The study was designed as an 18-month controlled intervention. The study was approved by the ethics committee of Otto-von-Guericke University, Magdeburg, and all subjects signed a written informed and they did not receive payment for their participation. Sixty-two healthy elderly individuals (63–80 years) recruited via announcements in local newspapers were screened for the study. The exclusion criteria were claustrophobia, tinnitus, metal implants, tattoos, diabetes mellitus, depression (Beck-Depressions Inventory, BDI-II > 13), cognitive deficits (Mini-Mental State Examination, MMSE < 27), neurological diseases and regular exercising (≥1 h/week). On the basis of these criteria, 10 subjects were excluded. The remaining 52 participants were then randomly assigned to either the dance or the sport group by using the website www.randomization.com and controlling for age, MMSE status and physical fitness. Assessments were performed at baseline, after 6 and after 18 months of training (Figure 1). Twenty-two participants completed the entire intervention and all measurements. Table 1 provides detailed demographic data for these participants. No group differences regarding the demographic data were found.

The interventions were separated into two periods. In the first period, the subjects trained twice per week in 90-min sessions for 6 months. For practical reasons (availability of participants and trainers), the second 12-month intervention period comprised a reduced training frequency of once per week. Both intervention programs were performed in a group context with music to control for psychosocial interactions. Conditional load was examined by recording the pulse values during the training sessions and by calculating the individual training heart rate according to Karvonen et al. (1957) with a factor 0.6 for extensive endurance training. Hence, the two training programs were comparable in terms of intensity, duration and frequency. Both were supervised by experienced instructors.

Cardiovascular fitness as measured by the PWC 130 (Table 2, relative physical capacity) did not differ between groups at baseline. Furthermore, the PWC 130 score did not increase significantly throughout the time course of either intervention.

A significant main effect of time was observed for verbal short-term memory (VLMT early recall; F_(2,19) = 6.438, p = 0.004, η² = 0.253), verbal long-term free recall (VLMT late recall; F_(2,19) = 3.387, p = 0.049, η² = 0.244), verbal long-term recognition (VLMT recognition; F_(2,19) = 5.352, p = 0.009, η² = 0.220) and attention reaction time (subtest flexibility; F_(2,19) = 19.156, p < 0.001, η² = 0.489). No significant time × group interactions emerged.

Post hoc pairwise comparisons showed significant improvements from baseline to 18 months and from 6 to 18 months in all three VLMT subcategories in both groups. Regarding TAP reaction times, a significant improvement was seen in the comparison of the baseline to the 18-month data in both groups.

Plasma levels of BDNF were analyzed in blood samples before the onset of training as well as 6 months and 18 months after the onset of training. Absolute BDNF plasma levels are summarized in Table 2. The intraindividual changes in BDNF level revealed a significant increase in the dancing group, whereas the intraindividual BDNF level remained constant after 6 months of training in the sport group. No further changes in BDNF occurred after 18 months in any group (Figure 2).

A significant group × time interaction was observed in the left precentral gyrus and the right parahippocampal. Post hoc t-tests between the baseline and 6-month data showed a significant increase in gray matter volume in the left precentral gyrus of only the dancers. In the comparison of the baseline and the 18-month data in addition to the precentral gyrus, the dancers exhibited a significant gray matter volume increase in the right parahippocampal gyrus, which was also the only significant change in the interval from 6 to 18 months. Thus, the volume increase in the precentral gyrus emerged after 6 months and remained stable over the remaining dance training interval, whereas the change in the parahippocampal gyrus occurred during the later training interval only (Figure 3).

The precentral gyrus is essential for the control of voluntary motor functions. The increase in gray matter volume in the precentral gyrus in the dance group may, therefore, have been based on the complex and ever-changing movement patterns that the dancers had to perform. These movements required the simultaneous coordination of several parts of the body in different directions and adjustment to the varying rhythms of the music (polycentric and polyrhythmic). Reflecting on these complex coordination requirements, Brown et al. (2006) have reported dance-induced activations in the putamen, the primary motor cortex and the supplementary motor area (SMA), as shown by PET. Other studies have indicated an association between coordination demands (e.g., balancing, juggling) and neuroplasticity in the precentral region (Boyke et al., 2008; Taubert et al., 2010). Hence, the dance-related volume increase in this area was consistent with expectations based on the literature.

The parahippocampal gyrus is part of the outer arc of the limbic system and plays an important role in working memory and episodic memory retrieval (Pantel et al., 2003). According to Bliss and Lomo (1973), the parahippocampal gyrus constitutes the interface between memory and the experiential consciousness of the present, because it is interconnected by the perforant tract both to regions of the frontal lobe, which are associated with working memory, and to the hippocampus, the central structure in episodic memory encoding and spatial navigation. Many VBM studies have reported an age-related volume loss in parahippocampal regions (Tisserand et al., 2002). Furthermore, Echávarri et al. (2011) have suggested that parahippocampal atrophy is an early biomarker of AD. The observation that engaging in a dance program for a longer period can induce neuroplastic processes in this crucial memory region, therefore, particularly encouraging in terms of developing prevention strategies.

