Mapping the peripheral immune landscape of Parkinson’s disease patients with single-cell sequencing

Abstract

Parkinson’s disease is most recognized for its impact on the CNS. However, recent breakthroughs underscore the crucial role of interactions between central and peripheral systems in Parkinson’s disease pathogenesis. The spotlight is now shifting as we explore beyond the CNS, discovering that peripheral changes such as inflammatory dysfunctions may predict the rate of disease progression and severity. Despite more than 200 years of research on Parkinson’s disease, reliable diagnostic or progression biomarkers and effective disease-modifying treatments are still lacking. Additionally, the cellular mechanisms that drive changes in immunity are largely unknown. Thus, understanding peripheral immune signatures could lead to earlier diagnosis and more effective treatments for Parkinson’s disease. Here, we sought to define the transcriptomic alterations of the complete peripheral immune cell compartment by single-cell RNA and T-cell-receptor sequencing with hopes of uncovering Parkinson’s disease signatures and potential peripheral blood biomarkers.

Following transcriptional profiling of 78 876 cells from 10 healthy controls and 14 Parkinson’s disease donors, we observed all expected major classes of immune cells; the myeloid (monocytes, dendritic cells) and lymphoid (T lymphocytes, B lymphocytes, natural killer) compartments were further analysed through bioinformatics re-clustering to obtain the final 38 cellular subtypes. Comparing immune cell subtypes and phenotypes between patients with Parkinson’s disease and healthy control subjects revealed notable features of Parkinson’s disease: (i) a significant shift of classical CD14⁺ monocytes towards an activated CD14⁺/CD83⁺ state; (ii) changes in lymphocyte subtype abundance, including a significant decrease in CD4⁺ naive and mucosal-associated invariant T-cell subtypes, along with an increase in CD56⁺ natural killer cells; (iii) the identification of several specific T-cell clones shared between multiple patients, suggesting the implication of common epitopes in Parkinson’s disease pathogenesis; and (iv) a notable increase in the expression of activation signature genes, including the AP-1 stress-response transcription factor complex, across all Parkinson’s disease cell types. This signal was not present in atypical parkinsonism patients with multiple system atrophy or progressive supranuclear palsy.

Overall, we present a comprehensive atlas of peripheral blood mononuclear cells from healthy and Parkinson’s disease donors which should serve as a tool to improve our understanding of the role the immune cell landscape plays in Parkinson’s disease pathogenesis.

Changes in the immune system have been linked to Parkinson’s disease. By analysing more than 78 000 immune cells, Moquin-Beaudry et al. identify shifts in immune cell subtype abundance and potential gene expression signatures linked to immune activation in patients. These findings may facilitate earlier diagnosis and treatment.

Introduction

Parkinson’s disease (PD) is a complex neurodegenerative disorder characterized by the progressive loss of dopaminergic neurons in the substantia nigra pars compacta, and the intracytoplasmic neuronal accumulation of the α-synuclein protein.^1-3 PD currently affects over 10 million people worldwide, with an increasing global prevalence estimated at 17 million by 2040.^4,5 Unfortunately, there is no biomarker for PD to date, and misdiagnosis occurs in up to 16% of cases.⁶ These misdiagnosed patients are deprived of necessary care, which impacts their disease prognosis.⁷ Moreover, there is no cure for PD and the available therapies only temporarily ease patients’ symptoms.⁸ Disease-modifying treatments are assumed to maximize benefits if introduced at an early stage of the disease, before the apparition of the motor symptoms.^9-11 Given the widespread peripheral changes observed early in PD, using easily accessible peripheral tissues for biomarker discovery is highly promising. Indeed, the involvement of multisystem inflammation in PD pathophysiology has been suspected for some time with the first report dating back to 1988 when McGeer et al.¹² noted microglial activation in the substantia nigra of PD post-mortem brains. The inflammation hypothesis is supported by the identification of pro-inflammatory cytokines in the brain and CSF of PD patients.^13,14 Subsequent studies have observed the infiltration of peripheral immune cells into the brain, believed to breach the blood–brain barrier (BBB) and amplify the microglial reaction.^15,16 These immune alterations are also present in the periphery, showing elevated levels of pro-inflammatory cytokines in serum and an altered blood immune cell profile.¹⁵ This connection between the central and peripheral immune system offers an opportunity to discover blood-based biomarkers for the diagnosis and prognosis of PD.

The rise of single-cell technologies has allowed for the characterization of complex systems at single-cell resolution, providing a novel unbiased way to identify gene signatures and functional alterations across the entire cellular landscape. Recently, multiple groups have applied such technologies to accelerate our understanding of PD pathogenesis.^17,18 However, no PD studies have interrogated the complete peripheral immune compartment in this fashion.^19,20 We believe that by studying these peripheral changes, we can enhance early diagnosis and help identify new therapeutic targets for PD.

In this study, we investigated the complete peripheral blood mononuclear cells (PBMC) compartment. These cells are ideal for investigation due to their pivotal roles in immune response, metabolism and intercellular communication, as well as their ease of collection from live subjects. By employing single-cell RNA (sc-RNAseq) and T-cell receptor (TCR) sequencing simultaneously, our aim was to uncover precise immune cell signatures unique to PD. Overall, our study provides a comprehensive atlas of PBMCs in both healthy individuals and PD patients. This resource enhances our understanding of the immune cell landscape in PD and may lead to novel insights into disease mechanisms and potential therapeutic targets.

Materials and methods

PBMCs isolation

Blood samples were obtained from 14 patients with sporadic PD, three with multiple system atrophy (MSA) and three with progressive supranuclear palsy (PSP), as well as 10 age- and sex-matched healthy donors after informed consent in accordance with the institutional Ethics Committee Protocol #20.367. Blood was recovered in ethylenediaminetetraacetic acid (EDTA) Vacutainer tubes and processed at room temperature within 4 h of harvest. Blood was diluted with PBS-2 mM EDTA in 50 ml tubes, deposited over 12 ml of Ficoll-Paque (GE Healthcare) and centrifuged as per the manufacturer’s instructions. The buffy coat was harvested, washed with PBS-2 mM EDTA and aliquoted for long-term storage in liquid nitrogen in a freezing solution containing 10% dimethyl sulfoxide (Sigma-Aldrich) in fetal bovine serum (FBS; Gibco).

Single-cell processing and sequencing

Cells were distributed evenly between 5-plex batches and were processed on ice and centrifugations operated at 4°C when possible. The 5-plex mixes were processed using 10x Genomics 5ʹ v.1.1 kit, overloading the Chromium Next GEM Chip G Lane with 7.2 × 10⁴ cells for an expected yield of 3 × 10⁴ recovered cells/mix. Gene expression (GEX), variable-diversity-joining (VDJ), and feature barcoding libraries were generated using the manufacturer’s protocol (Supplementary material, ‘Methods’ section).

