|Home | About | Journals | Submit | Contact Us | Français|
Amyloid burden and white matter hyperintensities (WMH) are two common markers of neurodegeneration present in advanced aging. Each represents a potential early indicator of an age-related neurological disorder that impacts cognition. The presence of amyloid is observed in a substantial subset of cognitively normal older adults, but the literature remains equivocal regarding whether amyloid in nondemented populations is deleterious to cognition. Similarly, WMH are detected in many nondemented older adults and there is a body of evidence indicating that WMH are associated with decreased executive function and other cognitive domains. The current study investigated amyloid burden and WMH in clinically normal older adult humans aged 65 to 86 (N=168) and examined each biomarker’s relation with cognitive domains of episodic memory, executive function, and speed of processing. Factors for each domain were derived from a neuropsychological battery on a theoretical basis without reference to the relation between cognition and the biomarkers. Amyloid burden and WMH were not correlated with one another. Age was associated with lower performance in all cognitive domains, while higher estimated verbal intelligence was associated with higher performance in all domains. Hypothesis-driven tests revealed that amyloid burden and WMH had distinct cognitive profiles, with amyloid burden having a specific influence on episodic memory and WMH being primarily associated with executive function but having broad (but lesser) effects on the other domains. These findings suggest that even prior to clinical impairment, amyloid burden and WMH likely represent neuropathological cascades with distinct etiologies and dissociable influences on cognition.
A subset of clinically normal older adults exhibit neuropathological changes associated with disorders such as Alzheimer’s disease (AD) and cerebrovascular disease (CVD). Approximately 30% of such individuals over the age of 60 have elevated amyloid burden, an indication of the amyloid plaques that are a hallmark of AD (Klunk, 2011; Sojkova & Resnick, 2011; Rabinovici & Jagust, 2009). Elevated levels of white matter hyperintensities (WMH) associated with CVD are prevalent among clinically normal older adults (Ylikoski et al., 1995). Amyloid accumulation and white matter abnormalities are two of the most commonly observed biomarkers of pathology among clinically normal older adults and represent chronically progressive neuropathologies with potential clinical implications. Some investigators have hypothesized a mechanistic relationship between amyloid and white matter abnormalities, possibly operating through vascular amyloid deposition observed as microbleeds as in cerebral amyloid angiopathy (CAA) with apolipoprotein ε4 as a common genetic predisposition (Brickman et al., 2009; Brun & Englund, 1986; Chen et al., 2006; Grimmer et al., 2012; Holland et al., 2008; Hommet et al., 2011; Horsburgh et al., 2011). Although associations between amyloid and white matter neuropathology have been observed in CAA and late-stage AD, emerging evidence suggests that the presence of amyloid burden and WMH are not always correlated and may represent distinct neuropathological cascades among cognitively normal older adults (Hedden et al., 2012; Marchant et al., 2012; Rutten-Jacobs et al., 2011). Here, we examine whether biomarkers of amyloid and WMH are related to distinct cognitive domains.
We have previously hypothesized that advanced aging is associated with multiple neuropathological cascades that may differentially impact cognitive domains (Buckner, 2004; Hedden & Gabrieli, 2004). Amyloid burden may primarily impact memory systems, whereas white matter degradation during aging may differentially impact executive control systems (Hedden et al., 2012). The literature on relationships between in vivo amyloid burden and cognition is large and varied, although studies with larger sample sizes or longitudinal methods tend to demonstrate a stronger association of amyloid burden with memory than with other cognitive domains (e.g., Jack et al., 2012; Klunk, 2011; Oh et al., 2012; Pike et al., 2011; Resnick et al., 2010; Roe et al., 2008; Storandt et al., 2012; though see Rodrigue et al., 2012). In contrast, the relationship between white matter degradation and cognition may be less specific, but executive control may be somewhat more affected than other cognitive domains (Gunning-Dixon & Raz, 2000; Oosterman et al., 2004).
Recent advances in imaging allow detection of amyloid burden and WMH in cognitively normal individuals, providing an opportunity to discover the correspondence of each to separable cognitive domains. One purpose of this endeavor is to better recognize each pathology as signified by a specific profile of cognition, insofar as domain-specificity can be identified. We hypothesized that amyloid burden would have a cognitive profile preferentially related to memory, whereas WMH would have a cognitive profile preferentially related to executive function. To test these hypotheses, we examined cognitively normal older individuals assessed with an extensive neuropsychological battery and imaging measures of amyloid burden and WMH.
