Tau Subtypes of Alzheimer’s Disease Determined in vivo Using Flortaucipir PET Imaging

Abstract

At autopsy, individuals with Alzheimer’s disease (AD) exhibit heterogeneity in the distribution of neurofibrillary tangles in neocortical and hippocampal regions. Subtypes of AD, defined using an algorithm based on the relative number of tangle counts in these regions, have been proposed—hippocampal sparing (relative sparing of the hippocampus but high cortical load), limbic predominant (high hippocampal load but lower load in association cortices), and typical (balanced neurofibrillary tangles counts in the hippocampus and association cortices) AD—and shown to be associated with distinct antemortem clinical phenotypes. The ability to distinguish these AD subtypes from the more typical tau signature in vivo could have important implications for clinical research. Flortaucipir positron emission tomography (PET) images acquired from 45 amyloid-positive participants, defined clinically as mild cognitive impairment or AD, aged 50–92 years, 56% female, and estimated to be Braak V-VI based on their PET pattern of tau pathology, were studied. By translating the neuropathologic algorithm to flortaucipir PET scans, patterns of tau pathology consistent with autopsy findings, and with a similar prevalence, were identified in vivo. 6/45 (13%) participants were identified as hippocampal sparing and 6/45 (13%) as limbic predominant AD subtypes. Hippocampal sparing participants were significantly younger than those assigned to the other two subtypes. Worse performance on delayed recall was associated with increased hippocampal tau signal, and worse performance on the trail making test B-A was associated with lower values of the hippocampus to cortex ratio. Prospective studies can further validate the flortaucipir SUVR cut-points and the phenotype of the corresponding AD subtypes.

Keywords

Hippocampal sparing limbic predominant subtype tau

INTRODUCTION

The presence of neurofibrillary tangles (NFT) composed of abnormal tau, alongside amyloid-β plaques, represents the defining central nervous system (CNS) pathology in Alzheimer’s disease (AD) [1]. Neuropathologic studies have suggested that tau pathology in AD spreads in a stereotypic sequence, beginning in the anteromedial temporal lobe (entorhinal cortex and hippocampus) before involving lateral and posterior temporal regions, areas of the association cortices and, finally, primary cortices. These observations have been codified in staging schemes that serve to index the degree of advancement of tau pathology at autopsy [2, 3]. However, the average or typical patterns of tau pathology that are often used as a conceptual reference belie heterogeneity across participants in the relative neuroanatomic density of NFT counts. In particular, variations in the amount of NFT in association cortices versus hippocampal regions have been noted and used as the basis for defining AD subtypes, namely hippocampal sparing (HpSp) AD and limbic predominant (LP) AD, in contrast to a more typical AD presentation [4, 5]. The HpSp AD group was defined as presenting with abnormally low NFT counts in the hippocampus but high counts in the cortex, whereas the LP AD subtype had the opposite signature—high counts in hippocampus but lower counts in the cortex. Interestingly, these subtypes are associated with distinct antemortem clinical phenotypes, with the HpSp AD group being younger and declining more rapidly [4].

The recent advent of positron emission tomography (PET) radiotracers enabling NFT to be imaged in vivo has opened the possibility of measuring individual patterns of tau pathology in the living patient [6 –11]. Initial clinical studies have shown that tau PET images correspond, at the group level, with expected distributions of tau pathology in AD [7, 12] and related disorders including logopenic variant primary progressive aphasia and posterior cortical atrophy [9]. However, a substantial degree of variability in the distribution and intensity of the tau PET signal across individuals has also been noted, especially in more advanced stage disease [7 , 12]. In AD, some of this variability may be explained in terms of the heterogeneity of AD subtypes described above. The ability to distinguish these subtypes from the more typical AD signature in living participants and earlier in the course of disease could have important implications for clinical research, enabling a biologically-driven means of segregating participants for research and clinical management.

The primary aims of the present study were: 1) to determine whether the AD subtype classification algorithm developed for use with NFT counts on neuropathology could be applied to regional tau PET standardized uptake value ratios (SUVR) derived from flortaucipir images; 2) to determine whether the AD tau subtypes were observed at a comparable frequency to the published neuropathology study [4]; and 3) to determine whether the algorithm recapitulates expected tau patterns and associated clinical phenotypes observed in the large neuropathology study [4].

In a recent study using a similar algorithmic approach on volumetric MRI scans to define AD subtypes [13], we found that a larger hippocampal-to-cortical volume ratio, reflecting increased cortical relative to hippocampal atrophy, was associated with poorer baseline executive function, but not memory. Based on the association between tau deposition and atrophy [14, 15], we hypothesized that we would observe an association between smaller hippocampal-to-cortical tau ratios, reflecting increased cortical relative to hippocampal tau deposition, and poorer executive function. Therefore, our secondary aims were to explore the relationship between the cortical and hippocampal tau PET SUVR values and their ratio, as continuous variables, to measures of cognition and brain amyloid load, and to determine potential cut-points enabling a subtype to be assigned prospectively to individual scans in future studies.

