Canonical correlation analysis of hippocampus and basal forebrain cholinergic brain structures with cognitive domain‐specific performance scores

Abstract Background Hippocampal atrophy is a diagnostic marker of AD and various compartments of the basal forebrain cholinergic (BFC) system also show volume reductions that correlate with cortical thinning and volume reduction in their innervated targets in the hippocampus. Moreover, bilateral volume loss of the nucleus basalis Meynert (Ch4) and spreading the atrophy to medial septum and vertical limb of the diagonal band of Broca (Ch12) in preclinical and prodromal AD show correlation with abnormal CSF levels of Aβ 1‐42 and phosphorylated tau. It is unclear, however, how volume integrity of hippocampus and various compartments of the cholinergic space (Ch12 and Ch4) associate with specific cognitive domains. Method Data from 402 participants were downloaded from the AD Neuroimaging Initiative [120 AD, 118 cognitively normal (CN), and 164 mild cognitive impairment (MCI)]. For each participant, domain‐specific performance scores were downloaded for: memory (ADNI‐MEM), executive function (ADNI‐EF), language (ADNI‐LAN) and visuo‐spatial ability (ADNI‐VS). In addition, all participants had 3D T1‐weighted structural MRI scans (3T) from which we computed the structural integrity measures for the left and right hippocampus (LHC, RHC), Ch4 (LCH4, RCH4) and Ch12 (LCH12, RCH12) BFC nuclei. Canonical correlation analysis (CCA) was used to determine the association between the 4 cognitive measures on the one hand and the 6 brain structural MRI measures on the other hand after controlling for age and sex. Result The first and second canonical correlation coefficients were found to be statistically significant: r 1 = .56 (p<.001) and r 2 = .23 (p = .011). The first pair of canonical correlation vectors (CCVs), indicated that predominantly ADNI‐MEM (Fig. 1) is associate with a bilateral hippocampal integrity (LHC and RHC) (Fig. 2) particularly on the right. Moreover, the corresponding cognitive and structural canonical variates differentiated between CN, MCI and AD (Fig. 3). The second pair of CCVs indicated that predominately ADNI‐LAN (Fig. 4) was associated with RHC>LHC asymmetry (Fig. 5). However, the corresponding canonical variates did not differentiate between groups (Fig. 6). Conclusion Bilateral hippocampal atrophy is associated with reduced ADNI‐MEM performance, and both differentiate CN, MCI and AD. Asymmetry of the hippocampal volumetric integrity is associated with ADNI‐LAN (LZ was supported by NIH RF1NS023945‐28).