The volume of the brain declines with aging, starting in midlife (Scahill et al., 2003). This reduction in brain volume is presumed to reflect decreases in neuron volume and number (Gogniat et al., 2021). One of the mechanisms underlying the brain-protective effects of PA is the preservation of brain volume, or even an increase in brain volume. Observational studies have found that brain volume, particularly that of the hippocampus, is preserved in more active people (Erickson et al., 2010; Flöel et al., 2010; Maasakkers et al., 2021). PA is also associated with preservation of cortical thickness (Walhovd et al., 2014; Lee et al., 2016).

Many interventional trials have also shown the positive impact of PA on brain volume. Some studies showed that PA interventions reduced the development of brain atrophy compared with controls (Best et al., 2015), and some studies even demonstrated an increase in brain volume after an exercise intervention (Colcombe et al., 2006; Tao et al., 2017) in older adults. A systematic review of the effects of exercise intervention on the hippocampus concluded that aerobic exercise helped to preserve hippocampal volume by preventing its decrease, while some of the included studies even showed an increased volume (Firth et al., 2018). The characteristics of these representative studies are summarized in Table 2.

Although the precise mechanisms remain to be elucidated, the neuroprotective effects of PA may be exerted through increased brain-derived neurotrophic factor (BDNF) and blood flow or reduced oxidative stress and amyloid accumulation (as discussed below). Recently, the MAPT study reported that more physically active individuals had lower blood concentrations of neurofilament light chain, a well-established biomarker of neurodegeneration (Raffin et al., 2021). These results suggest that PA may ameliorate neurodegeneration. Moreover, decades of research have shown that adult neurogenesis persists throughout life, although it declines with aging. Hippocampal neurogenesis is reported to be crucial in learning and memory in rodent experiments (Babcock et al., 2021). PA may accelerate neurogenesis, particularly that of the hippocampus (Leal-Galicia et al., 2019). The increased brain volume may thus be due to enhanced neurogenesis. However, human evidence, particularly that from older adults, is scarce.

White matter plays a crucial role in cognition by connecting different brain regions to enable efficient signal transmission. White matter in adult brains exhibits plasticity involving myelin formation and remodeling (Sampaio-Baptista and Johansen-Berg, 2017). Observational studies suggest that higher PA is associated with better white matter integrity as measured by diffusion tensor imaging on magnetic resonance imaging (MRI; Buchman et al., 2018; Franchetti et al., 2020; Wolf et al., 2020). Several interventional studies have found evidence of improved white matter integrity (Voss et al., 2013; Burzynska et al., 2017; Clark et al., 2019; Colmenares et al., 2021), but another study failed to show an effect (Venkatraman et al., 2020).

Small vessel disease is represented by white matter lesions (WMLs). T2-weighted or fluid-attenuated MRI images visualize WMLs as diffuse high-signal areas. WMLs have been linked to cognitive impairment (Alber et al., 2019). Although the reported results are not in complete agreement, several cross-sectional and longitudinal studies have shown that PA is associated with fewer WMLs (Torres et al., 2015; Moon et al., 2018). However, no relevant interventional studies have been reported.

A systematic review concluded that PA reduced stroke risk (Wendel-Vos et al., 2004). Therefore, PA may contribute to reducing the risk of vascular cognitive impairment.

Brain-derived neurotrophic factor is a neurotrophin that influences neuronal survival, differentiation, synapse generation, and long-term potentiation (Phillips et al., 2014). Decreased BDNF level is associated with neuropathological conditions, including mild cognitive impairment (MCI; Shimada et al., 2014). The blood concentrations of BDNF decrease with aging, and there is a strong correlation between the brain and blood levels of BDNF (Knaepen et al., 2010). A recent systematic review concluded that PA interventions increased the plasma level of BDNF in individuals with MCI or AD dementia (Ruiz-González et al., 2021). A pair of studies also showed that exercise increased plasma BDNF levels in cognitively normal older people (Müller et al., 2017; Rehfeld et al., 2018). The representative studies are summarized in Table 3.