MATERIALS AND METHODS

Clinical sample

Two hundred and twenty-three study participants or authorized representatives gave informed consent to participate in a phase II trial of flortaucipir (NCT 02016560). The protocol was approved by the relevant institutional review boards and procedures were in accordance with the ethical standards on human investigation of the institutions and the principles of the declaration of Helsinki. For the purposes of the current analysis, we restricted the sample to participants that were clinically diagnosed according to the study protocol procedures with mild cognitive impairment (age≥50, Mini-Mental State Exam (MMSE)≥24 and consistent with NIA-AA guidelines [16]), or probable or possible AD (age≥50, MMSE≥10 and consistent with NIA-AA guidelines [17]), that were determined to be amyloid positive based on a florbetapir PET scan and a florbetapir global SUVR > 1.1 [18], and that were estimated to have an in vivo estimated Braak stage of V or VI [7]. These criteria were designed to match those of the neuropathology reference study [4], and ensure both confirmation of elevated amyloid and that tau pathology has spread sufficiently to make the investigation of the association cortex signal meaningful. Participants with clinical presentations suggestive of non-AD pathologies were excluded from this clinical trial. The resulting sample for analysis in the current study comprised forty-five participants (see Supplementary Figure 1).

Clinical and cognitive assessments

Cognitive assessments included measures of global cognitive status, the MMSE and the Alzheimer’s Disease Assessment Scale for cognition (eleven item version, ADAS-Cog11), along with tests designed to assess more specific cognitive domains. Because we were interested in potential differential presentations relating to memory and executive function, we also examined the Wechsler logical memory delayed recall (story A) as a probe of memory function and trail making test B-A (Trails B-A) as a probe of executive function.

Image acquisition

The florbetapir scan comprised a ten-minute, two 5-min frames, whole brain acquisition from 50–60 min after injection of ca. 10 mCi florbetapir. The flortaucipir scan comprised four 5-min whole brain scans from 80–100 min following injection of ca. 10 mCi of flortaucipir and was acquired on a separate day than the florbetapir visit. All PET data were acquired on either General Electric, Philips, or Siemens scanners and reconstructed using scanner-specific protocols with the following parameter selections: iterative reconstruction algorithm (FORE, OSEM, or RAMLA), 3–6 iterations, 16–33 subsets, 2.0–2.67 mm×2.0–2.67 mm×2.0–4.25 mm voxel size. Post-reconstruction Gaussian filters of 3–5 mm were applied (except for Philips scanners where the ‘Normal’ or ‘Sharp’ relaxation parameter setting was used). Detector response modelling was not used. Time of flight reconstruction was used at one site. Three dimensional T1-weighted structural magnetic resonance imaging (MRI) scans were also acquired from each individual on either a 1.5T or 3T scanner and consisted of sagittal 3D MPRAGE (1.5T Siemens [TR/TE = 2400/3.5 ms] = 24.2%; 3T Siemens [TR/TE = 2300/2.98 ms] = 38.6%), sagittal 3D T1-TFE (1.5T Philips [TR/TE = 8.5/4 ms] = 5.4%; 3T Philips [TR/TE = 6.8/3.1 ms] = 8.5%), or sagittal 3D IR-FSPGR (1.5T General Electric [TR/TE = 9.1–9.6/3.9–4 ms] = 14.3%; 3T General Electric [TR/TE = 7.3/3 ms] = 9%) sequences with 1.2 mm thick slices (no gap) and a 0.9375–1.25×0.9375–1.25 mm² in-plane resolution.

Flortaucipir image processing

Individual study participants’ flortaucipir (formerly [¹⁸F]-AV-1451) image sequences (4×5 min) were motion corrected, averaged and corrected for injected dose and weight of study participant to create a single static standardized uptake value image. A correction was also applied to compensate for any deviations from the prescribed 80-min post-injection scan start, due to the tracer not reaching steady state condition during the scanning period [19, 20]. This correction comprised a linear interpolation of the image intensity values proportional to the difference between the actual scan start time and 80 min.

These images were spatially normalized to Montreal Neurological Institute (MNI) space by rigid body co-registration to the 3D T1 MRI scan from the same individual and study time point, calculating the spatial transform mapping the MRI scan to the MNI template and applying the same transformation to the flortaucipir image. Spatial normalization was calculated using SPM8 with nonlinear normalization and medium regularization. Images were converted into SUVR units using the cerebellar crus as a reference region.

Participants were staged for neuroanatomic distribution of tau load according to a recently published method [7] based on neuropathologic Braak staging [2]. Briefly, this method is based on sampling the 3D flortaucipir PET images in small regions of interest (ROIs) in the anterior temporal and occipital lobes that correspond closely to the slices specified in [2], and classification rules that match as accurately as possible the 6-stage operationalized neuropathologic staging scheme proposed by Braak et al. [2]. Participants with in vivo estimated Braak stage V or VI were retained, to exclude early stage participants that could present as LP AD simply due to lack of cortical involvement of tau.

Regions of interest for the subtype definition were chosen based on the Harvard-Oxford atlas to closely follow regions sampled in the reference neuropathology study [4]. Cortical regions were the superior temporal gyrus, lateral parietal lobe (angular gyrus) and lateral frontal lobe (mid-frontal gyrus). The hippocampal region (HP) was restricted to the anteriormost portion (head) to avoid potential confounding contribution from the choroid plexus (see Supplementary Figure 2). For each participant, the cortical composite value (CTX) was calculated as the average of the six individual cortical ROI values across both hemispheres, and the hippocampal value was calculated as the average of the two ROI values from each hemisphere.

When computing the average SUVR from each region, only voxels lying both within the region of interest and within the grey matter of the study participants were used. A grey matter mask for each individual was calculated from the individual grey matter tissue class probability map generated from the 3D T1 MR image as part of the SPM8 spatial normalization procedure by thresholding at 25%. The grey matter masking minimized bias in the SUVR signal due to atrophy, especially in the hippocampus (see Supplementary Figure 3).

