Abstract
Amyloid imaging is limited by an inconsistent relationship between cerebral cortex amyloid- β (Aβ) plaques and dementia. Autopsy studies suggest that Aβ plaques first appear in the cerebral cortex while subcortical plaques are present only later in the disease course. The presence of abundant plaques in both cortex and striatum is more strongly correlated with the presence of dementia than cortical Aβ plaques alone. Additionally, detection of striatal plaques may allow, for the first time, pathology-based clinical staging of AD. Striatal plaques are reportedly identifiable by amyloid imaging but the accuracy and reliability of striatal amyloid imaging has never been tested against postmortem histopathology. To determine this, we correlated the presence of histopathologically-demonstrated striatal Aβ deposits with a visually positive panel consensus decision of a positive [18F]flutemetamol striatal PET signal in 68 subjects that later came to autopsy. The sensitivity of [18F]flutemetamol PET striatal amyloid imaging, for several defined density levels of histological striatal Aβ deposits, ranged between 69% and 87% while the specificity ranged between 96% and 100%. Sensitivity increased with higher histological density thresholds while the reverse was found for specificity. In general, as compared with PET alone, PET with CT had slightly higher sensitivities but slightly lower specificities. In conclusion, amyloid imaging of the striatum with [18F]flutemetamol PET has reasonable accuracy for the detection of histologically-demonstrated striatal Aβ plaques when present at moderate or frequent densities. Amyloid imaging of the cerebral cortex and striatum together may allow for a more accurate clinicopathological diagnosis of AD and enable pathology-based clinical staging of AD.
INTRODUCTION
Amyloid imaging represents a major advance for Alzheimer’s disease (AD) research and clinical practice but staging is limited by an inconsistent relationship between cerebral cortex amyloid plaques and dementia. Recent amyloid imaging results [1–8] have confirmed longstanding neuropathological reports [9–20] that many non-demented elderly individuals have cortical amyloid plaques. Additionally, amyloid imaging cannot provide a direct estimate of the extent of neurofibrillary tangle (NFT) spread throughout the brain, and candidate NFT or tau imaging agents have not yet been validated against autopsy findings. The neuropathological diagnosis of AD, recently reformulated under the sponsorship of the National Institute on Aging and the Alzheimer’s Association (NIA-AA) requires the presence of a minimal brain load of both plaques and NFTs [21, 22].
Autopsy studies suggest that abundant striatal amyloid plaques may be mainly restricted to subjects in higher Braak NFT stages that are more likely to meet clinicopathological diagnostic criteria for AD; amyloid imaging of striatal plaques might therefore be expected to offer greater diagnostic accuracy for dementia due to AD than cortical amyloid imaging alone [23–27]. Additionally, by analogy with histopathological staging based on cerebral amyloid progression [24], striatal amyloid imaging could allow the separation of subjects that have cortically-restricted amyloid deposits from those with both cortical and striatal amyloid deposition; this would enable, for the first time, pathology-based clinical staging of AD. A small number of reports indicate that striatal amyloid plaques may be identifiable by amyloid imaging [28–33] but the accuracy and reliability of striatal amyloid imaging has never been tested against the gold standard of postmortem histopathology. This study was designed as a preliminary test of the ability of striatal PET amyloid imaging to detect histopathologically-demonstrated striatal amyloid plaques.
The agent [18F]flutemetamol [7, 34–36] has been recently approved by the US FDA and the EU EMA for the detection of brain amyloid (marketed as VIZAMYLTM by GE Healthcare). As a continuation of a Phase III clinical trial designed to test the concordance of antemortem [18F]flutemetamol positron emission tomography (PET) cortical amyloid imaging (by visual assessment) with postmortem histological amyloid presence [7], the [18F]flutemetamol PET striatal amyloid signal, together with histological amyloid-β (Aβ) staining, was evaluated in 68 subjects who came to autopsy. The results were used to obtain estimates of the sensitivity and specificity of [18F]flutemetamol PET striatal amyloid imaging for the presence of histologically-verified striatal amyloid deposits.
MATERIAL AND METHODS
Study design and participants
The initial study was a Phase III, multicenter PET study of Flutemetamol F18 injection for estimating cerebral cortex neuritic plaque density [7]. Local institutional review boards or ethics committees approved the study protocol before initiation. All subjects or their caregivers/relatives gave written informed consent before study procedures were performed. The ClinicalTrials.gov identifier is NCT01165554.