Libraries were sequenced on a NovaSeq 6000 S4 (Illumina) using the 10x recommended read parameters for 5ʹ v.1.1 libraries: paired-end, single indexing, Read 1: 26 bp; Read 2: 91 bp; i7: 8 bp. Libraries were multiplexed with a targeted sequencing depth of 50 000 reads/cell for GEX libraries and 5000 reads/cell for VDJ and feature barcoding libraries (CHU de Québec-Université Laval’s genomic center).

Data preprocessing

Count matrices were generated using Cellranger v.6.0.1 (10x Genomics) with the GRCh38 genome.²¹ GEX and feature barcoding libraries were processed using the ‘count’ function and VDJ libraries with the ‘vdj’ function.

Single-cell analysis tools

Analysis was performed in R (v.4.3.1) mainly using the single-cell analysis package Seurat v.4.3.0.1. Some figures were generated using ggplot2 v.3.4.2, scCustomize v.1.1.3, dittoSeq v.1.12.2 and ggpubr v.0.6.0.^22-26 Other R packages were employed as described in the relevant methods section.

Sample demultiplexing

Demultiplexing was done by associating hashtag oligonucleotides (HTO) to genotypes. HTO assignments were obtained from antibody feature barcoding using Seurat’s ‘HTODemux’ function, while the genotype demultiplexing tool souporcell v.2.0.40 was used to obtain genotype assignments.^27,28 Once genotypes were attributed to each donor, genotype demultiplexing was used to attribute cells to each donor due to a lower dropout rate than the HTO method. Doublets were filtered out (Supplementary material, ‘Methods’ section).

Initial quality control and clustering

Following standard cell quality control, cells with high or low feature count (<700 and > 5000) and high mitochondrial content (>10%) were filtered out. Principal component analysis and subsequent uniform manifold approximation and projection (UMAP) calculations were done using previously filtered integration features, to which TCR (^TR[ABDG][VDJ]*) and Immunoglobulin (^IG[KHL]V*) genes were additionally filtered out. Jackstraw plots were employed to assess the significance of principal components, aiding in the determination of the appropriate number of dimensions to use for clustering.²⁹

Cell annotation and filtering

A first automatic cell annotation was performed using Seurat’s reference mapping function to project our query dataset on the reference dataset’s UMAP projection. The reference dataset used in this case was the Human PBMC (162 000 cells with 228 antibodies) from Hao et al.²⁴ A second automatic cell type annotation procedure was run using SciBetR v.0.1.0 with the coronavirus disease 2019 (COVID-19) PBMC Ncl-Cambridge-UCL reference dataset from Stephenson et al.^30,31 Cell type annotations were further manually curated (Supplementary material, ‘Methods’ section).

Differential gene expression

Differential expression analysis was performed using Seurat’s ‘FindMarker’ function with standard parameters on library size normalized unique molecular identifiers (UMI) counts to identify differentially expressed genes (DEGs) between conditions within broad cell types. DEGs were defined as having an adjusted P < 0.05. Gene module score was calculated by Seurat’s ‘AddModuleScore’ function with the genes of interest.

Population composition analysis

Determination of differential cell population abundance between conditions was computed using the differential composition analysis transformed by a similarity matrix (DCATS) tool, a beta-binomial generalized linear statistical model used to compare annotated clusters.³² MiloR, which constructs a k-nearest neighbour graph for differential abundance testing of the various clusters, was also applied to validate differences between PD and control (CTRL) samples (Supplementary Fig. 1A and B).³³

TCR sequencing and analysis

Output from CellRanger’s ‘vdj’ function with reference = refdata-celranger-vdj-GRCh38-alts-ensembl-5.0.0 was analysed using scRepertoire v.1.3.5.³⁴ Clonotypes were defined as cells sharing the same VDJ-C genes and complementarity-determining region 3 (CDR3) nucleotide for TCRs α and β. To identify common antigen-specific TCRs between donors based on CDR3 sequence similarity, βCDR3 sequences were clustered using the GIANA tool: Geometry Isometry based TCR AligNment Algorithm v.4.1.³⁵ Cluster sequence visualization was generated using ggseqlogo v.0.1³⁶ (Supplementary material, ‘Methods’ section).

Single-cell trajectory analysis

Relevant clusters were subset from the Seurat object and directly transformed into a ‘cell_data_set’ object using SeuratWrapper v.0.3.1.³⁷ Subsequent processing was performed with Monocle3 v.1.3.3.^38,39 Cells were partitioned and re-clustered through the UMAP while the defined resolution was chosen to be comparable to previous clustering. Gene pseudotime trajectory inference was calculated using classical CD14⁺ monocytes as the starting state. Subsequent annotations used gene modules that most strongly correlated with each cluster while significantly driving trajectory signals (Q < 0.05).

Clinical assessments

All patients underwent a clinical assessment on the day of recruitment. Several clinical scales were extracted from the medical records to conduct correlation analyses between clinical and biological data (Supplementary Table 1). The medical evaluation was based on the MDS-UPDRS (Movement Disorder Society Unified Parkinson’s Disease Rating Scale).⁴⁰ These assessments included motor UPDRS, sleep UPDRS, ADL (Activities of Daily Living), total UPDRS and BREF (Batterie rapide d’efficience frontale) scores (Supplementary material, ‘Methods’ section).

Results

Comprehensive single-cell analysis of PBMCs in PD

Gaining detailed insight into PBMC subpopulations is challenging due to their heterogeneous composition, dynamic nature and diverse functional roles in the immune response. To identify cellular or molecular biomarkers for tracking and diagnosing PD, we generated a comprehensive high-quality sc-RNAseq dataset of PBMCs from 14 PD patients and 10 age- and sex-matched CTRLs (Fig. 1A and Supplementary Table 1). Quality consideration included removal of dead cells,⁴¹ as well as a cell-per-patient ratio that was maximized while keeping a high sequencing depth. Sample multiplexing was employed to maximize efficiency, and the samples from a single batch were demultiplexed using hashtag oligo-bound antibodies and genotypic computational approaches. Data integration was performed using Seurat to mitigate potential batch effects from sample processing (see the ‘Materials and methods’ section). Following doublet removal and cell quality control, we obtained 78 876 cells, with a median of 4349 UMIs and 1606 features per cell. TCR data was also obtained for 57 745 cells.

Figure 1.

