Age-Related Differences in Memory and Executive Functions in Healthy APOE ε4 Carriers: The Contribution of Individual Differences in Prefrontal Volumes and Systolic Blood Pressure

Abstract

Advanced age and vascular risk are associated with declines in the volumes of multiple brain regions, especially, the prefrontal cortex, and the hippocampus. Older adults, even unencumbered by declining health, perform less well than their younger counterparts in multiple cognitive domains, such as episodic memory, executive functions, and speed of perceptual processing. Presence of a known genetic risk factor for cognitive decline and vascular disease, the ε4 allele of the apolipoprotein E (APOE) gene, accounts for some share of those declines; however, the extent of the joint contribution of genetic and physiological vascular risk factors on the aging brain and cognition is unclear. In a sample of healthy adults (age 19–77), we examined the effects of a vascular risk indicator (systolic blood pressure, SBP) and volumes of hippocampus (HC), lateral prefrontal cortex (lPFC), and prefrontal white matter (pFWM) on processing speed, working memory (WM), and recognition memory. Using path analyses, we modeled indirect effects of age, SBP, and brain volumes on processing speed, WM, and memory and compared the patterns of structural relations among those variables in APOE ε4 carriers and ε3 homozygotes. Among ε4 carriers, age differences in WM were explained by increase in SBP, reduced FWM volume, and slower processing. In contrast, lPFC and FWM volumes, but not BP, explained a share of age differnces in WM among ε3 homozygotes. Thus, even in healthy older carriers of the APOE ε4 allele, clinically unremarkable increase in vascular risk may be associated with reduced frontal volumes and impaired cognitive functions.

Advanced age is associated with reduced regional brain volume and impaired performance in multiple cognitive domains (Raz & Rodrigue, 2006). These differences are exacerbated by multiple physiological and genetic variables. For example, hypertension, a common age-related vascular risk factor is associated with smaller brain volumes (Salerno et al., 1992), disproportionately smaller prefrontal cortices (Raz, Rodrigue, & Acker, 2003), and accelerated shrinkage of the hippocampus (Raz et al., 2005). Hypertension is also linked to reduced performance in multiple areas of age-sensitive areas of cognition: speed of perceptual-motor processing (Salthouse, 1996b), episodic memory (Brady, Spiro, & Gaziano, 2005; Harrington, Saxby, McKeith, Wesnes, & Ford, 2000; Singhmanoux & Marmot, 2005), and executive functions (Kuo et al., 2004). It is important to keep in mind, however, that hypertension is a categorical diagnosis based on cut-off value of arterial blood pressure (Chobanian et al., 2003), which is a continuous variable. Therefore, it is plausible that individual differences in blood pressure account for variance in brain volumes and cognitive performance, even in normotensive individuals.

A significant proportion of individual differences in brain parameters, cognitive performance, and vascular risk are explained by genetic makeup (see Jagust, 2009 for commentary and review; see Song, Stampfer, & Liu, 2004 for meta-analysis and review). Among the multitude of genetic variants that contribute to variance in brain and cognition, the best known is a polymorphism of a gene that controls a protein responsible for trafficking of lipids in the blood vessels, apolipoprotein E (see Hauser, Narayanaswami, & Ryan, 2011 for review). The protein, APOE, has three isoforms, E2, E3, and E4 that vary in their capability to transport lipids, and each isoform corresponds to a specific variant of APOE gene: ε2, ε3, ε4, and thus is differentially expressed in the carriers of the respective alleles. The ancestral allele of APOE, ε4 is an established risk factor both for vascular disease (Haan & Mayeda, 2010; Song et al., 2004) and Alzheimer’s dementia (AD; Roses et al., 1993; Lesser, Beeri, Schmeidler, Purohit, & Haroutunian, 2011), with ε4 homozygotes carrying a 10- to 12-fold risk for AD in comparison to ε3 homozygotes. Moreover, the deleterious effects of the ε4 allele may be exacerbated by advanced age (Haan & Mayeda, 2010; Haan, Shemanski, Jagust, Manolio, & Kuller, 1999; Nilsson et al., 2006). It stands to reason, therefore, that carriers of APOE ε4 may evidence reductions in cerebral regions and cognitive skills that are particularly vulnerable to AD and cerebrovascular disease. The extant literature, though not entirely consistent, provides guarded support for that prediction (see Buckner, 2004 for a review).

Several studies associated ε4 with smaller brain volumes, including prefrontal and medial temporal regions (Bartzokis et al., 2007; Espeseth et al., 2008; Honea, Vidoni, Harsha, & Burns, 2009; Wishart et al., 2006), although others have not found such a link (Cherbuin et al., 2008), and in some small samples even the reversed effect was observed (Striepens et al., 2011). A recent meta-analysis showed that possession of the ε4 allele by healthy adults is associated with reduced performance on age-sensitive cognitive tasks, such as executive functions and episodic memory, and that the effects worsen in old age (Wisdom, Callahan, & Hawkins, 2011). Some studies suggest that the negative effects of APOE ε4 allele on brain volume and cognition are mediated by phenotypic vascular risk factors, such as elevated systolic blood pressure (SBP; Peila et al., 2001), or age-associated endothelial cell dysfunction (Yavuz et al., 2008). In addition to phenotypic vascular risk, APOE ε4 may also interact with genes involved in neurotransmission to yield impairments in white matter volume and perceptual-motor speed (Espeseth et al., 2006).

Because cognitive abilities are correlated, it is unclear whether the effect of ε4 on episodic memory (Bondi et al., 1995; Wilson et al., 2002) is specific or reflects its association with other cognitive variables, such as processing speed, or working memory (WM) and executive processes (Blair, Vadaga, Shuchat, & Li, 2011; Parasuraman, Greenwood, & Sunderland, 2002; Reinvang, Winjevoll, Rootwelt, & Espeseth, 2010; Wetter et al., 2005). Moreover, genetic risk conveyed by APOE ε4 and physiological vascular risk expressed by various biomarkers may act in synergy in reducing cognitive performance. Vascular risk factors reported to interact with ε4 include elevated triglycerides (De Frias et al., 2007), homocysteine (Elias et al., 2008), and blood pressure (Haan & Mayeda, 2010; Haan et al., 1999; Kuller et al., 1998).

