Our main findings are: (i) in mild cognitive impairment and Alzheimer's disease, progression of hippocampal loss was detected over 6 months and accelerated over 1 year, whereas in the normal group hippocampal loss was detected over 1 year with no indication of acceleration; (ii) ApoE4 was associated with higher rates of hippocampal loss in Alzheimer's disease patients, irrespective of their level of cognitive impairments; (iii) higher rates of hippocampal loss correlated with a lower concentration of CSF Aβ1–42, irrespective of ApoE genotype, predominantly in mild cognitive impairment patients. Furthermore, we showed that the power of measuring hippocampal change can be improved by exploiting intrinsic correlations between successive MRI observations. In addition, we showed that site-to-site variations in MRI can effectively be brought to levels similar to single site settings using the rigorous methods of the ADNI.
We found significant hippocampal volume loss in mild cognitive impairment and Alzheimer's disease over only 6 months. All but one other MRI study reported hippocampal loss over such a short period and only for Alzheimer's disease (Barnes et al.
). Furthermore, the prior study conducted MRI at a single centre only, yielding no conclusions for multicentre trials. Nonetheless, our results in Alzheimer's disease and those from this single centre study are intriguingly similar. Our results (average of 0–6 and 6–12 months rate values) expressed as percentage annual change yield 4.5% hippocampal loss in Alzheimer's disease (2–6% within 95% CI) compared with the range 4.35–5.04% that the single centre study reported. For mild cognitive impairment, our results over 6 months yield 2.5% annualized hippocampal loss (2.0–3.3% within 95% CI), in good agreement with reports of most other longitudinal MRI studies of mild cognitive impairment (Jack et al.
, 2000; Du et al.
; Fox et al.
; Hashimoto et al.
) that used much longer scan intervals, including two multicentre trials (Jack Jr et al.
; van de Pol et al.
). The rates of hippocampal loss in our study are also within the range found in a large meta-analysis (Barnes et al.
). It is furthermore re-affirming that our results obtained with a semi-automated method for tracing the hippocampus are comparable with those that employed entirely manual methods (Jack et al.
). Taken together, our results imply that an assessment of hippocampal loss over 6 months is possible and this extends to multicentre trials. However, measurement power is clearly sacrificed at shorter intervals as indicated by the larger standard errors for rates from measurements over 6 months compared to those over 12 months. Limited spatial resolution of MRI is likely the main reason for incurring errors at shorter scan intervals. Therefore, longitudinal studies of hippocampal loss should benefit from higher MRI resolution, if it can be afforded. Since a quarter of the participants in the ADNI
will also be scanned at 3T parallel to 1.5T but at 20% higher resolution, the impact of image resolution on the power to measure brain volume loss can ultimately been tested. It is surprising that the rate of volume loss in mild cognitive impairment between 6 and 12 months is similar to that in Alzheimer's disease between 0 and 6 months, though some mild cognitive impairment subjects are destined to develop Alzheimer's disease. A possible explanation for this observation is that accelerated volume loss over time is a more prominent feature that separates Alzheimer's disease from mild cognitive impairment than constant volume loss. In addition, the annual percentage change from baseline is lower in mild cognitive impairment than in Alzheimer's disease.
The rates of hippocampal loss also correlated with rates of cognitive decline. In general, Alzheimer's disease and mild cognitive impairment patients with high hippocampal rates also had rapidly increasing ADAS-Cog scores, while the correlation with MMSE scores was weaker. It is possible that cognitive tests that are more specific for hippocampal function show stronger correlations with MRI. Additional analyses that are beyond the scope of this report are warranted to further explore the cognitive correlates of hippocampal volume changes.
We also found significant hippocampal loss in the normal group over 1 year, in agreement with several prior MRI studies (Fox and Schott, 2004
; Du et al.
; Jack et al.
) as well as with autopsy findings of neuronal loss in the ageing hippocampus (West, 1993
; Simic et al.
). However, to determine if hippocampal loss in normal subjects is already an indication of incipient Alzheimer's disease or other pathologies affecting the hippocampus requires clinical follow-up of the subjects to determine their cognitive decline and ultimate development of Alzheimer's disease.
We found accelerated hippocampal loss in the mild cognitive impairment and Alzheimer's disease groups over a period of 1 year. Accelerated rates of hippocampal loss have previously been reported for mild cognitive impairment (Jack et al.
) and for familial Alzheimer's disease (Ridha et al.