Comprehensive single-cell analysis of peripheral immune cells in PD. (A) Schematic of sample origin and processing. Cryopreserved peripheral blood mononuclear cells (PBMCs) from 14 Parkinson’s disease (PD) patients and 10 healthy controls (CTRL) were prepared, multiplexed and processed using 10x Genomics 5′ v.1.1 chemistry to obtain gene expression (GEX) and T-cell receptor libraries. Created in BioRender. Tetreault, M. (2025) https://BioRender.com/k11f569. (B and C) UMAP projection of 78 876 cells using 1606 features per cell after removing ribosomal, mitochondrial, T-cell repertoire and immunoglobulin genes. Cells are coloured by disease status (B) and broad cell type (C). (D) Dot plot of the top three differentially expressed genes (DEGs) used to define PBMC broad clusters: T cells, B cells, natural killer (NK) cells, monocytes, dendritic cells, other myeloid cells and haematopoietic stem cells (HSC). Clusters were annotated with SciBetR as well as manually curated. (E) Bar plot showing the cell abundance of broad clusters in PD and CTRL groups. Differential cell population abundance was computed using a beta-binomial generalized linear model. No significant percentage differences were observed: B cells (P = 0.23), dendritic cells (P = 0.21), monocytes (P = 0.16), NK cells (P = 0.45), T cells (P = 0.09). Error bars represent standard error of the mean. (F) DEGs in myeloid clusters: 99 in monocytes, 46 in dendritic cells, with 31 common genes (including: CXCR4, ZFP36, RPS26, JUNB, FOS, CMEC12A, SAMHD1, FGL2, TYMP, RNF213). (G) DEGs in lymphoid clusters: 44 in B cells, 33 in NK cells, 65 in T cells, with 10 common genes (RPS26, JUNB, NFKBIA, RPS2, NR4A2, DUSP1, ZFP36, FOS, BTG1, ZFP36L2).

Cell annotation was carried out using Seurat’s reference mapping procedure with datasets from Hao et al.³⁷ and SciBetR v.0.1.0, which used the COVID-19 PBMC Ncl-Cambridge-UCL reference dataset by Stephenson et al.³⁰ Following this process, we observed comparable cell distributions between PD and CTRL (Fig. 1B). We identified seven broad cell types, including T cells, B cells, natural killer (NK) cells, monocytes, dendritic cells (DC), haematopoietic stem cells (HSC) and other myeloid cells (basophils, platelets), confirmed through manual curation based on the expression of classic marker genes (Fig. 1C and D). There were no significant differences in the overall proportions of the broad PBMC populations between the PD and CTRL groups. However, we observed trends showing lower percentages of T cells (P = 0.09) and B cells (P = 0.23), and higher percentages of monocytes (P = 0.16) and NK cells (P = 0.45) in the PD group (Fig. 1E). Our transcriptomic analysis between PD patients and CTRL revealed significant DEGs across various immune cell types. Specifically, we identified 99 DEGs in monocytes, 46 in DC, 44 in B cells, 33 in NK cells and 65 in T cells (Supplementary Fig. 1C). When comparing DEGs between myeloid cells (monocytes and DC), 31 common DEGs were identified. Notable increases were observed in genes such as CXCR4, ZFP36, RPS26, JUNB and FOS, while CLEC12A, SAMHD1, FGL2, TYMP and RNF213 showed decreased expression (Fig. 1F). Within the lymphoid cell compartment (T cells, B cells and NK cells), 10 common DEGs were identified, all of which were upregulated in PD compared with CTRL. These genes include RPS26, JUNB, NFKBIA, RPS2, NR4A2, DUSP1, ZFP36, FOS, BTG1 and ZFP36L2 (Fig. 1G).

This analysis revealed generally comparable proportions of the main immune cell types between PD patients and CTRL. However, it also identified interesting DEGs across all immune cell types, indicating distinct regulatory patterns within the immune system in PD.

Myeloid cells reveal a shift of classical monocytes to an activated phenotype in PD

Myeloid cells function primarily in innate immunity through phagocytosis, inflammatory response modulation, antigen presentation and tissue repair.⁴² Through sc-RNAseq profiling, we identified various subpopulations within the myeloid lineage, which were categorized into specific DC and monocyte subsets (Fig. 2A). Initial clustering was performed using SciBetR, which identified 10 distinct clusters, including two separate CD14⁺ monocyte clusters and two CD16⁺ non-classical monocyte clusters. Manual curation of the CD14⁺ clusters revealed the presence of the activation marker CD83 in one cluster, which we annotated as CD14⁺/CD83⁺ or activated monocytes.^43,44 Similarly, within the CD16⁺ clusters, we identified a cluster with high expression of the C1 complement genes, which we designated as CD16⁺/C1⁺ monocytes (Fig. 2B). These subsets also included AXL⁺ SIGLEC6⁺ DC (ASDC, a transitional state), plasmacytoid dendritic cells (pDC), DC1, DC2, DC3, proliferating DC (DC prolif) (Fig. 2A and B). Differential composition analysis using DCATS revealed significant differences between PD and CTRL groups. Specifically, we observed a significant decrease in the abundance of classical CD14⁺ monocytes (P = 0.03), accompanied by a concomitant increase of activated CD14⁺/CD83⁺ monocytes (P < 0.01) in PD patients (Fig. 2C), which was validated by miloR (Supplementary Fig. 1A and B). Associated DEGs were extracted to investigate related molecular mechanisms, complemented by analysis of differences between PD and CTRL expression across all clusters (Supplementary Fig. 1C and Supplementary Tables 2 and 3). The ASDC population also showed a significant increase in abundance (P = 0.03). However, the cell count was extremely low (<1% of all cells), making further interpretation difficult.

Figure 2.

Single-cell transcriptomic profiling of myeloid cells revealed a shift of CD14⁺ classical monocytes to activated CD14⁺/CD83⁺ in PD. (A) UMAP projection of 18 169 cells from the myeloid population [monocytes, dendritic cells (DC)]. Automated annotation followed by manual curation identified 10 subclusters; AXL⁺ SIGLEC6⁺ DC (ASDC), plasmacytoid DC (pDC), DC1, DC2, DC3, DC prolif, CD14⁺ monocytes, CD14⁺/CD83⁺ monocytes, CD16⁺ monocytes, CD16⁺/C1⁺ monocytes. (B) Dot plot of the top three markers used to define the myeloid subclusters. Manual curation allowed the identification of the CD14⁺/CD83⁺ monocytes subcluster which are classical monocytes expressing the CD83 activation marker. The non-classical monocytes expressing the protein complement C1 (CD16⁺/C1⁺) were also identified through the same method. (C) Bar plot showing the cell abundance of myeloid cluster’s subpopulations in Parkinson's disease (PD) and control (CTRL). A significant difference was observed in CD14⁺/CD83⁺ monocytes with a higher abundance in PD (P < 0.01), with a concomitant lower abundance of CD14⁺ classical monocytes (P = 0.03). A significant increase in abundance was also observed for ASDC (P = 0.03) in PD. No significant difference was observed for pDC (P = 0.24), DC1 (P = 0.54), DC2 (P = 0.65), DC3 (P = 0.18), DC prolif (P = 0.50), CD16⁺ monocytes (P = 0.82), CD16⁺/C1⁺ monocytes (P = 0.59). Data were analysed using DCATS beta-binomial generalized linear model. Error bars represent standard error of the mean. (D) Re-clustering of the myeloid cells based on the same UMAP generated in A using Monocle 3 pseudoclustering inference. A new cluster of cytotoxic monocytes was identified with this method. Numbers on the UMAP correspond to pseudotrajectory endpoints. (E) Pseudotime analysis to identify the differentiation trajectories of the myeloid cells showing the evolution from classical monocytes to inflammatory states. (F) Overview of the main modules driving the cell differentiation. The top three genes in the most significant modules are highlighted. (G) Gene expression trajectory plot of top monocyte module genes along pseudotime. DCATS = differential composition analysis transformed by a similarity matrix; UMAP = uniform manifold approximation and projection.

