Artificial intelligence–based 3D segmentation of tangle‐associated TDP‐43 in neurodegeneration

Abstract

TAR DNA‐binding protein 43 (TDP‐43) inclusions are often associated with hyperphosphorylated tau, thus neurofibrillary tangles as the hallmark of Alzheimer's disease (AD) and primary age‐related tauopathy (PART). TDP‐43 in AD is associated with cognitive impairment, and while staging is known, the localization, cellular and inclusion characteristics of TDP‐43 are yet to be elucidated. We investigate relationships between TDP‐43 inclusions and the tangle maturation continuum in AD, PART, and co‐pathologies by multiplex immunostaining combined with artificial intelligence (AI)–based segmentation via object recognition, reconstruction, and quantification. We performed anti‐phosphorylated TDP‐43 immunofluorescence with phosphorylated tau labeling different stages and modifications of tangles (AT8, pS396, TauC3, MN423, GT38) in three controls, three cases with PART and TDP‐43 (PART‐TDP), five cases with high likelihood AD and TDP‐43 (AD‐TDP), and four cases of high likelihood AD with TDP‐43 and Lewy Body disease (AD‐TDP‐LBD). Confocal imaging was taken from eight regions: amygdala (amygdala‐BL and amygdala‐CM) and hippocampus (Cornu Ammonis (CA)‐1, CA2/3, CA4, dentate gyrus (DG), subiculum (SUB)), and entorhinal cortex (ERC) and quantified with AI segmentation to identify 3D spatial relations, thus the maturity of neurofibrillary tangle associated TDP‐43 (TAT) inclusions. TATs, which were either identified by pTDP‐43 and AT8 or pTDP‐43 and pS396 double positivity, were also investigated by Thioflavin S (ThioS) histochemistry. We found pS396 labeled mature TATs predominated in PART and AD in every region. Basolateral and centromedial amygdala displayed overall greatest number of pre‐TATs and mature TATs. Mature TATs were homogenously distributed among hippocampal subfields whereas CA4 and DG had the greatest mature TAT composition. ERC revealed closer numbers of pre‐TATs and mature TATs yet mature TATs predominated all groups. Unbiased AI‐based object identification, reconstruction, and TAT maturation analysis pipeline in conjunction with TDP‐43, tau, and ThioS multiplex immunostaining demonstrated unique aggregation and maturation patterns, highlighting region‐specific dynamics in the neurodegenerative processes of PART and AD.

Multiplex immunostaining of anti‐phosphorylated TDP‐43, tau (AT8, pS396, TauC3, MN423, GT38), and Thioflavin S combined with AI‐based object recognition, reconstruction, and TAT maturation analysis pipeline in limbic regions revealed that distinct aggregation and tangle maturation patterns underscored region‐specific dynamics in the neurodegenerative processes of PART and AD.

1. INTRODUCTION

Artificial intelligence (AI) quantification is an objective, standardized computational pathology method, which uses machine learning and deep learning models, such as convolutional neural networks, converting visual information into quantifiable outputs which are trained on large, complex datasets to learn variable features such as edges, texture, and shapes [1, 2]. On the other hand, semi‐quantification represents the traditional pathological burden assessment of a disease by a neuropathology expert's visual inspection of whole slide images [3, 4]. Although semi‐quantification data output is generally categorized as low‐resolution ordinal data [3] based on experts' subjective impressions of the observed pathology [4, 5] which may lead to high inter‐rater variability and inherent human bias, even among experienced experts [3, 4], 3D AI object recognition, reconstruction, and quantification outputs primarily consist of high‐resolution, continuous data [6].

Transactive response (TAR) DNA‐binding protein‐43 (TDP‐43) is an essential, highly conserved 414 amino acid heterogenous nuclear ribonucleoprotein (hnRNP), which primarily functions in the nucleus as a key regulator of RNA metabolism, including splicing and mRNA stability [7, 8, 9, 10]. Pathological TDP‐43 is characterized by its mislocalization from the nucleus to the cytoplasm, coupled with extensive post‐translational modifications (PTMs) including ubiquitination, cleavage, and hyper‐phosphorylation [7, 10, 11]. Aberrant phosphorylation at Serine 409/410 (S409/410) in the C‐Terminal domain, mediated by kinases like TTBK1/2 and CK1, is the most consistent and widely used hallmark of the pathology [12, 13].

TDP‐43 proteinopathy is the defining pathological feature of nearly all amyotrophic lateral sclerosis cases (~97%), half of frontotemporal lobar degeneration (FTLD) cases, and a common co‐existing pathology in AD (~20–75%) [9, 14] which correlates with increased tau aggregation and accelerating cognitive decline [7, 11, 15]. Furthermore, it may co‐exist with other age‐related tauopathies such as primary age‐related tauopathy (PART) [16, 17] in which TDP‐43 pathology has been named as limbic predominant age‐related TDP‐43 encephalopathy (LATE) neuropathologic changes [7, 18].

Tau pathology in AD and PART progresses through a defined sequence of phosphorylation, conformational change, truncation, and fibril stabilization inducing neurofibrillary tangle (NFT) formation [19, 20]. Early pre‐tangle tau shows abnormal phosphorylation at Ser202/Thr205, detected by AT8 [21], followed by additional phosphorylation at Ser396 recognized by pS396 as fibrils become more ordered [22]. Pre‐tangles are present in morphologically normal neurons with an intact nucleus, exhibiting diffuse or finely granular tau immunostaining in the cytoplasm and/or perinuclear regions [20, 23, 24]. In contrast, mature tangles are observed in neurons with either a reduced or mislocalized nucleus, accompanied by argyrophilic fibrils that adopt a “basket‐like” or “sling‐like” configuration in place of the normal cytoplasm [20, 25, 26]. NFTs exhibit neuron‐specific morphology, with mature tangles in pyramidal neurons typically manifesting as a flame‐like shape [20, 23, 27]. With further maturation, tau undergoes caspase‐ and protease‐mediated truncation, generating neoepitopes labeled by TauC3 (D421) and MN423 (E391), which mark late‐stage intracellular tangles and extracellular remnants [19, 28]. After neuronal death, only extracellular ghost tangles remain, enriched in truncation‐specific epitopes and representing the terminal stage of tau maturation [28]. The conformational antibody GT38 selectively binds aggregated AD‐type tau, distinguishing disease‐specific fibril structures [29].