Computation of AD subtypes

AD subtypes were defined following the Murray algorithm [4], except with regional flortaucipir SUVR values used instead of NFT counts (Supplementary Figure 4). In stage 1 of the algorithm, the ratio of the hippocampal to cortex SUVR values was split at the 25th and 75th percentiles of the distribution across the study participants. Participants whose HP:CTX ratio was < 25% were assigned to the HpSp AD category, while those whose HP:CTX ratio was > 75% were assigned to LP AD. In stage 2 of the algorithm, the individual HP and CTX SUVR values were considered separately and compared to the overall median value of each. Participants who were assigned to the HpSp AD category in stage 1 remained HpSp AD only if their HP SUVR was less than the median HP value and if their CTX SUVR was greater than the median CTX value. Similarly, participants who were assigned to the LP AD category in stage 1 remained LP AD only if their HP SUVR was greater than the median HP value and if their CTX SUVR was less than the median value.

Volumetric MRI analyses

Individual estimates of hippocampal volume and ventricular volume were quantified from the 3D T1 volumetric MRI images using a spatiotemporal regularization method [21]. Left and right hemisphere values were averaged for each individual. FSL/SIENAx was used to obtain a skull-constrained normalization scaling factor to correct for head size [22]. The individual volumes were multiplied with their respective scaling factor to produce a volume normalized for head size.

Statistical analysis

All statistical computations were carried out using the R statistical package (v. 3.3.1) [23]. Sex, AD subtype status (i.e., HpSp, typical, and LP), and genotype (APOE ɛ4+ and MAPT H1/H1) were treated as independent, categorical variables. Age, PET SUVR (i.e., HP tau, CTX tau, HP:CTX tau, and global amyloid SUVR), volumetric MRI, cognitive performance measures (MMSE, ADAS-Cog11_, memory (delayed recall) and executive function (Trails B-A)) were treated as continuous variables. There was an average time delay between florbetapir and flortaucipir scans of 3.7 days (range = 0–34), between florbetapir and cognitive scores of 34.6 to 39.2 days, and between flortaucipir and cognitive scores of 21.4 to 26.1 days, depending on the particular cognitive measure (see Supplementary Table 1). Categorical variables (i.e., genotype and sex) in each AD subtype were compared using Fisher’s exact test. Continuous variables were compared between AD subtypes using a one-way analysis of variance (ANOVA) and a Tukey’s multiple comparison test. Correlations between continuous variables were performed using Pearson’s product-moment correlation coefficient. To assess the relationship between each flortaucipir ROI SUVR metric (i.e., HP, CTX, and HP:CTX) and cognitive performance, four backward elimination multivariate linear regression models were built with cognitive measure (i.e., MMSE, ADAS-Cog11_, memory (delayed recall), and executive function (Trails B-A)) as the dependent variable, and HP, CTX, and HP:CTX SUVR as independent variables. Because of the known influence of age and sex on cognition, age and sex were also added to the models (Cognitive Score = Intercept + HP:CTX ratio + HP + CTX + Age + Sex + ɛ (where ɛ represents a residual unexplained error)). The statistical significance for all tests was set at a level of 0.05 (2-sided). For continuous variables, data are presented as mean±SD.

RESULTS

Study participant sample

The demographic, genetic, and clinical profiles of the overall cohort are summarized in Table 1. Of the 45 participants qualifying for the subtype analysis, 23 were clinically diagnosed as MCI and 22 as AD. The MCI participants had MMSE scores between 24 and 30 (inclusive), with an average of 26.7, and their age ranged from 50 to 92 years (inclusive), with an average of 72.9. The AD participants had MMSE scores between 12 and 28 (inclusive), with an average of 21.8, and their age ranged from 54 to 84 years (inclusive), with an average of 72. Five of the AD participants had MMSE scores lower than 20: two had a score of 18, two had a score of 13 and one had a score of 12. Fisher’s exact tests revealed no overall significant differences in APOE ɛ4+ (p = 0.876) or MAPT H1H1 (p = 0.090) genotypes between subtypes.

Table 1

Demographic, clinical, and genetic profiles for the full study sample and Alzheimer’s disease subtypes

Characteristics	Overall	Alzheimer’s disease Subtypes			p¹
		Hippocampal sparing	Typical	Limbic predominant
N	45	6 (13%)	33 (73%)	6 (13%)	–
Age (y)²	72.5±9.5	61.3±8.5	74.1±9.1	74.5±5.2	0.005, 0.030, 0.995
% Female	56%	100%	45%	67%	0.022, 0.4075, 0.4545
% APOEɛ4 + ³	69.8%	66.7%	67.7%	83.3%	0.876
% MAPT H1/H1⁴	48.7%	66.7%	37%	83.3%	0.090
MMSE²	24.3±4.2	22.5±5.9	24.5±4.2	25±2.5	0.527
ADAS-Cog11²	17.3±7.8	19±15.2	17.0±6.7	17.2±3.7	0.860
Delayed recall²	4.1±3.6	5.4±2.9	4.4±3.8	1.7±2.0	0.172
Trails B–A²	119±71.4	151±36	117±71	101±95	0.503
Amyloid SUVR²	1.51±0.16	1.59±0.19	1.49±0.16	1.48±0.16	0.369