Using Monocle 3, we re-clustered the myeloid cells based on gene expression, considering biological processes and alternative cell fates as contributors to the latter (Fig. 2D). This approach immediately split CD14⁺/CD83⁺ monocytes into finer subtypes, including a new cluster characterized by B-cell signalling and a cluster defined by known cytotoxic markers.⁴⁵ This re-clustering also refined the cells specifically associated to an activated monocyte trajectory (Fig. 2D). While statistical analysis of cell percentages did not reveal that specific subclusters were significantly driving the previously observed differences in CD14⁺ monocytes between PD and CTRL (Fig. 2C), cell ratio trends support the idea that PD samples have less classical CD14⁺ monocytes (P = 0.05) while exhibiting a higher proportion of activated (P = 0.13) and cytotoxic monocytes (P = 0.13) (Supplementary Fig. 2).

Pseudotime analysis was conducted to map the differentiation trajectories of these cells and to identify contributors to the progression from classical monocytes to more inflammatory profiles (Fig. 2E). To deduce finer identities of CD14⁺/CD83⁺ monocytes, gene modules most associated with each refined cluster were manually compared with the literature. The genes indicated common roles in specific immune pathways, such as cytotoxic functions (GNLY, PRF1, SPON2) for module 47,⁴⁵ B lymphocyte functions (CD79A, MS4A1, BLK) for module 27 or cytokine response (ISG15, PSME2, LY6E) for module 38. The latter cells also exhibit rising CD16 expression, suggesting an intermediate status. Additionally, several genes correlated with a stronger activation signal (CXCL8, FOS, FOSB, S100A8, S100A12, VCAN) in the remaining monocyte clusters (Fig. 2F and Supplementary Table 4). Gene expression trajectory for top markers illustrates how the genes transition across pseudotime, highlighting the specificity to their respective clusters (Fig. 2G).

Together, this analysis of myeloid cells revealed a shift of CD14⁺ classical monocytes to activated CD14⁺/CD83⁺ monocytes with diverging fates in PD, and highlights the dynamic changes in cell states and their potential roles in the inflammatory environment observed in PD.

Transcriptomic profiling of lymphoid cell subpopulations

Lymphoid cells primarily function in adaptive immunity by producing antibodies, coordinating immune responses, direct cell killing through cytotoxic mechanisms and generating immunological memory against specific pathogens.⁴⁶ As with the myeloid lineage, we used SciBetR followed by meticulous manual curation of cell markers to identify and categorize 24 distinct clusters within the lymphoid lineage (Fig. 3A and Supplementary Fig. 3A). This approach allowed us to classify the three main lymphoid populations: T cells [encompassing CD4⁺, CD8⁺, natural killer T cells (NKT) and mucosal-associated invariant T (MAIT) cells], B cells and NK cells, along with their respective subclusters.

Figure 3.

Transcriptomic profiling of lymphoid cell subpopulations. (A) UMAP projection of 52 954 cells from the lymphoid population [T cells, B cells, natural killer (NK) cells]. Automated annotation followed by manual curation identified 24 subclusters [B immature, B naive, B non-switched memory, B switched memory, B exhausted, plasmablast, CD4⁺ naive, CD4⁺ central memory (CM), CD4⁺ effector memory (EM), CD4⁺ IL22, CD4⁺ Th1, CD4⁺ Th2, CD4⁺ proliferating cells (CD4⁺ prolif), CD8⁺ naive, CD8⁺ EM, CD8⁺ T effector (TE), CD8⁺ proliferating cells (CD8⁺ prolif), gamma delta T (gdT), mucosal-associated invariant T (MAIT), regulatory T cells (Treg), natural killer T cells (NKT), NK16hi, NK56hi, NK proliferating cells (NK prolif)]. (B) Bar plot showing the cell abundance of lymphoid cluster’s subpopulations in Parkinson's disease (PD) and control (CTRL). Subpopulations with a cell percentage below 1% relative to the lymphoid cluster were excluded from our analysis (B immature, B non-switched memory, B exhausted, plasmablast, CD4⁺ Th1, CD4⁺ Th2, CD4⁺ prolif, CD8⁺ prolif, gdT, Treg, NK prolif). A significant difference was observed in CD4⁺ naive (P = 0.02) and MAIT (P < 0.01) which were less abundant in PD. Conversely, PD patients exhibited a significant increase in NK56hi (P = 0.03) abundance. No significant difference was observed for the other clusters: B naive (P = 0.11), B switched memory (P = 0.36), CD4⁺ CM (P =0.74), CD4⁺ EM (P = 0.23), CD4⁺ IL22 (P = 0.89), CD8⁺ naive (P = 0.47), CD8⁺ EM (P = 0.14), CD8⁺ TE (P = 0.56), NKT (P = 0.15), NK16hi (P = 0.72). Data were analysed using DCATS beta-binomial generalized linear model. Error bars represent standard error of the mean. (C) UMAP projection of clonal expansion observed from T-cell receptor (TCR) sequencing. (D) Scatter plot of TCR sequence composition of clusters identified by GIANA v.4.1 depicting cluster size, in number of CDR3 sequences (point size), percentage of sequences from PD donors (x-axis), number of individual donors in cluster (y-axis) and percentage of sequences coming from PD donors (point fill colour). Jitter (geom_jitter) was added to y-axis values to minimize overlaps. Yellow dots represent clusters containing CDR3 sequences only found in PD donors. 1976 total clusters were identified. 364 clusters were PD specific, shared at least by two donors which correspond to 1094 sequences, when excluding non-unique sequences. (E) Pie charts depicting predicted CDR3 sequence antigen specificity in CTRL versus PD cells by using the McPAS TCR database. Left column represents CDR3 sequences with exact matches in the database. Right column represents a unique exact method. Values expressed in percentage of total CDR3 sequences. (F) Peptide sequence logos and cell type composition for the top 12 TCR clusters with the most abundant shared PD-specific CDR3 sequences. Cell types of origin are illustrated along with the cluster sequences for top hits. DCATS = differential composition analysis transformed by a similarity matrix; UMAP = uniform manifold approximation and projection.

