This study demonstrates that 3D maps of the caudate nucleus localize statistically significant regions of volume reduction associated with (1) diagnosis and clinical and cognitive scores, (2) future decline in MMSE scores at 1-year follow-up in the MCI group, (3) conversion from MCI to AD 1 and 2 years after scanning, (4) baseline CSF concentrations of tau, and (5) additional clinical measures including BMI, Geriatric Depression, and abnormal gait. The variables most strongly associated with caudate atrophy were found using a reduced N analysis, namely (in this rank order): BMI in the AD group only, BMI in all subjects, CDR-SB scores, group difference between AD and controls, and MMSE scores at baseline. It is interesting to note that cardiovascular measures and CDR-SB scores, which rely on interviews from both patients and informants, were more powerfully associated with caudate volume reductions in our 3D maps, compared to seemingly more direct measures of AD such as the MMSE. Although some of these measures are highly correlated and may be viewed as redundant, we felt that it was important to take a broad approach initially, in analyzing multiple measures of cognition and health. This has practical significance for the planning of clinical trials, which currently use a variety of similar tests, and also allows direct comparison with other studies that each may use different tests.
Caudate atrophy was generally localized to anterior regions and was specific to the right caudate for several variables assessed. Caudate atrophy in AD (versus controls) was 2-fold greater on the right than the left. Further study is needed to confirm the greater vulnerability of the right caudate. The natural asymmetry of the caudate (right 3.9% larger than the left in controls) has been found in many but not all studies, and large samples are needed to detect it. Most literature supports a rightward (right larger than left) asymmetry in caudate nucleus volume (Giedd et al., 1996
; Ifthikharuddin et al., 2000
; Larisch et al., 1998
; Peterson et al., 1993
; Yamashita et al., 2009
), but some studies have found leftward, sex-specific, or no asymmetry (Gunning-Dixon et al., 1998
; Szabo et al., 2003
). Based on our current findings and previous literature, it may be possible to view asymmetry findings in brain structures as a general indicator of poor brain health. Psychosis has been linked to genes involved in the development of cerebral asymmetry (Laval et al., 2004
) and differences in laterality have been found to covary with, or predict, individual differences in susceptibility to stress pathology and drug sensitivity (Carlson et al., 1989
). The presence or change in direction of asymmetry has been reported (but not always consistently) in disorders ranging from schizophrenia (Crow et al., 1996
), ADHD (Castellanos et al., 1996
), bipolar (Javadapour et al., 2010
) and affective mood disorders (Baumann et al., 1999
), and Tourette's syndrome (Klieger et al., 1997
There may be some skepticism that the caudate is a natural place to look for clinical correlations in AD, but it may give additional power to detect associations with cognition and future decline, if used along with structures that are more commonly studied. For example, it is widely accepted that the hippocampus - the target of the most published AD morphometry studies - is challenging to segment reliably, with a notorious 2-fold difference in mean hippocampal volumes across studies and imaging centers (Frisoni et al., 2010). Prior studies using the same radial distance mapping technique to other brain structures in ADNI have also found correlations between hippocampal atrophy (Morra et al., 2009
), ventricular expansion (Chou et al., 2009
) and clinical variables such as MMSE, CDR, and CDR-SB scores. There are also differences between diagnostic groups for the volumes of all of these structures (AD versus controls, MCI versus controls). As shown in , and as might be expected, hippocampal volumes gave greatest effect sizes for detecting group differences, within ranges supported by previous studies (Shen at al., 2010
), and for detecting correlations with clinical scores. Effect sizes for lateral ventricle volumes also gave relatively strong effect sizes, although they were generally weaker than the hippocampal effect sizes. Caudate effect sizes, while significant, were generally lower than those for the ventricles and hippocampus. As one exception to this rank order, the caudate gave numerically marginally better effect sizes than ventricular volume for distinguishing MCI from controls. Even so, we did not performed statistical tests for differences in effect sizes across structures, as so many measures were made that a stringent multiple comparisons correction would be needed. Despite these generally weaker effects, the current study of the caudate also found significant associations between reduced volume and MCI-to-AD conversion, and with decline in MMSE scores after a one-year follow-up in the MCI group, whereas these associations were not detectable in studies of ventricular expansion (Chou et al., 2009
) or hippocampal atrophy (Morra et al., 2009a
), using the same ADNI subject pool (with 240 and 490 subjects, respectively); the latter study also used the same Adaboost machine learning algorithm, applied to hippocampal segmentation. This is an intriguing finding, suggesting caudate volume at baseline may predict later clinical changes relatively well, perhaps because caudate atrophy is only severe when the disease is rapidly progressing.
