The current study reveals an effect of the APOE
ε4 allele for greater hippocampal atrophy in subjects with a diagnosis of MCI or AD as compared to individuals without the ε4 allele in these diagnostic cohorts. That is to say, subjects with the APOE
ε4 genotype have significantly worse atrophic changes seen in the hippocampi as Alzheimer-spectrum pathology progresses. Furthermore, not only does presence of ε4 correlate with greater atrophy, but the extent to which it is greater is dose-dependent with respect to the number of copies of ε4 that are present, i.e. it is linearly proportional to allele load. Notably, we were unable to demonstrate this effect in CN controls. It is difficult to say why this is the case given several previous reports of APOE
ε4-driven effects in the hippocampus of cognitively normal controls (see 
for example). It is possible that this is due to insufficient statistical power; in fact, there were only n
5 CN ε4 homozygotes that met inclusion criteria for the study. This small number of homozygotes, in conjunction with the fact that our model was constrained to the assumption of dose-dependency with regard to allele load, could potentially explain why the results from the dose-dependent model () do not detect an overt APOE
ε4 effect. This result is consistent with findings of several recent studies including those of Crivello et al
who showed that there is no APOE
ε4 dose-effect on longitudinal hippocampal atrophy rates in healthy controls, and Tupler et al
whose study of 159 CN elderly subjects revealed no difference in baseline hippocampal volume according to APOE
ε4 status 
. However, given the discordance between our findings and those of several previous reports, it is ultimately unclear as to the reason an APOE
ε4-effect was not seen in CN individuals.
While effects of ε4 on the hippocampus have been found previously 
, the demonstration of a dose-response is new to our knowledge, and differs from three previous studies that have examined this issue: Filippini et al
. showed that the APOE
ε4 allele exerts a dominant rather than dose-dependent effect on decreased deep gray-matter volumes, and neither Lemaitre H. et al
nor Liu Y. et al.
found a specific dose-dependent relationship as we have defined it here 
. Issues related to sample size, variations in population ε4 allelic frequencies and demographic characteristics, non-uniformity across volumetric-derivation methods, and lack of fully-automated algorithms are possible sources of variation that could account for these discrepancies.
Our conclusions regarding ε2-driven effects are not as certain. For instance, though our model indicates a significant dose-dependent main effect of the ε2 allele for moderate hippocampal protection, this result seems to be driven almost exclusively by the only two ε2 homozygotes in the study, both CN subjects with exceptionally large hippocampi. Though statistically valid according to our analysis, it is difficult to generalize from such a small number of subjects. It is worth noting, however, that using a slightly altered model–which treats APOE
as a nominal or categorical (carrier
) variable–in order to re-analyze a sub-set of this data which excludes these two ε2 subjects: 1) does not significantly alter the ε4 effects seen, and 2) still generates results suggestive of a protective morphometric main effect of a single ε2 allele (an effect size of +505.84 mm3
, equivalent to a nearly 13% increase in hippocampal volume; Cohen’s d
0.14, effect size r
0.68)–though, this result is only marginally significant (p
0.07) (see for a full summary of the results for this modified model). Thus, though our results are highly suggestive and may serve to guide future inquiry into the question, it currently may be difficult to draw robust conclusions regarding the putative protective effect the ε2 allele has on hippocampal morphometry. These results differ somewhat from previous studies examining ε2 effects, none of which were able to find a significant difference in hippocampal volumes according to APOE
ε2. This includes both those that relied on manual-only volumetry 
as well as even more recent studies using fully-automated methods 
–the latter two both having reported non-significant trends for larger mean hippocampal volumes for ε2 carriers versus non-carriers.
Summary of effects for variables on hippocampal volume: APOE nominal model.
There are limitations to the current study. Firstly, the number of ε4 (n
5) and ε2 (n
2) homozygotes included in our analyses was less than ideal; as mentioned, the presence of only two ε2 homozygotes makes it difficult to generalize from any results regarding an ε2 dose-effect, and similarly, ε4-driven effects in CN subjects may have been below our threshold for detection given the low number of ε4 homozygotes in conjunction with the constraints imposed by the dose-dependent model. Secondly, the overall ε2 allele frequency in the present study (48/1324; 3.6%) was low for one which set out to test for a main effect of this allele. Thirdly, though the ADNI study allows for examination of large study cohorts, this particular group of subjects has previously been shown to have a higher proportion of whites, to be freer of co-morbid conditions, and to be more educated than community-based samples 
, which complicates generalizations of findings using ADNI data. Finally, there are the limitations imposed by the cross-sectional study design, namely, the limitation that there can be no inferences made with regard to causality. Specifically, we have used the term “atrophy” throughout our paper to refer to ε4-driven effects for smaller hippocampal volumes; although our data are consistent with other longitudinal data for atrophy rates, our cross-sectional design is unable to determine if such volume differences are due to accelerated atrophy, or simply a genetic trait marker.
Despite these limitations, the results of this study have implications for our understanding of the mechanisms underlying the well-documented APOE
ε4-mediated acceleration of AD pathophysiology. Specifically, the dose-effects shown here are consistent with the mechanisms described by Jiang et al.,
among others, in which the ε4 isoform of ApoE results in a protein markedly less able to fulfill its neuro-protective tasks–tasks that are much more readily accomplished by the ε2 and ε3 isoforms–such as the proteolytic degradation of extracellular amyloid-β in the brain 
Future directions should include re-examination of this issue with a more substantial and statistically robust cohort that includes a higher number of individuals possessing the alleles of interest. Also of further importance would be to explore this cohort for any dose-dependent effects of the ε4 and ε2 alleles on the longitudinal atrophy rates of these measures, with the goal of using APOE genotype to better interpret quantitative, longitudinal imaging data for prognostic purposes.