Abstract
Background
The rapidly progressive phenotype of Alzheimer’s disease (rpAD) remains
a rare and less-studied entity. Therefore, the replication of key
results from the rpAD brain and cerebrospinal fluid (CSF) is lacking.
Methods
A label-free quantitative LC-MS/MS analysis of proteins co-aggregating
with core-amyloid
[MATH: β :MATH]
plaques in fresh frozen tissue (FFT) from medial temporal regions of
rpAD (
[MATH: n=8 :MATH]
) neuropathologically characterized at the National Prion Disease
Pathology Surveillance Center (NPDPSC), compared with microdissected
amyloid plaques from formalin-fixed, paraffin-embedded (FFPE) tissue
blocks from patients with rpAD (
[MATH: n=22 :MATH]
) previously published from the NPDPSC cohort, was performed. Matched
rpAD CSF from the FFT cases were compared to a previously published
proteomic evaluation of CSF in the AD subtype with rapid progression.
Results
A total of 1841 proteins were characterized in the FFT study, of which
463 were consistently identified in every rpAD patient analyzed. One
thousand two hundred eighty-three proteins were shared between the FFT
and the prior FFPE study. FFT offered a more comprehensive proteomic
profile than the prior FFPE study and prominently included the immune
system pathways. Thirty-five proteins were shared in the FFT brain
tissue, matched CSF from the same subjects, in which biological
processes related to immune response were again notable. These results
were validated against prior published proteomic CSF AD data with a
faster rate of progression to identify the top 5 potential protein
biomarkers of rapid progression in AD CSF.
Conclusions
These results support a distinct immune-related proteomic profile in
both the brain and the CSF that can be explored as potential biomarkers
in the future for the clinical diagnosis of rpAD.
Supplementary Information
The online version contains supplementary material available at
10.1186/s13195-025-01767-x.
Keywords: Rapidly progressive Alzheimer’s disease, Proteomics,
Cerebrospinal fluid, Fresh frozen tissue, Formalin-fixed,
Paraffin-embedded tissue, Dense core amyloid plaque, Neuroinflammation,
Alzheimer’s disease, Dementia, Rapidly progressive dementia,
Proteomics, Dense core amyloid plaque, Autopsy, CSF
Introduction
Alzheimer’s Disease (AD) is among the defining public health concerns
of the 21 st century and is now a leading cause of disability worldwide
[[34]1, [35]2]. Although AD is classically described by
neuropathological changes in amyloid
[MATH: β :MATH]
plaques and tau neurofibrillary tangles in the brain [[36]3],
(Alzheimer’s Disease Neuropathologic Change, ADNC), there is
substantial clinical heterogeneity among patients with AD [[37]4,
[38]5]. The trajectory of clinical decline among AD patients can vary
widely from less than 1 year to over 10 years [[39]6]. This
heterogeneity, particularly in the rate of clinical decline, entails a
huge challenge in evaluating patients and developing a new generation
of therapies [[40]7–[41]9]. Factors such as age, sex, genetics,
socioeconomic status, medical comorbidity, and cognitive reserve are
thought to affect the rate of AD clinical progression [[42]10].
A rapidly progressive subtype of AD (rpAD), among autopsy-confirmed
cases with a survival duration of less than 3 years has been reported
by us in two US national cohorts, the National Prion Disease Pathology
Surveillance Center (NPDPSC) and the National Alzheimer’s Coordinating
Center (NACC), despite different recruitment biases [[43]9]. In prion
disease surveillance centers worldwide, AD is the most frequent
diagnosis of non-prion disease at autopsy, accounting for 14 to
[MATH: 50% :MATH]
of all non-prion cases, since rpAD is associated with a diversity of
neurological signs and can mimic Creutzfeldt-Jakob disease [[44]11,
[45]12]. Here, the rpAD phenotype has been well established as detailed
clinical data supporting a rapid disease course is available [[46]6,
[47]11, [48]13]. Several studies have noted differences in the allelic
frequency of APOE
[MATH: ϵ4 :MATH]
, structural organization of the A
[MATH: β :MATH]
species, tau seeding, and amyloid plaque composition between rpAD and
slower progressing sporadic AD (spAD) [[49]6, [50]11–[51]17]. Together
they support the view that rpAD represents an AD subtype with distinct
biological characteristics.
Despite increasing interest in rpAD, most studies have included a
modest number of rpAD cases, typically fewer than 30. This is likely
due to its rarity and the challenges of accurate characterization via
disease biomarkers in rapid progression. When establishing clinical
biomarker evaluation or therapeutic targets among cases of rpAD, these
studies are consequently limited by the relatively small number of
well-characterized rpAD cases. Furthermore, ensuring consistent results
across different cohorts and within the same cohort is crucial,
especially when varied sample preparation methods are likely to
influence the outcomes. This is also true in the context of proteomic
evaluation of amyloid core plaques, which represent one of the central
neuropathological hallmarks of AD among rpAD cases.