Eligible subjects were ≥55 years of age, terminally ill with a life expectancy ≤1 year, with general health adequate to undergo the study procedures. Pregnant/lactating women were excluded. Subjects were ineligible if they had known/suspected structural brain abnormalities, a contraindication for PET, a known or suspected hypersensitivity/allergy to Flutemetamol 18F injection (or any of its components), or had participated in any clinical study using an investigational agent within 30 days of signing consent.
A clinical diagnosis of AD, other dementing disorder, mild cognitive impairment, or no cognitive impairment was determined by chart review. Before PET imaging, enrolled subjects underwent diagnostic head X-ray computed Tomography (CT), unless images of research quality, obtained within 12 months of the screening visit, were available. The objective was to determine the sensitivity of blinded visual interpretations of striatal [18F]flutemetamol PET images for estimating amyloid plaque densities in the same regions.
PET imaging
Injection of Flutemetamol 18F injection was administered intravenously within 40 seconds; the dose was 185–370 MBq at the physician’s discretion, based on how long the patient could tolerate lying in the scanner. PET image data were collected over 10 minutes, starting 90 minutes after injection.
All PET images were randomized and approximately 10% of the images were duplicated and randomly combined with the other images to measure intra-reader reproducibility.
Reading methodology
Reading of the images was conducted by five readers, who underwent image-read methodology training that has formed the basis of an interactive electronic system which is now available online. The training instructed readers on recognizing anatomy in Vizamyl PET images and specified image orientation and intensity adjustments. Following this, readers were trained in techniques for reviewing the frontal and temporal lobes as well as the posterior cingulate/precuneus regions, insula and the striatum. Scrolling was emphasized and examples were presented and reviewed for trainee readers. Images were reviewed in the axial plane, except for the posterior cingulate gyrus and precuneus, where the parasagittal plane was used. For cortical assessment, a negative or normal appearance for cortical regions was characterized by a white matter pattern with the sulcal-gyral outtime apparent and with intensities tapered in the white to gray matter direction. A positive or abnormal appearance was characterized by the absence of a sulcal-gyral pattern, due to the merging of the white non amyloid-specific matter, tracer uptake and the amyloid based elevated uptake in the gray matter. The assessment of the striatum was performed by looking for a bridge in intensity between the thalamus and frontal white matter (positive) or by a clearly reduced region of intensity between the two structures (negative). This assessment was performed in an axial plane in slices adjacent to and co-planar with that defined by the anterior-posterior commissures. Examples of normal and elevated intensity are shown for the five reviewed regions in Fig. 1.
After suitable training in [18F]flutemetamol image assessments, five independent qualified readers blind to subjects’ clinical and postmortem status independently interpreted scans at a central image review center. Images of four cortical regions and of the striatum (Fig. 1) were each assessed as positive or negative: The cortical regions included standard levels of 1) frontal/anterior cingulate, 2) posterior cingulate/precuneus, 3) insula, and 4) lateral temporal lobe, while 5) the striatum was assessed at the level of the head of the caudate nucleus and putamen. The cerebral cortex and striatum regions were each classified as positive or negative based on the majority decision of the five readers. The case was considered positive if any region was positive.
After readers interpreted the PET images and locked their decisions, the images were randomized again and were re-read with CT images for anatomic guidance. Inter-reader agreement and intra-reader reproducibility were determined for the visual PET interpretations without and with CT images.
Postmortem histological procedures
Postmortem procedures were performed on 68 subjects who died during the study period and had postmortem brain examination. Immunohistochemical staining for Aβ was performed on brain tissue from subjects who died during the study and underwent brain autopsy. Postmortem histopathology analysis was performed at a central pathology laboratory. Tissue blocks were chosen to closely match the regions evaluated with [18F]flutemetamol PET. Therefore, the striatum was assessed at the level of the head of the caudate and putamen. The Aβ staining procedure utilized formic acid pretreatment and the monoclonal antibody 4G8 at 1:100 dilution (Covance, USA). Signal development employed a biotinylated secondary antibody (DakoCytomation E0354, UK) and visualization with the DABMap Kit (Ventana, USA). Striatal amyloid plaques (all plaque types, e.g., diffuse, cored, neuritic, etc., were considered together) were graded blind to clinical or neuropathological status by two neuropathologists (TGB and DT). Plaque density scores were obtained by assigning values of none, sparse, moderate and frequent, analogously to the published Consortium to Establish a Registry for Alzheimer’s Disease (CERAD) guidelines [37]. Disagreements between the two pathology-readers were resolved by alternating the decision between readers.