¹p-values: Female and genotype based upon Fisher’s exact test; all others unbalanced ANOVA overall p-value if not-significant, if significant (p < 0.05) then only post hoc Tukey-Kramer test (adjustment for multiple comparisons). p-values: Hippocampal sparing Alzheimer’s disease versus typical Alzheimer’s disease, Hippocampal sparing Alzheimer’s disease versus limbic predominant Alzheimer’s disease, typical Alzheimer’s disease versus limbic predominant Alzheimer’s disease. ²Continuous variables shown as mean±SD. ³Due to missing values N’s for APOE ɛ4+ are 43, 6, 31, 6. ɛ2/ɛ3 = 2; ɛ3/ɛ3 = 11; ɛ3/ɛ4 = 19; ɛ4/ɛ4 = 11; NA = 2. ⁴Due to missing values N’s for MAPT H1/H1 are 39, 6, 27, 6. H1/H1 = 19; H2/H1 = 15; H2/H2 = 5; NA = 6.

AD subtypes based on in vivo flortaucipir PET images

The distributions of SUVR values and the correlations between each component ROI are shown in Fig. 1A. ROI-mean SUVR values from the three individual cortical regions were highly correlated with each other (0.61 < R² < 0.76) but relatively weakly with the hippocampus (0.08 < R² < 0.18). It was thus considered reasonable to combine the three cortical ROIs into a single representative composite value. A joint histogram contour plot indicates that, in general, the participants classified as typical AD had roughly proportional SUVR values in the hippocampus and cortex (lying along the diagonal), whereas participants with SUVR values disproportionately higher in either the hippocampus or cortex were assigned to either the HpSp AD or LP AD subtype (Fig. 1B). These distributions are qualitatively consistent with what is expected from the algorithm since subject classification is based upon their regional tau distributions. The SUVR cut-points derived from this sample and used in the classification algorithm were HP:CTX = 0.702 and 1.035 (25th and 75th percentiles, stage 1) and HP = 1.406 and CTX = 1.646 (50th percentiles, stage 2) (Supplementary Figure 4). The algorithm assigned 6 participants (13%) to each of the HpSp AD and LP AD groups, with the remaining 33 (74%) identified as typical AD. The HpSp, typical AD and LP AD groups had mean HP and CTX SUVR values of 1.25 and 2.18, 1.45 and 1.75, and 1.63 and 1.42, respectively. Individual values in each region are shown in Supplementary Figure 5. This presentation was consistent with those found in the original neuropathology study (Fig. 2), although a higher inter-subtype range of tracer binding was observed in the cortical regions than in the hippocampus, whereas in the original neuropathology study a higher range of NFT counts was observed in the hippocampus. The whole-brain subtype-mean SUVR maps illustrate a pattern of stronger cortical binding throughout the association cortices in the typical AD and HpSp AD groups relative to the LP AD group. All three groups showed relative sparing of the primary sensory cortices, consistent with those regions being affected in the latest stages of tau spread. Individual SUVR maps illustrate the range of tau anatomical distribution within each subtype (Fig. 3).

Fig.1

Correlations between standardized uptake value ratio (SUVR) values in regions sampled. A) The signal in the three individual association cortex regions are highly inter-correlated (0.61 < R² < 0.76), but more weakly correlated with the signal in the hippocampus (0.08 < R² < 0.18). B) Two-dimensional contour plot of SUVR values in cortex and hippocampus for each individual, color-coded by subtype.

Fig.2

Comparison of histological and PET imaging presentations of the three subtypes. A) Representative slides stained for tau neurofibrillary tangles in the hippocampus and cortex from participants classified as hippocampal sparing, typical Alzheimer’s disease or limbic predominant [4] (left) and a plot of subtype-mean values of NFT counts in the hippocampus and cortex from ref. [4] (right). B) Average flortaucipir standardized uptake value ratio (SUVR) images from each of the subtypes identified in the present study (left) and a plot of subtype-mean values of SUVR in the hippocampus and cortical regions (shown as mean±standard error) (right).

Fig.3

Thumbnail individual standardized uptake value ratio (SUVR) maps for all participants assessed in the current study. Participants are grouped by subtype, with hippocampal sparing Alzheimer’s disease participants in the box at the top left, and limbic predominant Alzheimer’s disease participants in the box at the bottom right. Within each subtype, participants are ordered by increasing age, progressing vertically from top-left to bottom-right.

Participants in the HpSp AD category were younger (average 61.3 years) than those in both the typical AD (p = 0.005) and LP AD (p = 0.03) categories (average age 74.1 and 74.5 years, respectively) (Table 1). The frequency of female participants was higher in the HpSp AD category than typical AD (p = 0.022), and intermediate in the LP AD group. The scales of global cognition (MMSE and ADAS-Cog11), memory (delayed recall) and executive function (Trails B-A) did not differ significantly across the categories.

There was no significant difference in florbetapir global cortical SUVR across the subtypes (ANOVA across subtypes, p = 0.37), although the HpSp AD group exhibited a nominally higher value (mean = 1.59, in comparison with 1.49 in the typical AD group and 1.48 in the LP AD group). Of the volumetric MRI (vMRI) metrics, neither ventricular volume nor hippocampal volume were significantly different across subtypes, although hippocampal volume exhibited a trend relationship (p = 0.08) with hippocampal volume in the HpSp AD group (3817 mm³) being larger than in the typical AD (3391 mm³) and LP AD (3155 mm³) groups.

