Neurofibrillary tangle deposition in primary idiotypic isocortices using flortaucipir PET in histopathologically confirmed Alzheimer’s disease

Summary

Tau deposition in Alzheimer’s disease (AD) spreads throughout the brain in a stereotypical pattern captured by the Braak neurofibrillary tangle staging scheme. While involvement of primary isocortex defines the highest stage, it is unclear whether flortaucipir PET uptake differs among primary isocortical regions. We determined relationships between flortaucipir uptake in primary visual, auditory, motor, and somatosensory isocortices and Braak stage in 155 autopsy-confirmed participants with AD. Flortaucipir uptake was observed starting at Braak stage V, with visual and auditory isocortices showing greater and faster increases in uptake than somatosensory and motor; age and APOE ε4 influenced uptake. Flortaucipir uptake and change in primary isocortices were similar in the lateral temporal cortex but different in the entorhinal cortex. Findings suggest that tau deposition in primary isocortices in AD occurs earlier than currently emphasized, although there are differences among them. Given these findings, clinical correlations of primary isocortical involvement in AD need to be evaluated.

Graphical abstract

Highlights

• •Flortaucipir uptake in primary isocortex was observed at Braak stages V and VI
• •Visual and auditory isocortices showed greater uptake than sensorimotor
• •Flortaucipir uptake and rates in primary isocortices differed from the entorhinal cortex
• •Younger age was associated with higher uptake in primary isocortical regions

Health sciences; Medicine; Medical specialty; Internal medicine; Neurology; Natural sciences; Biological sciences; Neuroscience; Clinical neuroscience

Introduction

Alzheimer’s disease (AD) is a neurodegenerative entity pathologically characterized by the presence of two abnormal proteins in the brain, β-amyloid and tau.¹ β-amyloid is deposited in the brain as senile plaques, while tau is deposited as neurofibrillary tangles (NFTs).² Senile plaques are extracellular lesions composed of β-amyloid, with or without dystrophic neurites, while NFTs are intracellular inclusions within cell bodies of pyramidal cells composed of paired helical filaments of tau.³ Patients suffering from AD most often develop symptoms over age 65, including early loss of episodic memory,⁴ due to destruction of cellular and subcellular components of the limbic system of the brain.⁵ However, a relatively large proportion of patients with AD present with changes in comportment, language, praxis, or visuospatial/visuoperceptual function⁶ rather than early loss of episodic memory. These patients are labeled as having atypical AD^6,7,8; they tend to be younger than age 65 at onset and often show NFT pathology, predominantly in the isocortex with relative sparing of limbic system regions.^9,10,11

Over the past three and a half decades, different proposed staging schemes have tried to capture the spread/distribution of senile plaques and NFTs across different brain regions in AD.^12,13,14 Currently, the two most accepted staging schemes for β-amyloid and NFT distribution are the Thal β-amyloid stage¹⁵ and the Braak NFT stage,^16,17,18 respectively. According to the Thal staging scheme, β-amyloid deposition first occurs in the isocortex (Thal stage 1) before spreading to limbic regions (Thal stage 2), basal ganglia (Thal stage 3), brainstem (Thal stage 4), and finally to the cerebellum (Thal stage 5). In the Braak NFT staging scheme, which also employs a six-stage staging scheme, NFT deposition first appears in the transentorhinal cortex (Stage I), followed by involvement of the entorhinal cortex (ERC) and hippocampus (Braak Stage II), before spreading to temporal isocortex (Braak Stage III), basal forebrain (Braak Stage IV), association isocortex (Braak Stages V), and lastly, primary isocortex (Stage VI).

The isocortex, or neocortex, is a 6-layered areas (i.e., having 6 laminae) of the brain that include both primary idiotypic and association homotypic cortices. The primary idiotypic cortex encompasses raw visual, auditory, olfactory, gustatory, somatosensory (i.e., sensory), and motor processes (e.g., executing an arm toss of a football), while association cortex influences the planning, integration, and interpretation of modality function, such as programming, planning, and integrating the movements required for the arm toss. There are many different primary isocortical regions distributed across different lobes of the brain.¹⁹ These include the primary motor cortex in the frontal lobe (M1, Brodmann area 4), the primary somatosensory cortex in the parietal lobe (S1, Brodman area 3-1-2), the primary visual cortex (also known as the striate cortex in the occipital lobe; V1, Brodmann area 17), the primary auditory area (also known as Heschel’s gyrus in the posterior superior lateral temporal lobe; A1, Brodmann area 41, 42), as well as other areas for the primary functions of taste and smell. The primary motor and somatosensory isocortices have recently been further subdivided into smaller areas to provide more fine-grained functional and anatomical knowledge.²⁰

Knowledge about β-amyloid and tau deposition has primarily come from neuropathological studies,^21,22 which are considered the gold standard for identifying abnormal protein deposition in the brain at the time of death. Neuropathological studies, however, cannot address questions about protein deposition early in the disease course, nor can they address the rate of protein deposition over time. Over the past two decades, advanced ligand chemistry and positron emission tomography (PET) have aided the development of antemortem biomarkers indicative of β-amyloid/senile plaques and tau/NFTs burden.^{23,24,25,26,27,28} One of the most widely used biomarkers for NFT burden and distribution is the ligand flortaucipir (¹⁸F-AV1451, T807).²³ Autoradiographic studies have shown excellent correlation between flortaucipir uptake and Braak NFT stage^29,30,31; hence, flortaucipir PET is an excellent technique for assessing NFT deposition and behavior in AD brains. To date, only a few imaging-pathological studies have directly assessed relationships between antemortem flortaucipir PET and postmortem NFT pathology.^{32,33,34,35,36,37,38,39,40} Most of these studies used a single primary isocortical region, usually the primary visual cortex (V1), to determine Braak NFT Stage VI. Additionally, little is known about the rate of protein deposition over time.

In this study, we examine flortaucipir PET data in a cohort of 155 autopsy-confirmed AD participants to elucidate the behavior of NFT pathology and the Braak NFT stage across four different primary isocortical areas, two of which were further subdivided (M1, S1). We validate findings in 29 additional cases. We hypothesize that although involvement of primary isocortex has been linked to Braak NFT stage VI, implying pathological homogeneity of NFT behavior across primary isocortical regions, there are differences in burden and rates of accumulation of NFTs at and before Braak stage VI. Understanding involvement of these primary isocortical regions and how the spread of the disease involves these regions furthers our understanding of the biology of AD and better defines appropriate tau PET biomarkers to capture disease progression for clinical treatment trials. This knowledge also has implications for pathological staging and for implementation of standard clinical batteries to assess for correlates of primary isocortical dysfunction in AD.

Results

Cohort characteristics

We included 155 participants recruited from Mayo Clinic, Rochester, MN, who had flortaucipir PET imaging during life, underwent an autopsy evaluation between 8/28/2016 and 5/5/2024, and had AD⁴¹ (Thal Aβ phase of 1–5¹⁵ and Braak NFT stage 0-VI¹⁶). Exclusion criteria were absence of β-amyloid deposition (i.e., Thal phase = 0), a pathological diagnosis of frontotemporal lobar degeneration (FTLD),⁴² or another primary pathological diagnosis. All 155 participants had enrolled in one of three NIH-funded cohort studies, including the Mayo Clinic Alzheimer’s Disease Research Center (PI: Boeve), the Mayo Clinic Study of Aging (PI: Petersen), or the Neurodegenerative Research Group (PIs: Josephs/Whitwell). Participant characteristics are shown in Table 1. There were more males than females in the cohort (58%). The median age at death was 79 years (inter-quartile range 69–89), with the last available flortaucipir PET scans obtained a median 2.5 years (1.6–3.9) before death. At the last visit before death, 28 (18%) were cognitively unimpaired, 74 (48%) had typical amnestic Alzheimer’s type dementia (including 17 with amnestic mild cognitive impairment due to AD), 21 (14%) had a diagnosis of an atypical Alzheimer’s dementia (11 visual variants, 9 language [logopenic] variant, 1 behavioral variant), and 32 (21%) had a non-Alzheimer’s dementia diagnosis (15 dementia with Lewy bodies, 7 hard-to-classify dementia, 7 parkinsonian syndromes, 2 vascular dementia, 1 non-fluent variant of primary progressive aphasia). Of those with unimpaired cognition, the majority (79%) had a Braak stage of 0-IV. Most of the typical and atypical Alzheimer’s dementia participants had a Braak stage of V-VI (81% and 90%). In contrast, the other dementia participants were almost equally split between those with Braak stages 0-IV (53%) and Braak stage V-VI (47%).

