Longitudinal Positron Emission Tomography in Preventive Alzheimer's Disease Drug Trials, Critical Barriers from Imaging Science Perspective

Abstract

Recent Alzheimer's trials have recruited cognitively normal people at risk for Alzheimer's dementia. Due to the lack of clinical symptoms in normal population, conventional clinical outcome measures are not suitable for these early trials. While several groups are developing new composite cognitive tests that could serve as potential outcome measures by detecting subtle cognitive changes in normal people, there is a need for longitudinal brain imaging techniques that can correlate with temporal changes in these new tests and provide additional objective measures of neuropathological changes in brain. Positron emission tomography (PET) is a nuclear medicine imaging procedure based on the measurement of annihilation photons after positron emission from radiolabeled molecules that allow tracking of biological processes in body, including the brain. PET is a well‐established in vivo imaging modality in Alzheimer's disease diagnosis and research due to its capability of detecting abnormalities in three major hallmarks of this disease. These include (1) amyloid beta plaques; (2) neurofibrillary tau tangles and (3) decrease in neuronal activity due to loss of nerve cell connection and death. While semiquantitative PET imaging techniques are commonly used to set discrete cut‐points to stratify abnormal levels of amyloid accumulation and neurodegeneration, they are suboptimal for detecting subtle longitudinal changes. In this study, we have identified and discussed four critical barriers in conventional longitudinal PET imaging that may be particularly relevant for early Alzheimer's disease studies. These include within and across subject heterogeneity of AD‐affected brain regions, PET intensity normalization, neuronal compensations in early disease stages and cerebrovascular amyloid deposition.

Introduction

Abnormal neurodegenerative processes in Alzheimer's disease (AD) begin many years before the clinical diagnosis of dementia, a phase that refers to as preclinical (presymptomatic) AD where individuals have little or no cognitive alterations to satisfy the diagnosis criteria for AD or mild cognitive impairment (MCI) 4, 26, 28, 31, 34, 36, 53, 62, 63, 66, 76, 79, 81. The National Institute on Aging‐Alzheimer's Association workgroup (NIA‐AA) has proposed three successive stages of preclinical AD 76. While common semiquantitative FDG‐PET and amyloid beta (Aβ)‐PET methods can stratify these stages 33, 39 by use of discrete cut‐points for abnormal levels of Aβ (starting at stage 1) and neurodegeneration (starting at stage 2), there is considerably less experimental evidence about the rate of change in these biomarkers. There is a need for better understanding of the temporal progression of AD pathophysiology in preclinical phase that requires a more precise definition for testing new preventive therapeutic interventions. Recent Alzheimer's trials have recruited cognitively normal people at risk for Alzheimer's dementia 77. Be caused by the lack of clinical symptoms in normal population, conventional clinical outcome measures, that is, time‐to‐AD conversion, may not be suitable for these early trials 1. While several groups are developing new composite cognitive tests that could serve as surrogate outcome measures by detecting subtle cognitive changes in normal people 2, 15, there is a need for longitudinal brain imaging techniques that can correlate with temporal changes in these new tests and provide additional objective measures of neuropathological changes in brain. In particular, within‐individual longitudinal comparisons of imaging data may improve these evaluations by controlling for between subject variability. Several additional sources of variability need to be addressed and resolved to increase the accuracy of longitudinal measures of PET data. Previous research has addressed some of the biomedical and technical factors that could lead to variable PET imaging outcomes and made recommendations for controlling these parameters in longitudinal and cross‐sectional studies. This article gives a summarized overview on the previous work and adds new points of consideration that are relevant to longitudinal PET imaging at preclinical stages of Alzheimer's disease.

Materials and Methods

We reviewed 88 peer‐reviewed articles dating from 1985 to 2016 to prepare this study. We searched these articles either directly in relevant journals or through PubMed database. In some cases, we also used imaging data as illustrative examples to support the cited literature.

Imaging data used in the preparation of this article were obtained from the Alzheimer's Disease Neuroimaging Initiative (ADNI) database (adni.loni.usc.edu). The ADNI was launched in 2003 as a public–private partnership, led by Principal Investigator Michael W. Weiner, MD. The primary goal of ADNI has been to test whether serial magnetic resonance imaging (MRI), positron emission tomography (PET), other biological markers, and clinical and neuropsychological assessment can be combined to measure the progression of mild cognitive impairment (MCI) and early Alzheimer's disease (AD).

