Diagnostic effectiveness of quantitative [¹⁸F]flutemetamol PET imaging for detection of fibrillar amyloid β using cortical biopsy histopathology as the standard of truth in subjects with idiopathic normal pressure hydrocephalus

Despite the early description of Alzheimer’s disease (AD) in 1906 [1], the first set of globally accepted criteria for the clinical diagnosis of AD was only established in 1984 by the National Institute of Neurological and Communicative Disorders and Stroke and the Alzheimer’s Disease and Related Disorders Association (NINCDS-ADRDA) [2]. Prior to these criteria, a suspected clinical diagnosis of AD could only be confirmed with certainty by the post mortem observation of ‘senile plaques’ and neurofibrillary tangles in the brain, much as they were described to appear by Alois Alzheimer himself as he visualized them with low power microscopy using Bielschowsky silver stain [1]. While diagnosis in living patients represents a significant advance, when applied by expert clinicians the 1984 clinical criteria are reported to have only approximately 80% positive predictive value (PPV) and 60% negative predictive value (NPV) for a pathology-confirmed diagnosis [3]. In general, by clinical criteria, sensitivity (proportion of true positive results) increases with more permissive clinical criteria, and specificity (proportion of true negative results) increases with more restrictive clinical criteria. However, the opposite is true for neuropathologic criteria.

Appropriately, a proposal to develop and validate new biomarkers for AD was included in the 2011 National Institute on Aging-Alzheimer’s Association (NIA-AA) updated diagnostic recommendations [4–7]. Underpinning the 2011 recommendations is the recognition that the progression of AD biomarkers over time likely follows an ordered temporal sequence, and the Aβ biomarkers (e.g., low brain Aβ clearance, i.e., low cerebrospinal fluid [CSF] Aβ₄₂; and tracer retention on amyloid positron emission tomography [PET] imaging) are representative of upstream events [5].

While normally CSF flows out of the brain ventricles, idiopathic normal pressure hydrocephalus (iNPH) may be caused by CSF ‘backflow’ into the brain from the ventricles [8]. Studies of various analytes have shown that a concentration gradient exists between CSF taken from the ventricles (higher concentration) and CSF taken from the lumbar spine (lower concentration) [9–12]. Aβ oligomers self-assemble into the larger Aβ species [13], which are deposited as plaque in the brain; elevated soluble Aβ oligomers in CSF have been associated with AD, although the data are inconsistent and the culprit toxic oligomer still needs to be identified [14]. With a hypothetical leap, it may not be so surprising that AD-like Aβ pathology frequently occurs concomitantly with iNPH.

Clinically, patients with iNPH present with a classic triad of symptoms: cognitive impairment, gait abnormalities, and urinary incontinence. Untreated, iNPH is progressive [15]. The treatment for iNPH is surgical placement of a ventriculoperitoneal shunt. A small cortical biopsy from the site of shunt placement may be obtained during the procedure. Though iNPH itself may cause dementia, an identifiable neurodegenerative process often underlies the cognitive decline [16]. Concomitant AD pathology is not uncommon [17–19], up to 68% in one series [20], and may be associated with a poorer response to ventriculoperitoneal shunting [17, 18]. Biopsies from iNPH patients represent a unique opportunity to study correlations between PET amyloid imaging and pathology in living subjects.

The PET radioligand, Pittsburgh compound B (PiB), which is a neutral analog of thioflavin T (a stain which detects fibrillar amyloid) has been extensively studied both in living subjects and in autopsy tissue. PiB was designed to cross the blood-brain barrier, bind with high selectivity and nanomolar affinity to fibrillar Aβ, and clear rapidly from the brain [21]. Cortical [¹¹C]PiB uptake, as measured by PET in living subjects, correlates with fibrillar Aβ load measured subsequently by immunohistochemical stains for Aβ₄₀ and Aβ₄₂ post mortem [22–24]. Unfortunately, the short half-life of [¹¹C] (about 20 minutes) requires an on-site cyclotron for its production, limiting its use to academic medical and specialized imaging centers. Consequently, several fluorinated amyloid imaging agents have emerged to bridge this gap (F18 having a half-life of about 110 minutes): [¹⁸F]flutemetamol [25] and [¹⁸F]florbetapir [26] (Aβ diagnostics both approved by the Food and Drug Administration]) as well as [¹⁸F]florbetaben making the radiopharmaceutical available for widespread community use.