We conducted neuropsychological testing on 168 (95 female) cognitively normal, community-dwelling older adults (aged 65–86, M = 73.24, SD = 5.80). These individuals are participants in the Harvard Aging Brain Study, an on-going longitudinal study currently in the baseline assessment phase. Participants were generally well-educated (years of education: M = 16.02, SD = 3.03) with high estimated verbal intelligence (VIQ; M = 121.9, SD = 8.4 estimated from the American National Adult Reading Test, Ryan & Paolo, 1992) and high socio-economic status (SES) as measured by the Hollingshead (1957) scale (M = 28.37, SD = 15.51 – the scale ranges from 11–77 with lower scores indicating higher SES). A comparison sample of 38 younger adults (aged 19–32, M = 22.03, SD = 3.69; years of education: M = 14.58, SD = 1.72; estimated VIQ: M = 118.3, SD = 5.2) completed the same neuropsychological testing as the older adults. All participants had a Clinical Dementia Rating (CDR) of 0 (Morris, 1993), performed no worse than 1.5 SD units below the age- and education- corrected norm on the Logical Memory IIa subtest of the Wechsler Memory Scale-Revised (Wechsler, 1987), and scored 26 or above on the Mini-Mental State Examination (Folstein et al., 1975). Participants were excluded if previously diagnosed with a neurological or psychiatric condition or if they scored greater than 11 on the Geriatric Depression Scale (Yesavage et al., 1983). All participants had normal or corrected to normal vision. Participants provided informed consent in accordance with protocols approved by the Partners Healthcare Inc. Institutional Review Board. Fifteen of the 168 subjects (9%) were excluded from at least one imaging visit because of contraindications uncovered after initial enrollment. Because of the staged nature of the visits (all baseline visits must be completed within 6 months), positron emission tomography (PET) amyloid imaging and magnetic resonance imaging (MRI) estimates of WMH were currently available for 109 of the older adults. Data from 20 (18%) of these participants were included in previous analyses of relationships between amyloid or WMH and cognition (Hedden et al., 2009, 2012; Rentz et al. 2010, 2011).
Participants completed an extensive battery of neuropsychological and behavioral tasks selected primarily to represent executive function component processes, episodic memory, and processing speed. The variables selected from each task were chosen through discussion with neuropsychologists and on the basis of prior literature without reference to the correlation matrix or other data from the current sample. Executive function was made up of the component processes of fluency, working memory, and switching. The selected measures (grouped by hypothesized factor) were: Fluency: the sum of the words produced in 60 seconds for the letters F, A, and S (FAS; Benton et al., 1983); the sum of the words produced in 60 seconds for each of the categories animals, vegetables, and fruits (CAT; Monsch et al., 1992); Working Memory: the number of trials correctly completed for the Letter-Number Sequencing (LNS) subtest of the WAIS-III (Wechsler, 1997); the highest level completed for the Digit Span Backward (BDS) subtest of the WAIS-R (Wechsler, 1981); the number of errors produced on the Self-Ordered Pointing (SOP) task (adapted from Petrides & Milner, 1982 and Shimamura & Jurica, 1994); Switching: the across-block switching RT effect (a difference score of RTs during switch versus non-switch blocks) from the Number-Letter (NL) task (adapted from Rogers & Monsell, 1995 and Miyake et al., 2000); the compatibility RT effect (a difference score of RTs during incompatible versus compatible trials) from a modified Flanker task (adapted from Eriksen & Eriksen, 1974 and Ridderinkhof et al., 1999 to include a compatibility manipulation in which participants must switch between responding with the same or opposite direction as the central arrow as cued by a color cue); the Trail Making Test (TMT) time to complete Form B minus Form A (Reitan, 1979); Episodic Memory: the cued recall names (CRN) and cued recall occupations (CRO) scores from the Face-Name Associative Memory Exam (FNAME; Amariglio et al., 2012; Rentz et al., 2011); the delayed recall (DR) score from the 6-Trial Selective Reminding Test (SRT; Masur et al., 1990); List 2 free recall (FR2) from the Memory Capacity Test (MCT; H. Buschke, personal communication; Rentz et al., 2010); Processing Speed: the median RT from the non-switching blocks from the Number-Letter Task; the Trail Making Test time to complete Form A; and the number of items completed on the Digit-Symbol subtest of the WAIS-R (Wechsler, 1981).