Table 1.

Demographic features of the full cohort by Braak NFT stage

	All (n = 155)	Braak NFT stage
		0 (n = 2)	I (n = 2)	II (n = 17)	III (n = 13)	IV (n = 21)	V (n = 33)	VI (n = 67)
Sex, Female (%)	65 (42%)	0 (0%)	1 (50%)	4 (24%)	6 (46%)	5 (24%)	16 (48%)	33 (49%)
Age last flortaucipir scan (y)	76 (67, 86)	74 (70, 79)	65 (64, 67)	82 (74, 89)	85 (75, 88)	81 (72, 88)	81 (74, 87)	72 (62, 82)
Age at death (y)	79 (69, 89)	77 (72, 82)	66 (65, 68)	85 (76, 92)	89 (79, 92)	82 (76, 89)	84 (75, 89)	74 (64, 84)
Time from last FTP to death (y)	2.5 (1.6, 3.9)	2.6 (2.4, 2.9)	0.8 (0.8, 0.9)	2.6 (1.8, 4.1)	2.0 (0.9, 5.9)	2.6 (1.9, 3.6)	1.9 (1.4, 3.2)	3.1 (1.8, 4.2)
Mini-Mental State Examination	21 (16, 26)	26 (24, 27)	22 (19, 26)	25 (22, 28)	28 (27, 29)	22 (20, 26)	21 (18, 26)	18 (12, 21)
Clinical Dementia Rating	6 (1, 10)	4 (2, 7)	5 (2, 8)	2 (0, 9)	0 (0, 2)	2 (0, 4)	4 (1, 8)	8 (5, 11)
Race
American Indian/Alaska Native	1 (1%)	0 (0%)	0 (0%)	0 (0%)	1 (8%)	0 (0%)	0 (0%)	0 (0%)
Black	1 (1%)	0 (0%)	0 (0%)	0 (0%)	0 (0%)	0 (0%)	1 (3%)	0 (0%)
White	153 (99%)	2 (100%)	2 (100%)	17 (100%)	12 (92%)	21 (100%)	32 (97%)	67 (100%)
Education (y)	14 (12, 16)	14 (14, 14)	12 (12, 13)	16 (12, 16)	12 (12, 14)	13 (12, 16)	14 (12, 16)	16 (12, 18)
APOE ε4 genotype, N %	84/150 (56%)	1/2 (50%)	1/2 (50%)	5/17 (29%)	2/12 (17%)	12/21 (57%)	16/32 (50%)	47/64 (73%)
Clinical diagnosis at last visit
Normal cognition	28 (18%)	1 (50%)	1 (50%)	7 (41%)	9 (69%)	4 (19%)	5 (15%)	1 (1%)
Typical Alzheimer’s dementia/MCI	74 (48%)	0 (0%)	0 (0%)	4 (24%)	1 (8%)	9 (43%)	18 (55%)	42 (63%)
Atypical Alzheimer’s dementia	21 (14%)	0 (0%)	1 (50%)	0 (0%)	0 (0%)	1 (5%)	3 (9%)	16 (24%)
Other dementia	32 (21%)	1 (50%)	0 (0%)	6 (35%)	3 (23%)	7 (33%)	7 (21%)	8 (12%)
Thal phase
1	10 (6%)	1 (50%)	0 (0%)	3 (18%)	3 (23%)	2 (10%)	1 (3%)	0 (0%)
2	11 (7%)	1 (50%)	1 (50%)	2 (12%)	4 (31%)	2 (10%)	0 (0%)	1 (1%)
3	26 (17%)	0 (0%)	1 (50%)	7 (41%)	5 (38%)	5 (24%)	6 (18%)	2 (3%)
4	39 (25%)	0 (0%)	0 (0%)	4 (24%)	1 (8%)	10 (48%)	7 (21%)	17 (25%)
5	69 (45%)	0 (0%)	0 (0%)	1 (6%)	0 (0%)	2 (10%)	19 (58%)	47 (70%)

Data shown as median (25^th, 75^th percentiles) or N (%). APOE, apolipoprotein E; FTP, flortaucipir.

Regional correlations in flortaucipir uptake

Flortaucipir standardized uptake value ratios (SUVRs) were calculated for the motor, somatosensory, auditory, and visual isocortical regions using the human Brainnetome atlas. The motor cortex was also subdivided into four subregions (Brodmann area 4 head/face (A4hf), upper limb (A4ul), trunk (A4t), and tongue/larynx (A4tl) regions), and similarly, the somatosensory cortex was subdivided into four subregions (Brodmann area 1, 2 and 3 upper limb and head and face region [A1_2_3ulhf], tongue/larynx [A1_2_3tonla], trunk [A1_2_3tru], and Brodmann area 2 [A2]; Figure 1). We also assessed the ERC and inferior temporal lateral (ITL) gyrus, regions affected at early Braak NFT stages.¹⁶ Individual log-transformed SUVR data points and pairwise rank correlations with 95% confidence intervals (CIs) illustrate relationships between regional SUVRs calculated from the PET scan closest to death (Figure 2 and S1). Flortaucipir uptake in the ERC strongly correlated with uptake in the ITL (r = 0.87) but only moderately in the four primary isocortices (r = 0.65 to 0.74). Uptake in the ITL strongly correlated with uptake in the auditory isocortex (r = 0.86). We observed a very strong correlation between uptake in the motor and somatosensory isocortices (r = 0.96), although the strength of the correlations varied across subregions (r = 0.63 to 0.97). Strong correlations were observed between the auditory isocortex and the other primary isocortices (r = 0.78 to 0.89). Uptake in the visual cortex correlated well with uptake in the other primary isocortices and ITL (r = 0.74 to 0.78). We also assessed correlations only within participants with Braak stage VI, since the primary isocortices are assumed to be involved at that stage, and we observed similar patterns of relationships, albeit with slightly lower correlations (Figure S2).

Figure 1.

Three-dimensional brain renders depicting the location of the Brainnetome atlas regions of interest

The brain renderings were created using MRIcroGL (version 1.2.20220720). OPC = occipital polar cortex; Mot A4hf = motor Brodmann area 4 head/face; Mot A4ul = motor Brodmann area 4 upper limb; Mot A4t = motor Brodmann area 4 trunk; Mot A4tl = motor Brodmann area 4 tongue/larynx; Sen A1_2_3ulhf = somatosensory Brodmann area 1, 2, and 3 upper limb, head and face region; Sen A1_2_3tonla = somatosensory Brodmann area 1, 2, and 3 tongue/larynx; Sen A2 = somatosensory Brodmann area 2; Sen A1_2_3tru = somatosensory Brodmann area 1, 2, and 3 trunk; Aud A41_42 = auditory Brodmann area 41_42; ITL = inferior temporal lateral; ERC = entorhinal cortex; L = left; R = right

Figure 2.