[¹¹C]PiB, [¹⁸F]florbetapir, and [¹⁸F]florbetaben have similar fibrillar Aβ binding site affinities, and can be used in a comparable manner to assess brain amyloid density [27]. In this analysis, we used [¹⁸F]flutemetamol, which differs from PiB in structure by the addition of a single F18 atom [28].

Four clinical studies (GE067-008, -009, -010, and -011, called Studies A, B, C, and D, here (and published separately as [20, 29–31], respectively)) were undertaken in iNPH patients requiring surgical shunt procedures or intracranial pressure (ICP) monitoring, to determine how well cerebral fibrillar Aβ uptake of [¹⁸F]flutemetamol as quantified by PET imaging corresponded with immunohistochemical (IHC) and histochemical (HC) estimates of amyloid burden in biopsy samples taken during these procedures. Two of the studies (one in Europe and one in the United States [US]) were retrospective studies in which the biopsy was followed by the PET scan. Two of the studies (one in the Europe and one in the US) were prospective studies, in which the order of procedures was reversed.

[¹⁸F]Flutemetamol uptake was measured quantitatively in specific brain regions including the cortical area of (prospective studies) or surrounding (retrospective studies) the biopsy site (ipsilateral site) and the site in the contralateral hemisphere that corresponded to the biopsy site (contralateral site). A composite neocortical measure of [¹⁸F]flutemetamol uptake was also calculated by averaging the uptake from frontal cortex, anterior cingulate gyrus, posterior cingulate gyrus/precuneus, lateral-temporal cortex, and parietal cortex. Aβ plaque frequency was determined and scored in biopsy specimens stained with Bielschowsky silver stain and thioflavin S. The percentage of grey-matter area occupied by plaque was also assessed following IHC for the monoclonal antibody 4G8. Finally, based on all available stains/slides, an overall pathology assessment was rendered of the Aβ load in biopsy tissue grey matter. These 4 pathology endpoints served as the standard of truth (SoT) in comparisons with the [¹⁸F]flutemetamol PET data.

In a pooled analysis of these 4 studies as previously reported, for specific parameters, [¹⁸F]flutemetamol uptake in ipsilateral and contralateral sites as well as the composite cerebral cortical measure of [¹⁸F]flutemetamol uptake were significantly correlated with Aβ plaque burden [32]. This confirmed similar findings in an autopsy study using [¹⁸F]florbetapir [33].

The standard uptake value ratio (SUVR) is a quantitative measure of tracer uptake in a brain, normalized for the mean uptake in a reference region. Pooling data from Studies A, B, C, and D, using biopsy pathology as the SoT and cerebellar grey matter as the quantitative PET reference region in one set of data and pons in another, here we set out to find (1) which pair(s) of PET SUVR and pathology SoT endpoints matched best, (2) whether quantitative measures of [¹⁸F]flutemetamol PET were better for predicting the pathology outcome than majority [¹⁸F]flutemetamol PET visual-based image read, and (3) whether there was a better match between PET image findings in retrospective vs. prospective studies.

Given that the SUVR of the composite VOI does not reflect the overall uptake level in the composite VOI (i.e., the mean value is not corrected for VOI size and all regions are treated as though they were of equal importance) it is perhaps unexpected that the composite measure for SUVR referenced to the cerebellum for all 4 studies combined was a better match with biopsy pathology findings for 2 pathology SoTs (overall pathology and Bielschowsky silver) than was the SUVR (any combination) from the biopsy region itself or its contralateral homolog. However, when the retrospective and prospective studies were analyzed separately, for retrospective studies, composite-cerebellum was generally the best match for the SoTs, while for the prospective studies, contralateral-pons was generally the best match for the SoTs. This could have been due to the general study limitation that there was a clear cut difference between the retrospective and prospective studies in the size of the VOI assessed. We acknowledge the general limitation that our data were not analyzed without the partial volume correction.

The largest ROC AUC for the combination of SUVR type and pathology SoT (4 studies combined) was for composite-cerebellum SUVR/overall pathology SoT, the most inclusive measure for imaging and the most comprehensive measure for pathology, respectively. ROC AUCs for composite-cerebellum/and contralateral-pons/Bielschowsky silver were nearly as large.

