Single-nucleus transcriptomics reveals a distinct microglial state and increased MSR1-mediated phagocytosis as common features across dementia subtypes

Abstract

Background

Alzheimer’s disease (AD), dementia with Lewy bodies (DLB), and Parkinson’s disease dementia (PDD) collectively represent the majority of dementia cases worldwide. While these subtypes share clinical, genetic, and pathological features, their transcriptomic similarities and differences remain poorly understood.

Methods

We applied single-nucleus RNA-sequencing (snRNA-seq) to prefrontal cortex samples from individuals with non-cognitive impairment control (NCI), and dementia subtypes (AD, DLB, and PDD) to investigate cell type-specific gene expression patterns and pathways underlying pathological similarities and differences across dementia subtypes. SnRNA-seq findings were validated through RNAscope, immunohistochemistry, and additional biochemical analyses in human tissues and cellular models.

Results

SnRNA-seq analysis revealed elevated microglial proportions across all dementia subtypes compared to NCI. Further analysis of cell type-specific transcriptomes identified overlapping differentially expressed genes (DEGs) between microglia and oligodendrocytes across all dementia subtypes. While AD showed molecular similarities to NCI, PDD and DLB were clustered more closely together, sharing a greater number of DEGs and related pathways, predominantly associated with microglia. Investigation of interactions between microglia and oligodendrocytes revealed a distinct microglial state in all dementia subtypes. MSR1, a gene encoding a scavenger receptor, was upregulated in microglia across all dementia subtypes, along with its associated gene HSPA1A in oligodendrocytes. RNAscope supported the potential interaction between microglia and oligodendrocytes, where these cells were in closer proximity to each other in human cortical tissues of PDD compared to NCI. MSR1 expression was significantly increased in cortical primary microglia from PD mice compared with non-transgenic (NTg) mice. Additionally, the expression of myelin-associated genes (MBP, MOBP, and PLP1) was significantly upregulated in PD microglia compared to NTg, supporting the presence of the distinct microglia. Furthermore, MSR1-positive microglia colocalised with MBP in cortical tissue of PDD patients, suggesting a functional role of MSR1 in myelin debris clearance. Overexpression of MSR1 in microglial cells enhanced their phagocytic activity toward myelin, and reciprocally, myelin treatment upregulated MSR1 protein levels, indicating enhanced MSR1-mediated myelin phagocytosis.

Conclusions

Our findings provide novel insights into the cell type-specific role of microglial MSR1 in AD, DLB, and PDD, linking its increased phagocytic capacity to myelin defects as a common feature of neurodegenerative dementias.

Supplementary Information

The online version contains supplementary material available at 10.1186/s13073-025-01519-4.

Background

Dementia is a condition characterised by a significant decline in cognitive abilities, impairing a person’s ability to perform daily tasks, work, or engage in social activities compared to their previous level of functioning. According to a World Health Organization (WHO) report in 2022, approximately 55.2 million people were living with dementia worldwide. Currently, dementia ranks as the seventh most common cause of death and poses a significant burden in terms of disability and dependency among the elderly population. Alzheimer’s disease (AD) is the most prevalent form of neurodegenerative dementia, followed by Lewy body dementia (LBD), which encompasses both dementia with Lewy bodies (DLB) and Parkinson’s disease dementia (PDD) [1].

AD is pathologically characterised by the widespread presence of amyloid beta (Aβ) plaques and tau neurofibrillary tangles [2, 3]. Genetic risk factors for AD include apolipoprotein E (APOE), with the ε4 allele being the main genetic risk factor for late-onset AD [4]. Mutations in amyloid precursor protein (APP), presenilin 1 (PSEN1), and presenilin 2 (PSEN2) are associated with early-onset AD [5]. Both PDD and DLB are associated with Lewy body pathology, characterised by the accumulation of misfolded alpha-synuclein (α-Syn) in the presence of Lewy bodies (LBs) and Lewy neurites (LNs) [6, 7]. Some patients with PDD and DLB exhibit varying degrees of amyloid plaque and tau tangle pathology, while some AD patients may also present with LBs [7, 8]. Clinically, PDD and DLB share core diagnostic features, but they can be distinguished by the “one-year rule”, which pertains to the relative temporal onset of cognitive versus motor symptoms [9]. Genetic risk factors for PDD overlap with those of Parkinson’s disease, such as mutations in alpha-synuclein (SNCA) [10], APOE4 [11, 12], and leucine-rich repeat kinase 2 (LRRK2) [13]. For DLB, genetic risk factors include mutations in SNCA [14], APOE4 [15], and glucocerebrosidase (GBA) [16]. APOE4 is a common genetic risk shared across dementia subtypes, being the strongest known risk factor for AD and occurring at a higher frequency in DLB than in PDD [17]. Despite shared features across these dementia subtypes, differences in the extent of co-pathology, genetic risk variants, and the progression of pathological hallmarks contribute to the distinct diagnoses of AD, PDD, and DLB. These overlapping features complicate the understanding of their underlying pathophysiology, necessitating further research to elucidate their similarities and differences.

Microglia, as the resident immune cells in the central nervous system (CNS), are pivotal in maintaining CNS homeostasis and neuronal circuitry integrity through the clearance of damaged or unnecessary neurons, synapses, and cell debris [18, 19]. Microglia exhibit a pronounced phagocytic capacity underpinned by a diverse array of receptors, including Toll-like receptors (TLRs), Tyro3/Axl/MerTK (TAM) receptors, CD14, scavenger receptors, and triggering receptor expressed on myeloid cells 2 (TREM2) [20–22]. In AD, TREM2 is a well-characterised receptor expressed on microglia, which activates a disease-associated microglia phenotype that limits the propagation of Aβ and tau via a phagocytic mechanism [23, 24]. Moreover, microglial receptors such as TLR2, TLR4, CD36, and P2X7 have been implicated in mediating microglial activation in response to α-Syn, thereby boosting α-Syn clearance through enhanced microglial phagocytosis in PD [25–28]. Microglia are also recognised for their ability to engulf myelin debris, a process regulated by multiple receptors, such as TREM2, Mer tyrosine kinase (MerTK), and scavenger receptors [29]. A growing number of studies have highlighted the role of microglia in myelin phagocytosis during developmental myelination and in multiple sclerosis [30, 31]. However, the involvement of microglia in receptor-regulated myelin clearance in AD, DLB, and PDD remains unclear.

Macrophage scavenger receptor 1 (MSR1) belongs to the scavenger receptor family and is implicated in various pathophysiological processes, including atherosclerosis, AD, neuroinflammation, and cancer [32]. It is primarily expressed in phagocytes such as macrophages, microglia, and monocytes [33, 34]. MSR1 can bind to various ligands such as soluble and fibrillar Aβ [35, 36], APOE [37], α-Syn [38], and heat shock proteins [39, 40]. Numerous studies have demonstrated elevated MSR1 expression in transgenic AD mice [41, 42] and its role in facilitating Aβ clearance through microglial phagocytosis, which is beneficial in reducing the accumulation of Aβ plaques [43]. However, sustained activation of microglia during this process can lead to chronic neuroinflammation, characterised by the production of nitric oxide and reactive oxygen species, ultimately posing detrimental effects on neuronal health [35, 36]. Research on the role of MSR1 in myelin phagocytosis has predominantly focused on MS pathology, showing the involvement of MSR1 in the early uptake of myelin by microglia and attenuated demyelination after MSR1 knockout in the experimental autoimmune encephalomyelitis model [45]. The exact function of microglial MSR1 in myelin phagocytosis within AD remains unclear. Moreover, given the multi-ligand nature of MSR1 and the shared pathological features in dementia, it is hypothesised that MSR1 might contribute to DLB and PDD by mediating microglial phagocytosis.

