Post-Mortem Evaluation of Amyloid-Dopamine Terminal PET Dementia Classifications

Abstract

Clinical classification of early dementia and mild cognitive impairment is imprecise. We reported previously that molecular imaging classification of early dementia and mild cognitive impairment with dual amyloid and dopamine terminal positron emission tomography differs significantly from expert clinical classification. We now report pathologic diagnoses in a substantial subset of our previously imaged subjects. Among 36 subjects coming to autopsy, imaging classifications and pathologic diagnosis were concordant in 33 cases (κ=0.85). This approach enhanced specificity of Alzheimer disease diagnosis. The strong concordance of imaging based classifications and pathologic diagnoses suggests that this imaging approach will be useful in establishing more accurate and convenient classification biomarkers for dementia research.

Introduction

Clinical classification of dementias, particularly in early disease phases, is imprecise.¹ There are 3 common neurodegenerative dementias; Alzheimer disease (AD), Lewy body dementia (LBD), and Frontotemporal dementias (FTDs). Even expert clinical characterization does relatively poorly in differentiating AD from FTDs.² Clinical criteria for LBD possess good specificity but relatively poor sensitivity.³ Mild Cognitive Impairment (MCI), a common precursor of dementia, is a heterogeneous category associated with all major neurodegenerative pathologies and vascular etiologies. Imprecise classification of MCI and early dementia subjects is an obstacle to clinical research as heterogeneous study populations dilute power to detect effects of trial interventions or associations with potential biomarkers. The emergence of PET ligands identifying specific pathologic features of neurodegenerative disorders raises the possibility of minimally invasive characterization of MCI and early dementia subjects. We previously reported results of combined amyloid ([¹¹C]PIB) and dopamine terminal ([¹¹C]DTBZ) PET imaging in 102 MCI and early dementia subjects, demonstrating only moderate concordance (κ=0.41) between imaging based and expert clinical consensus classifications.^4,5 Our prior results raise the possibility that this imaging based approach to classification more faithfully reflects underlying pathologies than clinical characterization. We now report neuropathologic follow-up of a substantial fraction of our study subjects.

Methods

Study participants were individuals with MCI or relatively mild dementia (MMSE > 17) as described previously and enrolled in our prior imaging study from 2005 to 2009.^4,5 The purpose of the prior study was to compare amyloid-dopamine terminal PET based classification of early cognitive impairment subjects with expert clinical classification. Subjects with primary features of cognitive impairment were recruited from the University of Michigan Cognitive Disorders Clinic. Patients with primary neurological presentations involving non-cognitive domains (ataxia, parkinsonism, etc.) were excluded. Inclusion-exclusion criteria are described in prior publications; patients with possible vascular dementia (modified Hachinski score >4 or meeting NINDS-AIREN criteria or large infarcts on structural imaging) were excluded.⁴ Clinical classifications were established via expert consensus conference based on clinical and neuropsychological data accumulated at the time of visits for imaging as described previously.⁴ Enrollees agreed to follow-up autopsy. To date, 41 study participants died and autopsies were completed on 36. Autopsy results of one subject were reported previously.⁶ All autopsies were performed at the University of Michigan Health System. Neuropathology was assessed by standard methods and using standard diagnostic criteria.^7–11 The examining neuropathologists (AF-H, APL, SC-P) were blind to results of imaging studies. Thal scores of amyloid plaque density were compiled for 3 neocortical regions; mid-frontal (Brodmann’s areas [BA] 10 & 46), parietal (BA 7 & 39), and primary occipital (BA 17). Plaques were identified with Aβ immunohistochemistry (6F/3D; Leica Biosystems, 1:50). Thal scoring was available for all subjects. Regional [¹¹C]PiB binding was quantified as distribution volume ratios (DVRs) with the cerebellar gray matter as the reference region. Image based classifications established in our prior studies were used for categorical comparison with pathologic diagnoses.^4,5 Standardized DVR image datasets were classified qualitatively by an expert interpreter (KAF) familiar with the normal and pathologic distributions of these tracers and blind to all clinical and routine structural imaging data as described previously.⁴ In our prior study, use of parametric regional DVR thresholds for classification did not alter results.⁴ The unweighted Cohen’s Kappa statistic was used to estimate concordance between imaging based and pathologic classifications. Spearman’s rank-order correlation was used to compare amyloid burden assessed pathologically with the [¹¹C]PiB DVR estimates of regional amyloid burden. Sixteen subjects also underwent [¹⁸F]fluoro-deoxyglucose PET (FDG-PET) at the same time they underwent DTBZ-PiB imaging. These studies were interpreted by the same expert interpreter (KAF) blind to the clinical histories, and structural and PET imaging data.

