First, we searched PubMed¹ and Google Scholar² for articles with combinations of the following keywords: “fMRI,” “PET,” “neuroimaging,” “motor,” “finger,” “tapping,” “flexion,” “movement,” “hand,” “aging,” and “older.” We then selected only publications that met our criteria for inclusion in the ALE analyses (please see below). To ensure no other potential papers were overlooked, we searched these publications’ reference lists and reviewed publications that had cited these (via Google Scholar).

For the ALE analysis of both groups, young adults and older adults, activation foci were included if the publication met the following criteria: (1) subjects were healthy, right-handed adults; (2) the average ages of young adult cohorts were between 18 and 34 and/or the average ages of older adult cohorts were between 49 and 82 (while this range extends below the border between adulthood and older age indicated in neurocognitive aging (i.e., age 60; Reuter-Lorenz and Lustig, 2005), our range was chosen from the age ranges (or, when range was not reported, one standard deviation below/above the mean) of the individual subjects who participated in the four studies of the aging motor system (Calautti et al., 2001; Mattay et al., 2002; Riecker et al., 2006; Loibl et al., 2011)); (3) fMRI or PET was used; (4) results were reported in Talairach (Talairach and Tournoux, 1988) or Montreal Neurological Institute (MNI) stereotaxic space; and (5) activation foci were generated from whole-brain, rather than region-of-interest, analyses; (6) studies required subjects to perform some movement of the fingers (e.g., index finger flexion, multi-digit sequences) interspersed with a baseline (rest or fixation) period; (7) studies required subjects to respond to stimuli using their right hands only; and (8) studies employed a regularly paced external stimulus. To illustrate the restrictiveness of these criteria for the older adults, 42 studies remained after limiting the sample of studies to the first three inclusion criteria. We then removed 24 studies that did not report within-group coordinates, two studies that performed ROI analyses only, three studies with movements of non-finger or hand body parts (e.g., elbow or wrist), and four studies that employed self rather than external pacing. In addition, one reported coordinates only for a subset of the brain regions revealed by their between-group analysis. This left eight experiments with within-group data in older adults.

The number of studies fitting the above criteria was far greater for young than for older adults. Thus, we first found studies on older adults that fit the above criteria, and then selected studies on young adults to match on (1) stimulus frequency, (2) stimulus modality (i.e., auditory or visual), and (3) effectors used (e.g., right index finger or multi-digit sequence). All three of these aspects can modulate the amount of activation measured (Blinkenberg et al., 1996; Rao et al., 1996; Riecker et al., 2006; Witt et al., 2008), and the matching strategy minimized confounds that may otherwise have been introduced if the studies in older adults differed in this regard from the studies in younger adults. By the end of this procedure, our goal was to have two groups of studies, one for older and one for young adults, whose stimulus frequency, stimulus modality, and effectors used did not differ significantly between groups.

Our criteria for inclusion into our meta-analysis resulted in 16 experiments and 183 foci total: eight experiments with 86 foci in the young adult group (n = 94; mean age = 26.3 ± 3.66 years) and eight experiments with 97 foci in the older adult group (n = 93; mean age = 60.6 ± 5.69 years). Four of these experiments came from publications that directly addressed the aging motor system as described in the introduction (Calautti et al., 2001; Mattay et al., 2002; Riecker et al., 2006; Loibl et al., 2011), as these studies reported foci for the within-group analyses of the older adult group as well as the young adult group. Two other studies (Hutchinson et al., 2002; McGregor et al., 2009) were not included because they did not report the coordinates for within-group data. The four remaining older adult experiments were data from control groups reported in studies focused on disorders (Elsinger et al., 2003; Tombari et al., 2004; Sharma et al., 2009; Rehme et al., 2011). The four remaining young adult experiments (Kuhtz-Buschbeck et al., 2003; Agnew et al., 2004; Akhlaghi et al., 2012) came from the general motor literatures and were selected because their experimental conditions most closely matched the older adult studies already selected.