Although the precise cause of the exercise-associated plasma BDNF increase has not yet been fully elucidated, some studies have shown increased BDNF in samples from the internal jugular vein after acute training (Rasmussen et al., 2009) and chronic regular training (Seifert et al., 2010), possibly suggesting that exercise-associated BDNF may have a central origin. Rasmussen et al. (2009) estimated that 70%–80 of circulating BDNF was derived from the brain and that the remaining levels were derived from peripheral sources. Thus, the question of whether circulating BDNF levels reflect those in the human brain warrants further study.

Circulating insulin-like growth factor 1 (IGF-1) passes through the blood–brain barrier, exerts neuroprotective effects, and induces synaptic plasticity (Sonntag et al., 2005). The upstream IGF-1 signaling pathway also induces BDNF expression (Yan et al., 2011). Interventional trials demonstrated that aerobic exercise increased plasma IGF-1 in older adults (Kang et al., 2020), including those with MCI (Baker et al., 2010) and AD (Stein et al., 2021). A systematic review concluded that resistance training also increased IGF-1 levels in participants, including older adults (Jiang et al., 2020).

Cerebral blood flow decreases with aging and may be associated with cognitive decline (Tarumi and Zhang, 2018). This reduced blood flow may reflect a decreased cerebral metabolic rate (Marchal et al., 1992). Increased sympathetic nervous activity and impaired vasodilation may also contribute to the age-associated decline in cerebral blood flow. Cardiac output tends to decline with aging and may also be associated with reduced blood flow in the brain (Tarumi and Zhang, 2018). Regular exercise may help to counteract the effects of these aging-associated changes in cerebral blood flow.

Angiogenesis may also be involved in the ability of PA to increase blood flow (Swain et al., 2003). Vascular endothelial growth factor (VEGF) is an important regulator of angiogenesis, and several animal studies have shown increased VEGF levels with physical exercise. However, studies in humans have been inconclusive (Vital et al., 2014). Observational studies have found that sedentary older adults had lower cerebral blood flow than active older adults (Rogers et al., 1990; Thomas et al., 2013; Knight et al., 2021; Li et al., 2021). Recently, a small randomized controlled trial involving a 1-year aerobic exercise intervention reported increased cerebral blood flow in the exercise group (Tomoto et al., 2021).

Alzheimer’s disease is a major neurodegenerative disease and is the leading cause of dementia. The main pathological features of AD are accumulation of amyloid β (Aβ; senile plaques) and intracellular accumulation of hyper-phosphorylated tau (neurofibrillary tangles). PA has been associated with reduced risk of AD (Barnes and Yaffe, 2011). Extensive basic research results suggest that exercise may be associated with increased Aβ clearance (Vecchio et al., 2018), reduced Aβ production (Adlard et al., 2005), enhanced tau degeneration, and decreased tau phosphorylation (Brown et al., 2019). While animal studies have established that high PA leads to better amyloid profiles (Brown et al., 2019), human studies investigating the effects of PA on brain pathologies have been very difficult to do. However, recent advances in AD-related biomarkers are enabling research into the effects of PA on AD pathologies in human subjects.

Although cerebrospinal fluid biomarkers and amyloid PET imaging are the most frequently applied modalities in this field, the applicability of plasma biomarkers is gradually improving. Many observational studies involving measures of Aβ in cerebrospinal fluid (Liang et al., 2010; Brown et al., 2013, 2017; Law et al., 2018), plasma (Stillman et al., 2017), and PET (Head et al., 2012; Brown et al., 2013, 2017; Jeon et al., 2020; Treyer et al., 2021) have reported better profiles of amyloid-related biomarkers in active people but inconsistencies remain (de Souto Barreto et al., 2015; Jeon et al., 2020; Palta et al., 2020; Stojanovic et al., 2020). The representative studies are summarized in Table 4.

Observational studies, mainly retrospective in nature, have suggested that individuals with higher PA tend to have biomarker profiles indicative of lower Aβ deposition in the brain (Frederiksen et al., 2019). However, interventional studies with AD biomarkers are rare and have had relatively small sample sizes. Most obtained inconclusive results, although some showed favorable effects (Moniruzzaman et al., 2020). More studies are warranted.

Researchers have investigated the effects of PA on tau-related biomarkers. Some cross-sectional studies involving cerebrospinal fluid biomarkers (Liang et al., 2010) and tau PET (Brown et al., 2018) showed lower tau profiles in participants with high PA levels, but a longitudinal study did not find an interaction between PA and tau over time.

