Substantial research demonstrated that regular physical activity (PA) exerts beneficial effects on cognitive performance in the elderly (Kirk-Sanchez and McGough, 2014; Erickson et al., 2015; Prakash et al., 2015). More specifically, it has been shown that PA exerts a larger positive influence on tasks mainly tapping executive functions than other tasks focusing on speed of information processing or visuospatial processing (Colcombe and Kramer, 2003). Different authors suggested that executive functions are not a unitary function but rather an umbrella of inter-correlated functions such as controlled inhibition, cognitive flexibility and updating of working memory (Duncan et al., 1997; Miyake et al., 2000). Within this diversity of executive functions, controlled inhibition seems to benefit more selectively from PA than the other executive functions (Smiley-Oyen et al., 2008; Boucard et al., 2012; Predovan et al., 2012; Gajewski and Falkenstein, 2016). Controlled inhibition is the capacity to suppress irrelevant information or prepotent responses, and it plays a core role in the executive functions construct (Miyake et al., 2000). Deterioration in inhibitory processes may play a central role in age-related declines in several different cognitive functions (Hasher and Zacks, 1988). However, some recent meta-analyses failed to demonstrate consistent and reliable benefits of PA on executive functions (Angevaren et al., 2008; Kelly et al., 2014), despite the apparent consensus of the positive effects of PA on cognitive functioning in elderly populations. Differences in study designs, training programs or the methodology used to analyze the principal outcomes primarily explain these inconsistent results (Audiffren et al., 2011). Another way explaining these discrepancies is the variety of participant characteristics in these studies, such as gender and age (see Fagot et al., 2017), or education level. The genetic profile is also an important factor that influences cognition, but it has received very little attention.

Several brain molecular and cellular mechanisms, such as the synthesis of neurotrophic proteins, progressively change during the aging process (Grady, 2008). Brain-derived neurotrophic factor (BDNF) is one of the main proteins involved in cerebral functioning (Neeper et al., 1995; Adlard et al., 2005; Snigdha et al., 2014), and its secretion decreases during aging (Pang and Lu, 2004). BDNF release is associated with the genetic profile of its gene. In other respect, regular PA was also associated with higher concentration of BDNF in the brain and enhancement of brain plasticity (Cotman and Berchtold, 2002; Cotman et al., 2007).

A single nucleotide polymorphism in the BDNF gene, BDNFVal66Met, affects BDNF secretion, and it may be involved in the deleterious effect of brain aging (Miyajima et al., 2008). The Met allele of the BDNFVal66Met polymorphism is present on one or two gene copies in 30% of the United States population (Shimizu et al., 2004). The deleterious effect of this Met allele reduces BDNF secretion in response to neuronal stimulation in adults (Egan et al., 2003; Kleim et al., 2006). When brain BDNF levels decrease, a lower cognitive performance can thus be expected. As mentioned just earlier, a lower level of brain BDNF can be observed in sedentary people by comparison to active people (Szuhany et al., 2015; Dinoff et al., 2016) and/or in people with two Met alleles at codon 66 of the BDNF gene. The examination of a possible interaction between PA and BDNF polymorphism on different cognitive functions declining with aging, and more particularly executive functions, is then particularly important. For instance, it can be expected that the positive effect of PA on executive functions may counteract the negative effect of the BDNF polymorphism Met allele (Raz et al., 2009). In this perspective, several authors examined the interaction between BDNFVal66Met and PA in elderly participants on cognitive performance, brain volume and/or cerebral plasticity (Kim et al., 2011; Erickson et al., 2013; Brown et al., 2014; Canivet et al., 2015; Crispim Nascimento et al., 2015; Thibeau et al., 2016). Only the last study of Thibeau et al. (2016) tested this interaction on executive functions. These authors demonstrated in a longitudinal study that Val carriers without PA exhibited poorer executive functions performance at age 75 than their peers with higher PA levels. In contrast, the level of PA did not affect executive functions performance in Met-heterozygous. It is interesting to note that they used four tasks tapping executive functions including two tasks well-known to involve controlled inhibition, the Stroop task and the Hayling sentence completion test. The results observed by Thibeau et al. (2016) converge with those of a previous study conducted by our group (Canivet et al., 2015), which demonstrated that only Val-homozygous older adults benefited from regular PA and exhibited better episodic memory performance compared to their Val-homozygous inactive counterparts. However, a same polymorphism can lead to notable variations in protein secretion in different ethnic groups (see Tsai et al., 2008). Consequently, a replication of Thibeau et al. (2016) data in a different population is important to generalize the results.