Based on these selection criteria, the older adult and young adult experiments were well-matched for their experimental parameters. Both the older adult and young adult experiments consisted of seven fMRI studies and one PET study. In all experiments, subjects used their right hands to perform either individual finger movements (two older and three young adult experiments), multi-digit movements (four older and four young adult experiments), or whole-hand closures (two older and one young adult experiment) in response to visual (two older and three young adult experiments) or auditory stimuli (six older and five young adult experiments). For these studies, subjects lay supine in the scanner and pressed buttons as instructed, following training of the task prior to the scan [one study (Mattay et al., 2002) did not report whether training was done]. Subjects viewed stimuli on the screen or were presented with auditory stimuli that served to indicate to them to make the motor movement. The rate at which the stimuli were presented (and hence at which subjects responded) varied across the studies. However, the overall average stimulus frequency was equated for between the older and younger adult studies [t(14) = 0.00246; p > 0.99], to ensure that these experimental parameters did not influence the ALE result [one of the studies’ stimulus frequencies included in this calculation is an average value because the experiment employed a range of frequencies in a task used to generate motor activity (Riecker et al., 2006)]. All subjects in all studies were right handed, except for in one study (Lehéricy et al., 2006), where one out of 12 subjects was left-handed (and as such there was one left-handed subject in the entire group of 93 subjects). Details of the original published experiments used in the meta-analysis are provided in Table 2.

Three approaches were used: (1) group ALE maps of young and older adults separately, (2) a conjunction analysis identifying commonalities between these two ALE maps (Eickhoff et al., 2011), and (3) a contrast analysis identifying differences in ALE maps between groups (Eickhoff et al., 2011). We followed the procedures laid out in the user manual for GingerALE 2.3³ and briefly explained in a previous publication from our lab (Purcell et al., 2011). To ensure that multiple, tightly clustered foci from a single experiment did not bias the MA maps, we applied the Non-Additive method in which, for each voxel in the brain, only the probability associated with the focus with the shortest Euclidean distance from that particular voxel (i.e., the maximum probability) was used for the MA (Turkeltaub et al., 2012). Significance of the output ALE map was assessed to distinguish true convergence across studies from random convergence (i.e., noise) and was undertaken using the cluster-level inference method, recommended for optimal sensitivity and as an alternative to the more stringent family-wise error correction (Eickhoff et al., 2012). The cluster-level inference threshold was set to p < 0.05 with a cluster-forming threshold false discovery rate (FDR; Thomas Nichols) ⁴ of pID < 0.05, assuming independence or positive dependence, and 10,000 permutations. GingerALE was used to identify peak MNI coordinates associated with all clusters in young and older adult maps, transform these coordinates into Talairach space using icbm2tal (Lancaster), and label nearest gray matter using the Talarach Daemon⁵.

To identify brain areas likely to be activated in both groups, and brain areas that differed between the two groups, we generated a conjunction map as well as two contrast ALE maps (older adult > young adult and young adult > older adult; Eickhoff et al., 2011). The conjunction map showed voxels that survived correction in both the individual group maps for young and older adults with a minimum cluster size of 100 mm³. Contrast maps (older adult > young adult and vice versa) were generated by directly subtracting one group ALE map from the other. Specifically, GingerALE was used to perform the following steps: (1) randomly re-group the experiments constituting the group datasets for young and older adults into two new datasets of the same size; (2) create ALE images for each of the new datasets and subtract one from the other; (3) repeat this process 10,000 times (number of permutations) to generate an extensive null distribution; and (4) test the actual ALE subtraction against this resultant null distribution using a voxel-wise FDR correction (Laird et al., 2005) of pID < 0.05 and a minimum cluster size of 100 mm³.