To elucidate the differences and similarities among various types of dementia, we applied single-nucleus RNA-sequencing (snRNA-seq) to compare transcriptomes obtained from the postmortem prefrontal cortex of patients diagnosed with AD, as well as from the two primary subtypes of LBD: PDD and DLB. Our analysis revealed a dynamic interaction between microglia and oligodendrocytes, leading to the identification of a distinct microglial state, characterised by a high number of myelin transcripts. To enhance the robustness of our findings, we performed a comparative analysis between our dataset and previously published data, which yielded consistent results. Next, we demonstrated that MSR1 was one of the most significantly upregulated DEGs in microglial clusters across all dementia subtypes. We further experimentally validated its expression changes in cortical tissues from both human and mouse models of PD. Furthermore, we identified evidence of cell interactions between microglia and oligodendrocytes through transcriptomics, as well as colocalisation of MSR1-positive microglia with myelin proteins in the cortex of PDD patients. In parallel, microglia isolated from PD mouse models exhibited increased expression of both MSR1 and myelin-related transcripts. Lastly, overexpression of microglial MSR1 promoted myelin uptake, and exposure to myelin debris induced a subsequent elevation in microglial MSR1 level. Overall, our study sheds light on the intricate interplay between microglia and oligodendrocytes, offering valuable insights into the role of MSR1 in mediating myelin phagocytosis in dementia.

Methods

Postmortem human prefrontal cortical tissues

Postmortem human brain samples were obtained from the Parkinson’s UK Brain Bank, Division of Neuroscience, Imperial College London, as previously described [46]. Postmortem prefrontal cortical tissues from 6 individuals with DLB (three males and three females), 6 individuals with PDD (three males and three females), 4 individuals with AD (one male and three females), and 4 age-matched healthy non-cognitive impairment controls (NCI, two males and two females), whose ages ranged from 73 to 92, were analysed by snRNA-seq. The complete sample metadata is provided in Additional file 1: Table S1. The human prefrontal cortex samples included both white matter and grey matter from the middle frontal gyrus region. All tissues were obtained via a prospective donor scheme with fully informed written consent, and their collection was approved by the UK Human Tissue Authority (Approval Number 18/WA/0238). All methods and protocols used in this study were performed in accordance with institutional and relevant ethical guidelines and regulations. The cause of death, clinical diagnosis, and pathological features of each patient were carefully examined by clinicians and pathologists. The neuropathological diagnoses of the cases, including Braak stage, Thal score, CERAD score, Lewy body stage, and co-pathology, are provided in Additional file 1:Table S3. This study was also approved by the Institutional Review Board of the National Neuroscience Institute of Singapore.

Animals

Mutation in the LRRK2-G2019S variant is a known risk factor for Parkinson’s disease. Mice harbouring transgenes encoding the human mutant LRRK2-G2019S were utilised in accordance with our previously published papers [47–49]. Mice hemizygous for the C57BL/6J bacterial artificial chromosome (BAC) LRRK2-G2019S transgene were obtained from The Jackson Laboratory (#012467). In this model, the kinase is driven by the human LRRK2 promoter/enhancer regions on the BAC transgene. Animal care and experimental procedures were conducted in accordance with institutional guidelines and were approved by the Institutional Animal Care and Use Committee Animal Use Protocol (IACUC AUP #19,113 and #25,006) of NTU-LKCMedicine. The 15–17-month-old PD mice and age-matched non-transgenic (NTg) mice used in this study were littermates to ensure genetic consistency across groups. The mice were housed in a pathogen-free facility under a 12-h light/dark cycle with ad libitum access to food and water.

Nuclei isolation from frozen human postmortem brain tissues

Nuclei were isolated from frozen postmortem tissues as previously described with modifications [44]. The tissues were homogenised in chilled lysis buffer (10 mM Tris, 10 mM NaCl, 3 mM MgCl₂, 0.05% NP40) using a glass Dounce homogeniser, with 10 loose pestle strokes, followed by 5 tight pestle strokes. The homogenates were filtered through a 40-μm strainer and centrifuged for 5 min at 500 × g, 4 °C to pellet the nuclear fraction. The pellet was then resuspended in 10 ml of wash buffer (5% BSA, 40 units/ml Protector RNase inhibitor in 0.5 × PBS) and centrifuged for 5 min at 500 × g, 4 °C. The pellet was then resuspended in 1 ml wash buffer, mixed with an equal volume of 50% OptiPrep solution, and layered on top of 2 ml of 29% OptiPrep solution. The nuclei were separated by centrifugation at 1000 × g for 30 min at 4 °C. The nuclear pellet was washed three times with wash buffer (2% BSA, 40 units/ml Protector RNase inhibitor in 0.5 × PBS) and subsequently resuspended in 500 μl wash buffer to yield a clean single-nucleus suspension. For quantification, the collected nuclei were stained with DAPI and counted with Countess II Automated Cell Counter (Life Technologies).

SnRNA-seq library preparation

An estimated 10,000 human nuclei were processed using the Chromium Single Cell 3′ v3 Gene Expression Reagent Kits, following the manufacturer’s instructions (10 × Genomics). The generated snRNA-seq libraries were sequenced using an Illumina HiSeq 4000 at NovogeneAIT Genomics Singapore Pte Ltd.

SnRNA-seq data processing

Preprocessing, including debarcoding and unique molecular identifier (UMI) counting, was performed with Cell Ranger version 3.1.0 (10 × Genomics). For each sample, the UMI count file was read by Seurat (version 4.1.0) in R and then subjected to sample-wise quality control. Quality control included assessing the feature number, UMI number, and mitochondrial gene percentage. Potential doublets were identified and removed using the “scDblFinder” and “DoubletFinder” packages [107]. A total of 2000 variable features in the RNA assay of each sample were identified using the “vst” method in the Seurat package. Data integration was performed using the scVI method with 5000 common feature genes across all samples [108]. The 50 latent dimensions output by scVI were utilised for Uniform Manifold Approximation and Projection (UMAP) analysis, followed by clustering. For downstream analysis, the data were normalised using either “log normalisation” or “sctransform”, depending on the specific analysis being performed.

Cell type annotation

The cell types of each cluster were predicted using the CELLiD method, which is available within the DISCO database. In essence, the average gene expression profile of each cluster was compared with reference data in the DISCO database, and the most analogous cell type was assigned as the predicted cell type. Differentially expressed genes (DEGs) for each cluster were determined using the'FindAllMarkers'function within the Seurat package. Manual cell type annotation was performed based on the results of the automated annotation and the identified DEGs serving as marker genes.

Identification of phenotype-related DEGs

To identify phenotype-related DEGs, such as AD-specific DEGs in microglia, we evaluated two different approaches:

1. The"FindAllMarkers"function from the Seurat package was used with the MAST method for DEG identification between disease and control groups within specific cell types. MAST accounts for the fraction of genes expressed and allows the inclusion of covariates. In our analysis, sex and sample were included as covariates to mitigate the influence of nuclei correlation within samples. DEGs were considered significant if they had an adjusted p-value (BH method) < 0.01.

2. Pseudo-bulk data were generated for each cell type in each sample, and DESeq2 was used to identify phenotype-specific DEGs. Sex was included as a covariate in the model, and default parameters were used. Due to the limited sample size, we applied a soft threshold for DEG filtering, using a p-value < 0.05.

We compared the results of the two methods and observed substantial overlap between the identified DEGs (Additional file 2:Fig. S4). However, due to the limited sample size, DESeq2 identified significantly fewer DEGs compared to MAST. For downstream analysis, we included all DEGs identified by DESeq2 and any DEGs identified by MAST with an absolute logFC greater than 1 to ensure robustness.

GO and KEGG pathway enrichment

Gene set enrichment analyses were performed by the “enrichGO” and “enrichKEGG” functions provided by the “ClusterProfiler” R package using DEG lists. For “enrichKEGG”, gene symbols were first converted to Entrez IDs using the “bitr” function provided by ClusterProfiler. The enrichment results were visualised through either the"treeplot"or"dotplot"functions in the"enrichplot"R package.

