A Systems Approach to the Aging Brain

0 downloads 0 Views 33MB Size Report
at age 46 and (d) the same person at the age of 51; (e) an older adult age .... fortell impending cognitive decline (Dickerson et a!., 2001; Tapiola ... caudate nuclei in a sample of healthy adults of a wide age range ...... Foundation Sleep in America Survey. Journal .... Neuroscience and Biobehavioral Review, 30: 1004-1031.
A Systems Approach to the Aging Brain: Neuroanatomic Changes, Their Modifiers, and Cognitive Correlates Naftali Raz and Kristen M. Kennedy

Bear with my weakness; myoid

brain is troubled.

William Shakespeare, The Tempest, Act IV, Scene I

Abstract uccess in diagnosing and treating age-related brain disease depends on understanding normative and optimal aging, and noninvasive neuroimaging is a vital tool in advancing toward that goal. Studies of the brains of healthy adults reveal uneven, differential shrinkage of the parenchyma and expansion of the ventricular system, with the prefrontal cortices evidencing the largest magnitude of age-related differences, and the tertiary association (prefrontal and parietal) cortices, the neostriatum, and the cerebellum showing the greatest rate of shrinkage. Although findings vary across samples, reduced regional brain volumes and steeper longitudinal declines are usually associated with lower cognitive performance in specific domains. The observed pattern of differential brain aging is significantly modified by negative and positive factors. Although negative effects of vascular risk factors are apparent in the regions deemed most vulnerable to aging, the positive modifying influence of aerobic fitness is clearest in the same age-sensitive areas. Genetic variation may have a significant impact on age-related changes in brain and cognition, although the sparsity of evidence precludes evaluating the true magnitude of the effect of specific polymorphisms. In addition to (and in conjunction with) aerobic fitness, antihypertensive treatment and hormone replacement therapy (in women) may alleviate negative effects of aging on brain structure.

Introduction All living systems change with time and aging is as much an integral part of their developmental trajectory as is maturation. However, the trajectories that connect the initial and the final moments of life vary across species and among individuals. The reasons for such

variability are not clearly understood, and it is not easy to gauge how much of the individual differences in aging can be ascribed to its normal physiological course and how much is attributable to beneficial and pathological effects that constitute one's life history and are included with one's genetic endowment. In this chapter, we will attempt to summarize the current state of knowledge regarding adult development and aging of the central nervous system, and more specifically the brain. Although informed by investigations of animal models and postmortem human material, our narrative is focused on the in vivo studies of human brain aging, its cognitive correlates, and the pathological as well as therapeutic factors that alter the course of brain aging in a predictable way. Our main premise is that if the objective is alleviating the negative and accentuating the positive extremes of the normal aging process, it is imperative to understand the physiological and pathological underpinnings of that variability and its relation to cognition and behavior. In the past hundred years, aging of the human brain attracted the attention of neuroanatomists and neuropathologists, and postmortem (PM) studies revealed many gross and microanatomical characteristics of the aging brain (for a detailed account of postmortem neuropathology of aging see Dickstein, Morrison and Hof Chapter 3 in this volume). Gross anatomic investigations noted age-related reduction in brain weight and volume, ventricular and sulcal enlargement, reduction in the bulk and quality of the white matter (e.g., see Kaes [1907) for meticulous neuroanatomical work and Kemper [1994) for a comprehensive review of the topic). A series of elegant studies in primates revealed significant age-related alterations in the structure of the myelin sheath (Marner et al., 2003; Peters and Sethares, 2002). Loss and shrinkage of neurons, albeit not pervasive and global, is also observed (Haug, 1985). Neuroanatomical and neurophysiological studies demonstrate age-related differences in the basic cellular structure and function of the organism (Brunk and Terman, 2002; Lenaz et al., 2002). 43

/ 44

Imaging the Aging Brain

Thus, the postmortem investigations have demonstrated that aging affects the brain at virtually every level, and histological studies of postmortem material remain a valuable exploratory tool of modern neurogerontology. Nonetheless, for all the advantages afforded by the precise microanalyses, PM studies have several significant disadvantages. Although the PM tissue collected from sources with good quality control show excellent protein stability and RNA quality (Stan et al., 2006), the degree to which ante-mortem hypoxia, hyperpyrexia, and ischemia affect the results of most studies is unknown. Moreover, PM studies are inherently incapable of addressing several fundamental questions, such as how the brain is changing over time, what the neuroanatomical picture of optimal (as opposed to common) aging would be, and what cognitive consequences the observed changes entail.

