Predicting Imminent Progression to Clinically Significant Memory Decline Using Volumetric MRI and FDG PET

INTRODUCTION

One of the keys to slowing the rates of incident Alzheimer’s disease (AD) is to detect the disease early when interventions are likely to be most effective [1]. Even at the stage of mild cognitive impairment (MCI) [2], high levels of cerebral amyloid, hippocampal volume loss, and cognitive/memory impairment are already substantial [3]. Therefore, the best hope for delaying, slowing, or even preventing the onset or the progression of AD requires instituting interventions prior to the development of MCI, which in turn requires methods to identify sensitive biomarkers that precede and predict future symptomatic onset.

The present analysis examined imaging data approximately 2 years before clinical evidence of disease (baseline), with the long-term goal of developing predictive tools for clinicians using readily available neuroimaging biomarkers, such as volumetric MRI and FDG PET. Prior research suggests that brain atrophy accelerates about 2 years prior to clinical detection of MCI [4]. Neuroimaging studies predicting progression from normal to MCI/AD have varied in design, e.g., cross-sectional data [5, 6], group based comparisons [5, 7], longitudinal data with baselines greater than 2 years before clinical symptoms [8 –11] and/or reporting rate of change [4 , 12–14]. These studies suggest that progression to the clinical stages of AD can be predicted by FDG PET measured glucose metabolism reductions in posterior cingulate [8], medial temporal [11], parietal regions [11], hippocampus [10], and entorhinal cortex [9] or MRI measures of hippocampus [7 , 15], entorhinal cortex [5, 15], medial temporal lobe [7, 13], and temporal lobe thickness [6, 14]. Importantly, the time between baseline neuroimaging measures and clinically significant symptoms also varied among these studies, which may account for some of the differences in results. We therefore specifically focused on approximately 2 years before the onset of symptoms for this study. Our cohort of individuals with incident MCI is younger (early 70 s) than those in the Alzheimer’s Disease Neuroimaging Initiative (ADNI) study (late 70 s) and our study examines MRI and PET scans simultaneously at a timepoint that is early enough so emerging preclinical interventions might be more effective. While this longstanding study did not have PET or CSF biomarkers of amyloid or tau pathology, our study provides potential new insights regarding the best established brain imaging measurements of neurodegeneration in the preclinical stages of AD.

MATERIAL AND METHODS

RESULTS

Progressors included 13 APOE4 homozygotes, 3 APOE4 heterozygotes and 2 APOE4 non-carriers. APOE4-matched nonprogressors included 13 APOE4 homozygotes, 4 APOE4 heterozygotes, and 3 APOE4 non-carriers. The nonprogressor group not matched for APOE4 included 3 APOE4 homozygotes, 11 APOE4 heterozygotes, and 21 APOE4 non-carriers. The average age of aMCI/probable AD dementia diagnosis was 70.7±5.0 (range: 61 to 78 years). On average, the baseline scans for this prediction study were collected 1.8 years±0.8 (range: 1 to 4 years) prior to aMCI/AD diagnosis for the progressor group. The range of years that individuals in the progressor group had been followed since they entered the original prospective cohort study and the epoch before they developed cognitive impairment (i.e., time of baseline scan) was 2 to 14 years (average 7.6 years, SD = 4.4 years). The range of years that individuals in the nonprogressor groups had been followed after their scans was 4 to 12 years (average 6.8 years, SD = 2.3 years) in the APOE4 matched cohort and 4 to 17 years (average 8.7 years, SD = 3.7 years) in the cohort not matched for APOE4. Given our matching criteria, age, sex, and education were similar at the time of scan, as was APOE allele dose for the nonprogressor group that was matched for APOE4 allele dose. Progressors and both sets of nonprogressors also did not differ in their family history of dementia, i.e., the proportions of father, mother, or both parents. For both the APOE4-matched and non-matched groups, there were no significant differences between progressors and nonprogressors on WAIS-R subtests, Dementia Rating Scale (DRS), or Controlled Oral Word Association Test (COWAT). Significant differences were found on memory tests, as expected on the basis of receiving the amnestic MCI diagnosis— based in part on memory test decline— in the progressor group at the subsequent epoch. Group characteristics are reported in Table 1. Although group averages appeared below published norms for AVLT, individual progressors did not exhibit impaired scores on more than one memory test at this stage and appeared to be a result of normal intraindividual variability [38, 39], nor were there functional impairments. Differences in measures of memory were also evident at their first entry into the prospective cohort study even though group averages were completely normal, and there were no group differences at time of entry into study on measures of general intellect, language, working memory, mental speed, problem solving, or depression/anxiety.