Confirmatory factor analysis was conducted using AMOS 18.0 (Amos Development Corp., FL). Missing data were imputed with mean replacement. Only 6 of the 15 observed variables had missing data and a maximum of 6.5% of values were missing for any task. To reduce the reliance on individual subtests, which are necessarily influenced by measurement error, and to examine the shared variance between tasks, we used confirmatory factor analysis to model a hypothesized factor structure. We hypothesized that the cognitive variables under investigation would be organized into 3 primary factors: executive function, episodic memory, and processing speed. We further hypothesized that executive function would be represented by a 2nd-order factor consisting of the variance shared by three 1st-order factors: fluency, working memory, and switching. This factor structure was hypothesized based on prior work indicating that executive function has multiple component processes and is distinct from episodic memory and processing speed (Friedman & Miyake, 2004; Hedden & Yoon, 2006; Miyake et al., 2000; Park et al., 2002; Salthouse et al., 2003; Salthouse, 2011). The hypothesized model is depicted in Figure 1. Briefly, we hypothesized that each observed variable would be explained by a single factor, and that all correlations involving fluency, working memory, and switching would operate through their shared variance in the executive function 2nd-order factor. We also allowed for correlated errors between observed variables drawn from the same task.
Before examining relationships of amyloid burden and WMH to the cognitive factors, we evaluated and finalized the factor structure on the neuropsychological data without links to the imaging data. We then conducted a path analysis by simultaneously entering age, estimated VIQ, amyloid burden, and WMH as predictors of the cognitive factors. We hypothesized that age would be negatively related to all factors, VIQ would be positively related to all factors, that amyloid would be negatively related to episodic memory, and that WMH would be negatively related to executive function. Because of the strong directional nature of our hypotheses, we used one-tailed significance values when evaluating individual correlations and loadings in the path analyses. The number of tests relevant to each hypothesis was limited by the number of factors (3 for comparisons between executive function, memory, and speed; 3 for the component factors of fluency, working memory, and switching).
The primary measure of model fit used was the chi-square statistic, which should be non-significant for a well-fitting model. Relative change in chi-square was used when comparing alternative models. We also report two common fit indices: the Tucker-Lewis Index (TLI), for which values greater than .95 indicate good fit, and the Root Mean Square of Approximation (RMSEA), which should be less than .05 and accompanied by a non-significant p-value for closeness of fit (PCLOSE) (Hu & Bentler, 1999).
Amyloid burden was measured with N-methyl-[11C]-2-(4-methylaminophenyl)-6-hydroxybenzothiazole (Pittsburgh Compound B; PiB), which binds to fibrillar amyloid, and was prepared at Massachusetts General Hospital as described previously (Klunk et al., 2004; Mathis et al., 2003). Participants underwent PiB PET imaging as described previously (Gomperts et al., 2008; Hedden et al., 2009, 2012; Sperling et al., 2009). Briefly, data were acquired using a Siemens/CTI ECAT HR+ scanner (3-D mode; 63 image planes; 15.2cm axial field of view; 5.6 mm transaxial resolution and 2.4mm slice interval; 39 frames: 8x15s, 4x60s, 27x120s). After a transmission scan, 8.5 to 15mCi [11C]-PiB was injected as a bolus and followed immediately by a 60-minute dynamic acquisition. PET data were reconstructed and attenuation corrected, and each frame was evaluated to verify adequate count statistics and absence of head motion (inter-frame head motion, if present, was corrected prior to further processing).
The dynamic PET data were reconstructed with scatter correction using commercially available routines for 3-D PET data. The average of the initial 0–8 minutes of dynamic PET data was spatially normalized to a PET MNI template using SPM8. PET data were parameterized by the distribution volume ratio (DVR) computed using the Logan graphical analysis technique (Logan et al., 1990) applied to the frame data acquired 40–60 minutes after injection; this method has been fully validated for PiB imaging (Price et al., 2005). Time-activity curves were measured in each brain region under analysis (region of interest or voxel) and in a reference region in cerebellar cortex known to contain low levels of fibrillar amyloid. This approach has been applied to numerous PiB studies (e.g., Fagan et al., 2006; Johnson et al., 2007; Lopresti et al., 2005; Price et al., 2005) and yields data that are similar to arterial blood input methods (Lopresti et al., 2005).
