In a rodent study, Black et al. (1990) contrasted effects of aerobic exercise and motor learning. Specifically, while 30 days of wheel running produced an increase in capillary density, 30 days of traversing an acrobatic maze produced an increase in the number of synapses per purkinje cell (the large, primary nerve cell in the cerebellum). Subsequent studies have demonstrated similar changes in blood vessel density in motor cortex using both histological (Kleim et al., 2002b) and magnetic resonance imaging (MRI) techniques (Swain et al., 2003).

Some of the most compelling evidence of exercise-mediated brain changes has been found in the hippocampus, a brain structure involved with memory as well as stress regulation (Kim and Diamond, 2002; Small et al., 2011). As early as 1999 it was demonstrated that wheel running dramatically increased the number of new neurons in the hippocampus of mice (van Praag et al., 1999a). Pereira et al. (2007) replicated this finding and went on to demonstrate that the number of new neurons correlated with increases in CBV, measured using contrast-enhanced MRI. The same paper showed a similar increase in CBV in a small group of middle-aged human subjects after 12 weeks of exercise training.

Chaddock et al. (2010) showed that preadolescent children with higher V.O2max scores had larger hippocampal volumes (as determined by MRI scans) than those with lower V.O2max scores. Erickson et al. (2009) showed a correlation between the volume of the hippocampus and cardiovascular fitness (peak V.O2) in a cohort of 165 older adults. A follow-up study by the same authors (Erickson et al., 2011) demonstrated that 1 year of aerobic exercise increased the volume of hippocampus by 2% in elderly adults, while controls who underwent 1 year of stretching exercises exhibited a 1.4% decrease in hippocampal volume. Similarly, Pajonk et al. (2010) reported a 12–16% increase in hippocampal size in a small group of exercising schizophrenic patients as well as in matched controls.

In summary, aerobic activity reliably induces structural change in hippocampal volume and vasculature. This high degree of plasticity is perhaps not surprising given the hippocampus’ purported role in the rapid encoding of episodic memory (McClelland et al., 1995). It is also notable that the hippocampus displays dramatic volume changes in disease states such as Alzheimer’s disease and depression (Steffens et al., 2000; Butters et al., 2008).

All of the afore-mentioned studies, including the neuroimaging studies, limited their investigations to specific regions of interest. However, it is possible that some effects of aerobic exercise on brain structure occur via a diffuse mechanism (such as globally altered blood volume) that would not produce localized effects. Some whole-brain, voxel-based morphometry (VBM) studies have reported an association between physical activity levels (Flöel et al., 2010) or an exercise intervention (Colcombe et al., 2006) and large clusters of increased gray matter density in frontal, temporal, and cingulate areas of the brain. Another recent study has shown that aerobic activity levels in elderly human subjects correlate with both the number and tortuosity of blood vessels throughout the brain using the relatively novel technique of magnetic resonance angiography (Bullitt et al., 2009). These studies suggest that while some regions of the brain may show localized volume changes in response to aerobic exercise, there may also be more global changes.

The study of structural brain change in humans is currently limited to the relatively macroscopic measurements that are possible using neuroimaging techniques. This contrasts sharply with histological studies in animals that can measure cell number, density of dendrites, or even the size and shape of dendritic spines (See Figure 1). Many neuroimaging studies report changes in “gray matter density” or the overall volume of a given structure. These measures are commonly derived from segmenting T1-weighted MR images using the relative intensities of different voxels. Although these image intensities depend on the underlying tissue structure, they do not relate in a straightforward way to specific tissue components (e.g., cell density). The magnitude of these changes are typically in the range of 1–8% (Draganski et al., 2004; Scholz et al., 2009; Erickson et al., 2011), but it is not possible to draw conclusions on which microscopic structures within gray matter are exhibiting an increase in volume.

Some insight into this question can be gleaned from animal studies. Figure 2 provides an approximate breakdown of the components of cerebral gray matter by volume based on several histologic studies (Cherniak, 1990; Syková, 1997; Braitenberg and Schüz, 1998; Chklovskii et al., 2002; Kleim et al., 2007). Note that these ratios should be taken only as a rough guide. The composition of cortical gray matter varies considerably between brain areas (Chklovskii et al., 2002) and species (Herculano-Houzel et al., 2006; Sarko et al., 2009). However, it has been consistently shown that well over 50% of gray matter is composed of the tangle of axonal, dendritic, and glial processes known as the neuropil, while vasculature accounts for no more than 5% of gray matter volume. Cell bodies account for less than 20% and the interstitial space between cells, synapses, and vessels accounts for over 20%.

