Characterization of neurofibrillary tangle immunophenotype signatures to classify tangle maturity in Alzheimer's disease

Abstract

INTRODUCTION

Tau aggregation into neurofibrillary tangles in Alzheimer's disease (AD) is a dynamic process involving changes in tau phosphorylation, isoform composition, and morphology. To facilitate studies of tangle maturity, we developed an image analysis pipeline to study antibody labeling signatures that can distinguish tangle maturity levels in AD brain tissue.

METHODS

Using fluorescent immunohistochemistry, we co‐labeled AD brain tissue with four antibodies that bind different tau epitopes. Mean fluorescence intensity of each antibody was measured, and spectral clustering was used to identify tangle immunophenotypes.

RESULTS

Five distinct tangle populations were identified, and different tangle maturity immunophenotypes were identified with increasing Braak stage. Early tangle immunophenotypes were more prevalent in later affected regions and advanced immunophenotypes were associated with ghost morphology.

DISCUSSION

Our findings indicate that tangle populations characterized by advanced tau immunophenotypes are associated with higher Braak stage and more mature morphology, providing a new framework for defining tangle maturity levels using tau antibody signatures.

Highlights

• Populations of neurofibrillary tangles exist in Alzheimer's disease.

• The immunophenotype of neurofibrillary tangle populations relates to their maturity.

• The most advanced immunophenotypes are associated with higher Braak stage.

• The most advanced immunophenotypes are associated with ghost morphology.

• The most immature immunophenotypes are associated with later affected regions.

1. BACKGROUND

The deposition of neurofibrillary tangles composed of hyperphosphorylated tau protein is one of the hallmark features of Alzheimer's disease (AD). ¹ , ² The tau protein undergoes post‐translational modifications that lead to its pathological aggregation into insoluble fibrillary structures. ³ Emerging evidence suggests that tau aggregation is a dynamic and complex process, resulting in the accumulation of different forms of modified tau within tangles as they mature. ⁴ The composition of these forms of modified tau within tangles may result in distinct biochemical properties that contribute to disease progression and regional vulnerability. ⁴ , ⁵ Measuring the heterogeneity of modified tau species within tangles and how these relate to disease progression is necessary to decipher the mechanisms of AD pathogenesis.

In AD, tau tangle pathology develops and progressively affects more brain regions with a well‐defined sequence that correlates with symptom progression. ¹ , ⁶ , ⁷ The severity of tau pathology in AD can be classified by the Braak staging system, which classifies the spread of tau pathology into one of six stages, condensed into four stages (0, I–II, III–IV, and V–VI) in the National Institute on Aging–Alzheimer's Association (NIA‐AA) guidelines for the assessment of AD neuropathologic change. ² , ⁸ These stages are determined based on labeling for hyperphosphorylated tau using the AT8 antibody, which binds tau phosphorylated at the serine 202 and threonine 205 positions. ⁹ , ¹⁰

Tangles progress through levels of maturation that are distinguished by their morphology. ⁴ , ¹¹ Three levels of tangle maturity are commonly recognized, although maturation is a dynamic process and intermediary morphologies exist between these three levels. ⁴ The earliest level of tangle maturity is a pretangle morphology characterized by diffuse or discrete granular tau labeling. ⁴ , ¹¹ , ¹² As the tangles develop into a mature morphology the tau fibers become compact, taking on a characteristic flame‐like structure that often displaces the nucleus. ¹² The most advanced level of maturation is a “ghost” tangle morphology, evidenced by weaker tau labeling and no associated nucleus. ¹² These morphological classifications suggest tangle development is a progressive process of aggregation and transformation of tau protein within neurons.

In addition to tangle morphology, different modifications and isoforms of the tau protein are associated with tangle maturation. Phosphorylation at the S202 and T205 sites is one of the earliest modifications to the tau protein and the AT8 antibody that binds to this site primarily recognizes pretangle and mature tangle morphologies. ⁴ C‐terminal truncation at E391, recognized by the MN423 antibody, is a modification associated with advanced tangles that have mature and ghost morphologies. ¹³ , ¹⁴ Additionally, AD is classified as a 3R/4R tauopathy, in that tangles contain both 3R and 4R isoforms of tau. The distribution of isoforms in the AD hippocampus shifts from predominantly 4R tau in early Braak stages to predominantly 3R tau tangles in later Braak stages. ¹⁵ , ¹⁶ , ¹⁷ , ¹⁸ , ¹⁹ As “pretangle,” “mature tangle,” and “ghost tangle” are morphological descriptions, from here on when referring to tangle maturity level in the context of antibody labeling, we will refer to tangles as being “early” or “advanced.”

Braak staging is a useful measure of global tangle presence and thus pathology severity, but there has been limited assessment of tangle maturity in the context of Braak staging outside the medial temporal lobe. Quantitative approaches to assess tangle maturity are also needed. Tangle maturity is often scored by manually inspecting tau labeling and categorizing the frequency of each morphology. ⁵ , ¹¹ This is a time‐consuming and inherently subjective process that is complicated by the presence of intermediary morphologies. Quantitative characterization of tau antibody labeling in the context of tangle maturity is lacking in current literature, but accumulated knowledge of how these antibodies recognize different tangle maturity levels (reviewed by Moloney et al. ⁴ ) suggests that early versus advanced tangles have distinct antibody labeling signatures, known as immunophenotypes.

In this study, we investigated whether immunophenotypes can be used to distinguish tangle populations associated with varying maturity levels and morphologies. We examined tangle immunophenotype signatures in cases with varying Braak stages across different brain regions. In doing so we provide a high‐throughput approach to classifying tangle maturity based on fluorescent immunolabelling of four tau antibodies to provide a quantitative view of regional pathological severity.

