Skip to main content
NCBI home page
NIHPA Author Manuscripts logoLink to NIHPA Author Manuscripts
. Author manuscript; available in PMC: 2026 Sep 27.
Published in final edited form as: Parkinsonism Relat Disord. 2025 Sep 27;140:108066. doi: 10.1016/j.parkreldis.2025.108066

Epigenomic profile of GBA1 in Parkinson’s disease

Eloise Berson a,b,c, Raphael Zaghroun a, Matteo Santoro a, Syed Bukhari a, David Seong b,d, Chi-Hung Shu b,c,e, Amalia Perna a, Tomin James b,c, Kathleen S Montine a, Geidy E Serrano f, Thomas G Beach f, C Dirk Keene g, Howard Y Chang h,i, M Ryan Corces j,k,l, Brenna Cholerton a,*, Nima Aghaeepour b,c,d,†, Thomas J Montine a,†
PMCID: PMC12825421  NIHMSID: NIHMS2116230  PMID: 41033114

Abstract

Introduction:

While genome-wide association studies have identified GBA1 as a key gene contributing to disease severity and cognitive decline in PD, its molecular effects remain poorly understood.

Methods:

We used integrative bulk ATAC-seq across six brain regions from autopsied individuals with PD and varying genetic risk to characterize region- and cell type-specific molecular differences. Using Cellformer, an AI-based bulk ATAC-seq-deconvolution tool, we determined cell type-specific effects of GBA1 on PD disease progression and then validated our findings using whole transcriptome data from blood samples.

Results:

Epigenomic differences between PD with (“GBA+”; n=15) and without (“GBA-”, n=15) GBA1 variants were localized in substantia nigra. Nineteen chromatin-accessible regions strictly separated GBA+ from GBA-, including the promoter sites of key genes such as CACNA1C, EHMT1, and SLC25A48. The effect in GBA+ spanned the main cell types in brain, and chromatin differences between GBA- and GBA+ increased with neuropathologic progression of disease. Significant differences in the epigenomic profile in GBA+ were observed in neuronal cells (AUROC=0.8, AUPRC=0.8, P-value<0.0001). Validation in blood samples distinguished between GBA+ and GBA-subtypes, achieving AUROC values of 0.99. Over 5,000 transcripts in blood cells distinguished GBA+ from GBA-, validating key genes and pathways from our epigenomic analysis of brain regions.

Conclusion:

Our study provides novel insights into the cell type-specific epigenomic and transcriptomic landscape of GBA+ and its molecular divergence from other PD subtypes, and highlights potential therapeutic targets for this genetically defined subset of PD.

Keywords: GBA, epigenomic, genetics, Parkinson’s disease

INTRODUCTION

Parkinson’s disease (PD) is a common and debilitating neurodegenerative disease with complex etiology deriving from processes of aging, multiple genetic causes and risk factors, and environmental influences. Along with its cardinal motor symptoms, PD also is associated with numerous non-motor symptoms; of these, cognitive impairment is highly prevalent throughout the disease and can be particularly distressing for people with PD and their caregivers [1]. Efforts to identify the molecular signatures of cognitive decline in PD are thus an important next step toward targeted therapies for people with PD.

Among the strongest genetic predictors of PD risk as well as disease severity and progression are variants in GBA1, a gene encoding for the lysosomal enzyme, glucocerebrosidase (GCase). Patients with GBA1-associated PD (“GBA+”) have an earlier disease onset as well as a more severe motor and non-motor symptom profile, including dementia[2], supported by some studies showing a more diffuse pattern in Lewy body distribution than in idiopathic PD.[3] Further, GBA+ has been shown to be associated with greater effects outside substantia nigra (SN) when compared to those without a GBA1 risk variant (“GBA-“), including reduced GCase activity in the putamen, cerebellum, amygdala, and frontal cortex [4, 5], and a significant reduction in DNA methylation of SNCA intron 1 in the frontal cortex[6]. Risk variants in GBA1, including variants that are pathogenic and nonpathogenic for Gaucher disease, are found in 2%−31% of people with PD and confer a 5–30 fold increased risk of developing PD [2].

