Neocortical tau burden determines the degree of cognitive impairment in individuals with Braak stage V neurofibrillary degeneration

Abstract

Alzheimer disease neuropathologic change (ADNC) is considered to be the most common cause of cognitive decline and dementia worldwide. ADNC level is determined using the density of neuritic plaques in combination with the topographical distribution of β-amyloid (Aβ) plaques and hyperphosphorylated tau (p-tau)-positive neurofibrillary tangles (NFTs). While cognitive decline correlates with the level of ADNC, there remains a great deal of variation in cognitive outcomes between individuals that is unaccounted for by current neuropathologic evaluation metrics. We leveraged quantitative computer-assisted positive pixel assessments to establish the neocortical p-tau burden in the middle frontal and superior temporal gyri of 61 individuals with Braak NFT stage V who had a wide range of cognitive outcomes and trajectories. Frontal and temporal neocortical p-tau burden varied between 0.2% and 53.7%. Both frontal and temporal p-tau burden directly affected cognitive outcome and correlated with function of multiple cognitive domains, including measures of language/semantic memory and attention/working memory. In multivariable analysis, only p-tau burden and microinfarcts significantly impacted cognitive decline, while Aβ, limbic-predominant age-related TDP-43 encephalopathy, Lewy body pathology, and other measures of cerebrovascular disease did not. Additionally, individuals with low mean neocortical p-tau burden (≤ 13%) had significantly better longitudinal cognitive trajectories over the final 15 years of life compared to those with high burden (≥ 23.5%). These results suggest that while all individuals with Braak stage V have some degree of neurofibrillary degeneration in the neocortex, the significant variation in cognitive decline observed between these individuals can be partially understood as a reflection of the variation in quantitatively assessed neocortical p-tau burden, which had a greater impact on progression to dementia than common comorbid neuropathologies associated with dementia risk. This argues for the incorporation of the density of ADNC-related pathology, in addition to its regional location, as an adjunct to future staging systems for Alzheimer disease.

Supplementary Information

The online version contains supplementary material available at 10.1007/s00401-026-03031-4.

Introduction

Since the first description of Alzheimer disease (AD) in the early twentieth century [5], it has been recognized as the most common neuropathologic finding underlying cognitive decline in older adults [57]. Currently, AD is thought to be the 7th leading cause of death in the United States, affecting approximately 7.2 million Americans aged 65 or older, a number which is expected to grow to nearly 14 million by the year 2060 [1]. Alzheimer disease neuropathologic change (ADNC), the histopathologic substrate of AD, consists of the deposition of two interrelated protein deposits, hyperphosphorylated tau (p-tau) neurofibrillary tangles (NFTs) and β-amyloid (Aβ) plaques in relatively consistent patterns throughout the brain [19]. NFTs appear to begin initially in the brainstem early in life before progressing through the entorhinal cortex, hippocampus and limbic structures, and into the neocortex in successive Braak stages [2, 14–17], although the earlier appearance of locus coeruleus pretangles might also represent the first stage of primary age-related tauopathy (PART) or even a non-specific injury reaction [7, 27, 50]. Aβ plaque deposition begins in the neocortex and proceeds in the opposite direction, extending into the entorhinal cortex and hippocampal structures, basal ganglia, brainstem, and cerebellum in well-defined Thal phases [65]. These two measures, in conjunction with the density of neuritic Aβ plaques in the frontal, temporal, and parietal lobes (CERAD neuritic plaque [NP] score) [38], are used to determine the overall level of ADNC [41].

The level of ADNC largely correlates with the degree of global cognitive impairment [43], although there is significant variation in cognitive trajectories and clinical outcomes between individuals, even among those with the most severe ADNC [49], and a significant portion of cognitive variation is unaccounted for by recorded neuropathologic data [13, 21, 36]. This paradigm is further complicated by several major confounds. The first is the frequent presence of comorbid neurodegenerative pathologies, including limbic-predominant age-related TDP-43 encephalopathy (LATE) [44], Lewy body pathology [8], and various forms of cerebrovascular disease (CVD) [9, 48, 69], which may interact in an additive or synergistic manner with ADNC, or may mimic some of the symptoms typically associated with ADNC [3, 4, 9, 28, 29, 36, 48, 49, 52, 53, 60, 69, 72]. On the other end of the spectrum are cognitive resilience and reserve, which are generally defined as an individual’s ability to remain cognitively intact despite harboring pathology which would otherwise be expected to cause dementia and the factors which provide this apparent tolerance to pathology [40, 49, 62, 63, 67]. Also important is the concept of resistance, which is the absence of a certain pathologic process which might otherwise be expected at a given age, and some amount of resilience against ADNC might actually represent resistance to the development of certain common comorbid aging-related neuropathologies [11, 69].

Another factor which might lead to a disconnect between ADNC level and cognition is the low precision of widely used neuropathologic assessment protocols and the fact that the Braak staging system primarily relies upon the topographical distribution of neurofibrillary degeneration without fully accounting for the potential variation in NFT burden or density in a given region, although some authors have suggested semi-quantitative assessments for these protein deposits [2]. For example, a case may be classified as Braak NFT stage V in the presence of relatively few NFTs in a single area of the neocortex or with dense NFT burden throughout the majority of the neocortex. Numerous studies have demonstrated that the p-tau burden and NFT density may actually correlate better than Braak NFT stage in some circumstances (including in both ADNC and PART) [10, 12, 26, 31, 37, 43, 46, 47, 68], and this could explain some of the variation found within ADNC levels or Braak stages [43, 49]. In this study, we investigated neuropathologic features which contribute to variation in cognitive decline in a cohort of 61 individuals with Braak stage V. We leveraged computer-assisted positive pixel counts to quantitatively evaluate the overall p-tau burden (which included p-tau-immunoreactive neurofibrillary tangles, neuritic plaques, and neuropil threads) in the middle frontal and superior temporal gyri to understand how the density, in addition to the location, affects the cognitive status, in the context of a number of neurodegenerative comorbidities.