Continuous variable relationships

Across the whole sample, the HP:CTX ratio was strongly dependent on the age of the participants in the sample, with higher values of the ratio being associated with older age (R² = 0.42, p < 0.001) (Fig. 4A). This relationship was driven by a strong negative relationship between the CTX signal and age (R² = 0.53, p < 0.001)—that is, higher cortical tau binding was associated with younger age—whereas the HP signal evidenced no significant relationship with age (R² = 0.04, p = 0.19) (Fig. 4B, C).

Fig.4

Age dependence of (A) the HP:CTX ratio, (B) the hippocampus, and (C) the cortex SUVR values independently. HP, hippocampus; CTX, cortex; SUVR, standardized uptake value ratio.

The backward elimination regression analysis revealed a significant negative relationship between delayed recall and HP signal (R² = 0.14, p = 0.013), reflecting a decrease in delayed recall scores as hippocampal tau levels increase, with the optimal model identified as score = intercept + HP + ɛ (where ɛ represents residual unexplained error); and a significant negative relationship between Trails B-A and the HP:CTX ratio (R² = 0.13, p = 0.017) reflecting an increase in Trails B-A scores with decreases in the HP:CTX ratio, with the optimal model identified as score = intercept + HP:CTX ratio + ɛ (Fig. 5). We found no evidence of a sex or age effect in the final models. No significant association between MMSE or ADAS-Cog11 and the HP or CTX values, or the HP:CTX ratio, was found. Normality was evaluated and no data transformation was required [24].

Fig.5

Association between (A) delayed recall score (lower = worse performance) and hippocampus SUVR value, and (B) Trails B-A score (higher = worse performance) and the HP:CTX ratio. HP, hippocampus; CTX, cortex; SUVR, standardized uptake value ratio.

The HP:CTX ratio (p = 0.038) but not the HP (0.53) or CTX (0.11) values individually, was significantly associated with global cortical florbetapir SUVR. Lower values of the HP:CTX ratio were associated with higher florbetapir SUVR. Of the vMRI metrics, hippocampal volume was significantly negatively associated with HP flortaucipir SUVR (R² = 0.09, p = 0.05) and with the HP:CTX ratio (R² = 0.15, p = 0.01), but was not significantly associated with CTX SUVR. Ventricular volume was not significantly associated with HP or CTX SUVR or with the HP:CTX ratio.

DISCUSSION

Our main findings were 1) that an algorithm designed to classify autopsy-confirmed AD participants into disease subtypes based on tau neuropathology can be applied to segregate a living MCI/AD population into subtypes with a similar presentation, frequency and age profile using flortaucipir PET imaging; 2) that the age association of the subtypes was driven by neocortical and not hippocampal tau SUVR values; and 3) that while cognitive performance on a delayed recall task was associated with hippocampal tau load, performance on the executive function task Trails B-A was associated with the ratio of SUVR values in the hippocampus and cortex. Moreover, 4) we determined flortaucipir SUVR cut-points that can be prospectively applied to classify individuals into AD subtypes in other studies.

The characteristics of the subtypes determined from PET imaging are consistent with those identified in the original neuropathology study [4] in several respects. We found a frequency of 13% each for HpSp AD and LP AD, similar to the neuropathologically-determined 11% HpSp AD and 14% LP AD. The HpSp AD subtype was significantly younger than the other two subtypes in our study, with a mean age at scan of 61 years compared to 74 years (typical AD) and 75 years (LP AD), in comparison with the reported mean ages at onset of clinical symptoms in the three subtypes of 63, 69, and 76 years, respectively [4]. Overall, these findings are highly consistent, despite the substantially smaller sample size in the present study (45 versus 889). We note, however, that the typical AD subtype determined in the present study had a mean age close to that observed in the LP AD subtype, whereas in the larger neuropathology sample the mean age associated with typical AD at symptom onset was younger and intermediate between those in the HpSp AD and LP AD subtypes. There was no significant difference in amyloid burden across the three subtypes, although the mean value was nominally higher (SUVR difference of 0.1) in the HpSp subtype than in the other two. The relationship between cortical amyloid burden and tau subtype should be further investigated in larger samples, to assess whether there is a meaningful association.

Although the subtypes did not differ significantly in either the global or domain-specific cognitive tests in the present study, we found a significant association between worse performance on the delayed recall task and increased hippocampal flortaucipir SUVR. We also found a significant association between worse performance on Trails B-A, selected as a task indicating executive function, and smaller values of the HP:CTX ratio, reflecting increased cortical relative to hippocampal flortaucipir SUVR. These findings are consistent with the key role of the hippocampus in memory recall, and of the fronto-parietal association cortices in executive tasks, and with the notion that tau pathology is a key precursor to anatomically co-localized neuronal dysfunction and neurodegeneration. In other words, these observations corroborate the notion that higher levels of tau pathology in brain regions and networks subserving specific cognitive domains correspond to worse cognitive performance in those domains. They also align with the observation in [4] that the HpSp AD subtype presented with a higher frequency of atypical (i.e., non-amnestic) cognitive dysfunction.

The lack of significant associations between MMSE or ADAS-Cog11 and flortaucipir SUVR, and between florbetapir SUVR and cortical flortaucipir SUVR is in contrast to previously published results on the full dataset [24] that showed significant correlations between flortaucipir SUVR and both MMSE (R = –0.42, p < 0.0001) and ADAS-Cog11 (R = 0.45, p < 0.0001), and between flortaucipir and florbetapir SUVR (R = 0.64, p < 0.0001). This may reflect the fact that the present analysis only included a smaller sub-population of amyloid-positive subjects.