Furthermore, synergistic interactions between TDP‐43 and tau drive the mutual misfolding of both proteins, with pathological TDP‐43 contributing to tau mRNA instability and an altered 3R/4R tau ratio [7, 30]. This co‐occurrence in neurons as tangle‐associated TDP‐43 (TAT) inclusions [31], which are associated with increased tau deposition and serve as an accelerator of cognitive decline, leads to significantly more severe clinical outcomes than isolated proteinopathies [7].

Amygdala, hippocampal subfield tau, and TDP‐43 assessments revealed differences in AD and FTLD subtypes [32, 33]. In AD or PART cases (non‐FTLD brains), TDP‐43 pathology is centrally accumulating within the limbic system [34, 35], and differentiated into subtypes, notably TDP‐43 Type‐α (featuring classical TDP‐43 type A FTLD‐like inclusions) and TDP‐43 Type‐β (where TDP‐43 uniquely co‐localizes with neurofibrillary tangles, i.e., TATs) [34]. In either Josephs or LATE_NC staging, TDP‐43 staging schemes delineate a hierarchical spread beginning in the amygdala (Stage 1) and progressing to the hippocampus (Stage 2) [36, 37, 38]. This pathological spread is critically linked to clinical outcomes, demonstrating an independent association with significant cognitive impairment, especially memory loss, and profound medial temporal atrophy—amygdala and hippocampal volume loss—even when adjusting for coexisting AD pathology [39]. Finally, neither the molecular features of TATs nor how they differ from those of other FTLD‐TDP inclusions in AD have not been fully elucidated yet.

The basolateral amygdala (amygdala‐BL) is the primary nucleus that both sends and receives information from the entorhinal cortex (ERC) and hippocampal areas (CA1–CA3 fields, subiculum (SUB), and para‐subiculum), which are recognized as atrophic in individuals with AD [40]. While amygdala‐CM is the major output nucleus of the amygdaloid complex, which is responsible for the reward/aversive based learning and execution of fight, flight, or freezing behavior, it receives modulatory input from the amygdala‐BL and shows early atrophy in AD [36, 39, 41]. The CA of the hippocampus is segmented into multiple subfields (CA1, CA2, CA3, and CA4) [42, 43]. The CA1 and CA2/3 regions are frequently recognized for the early emergence of numerous NFTs associated with aging; the CA4 region also displays a rise in tangles, which correlates with synaptic loss and the severity of dementia in normal aging and in AD [42, 43, 44, 45, 46]. On the other hand, the DG, a subfield of the hippocampus involved in learning and memory, especially in pattern separation [47, 48] and adult hippocampal neurogenesis [47], exhibits significant age‐related functional deterioration in humans, non‐human primates, and rodents [47]. Additionally, as highlighted by Braak and Braak in the early 1990s [15, 49], the trans‐entorhinal region (TER), ERC [50, 51], and hippocampus (including subiculum and CA1) are some of the initial regions identified as being impacted by NFT pathology [15, 49, 50, 51]. TDP‐43 and tau concurrence in ERC have been associated with cognitive decline and most vulnerable neuronal populations [52, 53].

In this study, we used multiplex immunofluorescence and 3D AI object identification and quantification approaches to examine and evaluate the distribution of the maturity of TATs inclusions in limbic system subfields of AD and PART cases exhibiting distinct co‐pathologies. Morphological quantification could leverage 3D AI segmentation to drive complex metrics that would define the structural quality and phenotypic complexity of the TDP‐43 aggregates including TATs [31, 54]. We predict that TATs will exhibit more colocalization with ThioS histochemistry and be linked to pS396 labeled mature tangles more than AT8‐positive pre‐mature/early tangles.

2. METHODS AND PROTOCOLS

2.1. Patient recruitment

Between December 2020 and October 2023, we prospectively recruited participants including three controls, three PART‐TDP, five hAD‐TDP, and four hAD‐TDP‐LBD aggregates that were referred to and enrolled in one of several NIH‐funded studies conducted by the Neurodegenerative Research Group (NRG) at Mayo Clinic, Rochester, Minnesota. The participants who have died with autopsy and pathological confirmation of no neurodegenerative disease process were recruited as controls. All participants were evaluated by a behavioral specialist (KAJ) and neuropathologist (RRR). Informed consent was obtained from all subjects for participation in the study, which was approved by the Mayo Institutional Review Board. The experiments and cases have been reported in compliance with the STROBE statement [55].

2.2. Multiplex immunofluorescence studies

A total of eight regions of interest (ROIs) from two formalin fixed paraffin‐embedded (FFPE) human brain tissue blocks spanning the limbic system (amygdala and hippocampus) were selected for analysis and neuropathological examination. These included: (1) amygdala‐BL, (2) amygdala‐CM, (3) CA1, (4) CA2/3, (5) CA4, (6) DG, (7) SUB, and (8) ERC. We quantified these tau and TDP‐43 colocalizations to further elucidate the internuclei variability which might predict the associated pathological alterations [32]. Block sections were cut into 10 μm to include all cell types without interruption. The sections were permeabilized with 0.3% Triton‐X 100‐1X phosphate buffered saline (PBS) at room temperature for 30 min, then incubated with 10 mM sodium citrate (pH: 6.0) at 80°C for 30 min. The sections were incubated with unconjugated AffiniPure™ F(ab′)₂ Fragment Donkey Anti‐Human IgG (H + L) (Jackson Immunoresearch, catalog # 709‐006‐149), diluted 1:35 in 1X PBS overnight at 4°C. On the following day, the sections were incubated in blocking buffer (2% normal donkey serum and 0.1% bovine serum albumin (BSA) in 1X PBS) for 1 h, and then incubated with primary antibodies overnight at 4°C. After washing, the sections were incubated with secondary antibodies for 2 h (Jackson Immunoresearch, 1:200 in 1X PBS). Samples were then incubated with TrueBlack® lipofuscin autofluorescence quencher (Biotium, catalog #23007) for 3 min at room temperature, followed by 1X PBS washes for five mins three times. Stained sections were mounted on glass slides in antifade mounting medium (Vector laboratories, VECTASHIELD Vibrance® Antifade Mounting Medium (H‐1700)). Control sections omitting either primary or secondary antibodies showed no signal above background.