In this study, we examined whether the pattern and magnitude of contributions of individual differences in regional brain volumes and vascular risk to age-related differences in age-sensitive areas of cognition differ between the carriers of a risky APOE ε4 allele in a lifespan sample of healthy adults. Specifically, we hypothesized that in APOE ε4 carriers, higher indicators of physiological vascular risk (sub-clinically elevated SBP) would be associated with reduced brain volumes and impaired cognitive performance, whereas ε3 homozygotes would not demonstrate such vulnerability.

Method

Participants

Study data were collected as part of an ongoing study of the cognitive and neural correlates of healthy aging taking place in the Detroit metropolitan area. All participants provided informed consent, consistent with the University Human Investigations Committee guidelines. All participants completed a self-report health questionnaire to screen for health issues including medical diagnosis of hypertension and use of prescribed anti-hypertensive medication, as well as history of cardiovascular disease, neurological, or psychiatric disease. Participants denied history of cancer, head trauma accompanied by loss of consciousness for more than five minutes, thyroid disorder, and diabetes. Persons who acknowledged prior or current treatment for drug or alcohol abuse, or taking more than three drinks per day were not recruited for this study. All participants denied taking anticonvulsive, antidepressant, antihyperglycemic, antipsychotic, or anxiolytic medications. In addition, participants who reported taking anti-hypercholesterolemic medications were not included in the present study.

The participants were native English speakers, with at least high school diploma or equivalency; mean education corresponded to nearly a full four-year college: 15.8 ± 2.4 years. Participants completed a questionnaire (CES-D; Radloff, 1977; cutoff = 15) to screen for depressed state, and an experimenter screened participants for cognitive impairment using the Mini Mental Status Examination (MMSE; Folstein, Folstein, & McHugh, 1975; cutoff = 26). Participants were screened for near, far, and color vision problems (Optec 2000 Vision Tester, Stereo Optical Co., Inc., Chicago, IL) and speech-range hearing deficits (MA27 Screening Audiometer, Maico Diagnostics, Eden Prairie, MN). An experimenter measured participant blood pressure with an analog mercury sphygmomanometer (Model 12-525; Country Technology, Gays Mills, WI), using a left arm brachial cuff. Measurements were taken on three separate days prior to cognitive testing with the participant comfortably seated in a quiet room; measurements were averaged across occasions.

The sample was composed of 72 adults, 19 to 77 years of age (see Table 1 for sample descriptive statistics), including 50 women and 22 men. None of the participants reported diagnosis of hypertension or use of anti-hypertensive medication. Men and women sub-samples did not differ in mean age, years of education, MMSE scores, systolic and diastolic blood pressure, and proportion of participants reporting regular smoking or exercise. The sample overlapped with prior studies from our lab as follows: 65% sample overlap with (Raz, Rodrigue, Kennedy, & Land, 2009), 60% sample overlap with (Raz et al., 2008), 64% sample overlap with (Raz, Dahle, Rodrigue, Kennedy, & Land, 2011), and 64% sample overlap with (Dahle, Jacobs, & Raz, 2009).

Table 1.

Sample Demographic Characteristics

Variable	Men	Women	t or χ2a	p
	Mean ± SD	Mean ± SD
Age	49.0 ± 17.3	50.40 ± 12.9	0.36	0.72
Education	15.7 ± 2.7	15.8 ± 2.3	0.18	0.86
MMSE	28.8 ± 1.2	28.9 ± 1.1	0.45	0.65
SBP (mm Hg)	122.3 ± 10.2	119.6 ± 12.8	1.19	0.24
Diastolic BP (mm Hg)	75.6 ± 8.6	73.1 ± 6.5	1.36	0.18
Smokers	1 (4.5%)	8 (16.0%)	1.83	0.18
Exercise	18 (82.0%)	41 (81.8%)	0.00	0.99
Exercise Frequency	3.6 ± 2.2	3.6 ± 2.3	0.07	0.94
APOE ε4 Carriers	5 (22.8%)	15 (30.0%)	0.40	0.53
Non-whites	9 (40.9%)	10 (20.0%)	3.43	0.06

Variable	ε3 Homozygotes	ε4 Carriers	t or χ2a	p
	Mean ± SD	Mean ± SD
Age	48.3 ± 14.4	54.2 ± 13.2	1.56	0.12
Education	15.9 ± 2.6	15.6 ± 2.0	0.55	0.58
MMSE	28.7 ± 1.2	29.2 ± 0.7	1.90	0.06
SBP (mm Hg)	121.3 ± 11.9	119.5 ± 12.9	0.56	0.58
Diastolic BP (mm Hg)	74.1 ± 7.7	73.3 ± 6.3	0.39	0.70
Smokers	7 (13.4%)	2 (10.0%)	0.15	0.69
Exercise	44 (84.6%)	15 (75.0%)	0.80	0.43
Exercise Frequency	3.4 ± 2.0	3.9 ± 2.6	0.90	0.34
Non-whites	12 (23.1%)	7 (35.0%)	1.39	0.24

^asingle-degree of freedom chi-square test

Percentages are by group, not total

Genomic analysis

DNA was isolated from buccal cultures obtained in mouthwash samples with Gentra Autopure LS under the standard buccal cell protocol. DNA isolations and genotyping assays were conducted in the Wayne State University Applied Genomics Technology Center on an Applied Biosystems 7900. APOE (rs429358 and rs7412) polymorphisms were preamplified with forward 5′-CAATGCTACCGAGTTTTCTTCC-3′ and reverse primers 5′-TTCAGATTCTTCACAGATGCGTA-3′ in a 25 μl reaction containing 2.5 mmol/l MgCl₂, 0.5 μmol/l of the primers, 1.25 U AmpliTaq Gold polymerase, and 200 μmol/l dATP, dCTP, dGTP, and dTTP. The mixture was denatured at 95°C for 10 minutes and amplification achieved by 15 cycles of 94°C for 30 seconds, 58°C for 30 seconds, and 72°C for 1 minute, followed by a final extension at 72°C for 10 minutes. One μl of this reaction was subsequently used for rs429358 and rs7412 5′-nuclease assays under standard conditions. The primers and probes for the rs7412 assay were 5′-TCCGCGATGCCGATGAC-3′, 5′-CCCCGGCCTGGTACAC-3′, VIC-CAGGCGCTTCTGC-NFQ and FAM-CAGGCACTTCGC-NFQ. The primers and probes for the rs429358 assay were 5′-GCGGGCACGGCTGT-3′, 5′-GCTTGCGCAGGTGGGA-3′, VIC-CATGGAGGACGTGTGC-NFQ and FAM-ATGGAGGACGTGCGC-NFQ. DNA sequencing reactions was carried out using the 0.5X protocol for ABI PRISM BigDye Terminator Cycle Sequencing Ready Reaction Kit (Applied Biosystems). The sequencing extension products were purified utilizing Sephadex, and analyzed on an ABI PRISM 3700 DNA Analyzer with a 50 cm capillary array.