Differential composition analysis revealed decreases in CD4⁺ naive T cells (P = 0.02) and MAIT cells (P < 0.01), as well as an enrichment of NK cells in PD, with a marked increase in the NK56hi (CD16^low/CD56⁺⁺) subpopulation (P = 0.03) (Fig. 3B and Supplementary Fig. 1A and B). This subpopulation is known for its heightened immunoregulatory activity compared with the cytotoxic NK16hi (CD16⁺/CD56^low), and its function suggests a potential role in the immune response alterations observed in PD.⁴⁷

This analysis of the lymphoid cell compartments revealed decreases in subtypes of T cells and increased NK cells. This increased presence of NK56hi cells in PD patients could indicate an enhanced state of immune activation or a compensatory mechanism in response to inflammation.

TCR sequencing

T cells play crucial roles in adaptive immunity by recognizing and killing infected cells, coordinating immune responses and regulating the activity of other immune cells.⁴⁸ VDJ recombination is a process of genetic rearrangement that occurs in developing T cells and B cells. This process generates the diverse repertoire of antigen receptors necessary for recognizing a vast array of pathogens and are essential for a naive T cell to undergo clonal expansion and differentiate into effector subsets. To explore the diversity and specificity of the T-cell population in PD patients, we performed single-cell sequencing of the variable regions of TCRs and plotted the distribution of TCR expansion categories on the UMAP (Fig. 3C).

Identical TCR sequences indicate T-cell clonal expansion patterns and lineages, which are pivotal for recognizing endogenous and exogenous antigens presented by the major histocompatibility complex.⁴⁹ In both PD and CTRL groups, the most expanded clones were predominantly found in the effector CD8⁺ T and NKT clusters, reflecting their high immune activity. However, the majority of clonotypes were detected within the CD4⁺ T-cell populations, owing to their higher abundance (Fig. 3C). Clustering of all CDR3 sequences based on similarity using the GIANA algorithm (see the ‘Materials and methods’ section) generated 1976 clusters (Fig. 3D). Of those, 364 clusters (1094 sequences) were unique to PD cells and shared at least by two donors.³⁵

We assessed if the identified PD-specific CDR3 sequences had predicted specificity for antigens associated to disorders in the McPAS-TCR database.⁵⁰ The ‘all exact’ methodology is a clonal-expansion sensitive approach querying all CDR3 sequences for exact matches in the database on the overall avidity of our repertoire. The ‘Unique exact’ results were obtained by removing duplicated sequences from a single donor, to mitigate the impact of single-patient clonal expansion (Fig. 3E). While exact repertoire coverage in the database was low at 0.86% and 1.43% for CTRL and PD, respectively, it showed a striking enrichment of autoimmune-annotated CDR3 sequences in PD cells (Fig. 3E). The ‘Unique exact’ profiles were similar between CTRL and PD groups (Fig. 3E), suggesting that the enrichment of the autoimmune signal came from clonally expanded sequences in specific PD patients. Detailed analysis revealed that the enrichment in autoimmune-annotated sequences in the ‘all exact’ is driven by a single CDR3 sequence (CASSRDSNQPQHF) with a predicted reactivity to psoriatic arthritis. It was found in three PD individuals, with a high clonotype expansion in only one individual.

We next focused on epitope prediction of PD-specific CDR3 clusters shared between multiple donors. Interestingly, we identified clones shared by several PD patients, including two clones shared by 42.86% (6/14), two other clones shared by 35.71% (5/14) and 10 clones shared by 28.57% (4/14) of the patients (Fig. 3F). While the epitopes recognized by these PD-specific CDR3 sequence clusters are currently unannotated, these results suggest common and shared PD-specific epitopes which could better inform us on PD pathogenesis.

These results reflect the diversity of CDR3 sequences in PD, spanning over 300 clusters shared among varying numbers of donors. More importantly, our results suggest a consistent immune mechanism among our PD patients, where a high percentage share common TCR clusters despite PD’s known heterogeneity.

Increased activation signature gene expression in PD cell clusters

Our comparative DEG analysis revealed that several genes from the AP-1 transcription factor family as well as activation signature genes (ASG), recently described as stress-response genes,^51,52 were increased in both the myeloid and lymphoid cell populations in PD patients (Fig. 1F and G). Using Seurat’s module scoring tool, we grouped the ASGs and visualized their expression levels (Fig. 4A). This revealed elevated ASG scores in PD versus CTRL across most cell types. To determine if the signal was an artefact of sample processing, we evaluated atypical parkinsonism (AP) participants (three MSA and three PSP) that were included at the beginning of our study (Fig. 4B). Intriguingly, their ASG scores were significantly lower than CTRL across the cell populations. Moreover, all the individual ASGs exhibited higher expression in PD compared with CTRL, MSA and PSP (Fig. 4C). In line with this, the cell abundances from MSA and PSP patients also showed changes that were either comparable to CTRL or inverse to PD (Supplementary Fig. 4A and B). To identify which cell types contributed most to the ASG signal in PD, we examined the signal on the feature plot and the highlighted corresponding cell populations. The ASG signal was most elevated in CD14⁺/CD83⁺ activated monocytes, NK56hi cells, CD8⁺ effector memory (EM) T cells and MAIT (Fig. 4D). For insights into the biological processes underlying the ASG signature, we performed pathway analysis on DEGs, focusing on differential cell types between CTRL and PD; activated monocytes and NK56hi (Fig. 4E and F and Supplementary Fig. 3B–E). A non-weighted Kyoto Encyclopedia of Genes and Genomes (KEGG) analysis revealed a significant enrichment of two important immune pathways, the interleukin (IL) IL-17 signalling pathway in activated monocytes (Fig. 4E), and the nuclear factor kappa B (NF-κB) signalling in both activated monocytes and in NK56hi (Fig. 4E and F). Furthermore, the NK56hi subpopulation exhibited enrichment in the tumor necrosis factor (TNF) signalling pathway, another key pro-inflammatory pathway (Fig. 4F). A non-weighted gene ontology (GO) term and weighted KEGG pathway analysis also returned immune-relevant hits such as positive regulation of migration for CD83⁺ monocytes, innate immune response activation for NK56hi and various neurodegenerative disorders for both cell populations (Supplementary Fig. 3B–E).

Figure 4.