Structural Brain Aging: Volumetric Findings Cross-Sectional Studies of Age-Related Differences in Brain Volumes Although the cumulative results of more than two decades of cross-sectional studies of age differences in regional brain volumes may be variable and at times contradictory, they converge onto several general trends. According to the neuroimaging volumetric literature (for detailed reviews see Raz, 2000, 2004; Raz and Rodrigue, 2006), the prefrontal cortices emerge as the most vulnerable region of the aging brain, whereas sensory (e.g., primary visual) cortices and lower regions of the central nervous system (CNS) such as the ventral pons show little, if any, aging-related variability. In some regions of the brain, such as the hippocampus, temporal and parietal lobes, the volume declines seem to accelerate with age as nonlinear trends observed in several samples suggest (Cohen et al., 2006a; DeCarli et al., 2005; Jernigan et al., 2001; Lupien et al., 2007; Raz et al., 2004b). In the entorhinal cortex, age-related differences in volume are observed only in the later part of the lifespan (see Raz [2004] for a review and Miyahira et ai, [2004] as well as Siwak- Tapp et al. [2008] for findings in canines). Studies that assessed cortical thickness and sulcal depth rather than regional volumes report a similar pattern of age-related differences (Kochunov et al., 2005; Nesvag et al. 2008; Salat et al., 2004), although in a notable exception, at least one study reported a significant negative correlation between age and thickness of the pericalcarine cortex as well (Salat et al., 2004). Assessing volume differences in the brain is not an easy task. It requires many hours of tedious work by highly trained personnel. To alleviate the difficulties of manual measurement, semi-automated methods that use whole-brain data to estimate local differences in gray matter "density" or, with an appropriate correction, "volume," have been introduced in the past decade (e.g., voxel-based morphometry, VBM; Ashburner and Friston, 2000). VBM and related methods have several advantages over manual morphometry: perfect repeatability, no requirement of complex decision making on the part of highly trained operators, and examination of the whole brain, not just selected regions of interest (ROIs). However, the disadvantages become apparent when the issues of validity are considered. In VBM, the brain tissue is automatically segmented into gray matter, white matter, and cerebral spinal fluid (CSF), and errors of segmentation are possible when the image is less than perfect and contains artifacts. What a knowledgeable human operator would dismiss as an

