Forty healthy adults ages 51–91 were recruited from a larger dataset of community-dwelling adults enrolled in the Active Brain study, a University of Florida (UF) neuroimaging study aimed at investigating brain activity in healthy older adults. The sample size was chosen based on study group size in similar fMRI research showing significant findings (Ryan et al., 2012). Inclusion and exclusion criteria are listed in Table 1. Participant characteristics are listed in Table 2. Ten participants were gathered for age decades 50–59, 60–69, 70–79, and 80+. They were college educated, on average, 55% female, and predominantly Caucasian (N = 38/40). Participants were free from pre-existing dementia or other neurological disease, major psychiatric disorders, history of head injury with loss of consciousness greater than 15 min, and major visual impairments. Exclusionary criteria were verified via medical history questionnaires, interview, and testing of cognitive and visual function. All participants had or were corrected to 20/40 vision or better. Participants were also free of conditions or implants for which MRI is contraindicated, including claustrophobia. Participants were given information about the study and, if they were interested in participating, provided verbal and written informed consent to participate.

Visual stimuli were created using E-Prime 2.0 software based on a subset of those used by Swearer and Kane (1996), which we described previously (Seider et al., 2020). Participants were presented with simultaneous matching paradigms requiring perceptual judgments of either spatial location, shape, or velocity (Figure 1). For each task, three stimuli were presented, with two located side-by-side below the horizontal midline of the display and one located above the horizontal midline and centered on the vertical midline. The upper stimulus (target) was identical to one of the lower (sample) stimuli, and participants were asked to indicate which of the sample items was identical to the target via button press. The tasks varied over three levels of difficulty (easy, medium, and hard). For Location, the task was to decide which sample box below had a dot in the same location as the target box above. Boxes were 6.65-cm squares, and difficulty was based on how far the dot in the sample item was located from the dot in the target item, either 1.27, 0.79, or 0.48 cm from the target location. For Shape, the task was to decide which sample design below was identical to the target design above. Difficulty was based on judgment RTs, which were highly correlated to difficulty ratings, obtained during a pilot study. For Velocity, the task was to decide which sample set of lines below were moving at the same speed as the target set of lines above. Difficulty was based on the speed differences between the sample and target items, which differed by a factor of the just noticeable difference (JND) for that velocity (either nine, six, or three JNDs).

fMRI scanning took place at the UF McKnight Brain Institute (MBI) Advanced Magnetic Resonance Imaging and Spectroscopy (AMRIS) facility and lasted approximately 1 h. Prior to entering the scanner, participants were screened using the MRI safety screening form for research participants. They were given a clear and detailed explanation regarding the MR procedure and trained on the experimental paradigm. Those requiring glasses received MRI-compatible lenses to correct vision, with glasses prescription measured by lensometer. During scanning, participants were positioned supine, with cushioning surrounding their heads to reduce movement artifact. Structural imaging was acquired first, during which participants were instructed to be still and relax. This was followed by functional imaging, during which they completed the cognitive task.

To ensure that each participant understood the nature of the task and were comfortable performing it in the scanner, they completed a practice block in the MRI with nine trials, consisting of one trial for each of the three levels of difficulty within each of the three tasks. All stimuli were projected onto an LCD screen (BOLDScreen 32, Cambridge Research Systems) behind the scanner and were viewed via a mirror positioned above participants’ heads. Participants were instructed to indicate which of the two lower sample stimuli was identical to the upper target stimulus (Figure 1) by pressing one of two buttons on a response box, using their right middle and index fingers. They were asked to respond as quickly and as accurately as possible, but not to sacrifice accuracy for speed. They were given 3 s to respond. Accuracy and RT were recorded. Prior to each stimulus, a centrally located fixation cross was presented to direct attention the center of the screen.

fMRI data was acquired on a 3T Philips Achieva scanner (Philips Healthcare, Best, Netherlands) at the UF-MBI AMRIS facility using a 32-channel sensitivity encoding (SENSE) head coil. Anatomical images were acquired using a whole-brain high-resolution 3D MP-RAGE sequence: 176 1 mm thick sagittal slices, TR = 7 ms, TE = 3.2 mc, FA = 8°, matrix = 256 × 256, FOV = 256 mm. Functional images were acquired in the axial-oblique plane using a Phillips 3T scanner: 36 slices, TR = 2 s, TE = 30 ms, FA = 80°, FOV = 224 mm × 224 mm, acquisition matrix 64 × 64, isotropic voxels of 3.5 × 3.5 × 3.5 mm³. Prior to the collection of each run, a number of dummy scans were collected and discarded to allow for the signal to stabilize.