Gene set scoring

In certain analyses, including identification of cell types and signalling pathways, we calculated scores for specific gene sets using the"AddModuleScore"function within the Seurat package, using default parameters.

Cell–cell interactions

CellChat was used to identify cell-to-cell interactions in our dataset. Microglia and oligodendrocytes were divided based on disease types (NCI, AD, DLB, or PDD) and were analysed separately. The quantity of interactions and the cumulative weights assigned to these interactions were used for visualisation.

Plasmid construction

The human MSR1 plasmid was generously provided by Associate Professor Wang Penghua’s laboratory [45]. To generate a GFP-expressing construct, the human MSR1 gene was amplified by PCR and then inserted into the pEGFP-N1 vector (Promega). A 98-bp HA-P2A DNA sequence was amplified by overlapping PCR and inserted into the BamHI and AgeI sites between MSR1 and EGFP, which was in frame with the HA-P2A coding sequence. The obtained pEGFP-N1-MSR1-HAP2A-EGFP plasmid expresses the MSR1::HAP2A::EGFP cassette, which will be cleaved into the MSR1:HA and EGFP proteins at the self-cleaving 2 A peptide. The expression of MSR1 is not influenced by the cellular localisation of EGFP in the cells. The sequence of the resulting plasmid was confirmed by Sanger sequencing.

Myelin isolation

Myelin was isolated from adult mouse brains using a sucrose gradient as previously described [50]. The mice were euthanized by carbon dioxide (CO₂). Mouse brain tissues were homogenised with a glass Dounce homogeniser, with 5 strokes of the loose pestle and 5 strokes of the tight pestle in cold TrisCl buffer (20 mM TrisCl, 2 mM Na₂EDTA, 1 mM dithiothreitol, protease inhibitors, pH 7.4), and resuspended in 0.32 M Tris-buffered sucrose. The brain homogenate in 0.32 M sucrose was then layered on top of a 0.83 M sucrose gradient in an ultracentrifugation tube and ultracentrifuged for 30 min at 75,000 × g, 4 °C. The interface between 0.32 M and 0.83 M layers (myelin) was collected, subjected to osmotic shock (TrisCl buffer) twice, and centrifuged for 10 min at 12,000 × g, 4 °C. The obtained crude myelin pellet was resuspended in 0.30 M sucrose solution, layered on top of a 0.83 M sucrose gradient in an ultracentrifugation tube and ultracentrifuged for 30 min at 75,000 × g, 4 °C. The interface between the 0.32 M and 0.83 M layers (purified myelin) was collected, and its concentration was measured using the RC-DC Protein Assay (Bio-Rad). The purified myelin was frozen at − 80 °C until further processing. Before use, myelin was labelled with pHrodo (Thermo Fisher).

Treatment with human Aβ peptide, human α-Syn preformed fibril (Pff), and myelin protein

Human microglial HMC3 cells were maintained in DMEM/F12 medium supplemented with 10% FBS, 1% GlutaMAX and 1% penicillin/streptomycin under standard culture conditions (95% relative humidity with 5% CO₂ at 37 °C). The cells were treated with (1) 10 μM Aβ (Anaspec), (2) 10 μg/ml α-Syn Pff (StressMarq), (3) a combination of 5 μM Aβ and 5 μg/ml α-Syn Pff, and (4) 20 μg/ml myelin protein for 24 h. Following incubation, the cells were harvested for further analysis.

In vitro phagocytosis assay

The human microglial HMC3 cells were transfected with the MSR1 plasmid or an empty vector using the Lipofectamine LTX Kit (Thermo Fisher) for 48 h, followed by treatment with 20 μg of pHrodo-labelled myelin for 6 h. After incubation, the cells were imaged using a confocal microscope.

Isolation of primary murine microglia

Adult male LRRK2-G2019S PD mice (15–17 months old) and age-matched NTg mice were used for primary microglia isolation following an established protocol [51] with modifications. The mice were anaesthetised via intraperitoneal (i.p.) injection of a ketamine/xylazine and transcardially perfused with ice-cold PBS. Whole brains were harvested, and cortical tissues were dissected, finely minced and digested in 5 ml of Collagenase A (2 mg/ml, Sigma) at 37 °C for 15 min. The tissues were dissociated by pipetting up and down 20 times with a 5-ml serological pipette, and this step was repeated three times. The homogenates were then filtered through a 70-μm cell strainer to remove clumps. Debris was removed using a debris removal solution (Miltenyi Biotec) followed by treatment with a red blood cell removal solution (Miltenyi Biotec). Microglia were then isolated through magnetic sorting using CD11b microbeads (Miltenyi Biotec).

The isolated CD11b-positive microglia were cultured in adult glial medium supplemented with macrophage colony-stimulating factor (MCSF; 100 ng/ml; Peprotech) and granulocyte–macrophage colony-stimulating factor (GM-CSF; 50 ng/ml; Peprotech) for 5 days. On day 5, the microglia were fixed and subjected to immunostaining with antibodies against microglial markers (CD11b, IBA1, CD68, and TMEM119) as well as neuronal markers (NeuN and MAP2) to assess the purity of the isolated microglia.

RNAscope multiplex in situ hybridisation

The RNAscope assay was used to simultaneously detect the expression of human ITGAM, PTPRC, MBP, and PLP1 in FFPE sections of cortical tissues. The RNAscope probes used included Hs-ITGAM, Hs-PTPRC-C2, Hs-PLP1-C3, and Hs-MBP-C4, along with the RNAscope 4-Plex Negative Probe, all obtained from Advanced Cell Diagnostics (ACD). TSA Vivid fluorophores 520, 570, and 650 (ACD) and Opal 620 (Akoya Biosciences) were used for probe signal detection. DAPI was used to stain nuclei. Sample pretreatment, probe hybridisation, and signal amplification were carried out according to the manufacturer’s protocol (Multiplex Fluorescent Reagent v2 Kit, ACD). Slides were mounted with ProLong™ Gold Antifade Mountant (Invitrogen). Fluorescent signals were captured using a Vectra slide scanner.

RT-qPCR

Total RNA was extracted from cells using an RNeasy Mini Kit (Qiagen) and reverse-transcribed with an iScript™ cDNA Synthesis Kit (Bio-Rad). The generated cDNA was subjected to RT-qPCR using All-in-One™ qPCR Mix (GeneCopoeia). β-actin was used as a reference gene for mRNA expression normalisation. The relative expression of each gene was calculated using the 2^−ΔΔCT method relative to β-actin expression. The primers used for RT-qPCR are listed in Additional file 1: Table S4.

Western blotting

Proteins were extracted from cells using RIPA lysis buffer supplemented with a protease inhibitor cocktail and a phosphatase inhibitor cocktail (MedChemExpress). Protein concentrations were measured using an RC DC™ Protein Assay Kit II (Bio-Rad). A total of 20 μg of protein lysates was separated by SDS-PAGE and then transferred to polyvinylidene fluoride (PVDF) membranes (0.45 μm pore size, Millipore). The membranes were blocked with 5% milk in Tris-buffered saline with Tween 20 (TBST) for 1 h at room temperature and incubated with a 1:1000 dilution of primary antibodies (rabbit anti-MSR1, Invitrogen; mouse anti-β-actin, Santa Cruz) overnight at 4 °C on a rotating shaker. The membranes were then washed with TBST, followed by incubation with horseradish peroxidase (HRP)-conjugated secondary antibody (1:3000, Cell Signalling Technology). The protein bands were visualised using the Pierce™ ECL Western Blotting Substrate (Thermo Fisher) and imaged with the Bio-Rad ChemiDoc Imaging System. Band intensities were quantified using ImageJ software.