Results

There was overall excellent concordance of imaging based classifications with neuropathologic diagnoses; κ=0.85 (95% CI = 0.69–1.0; Table 1; details of pathologic results in Supplementary Table). Regional amyloid DVRs correlated well with neuropathologic scoring of amyloid burden in the selected neocortical regions (Figure). For mid-frontal cortex, rho = 0.72; for parietal cortex, rho = 0.79; for primary occipital cortex, rho = 0.64 (all p<0.05). There were 3 cases with discordant imaging-pathologic classifications. One subject had a clinical diagnosis and imaging classification of LBD but a pathologic diagnosis of AD. α–Synuclein immunoreactive Lewy bodies were found in midbrain neurons in this subject, suggesting the presence of mixed AD-LBD pathology. The second discordant subject had marked frontal and temporal atrophy secondary to multiple small infarctions and imaging classification as FTD. The final discordant case was classified by imaging as LBD but remarkable only for the presence of TDP-43-immunoreactive neurites in the frontal cortex and hippocampal formation. This was an unusual case in that there was marked unilateral striatal loss of [¹¹C]DTBZ binding. Three cases were assessed pathologically as meeting criteria for both AD and LBD. These individuals had imaging classifications as LBD with amyloid deposition and are assessed as concordant classifications. There was excellent concordance between imaging assessments of increased amyloid burden and pathologic results; all subjects found to have moderate to high amyloid plaque burden at autopsy were classified as amyloid positive in imaging classifications.

Table 1.

Expert consensus clinical classifications, clinical data, imaging classifications, and neuropathologic diagnoses.

Clinical Consensus Classification	Age at Initial Evaluation	Age at Imaging Classification	Age at Death	Disease Duration	PiB-DTBZ Imaging Classification	Pathologic Diagnosis	Comment
LBD	75	76	80	5	LBD	AD	Midbrain Lewy Bodies
AD	59	60	66	7	AD	AD
AD	66	67	70	4	AD	AD
LBD	68	70	71	3	LBD	LBD
LBD	74	74	78	4	LBD	LBD	Subacute Right Basal Ganglia Infarct
AD	76	78	80	4	AD	AD	Subacute Left Parahippocampal Infarct
mdMCI	78	78	83	5	AD	AD
FTD	71	72	75	4	FTD	Multiple Infarcts	Frontal and Temporal Atrophy, Multiple Small Infarctions, Mild Loss of Nigral Neurons
AD	64	65	67	3	AD	AD
AD	69	71	73	4	AD	AD
naMCI	66	67	71	5	FTD	FTD	C9ORF73 Mutation
LBD	65	65	67	2	LBD	LBD + AD
LBD	75	79	86	11	LBD	LBD
FTD	71	72	76	5	AD	AD
LBD	74	74	79	5	LBD	LBD
LBD	65	66	73	8	LBD	LBD
mdMCI	65	65	66	1	AD	AD
FTD	78	80	82	4	LBD	TDP-43+	TDP-43 Immunoreactive Neurites in Frontal Cortex and Hippocampal Formation
aMCI	77	80	83	5	AD	AD
FTD	62	62	63	1	AD	AD	Sparse TDP-43 Immunoreactive Inclusions in Frontal Cortex and Dentate Gyrus
AD	76	76	80	4	AD	AD + LBD
AD	64	65	69	5	AD	AD
naMCI	62	64	67	5	LBD	AD + LBD
FTD	67	67	71	4	FTD	PSP
AD	76	81	82	6	AD	AD
AD	79	79	82	3	AD	AD
AD	85	85	89	4	AD	AD
naMCI	78	79	82	4	FTD	PSP	Remote Cerebellar Infarct
AD	66	67	72	6	LBD	LBD
aMCI	79	80	84	5	AD	AD
CBS	62	66	71	9	AD	AD
AD	55	55	59	4	AD	AD
AD	65	68	72	7	LBD	LBD + AD
LBD	76	77	83	7	AD	AD
AD	56	59	63	7	AD	AD
AD	85	88	92	7	AD	AD	Remote Midbrain Infarct
	Mean = 70.3 yrs	Mean = 71.6 yrs	Mean = 75.2 yrs	Mean = 4.9 yrs

Disease duration = interval from age at initial evaluation to age at death. AD = Alzheimer disease, LBD = Lewy body dementia, FTD = Frontotemporal dementia, MCI = Mild Cognitive Impairment, aMCI = Amnestic MCI, naMCI = Non-Amnestic MCI, mdMCI = Multidomain MCI, CBS = Corticobasal syndrome, PSP = Progressive Supranuclear Palsy, PiB-DTBZ = [¹¹C]Pittsburgh B & [¹¹C]dihydrotetrabenazine Positron Emission Tomography, TDP-43 = Transactive response DNA binding protein 43 kDa.