Neuroimaging group differences, prespecified ROIs and the HCI

Statistical analysis of predefined ROIs confirmed that progressors had smaller volume in hippocampus, entorhinal thickness and parahippocampal thickness than both nonprogressor groups. Progressors also showed increased HCI but no significant differences from both nonprogressor groups in posterior cingulate metabolism (Fig. 1).

OPEN IN VIEWER

Fig.1

Neuroimaging group differences for a priori baseline regions of interest (18 Progessors and 20 APOE-matched Nonprogressors).

Neuroimaging group differences, voxel-based analysis findings

Compared to both groups of nonprogressors, those who progressed to incident aMCI/AD had lower (FWE corrected p < 0.05) FDG PET measured CMRgl in right and left parietal, left and right temporal, and occipital regions (Fig. 2). At uncorrected p = 0.005, glucose hypometabolism (lower CMRgl) in the progressors was found in additional brain areas, including angular, frontal and parietal regions. Results did not change when using cerebellum as the reference region instead of the global mean for FDG PET data; however, we found a greater magnitude and more wide-spread spatial extent differences between progressors and nonprogressors when using the pons as reference. MRI VBM analysis revealed that progressors had significantly lower gray matter volume compared to the nonprogressors (FWE corrected p < 0.05) in left and right hippocampal, left and right parahippocampal, left and right parietal, posterior cingulate, left and right precuneous, and left and right temporal regions (Fig. 3). At uncorrected p = 0.005, lower regional gray matter volume was found in additional brain areas, including frontal, amygdala, angular, calcarine, fusiform, lingual and caudate.

OPEN IN VIEWER

Fig.2

Lower baseline CMRgl in 18 Progressors compared to 20 APOE-matched Nonprogressors (left), and 35 Non-APOE-matched Nonprogressors (right) (voxel-based analysis, FDG PET, p = 0.005 uncorrected and without spatial clustering, with whole brain as reference).

OPEN IN VIEWER

Fig.3

Lower baseline regional grey matter in 18 Progressors compared to 20 APOE-matched Nonprogressors (left), and 35 Non-APOE-matched Nonprogressors (right) (voxel-based analysis, MRI, p = 0.005 uncorrected and without spatial clustering, with whole brain as reference).

DISCUSSION

This study utilized longitudinal clinical outcome data from a well-defined age, sex, education, and APOE4-dose balanced sub-cohort of cognitively unimpaired individuals who progressed to incident aMCI within approximately 2 years and “non-progressor” individuals who did not develop clinical symptoms (and who were observed for at least 4 years to remain asymptomatic). Our study is unique in that we directly compared the prognostic value of FDG PET and MRI in predicting imminent clinical progression for individual subjects. Statistical analysis of predefined ROIs indicated that progressors, compared to both APOE4-matched and non-matched nonprogressors, showed reduced hippocampal volume, entorhinal thickness, and parahippocampal thickness. Progressors also showed increased HCI and decreased right temporal CMRgl but no significant differences in posterior cingulate metabolism. We further established that among MRI measures of gray matter volume and cortical thickness and FDG PET regional CMRgl measures, hippocampal volume was the best predictor of progression to clinically significant decline occurring within approximately 2 years. Using a binary logistic regression classifier, hippocampal volume was predictive of clinical progression to MCI/AD for any given individual with 79% sensitivity and 78% specificity; combining FDG PET and MRI data did not improve the performance. Similarly, in the non-APOE4 matched cohort, left hippocampal volume was the single best predictor of progression and combining FDG PET and MRI did not improve the performance. Interestingly, despite our concerns about potential bias with including cognitive data in the model, cognitive measures did not out-perform hippocampal volume in predicting clinically significant memory decline.

