Abstract
Physical exercise is a robust lifestyle intervention known for its enhancement of cognitive abilities. Nevertheless, the extent to which these benefits can be transmitted across generations (intergenerational inheritance to F1, and transgenerational to F2 and beyond) remains a topic of limited comprehension. We have already shown that cognitive improvements resulting from physical exercise can be inherited from parents to their offspring, proving intergenerational effects. So, we set out to explore whether these enhancements might extend transgenerationally, impacting the F2 generation. In this study, we initially examined the behavioral traits of second generation (F2) male mice, whose grandfathers (F0) had an exercise intervention. Our findings revealed that F2 mice with physically active grandpaternal F0 progenitors displayed significantly improved memory recall, encompassing both spatial and non-spatial information when compared to their counterparts from sedentary F0 progenitors, and proving for the first time the transgenerational inheritance of physical exercise induced cognitive enhancement. Surprisingly, while F2 memory improved (as was the case with F1), adult hippocampal neurogenesis remained unchanged between experimental and control groups (unlike in F1). Additionally, our analysis of small RNA sequences in the hippocampus identified 35 differentially expressed miRNAs linked to important brain function categories. Notably, two of these miRNAs, miRNA-144 and miRNA-298, displayed a robust negative correlation with cognitive performance. These findings highlight the enduring transgenerational transmission of cognitive benefits associated with exercise, even after two generations, suggesting that moderate exercise training can have lasting positive effects, possibly orchestrated by a specific set of miRNAs that exert their influence across multiple generations.
- adult hippocampal neurogenesis
- behavioral phenotype
- in vitro fertilization
- microRNA
- moderate exercise
- transgenerational effects
Significance Statement
Physical exercise is well known by its positive effects on body health and specifically on brain functioning and health. Here, we test whether those effects are inherited from exercised grandparents to the second generation. We report here, for the first time, the transgenerational inheritance of moderate exercise-induced grandpaternal traits in grandson's cognition, even though some of the cellular changes induced in F1 vanish in F2, and suggesting that moderate exercise training has a longer-lasting effect than previously thought, most probably mediated by a small group of microRNAs acting across generations.
Introduction
The beneficial effects of exercise on body and brain health are very well known (Trejo et al., 2002; Llorens-Martín et al., 2008; Di Liegro et al., 2019; Cantelon and Giles, 2021; Hashimoto et al., 2021; McGillivray et al., 2021; Meli et al., 2021; Sutkowy et al., 2021; Umegaki et al., 2021; Mellow et al., 2022), as well as the adverse effects of a sedentary lifestyle (Martin et al., 2018; Rendeiro and Rhodes, 2018; Song et al., 2018). In fact, lifestyle interventions are known as potent drivers of neural plasticity in all types of organisms (Mora, 2013; Valero et al., 2016; Phillips, 2017; Maharjan et al., 2020). Besides, the heritability of these lifestyle effects to either the next generation (intergenerational inheritance) or the second and later generations (transgenerational inheritance; Jirtle and Skinner, 2007) has been largely explored in different taxa (Zare et al., 2018; Deas et al., 2019; Baugh and Day, 2020; Hoyer-Fender, 2020; Frolows and Ashe, 2021; Grishok, 2021; Mu et al., 2021; Chey and Jose, 2022). It has also been explored in different models, with a special focus on stress or diet (Lagisz et al., 2014; Blaze and Roth, 2015; Hime et al., 2021; Risal et al., 2021). However, the inheritance of the exercise-induced effects has been only recently addressed and mostly in intergenerational designs, either patrilineal (Short et al., 2017; Benito et al., 2018; McGreevy et al., 2019; Yeshurun and Hannan, 2019; Yang et al., 2021) or following maternal exercise during pregnancy (Eclarinal et al., 2016; Kusuyama et al., 2020). In these last designs, germline-mediated matrilineal epigenetic inheritance is not easily distinguished from a nongermline epigenetic inheritance, mainly because the role of maternal intervention cannot be ruled out (Champagne, 2008). Additionally, a study revealed the transgenerational transmission of paternal antidepressant effects after environmental enrichment housing, a paradigm different from pure exercise intervention (Yeshurun et al., 2017). Notably, this transmission occurred from F0 to F2 without affecting F1.
On the contrary, the inter- and transgenerational effects of stress have been largely described (Gapp et al., 2014; Yuan et al., 2016; Jawaid et al., 2018; Ambeskovic et al., 2019, 2020). Evidence on the direct and indirect consequences of stress and the mechanisms mediating its inheritance on next generations have paved the way to analyzing the potential positive effects of other lifestyle interventions, as well as to setting a gold standard of the methodology to address these questions in mice models (Bohacek and Mansuy, 2015).
We now know that the positive, procognitive effects and proneurogenic effects of moderate physical exercise in male mice are inherited by the males of the first generation (McGreevy et al., 2019; Vieira de Sousa Neto et al., 2021). Specifically, the former reported that litters from exercised fathers show better spatial and nonspatial cognitive abilities and increased adult hippocampal neurogenesis (AHN), along with changes in the brain mitochondrial activity and hippocampal gene expression. This includes the differential expression of target genes of several microRNAs, such as the 212/132 cluster, observed in the hippocampus of both fathers and litters (McGreevy et al., 2019). Similar procognitive results were reported after paternal environmental enrichment inherited by the first generation (Benito et al., 2018). In the latter study, the involvement of the microRNA 212/132 cluster, present in the male sperm of enriched animals, was elegantly demonstrated to be one mediating mechanism of the intergenerational transmission (Benito et al., 2018).
In this work, we address whether the exercise-induced effects in the parental (F0) and first-generation litter (F1) are also transmitted transgenerationally to the second generation (F2). This is a relevant question due to the current lack of evidence supporting the transgenerational transmission of positive outcomes resulting from lifestyle interventions targeting the brain and cognition. Addressing this question is crucial as it could provide valuable insights for designing more impactful public health policies for the population affected by the sedentary life pandemic (Guthold et al., 2018).
Materials and Methods
Animals
In this work, the F1 animals of a previous work (McGreevy et al., 2019) were used to obtain an F2 generation. All animals (c57BL6J) from the three generations (F0, F1, and F2) were housed under standard laboratory conditions according to the consensus for inter- and transgenerational studies (Bohacek and Mansuy, 2015; see below) and in accordance with the European Union Directive 2010/63/EU. All experiments were performed according to the European Community Guidelines (Directive 2010/63/EU), Spanish Guidelines (Royal Decree 53/2013), and related norms, and they were first validated by the Committee of Ethics and Animal Experimentation of the Cajal Institute (20/05/2016), subsequently favorably evaluated by the CSIC Ethics Committee (Subcommittee of Ethics) of the Spanish Research Council (07/27/2016), and eventually authorized by the competent authority, the Animal Protection Area of the Department of Environment of the Community of Madrid (10/26/2016 and 06/19/2020). All females used for crossings were sedentary. In this line of investigation, only males were analyzed to focus resources on a discrete, specific goal. Females will be the focus of future experiments. Both F2SED and F2RUN animals were obtained from two different litters each (litter effect explained below). The age of the F1 and F2 generations used here is presented in Figure 1.
Experimental design. Mice were derived from sedentary (F0SED) or exercised (F0RUN) grandfathers. The F0RUN animals were submitted to treadmill exercise of moderate intensity for 6 consecutive weeks, while the F0SED animals remained in their home cage without any access to wheel or treadmill. At the end of these 6 weeks, their sperm was collected and used to produce offspring (F1SED and F1RUN animals, respectively) via IVF methods. Both F1 groups lived until 4.5 months of age when their sperm was collected. The F1 animals remained sedentary throughout the entire period. Then, a new offspring was generated via IVF (F2SED and F2RUN animals). These litters were analyzed from 3.5 months of age when the behavioral tests were conducted. At the end of the behavioral testing, the animals were killed for histological and epigenetic analyses. Both groups (F2SED and F2RUN animals) remained sedentary throughout the entire experiment.
