Hippocampal dentate gyrus integrity revealed with ultrahigh resolution diffusion imaging predicts memory performance in older adults

Abstract

Medial temporal lobe (MTL) atrophy is a core feature of age-related cognitive decline and Alzheimer's disease (AD). While regional volumes and thickness are often used as a proxy for neurodegeneration, they lack the sensitivity to serve as an accurate diagnostic test and indicate advanced neurodegeneration. Here, we used a submillimeter resolution diffusion weighted MRI sequence (ZOOMit) to quantify microstructural properties of hippocampal subfields in older adults (63-98 years old) using tensor derived measures: fractional anisotropy (FA) and mean diffusivity (MD). We demonstrate that the high-resolution sequence, and not a standard resolution sequence, identifies dissociable profiles for CA1, dentate gyrus (DG), and the collateral sulcus. Using ZOOMit, we show that advanced age is associated with increased MD of the CA1 and DG as well as decreased FA of the DG. Increased MD of the DG, reflecting decreased cellular density, mediated the relationship between age and word list recall. Further, increased MD in the DG, but not DG volume, was linked to worse spatial pattern separation. Our results demonstrate that ultrahigh-resolution diffusion imaging enables the detection of microstructural differences in hippocampal subfield integrity and will lead to novel insights into the mechanisms of age-related memory loss.

Keywords: aging; dentate gyrus; hippocampus; magnetic resonance imaging; memory.

Figure 1:

(a) Shows the FA and MD data for each area of interest: CA1 (in blue), DG (in green), rhinal cortex (in yellow), and sulcus (in black) derived from the high-resolution DWI sequence. (b) Shows the same data points categorized into 4 clusters via unsupervised K-means clustering. We note the large degree of visual overlap (Normalized mutual information score =0.623). (c) Shows the actual FA and MD data plotted for each area of interest: CA1 (in blue), DG (in green), rhinal cortex (in yellow), and sulcus (in black) derived from the whole brain DWI sequence. (d) Shows the same data points clustered into 4 clusters using unsupervised K-means clustering. Here we note the minimal degree of overlap between the actual data and predicted data identities (Normalized mutual information score = 0.242)

Figure 2:

(a-b) Shows the results of CA1 structural differences across the age range. (c-d) Show the results of DG structural differences across the age range. (a) Shows the non-significant relation between age and CA1 FA (r = −0.13, p = 0.26). (b) Shows the positive relation between age and CA1 MD (r = 0.52, p < 0.0001). (c) Shows the negative association between age and DG FA (r = −0.47, p <0.0001). Finally, (d) shows the positive relation between age and DG MD (r = 0.66, p < 0.0001).

Figure 3:

Here we report the results of DG MD, FA, and volume predicting delayed recall performance. (a) Shows the significant association between DG MD and delayed recall performance (r = −0.54, p < 0.0001). (b) Shows the non-significant but trending relation between DG FA and delayed recall performance (r = 0.21, p = 0.064).

Figure 4:

Results of structural equation (mediation) model shows DG MD is a statistically significant mediator between age and delayed recall performance (Indirect Effect = −0.069*, C.I. = −0.13 to −0.0082). *Indicated p < 0.05, ** indicates p < 0.01, *** indicates p < 0.0001

Figure 5:

In a subset of participants, we tested whether DG MD might also predict performance on a Spatial Pattern Separation Task. (a) The Spatial Pattern Separation Task. During incidental encoding, subjects are shown a series of objects placed in different spatial locations on a screen. Subjects are asked during incidental encoding to rate the images as appearing as “Indoor” or “Outdoor”. During test, targets or objects appearing in the same location as encoding are presented as well as lures which present the same objects in slightly different locations. Here subjects are asked if the object was in a “New Location” or “Old Location”. (b) There is a statistically significant association between DG MD and LDI on the spatial task (r = −0.39, p = 0.0051). (c) However, there is not a statistically significant association between DG volume and LDI (r = 0.12, p = 0.40).