For each subject an index of PiB binding in cortical regions was calculated using the dynamic data via Logan graphical modeling within a large aggregate cortical region of interest (ROI) consisting of frontal, lateral parietal and temporal, and retrosplenial cortices (the FLR region). PiB retention in the FLR region is substantial in patients with diagnosed AD and has been used as a summary measure of PiB retention in previous studies (Johnson et al., 2007; Gomperts et al., 2008; Hedden et al., 2009, 2012). FLR DVR was log-transformed because of the non-normal distribution of PiB values and treated as a continuous variable in all analyses. Despite this transformation, non-normality of the distribution remains evident and individuals with very high values of PiB may have a disproportionate influence on the results.
Participants underwent MRI on a Siemens TrioTIM 3-Tesla scanner (Siemens, Erlangen, Germany) equipped with a 12-channel phased-array whole-head coil. Head motion was restrained with a foam pillow and extendable padded head clamps. Earplugs were used to attenuate scanner noise. High-resolution 3D T1-weighted multi-echo magnetization prepared rapid acquisition gradient echo anatomical images were collected with the following parameters: TR = 2200 ms, multi-echo TEs = 1.54 ms, 3.36ms, 5.18ms, and 7ms, FA = 7°, 4x acceleration, 1.2x1.2x1.2 mm voxels. Fluid attenuation inversion recovery (FLAIR) images for visualization of white matter lesions were collected with the following parameters: TR = 6000 ms, TE = 454 ms, TI = 2100 ms, 1x1x1.5 mm voxels.
WMH were identified from each individual’s FLAIR image with an automated fuzzy-connected algorithm previously validated against a visual grading system (Wu et al., 2006) and using methods described previously (Hedden et al., 2012). This automated WMH segmentation method involves (i) automatic identification of WMH seeds based on the intensity histogram of the FLAIR image, (ii) using a fuzzy-connected algorithm to segment the WMH clusters, (iii) iteratively updating the set of seeds, and (iv) combining the WMH clusters into a final WMH segmentation. The histogram of the skullstripped (BET; Smith, 2002) FLAIR image was used to define a threshold, which was set to be the mean ± 2.5 standard deviations. From the resulting WMH segmentation we extracted the total WMH volume in mm3 within a mask defined by the Johns Hopkins University White Matter Atlas (Wakana et al., 2004), which was reverse normalized to the native space of each individual’s FLAIR image. Because the automated algorithm is designed to detect regions of relative hyperintensity, this can lead to false detections, particularly in individuals with little or no areas of true hyperintensity (as in younger adults). Assumptions built into the algorithm, such as the threshold applied and the regions included in the mask, may introduce systematic bias; however, the classification and ordering of individuals by the resulting WMH estimates appear to be robust to changes in these assumptions. WMH volumes were log-transformed because of the non-normal distribution of WMH values and treated as a continuous variable in all analyses.
In confirmatory factor analysis, multiple individual task variables are used to estimate latent factors representing the shared variance among those tasks. Table 1 shows the scores and Table 2 displays the intercorrelations for the neuropsychological tests from which factors were derived. We applied an a priori factor structure based on previous work (Friedman & Miyake, 2004; Hedden & Yoon, 2006; Miyake et al., 2000; Park et al., 2002; Salthouse et al., 2003; Salthouse, 2011). To prevent tuning the factor structure to the imaging data, we hypothesized the factor structure in advance, did not allow modifications without strong theoretical justification, and finalized the factor structure on neuropsychological data without links to amyloid or WMH imaging data. We took a strict stance in evaluating our hypothesized model (depicted in Figure 1), allowing for only one non-hypothesized path from memory to the Self-Ordered Pointing (SOP) task, which involves repeated trials and therefore has an episodic memory component. This model fit the data well χ2(79) = 85.49, p = .29, TLI = .986, RMSEA = .022, PCLOSE = .95. The original model excluding the path from memory to the SOP task also fit the data well, χ2(80) = 101.11, p = .06, TLI = .954, RMSEA = .040, PCLOSE = .76, but the addition of this path resulted in a significant improvement in model fit, Δχ2(1) = 15.62, p < .001. Although we accepted this model for its failure to reject our a priori hypothesis of the specified factor structure, this model was also preferred relative to all tested alternative models (see Table 3); these alternatives were evaluated solely to confirm that our hypothesized factor structure provided a better fit to our data than do other likely hypothetical factor structures. This model was used to extract factor scores for each participant in the sample; these were carried forward for additional analysis.