Data were collected over three functional scans using a mixed block/event-related design (Figure 2) to allow for analysis in either a block- or event-related fashion. Scans began with a 17-s fixation cross presented in the middle of the screen, followed by a 2-s instruction screen that informed participants as to the type of discrimination task to follow (Location, Shape, or Velocity). Following the instruction screen, a block of seven stimuli was presented. Each stimulus displaying a simultaneous match-to-sample task was on the screen for 3 s. Responses were recorded during either this time or the fixation time that immediately followed. Fixation time between stimulus events (inter-stimulus intervals) was randomly jittered with a minimum of 1000 ms and an average of 2500 ms to increase the temporal resolution of the estimated hemodynamic response. Blocks of activity were separated by 17-s fixation periods. Another fixation period of 17 s was presented at the end of the block. The first two scan runs had 14 full blocks and ran a total time of 824 s, while the third scan run had 14 blocks, three of which only had five stimuli instead of seven, and it ran for a total of 792 s. Task types were evenly divided between blocks, so there were 14 blocks for each VAB task (Location, Shape, and Velocity). Scan order, block order, and stimulus presentation within the block were randomized. Each block contained a mix of task type and difficulty levels. There were 288 stimuli total, 32 stimuli for each of the nine conditions (three tasks with three levels of difficulty each).

fMRI data pre-processing and analyses were performed using Statistical Parametric Mapping (SPM12) software¹. Slice timing correction was used to correct differences in image acquisition time between slices using the middle slice as a reference. Motion correction in functional data was addressed via SPM realign, using a least squares approach and a six-parameter (rigid-body) spatial transformation; a two-pass procedure registered the images to the mean of the images after the first realignment. Realignment parameters were used to exclude runs that had substantial motion artifacts based on researcher consensus, and motion regressors were included in the first level to account for movement. Functional data were co-registered to the subject’s anatomical image using a rigid-body model. Anatomical data were segmented, bias-corrected, and spatially normalized to a standard Montreal Neurological Institute (MNI) template. These transformations were applied to the functional data, normalizing them to MNI space. Data were then spatially smoothed with a Gaussian kernel of 8 mm to suppress noise and effects due to residual differences in functional and structural anatomy during inter-subject averaging.

Scan runs with less than 68% accuracy (i.e., performances approaching chance) were excluded to avoid measuring functional activity during poor effort or attention. One participant was excluded entirely (not included in the N = 40 described above), three participants had one scan run excluded, and two had two runs excluded. One participant had only two scan runs collected due to mechanical difficulties that stopped the MRI during the third scan. Data was analyzed in an event-related fashion to allow for examination of each difficulty level.

Regions of interest were created using the MarsBaR toolbox for SPM 8, and analyses were conducted in a MATLAB environment (Math-works, Natick, MA, United States). Spherical ROIs were designed based on prior research. Figure 3 depicts the ROIs, and Table 3 lists the MNI coordinates on which they are centered and radii. The Location ROI was designed to be in the dorsal processing stream based on established findings (Mishkin et al., 1983; Livingstone and Hubel, 1988). MNI coordinates for the posterior parietal cortex (PPC) were extracted from a study that required participants to judge the distance between a dot and a line across two panels and respond as to whether the distances were the same or different (Zachariou et al., 2013). PPC activation compared to a control task (slide matching without distance estimation) centered at MNI coordinates −25 −60 49 with a radius of 9.2 mm for the left hemisphere and 22 −61 50 with a radius of 10.6 mm for the right hemisphere. Shape tasks consistently activate the lateral occipital region (Jiang et al., 2007; Pitcher et al., 2009; Silvanto et al., 2010; Zachariou et al., 2013); thus, the Shape ROI was located there. Coordinates were extracted from a study that subtracted activation during viewing of scrambled car images from activation during viewing of car images (Jiang et al., 2007). The left hemisphere ROI was centered at MNI coordinates −43 −78 −10 with a radius of 4 mm, and the right hemisphere ROI was centered at MNI coordinates 48 −75 −11 with a radius of 4 mm. The Velocity ROI was designed to be in V5/MT based on established findings (Zeki et al., 1991; Orban et al., 2003), but the coordinates and radii for the spherical ROIs created were based on a study of retinotopic organization of that area (Kolster et al., 2010). The left hemisphere ROI was centered at −48 −75 8 and had a 4-mm radius, and the right hemisphere ROI was centered at 22 −61 50 and had a 4.6-mm radius. The larger right vs. left hemisphere ROI was consistent with our findings that the velocity task uniquely elicited more right than left hemisphere activity (Seider et al., 2020).

To test whether there was age-related dedifferentiation, a differentiation index was created as a measure of discriminability. The measure is calculated by subtracting the mean functional activation for the non-preferred tasks in the task-associated ROI from that of the preferred task in that ROI and dividing by a measure of the mean variability of activation for all tasks within that ROI. This method has been described by Voss et al. (2008), and the formula used was reproduced from their paper (Figure 4). In effect, the differentiation index measures the degree to which an ROI is differentially activated during the task that is theoretically specialized to that region vs. other similar tasks. A bivariate correlation compared age with the differentiation indices for each of the three tasks (Location, Shape, and Velocity).