Methods

Cohort and case selection

600 cases were randomly selected from the National Alzheimer’s Coordinating Center (NACC) database, including 200 individuals with normal cognition, 200 with mild cognitive impairment (MCI), and 200 with dementia, irrespective of reported neuropathologic findings. For each of these cases, 10 unstained slides from the middle frontal gyrus, superior temporal gyrus, calcarine cortex, hippocampus at the level of the lateral geniculate nucleus (LGN), amygdala, and pons (with locus coeruleus) were requested from 23 Alzheimer’s Disease Research Centers (ADRCs). 538 total cases were received at Boston University from: Banner Health, Boston University, Columbia University, Emory University, Icahn School of Medicine at Mount Sinai, Johns Hopkins University, Massachusetts ADRC, Mayo Clinic, New York University, Northwestern University, Oregon Health & Science University, Rush University, University of California, Davis, University of California, Irvine, University of California, San Diego, University of California, San Francisco, University of Kentucky, University of Pennsylvania, University of Pittsburgh, University of Southern California, University of Washington, University of Wisconsin, and Washington University in St. Louis, provided with funding from the National Institute on Aging (NIA) (RF1 AG062348). Written informed consent was provided for each participant at each ARDC of origin. Of these cases, 61 met the inclusion criteria for this study: (1) Braak NFT stage V (originally diagnosed at their respective ADRCs of origin) and (2) last cognitive exam during which global Clinical Dementia Rating (CDR) and CDR Sum of Boxes (CDR-SB) were assessed within the final 24 months of an individual’s life (Table 1).

Table 1.

Demographic, genetic, and neuropathologic features in Braak stage V individuals with impaired and retained cognition

	All cases	Cognitively unimpaired	Cognitively impaired	p-valueg
Demographic features
n	61	20	41	–
Age	87.6 ± 0.8	88.7 ± 1.3	87.0 ± 1.0	0.3565
Sex (M|F)	33|28	9|11	24|17	0.3193
Education level (years)	15.5 ± 0.4	15.3 ± 0.8	15.6 ± 0.5	0.6806
Race (% white)a	88.5%	95.0%	85.4%	0.2678
Total cognitive evaluations	4.6 ± 0.4	6.2 ± 1.0	3.9 ± 0.4	0.0114
Time between final evaluation and death (months)	8.2 ± 0.8	8.6 ± 1.5	8.1 ± 1.0	0.7729
ADNC-related pathology
ADNC Level (0|1|2|3)b	1|0|10|47	1|0|4|14	0|0|6|33	0.2895
Thal phase (0|1|2|3|4|5)c	1|0|0|8|38|11	1|0|0|4|11|3	0|0|0|4|27|8	0.3165
CERAD NP score (0|1|2|3)	1|6|18|36	1|2|7|10	0|4|11|26	0.4293
Frontal PPP (%)	10.8 ± 1.5	5.3 ± 1.4	13.4 ± 2.0	0.0121
Temporal PPP (%)	22.9 ± 1.8	16.6 ± 2.3	25.8 ± 2.3	0.0167
Average PPP (%)	19.5 ± 1.7	12.7 ± 1.9	22.8 ± 2.1	0.0035
CAA (0|1|2|3)	18|20|16|7	7|6|4|3	11|14|12|4	0.7682
Other neuropathologic features
LBD stage (0|1|2|3)	41|6|5|9	12|4|2|2	29|2|3|7	0.2692
LATE stage (0|1|2|3)	39|2|20|0	16|0|4|0	23|2|16|0	0.1604
Hippocampal sclerosisd	20.0%	15.0%	22.5%	0.4936
Infarcts	23.0%	15.0%	26.8%	0.3024
Microinfarcts	21.3%	10.0%	26.8%	0.1319
Arteriolosclerosis (0|1|2|3)	12|19|22|8	3|5|9|3	9|14|13|5	0.6984
Genetic features
APOE statuse
≥ 1 ε2 Allele	14.3%	22.2%	9.7%	0.2264
≥ 1 ε4 Allele	61.2%	38.9%	74.2%	0.0145
ε2/2 | ε2/3 | ε2/4 | ε3/3 | ε3/4 | ε4/4	0|4|3|15|24|3	0|3|1|8|6|0	0|1|2|7|18|3	0.1127
Clinical featuresf
Global CDR	1.7 ± 0.1	0.4 ± 0.1	2.3 ± 0.1	< 0.0001
CDR sum of boxes	9.5 ± 0.8	1.7 ± 0.3	13.3 ± 0.7	< 0.0001
MMSE	19.7 ± 1.1	27.5 ± 0.5	13.8 ± 1.0	< 0.0001
Logical memory immediate	5.4 ± 0.7	9.6 ± 1.0	1.8 ± 0.5	< 0.0001
Logical memory delayed	3.8 ± 0.6	7.0 ± 1.2	1.0 ± 0.4	0.0003
Digit span forward	6.6 ± 0.4	7.9 ± 0.5	5.6 ± 0.4	0.0169
Digit span backward	4.5 ± 0.3	6.3 ± 0.4	2.9 ± 0.3	< 0.0001
Trail making test part A	73.8 ± 5.4	57.1 ± 6.8	91.7 ± 7.1	0.0188
Trail making test part B	221.0 ± 12.0	186.8 ± 21.1	272.2 ± 10.8	0.0223
WAIS digit substitution score	25.7 ± 2.0	33.7 ± 3.2	16.9 ± 2.0	0.0071
Animals	9.8 ± 0.7	13.3 ± 1.1	7.1 ± 0.8	0.0004
Vegetables	6.5 ± 0.6	9.6 ± 0.8	3.9 ± 0.5	< 0.0001
Boston naming test	19.4 ± 1.1	25.0 ± 0.9	14.4 ± 1.4	0.0003

Bold denotes statistical significance after correcting for multiple comparisons

^aGiven the paucity of non-white participants, statistics for race are calculated as "white" and "non-white"

^bADNC level is unavailable for 3 cases

^cThal phase is unavailable for 3 cases, in 25 cases Thal phase is recorded as “4 + ” due to absence of cerebellar sections, included here as Thal phase 4

^dHippocampal sclerosis is unavailable for 1 case

^eAPOE data is unavailable for 12 cases

^fClinical features were variably available at the final cognitive exam, see Supplemental Table 1 for each respective n

^gp-values were adjusted for multiple comparisons with the Benjamini, Krieger, and Yekutieli method, p-values less than 0.0038 are considered significant

Immunohistochemistry

Formalin-fixed paraffin-embedded (FFPE) tissue sections cut at 5 µm from each of the six regions were mounted on charged slides and baked at 70 °C. Luxol fast blue/hematoxylin & eosin (LFB/H&E) and immunohistochemical stains for hyperphosphorylated tau (p-tau; AT8) were applied to one section from each region. Immunohistochemical stains for Aβ (4G8) were applied to the middle frontal gyrus, superior temporal gyrus, and hippocampal sections. Immunohistochemical stains for phosphorylated TAR DNA binding protein 43 (p-TDP-43) were applied to the amygdala, hippocampus, and middle frontal gyrus sections. Immunohistochemical stains for phosphorylated α-synuclein were applied to the pons, amygdala, hippocampal, and middle frontal gyrus sections. Bielschowsky silver stains were applied to the amygdala, hippocampus, middle frontal gyrus, superior temporal gyrus, and calcarine cortex sections. Sections were stained on a Leica Bond III automated stainer (Leica Biosystems), according to the manufacturer’s protocols and imaged using an Akoya Bioscience Vectra Polaris Digital Slide Scanner at Boston University.