F0 male progenitors (grandfathers of F2 animals)
Animals were purchased and housed individually as in McGreevy et al. (2019). They were randomly assigned to the experimental conditions (sedentary vs exercised). The exercised group (F0RUN) underwent the exercise protocol for 6 weeks, while the sedentary group (F0SED) remained in the home cage. The exercise protocol was implemented as described previously (McGreevy et al., 2019). Briefly, to minimize the intersubject variability, we employed a moderately forced activity on a treadmill. Mice ran at 1,200 cm/min for 40 min, 5 d a week. Sedentary mice remained in the same room without running throughout the duration of the protocol. After treatment, the sperm of these animals was used to obtain F1 males via in vitro fertilization (IVF).
To analyze whether transgenerational exercise-driven effects were germline dependent, the progeny of both F1 and F2 was generated through IVF and embryo transfer (ET) by using the male sperm of F0 and F1, respectively. After birth, the litter sizes were homogenized between (foster) mothers, and after weaning, the subjects from different experimental groups were randomly mixed (Bohacek and Mansuy, 2015; see below).
F1 male progenitors (fathers of F2 animals)
F1SED animals were obtained from the sperm of F0SED progenitors, and F1RUN animals were obtained from the sperm of F0RUN progenitors using IVF. On postpartum day 21 (P21), a mixed weaning strategy was used to avoid group effects and minimize possible litter effects due to cohabitation. Litters were assembled with foster mothers to guarantee an equitable distribution of pups in each litter. Consequently, during the initial 3 weeks leading up to weaning, the litters were composed of pups from distinct biological mothers and different experimental groups (different fathers). This was achieved by several strategies, including dividing the large litters upon delivery into two or more smaller groups, and supplementing potentially underpopulated litters by incorporating additional littermates (or pups from other mothers and fathers within the same experimental group) that will not be used (e.g., in this case, with females). This process was repeated with F2 generation. The original litter size from each birth was similar in all cases without any sign of fertility issues. Nothing leads us to believe that the different number of male versus female pups between groups was anything but randomness.
At 4.5 months old, their sperm was used to obtain F2 animals via IVF. Both F1SED and F1RUN males remained sedentary during all the procedures.
F2 males (current generation)
The F2SED (n = 15) and F2RUN males (n = 8) were obtained from two F1SED or F1RUN progenitors, respectively, using IVF methods. On postpartum day 21 (P21), a mixed weaning strategy was used to avoid group effects and minimize possible litter effects due to cohabitation (see the process described for F1 generation). Behavioral testing started at 3.5 months of age. Animals were killed at 4.5 months old.
In vitro fertilization
IVF and ET were conducted at the Mouse Embryo Cryopreservation Facility of the National Centre for Biotechnology, Spanish National Research Council as described previously (McGreevy et al., 2019).
Behavioral assessment
All behavioral tests were performed during the light phase between 9 A.M. and 3 P.M. (lights on from 7 A.M. to 7 P.M.). All tests were either recorded in video or analyzed automatically by the test device. When the test score was obtained by analyzing the videos, the experimenter was blind to the experimental groups. The behavioral assessment took place for 4 weeks, week 1 for activity assessment, Week 2 for novel object recognition (NOR), Week 3 for novel object location, and Week 4 for contextual fear conditioning (CFC).
Activity assessment
A VersaMax Legacy Open Field activity box (Omnitech Electronics) was used to study locomotor activity. The F2 animals underwent a 2 day protocol (5 min in the activity cage per day). On the first day of the protocol, the spontaneous locomotor activity was measured for 5 min in a novel arena. The following day, the animals were placed in the same arena for another 5 min. Behavioral measures included total horizontal activity, horizontal activity per minute, total vertical activity, total distance moved, total time mobile, and total time in margins (as time in margin + center = total time).
NOR
A NOR protocol was applied in an open box of 42 × 32 × 31 cm. The test was applied as described in previous work (McGreevy et al., 2019). It consisted of four phases: habituation (Hab), training (TR), short-term memory evaluation (STM), and long-term memory evaluation (LTM; Fig. 2). In the habituation phase, the animals were left to freely explore the box without any object (5 min). Immediately after, during the training phase, each animal was placed in the arena containing two different objects in the center of the box (Objects A and B) and was allowed to explore the objects freely for 5 min. Next, each animal was removed and put back in the home cage in the behavior room. One hour after training, the animals were placed back in the NOR arena containing a familiar object (Object A) and a novel object (Object C) and allowed to explore them for 10 min. At the end of this phase, all animals were placed in their home cage. Twenty-four hours after training, the animals were tested again for 10 min in the arena containing the familiar object (Object A) and a novel object (Object D). The box was cleaned with 0.03% acetic acid solution between trials. The time exploring each object was manually scored from the recorded video. Discrimination indexes were calculated using the following formula for TR (in STM and LTM, the time exploring B is replaced by either C or D, respectively):
Behavioral analysis. a–e, Basal activity. The F2 animals (F2SED, n = 15; F2RUN, n = 8) were subjected to activity measurements in activity cages for 5 min. A scheme of the actimeter test is shown. Both groups showed similar levels of horizontal activity (a) in a novel environment (Day 1, d1) and a known environment (Day 2, d2). Also, both groups showed a significant decrease in activity in the known environment compared with the novel environment. These were also true for the vertical activity (b), total distance traveled (c), and time in movement (d). Both groups spent a similar time in the margins of the cage (e), with no significant differences between them in d1 or d2. The F2SED animals spent significantly less time in the margins on the second day of the test. f–j, NOR test. A scheme of the test is shown. Evaluation of discrimination indexes (DIs, F2SED, n = 15; F2RUN, n = 8) reports a better performance of the F2RUN animals (f). Groups did not have a preference for any object in the training phase (TR), as they had DIs proximal to zero. In the short-term phase (STM), the F2RUN animals showed a trend for a better discrimination than the F2SED animals, but only the mean DI of the F2RUN animals fulfilled the standard criteria to consider that discrimination has occurred (DI > 0.2, red line). In the long-term phase (LTM), the F2RUN animals also showed slightly higher DIs compared with the F2SED animals (not significant). F2RUN showed the trend of a higher DI compared with the training phase, while F2SED did not. Also, in the LTM phase, only the F2RUN animals fulfilled the standard criteria. The time spent exploring objects was also checked (g). The F2RUN animals showed higher levels of exploration in the LTM phase. The correlational analysis between exploration time and DIs in STM and LTM phases was performed either in all F2 animals (h) showing a positive and significant correlation (Spearman's rho; ρ = 0.597; p < 0.001) or by group (i, j). Only the F2SED animals showed a significant positive correlation between time exploring objects and DIs (i; Spearman's rho; ρ = 0.654; p < 0.001), and no correlation was found in the F2RUN animals (j; Pearson's coefficient; r = 0.276; p = 0.3). These results indicate that the higher exploratory behavior of the F2RUN animals does not explain their better performance in the NOR. An OL test was performed to check spatial pattern separation abilities (k–o). A scheme of the test is shown. Evaluation of DIs (F2SED, n = 12; F2RUN, n = 8) reports a better performance of the F2RUN animals (k). Both groups did not prefer any of the columns in the TR phase. In the test phase (TS), the F2RUN animals had significantly higher DIs than the F2SED group and fulfilled the standard criteria for discrimination. In this test, the groups showed similar levels of exploration in both phases (l). The correlational analysis did not show a positive relation between time exploring columns and DIs in this test (m). This was also true for analysis by group (n, o). CFC was applied to study memory and discrimination abilities in an emotional-aversive context (p). A scheme of the test is shown. Both groups (F2SED, n = 14, F2RUN, n = 8) had similar levels of freezing in the training phase (TR), showing similar abilities for contextual classic conditioning. They also exhibited similar levels of freezing when placed again in the conditioning chamber 24 h later (TS). For both groups, the time freezing in TS was significantly higher compared with TR, as expected. In the context change phase (CC), the animals were placed again in the conditioning chamber, but with different walls (plain white vs checkers pattern). In this phase, both groups showed context discrimination, as indicated by a significant reduction in freezing levels compared with the TS phase, although the time freezing in CC was significantly higher compared with TR in both groups, as expected. A CI (q) was calculated to summarize the cognitive performance. We can see how only F2RUNs show a CI above 0.2, illustrating a significantly higher cognitive performance than F2SED. All data are shown as mean ± SEM. For comparisons between independent groups: *p < 0.05, **p < 0.01, ***p < 0.001; *’0.05 > p < 0.099. For comparisons between dependent groups: +p < 0.05, ++p < 0.01, +++p < 0.001, +’0.05 > p < 0.09. Full statistical information can be found in the Extended Data tables (including all statistical information) labeled from Extended Data Tables 2-1 to 2-17.