The present study used the same pool of participants aged 55 years and older than the study of Canivet et al. (2015) and examined the interaction between PA and the BDNF polymorphism on controlled inhibition performance using a modified Simon-like reaction time task. Participants in Simon-like tasks respond to a relevant stimulus feature (e.g., an arrow), and the position of the stimulus is always irrelevant (Verbruggen et al., 2005). There is a dimensional overlap between the irrelevant stimulus feature, i.e., the position of the stimulus, and the response set, i.e., the direction of the arrow in the example. When the stimulus position corresponds with the response side (e.g., left key press to a stimulus that is presented on the left of the screen), the trial is considered congruent, and the responses are typically faster than incongruent trials, where there is a mismatch between the stimulus position and response side. The dual-route model (e.g., Kornblum et al., 1990) suggests that the relevant stimulus feature (e.g., the direction of the arrow) is processed via a controlled route and activates the correct response, and a second response code is activated by the position of the stimulus via an automatic route. Ridderinkhof (2002) proposed that the automatic activation of the incorrect response is selectively suppressed by a central active inhibitory mechanism that requires time and may explain why participants respond slower on incongruent trials. The performance of older adults is consistently impaired in this type of task, but PA potentially reduces this impairment. It is also important to note that there is a problem of task impurity related to the measurement of the effectiveness and/or efficiency of a specific cognitive process (Burgess, 1997; Phillips, 1997). Generally, a cognitive task well-known to tap a cognitive process such as executive functions, or more specifically controlled inhibition, also involves other non-executive cognitive processes, for instance semantic memory. Consequently, a low performance observed in a specific task can be due to the executive component of the task and/or to the non-executive component of the task. The impurity of cognitive tasks may explain, on the one hand, the weak correlation between tasks well-known to tap executive functions (Miyake and Shah, 1999) and, on the other hand, the weak reliability test–retest of these same tasks (Denckla, 1996). It is recommended to use several tasks tapping the same cognitive function in the same study in order to avoid this problem of impurity. Another approach is to replicate the results using one or more tasks tapping the same cognitive process than the replicated study; this was the approach chosen in the present study.

The results of Canivet et al. (2015) and Thibeau et al. (2016) suggest a PA × BDNF polymorphism interaction on cognitive performance such as the effect of PA on controlled inhibition will be significant only for the Val-homozygous carriers, and Met carriers will not be affected by PA. In this perspective, the present cross-sectional study aims to replicate the results obtained by Thibeau et al. (2016) on controlled inhibition but using a different task tapping the controlled inhibition function in a sample of French participants.

The ANOVA on RT data revealed no main effect of PA [F(1,110) = 2.23, p > 0.13] or BDNF polymorphism [F(1,110) = 0.46, p > 0.50]. There was a significant effect of trial type. Congruent trials produced shorter RTs (M = 614.97 ms; SD = 11.60 ms) than incongruent trials (M = 770.26 ms; SD = 16.41 ms) [F(1,110) = 314.45; p < 0.05; ηp2 = 0.77]. The PA × BDNF polymorphism × trial type interaction was marginally significant [F(1,110) = 3.184, p = 0.08], which suggests that only Val-homozygous active participants exhibited shorter RT on incongruent trials compared to their Val-homozygous inactive counterparts. However, the results of the ANCOVA revealed that this interaction was not significant when controlling for age, gender and BMI [F(1,102) = 1.88, p > 0.18]. A last ANCOVA on RT data with Visual acuity and Depression as additional covariates, pointed out no significant effect of Visual acuity [F(1,100) = 0.30; p > 0.58] and a marginal effect of Depression [F(1,100) = 3.51; p = 0.06] on RT (Table 2).