Correlations in flortaucipir uptake between regions

Array of pairwise scatterplots show log SUVR of one region versus log SUVR of another. The rank correlation for each scatterplot is shown in the top left corner of each image along with 95% confidence intervals. Gray grid lines are drawn at SUVR values of 1, 1.5, and 2. The black diagonal line represents the identity line, y = x. The top-right image (blue points) shows correlations between the primary isocortices, ERC and ITL, and the bottom-left image (gray points) shows correlations between the motor and somatosensory subregions. Mot A4hf = motor Brodmann area 4 head/face; Mot A4ul = motor Brodmann area 4 upper limb; Mot A4t = motor Brodmann area 4 trunk; Mot A4tl = motor Brodmann area 4 tongue/larynx; Sen A1/2/3ulhf = somatosensory Brodmann area 1, 2 and 3 upper limb, head and face region; Sen A1/2/3tonla = somatosensory Brodmann area 1, 2 and 3 tongue/larynx; Sen A2 = somatosensory Brodmann area 2; Sen A1/2/3tru = somatosensory Brodmann area 1, 2 and 3 trunk; Aud = Auditory; Mot = motor; Sens = somatosensory; Vis = visual; ITL = inferior temporal lateral; ERC = entorhinal cortex.

Change in flortaucipir uptake by Braak NFT stage

Flortaucipir SUVRs at the last available scan are plotted against Braak NFT stage for all regions in Figure 3. Median SUVR in all four primary isocortices was low at Braak stages 0-V and elevated at Braak stage VI. However, several participants showed elevated SUVR at Braak stage V in the auditory and visual isocortices (Figures 3 and 4). In contrast, SUVR in the ERC was slightly elevated at Braak stage II and then increased with increasing Braak stage. The ITL showed higher SUVR than all other regions at all Braak stages, with SUVR increasing at Braak stages IV to VI.

Figure 3.

Flortaucipir SUVR by Braak stage within each region

Flortaucipir SUVR at last scan by Braak stage within each region. Gray dots represent individual SUVR values which have been randomly jittered along the horizontal axis to reduce overlap. The blue line connects the geometric mean SUVR at each Braak stage, with vertical bars representing 95% confidence intervals. Mot A4hf = motor Brodmann area 4 head/face; Mot A4ul = motor Brodmann area 4 upper limb; Mot A4t = motor Brodmann area 4 trunk; Mot A4tl = motor Brodmann area 4 tongue/larynx; Sen A1/2/3ulhf = somatosensory Brodmann area 1, 2 and 3 upper limb, head and face region; Sen A1/2/3tonla = somatosensory Brodmann area 1, 2 and 3 tongue/larynx; Sen A2 = somatosensory Brodmann area 2; Sen A1/2/3tru = somatosensory Brodmann area 1,2 and 3 trunk; ITL = inferior temporal lateral; ERC = entorhinal cortex.

Figure 4.

Representative tau PET SUVR images for Braak V participants with and without uptake in the visual and auditory isocortices

Participant 1 shows uptake in both the auditory (Aud+) and visual (vis+) isocortex; participant 2 shows uptake in the auditory isocortex but not the visual isocortex; participant 3 shows uptake in the visual isocortex but not the auditory isocortex; and participant 4 shows no uptake in either isocortex.

We used generalized additive models (GAMs) to estimate the relationship between SUVR and Braak stage for the four primary isocortical regions (motor, somatosensory, auditory, visual), ERC, and ITL (Figure 5A; Table S1). The GAM-associated smoothing provides a non-linear curve that best fits the data. Figure S3 shows percentage differences in mean SUVR for a 1-unit increase in Braak stage, and Figure S4 shows differences in mean SUVR across regions for a given Braak stage. The first derivatives of the estimated mean SUVR based on these GAMs were then calculated and plotted (Figure 5B). The first derivative represents the instantaneous change of geometric mean SUVR with change of Braak stage at any given point on the curve and allowed us to assess change in SUVR and the rate of SUVR change as Braak stage increased, independent of the absolute SUVR. The SUVRs were relatively low for all primary isocortical regions between Braak stage 0 and IV (Figure 5A; Table S1), with no evidence for increasing SUVR between Braak stages 0 and I, I and II, II and III, or III and IV (Figure S3). The first derivatives were close to 0 between Braak stage 0 and IV but started to increase from 0 at Braak stage IV (Figure 5B). Indeed, SUVRs increased between Braak IV and V and between V and VI in all primary isocortices, with a greater difference in SUVR observed between Braak V and VI compared to between Braak IV and V (p < 0.05 for all four primary isocortical regions, with percentage-point differences ranging from 11 points in the motor isocortex to 22 in visual isocortex; Figure S3). The visual and auditory isocortices showed similar relationships with Braak stage, with greater SUVR and greater first derivatives than those observed for the somatosensory and motor isocortices from Braak stage II (Figures 5 and S4; Table S1). The SUVR of the somatosensory and motor isocortices were similar at all Braak stages, except at Braak stage VI, where SUVR was significantly higher in the somatosensory isocortex (Figure S4).

Figure 5.

Generalized additive model results modeling flortaucipir PET by Braak stage for each region

Estimated geometric mean SUVR by Braak stage and corresponding first derivative curves for the primary isocortices, ERC and ITL are shown in the top image, and the motor and sensory subregions are shown in the bottom image. Mot A4hf = motor Brodmann area 4 head/face; Mot A4ul = motor Brodmann area 4upper limb; Mot A4t = motor Brodmann area 4 trunk; Mot A4tl = motor Brodmann area 4 tongue/larynx; Sen A1_2_3ulhf = sensory Brodmann area 1,2 and 3 upper limb, head and face region; Sen A1_2_3tonia = sensory Brodmann area 1,2 and 3 tongue/larynx; Sen A2 = sensory Brodmann area 2; Sen A1_2_3tru = sensory Brodmann area 1,2 and 3 trunk; ERC = entorhinal cortex; ITL = inferior temporal lobe.

The ERC showed increasing SUVR as Braak stage increased (Figure 5A; Table S1), with differences in SUVR observed at each Braak stage interval from II and above (Figure S3). The ERC had higher SUVR than all other regions at Braak stages IV and V, except for the ITL (Figure S4). At Braak stage VI, SUVR in the auditory and visual isocortices catch up to the ERC, with no differences observed in SUVR. The first derivatives for the ERC were greater than the first derivatives of other regions at low Braak stages (Figure 5B), reflecting an early increase in SUVR. After Braak stage III, the SUVR steadily increased with increasing Braak stage, although the slope of the first derivatives was not as steep as the slopes of the other regions (Figure 5). By Braak stage VI, the first derivatives in the ERC were smaller than the first derivatives of the ITL, primary visual, auditory, and somatosensory isocortices, but similar to the first derivative of the primary motor isocortex (Figure 5B). The ITL showed increased SUVR between Braak stages III and IV, between IV and V, and between V and VI (Figures 5A and S3; Table S1). The ITL had greater SUVR than all other regions between Braak stage II and VI and all other regions except for visual isocortex at Braak stage I (Figure S4). The first derivatives of the ITL start to increase after Braak stage II and continue to increase through Braak stage VI, at which point the first derivatives resemble the visual and auditory isocortices (Figure 5B).

The SUVRs and the first derivatives by Braak stage for all subregions of the somatosensory and motor isocortices are shown in Figures 5C and 5D; Table S1. There were minimal differences within the motor and somatosensory subregions at Braak stages 0-IV (Figures S5 and S6). At Braak stage V and VI, the motor tongue/larynx subregion showed higher SUVR than the other three motor subregions (Figures 5C and S5; Table S1). The somatosensory Brodmann area 2 subregion showed higher SUVR than the somatosensory upper limb/head/face and trunk subregions at Braak stages V and VI and higher SUVR than the somatosensory tongue/larynx subregion at Braak VI (Figures 5C and S6; Table S1). Similarly, at Braak stage VI, the motor trunk subregion showed the lowest first derivatives of all the motor subregions, and the somatosensory Brodmann area 2 subregion showed the highest first derivatives of all the somatosensory subregions (Figure 5D).