Table 2-1
Detailed statistical tables of all data into the figure 2 are shown, including actimeters, novel object recognition test, object location test, contextual fear conditioning, and cognitive index. Information about the animals removed, statistical analysis performed for each case, and statistical results have been listed. Download Table 2-1, DOCX file.
Table 2-2
Detailed statistical tables of all data into the figure 2 are shown, including actimeters, novel object recognition test, object location test, contextual fear conditioning, and cognitive index. Information about the animals removed, statistical analysis performed for each case, and statistical results have been listed. Download Table 2-2, DOCX file.
Table 2-3
Detailed statistical tables of all data into the figure 2 are shown, including actimeters, novel object recognition test, object location test, contextual fear conditioning, and cognitive index. Information about the animals removed, statistical analysis performed for each case, and statistical results have been listed. Download Table 2-3, DOCX file.
Table 2-4
Detailed statistical tables of all data into the figure 2 are shown, including actimeters, novel object recognition test, object location test, contextual fear conditioning, and cognitive index. Information about the animals removed, statistical analysis performed for each case, and statistical results have been listed. Download Table 2-4, DOCX file.
Table 2-5
Detailed statistical tables of all data into the figure 2 are shown, including actimeters, novel object recognition test, object location test, contextual fear conditioning, and cognitive index. Information about the animals removed, statistical analysis performed for each case, and statistical results have been listed. Download Table 2-5, DOCX file.
Table 2-6
Detailed statistical tables of all data into the figure 2 are shown, including actimeters, novel object recognition test, object location test, contextual fear conditioning, and cognitive index. Information about the animals removed, statistical analysis performed for each case, and statistical results have been listed. Download Table 2-6, DOCX file.
Table 2-7
Detailed statistical tables of all data into the figure 2 are shown, including actimeters, novel object recognition test, object location test, contextual fear conditioning, and cognitive index. Information about the animals removed, statistical analysis performed for each case, and statistical results have been listed. Download Table 2-7, DOCX file.
Table 2-8
Detailed statistical tables of all data into the figure 2 are shown, including actimeters, novel object recognition test, object location test, contextual fear conditioning, and cognitive index. Information about the animals removed, statistical analysis performed for each case, and statistical results have been listed. Download Table 2-8, DOCX file.
Table 2-9
Detailed statistical tables of all data into the figure 2 are shown, including actimeters, novel object recognition test, object location test, contextual fear conditioning, and cognitive index. Information about the animals removed, statistical analysis performed for each case, and statistical results have been listed. Download Table 2-9, DOCX file.
Table 2-10
Detailed statistical tables of all data into the figure 2 are shown, including actimeters, novel object recognition test, object location test, contextual fear conditioning, and cognitive index. Information about the animals removed, statistical analysis performed for each case, and statistical results have been listed. Download Table 2-10, DOCX file.
Table 2-11
Detailed statistical tables of all data into the figure 2 are shown, including actimeters, novel object recognition test, object location test, contextual fear conditioning, and cognitive index. Information about the animals removed, statistical analysis performed for each case, and statistical results have been listed. Download Table 2-11, DOCX file.
Table 2-12
Detailed statistical tables of all data into the figure 2 are shown, including actimeters, novel object recognition test, object location test, contextual fear conditioning, and cognitive index. Information about the animals removed, statistical analysis performed for each case, and statistical results have been listed. Download Table 2-12, DOCX file.
Table 2-13
Detailed statistical tables of all data into the figure 2 are shown, including actimeters, novel object recognition test, object location test, contextual fear conditioning, and cognitive index. Information about the animals removed, statistical analysis performed for each case, and statistical results have been listed. Download Table 2-13, DOCX file.
Table 2-14
Detailed statistical tables of all data into the figure 2 are shown, including actimeters, novel object recognition test, object location test, contextual fear conditioning, and cognitive index. Information about the animals removed, statistical analysis performed for each case, and statistical results have been listed. Download Table 2-14, DOCX file.
Table 2-15
Detailed statistical tables of all data into the figure 2 are shown, including actimeters, novel object recognition test, object location test, contextual fear conditioning, and cognitive index. Information about the animals removed, statistical analysis performed for each case, and statistical results have been listed. Download Table 2-15, DOCX file.
Table 2-16
Detailed statistical tables of all data into the figure 2 are shown, including actimeters, novel object recognition test, object location test, contextual fear conditioning, and cognitive index. Information about the animals removed, statistical analysis performed for each case, and statistical results have been listed. Download Table 2-16, DOCX file.
Table 2-17
Detailed statistical tables of all data into the figure 2 are shown, including actimeters, novel object recognition test, object location test, contextual fear conditioning, and cognitive index. Information about the animals removed, statistical analysis performed for each case, and statistical results have been listed. Download Table 2-17, DOCX file.
Between-group comparisons were based on a criterion for proper novel object discrimination following a consensus standard (Bellantuono et al., 2020), consisting of having a group mean DI scores >0.20 plus a significant intragroup difference when comparing STM (or LTM) to training. A DI = 0.20 represents an exploration bias for novel versus known object of 60–40%. This criterion is useful in evaluating cognition between groups when one group meets the criteria but not the other group (difficult protocols), independently of whether the DI is significantly different between groups or if some animals from any group have individual DI scores <0.20. Whatever the case, between-group comparisons of DI were also made and described in the Results section. The rationale of using difficult versions of the NOR and NOL tests is because, first, these versions were proved as keys to detect cognitive improvements (McGreevy et al., 2019). The protocols were set at a level where control animals are not expected to meet the criteria, while the runners’ improvement is assessed. This approach is highly recommended when evaluating cognitive improvements of a group relative to the normal cognition of a control group, in contrast to an experimental design in which the goal is to test cognitive deficits, in which regular versions of the behavioral tests are recommended (Bellantuono et al., 2020). In this kind of behavioral paradigm, it is essential to conduct during the protocol setup a mandatory control test using the easy versions to confirm that the control animals can meet the criteria in those versions but not in the difficult ones. This validation process was established in our previous publication (McGreevy et al., 2019).
A minimal exploration time criterion was established, so that animals exploring both objects together for <2 s were removed from the test analysis. Very few animals failed in fulfilling the minimal exploration time criterion, while the vast majority of subjects explored in a much longer time, reason because we have selected this arbitrary exclusion time point of “<2 s.” The same criterion was independently used also for OL. This is the reason for the differences in N between the NOR and OL tests.
Object location test
The object location (OL) test protocol was applied in a PVC circular arena (35 cm in diameter and 20 cm high), as described in McGreevy et al. (2019). The test consisted of two phases (Fig. 2): training (TR) and test (TS). In the training phase, the animals were left to explore the arena for 4 min. Two identical cylinders of plastic material 15 cm high and 5 cm of diameter were placed symmetrically along the diameter of the arena, at the same distance from the walls. At the end of the phase, the animals were placed again in their home cage. Forty minutes after the beginning of the training phase, the animals were put back in the circular arena for 4 min (test phase). One of the columns was displaced a short distance (difficult version of the test compared with easiest versions where the column is displaced a long distance) from the original position. The arena was cleaned with 0.03% acetic acid solution between trials. Time exploring each column was manually scored from the recorded videos. Discrimination indexes were calculated using this formula:
Cognitive index
The cognitive index (CI) was calculated to summarize the cognitive ability and compare the overall performance in both spatial and nonspatial tests, both short- and long-term trials, between the experimental groups. The CI was calculated as the mean of the DI scores of the three tests:
Contextual fear conditioning
A fear conditioning apparatus (Ugo Basile Fear Conditioning 2.1, 46003 Mouse Cage) was used to test contextual aversive memory and context discrimination abilities. The test consisted of three phases: training (TR), test (TS), and context change (CC). In the training phase (5 min), the animals were left to freely explore the conditioning chamber (17 × 17 × 25 cm) that has walls with a checkers pattern (Context A) for 3 min. At minutes 3:00, 3:30, and 4:00, floor shock (0.5 mA, 2 s of duration) was administered through the floor grid. The animal was left another minute in the cage before the test ended. The test phase was conducted 24 h after the training phase. The animals were put back in the conditioning chamber with a checkers pattern (Context A) and left to explore for 5 min, no shock applied. Twenty-four hours after the test phase, the CC phase was performed. The animals were placed again in the conditioning chamber but with white walls (Context B). The freezing time was automatically scored with ANY-maze (V6.0).