Figure.

Scatter plot of parietal cortex Thal scores (Autopsy Rating) versus [¹¹C]PiB DVRs. r – Spearman’s rho.

We performed a more limited comparison of combined amyloid and dopamine terminal imaging classifications, neuropathologic diagnoses, and FDG-PET classifications. Approximately thirty percent of the FDG-PET classifications differed from final neuropathologic diagnoses (Table 2). There were 3 cases where the FDG-PET classification was FTD with pathologic diagnoses of AD and 2 cases where the FDG-PET classification was AD with pathologic diagnoses of LDB. In all cases with discrepant FDG-PET classifications and neuropathologic diagnoses, combined amyloid and dopamine terminal PET imaging correctly identified the pathologic diagnosis (Table 2).

Table 2.

Expert consensus clinical classifications, PiB-DTBZ imaging classifications, neuropathologic diagnoses, and FDG-PET classifications.

Clinical Consensus Classification	Age at Initial Evaluation	Age at Imaging Classification	Age at Death	Disease Duration	PiB-DTBZ Imaging Classification	Pathologic Diagnosis	FDG-PET Classification
AD	59	60	66	7	AD	AD	AD
LBD	68	70	71	3	LBD	LBD	AD
AD	64	65	67	3	AD	AD	FTD
AD	69	71	73	4	AD	AD	AD
LBD	75	79	86	11	LBD	LBD	LBD
LBD	74	74	79	5	LBD	LBD	LBD
LBD	65	66	73	8	LBD	LBD	AD/LBD
mdMCI	65	65	66	1	AD	AD	AD
aMCI	77	80	83	5	AD	AD	FTD
AD	76	81	82	6	AD	AD	AD
AD	85	85	89	4	AD	AD	FTD
AD	66	67	72	6	LBD	LBD	AD
aMCI	79	80	84	5	AD	AD	AD
AD	65	68	72	7	LBD	LBD + AD	LBD
LBD	76	77	83	7	AD	AD	AD
AD	56	59	63	7	AD	AD	AD

Disease duration = interval from age at initial evaluation to age at death. AD = Alzheimer disease, LBD = Lewy body dementia, FTD = Frontotemporal dementia, MCI = Mild Cognitive Impairment, aMCI = Amnestic MCI, mdMCI = Multidomain MCI, PiB-DTBZ = [¹¹C]Pittsburgh B & [¹¹C]dihydrotetrabenazine Positron Emission Tomography, FDG-PET = [¹⁸F]fluorodeoxyglucose-Positron Emission Tomography. FDG-PET classification criteria: AD – temporoparietal and posterior cingulate hypometabolism; LDB - temporoparietal and posterior cingulate hypometabolism plus occipital hypometabolism; FTD – frontal, anterior temporal, and anterior cingulate hypometabolism.

Discussion

Our results indicate that classifications based on combined amyloid and dopamine terminal PET imaging correlate well with neuropathologic diagnostic classifications. Of 36 subjects studied, there were 3 cases (8.3%) where imaging based and pathologic classifications differed. In our prior studies, in contrast, ~35% of participants had discordant expert clinical consensus and imaging diagnostic classifications.^4,5 One discordant case was classified as LBD on the basis of significantly reduced striatal [¹¹C]DTBZ binding. Though not meeting pathologic criteria for LBD, this subject had nigral Lewy bodies, suggesting mixed pathology.

Our results are consistent with other recent studies. In trials of anti-amyloid therapy of clinically classified early AD subjects where participants underwent amyloid imaging, ~15% of enrolled subjects had negative amyloid imaging, excluding AD.^12,13,14 These results likely underestimate diagnostic misclassifications as ~50% of LBD cases exhibit significant amyloid burden, likely leading to misclassification of some LBD subjects as AD.¹⁵ A clinicopathologic study using the National Alzheimer’s Coordinating Center (NACC) dataset found that ~15% of classified clinically AD subjects failed pathologic criteria for AD.¹⁶ In our prior studies, a major cause of discrepant clinical and imaging classifications were subjects classified clinically as FTD but with positive amyloid imaging results suggesting AD.⁴ Our results comparing amyloid/dopamine terminal imaging and clinical diagnostic classifications are similar to those reported by Beach et al. using the NACC dataset to compare clinical and neuropathologic diagnoses.¹⁷