Regional flortaucipir uptake across clinical diagnoses

To determine whether the relationships between the primary isocortices and Braak stage differ by clinical diagnosis, we plotted flortaucipir SUVR at the last scan by Braak stage for each clinical diagnosis (Figure 6; Table S2). At Braak stage VI, the atypical Alzheimer’s dementia group had the highest mean SUVR in the motor, somatosensory, and auditory regions (all p ≤ 0.01) and higher mean SUVR in the primary visual cortex region compared to typical AD (p = 0.01). In the ERC, no clear differences were observed in SUVR across clinical diagnoses at any Braak stage. Since the most variability in SUVR by diagnosis was observed predominantly at Braak stage VI, we performed linear regression analysis to evaluate how age, sex, APOE, and diagnosis correlated with SUVR at the last scan in only those participants with Braak stage VI (Figure 7; Table S3). After adjusting for other covariates, there was evidence of higher SUVR in the motor trunk in atypical Alzheimer’s dementia compared to typical Alzheimer’s dementia (p = 0.03). No differences were observed between the other-dementia group and either the atypical or typical Alzheimer’s dementia groups. Younger age correlated with greater SUVR in all regions and subregions (p < 0.001 for all). Male sex correlated with lower SUVR in the visual isocortex (∼18%, p = 0.01) and the somatosensory Brodmann area 2 subregion (p = 0.04). Lastly, differences due to APOE ε4 genotype tended to be minimal, although ε4 carriers had ∼9% higher SUVR in the ERC (p = 0.06).

Figure 6.

Geometric mean SUVR at the last scan by Braak stage within region, shown separately by diagnosis

Mot A4hf = motor Brodmann area 4 head/face; Mot A4ul = motor Brodmann area 4upper limb; Mot A4t = motor Brodmann area 4 trunk; Mot A4tl = motor Brodmann area 4 tongue/larynx; Sen A1_2_3ulhf = somatosensory Brodmann area 1,2 and 3 upper limb, head and face region; Sen A1_2_3tonla = somatosensory Brodmann area 1,2 and 3 tongue/larynx; Sen A2 = somatosensory Brodmann area 2; Sen A1_2_3tru = somatosensory Brodmann area 1,2 and 3 trunk; ITL = inferior temporal lateral; ERC = entorhinal cortex.

Figure 7.

Factors associated with SUVR in Braak VI participants

Mot A4hf = motor Brodmann area 4 head/face; Mot A4ul = motor Brodmann area 4upper limb; Mot A4t = motor Brodmann area 4 trunk; Mot A4tl = motor Brodmann area 4 tongue/larynx; Sen A1_2_3ulhf = somatosensory Brodmann area 1,2 and 3 upper limb, head and face region; Sen A1_2_3tonla = somatosensory Brodmann area 1,2 and 3 tongue/larynx; Sen A2 = somatosensory Brodmann area 2; Sen A1_2_3tru = somatosensory Brodmann area 1,2 and 3 trunk; ITL = inferior temporal lateral; ERC = entorhinal cortex; APOE = apolipoprotein E

The typical Alzheimer’s dementia group had a large enough sample size to allow us to fit GAMs assessing flortaucipir uptake across Braak stage (0-VI), once again assessing both SUVR (Figure S7; Table S4) and the first derivatives (Figure S7). We found similar patterns of change occurring across the Braak stages to those observed across the whole cohort. We did not, however, see flattening of the first derivatives in the ERC at Braak stages V-VI, as we did in the whole cohort. Instead, the first derivatives increased at a relatively linear rate with Braak stage and were higher than the motor cortex at Braak stage VI. At Braak stage VI, mean SUVR in the ERC appeared higher than in visual and auditory isocortices but was not significant (p = 0.82). The ITL showed increasing SUVR and, hence, positive first derivatives between Braak stage 0 and III, while the first derivatives for the ITL in the whole cohort were close to 0 at these low Braak stages. GAM estimates were plotted separately for the typical Alzheimer’s dementia participants and the participants with either atypical Alzheimer’s dementia or other-dementia diagnoses (Figure S8; Table S5), with no significant differences observed in any of the regions or subregions.

Longitudinal rates of change in flortaucipir in participants with multiple flortaucipir PET scans

Table S6 shows the demographic and clinical features of the 76 participants with longitudinal flortaucipir PET data (i.e., ≥2 flortaucipir PET scans). The number of serial scans ranged from 2 to 6. Rates of change in flortaucipir SUVR were modeled using linear mixed-effects models. No region demonstrated an increased rate of flortaucipir accumulation over time at Braak stages ≤ IV (Figure 8). At Braak stage V, increased rates of flortaucipir accumulation over time were observed in the ITL and visual isocortex only. At Braak stage VI, all regions showed increased rates of tau accumulation over time, with the lowest rates observed in the ERC. There was little evidence of difference in the rates of flortaucipir accumulation between the motor and somatosensory subregions.

Figure 8.

Estimated mean annual longitudinal rates of change by region within Braak stage

Mot A4hf = motor Brodmann area 4 head/face; Mot A4ul = motor Brodmann area 4upper limb; Mot A4t = motor Brodmann area 4 trunk; Mot A4tl = motor Brodmann area 4 tongue/larynx; Sen A1_2_3ulhf = somatosensory Brodmann area 1,2 and 3 upper limb, head and face region; Sen A1_2_3tonla = somatosensory Brodmann area 1,2 and 3 tongue/larynx; Sen A2 = somatosensory Brodmann area 2; Sen A1_2_3tru = somatosensory Brodmann area 1,2 and 3 trunk; ITL = inferior temporal lateral; ERC = entorhinal cortex.

Within Braak VI participants (Table S7), based on linear mixed-effects models adjusted for covariates, those with atypical Alzheimer’s dementia had slower rates of tau accumulation over time in the visual isocortex compared to those with typical Alzheimer’s dementia (p = 0.02). No significant differences in rates of tau accumulation over time were observed between the participants with other dementias and the typical and atypical Alzheimer’s dementia groups. Younger age correlated with faster rates of tau accumulation over time in the ERC (p = 0.02), motor isocortex (p = 0.003), somatosensory isocortex (p = 0.02), and auditory isocortex (p = 0.02), as well as several motor and somatosensory subregions. Male sex only showed an association with slower rates of tau accumulation over time in the motor trunk subregion (p = 0.06). The presence of the APOE e4 genotype correlated with slower rates of tau accumulation over time in the ERC (−3%/y, p = 0.03) and faster rates of tau accumulation over time in the somatosensory Brodmann area 2 (p = 0.05) and motor tongue/larynx subregion (p = 0.05).

Sensitivity analyses and internal validation

A sensitivity analysis of the cross-sectional GAMs compared GAM estimates across five models, with each model leaving out a different random subset of 20% of the sample. The GAM estimates of the relationship between SUVR and Braak stage were almost identical for the five different models across all regions (Figure S9). In addition, 29 new participants who met inclusion criteria were collected over one year between 5/6/2024 and 5/1/2025 and served as an internal-validation cohort. Table S8 lists the demographic characteristics of these validation participants. All Braak stages were represented in the validation cohort except for Braak stage IV. The sex ratio, age at death, and time from last scan to death of these 29 validation participants resembled the main cohort of 155 participants. The internal-validation cohort had a slightly higher frequency of participants with normal cognition, higher MMSE, better CDR, lower frequency of the APOE ε4 allele, and a higher proportion of lower Braak stages (0-IV). Flortaucipir SUVRs at the last available scan for these 29 validation participants are plotted by Braak NFT stage in comparison to the 155 participants in the original cohort (Figure S10). The SUVRs for the validation participants, including Braak stage 0 and Stage I cases, were within the range observed in the original cohort for all regions. Cross-sectional GAMs were then fit to estimate the relationship between SUVR and Braak NFT stage for 1) the original cohort of 155 participants combined with the 29 new validation participants (n = 184) and 2) just the validation cohort of 29 participants (Figure S11; Table S9). The modeled associations between SUVR and Braak NFT stage were identical in the combined (n = 184) and original (n = 155) models across regions. The modeled associations between SUVR and Braak NFT stage for the smaller validation cohort (n = 29) were very similar to the original model in the motor, sensory, and visual isocortices and the ERC. In the auditory isocortex and ITL, SUVR started to increase at lower Braak stages in the validation cohort compared to the original cohort.