Computer-assisted quantitative assessments of p-tau

Gray matter was segmented and annotated on each available section of frontal or temporal neocortex using Aperio ImageScope software by hand using the Pen Tool (Supplemental Figure S1). Defects in the sections (including artifacts, such as tears and tissue folds) which may cause false positive or false negative pixel counts were excluded where possible using the Aperio ImageScope Negative Pen Tool, as previously described [68]. P-tau burden was determined using the Aperio ImageScope positive pixel count (Version 9) using default parameters (intensity threshold for weak positive pixels [upper limit] = 220, intensity threshold for weak positive pixels [lower limit] = 175, intensity threshold for medium positive pixels [lower limit] = 100, intensity threshold for strong positive pixels [lower limit] = 0, hue value 0.1, hue width 0.5, color saturation threshold 0.04), and the same thresholds were used for each slide (Fig. 1) [25, 68]. The positive pixel percentage (PPP) for the selected region was calculated as: 100 × (number of strong positive + number of positive) / (number of weak positive + number of positive + number of strong positive + number of negative). For each case, separate values were calculated for the frontal neocortex, temporal neocortex, and an average, consisting of the PPP calculated over both of these regions. Notably, one case was missing usable superior temporal lobe tissue and 2 were missing usable middle frontal gyrus tissue.

Fig. 1.

Examples of low, intermediate, and high neocortical hyperphosphorylated tau (p-tau) burden quantification in individuals with Braak stage V. The top row of panels shows an overview of neocortical sections, the middle row shows a high-power view of select regions, and the bottom row shows examples of positive pixel percentage (PPP) quantification performed with Aperio ImageScope software. Scale bars = 4 mm in top row, 500 µm in bottom two rows

Additional NACC dataset data

Corresponding cognitive, genetic, and neuropathologic data for this study were downloaded with permission from the NACC, established with funding from the National Institute on Aging (NIA) (U24 AG072122). NACC data, including standardized Uniform Data Set (UDS) version 3 variables, Neuropathology (NP) Data Set version 11 variables, and Genetic Data Set variables, were obtained with permission on November 25, 2024 (Data Request Number 13611) [28, 49].

Demographic and cognitive variables

Demographic and cognitive variables for each individual were extracted from the UDS, and included each participant’s sex (UDS variable SEX), age at death (UDS variable NACCDAGE), age at cognitive exam (UDS variable NACCAGE), level of education in years (UDS variable EDUC), and race (UDS variable RACE). The number of cognitive exams was determined by the total number of all UDS visits made (UDS variable NACCAVST) and the specific UDS visit number (UDS variable NACCVNUM). The approximate time between each participant’s death and a given cognitive exam was calculated using variables for month of death (UDS variable NACCMOD), year of death (UDS variable NACCYOD), visit month (UDS variable VISITMO), and visit year (UDS variable VISITYR). Cognitive testing for each individual was extracted from the UDS at each clinical encounter for which a particular variable was available, and assessed using global CDR (UDS variable CDRGLOB), CDR-SB (UDS variable CDRSUM), Mini-Mental State Examination (MMSE) (UDS variable NACCMMSE), logical memory immediate recall (LMI) (UDS variable LOGIMEM), logical memory delayed recall (LMD) (UDS variable MEMUNITS), digit span forward (DSF) (UDS variable DIGIF), digit span backward (DSB) (UDS variable DIGIB), Trail Making Test Part A (TMT-A) (UDS variable TRAILA), Trail Making Test Part B (TMT-B) (UDS variable TRAILB), Wechsler Adult Intelligence Scale Digit Symbol Substitution Test (WAIS DS) (UDS variable WAIS), animal naming fluency (UDS variable ANIMALS), vegetable naming fluency (UDS variable VEG), and Boston Naming Test, 30 odd items (BNT) (UDS variable BOSTON), as previously described in detail [3, 4, 28, 29, 36, 49, 64].

Genetic and neuropathologic variables

APOE genotype (ε2ε2/ε2ε3/ε2ε4/ε3ε3/ε3ε4/ε4ε4) for each participant was determined using the variable NACCAPOE. Each neurodegenerative pathology was defined using NACC NP dataset variables, as previously described in detail [28, 36]. ADNC was determined using the variable NPADNC and confirmed using Braak NFT stage (NP variable NACCBRAA) [14], Thal phase (NP variable NPTHAL) [65], and CERAD NP score (NP variable NACCNEUR) [38]. These values were confirmed using the provided slides and ADNC level was reevaluated according to National Institute on Aging-Alzheimer’s Association consensus criteria where applicable [41]. Missing variables were reassessed where possible and filled in where appropriate. In some instances and for certain measures we did not have access to immunohistochemical stains to fully evaluate these pathologies; for example, restained Aβ immunohistochemistry was only available for neocortical and hippocampal sections, so for cases with missing Thal phase, only phase 0–4 could be assigned. Cerebral amyloid angiopathy (CAA) was categorized as none, mild, moderate, or severe (NP variable NACCAMY).