To assess the overall tau and TDP‐43 distribution across the samples, primary antibodies were obtained from the following sources and used at various dilutions: (1) anti‐AT8 phosphorylated tau (Thermo Fisher Scientific, catalog #MN1020, 1:200), which detects pre‐stable tangles; and (2) anti‐tau pS396 (Cell Signaling Technologies, F3S9T, catalog # 48856S, rabbit monoclonal, 1:200), which detects phospho‐S396 tau associated with mature forms [20], (3) anti‐TauC3 (Millipore, Anti‐Tau Antibody, Caspase Cleaved (truncated at Asp421), mouse monoclonal, catalog #MAB5430, 1:100), (4) anti‐tau MN423 (Absolute Antibody, Anti‐Tau [MN423], Mouse IgG1, Kappa, mouse monoclonal, catalog # Ab02389‐1.1, 1:200), (5) anti‐tau GT38 (Abcam, Clone GT‐38, mouse monoclonal, catalog #ab246808, 1:100), (6) anti‐phospho TDP‐43 (pS409/410) (CosmoBio, mouse monoclonal, catalog #TIP‐PTD‐M01, lot #11‐9‐G6, 1:400), (7) anti‐phospho TDP‐43 (pS409/410) (Proteintech, rabbit polyclonal, catalog #22309‐1‐AP, 1:200). Secondary antibodies were obtained from the following sources and used at 1:200 dilutions: Alexa Fluor 568 Donkey Anti‐Rabbit IgG (H + L) (Jackson Immunoresearch, 711‐575‐152), Alexa Fluor® 647 AffiniPure™ Donkey Anti‐Mouse Pig IgG (H + L) (Jackson Immunoresearch, 715‐605‐151).

2.3. Thioflavin S histochemistry

Once the immunolabeling protocols were performed, sections were incubated in freshly prepared and filtered 1% w/v aqueous Thioflavin‐S (ThioS) (Sigma‐Aldrich, Cat. # T1892) for 8 min at room temperature. Then, slides were washed two times for 3 min in decreasing percentages of 95% v/v, 80% v/v, 70% v/v, 50% v/v ethanol, and finally washed with three exchanges of distilled water (ddH20). Samples were then incubated with TrueBlack® lipofuscin autofluorescence quencher (Biotium, catalog #23007) for 3 min at room temperature and then washed in 1X PBS for five mins three times. Stained sections were mounted on glass slides in antifade mounting medium (Vector laboratories, VECTASHIELD Vibrance® Antifade Mounting Medium (H‐1700)).

2.4. Manual burden assessment and colocalized TAT counting

All multiplex images were prepared with spinning disc Zeiss 980 LSM under identical laser illumination and capture conditions. Specifically, ThioS immunofluorescence was detected via a 445 nm laser under the confocal microscope. Both the people taking and analyzing the images were blinded to the treatment conditions. For brain sections, confocal images were taken at three or four locations that are at least 100 μm apart from each other under 20× magnification. The number of images taken from each region was dependent on the size of the spanned area which was significantly affected by disease processes. The confocal images were z‐stacked, and each processed image was prepared using a stack of ten 1‐μm‐thick Z‐stage images. The images were opened by Fiji/ImageJ using BioFormats Importer, maximally projected into 2D, and channels were split. Each channel belonging to pTDP‐43, tau species, and ThioS was separately thresholded using Triangle function and percentage of area coverage was obtained as burden. For colocalization studies, each channel was then merged and superimposed. The number of colocalized particles was counted using Analyze Particles function, verified manually, and the ratio of colocalized particles was calculated for each pair of analysis.

2.5. 3D AI‐assisted object recognition, reconstruction, and relation count quantification

To investigate the pTDP‐43 inclusions and the spatial relationship between tau depositions, we employed 3D AI object recognition, reconstruction, and analysis workflow to quantify the number of colocalized particles as pre‐ and mature TAT inclusions in each group (Figure S1, Supporting Information). The resultant demonstrative images of pTDP‐43, AT8, and ThioS from single plane and 3D Z‐stack confocal imaging were also compared side by side in Figure S1C. Because of superior inclusion capture and 3D assessment capabilities, 3D constructed z‐stacks were opened in AIVIA 15.0 Software (Leica Microsystems). Alexa Fluor 647 and 568 labeled pTDP‐43, Alexa Fluor 647 labeled AT8, Alexa Fluor 568 labeled pS396, and ThioS channels were separately trained in the software to detect all positive particles with supervision. All AI pixel classifier training was conducted with four to five images from various regions of two to three cases per pathological group in a supervised fashion. We used randomly assigned, distinct ROIs from the same cases both to train the AI segmentation algorithm and analyze the TATs. The background and actual signals were defined manually. After saving each classifier file specific to immunolabeling, a custom‐made workflow (macro) was created to reconstruct objects in 3D, recognize the labeled ones, quantify the morphometric data and dual spatial relationships between pTDP‐43 and AT8, pTDP‐43 and pS396, pTDP‐43 and ThioS. AIVIA analyzed all raw czi Z‐stacks supervised via this workflow and saved them in a predetermined folder. We referred to pre‐tangle and mature tangle associated TDP‐43 inclusions as pre‐TAT and mature TATs to sustain coherence and simplicity.

2.6. Statistical analysis

The “n” of each study defined as the number of independent experiments from each case. Numerical data were expressed as median ± interquartile range (Q1–Q3), analyzed using Kruskal–Wallis's test where multiple groups were compared to each other. Where only two groups were compared, the Mann–Whitney U test was used. Simple linear regression was performed to investigate the total burden and TAT relation count relationship in pre vs. mature TATs. All data analysis was performed by individuals who were blinded to the experimental conditions.

3. RESULTS

3.1. Patient demographics

Across all 15 cases, the median age at presentation was 93 years with an average age of 87.5 years. All the PART and h‐AD cases were at similar ages at death (93 ± 5.5) whereas controls were younger (67 ± 12.5). Four patients (26.7%) were female. TDP‐43 pathological stage was similar in each disease group demonstrating low variability of distribution of the TDP‐43 inclusions [37, 39]. Braak staging were similar at death within AD group (5.2 ± 0.2) and participants diagnosed with PART (3 ± 0.57). Except for the control cases, all subjects had secondary tauopathies such as argyrophilic grain disease (AGD), and age‐related tau astrogliopathy (ARTAG) [56, 57, 58]. Patient demographics were outlined in Table 1.

TABLE 1.

Patient demographics.

Case no.	Biological sex	Age at death (years)	CERAD dx	CERAD score	Thal phase	Braak stage	HpScl	ADNC	TDP‐43 (LATE) stage	Final diagnosis
1	M	91	Def.	3	5	VI	−	A3B3C3	2	h‐AD
2	M	87	Moderate	2	5	V	−	A3B3C2	2	h‐AD
3	M	94	Def.	3	4	V	−	A3B3C3	2	h‐AD
4	F	93	Def.	3	5	V	−	A3B3C3	2	h‐AD
5	M	99	Moderate	2	5	V	−	A3B3C2	2	h‐AD
6	M	89	Def.	3	5	VI	+	A3B3C3	2	h‐AD
7	M	93	Def.	3	3	VI	+	A3B3C3	3	h‐AD
8	M	96	Moderate	2	4	IV	+	A3B2C2	2	h‐AD
9	F	93	Def.	3	4	V	−	A3B3C3	2	h‐AD
10	M	86	None	0	1	II	+	A1B1C0	2	Possible PART
11	F	93	None	0	0	II	−	A0B1C0	2	Definite PART
12	F	97	None	0	0	IV	+	A0B2C0	2	Definite PART

Note: Demographic characteristics of each patient from this study.