These ratios give us some feel for the magnitude each compartment would need to change to produce the volume changes reported in recent imaging. For example a 20% increase in the neuropil volume would be sufficient to produce the 8% global gray matter changes that have been reported. However if the change were limited to vasculature, a near doubling would be required to produce an 8% change in gray matter volume.

Angiogenesis in response to exercise is well-studied in muscle tissue (Prior et al., 2004). Angiogenesis can occur via a splitting process known as intussusception or a sprouting process in which a new branch sprouts from one capillary and merges onto another (Makanya et al., 2009). Studies on changes in the capillary density in rats were some of the first to distinguish between the effects of exercise versus learning or exposure to complex environments (Black et al., 1990). A replication and elaboration is provided in a follow-up study that dissociates increases in capillary number from increases in capillary density (Isaacs et al., 1992). Exposure to a complex environment produces an expansion of the volume of the molecular layer in the cerebellum, while the density of blood vessels remains constant. This necessarily involves an increase in the number of capillaries. However, the voluntary exercise condition results in an increase in the density of capillaries, while the volume of the molecular layer remains constant. Thus the ratio of blood vessel volume to other components of the layer has increased.

This dissociation between the effects of exercise and enrichment on angiogenesis was subsequently replicated in the rat motor cortex (Kleim et al., 1996, 2002a). The basic result of increased vasculature with exercise has also been replicated many times in motor cortex as well as striatum with both histological and imaging techniques (Swain et al., 2003; Ding et al., 2005, 2006). The magnitude of increased blood vessel density reported in the original Black et al. (1990) study was approximately 20%, but the subsequent studies using different quantification methods have reported increases from 50% to over 1000%. If these measures are accurate they could account for the gray matter changes discussed above entirely within the vascular compartment.

A recent study using a small sample of macaques replicated the increase in vascular volume fraction in motor cortex with 5 months of daily treadmill exercise. Interestingly, this effect was found only in older animals (15–17 years old) while middle-aged animals (10–12 years old) showed no change in vascular volume fraction (Rhyu et al., 2010).

The mechanism by which aerobic exercise results in increased angiogenesis in the brain is not fully understood. It is believed that mechanical sheer stress on the walls of the capillaries as well as a reduction in the degree of blood oxygenation (hypoxia) may play a role (Makanya et al., 2009). Independent of exercise, hypoxia itself is a potent stimulus of angiogenesis in the brain. Patt et al. (1997) demonstrated that rats exposed to progressive amounts of hypoxia over 130 days showed angiogenesis in a wide variety of brain areas including cerebral cortex, striatum, and hippocampus. Differences in capillary density in cerebellum and the medulla oblongata failed to reach significance due to the high variability in baseline vascular density in control animals. Contrary to Isaacs et al. (1992), Patt and colleagues also found increases in capillary diameter in cerebral cortex and cerebellum. Such changes could potentially also occur in association with the mild hypoxia that can occur during aerobic activity. Differences in vessel diameter associated with aerobic activity have also been shown in a magnetic resonance angiography study of elderly human individuals (Bullitt et al., 2009).

A considerable amount of work has looked at angiogenesis in the hippocampus; however, these studies have focused on the tight relationship between angiogenesis and the growth of new neurons. This work will be discussed in more detail below.

As revealed in the graph above, astrocytes and other glial cells consume a significant proportion of neuropil volume. While glial cells are similar to the vasculature compartment in that they play an important role in the metabolic support of neurons, they also have an active role in regulating the efficacy of synapses and even directly communicating with neurons via calcium signaling (see Zhang and Haydon, 2005 for a review). Given these dual roles, we might expect astrocytes to behave like neural processes and expand their volume only in response to learning. Alternatively, we might expect them to behave more like vasculature and respond to both exercise and learning. Anderson et al. (1994) addressed this question in rat cerebellar cortex by measuring the volume of astrocytes after a period of motor learning or voluntary exercise. The study showed that while the overall density of glial processes did not differ between groups, the glial volume per purkinje cell in rats in the motor learning condition was significantly greater than that of rats exposed to voluntary exercise alone. The motor learning group also showed an overall increase in the volume of the molecular layer. This suggests that the astroglia are more similar to the synaptic and neural structure than they are to vasculature in their pattern of growth.