Behavioral performance was based on median RT for a task, used instead of mean because of the tendency of RTs to be positively skewed (Van Zandt, 2002), as well as accuracy. Composite performance scores were created by averaging the mean of the z-score for accuracy and the reversed z-score for median RT for each of three visual discrimination tasks, difficulty levels collapsed. Difficulty levels were collapsed because the parameters underlying difficulty differed for each task, so comparing easy, medium, and hard items across tasks was not possible. To determine whether dedifferentiation was associated with performance, composite performance scores for each task were correlated with the degree to which associated ROIs were differentially activated for those preferred tasks (differentiation index). Additionally, correlation analyses compared performance composite scores for each task (difficulty levels collapsed) to percent signal change in its task-specific ROI.

Dedifferentiation of visuospatial abilities was first demonstrated on behavioral tests. Young adults had very differentiated perceptual and spatial abilities (Chen et al., 2000), suggesting considerable functional benefits of performing different types of visual discrimination in a very specific manner. This specificity also had potential functional anatomic implications, such as the possibility that focal and spatial visual processing involved different cortical systems that acted somewhat independently (e.g., ventral vs. dorsal visual processing pathways). In contrast, a principal components analysis of these abilities in older adults showed only one common factor, suggesting dedifferentiation of those skills (Chen et al., 2002). This pattern of dedifferentiation was reflected in imaging studies. Research led by Grady et al. (1992) examined PET results during face and dot-location matching tasks in older and younger participants. While, in younger adults, face matching activated the occipitotemporal cortex and dot-location matching the superior parietal cortex, older adults had more occipitotemporal activation during location matching and more superior parietal activation during face matching. Similarly, Park et al. (2004) measured activation during encoding of different categories of stimuli (faces, places, non-sense words, and chairs) and showed that voxels that responded to one of the stimulus categories were more likely to respond to multiple categories in older compared to younger adults, demonstrating dedifferentiation in the older adults.

Age was unrelated to the extent of neural recruitment in the current study, a finding that differs from the findings of Cabeza (2001), Cabeza et al. (2002) While the reason for this discrepancy is not clear, several methodological differences exist between these past studies and the current investigation, which may account for a discrepancy in findings. First, prior research compared older and younger groups. Our participant sample did not include younger adults, so there was a more restricted age range. However, among middle-aged and older adults, age may be less determinant of neural recruitment than other factors. Examining age as a continuous variable in only middle-aged and older adults, as the current study did, may highlight differences associated with other variables, such as cognitive reserve, rather than age. Another clear difference is that other studies measured overlapping activation by indexing voxels that pass thresholds of significance. This can lead to problems when comparing activation between tasks or groups, as groups or tasks may have different activation intensities, which would lead to different significance thresholds and potential masking of effects (Zarahn et al., 2006; Voss et al., 2008).

The differentiation index employed in the current study was previously employed by Voss et al. (2008), who had adapted the method from studies of functional response specificity to faces in the inferior temporal/fusiform face area (Afraz et al., 2006; Grill-Spector et al., 2007). This method uses signal detection theory to calculate discriminability (d’), or the degree of selectivity among the three posterior ROIs. This measure considers the difference between ROI activation for the associated and unassociated tasks, normalized by the variability of activation in that ROI for all three tasks. By calculating d’, this and the Voss study had a quantitative measure of neural specificity to the preferred stimulus relative to the other tasks that could be compared to a continuous age measure. Like in prior research, the Voss et al. (2008) study examined differences between younger (aged 18–35 years) and older (aged 55–80 years) adults. Unlike the other studies, they examined differentiation indices as a function of age in just older adults as well, an approach that is consistent with the current study. Although older adults were less differentiated than younger adults in the Voss data, there was no effect of age when examining only older adults. This finding is consistent with the results of the current study and suggest that cortical dedifferentiation is not associated with age when only middle-aged and older adults are studied.

That there were not stronger age effects may reflect the restricted age range of this dataset, since no younger adults were included. Many studies show that older adults have more variability in functional activation than do younger adults (Li and Lindenberger, 1999; Ryan et al., 2012), and only when they are divided by performance do differences between older and younger adults emerge (Cabeza et al., 2002; Daselaar et al., 2003; Nyberg et al., 2003; Wingfield and Grossman, 2006). Given that younger adults were not included in the current study, and that age-related change in dedifferentiation could not be compared to performance directly, a conclusion regarding the compensatory effect of dedifferentiation could not be made. However, results add to a literature that suggests dedifferentiation could be compensatory, and future research would benefit from use of the differentiation index to compare the relationships between age, performance, and functional activation among younger and older adults.

While comparing an older to a younger group may be useful, the current focus on middle-aged and older adults suggests that there are not substantial age effects on differentiation. Future research should not only compare the current data with those from younger adults, but studies may also continue to examine performance effects among older adults, as variability in aging may cause differences in recruitment for those who perform well and those who perform poorly. As described, an ideal way to determine whether functional activation is compensatory is to directly compare performance to activity. However, regions activated may vary substantially based on experimental conditions. As such, it may be more comprehensive to characterize cognitive abilities by examining performance on a battery of cognitive tests, or to use a composite score to divide high and low performers (Daselaar and Cabeza, 2005). Future studies may attempt to not only compare functional activity during visual tasks to performance on those tasks, but also to performance on other types of visual tasks and to other cognitive performances.