Activation signature genes. (A) Feature plot of activation signature gene (ASG) module score of all cells in Parkinson's disease (PD) and control (CTRL), showing an increased expression across most cell types in PD. (B) Feature plot of ASG module score of all cells in progressive supranuclear palsy (PSP) and multiple system atrophy (MSA). The expression of ASG genes based on the module score of these two phenotypes is lower compared with PD. (C) Dot plot showing the expression level of each ASG gene, in all phenotypes. In PD donors, most ASG genes exhibit a log2 fold change ≥ 1, owing to their lower expression in CTRL, PSP and MSA. (D) UMAP representing the ASG module hotspot (red dot lines), which are, respectively, the CD14⁺/CD83⁺ monocytes, the NK56hi, the CD8⁺ effector memory (EM) and the mucosal-associated invariant T (MAIT). (E and F) Non-weighted KEGG analysis of DEGs showing the top 20 most significant pathways. (E) Activated CD14⁺/CD83⁺ monocytes show enrichment in immune-related pathways (e.g IL-17 signalling pathway, NF-κB signalling pathway), in metabolism (lipid and atherosclerosis) and in autoimmunity (rheumatoid arthritis). (F) NK56hi pathway analysis also revealed immune-related pathways (NF-κB signalling pathway, TNF signalling pathway, T-cell receptor signalling pathway). UMAP = uniform manifold approximation and projection.

Together, these data reveal consistent dysregulation of ASGs across both myeloid and lymphoid cells in PD, but not in AP, markedly in overrepresented populations such as activated monocytes (Fig. 2C) and NK56hi cells (Fig. 3B).

Integrated results from clinical and biological data

Comparison with clinical features gives essential insights into patient-derived datasets. To deepen our insights into how our biological findings and clinical features correlate in our cohort, we performed Pearson correlations on clinical features (see the ‘Materials and methods’ section) and cell percentages from our main biological findings; increased CD14⁺/CD83⁺ monocytes, decreased CD4⁺ naive T and MAIT cells, as well as increased NK56hi and ASG score in PD (Supplementary Fig. 5 and Supplementary Table 5).

Correlation analysis of clinical features with cell types that showed significant differential ratios between CTRL and PD found several significant correlations. A significant positive correlation (r = 0.8, P < 0.01) between ASG score and CD14⁺/CD83⁺ monocytes was observed, which aligns with the observation of high ASG signal in CD14⁺/CD83⁺ monocytes (Fig. 4A and B). We also found a significant negative correlation (r = −0.59, P = 0.04) between ASG score and sleep UPDRS score. In addition, being male was significantly positively correlated with higher scores in sleep UPDRS (r = 0.66, P = 0.02), ADL (r = 0.61, P = 0.03) and total UPDRS (r = 0.61, P = 0.03), indicating poorer sleep quality, reduced autonomy and greater overall impairment. No significant correlation was observed between MAIT, CD4 naive or NK56hi cell percentage and clinical features (Supplementary Fig. 5A). To compare these results, correlation testing was computed between age, sex and the ASG score, as well as the CD14⁺/CD83⁺ monocytes ratio in the CTRL group. The ASG score and the CD14⁺/CD83⁺ ratio was highly and positively correlated (r = 0.8, P < 0.01). Age was negatively correlated with both the ASG score and the CD14⁺/CD83⁺ monocytes ratio (r = −0.74, P = 0.01; r = −0.6, P = 0.04) (Supplementary Fig. 5B and Supplementary Table 5).

To assess if any more easily accessible PBMC ratios vary with clinical features, we correlated clinical features with broad PBMC percentages (Fig. 1). This analysis revealed a positive correlation between total UPDRS and B-cells percentage (r = 0.70, P = 0.01) (Supplementary Fig. 5C and Supplementary Table 5). Age did not correlate to any of the cell populations, while being male was positively correlated to monocytes (r = 0.57, P = 0.04) (Supplementary Fig. 5A). In addition, we observed a significant negative correlation between NK cells and DC (r = −0.65, P = 0.02) as well as between NK cells and T cells (r = −0.79, P < 0.01). Monocytes were positively correlated with HSC (r = 0.74, P < 0.01). Both B cells and DC showed positive correlations with symptom duration (r = 0.71, P < 0.01; r = 0.76, P < 0.01, respectively). To assess if similar correlations were seen in CTRL cells, we followed the same approach for CTRL cell populations and observed a positive correlation between age and monocytes (r = 0.74, P = 0.01), while age demonstrated a negative correlation with NK cells (r = −0.79, P = 0.01). From a sex-difference perspective, being female was correlated to the percentages of HSCs (r = −0.67, P = 0.03). Finally, on the cell population analysis, we observed a negative correlation between T cells and monocytes (r = −0.79, P < 0.01) (Supplementary Fig. 5D and Supplementary Table 5).

Discussion

From broad clusters to specific subpopulations

Single-cell transcriptomics was used in a funnel-down manner to analyse the complete PBMC compartments simultaneously, initially focusing on broad clusters and progressively honing in on specific populations.

While the broad analysis found no significant difference in relative cell abundance between PD and CTRL, PD samples trended towards an enrichment in monocytes and a reduction in lymphocytes (Fig. 1E). Although none of the previous sc-RNAseq publications encompassed the entire PBMC compartment in PD, Tian et al.⁵³ reported similar findings, showing significant increases only in the ratio of PD patients monocytes using flow cytometry analysis on total PBMC samples. Harnessing the power of single-cell analysis, we refined our annotations of myeloid and lymphoid compartments to deepen our examination of each cell subtype and identify the subcluster driving those differences.

An activated state of monocytes in PD

Marker-based identification of the monocyte population allowed us to identify a subcluster of CD14⁺ monocytes marked by the CD83 immune cell activation marker (Fig. 2B).⁴⁴ CD83⁺ monocytes DEGs include several chemokine signalling genes such as CCL3, CXCL8, IL1B and CXCR4 (Supplementary Table 2). CXCR4 overexpression has been previously described in the substantia nigra in PD and is associated with microglial activation.⁵⁴ Therefore, the observed CXCR4 overexpression in the periphery could contribute to the activation of peripheral immune cells, specifically monocytes. This potential involvement of CXCR4 in PD suggests that other related chemokines, some of which are differentially expressed in PD, might also be implicated.⁵⁵

