Advances in sex-specific single-cell transcriptomic profiling in Parkinson's disease

Abstract

Parkinson's disease (PD) is a progressive neurodegenerative disorder characterized by the loss of dopaminergic neurons and the accumulation of misfolded alpha-synuclein. While traditional bulk RNA-sequencing has provided valuable insights into PD pathology, it fails to capture the complex cellular heterogeneity of the human brain. Advances in single-cell transcriptomics have revolutionized our ability to dissect this complexity, enabling the identification of rare, disease-associated cell populations, or the inference of dysregulated intercellular communication networks. In this review, we discuss methodological and analytical frameworks of single-cell RNA-sequencing and summarize key findings from recent studies using single-cell RNA-sequencing that advance our understanding of PD. We highlight how single-cell transcriptomics has refined our understanding of neuronal vulnerability and revealed critical contributions of non-neuronal cells, particularly microglia and oligodendrocytes, to disease pathology in both human postmortem tissue and experimental model systems. Finally, we discuss emerging evidence for sex-specific molecular alterations in PD and emphasize the importance of sex-aware study design and analysis in future single-cell PD research.

Plain language summary

Parkinson's disease is a brain disorder affecting movement control. It is mainly known for the loss of nerve cells that produce dopamine. However, many other cell types are also involved in the disease. Traditional research methods looked at average changes across all cell types in a tissue. This, however, masks cell type-specific effects. New technologies, like single-cell RNA sequencing, can now measure gene expression changes in individual cells. This enables the study of how different cell types, like neurons, microglia, and other supporting cells in the brain, change during Parkinson's disease progression and how they interact with each other. In this review, we explain how single-cell RNA-sequencing methods work and summarize what they have revealed about Parkinson's disease so far. We show that these studies have deepened our understanding of which neurons are most vulnerable to be lost and how non-neuronal cells contribute to the disease. We also highlight an important gap in current research: most studies have not examined whether the molecular changes identified by single-cell RNA sequencing differ between men and women, even though sex strongly affects the development and symptoms of Parkinson's disease. Understanding these sex-specific differences on single-cell level is important for developing future treatments that work equally well for both male and female patients. Finally, we will also discuss how researchers can consider biological sex in future single-cell studies to enhance our understanding of Parkinson's disease.

Introduction

Parkinson's disease (PD) is the second most common neurodegenerative disorder worldwide, affecting about 1% of the population over the age of 60.^1,2 Clinically, PD is primarily characterized by motor symptoms such as resting tremor, rigidity, bradykinesia, and postural instability. ³ In addition, patients often experience a wide range of non-motor symptoms such as REM sleep disorder, hyposmia, constipation or depression, which can precede motor impairments. ⁴ Importantly, PD also exhibits pronounced sex differences in incidence, symptoms, and progression, with men being more likely to develop PD and women often experiencing a faster disease progression. ⁵ The etiology of PD still remains largely elusive, involving a complex interplay of genetic predisposition, aging, sex, and environmental factors. ⁶ In line, familial cases of PD, associated with mutations in genes, such as SNCA, LRRK2, and GBA, account for only 5% of PD cases, whereas the vast majority are idiopathic and lack a clearly defined genetic cause. ⁷ Pathologically, PD is characterized by the progressive degeneration of dopaminergic neurons in the substantia nigra pars compacta ⁸ and the accumulation of the misfolded protein alpha-synuclein (aSyn) in Lewy bodies.^9,10 While dopaminergic neurons are particularly vulnerable to neurodegeneration in PD, other neuronal populations and non-neuronal cell types, especially glia cells, are also affected. These non-neuronal cell populations actively contribute to disease progression through mechanism such as sustained proinflammatory responses, oxidative stress, and impaired protein homeostasis, thereby shaping the local brain microenvironment and influencing neuronal survival. ¹¹ Consequently, PD should be regarded as a multicellular disorder in which a complex crosstalk between neuronal and non-neuronal cells plays a critical role in disease progression. Traditional bulk RNA-sequencing, however, masks the cellular heterogeneity in the human brain by averaging gene expression changes across cell populations, potentially obscuring cell type-specific pertubations.^12,13 In contrast, high-throughput single-cell transcriptomics—in particular single-cell and single-nucleus RNA-sequencing (scRNA-seq, snRNA-seq)—enable the profiling of gene expression changes at the resolution of individual cells. This approach enables the identification of diverse neuronal and glial subpopulations and their dynamic states, as well as the inference of gene regulatory networks, and altered cell communication events,^14,15 thus providing critical insights into the molecular landscape of PD.

Single-cell RNA technologies

scRNA-seq and snRNA-seq involve a series of experimental and computational steps, each influencing data quality and interpretation. From tissue dissociation to data processing and downstream analysis, each step can affect cell representation and gene detection, posing particular challenges in studies based on postmortem brain tissue. The first crucial step in scRNA-seq and snRNA-seq is the preparation of high-quality suspensions with individual and intact cells or nuclei. For single-cell transcriptomics, fresh tissue is required, which is typically dissociated by a combination of mechanical and enzymatic methods. ¹⁶ However, in the case of brain tissue, this process is particularly challenging due to high connectivity and fragile morphology of neuronal cells. As a result, tissue dissociation can introduce biases like the relative loss of neurons compared to more robust glial cell types, and can induce stress-related transcriptional artifacts.^17,18 Such dissociation-induced effects may be particularly relevant for studies focusing on vulnerable neuronal subtypes, including dopaminergic neurons, where low recovery rates can limit statistical power and contribute to variability in reported disease-associated transcriptional changes across studies. In contrast to single cells, nuclei are more robust to mechanical stress and less susceptible to dissociation-induced artifacts, making snRNA-seq a suitable alternative for brain tissue. In addition snRNA-seq is compatibile with a wide range of sample types, including postmortem, fresh-frozen, and fixed tissue, extending its applicability to human studies, where access to fresh tissue is often limited. ¹⁸ However, snRNA-seq only captures nuclear mRNA, thereby missing transcripts in distal neuronal subcellular compartements such as dendrites and synapses.^19,20 In addition, postmortem tissue poses the challenge of RNA degradation, with RNA quality and integrity declining tissue- and gene-specifically with increasing postmortem interval.^21–23 Importantly, these degradation effects are not uniform across cell types, with neuronal populations, being disproportionately affected. ²¹ As a consequence, differences in postmortem interval and tissue preservation across studies can influence both cell-type representation and gene detection sensitivity, complicating direct comparisons between datasets. Therefore, careful assessment of RNA integrity, transparent reporting of quality metrics, and the use of computational corrections are essential to ensure reliable downstream analyses and interpretation of results.