Participants identified as APOE ε2 carriers were excluded from analysis because APOE alleles ε2 and ε4 may exercise opposite effects on memory and cognition (Helkala et al., 1995; Small, Rosnick, Fratiglioni, & Bäckman, 2004), and there were too few APOE ε2 carriers for statistical comparison. The allelic distribution of all three polymorphisms fit the Hardy-Weinberg equilibrium (all χ² < 1). Mindful of reported differences among populations in prevalence of allelic distributions (Corbo & Scacchi, 1999), we evaluated APOE ε4 frequency among Caucasians and African-American participants. Within each group, the distribution conformed to Hardy-Weinberg equilibrium (χ² = 1.12, p > .1, and χ² = 0.93, p > .1, respectively) and there was no significant difference in APOE ε4 frequency between African-American and Caucasian participants: n = 15 (40%) vs. n = 53 (25%), χ² = 1.39, p > .1. There were too few participants from other ethnicities (n = 4) for additional analyses. Most of the participants were homozygous for ε3 allele (72%), whereas 25% were ε3/ε4 heterozygotes and less then 3% were homozygous for the ε4 allele; ε4 carriersmade up 28% of the sample.

MRI Acquisition and Processing

MRI Acquisition

T1-weighted magnetization-prepared rapid acquisition with gradient echo (MPRAGE) images were acquired using a 4T MRI system (Bruker Biospin, Ettlingen, Germany) equipped with an 8-channel RF head coil. Sequence parameters were as follows: TR = 1600 ms, TE = 4.38 ms, TI = 800 ms, FOV = 256 × 256 mm², in plane resolution = 0.67 × 0.67 mm², slice thickness = 1.34 mm, matrix size = 384 × 384, number of slices = 176.

MRI Processing

Image volumes were created from DICOM images and filtered to correct for field-inhomogeneity using Analyze software (Biomedical Imaging Resource, Mayo Clinic, Rochester, MN). Filtered volumes were adjusted to correct for variability in head pitch, rotation, and tilt using the oblique sections and orthogonal alignment tools in Analyze. Using a parasagittal view, separate images were generated aligned along the anterior and posterior commissure line (see Raz, 2004 for details), and perpendicular to the longitudinal axis of the hippocampus for measurement of frontal and hippocampal volumes, respectively.

Volumetry

Images were displayed on a 21″ LCD digitizing tablet (Wacom Cintiq 21UX, WACOM, Vancouver, Washington; see Raz, 2004 for details). All regions of interest (ROIs) were manually measured in the coronal plane using the ROI tool in the Analyze software, except for intracranial volume (ICV), which was measured in the axial plane. Regional volumes were calculated by multiplying the summed bilateral area by slice thickness. Each ROI was regressed on ICV and the slopes were compared for men and women prior to adjustment to ensure equivalence of regression slopes between sexes. The following equation was applied to each volume to correct for individual differences in head and body size: Volume_adj = Volume_raw − b(ICV − Mean ICV). Here, Volume_adj is adjusted volume, Volume_raw is raw volume, b is the unstandardized regression coefficient from regressing the ROI on ICV, and Mean ICV is taken from the entire sample.

ROI Demarcation Procedures

Inter- and intra-rater reliability for measurement of regional volumes was determined by an intraclass correlation formula that assumes random assignment and independence of raters, (ICC[2]/ICC[3]; P. E. Shrout & Fleiss, 1979). All regions were traced by six-seven reliable raters and ICC(2) values ranged from .94 to .99.

Lateral Prefrontal Cortex (lPFC)

The lPFC was traced on 10 to 14 coronal slices found within 40% of the distance between the frontal pole and first slice on which the genu was visible. Starting with the first slice, every other slice (1.34 mm) was traced for a total of 9 to 12 lPFC slices. The lPFC ROI included superior, middle, and inferior frontal gyri, and corresponds to Brodmann areas 9 and 46 with partial overlap with areas 8, 10, and 45. Figure 1a shows an example of the lPFC ROI, which is bounded superiorly by the frontal cortex’s most dorsomedial point and inferiorly by the lateral orbital sulcus.

Figure 1.

Representative examples of manual measurement of regions of interest. Images are in radiological orientation. A. Example of lateral prefrontal cortex ROI; B. Example of prefrontal white matter ROI (from same slice as A.); C. Depiction of hippocampal ROI.

Prefrontal White Matter (pFWM)

The pFWM was traced on the same slices as lPFC; all white matter was traced on each slice, whether or not it was contiguous with other visible white matter (Figure 1b). Care was taken to exclude the frontal horns of the lateral ventricles and other non-white matter from the measurements.

Hippocampus (HC)

The HC was traced in a para-coronal view, orthogonal to the longitudinal axis of the right HC. The slice on which the mammillary bodies were most visible marked beginning of the range of measurement, and the slices in which fornices emerge from the fimbria mark the posterior limit. HC measurements excluded amygdala from rostral tracings, and care was taken not to include the fimbria in the ROI. As shown in Figure 1c, HC tracings included the entire hippocampal formation, and the inferomedial boundary was the most medial aspect of the parahippocampal gyrus, just below the hippocampal sulcus. Temporal lobe white matter and the temporal horn of the lateral ventricles were the inferior and lateral boundaries, respectively.

ICV

Using the auto-trace function in Analyze, the ICV was traced in the axial plane starting at the cerebral vertices on the first slice where cortical gray matter was visible. A total of 10 slices were measured with a gap of 14 slices (9.4 mm) in between. This range corresponds to the majority of the cranial vault, excluding the extra-cortical cranial vertex and base of the brain below the level of the orbits. A seed point for the auto-trace was chosen on the edge of the cranium and the auto-limit adjusted to include the entire cranium in the ROI, and the auto-traced ROI was manually edited as needed to ensure consistency with apparent anatomy.

