Proteomic comparison of hippocampal neurofibrillary tangles in PART, intermediate Alzheimer’s disease and advanced Alzheimer’s disease

Abstract

Alzheimer’s disease (AD) is characterised by the intraneuronal aggregation of phosphorylated Tau (pTau) into neurofibrillary tangles and by the extracellular deposition of β-amyloid (Aβ). Tau pathology restricted to the hippocampal formation is frequently observed in the elderly brain in the absence of any Aβ deposition and considered as “primary age-related tauopathy” (PART). Here, we applied an unbiased proteomic approach to determine how concomitant Aβ pathology modifies the neurofibrillary tangle proteome. Neurofibrillary tangles were isolated by dissecting Tau pSer202/pThr205 “AT8” immunopositive neuronal profiles, combining chromogenic immunohistochemistry with laser capture microdissection, from hippocampal sections of 17 post-mortem brains spanning three groups: PART (n = 5; A0, B1–2, C0 scores), intermediate AD (n = 6; A1–2, B2–3, C1–2 scores) and advanced AD (n = 6; A3, B3, C3 scores). A label-free quantitative liquid chromatography–mass spectrometry based proteomic analysis, using data independent acquisition (DIA) on a Bruker timsTOF, was performed. A conserved core of 63 proteins was identified as enriched in tangles across all groups, mostly associated with “RNA binding” and “regulation of mRNA metabolic process”, based on the Gene Ontology database. Group-specific signatures were also observed: 33 proteins were significantly enriched only in tangles collected from PART cases and were predominantly linked to “structural molecule activity”, whereas Aβ-positive cases showed specific enrichment of “RNA binding” and “cytoplasmic translation” pathways—with intermediate AD cases displaying a transitional profile. Our findings are consistent with PART having distinct tangle proteomic features; however, the majority of its proteomic signature is in common with tangles within the AD continuum. By addressing how Aβ accumulation alters the tangle proteome, this study provides mechanistic insights into the expansion of Tau pathology, paving the way towards the identification of biomarkers and therapeutic strategies that would allow for stabilisation of Tau pathology in the elderly.

Supplementary Information

The online version contains supplementary material available at 10.1007/s00401-026-03064-9.

Introduction

Alzheimer’s disease (AD) is characterised by two types of neuropathological changes: the intraneuronal accumulation of phosphorylated Tau proteins (pTau) into neurofibrillary tangles [7, 25], in combination with the extracellular deposition of β-amyloid peptides (Aβ) into Aβ deposits of various morphology [15, 23]. These proteinaceous aggregates progress through the brain via seeding and spreading mechanisms in the manner of prions [11, 39, 51], following distinct stereotypical sequences. Tau pathology progression is described by the six Braak stages (0-VI) [3], correlated with AD symptom progression [43]: it gradually involves the transentorhinal/entorhinal cortices (Braak I–II), hippocampus (Braak III-IV) and neocortex (Braak V–VI). Aβ pathology progresses following the five Thal phases (0–5) [59]: it involves progressively the neocortex (Thal 1), hippocampus (Thal 2), basal ganglia (Thal 3), midbrain (Thal 4) and cerebellum (Thal 5). The presence of neurofibrillary tangles confined to the hippocampal formation is commonly observed in the elderly brain, in the absence of Aβ pathology and with only mild cognitive impairments, if any (Braak stage ≤ IV, Thal phase 0). These cases are referred to as “primary age-related tauopathy” (PART), although this concept remains debated [6, 13, 20, 44, 53].

Tau pathology progresses more slowly and is associated with distinct cognitive decline patterns when limited to the medial temporal lobe in the absence of any Aβ pathology, in comparison with cases presenting with Aβ deposition [12, 53, 54]. Although these observations justify a need to distinguish PART and AD from a clinical standpoint, neuropathological observations suggest that PART precedes AD along a biological continuum: (1) Tau pathology in the hippocampal formation is a consistent feature of AD pathology, (2) in the general population, the Gaussian-like distribution of post-mortem cases considered as PART across age groups illustrate a transient state, rather than a distinct age-related tauopathy and (3) Tau aggregate composition presents no morphological nor biochemical difference between PART- and AD-associated tangles, both exhibiting an even ratio of Tau 3R and Tau 4R isoforms with similar Tau phospho-epitope profiles [4, 5, 20, 21, 60]. Structural biology further supports this continuum: cryo-EM studies did not find any structural differences of Tau fibrils extracted from PART and AD cases, while establishing a highly-specific conformation-based classification of tauopathies over the past decade [56].

Recent observations suggest that, compared with AD, cases classified as PART are associated with a different topography of Tau pathology in the hippocampus (greater vulnerability of the CA2 field, rather than CA1), as well as with distinct genetic features (over-representation of the APOEε2 protective allele, instead of the APOEε4 risk allele) [53–55]. The proteomic comparison of hippocampal or parahippocampal regions collected from cases of various severity of Tau pathology highlights progressive protein changes occurring as early as Braak I–III—notably associated with cytoskeletal dynamics, synapse changes or translation dysregulation [28, 36]. However, whether these changes are directly related to Tau pathology, and differ in cases labelled as PART, remain elusive. Different proteomic signatures of tangle-bearing neurons were observed in cases labelled as PART or AD through the analysis of a pre-set panel of 70 proteins [52, 62]. Although a few unbiased proteomic approaches succeeded in describing the tangle proteome in the post-mortem human brain, none examined its changes across the PART-AD spectrum [19, 27]. More recently, an unique enrichment in proteins involved in cellular responses to reactive oxygen species was identified as associated with Tau oligomers extracted from PART, but not from advanced AD cases [30]. Altogether, these observations suggest that the proteomic composition of tangle-bearing neurons differs in the absence or presence of a concomitant Aβ pathology.

To investigate this hypothesis, the present study aims to take advantage of our unbiased and localised proteomic strategy, combining laser capture microdissection of neurofibrillary tangles with a label-free quantitative liquid chromatography-mass spectrometry based proteomic analysis [18, 19, 32, 50, 61]. Using this approach, we performed a comprehensive comparison of the tangle-bearing neuron proteome in cases presenting with varying severity of AD-related neurofibrillary changes, in the absence or presence of Aβ pathology.