2. METHODS

2.1. Human brain tissue processing

Human post mortem brain tissue was obtained from the Neurological Foundation Human Brain Bank at the University of Auckland, New Zealand. The tissue was donated with informed consent from the family prior to brain removal and all procedures were approved by the University of Auckland Human Participants Ethics Committee (Ref: 011654). The ethnicity data for the cases was not collected; however, both male and female cases were included (Table 1). All cases used in this study were assessed by an independent neuropathologist who determined the Braak staging and NIA‐AA score. The nine Braak stage I to II neurologically normal donor cases had no history of neurological abnormalities, and no other neuropathology was noted. The mean age (± standard deviation) of the normal cases was 86.6 ± 6.6 years and ranged from 76 to 94 years. The average post mortem delay was 18.3 ± 4.8 hours with a range of 12 to 26 hours (Table 1). The 10 Braak stage III to VI AD donor cases had a clinical history of dementia, and the clinical AD diagnosis was confirmed by an independent pathologist. The average age of AD cases was 82.4 ± 6.1 years and ranged from 73 to 89 years. The average post mortem delay was 16.6 ± 8.8 hours with a range of 5 to 31 hours (Table 1). The 12 Braak stage V to VI AD donor cases had a clinical history of dementia, and the clinical AD diagnosis was confirmed by an independent pathologist. The average age of AD cases was 79.4 ± 11.1 years and ranged from 58 to 94 years. The average post mortem delay was 11.2 ± 6.6 hours with a range of 3.5 to 22 hours (Table 1).

TABLE 1.

Case information for tissue used in this study.

Case	Pathology diagnosis	Age (years)	Sex	PMD (hours)	Braak stage	NIA‐AA score
H229	Normal	89	F	17	I–II	2‐1‐2
H241	Normal	76	F	12	I–II	3‐1‐1
H246	Normal	88	M	17	I–II	0‐1‐0
H249	Normal	77	F	17	I–II	NA
H250	Normal	93	F	19	I–II	NA
H252	Normal	94	F	13.5	I–II	1‐1‐0
H254	Normal	87	M	26	I–II	1‐1‐0
H256	Normal	92	M	24	I–II	3‐1‐2
H258	Normal	83	M	19	I–II	2‐1‐1
H251	Alzheimer's	77	M	11.5	III–IV	3‐2‐3
AZ084	Alzheimer's	82	M	18.5	III–IV	3‐2‐2
AZ102	Alzheimer's	84	F	14.5	III–IV	3‐2‐2
AZ103	Alzheimer's	87	M	<24	III–IV	3‐2‐2
AZ104	Alzheimer's	79	F	13	III–IV	3‐2‐2
AZ109	Alzheimer's	89	F	31	III–IV	3‐2‐1
AZ111	Alzheimer's	85	F	30	III–IV	1‐2‐2
AZ118	Alzheimer's	73	F	10.5	III–IV	3‐3‐1
AZ119	Alzheimer's	88	M	5	III–IV	3‐2‐2
AZ120	Alzheimer's	73	F	6.5	III–IV	3‐2‐2
AZ143	Alzheimer's	89	M	18.5	III–IV	3‐2‐2
AZ099	Alzheimer's	94	F	8.5	V–VI	3‐3‐2
AZ101	Alzheimer's	75	M	12.5	V–VI	3‐3‐2
AZ108	Alzheimer's	94	F	11.5	V–VI	3‐3‐2
AZ113	Alzheimer's	77	M	3.5	V–VI	3‐3‐2
AZ123	Alzheimer's	81	F	NA	V–VI	3‐3‐2
AZ125	Alzheimer's	89	F	NA	V–VI	3‐3‐2
AZ127	Alzheimer's	66	M	8	V–VI	3‐3‐2
AZ129	Alzheimer's	81	M	4.5	V–VI	3‐3‐2
AZ132	Alzheimer's	58	M	NA	V–VI	3‐3‐3
AZ134	Alzheimer's	69	F	22	V–VI	3‐3‐3
AZ135	Alzheimer's	85	F	19	V–VI	3‐3‐2
AZ137	Alzheimer's	84	F	NA	V–VI	3‐3‐2

Abbreviations: NIA‐AA, National Institute on Aging–Alzheimer's Association; NA, not available; PMD, post mortem delay.

The right hemisphere of each brain was fixed by perfusion of 15% formaldehyde in 0.1 M phosphate buffer through the cerebral arteries and the hemisphere was subsequently dissected into approximately 60 blocks as described previously. ²⁰ From each block, a 0.5 cm‐thick section was selected for paraffin embedding and the remaining tissue was snap frozen using dry‐ice snow and stored at −80°C. The brain tissue blocks were processed for paraffin embedding as described previously. ²⁰ The paraffin blocks were sectioned in the coronal plane at a thickness of 10 µm using a rotary microtome (Leica Biosystems RM2335). The sections were floated on a water bath set at 37°C (Leica Biosystems, HI1210) before being mounted on Superfrost Plus slides (MenzelGlaser) and air dried for 18 hours at room temperature.

2.2. Immunohistochemistry

We used fluorescent immunohistochemistry to label tissue with antibodies against the 4R tau isoform (4R), 3R tau isoform (3R), ²¹ tau phosphorylated at Serine202 and Threonine205 (AT8), and tau with C‐terminus truncation at glutamic acid 391 (MN423). One 10 µm‐thick section of the entorhinal cortex (EC), middle temporal gyrus (MTG), and superior frontal gyrus (SFG) were labeled from each case.

RESEARCH IN CONTEXT

1. Systematic review: The authors searched PubMed and Google Scholar for literature on Alzheimer's disease, Braak staging, neurofibrillary tangles, and neurofibrillary tangle maturity.

2. Interpretation: Braak staging is used to describe the severity of tau pathology across the brain. The overall maturity of tangles is often scored by manually assessing the frequency of pretangle, mature, or ghost tangle morphology in a region, which is a semi‐quantitative and subjective measure. Tangles are comprised of tau proteins with different isoforms and post‐translational modifications associated with different maturity levels. Our study proposes a new framework for classifying tangles based on their antibody labeling signature.

3. Future directions: We recommend that future studies use antibody labeling signatures to classify tangle maturity. This approach will allow the maturity level of individual tangles to be classified, which could facilitate more in‐depth studies of early and late‐stage tau neuropathology.