Immunohistochemistry

Slides of NCI and PDD cortical brain tissue sections were deparaffinised at 60 °C, followed by incubation in xylene (100%, twice) and a 1:1 mixture of xylene and isopropanol (once) for 10 min each. The slides were rehydrated in a descending ethanol series. This was followed by incubation in acidic antigen retrieval solution (pH 6.0) for 5 min and boiling for 10 min. The samples were cooled for 20 min and washed in distilled water for 1 min before subsequent permeabilisation in 0.1% Triton X-100 in PBS. Next, slides were incubated overnight at 4 °C with a 1:200 dilutions of primary antibodies (rabbit anti-MSR1, Invitrogen; rat anti-MBP, Merck Millipore; mouse anti-IBA1, Invitrogen; mouse anti-TMEM119, Cell Signalling Technology; mouse anti-CD11b, Bio-Rad; mouse anti-CD68, Bio-Rad; mouse anti-NeuN, Merck Millipore; chicken anti-MAP2, Novus Biologicals). After three washes with 1 × PBS, the samples were incubated with secondary antibodies (1:400, Alexa Fluor 488/555/647 goat anti-rabbit/mouse/rat/chicken, Invitrogen) for 2 h at room temperature. DAPI (Sigma) was used to counterstain the nuclei. Finally, the slides were washed, dried, and mounted using Hydromount Aqueous Nonfluorescing Mounting Medium (National Diagnostics), followed by imaging using a confocal microscope (FV3000 Olympus) at × 40 magnification.

Statistical analyses

For each experiment, at least three independent measurements were conducted. All statistical analyses were performed using GraphPad Prism 8 software. Data are presented as mean ± SEM for all statistical analyses. For two-group comparisons, Student’s t test was used. For multiple group comparisons under one experimental condition, one-way ANOVA with Tukey’s post hoc test was used to compare the differences between groups. For multiple group comparisons under two experimental conditions, two-way ANOVA with Tukey’s post hoc test was used. The statistical significance levels were set at *p < 0.05, **p < 0.01, and ***p < 0.001.

Results

SnRNA-seq transcriptomics revealed transcriptional profiles in AD, DLB, and PDD

To explore the disease-related cellular and molecular profiles associated with the dementia subtypes (AD, DLB, and PDD), we conducted single-nucleus RNA-sequencing (snRNA-seq) on postmortem prefrontal cortex tissue samples from 20 individuals obtained from Parkinson’s UK Brain Bank using the 10 × Genomics platform (Fig. 1a). Our study cohorts comprised 4 individuals with AD, 6 with DLB, 6 with PDD, and 4 age-matched NCI. There were no significant age differences among the dementia subtypes and NCI (Additional file 1:Table S2). Following quality filtering and doublet removal, 87,470 high-quality nuclei were retained and integrated. Unsupervised clustering and marker gene identification revealed 8 major cell types (Fig. 1b,c): excitatory neurons expressed SYT1 and CBLN2 (20%), inhibitory neurons expressed SYT1 and GRIP1 (22%), oligodendrocytes expressed ST18 (37%), oligodendrocyte precursor cells expressed PCDH15 (OPCs, 4%), astrocytes expressed SLC1A2 (10%), microglia expressed DOCK8 (4%), endothelial cells expressed FLT1 (1%), and ependymal cells expressed CFAP54 (1%) (Additional file 2: Fig.S1).

Fig. 1.

SnRNA-seq revealed an increase in microglial proportion in AD, DLB, and PDD. a Schematic diagram of experimental design. b Unbiased UMAP projection of 90,733 single nuclei, showing 8 major cell clusters. Each dot represents a single nucleus, colour-coded by cell type. c Dot plot showing the expression of specific marker genes for the 8 cell clusters. Dot size reflects the percentage of cells expressing each gene within the cluster, while colour represents average log-normalised gene expression. Marker genes were ordered to highlight the differences between cell types. d Unsupervised hierarchical clustering of samples (nuclei number > 200) based on cell type proportions. e Box plots showing microglial composition among NCI, AD, DLB and PDD groups

Subsequently, we assessed changes in cell type composition across the dementia subtypes. For each sample, we calculated the percentage of each cell type, with all samples processed in the same batch and underwent the same protocols to ensure comparability. The results showed varying percentages of cell clusters associated with NCI, AD, DLB, and PDD (Additional file 2: Fig. S2). Unsupervised hierarchical clustering of samples (nuclei number > 200) based on cell type proportions revealed two major patterns (Fig. 1d). NCI and AD exhibited similar cellular compositions, suggesting that AD-related cellular alterations may be more subtle or cell type-specific in the prefrontal cortex. In contrast, PDD and DLB clustered more closely together, indicating a greater overlap in their cellular composition, which is consistent with their shared pathological and clinical features, such as α-Syn aggregation and neurodegenerative changes in the basal ganglia and cortex.

A detailed examination of cell type proportions across disease groups (Additional file 2: Fig. S3a) revealed distinct alterations in specific cell populations. Notably, microglial proportions were elevated across all dementia subtypes compared to NCI, reinforcing the role of microglia in neurodegeneration (Fig. 1e). This observation aligns with previous findings suggesting that microglial activation contributes to neuroinflammation and disease progression in neurodegenerative disorders. Additionally, excitatory neurons were reduced across all three dementia subtypes, with the most pronounced decrease observed in DLB, whereas inhibitory neurons were more abundant in DLB and PDD. Other cell types exhibited relatively minor changes in proportion. These findings highlight the distinct cellular landscape associated with each dementia subtype and provide insights into disease-specific alterations in brain cell populations.

In addition, we performed a more refined annotation of neuronal cells using the BICCN atlas along with additional reference datasets. This analysis identified 10 neuronal subclusters: four inhibitory neuron subclusters (VIP IN, SV2C IN, PVALB IN, and SST IN) and six excitatory neuron subclusters (L2/3 EN, L4 EN, L4/5 EN, L5/6 EN, L5/6-CC EN, and L6 EN). We subsequently examined the composition of these subclusters across different dementia subtypes (Additional file 2: Fig. S3h–r). NCI and AD displayed similar overall neuronal compositions, whereas PDD and DLB cases were more similar to each other. Specifically, NCI and AD exhibited a higher proportion of lower-layer excitatory neurons. Among inhibitory neurons, SST IN and SV2C IN were specifically enriched in PDD and DLB, suggesting subcluster-specific vulnerabilities in these dementia subtypes.

SnRNA-seq revealed greater transcriptomic similarities between DLB and PDD

We next performed differential expression analysis for each cell type, comparing each dementia subtype with NCI. To ensure the robustness and reliability of the identified DEGs, we applied two different DEG identification methods and implemented stringent filtering criteria, as described in the “Methods” section. This approach minimised potential false positives and ensured that only high-confidence DEGs were retained for further analysis. The number of identified DEGs varied across the dementia subtypes, with 48 DEGs detected in AD, the majority of which (37 DEGs, 77%) were associated with astrocytes and oligodendrocytes (Additional file 2: Fig. S4-5). In contrast, DLB and PDD exhibited substantially higher numbers of DEGs, with 684 and 478 DEGs, respectively, the majority of which were identified in inhibitory neurons (Fig. 2a,b). Next, we investigated the top 50 most frequently occurring DEGs among all comparisons to identify shared dysregulated genes and pathways associated with dementia. Twenty seven of the top 50 DEGs exhibited consistent fold change directions. For example, HSPA1A, a gene from the heat shock protein family, consistently showed higher expression in PDD and DLB, likely reflecting the stressful environment inherent in dementia. In contrast, the remaining 23 common DEGs demonstrated divergent fold change directions across different dementia subtypes or cell types, suggesting their multifaceted roles in disease pathology (Fig. 2c). Interestingly, we discovered significant overlapping DEGs between the microglial and oligodendrocyte clusters, suggesting a conserved functional crosstalk between the two cell types in dementia (Fig. 2c). Additionally, a comparison of AD-associated DEGs with two previous studies [52, 53] revealed a 65% overlap (Additional file 2: Fig. S6 and Additional file 1: Table S10), supporting the reproducibility and relevance of our findings.