Materials and methods

Cases

All procedures were performed under protocols approved by Institutional Review Boards at New York University Alzheimer’s Disease Research Center (NYU ADRC, NY, USA) and Massachusetts Alzheimer's Disease Research Center (MADRC). In all cases, written informed consent for research was obtained from the patient or legal guardian and the material used had appropriate ethical approval for use in this project. All patients’ data and samples were coded and handled according to NIH guidelines to protect patients’ identities. A total of 18 cases of various Braak stages and Thal phases were included in this study. Cases were selected from donated brain tissue collected at the NYU ADRC and MADRC, based on their ABC score [41]. Three groups were constituted as follows: group 1 (A0, B1–2, C0, considered as PART, n = 6 cases), group 2 (A1–2, B2–3, C1–2, considered as intermediate AD, n = 6 cases) and group 3 (A3, B3, C3, considered as advanced AD, n = 6 cases); none of these cases had significant TDP-43 pathology. Cases were selected to present enough tangles in the hippocampal formation for laser capture microdissection. PART cases were identified based on prior published criteria [26, 31]. Other inclusion criteria involved the absence of any additional primary tauopathy. The presence of a concomitant Lewy Body disease was tolerated for n = 3, n = 2 and n = 2 cases per group, respectively, to increase our number of cases, as this co-pathology is common in the elderly population and because its even distribution among groups did not impact our comparative study design. Individual case information is detailed in Table 1 based on the information we had available (sex, age, post-mortem interval, ABC score, other neuropathological findings).

Table 1.

Study cohort

Case	Source	Sex	Age	PMI	ABC score	Aβ pathology burden score (hippocampus)	Tau pathology burden score (hippocampus)	Other neuropathological findings
Group 1 PART
1	NYU ADRC	Female	90	< 48 h	A0, B2, C0	–	+	CAA, Lewy body disease, Binswanger’s disease, and lacunar infarcts (multiple, diffuse and remote)
2	NYU ADRC	Male	90	< 24 h	A0, B2, C0	–	+	None
3	NYU ADRC	Female	96	< 24 h	A0, B1, C0	–	+	Hippocampal sclerosis
4	MADRC	Female	≥ 90	N/A	A0, B1, C0	–	+	Lewy body disease and cerebrovascular disease
5	MADRC	Female	≥ 90	N/A	A0, B1–2, C0	–	+	Lewy body disease and cerebrovascular disease
6	MADRC	Female	≥ 90	N/A	A0, B2, C0	–	++	CAA, cerebrovascular disease, hypoxic/ischemic injury (CA1) and microinfarct (occipital white matter)
Group 2 Intermediate Alzheimer’s disease
7	NYU ADRC	Female	84	9 h	A1, B2, C1	+	+	Lewy body disease and Binswanger’s disease
8	NYU ADRC	Female	81	< 24 h	A1, B2, C2	+++	+	Lewy body disease
9	NYU ADRC	Female	86	< 24 h	A1, B2, C1	+	+	None
10	MADRC	Male	63	N/A	A2, B2, C2	+++	+++	CAA and cerebrovascular disease
11	MADRC	Female	87	N/A	A2, B2, C1	±	++	Cerebrovascular disease
12	MADRC	Female	≥ 90	N/A	A2, B3, C1	±	+++	CAA, cerebrovascular disease and acute hypoxia
Group 3 Advanced Alzheimer’s disease
13	NYU ADRC	Female	89	38 h	A3, B3, C3	+	+	CAA, Lewy body disease and Binswanger’s disease
14	NYU ADRC	Female	90	< 48 h	A3, B3, C3	+++	+	CAA, Lewy body disease, hippocampal sclerosis and Binswanger’s disease
15	NYU ADRC	Male	78	N/A	A3, B3, C3	+	+	None
16	NYU ADRC	Female	104	< 24 h	A3, B3, C3	+	+	CAA and Binswanger’s disease
17	MADRC	Female	≥ 90	N/A	A3, B3, C3	+++	+++	CAA and cerebrovascular disease
18	MADRC	Female	≥ 90	N/A	A3, B3, C3	+++	+++	Cerebrovascular disease

The 18 cases included in our study are listed in this table. The latter discloses their source, sex, age at death in years, post-mortem interval in hours, ABC score reflecting the distribution of Tau and Aβ pathologies based on the criteria from Montine et al. [41], as well as a summary of any other relevant neuropathological findings, based on the information we had available. A semi-quantitative score reflects Aβ pathology burden in the hippocampus (attributed from the average raw values measured in the hippocampus after anti-Aβ 4G8 immunohistochemistry) and Tau pathology burden in the hippocampus (attributed from the average raw values measured in the hippocampus after anti-pTau immunohistochemistry, using the means obtained from all three pTau epitopes: pSer202/pThr205, pThr217 and pThr231). CAA, cerebral amyloid angiopathy; MADRC, Massachusetts Alzheimer's disease research center; NYU ADRC, New York university Alzheimer’s disease research center; PMI, post-mortem interval

Immunohistochemistry

Immunohistochemistry was performed on formalin-fixed paraffin-embedded 8 µm-thick sections of hippocampus, as previously described [34, 61]. Sections were deparaffinised and rehydrated through a series of xylene and ethanol washes. Antigen retrieval was performed by treatment with 88% formic acid for 7 min, followed by boiling in citrate buffer (10 mM sodium citrate, 0.05% Tween-20, pH 6). Sections were blocked with 10% normal goat serum, then incubated overnight at 4 °C with a primary antibody: anti-Tau pSer202/pThr205 (1:500, AT8, Thermo Fisher Scientific, #MN1020), anti-Tau pThr217 (1:250, Thermo Fisher Scientific, #44-744), anti-Tau pThr231 (1:250, AT180, Thermo Fisher Scientific #MN1040) or anti-Aβ (1:1000; 4G8, BioLegend, #800711), diluted in 4% normal goat serum. Sections were incubated for 1 h at room temperature with a biotinylated secondary antibody (1:1000, Vector Laboratories, #BA-1000 or #BA-2000), revealed with an avidin–biotin complex HRP detection kit (Vector Laboratories, #PK-6100) in combination with a solution of 3,3′-diaminobenzidine (Thermo Scientific, #34065), counterstained with Mayer’s hematoxylin (Millipore Sigma, #MHS16) and coverslipped (Thermo Scientific, #P36970). This technique was applied to 18 cases (n = 6 PART, n = 6 intermediate AD, n = 6 advanced AD; Table 1).