The sex distribution across the thee subtypes did not recapitulate that observed in [4]. We found that 100% of the HpSp AD subtype was female, followed by 67% in the LP AD subtype and 45% in the typical AD subtype, whereas in the neuropathology study the proportion of women increased from 37% in the HpSp AD group to 55% in the typical AD group to 69% in the LP AD group. Our observations are likely limited in generalizability due to the very small sample size (only six participants in each of the HpSp AD and LP AD groups). Sex profiles of PET imaging-determined subtypes should be further investigated in larger studies.

APOE ɛ4+ and MAPT H1H1 genotypes were not significantly different between subtypes. This is consistent with Murray et al. [4] for the APOE ɛ4+ genotype; however, the authors observed a significantly higher proportion of MAPT H1H1 genotype in LP AD (70%) than in HpSp AD (46%; p = 0.011) groups. In our study we did not find an overall significant difference in MAPT H1H1 genotype between subtypes, possibly because of the small number of subjects in our sample (hippocampal sparing AD = 6, typical AD = 27, limbic predominant AD = 6). However, nominally, we found a higher proportion of MAPT H1H1 genotype in LP AD (83.3%) than in HpSp AD (66.7%) groups.

Several characteristics of the HpSp AD subtype are consistent with features associated with (sporadic) early onset AD, typically defined as participants younger than 65 years old. In the current study, 8 of the 45 participants were younger than 65 years old at the time of scan, of whom 4 were classified as HpSp AD and 4 as typical AD; 4/6 of the HpSp AD participants were younger than 65. The more prominent cortical pathology associated with the HpSp AD subtype has also been identified as a feature of early onset AD in studies of brain atrophy, in contrast to a more hippocampal atrophy presentation in late onset AD [25 –27]. Moreover, early onset AD participants are usually characterized by greater deficits in non-memory domains such as visuospatial processing, attention, and language as well as executive function. We observed a significant association between lower values of the HP:CTX ratio (consistent with increased cortical, relative to hippocampal, pathology) and poorer performance on the executive function assessment. A more cortically-dominant presentation of tau pathology and associated neurodegeneration may thus be a factor underlying the clinical presentation of early onset AD, which was recently demonstrated in a large autopsied AD series [28]. Given the strong bias toward younger age in the HpSp AD subtype, the traditional definition of early onset AD based on an age cut-off of 65 years may in effect be a practical approximation to an underlying HpSp presentation of AD.

This notion is also reinforced by recent imaging findings with clinical subtypes of AD pathology that are usually younger at onset but characterized as distinct syndromes, such as posterior cortical atrophy and logopenic variant primary progressive aphasia. Participants with these conditions have been observed to have overlapping tau PET and brain atrophy patterns that preferentially affect brain regions corresponding to the clinical phenotype—the posterior, especially visual, cortices in the case of PPA and the left temporoparietal junction in the case of logopenic variant primary progressive aphasia [9 , 29]. The present study excluded these distinct syndromes, which may explain some of the differences observed between the clinically heterogeneous neuropathology cohort [4] and the present study.

The association between decreased hippocampal volume and increased hippocampal tau SUVR is consistent with the idea that the presence of tau pathology is closely linked with colocalized neurodegeneration as evidenced by brain atrophy. We note that by correcting for the grey matter content of the hippocampal PET ROI, it is unlikely that this relationship is merely a consequence of signal dilution due to hippocampal atrophy per se. However, to avoid the potential confounding contribution of occasional strong flortaucipir binding to the choroid plexus [30], the nature of which is not yet fully understood, we restricted our hippocampal PET ROI to the anterior head of the hippocampus only. The hippocampal volume measures from vMRI reflect, in contrast, the whole hippocampus. This may partially explain the relatively weak relationship (adjusted R² = 0.073).

This study has several limitations. Of particular note is the small sample size, which makes it difficult to make firm conclusions from our observations on the relative frequency of sex and genetic markers across the three subtypes; neither behaved in the monotonic fashion expected from the much larger neuropathology study previously reported. We also acknowledge that the subtype definition algorithm, while informed by neuropathologic observations and robustly characterized in large postmortem tissue studies [4, 5], is based on a cut-point selection scheme that is dependent on the specific study sample. Further studies are thus needed to prospectively test the cut-points identified in the present study in independent cohorts. The degree to which the subtypes imposed by the present algorithm agree with natural groupings in flortaucipir MCI and AD cohorts also remains to be determined and could be more robustly investigated in larger cohorts. The lower inter-subtype range of tracer binding in the hippocampus, relative to cortical regions, is in contrast to the original neuropathology findings showing a higher range of NFT counts in the hippocampus. This suggests that PET may be relatively less sensitive to hippocampal tau than neuropathology. Although the reasons for this remains to be fully elucidated, we can propose several possible explanations for this finding. First, the hippocampus is a small structure and approaches the natural resolution limit of PET scans (typically ∼0.5 cm, depending on scanner and image reconstruction options) at which point partial volume effects can artefactually reduce the apparent signal magnitude. Second, in the studied population, the hippocampus is likely atrophied to some degree, reducing its size further. Third, the possibility of off-target binding of flortaucipir has been noted in the choroid plexus. For this reason, we sought to minimize the impact of any choroid plexus signal by restricting our hippocampal ROI to the anteriormost portion of the hippocampal head. This, however, resulted in sampling a different portion of the hippocampus than was the case in the original neuropathology study [4], where the posterior hippocampus was examined. A further possible explanation is that if the tau tangles in the hippocampus are qualitatively different from those in the neocortex and this is reflected in a different relative number of flortaucipir binding sites, then this could result in a different apparent PET signal in the hippocampus relative to the cortex, for cases where the number of neuropathologically countable tangles is the same.