Our study was driven by the overarching goal of advancing our
comprehension of the biological signature in rpAD by seeing how
replicable proteomic results across different analysis techniques are
and if there is a shared proteomic signature between the brain and
cerebrospinal fluid (CSF) of interest to develop as potential
biomarkers. Our primary objective was the examination of proteins
co-aggregating with dense core-amyloid plaques in fresh frozen tissue
(FFT) samples from the medial temporal lobe in both phenotypes using
mass spectrometry techniques. We hypothesized that a similar range of
amyloid core plaque-related proteins from rpAD and spAD would be seen
in both FFT tissue and the previously published formalin-fixed
paraffin-embedded (FFPE) preparation techniques [[52]15]. Furthermore,
we explored whether there were overlapping proteins between the rpAD
amyloid core plaques and CSF from these same cases.
Methods
Ethics statement
All procedures were performed under protocols approved by the
Institutional Review Board at Case Western Reserve University. All
patients’ data and samples were coded and handled according to NIH
guidelines to protect patients’ identities.
National prion disease pathology surveillance center rpAD cohort
The rpAD cohort from the NPDPSC has been previously described [[53]6,
[54]13]. In brief, following the standard NPDPSC protocol, case records
were retrospectively analyzed by trained personnel for the time of
onset of symptoms noted by the physician and/or obtained from
caregivers via standardized consent forms. As there are no consistent
clinical criteria for rpAD, patients in the rpAD cohort met the
following inclusion criteria: (i) suspected cases of prion disease
referred to the NPDPSC; (ii) death within 3 years of initial symptoms;
(iii) exclusion of prion disease via histology, immunohistochemistry,
and western blot analyses; (iv) neuropathology and immunohistochemistry
of tau proteins and amyloid
[MATH: β :MATH]
with unequivocal classification as Alzheimer’s disease according to the
published guidelines [[55]18]; (v) absence of neuropathologic
comorbidity contributing to phenotype; (vi) absence of medical
comorbidities that could affect disease progression (e.g., renal
failure). Eight rpAD cases were included in the analysis, demographic
details are provided in Table [56]1. These rpAD patient brains using
FFT had the same sex distribution and range of disease duration as
prior published reports using FFPE [[57]15].
Table 1.
Subject demographics
Diagnosis Sex Age at symptom onset (years) Age at death (years)
Estimated symptom duration (years)
rpAD Male 84 85 1
rpAD Female 81 82 2
rpAD Male 88 89 1
rpAD Male 75 76 1
rpAD Male 59 60 1
rpAD Female 59 62 3
rpAD Female 80 80
[MATH: <1 :MATH]
rpAD Female 73 74 1
spAD Male NA 67
[MATH: >3 :MATH]
spAD Female NA 83
[MATH: >3 :MATH]
spAD Female NA 78
[MATH: >3 :MATH]
[58]Open in a new tab
Slower progressing sporadic AD (spAD) cohort
AD patients with predominant amnestic symptoms from the Case Western
Reserve University (CWRU) Memory and Aging Center brain bank were used
to constitute the spAD cohort. They met AD neuropathological evaluation
criteria according to published guidelines [[59]18]. Due to unavailable
symptom onset data, 3 spAD cases with amnestic symptoms were randomly
selected for estimated symptom duration from clinical records that
noted clinical history over 3 years. This smaller sample size was used
for exploratory analysis based on the variance between brains observed
in a prior published report [[60]15].
Individual-level data on Braak stage and Thal phase were not available
for all subjects for detailed analysis. Inclusion of these data will be
essential for future studies to enable more precise stratification and
interpretation.
Demographic details are provided in Table [61]1.
Materials
Protease inhibitor cocktail and High Select^™ Top14 Abundant Protein
Depletion Mini Spin columns were purchased from Thermo Fisher
Scientific (Waltham, MA). Lys-C was obtained from FUJIFILM Wako
Chemicals U.S.A. (Richmond, VA). All other chemicals were either
reagent grade or of the highest commercially available quality.
Amyloid core-plaque isolation and preparation
Compared with the previous study using laser-capture microdissection
(LCM) to isolate plaques or regions within plaques [[62]15], this study
focused on studying the properties and molecular composition at the
center of the plaques. Amyloid plaque cores were isolated as previously
described [[63]19]. Briefly, human brain gray matter were homogenized
in lysis buffer (2% SDS, 50 mM Tris-HCl pH 7.5, 50 mM DTT, 1x protease
inhibitor cocktail) using a Dounce glass homogenizer, boiled for 10
min, and centrifuged at 100 000
[MATH: × :MATH]
g for 1 h at
[MATH:
10∘ :MATH]
C. Then, the pellet was solubilized in fraction buffer (1% SDS, 50 mM
Tris-HCl pH 7.5, 50 mM DTT) and centrifuged at 100 000
[MATH: × :MATH]
g for 1 h at
[MATH: 10∘ :MATH]
C. The resulting pellet was suspended in a fraction buffer and loaded
on top of a discontinuous sucrose gradient (1.0, 1.2, 1.4, and 2.0M
sucrose in 50 mM Tris pH 7.5 containing 1% SDS), centrifuged at 220 000
[MATH: × :MATH]
g for 20 h at
[MATH:
10∘ :MATH]
C, and the interface region between 1.4 and 2.0M sucrose was collected.