Once a single-cell or nuclei suspension is obtained, the cells or nuclei are physically separated to label their transcriptome with unique cell-specific barcodes. The most popular separation methods are microfluidic-based approaches such as 10x Genomics Chromium, Drop-seq or inDrop.^24,25 There, single cells or nuclei are encapsulated together with barcoded beads in lipid droplets using microfluidics. Each bead carries oligonucleotides containing a unique barcode and a unique molecular identifier (UMI). After cell lysis within the droplet, poly-(A)-tailed RNA is captured and reverse transcribed, thereby labelling all mRNA from a single cell with the same barcode. The resulting cDNA is amplified, sequencing libraries prepared and next generation sequencing performed,^24–26 enabling high-troughput profiling of thousands or millions of cells (Figure 1A). In contrast, plate-based methods such as Smart-seq, Smart-seq2, MARS-seq or STRT-seq rely on cell or nuclei sorting in multiwell plates using Fluorescence-activated cell-sorting (FACS) or laser detection microdissection and generally achieve lower thoughput but higher sensitivity per cell.^27–30

Figure 1.

Overview of single-cell and single-nucleus RNA-sequencing workflows. (A) The experimental workflow of scRNA-seq and snRNA-seq beginns with the tissue dissociation and isolation of single cells or nuclei. In droplet-based approaches single cells or nuclei are encapsulated in droplets with barcoded beads using microfluidics. After cell lysis mRNA is captured and unique barcodes are added during reverse transcription, followed by cDNA amplification, library preparation and next generation sequencing. (B) Processing of raw single-cell sequencing data includes the alignment to the reference genome or transcriptome to generate a cell-by-gene count matrix. After stringent quality control and removal of empty droplets, doublets and low quality cells, data are normalized and highly informative features are selected. Dimensionality reduction such as PCA is applied, cells are clustered, and clusters annotated using reference datasets or marker genes. (C) Downstream analyses depend on the biological research question and can include differential gene expression analysis, pathway enrichment, estimation of cell type proportions, gene regulatory network analysis, trajectory inference, or analysis of cell-cell communication.

In addition to experimental protocols, scRNA-seq requires the application of a comprehensive and robust computational workflow to address challenges such as high technical noise and data sparsity. Initial processing of raw single-cell sequencing data comprises several steps including read trimming of sequences used for library preparation, demultiplexing to assign sequencing reads to individual cells based on their unique cell barcode, read alignment to the reference genome or transcriptome, and the quantification of UMIs to generate a cell-by-gene count matrix. ³¹ To ensure that only high quality cells or nuclei are used in downstream analysis stringent quality control and cell filtering steps are performed to mitigate specific technical challenges inherent to postmortem brain tissue, including RNA degradation, dissociation-induced artifacts, and low recovery of fragile neuronal populations. These quality control steps include removing empty droplets, droplets with more than one cell, and low quality cells (e.g., dying cells) with a low number of detected genes or a high fraction of mitochondrial, alongside doublet detection algorithms (e.g., DoubletFinder, Scrublet) ³² (Figure 1B). Gene expression counts are then normalized to account for differences in sequencing depth and technical variation. When reported, RNA integrity metrics are used to assess sample quality and often regressed out during normalization to mitigate postmortem interval- and degradation-related biases that can affect transcript detection and cell type proportions across studies. For cross-study comparability, batch correction and data integration methods such as Harmony, Seurat Integration, or LIGER are widely applied to harmonize datasets despite technical variations in postmortem intervall, tissue processing protocols, dissociation methods, and sequencing depth (reviewed in ³² ). Quality control and integration strategies thereby enable robust identification of disease-associated transcriptional changes, however, residual heterogeneity may persist. In this context, direct quantitative comparison of cell-type proportions and effect sizes between postmortem brain single-cell transcriptomic datasets is often limited due to inconsistent reporting of quality control metrics and filtering thresholds across studies.

After normalization, large datasets can also be reduced to most variable and informative genes to reduce technical noise and computational burden. ³³ Given the high dimensionality of scRNA-seq data, dimensionlity reduction is applied inlcuding principal component analysis (PCA), and non-linear methods such as uniform manifold approximation and projection (UMAP) or t-distributed stochastic neighbor embedding (t-SNE). Prior to UMAP or t-SNE visualization cells are clustered based on their gene expression profiles. Clustering algorithms such as graph-based community detection (e.g., Louvain or Leiden) are commonly applied and enable the identification of transcriptionally distinct cell populations. ³⁴ Cell clusters are further annotated for specific cell types using well-curated reference atlases or cell type-specific marker genes (Figure 1B). Depending on the research question downstream analyses include a differential gene expression analysis, and the estimation of cell type proportions. Given the technical heterogeneity across studies, including variations in postmortem interval, tissue processing, RNA integrity, and quality control reporting, direct quantitative comparisons of effect sizes between datasets remain challenging. Instead, replication of cell type-specific transcriptional signatures and pathway and gene set enrichments across independent cohorts provides a more robust foundation for disease-relevant conclusions. Besides pathway enrichment altered cell-cell communication events can be infered from receptor and ligand expression patterns, and gene regulatory networks can be reconstructed to identify transcription factors and target genes driving cellular identities or pathological states. scRNA-seq also allows for trajectory and pseudotime analyses to reconstruct dynamic processes such as differentiation pathways in health and disease (Figure 1C).