Based on the sum of sensitivity and specificity, BIE was a better tool for predicting the Bielschowsky pathology findings than was quantitation with SUVR (in 6 of the 6 SUVR/SoT pairs) but very similar to both quantitation with SUVR and BIE using the overall pathology judgment (i.e., high sensitivity and high specificity); the converse was true for thioflavin S (less sensitive and similarly specific) and 4G8 (substantially less sensitive and usually more specific). Overall, based on the sum of sensitivity and specificity, BIE and quantitation with SUVR appeared to be tied.

No particular advantage to using one SUVR reference region over the other (i.e., cerebellum or pons) was apparent. However, for all SUVR types, SUVR cut-off criteria using pons as the reference region were unexpectedly found to be less than 1. From the observed optimal cut-off values of approximately 0.4 (pons as reference region) and approximately 1.2 (cerebellum as reference region) in this set of iNPH patients, one would predict that on average, much more Aβ was deposited in the pons than in the cerebellum. Whereas in AD, enough Aβ for detection with tracer does not appear in the cerebellum or pons until Thal Phase 5 [37]. The significance of this finding will need to be clarified in future research.

We should note that a post-hoc Spearman correlation analysis was performed between cognitive status (Mini-Mental State Examination [MMSE] scores, for which individual subject values were previously published [32]), and measures for all of the pathologic markers of amyloid, including Bielschowsky, 4G8, thioflavin S, and overall pathology judgment. No statistically significant correlation was found, consistent with previous results that found no significant correlation between an imaging marker for amyloid load ([¹¹C]PIB binding potential with PET) and MMSE [46].

While neocortical pathological Aβ changes in affected iNPH patients are similar to those in AD, and it is tempting to apply findings from one condition to the other, it is important to recognize that our understanding of the molecular biology of amyloid is incomplete [47]. We biopsied only cortical samples for use as our SoT. While a reduction in brain tissue is characteristic of both AD and iNPH, the reduction is due to atrophy in AD, whereas the sulci are less enlarged relative to the degree of ventricular dilatation in iNPH. However, the fact that the overall pathology SoT fared so well in the results may indicate that neocortical areas where amyloid was present in our iNPH patients might have been mimicking where Aβ pathological changes are seen in AD, i.e., widespread vs. focal. Others confirm the validity of diagnosis from a single cerebral biopsy sample [48].

To put the 4G8 antibody data into context, variable processing of the membrane-bound amyloid precursor protein (APP) produces Aβ and other APP fragments including p3 [47]. Aβ is a 39-43 amino-acid peptide; the 4G8 antibody detects amino acids 17-23 [49] or 17-24 [37]. The morphology of deposits in AD brain detected with 4G8 have been described with photomicrographs in exquisite detail as fleecy (fine fibrillar [49]), lake-like, and subpial band-like amyloid; diffuse and cored plaques; and core-only (burnt-out [50]) plaques, as well as white matter plaques; some of which types are further sub-divided into neuritic and non-neuritic [51]. From neocortex (Thal Phase 1), Aβ as detected using silver stain and 4G8 appears progressively over time in specific connected brain structures in an ordered anterograde sequence [51, 37].

Ikonomovic et al. demonstrated co-localization in an AD brain of 6-CN-PiB (thioflavin T derivative) and thioflavin S to cored plaques and core-only plaques in temporal cortex. Thioflavin S was not as selective and also stained numerous neurofibrillary tangles in this region, whereas 6-CN-PiB detected only an occasional isolated tangle [22].

Silver staining and IHC for Aβ provide complementary information. Silver, which stains both plaques and tangles (which appear in neuritic plaques), renders stable and reproducible results [52] and is recommended for quantitation of neuritic plaques [38, 43]. However, ‘argyrophilia’ is not a homogeneous phenomenon with respect to amyloid [51], and other subtle lesion-dependent variations between silver methods have been described [52].

While the newest NIA-AA criteria for the neuropathologic assessment of AD [38] also recommend IHC for ‘Aβ score’ (not plaque count) and refer to Thal [37], the NIA-AA criteria stop short of specific Aβ IHC reagent recommendations. Technical uncertainties with respect to Aβ IHC standardization may preclude its use as a standard for neuropathologic diagnosis, at least when deposits are quantified as with histological diagnostic criteria for AD [52]. For the neuropathologic changes of AD, the new criteria recognize the qualitative importance of the location and morphology of Aβ deposits as separate and different from quantification by neuritic plaque counts as in the CERAD scheme [38] or, by extension, as in the percentage area positivity we measured with 4G8 in this study of iNPH cortical biopsies.