Fig. 2.

Cell type-specific transcriptomic profiling demonstrated shared DEGs between microglia and oligodendrocytes across dementia subtypes. a Heatmap and bar plot displaying the number of upregulated (right panel) and downregulated (left panel) DEGs (log₂ FC > 1, FDR < 0.01) across the eight major cell types in AD, DLB, and PDD. The colour intensity within each cell type represents the abundance of DEGs identified in that specific cell type when comparing dementia subtypes to NCI. b Venn diagrams illustrating the overlap of DEGs shared among the three dementia subtypes compared to NCI, as well as the unique DEGs specific to each subtype. c Heatmap showing the log₂ FC of the most significantly upregulated (red) or downregulated (blue) DEGs shared across the three dementia subtypes for each cell type. d Heatmaps showing hierarchical clustering of samples based on the expression of DEGs across selected cell types when comparing AD, DLB, or PDD to NCI. Each column represents a DEG identified in a specific cell type, and each row corresponds to a sample. Only cell types with significant DEGs are shown. The clustering highlights the separation between disease and control samples based on cell type-specific transcriptional changes. Cell type annotations are indicated by the colour bar below each heatmap. e Clustering of shared pathway enrichments using the DEGs identified from each cell type in DLB and PDD through Gene Ontology (GO) term analysis

Clustering based on phenotype-related DEGs in each dementia subtype highlighted the segregation of disease groups according to their transcriptional profiles, demonstrating the accuracy and stability of the identified DEGs. This consistent clustering supports the robustness of our differential expression analysis in distinguishing dementia subtypes. However, we also observed some heterogeneity among individual samples. Notably, one DLB sample (DLB_6) exhibited a distinct oligodendrocyte signature, suggesting potential variability in oligodendrocyte-associated pathology within the DLB group. Similarly, one PDD sample (PDD_6) clustered more closely with the NCI group, indicating individual variation that may reflect differences in disease progression, molecular profile, or underlying pathology (Fig. 2d). These findings emphasise the complexity of dementia-related transcriptional changes and suggest that further investigation into inter-individual variability may provide deeper insights into disease mechanisms.

To better understand the functional implications of these transcriptional changes, we performed Gene Ontology (GO) enrichment analysis on DEGs across various cell types in PDD and DLB (Fig. 2e). Given that AD had only 48 DEGs, which is insufficient for robust analysis, it was excluded from this analysis. The analysis identified five main categories of enriched biological processes: neuronal development and growth, synaptic plasticity and cognition, synapse and cell adhesion, protein and lipid homeostasis, and cellular signalling and transport, highlighting key neurobiological pathways affected in PDD and DLB (Fig. 2e). When comparing astrocytes between PDD and DLB, we found that DLB astrocytes were enriched in fewer terms across these five categories, whereas PDD astrocytes were primarily enriched in protein homeostasis and cellular signalling and transport. For inhibitory neurons, both PDD and DLB were similarly enriched in terms related to neuronal development, cognition, synapse assembly and organisation, and regulation of membrane potential. For microglia, both DLB and PDD showed enrichment in synapse organisation, learning or memory, synaptic transmission, axonogenesis, and neuron projection development. Notably, DLB microglia were exceptionally enriched in pathways related to lipid localisation and transport, as well as receptor-mediated endocytosis, indicating disrupted lipid metabolism that may impair microglial functions such as migration, phagocytosis (e.g., myelin debris clearance), and intercellular communication within the CNS. Interestingly, the negative regulation of macrophage-derived foam cell differentiation was enriched in DLB oligodendrocytes but not in PDD. This suggests a dynamic interaction between oligodendrocytes and microglia in DLB, where oligodendrocytes may help control excessive inflammation by limiting foam cell formation, while microglia actively respond by clearing debris and maintaining lipid balance.

The substantial overlap in DEGs between DLB and PDD, reflecting their shared pathological features such as α-Syn aggregation, synaptic dysfunction, and neuroinflammation, prompted us to conduct two additional analyses to better understand their molecular commonalities and distinctions. First, we grouped DLB and PDD together under the classification of Lewy body dementia (LBD) and compared LBD with AD and NCI to identify common transcriptional changes associated with Lewy body pathology. Second, we compared DLB and PDD, enabling the identification of disease-specific DEGs that distinguish these two dementia subtypes.

The comparison between NCI, AD, and LBD (consisting of DLB and PDD) identified 460 DEGs, with the majority associated with inhibitory neurons, suggesting a role for altered inhibitory signalling in LBD (Fig. 3a). Interestingly, clustering based on these DEGs unexpectedly grouped AD with NCI, rather than distinguishing AD from the dementia subtypes. This finding suggests that the transcriptional landscape of AD shares certain similarities with NCI at the molecular level. In contrast, the transcriptional profile of LBD significantly differs from AD, indicating that the molecular mechanisms driving LBD are distinct from those of AD. This reflects differences in the underlying neuropathological processes between LBD and AD.

Fig. 3.

Transcriptional profiling revealed shared and distinct transcriptional signatures in DLB and PDD. a Heatmap illustrating transcriptional changes among NCI, AD, and LBD (consisting of DLB and PDD). b Heatmap comparing transcriptional changes between DLB and PDD. c Pathway enrichment analysis of microglial DEGs in PDD and DLB

The comparison between DLB and PDD identified 294 DEGs, with the majority being associated with microglia (162 DEGs, 55%), highlighting microglia as a key contributor to their pathophysiological distinctions (Fig. 3b). Pathway analysis (Fig. 3c) identified key processes, including synapse organisation, regulation of neuron projection development, and cell–cell adhesion via plasma membrane adhesion molecules, pointing to microglial involvement in synaptic maintenance and neuronal connectivity as a distinguishing factor between these two diseases. Furthermore, pathways related to ion transmembrane transport, regulation of membrane potential, and metal ion transport were significantly enriched, suggesting differences in microglia-mediated ion homeostasis and neuroimmune signalling in DLB compared to PDD.

Identification of a distinct state of microglia across dementia subtypes

Neurodegenerative diseases are characterised by progressive cognitive decline, accumulation of pathological protein aggregates, and neuroinflammation. A growing body of evidence highlights the interplay between microglia and oligodendrocytes in these diseases, particularly in relation to myelin integrity, neuroinflammation, and synaptic dysfunction [54–56]. Upon re-clustering all microglia and oligodendrocytes, we identified a distinct state of microglia characterised by the simultaneous expression of microglial markers (P2RY12, PTPRC, APOE, and CD74) and oligodendrocyte markers (PLP1, MBP, and ST18) (Fig. 4a–c, Additional file 2: Fig. S7a–b). The distinct microglia was more enriched in AD, DLB, and PDD than in NCI, with a notably higher proportion in DLB (Fig. 4d). Furthermore, similar results were obtained when analysing data from a previously published snRNA-seq study using the same parameters [59] (Additional file 2: Fig. S8).

Fig. 4.