Single-cell transcriptomics in PD research

Selective neuronal vulnerability and compositional changes in PD

One of the defining pathological hallmarks of PD is the progressive loss of dopaminergic neurons in the substantia nigra. This degeneration, however, is not uniform across all neuronal subpopulations. Some dopaminergic populations display pronounced vulnerability, whereas others survive even into late PD stages.^35,36 The molecular basis underlying this selective vulnerability remains poorly understood. Single-cell RNA-seq is a powerful approach to uncover distinct transcriptional signatures that may underly selective susceptibly in PD and predispose subpopulations to degeneration. For example, in 2022 Kamath et al. identified a highly vulnerable subtype of dopaminergic neurons in the substantia nigra showing a significant depletion in PD and Lewy body dementia (LBD) patients, which is characterized by expression of AGTR1. ³⁷ These AGTR1-positive neurons are enriched for several known PD risk genes, including SNCA, MAPT, and GAK, suggesting that genetic risk factors preferentially act in the most vulnerable neurons. Transcriptomic analyses further implicate perturbations in several canonical cellular stress pathways regulated by the transcription factors TP53 and NR2F2, both associated with neurodegeneration. Spatial mapping has revealed that AGTR1-positive dopaminergic neurons are predominantly localized in the ventral tier of the substantia nigra, ³⁷ corresponding to the region most severely affected by neurodegeneration in PD. ³⁸ Consistently, two independent studies demonstrate that the subpopulation of dopaminergic neurons lost in PD patients is characterized by high expression of AGTR1,^39,40 and their transcriptomic profiles point to dysregulation in pathways related to unfolded protein response (UPR), oxidative stress, mitochondrial energy production, cholesterol metabolism, and iron transport, processes implicated in PD pathogenesis. ³⁹ Other studies point to the existence of additional vulnerable neuronal clusters within the substantia nigra. A distinct neuronal population with high vulnerability in PD patients is marked by the expression of RIT2, a PD risk gene. The presence of RIT2-enriched neurons has been shown not only in postmortem brain tissue but also in midbrain organoids. RIT2-enriched neurons are heterogenous in terms of TH-expression, implicating that degeneration is not restricted to dopaminergic neurons but also affects other neuronal cell types in the substantia nigra of PD patients. ⁴¹ Additionally, another neuronal cell cluster exclusively present in PD samples is characterized by high expression of CADPS2 and TIAM1, the later controlling differentiation of dopaminergic neurons. The cluster is further characterized by low expression of TH, suggesting that these cells might correspond to neuronal cells which lost their dopaminergic identity. ⁴² Notably, the selective vulnerability in PD patients is not confined to dopaminergic neurons in the substantia nigra. Transcriptomic profiling of the prefrontal cortex of idiopathic and monogenic PD patients revealed that deep layer 6 excitatory neurons expressing ADRA2A show pronounced PD-associated alterations, marked by highest levels of differential gene expression and a global reduction in transcriptional output, with pathway enrichment highlighting impairments in protein homeostasis among vulnerable neuron populations. ⁴³ Beyond these subtype-specific effects, PDE10A was consistently downregulated across cortical neurons, a finding validated by immunohistochemistry and potentially linked to non-motor symptoms in PD. ⁴³ Importantly, selective vulnerability has also been observed in human in vitro models of PD. For example, in a recently developed midbrain-hindbrain assembloid model, the spread of aSyn from the hindbrain induces early synaptic dysfunctions and increases the vulnerability of dopaminergic neurons to degenerate in the midbrain. ⁴⁴ Furthermore, human iPSC-derived dopaminergic neurons display subtype-specific sensitivity to MPP⁺-induced cytotoxicity ⁴⁵ and to oxidative and endoplasmic reticulum stress ⁴⁶ (Table 1).

Table 1.

Overview of published human single-cell and single-nucleus RNA sequencing datasets in PD research. Samples correspond to the number of samples per group (PD patients and healthy controls). Sex column represents number of samples for F = Females, M = Males, or NA = no information available.

Sample origin	Samples	Sex	Tissue	Number of cells/nuclei	Sequencing strategy	Reference
Human ex-vivo	9 PD patients	9 F, 5 M	dorsolateral prefrontal cortex	101,691 cells	scRNA-seq (Singleron Biotechnologies)	Mirzac et al., 2025 47
	5 non-PD patients with neurological diseases
Human postmortem	15 idiopathic PD patients	4 F, 11 M	substantia nigra	83,484 nuclei	snRNA-seq (10x Genomics)	Martirosyan et al., 2024 39
	14 controls	5 F, 9 M
Human postmortem	6 PD patients	3 F, 3 M	substantia nigra	57,270 nuclei	snRNA-seq & snATAC-seq (10x Genomics)	Lee et al., 2023 40
	6 controls	3 F, 3 M
Human postmortem	23 idiopathic PD patients	6 F, 17 M	substantia nigra	315,867 nuclei	snRNA-seq (10x Genomics)	Wang et al., 2024 41
	9 controls	5 F, 4 M
Human postmortem	10 PD and LBD patients	4 F, 6 M	midbrain	387,483 nuclei	snRNA-seq after FANS (10x Genomics)	Kamath et al., 2022 37
	8 controls	6 F, 2 M
Human postmortem	14 PD patients	1 F, 13 M	midbrain	69,289 nuclei	snRNA-seq & snATAC (10x Genomics)	Adams et al., 2024 48
	9 young controls (avg. 24 years old)	5 F, 4 M
	8 old controls (avg. 75 years old)	5 F, 3 M
Human postmortem	5 idiopathic PD patients	1 F, 4 M	midbrain	41,435 nuclei	snRNA-seq after FANS (10x Genomics)	Smajic et al., 2022 42
	6 controls	1 F, 5 M
Human postmortem	4 PD patients	2 F, 2 M	striatum	30,908 nuclei	snRNA-seq (10x Genomics)	Xu et al., 2023 49
	4 controls	2 F, 2 M
Human postmortem	73 PD patients	28 F, 45 M	dorsal nucleus of vagus nerve, globus pallidus, primary motor cortex, prefrontal cortex, visual cortex	2,096,155 nuclei	snRNA-seq after FANS (10x Genomics)	N M et al., 2024 50
	24 controls	9 F, 15 M
Human postmortem	6 idiopathic PD patients	3 F, 3 M	prefrontal cortex	77,384 nuclei	snRNA-seq (10x Genomics)	Zhu et al., 2024 51
	6 controls	3 F, 3 M
Human postmortem	20 idiopathic PD patients	8 F, 12 M	prefrontal cortex	299,582 nuclei	snRNA-seq (10x Genomics)	Nido et al., 2025 43
	7 LRRK2-PD patients	5 F, 2 M
	6 MSA patients	3 F, 3 M
	13 controls	10 F, 3 M
Human postmortem	2 LRRK2-PD patients	all M	prefrontal cortex, anterior cingulate	88,876 nuclei	snRNA-seq (10x Genomics)	Dehestani et al., 2024 52
	2 GBA1-PD patients
	2 controls
Human PBMCs	48 PD patients	23 F, 25 M	PBMCs	∼ 591,000 cells	scRNA-seq (DNBelab C4 system)	Grandke et al., 2025 53
	135 controls	82 F, 53 M
Human PBMCs	2 early PD patients	all F	PBMCs	48,523 cells	scRNA-seq (10x Genomics)	Xiong et al., 2024 54
	2 late PD late patients
	2 controls
Human PBMCs	8 PD patients	4 F, 4 M	PBMCs	103,365 cells	scRNA-seq (10x Genomics) of FACS sorted CD3+ T cells	Wang et al., 2021 55
	13 controls	3 F, 3 M, 7 NA
Human PBMCs	8 PD patients	4 F, 4 M	PBMCs	10,466 cells	scRNA-seq (10x Genomics) of FACS sorted CD19+ B cells	Wang et al., 2022 56
	6 controls	2 F, 4 M
Human iPSC-derived organoid	PD donor (SNCA triplication)	F	midbrain dopaminergic organoid	19,157 cells	scRNA-seq (10x Genomics)	Patikas et al., 2023 57
	isogenic control (SNCA duplication)	F
	control	M
Human iPSC-derived neurons	MPP + treated cells and controls	M	dopaminergic neurons	1418 cells	scRNA-seq (SMART-seq2 technology)	Cardo et al., 2023 45
Human iPSC-derived neurons	3 PD GBA-N370S patients	3 M	dopaminergic neurons	146 cells	scRNA-seq (SMART-seq2 technology)	Lang et al., 2019 58
	3 controls	1 F, 2 M
Human iPSC-derived neurons	Isogenic SNCA-A53 T mutant and control	M	dopaminergic neurons	15,325 cells	scRNA-seq (10x Genomics)	Fernandes et al., 2020 46