Critical Barriers in Longitudinal Pet Imaging

Within and across subject heterogeneity of affected brain regions in AD

FDG‐PET

While from previous research, we know that AD hypometabolic regions include mainly posterior cingulate and parieto‐temporal cortices 23, 35, 52, 70, there are also some differences among individual subjects 14, 41, 43. We refer to the study of Mosconi and colleagues 55 to provide illustrative and quantitative examples of differences in the topographic progression of hypometabolism across four cognitively normal older adults who eventually converted to AD, as verified by their post‐mortem brain autopsy. Within this small cohort, we can see regional differences in the reduction of FDG‐PET standardized uptake value ratio (SUVR) values. For example, some subjects develop initial hypometabolism in posterior cingulate cortex while others exhibit hypometabolism in temporal or parietal regions in addition to or without the posterior cingulate involvement. Some subjects show metabolic decline on both hemispheres whereas in others one side of the brain is more affected than the other side. This regional heterogeneity can complicate the evaluation of FDG‐PET changes in a longitudinal study (e.g., drug evaluation). Previous research addressed this problem by defining composite regions of interest (ROIs) that included all AD signature areas 41, 45, 50. While this approach allows monitoring the same composite ROI for every subject (treats all subjects equally), it can attenuate the SUVR rate of change by averaging radiotracer activity values from rapidly changing regions with slower or nonchanging areas within the composite ROI. This will increase the variance of the SUVR mean value and reduce the longitudinal effect size.

Aβ‐PET

Previous research 21, 22, 53, 68 has found that certain brain regions, such as the frontal, cingulate gyrus, precuneus and lateral parieto‐temporal areas, show higher amyloid radiotracer uptake than other regions. Similar to the FDG‐PET's composite ROI concept, many Aβ‐PET studies calculate the mean SUVR value over these regions (sometimes over the entire neocortex), which is commonly referred to as the global cortical uptake. The global uptake has been used to classify subjects as amyloid positive or negative 33, 39 or to test the treatment efficacy of drugs 57, 64, 71. Similar to FDG‐PET's AD‐signature ROI, the cortical Aβ‐PET SUVR mean values are associated with large standard errors 74. In a single subject, the SUVR standard deviation indicates voxel intensity variations across the global ROI. Even with the hypothetical presence of uniform radiotracer uptake, the voxel intensities would be subject to some variability that is characteristic of the PET scanner's noise performance. In Aβ‐PET images, the radiotracer uptake is not uniform. There are additional voxel intensity variations due to specific (presence of classical Aβ plaques), nonspecific (retention in healthy tissue) and pseudo‐specific (e.g., presence of cerebral amyloid angiopathy) 46 bindings. In a group of subjects, there is an additional across‐subject variability in spatial uptake profiles due to variabilities in the Aβ patterns 85, even for those with similar total amyloid burden, as noted by Braak and Braak 4. We used Figure 1 to demonstrate regional differences in amyloid‐PET uptake profile in two cognitively normal ADNI subjects (both amyloid positive).

Figure 1.

Two cognitively normal ADNI subjects imaged with [11C]PiB PET: We used 4 axial slices of their PET images (locations shown in corresponding MRI) to show differences in regional radiotracers uptake profiles.

Intensity normalization of PET images

While the “gold standard” for PET image quantitation necessitates full kinetic modeling with dynamic acquisition and arterial input function 56, 61, the semiquantitative SUVR metric is frequently used as a substitute method in clinical settings because it does not require long dynamic scans (subject to patient motion and discomfort) or the need to measure the arterial input function (invasive). The SUVR method calculates the ratio of the voxel intensity to a reference region intensity to correct for nonspecific radiotracer uptake and enables comparisons between different scans and subjects.

FDG‐PET

Different local and global normalization schemes are utilized in FDG‐PET for brain studies and the selection of the optimal method has been debated 6, 16, 32, 51, 54. Ideally, metabolism in the reference region should be unaffected by AD. However, several non‐AD conditions prevalent in elderly populations, such as brain injuries or vascular diseases could also alter the FDG uptake in a reference region 7, 42, 58. Given that approximately 30% of elderly people have silent infarcts without clinical manifestations 47, alterations in reference region metabolism are expected to occur in this aged population. We used Figure 2 to illustrate how longitudinal changes of regional FDG‐PET SUVR values are affected by the reference region selection. This example shows longitudinal SUVR trajectories of two regions (posterior cingulate and occipital) in a cognitively normal ADNI subject. The SUVR trajectories are obtained by applying two different normalization schemes (pontine and cerebellar) to show how the selection of a reference region can change these trajectories. Global normalization, where cerebral global mean is used to calculate the uptake ratio, has also found wide applications in FDG‐PET studies of Alzheimer's disease. While the underlying rationale is that the global fluctuation in brain metabolism is not due to the changes produced by AD, local metabolic changes could contribute to the global signal variations, thus affecting this proportional scaling at both severe and early stages of the disease 84.