Immunohistochemistry quantification

Tau pathology

Tau pathology was quantified in the hippocampus after anti-Tau pSer202/pThr205 “AT8”, anti-Tau pThr217 and anti-Tau pThr231 immunohistochemistry, as previously described with minor changes [61]. Slides were scanned at a 20 × magnification with the Aperio VERSA 8 scanner and analysed with Aperio ImageScope 12.4.3.5008 (Leica Biosystems). For each case and staining, Tau pathology burden was quantified from regions of interest (ROI) covering the various fields of the hippocampus: CA4, CA3, CA2, CA1 and subiculum. Hippocampal subfields were determined using established histo-anatomical criteria [49]. Briefly, subfields were identified according to anatomical landmarks within the hippocampal formation, as well as pyramidal neuron size, density and overall organisation. To limit the risk of error, transitional regions between subfields were excluded from the defined regions of interest. The total burden of pTau immunoreactive material was obtained by running the open source "Positive Pixel Count 2004-08-11" algorithm on each ROI, with the colour saturation threshold set at 0.20 (Tau pSer202/pThr205, Tau pThr217) or 0.25 (Tau pThr231). Raw data were exported to Excel to calculate the averaged percentage of immunopositive pixels out of the total number of pixels for each ROI. Two-way ANOVA followed by Tukey’s multiple comparison tests were performed on GraphPad Prism 9.5.1 to compare these ratios among hippocampal subfields and across groups at a risk level of α = 0.05.

Aβ pathology

Aβ pathology was quantified in the hippocampus after anti-Aβ 4G8 immunohistochemistry. Slides were scanned at a 20 × magnification with the Aperio VERSA 8 scanner. A pixel classifier was trained in QuPath-0.5.1 to recognise Aβ deposits and was used to generate deposit annotations within the same ROIs used for Tau pathology analysis: CA4, CA3, CA2, CA1 and subiculum. Aβ deposit size was restricted to > 200 µm² to avoid inclusion of artefactual patterns. Annotation measurements were exported and used to calculate the Aβ load within each region. Two-way ANOVA followed by Tukey’s multiple comparison tests was performed on GraphPad Prism 9.5.1 to compare these ratios among hippocampal subfields and across groups at a risk level of α = 0.05.

Laser capture microdissection (LCM)

LCM was performed using our published protocol with minor changes [18, 19]. Formalin-fixed paraffin embedded sections of hippocampus (8 µm thick) were collected onto LCM-compatible PET Frame Slides (Leica, #11505190). Tangles were visualised using chromogenic anti-Tau pSer202/pThr205 “AT8” immunohistochemistry. Briefly, sections were deparaffinised and rehydrated by a series of xylene and ethanol washes. All the subsequent steps were performed in phosphate-buffered saline (PBS), without any detergent to avoid the tissue section falling off the slides. Sections were treated with H₂O₂ (0.3% in PBS for 20 min at room temperature), then blocked (10% normal goat serum in PBS for 1 h at room temperature) and incubated with an anti-phosphorylated Tau pSer202/pThr205 antibody (1:500, AT8, Thermo Fisher Scientific, #MN1020; diluted in 4% normal goat serum; overnight at 4 °C). Sections were then incubated with an anti-mouse biotinylated IgG secondary antibody (Vector Laboratories, #BA-2000), a horseradish peroxidase kit solution (Vector Laboratories, #PK-6100; 1 h at room temperature) and finally with a solution of metal enhanced 3,3′-diaminobenzidine substrate (Thermo Fisher Scientific, #34065). Sections were counterstained with Mayer’s hematoxylin, then rinsed thoroughly with double distilled H₂O and air dried prior to LCM. AT8-positive neuronal profiles (“tangle” samples) were manually dissected from the hippocampus from each case (encompassing CA4, CA3, CA2, CA1 and subiculum), using a LMD6500 microscope at 20X magnification (Leica). Note that proteomic results, therefore, reflect proteins present in pre-tangles, tangles and ghost tangles. An ellipse of an equivalent average area was dissected in the adjacent neuropil for each tissue sample (“non-tangle” samples), to constitute an internal control of reference required to conduct pairwise analyses, while controlling for the inclusion of the tangle immediate microenvironment in our samples, as previously detailed by our group [35]. A total area of 1.5 mm² was collected, per sample and for each case, into PCR tubes containing high-performance liquid-chromatography (HPLC) grade H₂O. This total area of 1.5 mm² corresponded to an average of 726.8 manually dissected objects (tangles or non-tangles). As the LCM collection of 1.5 mm² total area was not feasible in one sitting due to water evaporation, three samples of 0.5 mm² total area were collected separately. After collection, samples were centrifuged at 14,000g for 2 min and stored at − 80 °C until peptide extraction. This procedure was applied on 18 cases (n = 6 PART, n = 6 intermediate AD, n = 6 advanced AD; Table 1).

Proteomics

Peptide extraction

The three replicates of 0.5 mm² dissected area collected separately were combined prior to peptide extraction according to our previously published method, with minor modifications [19]. Water was removed by vacuum centrifugation and proteins were extracted and digested using a SPEED workflow [16]. Proteins were extracted in 100% trifluoroacetic acid (TFA) at 73 °C (10 µL/sample). TFA was neutralised by a 1:10 (v/v) dilution with Tris base (2 M Tris, 20 mM 2-chloroacetamide (CAA) and 10 mM Tris(2-carboxyethyl)phosphine (TCEP)), in which samples were incubated for 1 h at 90 °C. Samples were diluted at 1:5 (v/v) with water containing 0.2 µg of sequencing grade trypsin, allowing for protein digestion overnight at 37 °C. The resulting peptides were desalted and concentrated on Evosep Pure C18 tips, then analysed by liquid chromatography coupled with tandem mass spectrometry (LC–MS/MS).