The plaque-core-enriched fraction was resuspended in water and
centrifuged at 100 000
[MATH: × :MATH]
g for 1 h at
[MATH:
4∘ :MATH]
C. The resulting pellet was dissolved in urea buffer (7 M urea, 2 M
thiourea, 4% CHAPS, 30 mM Tris, 5 mM magnesium acetate, pH 8.5, 1
[MATH: × :MATH]
protease inhibitor cocktail, 1
[MATH: × :MATH]
phosphatase inhibitor cocktail), centrifuged at 18 000
[MATH: × :MATH]
g for 30 min, and the supernatants were collected. The protein
concentration was estimated with a Bradford assay.
Cerebrospinal fluid (CSF) samples
CSF from patients was obtained before death during their clinical
diagnostic workup to rule out the existence of prion disease in rpAD
cases.
Per the NPDPSC protocol, 2 mL of CSF (1 mL minimum), avoiding a bloody
tap, was recommended to be collected in a polypropylene, low-binding,
sterile collection tube. The samples were immediately frozen, at least
in a
[MATH:
-20∘ :MATH]
C freezer, and shipped to the NPDPSC on dry ice. Samples were stored in
a
[MATH:
-80∘ :MATH]
C freezer at NPDPSC after the initial analysis for prion disease, which
was ruled out using RT-QuIC for prion proteins. AD diagnosis was
confirmed at autopsy.
Proteomic analysis
The isolated amyloid plaque 10
[MATH: μ :MATH]
g was subjected to SDS-PAGE approximately
[MATH: ∼ :MATH]
1 cm into a 4–20% Mini-PROTEAN^® TGX^™ precast protein gel (Bio-Rad,
Hercules, CA). The top 1 cm of the gel was then excised and in-gel
digested by Lys-C [[64]20]. A 50
[MATH: μ :MATH]
L sample of CSF was mixed directly with 1
[MATH: μ :MATH]
L of 10
[MATH: × :MATH]
protease inhibitor cocktail, and the most abundant 14 proteins were
removed using a High Select^™ Top14 Abundant Protein Depletion Mini
Spin column according to the manufacturer’s instructions. The resulting
CSF was concentrated to approximately 100
[MATH: μ :MATH]
L using a 3 kDa molecular weight cut-off Amicon centrifuge filter unit
(Millipore Sigma, Burlington, MA) and dried in a SpeedVac concentrator.
The dried CSF was then subjected to SDS-PAGE and in-gel digested with
Lys-C as described above. Following digestion, LC-MS/MS was performed
using the Fusion Lumos^™ Orbitrap Mass Spectrometer (Thermo Fisher
Scientific). HPLC was carried out using a Dionex 15 cm
[MATH: × :MATH]
75
[MATH: μ :MATH]
m id Acclaim Pepmap C18, 2
[MATH: μ :MATH]
m, 100Å reversed-phase capillary chromatography column. Peptides eluted
from the column in an acetonitrile/0.1% formic acid gradient (flow rate
= 0.3
[MATH: μ :MATH]
L) were introduced into the microelectrospray ion source of the mass
spectrometer, which was operated at 2.5 kV. Samples were analyzed using
a data-dependent method with CID fragmentation. Proteins were
identified by comparing all experimental peptide MS/MS spectra against
the UniProt human database using the Andromeda search engine integrated
into the MaxQuant version 1.6.3.3 [[65]21, [66]22].
Carbamidomethylation of cysteine was set as a fixed modification,
whereas variable modifications included oxidation of methionine to
methionine sulfoxide and acetylation of N-terminal amino groups. For
peptide/protein identification, strict Lys-C specificity was applied,
the minimum peptide length was set to 7, the maximum missed cleavage
was set to 2, and the cutoff false discovery rate was set to 0.01.
Match between runs (match time window: 0.7 min; alignment time window:
20 min) and label-free quantitation (LFQ) options were enabled. The LFQ
minimum ratio count was set to 2. The remaining parameters were kept as
default. Protein quantitation was accomplished using Perseus [[67]23].
LFQ values were log2-transformed, and missing values were imputed using
the “Replace missing value from normal distribution” function on the
entire matrix using default parameters (Please refer to Additional
Files 2 and 3 for the data generated from amyloid plaque cores and CSF,
respectively).
