease of interpretation and modeled only Trials without distinguishing by days. In the second model, Trials were analyzed as Trial 1 through Trial 15 rather than Day 1 through 3 each with Trial 1 through Trial 5. Estimated marginal means and confidence intervals from these models were used to graph the adjusted performance values (see Figures 5, 6). These models adjusted for the covariates at Body Mass = 54.60 g and Age = 2.94 yrs. The repeated measure MANOVA was also conducted separately for each colony in order to incorporate dominance ranking as a covariate. These colony specific analyses were done because no dominance testing can be performed between animals of different colonies, as this testing results in lethal confrontations. The colony-specific models used body mass, age, and dominance ranking as covariates; sex as a fixed factor; and latency and distance as dependent measures. These analyses were also performed in two separate ways, one with two repeated measures Day and Trial and the second with Trials as the only within-subject factor. Colony An estimated marginal means were computed using: Mass = 54.63 g, Age = 2.35 years, Rank = 957.40. Colony B estimated marginal means were computed using: Mass = 54.56 g, Age = 3.49 years, Rank = 1002.96. All aforementioned statistical analyses were also conducted as separate repeated measures ANOVAs with speed as the dependent variable to complement the latency and distance analyses. These models were conducted separately because speed is equal to distance divided by latency. Bonferroni correction was utilized for post hoc pairwise comparison for all Day and Trial effects.

Performance percentages on memory probe trials were computed for both maze configurations (non-continuous and continuous), comparing performance (latency and distance traveled) from individual memory probe days to the last training day. Latency and distance traveled were analyzed for both continuous and non-continuous maze designs, whereas errors were analyzed only for the non-continuous maze design. These were calculated such that the value from the memory probe trial (1, 2, or 3) was divided by the last training day trial (Trial 5 on training Day 3; Figure 2 and Table 3). A Repeated Measures MANOVA using latency performance and distance performance as dependent measures; body mass and age as covariates; sex and colony as fixed factors; and memory probe trial (1, 2, or 3) as the within-subjects factor. This analysis was conducted for each colony separately and combined. This set of statistical analyses for memory probe trials were also conducted as repeated measures ANOVAs with speed as the dependent variable in lieu of latency and distance.

We tested different motivators in order to optimize subsequent spatial maze testing performance (Figure 1C). Using paired sample t-tests for the raw values of latency, dwell time, and dwell time ratio compared to the empty chamber, similar motivators proved to be significant (Table 2 and Figure 3). Latency was enhanced with the presence of home colony bedding [t(7) = 5.904, p = 0.003] or foreign colony bedding [t(7) = 5.352, p = 0.005], with heat approaching significance [t(7) = 3.21, p = 0.075]. Dwell Time Ratio was not significant for any motivator tested, but Dwell Time was significantly greater when home colony bedding [t(7) = −9.623, p < 0.001] or foreign colony bedding [t(7) = −7.904, p < 0.001] were present in the reward chamber. Based on these sets of analyses, heat and the presence of either bedding type (foreign or home bedding) were the most effective motivators when presented independently. Thus, both heat and home bedding were chosen for the reward chamber in subsequent continuous maze testing. Foreign colony bedding, though also effective, was not selected based on the potential risk of an experimental animal being attacked upon reintroduction to the home colony due to smelling like an intruder (Braude et al., 2021).

For the preliminary data set collected using a non-continuous maze design (n = 8; Figure 1A), no learning curves were evident across training trials for latency to the reward chamber (Figure 4A), total number of errors (wrong turns) made (Figure 4B), or total distance traveled (Figure 4C). Fail rate was also high (Table 1). The poor performance of naked mole-rats in this non-continuous maze design, along with the observation that the non-continuous maze design may have created a confound (i.e., dead-ends proving to be strong distractors that obscured the true potential for naked mole-rats to exhibit spatial learning, and memory improvements), led us to redesign the maze in a continuous layout with no dead-ends.

Both colonies (Colony A and Colony B) were analyzed together for maximum power with the continuous maze design (n = 27; Figure 1B), and using a repeated measures MANOVA that did not model the effect of testing on separate days for simplicity of interpretation. This analysis revealed a distinct learning curve across the 15 training trials (Figure 5A) that was absent in the non-continuous maze. There was a multivariate effect of trial number on performance [F_{(28, 588)} = 3.074, p < 0.001, Figure 5A and Supplementary Table 1]. The multivariate effect of Trial was driven by main effects on both latency and distance. Repeated pairwise comparisons of Trial 1 with Trials 2–15 revealed that latency (Bonferroni corrected p < 0.05, Figure 5A and Supplementary Table 1) and distance (Bonferroni corrected p < 0.05, Figure 5B and Supplementary Table 1) decreased significantly by the 3rd trial, whereas speed increased significantly by the 6th trial (Bonferroni corrected p < 0.05, Figure 5C and Supplementary Table 3). In addition, female animals had a significantly larger increase in speed across learning trials than males [F_{(14, 294)} = 2.811, p = 0.001, Supplementary Table 3], though there were no significant differences between sexes for latency and distances across trials.