Heat‐induced epitope retrieval was performed using trisEDTA with 0.05% Tween 20, pH 9.0 in a pressure cooker (2100 Antigen Retriever, Aptum Biologics Ltd.) for 20 minutes at 121°C and left to cool for 1.5 hours. Ninety‐nine percent formic acid was then applied for 3 minutes at room temperature, followed by permeabilization with phosphate‐buffered saline (PBS) + 0.5% Triton X100 for 15 minutes at room temperature. Autofluorescence was quenched  using Trueblack (23007, Biotium). Ten percent goat serum was used to block non‐specific binding of the secondary antibodies. Primary and secondary antibody information is presented in Table 2. Sections were incubated with the 4R tau primary antibody diluted 1:500 in 1% goat serum overnight at 4°C. The 4R tau antibody was visualized using tyramide signal amplification (TSA) with the Alexa Fluor 488 Tyramide Superboost Kit (B40941, ThermoFisher Scientific). The 4R primary antibody was then removed from the tissue using LiCor NewBlot Nitro stripping buffer (928‐40030, LiCor) and heat‐mediated epitope retrieval using Citrate buffer pH6.0, leaving the Alexa Fluor 488 labeling intact. The 3R tau antibody was then diluted 1:500 in 1% goat serum and applied to the sections overnight at 4°C. The 3R tau antibody was visualized using the Alexa Fluor 647 Tyramide Superboost Kit (B40916, ThermoFisher Scientific). The 3R tau antibody was also removed using LiCor NewBlot Nitro stripping buffer (928‐40030, LiCor), leaving the Alexa Fluor 488 and 647 labeling intact. Finally, the AT8 and MN423 antibodies were diluted 1:500 in 1% goat serum and applied to the sections overnight at 4°C. Species and immunoglobulin G subtype‐specific secondary antibodies conjugated to Alexa Fluor 546 and Alexa Fluor 594, as well as Hoechst (1:20,000 in PBS, Molecular Probes, #33342), were incubated at room temperature for 3 hours and then coverslipped. The sections were imaged on a Zeiss Z2 Axioimager (10X /0.45 NA objective) using MetaSystems VSlide acquisition software and MetaCyte stitching software. The Zeiss Z2 Axioimager is equipped with a Colibri 7 solid‐state light source with LED lamps and the following filter sets to enable spectral separation of the five fluorophores per round (Ex peak [nm]; Em [nm]/bandpass [nm]): Hoechst (385; 447/60), AlexaFluor® 488 (475; 550/32), AlexaFluor® 546 (555; 580/23), AlexaFluor® 594 (590; 628/32), and AlexaFluor® 647 (630; 676/29).

TABLE 2.

Antibodies used for immunohistochemistry.

SPrimary antibody	Target epitope	Visualization
4R tau (005‐804, Merck)	4R tau isoform	Goat anti‐mouse TSA Alexa Fluor 488 (B40941, ThermoFisher)
3R tau (005‐803, Merck)	3R tau isoform	Goat anti‐mouse TSA Alexa Fluor 647 (B40916, ThermoFisher)
AT8 (MN1020, ThermoFisher)	Tau phosphorylated at Serine 202, Threonine205	Goat anti‐mouse IgG1 Alexa Fluor 546 (A21123, ThermoFisher)
MN423 (ab02389‐23, Absolute Antibody)	Tau cleaved at glutamic acid 391	Goat anti‐rabbit Alexa Fluor 594 (A11037, ThermoFisher)

Abbreviation: TSA, tyramide signal amplification.

Validation experiments were conducted to ensure that the sequential 4R and 3R tau labeling using TSA did not result in cross‐reactivity (Figure S1 in supporting information).

2.3. Image processing and mean gray intensity measurement

Images were exported from the VSviewer software as uncompressed 8‐bit TIFF files per channel for processing and analysis using the FIJI distribution of ImageJ (version 1.52p).

The image segmentation and measurement of mean gray intensity (MGI) were conducted as follows (summarized in Figure S2 in supporting information):

1. Background subtraction: For each full section image, a 50 pixel² region was defined in an area of background on the Hoechst image. For each separate channel, the MGI of this background region was measured, and the value subtracted from the entire channel image. This approach was selected to mitigate the high variability in background labeling and the amount of lipofuscin between cases.

2. Region of interest selection: The region of interest (ROI) for each section was defined using the Hoechst image. A 2 mm wide strip of gray matter that included all cortical layers was delineated for each tissue section.

3. Tangle mask: For each background subtracted tau antibody image (4R, 3R, AT8, MN423), a binary mask was created that included all pixels with gray value > 10. The binary mask from each antibody image was added together to create a “tau master mask.” Individual tangles were defined using the “Analyze” function with a minimum size of at least 500 pixels.

4. Measurement of MGI: The perimeter ROI for each individual tangle was saved to the ROI manager. Each of the tau antibody images was then loaded and for each tangle ROI, the MGI of the tau labeling was measured. We used MGI as a proxy for protein expression level of each of the tau species identified by the different antibodies.

5. Removal of false positive objects: Due to the high amount of lipofuscin, particularly in the Braak stage V to VI AD cases, some false positive tangle objects were included in the total tau masks. To account for this, we removed objects from the dataset if they did not have an MGI value higher than a determined threshold for at least one of the four tau antibodies. The threshold for each tau antibody was determined based on the peak of the histogram of MGI values for that antibody.

6. To standardize the MGI measurements between images we used a previously reported approach. ²² That is, as MGI measurements of protein expression tend to follow log‐normal distributions, we applied a natural log transformation and determined the Z score of the log‐transformed MGI values. These values have been reported for the expression level plots and were used for the spectral clustering analysis.

7. Spectral clustering: Log Z–transformed intensity values were grouped by spectral clustering. Spectral clustering is an unsupervised machine‐learning technique that divides data points in large data sets into several groups. ²³ We used this method as it can identify convex clusters with different variances common in multiplex immunohistochemistry. ²⁴ The Spectrum package in R computes pairwise similarity matrices using an adaptive density‐aware kernel, then combines the matrices by calculating a cross‐view tensor product graph from each pair of individual graphs. ²⁴ The Spectrum package uses the Ng spectral clustering method on the resulting matrix to calculate the number of clusters present in the data using an eigengap heuristic. The final eigenvector matrix is clustered by Gaussian mixture modeling. ²³

8. Data visualization: The clustered tangles were plotted as a heatmap ordered by their assigned cluster, then by the corresponding Braak stage for the case the tangle came from, and then by the brain region the tangle was recorded from. Plotting was performed in RStudio using the ComplexHeatmap package. ²⁵ The intensity values were normalized to 100 for easier visualization. The distribution of the clustered tangles among different Braak stages, regions, or cases was plotted in proportionate bar graphs using the ggplot package.