Spillantini et al. reported finding “a substantial increase in the number of stained structures and the intensity of staining” with 4G8 after pre-treatment with formic acid [53], although data relative to no pre-treatment were not shown. Shankar et al. believe that formic acid solubilizes insoluble Aβ plaque cores isolated from human AD brains and releases their constituent dimers and monomers [54].

Plaques, composed largely of fibrillar Aβ, are dynamic structures and likely act as local reservoirs of smaller diffusable Aβ oligomers thought to be in equilibrium with plaques [55]. Disruption of the electrophysiological and microanatomical correlates of memory formation (i.e., inhibition of long-term potentiation, facilitation of long-term depression, reduced dendritic spine density, synapse loss) are associated with the smaller, soluble Aβ species [55], which are suspected of eventually [56] tipping the gain-of-toxic-function [47] past steady-state in favor of Aβ conglomeration and towards subsequent downstream events. Importantly, this paradigm provides that plaques may be present in cognitively (i.e., phenotypically) normal individuals. Clinically identifying patients on this cusp (cognitively normal/abnormal) and monitoring their disease progression (i.e., locating amyloid plaques in living patients) is of tremendous relevance and urgent importance to the testing of drugs that exploit the molecular biology of Aβ for the treatment of AD.

In this study, we found that while the numbers of FN and FP results were low (lowest for the overall pathology SoT), for each of the 3 stains separately, the proportions of patients with FP was substantially greater than the proportions of patients with FN. Interestingly, in our experience based on a large autopsy study of [¹⁸F]flutemetamol (68 brains, 43 Aβ positive, 25 Aβ negative) [report in process], cases with sparse to moderate diffuse and neuritic cortical plaques (Bielschowsky silver stain) may lead to FN or FP PET readings (similar to those discussed in a recent review [57]). This ‘cusp’ phenomenon may reflect the (currently unsettled) rest of the story of the molecular biology of Aβ, what triggers and perpetuates its neuronal anterograde trek [51, 37] and under what conditions (including the inflammatory component of the pathology outside the scope of this report), and what determines the point at which the recognizable phenotype becomes apparent (i.e., the presumed point of irreversible functional damage).

The 4G8 antibody has been used before in association with amyloid PET ligand uptake [33, 27]. The 4G8 component of our analysis can be interpreted both in an isolated manner and in context. We questioned the acceptability of pooling the 4G8 data from Study A (obtained using an antibody dilution of 1:500) with that from Studies B, C, and D (obtained using an antibody dilution of 1:100). We noticed that Clark et al. used 4G8 at a dilution of 1:2000 and we think reported similar correlations (the use of Bonferroni ρ was unclear) to what we previously reported for our pooled data (using Pearson’s r) [32] in their [¹⁸F]florbetapir PET autopsy study in subjects with and without AD or other age-related pathologies [58, 33]. And, Thal [49, 37] and Braak [59] used a dilution of 1:5000. In a systematic inter-laboratory study, Alafuzoff et al. recommended that in order to achieve reproducible results with 4G8 IHC, a dichotomized assessment (they suggested present/absent) rather than quantification should be applied [60]. Our raw data showed that in Study A 4/7 biopsy samples had 4G8 present; in Studies B, C, and D, respectively, 6/9, 13/15, and 16/16 biopsy samples had 4G8 present. Given the small sample sizes, it is impossible to know how different ‘4G8 present/absent’ in our studies truly was, and therefore whether or not pooling of data on this basis is indicated.

Our data showed that the 4G8 SoT did not perform as well as the other SoTs in terms of the sum of sensitivity and specificity; it showed good sensitivity and poor specificity (indicating too many FP). Tissue preparation may have been the reason for this, in that the threshold for abnormal may not have been optimized. If the threshold for abnormal were set lower than the 2.5% we used, sensitivity would have decreased and specificity would have increased. The fact that the 4G8 dichotomous assessment did not match the overall pathology SoT in many cases (i.e., 4G8 resulted in the most overall pathology misclassifications) supports the interpretation that the selected threshold was not optimal.