A distinct microglial state was enriched across dementia subtypes. a UMAP projection illustrating the re-clustered microglial and oligodendrocyte populations. The microglial cluster is highlighted in red, the oligodendrocyte cluster in orange, and the distinct microglia in blue. b Diffusion maps and density plots based on the signature scores per cell for microglial and oligodendrocyte clusters. c Violin plots showing the expression of oligodendrocyte markers (orange), microglial markers (red), and co-expressed markers indicative of distinct microglia (blue). d Box plot quantifying the percentage of the distinct microglia across dementia subtypes. e,f Bar plots demonstrating the count and weight of interactions between microglial and oligodendrocyte clusters across dementia subtypes. g Pathway analysis of upregulated DEGs in DLB microglia compared to NCI microglia. h Violin plot showing significantly elevated MSR1 expression in microglia from individuals with AD, DLB, and PDD compared to those from NCI. i Violin plots displaying increased HSPA1A expression in oligodendrocytes from AD, DLB, and PDD compared to those from NCI

Subsequently, we conducted cell-to-cell interaction analysis to examine the potential interactions between microglia and oligodendrocytes across various dementia subtypes. Our findings revealed an increase in both the predicted number of interaction pairs and the interaction strength in all dementia subtypes compared to NCI, suggesting a potential enhancement of microglial–oligodendroglial communication in dementia (Fig. 4e,f). To validate the interaction between microglia and oligodendrocytes, we conducted RNAscope multiplex in situ hybridisation on human cortical tissues from PDD and NCI, using microglial markers (ITGAM and PTPRC), oligodendrocyte markers (MBP and PLP1), and negative control probes (Additional file 2: Fig.S9), to assess the spatial relationship between the two cell types. As shown in Fig. 5, in both the NCI and PDD, proximal distribution between microglia (indicated by the colocalisation of ITGAM and PTPRC, circled in dotted lines) and oligodendrocytes (indicated by colocalisation of MBP and PLP1, circled in solid lines) was observed in several regions of both white matter (WM) and grey matter (GM). Notably, this distribution pattern was more frequent and appeared to occur at closer proximity in PDD compared to NCI, suggesting an elevation of potential interactions between these two cell types in dementia.

Fig. 5.

RNAscope supported potential interactions between microglia and oligodendrocytes in PDD. Representative RNAscope images showing the expression of microglial markers (PTPRC, green; ITGAM, red), oligodendrocyte markers (MBP, cyan; PLP1, grey), and nuclei (DAPI, blue) in the white matter and grey matter of human cortical tissues from NCI and PDD cases. ITGAM colocalised with PTPRC, and PLP1 colocalised with MBP. Higher-magnification images illustrate that microglia (nuclei circled with dotted lines) and oligodendrocytes (nuclei circled with solid lines) were in closer proximity, suggesting potential interactions. Dashed lines in the overview panels indicate the boundary between white matter (W) and grey matter (G). Scale bars: left panel, 50 μm; right panel, 20 μm

To further explore this issue, we identified DEGs between microglia in dementia subtypes and those in NCI and conducted an enrichment analysis. Our analysis revealed that these upregulated DEGs were enriched in five main groups: cell differentiation, migration and localisation; synapse development and maturation; lipid metabolism and transport; and synapse plasticity and cognitive function; and antigen processing and presentation (Fig. 4g). Notably, macrophage scavenger receptor 1 (MSR1) stood out as one of the most significantly upregulated DEGs in microglia across different dementia subtypes. Compared to NCI, MSR1 was highly expressed in microglia from AD, with a more pronounced increase observed in microglia from DLB and PDD (Fig. 4h). In addition, we detected elevated expression of heat shock 70 kDa protein 1 A (HSPA1A), encoding an interacting protein of MSR1, in oligodendrocytes from all three dementia subtypes (Fig. 4i, Additional file 2: Fig. S7c–d, reinforcing a potential functional correlation between microglia and oligodendrocytes. This observation further supports the findings on microglial–oligodendroglial interactions in dementia and suggests the involvement of microglial MSR1 in these processes.

Microglial MSR1 contributes to myelin phagocytosis in dementia

To validate the putative changes in MSR1 expression, we assessed its levels in cortical tissues from postmortem human brains and LRRK2-G2019S PD mice. Our findings revealed a significant upregulation of MSR1 expression in human AD, DLB, and PDD compared to NCI (Fig. 6a). Furthermore, both protein levels and mRNA expression of MSR1 were found to be significantly elevated in PD mice compared to NTg (Fig. 6b, e). To confirm that MSR1 expression was specifically upregulated in microglia, pure primary microglia (Additional file 2: Fig. S10) were isolated from mouse cortices for gene expression analysis. Consistently, MSR1 mRNA expression was significantly higher in cortical microglia from PD mice compared to NTg mice (Fig. 6c). Concomitantly, the expression of myelin-associated genes, including MBP (a marker for oligodendrocyte differentiation and myelination), MOBP (a marker for structural maintenance of the myelin sheath), and PLP1 (a major protein constituent of CNS myelin), was significantly upregulated in cortical microglia from PD mice (Fig. 6c), indicating myelin enrichment in microglia during PD pathology. Given MSR1’s established role in myelin phagocytosis in MS pathology [45], we postulated that the enriched myelin transcripts in the distinct microglia may be derived from phagocytosed RNA remnants within myelin debris [54–56], implicating that myelin loss could be a shared pathological feature in dementia. Supporting this, RT-qPCR analysis of myelin-associated markers showed a significant decrease in the expression of MOBP, MAG (a marker for long-term axon survival and glial–axon interaction), and MBP across AD, DLB, and PDD compared to NCI (Fig. 6d). Similarly, reduced expression of MOBP, MOG (a differentiation marker for oligodendrocyte maturation), and MBP was observed in PD mice compared to NTg mice (Fig. 6f). Together, these findings support our hypothesis regarding the perturbation of myelination in dementia pathology.

Fig. 6.

Elevated expression of MSR1 and myelin-associated genes in microglia revealed myelin dysregulation in dementia. a MSR1 mRNA expression in postmortem prefrontal cortex from 20 individuals with NCI, AD, DLB, and PDD. b Representative western blot and quantification of MSR1 protein levels in cortical tissues from LRRK2-G2019S PD mice and non-transgenic (NTg) mice. c mRNA expression of MSR1, MBP, MOBP, and PLP1 in CD11b-positive primary microglia isolated from PD and NTg mice. d mRNA expression of MOBP, MAG, and MBP in postmortem prefrontal cortex from NCI, AD, DLB, and PDD individuals. e,f mRNA expression of MSR1, MOBP, MOG, and MBP in cortical tissues from PD and NTg mice. Data are presented as mean ± SEM, 3 mice per group, aged 15–17 months. *p < 0.05, **p < 0.01, and ***p < 0.001 by Student’s t test for two-group comparisons or one-way ANOVA with Tukey’s post hoc test for multiple comparisons

Collectively, these findings led us to hypothesise that the distinct microglia may constitute a population of microglia actively involved in myelin phagocytosis, a process potentially driven by MSR1 in dementia pathogenesis. To experimentally test this hypothesis, we first examined the spatial relationship between MSR1-positive microglia and myelin in cortical brain tissues. We performed immunostaining using antibodies against IBA1, MBP, and MSR1 on cortical brain tissues from NCI and PDD patients. As shown in Fig. 7a, MSR1-positive microglia were observed to intermingle and/or co-localise with MBP in both the WM and GM of the PDD cortex, while in NCI samples, MSR1-positive microglia were in closer proximity to myelin but did not overlap. These observations suggest that MSR1-positive microglia more actively phagocytose myelin debris in the WM and GM of PDD compared to NCI. The myelin debris, likely derived from degenerated myelin sheaths or oligodendrocyte dysfunction, may accumulate over time during the ageing and pathological processes, thereby triggering microglial activation [31]. To further evaluate the functional role of MSR1 in myelin phagocytosis, we employed MSR1-overexpressing microglia as a model to mimic the phagocytic behaviour of the distinct microglia. We transfected human microglial HMC3 cells to overexpress MSR1 and exposed them to purified myelin isolated from mouse brains. After 6 h of exposure to pHrodo-labelled myelin, internalised myelin proteins were detected in both control and MSR1-overexpressing cells (Fig. 7b). A trend toward increased myelin uptake was observed in the MSR1-overexpressing cells compared to controls (Fig. 7c), suggesting that MSR1 may promote myelin phagocytosis in microglia. Notably, as a highly sensitive pattern recognition receptor, microglial MSR1 has been implicated in recognising various extracellular brain components such as Aβ, α-Syn Pff, and myelin, thereby stimulating clearance processes [79, 87–89]. Here, we exposed HMC3 microglia to these components to identify the potential pathological triggers responsible for inducing MSR1 upregulation. A significant increase in MSR1 protein levels was observed specifically in response to myelin exposure (Fig. 7d), suggesting a myelin-specific phagocytic mechanism, whereby myelin stimulation enhances microglial capacity to clear myelin debris.