Bioinformatics analysis
A log transformation was initially applied to the protein-level
identifications to reduce systematic variations in recorded
intensities. Additionally, in line with our assumption of
homoscedasticity within this multi-group design, we employed a joint
adaptive mean-variance regularization procedure using the R package MVR
[[68]24]. This approach not only tackles the challenge posed by the
[MATH: n≫m :MATH]
problem, where the number of observations (n) significantly exceeds the
number of variables (m), potentially resulting in statistical
complexities and overfitting but also addresses the common issue of
variance-mean dependence often encountered in expansive omics datasets.
Public data access
FFPE [[69]15] and CSF [[70]25] proteomics data were downloaded from the
corresponding articles’ supplementary data. In this study, we utilized
protein identifications from these studies as the authors reported
them. The signaling pathway database was downloaded from Reactome
Database [[71]26].
Differential expression analysis
Pairwise differentially expressed proteins were identified using ROTS
[[72]27], followed by Benjamini-Hochberg (BH) FDR correction.
Differential expression was presented in volcano plots and heatmaps,
generated with the ggplot2 package in R.
GO and pathway enrichment analyses
To characterize proteins based on the GO annotation, we employed the
clusterProfiler R package. The pruned output of Fisher’s exact test was
utilized for over-representation analysis, and the resulting z-scores
were visualized through a custom R script. The background proteome
comprised all proteins within each specific dataset. As previously
outlined, enrichment analysis for each module was conducted by
cross-referencing the respective gene symbols with sample gene lists.
The significance of enrichment within each dataset was determined using
a one-tailed Fisher’s exact test and corrected for multiple comparisons
using the Benjamini-Hochberg (BH) false discovery rate (FDR) method.
For the analysis of Reactome pathway databases, we utilized
Over-Representation Analysis (ORA). The pathway database was queried at
various levels of granularity by pruning the pathway tree at
incremental depths in terms of pathway hierarchy. For instance, level 2
(L2) represented major pathways and their respective sub-pathways. We
calculated one-tailed Fisher’s exact tests and adjusted p-values using
the BH method. The enrichment plots were generated using in-house
scripts to visualize the comparison of enrichment in two datasets.
Principal component analysis of shared proteins
Principal component analysis (PCA) was conducted to identify major axes
of variation within the proteomic data. We focused on the first
principal component (PC1) as a potential marker of disease-related
proteomic alterations. Spearman’s rank correlation coefficients were
calculated to assess the associations between PC1 and the expression
levels of microtubule-associated protein tau (MAPT) and neurofilament
light chain (NEFL), proteins commonly associated with tau pathology and
axonal injury, respectively.
Results
Proteomic analysis and differential expression in rpAD and spAD patients
rpAD patients had a comparable mean age in years at autopsy to spAD
patients [76.0 (SD 10.4) vs 76.0 (SD 8.2)]. Four of 8 rpAD cases were
female, while 2 of 3 spAD cases were female. The nature and the
distribution, count, and morphology of amyloid plaques were not
uniquely evaluated for each of the brains. Subject demographics are
noted in Table [73]1.
The dataset contained 1841 proteins, with 199 shared across both
phenotypes and 463 unique to rpAD samples. Data imputation was used to
address missing values, enhancing the reliability of protein expression
profiles and improving the accuracy of subsequent analyses.
Differential expression analysis was conducted on the dataset
normalized using the Systematic Variation Normalization (SVN) method,
employing the Reproducibility Optimized Test Statistic (ROTS) [[74]27],
which identified proteins with significantly altered abundance levels
[MATH: (adj-p-value<0.05) :MATH]
between spAD and rpAD cases (Supplementary Table [75]S1). A total of
243 were significantly differentially altered, with 160 proteins
showing decreased abundance and 83 proteins showing increased abundance
in rpAD compared to spAD. In this comparison, we identified key
AD-related genes, such as microtubule-associated protein tau (MAPT),
beta-synuclein (SNCB), and clusterin (CLU)
[MATH: adj-p-value<0.05
:MATH]
) (Fig. [76]1A and B) [[77]28–[78]31]. Among the expressed genes, MAPT
and amyloid-beta precursor protein (APP) levels were strongly
correlated with the rpAD disease phenotype. As shown in Fig. [79]2A and
B, unsupervised cluster analysis of MAPT and APP correlated protein
expression grouped samples by disease phenotypes, rpAD and spAD.
Fig. 1.
[80]Fig. 1
[81]Open in a new tab
A Heatmap of differentially expressed proteins and their association
with AD. Heatmap of differentially expressed proteins
[MATH: (adj-p-value<0.05) :MATH]
. Genes known to be associated with AD are labeled on the right. The
red color indicates increased expression. The top color bar labels rpAD
(orange) and spAD (cyan) samples. B Volcano plot showing the
differential expression of proteins between conditions. Volcano plot
showing significantly different expression of proteins. Blue circles
highlight proteins with statistically significant expression
differences
[MATH: (adj-p-value<0.05) :MATH]
. Increased expression in rpAD is shown on the right
Fig. 2.