artifactual variation in intensity and a mere nuisance is treated by a segmentation algorithm as a legitimate difference in tissue density. To compensate for significant individual differences in brain geometry, the images are normalized, i.e., fitted into a standard or study-specific template. In the process, the images are filtered, and as a result, while gaining signal-to-noise ratio, they lose resolution, sometimes by a factor of 103. Reduction of resolution and improvement in the signal-to-noise ratio may result in increased sensitivity to global group differences at the expense of variability in small structures (Bcokstein, 2001; Crum et al., 2003; Davatzikos, 2004). Refining the methods of data acquisition and tailoring sequences for visualizing and assessing specific structures may improve the validity of automated methods (van der Kouwe et al., 2008). Although VBM and similar methods produced results that are broadly consistent with those of manual morphometry (e.g., Good et al., 2001; see also Raz and Rodrigue [2006] for a review), several comparisons between semi-automated and manual methods indicate those methods are not interchangeable (e.g., Allen et al., 2005; Cardenas et al., 2003; Dorion et al., 2001; Douaud et al., 2006; Good et al., 2002; Gong et al., 2005; Kennedy et al., 2008a; Tisserand et al., 2002). One striking feature of those comparisons is that VBM analyses are more likely than the manual measures to find larger age differences in the regions bordering CSF such as the insula, the cingulate gyri, and the superior temporal and superior parietal cortex, i.e., regions that show minor though significant age-related differences in volumetric studies (Kennedy et al., 2008a; Raz, 2004). A comparative study of VBM and manual volumetry has suggested that statistically defined peak values of modulated density produced by VBM may indicate stronger age differences than manual measures do, and they may identify age effects in a different set of regions (Kennedy et al., 2008a). Whether the larger age effects reflect true differences in the aging brain structure, or stem from exaggeration of local partial volume artifacts at the border of CSF and parenchyma or gray and white matter is unclear. It is important to note that when VBM-derived measures are aggregated over anatomically defined regions (i.e., ROI masks) the pattern of age differences was more similar to that produced by the manual measures (Kennedy et al., 2008a). Semi-automated VBM methods may be more likely to report nonlinearities in age-volume relationships (Kennedy et al., 2008a; Walhovd et al., 2005a). There are no neurobiological reasons why a specific cluster of voxels is agedependent whereas its neighbor, within the cytoarchitectonically identical area, is not. Moreover, statistically, the fact that "red pixels" on color-coded maps of the gray matter differences in local densities significantly differ from zero does not mean that they also significantly differ from "orange" or even "yellow" pixels on the same map of age difference, a problem highlighted previously in regards to fMRI activation maps (Jernigan et al., 2003). In the context of the outlined concerns, it is important to note that when voxel-based analysis provides information that is biologically interpretable, it may open new windows to understanding of the neuroanatomy of aging. For example, a recent study of older adults revealed amplified age-related differences in the hippocampal regions that have long been suspected of excessive vulnerability to ischemia (i.e., CAI field) in conjunction with age-related differences in hippocampal volume revealed by manual volumetry (Xu et al., 2008). In sum, the magnitude and location of age-related differences in the brain volume may depend on the method of measurement. , Whereas VBM may be a useful first-pass approach to the analysis of a large set of brains, findings from automated procedures should not be treated without question as replications of manual studies.

4. A Systems Approach to the Aging Brain

45

Automated first-pass analyses should be followed by manual measures in the regions identified as the candidates for age-related differences, and in the regions that can be missed due to the method limitations, especially in the locations with intricate geometry and high risk of artifacts (e.g., entorhinal cortex). In addition, voxel-based methods may add to current knowledge by providing information about regional differences within structures that are difficult to demarcate on the basis of external landmarks. Although development of new methods more sensitive to local geometry of the brain parenchyma appears to add neuroanatomic validity to semi-automated morphology (for more details see Van Horn and Toga, Chapter 21 in this volume) a well-trained human operator guided by top-down knowledge of neuroanatomy will pro ably remain an instrument of choice in cases of analysis of less-than-perfect and artifact-ridden images Longitudinal Studies of Brain Volume Change in Adulthood Thus far, the discussion of age-related differences in brain volumes has been limited to cross-sectional studies. Such studies indeed represent the most popular way of estimating age effects. However, the cross-sectional design has major limitations. Although ideally, cross-sectional investigations should yield estimates of age-related change, this is rarely the case. Cross-sectional studies provide a snapshot of individual variability and in doing so they confound individual differences in brain and cognition, which are brought into the study from years of previous development, with age-related variance. Thus, findings from cross-sectional investigations should be interpreted with caution, taking into account the problem of variance commonality among multiple variables, which is impossible to sort out without introducing the time dimension (Kraemer et al., 2000; Lindenberger and Potter, 1998). Granted, the longitudinal approach has its own share of problems such as "3M" -the mobility, morbidity, and mortality of the participants (Raz, 2004). Nonetheless, the results of longitudinal studies, unlike those of cross-sectional investigations, can be interpreted as an indication of true change or the lack thereof, within the limits of generalizability predicated on the nature of the sample. Indeed, longitudinal studies of brain volume show that for some regions, cross-sectional estimates are quite accurate, whereas for others they are wide off the mark (e.g., Raz et al., 2005). One of the most frequently used indices of brain health is the volume of the cerebral ventricles, and in many longitudinal studies this is the only neuroanatomical measure. The volume of the cerebral ventricles reflects many diverse alterations in the volume and pressure of the CSF, and represents a summary measure of change in the whole CNS. The extant studies of ventricular enlargement are consistent in showing significant age-related declines, which proceed at an average pace of almost 3% per year. Moreover, the annual rate of ventricular expansion increases with age, exceeding 4% per annum in the 6th and 7th decades of life (Carmichael et al., 2007; Raz, 2004). Even in healthy adults, ventricular expansion over a 5-year period may be visible with the naked eye, as illustrated in Figure 4.l. In contrast to ventricular size, the total brain size changes at a significantly slower pace (Raz, 2004). This discrepancy may be rooted in the fact that the boundaries of the fluid-filled system of connected CSF cavities (ventricles, cisterns, etc.) are not fixed because they are defined by the bulk of surrounding tissue. Thus, ventricular expansion occurs in response to loss of brain tissue anywhere in the CNS, and barring significant pathological changes in CSF production and drainage, loss of tissue in multiple locations

