Reduced Correlations Associate with Decreased White Matter Integrity in Older Adults

The above results show decreased functional correlations across the default network and the dorsal attention system in advanced aging. Advanced aging is accompanied by compromised brain structure, including reduced integrity of white matter [for a review see Sullivan and Pfefferbaum (2006)]. Accordingly, we hypothesized that more severe white matter compromises would be associated with reduced functional correlations. Since the anterior-posterior correlations in the default network were most robustly affected by age, we chose this measure as our dependent variable in the following analyses. For each older subject, the mean anisotropy within a large white matter region was measured and correlated with the functional correlation strength between the mPFC and pC/rsp (Figure 7). Results revealed that reductions in functional correlations were associated with reduced diffusion tensor anisotropy measured by DTI (Figure 7A: r = 0.39, p < 0.01). This relationship persisted in the older adult group after controlling for the effect of age on both variables (Figure 7B: partial r = 0.33, p < 0.05), suggesting that the effect accounted for variance between individuals beyond that accounted for by age alone.

Open in a separate window

Figure 7

Functional Correlations Relate to White Matter Integrity in Older Adults

(A) Mean anisotropy for a white matter region shown in yellow on one transverse slice was extracted for each participant. The correlation coefficients resulting from correlating the mPFC and pC/rsp time course are plotted against the mean anisotropy (A_σ) (r = 0.39, p < 0.05).

(B) A linear regression was performed using the same measures, after controlling for the effect of age on each measure. When doing so, the strength of the relationship between the anterior-posterior functional correlations and white matter integrity remained significant (partial r = 0.33, p < 0.05).

Measuring Disruption of Large-Scale Brain Systems in Older Adults using fMRI

The present analyses explored functional integrity in a large sample of young and old adults by measuring fMRI signal correlations between regions within large-scale brain systems. Extracting estimates of activity correlations is a powerful means of measuring disruption of brain systems in advanced aging. Using estimates of effective connectivity, Grady et al. (2003) have recently observed that functional correlations between groups of older and younger adults differ in ways that relate to cognition. Here we found that the anterior-posterior functional integrity of the default network, measured using correlation coefficients, dropped from 0.43 to 0.00 in older adults. Before discussing the implications of these results, it is worth noting some of the strengths and weaknesses of our approach.

Low-frequency activity correlations are present and detectable in many brain systems by using fMRI. Originally described by Biswal et al. (1995) in the sensorimotor system, spontaneous correlations have since been observed in young adults within brain systems important to memory (Greicius et al., 2003; Vincent et al., 2006) and attention (Fox et al., 2006). While the origins of the activity events driving spontaneous correlations remain poorly understood, they are present during anesthesia and have been shown to track direct and indirect anatomic connectivity between regions (Vincent et al., 2007).

Several methodological concerns need to be addressed before the present observations can be interpreted. Older adults differ from young adults in brain anatomy and brain motion during imaging, both of which cause differences in variance properties that influence measured functional correlations between brain regions. A number of control analyses were performed to address concerns that our observed effects were simply due to systematic differences in the fMRI signal (Buckner et al., 2000; D’Esposito et al., 2003). First, the effects persisted after controlling for variance, and similar results were obtained when controlling for motion. Second, the effects also persisted in variance-matched subsets of participants (see Supplementary Materials). Finally, correlations between right and left visual brain regions that were predicted to be minimally affected in aging were largely conserved in our analyses. Thus, while measurement differences including variance in the fMRI signal differ between age groups, such differences almost certainly do not account for the present results. Moreover, the observation that reduced blood oxygen-level dependent (BOLD) correlations were associated with both DTI evidence of white matter differences and impaired cognitive performance reinforces the impression that the brain normally loses functional integrity in the course of advanced aging.

Image Acquisition

All images were acquired using a 1.5T Siemens Vision Scanner (Erlangen, Germany). Details regarding head stabilization, vision, and communication tools are described in Lustig and Buckner (2004) and only briefly described here. A high-resolution structural T1-weighted MPRAGE image was collected for each participant (1 × 1 × 1.25 mm, TR = 9.7 s, TE = 4 s, flip angle = 10°, TI = 20 ms, TD = 200 ms) followed by rapid event-related functional imaging using an asymmetric spin-echo sequence sensitive to BOLD contrast (TR = 2.5 s, T2* evolution time = 50 ms, 180 offset = −25 ms). Two relevant functional runs were collected (each run: 116 whole-brain acquisitions, 3.75 × 3.75 mm in-plane resolution, 16 slices 8 mm thick, aligned to the anterior-posterior commissure plane).

During the two BOLD runs used in this study, participants were engaged in a semantic classification task. Specific task details are outlined in Lustig and Buckner (2004). Each run began with 10 s of passive fixation, followed by 116 trials, each lasting 2500 ms. Two-thirds of trials included a semantic classification task where subjects made a button response to classify visually presented words as living or non-living objects. One-third of trials were passive fixation trials. Each 2500 ms trial was composed of 2000 ms of task or fixation followed by a 500 ms intertrial interval consisting of a fixation crosshair. Trials were randomly intermixed and counterbalanced across participants.