Results were visualized using the Mango software package⁶ with the Colin brain template in MNI space (Holmes et al., 2014). Functional motor regions (e.g., SM1, SMA, etc.) were labeled using the Human Motor Area Template (HMAT) depicted in axial slices (Mayka et al., 2006). This template was derived by implementing the ALE method on 126 fMRI or PET studies involving motor control, and demarcates three main divisions of motor areas [SM1, medial premotor cortex (MPMC), and LPMC] as well as their subdivisions [primary motor cortex (M1) and primary somatosensory cortex (S1) for SM1, SMA and pre-SMA for MPMC, and PMv and PMd for LPMC]. For our reporting, we differentiated the subdivisions of MPMC and LPMC, but not SM1, because most of the studies used in this meta-analysis showed activation in M1 and S1, and those activations registered in a single cluster. This is different from activations in MPMC and LPMC, which often registered in one but not both subdivisions. The areas demarcated by the template are inherently probabilistic in nature due to the ALE algorithm used to identify them. Probability maps for main and subdivisions represent the likelihood that an activation focus falls within a given area. We reported functional regions based on a 95% probability; i.e., if a coordinate fell within the 95% bounds for only one main or subdivision, we labeled that coordinate as that main or subdivision. Boundary zones (i.e., regions of the map in which main or subdivisions overlapped) manifest U-shaped probabilities, where the nadir represents an equal probability that an activation focus lies in adjacent main or subdivisions. We labeled coordinates in these boundary zones as the main or subdivision with the higher probability. Any coordinates that fell almost perfectly at the nadir were labeled with both names (e.g., pre-SMA/SMA).

A full list of ALE peaks (in MNI stereotaxic space) for the young and older adult groups is reported in Table 3, along with anatomical and functional labels. Superscript markings appending certain functional motor brain regions indicate that the peak is in a boundary zone between two main divisions (%1), e.g., between SM1 and LPMC; or subdivisions (%2), e.g., between PMv and PMd, as demarcated by the HMAT (see Materials and Methods).

Areas most likely activated by both young and older adults, as revealed by conjunction analysis, included left SM1 extending into left PMd^%1, left SMA extending into right SMA^%2 and bilateral pre-SMA^%2, bilateral insula, left thalamus, and right anterior cerebellum (Figure 1C; Table 4).

A direct between-group comparison revealed that older compared with young adults had a greater likelihood of activation in right SM1 extending into PMd^%1, right supramarginal gyrus (SMG), and left postero-ventral SMA extending into right SMA (Figure 2; Table 5). The reverse comparison revealed significantly lower ALE values in older compared with young adults in right pre-SMA, extending into left pre-SMA^%2, and left SMA. Older adults also had significantly lower ALE values in right posterior cerebellum (Figure 3).

Our ALE results for Movement > Rest contrasts in young adults were similar to those from a prior meta-analysis of 23 published studies restricted to right-hand index finger movements [one of several meta-analyses reported by Witt et al. (2008), with the main meta-analysis of 38 studies including left hand movements]. They too observed left SM1, left PMd, bilateral SMA, right insula, left basal ganglia, and bilateral anterior cerebellum. It should be noted that our ALE map additionally identified left IPL, left insula, and left thalamus, whereas we did not identify some areas reported by Witt et al. (2008), namely right PMd, left PMv, right IPL, bilateral claustra, right IFG, and bilateral DLPFC. There are three notable methodological differences between the meta-analysis presented here and the comparable analysis presented by Witt et al. (2008). The first is the number of experiments entered into the analyses (eight for ours and 23 for Witt et al., 2008). Second, our meta-analysis included some experiments in which subjects performed finger movements involving multiple fingers, whereas the subset meta-analysis (23 studies) reported by Witt et al. (2008) only included index finger movements. And third, our ALE maps were generated from studies that used only externally paced stimuli (i.e., externally guided by auditory or visual stimuli), which, based on their predominant representation in Witt et al. (2008), seems to be the more common approach found in the literature. In contrast, Witt et al. (2008) generated their ALE map for right-hand index finger movements from 23 studies comprising experiments using both externally paced (23 experiments) and internally paced (self-guided with no external stimuli, 11 experiments) finger movements, while in another aspect of their study showing that likelihoods of activation seem to differ depending upon whether the movement is externally or internally driven.