Brains were classified according to the level of National Institute on Aging – Alzheimer’s Association (NIA-AA) AD neuropathological change (not, low, intermediate or high) [21, 22], by the amyloid phase [24] and the Braak neurofibrillary degeneration stage [38].
Statistical analysis
The blinded visual interpretations of PET images, with or without concurrent reference to CT images for anatomical clarification, were used to calculate the sensitivity and specificity of positive imaging decisions for varying levels of Aβ plaque density. Sensitivity and specificity were determined as point estimates with exact two-sided 95% confidence intervals (CIs). Inter-reader agreement and intra-reader reproducibility in the blinded visual image assessments were determined as percentage agreement and kappa coefficient. Differences between group means were tested with unpaired, two-tailed t-tests or Mann-Whitney U-tests as appropriate.
RESULTS
Characteristics of study subjects
The clinical and neuropathological characteristics of the 68 autopsied subjects are shown in Table 1. Subjects died at a median age of 82. Imaging with [18F]flutemetamol PET was performed a mean of 3.5 months (range, 0 to 13 months) before death. Fifty-eight subjects were demented while 10 were non-demented. Of those with dementia, 30 had been clinically diagnosed with AD dementia (3 of these also had Parkinson’s disease), 19 were classified as dementia, not otherwise specified, 3 were Parkinson’s disease with dementia, 1 was corticobasal degeneration, 1 was dementia with Lewy bodies or Parkinson’s disease, 2 were classified normal pressure hydrocephalus with dementia, 1 was Pick’s disease, and 1 was vascular dementia. There were 10 non-demented subjects; two of these had Parkinson’s disease while the other 8 had no other neurological diagnosis.
In terms of AD histopathology, subjects had appreciable amyloid plaque and NFT burdens. For all subjects considered together, the median amyloid phase was 4, the median Braak neurofibrillary degeneration stage was 4, the median CERAD neuritic plaque density was moderate and the median NIA-AA neuropathological change level was intermediate.
Striatal [18F]flutemetamol PET readings
Thirty-seven subjects were classified as positive for striatal amyloid (Figs. 1–3) on [18F]flutemetamol PET striatal majority reads while 40 were classified as striatal positive using [18F]flutemetamol PET with CT (Table 2). The three additional positive cases ascertained with PET/CT were confirmed as positive with immunohistochemical Aβ staining on postmortem sections of striatum, indicating that the usage of CT may increase the sensitivity for striatal amyloid detection by PET. Subjects positive for striatal amyloid using [18F]flutemetamol PET with CT were significantly older (82.7 versus 78.1; p < 0.05) and had significantly higher amyloid phase, Braak NFT stage, CERAD cortical neuritic plaque density, and NIA-AA AD neuropathological change level, compared to subjects negative for striatal amyloid (p < 0.0001).
There were only two cases in this group of subjects where cortex and striatum were not both positive or both negative by majority assessment. These two cases both had a positive cortical signal and a negative striatal signal. All cases with a positive striatal signal also had a positive cortical signal.
Pairwise between-reader agreement was ≥90% with CT and ≥87% without CT, except for comparisons involving Reader 3 (72% –84% without CT and 78% –82% with CT); kappa scores were also lower for Reader 3. Kappa scores were higher with CT. Within-reader reproducibility ranged from 83% –100% both with and without CT images, and kappa ranges were 0.67–1.00.
Antemortem-postmortem correlations for striatal amyloid
The hypothesis was that a positive striatal amyloid image assessment by the majority of trained readers would predict, with high sensitivity and specificity, the presence of histologically-demonstrated striatal amyloid plaques at or above specified densities. In the examination of this hypothesis, different levels of histological striatal amyloid plaque densities (Fig. 4) were used as the “gold standard” or “standard of truth” as it is assumed that striatal amyloid image reading will not be as sensitive to the presence of amyloid as the histological stains. As amyloid angiopathy was only observed in 5 subjects and was present at only sparse densities, the PET signal in the striatum is assumed to be due entirely to plaques.