Cognitive Tests

Participants completed a series of cognitive tests of age-sensitive abilities including WM, speed of perceptual processing, and episodic memory. Five separate measures of WM included spatial, verbal, and non-verbal stimuli, using a combination of task designs (e.g., span tasks, n-back tasks); this heterogeneity improves construct validity. Similarly, speed of processing was assessed with well-validated verbal and nonverbal tasks (Salthouse, 1996b). In light of disproportionate age-related impairments in associative recognition, we administered the task used by Naveh-Benjamin (2000) in his test of an associative deficit hypothesis.

Working Memory: Size Judgment Span

Designed by Cherry & Park (1993) this measure requires participants to hold items in WM, make comparisons based on semantic features, and order them by ascending physical size for subsequent verbal report. In each trial an experimenter reads lists of items, and there are three trials per set. The test initially presents two items per trial and the number of items increases by one upon successful completion of each three-trial set. The task’s estimated reliability coefficient is .79 (Cherry & Park, 1993), which corresponds to the disattenuated test-retest correlation (Spearman, 1904) observed in our longitudinal study (Raz, unpublished data) further confirming the measure’s reliability in our larger sample.

Working Memory: Spatial Recall

A modified, computerized test adapted from the task described by Salthouse (1974, 1975; Salthouse, Kausler, & Saults, 1988), was used to assess spatial WM. A computer program displayed 30 5 × 5 matrices, each with seven darkened cells for 3 s. The experimenter provided participants with response forms that included three pages of two columns of five blank matrices. The experimenter informed participants that following the presentation of each matrix, they were to recall the locations of the darkened squares by drawing an X in seven cells on corresponding blank matrix on the response form; participants were told to guess if necessary. The task provided five practice trials, followed by 25 test trials. Test performance was calculated by averaging the total number correct across the 25 test trials. Cronbach’s alpha (α) was .89, as computed across the 25 test trials.

Working Memory: Listening Span

The Listening Span (LSPAN; Salthouse, Mitchell, Skovronek, & Babcock, 1989) task requires participants to listen to simple sentences, answer a multiple choice question about the sentence, and freely recall its final word (see Raz, Gunning-Dixon, Head, Dupuis, & Acker, 1998 for a full description). The LSPAN is organized into of seven blocks of three trials each. Starting with the first block, participants complete one item per trial, and the number of items increases by one for each successive block. Following presentation of all items in a trial, participants are told to recall as many of the final words as possible, in the original order of presentation. Participants are required to correctly answer questions in order to receive credit for reporting final words from the presented sentences. Participants receive one point for each correctly recalled and ordered final words. The absolute span (AS) is the sum of trials in errorless blocks, and has been used previously (Raz et al., 1998) as a measure of WM capacity.

Working Memory: N-Back Tests

An experimenter administered computerized n-back tests to assess WM storage and maintenance; separate tasks assessed verbal and nonverbal performance (modeled after Dobbs & Rule, 1989; Hultsch, Hertzog, & Dixon, 1990). The verbal n-back test displayed single-digit numbers on a CRT monitor, and the nonverbal n-back test presented abstract shapes. Participants were tested separately on 1-, 2-, and 3-back tests; sub-test order was counterbalanced across participants in a Latin square design. After all items in a trial were presented, the participants selected the item presented in the specified position for a given task. Number of correct responses (out of 20) was the performance index for both tasks in the present study. The tasks’ estimated reliability coefficients are .91 for the verbal and .88 for the nonverbal tests (Salthouse, 1996c). The present study only used data from the 3-back verbal and nonverbal tasks.

Processing Speed

Participants completed Letter Comparison and Pattern Comparison tests (Salthouse, 1996b) to assess speed of perceptual processing. Both tests required participants make same/different judgments on two pages of side-by-side letter strings or line patterns. An experimenter instructed participants to respond quickly and accurately; participants were given 30 seconds to complete as many items as possible on each page. The total number correct for both pages, divided by time allotted (number correct/60 s) is the index of performance. The estimated reliabilities for letter and pattern comparison are .77 and .87, respectively (Salthouse & Meinz, 1995).

Episodic Memory

Participants were administered a recognition test for word pairs (see Bender, Naveh-Benjamin, & Raz, 2010 for details; Naveh-Benjamin, 2000). Task conditions were intentional: an experimenter told participants to study and remember both the individual words and the pairs, and that they would be tested on both. Custom testing software written by laboratory staff in Visual Basic serially presented each participant with 26 pairs of unrelated words, at rate of 5.5 s per pair with a 200 ms inter-stimulus interval. After the study phase, participants were given a randomly generated 900 number and were told to count backwards by threes for 60 s to minimize rehearsal. Following study and distractor phases, participants completed separate recognition tests for items (individual words) and associations (word pairs), the order of which was counterbalanced across the sample. Both tests used a single item, yes/no design. The item test presented 16 individual words (8 targets/8 foils), and the associative recognition test presented 16 pairs (8 intact pairs/8 recombined pairs); participants indicated via keyboard button press whether or not a word was studied and whether pairs were intact or recombined. Upon completion of both tests the process was repeated with a second list of 26 new word pairs. Lists were randomly selected from a pool of six possible lists, and the order was randomized across the sample.

Data Conditioning

Participant age, mean SBP, and the ICV-adjusted lPFC, pFWM, and HC volumes were centered at their respective sample means and standardized to z-scores.

Working Memory Composite

The five performance indices from the corresponding tests of WM were screened for deviations from normality. A log transformation was applied to the AS performance index from LSPAN to correct for skewness in the distribution; no other transformations were required for the other WM measures. Results of an unrotated principal components analysis (PCA) showed that all five WM indices loaded onto a single common component. Component loadings are as follows: size judgment span = .78, spatial recall = .61, LSPAN = .79, 3-back verbal = .61, 3-back nonverbal = .76. A WM composite was therefore created, consisting of the averaged, standardized scores from the five measures.

Processing Speed

A PCA performed on the two scores from the processing speed measures showed both loaded onto a common component (both loadings > .90). Therefore, the scores were standardized and averaged to yield a single composite.