In conclusion, AD subtypes based on neuropathologic distributions of tau pathology in the hippocampus and association cortices can be identified in living MCI/AD participants using flortaucipir PET imaging. Smaller values of the HP:CTX ratio were associated with worse performance on the Trails B-A, a probe of executive function, whereas higher values of hippocampal SUVR were associated with worse performance on delayed recall, a probe of memory. Regional SUVR and HP:CTX ratio cut-points were determined and can be evaluated in prospective studies.

Footnotes

ACKNOWLEDGMENTS

Data acquisition and analysis was funded by Eli Lilly and Company. We thank Drs. Michael J Pontecorvo (Avid Radiopharmaceuticals, Philadelphia, PA, USA) and Emily C. Collins (Eli Lilly and Company, Indianapolis, IN, USA) for critically reviewing the manuscript and their constructive comments.

Authors’ disclosures available online ().

The supplementary material is available in the electronic version of this article: .

References

Scheltens

, Blennow

, Breteler

MMB

, de Strooper

, Frisoni

, Salloway

, Van der Flier

(2016) Alzheimer’s disease. Lancet 388, 505–517.

Braak

, Alafuzoff

, Arzberger

, Kretzschmar

, Del Tredici

(2006) Staging of Alzheimer disease-associated neurofibrillary pathology using paraffin sections and immunocytochemistry. Acta Neuropathol 112, 389–404.

Braak

, Braak

(1991) Neuropathological stageing of Alzheimer-related changes. Acta Neuropathol 82, 239–259.

Murray

, Graff-Radford

, Ross

, Petersen

, Duara

, Dickson

(2011) Neuropathologically defined subtypes of Alzheimer’s disease with distinct clinical characteristics: A retrospective study. Lancet Neurol 10, 785–796.

Janocko

, Brodersen

, Soto-Ortolaza

, Ross

, Liesinger

, Duara

, Graff-Radford

, Dickson

, Murray

(2012) Neuropathologically defined subtypes of Alzheimer’s disease differ significantly from neurofibrillary tangle-predominant dementia. Acta Neuropathol 124, 681–692.

Okamura

, Furumoto

, Fodero-Tavoletti

, Mulligan

, Harada

, Yates

, Pejoska

, Kudo

, Masters

, Yanai

, Rowe

, Villemagne

(2014) Non-invasive assessment of Alzheimer’s disease neurofibrillary pathology using 18F-THK5105 PET. Brain 137, 1762–1771.

Schwarz

, Yu

, Miller

, Shcherbinin

, Dickson

, Navitsky

, Joshi

, Devous

, Mintun

(2016) Regional profiles of the candidate tau PET ligand18F-AV-1451 recapitulate key features of Braak histopathological stages. Brain 139, 1539–1550.

Chiotis

, Saint-Aubert

, Savitcheva

, Jelic

, Andersen

, Jonasson

, Eriksson

, Lubberink

, Almkvist

, Wall

, Antoni

, Nordberg

(2016) Imaging in vivo tau pathology in Alzheimer’s disease with THK5317 PET in a multimodal paradigm. Eur J Nucl Med Mol Imaging 43, 1686–1699.

Ossenkoppele

, Schonhaut

, Scholl

, Lockhart

, Ayakta

, Baker

, O’Neil

, Janabi

, Lazaris

, Cantwell

, Vogel

, Santos

, Miller

, Bettcher

, Vossel

, Kramer

, Gorno-Tempini

, Miller

, Jagust

, Rabinovici

(2016) Tau PET patterns mirror clinical and neuroanatomical variability in Alzheimer’s disease. Brain 139, 1551–1567.

10.

Villemagne

, Okamura

(2014) In vivo tau imaging: Obstacles and progress. Alzheimers Dement 10, S254–264.

11.

Okamura

, Harada

, Furumoto

, Arai

, Yanai

, Kudo

(2014) Tau PET imaging in Alzheimer’s disease. Curr Neurol Neurosci Rep 14, 500.

12.

Schöll

, Lockhart Samuel

, Schonhaut Daniel

, O’Neil James

, Janabi

, Ossenkoppele

, Baker Suzanne

, Vogel Jacob

, Faria

, Schwimmer Henry

, Rabinovici Gil

, Jagust William

(2016) PET imaging of tau deposition in the aging human brain. Neuron 89, 971–982.

13.

Risacher

, Anderson

, Charil

, Castelluccio

, Shcherbinin

, Saykin

, Schwarz

, Alzheimer’s Disease Neuroimaging Initiative (2017) Alzheimer disease brain atrophy subtypes are associated with cognition and rate of decline. Neurology 89, 2176–2186.

14.

Gordon

, McCullough

, Mishra

, Blazey

, Su

, Christensen

, Dincer

, Jackson

, Hornbeck

, Morris

, Ances

, Benzinger

TLS

(2018) Cross-sectional and longitudinal atrophy is preferentially associated with tau rather than amyloid beta positron emission tomography pathology. Alzheimers Dement (Amst) 10, 245–252.