Data Preprocessing

Functional data were first preprocessed to reduce image artifacts and remove timing differences between slices that resulted from asynchronous slice acquisition. Odd/even slice intensity differences resulting from acquisition of interleaved slices were eliminated, and head movement within and across runs was removed using rigid body translation and rotation. Anatomical and functional data were transformed to atlas space (Talairach and Tournoux, 1988; Maccotta et al., 2001) using an atlas target composed of a merged young-adult and older-adult reference (Buckner et al., 2004), and data were resampled to 2 mm isotropic voxels. Finally, drift was corrected by removing the linear slope on a voxel-by-voxel basis, and whole-brain signal intensity was normalized to a mean of 1000 for each functional run.

Functional Correlation Analysis between Regions

Several additional steps, described in previous studies (Vincent et al., 2006; Fox et al., 2005), were taken to prepare the data for correlation analysis. First, low- and high-frequency components of the atlas-aligned BOLD data were removed using a temporal band-pass filter retaining 0.009 Hz< f< 0.08Hz. Next, data were spatially smoothed using a Gaussian kernel of 6 mm full-width at half-maximum. Removal of several spurious or nonspecific sources of variance was accomplished by regression of the following variables: (1) the six movement parameters computed by rigid body translation and rotation in preprocessing, (2) the mean whole-brain signal, (3) the mean signal within the lateral ventricles, and (4) the mean signal within a deep white matter ROI. The first temporal derivative of each time course was also included in the regression procedure to account for temporally shifted nuisance waveforms. Regression of each of these signals was computed simultaneously and the residual time course was retained for the following correlation analysis.

Two forms of analysis were performed to yield (1) correlation strengths between pairs of regions and (2) correlation maps between a seed region and all voxels across the whole brain. For analysis using a priori regions, the time course was extracted from each region and the correlation coefficient was computed using Pearson’s product-moment formula. Whole-brain exploratory analysis similarly involved computing the correlation coefficient between the averaged time course at the seed region and the time course for each voxel across the whole brain. The resulting correlation maps were converted to z values using Fisher’s r-to-z transformation (Zar, 1996). We also evaluated voxel-wise correlation strengths on lines connecting two regions along the anterior-posterior axis. Such plots are particularly informative for visualizing the correlation profile across a specific anatomical cross-section of the data. Significance testing of between-group effects was performed using random-effects analyses.

ROI Definition

To create regions in an independent and unbiased manner, an initiating pC/rsp seed was selected from a conjunction analysis of several functional studies (Buckner et al., 2005) and applied to a separate dataset (Buckner et al., 2000) of eight young (21.3 years; four male) and eight old (77.4 years; four male) adults. While this initiating seed was somewhat arbitrary, similar results are obtained using almost any of the regions that are commonly part of the default network (e.g., Shulman et al., 1997; Mazoyer et al., 2001). The Buckner et al. (2000) dataset was then examined for correlations with the pC/rsp seed. Based on the pattern of the correlations, we extracted peak coordinates corresponding to the mPFC, the HF, and the PHC. Regions were defined around these peaks as all contiguous voxels exceeding a z-transformed correlation coefficient of 0.1 within a radius of 4–8 mm. The peak coordinates corresponding to the LatPar and a newly defined pC/rsp ROI were selected in a similar manner using the correlation map for a seed in the HF, obtained in the above analysis. We have demonstrated in a previous study that the correlational pattern of both left and right HF is highly similar (Vincent et al., 2006). Thus, in order to increase power in the present analyses, we used bilateral, as opposed to unilateral, ROIs. Using this procedure, the following a priori ROIs were created (coordinates are in the atlas space of Talairach and Tournoux, 1988): mPFC (1, 40, 16), pC/rsp (−1, −50, 26), LatPar (L: −45, −67, 26; R: 53, −65, 26), HF (L: −23, −25, −12 ; R: 23, −25, −12), and PHC (L: −25, −39, −10; R: 25, −39, −10).

External attention system ROIs were defined using identical methods to the ones described above. Specifically, peak coordinates in selected regions of the cortex were identified on the group-averaged Fischer’s r-to-z transformed correlation map for an 8 mm (radius) spherical seed in the IPS, a region selected from a previous functional correlation study (Fox et al., 2006). Regions were defined around peak coordinates in or near the vIPS and FEF, and in and around the superior precentral sulcus, the PrCeS, and the MT⁺. Similarly, the peak coordinates corresponding to the IPS were selected using the correlation map for a seed in the vIPS, obtained from the above analysis.