Figure 4-1 Ventriculomegaly is a hallmark of the aging brain reflecting nonspecific, total expansion of CSF and contraction of brain parenchyma. The images were acquired five years apart. (a) Young adult age 24 and (b) the same person, age 29 years; (c) a middle-aged adult at age 46 and (d) the same person at the age of 51; (e) an older adult age 75 and (f) the same person at the age of 80. Notice that even in healthy middle- and older-aged adults, ventricular expansion is evident across 5 years.

adds up to fluid expansion to fill the newly created space. Hence, specificity of ventricular size as an index of brain health is low but its sensitivity is high. In contrast, differential and localized changes in the brain parenchyma coexist with localized preservation (or even enlargement), and the total brain volume may represent an averaged net change. Longitudinal studies that examined distinct brain regions reveal significant local differences in shrinkage rates. Association cortices, such as prefrontal and inferior parietal, tend to show the greatest rate of change, with temporal cortex following closely behind, and occipital (especially primary visual) cortex showing no or minimally significant changes (Pfefferbaum et al., 1998; Raz et al., 2005; Resnick et al., 2003). Note that in contrast to cross-sectional studies, longitudinal investigations tend to show

46

Imaging the Aging Brain

significant decline in parietal volume, a region that is marked with significant individual variability (Raz et al., 2005). Because the decline of the medial temporal structures (the hippocampus and the entorhinal cortex) is believed to be a harbinger of Alzheimer's disease (AD, see chapters 10, 12 for a detailed discussion), they have been measured in several longitudinal samples. These investigations reveal significant shrinkage of the hippocampus (HC), usually, though not always, with a nonlinear, age-accelerated trend (Jack et a!., 2000; Liu et aI., 2003; Raz et a!., 2004a, 2005; Rusinek et a!., 2003). The implication of that trend is that restricting samples to persons of advanced age would increase the likelihood of finding hippocampal shrinkage that may be missed in investigations of younger adults. This expectation is illustrated by the current literature on the volume of the entorhinal cortex. To date, most of the longitudinal studies of the entorhinal cortex have been conducted on samples of older adults, and their results indicate significant shrinkage comparable to that of the hippocampus (Cardenas et a!., 2003; Du et a!., 2003; Jack et al., 2007). By comparison, in a sample covering a wide age range, entorhinal shrinkage is significantly smaller than decline in hippocampal volume and becomes evident only in some individuals at the sixth and seventh decade oflife (Raz et al., 2004a, 2005). When it occurs, entorhinal shrinkage is more likely to happen in persons with lower cognitive aptitude (Raz et al., 2008) and may fortell impending cognitive decline (Dickerson et a!., 2001; Tapiola et al., 2008). A few other key structures deemed age-sensitive in crosssectional studies were investigated in very few longitudinal studies. Longitudinal investigations found significant shrinkage of the caudate nuclei in a sample of healthy adults of a wide age range (Raz et al., 2003a, 2005), while slower shrinkage was observed in the putamen and globus pallid us (Raz et al., 2003a). Substantial longitudinal shrinkage of the cerebellum that exceeded the rate estimated from cross-sectional data has been observed in samples with adequate age range (Raz et a!., 2003b, 2005), whereas shrinkage of the pons is minimal as predicted by cross-sectional studies (Raz et a!., 2003b). Apparently, when individual differences are controlled, the caudate nucleus and the cerebellum emerge as the leading candidates for the dubious honor of the most vulnerable regions of the aging brain. The degree to which the observed age-related changes and differences reflect the normal aging picture is unknown. Some studies in nonhuman primates reported that by contrast to humans there is little or no neuronal loss (e.g., Keuker, Luiten, and Fuchs, 2003), although such a conclusion frequently reflects the lack of statistical power. For example, "conservation of neuron number" in rhesus monkey entorhinal cortex corresponds to an age effect of d = 0.6 standard deviation in the most vulnerable layer II, with a smaller effect of d = 0.28 in layer III (Merrill, Roberts, and Tuszynski, 2000). In some samples and for some types of cells, the age effects are so dramatic that no statistical analyses are necessary to reveal them (Smith et aI., 1999). There are significant individual and interstrain differences in brain structure even among animals that are bred and reared under standard and controlled conditions (e.g., Chen et al., 2005). Although those differences may be not as large as among humans, it would take more than eight animals to show effects that are not visible to the naked eye. In vivo neuroimaging studies of nonhuman primates will shed more light on the notion of greater preservation and lesser declines in the animals protected from virtually all pathogens associated with even the most optimal human existence. Such studies are associated with significant logistic effort and expenses.