Turning to the older adults, we found likelihoods of activation in this group bilaterally in a number of regions including SM1, LPMC, SMA and pre-SMA, posterior parietal cortex, putamen, and anterior cerebellum. These findings are consistent with two studies of older adults not included in our meta-analysis (because the coordinates of activation foci were not reported), which reported activation in similar ipsilateral, right cortical motor areas (Hutchinson et al., 2002; McGregor et al., 2009).

A conjunction analysis revealed that older and young adults are both likely to recruit left SM1, left SMA, bilateral insula, left thalamus, and right anterior cerebellum. However, the between-group comparison revealed that older adults show a greater likelihood of engaging right SM1 (and extending into right PMd) and SMG, as well as bilateral postero-ventral SMA. When considered in the context of the results from the young adult within-group map, it becomes evident which of these regions are uniquely recruited by older adults and which regions are also present in young adults but demonstrate a relatively greater concordance in older compared to young adults. Interestingly, the right SM1 and right SMG, observed in both the older adults within-group map and the older versus young adults contrast, were not observed in the young adult map, indicating that these regions are uniquely recruited in older adults.

Two main theories have been proposed to explain the additional activations observed in older compared with young adults. The first is compensation, in which better task performance corresponds to activation in additional brain areas (Mattay et al., 2002; Heuninckx et al., 2008). Indeed, Mattay et al. (2002) reported positive correlations between performance and bilateral SM1, LPMC, SMA, and left parietal cortex and cerebellum (Mattay et al., 2002). This pattern is not limited to fMRI data, as it has been shown that the release from transcallosal inhibition between right and left SM1 (described earlier) also seems to contribute to better motor performance in older adults (Fling and Seidler, 2012; Fling et al., 2012). More recently, others have interpreted age-related increases in sensorimotor functional connectivity (FC) measures as also reflecting compensatory processes (Tomasi and Volkow, 2012; Mathys et al., 2014; Seidler et al., 2015). However, most FC analyses of the aging motor system have examined resting-state, rather than task-based data, the latter being more pertinent to the goal of this study (i.e., to examine age-related differences in the motor system as it mobilizes to perform a motor task). The second interpretation is de-differentiation (Carp et al., 2011; Bernard and Seidler, 2012), in which additional brain areas in older adults are recruited non-selectively after age-related degeneration of connections. In this case, additional activations do not correlate with performance or task demands (Riecker et al., 2006), or are possibly negatively correlated (Loibl et al., 2011). This controversy is not limited to the motor system, as it extends into vision (Park et al., 2004, 2012) and cognition (for a review, please see Grady, 2012). For example, the observation that older adults tend to show bilateral PFC activity during working memory tasks that elicit unilateral PFC activity in young adults has been attributed to compensation (Cabeza, 2002); however, to the best of our knowledge, the possibility of age-related changes in inter-hemispheric inhibition as we described in Section “Primary Sensorimotor Cortex” has not yet been investigated in the cognitive domain. This debate is also not limited to aging; interpretations of developmental changes are also controversial (Poldrack, 2010), and it is unclear whether differences in activation in children of different ages reflect compensation for immaturity or are a result of that immaturity.

As we cannot quantify performance or task demands for all studies in this meta-analysis, resolving the controversy between compensation and de-differentiation is beyond the scope of this study. Nevertheless, our observation of unique likelihoods of activations in right SM1 (extending into PMd) and SMG, three brain areas whose connections have been characterized in the animal literature (Geyer et al., 2000), suggests that right SM1 might be activated in older adults by input from SMG via PMd. This compensation by an aging-specific ipsilateral network would be the most parsimonious interpretation of our cortical results. Nevertheless, we would not exclude the possibility—as discussed earlier—that degeneration of transcallosal inhibition, as would be one basis for de-differentiation, also contributes to the unique likelihood of activation in SM1 specifically, which would indicate that both compensation and de-differentiation occur in aging.