[82]Fig. 2
[83]Open in a new tab
Protein expression profiles of (A) MAPT-correlated and (B)
APP-correlated proteins are shown. Pearson’s correlation is calculated
based on all sample profiles, and in these figures, any proteins that
have an absolute correlation greater than 0.85 are shown (See
Supplementary Fig. S1 for a lower threshold). The proteins are then
subjected to unsupervised clustering. rpAD (rpA) and spAD (spA) samples
separate based on their expression profiles
Functional analysis of proteomic changes between the rpAD and spAD groups
Gene ontology (GO) analysis of the 243 significantly altered proteins
revealed strong links to DNA Double-Strand Breaks (DSBs), Epigenetic
Alterations, Thyroid Dysfunction, and Telomere Attrition (Supplementary
Fig. [84]S2 and Supplementary Table [85]S2). Based on the annotation
terms and shared genes, we can group significant terms into 3 essential
functional groups: 1. Gene Silencing and Epigenetic Regulation, 2. DNA
Repair and Telomere Organization, 3. Metabolic Processes and RNA
Regulation (Fig. [86]3). These 3 groups show significant candidate
pathway differences worthy of exploration in rpAD compared to spAD in
future studies.
Fig. 3.
[87]Fig. 3
[88]Open in a new tab
Significant Gene Ontology (GO) terms grouped by biological function
based on ROTS statistics: Each bar represents the mean ROTS statistic
for genes annotated under a specific GO term, grouped into three
categories: Gene Silencing and Epigenetic Regulation, DNA Repair and
Telomere Organization, and Metabolic Processes and RNA Regulation. Bars
are colored by trend in expression: UP (blue) for positively regulated
GO terms and DOWN (red) for negatively regulated ones in rpAD compared
to spAD. Error bars indicate ±1 standard deviation. GO term labels
include both the identifier and a descriptive name
FFT vs FFPE comparison
A total of 1841 proteins were characterized in the current FFT study,
of which 1283 proteins were shared between the current FFT and the
prior FFPE study (Supplementary Fig. [89]S1a). There were only 14
proteins that were significantly differentially expressed in both
studies. The small overlap shows that the results of the two studies
could likely vary based on the tissue preparation method and suggest
variability in protein preservation in the different methods or samples
used. To further understand the differences, we investigated the
ontologies and pathways related to significantly altered proteins in
both studies (Fig. [90]4). While we observe some overlap, the FFT
dataset identified more pathways. The pathways identified are
categorized into various biological processes (at top-level categories
in the Reactome pathway database), including vesicle-mediated
transport, signal transduction, sensory perception, programmed cell
death, organelle biogenesis and maintenance, neuronal system, muscle
contraction, metabolism of RNA and proteins, immune system, gene
expression, disease, developmental biology, cellular responses to
stimuli, cell-cell communication, and cell cycle.
Fig. 4.
[91]Fig. 4
[92]Open in a new tab
Pathway enrichment scores are shown for significant pathways identified
from FFT and FFPE [[93]15] datasets. The pathways database is collected
from the Reactome Pathway Database, where only top-level pathways are
considered. While there is an overlap in identified pathways, there are
more annotations identified in our study
Identifying a larger number of differentially expressed proteins and
associated pathways in FFT compared to FFPE suggests that FFT preserves
a broader range of protein co-aggregates, potentially offering a more
comprehensive, albeit complex, molecular snapshot of the plaque
environment, rather than necessarily a clearer distinction in all
aspects (Supplementary Fig. [94]S5).
Similarly, the gene ontology enrichment comparison between FFT and FFPE
samples reveals that FFT samples offer more granular functional
annotations and a more transparent disease comparison (rpAD vs spAD)
(Supplementary Fig. [95]S6).
rpAD FFT samples with matched CSF
We next evaluated CSF from rpAD cases with matched FFT brain tissue and
compared the proteomic differences against a previously published CSF
proteomic evaluation in AD subtypes identified by data-driven
clustering methods on CSF proteins and controls [[96]25]. Supplementary
Fig. [97]S7 A depicts the intersection of gene clusters enriched in
both studies and those unique to each (120 for rpAD CSF only, 874 for
CSF of Tijms et al. [[98]25] and 435 shared across). We further
investigated the enriched biological processes within these gene
clusters, highlighting the functional differences between the two
sample groups. Supplementary Fig. [99]S7B shows a dot plot comparing
gene enrichment between three groups: genes shared by both CSF and
Tijms et al. [[100]25] datasets, and genes unique to each study are
shown. Both datasets share enrichment in biological processes like
complement activation, humoral immune response, and extracellular
matrix organization, indicating a common biological theme.
Additionally, we compared the rpAD amyloid plaque-related proteins from
FFT samples to those of matched CSF samples and a CSF proteomic AD
subtype with rapid progression identified in Tijms et al. [[101]25]
(Fig. [102]5). 35 proteins were identified across all three groups Fig.