Though we subsequently introduced amyloid PET and tau PET to this overall study, we did not have these data for this specific sub-cohort available at baseline. Given the emerging convention to characterize preclinical AD based on brain imaging or fluid biomarker evidence of amyloid, tau pathology, and neurodegenerative pathology, this additional information is likely to impact the prediction of subsequent progression to the clinical diagnosis of aMCI or AD dementia— and to clarify the extent to which of these measures specifically predict progression to those forms of clinically diagnosed aMCI and AD dementia that are associated with significant amyloid pathology [40]. However, our findings may be of particular value to those researchers and clinicians who are limited to available MRI or FDG PET data. Despite significant baseline FDG PET group differences between progressors and nonprogressors in brain regions preferentially affected by AD and the HCI, FDG PET changes probably occur earlier than hippocampal atrophy [3 , 42]. We previously showed that MRI changes occur later than FDG PET changes [43, 44]. Whether the earlier FDG PET changes reflect a progression of neuronal loss versus very early developmental metabolic reductions in APOE4 carriers, FDG PET may not have additional value at the stage of imminent clinical progression. Future studies should examine the predictive power of earlier baseline data, e.g., 10 years prior to diagnosis, and in those cases the functional imaging data and/or amyloid or tau biomarkers may contribute more information than the MRI data alone.

Our study complements and partly contradicts findings from a similar multimodal prediction study that utilized longitudinal data from the ADNI data base [11]. Ewers and colleagues found that FDG PET, not MRI, was predictive of progression to MCI/AD. The different result can be explained by several factors. First, both studies had a relatively small n (the ADNI study had 11 progressors and we had 18 progressors), which unfortunately limits generalizability and consistency of results and underscores the need to have additional studies, including this report. Second, their population was on average 10 years older (79 years for progressor and 74 years for non progressor versus ours at 69 years for progressor and 67 and 68 years for nonprogressor groups), had more men than ours (63% versus 38%), and their progressor and nonprogressor groups were not as closely matched for age and sex as ours. We had an APOE enriched sample and 2 nonprogressor groups, one matched for APOE4 and the other with significantly less APOE4 carriers, though our non-APOE matched nonprogressors were comparable to the proportion of APOE4 carriers in the ADNI study’s progressor group. Their nonprogressors were followed a minimum of 3 years, whereas ours were followed a minimum of 4 years. Additionally, their baseline was 3 to 4 years prior to progression, whereas we specifically targeted 2 years prior to progression. However, focusing on either 2 or 3 years prior to clinical progression is practical for the clinical setting. Both studies and other like them lay the groundwork for development of a clinical tool to aid in predicting imminent clinically significant decline using biomarker data such as MRI and/or FDG PET.

Our sample was enriched for APOE4 allele dose, well defined, and assessed with a wealth of additional clinical and cognitive measurements. Thus far, 8 were later diagnosed as definite AD, 6 developed probable AD, 3 remain aMCI likely due to AD, and 1 improved in memory and function at the epoch subsequent to the aMCI diagnosis but again declined to conform with an aMCI diagnosis at the most recent epoch. The one individual who improved had the largest hippocampal volume of all the progressors, consistent with our conclusion that hippocampal volume is a good predictor of imminent clinically significant decline. When the one participant who showed improvement in the epoch after MCI diagnosis was removed from the analysis, the prediction for combined right and left hippocampal volume improved to 84% specificity and 82% sensitivity; otherwise the results were unchanged. This underscores the utility of having sufficient follow-up data or other biomarker evidence to increase confidence that the groups are accurately defined. More sensitive methods to examine subregional hippocampal surface differences [45, 46] may also allow for more accurate classification of imminent progression than hippocampal volume, and we are currently investigating those methods.

Prior research, largely from cross-sectional data, supports the hypothesis that hippocampal volume is predictive of progression from MCI to AD [47 –51] at least 18 months before MCI symptom onset [3, 52]. Cross-sectional data from the Dominantly Inherited Alzheimer’s Network prospective longitudinal study [53] suggest that hippocampal atrophy can be detected 15 years before expected symptom onset. Another cross-sectional study comparing Autosomal Dominant AD mutation carrier biomarkers to noncarriers [41] indicates that CSF and PET measures of Aβ pathology 20 years before symptoms are followed by CSF measures of tau pathology and regional CMRgl decline, followed by hippocampal atrophy 6 years before symptoms, followed by clinical progression. Finally, a biomarker study [42] of late-onset sporadic AD also suggested that amyloid PET changes 17 years before, hippocampal changes 4 years before, and memory impairment at 3 years before a CDR of 1.