To assess how model fit may be affected by our sample characteristics, we estimated the accepted model in two subsamples. The first was a subsample of 143 subjects after randomly removing 25 individuals aged < 70 years to account for oversampling of individuals in this age range in our sample. The second was the subsample of 109 subjects with both amyloid and WMH imaging data, although it should be noted that fit in this subsample was evaluated without reference to the imaging data. As shown in Table 3, good fit was obtained for both subsamples and the fit within these subsamples was comparable to that in the full sample. No major qualitative differences were observed in the factor loadings or inter-factor correlations between these subsample models and the accepted model. These results indicate that the model fit and resulting factor scores are unlikely to be affected by oversampling of younger subjects or the inclusion of individuals without imaging data types.
To estimate the cross-sectional age effects on the factor scores across the adult lifespan, we used data collected from a comparison sample of 38 younger adults and computed the factor scores for executive function and its components, memory, and processing speed using the factor weights derived from the sample of 168 older adults. This analytic strategy assumes that the factor structure is the same for younger and older adults; we are unable to test this assumption with confidence given the small sample of younger adults. These younger adults had significantly lower estimated VIQ scores than did the 168 older adults, t(204) = 2.51, p = .01, yet outperformed the older adults on all factors – executive function: t(204) = 7.80, p < .001; fluency: t(204) = 5.80, p < .001; working memory: t(204) = 5.30, p < .001; switching: t(204) = 6.25, p < .001; memory: t(204) = 10.89, p < .001; processing speed: t(204) = 10.20, p < .001. These results exemplify the sensitivity of the factor scores for detecting cross-sectional age differences between younger and older adults. Note that because the factor scores were derived from the older adults and applied without additional tuning to the younger adults, methodological influences leading to misfit in out-of-sample data would be expected to work against detecting such cross-sectional age differences.
Within the 168 older adults, cross-sectional age effects were observed on all factors – executive function: r = −0.26, p = .001; fluency: r = −0.22, p = .004; working memory: r = −0.19, p = .01; switching: r = −0.27, p < .001; memory: r = −0.30, p < .001; processing speed: r = −0.24, p = .002. The consistency of the relations to age across domains is primarily due to the selection of domains known to be sensitive to age effects.
In addition to the observed age effects, we examined a proxy of crystallized intelligence (estimated VIQ) as a predictor of cognitive performance among the older adults. Within the 168 older adults, VIQ had a significant positive relationship to all factors – executive function: r = 0.50, p < .001; fluency: r = 0.50, p < .001; working memory: r = 0.50, p < .001; switching: r = 0.44, p < .001; memory: r = 0.33, p < .001; processing speed: r = 0.32, p < .001. VIQ and age were not correlated in this sample, r = 0.04, p = .65.
Within the sample of 109 older adults with both amyloid and WMH data available, amyloid burden (PiB FLR) was M = 1.24, SD = 0.20 in DVR units and M = 0.20, SD = 0.15 in ln(DVR) units. WMH values were M = 2407, SD = 2727 in volumetric units and M = 7.34, SD = 0.94 in ln(WMH) units. Raw units are presented for descriptive purposes only; log units were used in all analyses because of the non-normal distribution of both amyloid burden and white matter hyperintensities. Age was not significantly related to amyloid burden, r = 0.13, p = .09, whereas age was significantly related to white matter hyperintensities, r = 0.41, p < .001.
We examined the relationship between amyloid and WMH and found no significant relationship (r = 0.10, p = .14; controlling for age: r = 0.06, p = .28; Figure 2). This corresponds to our previous finding (Hedden et al., 2011) of a nonsignificant relationship between amyloid and WMH in a smaller sample and with independent results (Marchant et al, 2012; Rutten-Jacobs et al. 2011).