Lewy body stage was assessed using the NP variables NACCLEWY and NPLBOD and categorized as no Lewy body pathology (stage 0), brainstem-predominant (stage 1), limbic [transitional] (stage 2), or diffuse neocortical stage (stage 3) [8]. The presence/absence of frontotemporal lobar degeneration with TDP-43 (FTLD-TDP) and amyotrophic lateral sclerosis (ALS) was determined with the NP variables NPFTDTDP and NPALSMND, respectively. LATE stage was derived from the NP variables NPTDPB, NPTDPC, and NPTDPE and categorized as no TDP-43 pathology (stage 0), amygdala only (stage 1), + hippocampus (stage 2), or + middle frontal gyrus (stage 3) [44, 45]. Presence/absence of hippocampal sclerosis was determined with the NP variable NPHIPSCL and dichotomized as negative vs positive (including unilateral/bilateral/present, but laterality not assessed). Presence/absence of Pick’s disease (PiD) was determined with the NP variable NACCPICK. Presence/absence of progressive supranuclear palsy (PSP) pathology was determined using the NP variable NACCPROG. Presence/absence of corticobasal degeneration (CBD) pathology was determined using the NP variable NACCCBD. Various measures of cerebrovascular disease (CVD) were evaluated according to the presence/absence of infarcts/lacunes (NP variable NACCINF), presence/absence of microinfarcts (NP variable NACCMICR), and severity of arteriolosclerosis (NP variable NACCARTE) (none, mild, moderate, or severe) [28, 36, 69]. Missing variables were assessed and documented values were confirmed using the provided slides where possible.

Statistics and data analysis

Mixed-effects regression modeling with random intercepts and random slopes for time per participant was performed in R to compare slope differences of cognitive trajectories over time between individuals with low, intermediate, and high p-tau burdens, as previously described in detail [3, 4, 49]. Cognitive tests (Global CDR, CDR-SB, and MMSE) were considered dependent variables, while age and group were included as fixed effects to test for overall differences between groups. To test for slope differences, an interaction group*time was included in the model. Cases with missing values in any predictor variable were excluded from analysis. The "readxl" package was used to import data, the "dplyr" package was used for data wrangling, the "lme4" package was used to fit mixed-effects models that account for repeated measurements from the same participant over time, the "lmerTest" package was used to provide significance testing for the mixed effects models, the "emmeans" package was used for post-hoc analysis, and the "broom.mixed" package was used for output formatting. The "lstrends" function from the "emmeans" package was used to compare pairwise slopes.

Multivariable logistic regression analysis was performed with MedCalc (MedCalc Software Ltd, Ostend, Belgium), adjusted for age at death, sex, years of education, time between final cognitive evaluation and death, and ADRC of origin. Low p-tau burden was used as the reference category for p-tau burden. All other univariate analyses were performed using GraphPad Prism and all graphs were made using GraphPad Prism version 10 (GraphPad Software, Inc., La Jolla, CA, USA). Differences in categorical variables were calculated using the Chi-squared test. Differences in continuous variables were evaluated using multiple t-tests. Correlations between p-tau burden, additional neuropathologic features, and cognitive testing were made with linear regression modeling using the Pearson correlation coefficient [54, 68]. The sample size for each comparison is given in Supplemental Table S1. The level of significance for p-values for these tests was adjusted for multiple testing using the two-stage linear step-up procedure of Benjamini, Krieger, and Yekutieli method. Longitudinal trajectories were visualized with spaghetti plots and spline-smoothed fit lines.

Results

Cohort description

Represented in this cohort were 33 males and 28 females with an average age of 87.6 years, 88.5% of whom were Caucasian, and on average displayed significant impairments in measures of global cognition and across cognitive domains (Table 1), although there was a wide range of cognitive variability in these Braak V cases (Supplemental Figure S2). According to our defined inclusion criteria, all cases were Braak NFT stage V. A single case was completely devoid of Aβ, but all other cases had either “intermediate” or “high” level ADNC according to consensus criteria [41], and had a 19.5 ± 1.7% average neocortical p-tau burden across assessed frontal and temporal lobe sections. The distribution of APOE alleles was consistent with known ranges for individuals with ADNC [50]. Comorbid neuropathologies were also frequent; 32.8% of cases had LATE Stage 2–3, 23.0% of cases had either limbic-predominant or diffuse neocortical Lewy body pathology, 49.2% of cases had moderate or severe arteriolosclerosis, and 37.7% of cases had some form of infarct or microinfarct (Table 1). There was no evidence of PiD, PSP, CBD, FTLD-TDP, or ALS in any included case.

Frontal and temporal neocortex p-tau burden assessment

Comparison between temporal and frontal neocortical p-tau burden was possible in 58/61 cases. The p-tau burden (mean positive pixel percentage [PPP]) was significantly higher in the temporal (22.9 ± 1.8%) compared to the frontal neocortex (10.8 ± 1.5%; p < 0.0001) (Fig. 2a). In the vast majority of individuals (84.5%; 49/58 cases) the temporal p-tau burden was higher than the frontal burden (Fig. 2b), although there was significant variability in this (Fig. 2b–d), with an overall average difference of 13.6 ± 1.2% between the two regions.

Fig. 2.

Hyperphosphorylated tau (p-tau) quantification between temporal and frontal neocortical sections in individuals with Braak stage V. a positive pixel percentage (PPP) quantification demonstrated significantly higher mean temporal p-tau burden compared to mean frontal p-tau burden (p < 0.0001). b paired temporal-frontal PPP quantification demonstrated 84.5% of cases (49/58) with higher temporal lobe p-tau burden. Red lines represent cases in which temporal lobe p-tau burden was higher than frontal lobe p-tau burden, while blue lines represent cases in which frontal lobe p-tau burden was higher than temporal lobe p-tau burden. c-d Examples of paired temporal-frontal neocortical sections with varying p-tau burden between the two regions. Scale bars = 5 mm

Global cognitive and neuropsychological associations of frontal and temporal neocortical p-tau burden

In unadjusted univariate analysis, global CDR was significantly correlated with the neocortical p-tau burden in the frontal lobe (p = 0.0003), temporal lobe (p < 0.0001), and the average overall p-tau (p < 0.0001) (Fig. 3a). CDR-SB was also significantly correlated with the neocortical p-tau burden in the frontal lobe (p = 0.0018), temporal lobe (p < 0.0001), and the average overall p-tau (p < 0.0001) (Fig. 3b). Similarly, MMSE was significantly correlated with the neocortical p-tau burden in the frontal lobe (p = 0.0253), temporal lobe (p = 0.0027), and the average overall p-tau (p = 0.0012) (Fig. 3c). A similar assessment was performed for each neuropsychological test. In univariate analysis, significant correlations were identified between average p-tau burden and LMI (p = 0.0028), LMD (p = 0.0111), DSF (p = 0.0094), DSB (p = 0.0020), TMT-A (p = 0.0372), animal naming (p = 0.0002), vegetable naming (p = 0.0032), and BNT (p = 0.0012). Additional significant correlations were identified between temporal p-tau burden and LMI (p = 0.0255), DSF (p = 0.0167), DSB (p = 0.0038), animal naming (p = 0.0004), vegetable naming (p = 0.0479), and BNT (p = 0.0011). Frontal p-tau burden correlated with DSF (p = 0.0003), DSB (p = 0.0217), animal naming (p = 0.0003), vegetable naming (p = 0.0126), and BNT (p = 0.0227). No measure of neocortical p-tau burden associated with TMT-B or WAIS digit substitution testing (Supplemental Figure S3).