Recognition Tests

Prior to analysis of performance on the recognition tests, trials with response times shorter than 200 ms or longer than 10 s were excluded as such responses were likely due to error or may reflect other memory processes. The performance index for tests of recognition was A′, a non-parametric index of discriminability (Pollack & Norman, 1964); A′ was calculated using hit rate and false alarm rate, or proportions of correct target and incorrect lure responses, respectively (Stanislaw & Todorov, 1999). An arcsine transformation was applied to the A′ scores to correct for significant skewness in the distribution. Item and associative recognition A ′ scores were averaged across the two lists and standardized to create composite measures of item and associative recognition performance.

Statistical Analyses

The present study used a structural equations modeling (SEM) approach to conduct path analysis of statistical mediation. However, path and mediational analyses can provide valid estimates of neither direct causation nor change. Recent critical analyses confirmed both that correlational methods cannot inform causation and that cross-sectional regression and path coefficients cannot establish age-related change (Lindenberger & Pötter, 1998; Hofer et al., 2006; Maxwell & Cole, 2007; Lindenberger et al., 2011; Raz & Lindenberger, 2011). Thus, although we use familiar SEM terminology, “mediation” should only be considered as demonstrating consistent hierarchical partitioning of variance/covariance whereby individual differences in one variable are associated with individual differences in another.

To investigate the role of SBP as a contributor to the associations of age, hypothesized regional brain volumes, and different cognitive abilities, we used grouped path analysis in Mplus 6.0 (Muthén & Muthén, 2010). This approach involves specification of an initial model containing multiple hypothesized paths reflecting heterogeneous partitioning of variance based on group membership. Paths that are not hypothesized for a given group are constrained to zero to allow groups to feature different patterns of associations among the variables nested under a common model.

Because of statistical power limitations, the role of SBP as a mediator of the effects of age was modeled separately for WM via speed of processing, and for recognition of items and associations. In all models we used maximum likelihood (ML) estimation. Path analyses employed bias-corrected (BC) bootstrap resampling with 1000 draws to generate estimates and 95% confidence intervals (CIs) of indirect effects (Cheung & Lau, 2007; MacKinnon & Fairchild, 2009; Williams & MacKinnon, 2008); indirect effects are calculated as the product of two or more direct paths. BC bootstrapped indirect effects are widely considered superior estimates of mediation (MacKinnon & Fairchild, 2009; MacKinnon, Lockwood, Hoffman, West, & Sheets, 2002; Patrick E. Shrout & Bolger, 2002; Williams & MacKinnon, 2008) over other commonly used approaches such as those described by Baron and Kenny (1986) or Sobel (1982). Because this approach produces 95% CIs, any significant, non-zero indirect effect indicates mediation. Indirect effects were specified to test hypothesized mediation of age-volumetry and age-cognition associations. Model fit was assessed using several indices: the comparative fit index (CFI) and Tucker-Lewis Index (TLI) compare model fit to that of a null model, and values of .95 were cutoffs for both the CFI and TLI. For chi-square (χ²) tests of model fit, a nonsignificant (p > .05), smaller χ² value indicates acceptable fit in comparison to a null model. A related, more informative fit statistic, χ² divided by degrees of freedom (Jöreskog & Sörbom, 1993) used a fairly conservative cut-off value of ≤ 2.0 (Mueller, 1996). In addition, model fit was evaluated by inspection of root mean square error of approximation (RMSEA) and square root mean square residual (SRMR), measures of model misspecification and explained variance; acceptable fit was indicated by values of .05 and below for both the RMSEA and SRMR.

The first set of models assessed whether SBP mediates relationships between age and lPFC and pFWM volumes, processing speed, and WM. The initial model specified the following paths: age to the other five variables, SBP to both brain volumes, WM and processing speed, paths from both regional volumes to both processing speed and WM, and a path from processing speed to WM. The model was constrained differently for each group, consistent both with the hypothesis that SBP mediates the effects of age for carriers of the ε4 allele, but not for non-carriers, and with zero-order correlations (Table 2). Thus, for ε4 carriers, the direct effects of age on all measures except for SBP were constrained to zero; similarly, the direct effects of SBP on the two brain volumes and two cognitive composites were constrained to zero for non-carriers. Additional zero-constraints were also imposed to remove unreliable paths from the initial or nested models, yielding reduced grouped (mediational) path models (Figure 3a). Indirect effects were specified to test potential statistically causal paths, by which multiple variables are linked in mediating various associations therein. Indirect effects from Age to WM, speed, and pFWM and lPFC volumes freely estimated all possible indirect effects in the model.

Table 2.

Zero-order Correlations among Study Variables

Total Sample
Variable	1	2	3	4	5	6	7	8
1. Age	–
2. SBP	.36**	–
3. HC Volume	−.46***	−.16	–
4. lPFC Volume	−.61***	−.34**	.47***	–
5. pFWM Volume	−.49***	−.32**	.44***	.81***	–
6. Processing Speed	−.51***	−.31**	.36**	.50***	.47***	–
7. Working Memory	−.53***	−.36**	.41***	.51***	.49***	.67***	–
8. Item Recognition	−.24*	−.21	.12	.22	.21	.42***	.35**	–
9. Assoc. Recognition	−.42***	−.40***	.13	.36**	.27*	.48***	.61***	.62***

ε4 carriers
Variable	1	2	3	4	5	6	7	8
1. Age	–
2. SBP	.62**	–
3. HC Volume	−.74***	−.61**	–
4. lPFC Volume	−.47*	−.53*	.37	–
5. pFWM Volume	−.49*	−.49*	.57**	.76***	–
6. Processing Speed	−.65**	−.74***	.63**	.80***	.78***	–
7. Working Memory	−.58**	−.75***	.69***	.63**	.65**	.87***	–
8. Item Recognition	−.21	−.52*	.12	.10	.08	.32	.35	–
9. Assoc. Recognition	−.48*	−.62**	.24	.56*	.45*	.69***	.58**	.74***

ε3 homozygotes
Variable	1	2	3	4	5	6	7	8
1. Age	–
2. SBP	.29*	–
3. HC Volume	−.34*	.05	–
4. lPFC Volume	−.65***	−.29*	.51***	–
5. pFWM Volume	−.47***	−.28*	.37**	.83***	–
6. Processing Speed	−.47***	−.12	.23	.40**	.36**	–
7. Working Memory	−.51***	−.20	.27*	.47***	.42**	.58***	–
8. Item Recognition	−.26	−.08	.13	.26	.26	.46***	.35*	–
9. Assoc. Recognition	−.41**	−.32*	.09	.31*	.22	.41**	.62***	.58***

Notes:

^*p < .05,

^**p < .01,

^***p < .001

Figure 3.