Nonetheless, in a recent investigation of 19 rhesus monkeys, the pattern of aging derived from a VBM analysis was quite similar to the one observed in human studies (Alexander et al., 2008). Dorsolateral and orbital frontal regions evidenced the largest negative effect of age; smaller or none were observed for occipital, parietal cortices as well as globus pallidus. Contrary to human studies, including a small-sample study from the same group (Alexander et al., 2006), the cerebellum was also relatively intact (Alexander et aI., 2008). The dearth of MRI studies on normal aging animals and the virtual absence of combined MRI-histology studies precludes inferences about the validity of MRI evidence for mammalian aging in general. As a rule, as almost all primate studies do, the combined imaging-histological investigations have such low statistical power that when they essentially replicate effects reported in human studies (e.g., a correlation between age and HC volume r=-O.32, Shamy et a!., [2006]), they cannot demonstrate their statistical significance. One comparative cross-sectional study stands out, as it combines stereology, in vivo neuromorphometry (MRI) and metabolic (magnetic resonance spectroscopy, MRS) assessment with multiple behavioral tests in a sample of rats within a typical adult life span (Driscoll et a!., 2006). In that study, lower hippocampal cell density and smaller hippocampal volumes were observed in old animals compared to the young and middle-aged. Notably, presence of immature neurons (adult neurogenesis) predicted both larger HC volumes and better performance on the HC-sensitive tasks. The link between learning, environmental enrichment, and neurogenesis in rodents has been suggested before (Drapeau et al., 2003, 2007; Kempermann, Gast, and Gage, 2002, but see Merrill et al., 2003). Studies in rodents indicate that hippocampal neurogenesis is significantly reduced with age (Kuhn, Dickinson-Anson, and Gage, 1996), but can be brought to the young-age levels by environmental manipulations (Cameron and McKay, 1999; Kempermann, Kuhn, and Gage, 1998). Thus, neurogenesis and other plasticity-related phenomena may emerge as an important process shaping mammalian brain aging (Burke and Barnes, 2006). Currently it is impossible to assess neurogenesis in vivo and to directly measure neural plasticity in humans, although one study on terminally ill persons supports the claims of adult neurogenesis in the dentate gyrus (Eriksson et al., 1998). Moreover, the rates of neurogenesis vary dramatically across and within rodent species (Kempermann, Kuhn, and Gage, 1997), and the researchers are still struggling to establish reliable methods of assessing neurogenesis across various brain areas and various species (Gould, 2007). Thus, the implications of brain plasticity observed in rodents for human aging are still unclear. If cross-sectional comparative studies combining imaging and histology are rare, a longitudinal one is truly unique. One such study of a typical rodent revealed neither age-related regional shrinkage, nor ventricular enlargement over the entire lifespan (Sullivan et al., 2006a). The study showed a steady increase of the corpus callosum area, as well as volumes of the hippocampus and the cerebellum in two samples of rats. Although cerebellar volume leveled off toward the end of the life span, it did not decline. There are several caveats in interpreting these findings. First, the sample was very small and, although no strong trends were observed, it is unclear whether the observed age trajectories reflect the state of affairs in the population. Second, a rodent model of aging, with all its usefulness, still may not account for some specific features of the primate brain. Third, age-related changes could still happen in the regions not sampled in that study.