New neurons are generated in the vertebrate brain throughout the life span. (See Colucci-D’Amato and di Porzio, 2008 for an account of the rise and fall of the “no new neurons” dogma over the course of the twentieth century). There is general agreement that neurogenesis occurs in two areas of the mammalian brain: the sub-granular zone of the dentate gyrus within the hippocampus and the subventricular zone located in the lateral ventricles. Neurogenesis in the hippocampus, like angiogenesis, has been shown to increase with physical activity (van Praag et al., 1999a), but the creation of new neurons in the hippocampus is a delicate process. A large fraction of the new neurons created do not survive (Kempermann et al., 2010). Kronenberg et al. (2003) showed that while exercise increases the rate of neurogenesis, environmental enrichment increases the ratio of new neurons that survive and get incorporated into the lattice of the existing neural network.

Some studies have suggested that both new neural cells and new endothelial cells (used to build new capillaries) derive from the same pool of stem cells (Palmer et al., 2000). Pereira et al. (2007) demonstrated a correlation between the number of new neurons generated in mouse dentate gyrus and the degree of increase in regional CBV as measured by contrast-enhanced MRI. In a small sample of human subjects, they demonstrated a significant increase in regional cerebral blood volume (rCBV) after 3 months of aerobic exercise. Changes in rCBV were shown to correlate with increases in V.O2max scores as well as post-exercise performance on a verbal learning task. The authors argue that these data suggest that rCBV can be used as a proxy for measuring neurogenesis in mice as well as humans. This exciting possibility must be reconciled with the fact that there are several regions of the brain in which angiogenesis has been observed, such as the cerebral cortex and cerebellum, which most researchers believe are not capable of neurogenesis. Some authors have argued that neurogenesis does occur in the cerebral cortex (Gould et al., 1999) and in the cerebellum (Carletti and Rossi, 2008). This would provide a possible resolution to this apparent contradiction in Pereira and colleagues’ theory. But the idea of neurogenesis outside of the hippocampus and the subventricular zone remains very controversial (Rakic, 2002; Ming and Song, 2005).

The majority of the literature on structural change with aerobic activity has focused on the region of the brain that contains the majority of cell bodies, synaptic connections and vasculature, i.e., the “gray matter.” The effects of exercise on “white matter,” the large mass of mostly myelinated axons near the center of the brain, have been overlooked until very recently. An early attempt to look at white matter volume was conducted by Colcombe et al. (2004) using VBM techniques on segmented white matter images. They reported an anterior cluster of increased white matter after 6 months of exercise in a group of elderly adults. However, the use and interpretation of VBM on white matter volumes are somewhat controversial (Draganski et al., 2006; Smith et al., 2006).

The past 5–10 years have witnessed significant advancements in our ability to study the connectivity between brain areas embodied by the white matter. The advent and popularization of diffusion tensor imaging (DTI) have allowed researchers to study the structural integrity of white matter directly (see Johansen-Berg, 2010 for review). Concurrently, another area of study known as “functional connectivity” has grown in popularity and sophistication. This technique uses functional MRI data to explore temporal correlations between brain areas (see Smith et al., 2011 for review). The strength of these correlations is thought to provide a measure of the degree to which different regions of the brain are working in concert as a “functional network.”

DTI changes associated with aerobic activity have recently been examined in a 1-year fitness training study on elderly adults. Heo and Kramer (2010) show a correlation between white matter integrity and changes in V.O2max scores in frontal and temporal white matter tracts. Interestingly, the change in white matter integrity for the aerobic training group did not significantly differ from a control group that participated in 1 year of non-aerobic exercise, suggesting that aspects other than aerobic exercise contributed to the observed change. Differences in resting functional connectivity associated with fitness level were reported in a recent study by Voss et al. (2010). Aging has been previously shown to result in a decrease of functional connectivity in the so-called “default mode network” (DMN, Andrews-Hanna et al., 2007). The DMN is a well-studied set of brain regions that show a decrease in activity when external processing demands are increased. Voss and colleagues demonstrate that some of the functional connections within the DMN (particularly the connection between the posterior cingulate gyrus and the middle frontal gyrus) exhibit a positive correlation with V.O2max score, controlling for age. They also demonstrated several positive correlations between functional connectivity and cognitive measures of executive function and spatial memory.