Fig. 3.

Cognitive impairment correlated with neocortical hyperphosphorylated tau (p-tau) burden quantification in individuals with Braak stage V. a Global CDR, b CDR-SB, and c MMSE were highly correlated with the level of frontal, temporal, and overall average PPP using linear regression models

Longitudinal cognitive assessment

Next, we divided the cohort into three groups, those with “low” p-tau burden (the lowest 1/3 of average p-tau PPP; 0–13% PPP; n = 20), “intermediate” p-tau burden (the middle 1/3 of average p-tau PPP; 13–23.5% PPP; n = 20), and “high” p-tau burden (the highest 1/3 of average p-tau PPP; 23.5–51.2% PPP; n = 21). Those with low p-tau burden were significantly older than those with intermediate (p = 0.0009) or high (p < 0.0001) p-tau burden, but no differences were noted in terms of participant sex, level of education, race, overall ADNC level, Thal phase, CERAD NP score, LATE stage, Lewy body stage, CAA, cerebrovascular disease, or APOE allele status (Table 2). Individuals with low p-tau burden had significantly better global CDR (p < 0.0001), CDR-SB (p < 0.0001), and MMSE (p = 0.0004) at the final cognitive evaluation compared to those with high p-tau burden, but no significant difference was noted between those with low and intermediate p-tau burden or intermediate and high p-tau burden after adjusting the level of significance for multiple comparisons. Individuals with low p-tau burden also had significantly better animal naming (p = 0.0003) and Boston naming test (p = 0.0006) scores compared to those with high tau burden, but no other significant differences were noted after adjusting for multiple comparisons (Table 2).

Table 2.

Demographic, genetic, neuropathologic, and clinical features in Braak stage V individuals with low, intermediate, and high average p-tau positive pixel percentage

	Low p-Tau PPP	Intermediate p-Tau PPP	High p-Tau PPP	p-valueg (Low vs Int.)	p-valueg (Low vs High)	p-valueg (Int. vs High)
Demographic features
n	20	20	21	-	-	-
Age	92.5 ± 1.4	85.8 ± 1.2	84.6 ± 0.9	0.0009	< 0.0001	0.4155
Sex (M|F)	11|9	10|10	12|9	0.7515	0.8901	0.6466
Education level (years)	14.6 ± 0.6	16.0 ± 0.9	15.9 ± 0.7	0.2227	0.1962	0.9340
Race (% white)a	100%	85.0%	81.0%	0.0717	0.0399	0.7306
Total cognitive evaluations	5.8 ± 1.0	3.8 ± 0.5	4.3 ± 0.7	0.0716	0.2316	0.4994
Time between final evaluation and death (months)	7.9 ± 1.7	9.5 ± 1.4	7.5 ± 1.3	0.4696	0.8616	0.3178
ADNC-related pathology
ADNC level (0|1|2|3)b	1|0|6|12	0|0|2|18	0|0|2|17	0.1239	0.1450	0.9568
Thal phase (0|1|2|3|4|5)c	1|0|0|5|10|3	0|0|0|2|15|3	0|0|0|1|13|5	0.3529	0.2072	0.6215
CERAD NP score (0|1|2|3)	1|3|7|9	0|2|5|13	0|1|6|14	0.5201	0.3703	0.8037
Frontal PPP (%)	2.5 ± 0.8	6.7 ± 1.2	21.9 ± 2.6	0.0074	< 0.0001	< 0.0001
Temporal PPP (%)	8.0 ± 1.1	22.2 ± 1.1	36.9 ± 2.3	< 0.0001	< 0.0001	< 0.0001
Average PPP (%)	5.9 ± 0.8	18.3 ± 0.8	33.6 ± 2.0	< 0.0001	< 0.0001	< 0.0001
CAA (0|1|2|3)	4|7|7|2	6|6|5|3	8|7|4|2	0.7988	0.5462	0.8850
Other neuropathologic features
LBD stage (0|1|2|3)	14|3|2|1	12|2|2|4	15|1|1|4	0.5411	0.3698	0.8070
LATE Stage (0|1|2|3)	14|0|6|0	12|1|7|0	13|1|7|0	0.5404	0.5798	0.9922
Hippocampal sclerosisd	15.0%	20.0%	25.0%	0.6773	0.4292	0.7050
Infarcts	25.0%	20.0%	23.8%	0.7050	0.9293	0.7683
Microinfarcts	30.0%	20.0%	14.3%	0.7303	0.2243	0.6269
Arteriolosclerosis (0|1|2|3)	4|5|7|4	3|9|7|1	5|5|8|3	0.3786	0.9607	0.4425
Genetic features
APOE Statuse
≥ 1 ε2 Allele	29.4%	0%	12.5%	0.0185	0.2350	0.1441
≥ 1 ε4 Allele	41.2%	75.0%	68.8%	0.0494	0.1119	0.6942
ε2/2 | ε2/3 | ε2/4 | ε3/3 | ε3/4 | ε4/4	0|4|1|6|6|0	0|0|0|4|12|0	0|0|2|5|6|3	0.0608	0.1162	0.0684
Clinical featuresf
Global CDR	0.9 ± 0.2	1.6 ± 0.2	2.5 ± 0.2	0.0237	< 0.0001	0.0130
CDR sum of boxes	5.0 ± 1.2	9.4 ± 1.4	14.0 ± 1.1	0.0105	< 0.0001	0.0063
MMSE	25.5 ± 1.1	19.0 ± 1.6	13.1 ± 2.0	0.0116	0.0004	0.0920
Logical memory immediate	9.2 ± 1.2	3.9 ± 1.0	2.6 ± 0.9	0.0144	0.0138	0.5012
Logical memory delayed	7.1 ± 1.3	2.2 ± 0.8	1.9 ± 1.1	0.0137	0.0610	0.8484
Digit span forward	7.8 ± 0.4	6.5 ± 0.6	5.3 ± 0.8	0.1806	0.0522	0.3544
Digit span backward	6.5 ± 0.5	3.7 ± 0.5	3.1 ± 0.4	0.0048	0.0038	0.5766
Trail making test part A	60.7 ± 5.0	70.8 ± 10.5	106.8 ± 10.5	0.5074	0.0121	0.1415
Trail making test part B	185.3 ± 21.8	227.4 ± 21.4	292.5 ± 3.3	0.3310	0.0534	0.2087
WAIS digit substitution score	31.7 ± 3.1	26.3 ± 3.7	13.3 ± 2.5	0.4712	0.0540	0.1677
Animals	13.6 ± 1.1	9.2 ± 1.2	5.2 ± 0.7	0.0322	0.0003	0.0534
Vegetables	9.8 ± 0.9	5.2 ± 0.9	4.1 ± 0.7	0.0054	0.0038	0.5140
Boston naming test	24.6 ± 1.3	19.0 ± 1.9	11.1 ± 1.8	0.0725	0.0006	0.0655