[103]S8). GO enrichment analysis of these proteins showed that they are
enriched in biological processes related to the immune response,
including defense response, humoral immune response, complement
activation, and acute inflammatory response
[MATH: (adj-p-value<0.05) :MATH]
.
Fig. 5.
[104]Fig. 5
[105]Open in a new tab
A This Venn diagram illustrates the overlap and differences in protein
expression between cerebrospinal fluid (CSF) samples from the rapid
subtype from [[106]25], rpAD Amyloid Plaque FFT (AP), and rpAD CSF
samples. The overlapping region indicates that 35 proteins are
significantly altered in AP (amyloid plaque) samples and observed in
CSF samples. B The dot plot visualizes the enrichment of biological
processes within these 35 proteins. The y-axis lists the enriched
processes, while the x-axis represents the GeneRatio, indicating the
proportion of genes within the set associated with each process. The
size of the dots corresponds to the number of genes involved in each
process, and the color represents the statistical significance of the
enrichment (lower p-values indicated by warmer colors)
Among the 35 genes, five genes—PGAM1, YWHAG, DLD, PARK7, and EPDR1—were
upregulated, while the remaining 30 were downregulated (Fig. [107]S8).
Functional annotation demonstrated that the upregulated genes are
mainly involved in glycolysis (PGAM1), mitochondrial metabolism (DLD),
redox regulation (PARK7), stress signaling (YWHAG), and potential
lysosomal and adhesion-related roles (EPDR1) [[108]32–[109]34]. In
contrast, the downregulated set includes proteins critical to
proteostasis (CLU, HSPA5), synaptic adhesion (NRXN1), complement
regulation (C3, CD55), antioxidant defense (APOD, HP, GGCT), and
extracellular matrix stability (LAMB1, NID2, FBLN1, SPON1)
[[110]34–[111]39].
Pathway enrichment analysis based on these proteins underscores a
widespread failure in multiple protective systems. There is
downregulation in the complement cascade (C3, CD55), ER stress
responses (HSPA5), lysosomal function (CTSB, CTSS, NPC2), A
[MATH: β :MATH]
modulation (CLU, SPON1, A2M), and ECM organization (LAMB1, NID2)
[[112]38–[113]40]. Notably, synaptic structural proteins (NRXN1, TPM1)
are reduced, suggesting rapid loss of neuronal connectivity [[114]41].
Cathepsin S (CTSS) is a lysosomal cysteine protease that plays a key
role in antigen presentation (via MHC class II pathway), proteolysis of
extracellular matrix proteins, and microglial activation. The
downregulation of CTSS in rpAD—contrasting its established upregulation
in typical AD—suggests a distinct neuroinflammatory and proteostatic
state is likely from early versus late stage of AD-dependent expression
levels. This possibility of differential expression by AD stage is also
seen with other CSF inflammatory proteins [[115]42, [116]43]. In
late-stage AD, compared to early-AD, microglia may shift from a
pro-inflammatory to a dystrophic or exhausted state, leading to reduced
CTSS expression. In spAD models, CTSS drives microglial M1 activation
via the CX3 CL1-CX3 CR1-JAK2/STAT3 axis [[117]44], and pharmacologic
CTSS inhibition enhances BDNF/TrkB-mediated synaptic plasticity and
cognitive performance in mice [[118]45]. Its reduction in rpAD may
reflect either a “burned-out” inflammatory response or loss of
CTSS-expressing cell populations.
In parallel, YWHAG, a 14-3-3 protein, exacerbates tau pathology in AD
by promoting tau hyperphosphorylation through its interaction with tau
and tau-phosphorylating kinases [[119]46]. It also stabilizes
hyperphosphorylated tau, preventing its degradation and contributing to
toxic aggregate accumulation and neuronal dysfunction [[120]46].
Additionally, YWHAG upregulation disrupts protein homeostasis and
promotes calcium dysregulation and ER stress via pathways like C/EBP
[MATH: β :MATH]
-TRPC1-SOCE, further aggravating tau aggregation and neurodegeneration
[[121]47].
The downregulation of GGCT, TPM1, and SPON1 points to synergistic
vulnerabilities. GGCT’s role in glutathione metabolism implies that its
loss may weaken antioxidant defenses, heightening oxidative stress
damage [[122]48]. TPM1 reduction likely destabilizes actin filaments in
dendritic spines, accelerating synaptic collapse and cognitive decline
[[123]49]. Lowered SPON1 removes an endogenous inhibitor of BACE1
cleavage of APP, potentially driving the A
[MATH: β :MATH]
accumulation in rpAD [[124]50]. Increased CSF levels of spondin 1 in
the AD compared to normal controls have been noted [[125]51]. An
association between the minor allele (A) of rs11023139 in the SPON1
gene and reduced rates of cognitive decline during the later stages of
AD has been reported in the Alzheimer’s Disease Neuroimaging Initiative
(ADNI) study [[126]52]. Additionally, the same locus has been
implicated as a factor in cognitive decline during the preclinical
stages of AD in the Australian Imaging, Biomarkers and Lifestyle Study
of Aging (AIBL) [[127]53]. Older people with SPON1 variant at rs2618516
also had significantly milder clinical dementia scores and lower risk
of AD [[128]54].