Nested path models for ApoE ε4 carriers and ApoE ε3 homozygotes from reduced grouped mediational models. Only significant direct paths are shown. A. Models for mediation of associations with processing speed and working memory; B. Models for mediation of associations with recognition memory for items and associations. * p < .05, ** p < .01, *** p < .001

The second set of models examined potential mediation of the associations between age and recognition of items and associations by SBP, lPFC and hippocampal volumes. The path analysis included the two arcsine-transformed recognition memory scores, A′ (for item and association), age, SBP, lPFC and HC volumes. The initial model specified paths from age to SBP, lPFC and HC volumes, and the two recognition indices; direct paths were also specified from SBP to both brain volumes and both recognition indices, as well as from HC and lPFC volumes to the two recognition indices. In the nested model for carriers of the ε4 allele, paths from age to brain volumes and from age to associative memory were constrained to zero, whereas in the nested model for ε3 homozygotes paths from SBP to item recognition, prefrontal volume, and hippocampal volume were all constrained to zero. Thus, because paths were differentially constrained to zero by genotype, different indirect effects were possible in the nested models. As with the prior model of WM and speed, unreliable paths were constrained to zero; the resultant reduced grouped mediational path model (Figure 3b) included additional degrees of freedom due to estimation of fewer parameters. Indirect effects from Age to item and associative recognition, HC and lPFC volumes were estimated from the paths specified in the model.

Age-related differences in blood pressure or brain volumes may reflect individual differences in memory or cognition because those variables serve as proxies for another variable not included in the models. Mindful of this possibility, for both models we also tested two alternate path analyses: a reversed-mediation model and an exclusively correlational model to test the possibility that the relationships without specific directionality may fit the data as well. Both models were specified based on the final path model with no re-specification except for reversing the paths between blood pressure, brain volumes, and cognitive variables in the reversed-mediation model and substitution of correlations for direct paths in the correlational model. In the former, paths from age were specified as bidirectional correlational relationships. Thus, three separate models were evaluated for both speed/WM and recognition memory: a reduced group mediational model, a reversed-mediation model, and a correlational model.

Results

To examine genetic differences in the associations of age and SBP with regional brain volumes and cognitive measures, we compared zero-order correlations transformed via Fisher r-to-z formula (Table 2) in APOE ε4 carriers and ε3 homozygotes. We found that the relationship between age and hippocampal volume was stronger among ε4 carriers (r = .62, p < .001) than ε3 homozygotes (r = .29, p < .05; z = 1.5, p < .05). In addition, whereas among ε4 carriers higher SBP was significantly associated with smaller hippocampal volumes (r = −.61, p < .01), slower processing speed (r = .74, p < .001), and reduced WM capacity (r = −.75, p < .001), no such associations were evident for ε3 homozygotes (p > .1 for all). The scatter plots and regressions illustrating age differences in regional volumes in APOE ε4 carriers and ε3 homozygotes are presented in Figure 2.

Figure 2.

Age differences in regional volumes for APOE ε4 carriers (empty circles, dashed line) and ε3 homozygotes (filled circles, sold line).

Path Analysis - Processing Speed and Working Memory

We proceeded to compare the patterns of associations among the variables in two APOE ε-variant groups via path analyses. Evaluation of the three path models (reduced grouped mediational, reversed-mediational, correlational) showed that the reduced grouped mediational model (Figure 3a) fit the data well, according to all goodness-of-fit indices (Table 3). In contrast, neither the reversed-mediational model nor the exclusively correlational model fit the data well. Therefore, we will focus only on the reduced grouped mediational path model in which there were several significant indirect effects. Among APOE ε3 homozygotes, WM was independently associated with lPFC volume and speed of processing (Table 4). Greater age was associated with higher SBP, smaller pFWM and lPFC volumes, and slower processing; however, SBP was unrelated to volumes or cognitive variables. Carriers of the APOE ε4 allele displayed a different pattern of associations: SBP through pFWM volume, and via speed of processing mediated the effects of age on WM. That is, older ε4 carriers with higher SBP had smaller pFWM volumes, and persons with smaller pFWM had slower perceptual-motor speed, which was in turn associated with poor performance on WM tasks. It is important to note this larger indirect effect subsumes multiple other significant indirect effects in the model (Table 4), each indicating a different mediation effect. Furthermore, age differences in WM were associated with individual differences in speed, which were associated with individual differences in SBP. As shown in Figure 3a, the model showed both a significant direct effect of SBP on processing speed and a significant indirect effect in which individual differences in SBP were associated with individual differences in pFWM, which in turn explained variance in processing speed. Significant indirect effects also demonstrated that only among ε4 carriers, age differences in pFWM and lPFC volume were explained in part by individual differences in SBP.

Table 3.

Goodness-of-Fit Indices

Model	χ2	df	χ2/df	p	CFI	TLI	RMSEA	SRMR
SBP Mediating Age-Speed-WM
PFC/FWM Mediation	14.8	15	1.0	.464	1.00	1.00	.000	.055
PFC/FWM Reversed Mediation	40.6	8	5.1	.000	0.85	0.42	.337	.205
PFC/FWM Correlation	53.6	8	6.7	.000	0.79	0.19	.398	.267
SBP Mediating Age-Recognition
PFC/HC Mediation	13.1	14	0.9	.519	1.00	1.01	.000	.051
PFC/HC Reversed Mediation	39.2	14	2.8	.000	0.75	0.50	.223	.150
PFC/HC Correlation	21.9	14	1.6	.082	0.94	0.88	.125	.134

Notes: CFI = Comparative Fit Index; TLI = Tucker-Lewis Index; RMSEA = Root mean square error of approximation; SRMR = standardized root mean residual.

Table 4.