Abbreviations: ADNC, Alzheimer's disease neuropathologic change; CERAD, Consortium to Establish a Registry for Alzheimer's disease; h‐AD, high likelihood of Alzheimer's disease; HpScl, hippocampal sclerosis; LATE, limbic‐predominant age‐related TDP‐43 encephalopathy; PART, primary age‐related tauopathy; TDP, transactive response DNA‐binding protein 43.

3.2. TDP‐43 and tau species spatial relationship in neurodegeneration

We immunolabelled sections passing through amygdala with pTDP‐43 (red), and all tau species in cyan: pre‐tangle marker AT8, mature tangle marker pS396, caspase cleaved tau‐C3, mature confirmation specific tau MN423 clone, and ghost tangle marker GT38 in PART‐TDP, AD‐TDP, and AD‐TDP‐LBD cases (Figure 1). AT8 and pS396 demonstrated the greatest pTDP‐43 colocalizations in all neurodegenerative groups by visual assessment. TauC3 (Figure 1C) and GT38 (Figure 1E) did not show much immunolabeling in amygdala‐BL whereas MN423 displayed occasional staining and colocalization with pTDP‐43. It should also be noted that although MN423 labeling was morphologically comparable to pTDP‐43 aggregates, the total number of colocalizations was greater in AT8 and pS396 labeled sections (Figure 1D vs. Figure 1A,B).

FIGURE 1.

Neurofibrillary tangle maturation and TDP‐43 co‐assessment in amygdala‐BL. Each row represents merged images of immunofluorescent staining of the controls, PART‐TDP, AD‐TDP, and AD‐TDP‐LBD with phosphorylated TDP‐43 (pTDP‐43) (red), (A) pre‐tangle marker phosphorylated tau (AT8) (cyan), (B) mature tangle marker pS396 (cyan), (C) mature tangle marker TauC3 (cyan), (D) mature/ghost tangle marker tau isotype MN423 (cyan), and (E) ghost tangle marker conformational tau GT38 (cyan) to demonstrate spatial relationship of pTDP‐43 aggregates in each condition. White rectangles show where the inserts are taken. Scale bars = 20 μm.

3.3. Neurofibrillary TAT assessment in amygdala‐BL

First, we immunolabelled amygdala sections with pTDP‐43 (red), pre‐tangle marker AT8 (magenta), mature tangle marker pS396 (cyan), and aggregated protein marker ThioS (green) in PART‐TDP, AD‐TDP, and AD‐TDP‐LBD cases (Figure 2A). We trained the AI‐based detection software to capture tau and TDP‐43 inclusions in an unbiased manner (Figure S1A,B). Then, we assessed whether single‐plane or 3D Z‐stack confocal imaging should be performed to identify the relationship among pTDP‐43, tau, and ThioS. 3D z‐stack confocal imaging helped us capture the whole slide and create 3D objects which gave us more accurate results of where these proteins aggregate in tissue (Figure S1C). We quantified AT8 and pS396‐labeled TATs manually and through AI‐based object recognition and reconstruction pipeline (Figures 2 and S1). In all groups, manual assessment of TATs in amygdala‐BL demonstrated that pTDP‐43 inclusions had more colocalization counts with mature tangle marker pS396 than AT8‐labeled pre‐tangles (Figure 2B–D). Interestingly, while overall AT8 immunoreactivity was more prominent in AD‐TDP‐LBD cases than in both groups, pTDP‐43 colocalized pS396 inclusions were found to be higher (Figure 2D). Relative relation count quantifications (Figure 2E), which assessed the relative burden of AT8 and pS396 labeled pTDP‐43 inclusions, showed a trend in increase in the percentage of mature TATs from PART‐TDP through AD‐TDP and AD‐TDP‐LBD. In other words, PART‐TDP exhibited the lowest AT8 to pS396 ratio, AD‐TDP‐LBD group had a higher number of pre‐tangle TATs (Figure 2E). Additionally, although pTDP‐43 colocalizations with pS396 and ThioS (Figure S2B,C) did not reveal any difference among these 3 groups, the percentage of AT8 and pTDP‐43 double positive inclusions was shown to differ between PART and AD groups (Figure S2A). We implemented our AI‐based object recognition, reconstruction, and analysis pipeline and verified that the results were comparable to manual counting of TATs (Figure 2F–I). The percentage of counts was greater in AI‐based assessment because the software quantifies each particle and every spatial relationship in 3D fashion (Figure 2F–I). Thus, verification led us to use this paradigm in the rest of the manuscript.

FIGURE 2.

Neurofibrillary tangle associated TDP‐43 (TAT) assessment in amygdala‐BL. (A) Immunofluorescent staining of the controls, PART‐TDP, AD‐TDP, and AD‐TDP‐LBD with phosphorylated TDP‐43 (pTDP‐43) (red), pre‐tangle marker phosphorylated tau (AT8) (magenta), mature tangle marker pS396) (cyan), ThioS (green). Each row represents merged images of pTDP‐43 and AT8, pS396 and ThioS to demonstrate spatial relationship of pTDP‐43 aggregates in each condition. White rectangles show where the inserts are taken. Scale bars = 20 μm. TAT manual counting performed in (B) PART‐TDP, (C) AD‐TDP, and (D) AD‐TDP‐LBD cases. (E) Relative relation count (AT8 vs. pS396) in each group. AI segmentation and recognition of 3D objects from confocal z‐stack in each staining are performed. The relation counts between pTDP‐43 and AT8 are compared to pTDP‐43 and pS396 in (F) PART‐TDP, (G) AD‐TDP, and (H) AD‐TDP‐LBD cases. (I) Relative relation count (AT8 vs. pS396) in each group. *p < 0.05, **p < 0.01, ns = nonsignificant, Kruskal–Wallis for grouped variables, Mann–Whitney U for pairwise comparisons.