Fig. 7.

Microglial MSR1 contributes to myelin phagocytosis in dementia

Discussion

In this study, we conducted snRNA-seq using postmortem prefrontal cortex samples obtained from individuals with NCI, AD, DLB, and PDD. Our analysis revealed a substantial overlap in DEGs and pathways between DLB and PDD across various cell types. Further analysis led to the identification of a distinct microglial state, originating from the dynamic interaction between microglia and oligodendrocytes. This unique cluster was pervasively present across dementia subtypes compared to NCI, with the most pronounced presence observed in DLB. Our investigation highlighted MSR1, which encodes a class A macrophage scavenger receptor, as one of the most significantly upregulated DEGs within microglia across all dementia subtypes compared to NCI. Furthermore, we consistently observed increased expression of HSPA1A, encoding an MSR1-interacting protein, in oligodendrocytes across all dementia subtypes compared to NCI. We validated the snRNA-seq findings using RNAscope, immunohistochemistry, and other biochemical analyses in human tissues and cellular models. This study provides evidence for the function of microglial MSR1 in myelin phagocytosis and is the first to reveal that microglial MSR1 participates not only in the pathogenic processes of AD but also in other Lewy body diseases such as DLB and PDD.

To date, several studies have examined gene expression profiles in cortical tissues of individuals with neurodegenerative diseases such as AD, DLB, and PDD at the single-cell level. Previous transcriptomic studies have predominantly focused on comparing individual diseases with NCI or two different types of dementia in various cortical regions, including the prefrontal cortex, entorhinal cortex, and anterior cingulate cortex, using bulk-seq and/or snRNA-seq approaches [57–61]. Our study represents a pioneering effort in collectively profiling these dementia subtypes, revealing 854 cell type–specific DEGs across dementia subtypes (Fig. 2a). Notably, the similar differential transcriptional profiles observed between DLB and PDD across various cell types may contribute to their shared neuropathological and clinical characteristics, such as motor symptoms resulting from Lewy body aggregation. Beyond the diverse transcriptional responses of different cell types to pathological processes, this phenomenon may also stem from shared mechanisms among dementia subtypes, particularly involving dysregulation of microglia and oligodendrocytes. Upon deeper examination, potential interactions between microglia and oligodendrocytes were uncovered through reclustering of these cell types (Fig. 4e,f). This was further supported by RNAscope, which showed that microglia and oligodendrocytes were in closer proximity in PDD compared to NCI. These findings imply the involvement of microglial–oligodendroglial crosstalk both functionally and spatially in the pathophysiology of dementia. Notably, the importance of this interplay has been increasingly recognised in the pathogenesis of various neurodegenerative disorders, especially in the context of demyelinating diseases [62, 63, 99, 100]. Within this dynamic interaction, microglia play a dual role in regulating oligodendrocytes. These cells secrete a plethora of cytokines and growth factors that can either damage oligodendrocytes or promote their repair and regeneration, thereby influencing myelin formation and maintenance during development and disease progression. In response, injury or dysfunction of oligodendrocytes triggers the release of immune mediators that recruit and activate microglia to phagocytose damaged cells and debris. Complement-activated oligodendrocytes, along with activated microglia, have been detected in the cortical WM of patients with AD and PD [64–66]. These cellular interactions likely contribute to broader structural changes observed in neurodegeneration. WM degeneration in AD is characterised by the expansion of oligodendrocyte lineage cells and free radical injury to myelin and axons [67]. Emerging evidence indicates that both ageing brains and those with AD show WM deterioration accompanied by increased microglial activation, with volume loss and neuroinflammation more pronounced in WM than in GM [68]. Furthermore, although much research has focused on WM, it is notable that its damage correlates with GM atrophy [69]. While increased microglial activation is associated with relatively preserved GM in patients with mild cognitive impairment [70], it coincides with GM atrophy in AD [71]. Similarly, PDD patients exhibit greater GM atrophy than PD patients, suggesting that GM volume loss is predictive of cognitive decline in PD [72]. However, the implications of GM alterations in Lewy body pathology (i.e., PDD and DLB) remain poorly understood. Building on these observations, our findings provide novel insights into microglial–oligodendroglial interplay in LBD, advancing our understanding of both WM and GM pathology across dementia subtypes.

In the CNS, myelin functions as a multilayered membrane protective sheath that encases axons and is crucial for sustaining axonal structural integrity and survival while facilitating nerve impulse propagation [73]. Myelin is synthesised by mature oligodendrocytes, which extend processes to wrap around nerve fibres and form compact myelin layers under the regulatory influence of microglia, astrocytes, and OPCs [74]. Its significance in axonal function has been implicated in various neurological diseases, including AD, MS, PD, Huntington’s disease, and amyotrophic lateral sclerosis. Myelin loss has been detected in individuals with mild cognitive impairment, AD, and vascular dementia [75], as well as in ageing. Recent studies have linked reduced myelin content to accelerated cognitive decline, prompting further investigation into its significance in dementia subtypes [76–78]. As a crucial step in microglial–oligodendroglial communication, myelin phagocytosis by microglia occurs at sites of myelin damage and plays a critical role in effectively restoring myelin integrity by clearing toxic myelin debris and supporting oligodendrocyte function [101, 102].

In our study, the newly identified microglia state may represent a specialised activated state of microglia tasked with engulfing degenerated or damaged myelin or oligodendrocytes (Fig. 4a–b). This was corroborated by our validation results, demonstrating myelin-related defects in AD, DLB, and PDD patients (Fig. 6d), as well as in PD mice (Fig. 6f), suggesting a potential association with the shared cognitive symptoms across these dementia subtypes. The identification of this distinct microglia state is supported by evidence from 5XFAD mice, which exhibit loss of function of the myelin architectural proteins CNP and PLP, leading to demyelination [56]. Additionally, it was demonstrated that compromised myelin integrity could contribute to amyloid deposition by engaging more microglia in myelin phagocytosis, thereby diverting their attention from Aβ plaques [56]. Our findings not only provide further evidence of its existence in the context of AD patients but also extend this feature to PDD and DLB, indicating myelin dysfunction and its potential influence on toxic protein deposition in LBD pathology. However, we should note that excessive or inappropriate uptake of intact myelin may lead to pro-inflammatory polarisation and demyelination, as observed in multiple sclerosis [79]. Moreover, the phagocytic capacity of microglia may become overwhelmed by myelin debris, resulting in reduced efficiency and responsiveness to subsequent insults [103, 104]. Therefore, it is crucial to comprehend the cellular and molecular intricacies of these interactions, which may exert both neuroprotective and detrimental effects during damage and repair processes.

A higher number of distinct microglia in DLB compared to AD and PDD may reflect distinct cellular and pathophysiological mechanisms, potentially driven by the co-occurrence of AD and PD pathologies. The synergistic effect of α-Syn, p-Tau, and amyloid beta (Aβ) pathologies creates a destructive feed-forward loop that accelerates neurodegeneration [80, 81], resulting in a more rapid and severe progression compared to AD or PD pathology alone. Furthermore, the co-occurrence of these protein aggregates has been shown to trigger both innate and adaptive immune responses, leading to microglia activation in DLB and AD [82, 83]. The aggregation of Aβ into neuritic plaques has also been shown to induce myelin lipid loss and myelin degeneration [84]. More severe neurodegeneration, heightened immune responses, and increased microglial activity could contribute to myelin and axonal damage [85]. This could explain the elevated presence of distinct microglia in DLB, where the need for myelin clearance is more pronounced. Additionally, pathways related to lipid localisation and transport, as well as receptor-mediated endocytosis, were enriched in DLB but not PDD microglia (Fig. 2e), indicating disturbed lipid metabolism in DLB microglia. This disruption adversely affects myelin homeostasis and impairs remyelination [86]. Inefficient clearance or the accumulation of lipid-rich debris could result in an increase in the distinct microglia as the microglia attempt to clear the accumulating myelin debris.