Bold denotes statistical significance after correcting for multiple comparisons

^aGiven the paucity of non-white participants, statistics for race are calculated as "white" and "non-white"

^bADNC level is unavailable for 3 cases

^cThal phase is unavailable for 3 cases, in 25 cases Thal phase is recorded as “4 + ” due to absence of cerebellar sections, included here as Thal phase 4

^dHippocampal sclerosis is unavailable for 1 case

^eAPOE data is unavailable for 12 cases

^fClinical features were variably available at the final cognitive exam, see Supplemental Table 1 for each respective n

^gp-values were adjusted for multiple comparisons with the Benjamini, Krieger, and Yekutieli method, p-values less than 0.0017 are considered significant

We then assessed the longitudinal cognitive trajectories between these three groups using a combination of spaghetti plots and spline-smoothed fit lines to visualize cognitive decline over the final 15 years of life (Fig. 4), as well as mixed-effects linear regression modeling (Supplemental Table S2). The individuals with high p-tau burden had significantly more rapid and significant cognitive decline compared to those with low p-tau burden in terms of global CDR (p < 0.0001), CDR-SB (p = 0.0009), and MMSE (p = 0.0004). Those with intermediate tau burden had more rapid and significant cognitive decline compared to those with low p-tau burden in terms of MMSE only (p = 0.0136) after Tukey correction for multiple comparisons, although notably there were trends toward more rapid cognitive decline in terms of global CDR (p = 0.0747) and CDR-SB (p = 0.0764) as well.

Fig. 4.

Cognitive trajectories depend on the overall neocortical p-tau burden in individuals with Braak stage V. Spaghetti plots show individual cognitive trajectories and spline-smoothed fit lines show average cognitive trajectories for those with low p-tau burden (bottom 1/3; 0–13% PPP; n = 20), intermediate p-tau burden (middle 1/3; 13–23.5% PPP; n = 20), and high p-tau burden (top 1/3; 23.5–51.2% PPP; n = 21) in terms of a global CDR, b CDR-SB, and c MMSE. For statistical differences between groups, see Supplemental Table S2

We also divided the cohort into two groups, consisting of those without significant cognitive impairment (global CDR 0–0.5; n = 20) and those with some degree of cognitive impairment (global CDR 1–3; n = 41). After correcting for multiple comparisons, there were significant differences in the mean values for global cognitive tests and all neuropsychological tests except for DSF, TMT-A, TMT-B, and WAIS DS (Table 1). Interestingly, there were no significant differences in participant age, sex, level of education, race, overall ADNC level, Thal Phase, CERAD NP score, level of CAA, Lewy body stage, LATE stage, frequency of hippocampal sclerosis, or cerebrovascular disease variables. The cognitively impaired subset was enriched for the presence of APOE ε4 alleles compared to those without cognitive impairment (p = 0.0145), but this did not survive adjustment for multiple comparisons. The only significant difference in pathologic variables identified between these groups was the level of overall average neocortical p-tau burden (p = 0.0035) (Table 1).

Assessment of neuropathologic comorbidities

To determine if the neocortical p-tau burden was a primary factor impacting cognition in this cohort of individuals with Braak stage V, we sought to make similar correlations with additional neuropathologic comorbidities. As expected, the three quantitative measures of p-tau in the neocortex were each significantly correlated with one another, Thal phase and CERAD NP score were positively correlated with one another, and LATE stage and hippocampal sclerosis were tightly correlated with one another, but there were few other significant correlations between the pathologies (Supplemental Figure S4). After adjusting for multiple comparisons, temporal and overall average p-tau burden significantly affected global CDR, CDR-SB and animal naming (part of language/semantic memory), while frontal p-tau burden was associated with global CDR, DSF (part of attention/working memory), and animal naming (Fig. 5a), although numerous additional clinico-pathologic correlations were identified prior to FDR correction (Supplemental Figure S5). Using multivariable logistic regression analysis, p-tau burden grouping (low, intermediate, and high) was significantly associated with the risk of cognitive decline (p = 0.0175) as was the presence of microinfarcts (p = 0.0409), while Thal phase, CERAD NP score, and all other comorbidities were not (Fig. 5b).

Fig. 5.