Tissue collection
The F2SED and F2RUN animals were deeply anesthetized with 10 mg/kg bw of pentobarbital (Euta-Lender) and transcardially perfused with 0.9% saline. Next, the brain was removed and dissected down the midline into two hemispheres. The left hemisphere was fixed by immersion in 4% paraformaldehyde in phosphate buffer (PB) for 24 h at room temperature. The following day, each hemisphere was washed with PB, stored at 4°C, and used for immunohistochemistry. The other hemisphere was used for miRNA analysis.
Histology
Serial coronal brain sections (50 µm of thickness) containing hippocampal formation were obtained from each hemisphere on a Leica VT1000S Vibratome and individually collected in a 96-multiwell plate filled with 0.1 M PB. The plates were kept at 4°C until further analysis.
Immunohistochemistry
Each series of sections used for the immunohistochemical analysis was composed of the systematic sampling of the hippocampus in the rostrocaudal axis (eight–nine hippocampal sections per animal, 50 µm of thickness each section, 400 µm apart from each other). For single or double staining, slices were initially preincubated in PB with 1% Triton X-100 and 1% bovine serum albumin (BSA; PBT-BSA). Primary antibodies (DCX, Goat anti-doublecortin, 1:500, Santa Cruz; SOX2, Goat anti-sex determining region Y-box 2, 1:200, R&D Systems; CLR, Rabbit anti-calretinin, 1:3000, Swant; GFAP, Rabbit anti-glial fibrillary acidic protein, 1:2,000, Abcam; pH3, Rabbit anti-phosphohistone H3, 1:500, Millipore) were incubated in agitation with PBT-BSA for 1 h at room temperature and 72 h at 4°C. Secondary antibodies (Donkey anti-goat, Alexa Fluor 594; Donkey anti-rabbit, Alexa Fluor 594; Donkey anti-rabbit, Alexa Fluor 488; Donkey anti-rat, Alexa Fluor 594; Invitrogen, 1:1,000) were incubated with PBT-BSA for 1 h at room temperature and 24 h at 4°C. Cell nuclei were counterstained with 4′,6-diamino-2-phenylindole (DAPI, Sigma-Aldrich, 1:1,000). This set of markers of neurogenesis was selected because it covers the range of different stages of the AHN (from neural stem cells to in situ—granule cell layer and subgranular zone (SGZ)—proliferation, to progenitors and immature differentiating neurons), with the latter subpopulation reflecting not only the effects on progenitor proliferation but also the survival of more differentiated cells (DCX+/CLR+ and DCX−/CLR+).
Stereology
The number of neural precursor cells was estimated using seven physical disectors per animal and obtained with confocal microscopy (Leica TCS SP5, oil immersion 40× objective). The disectors were randomly positioned in the rostrocaudal sections of the granule cell layer of the dentate gyrus visibly containing the SGZ. Neural precursor cells were identified following these criteria: (1) SOX2+/GFAP+ staining; (2) cell body positioned in the SGZ; and (3) radial glia-like morphology with a long process across the GCL. They were then counted using ImageJ (Fiji 1.46). The density per disector and the total number of neural precursors along the hippocampus were calculated in the same manner as described below for DCX/CLR cells.
Phosphohistone H3+ (Ph3+) cells were identified and counted for each animal in a complete series of hippocampal sections using a fluorescence microscope (Leica DMI6000 B). The total number of positive cells was estimated through the optical fractionator method, multiplying the total cell number by the sampling fraction (i.e., 1/8 of every hippocampal section of an animal was counted; thus, the sampling fraction was 8). Only cells stained for pH3+, positioned in the subgranular zone and showing mitotic morphology, were counted.
Doublecortin- and/or calretinin-expressing cells (DCX+/CLR+, DCX+/CLR−, and DCX−/CLR+ cells) were counted using ImageJ (Fiji 1.46) on the seven physical disectors obtained with confocal microscopy (Leica TCS SP5, oil immersion 63× objective) in each animal. The disector were randomly positioned in the rostrocaudal sections of the granule cell layer of the dentate gyrus visibly containing the SGZ. The density of the DCX+/CLR+, DCX+/CLR−, and DCX−/CLR+ cells was calculated per disector. The total number of each type of cell per animal was estimated by multiplying the cell density by the area of the SGZ of each animal. The area of the SGZ was estimated for each animal using the Cavalieri method (Howard and Reed, 2005). The length of the SGZ was measured in a different series of Nissl staining with Neurolucida software. This length was then multiplied by the distance between two systematic-sampled slices (400 µm).
miRNA analysis
The hippocampus regions were dissected and immediately frozen at −80°C. miRNA isolation was performed using a microRNA micro kit (Qiagen, Qiagen GmbH) following manufacturer protocol. Briefly, the samples were homogenized in a QIAzol Lysis Reagent, cleaned by a Sepharose column, and eluted in RNase-free water. The total RNA quality was assessed using vertical electrophoresis (Qsep100, BiOptic), and cleanness and concentration were evaluated using NanoDrop One (Thermo Fisher Scientific). Samples were remitted to Novogene for microRNA isolation and sequencing. Libraries were prepared using NEBNext Multiplex Small RNA Library Prep Set for Illumina (Set 1; catalog #E7300). Libraries were then sequenced on Illumina NovaSeq 6000 to generate 50 bp single reads according to the manufacturer's protocol. The volume of data was 10 million reads equivalent to 0.5 GB of data. The quality score of at least 85% of bases was >Q30 (Phred Quality Score, 30).
After quality checking with FastQC (Wingett and Andrews, 2018) and miRTrace (Kang et al., 2018) software, the reads were analyzed using miRge3.0 v0.1.2 (Patil and Halushka, 2021), aligning against the MirGeneDB database. Normalization of miRNA raw counts and differential expression analysis were performed with DESeq2 R package (Love et al., 2014). The software DIANA-miRPath v.3.0 (Vlachos et al., 2015) was used to associate differentially expressed miRNAs to target genes (with TarBase v7.0) in order to determine the associated and potentially affected GO terms and KEGG metabolic pathways.
For the miRNA analysis comparisons, we found a (expected) biological diversity among the samples, and due to the small number of microRNAs detected (therefore, a low number of comparisons performed in the analysis compared with the RNAseq, for example), the obtained p values were not mathematically corrected for multiple testing. Instead, two thresholds of confidence were applied: a p value <0.05 and a fold change runner/sedentary >0.5.
For the miRNA transgenerational coincidences, we used two approaches. First (Fig. 5), we obtained from the RNAseq of F0 and F1 (McGreevy et al., 2019) two lists of microRNAs with putative target genes that were enriched in the RNAseq GSEA filtered by a false discovery rate <0.05, one for F0 and one for F1. Those lists were compared with the list of the present 35 significantly differentially expressed (sDE) miRNAs. The second approach (Table 1) consisted of filtering the target genes of all 35 sDE miRNAs on F2 (via mirdb.org; Chen and Wang, 2020) by a threshold of 80 on the target score and comparing this list against the sDE genes on the F0 and F1 RNAseq (McGreevy et al., 2019). Nine genes were found in common between F2 and either F0 or F1. Lastly, we submitted every gene target of mir-144 and mir-298 to the GSEA (Subramanian et al., 2005) to compute overlaps, checking collections M1, M2, M3, M5, M8, and MH. We found the following enriched gene sets: MIR_298_5P, MIR_144_5P, MIR_451A, MIR_6951_3P, GOCC_SYNAPSE, GOCC_NEURON_PROJECTION, MIR_466D_5P, MIR_466K, GOBP_NEUROGENESIS, and MIR_466I_5P.
List of F2 sDE miRNAs whose target genes are sDEGs in the RNAseq of F0 or F1 (David analysis)
Litter effect
We used two different litters per experimental group. Nevertheless, we analyzed the potential impact of the litter effect, although both experimental groups included more than one litter. We compared the mean standard deviation of every litter for each of the 25 behavioral variables (from the four behavioral tests) to the mean standard deviation of the whole experimental group and to the mean standard deviation of all animals. By comparing the magnitude of the variability of a litter with that of its experimental group and that of the whole group, an estimation of the impact of the litter effect can be made. We found (data not shown) that 23 of the 25 variables had ≥1 litter with an intralitter variability higher than that of its group or that of all animals and 14 of the 25 (including variables from all the four tests analyzed) had ≥2 litters with an intralitter higher variability than that of its group or that of all animals. Besides, 12 of the variables found with two litters with a higher variability than that of its group of all animals had one litter of each experimental group. A high intralitter variability precludes any litter effect.