), but over periods that ranged from 2 to 5 years. Our finding is also consistent with reports of accelerated loss of whole brain (Chan et al.
; Carlson et al.
) and clinical studies of accelerated cognitive decline in mild cognitive impairment and early Alzheimer's disease (Mungas et al.
; Boyle et al.
). To estimate accelerated hippocampal loss, we used a simple quadratic expansion for change that may not accurately reflect the true progression of hippocampal loss. Observations at much more than three time- points, as planned in the ADNI
, should help to better characterize the trajectory of accelerated loss. Nonetheless, the finding of accelerating hippocampal loss is important for understanding the natural history of Alzheimer's disease and emphasizes the need for early diagnosis and therapeutic intervention. The fact that we were able to detect accelerated rates over 1-year period has consequences for longitudinal MRI studies with only two time-points. First, such studies may overestimate or underestimate hippocampal loss rates since two serial measurements are indifferent to accelerations. Second, the ability to detect differences may be limited if accelerations vary among subjects. The ADNI
is funded to scan each subject multiple times over 3 years to provide more information on longitudinal change.
Our finding that ApoE4 is a modulator of hippocampal loss rates is consistent with other MRI studies (Moffat et al.
; Cohen et al.
; Jack Jr et al.
; van de Pol et al.
), but some studies found no ApoE effect (Laakso et al.
). Furthermore, whereas others reported an ApoE4 effect on hippocampal rates for mild cognitive impairment (Jack Jr et al.
), we found it limited to Alzheimer's disease. In a recent cross-sectional MRI study of 676 ADNI
subjects using tensor-based morphometry (Hua et al.
), over half of the Alzheimer's disease and mild cognitive impairment subjects carried the ApoE4 gene, and they showed greater hippocampal and temporal lobe deficits than non-carriers. Around one-sixth of the controls carried the protective ApoE2 gene and showed reduced ventricular expansion, perhaps reflecting a lesser degree of overall brain atrophy. We did not find a significant dose effect of ApoE4 on hippocampal loss rates in contrast to another study (van de Pol et al.
). Given our sample size of 226 mild cognitive impairment patients and ~10% within subject variation, we should have been able to detect about 5% difference between carriers and non-carriers with 80% power and at α = 0.05 significance. One possible explanation for the discrepant ApoE findings in mild cognitive impairment is the notorious heterogeneity of this group, which—in absence of histological evidence for Alzheimer's disease pathology—may include other causes of cognitive complaints, such as mood disorders or cerebrovascular disease that both can impact the hippocampus (Lloyd et al.
; Mungas et al.
). Nonetheless, our finding suggests that presence of ApoE4 exacerbates the impact of Alzheimer's disease on the hippocampus. Alzheimer's disease cohorts of future therapeutic trials could be enriched by including specifically patients who carry ApoE4 and are likely to have small hippocampi and high rates of hippocampal loss.
We also found a marked association between higher rates of hippocampal loss and decreased concentration of Aβ1–42
in CSF, predominantly in the mild cognitive impairment group, whereas for t-tau we found only a trend in the Alzheimer's disease group and for p-tau no significant association in any of the groups. Aβ1–42
and tau concentrations in CSF represent the earliest and most intensely studied biochemical markers of Alzheimer's disease (Frank et al.
; Grossman et al.
). How well these biomarkers reflect an autopsy-confirmed dementia diagnosis has intensely been studied (Clark et al.
). Both proteins are directly linked to the two hallmark lesions of Alzheimer's disease, Aβ1–42
with amyloid plaques and tau (t-tau and p-tau) with neurofibrillary tangles (Clark et al.
). Numerous studies have documented reduced Aβ1–42
and increased p-tau in CSF in Alzheimer's disease patients (see Clark et al.
for review). But only two prior MRI studies compared CSF biomarkers with hippocampal loss rates in a small pool of subjects. Consistent with our findings, de Leon et al.
) reported greater hippocampal loss with greater Aβ1–42
decrease. In contrast to our results in mild cognitive impairment, they also found significantly greater hippocampal loss with increased p-tau but their study included only nine mild cognitive impairment subjects. Hampel et al.
), studying Alzheimer's disease, found increased hippocampal rates correlated with increased p-tau in 22 patients, while we found only a trend. Our finding that increased rates of hippocampal loss correlate with decreased Aβ1–42
in a large number of mild cognitive impairment patients is particularly interesting, because Aβ1–42
is directly related to Alzheimer's disease pathology. It is also important to note that the association between high hippocampal rates and reduced CSF Aβ1–42
was independent of the ApoE profile, implying that CSF Aβ1-42
levels and rates of hippocampal loss are directly linked. The results support the view that high rates of hippocampal loss in mild cognitive impairment indicate Alzheimer's disease pathology. Since the hippocampus is spared from early amyloid burden (Silbert et al.