Discussion

This study investigated tau deposition on flortaucipir PET in four primary isocortical regions (motor, somatosensory, auditory, and visual isocortices) as well as subregions of the primary motor and primary somatosensory isocortex, in a large, autopsy-confirmed cohort of 155 participants with primary AD pathology. Flortaucipir uptake in these primary isocortices was compared to uptake in the ERC and ITL, regions involved earlier in the Braak NFT staging scheme. Many novel findings from this study have not been previously reported. First, we found evidence that all four primary isocortices are affected by Braak stage V. Second, we show that the auditory and visual isocortices have higher SUVRs and faster accumulation of tau across the Braak spectrum based on first derivatives, a novel technique applied to flortaucipir PET data, compared to the somatosensory and motor isocortices. Third, we show that the relationship between flortaucipir SUVR and Braak NFT stage in these primary isocortices differed from that observed in the ERC and ITL. These results were similar in the sensitivity analysis and in the validation cohort of 29 participants. Fourth, we subdivided the primary motor and somatosensory isocortices using the Brainnetome atlas, which provided evidence of subregional differences in tau accumulation in these two primary isocortical regions. In addition to those novel findings, we also validated previous work and showed that the APOE ε4 genotype trended toward higher SUVRs but correlated with lower rates of accumulation over time in the ERC, while younger age correlated with higher SUVRs and faster rates of tau accumulation over time in all regions.

Our data showed low SUVR in the primary isocortical regions between Braak stages 0 and IV, suggesting little-to-no involvement of these regions at these stages, which is consistent with neuropathological findings.¹⁶ At Braak stage IV, median SUVR was still relatively low across regions, but the first derivatives started to increase after Braak stage IV, suggesting a subtle increase in the rate of tau accumulation. In fact, we observed increasing SUVR between Braak stages IV and V as well as between V and VI in all four primary isocortical regions and subregions. Hence, there is evidence of involvement of the primary isocortical regions at Braak stage V, earlier than currently emphasized in the literature. This is a novel finding. Since we did not assess association cortex in this study, we cannot comment on whether the primary isocortical regions are affected at a similar point or later in the disease than associated cortical regions. However, we can conclude that the primary isocortical regions are apparently affected later than the ERC and ITL. Previous voxel-level studies of flortaucipir in clinical cohorts from our group and others have suggested that the primary isocortices are the least-affected regions in AD.^11,43,44 This is unsurprising given that these regions remain unaffected until Braak stage V. Our findings suggest that neuroimaging studies aiming to assess mechanisms and patterns of spread of tau pathology in AD should include the primary isocortical regions. The findings could also have clinical implications for patients with underlying AD, suggesting that signs and symptoms linked to primary visual, auditory, motor, and sensorimotor function may become impaired. However, patients with AD at Braak stage V and VI likely have marked dementia; hence, clinical examinations assessing signs linked to primary isocortical areas, such as hearing or somatosensory loss, will be difficult in this group.

We observed crucial differences between the primary isocortical regions. The auditory and visual isocortices showed similar relationships to Braak stage, which differed from the motor and somatosensory isocortices (motor and somatosensory isocortices showed similar relationships). The auditory and visual isocortices showed consistently higher SUVRs and steeper increases in SUVR through later Braak stages compared to the motor and somatosensory isocortices. The higher SUVRs in the auditory and visual isocortex at low Braak stages may reflect low levels of off-target binding or possibly binding to tau fibrils in neuritic plaques. These explanations would also apply to the relatively high SUVRs observed in the ITL at low Braak stage. Nevertheless, a recent PET-pathology study found no evidence for flortaucipir binding to tau neurites in neuritic plaques.³⁴ Another possibility is that flortaucipir binds to β-amyloid. Binding to β-amyloid in dense-core plaques in the absence of neuritic tau has been reported,⁴⁵ although findings from autoradiographic studies do not support that explanation.^29,31,45 However, β-amyloid is deposited in the isocortices prior to being deposited in the ERC,¹⁵ which would fit with our findings. Regardless of the exact explanation for increased SUVR at low Braak stages, it appears that once SUVR begins increasing in the auditory and visual isocortices, the rate of increase by Braak stage is faster than the rates in the motor and somatosensory isocortices, and the larger differences in SUVR at Braak VI are most likely due to greater NFT burden. We cannot use the GAMs to conclude definitively that any primary isocortical regions are affected earlier than others, and this study did not intend to address the current Braak NFT staging scheme. Our findings, however, do support NFT deposition in the auditory and visual isocortices, predating deposition in the somatosensory and motor isocortices. One possible explanation for this may relate to the combination of laminar differences in cell density and composition across different primary isocortices and regional selective vulnerability to NFT deposition. NFTs are primarily found in pyramidal cells in laminar III and laminar V,^46,47 and differences in NFT deposition in laminar III versus laminar V depend on the isocortical region assessed.⁴⁶ As a result, differences in cell density and cell type in laminar III vs. laminar V across different primary isocortices could lead to differences in NFT burden and rate of deposition. Laminar V of the primary motor isocortex is the most developed layer, consisting of specialized, large cells known as Betz cells. Yet, Betz cells are resistant and do not show NFT pathology,⁴⁸ which could explain why the primary motor isocortex is the least and last region to be affected. Another factor that may be the number of synapses required for tau to spread from the limbic system to each of the four primary isocortices. Third, the relative proximity of one primary isocortex to another and density of interconnecting fibers may play a role in trans-synaptic tau propagation.^49,50 As such, the primary motor and somatosensory isocortices, which are located next to each other and have dense reciprocal connections, may behave more similarly to each other than to the primary visual and auditory isocortices located further away.

The ERC and ITL behaved somewhat differently from the primary isocortical regions. The ERC showed evidence for increasing SUVR with Braak stage even at early Braak stages, as we have previously reported,³³ and supports the ERC becoming affected pathologically at Braak stage II.¹⁶ SUVR increases concordantly with increasing Braak stage, and the rate of tau accumulation with each Braak stage (i.e., the first derivatives) also increases as Braak stage increases, like all the other regions. However, the rate of acceleration of tau accumulation appears fastest between Braak stages III and IV, with a slower change in the rate of acceleration observed after Braak stage IV. This slowing may be due to an approaching ceiling effect, possibly due to the limitation of PET spatial resolution. Positron emission tomography has a spatial resolution of ∼5 mm, and therefore brain regions of that size or smaller produce lower overall count statistics in the scan data. This means that intensity in small regions will always be limited. Hence, as we previously discussed, the ERC may be closer to its maximum capacity of detected tau deposition at a timepoint when isocortical regions just begin to accumulate tau.³³ This differs from what we observed for the ITL and primary isocortical regions, where the rate of change did not slow. In fact, the rate of tau accumulation over time in the ERC was also slower than the other regions at Braak stage VI. Hence, the change in the rate of tau deposition in the ERC is slowing at higher Braak stages. These findings have implications for treatment trials and show that outcome measures that consider the rate of change in SUVR values are influenced by the Braak NFT stage of the enrolled participants. The ITL showed similar trajectories of SUVR and first derivatives by Braak stage to the primary isocortical regions, although there was evidence for tau accumulation at earlier Braak stages, again consistent with the ITL being affected earlier pathologically than the primary isocortical regions.¹⁶ In summary, tau and NFTs in the isocortices, regardless of whether it is homotypic cortex (e.g., ITL) or primary isocortex, show similar relationships to Braak stage. However, the relationships differ in the ERC, which, although possessing a six-layer cortex, has sparse cell burden in lamina I-IV, and none of its six layers corresponds to the six layers of isocortex.⁵¹

We observed some subtle differences across subregions of the motor and somatosensory isocortices. The highest SUVR and first derivatives at Braak stage VI in the primary somatosensory isocortex were observed in the Brodmann area 2 subregion and in the primary motor cortex in the head/face and tongue/larynx subregions. Hence, tau does not deposit uniformly throughout the entire region of these two primary isocortices. This has implications for pathological sampling given that sampling within these primary isocortices will likely capture only one subregion, and different centers are likely to sample different subregions. In addition, pathological sampling in one subregion may not be representative of NFT burden across the primary isocortex.