Statistical analysis
The following procedure was applied to all data: we removed animals whose behavior in a test does not meet the quality criteria (e.g., not enough exploration time in the NOR, OL test, immobility, etc.). These quality criteria were blindly applied to all samples. The vast majority of animals fulfilled the criteria to be included, except for very rare cases when, for example, the animal hardly explored the objects. Next, we identified extreme outliers and removed them from the data (we considered extreme outliers of those data out of the third quartile + 3*interquartile range or the first quartile − 3*interquartile range), followed by a test for normality of distribution and a test for equality of variances. The only exception to this protocol is the behavior analysis data (Fig. 2) that are reported after analyzing them without eliminating outliers and that are represented in the figure with all the experimental subjects and including the statistical outliers. The difference in sample size between groups was not desirable, but we ruled out the idea of eliminating animals from the F2SED to reach uniformity between experimental groups, as we tend to work with as many animals as available. Having this in mind, this variation in sample sizes handicap has been handled following good statistical standards: testing for normality and for equality of variances and adjusting p values with robust corrections when multiple comparisons are applied to all behavioral tests.
Next, we ran parametric or nonparametric tests depending on results of the previous tests. Parametric tests were applied only when assumptions for each test were met. If not, nonparametric tests were applied. Finally, when post hoc comparisons were calculated, such in ANOVA, p values were always corrected for multiple comparisons by Bonferroni’s adjustment.
For comparisons between two independent groups, the t test was applied when the dependent variable was normally distributed. The Mann–Whitney U test was used in the case of non-normal distribution. For comparisons between two dependent groups, paired sample t test was applied if the dependent variable was normally distributed. When not, the Wilcoxon signed-rank test was used. For behavioral tests that contained between-subject and within-subject measures with three levels, mixed ANOVA was applied when the normality and homogeneity assumptions were met. If the normality and/or homogeneity of variance assumptions were violated, a Friedman test was used, followed by a post hoc Wilcoxon signed-rank test. To study correlations, we calculated Pearson’s correlation when data followed normal distribution. The Spearman correlation was used with non-normal distributions. For microRNAseq analysis, due to the low number of comparisons performed in the analysis (because of the small number of microRNAs detected) and the expected biological diversity among the samples, the obtained p values were not corrected for multiple testing. Instead, two thresholds of confidence were applied: a p value <0.05 and a fold change ratio runner/sedentary >0.5. All data were analyzed using SPSS Statistics (IBM, v.27.0.0). It is important to point that ANOVAs were calculated using SPSS 27, and that to perform ANOVAs, SPSS uses type III sum of squares by default. SS type III is the most recommended type when analyzing unbalanced data. Also, it corresponds to Yates’ weighted squares of means analysis, which was specifically introduced for testing the effects of two factors in unbalanced models that included possible interaction effects. Additionally, to perform every pairwise comparison, we used a custom code in SPSS, where we also specified type III sum of squares. Data are shown as mean ± SEM. To test normality, we applied the Shapiro–Wilk test. Intergroup differences are as follows: *p < 0.05, **p < 0.01, ***p < 0.001. Trends in intergroup differences: 0.05 ≥ *’p < 0.099. Within-group differences are as follows: +p < 0.05, ++p < 0.01, +++p < 0.001. Trends in intragroup differences 0.05 ≥ +’p < 0.099. All graphs were created in GraphPad Prism 5. Data tables outlining the statistics conducted for each variable presented in Figure 2 of the behavior section are included as Extended Data.
We finally ran a statistical power test by using G*Power software to measure the statistical power calculated post hoc for the most important behavioral test: differences in CI, obtaining a statistical power of 0.8398.
Results
We analyzed here through a patrilineal design the behavioral phenotype and the AHN of sedentary F2 adult male mice, bred from sedentary F1 animals by IVF and ET, which in turn were bred either from sedentary or exercised F0 mice by IVF and ET (Fig. 1). In other words, the grand-offspring from sedentary F0 animals (F2SED) were compared with those from exercised F0 mice (F2RUN). We found a similar, normal exploratory activity in both groups as measured by an automated actimeter, showing no significant differences neither in horizontal (Fig. 2a; Day 1, independent samples t test, t(21) = 0.211, p = 0.835; Day 2, independent samples t test, t(21) = 0.529, p = 0.602) nor in vertical (Fig. 2b; Day 1, independent samples t test, t(21) = −0.067, p = 0.947; Day 2, independent samples t test, t(21) = 0.807, p = 0.429) activity, no differences in the total distance covered (Fig. 2c; Day 1, independent samples t test, t(21) = 0.823, p = 0.420; Day 2, independent samples t test, t(21) = 0.776, p = 0.446) or time in movement (Fig. 2d; Day 1, independent samples t test, t(21) = 1.254, p = 0.224; Day 2, independent samples t test, t(21) = 0.553, p = 0.586). No differences were found in the time spent exploring the margin (Fig. 2e; Day 1, independent samples t test, t(21) = 0.046, p = 0.963; Day 2, independent samples t test, t(21) = −0.331, p = 0.744) or the center of the arena (complementary value; data not shown). Both groups showed a significant decrease in activity in Day 2 compared with Day 1, as expected when animals re-explore an already explored environment. This was seen in the horizontal activity (Fig. 2a; F2SED, paired sample t test t(14) = 9.588, p < 0.001; F2RUN, paired sample t test t(7) = 5.856, p < 0.001), vertical activity (Fig. 2b; F2SED, paired sample t test t(14) = 7.511, p < 0.001; F2RUN, paired sample t test t(7) = 5.805, p = 0.001), total distance traveled (Fig. 2c; F2SED, paired sample t test t(14) = 8.624, p < 0.001; F2RUN, paired sample t test t(7) = 4.894, p = 0.002), and mobility time (Fig. 2d; F2SED, paired sample t test t(14) = 8.582, p < 0.001; F2RUN, paired sample t test t(7) = 5.515, p = 0.001). F2SED also spent less time in margins in Day 2 compared with Day 1 (paired sample t test t(14) = −2.381; p = 0.032). Thus, both F2SED and F2RUN animals showed normal values in all behavioral variables measured in the activity cage.
Next, we analyzed cognition using the NOR test, the OL test, and the CFC. A complete data list outlining the statistics conducted for each variable presented for the behavior analysis is included as Extended Data supporting Figure 2 (labeled as Extended Data Fig. 2-1). In the NOR, we used a version specifically designed to make it difficult to distinguish the replacement of a known object by a novel one, as previously described (Fontán-Lozano et al., 2007; McGreevy et al., 2019; see Materials and Methods section, NOR for the rationale to use difficult vs easy versions of the task). We found that while the control F2SED (grandsons whose grandparents were sedentary) did not meet the discrimination criteria as expected (Fig. 2f; see definition below and in Materials and Methods), the F2RUN (grandsons whose grandparents were runners) mice fulfilled the criteria for short-term (1 h, STM) and a trend for long-term (24 h, LTM) memory tests [DI higher than 0.2, and training vs test comparisons significantly different in STM (mixed ANOVA F(1,21) = 8.035, p = 0.01, ηp2 = 0.277; post hoc comparisons adjusted by Bonferroni p = 0.01) and a trend in LTM (p = 0.066)]. The F2RUN animals explored both the known and novel objects longer than the F2SED (Fig. 2g; Mann–Whitney U test, training U = 21, p = 0.025; STM U = 27, p = 0.076; LTM U = 15, p = 0.006). To test if the differences in exploration time could explain the differences observed in the DIs, we performed a correlational analysis between exploration time and DIs in STM and LTM phases in all F2 animals. Indeed, it may be relevant because a longer exploration time correlates with higher DIs when considering all animals (Fig. 2h; Spearman's rho, ρ = 0.597, p < 0.001), but this correlation is only significant for F2SED animals (Fig. 2i; Spearman's rho ρ = 0.654, p < 0.001) and not for F2RUN mice (Fig. 2j). These results indicate that a higher exploratory behavior of the F2RUN animals cannot explain their better performance in the NOR.