Using this procedure, the following a priori ROIs were created for further analysis of the dorsal attention system: IPS (L: −21, −71, 44; R: 23, −63, 50), vIPS (L: −27, −75, 24; R: 31, −79, 22), FEF (L: −27, −7, 50; R: 27, −1, 54), PrCeS (L: −45, −1, 34 ; R: 45, 5, 34), and MT⁺ (L: −47, −75, −4; R: 49, −69, 4).

Using an initiating visual seed from a separate study (Konishi et al., 2000), two control ROIs were also created: one centered in the left visual cortex (−19, −95, 2) and another centered in the right visual cortex (19, −95, 2). All of the ROIs were then carried forward as a priori seed and target regions for further analysis in the present dataset.

DTI

A subset of 44 older adults (76.2 years; range = 60 to 93; 13 male) were imaged in an additional data session that included DTI. The DTI acquisition protocol was identical to Head et al. (2004), using a spin-echo echo-planar imaging sequence (1 × 1 × 1.25 mm, TR = 7.2 s, TE = 108 ms, 36 slices at 4 mm at 2.5 mm² in-plane resolution, interpolated to 1.25 mm²). Four acquisitions were acquired for each participant and averaged. Diffusion-weighted data were coregistered to the structural data, and the diffusion tensor at each voxel was computed using standard least-squares techniques, with the derived measure of mean diffusion anisotropy used for this analysis being A_σ, ranging from 0 to 1 (Conturo et al., 1996; see Head et al., 2004 for implementation as used here). Diffusion anisotropy represents the degree to which water molecules diffuse along a particular axis, with values closer to the upper limit of 1 indicating strong directionality (typical of healthy myelinated white matter) and values closer to 0 indicating isotropic diffusion (indicating degraded white matter).

For each participant, the mean diffusion anisotropy was calculated for a large ROI (Figure 7A; Figure S2) that was manually traced using Analyze-AVW Software (Mayo Medical Foundation, Rochester, MN) on a mean anisotropy map from an independent sample of 44 older adults (76.6 years; range = 60 to 91; 14 male) otherwise not analyzed in the present study. Specifically, the ROI was traced on the mean diffusion weighted anisotropy image as opposed to the T1-weighted structural image in order to minimize the inclusion of gray matter and because of distortion of the DTI volumes. The ROI was demarcated to include white matter within the medial part of the centrum semiovale for each hemisphere on all slices from z = 21 to slice z = 41. The rostral and caudal y coordinates for the ROI in each hemisphere were y = 34 and y = −65, while the x coordinates in the left hemisphere ranged from x = −29 to x = −14, and in the right hemisphere, from x = 15 to x = 28. The bilateral ROI included several white matter bundles such as the corona radiata, superior longitudinal fasciculus, and cingulum bundle. These regions have been shown to exhibit age-associated white matter disruption (Pfefferbaum et al., 2000, 2005; Head et al., 2004), and our selection of this relatively large region allows for a stable estimate of white matter disruption in individual participants. White matter disruption in aging is diffuse with an anterior-to-posterior gradient. Our region captures the overall health of white matter within a participant, but it is not meant as a selective measure of specific fiber bundle integrity. Mean anisotropy values within this ROI were calculated for each available participant in the present data set and were entered into regression analyses that compared the DTI measures to the strength of functional correlations.

Neuropsychological Measures

A subset of 40 older adults (76.4 years; range = 60 to 93; 13 male) completed a battery of psychometric tests as part of more ongoing assessments of structure and cognition in our older adult participant pool. Individual psychometric tests were combined based on functional domains into three composite scores that assessed executive function, memory, and processing speed ability. Table 1 shows the individual test means as well as their grouping into composite scores. To compute composite scores, raw test scores for each individual were converted to z-scores using the group mean for each test, and the z-scores for all tests within each composite were summed for each individual. The executive composite combined test scores from the Category Fluency Test (Animals), the Word Fluency Test, the Trail-making Test B, and the Weschler Memory Scale-Revised (WMS-R) Digit Span Test (backward). The memory composite included the WMS Associate Memory Recall Test (hard items), the WMS-R Logical Memory Recall Test, and the Benton Visual Retention Test (Form C, with a delay). The processing speed composite included the Trailmaking Test A, the Crossing Off Test, and Wechsler Adult Intelligence Scale (WAIS) Digit Symbol Test (see Table 1 for test scores). These composite scores were entered into regression analyses that compared the cognitive measures to the strength of functional correlations.

We thank the Washington University ADRC for participant recruitment, and Mark Mintun for PIB amyloid imaging. Martha Storandt provided neuropsychological test scores. Michael Fox provided valuable discussion concerning use of functional correlations, and Dale Stevens, Gagan Wig, Itamar Kahn, and Yun-Ching Kao provided additional valuable discussion. We also thank three anonymous reviewers for their thoughtful comments. This work was supported by NIH grants AG05886, AG03991, AG00030, P50 AG05681, the Alzheimer’s Association, and the Howard Hughes Medical Institute.