Figure 2.

Mean FDG‐PET SUVR values in a subject's posterior cingulate (left) and occipital lobe (right). For a given region, the cerebellar and pontine normalizations result in different trajectories (black and green) whereas the trajectories of two different regions normalized by the same reference region also look similar, indicating potential impact of reference region selection on SUVR trajectories.

Aβ‐PET

Similar to FDG‐PET, semiquantitative SUVR values are used in clinical Aβ‐PET image analyses with radiotracers, such as [¹¹C]PiB and [¹⁸F]Florbetapir 8, 29, 48. As these radiotracers target predominately the classic cored and neuritic Aβ plaques, which are not frequently present in the cerebellum 11, 30, 37, the whole cerebellum or the cerebellar gray matter have served as reference regions for the majority of Aβ‐PET studies to date. However, recent research 5, 10, 18, 44, 80, 83, including our own study 75, have raised concerns about the reliability of the cerebellar normalization for longitudinal [¹⁸F]Florbetapir PET. It is not quite clear why the cerebellar normalization could be associated with increased variability of SUVR measures. Previous quantitative human Aβ‐PET studies with [¹¹C]PiB 13, 17, 24, 40, 48, 61, [¹⁸F]AZD4694 67, [¹⁸F]Florbetaben 3 and [¹⁸F]Florbetapir 82 have clearly demonstrated low cerebellar retention in both AD and controls. It has been hypothesized that the axial location of the cerebellum, which increases the likelihood of scattered coincident events, and its shift between serial scans (Figure 3) may induce additional variability in cerebellar SUV 72. Clinical PET data undergo validated attenuation and scatter corrections to ensure quantitative accuracy across the field of view 86, 87. If the cerebellar axial position causes variable longitudinal measures, then it should be carefully evaluated with experimental studies to help improve the quality control of PET scans for brain studies. Pons/brainstem is another reference region that also falls often within the edges of the scanner's axial field‐of‐view, but appears to be a more stable reference region than cerebellum 75. For comparison, Landau and colleagues 44 found that using a combination of multiple reference regions, including the subcortical white matter and cerebellum and pons, into a composite reference region provides longitudinal changes that are biologically more plausible than trajectories obtained from cerebellar normalization alone. The reference region selection would be particularly important for monitoring earliest changes in Aβ‐PET when the uptake ratio between target region and reference region is small, thus, more susceptible to potential variations.

Figure 3.

Two FDG‐PET images acquired from an ADNI subject with slight shift in axial head position between baseline and follow‐up.

Neuronal compensations in preclinical AD

There is an emerging evidence based on both FDG‐PET 12 and functional MRI 19 that supports the existence of neural compensations in elderly people. The presence of positive correlation between metabolism and amyloid accumulation in early stages of the disease may indicate that the increased metabolism is a compensatory mechanism in the setting of initial AD pathology and a marker of impending neurodegeneration 12. Therefore, longitudinal PET analyses in preclinical AD should account for both abnormal increase and decrease of FDG‐PET signals in different brain regions, thereby providing new ways of assessing disease progression in early stages where neuronal compensations are more likely present.

