Correlations between FDG PET glucose uptake-MRI gray matter volume scores and apolipoprotein E ε4 gene dose in cognitively normal adults: A cross-validation study using voxel-based multi-modal partial least squares

Abstract

We previously introduced a voxel-based, multi-modal application of the partial least square algorithm (MMPLS) to characterize the linkage between patterns in a person's complementary complex datasets without the need to correct for multiple regional comparisons. Here we used it to demonstrate a strong correlation between MMPLS scores to characterize the linkage between the covarying patterns of fluorodeoxyglucose positron emission tomography (FDG PET) measurements of regional glucose metabolism and magnetic resonance imaging (MRI) measurements of regional gray matter associated with apolipoprotein E (APOE) ε4 gene dose (i.e., three levels of genetic risk for late-onset Alzheimer's disease (AD)) in cognitively normal, late-middle-aged persons. Coregistered and spatially normalized FDG PET and MRI images from 70% of the subjects (27 ε4 homozygotes, 36 ε4 heterozygotes and 67 ε4 non-carriers) were used in a hypothesis-generating MMPLS analysis to characterize the covarying pattern of regional gray matter volume and cerebral glucose metabolism most strongly correlated with APOE-ε4 gene dose. Coregistered and spatially normalized FDG PET and MRI images from the remaining 30% of the subjects were used in a hypothesis-testing MMPLS analysis to generate FDG PET-MRI gray matter MMPLS scores blind to their APOE genotype and characterize their relationship to APOE-ε4 gene dose. The hypothesis-generating analysis revealed covarying regional gray matter volume and cerebral glucose metabolism patterns that resembled those in traditional univariate analyses of AD and APOE-ε4 gene dose and PET-MRI scores that were strongly correlated with APOE-ε4 gene dose (p < 1 × 10^− 16). The hypothesis-testing analysis results showed strong correlations between FDG PET-MRI gray matter scores and APOE-ε4 gene dose (p = 8.7 × 10^− 4). Our findings support the possibility of using the MMPLS to analyze complementary datasets from the same person in the presymptomatic detection and tracking of AD.

Introduction

Patients with Alzheimer's disease (AD) have abnormally low [18F] fluorodeoxyglucose (FDG) positron emission tomography (PET) measurements of the cerebral metabolic rate for glucose (CMRgl) in the precuneus, posterior cingulate, parietotemporal, and frontal cortex (Alexander et al., 2002, Landau et al., 2009, Langbaum et al., 2009, Minoshima et al., 1994). In structural magnetic resonance imaging (MRI)measurements, AD patients have reduced volume in the entorhinal cortex (Du et al., 2004, Jessen et al., 2006, Killiany et al., 2002), medial temporal lobe (Barta et al., 1997, Convit et al., 1997, Jack et al., 1998), hippocampus (Ball et al., 1985, Basso et al., 2006), and whole brain volume (Fox et al., 1996). Decreased CMRgl (Drzezga et al., 2003, Herholz et al., 2007) and smaller hippocampal volume (Desikan et al., 2009, Grundman et al., 2002) have also been reported in patients with mild cognitive impairment (MCI) who subsequently developed AD.

We have used FDG PET and volumetric MRI in an ongoing longitudinal study to detect and track changes in brain function and brain structure in cognitively normal, late-middle aged persons with two copies, one copy, and no copies of the apolipoprotein E (APOE-ε4) allele, a major susceptibility gene in persons with AD with onset of dementia after age 60, representing at three levels of genetic risk for late-onset Alzheimer's disease. Our group, and others, have previously shown that ε4 homozygote (HM) and heterozygote (HT) groups have reduced cerebral glucose metabolic rates compared to noncarriers (NC) in the same brain regions as patients with probable AD and higher rates of the cerebral glucose metabolic rate decline in these and other brain regions (Higuchi et al., 1997, Reiman et al., 1996, Reiman et al., 2001, Reiman et al., 2005, Rimajova et al., 2008). Additionally, we have shown tendencies for smaller hippocampal volumes (Reiman et al., 1998) and whole brain atrophy rates (Chen et al., 2007, Reiman et al., 1998) in ε4 carriers compared to non-carriers. Similar findings have been reported in mild cognitive impairment (MCI) and AD patients who are ε4 carriers (Drzezga et al., 2009, Filippini et al., 2009, Fleisher et al., 2005, Geroldi et al., 2000, Hashimoto et al., 2001, Liu et al., 2010, Pievani et al., 2011).

We previously developed a method, multi-modal partial least square (MMPLS), to capitalize on complementary information for two or more complex data sets (e.g., FDG PET CMRgl images of volumetric MRI gray matter maps (Chen et al., 2009)). The resulting spatial patterns were similar to those reported in conventional voxel-based univariate statistical analyses. The MMPLS derived global index of FDG PET CMRgl-MRI gray matter scores, distinguished healthy young and older subjects with significantly increased power compared to using either data alone.

In this study, we used the MMPLS and a cross-validation scheme to demonstrate a strong correlation between FDG PET-MRI gray matter MMPLS scores and APOE-ε4 gene dose, reflecting three levels of genetic risk for late-onset AD, in cognitively normal, late-middle-aged persons. To examine the increased strength of such correlation, we also estimated the correlation of APOE ε4 gene dose with FDG PET score alone or with MRI gray matter alone using unimodal PLS (Chen et al., 2008, McIntosh et al., 1996). The PET-MRI MMPLS scores and their linked covarying patterns were initially generated in a training data set and then confirmed in a test set. Another extension to the previous method is the application of a permutation test to assess type-I error of the generated covarying patterns.