When utilizing a model that did incorporate separate testing days into the model similar results were obtained. There was a multivariate effect of Trial [F_{(8, 168)} = 3.732, p < 0.001, Supplementary Table 2] driven by significant main effects of trial on both latency [F_{(4, 84)} = 5.494, p = 0.001, Supplementary Table 2] and distance [F_{(4, 84)} = 2.809, p = 0.031, Supplementary Table 2], but not speed [F_{(4, 84)} = 2.396, p = 0.051, Supplementary Table 4]. Bonferroni-corrected pairwise comparisons showed that animals traveled significantly less during Trials 2, 3, and 5 of any given day vs. Trial 1 (−1.888, −2.4843, −1.910, and −2.574 m respective changes; p < 0.05; Figure 6C and Supplementary Table 2). Similarly, latency significantly decreased from Trial 1 to Trials 2, 3, and 5 (−43.515, −61.82, and −72.84 s respective changes; p < 0.05; Figure 6A and Supplementary Table 2).

There was no multivariate effect of Day, independent of trial number. There was also no univariate main effect of Day on latency, speed, or distance traveled to reach the reward chamber. However, pairwise comparisons demonstrated that distance decreased significantly between Day 1 and 2 of training (−1.768 m, Bonferroni corrected p = 0.002; Figure 6D and Supplementary Table 2). There was no significant decrease between distance traveled Day 1 and 3 of training. The average latency across trials on Day 1 of the learning period was significantly longer than latencies on Day 2 (70.376 s, p < 0.05; Figure 6B and Supplementary Table 2) and Day 3 (76.556 s, p < 0.05; Figure 6B and Supplementary Table 2).

There was a significant multivariate Day by Trial interaction [F_{(16, 36)} = 3.534, p < 0.001, Supplementary Table 2] driven by significant interactions for both latency [F_{(8, 168)} = 6.006, p < 0.001, Supplementary Table 2] and distance [F_{(8, 168)} = 4.084, p < 0.001, Supplementary Table 2]. There was also a significant interaction of Day and Trial on speed [F_{(8, 168)} = 2.40, p = 0.018, Supplementary Table 4]. Both latency and distance demonstrated the greatest performance improvements across trials on Day 1 of training with performance improvements plateauing across trials in Days 2 and 3, showing that most learning occurred on Day 1.

There was no multivariate interaction between Day and Body Mass, however, there was a univariate effect on latency [F_{(2, 42)} = 4.137, p = 0.023, Supplementary Table 2], but not distance. Larger animals had higher latencies to reach the reward chamber in early days of training with these differences decreasing across days despite all animals traveling comparable distances.

No multivariate interaction of Trial by Age was present. However, there was a univariate effect of Trial by Age on distance [F_{(4, 84)} = 2.580, p = 0.043, Supplementary Table 2] indicating better performance for younger animals in Trials 2 and 3, but for older animals in Trials 4 and 5. There was also a significant multivariate 3-way Day by Trial by Body Mass interaction [F_{(16, 336)} = 1.988, p = 0.013, Supplementary Table 2] driven by significant interactions on both latency [F_{(8, 168)} = 3.245, p = 0.002, Supplementary Table 2] and distance [F_{(8, 168)} = 2.398, p = 0.018, Supplementary Table 2]. This effect indicated that the shape of the learning curve across days and trials was impacted by body mass with smaller animals showing faster learning initially, with more variable performance in later training trials.

There was no Day by Trial by Sex multivariate interaction on performance, nor univariate effects on latency and distance. However, there was a significant 3-way univariate interaction on speed. Female animals increased speed throughout learning trials significantly more than males [F_{(8, 168)} = 3.443, p = 0.001, Supplementary Table 4].

No other multivariate or univariate effects or interactions (Day by Age, Day by Sex, or Day by Colony, Trial by Mass, Trial by Sex, and Trial by Colony) were significant for latency to reach the reward chamber, distance traveled, or speed.

In order to incorporate dominance ranking (a colony specific measure), into the GLM, the data set was split by colony for analysis. There were no significant effects of dominance rank on latency or distance across trials and days, nor were there any interactions of dominance rank with any of the other variables in the model (for detailed statistics see Supplementary Tables 1, 2). Speed traveled varied by dominance rank for Colony B with subordinate animals traveling faster across trials than dominant animals (for detailed statistics see Supplementary Tables 3, 4).

These findings indicate that naked mole-rats exhibited significant spatial learning in the continuous maze. The bulk of the performance improvements occurred within the first 3 trials, indicating rapid spatial learning. Both latency and distance calculations were informative in assessing spatial learning gains in this particular maze configuration in naked mole-rats, suggesting that naked mole-rats were able to rapidly learn to navigate the continuous maze design.

Mean performance was analyzed across memory probe trials for both the non-continuous maze and the continuous maze (Table 3). There were no significant differences between the last trial of the last learning day vs. memory probe trials 1, 2, or 3 in assessing latency (Figure 7A) or distance traveled (Figure 7B), nor any effects of age, sex, ranking, or mass. This indicates that spatial memory was retained for at least 1 month in naked mole-rats.