The cognitive differences between F2SED and F2RUN were also revealed in another memory task: the OL test. Only the F2RUN animals significantly discriminated the small change in the spatial location of the objects in a difficult version of the test (Fig. 2k; related sample t test; t(7) = −3.252; p = 0.005). Controls showed no preference for the known-location object (neither significant nor lower than −0.2). Besides, in this case, no significant differences in exploration time were found between groups (Fig. 2l), and there was no significant correlation between exploration time and DIs overall (Fig. 2m) nor segregating per group (Fig. 2n,o).
In the contextual fear conditioning test, an emotional-aversive task, we found no significant differences between experimental groups in neither the acquisition of the task nor the change in behavior after distinguishing a different context (Fig. 2p). Thus, both groups under study learned and recalled the task to the same extent. Both groups showed significantly elevated freezing 24 h after training when exposed again to the same training context (test phase, TS) and even when exposed to a slightly different context (context change phase, CC), in which patterned walls from the training phase were changed for white ones [F2SED, Friedman test (χ2 = 26, p < 0.001; post hoc comparison training vs TS, p < 0.001; training vs CC, p = 0.032; STM vs CC, p = 0.032); F2RUN, repeated-measures ANOVA (F(2,14) = 25.171, p < 0.001, ηp2 = 0.782; post hoc comparisons training vs TS, p < 0.001; training vs CC, p = 0.019; STM vs CC, p = 0.037)]. Together, these results indicate that F2SED had no impairment in learning capabilities, and that F2RUN had an improvement in cognition that is detected when tests are tuned to evaluate fine differences in cognition as the case of difficult protocols for the NOR and OL, as we previously demonstrated when testing F1 generation (McGreevy et al., 2019).
To check that the results were comparable and replicable between the OL and NOR, we elaborated a global CI, considering the test phase of the OL and both the STM and LTM phases (NOR). Only the F2RUN animals significantly discriminated both the change in the spatial location of objects and the objects themselves on the difficult version of both tests (Fig. 2q; t test; t(21) = −2.828; p = 0.005).
Next, we characterized the AHN of the animals by a battery of histological markers in the dentate gyrus. We found no significant differences in neural stem cell numbers, as measured by Sox2/GFAP labeling (Fig. 3a,b; independent samples t test; t(20) = 0.635; p = 0.635). There were also no differences in cell proliferation in the granule cell layer, as indicated by pH3 labeling (Fig. 3c,d; independent samples t test; t(17) = 0.909; p = 0.376). Additionally, there were no differences in the number of cell progenitors or differentiating, immature newborn neurons, measured by DCX/CLR (Fig. 3e,f; DCX+/CLR− independent samples t test, t(19) = −0.582, p = 0.567; DCX+/CLR+ independent samples t test, t(19) = 0.778, p = 0.728; DCX−/CLR+ independent samples t test, t(19) = 0.528, p = 0.519). As expected, no significant changes were found in the area occupied by the dentate subgranular layer (Fig. 3g; independent samples t test; t(20) = 0.215; p = 0.321).
Histological analysis. Immunohistochemistry against a battery of markers was used to characterize the AHN. Total cell numbers were estimated by stereological unbiased protocols (F2SED, n = 13; F2RUN, n = 8). Mean total cell numbers are shown. Neural stem cells/precursors (a, b) were analyzed by the Sox2+/GFAP+ cell number in the subgranular zone of the hippocampal dentate gyrus (a). No significant differences between experimental groups were found (b). Cell proliferation (c,d) in the dentate gyrus granule cell layer was assessed by pH3 immunohistochemistry. pH3+ cells shown in c were counted, but no significant differences between groups were found (d). The immature, differentiating cell subpopulation in the granule cell layer (e,f) showed no significant differences in DCX+/CLR−, DCX+/CLR+, and DCX−/CLR+ cell subpopulations (f, from left to right respectively). Similarly, no differences were found in the subgranular zone area (g).
We next analyzed the differential expression of microRNAs by a smallRNAseq of the F2 hippocampus (n = 6 per group, randomly chosen from each group). We found 35 sDE microRNAs, 15 microRNAs with a lower expression, and 20 microRNAs with a higher expression in exercised F2 compared with sedentary F2 (Fig. 4a). Relative homogeneity of the differential microRNA expression of both experimental groups can be observed when the heat map is clustered according to the expression level (Fig. 4a). This clusterization is more informative than simply arranging individuals by experimental group as it sorts them based on their similarity. This method reveals an almost complete segregation of both groups, with only two F2SED animals interspersed with the F2RUN group. The global expression change can be observed in a volcano plot (Fig. 4b) with indication of fold change and statistical significance to appreciate which microRNAs are with their relative differential expression, as well as the most represented gene ontology categories (Fig. 4c). Interestingly, we found that the expression level of two microRNAs was significantly negatively correlated with the F2 CI: 298_5p (R −0.64; p = 0.025) and 144-v1_5p/144-v2_5p (R −0.68; p = 0.016). The lower the expression levels of these miRNAs induced by the grandparents’ exercise, the better the CI of the F2 litters (Fig. 4d,e). The overlapping target genes of these two miRNAs appeared in GO annotations “GOCC_SYNAPSE,” “GOCC_NEURON_PROJECTION,” and “GOBP_NEUROGENESIS.” A clear separation of experimental groups was observed by means of a principal component analysis of the miRNAseq (data not shown).
smallRNA analysis. miRNAseq differential expression analysis between the F2 mice, grandsons of either sedentary or exercised grandfathers. a, Clustered heat map of differentially expressed microRNAs between the two groups (p < 0.05). Samples labeled with “R” are from grandsons of exercised grandfathers, and samples labeled with “S” are from grandsons of sedentary grandfathers. All samples have been grouped by unsupervised hierarchical clustering (Pearson's correlation) according to the expression level. Note that the CI of each mouse is depicted above the heat map. b, Volcano plot showing the global expression change between the two groups. microRNAs with a p value <0.05 and positive log2 fold change (upregulated) are represented as blue dots. microRNAs with a p value <0.05 and negative log2 fold change (downregulated) are represented as red dots. The x-axis represents the log2-transformed fold change of the microRNA expression (runner/sedentary). The y-axis is the p value (−log10-transformed). c, Gene Ontology (GO) overrepresentation analysis (ORA) on the differentially expressed miRNA gene targets. Bar plot shows the top 10 enriched GO terms with a p < 0.05. d–e, Significant negative correlation (Pearson's) plots between CI and microRNA expression levels of 144-v1_5p/144-v2_5p* (R −0.68, p < 0.01) and 298_5p (R −0.64, p = 0.025). f, PCA analysis plot showing the separation of both experimental groups.
Finally, we conducted a descriptive analysis to compare the results of the F2 smallRNAseq with our previous results of RNAseq for both F0 and F1 [after GSEA analysis of the RNAseq and the Molecular Signatures Database collection MIR: microRNA targets, or DAVID analysis, both published in McGreevy et al. (2019)], to check for coincidences between generations at the gene expression level. To do this, we followed two different approaches (Fig. 5a). In the first one, we crossed the sDE miRNA list from F2 with the microRNA-related significantly enriched gene set lists from the F0–F1 GSEA analysis. When the GSEA marks a miRNA-related gene set as “enriched,” it means that genes that are putative targets of a given miRNA are overrepresented in the upper or lower parts of the list of genes when sorted by expression levels (i.e., the genes that are putative targets of a specific miRNA accumulate expression changes as a whole group, although these expression differences might not be significant). We found coincidences for 11 of the F2 sDE miRNAs (Fig. 5), a finding that points them as interesting targets to be further explored and as potential drivers of epigenetic inheritance across the three generations in future works. Some of the sDE microRNAs (Fig. 4) present both 3p and 5p arms that are coexpressed and apparently functional (for that reason, they add up to 35). Instead, in Figure 5, each microRNA is represented only one time, thereby adding 27.