3.4. Total and TAT burden assessment in amygdala‐BL

We wanted to investigate if the total burden of the proteins of interest was correlated with TAT burden or colocalization counts. We first quantified pTDP‐43 (Figure 3A), AT8 (Figure 3B), pS396 (Figure 3C), and ThioS (Figure 3D) burden in amygdala‐BL. AD‐TDP‐LBD group demonstrated greatest burden of all proteinopathies and ThioS. AT8+ pre‐TATs, pS396+ mature TATs, and ThioS‐labeled mature TATs all revealed the similar pattern that AD‐TDP‐LBD and PART‐TDP showed greatest and lowest TAT burden, respectively (Figure 3E–G). Linear regression analysis demonstrated that pre‐TATs and mature TATs separated as groups; however, we could not demonstrate a significant relationship between total burden of tau and TAT relation counts in both groups where Braak stages were comparable (Figure 3H).

FIGURE 3.

Correlative pTDP‐43, AT8+ tau, pS396+ tau, ThioS, and neurofibrillary TAT burden assessment in amygdala‐BL. Total burden assessment of (A) pTDP‐43, (B) AT8+ tau, (C) pS396+ tau, and (D) ThioS in amygdala‐BL of PART‐TDP, AD‐TDP, and AD‐TDP‐LBD cases. TAT burden assessment of (E) AT8+ pre‐TATs, (F) pS396+ mature TATs, (G) ThioS+ mature TATs. *p < 0.05, **p < 0.01, ns = nonsignificant, Kruskal–Wallis for grouped variables, Mann–Whitney U for pairwise comparisons. (H) Linear regression analysis of pre vs. mature TAT comparison in total burden and TAT relation count correlations. Red circle: pre‐TAT, green square: mature TAT.

3.5. Neurofibrillary TAT assessment in amygdala‐CM

Moving forward, amygdala‐CM demonstrated slightly different results (Figure 4). Although mature tangles were associated with pTDP‐43 in higher percentages than pre‐tangles like amygdala‐BL (Figure 4A–C), the AD‐TDP‐LBD group had shown closer numbers of pre and mature TATs (Figure 4D) which indicated increased early and mature tau aggregation in pTDP‐43 inclusions. Compared to amygdala‐BL, amygdala‐CM also revealed a pattern of pre‐tangle/mature TAT distribution (Figure 4E) in which PART‐TDP demonstrated the highest percentage of pS396 labeled mature TATs. Whereas the percentage of AT8‐labeled pre‐TATs were least in PART‐TDP, higher in AD‐TDP and highest in AD‐TDP‐LBD (Figures 4E and S2D), pS396 and ThioS labeled mature TATs were similar in every group (Figure S2E,F).

FIGURE 4.

Neurofibrillary TAT assessment in amygdala‐CM. (A) Immunofluorescent staining of the controls, PART‐TDP, AD‐TDP, and AD‐TDP‐LBD with pTDP‐43 (red), AT8 (magenta), pS396 (cyan), ThioS (green). Each row represents merged images of pTDP‐43 and AT8, pS396 and ThioS to demonstrate spatial relationship of pTDP‐43 aggregates in each condition. White rectangles show where the inserts are taken. Scale bars = 20 μm. AI segmentation and recognition of 3D objects from confocal z‐stack in each staining are performed. The relation counts between pTDP‐43 and AT8 are compared to pTDP‐43 and PS396 in (B) PART‐TDP, (C) AD‐TDP, and (D) AD‐TDP‐LBD cases. (E) Relative relation count (AT8 vs. pS396) in each group. *p < 0.05, **p < 0.01, ns = nonsignificant, Kruskal–Wallis for grouped variables, Mann–Whitney U for pairwise comparisons.

3.6. Neurofibrillary TAT assessment in hippocampal subfields

We dissected hippocampal subfields to assess whether co‐pathologies differentiated participants in the TAT maturation process. CA1, CA2/3, CA4, DG, and SUB harbored a higher number of mature TATs than pre‐TATs in all groups, although they demonstrated different sized and shaped inclusions with variance in burden (Figures 5, 6, and S3–S7). Looking into AT8, pS396, and ThioS labeled TATs, mature TATs were always more abundant than pre‐TATs in all groups (Figure 5B–D). Surprisingly, relative relation counts did not distinguish the groups in any of the hippocampal subfields as well (Figures 5, 6, and S3–S7), and all the other parameters failed to show unique pathological processes regarding tau maturation in TDP‐43 inclusions (Figures S4 and S7).

FIGURE 5.

Neurofibrillary TAT assessment in CA1 of hippocampus. (A) Immunofluorescent staining of the controls, PART‐TDP, AD‐TDP, and AD‐TDP‐LBD with pTDP‐43 (red), AT8 (magenta), PS396 (cyan), ThioS (green). Each row represents merged images of pTDP‐43 and AT8, pS396 and ThioS to demonstrate spatial relationship of pTDP‐43 aggregates in each condition. White rectangles show where the inserts are taken. Scale bars = 20 μm. AI segmentation and recognition of 3D objects from confocal z‐stack in each staining are performed. The relation counts between pTDP‐43 and AT8 are compared to pTDP‐43 and pS396 in (B) PART‐TDP, (C) AD‐TDP, and (D) AD‐TDP‐LBD cases. (E) Relative relation count (AT8 vs. pS396) in each group. *p < 0.05, ns = nonsignificant, Kruskal–Wallis for grouped variables, Mann–Whitney U for pairwise comparisons.

FIGURE 6.

Neurofibrillary TAT assessment in CA2/3 of hippocampus. (A) Immunofluorescent staining of the controls, PART‐TDP, AD‐TDP, and AD‐TDP‐LBD with pTDP‐43 (red), AT8 (magenta), pS396 (cyan), ThioS (green). Each row represents merged images of pTDP‐43 and AT8, PS396 and ThioS to demonstrate spatial relationship of pTDP‐43 aggregates in each condition. White rectangles show where the inserts are taken. Scale bars = 20 μm. AI segmentation and recognition of 3D objects from confocal z‐stack in each staining are performed. The relation counts between pTDP‐43 and AT8 are compared to pTDP‐43 and pS396 in (B) PART‐TDP, (C) AD‐TDP, and (D) AD‐TDP‐LBD cases. (E) Relative relation count (AT8 vs. pS396) in each group. *p < 0.05, **p < 0.01, ns = nonsignificant, Kruskal–Wallis for grouped variables, Mann–Whitney U for pairwise comparisons.

