We present evidence for the cumulative diffusion model as a possible mechanism for AD-specific brain atrophy. In this model, rates of atrophy behave nonlinearly (ie, with a sigmoidal pattern), increasing initially, peaking at the inflection point, and decreasing thereafter.
Rates of atrophy in AD-vulnerable cortical ROIs and in the hippocampus in presymptomatic individuals with an AD-like CSF molecular profile were statistically indistinguishable from those of healthy controls. However, the former individuals exhibited an elevated profile, suggesting that these 2 groups can be potentially discriminated with a larger sample size. Of interest, the mean thickness of AD-vulnerable ROIs was significantly reduced in this presymptomatic group, but hippocampal volume was not. This finding suggests that cortical thickness in AD-vulnerable ROIs may be a sensitive biomarker in the earliest stages of the disease process. Longitudinal rates of atrophy in AD-vulnerable cortical and hippocampal regions, however, seem unlikely to be useful for tracking disease progression during the presymptomatic period. These results are further supported by a supplemental analysis of a small number of individuals (n=10) who were CN at baseline but had progressed to aMCI or AD within 3 years of follow-up.
Our results from the CN group add to the growing body of literature that demonstrates cortical thickness measurements from select ROIs as sensitive markers of very early AD.39–41
This observation does not necessarily contradict the widely accepted pathophysiology of AD that is characterized by early neurofibrillary tangle deposition in the medial temporal lobe (including the hippocampus) and associated atrophy in these regions because volume measurements of a structure as large as the hippocampus may be less sensitive to subtle and localized atrophy than those of ROI-based cortical thickness. Moreover, our analyses suggest that AD-specific cortical thinning in CN individuals is mostly associated with β-amyloid and not tau. The recently demonstrated tight coupling between hippocampal atrophy and episodic memory impairment42
further suggests that significant volume loss in the hippocampus may be associated with clinical symptoms and therefore is unlikely to occur during a preclinical phase. That is, hippocampal atrophy may mark the transition to cognitive symptoms.
Thickness in AD-vulnerable cortical ROIs and hippocampal volume are significantly reduced during aMCI and clinical AD. Consistent with data from prior studies,5–8,18
rates of AD-specific cortical thinning and hippocampal volume loss also are significantly elevated and correlate with concurrent cognitive decline.
Consistent with the sigmoidal pattern, the rate of cortical thinning accelerates throughout the presymptomatic and aMCI stages, starting from levels indistinguishable from those of healthy controls and reaching its fastest pace at approximately the MMSE score of 21. Although continuing to progress, AD-specific cortical thinning starts to slow beyond this point. This characterization was consistent across all 7 AD-vulnerable cortical ROIs that we examined. Hippocampal atrophy rates, however, exhibit a progressively increasing pattern without a clearly discernible peak before the MMSE score of 15.
Our longitudinal observation is in broad agreement with a recent cross-sectional characterization of the dynamics of AD biomarkers,24
in which a sigmoidal pattern of hippocampal atrophy was demonstrated. Building on data from prior serial imaging studies,7,14,18
our results demonstrate that AD-specific brain atrophy is characterized by early acceleration, possibly driven by cumulative insults, such as amyloid toxicity, tangle deposition, and neuronal and synaptic dysfunction, followed by late deceleration, constrained by the diminishing residual intact tissue.
The nonlinear, sigmoidal pattern has important implications for clinical trials. First, a linear characterization of brain atrophy can lead to incorrect sample size estimates and underpowered clinical trials. Second, the bell-shaped derivative of the sigmoid implies that early in the disease process, atrophy rates are likely to be indistinguishable from those of controls and therefore probably will be of limited use in tracking progression. Finally, the natural deceleration observed in later disease stages needs to be carefully considered when assessing a disease-modifying therapeutic effect in an AD trial.
The present study uses longitudinal neuroimaging data collected from multiple sites and thus demonstrates the potential use of these biomarkers in multicenter clinical trials. Another important aspect of the study is the use of cortical ROIs defined with an independent sample. In contrast with traditional methods that use anatomical landmarks, this approach identifies a disease-specific effect through an exploratory analysis,39
which yields sensitive markers of disease.
The present study has several limitations. One issue involves the use of a cross-sectional design in which cohort effects can confound results. A second concern pertains to the usefulness of a CSF-based cutoff to select participants who have a molecular profile consistent with AD and our assumption that the conditions of these individuals exist on the same disease trajectory. Hence, our findings are contingent on the validity of this hypothetical trajectory. Another limitation involves the examination of longitudinal atrophy during a 1-year period. It is possible that acceleration and deceleration patterns of cortical regions vary substantially across individuals. Furthermore, the limited number of patients in the severe stages of dementia (eg, with an MMSE score <20) may have biased our computation of where cortical thinning rates peak. As additional longitudinal MRI data during multiyear periods become available, future studies will examine the validity of these findings. Finally, the apparent slowing of AD-specific cortical thinning may not occur due to the underlying biology but may be a consequence of the technical difficulty of resolving thickness changes around and beyond the voxel resolution. Although our characterization of AD-specific cortical thinning as a dynamic biomarker still will be valid, in the interest of understanding the underlying biology of this region, this potential confounder needs to be examined in future work.