CSF Biomarkers in Prediction of Cerebral and Clinical Change in Mild Cognitive Impairment and Alzheimer's Disease

Sample.

ADNI eligibility criteria are described at http://www.adni-info.org. Briefly, participants are 55–90 years of age, had an informant able to provide an independent evaluation of functioning, and spoke either English or Spanish. All subjects were willing and able to undergo test procedures including neuroimaging and agreed to longitudinal follow up. Specific psychoactive medications are excluded. General inclusion/exclusion criteria are as follows: (1) normal subjects: Mini-Mental State Examination (MMSE) (Folstein et al., 1975) scores between 24 and 30 (inclusive), clinical dementia rating (Morris, 1993) of 0, nondepressed, non-MCI, and nondemented; (2) MCI subjects: MMSE scores between 24 and 30 (inclusive; exceptions made on a case-by-case basis), memory complaint, objective memory loss measured by education adjusted scores on Wechsler Memory Scale Logical Memory II, CDR of 0.5, absence of significant levels of impairment in other cognitive domains, essentially preserved activities of daily living, and an absence of dementia; (3) mild AD: MMSE scores between 20 and 26 (inclusive; exceptions made on a case-by-case basis), CDR of 0.5 or 1.0, and meets National Institute of Neurological and Communicative Disorders and Stroke/Alzheimer's Disease and Related Disorders Association criteria for probable AD. Only ADNI subjects for whom adequate processed and quality checked MR and CSF baseline data were available by February 2009 were included. Of the 105 controls, one converted to MCI after 1 year and an additional three converted to MCI over 2 years. For the MCI group, 33 of 175 converted to AD over 1 year and an additional 19 converted to AD over 2 years. Because prediction of atrophy and CDR-sb change from baseline values of CSF biomarkers and MR morphometry is the theme of the paper, all participants were included in the analyses based on their baseline diagnostic status. Sample characteristics are presented in Table 1.

MR acquisition and analysis.

All scans used for the present paper were from 1.5 T scanners. Data were collected across a variety of scanners with protocols individualized for each scanner, as defined at http://www.loni.ucla.edu/ADNI/Research/Cores/index.shtml. Raw DICOM MRI scans (including two T1-weighted volumes per case) were downloaded from the public ADNI site (http://www.loni.ucla.edu/ADNI/Data/index.shtml) and processed as described previously (Fennema-Notestine et al., 2009). Briefly, these data were reviewed for quality, automatically corrected for spatial distortion attributable to gradient nonlinearity (Jovicich et al., 2006) and B₁ field inhomogeneity (Sled et al., 1998), registered, and averaged to improve signal-to-noise. Scans were segmented as described by Fischl et al. (2002), yielding volumetric data for 15 different subcortical structures, including the hippocampal formation [consisting of the dentate gyrus, CA fields, subiculum/parasubiculum and the fimbria (Makris et al., 1999)]. The procedure (Fischl et al., 2002, 2004) uses a probabilistic atlas and applies a Bayesian classification rule to assign a neuroanatomical label to each voxel. The cortical surface was reconstructed to measure thickness at each surface point using a semiautomated approach described previously (Dale and Sereno, 1993; Dale et al., 1999; Fischl et al., 1999a,b; Fischl and Dale, 2000; Salat et al., 2004). Thickness measurements were obtained by reconstructing representations of the gray matter/white matter (WM) boundary (Dale and Sereno, 1993; Dale et al., 1999) and the pial surface and then calculating the distance between those surfaces at each point across the cortical mantle. The measurement technique used here has been validated via histological (Rosas et al., 2002) as well as manual (Kuperberg et al., 2003) measurements. The cortical surface was parcellated as described by Fischl et al. (2004), labeling 33 cortical sulci and gyri (Desikan et al., 2006), and thickness values were calculated in these regions of interest (ROIs).