Our limited evaluation of FDG-PET classifications suggests that this method is less precise than combined amyloid and dopamine terminal PET imaging. These types of FDG-PET misclassifications are well described in prior literature. Disproportionate frontal amyloid deposition may give rise to frontal predominant hypometabolism.¹⁸ The canonical pattern of cerebral metabolic deficits in LDB is the pattern of temporoparietal and frontal deficits found in AD plus occipital hypometabolism, but the distinguishing occipital metabolic deficits are absent is a significant fraction of patients.¹⁹

Amyloid imaging is accepted as a useful biomarker of fibrillar amyloid deposition. The high prevalence of amyloidopathy in LBD, however, indicates that increased amyloid burden is not a unique AD biomarker. Combining amyloid imaging with a dopamine terminal marker enhances accuracy. Our study, and this approach in general, have some limitations. Our number of autopsied subjects is relatively small. Because amyloid imaging is relatively sensitive for detecting AD, and dopamine terminal imaging allows exclusion of LBD, this method is arguably best at improving identification of AD. Two of the imaging misclassifications assessed subjects as LBD were found at autopsy to have another diagnosis. This result and the existence of nigrostriatal pathology in FTD and related syndromes indicates that dopamine terminal imaging possesses good sensitivity but less specificity for detection of LBD. In amyloid negative individuals, substantial nigrostriatal terminal loss could indicate either LBD or FTD, as some FTD patients develop parkinsonism with nigrostriatal degeneration, particularly those with MAPT or GRN mutations, decreasing the specificity of this approach of accurate classification of LBD.²⁰ Identification of FTD is most problematic as our classification of FTD is based on negative imaging results – the absence of pathologic amyloid or dopamine terminal imaging changes. This may be misleading as there are multiple potential causes of cognitive impairment without amyloid or nigrostriatal pathology, for example, our case where the neuropathologic evaluation revealed multiple small infarcts instead of neurodegeneration. Positive imaging markers for tau deposition and other FTD associated pathologies would be useful additions to this imaging approach.²¹

Our results point to another problem secondary to use of the trinary classification scheme. This conventional approach is artificial in that mixed pathologies are common, though the presence of other pathologies does not confound amyloid ligand binding.²² Mixed pathologies are seen in our dataset with 3 subjects with both AD and LBD, and the discordant subject who met neuropathologic criteria for AD and had midbrain Lewy bodies. Identification of individuals with both AD and LBD is particularly difficult, both with our dual tracer approach and with conventional clinical classifications. This is an area where addition of a tau tracer may be valuable.

Our study has significant advantages. Our subjects were enrolled during relatively early disease stages, either MCI status or relatively mild dementia (MMSE >17). Prior studies correlating amyloid imaging results with neuropathology enrolled individuals with advanced dementia.^23–26 Our study population is more typical of clinical research studies and offers reassurance that prior imaging-pathologic correlation studies of amyloid imaging are relevant to earlier phases of neurodegeneration. Our study design may underestimate the utility of this multi-tracer approach. Imaging classifications were made in the absence of clinical and structural (CT or MRI) imaging information. Conversely, our clinical classifications were made without the PET results. Use of this multi-tracer PET approach in conjunction with clinical and structural imaging data would likely enhance accuracy of classifications. These methods may provide additional useful data. We showed previously that regional cerebral blood flow data derived from [¹¹C]DTBZ PET closely mimics the patterns of regional cerebral metabolism visualized with [¹⁸F]FDG-PET imaging.^27,28 Analysis of [¹¹C]DTBZ-based regional perfusion data would add a functional dimension to analysis and might further enhance classifications. Individuals, for example, with abnormal striatal [¹¹C]DTBZ ligand binding could have either LBD or FTD. These syndromes exhibit distinctive regional cerebral metabolic-perfusion deficit patterns which could be helpful in classifying LBD and FTD subjects more accurately.

We do not suggest that this approach to classification would be broadly useful in clinical practice. It is more plausible that this approach, or approaches using similar tracers or incorporating additional tracers such as a tau ligand, will be useful in clinical research. These methods may allow purer subject samples or better subject stratification, particularly for selection of AD subjects, for clinical research studies. This approach may be useful in establishing the utility of more accessible classification biomarkers. Rather than waiting years for autopsy results, this dual tracer approach or similar methods could be used as surrogates to validate more convenient classification biomarkers.