Correlation between cognitive and neuropsychological measures and neuropathologic features. a After the level of significance for p-values was adjusted for multiple testing using false discovery rate (FDR) correction with the two-stage linear step-up procedure of Benjamini, Krieger, and Yekutieli method (p-values less than 0.0007 are considered significant), frontal positive pixel percentage (PPP) was significantly correlated with global CDR, digit span forward test, and animal naming, while temporal and average overall PPP are each significantly correlated with global CDR, CDR-SB, and animal naming. No other comorbid neuropathologic features were significantly associated with cognitive or neuropsychological testing. For sample size for each comparison, see Supplemental Table S1. For unadjusted results, see Supplemental Figure S5. b Forest plots demonstrating odds ratios (OR) for multivariable logistic regression modeling of neuropathologic features associated with cognitive decline. This model was adjusted for age at death, sex, years of education, time between final cognitive evaluation and death, and ADRC of origin. Low p-tau burden was used as the reference category for “p-tau burden (Low/Int/High)”

Investigation of outliers

There were a number of cases with p-tau burden/cognitive decline mismatches included in this study. Two individuals with high overall neocortical p-tau burden scores were relatively cognitively unimpaired (CDR-SB of 1.5 and 3.5), but very little distinguishes these participants from their counterparts. This group included one male and one female participant with an average age of 85.5 years and an average education level of 18 years. Both had relatively high p-tau burdens in both temporal (29.5% and 30.3%) and frontal (18.8% and 20.7%) lobes. Both of these cases had high levels of ADNC. One had APOE ε3/ε3 while the other had APOE ε2/ε4. There was a trend toward having slightly fewer comorbidities in these cases compared to their counterparts: one had moderate arteriolosclerosis and the remaining case had limbic-predominant Lewy body pathology, but no other pathologies were noted, including any level of LATE or other forms of CVD. As noted by Montine et al., however, significant levels of ADNC may precede symptom onset by years [41].

At the other end of the spectrum, 25% (5/20), 15% (3/20), and 5% (1/20) of cases with low overall neocortical tau burden had mild, moderate, or severe cognitive impairment, respectively. No significant differences were noted between these individuals and those with no cognitive impairment in terms of frontal, temporal, or average neocortical p-tau burden, participant age, level of education, overall level of ADNC, Thal phase, CERAD NP score, Lewy body stage, LATE stage, frequency of hippocampal sclerosis, frequency of infarcts/microinfarcts, severity of arteriolosclerosis, or APOE allele frequency.

Also notable was a single individual with Braak stage V without Aβ deposition. This participant was a 92-year-old male with only questionable to very mild cognitive impairment (Global CDR 0.5, CDR-SB 3.5, and MMSE 25), APOE ε2/ε3, 14 years of education and an absence of any neuropathologic comorbidity except for moderate arteriolosclerosis and aging-related tau astrogliopathy (ARTAG) [35]. This individual also had a relatively low neocortical p-tau burden (temporal p-tau PPP 10.4% and frontal p-tau PPP 0.4%). The neuropathologic classification for this man is unclear; while there was complete absence of Aβ staining, primary age-related tauopathy (PART) generally remains in the medial temporal lobe without extension into the neocortex, although some rare exceptions have been noted [23, 70, 71], and review of the hippocampal p-tau staining pattern did not reveal a CA2-predominant pattern, which has been noted in some studies of PART [32, 56, 68, 70], while others have not found this pattern [24].

Discussion

ADNC is the most common pathology underlying dementia and cognitive impairment and represents a significant public health concern which is predicted to have an ever increasing impact on society in the coming decades [1, 57]. The level of ADNC depends on three factors, the density of neuritic plaques in the neocortex (CERAD NP score), the topographical distribution of Aβ (Thal phase), and the topographical distribution of p-tau (Braak stage) [30, 41]. Thal phase and Braak stage do not meaningfully capture the overall burden of pathology in any region; there may be a wide range of regional p-tau density within a single Braak stage, for example (Fig. 1). While cognitive decline generally correlates with ADNC level, and more specifically with CERAD NP score and Braak stage [43], there remains a great deal of variation in cognitive trajectory and cognitive outcome among individuals even with the highest level (and narrowest range) of ADNC [49], and there may be distinct AD subtypes with differing neuropathologic progression patterns and rates of cognitive decline [42]. While some of this variation is accounted for by the presence of neurodegenerative comorbidities (most commonly LATE, Lewy body pathology, and CVD), a significant portion of this cognitive variation remains unexplained at the population level [13, 21, 36]. To evaluate the effect of neocortical p-tau on cognition in a more granular manner than that which is currently assessed and reported under current NIA-AA guidelines, we investigated clinico-pathologic correlations in 61 individuals with Braak stage V and wide variation in both cognitive trajectory and neocortical p-tau burden [49].

Temporal lobe p-tau burden was higher than frontal in the vast majority of participants, and while both correlated with cognitive decline and dementia, this correlation was somewhat more robust in the temporal lobe (Figs. 2–3). The average p-tau burden (incorporating the p-tau burden of both middle frontal and superior temporal gyri) predicted the degree of end-of-life cognitive decline experienced by these individuals (Fig. 4), and notably, only quantitative p-tau burden (including all forms of p-tau-immunoreactive deposits) and microinfarcts significantly correlated with cognitive decline, while Aβ and comorbidities, such as LATE, Lewy body pathology, infarcts, and arteriolosclerosis, did not have significant effects with multivariate analysis (Fig. 5). Although this study represents a limited cohort and is likely underpowered to detect the cognitive effects of some of these additional neurodegenerative processes, p-tau burden (particularly temporal lobe p-tau burden) correlates strongly with cognitive decline suggesting that increasing neocortical p-tau burden is a significant factor underlying the development of cognitive impairment with age. Importantly, there are a number of outliers, including both those with resilience to high neocortical p-tau burden and those with cognitive impairment despite a low neocortical p-tau burden (and in the absence of significant comorbid neuropathologic contributors), which suggests that there remains additional nuance to uncover in our understanding of how neurodegenerative processes translate into clinical symptoms.

The finding that neocortical p-tau burden is a major contributor to cognitive decline in individuals with Braak stage V has significant implications for the future of ADNC staging. This study provides evidence that the pattern of p-tau distribution at autopsy in Braak stages does not provide sufficient information to fully and accurately predict cognitive impairment, even when combined with grading systems of other ADNC measures and other neuropathologic entities. This has further implications for a wide range of studies; researchers simply using Braak stage or ADNC level to define pathology and disease severity may miss a great deal of histopathologic variation and granularity which may have important biological relevance (i.e., a case diagnosed as Braak stage V, but only mild neocortical p-tau may be functionally more similar to a case with Braak stage IV than it is to a case with Braak stage V and widespread, high neocortical p-tau burden). The inclusion of all forms of p-tau which AT8 captures in our model (including pre-tangles, NFTs, p-tau-positive neurites, and neuritic plaques) also forms an important correlate with neuroimaging models of ADNC. Recent studies using various forms of [¹⁸F]-flortaucipir positron-emission tomography (PET) have demonstrated the ability to detect and diagnose Alzheimer’s-type neocortical p-tau in living participants, even recapitulating Braak stage in some cases, but this technology is not yet granular enough to reliably distinguish between all forms of p-tau in ADNC, including NFTs, neuropil threads, and neuritic plaques [20, 33, 55, 58, 59, 66]. Given the impact of postmortem neocortical p-tau burden on cognition, radiographic and biomarker studies could be used to better trace this overall p-tau accumulation in vivo and correlate this with cognition and postmortem neuropathology to fully understand the effect of both the extent of spread and the magnitude of overall and regional p-tau burden during aging [18].