According to clinical diagnosis, flortaucipir SUVR did not differ at low Braak stages, but we did observe variability at Braak stage VI. The participants with atypical Alzheimer’s dementia had higher mean SUVR than the other clinical groups across isocortical regions, although only the motor-trunk subregion association persisted when we entered age, sex, and APOE genotype into the models. Atypical Alzheimer’s dementia has been associated with higher tau NFT burden^9,52 and flortaucipir uptake^53,54,55,56 in cortical regions compared to typical Alzheimer’s dementia, although atypical Alzheimer’s dementia is associated with a younger age at onset.^7,8 Younger age also correlates with higher flortaucipir uptake,^55,57,58 so age could have driven the higher uptake observed in the motor-trunk subregion in atypical Alzheimer’s dementia in our cohort. The other-dementia group also showed relatively high SUVR in the visual cortex compared to other regions. This may be due to the subset of participants clinically diagnosed with dementia with Lewy bodies, which shows higher flortaucipir uptake in this region.^59,60 The GAMs were also fit just among the typical Alzheimer’s dementia participants. The modeled relationships between Braak stage, SUVR, and the first derivatives resembled the main GAMs run in the whole cohort for all regions of interest. Hence, the behavior of flortaucipir observed across the entire cohort is unlikely to have been biased by any of the clinical groups included in the study and therefore generalizes to typical Alzheimer’s dementia. The Braak VI atypical Alzheimer’s dementia participants showed lower rates of tau accumulation over time than the Braak VI typical Alzheimer’s dementia participants in the visual isocortex. This concurs with previous findings that the occipital lobe has one of the fastest rates of tau accumulation over time in typical Alzheimer’s dementia, while little tau accumulation was observed in this region in atypical Alzheimer’s dementia.⁵⁶ We hypothesize that tau has “filled up” occipitoparietal regions in atypical Alzheimer’s dementia early and subsequently spreads to new regions of the brain. Conversely, in typical Alzheimer’s dementia, tau accumulation shifts from mesolimbic regions to the occipital lobe.

This is the first study to assess rates of flortaucipir accumulation over time in participants with multiple PET scans by Braak NFT stage in an autopsy cohort. We found no evidence for increased rates of tau accumulation over time at low Braak stages. Increased rates of tau accumulation over time were first observed in the ITL and in the visual isocortex at Braak stage V, with all other regions showing increased rates at Braak VI. Given that the rates of accumulation were calculated using all available PET scans and, hence, covered the period leading up to the final PET scan used in the GAMS, the primary visual isocortex is the earliest primary isocortical region where NFT deposition occurs. There was evidence at Braak stage VI that rates of tau accumulation in the visual isocortex were slower and rates in a motor subregion were faster in participants with atypical Alzheimer’s dementia compared to those with typical Alzheimer’s dementia. This may indicate that the participants with atypical Alzheimer’s dementia reach a maximum NFT burden in the visual isocortex earlier, with the rate of accumulation beginning to plateau, while participants with typical Alzheimer’s dementia continue to accumulate tau in this region. Conversely, tau burden in the motor cortex still accumulates in participants with atypical Alzheimer’s dementia given its later involvement in NFT distribution. This finding could have been driven by participants with the visual variant of Alzheimer’s dementia, since they often develop motor features late in the disease course.^61,62

Younger age correlated with higher SUVRs at the last scan before death in all regions, along with faster rates of tau accumulation in the motor, somatosensory, and auditory isocortices and the ERC. Previous studies have similarly observed higher SUVRs,^55,57,58 and faster rates of tau accumulation^63,64,65 over time in younger participants with Alzheimer’s dementia, reflecting more aggressive tau accumulation. Our study demonstrates that age similarly affects the primary isocortical regions. We also observed that males showed lower SUVR in the visual isocortex and a somatosensory subregion, and trends in the somatosensory and auditory isocortex. The association between men and lower SUVR in the visual cortex may relate to the fact that only a third of participants with the visual variant of Alzheimer’s dementia were men. However, that explanation does not account for the other associations. Associations between sex and rates of tau accumulation were less clear, but we did find an association between male sex and slower rates of tau accumulation in one motor subregion. These are novel findings but consistent with greater tau SUVR and faster rates of tau accumulation over time observed in females,^{65,66,67,68,69} which may be due to perimenopausal and menopausal hormonal effects.⁷⁰ The presence of the APOE ε4 allele correlated with greater tau SUVR and slower rates of tau accumulation in the ERC. Conversely, faster rates of tau accumulation in some somatosensory and motor subregions were associated with the presence of the APOE ε4 allele. Analyzing the somatosensory and motor subregions allows us to investigate a more fine-grained assessment of APOE ε4 associations. Our findings regarding the ERC are consistent with results from previous literature. APOE ε4 carriers show greater tau accumulation in the medial temporal lobe.^71,72,73,74 Also, an independent cohort demonstrated that APOE ε4 carriers show slower rates of tau accumulation in the medial temporal regions, hypothesized to be due to tau accumulation in non-carriers catching up to the carriers.⁷¹ However, our findings regarding the primary isocortical subregions are novel and have not been previously reported. It is interesting that APOE ε4 is related to faster tau accumulation in the primary somatosensory and motor subregions but not with the ITL, primary visual, or auditory isocortices. Hence, the associations appear specific to the primary somatosensory and motor isocortices. The reason for this is unclear but unlikely to indicate a region-specific APOE ε4 effect. Instead, the association of APOE ε4 with slower ERC rates but faster rates of tau accumulation in somatosensory and motor subregions may be due to what is happening in these regions at Braak stage VI. At Braak stage VI, the ERC has likely neared NFT maximum burden, while NFT burden in the primary somatosensory and motor isocortices is at a relative minimum and hence likely to have faster rates of NFT deposition. It is important to note that the associations we observed with age, sex, and APOE were identified in participants with a Braak stage of VI and, hence, predominantly in participants with dementia.

The degree of uptake on flortaucipir tended to correlate between the different primary isocortical regions, although the strength of the correlations varied. Very strong correlations were observed between SUVR in the motor and somatosensory isocortices. As mentioned above, this may be due to the proximity of these two regions. In further support, neighboring subregions within both the motor and somatosensory isocortex also tended to show the highest correlations. Flortaucipir PET signal may also bleed into adjacent regions, increasing local correlations. High correlations may also be driven by regions showing involvement at similar disease stages. There may also be a biological explanation for the strong correlation between these isocortices related to cellular and laminar structure, as mentioned earlier in the discussion. However, the density of the cell types within the six different laminae does differ between the motor and somatosensory isocortex, which argues against this latter conjecture. The ERC may only strongly relate to the ITL because both are associated homotypic cortices involved early in Braak stage and, presumably, earlier in the disease process.

We assessed two different types of rate measures. In our derivative method, we modeled the rate of SUVR changes between Braak stages based on non-linear mean curves across Braak stages generated from cross-sectional SUVR data. Hence, in the derivative methods, Braak stages are not treated independently. In the longitudinal analysis, we use multiple SUVRs collected over time within a participant and generate rates of SUVR change per participant, given the participant’s Braak stage. Hence, in the longitudinal analysis, results at a specific Braak stage are independent of what is happening at another Braak stage. Both methods provide complementary information but should not be expected to be identical. For example, while we saw evidence of increased rates of tau accumulation over time in the primary visual isocortex in both methods at Braak V, we did not detect increased rates of tau accumulation over time in the other primary isocortices at Braak V in the longitudinal method, although the first derivates showed evidence for an increasing rate of change. This difference could be due to the inherent variability present in calculating longitudinal measurements within a participant, which will affect sensitivity of longitudinal flortaucipir measurements. In fact, we previously found cross-sectional differences to be superior to longitudinal differences in distinguishing AD from frontotemporal lobar degeneration.⁴⁰

Over the past decade, clinical-imaging studies have indirectly assessed the behavior of flortaucipir uptake with Braak NFT stage in AD^44,75; however, only a handful of studies have assessed flortaucipir behavior in actual pathologically confirmed cases.^{32,33,34,35,36,37,38,39,40} In these imaging-pathological studies, investigators focused primarily on whether and at what Braak NFT stage flortaucipir uptake can be appreciated, or they tried to understand the significance of flortaucipir uptake. However, no imaging-pathological studies have investigated relationships between Braak NFT stage and flortaucipir uptake in primary isocortex or determined when primary isocortex is involved compared to other regions of interest. Unsurprisingly, clinical deficits in patients with AD due to involvement of the primary isocortex are underappreciated. Furthermore, standardized testing of neurological functions linked to the primary isocortical regions is seldom performed in patients with AD, either due to the notion that it is not important or due to the difficulty in accurately examining function due to dementia severity.