The identification of vulnerable neuronal subtypes raises the question of whether these molecular susceptibilities translate into measurable changes in cell type abundance at tissue level. While bulk RNA-seq studies already suggest that disease-associated transcriptional changes might largely reflect shifts in cell type composition, ⁵⁹ scRNA-seq enables direct quantification of the proportions of neurons and glial cells in PD brains. Findings regarding compositional changes, however, remain inconsistent across studies. Several analyses of the human midbrain report a reduction in neuronal cells in PD,^39,40 particularly in TH-enriched dopaminergic neurons, which represent the cell population most affected in the disease. ³⁹ Similar reductions are also described for glutamatergic and GABAergic neurons in the midbrain, ⁶⁰ suggesting that neuronal loss in PD is not restricted to a single neurotransmitter system. In contrast, other studies find no significant differences in cell type composition between PD patients and healthy controls,^43,51,52,61 nor between PD and non-PD neurological conditions such as multiple sclerosis. ⁴⁷ These inconsistencies between studies likely arise from a combination of biological, clinical, and methodological factors. Clinical heterogeneity, including differences in disease stage, age, postmortem interval, and thus overall tissue quality, can substantially influence cell recovery and the detectability of subtle compositional shifts. Region-specific vulnerability also plays an important role as a loss of neurons is detected in the midbrain, whereas cortical regions show preserved neuronal proportions. In addition, analytical differences, e.g., in clustering resolution, doublet removal, or filtering thresholds, can shift relative cell type abundance. Finally, methodological differences in tissue processing and cell or nuclei isolation further contribute to discordant results. For example, a meta-analysis of three snRNA-seq datasets from the substantia nigra^42,62,63 reports no reduction in the proportion of dopaminergic neurons. ⁶¹ However, the overall number of captured dopaminergic nuclei was low, likely due to technical limitations in sampling. Low cell numbers might result from the relatively large nuclei of dopaminergic neurons, which are prone to loss during tissue sectioning. ⁴² Supporting this interpretation, quantitative immunofluorescence imaging of TH- or neuromelanin-positive neurons confirms a significant depletion of dopaminergic neurons in PD tissue. ⁴² To overcome these limitations, novel enrichment strategies need to be developed. Kamath et al. (2022) introduced a flow cytometry–based protocol that selectively isolated dopaminergic neurons from postmortem midbrain tissue for snRNA-seq based on the cell type-specific expression of the transcription factor NR4A2 (NURR1), achieving a 70-fold enrichment in the number of profiled dopaminergic neurons. ³⁷ Such targeted approaches may help resolve inconsistencies among existing datasets and improve the detection of neuron-specific alterations in PD.

Transcriptomic alterations in non-neuronal cells and changes in cell-cell communication

While the degeneration of neuronal cells is the pathological hallmark of PD, single-cell transcriptomic studies increasingly highlight the active contribution of non-neuronal cells in PD pathogenesis. Multiple datasets point to higher proportions of glial cells in PD patients, particularly microglia,^39,42,51,61 and astrocytes,^42,61 validated by immunostainings ⁴² and consistent with an enhanced neuroinflammatory state in PD. ⁶⁴ Furthermore, selective vulnerability to degeneration is not confined to neurons. Disease-associated glial subpopulations, particularly TH-enriched glial cells expressing dopamine metabolism-related genes, are depleted in PD tissue and exhibit transcriptomic signatures of protein-folding stress, oxidative stress, and apoptosis. ³⁹ Beyond such subtype-specific alterations, converging evidence across several studies supports the concept of a “pan-glial activation,” characterized by shared stress and inflammatory responses across all glial cell types.^42,47 The most consistently dysregulated pathways across studies and brain regions involve protein folding stress, mitochondrial dysfunction, and immune activation. For instance, Mirzac et al. (2025) identified coordinated dysregulation of genes involved in neuroinflammation, mitochondrial dysfunction, and protein folding stress in astrocytes, microglia, and oligodendrocyte precursor cells (OPCs) from ex vivo tissue of PD patients undergoing deep brain stimulation. The molecular chaperone HSP90 was the most strongly upregulated gene across all cell types, and pharmacological interaction analysis highlights the HSP90 inhibitor NVP-BEP800 as a potential therapeutic candidate. ⁴⁷ Similarly, Smajic et al. (2022) reported activated microglial subpopulations in PD patients enriched for cytokine secretion and unfolded protein response pathways, paralleled by reactive astrocyte states. ⁴² These findings suggest that glial-mediated responses to misfolded proteins might be dysregulated in PD. Interestingly, the upregulation of genes involved in protein folding pathways, including HSP90AA1, HSP90AB1 and co-chaperone FKBP4, appears to be specific to PD microglia and is not observed in microglia in other neurodegenerative diseases like Alzheimer's disease. ⁴⁹ However, in PD, increased protein folding stress is not limited to glial cells but has been reported across all cell types and brain regions.^43,60,65