The caudate has previously been implicated in AD through multiple lines of evidence including histological findings of tau and amyloid accumulations (Braak and Braak, 1990
). Volumetric studies find reductions in caudate nucleus volume with normal aging (Jernigan et al., 2001
; Krishnan et al., 1990
; Raz et al., 2003
) in AD versus controls (Rombouts et al., 2000
), and in many other or neurological, psychiatric or neurodevelopmental conditions including autism (Turner et al., 2006
), Fragile X syndrome (Gothelf et al., 2008
), elderly depression (Butters et al., 2008
), and HIV (Becker et al., 2006
). Caudate lesions can produce deficits in executive control and cognitive processing speed (Rubin, 1999
). Interestingly, a recent study also found subcortical brain volume reductions, encompassing the caudate nucleus, in elderly individuals with a higher body mass index (BMI, a measure of obesity; Raji et al., 2009
The anterior caudate head is the most closely involved in traditionally studied cognitive processes and integrates major inputs from the dorsolateral and orbitofrontal cortices, involved in attention and planning (Cummings et al., 1995
). Disturbances in the hemispheric asymmetry of the caudate head are associated with ADHD and schizophrenia (Ballmaier et al., 2008
; Blanton et al., 1999
; Hynd et al., 1993
). One study found significant caudate volume reductions in depressed elderly and created 3D maps using the same technique here, except on manually-derived segmentations; however, their maps were not significant after permutation-based correction for multiple comparisons (Butters et al., 2008
Beyond the caudate and the medial temporal lobes, additional basal ganglia and subcortical gray matter structures are implicated in AD and age-related neurodegeneration. Putamen and thalamus volumes are reduced in AD, and volume decreases correlate with cognitive deficits (de Jong et al., 2008
). Putamen volumes are reduced in frontotemporal lobe degeneration, but not in AD (Looi et al., 2009
). A recent study of individuals with Down's Syndrome - a group at heightened risk for developing AD - found significant associations in those who developed AD and bilateral volume reductions in hippocampus and caudate and in the right amygdala and putamen (Beacher et al., 2009
). In vivo
imaging with [11C]-PIB PET, which is thought to map the profile of amyloid deposition in the living brain, has correlated increased amyloid signal with increased atrophy in the hippocampus and amygdala (Frisoni et al., 2009
). In addition, post-mortem studies have localized diffuse amyloid plaques in the caudate nucleus in AD (Ikonomovic et al., 2008
In a PET study of dopamine D1 and D2 receptors, researchers found a 14% reduction in mean signal for D1 receptors for the caudate and putamen in AD, but no difference in D2 receptor signal (Kemppainen et al., 2000
). The basal ganglia also have the highest iron content of any brain region, which is interesting because AD patients exhibit disrupted iron metabolism that may relate to oxidative damage and higher iron levels are found in the caudate and putamen in AD patients compared to controls (Bartzokis et al., 2000
A common limiting factor in neuroimaging studies of AD is the labor-intensive task of manually segmenting subcortical structures on large MRI datasets, requiring long-term effort from expertly trained neuroanatomists. Fortunately, automated methods now exist to efficiently segment many subcortical structures including the hippocampus and caudate nucleus (Fischl et al., 2002
; Powell et al., 2008
; Yushkevich et al., 2006
). Here we used an automated method based on adaptive boosting, that we recently developed and validated (Morra et al., 2008
). Using this method, we efficiently created caudate nucleus segmentations in 400 ADNI brain MRI scans. The resulting 3D maps had enough statistical power to detect local volume differences associated with a wide variety of clinical variables in AD. Our maps detected associations between caudate atrophy and CSF levels of tau, which have not always been associated with other MRI-derived measures in the same cohort (Chou et al., 2009
). CSF levels of tau are significantly associated with higher rates of temporal lobe atrophy (Leow et at., 2009
), but only showed a trend level of association with the rate of hippocampal atrophy (Schuff et al., 2009
). In prior amyloid PET studies, amyloid deposition has been noted in the caudate nucleus (Ikonomovic et al., 2008
). Here we did not detect a relation between the level of CSF amyloid and the volume of the caudate. In future, a direct PET-based brain measure of amyloid in the caudate may correlate better with atrophy than CSF-based measures of amyloid burden. Limited sample sizes may have also made it hard to detect an association; even so, the sample was large enough for Tau levels to be reliably correlated with in maps of atrophy for the right caudate nucleus in the pooled sample and bilaterally for the volumetric summaries in the AD group.