Classical CD14⁺ monocytes can be triggered to express CD83 when treated with the cytokine interferon (IFN)-α. These activated monocytes exhibit phagocytic behaviour similar to CD14⁺ monocytes and do not adopt a dendritic morphology, distinguishing them from DCs.⁵⁶ IFN-α is a key cytokine in the response to viral infections, initiating the innate immune response. Under physiological conditions, lower levels of IFN-α/β are present without viral infections, resulting in a weak signal that primes cells for an amplified response during viral infections and increases their sensitivity to other cytokines.⁵⁷ Based on our findings, it is possible that PD patients may have elevated baseline levels of circulating IFN-α, which could lead to a higher abundance of activated monocytes. Furthermore, Cao et al.⁵⁸ demonstrated the presence of preformed CD83 within monocytes using western blotting and flow cytometry and showed that CD83 could be induced on the surface of monocytes following lipopolysaccharide (LPS) stimulation, with expression peaking 2–4 h post-stimulation and disappearing after 24 h. This indicates that CD83 is not constitutively expressed but can be rapidly upregulated in response to specific stimuli. These observations suggest that PD patients might exhibit heightened sensitivity to stimulation and stress compared with healthy controls, resulting in a higher expression of CD83 in CD14⁺ monocytes. Grozdanov et al.⁵⁹ analysed the monocyte population from PD patients using fluorescence-activated cell sorting (FACS) and RNA sequencing. FACS results showed an enrichment of classical CD14⁺ CD16⁻ monocytes. In this study, the researchers aimed to identify only classical, intermediate and non-classical monocytes, thus it is unknown if they had increased proportions of CD14⁺/CD83⁺ monocytes specifically. The CD14⁺/CD83⁺ monocytes may have been included in the classical monocyte cluster, and an abundance of these activated monocytes may have led to the enrichment of the CD14⁺ monocyte population identified by FACS. They also conducted LPS stimulation on the classical monocytes, which demonstrated a hyperactivation state through excessive production of pro-inflammatory cytokines compared with the control. Our finding of an activated state in monocytes may correlate with this hypersensitivity, suggesting a greater immune response capacity in PD patients.

Consistent with these findings, our non-weighted KEGG pathway analysis in the activated CD14⁺/CD83⁺ monocytes revealed enrichment in multiple immune-related processes including some previously linked to PD such as IL-17 and NF-κB (P < 0.05) (Fig. 4E).⁵⁹ The transcriptomic data from Grozdanov et al.⁵⁹ revealed a dysregulation of inflammatory pathways, including IL-17A signalling and regulation. IL-17 is a pro-inflammatory cytokine involved in chronic inflammation and present in the pathogenesis of autoimmune diseases such as rheumatoid arthritis.^60-63 In accordance with this, an increase of IL-17 producing cells as well as cytokine concentration in PD patient serum was reported by Yang et al.⁶⁴ A recent transcriptome study on whole blood found a similar upregulation in pathways associated with positive regulation of IL-17 in PD compared with CTRL.⁶⁵ Intriguingly, the pro-inflammatory NF-κB signalling pathway, which also emerged from the KEGG analysis, is activated by IL-17.⁶⁶

In the CNS, Batchu et al.⁶⁷ observed a higher abundance of monocytes in the prefrontal cortex compared with controls, suggesting a possible infiltration of these monocytes into the cortex. Given this, peripheral activation of monocytes could be mirrored in the brain, as monocytes may have the ability to cross the BBB.

Together, our results indicate that the activated state of CD14⁺/CD83⁺ monocytes may reflect baseline immune activation or immune dysregulation in PD. In addition, our findings provide additional evidence for the importance of inflammation and autoimmunity in PD pathogenesis, highlighting the specific role CD14⁺/CD83⁺ activated monocytes may play.

Higher activation of NK56hi cells in PD

NK cells represent the innate immune system lymphocytes. Besides their cytotoxic- and cytokine-producing functions, they have an immunoregulatory role through their interaction with other immune cells.⁶⁸ Here, we found a significant increase of regulatory NK56hi cells in PD compared with CTRL (P = 0.01) (Fig. 3B). Similar results were described in a flow cytometry study assessing the expression of the activating receptor NKG2D on NK cells between CTRL and PD subgroups.⁶⁹ In this study, they used motor UPDRS to stratify a mild symptoms group and moderate to severe symptoms group.⁴⁰ Interestingly, they found an increased NKG2D mean fluorescence intensity on NK56hi cells of PD patients only in the moderate/severe group.⁶⁹ Our correlation analysis did not reveal an association between motor impairment and the percentage of NK cells (Supplementary Fig. 5C). However, it is important to note that 10 of 12 patients in our cohort belong to the moderate/severe group as defined by Weber et al.⁶⁹ To analyse whether NK cell ratio could contribute to disease severity, a larger cohort evenly distributed across severity groups will be required to ensure sufficient statistical power.

The significant enrichment of GO terms relating to immune activation, along with the increased cell abundance, suggests a state of hyperactivation of the NK56hi cells, possibly in response to an immune dysregulation. Noticeably, our KEGG analysis identified significant enrichment of the NF-κB and TNF-α pathways (Fig. 4F). TNF-α is a pro-inflammatory cytokine which mostly leads to cell death, and its neutralization is used in chronic inflammatory and autoimmune disease treatment.⁷⁰ Interestingly, NK cells produce TNF-α,⁷¹ which is intertwined with NF-κB as the latter is activated by TNF in most canonical NF-κB signalling pathways.^72,73

Overall, this suggests a specific increase in activation of the immunoregulatory NK56hi in PD, which supports the hypothesis of immune activation or dysregulation in patients.

Common TCR in PD

Recently, the notion of PD as a single entity has been challenged in favour of a more heterogeneous disease.⁷⁴ Despite this heterogeneity, we uncovered 1094 PD-specific sequences including several TCR clonotypes shared by up to 42% of our cohort, supporting the involvement of peripheral immune mechanisms irrespective of genetic backgrounds and environmental factors. Examining annotations associated with the GIANA ‘all exact’ clusters showed that the TCR sequences in PD patients were enriched for ‘autoimmunity’ (Fig. 3E). Indeed, three patients shared a sequence, which was expanded in only one of them, associated with psoriatic arthritis. None of these patients exhibited symptoms of psoriatic arthritis at the time of enrolment. While no biological link between psoriatic arthritis and PD has been established, epidemiological studies suggest that patients with psoriasis have a higher risk of developing PD.^75-77 Our findings could pave the way for further exploration into the biological mechanisms linking PD and psoriatic arthritis or other autoimmune diseases.

Of note, databases such as McPAS-TCR still contain a limited amount of data. Thus, the low cluster match score (∼1%) suggests a high prevalence of CDR3 sequences with unidentified related pathology (Fig. 3E), highlighting the need to enrich annotation databases.

Wang et al.,¹⁹ data taken from Gate et al.,⁷⁸ reported a similar identification of 67 PD-specific TCR groups. Using machine learning, they screened these TCRs and their candidate epitopes to discover that predicted antigens could induce helper and cytotoxic T-cell responses in PD patients. The identification of PD-specific TCRs in both our studies supports the hypothesis of PD-enriched TCR sequences and common immune dysregulation. Further analysis of our shared PD epitopes with emerging algorithms, coupled with future referencing with HLA-peptide/TCR experiments, could enable deeper insights into what triggers common immune responses in PD patients.^79,80

T-cell population imbalance

Sc-RNAseq studies in PD PBMCs have been limited to preselected subpopulations. Wang et al.^19,20 conducted two single-cell studies, focusing separately on T-cell populations with TCR sequencing, and on B cells with BCR sequencing. Their T-cell analysis found increased CD4⁺ and CD8⁺ populations in PD samples.¹⁹ Similarly, we did not find a differential abundance in B cells. In contrast, Wang et al.²⁰ reported a significant increase of unswitched memory B cells, along with a significant decrease of naive B cells.