Cerebrovascular amyloid deposition

Previous studies with [¹¹C]PiB have detected elevated tracer uptake due to the contribution of cerebral amyloid angiopathy (CAA) in addition to the parenchymal amyloid 38, 46. CAA involves the deposition of β‐amyloid in the media and adventitia of small and mid‐sized cerebral arteries and veins 59. Under normal conditions, interstitial fluid and solutes drain from the brain parenchyma into cervical lymph nodes along basement membranes of small arteries, veins and capillaries by reverse transport, which is stimulated by the pulsatile flow in these vessels. The pathogenesis of CAA results from impaired clearance of β‐amyloid peptides from the perivascular basement membranes, resulting in accumulation of these peptides in the cerebral vessels 20. CAA can be easily differentiated from parenchymal amyloid plaques in post‐mortem tissue. However, in vivo image analytical methods have not as yet been developed for evaluating the development and relative contribution of CAA and parenchymal amyloid deposition to overall amyloid burden in AD brains studied longitudinally. CAA lesions can be found in 80%–85% of post‐mortem brains of patients with AD and Down's Syndrome, and extensive CAA is seen in approximately half of AD patients 9. It is possible that both CAA and parenchymal amyloid contribute to the clinical AD progression 49. Additionally, APOE e2 and e4 alleles confer increased risk for CAA, as well as for AD 25, 60. Thus, CAA lesions may present an important factor in the pathogenesis of general amyloid deposition in the brain, and in the development of senile plaque formation. The ability to detect CAA in vivo, utilizing sophisticated image analysis of Aβ‐PET scans in individuals with mild neurocognitive impairment or normal individuals with subjective memory complaints (SMC), may improve our understanding of the contribution of CAA to overall brain amyloid deposition. Different AD drugs, that is, immunotherapies, may even increase the CAA, thus, preventing the complete removal of Aβ from brain 69. Current in vivo Aβ‐PET imaging techniques cannot differentiate between parenchymal and vascular amyloid accumulations. Therefore, it is important to find new ways to evaluate the progression of these two co‐occurring pathologies during the course of Alzheimer's disease and in response to treatments.

Discussion

While early AD drug trials hold the promise of slowing abnormal disease processes before the onset of irreversible brain damage, the evaluation of their efficacy can be challenging due to the absence of clinical symptoms in tested subjects. Longitudinal imaging studies are becoming increasingly important to make objective assessments of abnormal changes in brain. Both Aβ‐PET and FDG‐PET are used in AD drug trials. Current longitudinal FDG‐PET findings correlate well with clinical outcomes and cognitive tests of subjects with Alzheimer's dementia and mild cognitive impairment 45, 73, 78. There is a need for imaging studies that can find similar associations between PET measures and subtle cognitive alterations in normal\preclinical AD population. This is a challenging task because longitudinal imaging measures are subject to different sources of variability. The Influence of technical and biological factors on Aβ‐PET and FDG‐PET imaging in Alzheimer's disease has been discussed previously 17, 27, 72. This review gives a summarized overview on the previous work and added new points of consideration that are relevant to longitudinal PET imaging at preclinical stages of Alzheimer's disease.

Some of the FDG‐PET related critical barriers were addressed in our previous studies where we developed a new technique for FDG‐PET analysis, the regional FDG‐PET time correlation coefficient (rFTC) 73 that measures similarities (correlations) between subject's serial scans. The correlation analysis can be performed on clinically acquired PET data (short static PET scans) and the correlation decline can be used as a marker of metabolic changes occurring anywhere in the brain. Therefore, this method provides a trajectory of metabolic changes using a single variable without the restriction of monitoring FDG changes in a single region or averaging FDG activities from multiple regions into a composite activity value. The correlation calculation does not require normalized PET images and the decline is triggered by both increases and decreases in brain metabolism, thus, providing a unique way of assessing the disease progression at the earliest stages where neuronal compensations can occur as early adapting mechanisms to the pathology. In comparison to region‐based analyses, voxel‐based methods do not require composite ROIs, and they can be applied on FDG‐PET 52 and Aβ‐PET 88. However, these techniques require the provision of means and standard deviations from a normal reference population (heterogeneity in reference population) to calculate the corresponding z‐scores of the brain voxels. Finding a reference population for preclinical AD studies where test subjects are cognitively normal can be challenging. Also, voxel‐based methods are typically performed on smoothed PET images that are spatially transformed into a template space. While these preprocessing steps enable cross‐sectional comparisons as well as enhance the uniformity of PET images for qualitative assessments, their utility in longitudinal within‐subject comparisons is debatable due to potential loss of image information 72. Regional analyses can be calculated in subject's native space. By selecting specific regions, they may provide more sensitive measures than global cortical averages 65, 85. Our group has also developed an alternative method to regional and voxel‐based methods that enables monitoring different subjects with different topographic Aβ‐PET radiotracer uptake profiles without averaging the voxel values (regional analysis) or requiring a reference population (voxel‐based analysis). This method 74 is based on two‐point correlation functions that capture longitudinal changes in Aβ‐PET by detecting subtle changes in spatial radiotracer uptake patterns that are referred to as increased clustering or flocculence.