The observed volume increases in the two brain regions developed at different times. Dancing led to a volume increase in the motor areas after 6 months (see also Rehfeld et al., 2016), which remained stable during the subsequent 12 months. The dance-associated volume increase in the parahippocampal gyrus emerged later and was observed only in the 18-month data. The different temporal dynamics in the evolution of the two brain regions may be related to differences in the underlying cellular mechanisms. Animal research has suggested that angiogenesis and the formation of new dendrites (Thomas et al., 2012) occur rapidly, whereas changes in the neuropil occur much slower (Black et al., 1990). In humans, rapid increases in gray matter volume in the prefrontal regions have been observed after only 2 weeks of motor learning in younger adults (Taubert et al., 2010). To induce neuroplasticity in the hippocampus, longer training periods have been recommended (Erickson et al., 2012; Niemann et al., 2014).

Interestingly, the aforementioned volume increases from our study are in contrast to the results of Hänggi et al. (2010), which revealed that professional ballet dancers have decreased gray matter volumes in the left premotor cortex, SMA, Putamen and superior frontal gyrus and the results of Hüfner et al. (2011), which report reduced volumes in several brain regions including the anterior hippocampal formation in professional dancers and slackliners compared with non-professionals. However, these studies were cross-sectional observational studies which compared the brains of professional dancers to those of non-professionals. In addition, our dance group consisted of old aged novices. It has been shown that learning of a new skill first leads to recruitment of additional neural resources. Later, when the skill becomes more automatic, less neuronal resources are needed which may lead to volume decreases in those with long-term experience. Based on these initial conditions (novices, older adults) and our special designed dancing training program, which required constantly new learning of movement patterns, it is possible that effects of specialization induced volume decreases as reported by Hänggi et al. (2010) are not observed.

It is generally thought that motor training initially induces brain volume increases. However, prolonged training leads to automatization, which may have the opposite effects on cortical volume, because less cortical control is needed after the motor skills have been fully established (Taube, 2008). Our dance training program, therefore, was specially designed to avoid such automatization, which may explain why, at least within 18 months, no cortical volume decreases were observed in our study.

Cognitive functions also showed nonlinear development, whereby verbal memory increased only during the second training period. Regarding attention performance, significant improvements were observable after only 6 months in both groups (Rehfeld et al., 2016). These findings support prior reports regarding the beneficial effects of physical interventions on neuropsychological tests (Bamadis et al., 2014). However, in cognitive ability data, in contrast to the brain data, no group differences emerged. Others have reported superior effects of combined cognitive and physical training as opposed to single interventions (Oswald et al., 2006). We will extend our interventions further to test whether group differences in cognition might emerge at even later time points.

Although VBM is an imaging modality that reveals volume changes in the brain, this technique does not allow causal conclusions regarding the underlying neurophysiological processes. Neurogenesis, synaptogenesis and angiogenesis are just a few of the mechanisms that have been suggested to be the basis of brain volume changes (Zatorre et al., 2012). As mediators of the effects of cardiovascular fitness on the brain, growth factors such as BDNF, insulin-like growth factor (IGF) and nerve growth factor (NGF) are being studied (Kirk-Sanchez and McGough, 2014). However, in our study, in contrast to previous ones (Erickson et al., 2011; Maass et al., 2015), no differences in cardiovascular fitness were present between groups, and the fitness levels did not change during the interventions. The latter observation was probably related to our control of the individual heart frequency, which we aimed to maintain in the aerobic zone. However, BDNF changes have also been associated with physical activity, social interaction and positive stress (Mattson, 2008), and not all studies have observed a BDNF increase after cardiovascular training (Vital et al., 2014). Finally, animal research has suggested that coordination but not endurance training induces synaptogenesis and glial changes (Black et al., 1990). Together, the named additional factors that drive BDNF secretion may have been more crucial during dancing than during fitness activities, thus explaining why only dancers showed a BDNF increase in the first 6 months. The observation that BDNF levels returned to baseline in the following 12 months while volume increases were simultaneously observed in the parahippocampal gyrus, however, indicates that there must be other factors involved in adult brain plasticity than the ones represented by BDNF levels in the peripheral blood.

Regarding neurobiological mechanisms of exercised-induced plasticity also concept of brain reserve (Satz et al., 2011) should be considered. The concept of brain reserve describe individual differences in an increased baseline adaptive neuroplasticity, which provide greater dynamic capacity for remodeling cortical circuits to different stressors (Barulli and Stern, 2013; Freret et al., 2015).

The results of our study suggest that a long-term dancing intervention could be superior to repetitive physical exercise in inducing neuroplasticity in the aging human brain. We presume that this advantage is related to the multimodal nature of dancing, which combines physical, cognitive and coordinative challenges. To our knowledge, this is the first longitudinal, randomized study to recommend dancing programs as a means of preventing gray matter and cognitive decline in the elderly. Further research is needed to clarify in greater detail the temporal dynamics and the underlying neurobiological mechanisms of dance-induced neuroplasticity and whether this intervention truly has the potential to reduce the risk of neurodegenerative diseases such as Alzheimer’s.