2.4. Tangle morphology classification

The morphology analysis was only conducted on the ROI images from the EC. The FIJI distribution of ImageJ (version 1.52p) was used for this analysis. First, a total tau image was created by adding the 4R, 3R, AT8, and MN423 background‐subtracted images. This total tau image was then overlaid with the Hoechst image to allow identification of cell nuclei within a tangle. The perimeter ROI for each individual tangle was imported into the ROI manager. An investigator then manually determined whether each of the tangle objects fit the morphological criteria for a pretangle, mature tangle, or ghost tangle. These criteria were based on previously reported definitions of tangle morphology: ⁴ , ¹¹ Pretangles had diffuse or punctate cytoplasmic tau labeling, surrounding a Hoechst‐positive nucleus. Mature tangles had the characteristic dense, flame‐shaped cytoplasmic morphology, surrounding a Hoechst‐positive nucleus which may or may not be displaced or distorted. Ghost tangles were defined as loosely bundled tau‐positive fibers with no associated nucleus. Tangle objects that did not fit any of the above criteria were not classified or included in the analysis. Two investigators (D.H. and H.C.M.) performed the morphology classification. Tangles were included in this analysis if both investigators independently gave them the same classification. A total of 451 tangles across all cases were classified.

2.5. Statistical analyses

Differences in the proportions of clusters between Braak stages, regions, and morphology classes were tested using a chi‐square (χ ²) test for trend in GraphPad Prism (version 9.3.1). To determine whether the labeling intensity of the tau antibodies differed between clusters, Braak stage groups, regions, or tangle morphology classes we compared the MGI Z scores for each tau antibody between groups as per a previously reported approach. ²² As the MGI Z scores from each tangle belonging to the same case are not independent observations, we used mixed effects regression models with cluster, Braak stage group, region, or tangle morphology class as the fixed effect and case ID as a random effect. The “lmer” function from the lme4 package in R (version 4.1.1) was used to fit the model with restricted maximum likelihood (REML) estimation. To further explore the differences between the levels of the fixed effect variable (cluster, Braak stage group, region, or tangle morphology class), post hoc pairwise comparisons were conducted using the “emmeans” and “pairs” functions from the emmeans package. The P values for the pairwise comparisons were adjusted using Satterthwaite's method to control for multiple comparisons. The significance level was set at α = 0.05 for all tests.

3. RESULTS

3.1. Tau isoform diversity increases as AD pathology increases

To establish whether tau epitope abundance differs with increasing Braak stage, we used tau epitope‐specific antibodies and labeled the EC, MTG, and SFG from AD cases with different Braak stages. We used antibodies raised against the 3R and 4R tau isoforms, phosphorylation at S202‐T205 (AT8 antibody), and a truncated tau variant (MN423 antibody, Figure 1). Tangles in the EC of Braak I to II cases were predominantly only AT8‐positive, and only the EC (not the MTG or SFG) showed clear tangle labeling (Figure 1A‐C). Tangles in the EC and MTG of Braak stages III to IV were labeled with both 3R and MN423 antibodies (Figure 1D‐F). Tangles that were predominantly labeled with the 4R antibody were most abundant in the MTG and SFG of Braak III to IV and V to VI cases (Figure 1G‐I). Unexpectedly, there was a range of tangles with different combinations of antibody labeling and different labeling intensities.

FIGURE 1.

Representative images of tau antibody labeling by Braak stage and region. Representative fluorescence microscopy images from the EC, MTG, and SFG of cases classified as Braak stage I to II, III to IV, and V to VI showing fluorescent co‐labeling of four tau antibodies: AT8, 4R tau, 3R tau, and MN423. AT8 labeling is abundant in the EC of Braak stage I to II cases (A), but tangles are not observed in the MTG (B) or SFG (C). Tangles that were co‐labeled with all four antibodies were observed in the EC of Braak stage III to IV cases, with an abundance of those co‐labeled with 3R and MN423 antibodies (D). Conversely, 4R tau was more abundant in tangles in the MTG (E) and SFG (F). In advanced Braak stage V to VI cases, all tangles were abundantly labeled with all four tau antibodies, with a greater abundance of tangles in the EC (G) and more frequent labeling of 4R tau in the MTG (H) and SFG (I). Scale bar 50 µm. EC, entorhinal cortex; MTG, middle temporal gyrus; SFG, superior frontal gyrus.

3.2. Five distinct tangle immunophenotypes are present across cases from different Braak stages

Tau epitope‐specific antibodies revealed diverse tangle morphology in all Braak stages, but stages III to VI expressed the most diverse combination of AT8, 4R, 3R, and MN423 labeling. To accurately categorize tangles, we applied a computational approach to cluster distinct tangle populations based on tau co‐labeling signatures. We identified individual tangle objects within the image using image segmentation pipelines and within each tangle object, we measured the mean gray value for each antibody. We then applied a natural log transformation and obtained the Z score of these log‐transformed MGI values. This data processing provided values indicative of the fluorescent labeling intensity that could be compared between cases and used for subsequent analyses. From here on, we will refer to the log‐transformed MGI Z scores simply as “intensity scores” comparing the fluorescent intensity of the antibody labeling between groups.

To identify populations of tangles with distinct immunophenotypes within the tissue we applied unsupervised spectral clustering on the intensity scores for each tangle. This approach performs dimensionality reduction and divides the data into network clusters based on similarity. It has previously been applied to cell populations identified using multiplexed immunohistochemistry in which it outperformed traditional clustering algorithms. ²² We identified five distinct tangle immunophenotypes based on eigengap analysis, which signifies the gap between successive eigenvalues and indicates the inherent cluster structure in the data. Selecting five clusters optimized the separation between the clusters while minimizing the within‐cluster variance.

Four of the clusters (comprising 60.8% of all tangles) showed elevated labeling intensity for one or more of the four antibodies, while the fifth cluster (comprising 39.2% of all tangles) contained tangles with relatively high labeling intensity for all four markers (Figure 2A,B; Table S1 in supporting information). We named each immunophenotype according to the antibodies that showed the highest labeling intensity: “AT8 high,” “4R Tau high,” “3R Tau high,” “3R Tau‐MN423 high,” and “Mixed.” Examples of tangles from each of these clusters were qualitatively observed across the different regions and cases included in our study (Figure 2C).

FIGURE 2.