Acute exercise increases oxidative stress, but regular PA is expected to regulate the cellular redox state of the brain, and PA-induced redox adaptation may contribute to the neuroprotective effects of PA (Radak et al., 2016). Animal studies showed that PA protected the brain from oxidative stress (Radák et al., 2001; Radak et al., 2006). A systematic review concluded that antioxidant indicators tend to increase and pro-oxidant indicators tend to decrease after resistance training in humans (de Sousa et al., 2017). However, another systematic review found no effects of resistance training on molecular oxidation and antioxidant capacity markers (Cuyul-Vásquez et al., 2020).

Nonetheless, studies of individuals with dementia or MCI are rare. Only Jensen et al. (2019) have measured the plasma level of 8-isoprostane, an oxidative stress marker, in AD participants before and after an exercise intervention and reported only non-significant changes.

Inflammation is one of the major mechanisms of neurodegeneration (Onyango et al., 2021). PA may have anti-inflammatory potential (Mee-Inta et al., 2019). Cross-sectional observational studies have reported lower levels of inflammatory markers (IL-6 and C-reactive protein) in active older adults (Reuben et al., 2003) and determined that active women had lower levels of the plasma inflammatory biomarker TNF-α (Castells-Sánchez et al., 2021). Another observational study also found low levels of IL-6, C-reactive protein, and TNF-α (Colbert et al., 2004).

Plasma IL-6 levels decrease in response to aerobic exercise in both MCI (Nascimento et al., 2014) and AD (Abd El-Kader and Al-Jiffri, 2016). In addition, levels of C-reactive protein, a representative inflammatory marker, decrease in response to exercise interventions (Muscari et al., 2010; Alghadir et al., 2016). A systematic review concluded that exercise was associated with a decrease in C-reactive protein levels regardless of age or sex (Fedewa et al., 2017). Another recent systematic review of exercise interventions in people with MCI or dementia reported that exercise significantly decreased the levels of IL-6 and TNF-α (Huang et al., 2021).

Functional MRI enables functionally connected brain regions to be identified by measuring simultaneous activations via the blood oxygen level-dependent (BOLD) signals of spatially distinct regions. A systematic review by Li et al. (2017) suggested that aerobic exercise increases functional connectivity in the default mode network, which is associated with memory and abstract thinking. Recently, Bray et al. (2021) reviewed the effect of physical exercise on functional brain network connectivity in older adults with and without cognitive impairment and concluded that physical exercise increases functional brain network connectivity.

Physical activity might somehow help to recruit new neuronal circuits that would be involved in processing tasks. An interesting fMRI study suggested that 6-week dance exercise training led to the involvement of the motor-related network during highly cognitive-demanding memory tasks, possibly as a compensatory mechanism. Exercise may accelerate the involvement of new networks in the cognitive process (Ji et al., 2018).

Event-related potentials are generated in response to specific events or stimuli such as audio sound. Studies involving the P3 component, the major focus of studies of event-related potentials, showed that exercise increases the amplitude of P3, possibly suggesting that exercise enhances the allocation of neural resources as part of a compensatory mechanism (Chang et al., 2015). However, the results are controversial (Gajewski and Falkenstein, 2018; Alatorre-Cruz et al., 2020).

Exercise may increase neuronal efficiency. An fMRI study demonstrated that an exercise group had lower BOLD signals in the hippocampus and para-hippocampal gyrus compared with a non-exercise group during a memory-encoding task. Because the BOLD signal reflects brain activity in a specific region, lower BOLD signals during a certain task mean that the task was performed with less burden, which probably reflects higher neuronal efficiency in the hippocampus and para-hippocampal gyrus in the exercise group (Friedl-Werner et al., 2020). Another trial also showed that a 12-week exercise intervention reduced neuronal activation in several brain regions, including the prefrontal cortex, during a memory task (Nishiguchi et al., 2015).

Synaptic plasticity (long-term potentiation and long-term depression) is a biological model for learning and memory processes. Animal studies have demonstrated that PA controls synaptic plasticity (Bettio et al., 2019). Synaptic plasticity can be non-invasively measured in humans via a combination of transcranial magnetic stimulation and recording of motor-evoked potentials. A recent study reported that PA was associated with long-term depression-like neuroplasticity in older adults (Smith et al., 2021).