), the correlation between high rates of hippocampal loss and decreased CSF Aβ1–42
, further implies that the two measures provide complimentary information about the presence of Alzheimer's disease pathology. However, the diagnostic value of the measures used together remains unclear, because each measure can also change in other pathological conditions, such as Lewy body dementia (Clark et al.
Using a Markov chain model to analyse hippocampal change, we showed that intrinsic correlations between successive MRI observations exist and can be exploited to reduce within subject variability and consequently improve measurement power. The impact on power was substantial despite the fact that we performed hippocampal tracing independently for each time-point. The results imply that image processing algorithms that utilize correlations between observations, i.e. by using the hippocampal boundaries from past MRI scans as priors for tracing the boundaries in a new MRI scan, should be superior to those algorithms that do not employ priors. However, the approach also has limitations, as our results indicate. For instance, the Markov chain approach did not benefit the analysis of data from the normal group as much as the analysis of the mild cognitive impairment or Alzheimer's disease group, presumably because errors in assessing the small volume changes in normal subjects were random and dominated systematic biological changes. At the other end, the benefit of a Markov chain analysis was also less effective for Alzheimer's disease than for mild cognitive impairment data, presumably because the volume changes in Alzheimer's disease patients were sufficiently large to incur fewer errors to begin with. Since each subject in the ADNI study will ultimately have multiple successive MRI scans, it will be possible to evaluate the benefit of analysis using Markov chain models in more detail.
Our results demonstrate that for studies of hippocampal rates site-to-site variations in MRI can effectively be controlled using the rigorous methods of the ADNI. The result is important, because multiple MRI centre settings are indispensable for large studies, such as clinical trials with hundreds of subjects. Since additional variability is inevitably introduced in multiple MRI site settings compared to a single site setting, which is more common for investigational study use, we were also interested in comparing the powers of multisite and single site studies to detect a certain level of atrophy rates. Based on the site-to-site variations shown in , we estimated that 80% power of this multisite study of 47 MRI sites translates roughly into 87% power if the same study with the same population and number of subjects was conducted at the ‘best’ single MRI site in this study (#116 in ) with the least variability of all sites. Although the result implies a benefit in power from a single site, as expected, the gain overall seems small, especially if one considers that the other participating sites in this study have smaller benefit margins since they show greater variability than the ‘best’ single site. In summary, the result attests to the effectiveness of the rigorous control methods developed by the ADNI. Our power estimations also show that MRI consistently provides greater power to measure progression than cognitive tests, such as ADAS-Cog or MMSE. Furthermore, some additional power can be gained by measuring hippocampal change at more time-points and by considering whether patients carry ApoE4.
Several limitations ought to be mentioned: First, mild cognitive impairment and normal subjects have not been followed long enough to determine the incidence of incipient Alzheimer's disease in each respective group. The rates and accelerations for these groups may therefore be biased toward higher values if many subjects with preclinical Alzheimer's disease were included. Furthermore, clinical criteria are always imperfect and some subjects in a mild stage of disease can be difficult to classify. However, the ADNI
utilizes the rigorous diagnostic criteria of therapeutic trials which are one of the best available. The chance of a major bias of the results due to clinical misdiagnosis is small. Second, white matter lesions, an indication for cerebrovascular disease, were not accounted for. As previous studies showed white matter lesions can be associated with hippocampal atrophy (Fein et al.
), our results could at least in part be related to vascular disease. Third, since the algorithm to measure the hippocampus utilizes information from the rest of the brain, changes in other brain regions as well as image artefacts could have mimicked hippocampal variations. Completely manual measurements of the hippocampus might therefore lead to a different outcome.
In conclusion, the demonstration of hippocampal loss in mild cognitive impairment and Alzheimer's disease patients over 6 months and accelerated loss over 12 months illustrates the power of MRI to track morphological brain changes over time in a large multisite setting. Furthermore, our finding of higher hippocampal loss in presence of ApoE4 and reduced CSFAβ1–42 supports the concept that increased hippocampal loss is an indicator of Alzheimer's disease pathology and a potential marker to assess therapeutic interventions in Alzheimer's disease.