LC–MS/MS data acquisition

LC–MS/MS analyses were performed using an Evosep One liquid chromatography system (Evosep Biosystems, Odense, Denmark) coupled to a Bruker timsTOF HT mass spectrometer (Bruker Daltonics, Bremen, Germany). Peptide samples were loaded onto Evotips according to the manufacturer’s instructions. Peptides were separated using a predefined Evosep gradient (15SPD, 88 min LC gradient). Mobile phase A consisted of 0.1% formic acid in water, and mobile phase B consisted of 0.1% formic acid in ACN. Peptides were eluted directly into the mass spectrometer. MS data were acquired on the timsTOF HT instrument operated in positive ion mode using data-independent acquisition with parallel accumulation–serial fragmentation (diaPASEF). Peptide ions were accumulated and separated in the TIMS analyzer with an accumulation time of 100 ms and a ramp time of 100 ms. Each TIMS frame consisted of 936 scans, with an effective scan window of approximately 25.66–26.70 ms per frame. MS1 spectra were acquired over a mass-to-charge ratio range of 400–1200 Da. Fragment ion spectra were acquired using a diaPASEF acquisition scheme comprising multiple predefined isolation window groups distributed across the ion mobility (0.6–1.6 1/Ko) and mass (400–1200 Da) dimensions. Each window group contained two isolation windows per TIMS ramp, with isolation widths adapted to precursor density. Fragmentation was performed using window-specific, mobility-dependent collision energies ranging from approximately 20 eV for lower 1/Ko windows to 59 eV for higher 1/Ko values. Each diaPASEF acquisition cycle consisted of one MS1 TIMS frame followed by multiple DIA-PASEF MS/MS frames, resulting in a total cycle time of approximately 1.8 s. Raw data were stored in Bruker.d format.

Data analysis

The MS DIA data were searched with the Spectronaut software in directDIA (library-free) mode. The enzyme specificity was set to trypsin with the maximum number of missed cleavages set to 1. Oxidation of methionine was searched as variable modification; carbamidomethylation of cysteines was searched as a fixed modification. The false discovery rate (FDR) for peptide, protein, and site identification was set to 1% and two different peptides per protein were required for a protein to be considered identified. Quantification was performed on MS2 level. One outlier was removed from the analysis (case 6, PART group), due to an unusually low number of identified protein groups observed in both “tangle” and “non-tangle” samples, in comparison to the rest of the analysis (n_tangles_case#06 = 1043 and n_non-tangles_case#06 = 1292 versus n_average_tangles = 2755.3 ± 114.0 and n_average_non-tangles = 2754.4 ± 107.1). We interpreted this anomaly as the result of protein degradation within the brain tissue related to the peri-mortem or post-mortem conditions, although we could not identify the exact cause based on information we had available. The dataset was further filtered to remove manually all non-human proteins as well as proteins known as experimental contaminants, such as keratins. Label-free quantification (LFQ) protein intensity values were log₂-transformed. Protein intensities were then cross-run normalised to account for any systemic bias in overall peptide amounts injected on LC–MS/MS system. For this, intensity distributions within each sample were shifted to get the same median value across all the samples. Missing values were imputed in Perseus based on a normal distribution prior to the principal component analysis (PCA); a non-imputed dataset was preferred otherwise. Pairwise comparisons were performed on Perseus, within each group, between “tangle” and “non-tangle” samples (paired t test). Proteins with a fold change (FC) ≥ 1.5 and a p value (p) < 0.05 were considered as significantly enriched in tangles and further analysed, consistent with our previous studies [17, 33]. Additionally, pairwise comparisons were also performed on “non-tangle” samples across disease groups (unpaired t test). Proteins with a fold change (FC) ≥ 1.5 and a p value (p) < 0.05 were considered as differentially abundant in non-tangle samples and further analysed. Venn diagram was obtained from InteractiVenn. Functional enrichment analyses were conducted on STRING 12.0 and Cytoscape 3.10.2, using the Gene Ontology terms “molecular function” and “biological process”, with a high confidence interaction score set at 0.70 excluding text mining-based evidence, a focus on physical interactions and a redundancy cut-off of 0.25 (Online Resources 1–4).

Results

Neuropathology of the hippocampus

Immunohistochemistry was performed on hippocampal sections from all cases, using antibodies against Aβ (4G8), Tau pSer202/pThr205 (AT8), Tau pThr217 and Tau pThr231 (Fig. 1a–l). Overall, the quantitative analysis of the corresponding lesion burden across hippocampal subfields did not show any statistically significant group differences, likely due to high inter-individual variability. Aβ deposition involved only the hippocampus of the intermediate and advanced AD cases, confirming the absence of hippocampal Aβ pathology in our selection of PART cases (Fig. 1m). One advanced AD case was excluded from Tau pathology quantitative analysis as the observation of astrocytic accumulations of pTau—interpreted as ARTAG—confounded automated pixel detection (case 13). As expected, a progressive increase of pTau immunopositive material was observed across groups, following a similar distribution within the hippocampal subfields and for all three examined pTau epitopes: CA4 and CA3 were least affected by Tau pathology, while CA2, CA1 and the subiculum showed greater involvement. These results are consistent with an evolution of Tau pathology across groups, rather than a distinct pathological signature (Fig. 1n–p).

Fig. 1.