A possible confounding factor in some voxel-based morphometric studies of the caudate is the potential for mis-registration of anatomy across subjects along the ventricles, especially in elderly subjects with substantial brain degeneration. Since the caudate nucleus is a gray matter structure that lies just below the boundary of the lateral ventricles, some volumetric studies may be confounded by poor or biased registration along the lateral ventricular surface. This effect is unlikely in the current study as our measures of radial atrophy are intrinsic, meaning they are independent of whether the structure is translated in space. They are computed from a centerline traced down the center of the structure, and do not rely on the correct registration of images across subjects, as some voxel-based averaging methods do.
The etiology of caudate atrophy in AD and MCI is of interest. One probable explanation is that we are assessing neuronal atrophy, secondary to the accumulation of amyloid plaque pathology and tau neurofibrillary tangles (Braak and Braak, 1990
). The caudate may also experience a loss of afferent projections from other brain regions. As such, a limitation of the study is that the associations between caudate atrophy and cognition may in reality be mediated by atrophy occurring in other brain structures, such as the hippocampus. The intent in this study was to identify correlates of caudate atrophy, but it is likely that the more immediate causes of the clinical differences may be atrophy (or functional compromise) of structures elsewhere in the brain, not necessarily those assessed here.
A further limitation of the current study is that the available sample sizes made it necessary to use a pooled sample including both patients and controls for several analyses. Pooling subjects allowed us to achieve sufficient statistical power to detect important findings, but there is also a risk of recovering group sampling effects due to the differences between diagnostic groups in cognitive measures. Where sample sizes and effect sizes were sufficient for within-group correlations to be detected, we did find statistically significant associations in one-year change in MMSE score in the MCI group and in BMI in the AD group. It is possible that there are additional within-group associations for these and other measures that did not survive the sample size reduction. It can also be argued that AD, MCI, and healthy elderly controls represent stages on a continuum - in that case, binning the subjects into diagnostic groups could limit the power of correlations found across the spectrum and lead to false negative results.
In this study, we report correlations between atrophy and cognitive or CSF-derived measures in a pooled ADNI sample (combining patients with AD, MCI and controls), while also reporting several within-groups correlations split by diagnosis (“disaggregated” analyses of AD subjects only, for example). Both types of analysis are complementary and each has its own limitations. When analyzing a mixed cohort of subjects, our goal is to determine (1) the cognitive correlates of atrophy in the entire sample, and (2) whether the chosen biomarker of disease burden is linked with decline across the full spectrum of controls, MCI, and AD subjects. As the whole cohort is arguably a continuum, it is vital to look beyond diagnostic categories to analyze relationships between atrophy and particular measures of cognitive function. A within-group analysis, of AD subjects only, for example, may not be able to detect these correlations due to a “restricted range” effect seen when looking at a limited range of differences in brain structure and cognitive performance. By running split analyses only, many important correlations may be missed. For instance, brain atrophy correlates well with CSF-derived measures of tau pathology across the continuum from healthy aging to MCI and to AD. However, it is possible that no correlation will be detected if we sub-select a group such as MCI, with a very narrow range of disease burden. Furthermore, it is a fallacy to pre-select groups that are defined partly on cognitive measures, and then later test for correlations with these same cognitive scales. If the selection criterion for the group correlates with the variable of interest, we will find many false negatives due to the truncated range, making the results uninterpretable.
Pooled analyses have limitations to consider, as well. When cognitive measures are correlated with diagnosis, correlations in a pooled cohort may show similar patterns to those found when directly comparing AD subjects with controls. Another consideration when using pooled cohorts such as ADNI, is that correlations may be influenced by the proportion of subjects in each diagnostic group. By design, ADNI oversampled subjects with cognitive decline, including a 1:2:1 ratio of AD:MCI:controls. Correlations in this particular sample may not be detected to the same degree in other population studies with different proportions of subjects or within each diagnostic group. For these reasons, we chose to use a combination of both pooled and split analyses, which have complementary value in understanding the cognitive and pathological correlates of atrophy.
Since it can be difficult to tease apart contributions of each individual brain region to a disease such as AD, studying effects of the disease on specific anatomical structures, beyond those most commonly targeted, may help to piece together relevant circuitry affected by AD pathology. Multiple AD biomarkers can be used in a mutually reinforcing way to identify subjects most likely to decline to AD in a sample (Kohannim et al., unpublished results), so caudate atrophy may inform the classification of AD, MCI or the prediction of future decline. Another line of work involves genome-wide association studies of caudate volume in large subject cohorts such as ADNI. This information will improve our understanding of the mechanistic processes involved in brain aging and AD.