A critical step in translating genetic associations to targeted therapeutics is elucidating their molecular signatures in humans that can then be tested for cause-and-effect relationships in experimental models. One important molecular signature for developing new therapeutics is DNA accessibility within chromatin as determined by Assay for Transposase Accessible Chromatin by sequencing (ATAC-seq) [7]. ATAC-seq maps the chromatin regulatory state and altered mechanisms guiding gene expression in disease [8]; an added advantage of ATAC-seq with brain autopsy is that DNA is a much more stable macromolecule than messenger RNA [9]. Here, we analyzed bulk ATAC-seq data from multiple regions of brain to determine the chromatin regulatory state in people with PD with and without a GBA1 variant in order to determine whether group differences are found primarily in the SN or extend to regions outside the SN. We then applied a deep learning method called Cellformer to investigate cell type-specific expression [10], and validated our epigenomic findings using human blood samples assayed with long-read RNA-seq.

METHODS

An overview of the study methods is provided in Figure 1 and detailed below. References for specific model analysis methods are provided in the Supplement.

Figure 1. Study overview.

Figure 1.

Open in a new tab

(Top) Bulk ATAC-seq data was collected from six brain regions of individuals with Parkinson’s disease (PD) (n=30). Variance and univariate analyses were conducted from bulk data to identify key brain region-specific epigenomic distinctions among PD subgroups. AI-based deconvolution was subsequently applied to uncover cell-type-specific molecular signatures of GBA1. (Bottom) Leveraging the large-scale cohort from the Parkinson’s Progression Markers Initiative (n=372), we validated our findings and provided a comprehensive overview of the molecular differences between PD subgroups in blood.

Abbreviations: AST, astrocyte; CAUD, caudate; GBA, glucoceribrosidase gene; HIPP, hippocampus; MDFG, middle frontal gyrus; MIC, microglia; NEU, neuronal cell; OLD, oligodendrocyte; OPC: oligodendrocyte precursor cells; PD, Parkinson’s disease; PCA, principle components analysis; PTMN, putamen; SMTG, superior and middle temporal gyri; SN, substantia nigra

Brain autopsy samples

Brain autopsy samples from participants with PD were obtained at Stanford University, the University of Washington, and the Arizona Study of Aging and Neurodegenerative Disorders and Brain and Body Donation Program at Banner Sun Health Research Institute in Sun City following informed consent procedures and Institutional Review Board approval. The sample size was based on the available biological material. Clinical diagnosis of PD and cognitive status proximate to death, including dementia and mild cognitive impairment were determined using current consensus criteria [11]. Neuropathologic assessment of Lewy body disease (LBD) and Alzheimer’s disease neuropathologic change (ADNC) were likewise determined using current harmonized consensus guidelines; other co-morbidities were excluded [12].

DNA screening

DNA sequencing for GBA variants was performed on blood cells or cerebellar cortex by the tissue procurement site using Sanger sequencing as previously described.[13] GBA1 risk variants include pathogenic mutations, which are defined as those mutations known to be associated with Gaucher’s disease, and variants that do not increase the risk for Gaucher’s disease but do increase the risk for PD and cognitive decline (Table S1).

Bulk ATAC-seq data processing

Bulk ATAC–seq were sequenced using an Illumina HiSeq 4000 System with paired-end 75-bp reads. Bulk ATAC-seq samples were processed using the ENCODE Data Coordination Center (DCC) ATAC–seq pipeline (https://doi.org/10.5281/zenodo.211733) (v.1.1.7). starting from FASTQ files and aligned using GRCh38 reference genome assembly preserving the primary chromosomes chr1–chr22, chrX, chrY, chrM. Technical replicates were first merged using the merge function from SAMtools, and ENCODE pipeline with default parameters was applied on all the biological replicates from the same brain region and same PD groups. The final set of chromatin accessible regions (CAR) was obtained by merging group and brain region-specific idr reproducible peaks using bedtools using a Maximum distance between features of 100. Quality control across different brain regions and PD groups indicated a Fraction of Reads in Peaks (FRiP) score ranging between 0.2 and 0.3, confirming the high quality of the data and their suitability for downstream analysis (Figure S1). Once the final set of CARs was derived, the peak matrix was obtained using featureCounts and preserving peaks that account for 99% of the variance, leading to 131833 reproducible CARs.