Comparative distribution of Aβ and Tau pathologies in the hippocampus of PART, intermediate AD and advanced AD cases. a–l Representative overviews of the neuroanatomical segmentation of the hippocampus after anti-Aβ (4G8; a, e, i), anti-Tau pSer202/pThr205 (AT8; b, f, j), anti-Tau pThr217 (c, g, k) or anti-Tau pThr231 (d, h, l) immunohistochemistry, conducted on cases from PART (a–d), intermediate AD (e–h) or advanced AD groups (i–l); scale bar: 5 mm. Bottom-right insert: higher magnification in the CA1 field showing a representative illustration of each associated neuropathological changes analysed, when present. Note the absence of Aβ deposition in the PART case. Scale bar: 50 µm. m–p Quantitative analysis of Aβ and Tau pathologies conducted on formalin-fixed paraffin-embedded hippocampal sections, in a total of n = 6 PART cases (A0, B1–2, C0 scores; blue), n = 6 intermediate AD cases (A1–2, B2–3, C1–2 scores; orange) and n = 5 advanced AD cases (A3, B3, C3 scores; red). The lesion burden associated with Aβ or Tau pathology was assessed in ROIs encompassing each hippocampal subfield, when identifiable (CA4, CA3, CA2, CA1 and subiculum). m Quantification of the percentage area occupied by Aβ extracellular deposits was performed on each ROI after anti-Aβ 4G8 immunohistochemistry with DAB revelation. n–p Tau pathology was evaluated by quantifying the percentage of immunopositive pixels for each ROI after n anti-Tau pSer202/pThr205, o anti-Tau pThr217 and p anti-Tau pThr231 immunohistochemistry with DAB revelation. Ordinary two-way ANOVA with Tukey’s multiple comparison tests, * for p < 0.05, *** for p < 0.001 and **** for p < 0.0001. AD, Alzheimer’s disease; p, p value; PART, primary age-related tauopathy; Sub., subiculum

Proteomics overview

Tangles and adjacent non-tangle regions were microdissected from FFPE hippocampal sections in three groups: group 1 (A0, B1–2, C0, considered as PART, n = 5 cases), group 2 (A1–2, B2–3, C1–2, considered as intermediate AD, n = 6 cases) and group 3 (A3, B3, C3, considered as advanced AD, n = 6 cases; Fig. 2a and Online Resource 5). A total of 2413 proteins were detected after removing non-human proteins and proteins known as experimental contaminants; 2310 of these proteins were detected in at least half of one group for one sample type (tangles or non-tangles). PCA was performed on imputed data (Online Resources 6a-6d), illustrating proteome differences among groups (PART, intermediate AD or advanced AD) and sample types (tangles or non-tangles). Advanced AD samples clustered separately from PART samples (p < 0.001) and from intermediate AD samples in PC1 (principal component 1; p < 0.05; one-way ANOVA with Tukey’s multiple comparison test; Online Resources 6a and 6b). Although there was no significant segregation by sample type, a trend could be observed in PC2 (principal component 2; tangles versus non-tangles, p > 0.05; paired t test; Online Resources 6a and 6c). Paired t tests identified 119, 111 and 138 proteins significantly enriched in tangle versus non-tangle samples from PART, intermediate AD and advanced AD cases, respectively (p < 0.05, FC ≥ 1.5). Tau was abundantly detected in all samples, but missed the threshold to be considered as significantly enriched in tangles, possibly due to the presence of physiological Tau in “non-tangle” samples [27]. A focus on Tau peptide-level data, however, confirmed the enrichment of pathological Tau peptides phosphorylated on Thr217 in tangle versus non-tangle samples (Online Resource 7) [10]. Several proteins previously identified as pTau interactors were found among proteins significantly enriched in tangles in all groups, validating our method (e.g. ADAR, MYEF2, PURA, SRSF1, TRA2B; Fig. 2b–e and Online Resource 1) [19, 61]. Additionally, unpaired t tests across non-tangle samples identified significant differences among PART, intermediate AD, and advanced AD groups—with the greatest number of differentially abundant proteins observed between PART and advanced AD. A total of 126 proteins were increased in PART when compared to both intermediate AD and advanced AD groups. These 126 proteins were associated with a significant network functional enrichment (PPI enrichment p < 1.0 × 10^–16), spanning diverse functions, such as nucleotide binding (46/126 proteins, #1 GO term “molecular function”, GO:0000166, FDR = 1.9 × 10^–10) and generation of precursor metabolites and energy (25/126 proteins, #1 GO term “biological process”, GO:0006091, FDR = 6.0 × 10^–13; see Online Resources 3, 4 and 8 for details).

Fig. 2.

Overview of the hippocampal neurofibrillary tangle proteome of PART, intermediate AD and advanced AD cases. a Experimental workflow, illustrated with BioRender.com. Tangles (AT8 positive neuronal profile) and non-tangles (neighboring neuropil region of similar averaged area, to control for tangle immediate environment) were laser capture microdissected from FFPE hippocampal sections (CA4, CA3, CA2, CA1 and subiculum). Samples were obtained from a total of 17 post-mortem cases, including n = 5 PART cases (A0, B1–2, C0 scores), n = 6 intermediate AD cases (A1–2, B2–3, C1–2 scores) and n = 6 severe AD cases (A3, B3, C3 scores). After collecting a total area of 1.5 mm² per sample type (tangles or non-tangles), samples were analysed by label-free quantitative mass spectrometry to identify protein differences. b–d Volcano plots showing all detected proteins based on their p value (− log₁₀ transformed) and fold change (log₂ transformed), obtained from a paired t test comparative analysis of “tangles” versus “non-tangles” datasets, for each group. Paired t tests identified 119, 111 and 138 proteins enriched in “tangle” versus “non-tangle” samples from the b PART, c intermediate AD or d advanced AD group, respectively (p < 0.05 and fold change ≥ 1.50; the top 25 proteins enriched are annotated on their respective plot). e Venn diagram showing the distribution of proteins found significantly enriched in tangles across all groups. AD, Alzheimer’s disease; FFPE, formalin-fixed, paraffin-embedded; PART, primary age-related tauopathy