Reduced CLU and HSPA5 indicate impaired extracellular and ER
proteostasis, respectively, critical for protein folding and misfolded
protein clearance [[129]55, [130]56].
SMOC1, a matricellular protein enriched in amyloid plaques in
early-onset AD and Down syndrome-associated AD [[131]57], is
significantly downregulated in our rpAD cohort. Its reduction may
impair extracellular matrix remodeling or limit protective signaling
cascades associated with plaque containment and stabilization,
potentially facilitating a more permissive environment for neurotoxic A
[MATH: β :MATH]
species aggregation.
A2M inhibits A
[MATH: β :MATH]
fibril formation and promotes its clearance via LRP1 [[132]58]. When
A2M and LRP1 are downregulated in rpAD, they could synergistically
impair A
[MATH: β :MATH]
clearance. Decreased A2M levels would result in fewer A
[MATH: β :MATH]
-A2M complexes, while reduced LRP1 expression would limit the capacity
to transport these complexes across the blood-brain barrier.
Haptoglobin (HP), an apolipoprotein E (APOE) antioxidant, binds APOE
and A
[MATH: β :MATH]
, facilitating its clearance, as previously reported [[133]59]. NRXN1,
a key synaptic adhesion molecule reduced in rpAD, indicating impaired
synaptic integrity [[134]60]; and CTSB/CTSH, lysosomal enzymes whose
decline may impede proteolytic clearance of toxic aggregates.
Downregulation of NID2, LAMB1, and FBLN1, all ECM structural genes,
underscores loss of matrix integrity and vascular support, while
suppression of C3, CD55, and SERPINA3 reveals broad immune and
complement dysfunction—potentially leading to inefficient clearance and
increased vulnerability to inflammatory damage [[135]38, [136]61].
Coordinated proteomic downregulation parallels reduced CSF tau in rpAD
Principal component analysis of the 35-protein signature yielded a
first component (PC1) that separates rpAD from spAD cases. Spearman’s
correlation shows that PC1 is inversely associated with MAPT (
[MATH: ρ :MATH]
= −0.62, p < 0.048), supporting the interpretation that samples
exhibiting the greatest coordinated down-regulation of these proteins
tend to have the lowest tau levels. By contrast, the relationship
between PC1 and NEFL is also negative but does not reach significance (
[MATH: ρ :MATH]
= −0.44, p = 0.18), indicating that this proteomic axis tracks more
closely with tau reduction than with axonal-injury marker variation.
Discussion
This study compared proteins in rpAD using two methods, FFT and FFPE
tissue analysis. Despite different tissue preparation methods, we
identified overlapping protein coverage, including proteins associated
with the immune and nervous systems (Fig. [137]4). However, the methods
also revealed differences. The FFT samples yielded a significantly
larger number of proteins not identified as differentially expressed in
the FFPE analysis. This list of unique proteins in FFT included many
AD-related genes, and GO analysis provided novel biological function
differences. Using FFT samples also provided a more precise disease
ontology mapping. FFT tissue preparation significantly enhanced our
proteomic insights and is interesting for follow up in future analyses.
Our analysis identified three main biological themes separating rpAD
from spAD.
* Gene Silencing and Epigenetic Regulation: Epigenetic modifications,
like DNA methylation and histone modifications, regulate gene
expression, leading to activation or silencing and impacting
various cellular processes, have been postulated in AD
pathophysiology [[138]62]. Most differentially methylated positions
in neuropathological ‘bulk’ cortical tissue are thought to reflect
DNA methylation differences occurring in non-neuronal cells
[[139]63].
* Metabolic Processes and RNA Regulation: Epigenetic modifications
affect post-transcriptional RNA processing, and RNA modifications
reciprocally regulate gene expression transcriptionally by
influencing the epigenome [[140]64].
* DNA Repair and Telomere Organization: DNA repair mechanisms are
crucial for maintaining genome stability. DNA double-strand breaks
are among the most serious types of DNA damage, and their signaling
and repair are critical [[141]65]. Telomeres are the genomic
portions at the ends of linear chromosomes. Telomeric DNA in
vertebrates protects them from being recognized as DNA damage that
triggers a DNA damage response [[142]66].
Proteomic profiling of eight rpAD and three spAD cases revealed a clear
dichotomy among proteins that are altered in rpAD relative to spAD.
Additionally, overlapping proteins in rpAD amyloid plaques and CSF
informed our mechanistic model of rapid AD. The rapid progression of
rpAD is driven by a distinct molecular signature characterized
primarily by a widespread failure of essential cellular maintenance and
defense systems, rather than the hyperactivation of specific
detrimental pathways often associated with typical spAD [[143]67].