Unsupervised spectral clustering reveals five distinct tangle immunophenotypes. (A) Heatmap demonstrates the spectral clustering of 14,975 tangles based on the MGI Z scores for the labeling of four tau antibodies: AT8, 4R, 3R, and MN423. Five distinct clusters were identified: “AT8 high,” “4R Tau high,” “3R Tau high,” “3R Tau‐MN423 high,” and “Mixed.” For each tangle, the scaled MGI Z score for each antibody was plotted on the heatmap. The tangles were ordered on the heatmap by their assigned cluster, then by the corresponding Braak stage of case the tangle came from, and then by the brain region the tangle was recorded from. (B) Violin plots illustrate the distribution and median of the MGI Z scores for each tau antibody in the tangles assigned to each cluster. Median is indicated by a solid line and quartiles are indicated by dotted lines. The MGI Z scores for AT8 and 4R were highest in the “AT8 high” and “4R Tau high” clusters, respectively. The MGI Z scores for 3R were highest in the “3R Tau high” and “3R‐MN423 high” clusters. MN423 MGI Z scores were the highest for the “3R‐MN423 high cluster” and the “Mixed” cluster showed similar MGI Z scores for all four antibodies. (C) Images of representative tangles from each of these clusters, showing the fluorescent co‐labeling of AT8, 4R, 3R, and MN423 antibodies. Scale bar 20 µm. EC, entorhinal cortex; MGI, mean gray intensity; MTG, middle temporal gyrus; SFG, superior frontal gyrus.

3.3. Immunophenotype proportions differ between Braak stages in the EC

Next, we assessed whether specific tangle immunophenotypes are associated with AD pathological severity. Examining individual tangles, we found the relative antibody labeling intensities differed between Braak stages. As the number of tangles outside the EC was minimal for the Braak stage I to II cases, only the EC was included for this analysis.

Qualitatively, 4R, 3R, and MN423 labeling was rarely observed in the EC of Braak I to II cases but was observed in the Braak III to IV and V to VI cases (Figure 3A). We compared the intensity scores between Braak stage groups to assess the difference in antibody labeling intensity. Because tangles belonging to the same case are not independent observations, we applied mixed‐effects regression models with Braak stage as a fixed effect and the case number as a random effect to assess the statistical significance of the differences in intensity score between Braak stage groups. For AT8, there was a statistically significant increase in intensity score for the Braak stage I to II group compared to the Braak stage V to VI group. There was also a statistically significant decrease in 3R tau mean intensity score for the Braak stage I to II group compared to the Braak stage V to VI group (Figure 3B; Table S1).

FIGURE 3.

Comparison of tangle immunophenotypes between different Braak stage groups in the EC reveals an increased proportion of AT8 high immunophenotype tangles in Braak I to II cases. (A) Representative images of co‐labeling for AT8, 4R, 3R, and MN423 antibodies in the EC from cases with different Braak stage classifications. Scale bar 50 µm. The tangles observed in the Braak I to II cases had high‐intensity AT8 labeling, with minimal 3R and MN423 labeling, while tangles in the Braak III to IV and V to VI cases showed high‐intensity labeling for 4R, 3R and MN423. (B) Violin plots illustrate the distribution of the MGI Z scores (intensity scores) for each tau antibody in the EC tangles across different Braak stage groups. Median is indicated by a solid line and quartiles are indicated by dotted lines. Mixed effects regression models with Braak stage as a fixed effect and the case number as a random effect were used to assess the statistical significance of the differences in intensity score among Braak stage groups. AT8 intensity scores were significantly increased in Braak I to II cases compared to Braak V to VI. 3R tau intensity scores were significantly increased in Braak V to VI cases relative to Braak I to II cases. The statistical significance of differences between Braak stage groups is represented by **P ≤ 0.01. (C) Stacked bar plot illustrating the percentage of total tangles that each immunophenotype comprises, by Braak stage group in the EC. A chi‐square (χ ²) test was used to assess the statistical significance of differences in proportions of clusters between groups. The percentage of tangles with the “AT8 high” immunophenotype was significantly increased in the Braak I to II group relative to the III to IV and V to VI groups. The immunophenotype proportions were similar between the Braak III to IV and V to VI groups. EC, entorhinal cortex; MGI, mean gray intensity.

The proportions of the five tangle immunophenotypes were then compared between Braak stage groups (Figure 3C). Of the tangles in the EC of Braak stage I to II cases, 75.1% were classified as the “AT8 high” immunophenotype; 15.8% were classified as the “Mixed” immunophenotype, followed by 2.1% as “4R high,” 1.6% as “3R high,” and 5.3% as “3R Tau‐MN423 high” immunophenotypes.

Conversely, the “3R Tau‐MN423 high” and “Mixed” immunophenotypes were the most abundant in the EC of Braak stage III to IV and V to VI cases. The “Mixed” immunophenotype accounted for 40.7% of tangles in Braak III to IV cases and 41.1% of tangles in Braak V to VI cases. The “3R Tau‐MN423 high” immunophenotype accounted for 29.8% of tangles in Braak III to IV cases and 26.7% of tangles in Braak V to VI cases. The remaining tangles were classified as “AT8 high” (13.6% for Braak III–IV, 10.4% for Braak V–VI), “4R tau high” (6.0% for Braak III–IV, 7.9% for Braak V–VI), and “3R Tau high” (10.0% for Braak III–IV, 13.9% for Braak V–VI). The difference in immunophenotype proportions between Braak stage groups was statistically significant (χ ² [8, N = 6988] = 1197, P < 0.0001). Plotting the proportion of EC tangles in each cluster for each individual case confirmed that these trends were consistent across cases within a Braak stage group (Figure S3 in supporting information). Overall, the proportion of each immunophenotype was similar between the Braak III to IV and V to VI groups, suggesting a distinction between those cases neuropathologically classified as AD and those that were not (Braak I–II).

3.4. 4R Tau high tangles are more abundant in the SFG than in the MTG or EC of Braak stage V and VI cases

Different brain regions are affected at different stages of AD progression. We hypothesized that regions affected later in the disease progression, such as the SFG, would contain more early tangles and thus have a different tangle immunophenotype profile than regions affected earlier in the disease progression such as the EC or MTG. As the number of tangles in the SFG was minimal for the Braak stage I to II and III to IV cases, only the Braak stage V to VI cases were included for this analysis.