The activation of microglia emerges as a particularly central component of PD-associated inflammation. Recent scRNA-seq identified increased type I interferon activity and upregulation of the IFN-I signaling cascade in microglia in PD patients, which was validated in MPTP-treated mice. Furthermore, this pro-inflammatory phenotype was shown to be regulated by the transcription factor NFATC2. In silico perturbation analyses and functional assays demonstrated that NFATC2 directly regulates interferon–associated genes, with NFATC2 knockdown reducing IFN-β production and attenuating NF-κB activation in MPP⁺-treated microglial cells. Suppression of NFATC2 also mitigated microglia-induced toxicity in co-culture assays, underscoring its central role in driving aberrant type I interferon responses in PD. ⁶⁶ Additionally, microglia in PD are implicated in the integrated stress response (ISR), with ISR-related genes such as DDIT4, GNA13, HSPA1B, and SLC7A5 being differentially expressed and linked to immune signaling pathways and neuroinflammation including JAK–STAT and NF-κB signaling. Functional experiments in vitro and in MPTP-induced mouse model of PD further demonstrate that DDIT4 knockdown reduces aSyn aggregation and neurotoxicity. ⁶⁷ Integration of scRNA-seq with GWAS studies of known PD risk genes provides further evidence for a contribution of microglia in PD pathology. Several studies demonstrate a significant enrichment of PD risk genes, including LRRK2 and SNCA, in microglia^40,42 and PD-associated SNPs are significantly enriched in regulatory elements of microglia. ⁴⁰

Besides microglia and astrocytes, transcriptional alterations in oligodendrocytes are implicated in PD pathology. Several scRNA-seq studies report that oligodendrocytes represent the major cell type in the human midbrain accounting for about 75% of all cells.^42,48,62 Although this might partially reflect technical biases introduced during tissue dissociation, some studies nonetheless report a reduced proportion of oligodendrocytes in PD patients compared to controls, both in midbrain and prefrontal cortex.^42,51 Beyond compositional shifts, transcriptomic analyses indicate that oligodendrocytes and their precursor cells are closely linked to PD pathology. They are significantly associated with known PD risk genes, ⁵² with one study finding the highest number of known PD risk-genes being differential in oligodendrocytes including e.g., MAPT and FBXO7. ⁴⁰ Clinically, transcriptomic changes in oligodendrocytes are associated with depression severity in the anterior cingulate cortex, while changes in OPCs correlate with motor impairment in the substantia nigra. Furthermore, a distinct OPC subtype is seemingly increased in PD patients and strongly predicts motor symptoms. ⁵² Although most oligodendrocytes retain a healthy transcriptional profile during aging and PD, a PD-associated oligodendrocyte subtype has been identified, that is characterized by increased expression of stress response genes such as HSP90AA1 or MAPT and reduced expression of myelination-linked genes (MBP, MOBP), validated by RNA-FISH ⁴⁸ and suggesting that this subpopulation might lose myelination function, contributing to axonal degeneration and neuronal cell death. Trajectory analysis also identified a subpopulation of oligodendrocytes differentiating from a population of high OPALIN-expressing cells to a subpopulation of high S100B-expressing cells, the latter implicated in glial stress responses. ⁴² Along this trajectory downregulated genes are associated with neuronal maintaining pathways, while upregulated genes are related to the response to unfolded protein pathways, ⁴² suggesting a shift from a myelination-supportive to a stress-responsive state that may compromise neuronal support and contribute to neurodegeneration in PD.

Beyond cell intrinsic transcriptional dysregulation, the impact of non-neuronal cells in PD is also mediated through altered interactions with neighboring neurons and glia cells. Mapping cell–cell communication patterns can help to reveal how these intercellular signaling pathways contribute to pathogenesis. Several studies consistently report a global decrease in communication events among neurons in PD patients compared to controls.^41,51 A similar reduction in interaction frequency is observed focusing on neuron-astrocyte signaling, with disrupted signaling pathways including e.g., the SIRPA/CD47 signaling, pointing to an abnormal neuroinflammation process in PD. ⁵¹ In addition, while the global quantity of cell-cell communication is reduced in PD patients, one study reports an increased strength of communication events, potentially reflecting a compensatory response to overall loss of interaction quantity. The affected communication pathways include growth and differentiation factors, chemokines and neurotrophic pathways. ⁶¹ In contrast, another study reports more communication events in PD patients compared to controls, largely driven by increased crosstalk between microglia and neurons undergoing necroptosis. These interactions are mediated by the CLDN1 and other immune-related pathways, suggesting that microglia activated by necroptotic neurons might play an important role in the clearance of dysfunctional neurons. ⁶⁰

It is important to note that these observations rely solely on computational inference and should be interpreted with caution which might contribute to partly conflicting results across studies. Current cell-cell communication analyses are based on transcriptomic profiles of ligands and receptor without any spatial information and transcript levels do not necessarily reflect receptor protein abundance, surface availability, or activation state. Moreover, the sparsity of scRNA-seq data can lead to falsely missing signalling interactions as dropouts particularly affect lowly expressed ligand and receptor. ⁶⁸ Recent benchmarking studies also indicate that predictions can vary depending on the computational method and the ligand–receptor database used.^69–72 These limitations highlight the need for complementary experimental validation, for example using spatial transcriptomics, multiplexed imaging, or proteomic approaches such as co-immunoprecipitation, to confirm the spatial co-localization and physical interaction of predicted interactions in PD tissue. Nevertheless, the current communication analyses in PD patients offer useful insights into potential intercellular signaling changes, highlighting the key involvement of non-neuronal cells and neuroinflammatory pathways in PD, which can help guide future mechanistic studies.