Interestingly, upon investigation of DEGs in the microglial cluster, we discovered a significant upregulation of MSR1 across all dementia subtypes (Fig. 4h). MSR1, predominantly expressed by microglia in the CNS, serves as a primary phagocytic receptor for the clearance of Aβ in AD [87, 88], as well as myelin debris in multiple sclerosis (MS) [79] and spinal cord injury (SCI) [89]. Moreover, emerging independent studies have reported increased MSR1 expression in microglia and reduced myelin content in AD [78, 87, 90], which is consistent with our observations in PDD patients and PD mice compared to controls (Fig. 6). The reduced expression of myelin transcripts may result from neuronal loss and the subsequent decline in myelination demand in degenerating tissues. Another possible explanation for the decrease in myelin transcripts could be oligodendrocyte dysregulation and impaired axon–glial junctions [91]. In mouse models with loss-of-function mutations in myelin genes such as Cnp [91] and Plp1 [92], altered oligodendrocyte gene expression was observed even before the onset of neurodegeneration. However, the role of MSR1-mediated myelin phagocytosis in dementia subtypes, particularly in DLB and PDD, remains poorly understood. Our findings contribute to filling this knowledge gap and revealed that MSR1-positive microglia actively phagocytose myelin debris in both WM and GM of PDD cortex compared to NCI (Fig. 7a). Additionally, the positive regulation of microglial MSR1 in phagocytic activity toward myelin in vitro (Fig. 7b,c) further supports this finding. Notably, in an AD mouse model coupled with myelin damage, a large cluster of microglia was observed to prioritise the clearance of myelin debris over Aβ plaque [56]. However, how myelin dysfunction impacts microglial selective clearance of myelin, Aβ, and α-Syn deposits in dementia remains largely unexplored. We revealed distinct microglial response patterns to myelin, as evidenced by an elevated protein level of MSR1 (Fig. 7d), highlighting the context-dependent nature of microglial activation. These observations further underscore the specific role of myelin in modulating MSR1 function, implicating it as an upstream factor in dementia pathogenesis. The upregulation of MSR1 caused by exposure to myelin debris might lead to a subsequent increase in phagocytic capacity toward myelin. This myelin–MSR1 positive feedback loop may underlie a shift in microglial phagocytic priority away from Aβ plaques. Further experiments are required to verify whether the distinct microglia is part of a response to myelin dysfunction or if they serve as drivers of myelin loss by impairing oligodendrocyte function. Collectively, these novel findings suggest that myelin deficiency may represent a pathological hallmark of DLB and PDD, potentially attributable to microglial MSR1-mediated myelin phagocytosis.

Another intriguing discovery is the upregulation of MSR1-associated genes in oligodendrocytes. HSPA1A, an isoform of HSP70, has been noted for its role in regulating MSR1 functions such as internalisation, cell-surface expression, and signal transduction in atherosclerosis [106]. HSPA1A upregulation has been reported in multiple brain regions, including the prefrontal cortex of AD patients, suggesting its potential as a biomarker linked to cognitive decline [58, 93–96]. Additionally, there is evidence of its extracellular release functioning as an endogenous ligand for pattern recognition receptors in other diseases [105]. The elevated expression of HSPA1A in oligodendrocyte clusters suggests a potential cooperative interaction with microglia (Fig. 4i), further reinforcing the occurrence of their crosstalk in dementia patients. Elucidating how this crosstalk influences ubiquitination and deubiquitination may help decipher the mechanisms underlying toxic protein degradation, thereby aiding in the development of novel therapeutics targeting protein misfolding in dementia.

Conclusions

In summary, we profiled the transcriptomic landscapes of postmortem prefrontal cortex tissues from individuals with dementia subtypes, including AD, DLB, and PDD. Our study highlights microglial–oligodendroglial interactions, the presence of a unique microglial state and increased microglial MSR1-mediated myelin phagocytosis across dementia subtypes. Future investigations are warranted to elucidate the functional role of the distinct microglia in demyelination in dementia.

Abbreviations

Aβ

Amyloid beta

AD

Alzheimer’s disease

APOE

Apolipoprotein E

α-Syn

Alpha-synuclein

MAG

Myelin-associated glycoprotein

MBP

Myelin basic protein

MOBP

Myelin-associated oligodendrocyte basic protein

MOG

Myelin oligodendrocyte glycoprotein

CNS

Central nervous system

DEGs

Differentially expressed genes

DLB

Dementia with Lewy bodies

GBA

Glucosidase Beta Acid

GM

Grey matter

HSPA1A

Heat shock 70 kDa protein 1A

LBD

Lewy body dementia

LRRK2

Leucine-rich repeat kinase 2

LNs

Lewy neurites

MSR1

Macrophage scavenger receptor

NCI

Noncognitive impairment

NTg

Non-transgenic

PDD

Parkinson’s disease dementia

Pff

Preformed fibrils

PLP1

Proteolipid protein 1

RT-qPCR

Reverse transcription-quantitative polymerase chain reaction

UMAP

Uniform manifold approximation and projection

SNCA

Synuclein Alpha

SnRNA-seq

Single-nucleus RNA sequencing

WM

White matter

Authors’ contributions

ZL and CJ conceived and supervised the entire study. CSY conducted experiments, interpreted the data, and wrote the manuscript. LM performed the computational analyses of the snRNA-seq data and edited the manuscript. LZ carried out the validation experiments, analysed the data, and edited the manuscript. HT assisted with plasmid generation, improved the layout of figures, and provided valuable advice for the manuscript. JLWL optimised single nuclei isolation from postmortem tissues, and prepared snRNA-seq library. QL isolated mouse primary microglia. LJ assisted with data analysis. RR provided materials and technical expertise. SA provided slide scanner for RNAscope. QL, ASN, and EKT contributed to the critical revision of the manuscript. All the authors have read and approved the final manuscript.

Funding

This research was supported by a National Medical Research Council Open Fund-Large Collaborative Grant (LCG002–SPARK II, NMRC/OFLCG/002/2018), a Clinician Scientist Individual Research Grant (MOH-CIRG21nov-0001, CIRG19may0052), Open Fund-Individual Research Grant (OFIRG23jul-0075), MOH-STaR19nov-0002, A*STAR PEC21-H22P0M0003, Duke-NUS, and SingHealth AMC core funding, as well as the Singapore Ministry of Health’s NMRC under its Centre Grant Program (MOH-000988).

Data availability

The processed snRNA-seq data can be downloaded from Zenodo (https://zenodo.org/records/11107597) [97]. The codes for reproducing can be found at https://github.com/JinmiaoChenLab/NNI_paper [98].The raw data is available at the NCBI GEO under the accession number GSE303823 [109].

Declarations

Ethics approval and consent to participate

Our study was approved by the SingHealth Institutional Review Board Committee (CIRB Ref# 2013/656/1 and #2025–0095) for the use of postmortem tissues and was conformed to the principles of the Declaration of Helsinki. Informed written consent was approved by the UK Human Tissue Authority (#18/WA/0238).

All experimental procedures and animal care were approved by the Institutional Animal Care and Use Committee Animal Use Protocol (IACUC AUP #19113 and #25006) of NTU-LKCMedicine Animal Research Facility. All experiments were carried out in accordance with the approved guidelines and regulations.

Consent for publication

Not applicable.

Competing interests

The authors declare that they have no competing interests.