Qualitatively, we observed less 3R tau and MN423 labeling in the SFG of Braak V to VI cases compared to the EC and MTG (Figure 4A). To assess the difference in antibody labeling intensity, we compared the intensity scores among the three regions for Braak stage V to VI cases. As described above, we applied mixed‐effects regression models with region as a fixed effect and the case number as a random effect to assess the statistical significance of the differences in intensity scores between regions. There was a statistically significant increase in the intensity scores for AT8 and 4R tau in the SFG region compared to the MTG. The intensity scores for 4R tau were significantly increased in the SFG compared to the EC. There was also a statistically significant decrease in the 3R tau and MN423 intensity scores in the SFG compared to the EC (Figure 4B; Table S1).

FIGURE 4.

Comparison of tangle immunophenotype between early‐ and late‐affected brain regions in Braak stage V to VI cases reveals a higher proportion of 4R tau high tangles in the SFG. (A) Representative images of co‐labeling for AT8, 4R, 3R, and MN423 antibodies in the EC, MTG, and SFG of Braak stage V to VI cases. Scale bar 50 µm. The number of tangles labeled for 3R and MN423 is lower in the SFG than in the EC and MTG. (B) Violin plots illustrate the distribution of the MGI Z scores (intensity scores) for each tau antibody in tangles from the EC, MTG, and SFG of Braak stage V to VI cases. Median is indicated by a solid line and quartiles are indicated by dotted lines. Mixed effects regression models with region as a fixed effect and the case number as a random effect were applied to assess the statistical significance of the differences in intensity scores between regions. There were significant differences in intensity scores for all four antibodies among the three regions. Notably, the intensity scores for 4R tau were significantly increased in the SFG compared to the MTG and EC. Conversely, the intensity scores for 3R and MN423 were significantly reduced in the SFG compared to the MTG and EC. The statistical significance of differences among regions is represented by: ****P ≤ 0.0001, ***P ≤ 0.001, **P ≤ 0.01, *P ≤ 0.05. (C) Stacked bar plot illustrating the percentage of total tangles that each immunophenotype comprises, in each region of Braak stage V to VI cases. A chi‐square (χ ²) test was used to assess the statistical significance of differences in proportions of clusters between regions. The percentage of tangles with the “4R Tau high” immunophenotype was increased, while the “3R Tau high” and “3R‐MN423 high” immunophenotypes were decreased in the SFG relative to the EC and MTG. The immunophenotype proportions were similar between the EC and MTG. EC, entorhinal cortex; MGI, mean gray intensity; MTG, middle temporal gyrus; SFG, superior frontal gyrus.

The abundance of five tangle immunophenotypes were then compared among regions (Figure 4C). The “4R Tau high” immunophenotype comprised the highest percentage of tangles in the SFG (37.1%, compared to 7.9% for the EC and 12.5% for the MTG), while the “Mixed” immunophenotype was abundant across all three regions (EC: 41.1%, MTG: 39.2%, SFG 35.7%). In the EC and MTG regions, the “3R Tau high” and “3R Tau‐MN423 high” immunophenotypes were more abundant than in the SFG (EC: 26.7%, MTG: 26.1%, SFG: 35.7%). The difference in immunophenotype percentages between regions was statistically significant (χ ² [8, N = 10,230] = 1213, P < 0.0001). Plotting the percentage of Braak stage V to VI tangles in each cluster by case confirmed that these trends were consistent across cases (Figure S4 in supporting information).

3.5. Tangle immunophenotypes are associated with distinct tangle morphologies

Last, we sought to determine whether tangle immunophenotypes could be used to infer the relative maturity level of individual tangles. To do this we investigated whether tangles defined as having pretangle, mature, or ghost morphology had distinct immunophenotypes. We first manually classified the morphology of tangles from the EC across all Braak stage groups. We observed a subset of tangles that fit the previously defined criteria for pretangles, mature tangles, and ghost tangles ⁴ , ¹¹ (Figure 5A). All three morphologies were observed across all Braak stage groups, although the proportion of tangles with pretangle morphology was highest in Braak I to II cases and the proportion of tangles with ghost morphology was highest in Braak V to VI cases (Figure S5 in supporting information). Overall, we found that most tangles did not clearly fit the criteria for any one of these morphologies, either due to their position in the plane of the tissue section or due to having an intermediary morphology, and these tangles were not classified (Figure S6 in supporting information).

FIGURE 5.

Tangle immunophenotypes are associated with specific tangle morphologies in the EC. (A) Representative images of tangles classified as having pretangle, mature, and ghost morphology. Labeling from all four tau antibodies has been flattened into one channel (gray) and overlaid with the Hoechst channel (blue). Scale bars = 10 µm. (B) Stacked bar plot illustrating the percentage of each tangle immunophenotype for each tangle morphology group. A chi‐square (χ ²) test was used to assess the statistical significance of differences in proportions of clusters between morphologies. Pretangle tangles are predominantly the “AT8 high” and “Mixed” immunophenotypes, while the highest proportion of mature tangles is the “Mixed” immunophenotype and the highest proportion of ghost tangles are the “3R Tau‐MN423 high” immunophenotype. (C) Violin plots illustrate the distribution of the MGI Z scores (intensity scores) for each tau antibody in tangles with pretangle, mature, or ghost morphology. Median is indicated by a solid line and quartiles are indicated by dotted lines. Mixed effects regression models with morphology as a fixed effect and the case number as a random effect were applied to assess the statistical significance of the differences in intensity scores between morphologies. There was a statistically significant decrease in AT8 and 4R intensity scores for the ghost tangles relative to the pretangle and mature tangles. There were significant differences in intensity scores for 3R and MN423 among all three tangle morphologies. Statistical significance of differences among Braak stage groups within a region is represented by: ****P ≤ 0.0001, ***P ≤ 0.001, **P ≤ 0.01, *P ≤ 0.05. EC, entorhinal cortex; MGI, mean gray intensity; MTG, middle temporal gyrus; SFG, superior frontal gyrus.