In addition to inflammation-related changes within the central nervous system, recent single-cell transcriptomic studies highlight that peripheral immune dysregulation might play an important role in PD pathology. The availability of blood and cerebrospinal fluid samples further enables the investigation of immune alterations that may arise early in disease progression and could serve as potential biomarkers. Single-cell profiling of peripheral blood mononuclear cells (PBMCs) has uncovered distinct perturbations in both the adaptive and innate immune system of PD patients. For instance, several studies highlight pronounced changes in T cell populations. PD patients show an increased proportion of CD8⁺ T cells, ⁵⁵ particularly in male patients, ⁵³ and a reduction in CD4⁺ T cells. T cell receptor sequencing further revealed a decreased T cell receptor diversity in PD, suggesting clonal expansion and more restricted antigen specificity. Pseudotime analysis identified a shift in CD8⁺ T cell populations from central memory to terminal effector phenotypes, consistent with chronic antigen exposure and progressive T cell activation in PD. ⁵⁵ In addition, several studies report altered B-cell composition and activation. PD patients exhibit increased proportions of memory B cells^56,73 and reduced proportions of naive B cells alongside elevated IgG and IgA isotypes and higher class-switch recombination rates, reflecting enhanced humoral activation in PD patients. ⁵⁶ Consistently, clonally expanded memory B cell population in PD show an upregulation of MHC II genes and transcription factor activator protein AP-1, consistent with antigen-driven activation, suggesting activated humoral immune response in peripheral blood of PD patients. ⁵⁶ Furthermore, natural killer cells (NK cells) also appear to be dysregulated in PD as a reduced proportion of NK cells has been reported in early and advanced stages of PD, a finding validated by immunostainings. ⁵⁴ In addition, XCL2, a chemokine specifically expressed in NK cells, shows markedly increased expression in PD patients, which was confirmed by RT–qPCR. Notably, XCL2 expression correlates positively with PD severity and is proposed as a potential diagnostic biomarker for PD. ⁵⁴

Central and peripheral immune alterations appear to co-occur in PD, raising the question of whether these processes are mechanistically linked. Multiple lines of evidence suggest potential interactions between the peripheral immune system and central neuroinflammation. Several studies report the infiltration of peripheral immune cells into PD-affected brain regions contributing to neurodegeneration^74,75 with increased blood-brain barrier permeability during inflammation facilitating immune cell entry into the central nervous system under disease conditions. ⁷⁶ In line, recent single-cell studies have shown elevated gene expression of molecules associated with blood-brain barrier transmigration in cytotoxic T cells from PD patients. ⁵⁵ In addition, ligand–receptor analyses have identified potential interaction pairs between infiltrating T cells and neuronal and glial cells, ⁵¹ supporting the possibility of direct immune-neural communication in PD. Recent single-cell analyses have also identified NK cell infiltration into the amygdala and motor cortex correlating negatively with TH levels, ⁵⁴ supporting potential bidirectional communication between the peripheral immune system and CNS. Nevertheless, direct causal relationships between peripheral immune changes and central glial activation remain difficult to establish. Addressing this question will require longitudinal study designs, paired profiling of blood, cerebrospinal fluid, and postmortem brain tissue from the same individuals, and integrated single-cell analyses. Such approaches will be critical to determine whether peripheral immune signatures merely reflect, actively contribute to, or are independent of central neuroinflammatory processes in PD.

Together, these findings suggest that non-neuronal cells play an active and multifaceted role in PD. In postmortem brain tissue, glial cells exhibit shared stress, mitochondrial, and inflammatory responses, with microglia, astrocytes, and oligodendrocytes displaying disease-specific alterations linked to disrupted protein homeostasis and immune signaling. These changes, coupled with impaired intercellular communication and peripheral immune dysregulation, highlight a complex, system-wide network of neuronal, glial and immune interactions that might drive or contribute to PD pathogenesis.

Neuronal and glial transcriptomic alterations in PD mouse models

While scRNA-seq studies of postmortem human brain tissue have provided key insights into the cellular landscape of PD, the limited availability of human tissue and their representation of largely the terminal disease stage, make animal models indispensable to investigate transcriptional perturbations underlying PD pathology, particularly at early disease stages. Several studies show that the selective neuronal vulnerability observed in PD patients is also evident in animal models (Table 2). In 6-month-old transgenic mice expressing human A53 T aSyn, dopaminergic neurons display downregulation of ion channel components such as Cacna1a and Cacna1b, pointing to impaired neurotransmitter release in early disease stages. ⁷⁷ Also, in toxin-based models, specific neuronal subpopulations exhibit high susceptibility to degenerate. For example, dopaminergic neurons expressing Sox6 or Pcsk6 are particularly vulnerable to 6-OHDA–mediated degeneration, ⁷⁸ while glutamatergic neurons in the midbrain are transcriptionally reprogrammed in paraquat-treated mice that develop depression-like behaviors. ⁷⁹ In MPTP-treated mice, Guo et al. (2022) report a distinct cluster of medium spiny neurons not observed under control conditions, also suggesting disease-specific neuronal states and subtypes. ⁸⁰

Table 2.

Overview of published single-cell and single-nucleus RNA sequencing datasets from PD mouse models.