We examined amyloid burden and WMH for their relationship to cognitive performance by using the factor scores for each of the 109 participants with these imaging data types. Factor scores were extracted from the model of the full sample of 168 older adults (nearly identical results obtain if factor scores are extracted instead only from the 109 participants with full imaging data). These factor scores were entered as observed variables in a path analysis using age, VIQ, amyloid burden and WMH as predictors. In such models, the effect of each predictor on each outcome variable represents the unique influence of that predictor while simultaneously controlling for all other predictors. Because our previous model indicated that relationships between the executive function components (fluency, working memory, and switching) and other variables operated through the 2nd-order executive function factor, we entered only this 2nd-order factor as an outcome variable in the current analysis. Note that among the predictor variables, only age and WMH were modeled as being correlated, as all other correlations could be constrained to zero without impacting model fit.
This analysis is depicted in Figure 3 and produced a well-fitting model, χ2(5) = 6.02, p = .30, TLI = .985, RMSEA = .043, PCLOSE = .45. (When factor scores from the 109 participants are used, overall model fit was identical, χ2(5) = 6.02, p = .30, and no individual standardized path weight or correlation changed by more than .03.) However, we are less concerned with the full model fit and more so with the relative weightings and significance of individual paths from the four predictors to each factor score. As can be seen in Figure 3 and confirming the above analyses, the paths from VIQ to executive function, memory, and speed were all significant, as were the paths from age to each of the factors. The path from amyloid burden (PiB FLR) to memory was significant, p = .01, but the paths from amyloid burden to executive function and to speed were not significant (p = .50 and p = .36, respectively). These latter two paths can be constrained to equal zero without impacting model fit, Δχ2(2) = 0.47, p = .79. Further, the paths from amyloid burden to memory and from age to memory have equivalent weightings, as they can be fixed to be equal without impacting model fit, Δχ2(1) = 0.98, p = .75, suggesting that the influence of amyloid on memory performance is of similar size to that of age on memory.
The paths from WMH to the cognitive factor scores display a distinctly different pattern. The path from WMH to executive function was significant, p = .04. Further, the influence of WMH on executive function and of age on executive function are of approximately the same magnitude, as these paths can be fixed to be equal without impacting model fit, Δχ2(1) = 0.13, p = .72. In contrast, the paths from WMH to memory (p = .08) and to speed (p = .13) did not reach significance and constraining these paths to equal zero did not impact model fit, Δχ2(2) = 0.47, p = .79. Nonetheless, we cannot unreservedly conclude that WMHs have no influence on memory and speed, as constraining the paths from WMH to EF, to memory, and to speed to all be equivalent also did not impact model fit Δχ2(2) = 0.77, p = .68.
The influences of amyloid and WMH on each cognitive variable are illustrated in Figure 4, which shows the scatterplot of each cognitive factor score after regressing out age and VIQ as a function of amyloid burden or WMH. These results are similar to those from the preceding path analysis except that the simultaneous influences of amyloid and WMH are not explicitly modeled. As can be seen in the left panel, the effect of amyloid is selective to memory (r = −0.21, p = .01), with minimal impact on either executive function (r = 0.00, p = .50) or speed (r = −0.04, p = .34). In the right panel, it is apparent that although only the relationship between WMH and executive function is significant (r = −0.17, p = .04), this relationship is approximately the same as that with memory (r = −0.14, p = .08), and speed (r = −0.11, p = .13). These figures demonstrate that although significant relationships between amyloid, WMH, and cognition can be observed in cross-sectional data on cognitively normal older adults, the effect sizes are small and there remains substantial unexplained variance in cognitive performance among this sample.
We note that the selected model depicted in Figure 3 cannot be distinguished statistically from a model replacing the covariance among factors with a higher-order factor (general function, or Gf) that explains the shared variance in the executive function, memory and, speed factors. Using such a model, amyloid did not have a significant path to Gf (p = .45), and did have a significant path to memory (p = .001). In contrast, WMH had a significant relationship to Gf (p = .04), but did not have a significant relationship to executive function, memory, or speed after controlling for the relationship to Gf. These results conform with the above and suggest that amyloid burden has a specific association with memory performance, whereas the effects of WMHs on cognition are likely more general in nature.