For the tangles classified as having ghost morphology, the immunophenotypes with the high intensity of 3R tau labeling were predominant (Figure 5B). The “3R Tau‐MN423 high” immunophenotype was most abundant (61.5%) followed by the “Mixed” immunophenotype (21.8%) and “3R Tau high” immunophenotype (16.0%). Only one tangle was classified as “AT8 high” (0.6%), and none were “4R Tau high.” For the mature morphology tangles, all five immunophenotypes were represented; however, the “Mixed” immunophenotype was most abundant (64.5%). The tangles classified as having pretangle morphology were predominantly the “AT8 high” immunophenotype (44.9%), followed by the “Mixed” (42.9%) and “4R Tau high” immunophenotypes (12.2%). Notably, the “3R Tau high” and “3R Tau‐MN423 high” immunophenotypes were not represented in the pretangle morphology tangles. The difference in immunophenotype proportions between morphology classes was statistically significant (χ ² [8, N = 451] = 282.1, P < 0.0001). Overall, the pretangle morphology was predominantly associated with “AT8 high” and “4R Tau high” immunophenotypes, while the most advanced ghost morphology was predominantly associated with “3R Tau‐MN423 high” immunophenotypes.

Supporting this, we applied mixed‐effects regression models with morphology as a fixed effect and the case number as a random effect to assess the statistical significance of the differences in intensity scores between morphologies. 3R and MN423 intensity scores were higher for the ghost morphology compared to the mature morphology and the pretangle morphology, and for the mature morphology compared to the pretangle morphology. The intensity scores for AT8 and 4R tau were higher for the pretangle morphology compared to the ghost morphology and higher for the mature morphology compared to the ghost morphology (Figure 5C; Table S1).

As tangle morphology levels are most well described in hippocampal pyramidal neurons, we qualitatively investigated whether the morphologies associated with each immunophenotype in the EC were also observed in the CA1 region. We observed similar tangle morphologies and immunophenotypes in the CA1 region as what we saw in the EC. That is, tangles with pretangle morphology had high‐intensity labeling for AT8 or 4R tau, but not 3R or MN423. Conversely, ghost morphology tangles had high‐intensity labeling for 3R and MN423, but not AT8 or 4R. Tangles with a mature morphology were observed to have AT8 high, 4R high, mixed, or 3R high immunophenotypes (Figure S7 in supporting information).

4. DISCUSSION

Our findings demonstrate that tangle immunophenotypes can distinguish distinct tangle populations that reflect different maturity levels. By co‐labeling a panel of four tau antibodies that bind epitopes associated with early and advanced tangle maturity, we identified five distinct tangle populations across three brain regions affected at different stages of AD. The immunophenotypes support established literature on the timeline of tau modifications and progression of tangle morphologies with increasing disease severity (reviewed by Moloney et al. ⁴ ). We show that neurologically normal cases with age‐related tau pathology (Braak stage I–II) have a different tangle immunophenotype to cases with established AD neuropathological change (Braak stage III–VI). We also show that the 4R Tau high tangle immunophenotype is more abundant in the SFG, a brain region affected later in AD progression, and the most advanced tangle immunophenotypes (“3R Tau high” and “3R Tau MN423 high”) are associated with ghost tangle morphology. Therefore, our findings indicate that tangle immunophenotype classification is a quantitative approach that can be used to investigate the severity of tau pathology across the brain in AD.

Our pipeline of multiplexed tau antibody labeling paired with single‐object analysis offers several advantages over conventional approaches to quantifying tangle maturity. Co‐labeling multiple antibodies targeting different tau epitopes allowed us to visualize the heterogeneity of tangles on a single tissue section while maintaining spatial context. The main objective of our clustering analysis was to identify meaningful clusters in our data relating to distinct immunophenotypes that could be readily identified in future studies by assessing the relative intensity of the antibody labeling, without the need for repeated clustering analysis. Automated segmentation, immunophenotype classification, and quantification can improve studies of tangle maturity through increased throughput and reproducibility. Of the few studies to examine the relationship between tangle maturity and specific tau post‐translational modifications, tangle maturity was classified manually by semi‐quantitatively scoring the abundance of each morphology based on labeling of one tau antibody. ⁵ , ¹¹ , ²⁶ , ²⁷ This classification of tangle maturity is subjective, complicated by the presence of intermediary morphologies and the preferential labeling of certain morphologies by different antibodies. ¹¹ Our automated pipeline allows fast, unbiased quantification of many images and provides quantitative data at the single‐object level.

We identified five tangle immunophenotype clusters with labeling signatures that are representative of tangle maturity (Figure 6). The immunophenotypes we observed align with accumulated knowledge of the sequence of post‐translational modifications that occur with increasing tangle maturity. Recent proteomic analysis suggests post‐translational modifications first accumulate in the proline‐rich region of the tau protein and that 4R tau isoforms adopt a structure prone to fibril formation when phosphorylated in this region. ²⁸ This is coherent with our identification of predominantly AT8 high and 4R high immunophenotypes in tangles with pretangle morphology and in the later affected SFG region of Braak V to VI AD cases.

FIGURE 6.

Summary of tangle maturity based on immunophenotype from early to advanced. We identified five immunophenotypes that are representative of tangle maturity levels. Tangles with an “AT8 high” or “4R Tau high” immunophenotype had predominantly pretangle or mature tangle morphologies which align with current literature and indicate these immunophenotypes are indicative of early tangle maturity levels. ⁴ Tangles with a “Mixed” immunophenotype were observed with all types of morphology, but predominantly mature morphology. “3R Tau high” and “3R Tau‐MN423 high” tangles had predominantly ghost morphology and are considered more advanced tangles. Scale bar 10 µm.

In AD, tangles transition from containing majority 4R tau isoform, to mixed 4R and 3R, and finally to only 3R at the most advanced levels. ¹⁵ , ¹⁶ , ¹⁷ , ¹⁹ While the dynamics of this transition are unknown, histological studies suggest that the 3R isoform accumulates within the dendrites, followed by a gradual retraction of 4R tau first from the dendrites and then the soma. ¹⁹ Therefore, our 4R high cluster likely represents an early tangle immunophenotype and the mixed cluster likely represents a more advanced tangle immunophenotype. ⁴ , ¹⁷ , ¹⁹ The “3R Tau‐MN423 high” immunophenotype identified the most advanced tangles as truncation at E391, recognized by the MN423 antibody, is a modification reported in ghost tangles. ⁴ , ¹⁴ , ²⁹ Our morphology analysis supported this as the “3R Tau high” and “3R Tau‐MN423 high” immunophenotypes were most represented in the ghost morphology tangles. When interpreting these immunophenotypes, it should be noted that our labeling captures a static view of the dynamic process by which tangles mature. The “Mixed” immunophenotype likely encompasses the many intermediary tangle structures that occur during maturation. However, the “4R Tau high,” “3R Tau high,” and “3R Tau‐MN423 high” immunophenotypes provide a window into the extreme extents of the tangle maturity spectrum. Future studies should investigate the dynamics of this shift between immunophenotypes to better understand the mechanisms by which tangles mature. It should also be investigated whether these immunophenotypes are conserved in other mixed 3R/4R tauopathies such as chronic traumatic encephalopathy and primary age‐related tauopathy (PART).