We used global normalization, i.e., whole brain reference, for the FDG PET analyses, based on significant findings with this method in our previous publications on cognitively unimpaired individual carrying 0, 1, or 2 copies of APOE4 allele [24 , 54], or on AD patients [26, 27]. However, after noticing the relatively poor performance of the posterior cingulate ROI, we performed post-hoc additional SPM and ROI analyses using separately the cerebellum and pons as reference regions. Although not numerically better than HCI, we indeed found better performance of posterior cingulate when using pons as the reference region. Moreover, for the voxel-wise analysis, we found greater magnitude and more widespread spatial extent differences between progressors and nonprogressors when using the pons as reference region. The results from these exploratory analyses underscore the need for further study to establish best methods for analyses of pre-clinical FDG PET biomarkers.

This study has limitations. First, all participants were selected to have a first-degree family history of dementia, so results may not generalize to those without this family history. Second, although we found similar results regardless of whether the nonprogressors were APOE4 carriers, the progressors were largely APOE4 carriers and results may not generalize to APOE4 noncarrier progressors. Third, the absence of baseline PET or CSF biomarker measurements of amyloid or tau pathology for the entire sample made us unable to use those biomarkers either in the predictive model or for further characterizing our two groups. This is of particular relevance in light of a recent study reporting that abnormal levels of amyloid and tau, independent of APOE, appear to predict subsequent cognitive decline (though with a much larger age range and younger subject group than in our study) [55]. We know that there is heterogeneity in amyloid positivity or negativity among individuals with a clinical diagnosis MCI and dementia due to possible or probable AD, so having confirmatory amyloid or tau data would have strengthened the quality of our training set. We are acquiring those measurements in an increasing number of subjects, so future studies from this cohort should be able to use such data. Fourth, the sample size of the progressor group was relatively small and may not have the power to detect all of the relevant differences between groups. Fifth, the MRI data used here were from a 1.5T scanner; greater sensitivity and specificity might be achieved with 3T MRI data, especially if hippocampal subregions are tested. We did not use MRI data to correct the partial volume effects of the FDG PET data, but we have previously found comparable results with and without such correction [54]. Sixth, our FDG PET data were from 2 different scanners. Although we adjusted for scanners in the analysis and we also found similar, yet reduced statistical significance due to smaller sample size, results in an analysis of the subset of subjects from a single scanner, we cannot exclude the possibility of the difference of the HR+ and 951/31. The latter had a limited field of view and very low spatial resolution and was decommissioned on 3/27/2004. Our design differs from others that analyzed baseline scans of individuals who subsequently did or did not develop cognitive impairment at various time points. We specifically targeted neuroimaging markers of progression to impairment within approximately 2 years, and our results must be interpreted accordingly. For example, it is likely that prediction models from several years prior to clinical progression may benefit more from combining FDG PET data with MRI data. Finally, in order to minimize Type 1 error associated with multiple regional comparisons, modality specific region selections were based on cross-sectional group differences with the assumption that these regions should also yield the best disease progression prediction. These are two different questions and potentially different brain regions are involved in static disease state differences and dynamic disease progression. It is possible that if a non-biased data driven approach were used to compare predictive value of two imaging modalities, FDG may have more added value.

The major strengths of this study include our use of a very well-defined, unique APOE4 allele dose matched and nonmatched cohorts, younger than in many other studies, in which the subsequent diagnosis of AD was likely in the majority of the clinical progressor group, consideration of FDG PET and MRI simultaneously in the prediction models, and implementation of pre-specified brain regions based on prior research from independent data sets. Even though those in the nonprogressor group may evidence disease in the future, we know that none of these individuals developed cognitive impairment for at least 4 years, and our goal was to determine imaging predictors of imminently clinically significant decline. Further research is needed to determine how far in advance brain imaging measurements, other biomarkers, cognitive measurements, APOE4 allele dose and other risk factors can be used in cognitively unimpaired adults to predict progression to the clinical stages of AD.