3.7. Neurofibrillary TAT assessment in ERC

We identified pre and mature TATs in entorhinal areas as well (Figure 7). Apart from the amygdala and hippocampus, ERC had shown a high number of pre‐TATs on top of mature TATs (Figures 7A–D and S7). Although demonstrating a greater number of pre‐TATs and AT8 burden, mature TATs in ERC were more prominent in all groups (Figure 7B–D). Interestingly, like other regions we analyzed, relative relation count quantification also failed to show differences because of high variability among groups (Figure 7E).

FIGURE 7.

Neurofibrillary TAT assessment in ERC. (A) Immunofluorescent staining of the controls, PART‐TDP, AD‐TDP, and AD‐TDP‐LBD with pTDP‐43 (red), AT8 (magenta), pS396 (cyan), ThioS (green). Each row represents merged images of pTDP‐43 and AT8, pS396 and ThioS to demonstrate spatial relationship of pTDP‐43 aggregates in each condition. White rectangles show where the inserts are taken. Scale bars = 20 μm. AI segmentation and recognition of 3D objects from confocal z‐stack in each staining are performed. The relation counts between pTDP‐43 and AT8 are compared to pTDP‐43 and pS396 in (B) PART‐TDP, (C) AD‐TDP, and (D) AD‐TDP‐LBD cases. (E) Relative relation count (AT8 vs. pS396) in each group. *p < 0.05, **p < 0.01, ns = nonsignificant, Kruskal–Wallis for grouped variables, Mann–Whitney U for pairwise comparisons.

3.8. Comparative pre and mature TAT assessment in limbic system

Comparing all subfields had elucidated that pre‐TATs were lowest in CA4 and DG, whereas amygdala‐BL and ERC had the highest percentage of them in each PART and AD group (Figure S8A–C). However, only the AD‐TDP‐LBD group demonstrated the amygdala‐BL's prominent expression of pre‐TATs (Figure S8C). Strikingly, the amygdala‐BL of the AD‐TDP group showed the highest percentage of pS396‐labeled mature TATs among the limbic areas, whereas CA4 was distinguished in the AD‐TDP‐LBD group (Figure S8E).

4. DISCUSSION

Our previous studies in AD, PART, and FTLD demonstrated the importance of tau and TDP‐43 in neurodegeneration and disease progression [16, 31, 33, 34, 36, 37, 39]. In this study, besides the molecular machinery causing post‐translational modifications, aggregations, and seeding of TDP‐43 and tau in neurodegenerative disorders, the amygdala, as the first region in which TDP‐43 emerges in AD, has been our main target to assess the maturity of TATs [36, 37]. Because of severe neurodegeneration in AD, the architectural integrity of the amygdala nuclei was substantially compromised, limiting our analysis to two principal regions: amygdala‐BL and amygdala‐CM [41]. As it has been known that amygdala‐BL is responsible for the evaluation of emotions and sensory information, such as fear/reward mechanisms, whereas the amygdala‐CM leads to the execution of emotional reactions (such as freezing, fight‐or‐flight) by modulatory input from the amygdala‐BL. We analyzed the tangle maturation continuum versus TDP‐43 inclusion accumulation patterns in the amygdala‐BL, as well as comparative correlation between TAT, tau, and TDP‐43 burden with multiplex immunostaining of pre‐mature/early, mature, and hypermature/ghost tangle tau markers labeling different tau phosphorylation sites and conformation, phosphorylated TDP‐43, and amyloidogenic protein aggregation via ThioS [20, 59].

Our findings demonstrate that biochemical staging and morphological assessment do not contribute equally to identifying tau pathology in the amygdala. Although the maturation sequence outlined in the introduction predicts increasing specificity from AT8 and pS396 toward late‐stage truncation markers such as TauC3, MN423, and the AD‐selective conformational antibody GT38, our data show that early and mid‐mature stage phospho‐tau species remain the most informative for detecting pathological convergence with pTDP‐43. AT8 and pS396 exhibited the highest colocalization across PART‐TDP, AD‐TDP, and AD‐TDP‐LBD, underscoring their sensitivity to shared pathogenic environments. In contrast, TauC3 and GT38 showed minimal labeling in the amygdala‐BL, consistent with the scarcity of late intracellular tangles and extracellular ghost tangles in this region. Although MN423 displayed morphology resembling pTDP‐43 aggregates, its lower colocalization frequency highlighted that structural similarity alone did not reliably indicate the extent of biochemical overlap. We also noted a variety of morphologies in TAT inclusions, frequently resembling a bitten apple or having a flame‐like appearance [23, 27, 31]. Additionally, previous observations revealed heterogeneous staining patterns in certain lesions, with regional variability in staining intensity [31]. While the underlying cause remains unclear, this variability may relate to the relative abundance of TDP‐43 and tau present in the examined regions. Thus, consistent 3D AI‐based analysis together with thicker‐than‐standard tissue sections (10 μm), allowing inclusion of at least one intact neuron within each 3D plane helped us to address this variability. Collectively, these results argue that phospho‐tau markers outperform morphology in capturing early pathological intersections between tau and TDP‐43, particularly in regions where late‐stage tangles are limited.

We determined the colocalization of 3D AI reconstructed particles, thereby removing the subjective nature of pathology assessment and resulting in low inter‐rater variability along with reduced selection/diagnosis bias. We hypothesized that automated 3D AI segmentation and relation count quantifications would also eliminate the possibility of accidental colocalizations compared to 2D maximally projected images which have been analyzed through Pearson Correlation Coefficient (PCC) [7, 60]. We compared AI segmentation efficiency using a single optical plane versus the full multi‐z‐stack reconstruction 3D confocal imaging and by manually counting TATs along with quantifying burden of these proteins. Incorporating the full z‐stack markedly enhanced detection and classification accuracy for tau, TDP‐43, and ThioS‐labeled inclusions. The improvement was especially pronounced for pre‐ and mature TATs, where capturing 3D structure across the tissue depth was essential for correct morphological identification. Moreover, the extensive image volume required for such analyses renders this manual approach impractical for routine clinicopathological application. By contrast, the automated AI‐based workflow enabled rapid and unbiased quantification, making it more suitable for large‐scale and translational studies [54] without semi‐quantitative expert evaluation.

By implementing AI‐based 3D segmentation that was comparable to manual counting, we showed a predominance of pS396 labeled mature TATs in PART and h‐AD. Amygdala‐BL became a denominating region in all TAT analyses by demonstrating the highest percentage of mature TATs in the AD‐TDP group and pre‐TATs in the AD‐TDP‐LBD. Meanwhile, amygdala‐CM also displayed the propensity to have a greater proportion of mature TATs, suggesting that the pathological maturation of TATs may occur earlier in this region. Consequently, the ongoing degenerative process, thus atrophy, and high number of pre and mature TATs in amygdala‐BL might account for explicit memory deficits in encoding stimuli in cases with AD‐TDP [40, 41].