There are many strengths to our study. Strengths of the study include the relatively large cohort of participants (n = 155) who completed flortaucipir PET, all of whom underwent brain autopsy with comprehensive neuropathological examination that included Braak NFT staging. Also, almost half the cohort underwent multiple flortaucipir PET scans, which permitted longitudinal analyses. The assessment of four subregions of the primary motor and primary somatosensory isocortex was also a strength, since it allowed for more fine-grained analyses of the primary isocortical regions. We also performed a sensitivity analysis and added an internal-validation cohort to validate our findings and improve generalizability.

Limitations of the study

Weaknesses of the study included the very low number of participants with Braak stages of 0 and I, potentially reducing confidence in the model estimates at these Braak stages. SUVRs in the few Braak 0 and Braak 1 cases in the validation cohort were similar; however, the model estimates for the transition from Braak 0 to II should be interpreted with caution. There were also relatively few participants with an atypical Alzheimer dementia clinical diagnosis or other types of dementia. We could not assess changes in the primary gustatory and primary olfactory isocortices, as these regions are not included as distinct regions in the BrainNetome atlas. We were also unable to assess clinical correlates of involvement of the primary isocortical regions, since the participants in the cohort had not undergone standardized testing for primary motor, somatosensory, auditory, or visual symptoms. We must note that our cohort, and the Braak NFT stages, although consistent with other pathological study cohorts of participants with AD, may not be perfectly representative of the general population. Nevertheless, within a Braak NFT stage, we believe that our estimates are generalizable and our conclusions broadly applicable. We do show, however, that mean cohort SUVRs will depend somewhat on the ratio of typical to atypical AD cases included in the cohort. Lastly, we acknowledge that the lack of quantitative tau burden data may have prevented us from making further conclusions within each Braak stage.

In conclusion, this study provides further insight into tau deposition in primary isocortical regions and how deposition in the late-affected regions in the Braak staging scheme compares to tau deposition in the early-affected ITL and the ERC. Demonstrating that the primary isocortical regions appear to be affected earlier than previously recognized is novel and affects neuroimaging studies that indirectly try to recapitulate the Braak NFT staging scheme. Evidence from this study supports primary motor and somatosensory isocortex becoming affected after primary auditory and visual isocortices.

Resource availability

Lead contact

Requests for further information and resources should be directed to and will be fulfilled by the lead contact, Professor Keith A. Josephs (josephs.keith@mayo.edu).

Materials availability

This study did not generate new unique reagents.

Data and code availability

• •Data associated with this work are publicly available in the Mendeley Repository, https://data.mendeley.com/datasets/c24rs9zy4h/1
• •This paper does not report original code.
• •Any additional information required to reanalyze the data reported in this paper is available from the lead contact upon request.

Acknowledgments

This study was funded by the National Institutes of Health grants RF1-NS112153, R01-AG50603, P30-AG062677, R37-AG011378, U01-NS100620 and U01-AG 006786, the Elsie and Marvin Dekelboum Family Foundation, and the Oxley Foundation. We thank AVID Pharmaceuticals for support in supplying the AV-1451 precursor, chemistry production advice and oversight, and the US Food and Drug Administration regulatory cross-filling permission and documentation needed for this work. We also thank Ms. Nirubol Tosakulwong, B.S., for extracting data that went into the analyses and Dr. David Jones, Dr. Hugo Botha, Dr. Vijay Ramanan, and Dr. Stuart McCarter for patient care of participants.

Author contributions

Conceptualization, K.A.J., J.L.W., and D.W.D.; data curation, S.D.W.; formal analysis, S.D.W. and A.T.; investigation, R.S., A.L., D.W.D., A.N., M.M.M., and V.J.L.; methodology, R.S., S.D.W., A.L., and A.T.; software, R.S.; resources, R.C.P., C.R.J., D.S.K., B.F.B., K.K., and J.G.R.; visualization, R.S. and S.D.W.; funding acquisition, K.A.J., J.L.W., R.C.P., and V.L.; supervision, K.A.J. and J.L.W.; writing – original draft, K.A.J. and J.L.W.; writing – review and editing, S.D.W., A.T., R.S., A.L., J.G.R., M.M.M., K.K., B.F.B., D.S.K., A.N., R.R.R., D.W.D., R.C.P., V.J.L., and C.R.J.

Declaration of interests

Dr. Petersen serves as a consultant for Roche Inc., Merck Inc., and Biogen, Inc., serves on the Data Safety Monitoring Board for Genentech, Inc., and receives royalties from Oxford University Press and UpToDate. Dr. Lowe consults for Bayer Schering Pharma, Piramal Life Sciences, Life Molecular Imaging, Eisai Inc., AVID Radiopharmaceuticals, and Merck Research and receives research support from GE HealthCare, Siemens Molecular Imaging, and AVID Radiopharmaceuticals. Dr. Boeve receives honoraria for SAB activities for the Tau Consortium; is a site investigator for clinical trials sponsored by Alector, Cognition Therapeutics, EIP Pharma, and Transposon; and receives research support from NIH. Dr. Graff-Radford serves on the DSMB for STROKENET; is a coinvestigator for studies sponsored in part by Eisai and Cognition Therapeutics; and receives research support from NIH. The other authors declare no financial or other conflict of interest.

Declaration of generative AI and AI-assisted technologies in the writing process

No AI generative tools were used in the preparation of this manuscript.

STAR★Methods

Key resources table

REAGENT or RESOURCE	SOURCE	IDENTIFIER
Antibodies
Mouse monoclonal AT8 clone	Thermo Fisher Scientific,	Cat#: MN1020; RRID: AB_223647
Mouse monoclonal 6F/3D clone	DAKO, Carpinteria, CA	Cat#: M0872; RRID: AB_2056966
Biological samples
Adult brain tissue	Mayo Clinic, Rochester, MN	N/A
Chemicals,peptides, andrecombinantproteins
18F-Flortaucipir	AVID Pharmaceuticals	N/A
Depositeddata
Mendeley	https://data.mendeley.com/datasets/c24rs9zy4h/1	N/A
Software andalgorithms
R: A Language and Environment for Statistical Computing	https://www.R-project.org/	Version 4.5.1
mgcv package for R	https://doi.org/10.32614/CRAN.package.mgcv	Version 1.9–4
numDeriv package for R	https://doi.org/10.32614/CRAN.package.numDeriv	Version 2016.8–1.1
lme4 package for R	https://doi.org/10.32614/CRAN.package.lme4	Version 1.1–38
brms package for R	https://doi.org/10.32614/CRAN.package.brms	Version 2.23.0

Experimental model and study participant details

We identified all individuals at Mayo Clinic that had undergone flortaucipir PET during life, died between 8/28/2016 and 5/5/2024, undergone an autopsy evaluation and had a neuropathological diagnosis of Alzheimer’s Disease Neuropathologic changes (ADNC) as the primary diagnosis. Participants were excluded if Aβ deposition was absent (Thal phase = 0) or if the participant had a primary pathological diagnosis of a frontotemporal lobar degeneration (FTLD) (including FTLD-tau, FTLD-TDP and FTLD-FET), vascular disease, Lewy body disease, or any other neurodegenerative disease. Participants had been enrolled and followed during life in one of the NIH funded enterprises including the Mayo Clinic Alzheimer’s Disease Research Center, the Mayo Clinic Study of Aging/Patient Registry,⁷⁶ or the Neurodegenerative Research Group (NRG). A total of 155 participants who met this inclusion criteria were identified. The age, sex, race and ethnicity of these participants are shown in Table 1. In addition to this original cohort, we identified 29 participants who met inclusion criteria and had died between 5/5/2024 and 5/1/2025. This new cohort served as an internal validation cohort. The age, sex, race and ethnicity of these 29 participants are shown in Table S8.