A novel procedure for longitudinal analysis of volumetric data was used. For each participant, dual three-dimensional follow-up structural scans were rigid-body aligned, averaged, and affine aligned to the participant's baseline. A deformation field was calculated from a nonlinear registration according to Holland et al. (2008). The images are heavily blurred (smoothed), making them almost identical, and a merit or potential function was calculated. This merit function expresses the intensity difference between the images at each voxel and depends on the displacement field for the voxel centers of the image being transformed; it is also regularized to keep the displacement field spatially smooth. The merit function by design will have a minimum when the displacement field induces a good match between the images. It is minimized efficiently using standard numerical methods. Having found a displacement field for the heavily blurred pair of images, the blurring is reduced and the procedure repeated, thus iteratively building up a better displacement field. Two important additions to this are as follows: (1) applying the final displacement field to the image being transformed, then nonlinearly registering the resultant image to the same target, and finally tracing back through the displacement fields thus calculated to find the net displacement field; and (2) restricting to regions of interest and zooming when structures are separated by only a voxel or two. These additional features enable very precise registration involving large or subtle deformations, even at small spatial scales with low boundary contrast. Although large deformations are allowed by multiple nonlinear registration (or relaxation) steps, nonphysical deformations are precluded because at each level of blurring the image undergoing deformation is restricted to conform to the target. Note that calculating the deformation field does not depend on initially segmenting tissue. This deformation field was used to align scans at the subvoxel level. The follow-up aligned image underwent skull stripping and subcortical segmentations, with labels applied from the baseline scan. For the cortical reconstructions, surface coordinates for the white and pial boundaries were derived from the baseline images and mapped onto the follow-up images using the deformation field. Parcellations from the baseline image were then applied to the follow-up image. This resulted in a one-to-one correspondence between each vertex in the base image and the follow-up image. The procedure produces an estimate of the percentage cortical volume loss at each vertex and within each ROI. To the extent that regional areas are relatively stable across time points, the volume change is likely driven almost exclusively by changes in thickness. Estimated intracranial volume (ICV) was used to correct the volumetric data. This was calculated by use of an atlas normalization procedure described by Buckner et al. (2004), in which the atlas scaling factor is used as a proxy for ICV, shown to correlate at a level of 0.93 with manually derived ICV.

CSF acquisition and analysis.

CSF samples were obtained by lumbar puncture using a standardized protocol, as described in the ADNI procedures manual at the participating clinical sites. The CSF samples were transferred into polypropylene transfer tubes, frozen on dry ice within 1 h of collection, and shipped overnight to the ADNI Biomarker Core Laboratory at the University of Pennsylvania Medical Center. The CSF aliquots were stored in bar code-labeled polypropylene vials at −80°C. All CSF samples were analyzed over a 14 d period. Test/retest analyses of a subset of the samples showed excellent analytical performance, with r² values for test/retest results of 0.85–0.98 (Shaw et al., 2009).

t-tau, p-tau, and Aβ_1-42 levels in CSF were determined by the Luminex xMAP technology using the INNO-BIA AlzBio3 kit (Innogenetics), as described previously in detail (Olsson et al., 2005). In brief, the method is based on flow cytometric separation of antibody-coated microspheres that are labeled with a specific mixture of two fluorescent dyes. After binding of a biotinylated reporter antibody, quantification is made by binding of a third fluorochrome coupled to streptavidin. The method allows simultaneous quantification of several proteins the same sample and has been shown to have high analytical precision (Olsson et al., 2005). Mean values are presented in supplemental Table 1 (available at www.jneurosci.org as supplemental material).

Statistical analyses.

Cortical thickness and volume change data were analyzed on a vertex-by-vertex basis, covering the entire cortical mantle, by use of general linear models (GLMs). In addition, 30 ROIs were chosen for further analyses (Fig. 1). To reduce the number of comparisons, MR ROIs were averaged across hemispheres. ICV, sex, and age were regressed out from all subcortical ROI measures before they were entered into statistical analyses, and sex and age were used as covariates in all surface-based analyses. Because sex does not influence levels of CSF biomarkers and age and CSF biomarker levels are not correlated in the present sample (p > 0.60), sex and age were not regressed out from the CSF biomarker values unless specified. Analyses were performed for t-tau, and p-tau, Aβ₄₂, as well as the ratios between each of the tau measures and Aβ₄₂. Primary focus is on tau and Aβ₄₂, but the ratio scores were included to allow comparisons with previous studies.