There are a number of limitations and strengths in this study. The overall cohort size was relatively small, as necessitated by the limitations of grant funding and initial selection criteria, which decreases the power and may limit our ability to detect some associations between p-tau burden and more subtle neurocognitive deficits as well as comorbid neurodegenerative pathologies and cognitive tests. Further, the participants in the general NACC cohort are not necessarily representative of the general population, as it is enriched for individuals with rare diseases, higher levels of cognitive impairment, higher levels of education, more frequent APOE ε4, and white and non-Hispanic individuals [6]. The longitudinal modeling of cognitive decline allows for detection of more subtle changes over time as opposed to simply evaluating the final cognitive state as a single variable, which is particularly useful in determining cases which are resilient over the full time course [3, 34, 49, 51]. There was no significant difference between cognitively impaired and cognitively intact individuals in terms of time between final cognitive evaluation and death (Table 1), although the fact that participants with the most severe dementia may be lost to follow-up earlier in their disease course may introduce a source of bias, which we partially accounted for by adjusting for time between final cognitive evaluation and death in multivariate models (Fig. 5).

Our evaluations grouped p-tau structures together, limiting the ability to understand the contributions of each specific deposit. On the other hand, we were able to tightly control for various neuropathologic measures and had access to LFB/H&E and key immunohistochemical stains from key brain regions to evaluate, allowing us to perform the computer-assisted quantitative p-tau assessments, as well as verify reported neuropathologic data and fill in additional metrics, particularly including LATE stage, as < 50% of cases had TDP-43 data provided from their original institutions. P-tau burden was assessed exclusively using AT8 immunohistochemistry, which is a common antibody that is widely used among ADRCs and other research centers for evaluating and quantifying p-tau pathology; however, this does introduce a limit as AT8 does not assess the full spectrum of neurofibrillary degeneration, which should be addressed in future studies [22, 39]. Another limitation, inherent to this study, is that the computer-assisted quantitative assessment of pixels in scanned slides may only be available at major research centers and in research settings, and, therefore, may not yet be suitable for routine diagnostic practice. Non-p-tau proteinopathies are still assessed using classic topographical categorical staging systems rather than burden-based quantitative measures, and for some measures, such as certain forms of cerebrovascular disease, a full assessment was still not possible as we only had access to histology for six brain regions, targeted at staging common neurodegenerative processes, but precluding full gross and microscopic evaluation for infarcts, hemorrhages, and white matter pathology. Additionally, APOE genotyping was only available in 80.3% of cases (49/61), which may have limited our ability to identify significant cognitive or pathologic associations with this variable.

Even with these limitations, this study demonstrates that measuring and incorporating neocortical p-tau burden could be a useful adjunct to traditional Braak and ADNC staging systems which could add granularity and tighten the correlation between our assessment of ADNC (and other neurodegenerative pathologies) and cognitive decline [2]. Given that neocortical p-tau burden correlates with cognitive decline better than any other component of ADNC or neurodegenerative comorbidity, it is reasonable to conclude that the density of p-tau is a major driver of cognitive decline, and that the association between Braak stage and dementia is at least partially reflective of the increasing burden of p-tau accumulation (within a given Braak stage and with progression of Braak stage). This finding is given weight by recent tau PET imaging studies [61] and studies of hippocampal p-tau burden in PART and in lower levels of ADNC [31]. Further studies in larger cohorts analyzing the density as well as the topographical distribution of ADNC and other neurodegenerative proteins, in additional regions, in conjunction with imaging and fluid biomarker studies, will be crucial to understanding how these disorders progress and interact, and will provide insight into clinical trial stratification and ultimately therapeutic development. These data provide a significant step forward in understanding how ADNC progresses in its later stages and how this progression translates into clinical symptoms, which in turn offers insight into a window of opportunity for disease-modifying intervention.

Supplementary Information

Below is the link to the electronic supplementary material.

Author contributions

Conception of the work: TER and JMW; major role in acquisition of data: TER, JC, KFB, KF, MU, TDS, TGB, MMC, BDD, MEF, MPF, MG, LTG, LAH, DH, EH, CDK, JK, EBL, PTN, DHO, RJP, RAR, SS, JAS, GES, AFT, JCT, TW, RLW, JFC, DWD, and ACM; analysis or interpretation of data, including statistical analysis: TER, JC, SK, FTZ, SKR, SH, CMD, YT, and JMW; design of figures: TER, SK, and JMW; Writing, first draft: TER and JMW; Writing, substantial revisions: TER, JC, KFB, KF, MMH, MBM, TDS, JFC, DWD, ACM, and JMW; all authors have reviewed and approved of the final draft.

Funding

This work was supported by grant funding from the National institutes on Aging (NIA) RF1 AG062348, provided to ACM, DWD, and JFC. TER and JMW are supported in part by R21 AG078505 and Texas Alzheimer’s Research and Care Consortium (TARCC) grants 957581 and 957607. MBM is supported in part by NIA R01 AG082346. MEF is supported by P30 AG066546, U24 NS133945, R01 AG072080, and R01 AG082118. LTG is funded by K24 AG053435. JFC is funded in part by R01 NS095252, R01 AG054008, Rainwater Charitable Foundation/Tau Consortium, and Cure Alzheimer’s Fund. These funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.

Data availability

A portion of the data presented in this manuscript was derived from the National Alzheimer’s Coordinating Center (NACC) dataset, and is available upon request from https://naccdata.org/. Additional data and analysis are available from the corresponding authors upon reasonable request.

Declarations

Conflict of interest

T.E.R. has been a consultant for Servier Pharmaceuticals. The authors declare that these disclosures are unrelated to the present work, and that they have no additional competing interests, conflicts of interest, or other relevant disclosures.

Ethical approval and consent to participate

Not applicable.

Consent for publication

Not applicable.