The discrepancy between the results from Wang et al. and ours may be attributable to differences in methodology such as the inclusion of preprocessing selection steps which may limit the reliability of quantitative population comparisons or enrich for subtle changes not detectable in a larger cell population. From a clinical perspective, disease stage and duration may influence the immune profile of patients.⁸¹ However, no correlation was observed between T lymphocyte percentage and symptom duration or total UPDRS in our cohort (Supplementary Fig. 5C). Unfortunately, Wang et al. did not present the clinical characteristics of their population, and the differences in the results of our studies may be attributed to variations in the disease features across our respective populations.

Activation signature genes

In a 2022 publication, Marsh et al.⁵² identified a signal in their single-cell transcriptomics of PBMCs, which seemingly correlated to their tissue processing method. They showed that if you add an enzymatic digestion at 37°C to a standard PBMC isolation protocol, cells upregulate several stress-response genes such as JUNB, JUN, FOS, DUSP1, NFKBIA, ZFP36, DDIT4, CXCR4 and CCL4, which they termed ASGs. The article recommends using transcription and translation inhibitors (TTIs) immediately upon blood draw to mitigate these ‘artefacts’. Subsequently, we found that ASGs were increased by standard PBMC isolation protocols, even without enzymatic digestion, and that adding TTIs or keeping cells cold (4°C) during processing could abrogate this signal.⁵¹ Moreover, we demonstrated that cells treated with TTIs no longer respond to immune stimulation and thus keeping cells cold would be more advantageous due to its flexibility and cost-effectiveness.

Surprisingly, while we observed increased ASG signal in all cells, it was significantly higher in PD samples. When compared with other AP donors processed at the same time, this increase was not present; instead, lower ASG levels than CTRL samples were observed in both AP groups (Fig. 4A–C). Despite the global overexpression of ASGs across PD-cell populations, the signal peaks correlate to previously described activated monocytes and NK56hi clusters (Fig. 4D).

Interestingly, recent studies have linked ASGs and AP-1 complex genes to ageing immune-cell phenotypes. Karakaslar et al.⁸² described an epigenetically modulated age-related signature involving the AP-1 complex genes JUN, JUNB and FOS. Human blood assay for transposase-accessible chromatin (ATAC) sequencing revealed increased binding sites for these genes with age, potentially triggering the production of pro-inflammatory IL-6 and contributing to the inflammaging process.⁸³ This raises questions about the pathophysiology of immune dysregulation in PD, whether it is due to an impairment in the physiological process of inflammaging, or a reaction due to other disease-related mechanisms.

It is important to emphasize that all the samples in our study were processed identically, and each processing batch contained samples from all phenotypes. Therefore it is unlikely that processing artefacts led to the differential ASG signal, strengthening the hypothesis of a heightened state of activation in PD. What remains to be determined is whether a higher baseline signal can be observed in vivo or if differential levels arise from stress responses during PBMC isolation and/or sc-RNAseq processing.

Correlation of biological signatures with clinical features

Immune cell abundance plays a crucial role in shaping the immune state of patients. We investigated the correlation between immune cell profiles and clinical features, seeking valuable insights into the relationship between immune responses and disease progression (Supplementary Fig. 5).⁸⁴ From our single cell dataset, we observed a significant positive correlation between ASG score and CD14⁺/CD83⁺ monocytes from PD patients (Supplementary Fig. 5A), which further highlights the higher state of activation in PD cells. We also found a significant negative correlation between ASG score and sleep UPDRS score, implying better sleep when ASG are high. Additionally, ASG score was positively correlated with age in PD samples (Supplementary Fig. 5A) and negatively correlated with age in CTRLs (Supplementary Fig. 5B).

Inflammaging is a process characterized by chronic, low-grade inflammation associated with ageing. This phenomenon may explain the strong correlation we observed between age and monocyte percentages in our CTRLs, as monocytes are known to become dysregulated during ageing. In contrast, we did not observe an age-related change in monocyte or CD14⁺/CD83⁺ ratio in the PD group (Supplementary Fig. 5A and C). This may suggest that PD patients harbour inflammatory processes that are distinct from inflammaging. This hypothesis is further supported by our finding of a significant negative correlation between the ASG score and age in CTRLs (Supplementary Fig. 5B). Indeed, unlike CTRL, PD patients show a positive correlation between age and activated monocytes (Supplementary Fig. 5A). These findings highlight a distinctly different pattern of activation signatures between CTRL and PD patients.

Our assessment of the correlation of broad cell type abundances with clinical features revealed some cell populations with significant correlations including with disease duration and UPDRS score (Supplementary Fig. 5C). Exploring these immune cell dynamics in pre-symptomatic and early-stage PD could help identify biomarkers to predict who converts to PD as well as the timing of conversion from non-symptomatic to symptomatic PD. This has particular relevance for REM sleep behaviour disorder (RBD) patients, who are at high risk of developing PD.⁸⁵ A longitudinal study spanning the progression of PD, including the pre-symptomatic phase, could provide deeper insights into immune cell count dynamics and their relationship with disease pathophysiology. Along these lines, increased cytokine levels in blood are observed in RBD that persist in PD suggesting they may precede symptom development.⁸⁶ In our cohort, we did not observe a significant correlation between symptom duration and ASG score or CD14⁺/CD83⁺ monocytes, which could suggest that these immune markers remain stable throughout the progression of the disease, indicating their potential as biomarkers could be relevant across disease stages (Supplementary Fig. 5A).

Overall, this intricate molecular and cellular signature in peripheral tissues highlights systemic immune dysregulation in PD patients, suggesting that targeting this hyperreactive immune state could be a promising therapeutic strategy. Integrating these immune signatures with clinical data is essential and will be invaluable in identifying robust biomarkers for early diagnosis and prognosis of PD. This comprehensive PBMC atlas aims to serve as a valuable resource for researchers, enhancing our grasp of the disease and facilitating the identification and validation of potential biomarkers and drug candidates.

Data availability

Raw sc-RNAseq data are available on the National Center for Biotechnology Information (NCBI), under the SRA: SRP525054, BioProject ID: PRJNA1145007.

Funding

This project was funded by the Courtois Foundation and the Weston Family Foundation (#211112, obtained jointly with Diana Matheoud). L.A. received a PhD scholarship from the Schlumberger Foundation. S.A. received a PhD scholarship from the Fonds de Recherche du Quebec Sante (FRQS) and the Canadian Institutes of Health Research. M.T. received a Junior 2 salary award from the FRQS.

Competing interests

The authors report no competing interests.

Supplementary material

Supplementary material is available at Brain online.