The ANOVA on error rates revealed no main effect of PA [F(1,110) = 2.90, p > 0.09] or BDNF polymorphism [F(1,110) = 1.49, p > 0.22]. The effect of trial type was significant [F(1,110) = 337.34; p < 0.05; ηp2 = 0.75]. Participants made fewer errors for congruent trials (M = 1.24%; SD = 0.24%) than incongruent trials (M = 7.61%; SD = 0.60%). The PA × BDNF polymorphism × trial type interaction reached significance [F(1,110) = 9.75, p < 0.05, ηp2 = 0.08]. The results of the ANCOVA revealed that this interaction remained significant after controlling for age, gender and BMI [F(1,102) = 6.72, p < 0.05, ηp2 = 0.06]. The statistical power of this interaction was 72.8%. Post hoc analyses confirmed a significant difference in error rate between Val/Val active and Val/Val inactive participants only for the incongruent trials (p < 0.01) and a significant difference between Val/Val inactive participants and Met carrier inactive participants only for the incongruent trials (p < 0.01) (see Figure 3). A last ANCOVA on error rates with Visual acuity and Depression as additional covariates, pointed out no significant effect of Visual acuity [F(1,100) = 0.25; p > 0.61] and a significant effect of Depression [F(1,100) = 4.03; p < 0.05; ηp2 = 0.05]. The interaction between PA × BDNF polymorphism × trial type remained significant despite controlling for these two additional covariates [F(1,100) = 6.70; p < 0.05; ηp2 = 0.06].

This study examined the interaction between PA and the BDNFVal66Met polymorphism on controlled inhibitory performance in older adults. Our hypothesis was that PA would interact with the Val allele on cognitive performance that involved conflict monitoring and controlled inhibition (i.e., the incongruent trials in our experimental task). The classical detrimental effect of no-dimensional overlap between the direction and location of the stimuli (i.e., an arrow) was clearly reproduced and demonstrated using RT and error rates. The dual-route model states that incongruent trials require more time to inhibit an automatic response, which leads to slower RT and a higher percentage of errors because of the failure of the inhibition process in some trials (Ridderinkhof, 2002). However, we did not observe a clear significant interaction between PA and polymorphisms on the average RT for trials that involved controlled inhibition, but the interaction reached significance on error rates. Specifically, the interaction demonstrated that inactivity was associated with a significant increase in error rate only for inactive Val-homozygous participants in the incongruent condition. This interaction confirms that PA modulates the relationship between the BDNF polymorphism and controlled inhibition as measured using error rates for incongruent trials. More precisely, it suggests that PA and BDNF share a common neurophysiological mechanism that influences controlled inhibition. The most plausible mechanism that could play this role is the neurotrophic influence of BDNF on prefrontal neurons. Both exercise and the Val allele of the BDNF polymorphism increase the concentration of BDNF in the brain (Cotman and Berchtold, 2002), consequently they must interact at a structural and/or a functional level in some regions of the brain, in this case prefrontal regions. Then, the interaction between PA and BDNF polymorphism observed in the present study gives a supplementary argument for the mediation role of BDNF in the causal link between exercise and cognition.

In addition, the present results confirm and extend the results obtained by Thibeau et al. (2016): inactive Val-homozygous exhibited a poorer performance in controlled inhibition than Met carriers and active Val homozygous. In the current study, the degraded performance corresponded to a higher error rate for incongruent trials in a Simon-like task. Thibeau et al. (2016) observed very similar results using a Stroop task and a Hayling sentence completion task. These two studies strongly suggest that Val-homozygous individuals are more sensitive to PA level. Inactivity seems very deleterious for Val-homozygous carriers. Our results should be interpreted in light of the fact that physical inactivity is a new phenomenon from an evolutionary point of view. In contrast, the human genome is a product of a long evolutionary process that has remained stable for 10.000 years (Ratey and Loehr, 2011) and was built for hunters with periods of intense PA balanced with periods of rest. Therefore, the wild type Val/Val individuals could suffer more strongly from the effects of a sedentary lifestyle than Met carriers.