Common core of tangle proteome detected across all groups

A large proportion of tangle-associated proteins were shared across all three groups of various severities of AD-related pathology (n = 63 proteins; Fig. 2e). In this network, 17/63 proteins were previously identified as pTau interactors by our group [19, 61], while 44/63 proteins were previously detected as dysregulated in the AD brain based on the NeuroPro online repository, compiling proteomic datasets obtained from post-mortem AD brain studies [2]. Only 2/63 proteins were identified as associated with the AD brain for the first time in this study: CCDC124 and SUGP2. A significant network functional enrichment was associated with these 63 proteins (PPI enrichment p < 1.0 × 10^–16). This network was predominantly composed of proteins involved in RNA binding (58/63 proteins, #1 GO term “molecular function”, GO:0003723, FDR = 4.1 × 10^–52) and associated with the regulation of mRNA metabolic process (14/63 proteins, #1 GO term “biological process”, GO:1903312, FDR = 1.8 × 10^–16). A detailed analysis of the top significant functional enrichments associated with this network, ranked by false discovery rate, confirmed a large enrichment of functions associated with RNA processes (Fig. 3 and Online Resource 2a). Correlation analyses of “tangles” versus “non-tangles” fold changes were performed for proteins significantly altered in at least one group (p < 0.05, Student paired t test of “tangles” versus “non-tangles” samples for each group): 88.7% of proteins had their fold change going in the same direction—enriched or decreased—between PART and intermediate AD (n = 633/714 proteins, r = 0.8205, p < 0.0001, Pearson correlation test, Fig. 5a); 91.8% between intermediate AD and advanced AD (n = 473/515 proteins, r = 0.8361, p < 0.0001, Pearson correlation test, Fig. 5b); and 84.3% between PART and advanced AD (n = 585/694, r = 0.7386, p < 0.0001, Pearson correlation test; Fig. 5c). These results demonstrate strong similarities between proteins detected across our three experimental groups, which share a large core of common proteins significantly enriched in tangle samples—irrespective of the severity of Tau and Aβ neuropathological changes.

Fig. 3.

Common core of neurofibrillary tangle-associated proteins observed at any severity of Tau and Aβ neuropathological changes. Network representation of the 63 proteins identified as significantly enriched in tangles in all groups of various severities of Tau and Aβ pathologies (p < 0.05, FC ≥ 1.5, paired t tests of “tangles” versus “non-tangles” datasets). Each protein is represented by its gene ID as a node. Physical interactions are represented by edges, with a high confidence interaction score set at 0.70 (STRING). The node colour reflects the protein status regarding previous proteomic studies conducted in the AD brain, as detailed in the legend [2, 19, 61]. The node outline reflects the main functional enrichments attributed to this network: proteins associated with the #1 GO term “molecular function” are shown in blue and those associated with the #1 GO term “biological process” in purple, as detailed in the legend. AD, Alzheimer’s disease; FC, fold change; p, p value; PART, primary age-related tauopathy

Fig. 5.

Divergent fold changes of proteins gradually decreased or enriched in the neurofibrillary tangle proteome across the Alzheimer’s pathology spectrum. a–c Linear regressions of the fold changes associated with proteins significantly altered in at least one the two groups from each pairwise comparison, as detailed in the legend (p < 0.05, paired t tests of “tangles” versus “non-tangles” datasets). Pearson correlation factors are indicated for each pairwise comparison, computed between a PART versus intermediate AD groups (r = 0.8205, p < 0.0001), b intermediate AD versus advanced AD groups (r = 0.8361, p < 0.0001) and c PART versus advanced AD groups (r = 0.7386, p < 0.0001). d–f Histograms plotting the individual fold changes calculated for a subset of proteins of interest (QuantityNorm_tangles/QuantityNorm_non-tangles; Online Resource 1), significantly increased in tangles of only d PART, e intermediate AD or f advanced AD cases. Only proteins presenting with a divergent fold change in at least one of the other two groups of comparison are shown. Two-way ANOVA with Dunnett’s multiple comparisons test, * for p < 0.05, ** for p < 0.01, *** for p < 0.001 and **** for p < 0.0001. AD, Alzheimer’s disease; PART, primary age-related tauopathy; Inter. AD, intermediate AD; Adv. AD, advanced AD; p, p value; r, Pearson correlation factor

Differences of the tangle proteome in Aβ negative or positive cases

Tangle-associated proteins in Aβ negative cases (PART)

Interestingly, 33 proteins were found to be significantly enriched uniquely in tangles collected from PART cases, presenting with Tau pathology, but no Aβ deposition (p < 0.05, FC ≥ 1.5; paired t test comparing “tangles” versus “non-tangles” samples in PART cases; Fig. 2e). Note that none of these 33 proteins were previously reported as a pTau interactor by our group [19, 61]. Although 21/33 proteins were previously found dysregulated in the AD brain, a total of 12/33 proteins were not known to be associated with AD based on the NeuroPro proteomic database [2]: AASS, CPA4, HOXD10, LDB3, MCM3, NID1, PACSIN3, PI3, PLBD1, PRSS3, SPRR1A and TINAGL1. A significant network functional enrichment was associated with these 33 proteins, despite the presence of many singletons (28/33 proteins found isolated in the network, PPI enrichment p = 8.3 × 10^–3). Only one functional enrichment term was attributed to this protein network, predominantly composed of proteins involved in structural molecule activity (10/33, #1 GO term “molecular function”, GO:0005198, FDR = 1.9 × 10^–3; Fig. 4a and Online Resource 2b). An examination of these 10 proteins showed an association with structural maintenance encompassing various scales: cell–extracellular matrix interactions and tissue organisation (LAMB2, NID1), structural integrity of the ribosome (RPL27A, RPL3, RPS15A, RPS3A, and RPS14), as well as regulation of proteolytic processes (PI3, TINAGL1).

Fig. 4.

Different neurofibrillary tangle-associated proteins associated with the absence or presence of Aβ pathology. a–c Protein networks identified as specifically enriched in tangles collected from Aβ negative cases (a for PART, n = 33 proteins) or Aβ positive cases (b for intermediate AD, n = 34 proteins and c for advanced AD, n = 66 proteins; p < 0.05, FC ≥ 1.5, paired t tests of “tangles” versus “non-tangles” datasets). Each protein is represented by its gene ID as a node, whose size reflects its relative abundance in “tangles” versus “non-tangles” samples (log₂(fold change tangles/non-tangles)). Edges illustrate physical interactions with a high confidence interaction score set at 0.70 (STRING). The node colour highlights the protein status regarding previous proteomic studies performed in the AD brain, as shown in the legend [2, 19, 61]. The node outline reflects the main functional enrichments attributed to each network: proteins associated with the #1 GO term “molecular function” are shown in blue and the ones associated with the #1 GO term “biological process” in purple, as detailed in the legend. AD, Alzheimer’s disease; FC, fold change; p, p value; PART, primary age-related tauopathy

We compared the level of detection of these proteins associated with the tangle proteome of PART cases across groups, to highlight any differences that could be attributed to increased levels of Aβ deposition. We were interested to see that some of these 33 proteins significantly enriched in tangles collected from PART cases tended to be decreased in tangles obtained from intermediate and advanced AD cases, in comparison with “non-tangles” samples: ATP5ME, MAP1LC3B2, NID1 and PRSS3 tended to be decreased in “tangles” versus “non-tangles” samples collected in the advanced AD group only, while GRN, NQO1 and TINAGL1 tended to be decreased in “tangles” versus “non-tangles” samples collected in both intermediate and advanced AD groups (FC < 1; Fig. 5a, c, d). Interestingly, BANF1, FMC1, HOXD10, PI3, SNRPE and SPRR1A could not be detected in enough samples from the advanced AD group to compute a fold change from a paired t test comparison (PI3 being additionally absent from all, but one tangle sample from the intermediate AD group). These results suggest a distinct proteomic signature of tangles in PART cases, associated with proteins downregulated as the severity of Aβ pathology increases.