Comparison of F2 smallRNAseq results with F0–F1 RNAseq. a, Illustration of the two different approaches taken to compare F0–F1 with F2. Approach 1 compares putative F0–F1 miRNAs (from the RNAseq GSEA) to F2 miRNA (from miRNAseq). Approach 2 compares F0–F1 enriched genes (from the RNAseq DAVID) with high-probability target genes of the F2 sDE miRNA (from miRNAseq). b, The gene sets from “Molecular Signatures Database collection MIR: microRNA targets” enriched after a GSEA analysis of F0 and F1 RNAseq were used to detect coincidences with the F2 sDE miRNAs. We found the number of genes represented in the Venn diagram for each of the generations. Eleven F2 miRNAs or their target genes in F0 and F1 appeared consistently different across three generations. This diagram shows 27 microRNAs because each sDE miRNA from Figure 4 is represented here just one time for clarity, while Figure 4 shows 35 because all the different sDE miRNA arms are represented (e.g., Fig. 4 represents both 3p and 5p arms of some miRNA when both forms have been found as sDE, while Fig. 5 represents one time that of miRNA). Venn diagram licensed by BioRender.
In a second approach, we compared a filtered list of the target genes of all 35 sDE miRNAs on F2 to the list of sDE genes on the F0 and F1 RNAseq (McGreevy et al., op. cit.), finding that the target genes of 6 of those 35 F2 sDE microRNAs had been differentially expressed (sDEGs, DAVID analysis) also either in F0 or F1 (Table 1, here all the coexpressed, coregulated forms are represented, thereby adding nine different forms), pointing them as potentially regulated genes to explore in further studies as important for the procognitive phenotype. Overall, this comparative analysis across generations shown in Figure 5 and Table 1 reveals interesting coincidences between F0, F1, and F2 and with works by other authors, suggests targets to be further explored in future studies to assess the causality of miRNAs in the transmission of the cognitive phenotype to the F2 generation.
Discussion
The behavioral phenotype and AHN marker expression were analyzed in the second generation from either sedentary or exercised male mice (patrilineal design). Only the F0 generation (grandfathers) experienced moderate, forced physical exercise training, while the F1 and F2 generations (the latter, analyzed here) were all sedentary (results of the effects of exercise in F0 and F1 were reported previously; McGreevy et al., 2019). Grand-offspring (F2SED) from sedentary F0 animals were compared with those (F2RUN) from exercised F0 mice. As explained in McGreevy et al. (op. cit.) the original F0 consisted of five animals per group.
The F2RUN mice learned to distinguish a novel object from a known object 1 h after training and were able to recall this information after 24 h in a difficult nonspatial NOR design, which the F2SED animals were not able to accomplish fully. Independently, the F2RUN animals spent longer times exploring both objects. However, this specific (object-driven) exploratory activity does not significantly correlate with the F2RUN DIs (while it is the case for F2SED). Together, these results show that neither the different unspecific, spontaneous exploratory activity nor a specific (object-driven) higher exploratory activity in the F2RUN animals can account for the higher discrimination abilities between known and novel objects. This points to a specific improved ability in the F2RUN mice to distinguish a difficult object replacement, apart from an increased interest in exploring the environment. These findings are similar to those reported in intergenerational experiments analyzing the first generation (McGreevy et al., 2019). It is also relevant to take into account that between-group comparisons were based on criteria set for proper object discrimination (explained above and in the Materials and Methods) and not only in direct comparisons of DI between groups.
Our criterion for evaluating discrimination in all object tests (NOR and novel object location, see below) is threefold, as we look at three possible differences as follows: (1) significant intergroup differences in DI in STM (and LTM in those cases where it is measured; (2) that the mean of one group (and a substantially high number of subjects in those groups) is above the cutoff of 0.2, while the other group is not; and (3) significant intragroup differences between STM and its previous training (and LTM vs training, in those cases where it is measured) in one of the groups, while the other group did not. Any of these three possible differences is sufficient and used to be reported in the field. Obviously, the more these three possible differences are met, the more powerful the result will be, but it is considered not necessary that all of them be met. In the particular case of these experiments and the NOR test, we found that in LTM, the F2RUN group has a mean DI (and the vast majority of the subjects in the group) above 0.2 (while the F2SED group does not), and the DI of the F2RUN is significantly different (greater) than their own training, while the F2SED group is not. For this reason, we consider this result, namely, the F2RUN group performing better in this test than the F2SED group, to be solid. Regarding the technical value of 0.2 as a cutoff level, this value is a field consensus established, as mentioned above, from the consideration that a DI >0.2 means that the subjects have explored the new object at least 60% of the time or longer, and 40% of the time or shorter, the known object, which is considered a nonspurious discrimination.
The F2RUN animals discriminated better than F2SED mice in a spatial, difficult version of the OL test by fulfilling the discrimination criteria, while the controls did not. Also, control mice did not match the criterion for the immobile object (the one in its previous location), thereby displaying no preferences by either object, suggesting these animals did not appreciate any change in the environment. The results in both tasks (NOR, but specially more evident for the OL) point to a general improvement in both spatial and nonspatial abilities caused by the intervention. Again, similar results have been reported previously in both F1 and F0 generations (McGreevy et al., 2019).
It is relevant to consider that the experiments conducted here were designed to be sensitive to the difficulty level, although the findings might also be task dependent. Certainly, both difficulty level- and task-dependent results have been found in intergenerational experiments with interventions like environmental enrichment (Benito et al., 2018) and physical exercise (McGreevy et al., 2019). Here, we have shown, first, no significant differences between groups in a different cognitive task like CFC (test trial) and second, that a salient change in the features the animals use to distinguish the contexts in CFC (easy versions) reveals no differences (change of context trial). Nevertheless, the aversive stimulus in the CFC, a repeated electric shock, is quite reinforcing. Although the intergenerational inheritance of exercise-induced effects on conditioned fear has been previously shown (Short et al., 2017; Yeshurun and Hannan, 2019), we found no differences in F2 generation (transgenerational). Nevertheless, these works and others demonstrate that the inter-/transgenerational inheritance related to fear conditioning is limited to very specific circumstances. Specifically, Dias and Ressler (2014) reported that the intergenerational effect of odor fear conditioning in F0 to F1 increase in sensitivity to the F0 conditioned odor (relevantly through CpG hypomethylation of some olfactory gene). Arai et al. (2009) reported that in a genetic model unable to recall CFC information, the F0 environmental enrichment is able to normalize F1 freezing levels only in females. No normalization was found in F2 (as in the present work), so that effect is intergenerational (not transgenerational) only in females and transitory. Debiec and Sullivan (2014) used also a model of intervention in mothers (stress) to analyze fear conditioning, finding an intergenerational transmission of conditioned fear after mothers expressed fear to a specific smell in pups’ presence. Short et al. (2017) also found altered CFC (only extinction) in F1 P15 females after F0 administration of high doses of glucocorticoids. Therefore, most solid effects in the literature have been found after interventions inducing changes in methylation, which are known to induce lasting effects, generally associated to high stress interventions. When interventions that induce a not-so-profound effect (like physical exercise or environmental enrichment) were used, the authors found intergenerational effects in fear conditioning restricted to specific aspects of the process (extinction, not tested in the present work), or only in one sex, or during a short time window, and not transgenerational effects (even present in F1, but not in F2 in one work). Unfortunately, we do not know about the outcome of CFC in our F1 fathers, so we cannot comment on whether the inter-/transgenerational leap is different from that previously reported. Nevertheless, we cannot rule out that because CFC tests are not only cognitive, but also include other aspects (as pain sensitivity for example), it is possible that the threshold of our task protocol was not sensitive enough to detect differences between experimental groups in F2, although this result is not much different than those found in the literature. In summary, what we report here is a transgenerational transmission of exercise-induced effects specific to some behavioral tasks, but not others.