In a separate analysis (Study C) of iNPH patients included in this pooled analysis and who also underwent [¹¹C]PiB PET imaging, ipsilateral, contralateral, and composite SUVRs for both [¹⁸F]flutemetamol and [¹¹C]PiB correlated significantly with Aβ biopsy specimen levels evaluated by 4G8, thioflavin S, and Bielschowsky silver stain [30]. Our findings using the pooled [¹⁸F]flutemetamol data in iNPH patients are also consistent with findings for [¹⁸F]florbetapir where a correlation was shown between PET brain labeling and grey matter plaque density not only by 4G8 as alluded to above, but also as measured by silver stain at autopsy in subjects with and without AD or other age-related pathologies [58, 33]. To our knowledge, no biopsy or autopsy data for [¹⁸F]florbetaben have been published yet.

Whereas our study was limited by the logistical constraints associated with the collection of comparatively primitive neuropathologic data from clinical trials at multiple sites in the setting of rapidly changing (increasingly refined) pathologic diagnostic criteria, we propose that an ideal experiment might improve upon a similar basic study design to that of Ikonomovic et al. [22] and include the following elements: multiple patients with AD and brain [¹⁸F]flutemetamol imaging near life’s end, followed by post mortem examination of adjacent sections treated with cyano-flutemetamol vs. e.g., IHC for selected Aβ epitopes and tau epitopes (plus routine conventional staining), with the Aβ and neurofibrillary tangle pathology results described according to current nomenclature for neuropathologic changes of AD [38]. Other desirable experimental elements include a selected battery of sections of brains associated with all phases of AD severity, sections thick enough to allow confocal microscopy through entire cells, collection of information with respect to individual genetic risk association factors, thorough history and timeline of medical conditions with any inflammatory component, duration of cognitive impairment, and results of recent neuropsychiatric assessments. The importance of careful archiving of tissue samples and associated patient data to test for future findings cannot be overstated. Paradoxically, owing to the relative rarity of appropriate post mortem handling and disposition for these purposes, the human brain may be as valuable an asset in death as it is in life.

The performance characteristics and diagnostic efficacy of the Aβ ligands when used alone, and more recently when used with other imaging modalities (e.g., structural MRI and fluorodeoxyglucose PET) [e.g., [42, 61] or in conjunction with the assessment of the well known AD risk variant of apolipoprotein E (ϵ4 allele) [e.g., [62], have been described in over a decade of literature. ApoE type has not yet added clinically useful diagnostic information in conjunction with imaging; however, algorithms which include multiple biomarkers have been clearly shown to increase the power of studies and reduce the number of patients required to demonstrate the statistical significance of findings [61] and are encouraged by the Food and Drug Administration [63] and the Alzheimer’s Disease Neuroimaging Initiative (http://www.nia.nih.gov/research/dn/alzheimers-disease-neuroimaging-initiative-adni), which freely shares data. While we did not pre-specify the collection any CSF biomarker data in our 4 clinical trials (and ApoE genotype was published for only 18 subjects in our pooled dataset [31]), the relationship of certain CSF biomarkers to brain biopsy findings was recently described for a large series of patients (53 iNPH, 26 AD, and 23 other) at Kuopio University in Finland; quantified biopsy Aβ load showed a negative correlation with both ventricular and lumbar CSF Aβ₄₂ while levels of Aβ₃₈ and Aβ₄₀ showed no correlation [64]. In the near future we hope to see clinical imaging with Aβ ligands combined with assessments of other recently identified AD risk genes (up to and including IHC for their protein products at autopsy) such as for specific variants of clusterin (CLU, Apo J), complement component receptor 1 (CR1), phosphatidylinositol binding clathrin assembly protein (PICALM) [61], until a more powerful algorithm is achieved that will almost certainly inform the identity of one or more useful drug targets (drugs) for the slowing, prevention, and/or arrest of this ultimately devastating pathology.

In summary, both quantitative assessment and BIE of [¹⁸F]flutemetamol images in this series of iNPH patients showed good agreement with cortical biopsy histopathology. The primary diagnostic effectiveness (as measured by ROC AUC, Youden index, and sum of sensitivity and specificity) for [¹⁸F]flutemetamol PET was best when the composite SUVR measure using cerebellum as the reference region was paired with the overall pathology SoT.