This accelerated trajectory may result from a synergistic collapse
across multiple cellular systems, as defined by a unique protein
expression signature. Key factors include the catastrophic failure of
proteostasis, where impaired chaperoning and reduced
clearance/modulation allow toxic species to accumulate rapidly.
Additionally, lysosomal dysfunction further hinders aggregate
clearance, while compromised inflammatory and immune responses fail to
effectively tag debris for clearance or mount a typical inflammatory
response.
Accelerated synaptic failure and loss of extracellular matrix integrity
in rpAD create an unstable environment. Overwhelmed stress responses
and metabolic crises generate damage that antioxidant and chaperone
systems cannot effectively neutralize. This synergy of failures,
resulting in a distinct molecular subtype, potentially drives the rapid
progression of rpAD via a catastrophic, widespread breakdown of
cellular maintenance, defense, and structural systems. This systemic
collapse leads to rapid decline as the brain’s impaired resilience and
repair capacity allow toxic conformers or baseline damage processes to
proceed unchecked.
Our analysis indicates that the primary axis of variation in CSF
proteomic profiles (“[144]Coordinated proteomic downregulation
parallels reduced CSF tau in rpAD” section) is significantly associated
with MAPT levels, a key marker of tau pathology in AD. The lack of a
significant correlation between shared CSF proteins and NEFL suggests
that axonal injury, while relevant, may not be the predominant factor
driving the observed proteomic differences between rpAD and spAD cases.
This aligns with previous studies highlighting the central role of tau
pathology in AD progression [[145]68–[146]72].
This study has some limitations. Firstly, the small sample size in this
study is exploratory and needs future validation in larger samples. The
limited number of spAD cases reflects the stringent matching criteria
employed based on age and post-mortem interval relative to the rare
rpAD cohort available from our center during the study period. While
obtaining additional spAD cases is generally feasible, securing cases
that precisely matched the specific post-mortem characteristics of the
available rpAD group proved challenging. Additionally, there is limited
data on the severity of AD pathology by Braak stage in spAD cases. The
absence of APOE
[MATH: ϵ4 :MATH]
status data in both rpAD and spAD groups further constrains our
analysis. Finally, rpAD and spAD were identified in two distinct
autopsy cohorts, which may introduce related biases. Replication in
larger cohorts is necessary to validate these findings. Furthermore,
this study did not include a direct comparison to non-demented control
cases, which limits the ability to discern AD-specific changes from
normal aging or other non-AD pathologies.
While our study focused on plaque co-aggregates and broad CSF
proteomics, future analyses should directly compare these findings with
established AD fluid biomarkers like phosphorylated Tau (pTau), total
Tau (tTau), and neurofilament light chain (NfL) levels in the same
cohort. Such comparisons would help contextualize the identified
proteomic signatures within the known spectrum of AD pathology and
progression markers, potentially revealing how plaque-associated
changes relate to neurodegeneration and tauopathy severity in rpAD
versus spAD.
Conclusion
Shared amyloid plaque-related protein compositions can reliably
differentiate rpAD and spAD across two brain tissue preparation
methods. Differences in protein signatures related to the innate immune
system and cellular transport of small molecules were noted between the
two brain tissue preparation methods. A shared protein signature could
be identified between CSF and amyloid plaque proteomics enriched for
neuroinflammation-related proteins. The results support the notion that
rpAD has a distinct biological profile compared to spAD, and altered
immune changes could be key markers of these AD subtypes. If confirmed
in larger studies, a distinct CSF profile in rpAD could aid in
prognostication and AD clinical trial evaluation.
Supplementary information
[147]13195_2025_1767_MOESM1_ESM.pdf^ (1.3MB, pdf)
Additional file 1. This file contains supplementary tables and figures
related to the study. The supplementary materials include: Table S1:
Pairwise differentially expressed proteins identified using ROTS, with
statistical significance and FDR-adjusted p-values. Table S2: Enriched
Gene Ontology (GO) terms associated with differentially expressed
proteins. Figures S1-S8: Figures illustrating protein expression
profiles, gene ontology terms, Venn diagrams, and enrichment of disease
ontology categories.
[148]13195_2025_1767_MOESM2_ESM.csv^ (123.6KB, csv)
Additional file 2. Proteome Analysis of Amyloid Plaque from FFT
samples: This file contains the proteomic data of rpAD and spAD amyloid
plaque samples. The data includes log2 LFQ intensity values for various
proteins across multiple replicates and additional information such as
peptide counts, sequence coverage, molecular weight, and protein
identifiers.
[149]13195_2025_1767_MOESM3_ESM.xlsx^ (632.6KB, xlsx)
Additional file 3. Proteome Analysis of Cerebrospinal Fluid (CSF): This
file contains the proteomic data of rpAD cerebrospinal fluid (CSF)
samples. The data includes log2 LFQ intensity values for various
proteins across multiple replicates and additional information such as
peptide counts, sequence coverage, molecular weight, and protein
identifiers.
Acknowledgements