We found no differences in the adult neurogenesis rate in the hippocampal dentate gyrus, neither in the neural stem cell number, nor in proliferation, nor in immature newborn neuron numbers. The neurogenic niche has been largely reported as highly sensitive to lifestyle interventions (Llorens-Martín, 2018) and a paradigmatic example of metaplasticity (García-Segura, 2009; Llorens-Martín et al., 2009). It is not surprising that a highly sensitive process like AHN responds to direct physical activity (for a recent review, see Bettio et al., 2020) and is intergenerationally inherited by the next generation (F1; Benito et al., 2018; McGreevy et al., 2019), but is no longer present transgenerationally, in a generation (F2) whose fathers (F1) were sedentary as well (present results). Nevertheless, it cannot be ruled out that more subtle changes in the neurogenic subpopulation of the hippocampus (like synaptic boutons, for example) might be altered by the intervention, but not detected by the analyses performed here or in previous works, as suggested by the GO annotation (see below) of the sDE miRNAs found in the present work. Anyway, it is important to note that even though AHN has been long associated to cognitive performance, it is not the only significant factor. Other factors, such as synaptic plasticity, neuronal pruning, and the balance of excitatory and inhibitory neurotransmission, among many others, also play key roles in cognitive function. Therefore, while AHN is a key component, it is part of a complex interplay of factors that contribute to cognitive performance. With no doubt, some other variables might be influenced by the interventions contributing to the inter- and transgenerational transmission of behavioral effects such as mitochondria or microbiota. While it is not the target of the present work to investigate the effect of exercise on these two biological elements, the behavioral outcome found previously in F1 and here in F2 should depend to some extent, but not solely, on AHN, and this might be one explanation why behavioral changes are partially vanished from F1 to F2 in parallel with no AHN changes in F2 compared with F1. Also, the mechanisms underlying inter- but not transgenerational transmission of effects on AHN should be further addressed in the future. A tempting research avenue might be parallel with the analysis performed for the inter- and transgenerational transmission of stress-induced effects. In this paradigm, the brain glucocorticoid receptor has been reported as a target of epigenetic modifications mediating the outcome of stress intervention and vanishing through generations probably depending on the intensity of the original stimulus (Yehuda et al., 2014, 2016). This model might be used to analyze brain IGF1, BDNF, or VEGF receptors (among other factors) as these have been long reported as targets of exercise interventions mediating the effects on AHN (for a recent review, see Kraemer and Kraemer, 2023).
Previous studies have found that the procognitive effects of physical exercise rely on changes on different miRNAs (Benito et al., 2018; Goldberg et al., 2021). Therefore, to explore potential mediating mechanisms of these results, we performed smallRNAseq of the hippocampus of adult F2SED and F2RUN animals. We also compared the results of this smallRNAseq with the RNAseq of F0 and F1 (McGreevy et al., 2019). We found (Fig. 4a,b) 35 sDE miRNAs in the miRNAseq (either over- or underexpressed between F2SED and F2RUN) involved in relevant functions such as structure development, cell differentiation, and ion binding, among others (Fig. 4c). This result is relevant as few more than 500 different miRNAs were detected (Table 2), pointing to 6.8% of significant genes in relation to the total detected. The finding is especially interesting after comparing to the RNAseq of F0 and F1. As the target genes of many miRNAs may be either directly regulated by a specific miRNA or indirectly by a complicated network of different coregulatory group of miRNAs, we made a comparative analysis of the F0 and F1 RNAseq (McGreevy et al., 2019) and the present F2 microRNAseq. Eleven of the 35 F2 sDE miRNAs were present in the two lists (for F0 and F1) of microRNAs whose putative target genes were enriched after a GSEA analysis (“microRNA targets”) of the RNAseq, pointing to a small group of active miRNAs involved across generations in the determination of the traits analyzed here. Moreover, specific target genes of six sDE miRNAs in F2 were also sDEGs either in F0 or in F1 after DAVID analysis, pointing to a group of specific genes involved in this transgenerational inheritance. The literature contains relatively little information about most of these miRNAs, and there is conflicting information specifically regarding their function in cognition (see below). In any case, it is especially noticeable that the miRNA 144, one of the two sDE miRNAs significantly correlated to the CI in F2 (the other is 298), is found in both the F0 and F1 lists of enriched microRNA target genes (McGreevy et al., 2019), and two of its target genes are also sDEGs in F1 (Srs10 and Hnrnpl). Finally, the GSEA analysis revealed that target genes of miRNAs 144 and 298 are involved in the biological function of neurogenesis and the cellular components of synapse and neuron projection. The specific role of these miRNAs in brain-associated parameters is, to our knowledge, far to be well stablished. miRNA144 has been reported both to be associated to worsening the lesion-induced outcome (Sun et al., 2017) or depression-like symptomatology (van der Zee et al., 2022) and to show a beneficial, inverse correlation to depression rating scales (Wang et al., 2015) or association with fear extinction (Murphy et al., 2017). Similarly, miRNA298 has been reported both to be related to pharmacological antidepressant effects (Zou et al., 2020) and to the induction of neurodegeneration in different models (Wallach et al., 2021). As the vast majority of these reports address very specific paradigms associated with lesions or diseases, further research is needed to clarify the role of these miRNAs in cognition of healthy individuals, focusing on miRNA levels in the brain, not only in plasma. Together, these findings suggest common pathways of epigenetic/genetic regulation leading to the phenotype described here. This study represents the first instance, to our knowledge, where a certain cognitive functionality has been assigned to these miRNAs.
Number of miRNAs detected per animal in the miRNAseq and the number of sDE miRNAs
Our findings are relevant because, to our knowledge, only adverse outcomes have been largely reported to be transgenerationally inherited in mammals (Gapp et al., 2014; Bohacek and Mansuy, 2015; Jawaid et al., 2018; Cunningham et al., 2021; Pang et al., 2021) and the transgenerational inheritance of memories has been long described in worms (Deshe et al., 2023). On the contrary, only the intergenerational inheritance of beneficial, procognitive effects has so far been described after physical–cognitive activity (Benito et al., 2018; McGreevy et al., 2019). Here we describe, for the first time, a transgenerational inheritance of the positive effects of moderate, forced physical exercise training in male mice on cognition. We found that some, but not all, effects intergenerationally inherited by the first generation of sedentary litters (McGreevy et al., 2019) are transgenerationally inherited in the second generation (present results), pointing to a partial vanishing of the influence of the parental exercise when the lifestyle intervention is removed across generations. It is relevant to take into account that litter numbers in our study are low and might be considered as a weakness. We checked for a potential litter effect (by comparing the different intralitter vs interlitter variability for each and every one of the parameters; data not shown) to find no effect, provided that the intravariability that was slightly higher than the interlitter variability negates the presence of any significant litter effect. Besides, we also ran a post hoc G*Power calculation for the differences in the CI, to find a 0.8398 score, which was considered robust. This information suggests that, even with a small sample size, our study has sufficient statistical power to detect meaningful differences in the CI between sedentary and exercised mice.
In the present work, the grandfather exercise-induced cognitive improvement of grandsons is revealed in both spatial and nonspatial tests and seems to be task-specific and difficulty level sensitive. Because sperm miRNAs as an epigenetic mechanism of inter-/transgenerational inheritance of exercise/enrichment effects are gaining position in the field (Benito et al., 2018), the small group of miRNAs and target genes found here might be considered as a potential epigenetic/genetic mechanism accounting for this effect (also considering that no changes in sperm methylation were found in McGreevy et al., 2019), but this should be further demonstrated. These findings might also be interpreted as suggesting that sedentary lifestyle effects (adverse body and brain health effects) can be transmitted to next generations.
Together, these results point to an unexpected heritability of the beneficial effects of a moderate exercise program on cognition, clearing the way for the further exploration of the molecular mechanisms mediating these effects (e.g., specific microRNAs reported here) that might be used as pharmacomimetics of healthy lifestyles. Nevertheless, these findings may be valuable data supporting evidence-based health policies in contexts such as development, adult disease, and aging.
Data Availability
All materials, data, and associated protocols used in this work are available to interested readers. microRNAseq data are available uploaded at GEO accession GSE217487.
Footnotes
We are grateful to Laude Garmendia from the Animal House at the Cajal Institute for her unpayable help and advice and to the Image Analysis Unit of the Cajal Institute. E.C. and P.M. were funded by a predoctoral fellowship (FPI) grant and P.T. by a predoctoral fellowship (FPU). This work was supported by project grants H2020-INFRADEV-2016-2017 730879 (to L.M.), BFU2013-48907-R, BFU2016-77162-R, BES-2017/080415 (from the Spanish Ministry of Economy and Competitiveness), PID2019-110292RB-100, PID2022-136891NB-I00, and PRE2020/093032 (from the Spanish Ministry of Science and Innovation; to J.L.T.), and Ministerio de Universidades (18/00069).
↵*E.C. and P.T. contributed equally to this work.
The authors declare no competing financial interests.
- Correspondence should be addressed to José Luis Trejo at jltrejo{at}cajal.csic.es.