Tangle-associated proteins in Aβ positive cases

In contrast, 34 and 66 proteins were found significantly enriched only in tangles collected from cases with Tau pathology associated with Aβ pathology (intermediate AD and advanced AD groups, respectively; p < 0.05, FC ≥ 1.5; paired t tests comparing “tangles” versus “non-tangles” samples). A total of 12 proteins was shared between those two groups (CELF2, DDX39B, EIF3A, H2AC20, HP1BP3, LMAN1, RALY, RPS17, RPS20, RPSA, SRRM2 and TMED10). Note that all these 12 shared proteins tended to be also increased in tangles collected from PART cases, but missed the thresholds of significance (Fig. 2e).

Intermediate AD group A total of 5 out of the 34 proteins enriched in tangles from intermediate AD, but not PART, were previously reported as pTau interactors by our group [19, 61]. Interestingly, 10/34 proteins were not known to be associated with the AD brain based on the NeuroPro proteomic database [2]: CPSF7, KTN1, NOMO3, NTPCR, RPS17, RPS24, RPSA, S100P, SART3 and THRAP3. A significant network functional enrichment was associated with these 34 proteins (PPI enrichment p = 2.1 × 10^–5); these proteins were mostly involved in RNA binding (22/34 proteins, #1 GO term “molecular function”, GO:0003723, FDR = 3.0 × 10^–12), as well as cytoplasmic translation to a lesser extent (5/34 proteins, #1 GO term “biological process”, GO:0002181, FDR = 1.7 × 10^–3; Fig. 4b and Online Resource 2c). Among the 22 proteins significantly enriched only in tangles collected from intermediate AD cases, we observed some candidates illustrating a gradient across our study groups: indeed, 5 of these 22 proteins showed a tendency to be decreased in tangles collected from PART cases (FC < 1), but tended to be increased in tangles obtained from advanced AD cases (FC > 1; PLTP, EEF1A2, NT5E, NTPCR and RPS24). Interestingly, most of these 22 proteins associated with the tangle proteome from intermediate AD cases tended to be also increased in tangles of the advanced AD group (FC > 1 for all proteins except CPSF7, undetected). These observations support a stronger similarity of the tangle proteome identified in intermediate AD cases with the tangle proteome observed in advanced AD, compared to PART (Fig. 5a, b, e).

Advanced AD group. A total of 6 out of the 66 proteins enriched in tangles from intermediate AD, but not PART, were previously identified as pTau interactors in our team [19, 61]. Only 10/66 proteins were reported for the first time as associated with AD, according to the NeuroPro proteomic database [2]: CTSG, DDOST, EVPL, LACRT, PRPS1, RPL26, RPL4, RPS17, RPSA and SERPINB5 (RPS17 and RPSA being common to the intermediate AD group). A significant network functional enrichment was attributed to these 66 proteins (PPI enrichment p < 1.0 × 10^–16). As for the intermediate AD group, this network was composed of some proteins involved in RNA binding (5/66 proteins, #1 GO term “molecular function”, GO:0019843, FDR = 2.1 × 10^–3), but was mostly associated with cytoplasmic translation (24/66 proteins, #1 GO term “biological process”, GO:0002181, FDR = 5.6 × 10^–31; Fig. 4c and Online Resource 2d). Some of the 54 proteins significantly associated only with the tangle proteome of the advanced AD group appeared decreased in the PART and intermediate AD groups, consistent with a gradient associated with tangle maturation: 9 of these 54 proteins tended to be decreased in tangles dissected from PART cases (FC < 1; COPB2, CTSG, GSTM3, LACRT, PANK4, PRPS1, PSMD11, PSMD14 and SERPINA3), while 5 of these 54 proteins showed a tendency to be decreased in tangles obtained from intermediate AD cases (FC < 1; CAST, CTSG, EIF6, EVPL and IGHG2; Fig. 5b, c, f). Note that GSTM3 and LACRT were not detected in enough samples of the intermediate AD group to compute a fold change from a paired t test comparison.

Discussion

We provided a comprehensive comparison of the tangle-bearing neuron proteome across PART, intermediate AD and advanced AD. A core set of 63 proteins was consistently enriched in tangles, regardless of the severity of Tau and Aβ pathologies. Although some proteins may reflect general neuronal content due to our experimental design limitations (i.e. “tangle” versus “neuropil” protein enrichment), comparative analyses with previous proteomic datasets identified 61 of them as dysregulated or associated with the AD brain, including 17 pTau interactors, thereby underlining their pathological relevance [2, 19, 61]. Interestingly, some differences were observed; although they could reflect some dynamic and potentially transient protein pool associated with tangles, some reflected gradual changes consistent with pathological-driven changes. Importantly, some proteins were solely associated with PART tangles, revealing a distinct molecular signature. Our study is consistent with tangles in PART and AD being highly related, despite having important distinctions. These findings support a scenario in which Aβ accumulation is permissive to tangle spread beyond the hippocampal formation, marked by an increasing proteostasis stress as neuronal resilience declines.