Significant Indirect Effects From Model 3

Indirect Effect	Std. Estimate	p	95% CI
Model of SBP Mediating Age Differences in Speed of Processing and WM
ε3 homozygotes
Age → lPFC → WM	−.18	.025	−.35 to −.02
Age → Speed → WM	−.18	.045	−.35 to .00
ε4 carriers
Age → SBP → pFWM→ Speed → WM	−.15	.006	−.25 to −.04
Age → SBP → Speed → WM	−.25	.004	−.42 to −.08
Age → SBP → Speed	−.28	.010	−.48 to −.10
SBP → pFWM→ Speed → WM	−.24	.001	−.38 to −.09
SBP → Speed → WM	−.40	.000	−.57 to −.24
pFWM → Speed → WM	.48	.000	.31 to .67
Age → SBP → pFWM	−.30	.008	−.53 to −.08
Age → SBP → pFWM→ Speed	−.17	.003	−.28 to −.02
SBP → pFWM→ Speed	−.27	.000	−.42 to −.12
Age → SBP → lPFC	−.33	.020	−.06 to −.05
Model of SBP Mediating Age Differences in Recognition Memory
ε4 carriers
Age → SBP → Assoc. Recognition	−.21	.021	−.40 to −.03
Age → SBP → lPFC → Assoc. Recog.	−.15	.040	−.03 to −.01
SBP → lPFC → Assoc. Recognition	−.25	.016	−.44 to −.05
Age → SBP → Item Recognition	−.32	.004	−.53 to −.11
Age → SBP → lPFC	−.34	.007	−.59 to −.09

Notes: SBP=Systolic Blood pressure; lPFC = lateral prefronal volume; pFWM = prefrontal white matter volume; WM = WM composite; Speed = processing speed

Path Analysis - Item and Associative Recognition Memory

Inspection of the fit indices in Table 3 shows that the reduced grouped mediational path model (Figure 3b) fit the data well. As with the speed-WM models, neither the reversed-mediational nor the correlational models fit the data well (Table 3). There were several significant indirect effects in the reduced grouped mediational model.

Among APOE ε3 homozygotes, greater age was associated with higher SBP, which was in turn associated with poor associative memory; however, the indirect effect linking the three was not significant (p > .15). Age was directly, negatively associated with recognition of both items and associations, which were significantly correlated. Similarly, greater age was also directly associated with smaller HC and lPFC volumes. However, there were no indirect effects for ε3 homozygotes. In contrast, older carriers of the APOE ε4 allele with higher blood pressure had smaller lPFC volumes and poorer associative recognition (Table 4). In addition, SBP mediated the effects of age on recognition of items and associations, independent lPFC volume. Notably, SBP mediated the effects of age on lPFC volume but not hippocampal volume.

Discussion

In a sample of healthy adults, genotypic and phenotypic vascular risk factors were associated with smaller prefrontal volumes, reduced speed of perceptual processing, poorer WM performance, and weaker verbal recognition memory. It appears that for the carriers of APOE ε4 allele the effects of elevated vascular risk may have more dire consequences: older ε4 carriers with high-normal SBP had smaller prefrontal volumes than those who had lower SBP, as indicated by the negative paths in Figure 3a. Moreover, the overall model fit (Table 3) supported the absence of such a mediational pattern among ε3 homozygotes. The combination of a relatively mild elevation in physiological indicators of vascular risk with a genetic risk factor is associated with poorer state of brain and cognition than would be expected. The observed synergy between genetic and physiological risks in explaining cognitive deficits is in accord with previous reports regarding APOE ε4 (De Frias et al., 2007; Elias et al., 2008; Fuzikawa, Peixoto, Taufer, Moriguchi, & Lima-Costa, 2008; Peila et al., 2001) and the met allele of BDNF val66met polymorphism (Raz et al., 2008). Thus, this finding is in accord with the growing evidence that “normal” should be viewed in the context of individual’s genetic predisposition (Raz, 2011).

Previous reports indicated greater ventriculomegaly and reduced performance on tests of episodic memory and executive functions in ε4 carriers with elevated SBP, in comparison to participants with only one of the risk factors (Zade et al., 2010). In community samples of typical older adults, the APOE ε4 allele is associated with increased burden of white matter abnormalities when blood pressure is elevated (de Leeuw et al., 2004) and with reduced brain volume and ventriculomegaly when overt cardiovascular disease is present (DeCarli et al., 1999). However, this is the first report of synergistic contribution of vascular risk and ε4 to age-related differences in volume of a selected brain region in a sample of healthy, normotensive individuals within a broad age range. In the present sample, SBP mediated the effects of age on prefrontal cortical and white matter volumes only among ε4 carriers.

The results reported here add to the existing literature on the associations among age, brain, and cognition. Individual differences in lPFC volume explained some of the age differences in WM performance in one previous study (Head, Rodrigue, Kennedy, & Raz, 2008), whereas in another sample, we observed no association of age differences in WM with individual differences in lPFC volume (Raz et al., 1998). There may be many reasons for such inter-study discrepancies, including the demographics of the sample, and health screening. One important distinction may be the selection of measurement for WM. Various tests, conceived to measure WM are actually not highly associated and may not form a well-defined construct (Kane, Conway, Miura, & Colflesh, 2007; Roberts & Corkin, 1997). However, all WM tasks used to form the WM composite in the present study, including both n-back and span task indices, loaded onto a single component. Thus, a strength of the present study is enhanced WM construct validity.

The present findings demonstrate markedly different patterns of associations between ε4 carriers and ε3 homozygotes. Despite previous reports of reduced processing speed and impaired cognitive resource allocation by ε4 carriers relative to ε3 homozygotes (see Parasuraman et al., 2002 for a review), we did not observe such differences in mean performance by APOE genotype (speed: t = .56, p > .1; working memory: t = .44, p > .1). In contrast to the findings reported by Liu and colleagues (2010), we found that old age and ε4 did not independently influence cognition. Rather, the present results showed that individual differences in SBP explained age-related variance in speed of perceptual processing and verbal recognition memory, only among ε4 carriers. However, Liu et al. did not model interactions between age, APOE genotype, and vascular risk on cognitive performance. Moreover, our results were derived from a lifespan sample of healthy adults, and whereas Liu and colleagues used a population-based sample of typical older individuals. Similar to its effects on regional brain volume, elevated SBP in midlife is associated with poorer cognitive outcomes (Qiu, Winblad, & Fratiglioni, 2005), but particularly for ε4 carriers (Peila et al., 2001). Thus, the ε4 allele appears to predispose individuals to the deleterious effects of small elevations in physiological vascular risk which become manifest in older age.