Inter-reader agreement (TGB and DRT) for the histopathological grading of striatal amyloid plaque density, as none, sparse, moderate or frequent, was 60/68 subjects (88.2%). For the 8 subjects for whom there was a disagreement, the decision was made by alternating between the two readers.
The sensitivity of [18F]flutemetamol PET imaging for detecting varying densities of striatal amyloid plaques ranged from 68.5 to 86.7% while the specificity ranged from 96 to 100% (Table 3) depending upon which histologic threshold was applied. Using a higher plaque density to define the true positive state resulted in higher sensitivity but lower specificity. Imaging with CT to assist with anatomical localization gave slightly higher sensitivity but slightly lower specificity.
DISCUSSION
Amyloid imaging is an excellent diagnostic tool for determining whether dementia is due to AD or not [39]. However, due to the high prevalence of amyloid plaques in non-demented elderly subjects, amyloid imaging cannot, in the absence of clinical information, distinguish between clinically relevant and irrelevant cortical amyloid plaque loads. Amyloid imaging does not indicate the extent of NFT spread throughout the brain and this also limits its usefulness, as it is widely accepted that a minimal brain load of both plaques and tangles is necessary to cause dementia [21, 22]. Autopsy studies, however, have suggested that the amount of striatal Aβ deposition might be a useful marker of the presence of clinicopathological AD as abundant striatal plaque densities occur only at later histopathological stages of AD, and largely after dementia onset [23–27].
This is the first study to evaluate the accuracy of striatal PET amyloid imaging for the detection of striatal amyloid plaques. The results indicate that [18F]flutemetamol PET imaging of the striatum has approximately 77 to 83% sensitivity and 100% specificity for the detection of striatal Aβ plaques when they are present at moderate or frequent densities. Sparse striatal plaque densities were not detected. The ability of striatal [18F]flutemetamol PET imaging to provide a reasonably accurate prediction of the presence of striatal amyloid plaques allows, for the first time, pathology-based clinical staging of AD, as subjects with a positive striatal amyloid signal will be at a more advanced stage of AD than subjects with only a positive cortical amyloid signal [24]. This may be of considerable importance for the analysis of the effects of experimental agents in clinical trials for AD, or for AD prevention, as it is likely that therapeutic agents would be more likely to be efficacious in subjects at lower rather than higher histopathological stages of AD.
In this study, there were only two subjects with a positive cortical amyloid PET signal but a negative striatal signal. This scarcity may have been due to an under-representation of non-demented subjects (10/68) and subjects with lower amyloid phases. Data from large autopsy studies with a greater representation of non-demented subjects indicate that cortical amyloid is often present in the absence of striatal amyloid, potentially allowing a practically useful clinical separation of two pathological stages of AD. Large autopsy-based studies also indicate that cortical amyloid imaging may have reasonably high sensitivity, over 90%, but low specificity, between 50% and 70%, for distinguishing demented AD subjects from non-demented subjects while striatal amyloid imaging (based on histopathology) should have similar sensitivity but much greater specificity, about 85% [26].
In conclusion, amyloid imaging of the striatum has reasonable accuracy for the detection of histologically-demonstrated striatal Aβ plaques present at moderate or frequent densities. Amyloid imaging of the cerebral cortex and striatum together may allow for a more accurate clinicopathological diagnosis of AD and for pathology-based clinical staging of AD but this will require further testing in studies that evaluate both striatal and cortical amyloid in large numbers of non-demented and demented subjects.
Footnotes
ACKNOWLEDGMENTS
GE Healthcare sponsored the original study (GE067-007), and participated in designing, monitoring, analyzing, and reporting the current study – of which GE Healthcare had oversight. TGB and DRT were consultants to GE Healthcare. MZ, AS, and CB are employees of GE Healthcare. Additionally, TGB is a consultant for Avid Radiopharmaceuticals and receives institutional funding from Avid Radiopharmaceuticals and Navidea Biopharmaceuticals. DRT received consultancies from Simon-Kucher and Partners (Germany) and Covance Laboratories (UK), received a speaker honorarium from GE Healthcare (UK) and collaborated with Novartis Pharma Basel (Switzerland).