Samples	Tissue	Number of cells/nuclei	Sequencing strategy	Reference
Six female 6-OHDA treated mice and controls	midbrain	68,914 nuclei	snRNA-seq after FANS (10x Genomics)	Salmani et al., 2024 78
Eight male 10-weeks old MPTP-treated mice and controls	substantia nigra	22,983 nuclei	snRNA-seq (10x Genomics)	Liu et al., 2024 82
Male 8-weeks old MPTP-treated mouse and control	pooled cerebral cortex, hippocampus, striatum and cerebellum	19,531 nuclei	snRNA-seq (10x Genomics)	Guo et al., 2022 80
Three male 6–8 weeks old Paraquat-treated mice and controls	whole brain	55,779 nuclei	snRNA-seq (10x Genomics)	Weng et al., 2024 79
Three male 10-weeks old MNPQ-treated mice and controls	substantia nigra	962 cells	scRNA-seq (10x Genomics)	Khan et al., 2022 81
Male 6 months-old WT and A53 T aSyn TG mice	midbrain, striatum, cerebellum	45,996 nuclei	snRNA-seq (10x Genomics)	Zhong et al., 2021 77

Integration with human PD risk genes further strengthen the translational relevance of PD mouse models. Across multiple mouse models of PD transcriptional signatures show substantial overlap with PD GWAS loci, including LRKK2, FYN, and CACNA1C. This overlap is not limited to dopaminergic neurons, but extends to astrocytes, OPCs, and microglia, highlighting that genetic susceptibility likely involves multiple brain cell types.^77,81,82 Mouse models also revealed early alterations in non-neuronal cell types that mirror the glial activation observed in humans. In young A53 T transgenic mice transcriptional activators of the NF-κB signaling pathway are markedly elevated in astrocytes and microglia in the midbrain, indicating early pro-inflammatory glial activation. ⁷⁷ Similarly, in MPTP-induced models, microglia show the highest enrichment for PD risk genes and activation of inflammation-related pathways, with Nfe2l2 and Runx1 identified as key transcriptional regulators mediating microglial activation. ⁸² In addition, PD-specific astrogliosis is observed in MPTP-treated mice, with the transcription factors Dbx2 and Sox13 emerging as novel regulators of astrocyte activation. ⁸⁰ At the same time, PD mice exhibit a loss of endothelial cells, which might enhance permeability of the blood-brain barrier and promote the entry of MPP + toxins into the brain environment. ⁸⁰ Across multiple brain regions, dysregulation of stress-response pathways emerges as a reoccurring pattern, mirroring findings from human data. Upregulation of heat shock protein genes, proteasomal genes, and ubiquitin-related genes in both neurons and glia cells points to widespread proteotoxic stress. ⁷⁷ Furthermore, network analysis from a toxin-induced model revealed alterations of extracellular matrix-related processes with PD, with Cacna1c identified as a central hub gene. ⁸¹ Inference of cell-cell communication, similar to findings from human datasets, point at a reduction in the global quantity of interactions but increased interaction strength among PD-specific cells, with NRXN and NEGR signaling pathways being notably enhanced in PD mouse model. ⁸⁰ Together, this demonstrates that PD mouse models not only recapitulate key transcriptional features observed in human postmortem tissue but also extends them by capturing early stages of the disease.

Sex-specific considerations in single-cell transcriptomics studies of PD

PD is characterized by pronounced sex differences in prevalence, progression, and clinical symptoms, yet sex-dependent effects are often overlooked in study design and analysis. Men have an approximately 1.5–2 times higher risk of developing PD than women and often experience an earlier age of onset.^83–87 Contrary, women tend to have a faster disease progression and higher mortality rates.^87,88 Sex-related disparities also extend to both motor and non-motor symptoms. Female PD patients are more likely to present with tremor and are at a greater risk of experiencing depression, whereas males are more prone to cognitive decline and rigidity.^87–90 However, the underlying mechanisms driving these sex-specific variations remain incompletely understood. Previous studies suggest that sex-specific hormones, particularly estrogen, may have neuroprotective effects on dopaminergic neurons, possibly contributing to lower PD risk in women.^91–93 In addition, intrinsic molecular sex differences, such as higher SNCA expression in male dopaminergic neurons, or sex-dependent expression of inflammatory factors might contribute to sex-specific effects.^88,94 Lifestyle factors, including gender-biased behaviors such as smoking, caffeine intake, and occupational exposures, like pesticide exposure, may further modulate PD risk and pathology in a sex-dependent manner, adding another layer of complexity to sex differences in PD patients. ⁹⁵

Despite the well-known sex differences at clinical level, the underlying transcriptional basis remains poorly characterized, particularly at single-cell level. On bulk level, several meta-analyses revealed sex-specific transcriptomic differences related to typical PD hallmarks, including oxidative stress, or neuroinflammation,^96–98 emphasizing the importance to consider sex as a biological variable. For instance, male PD patients exhibit more differential expression related to disturbed mitochondrial and dopamine metabolism, with NR4A2 identified as key transcriptional regulator. ⁹⁷ In contrast, female PD patients exhibit increased expression of proinflammatory genes.^97,98 However, it is unclear which cell types contribute to these observed sex-dependent effects. To date, only a very limited number of scRNA-seq studies explored cell type-specific sex differences in postmortem PD tissue. The analysis of two independent snRNA-seq datasets from the substantia nigra revealed a higher number of male-specific differential expression than in female PD patients. ⁹⁷ Transcriptional changes are particularly pronounced in oligodendrocytes and astrocytes, suggesting that these could be the main cell types affected by sex-dependent changes. Furthermore, male-specific oligodendrocyte and sex-dimorphic astrocyte genes, showing opposite direction of expression changes between sexes, are enriched for mitochondrial-linked pathways, apoptosis, and neuroinflammation, particularly cytokine signaling and NF-κB signaling, ⁹⁷ highlighting sex-dependent immune-related perturbations in glial cells. Complementary insights from peripheral immune cells further support the presence of sex-specific disease signatures. A recent large-scale analysis of scRNA-seq data from PBMCs demonstrate divergent shifts in immune cell composition between male and female PD patients. ⁵³ Specifically, males exhibit increased proportions of CD8⁺ T cells and plasma cells, whereas females show the opposite pattern, along with an enrichment of B cells in female PD patients with mild cognitive impairment. Importantly, such sex-specific compositional changes were not detected in earlier studies lacking sex-stratified analyses.^55,56 In addition, females display a higher number of differentially expressed genes overall, and transcriptional changes are often anti-correlated between sexes, highlighting distinct immune responses in male and female PD patients. ⁵³

However, to date, the vast majority of single-cell transcriptomic studies have not systematically examined sex-specific differences and genes on sex chromosomes are excluded from the analysis. Most published datasets combine male and female samples without performing a sex-stratified analysis (Table 1), primarily because their limited sample sizes do not support statistically robust subgroup analyses. In some cases, the sex of donors is not even reported and for many studies there is a high imbalance in male and female samples between PD and control groups, potentially skewing statistical power and obscuring sex-dependent differential expression. For instance, unequal representation, with high numbers of male samples, might lead to biased detection of differential expression and higher numbers of male-specific differential genes, which might reflect differences in statistical power rather than true biological differences. Subsampling approaches, such as reanalyzing randomly selected male subsets to match female sample numbers, can help to confirm interpretability of sex-specific findings. ⁹⁷ To date, only two large single-cell studies, namely by Grandke et al. on PBMCs with 183 donors ⁵³ and a meta-analysis by Tranchevent et al. on the substantia nigra of 82 donors, ⁹⁷ offer sufficient statistical power and have therefore been able to report robust sex-dependent transcriptional differences. This highlights the need for larger and more balanced cohorts in future studies. In addition, meta-analytic approaches for existing single-cell datasets are a promising approach to increase statistical power and enable more reliable detection of sex-specific transcriptional changes in PD.