We observed that across all hippocampal subfields, mature TATs markedly outnumbered pre‐TATs in PART and AD groups. Given that phosphorylated tau burden generally increases with advancing Braak stages [15], with evidence of plateauing at later stages, we tended to speculate that a higher proportion of mature TATs reflects accelerated neurodegeneration within the hippocampus across disease groups [40, 41]. In parallel with prior studies demonstrating regional vulnerability—particularly the increased tangle burden in CA1 and SUB detected by ThioS in AD—our findings showed that ThioS–labeled mature TATs predominated across hippocampal subfields as tangle pathology advances [20].

In one of the interesting results, while CA4 and DG displayed the lowest pre‐TAT percentages across all co‐pathologies, CA4 emerged as a distinguishing region in the AD‐TDP‐LBD group displaying the highest proportion of mature TATs. Although the pathological alterations in the CA4 region have not been as extensively researched as those in CA1 and CA2/3, the accumulation of NFTs, thus comparatively greater loss of neurons in the CA4 region correlated with synaptic loss and the severity of dementia as well as in normal aging [61]. Surprisingly, even though α‐synuclein aggregation in the form of Lewy bodies is classically associated with synucleinopathies such as dementia with Lewy bodies, emerging evidence indicates that α‐synuclein also influences TAT maturation within limbic regions in AD cases [62]. Similarly, DG, a neurogenic niche supporting adult hippocampal neurogenesis, exhibits a significant increase in mature TATs [47]. Nevertheless, the magnitude of the transition from pre‐ to mature TATs in the DG was less pronounced than that observed in CA4, suggesting region‐specific differences in TAT maturation dynamics. Despite subtle regional heterogeneities in the hippocampus, CA4 and DG exhibited the lowest pre‐TATs whereas the Amygdala‐BL and CA4 had the highest percentage of mature TATs. Presence of α‐synuclein aggregation as Lewy bodies in h‐AD favored AT8‐labeled pre‐TAT accumulation and subfield differentiation while pS396‐labeled mature TATs showed regional differences in cases with no LBD.

Concurrent higher percentages of pre and mature TATs in ERC of PART and h‐AD cases were identified although mature TATs still predominated across pathological groups. In contrast, h‐AD with Lewy bodies exhibited the highest burden of pre‐TATs both in Amygdala‐BL and ERC. This observation is consistent with the established role of the ERC as one of the earliest regions affected by neurofibrillary tangle pathology [15, 21, 27, 32, 52]. Also considering the fact that AT8 immunostains both pre‐tangles and mature tangles as shown in prior studies while pS396 primarily labels mature tangles [22, 63, 64, 65], the elevated proportions of both pre‐TATs and mature TATs in the Amygdala‐BL and ERC may reflect a distinct trajectory of disease progression compared with other limbic regions. In other words, tau maturity of TATs in amygdala and ERC differentiated the co‐pathologies in AD by showing high AT8 and pS396 labeling in TATs, a relationship that pointed out the ongoing widespread disease progression [52, 66].

4.1. Strengths and limitations

Considering the strengths and limitations of our study, to our knowledge, this study represents the first co‐application of multiplex immunostaining of TDP‐43 combined with colocalized assessment of pre‐tangles, mature and hypermature/ghost tangles using several tau markers and ThioS, integrated with 3D AI‐based object training, recognition, reconstruction, and quantification. It is worth mentioning that this approach enabled detailed analysis of tau‐TDP‐43 colocalizations and thus TAT maturation across limbic subfields, revealing not only a tight association between mature tangles and TDP‐43 aggregation but also region‐specific differences in their distribution. Importantly, our use of unbiased sample selection among advanced participants along with blinded AI‐driven measurements minimized observational bias, the subjectivity inherent to semi‐quantitative analyses, thus increasing the generalizability of our results in a similar cohort with minimal confounding variables.

Limitations of the study include the absence of comprehensive burden and morphological particle analyses in every region with multiple tau antibodies, and the lack of evaluation of how co‐pathologies such as AGD and ARTAG influence TAT maturation. Braak staging, tangle and TAT progression models were not analyzed because the cases were selected to underline how TDP‐43 interacts with tau in a homogenous group. In other words, the relative homogeneity of Braak stage in this cohort, because of the fact that our focus was to identify the patterns of how tau and TDP‐43 were overlapping, limits the utility of Braak‐based comparisons and stratification.

Also, our cohort is composed of relatively older participants. It should be noted that AD and PART cases with TDP‐43 pathology are known to age further than TDP‐43‐negative counterparts [36, 37, 39, 67, 68]. Thus, we followed STROBE statement guidelines to give details of our participant selection criteria [55].

Additionally, molecular assessments of tau and TDP‐43, such as co‐immunoprecipitation or phospho‐proteomic studies, were not performed and could provide further insight into the mechanistic interactions between these proteins in neurodegeneration. These observations highlight the need for further studies to fully elucidate the characteristics of tau–TDP‐43 interactions in the context of neuronal biology and disease progression.

5. CONCLUSION

To conclude, our study provides a detailed characterization of the maturation of neurofibrillary tangles associated with pathological TDP‐43 aggregation in limbic regions of PART and h‐AD using advanced, innovative 3D AI segmentation tools. Multiplex immunostaining combined with unbiased AI‐based object recognition, reconstruction, and TAT maturation analysis pipeline revealed distinct aggregation and maturation patterns, underscoring region‐specific dynamics in the neurodegenerative processes of PART and AD.

AUTHOR CONTRIBUTIONS

GU performed fluorescence immunohistochemical studies, as well as quantitative and statistical analyses, collected and analyzed the neuropathological and immunofluorescent data. GU and YH drafted the original manuscript. GU, YH, and RG contributed significantly to the drafting of the final manuscript. ND contributed to the table and figures. RRR provided neuropathological data samples. KJ and JW provided clinical information and expertise in data interpretation. KJ and JW also conceptualized, supervised, and funded the study. All authors read and approved the last version of the manuscript.

FUNDING INFORMATION

This work was supported by the National Institutes of Health (NIH) under grants: NIH: R01‐AG37491.

CONFLICT OF INTEREST STATEMENT

The authors declare no conflicts of interest.