4. A Systems Approach to the Aging Brain

In sum, cross-sectional and longitudinal studies of neuroanatomical aging indicate a substantial age-accelerated expansion of CSF-filled cavities, mild shrinkage of the cerebral parenchyma, and a pattern of differential regional changes. Association cortices (e.g., prefrontal and inferior parietal), the neostriatum, the hippocampus and the cerebellum appear more sensitive to aging than do the primary sensory cortices, entorhinal cortex, the paleostriatum, and the pons. Little is known what neurobiological changes are reflected in the apparent shrinkage observed on the MR images. Given the ethical impossibilities, comparative studies, especially in primates (for an account of cross-species studies focused on AD see Chapter 8), are. of critical importance in that area of research. However, such studies must adhere to the design standards applicable to the human studies, especially those ensuring adequate statistical power. Age-related differences and changes in the white matter. Unlike most of the regional cortical volumes, the gross volume of the white matter shows a nonlinear relationship with calendar age (Bartzokis, 2004; Walhovd et al., 200Sa). It increases from childhood to young adulthood, remains stable throughout middle age, and exhibits linear decline in the late years. Thus, the results of a particular study depend on the age range of the subjects. Observations on the developing brain reveal steady increases in white matter volume (see Lenroot and Giedd, 2006 for a review); studies with a significant proportion of older participants are likely to find decline in white matter volume (e.g., Ikram et al., 2008), whereas those that examine the whole adult age range find little if any age differences in that cerebral compartment (e.g., Raz et al., 1997, 2004c). Notably, complexity of the white matter structure, as indexed by fractal dimensionality, is lower in older adults even in the absence

47

of gross volume differences (Zhang et al., 2007), a finding that is in accord with age-related reduction of structural complexity observed in cytoarchitectonic studies (Dickstein et al., 2007). It is unknown how microstructural complexity decline is related to loss of tissue volume, and the relation between the two, especially with regards to temporal precedence, needs to be examined. However, the magnitude of cross-sectional age differences and longitudinal shrinkage in a specific brain region is associated with the ontogenetic status of the region with respect to myelination as described by Flechsig (1901). This relationship, first noted in Raz (2000), is depicted in Figure 4.2. In fact, myelination precedence order accounts for 36% of the variance in magnitude of age-differences in volume and 38% of volume shrinkage. One of the most commonly used indicators of white matter health is the burden of white matter hyperintensities (WMH) observed on T2-weighted MRI scans, illustrated in Figure 4.3 (for a more detailed account see Chapter 17 in this volume). What appear as WMH on the MRI are localized and circumscribed areas of extreme reduction in white matter density. Most WMH are of pathological origin and are believed to reflect hypoperfusion (Fernando et aI., 2006; Holland et al., 2008), microbleeds, and infarcts (De Leeuw, De Groot, and Breteler, 2001). A larger, irregularly shaped WMH may represent "silent" lacunar infarcts that are found in up to 28% of asymptomatic and ostensibly healthy older adults (Vermeer, Longstreth, and Koudstaal, 2007), and are associated with other signs of brain aging such as reduction of cortical metabolism, especially in the prefrontal regions (Reed et al., 2004; Tullberg et aI., 2004). Age differences in WMH mirror the pattern of age-related differences in the white matter volume. As expected for healthy

Flechsig's (1901) Myelination Precedence Rank and Age-Related Regional Shrinkage of Gray Matter .lh8ri.tutJerf.~D/frM_am.~.f,~"';'

en;".

.\.,."",r&,lWtl