Beyond study design, analytical decisions can significantly constrain the detection of sex-dependent effects. One common issue concerns the handling of sex-chromosome-linked genes. Many studies simplify their analyses by removing sex chromosomes to avoid confounding effects, thereby potentially overlooking meaningful biological differences related to sex. Indeed, previous bulk transcriptomic studies identified X-linked genes such as ARMCX2, which was upregulated in female PD patients and downregulated in males, underscoring the importance of these loci in PD.^96,97 Likewise, the Y-chromosome-linked gene SRY was shown to regulate dopamine synthesis, and was upregulated in vitro and in toxin-induced animal models of PD. Reducing nigral SRY expression diminished the male bias in dopaminergic degeneration, mitochondrial dysfunction, and neuroinflammation in a 6-OHDA rat model, suggesting that Y-chromosome-linked genes might contribute to higher PD susceptibility of males.^99,100 Another common practice in scRNA-seq data analysis is to include sex only as a covariate in the differential model to account for sex effects. While this approach adjusts for average differences between sexes, it collapses male and female samples into a single analysis, thus masking genes that are differentially regulated by sex. Importantly, a sex-stratified analysis goes beyond identifying male- and female-specific differentially expressed genes, it also captures sex-dimorphic or sex-modulated changes, where genes exhibit opposite directions of regulation or similar directional changes with different magnitudes between sexes. ¹⁰¹ Consequently, sex-stratified analyses or differential models incorporating interaction terms between sex and disease are crucial to uncover sex-dependent transcriptional patterns.

In line with these considerations, several computational resources now explicitly support sex-aware omics analysis. For bulk and single-cell transcriptomic as well as other omics data, the R package XYOmics was developed to provide an end-to-end workflow for disease-specific molecular sex differences, including linear models with sex-by-disease interaction terms or sex-stratified analyses, to detect sex-specific, sex-dimorphic and sex-modulated genes, followed by downstream pathway enrichment and gene regulatory network analyses. ¹⁰² Beyond this, standard differential expression frameworks (e.g., limma, DESeq2, seurat) can be easily adapted for sex-aware analysis by incorporating sex in the differential model or by performing sex-stratified analyses, as discussed in a recent bioinformatics overview. ¹⁰¹ However, sex-stratified approaches double the multiple testing burden and reduce statistical power, since each comparison is performed twice on smaller per-sex sample sizes. Recent methodological work has therefore proposed permutation-based strategies that shuffle sex labels to test if observed sex effects exceed chance levels, yielding empirical significance values, that provide a robust framework for validating molecular sex differences in high-dimensional omics data. ¹⁰³ Finally, position papers on sex annotation in omics resources emphasize the importance of recording and reporting sex, performing both combined-sex and sex-stratified analyses where possible, and critically evaluating sex composition and annotation in public datasets and pathway databases used for PD studies. ¹⁰⁴

Collectively, these considerations highlight the need for sex-balanced study designs, transparent reporting of the donor sex, and sex-aware analytical frameworks in single-cell studies of PD (Figure 2). Incorporating these principles will be essential to elucidate the molecular bases of sex differences in PD vulnerability and progression. Importantly, to confirm the biological relevance of sex-specific findings from single-cell transcriptomic studies, validation across both female and male samples, including replicating key observations in independent, sex-balanced cohorts, is essential. Furthermore, subsequent validation should occur on multiple levels including gene, protein, or epigenetic level, e.g., by snATAC-seq to confirm accessibility of key regulatory genes, or through high-throughput spatial approaches, such as spatial transcriptomics, MERFISH, or RNA-FISH (reviewed in ¹⁰⁵ ). Functional consequences of candidate genes could also be interrogated using CRISPR-based perturbation approaches, e.g., in vitro in iPSC-derived neurons or glial cells. Importantly, these experiments should include both male and female models to capture sex-dependent effects. Integrating these complementary approaches will be critical to establish the mechanistic relevance of single-cell transcriptomic findings and to develop targeted therapies for male and female PD patients.

Figure 2.

Sex-specific considerations across experimental and computational steps in single-cell transcriptomics studies. Key steps at which sex as a biological variable should be incorporated, including sample collection, library preparation and sequencing, sex-stratified preprocessing, and sex-aware differential expression analysis.

Conclusion

In recent years, single-cell transcriptomics has transformed our ability to interrogate the molecular landscape of PD by enabling the detection and characterization of vulnerable neuronal subpopulations. Furthermore, scRNA-seq strengthened our understanding that PD is a multicellular disease and non-neuronal cells, particularly microglia, astrocytes, and oligodendrocytes, actively contribute to disease pathology, and show a consistent dysregulation of common PD-associated pathways, such as protein folding stress, mitochondrial dysfunction, neuroinflammatory signaling, and disturbed communication events. Beyond the brain, single-cell profiling of peripheral immune cells highlights systemic alterations that might serve as early biomarkers of disease. Integrating insights from bulk transcriptomics and initial sex-stratified single-cell analyses further suggest that PD manifests in a sex-specific manner, with distinct cell types and molecular pathways affected in males and females. These findings underscore the importance of incorporating sex as a biological variable in future study designs and analyses of scRNA-seq data. Despite the advances of single-cell sequencing, important challenges remain, including technical limitations related to sample quality, and data sparsity. Crucially, functional validations are required to confirm transcriptomic studies and to translate them into a deeper mechanistic understanding of PD pathogenesis and the development of targeted therapies.