Supporting information

Figure S1. Experimental paradigm and AIVIA 3D object recognition, reconstruction, and analysis. (A) Schematic diagram demonstrating the experimental paradigm workflow. (B) Immunofluorescent staining of the AD‐TDP‐LBD with pTDP43, AT8, and ThioS. Each detected 3D particle automatically pseudo colored to demonstrate their spatial relationship in x‐y‐z axes. White rectangle shows where the insert is taken. Scale bars = 500 μm. AI segmentation and recognition of 3D objects from confocal z‐stack in each staining are performed. (C) The representative single plane and maximally projected 3D Z‐stack confocal imaging results of AD‐TDP‐LBD with pTDP43, AT8, and ThioS were shown. Scale bars = 20 μm.

Figure S2. Related to Figures 1 and 2. Comparative neurofibrillary TAT assessment in amygdala. AI segmentation and recognition of 3D objects from confocal z‐stack images taken from pTDP‐43, AT8, PS396, and ThioS in cases with PART‐TDP, AD‐TDP, AD‐TDP‐LBD. The relation counts between (A) pTDP‐43 and AT8, (B) pTDP‐43 and PS396, (C) pTDP‐43 and ThioS in amygdala‐BL are shown. The relation counts in amygdala‐CM are assessed in D‐F. *p < 0.05, ns = nonsignificant, Kruskal–Wallis for grouped variables, Mann–Whitney U for pairwise comparisons.

Figure S3. Neurofibrillary TAT assessment in CA4 of Hippocampus. (A) Immunofluorescent staining of the controls, PART‐TDP, AD‐TDP, and AD‐TDP‐LBD pTDP‐43 (red), AT8 (magenta), PS396 (cyan), ThioS (green). Each row represents merged images of pTDP‐43 and AT8, PS396, and ThioS to demonstrate spatial relationship of pTDP‐43 aggregates in each condition. White rectangles show where the inserts are taken. Scale bars = 20 μm. AI segmentation and recognition of 3D objects from confocal z‐stack in each staining are performed. The relation counts between pTDP‐43 and AT8 are compared to pTDP‐43 and PS396 in (B) PART‐TDP, (C) AD‐TDP, and (D) AD‐TDP‐LBD cases. (E) Relative relation count (AT8 vs. PS396) in each group. *p < 0.05, **p < 0.01, ns = nonsignificant, Kruskal–Wallis for grouped variables, Mann–Whitney U for pairwise comparisons.

Figure S4. Related to Figures 5, 6, and S3. Comparative neurofibrillary TAT assessment in CA1, CA2/3, CA4 of Hippocampus. AI segmentation and recognition of 3D objects from confocal z‐stack images taken from pTDP‐43, AT8, PS396, and ThioS immunolabeling in cases with PART‐TDP, AD‐TDP, AD‐TDP‐LBD. The relation counts between (A) pTDP‐43 and AT8, (B) pTDP‐43 and PS396, (C) pTDP‐43 and ThioS in CA1 are shown. The relation counts in CA2/3 are assessed in (D–F) while CA4 is demonstrated in (G–I). *p < 0.05, ns = nonsignificant, Kruskal–Wallis for grouped variables, Mann–Whitney U for pairwise comparisons.

Figure S5. Neurofibrillary TAT assessment in DG of hippocampus. (A) Immunofluorescent staining of the controls, PART‐TDP, AD‐TDP, and AD‐TDP‐LBD with pTDP‐43 (red), AT8 (magenta), PS396 (cyan), ThioS (green). Each row represents merged images of pTDP‐43 and AT8, PS396 and ThioS to demonstrate spatial relationship of pTDP‐43 aggregates in each condition. White rectangles show where the inserts are taken. Scale bars = 20 μm. AI segmentation and recognition of 3D objects from confocal z‐stack in each staining are performed. The relation counts between pTDP‐43 and AT8 are compared to pTDP‐43 and PS396 in (B) PART‐TDP, (C) AD‐TDP, and (D) AD‐TDP‐LBD cases. (E) Relative relation count (AT8 vs. PS396) in each group. *p < 0.05, **p < 0.01, ns = nonsignificant, Kruskal–Wallis for grouped variables, Mann–Whitney U for pairwise comparisons.

Figure S6. Neurofibrillary TAT assessment in SUB of Hippocampus. (A) Immunofluorescent staining of the controls, PART‐TDP, AD‐TDP, and AD‐TDP‐LBD with pTDP‐43 (red), AT8 (magenta), PS396 (cyan), ThioS (green). Each row represents merged images of pTDP‐43 and AT8, PS396 and ThioS to demonstrate spatial relationship of pTDP‐43 aggregates in each condition. White rectangles show where the inserts are taken. Scale bars = 20 μm. AI segmentation and recognition of 3D objects from confocal z‐stack in each staining are performed. The relation counts between pTDP‐43 and AT8 are compared to pTDP‐43 and PS396 in (B) PART‐TDP, (C) AD‐TDP, and (D) AD‐TDP‐LBD cases. (E) Relative relation count (AT8 vs. PS396) in each group. *p < 0.05, ns = nonsignificant, Kruskal–Wallis for grouped variables, Mann–Whitney U for pairwise comparisons.

Figure S7. Related to Figures S5, S6, and 7. Comparative neurofibrillary TAT assessment in DG, SUB, and ERC. AI segmentation and recognition of 3D objects from confocal z‐stack images taken from pTDP‐43, AT8, PS396, and ThioS in cases with PART‐TDP, AD‐TDP, AD‐TDP‐LBD. The relation counts between (A) pTDP‐43 and AT8, (B) pTDP‐43 and PS396, (C) pTDP‐43 and ThioS in DG are shown. The relation counts in subiculum are assessed in (D–F) while Entorhinal Cortex is demonstrated in (G–I). *p < 0.05, ns = nonsignificant, Kruskal–Wallis for grouped variables, Mann–Whitney U for pairwise comparisons.

Figure S8. Comparative neurofibrillary TAT assessment. AT8 and PS396 labeled pre or mature TAT quantifications. pTDP‐43 and AT8 relation counts in amygdala‐BL, amygdala‐CM, CA1, CA2/3, CA4, DG, SUB, ERC of cases with (A) PART‐TDP, (B) AD‐TDP, and (C) AD‐TDP‐LBD. pTDP‐43 and PS396 relation counts in (D) PART‐TDP, (E) AD‐TDP, and (F) AD‐TDP‐LBD. *p < 0.05, **p < 0.01, Kruskal–Wallis for grouped variables, Mann–Whitney U for pairwise comparisons.

DATA AVAILABILITY STATEMENT

All aggregated data is available in the main text and figures or the Supporting Information. Primary data (e.g., images and spreadsheets) will be provided upon written request to the corresponding author. This paper does not report any original code. Any additional information can also be requested directly from the corresponding author.