In the above analyses, we examined the influence of amyloid and WMH on the 2nd-order factor for executive function under the assumption that these large-scale pathological measures would not selectively affect executive function component processes. We additionally examined the correlations between each marker of neuropathology and the executive function component factor scores controlling for age and VIQ. For amyloid burden, we found no significant relationship with fluency (r = 0.05, p = .31), working memory (r = −0.06, p = .26), or switching (r = 0.03, p = .38). For WMH, we found a significant relationship for switching (r = −0.29, p = .001), but the association with fluency (r = −0.14, p = .07) and working memory (r = −0.01, p = .48) did not reach significance. The magnitude of the association between WMH and switching was roughly twice as large as that between WMH and the 2nd-order executive function factor described above (r = −0.17), while the association between WMH and fluency is of approximately the same size as that effect. This suggests that WMH may have a selective influence on certain component processes of executive function above and beyond the more general effects involving large-scale cognitive functions (Oosterman et al., 2004), but this finding must be interpreted cautiously as it was not predicted.
In a sample of cognitively normal older adults, we found an association between amyloid burden and memory and between WMH and executive function, even after controlling for age and estimated VIQ. The association between amyloid burden and memory appears to be somewhat specific, as no relationship between amyloid and executive function or speed of processing was observed. The influence of WMH on cognition may be more general, however, as paths from WMH to memory and speed of processing did not significantly differ from that of WMH to executive function.
Amyloid burden and WMH were not significantly associated with one another, suggesting that these may result from separable neuropathological cascades among cognitively normal older adults. This corresponds to recent findings in cognitively normal samples (Hedden et al., 2012; Marchant et al, 2012; Rutten-Jacobs et al. 2011). However, it must be noted that failure to observe a relationship between amyloid and WMH in samples restricted to cognitively normal older adults may represent selection bias (those suffering from a “double hit” may be more likely to be classified as cognitively impaired) and we therefore cannot rule out the possibility of an association between these pathologies that may become evident at later disease stages (see Brun & Englund, 1986; Grimmer et al., 2012; Horiuchi et al. 2012; Horsburgh et al., 2011; Polvikoski et al., 2010).
The impact of amyloid burden on cognition among cognitively normal individuals has become a topic of immense interest since the advent of in vivo amyloid imaging and the ability to measure CSF biomarkers of amyloid (Klunk et al., 2004; Scheuner et al., 1996). Results within cognitively normal adults have been equivocal due to small sample sizes and inconsistent methods, although recent studies with larger sample sizes appear to be converging on memory as an important cognitive factor that is impacted by amyloid (Klunk, 2011). However, not all studies show a selective effect of amyloid burden on memory, also finding significant effects on global cognition and executive function (e.g., Bennett et al., 2012; Rodrigue et al., 2012).
The current study adds to this emerging picture, indicating that in a highly selected cross-sectional sample, in vivo imaging of amyloid burden is associated with memory function, but minimally so with executive function or speed of processing. We used multiple measures of each cognitive domain modeled with confirmatory factor analysis so that measurement error associated with individual tasks was minimized. In addition, our measures of memory included tasks that were designed to be more challenging (H. Buschke, personal communication; Rentz et al., 2010, 2011) than many standard neuropsychological tests of memory. Hence, these measures may be more sensitive to subtle changes in memory function than have been observable in some previous studies.
WMH appear to be associated with a different cognitive profile from amyloid burden in this sample. In particular, WMH were associated most strongly with executive function. When specific domains of executive function were examined, WMH were more strongly associated with task switching than with fluency or working memory. However, WMH also had non-significant but non-trivial relationships to memory and speed of processing, suggesting that the effects of WMH on cognition may be less specific in nature. These results accord well with previous findings and particularly with the results of two meta-analyses of associations between WMH and cognition (Gunning-Dixon & Raz, 2000; Oosterman et al., 2004). WMH may therefore represent a separate neuropathological cascade from amyloid accumulation and may be associated with cerebrovascular disease (Reed et al., 2007).