The results reported here are in accord with previous findings that older APOE ε4 carriers show increased white matter degradation and slower cognitive processing speed, (Espeseth et al., 2006) particularly in frontal lobes, as compared to non-carriers (Bartzokis et al., 2007). Age-associated differences in processing speed mediate individual differences WM performance, which can in turn explain variance in episodic memory (Salthouse, 1991, 1996a). However, the roles of and interactions between genetic and phenotypic vascular risk factors in modifying these relations have not been previously examined.

The ε4 allele of APOE is associated with impairments in episodic memory (Bondi et al., 1995; Honea et al., 2009). In comparison to non-carriers, older ε4 carriers show greater deficits in recall than recognition (Nilsson et al., 2006), poorer verbal recall (Helkala et al., 1996) and greater longitudinal decline in memory for faces and words (Small, Basun, & Backman, 1998). In addition, ε4 is associated with increased longitudinal decline in episodic memory relative to performance on measures of executive function (Wilson et al., 2002). However, of two meta-analyses evaluating the impact of APOE genotype on cognition, only Wisdom and colleagues (2011) demonstrated negative effects of ε4 on processing speed, executive functioning, and episodic memory. In an earlier meta-analysis of the effects of ε4 on cognition, Small and colleagues (2004) found an association between ε4 and reduced performance on executive function and episodic memory, although the effect sizes were very small. Thus, it is possible that individual differences in vascular risk factors may have explained such effects. However, such data are infrequently collected as the normal course of such investigations and are rarely accounted for statistically.

Limitations and Future Directions

Analytic approaches such as path analysis are based on variance/covariance partitioning. Therefore, such methods can only elucidate of the role of individual differences and cannot be interpreted as estimates of age-related change or causal relations among the variables (Lindenberger, von Oertzen, Ghisletta, & Hertzog, 2011). Rather, the present findings underscore the contributions of other statistical correlates of age-related differences in indices of regional brain volume and cognitive performance. Furthermore, the present findings suggest that regardless of age, normotensive carriers of the APOE ε4 allele with elevated SBP are more likely to also possess smaller prefrontal volumes, and have slower processing speed, lower WM capacity, and reduced verbal recognition than ε4 carriers with lower SBP. Although advanced age is negatively associated with the outcome variables, we cannot be clear if such deficits are the result of the aging process; only longitudinal analysis can shed light on individual differences in age-related decline. In addition, although bias-corrected bootstrap resampling in generation of confidence intervals around indirect effects is acknowledged as the most rigorous method for establishing statistical mediation in SEM-based models (MacKinnon et al., 2002; Preacher & Hayes, 2008), the stability of multi-mediator associations has not been clearly established.

Even beyond the matter of extrapolating from cross-sectional differences to age-related change, many extant studies exhibit an additional flaw, for which they have been recently subjected to a systematic critique criticized, i.e., many fail to consider alternative hypotheses about the age – brain – cognition relations (Salthouse, 2011). Here we ruled out alternate explanations, such as “reversed causation” or simple correlational association and showed that individual differences in prefrontal volume and SBP explain age-related differences in cognition, and not the other way around.

In the present study, we did not evaluate many markers of vascular risk previously shown to interact with APOE genotype in vascular risk and correlate with cognitive performance decrements, such as homocysteine (Elias et al., 2008), triglycerides (De Frias et al., 2007), or LDL cholesterol (Fuzikawa et al., 2008). In addition, according to some studies, APOE ε4 may be associated with increased hippocampal atrophy in older women, but not men (Cohen, Small, Lalonde, Friz, & Sunderland, 2001). According to others, APOE ε4 may be associated with poorer WM and cognitive control among older men than women (Reinvang et al., 2010). However, sample size limitations precluded analysis of sex differences in the present study. Thus, additional work is needed to evaluate the effects of and interactions between APOE genotype, sex, and individual differences in sex hormones, age, regional brain volumes, and cognitive performance. The small number of ε4 carriers in this study precluded modeling of additional variables of interest and of more complex interactions.

Peila et al. (2001) reported that carriers of the APOE ε4 allele are more susceptible to the negative effects of hypertension on cognition than non-carriers, and suggest that antihypertensive medication can assuage such decline. Because genetic factors may alter what may be considered an appropriate threshold for diagnosis and treatment of hypertension, such an approach may be warranted for future studies. It is possible reduced brain volumes associated with high-normal blood pressure in ε4 carriers reflects target-organ damage commonly associated with hypertension. Although according to longitudinal investigations, treatment of hypertension may not eliminate brain shrinkage (Raz et al., 2005; Jennings et al., 2011), to date, no studies have evaluated the effects of interventions aimed to reduce blood pressure and cholesterol levels on brain and cognitive decline in otherwise healthy or younger samples of ε4 carriers. Such intervention should not be limited to pharmacotherapy. Recent evidence indicates that increased physical activity may ameliorate the effects of ε4 on brain activity associated with semantic memory processing (Smith et al., 2011). Thus, for at-risk individuals (ε4 carriers) exercise regimen may bring benefits if started early enough. The present findings support calls for better understanding of the contributions of vascular risk to age-related brain atrophy (Wright & Sacco, 2010), and emphasize the need for future studies to evaluate interactions between age and both genotypic and manifest vascular risk factors on neural and cognitive correlates.

In conclusion, possession of the APOEε4 allele appears to predispose older adults to the negative effects of minor elevations in SBP on prefrontal brain volumes, processing speed, WM, and verbal recognition memory. In comparison to APOEε3 homozygotes, individual differences in SBP among normotensive carriers of the APOEε4 mediate age-related variance in prefrontal white matter volume and lateral prefrontal cortical volumes. Furthermore, the present findings show that for APOEε4 carriers the variance-covariance structure and statistical linkage between age, vascular risk, regional brain volumes, and associated cognitive performance differs from those of the APOE ε3 homozygotes. Multiple risk factors for cognitive decline act in concert and the impact of a physiological measure may depend on individual genetic contexts.

Highlights.

• Studied: effects of genetic and physiological vascular risk on brain and cognition.

• Examined links of age, systolic BP, brain volumes, and cognition with path analysis.

• Found smaller prefrontal volumes in APOE ε4 carriers with elevated systolic BP.

• Observed reduced cognitive performance in ε4 carriers with high-normal BP.

• APOE ε4 and physiological vascular risk act in synergy to reduce performance. 1/5/2012