The classification of PART as a discrete entity or as part of the AD continuum remains debated [6, 44]. Whether PART represents a distinct age-related tauopathy or an early, relatively benign and stable stage of AD, it unquestionably offers a unique window into potential mechanisms of resilience against Tau propagation, beyond a restricted brain distribution. Clinically, PART is common in the elderly and is associated with both a limited progression of Tau pathology and minimal cognitive decline, compared with cases exhibiting Aβ deposition [12, 26, 54]. Biologically, our proteomic approach highlights shared tangle features [6, 20, 56], but also important distinctions between PART and AD groups. We identified 33 proteins specifically associated with PART tangles, mostly involved in protein structure regulation. About one-third of these proteins were reduced or undetected in intermediate AD and advanced AD groups, with functions compatible with a progressive loss of defence mechanisms against Tau pathology expansion, such as: autophagosomal pathway (MAP1LC3B2/LC3) [38], oxidative stress response (NQO1) [9] or mitochondrial ATP production (ATP5ME) [1]. GRN particularly stood out as a protein of interest, as the deficiency of this lysosomal regulator is well described in the context of neurodegenerative proteinopathies [48]. A recent in vivo study suggests that the relative enrichment of GRN in tangles collected from PART, along with its decline observed as AD pathology progresses, could reflect a resilient lysosomal pathway limiting early Tau changes and spread [57]. Altogether, these functions coincide with the ones altered in PART versus AD cases in a recent proteomic study, despite a limited overlap of detected proteins due to different technical approaches (tangle proteome versus Tau interactome) [30].

The exact effect of Aβ pathology onset on Tau pathology could not be fully addressed with our study design, as the distribution of Tau pathology in PART cases (A0, B1–2, C0) and intermediate AD (A1–2, B2–3, C1–2) was not perfectly matched due to the limited availability of cases fitting our inclusion criteria. By demonstrating both commonalities and distinctions between tangles from PART and those from intermediate and advanced AD, our results support a reorganisation of the tangle proteome as Aβ pathology progresses. Our neuropathological evaluation did not show a clear topographical change of Tau pathology within the hippocampus across PART, intermediate and advanced AD—contrasting with recent reports of a selective CA2 vulnerability in PART, although this observation was based on a larger cohort [55, 63]. In intermediate and advanced AD, tangle-associated proteins were particularly enriched in RNA-binding and translational functions, suggestive of an over-representation of the translational stress response as AD pathology progresses [14]. Among the proteins enriched exclusively in advanced AD tangles, several were reduced or undetected in PART and intermediate AD groups and associated with an exacerbated Tau pathology. The enzyme GSTM3, involved in the oxidative stress response and colocalised within neurofibrillary tangles, was of particular interest as its polymorphism is a known AD risk factor [29, 58]. Two 26S proteasome subunits, PSMD14 and PSMD11, also stood out; their capacity to process ubiquitinated proteins reportedly declines as they physically interact with pathological Tau in vivo [42]. SERPINA3/ACT, which is overexpressed in the AD brain, promotes Tau hyperphosphorylation in neurons cultured in vitro [47]. Notably, CTSG—whose detection levels across groups mirror those of its inhibitor SERPINA3/ACT—belongs to the cathepsin family, a group of enzymes responsible for the GRN cleavage associated with Tau accumulation [37, 40, 57]. The latter observations converge to suggest an altered lysosomal function in tangles, comparing PART to advanced AD.

Limitations

Several limitations should be considered when interpreting these findings. The cohort size is limited (n = 5–6 per group), due to a stringent case selection to control for inter-individual variability (e.g. tau burden, co-pathology, sex ratio and brain bank source). While this improves internal consistency, it reduces statistical power, hence the need for future complementary studies and orthogonal validation strategies—particularly as a transcriptomic study recently emphasized the complexity of tangle studies, by highlighting tangle heterogeneity within one sample, at the single-cell level [45]. In addition, cases were selected to present enough tangles in the hippocampus for LCM, as this approach necessitates a minimal amount of dissected objects for downstream proteomic analyses (on average, 726.8 objects per sample). This may introduce a selection bias by narrowing pathological variability across groups, although this bias could underestimate, rather than exaggerate, differences between groups. Methodologically, the use of non-tangle regions as internal controls—instead of tangle-free neurons, which were too rare in the hippocampus of advanced AD cases to be collected—may explain why we did not observe any enrichment of Tau proteins in tangle samples versus non-tangle samples. Our peptide extraction conditions may underestimate insoluble Tau detection in tangle samples and preferentially digest more soluble Tau species overall, which are expected to be present both inside and outside tangles, as the neuropil is made of neuronal processes containing physiological Tau proteins. A focus on Tau peptide-level data, however, showed an enrichment of Tau peptides phosphorylated on Thr217 in tangle versus non-tangle samples, despite a very limited amount of LCM material available—without a specific method, phosphorylated peptides are generally present at relatively low abundance and are, therefore, less likely to be consistently identified and quantified. Our proteomic workflow might, therefore, better reflect the molecular changes associated with the soluble Tau fraction present in tangle-bearing neurons, than those associated with the more insoluble Tau fibrils—although considered to be less biologically active and toxic [24, 46].

Perspectives

Overall, our results are consistent with a remodelling of the tangle proteome across groups of varying severity of Tau and Aβ pathologies. These changes could lead to a proteostatic collapse that disrupts Tau pathology microenvironment, allowing further Tau propagation. The resulting loss of neuronal resilience, promoted by Aβ deposition, may underlie the transition from clinically benign PART to symptomatic AD. Future validation studies are needed to better understand the mechanisms underlying the proteostasis disruption. By investigating the molecular changes associated with Tau pathology in the absence or presence of Aβ pathology, this novel human-based and unbiased proteomic dataset will drive biomarker discovery to discriminate PART cases while supporting innovative therapeutic strategies designed to stabilise Tau pathology in the elderly and prevent the associated cognitive decline, along with anti-Aβ immunotherapies [8, 22].

Supplementary Information

Below is the link to the electronic supplementary material.

Author contributions

MT performed studies and help write the manuscript; TK performed studies; KB performed studies; LT performed studies; DL performed studies; EK performed studies; CW performed studies; DO provided tissue and information used in studies; BH provided tissue and information used in studies; BU performed studies and helped write the paper; ED performed studies and helped write the paper; TW performed studies, helped write the paper and conceived the study.

Data availability

The mass spectrometric raw files are accessible at https://massive.ucsd.edu under accession MassIVE MSV000100478.