Given that AD is described as a mixed 3R/4R tauopathy, the “AT8 high” immunophenotype was unexpected. Based on the proposed mechanisms of tangle maturity we expected that all AT8‐labeled tangles would also be strongly labeled with some combination of 4R and 3R tau antibodies. ⁴ The low intensity of 4R and 3R labeling in the “AT8 high” tangles suggests that these antibodies must recognize an epitope that is not readily available on these tangles and has been noted in a previous study, although not overtly discussed. ¹⁷ Phosphorylation at S202 and T205 is an early event in tangle formation, although AT8 has been reported to label all tangle morphologies in chromogenic labeling studies. ²⁶ , ²⁷ , ³⁰ , ³¹ Based on our findings that “AT8 high” tangles were mainly identified in the EC of Braak stage I to II cases and a high proportion of this immunophenotype have pretangle morphology, we propose that this is an early tangle maturity immunophenotype. The Braak stage I to II cases were neurologically normal individuals classified as having low AD neuropathological change, with a small number of tangles confined to the hippocampus considered to be age‐related and not classified as PART. Therefore, observing predominantly early “AT8 high” tangles in this group suggests that age‐related tauopathy may have a different immunophenotype signature than that of AD, and the relative co‐labeling of AT8, 4R, and 3R antibodies could help distinguish AD‐related tangles from age‐related tangles. Alternatively, the “AT8 high” immunophenotype may identify a very immature population of tangles that are less prevalent by the time Braak stage III to IV pathology has developed. This would suggest that “AT8 high” tangles eventually accumulate more tau proteins which are recognized by the 4R antibody. Further studies are required to determine whether the “AT8 high” cluster represents the earliest stages of tangle development or a divergent pathway of tangle formation.

It is well established that tangles progressively accumulate in more brain regions and transition from 4R to 3R in the hippocampus with increasing Braak stage. ¹ , ⁷ , ¹⁵ , ¹⁶ , ¹⁸ , ¹⁹ However, tangle maturity in regions outside of the medial temporal lobe is largely unexplored. When we compared the relative proportions of each tangle immunophenotype between the EC, MTG, and SFG of Braak V to VI cases, we observed that early maturity immunophenotypes were more abundant in the later affected SFG region. Specifically, the proportion of “4R Tau high” tangles was increased rather than the “AT8 high” immunophenotype, which supports the idea that early AD‐related tangles have distinct immunophenotype. Our data also support a progressive distribution of tangles as proposed by Braak staging where the quantity and maturity level of tangles increases within a brain region with increasing Braak stage, and regions affected at the earliest stages of the disease have more mature tangles.

The heterogeneity of tau tangles has potential implications for the interpretation of tau positron emission tomography (PET) imaging as the composition of tangles at different levels of maturity likely influences tau PET tracer binding. ³² PET tracers that bind individual isoforms may be limited as the tangle pathology changes over the course of the disease, and a mixed isoform tracer may miss early 4R or advanced 3R predominant pathology. ⁴ , ¹⁷ Multiple tracers may be required for differential disease diagnosis or to determine AD progression. Fluid biomarkers that detect the molecular diversity of tau tangle populations could also be useful indicators of pathological severity. Recent studies of cerebrospinal fluid from AD patients indicate that site‐specific phosphorylation of tau occurs at different stages of disease progression with some species such as p‐tau217 and p‐tau181 accumulating alongside amyloid beta decades before tau tangles form, while others such as p‐tau205 increase closer to symptom onset. ³³ These phosphorylated tau species have also been associated with pretangle morphology in post mortem tissue studies. ¹¹ Our multiplexed antibody panel could be used to explore the associations among p‐tau species, tau isoform composition, and tangle maturity level in future post mortem tissue studies to aid future interpretation of fluid biomarkers.

Last, methodological limitations contributed to our inability to classify the majority of tangles according to the morphology criteria defined in previous studies. ⁴ , ¹¹ Previous studies classified tangle morphology using a single antibody; however, to limit observer bias we merged our four antibody labels into one grayscale image, which likely affected the morphologies we observed. The use of thin tissue sections further limited our analysis as the bulk of some tangles will be outside the plane of the section.

In conclusion, we performed an immunohistochemical analysis of four tau antibodies that are established markers of early and late tangle maturity. We identified five distinct immunophenotypes that distinguish tangles at different maturity levels and correlate with Braak staging of AD pathology and standard tangle morphology classifications. Our tangle immunophenotype classification is a reproducible and efficient method of quantifying the maturity of individual tangles and can be used for future studies of regional tau pathology.

AUTHOR CONTRIBUTIONS

HCM, DH, and MAC designed the experiments. Tissue processing was performed by MAC, RLMF, HCM, and DH. Pathological examination was performed by CT. Experiments were carried out by DH. Image analysis and data processing were conducted by DH and HCM. Spectral clustering analysis was optimized and conducted by CR. The manuscript was prepared by DH and HCM, with critical revision by MAC and feedback from all authors. All authors have approved the final manuscript.

CONFLICT OF INTEREST STATEMENT

The authors declare no potential conflicts of interest. Author disclosures are available in the supporting information.

CONSENT STATEMENT

The tissue used in this study was donated with informed consent from the family prior to brain removal and all procedures were approved by the University of Auckland Human Participants Ethics Committee (Ref: 011654).

Supporting information

Supporting Information

Supporting Information

Supporting Information

Hamlin D, Ryall C, Turner C, Faull RLM, Murray HC, Curtis MA. Characterization of neurofibrillary tangle immunophenotype signatures to classify tangle maturity in Alzheimer's disease. Alzheimer's Dement. 2024;20:4803–4817. 10.1002/alz.13922

DATA AVAILABILITY STATEMENT

The data used for this study are available from the corresponding author upon reasonable request.