15.

Xia

, Makaretz

, Caso

, McGinnis

, Gomperts

, Sepulcre

, Gomez-Isla

, Hyman

, Schultz

, Vasdev

, Johnson

, Dickerson

(2017) Association of in vivo [18F]AV-1451 tau PET imaging results with cortical atrophy and symptoms in typical and atypical Alzheimer disease. JAMA Neurol 74, 427–436.

16.

Albert

, DeKosky

, Dickson

, Dubois

, Feldman

, Fox

, Gamst

, Holtzman

, Jagust

, Petersen

, Snyder

, Carrillo

, Thies

, Phelps

(2011) The diagnosis of mild cognitive impairment due to Alzheimer’s disease: Recommendations from the National Institute on Aging-Alzheimer’s Association workgroups on diagnostic guidelines for Alzheimer’s disease. Alzheimers Dement 7, 270–279.

17.

McKhann

, Knopman

, Chertkow

, Hyman

, Jack

Jr. , Kawas

, Klunk

, Koroshetz

, Manly

, Mayeux

, Mohs

, Morris

, Rossor

, Scheltens

, Carrillo

, Thies

, Weintraub

, Phelps

(2011) The diagnosis of dementia due to Alzheimer’s disease: Recommendations from the National Institute on Aging-Alzheimer’s Association workgroups on diagnostic guidelines for Alzheimer’s disease. Alzheimers Dement 7, 263–269.

18.

Joshi

, Pontecorvo

, Lu

, Skovronsky

, Mintun

, Devous

Sr (2015) A semiautomated method for quantification of F 18 Florbetapir PET images. J Nucl Med 56, 1736–1741.

19.

Shcherbinin

, Schwarz

, Joshi

, Navitsky

, Flitter

, Shankle

, Devous

Sr. , Mintun

(2016) Kinetics of the tau PET tracer 18F-AV-1451 (T807) in subjects with normal cognitive function, mild cognitive impairment and Alzheimer’s disease. J Nucl Med 57, 1535–1542.

20.

Baker

, Lockhart

, Price

, He

, Huesman

, Schonhaut

, Faria

, Rabinovici

, Jagust

(2017) Reference tissue-based kinetic evaluation of 18F-AV-1451 in aging and dementia. J Nucl Med 58, 332–338.

21.

Guizard

, Fonov

, Garcia-Lorenzo

, Nakamura

, Aubert-Broche

, Collins

(2015) Spatio-temporal regularization for longitudinal registration to subject-specific 3D template. PLoS One 10, e0133352.

22.

Smith

, Zhang

, Jenkinson

, Chen

, Matthews

, Federico

, De Stefano

(2002) Accurate, robust and automated longitudinal and cross-sectional brain change analyses. Neuroimage 17, 479–489.

23.

Team

RDC

(2011) R: A Language and Environment for Statistical Computing, The R foundation for statistical computing, Vienna, Austria.

24.

Pontecorvo

, Devous

Sr. , Navitsky

, Lu

, Salloway

, Schaerf

, Jennings

, Arora

, McGeehan

, Lim

, Xiong

, Joshi

, Siderowf

, Mintun

, 18F-AV-1451-A05 investigators (2017) Relationships between flortaucipir PET tau binding and amyloid burden, clinical diagnosis, age and cognition. Brain 140, 748–763.

25.

Ossenkoppele

, Mattsson

, Teunissen

, Barkhof

, Pijnenburg

, Scheltens

, van der Flier

, Rabinovici

(2015) Cerebrospinal fluid biomarkers and cerebral atrophy in distinct clinical variants of probable Alzheimer’s disease. Neurobiol Aging 36, 2340–2347.

26.

Ossenkoppele

, Cohn-Sheehy

, La Joie

, Vogel

, Moller

, Lehmann

, van Berckel

, Seeley

, Pijnenburg

, Gorno-Tempini

, Kramer

, Barkhof

, Rosen

, van der Flier

, Jagust

, Miller

, Scheltens

, Rabinovici

(2015) Atrophy patterns in early clinical stages across distinct phenotypes of Alzheimer’s disease. Hum Brain Mapp 36, 4421–4437.

27.

Frisoni

, Pievani

, Testa

, Sabattoli

, Bresciani

, Bonetti

, Beltramello

, Hayashi

, Toga

, Thompson

(2007) The topography of grey matter involvement in early and late onset Alzheimer’s disease. Brain 130, 720–730.

28.

Liesinger

, Graff-Radford

, Duara

, Carter

, Hanna Al-Shaikh

, Koga

, Hinkle

, DiLello

, Johnson

, Aziz

, Ertekin-Taner

, Ross

, Dickson

, Murray

(2018) Sex and age interact to determine clinicopathologic differences in Alzheimer’s disease. Acta Neuropathol 136, 873–885.

29.

Ossenkoppele

, Mattsson

, Teunissen

, Barkhof

, Pijnenburg

, Scheltens

, van der Flier

, Rabinovici

(2015) Cerebrospinal fluid biomarkers and cerebral atrophy in distinct clinical variants of probable Alzheimer’s disease. Neurobiol Aging 36, 2340–2347.

30.

Ikonomovic

, Abrahamson

, Price

, Mathis

, Klunk

(2016) [F-18]AV-1451 PET retention in choroid plexus: More than “off-target” binding. Ann Neurol 80, 307–308.