The group contrasts of interest were NC versus MCI and NC versus AD. We did not focus on the MCI versus AD contrast, because only a subgroup of the MCI patients will eventually develop AD, making it difficult to interpret the differences between the MCI group and the AD group. Furthermore, because AD has an insidious onset, cutoff will necessarily be somewhat arbitrary, with some MCI patients having mild AD. This also contributes to making the interpretations difficult. Nevertheless, we will present results from MCI versus AD comparisons as supplemental data (available at www.jneurosci.org as supplemental material). For the surface-based analyses, a false discovery rate (FDR) < 0.05 was used as the statistical criterion of significance, corrected for multiple comparisons, whereas an α value of 0.01 was used for the ROI analyses. FDR was used to decide α level for the left hemisphere (arbitrary chosen), and then the same level was used to threshold right hemisphere data. In addition to surface-based GLMs, t tests were used for the ROIs and CSF biomarker values to perform between-group comparisons. Effect sizes for all measures were quantified by use of Cohen's D (group difference divided by the pooled SD).

The first question regarded whether CSF biomarker levels can explain the morphometric differences between NC and MCI and between NC and AD. First, CSF biomarkers were related to baseline morphometry by surface-based GLMs point by point across the cortical surface and by linear regressions. Next, GLMs were used to assess difference in cortical across the cortical surface both with and without the three CSF measures used as simultaneous covariates. The results were compared with an analysis performed without the CSF biomarkers as covariates, so that a group effect might indicate that the CSF measures influenced the relationship between diagnosis and baseline thickness. Third, separate regression analyses were run using each MR ROI as a dependent variable, with diagnosis, sex, and age as predictors. The regressions were run both with and without the three CSF measures entered simultaneously as covariates. A substantial decrease in β values for diagnosis when CSF variables are included among the predictors would indicate that the effect of diagnosis on morphometry was influenced by differences in CSF biomarker levels. These analyses were performed only for those ROIs in which significant correlations with CSF biomarkers were found in the MCI or the AD group.

The second question regarded how baseline brain morphometry and CSF biomarkers are related to longitudinal brain atrophy. Atrophy rates in the diagnostic groups were examined and tested with linear mixed-effect models with change (both 1 and 2 years included in the same analysis) × three diagnostic groups (NC, MCI, and AD) for each of the 30 ROIs. Baseline cortical thickness was used as a regressor on cortical change at 1 year point by point in the MCI sample, and linear mixed-effects models were run to test the relationship between baseline morphometry and change in each of the 30 ROIs. The relationship between CSF biomarker levels and atrophy was tested by surface-based GLMs and by running linear mixed-effect models for CSF biomarker values with percentage change in volume for each of the ROIs. For 12 selected ROIs in which CSF biomarkers were related to atrophy, rate of change was compared between NC, MCI with highest (high-risk) and MCI with lowest (low-risk) levels of p-tau, and highest (low-risk) and lowest (high-risk) levels of Aβ₄₂, based on median splits. The split values were almost identical to the medians for the MCI group reported in the major ADNI paper on CSF biomarkers (Shaw et al., 2009). Dividing the MCI patients in two groups yielded the opportunity to compare pairwise the NC group and each of the MCI groups. The relationships were tested by use of linear mixed-effects models, and the results were shown as surface overlays and plotted as atrophy graphs with error bars. A mixed linear model analysis was run to test whether apolipoprotein E (APOE) genotype had an effect on rate of change in the MCI group, either alone or in interaction with levels of CSF biomarkers. APOE was coded as presence of zero, one, or two ε4 alleles.

To answer the third question, of whether CSF biomarkers and morphometry at baseline are related to change in CDR-sb, the relationship between the CSF biomarkers, the 30 ROIs, and CDR-sb at baseline, 1 and 2 years were tested with linear mixed-effect models. In addition, Pearson's correlations between CSF biomarkers, brain morphometry and change in CDR-sb scores were used to test the relationships at each time point separately.

For questions 2 and 3, the analyses of interest were performed in the MCI group, because predictors of brain atrophy and clinical change (as indexed by CDR-sb) are especially important in MCI as it is suggested to represent a transitional phase between NC and AD (Petersen, 2004; Petersen and Bennett, 2005). Additionally, the MCI group was approximately twice the size of the other groups, ensuring adequate statistical power. The analyses were repeated for AD, and the results are presented as supplemental online data (available at www.jneurosci.org as supplemental material) and briefly commented on below.