We note that the effects of both amyloid on memory and WMH on executive function were small, each accounting for less than 5% of the variance in cognitive function. However, each had a comparable effect size as that of age on the respective cognitive domains, indicating that this is a non-trivial influence when accounting for age-associated individual differences in cognitive function. The clinical implications of such small effects may be minimal, although it should be remembered that the effects observed here likely represent the earliest stages of neuropathology; as individuals progress toward clinically detectable impairment, effect sizes are likely to increase. The small effect size may, in part, account for the variation in the literature relating amyloid burden to cognition, especially among earlier studies with small samples. It is also possible that APOE genotype or other genetic factors may influence the relationship between pathology and cognition (Kantarci et al., 2012). It is important to point out that our sample was highly select in terms of education, estimated intelligence, and socio-economic status. Given that education and intelligence may provide a buffer against some effects of neurodegeneration on cognition (van Veluw et al., 2012), our results may not be fully generalizable to the population at large and the effect sizes may differ from those in a less homogeneous sample. Nonetheless, these results indicate that differential relations between amyloid burden and WMH with cognition are likely to be subtle but observable among cognitively normal older adults when using challenging tests with relatively large samples.
The results additionally indicate that the unique contribution of age to all cognitive domains remained significant even after accounting for the influence of both amyloid burden and WMH. Hence, there remains a substantial age-associated influence on cognition that is not explained by these two common markers of neuropathology. The residual relationship between age and cognition may be better explained by other age-related neurobiological changes, such as neurotransmitter depletion, gray matter atrophy, structural changes in white matter tracts, or declines in functional connectivity. It is also possible that these other neurobiological changes mediate the relationships between amyloid, WMH, and cognition, which would be consistent with the small effects identified in our analysis. Gray matter atrophy, for example, may be a more proximate indicator of neural dysfunction in advanced aging and may mediate the small influences of amyloid burden or WMH on cognition (He et al., 2012; Head et al., 2008; Mormino et al., 2008). Simultaneous measurement of multiple markers of age-related neurobiological change will be necessary to fully understand the complex interactions that underlie cognitive decline in advanced aging (Salthouse, 2011).
In sum, these results indicate that within cognitively normal older adults, increased amyloid burden and increased WMH are associated with different cognitive sequelae, adding to the previously reported dissociation of the two biomarkers, which typically occur in different individuals within the cognitively normal range (Hedden et al., 2012; Marchant et al, 2012; Rutten-Jacobs et al. 2011). This supports the hypothesis that advanced aging is associated with multiple neuropathological cascades that differentially impact systems related to memory and executive function (Buckner, 2004; Hedden & Gabrieli, 2004; Hedden et al., 2012). That differential cognitive profiles of biomarkers of neuropathology can be observed cross-sectionally in a healthy sample indicates that the earliest stages of neuropathological progression may have a small but measurable cognitive impact and provides hope that such profiles could be used for early disease detection and to measure the success of potential interventions. Most directly, the present results imply that distinct intervention strategies may be required to mitigate early decline in memory and executive function given that they may arise from distinct neuropathological cascades.
We thank Lauren Wadsworth, Natacha Lorius, and Emily Shire for assistance during subject recruitment and data collection. The Massachusetts Alzheimer’s Disease Research Center, Gad Marshall, Brendon Boot, and Mykol Larvie provided assistance with clinical and neuropsychological characterization of participants. The Massachusetts General Hospital Molecular Imaging PET Core provided assistance with amyloid imaging. Christopher Gidicsin, Jacqueline Maye, Bill Klunk, and Chet Mathis provided assistance with PiB imaging. Minjie Wu provided software for labeling WMH and Koene Van Dijk advised on the adaptation of this software. Donald McLaren provided valuable discussion. Dr. Herman Buschke and the Albert Einstein College of Medicine of Yeshiva University provided the Memory Capacity Test. This research was carried out in part at the Athinoula A. Martinos Center for Biomedical Imaging at the Massachusetts General Hospital, using resources provided by the Center for Functional Neuroimaging Technologies, P41RR14075, a P41 Regional Resource supported by the Biomedical Technology Program of the National Center for Research Resources (NCRR), National Institutes of Health. This work also involved the use of instrumentation supported by the NCRR Shared Instrumentation Grant Program and/or High-End Instrumentation Grant Program; specifically, grant numbers 1S10RR023401, 1S10RR019307, and 1S10RR023043. Funding was provided by National Institute on Aging grants P01 AG036694, P50 AG0513421, R01 AG034556, R01 AG027435, P41 RR14075 and K01 AG040197; The Alzheimer’s Association grant IIRG-08-90934, and by the Howard Hughes Medical Institute.
Conflict of Interest: None