But why Met carriers did not demonstrate any positive effect of PA or negative effect of inactivity? The Met allele of the BDNF polymorphism is generally viewed as a deleterious variant because the Met allele reduces BDNF secretion in response to neuronal stimulation (Egan et al., 2003; Kleim et al., 2006). It is important to note that this deleterious effect of the Met allele was not observed in the present experiment. However, it could be hypothesized that BDNF secretion level in Met carriers may not increase sufficiently in response to regular PA, and the BDNF rate would remain too low to benefit cognitive performance. Another mechanism may explain the similar performance of controlled inhibition in active and inactive Met carriers. Brown’s hypothesis (Brown et al., 2014) is that the mature BDNF form is altered and inefficiently binds to the BDNF receptor (TrkB) in BDNF Met carriers, but the precursor form of BDNF (proBDNF) remains unaltered in BDNF Met carriers. Unfortunately, proBDNF binds its receptor (p75), and this link induces apoptosis. Therefore, Met carriers with high levels of PA may undergo higher levels of apoptotic changes in the brain because of the increasing proBDNF levels. Increasing mature BDNF levels in response to increased PA will not be sufficient to improve cognition. Notably, the Caucasian population exhibits an approximately 70% distribution of Val-homozygous (Shimizu et al., 2004). Val-homozygous carriers in our population exhibited the greatest benefit from the effects of PA, and 30% of the population (the Met carriers) was less or not sensitive to the effects of PA on controlled inhibition performance. Therefore, the cognitive improvements reported by epidemiological and interventional studies in primarily Western countries may be the result of the high percentage of BDNF Val-homozygous allele carriers in this population.

The present study emphasizes the interaction between BDNF polymorphism and PA as a set of life habits that is recognized as beneficial for brain health. We demonstrated that PA interacted with BDNFVal66Met on controlled inhibitory performance. Therefore, it is imperative to control for PA level to examine the putative impact of the BDNFVal66Met polymorphism on cognition in elderly populations.

Five main limitations may be noted. First, this study was cross-sectional. The interpretation of these results as a causal relationship between PA and cognition is limited by the study design; cross-sectional study do not allow establishing a causal relationship because better cognitive functioning may promote an individual to engage in PA. Future studies should use a randomized-controlled trial approach to determine whether BDNF Val-homozygous carriers benefit from exercise training programs. Second, the measure of PA was based on the self-report of participants. These factors induce the possibility of a biased reporting from the subjective perception of participants about their PA level. However, the HLAQ questionnaire and the NASA/JSC Physical Activity Scale are widely recognized and well-validated instruments. Third, some haplogroups may present protective or compensative effects to certain alterations in protein secretion that is influenced by a gene polymorphism. Tsai et al. (2008) demonstrated that the Met allele protected Asian-type populations from the deleterious effects of cognitive aging. It could be important to conduct replication of the present study and Thibeau’s et al. (2016) study in aging people from other ethnic group such as Asian-type population. Fourth, only one gene related to executive control was examined in the present study. However, executive control and the association between physical fitness and cognition are likely to be underpinned by many genes. Several genes have been identified in addition to BDNF: DRD4, DAT1, MAOA, COMT, APOE, each with a relatively small effect (for a review, Goldberg and Weinberger, 2004). It could be interesting to examine the synergistic and antagonistic actions of these different genes on cognition, and focus on those that share a common mechanism that may explain the causal link between exercise and cognition. However, these multi-polymorphisms studies require very large sample of participants.

The present report showed a significant interaction between PA and a BDNF gene polymorphism on controlled inhibitory performance. These data demonstrated that the association between PA and controlled inhibition was modulated by a BDNF polymorphism. Inactivity appeared more deleterious for Val-homozygous than Met carriers. This result improves our understanding of the neurophysiological mechanisms underlying the positive effect of PA on cognitive decline and provides brain BDNF as a convincing modulating candidate in this relationship.

MA: Scientific coordinator of the study, design of the protocol, data collection, data treatment, writing of the article, statistical analyses. AC: Data collection, data treatment, writing of the article, statistical analyses. CA: Design of the protocol, data collection, data treatment, writing of the article. MR-B: Data collection, determination of genetic polymorphism, writing of the article. CC and DF: Design of the cognitive task, data treatment, writing of the article. NA: Design of the protocol, data treatment, writing of the article.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.