The study was approved by the Mayo Clinic IRB (IRB numbers: 18–001813; 14–004401; 712–98). All participants provided written informed consent to participate in this study.

Method details

Clinical evaluations

All participants had undergone detailed neurological and neuropsychological evaluations,⁷⁶ including the Mini-Mental Status Examination (MMSE)⁷⁷ for global cognition and the Clinical Dementia Rating Scale (CDR)⁷⁸ for global functional status, and had been evaluated and diagnosed by a specialist in behavioral neurology (BFB, DSK, JGR, KAJ, RCP). For this study, we classified participants into four diagnostic groups: 1) normal cognition: participants judged to have no cognitive impairment according to published criteria⁷⁸; 2) typical amnestic Alzheimer’s dementia: participants meeting published criteria for amnestic Alzheimer’s dementia⁴ or mild cognitive impairment (MCI)⁷⁹; 3) atypical Alzheimer’s dementia: participants diagnosed with an atypical presentation of AD, including the visual (posterior cortical atrophy⁸⁰), language (logopenic progressive aphasia⁷), behavioral⁸¹ variants of AD; and 4) other dementia: participants who do not fulfill criteria for one of the three main diagnostic groups, including participants meeting clinical criteria for dementia with Lewy bodies,⁸² vascular dementia,⁸³ frontotemporal dementia,⁸⁴ primary progressive aphasia,^85,86 parkinsonian dementia syndromes,^87,88 and participants with a dementia that was hard to classify.

Apolipoprotein E testing was performed as previously described.⁸⁹

Neuropathological evaluations

All participants underwent standard neuropathological examination performed by a board-certified neuropathologist (DWD, RRR or AN) as previously described in detail.³² Each case had been assigned a Braak neurofibrillary tangle (NFT) stage¹⁶ using anti-tau antibodies (mouse monoclonal clone AT8, 1:1,000 dilution; Thermo Fisher Scientific),¹⁶ and a Thal phase¹⁵ using antibodies to β-amyloid (mouse monoclonal clone 6F/3D, 1:10 dilution; DAKO, Carpinteria, CA). Using the National Institute of Aging-Alzheimer’s Association recommendations,^41,90 we classified participants as having Alzheimer’s disease if the Thal β-amyloid phase was 1–5¹⁵ and the Braak NFT stage was 0-VI. Hence, all participants in the study had evidence of β-amyloid deposition in the brain and met criteria for either low, intermediate or high AD. Participants were excluded if they had neuronal loss and gliosis affecting frontal and/or temporal lobe consistent with frontotemporal lobar degeneration⁴² regardless of subtype.⁹¹

Neuroimaging analysis

All participants underwent a structural head MRI and flortaucipir PET using consistent protocols. Flortaucipir PET scans were acquired using a PET/CT scanner. Approximately 370MBq (range 333–407 MBq) of [¹⁸F]-flortaucipir was administered, followed by a 20-min PET acquisition performed 80 min after injection. The scans consisted of four 5-min dynamic frames following a low dose CT transmission scan. The PET images were reconstructed on-scanner using iterative algorithms (OSEM) with standard corrections and a 5 mm Gaussian post-reconstruction filter. All participants also underwent a 3T head MRI protocol using GE scanners that included a magnetization prepared rapid gradient echo (MPRAGE) sequence (TR/TE/TI, 2300/3/900 ms; flip angle 8°, 26-cm field of view (FOV); 256 × 256 in-plane matrix; phase FOV of 0.94; slice thickness of 1.2 mm). The MRI scans were performed a median of one day from the flortaucipir PET scans.

The MRI and PET images were processed on the high-performance computing environment (mForge) hosted by the National Center for Supercomputing Applications using NRG in-house developed pipelines (RS). The four-frame PET images were averaged after realignment and then rigidly registered to the MPRAGE. The MPRAGE scans were segmented into gray and white matter using templates and settings from the Mayo Clinic Adult Lifespan Template (MCALT: https://www.nitrc.org/projects/mcalt/) and SPM12. The Human Brainnetome atlas was transformed to MPRAGE space and used to output median flortaucipir PET uptake in the four primary isocortices of interest: motor, somatosensory, auditory (Brodman area 41_42) and visual (occipital polar cortex, OPC) isocortex. The motor cortex was also subdivided into four sub-ROIs (Brodman area 4 head/face (A4hf), upper limb (A4ul), trunk (A4t), and tongue/larynx (A4tl) regions) and similarly the somatosensory cortex was subdivided into four sub-ROIs (Brodman area 1, 2 and 3 upper limb & head and face region (A1/2/3ulhf), tongue/larynx (A1/2/3tonla), trunk (A1/2/3tru), Brodman area 2 (A2)) (Figure 1). We also assessed the ERC and ITL gyrus given that flortaucipir PET uptake is observed in these regions early in ADNC. We did not analyze the olfactory isocortex (Brodman area 28) since the atlas included Brodman area 28 as part of the ERC ROI. We did not analyze the gustatory isocortex (Brodman area 43) since the atlas does not include an ROI for this region. Median uptake within the gray and white matter masks for each ROI were averaged between left and right hemispheres and divided by uptake in the cerebellar crus gray matter to create SUVRs.

Quantification and statistical analysis

We used generalized additive models (GAMs) to estimate mean log-transformed SUVR as a smooth function of Braak stage.⁹² We fit separate GAM models for each region of interest and set the basis dimension to be k = 6; this allowed us to estimate the underlying relationship between Braak stage and SUVR in a data-driven way without prespecifying a particular functional form (e.g., linear or quadratic). The model estimates were then exponentiated and interpreted as the geometric mean SUVR. We calculated empirical first derivatives of the resulting mean curves to further characterize the change in slope of the GAM estimates over Braak NFT stages. One benefit of this novel approach is that it characterizes the relationship between increasing SUVR and Braak stage that is independent of SUVR level. This facilitates comparisons across regions which may have different average SUVR values. One benefit of this summary is that it is independent of SUVR level, thus facilitating comparisons across regions which may have different average SUVR values. We used the variance-covariance estimates from the fitted model to contrast adjacent Braak levels; the contrasts are summarized with estimates, 95% CIs, and p-values using the back-transformation 100 × (exp(x) – 1). We also fit all pairwise bivariate-outcome GAMs to enable us to directly compare estimates between any two regions. We used linear regression models of log-transformed SUVR at the last scan to evaluate associations between age, sex, APOE ε4 carriership, and diagnosis and tau PET SUVR. These models included time from last scan to death to control potential confounding due to differences in this interval. We analyzed longitudinal change in SUVR using Bayesian linear mixed effects models with log-transformed SUVR as the outcome and using individuals with two or more tau PET scans. We fit separate models for each region and modeled Braak stage as a random effect with uncorrelated intercepts and slopes.⁹³ This treatment of Braak stage allowed us to get stable estimates of geometric mean SUVR at the last scan and mean annual percentage change in SUVR over time for all Braak levels. These models also included uncorrelated random intercepts and random slopes for participants. Finally, we fit region-specific linear mixed effects models among Braak VI individuals with at least two scans. These models were estimated via restricted maximum likelihood (REML) and included uncorrelated random intercepts and slopes for participants and included age, sex, APOE ε4 carriership, and diagnosis, and controlled for time from last scan to death. All analyses were performed using R version 4.4.2. GAM models were estimated using the mgcv package, Bayesian linear mixed effects models were estimated using the brms package, and linear mixed effects models estimated via REML were fit via the lme4 package.