Cell type-specific epigenomic profile using Cellformer

Cell type-specific expression was predicted using the Cellformer model, trained and evaluated as described previously [10] (see https://github.com/elo-nsrb/Cellformer). To train the Cellformer model, we used processed single-cell ATAC-seq fragment files. Fragments were processed using the recommended ArchR workflow: after doublet removal, regional and cell type-specific pseudo bulk replicates were created from snATAC-seq and used for peak calling conducted using the integrated MACS2 library from ArchR. Only regional and cell-type-specific significant CARs were conserved for downstream analysis (FDR < 0.001 and FC > 2). In total, we defined a set of 42352 CARs. The count normalized matrix, combining CAR from all the bulk samples, was then derived using featureCounts. CARs were annotated using ChIPseeker with the default parameters following ATAC-seq data processing guidelines and Harvard bioinformatics recommendations. All replicates were used for downstream analysis. As described in Berson et al. [10], a synthetic dataset of synthetic bulk profiles with known composition was created using cell-type specific pseudo-bulks derived from these single-nucleus ATAC-seq.

The Cellformer model was then trained and evaluated using GroupKFold cross-validation training, splitting individual samples into five non-overlapping groups. Low predictable cell type-specific CARs across cross-validation iterations (Rho<0.2) were removed for downstream analysis [10]. Validation was performed using Pearson and Spearman correlation values between the predicted cell type-specific profiles and the ground truth using Scipy library (v 1.11.4). We also assessed the ability of the model to predict non-zero CAR by computing the Area Under the Receiver Operating Characteristics (AUROC) and the Area Under Precision-Recall curve (AUPRC) after binarization of both the predicted and ground truth ATAC-seq profiles.

The model was also validated by computing the Spearman correlation between predicted cell type-specific signals from technical replicate samples across the entire cohort. To assess the significance, the correlations from technical replicates were compared to those from random replicates. More precisely, for each bulk sample, Spearman correlation was computed between the model’s output of this sample and a random replicate, arbitrarily selected from the same brain region, from the phenotype group, or both the same brain region and phenotype group. P value was derived by comparing the mean correlation between true replicates and random replicates using Bonferroni corrected two-sided Mann-Whitney test.

Bulk long-read RNASeq data processing

Human blood samples were obtained from people with PD who volunteered in the Parkinson’s Progression Markers Initiative (PPMI) [14]. Preprocessed whole-transcriptome long-read RNA-seq data from all available PD cases along with the curated metadata of study participants (version 2024–07-29, B38), demographic and diagnosis information (version November-2024) were downloaded from www.ppmi-info.org/access-dataspecimens/download-data. Cases with multiple genetic risk variants were removed. All PPMI participants provide written informed consent in accordance with site IRBs [14].

Machine learning analysis

LASSO, Ridge, Random Forest, and Xgboost models were tested for the prediction of PD phenotypes. The input to the models was cell type-specific and brain-specific profile data. A two-loop nested cross-validation framework, with AUROC and AUPRC criteria, was exploited to reduce the estimated error bias while selecting the optimal model and number of features that maximize the inner loop criteria. The best model was then tested on samples from unseen individuals in both the inner and outer loop using the GroupKFold function from scikit-learn. Generalization and robustness of model performance were ensured using a 5-iteration repeated the 10-fold cross-validation scheme for the outer loop and 5-fold inner loop scheme. All the tested models were run on Python using scikit-learn packages (v 1.5.2) with default hyperparameters. The Mann-Whitney U test was used to assess the significance of model prediction between different groups. Holm-Bonferroni was applied to correct for multi-testing using statsmodel package (v0.14.4).

Statistical analysis

Variance analysis was computed using Python decoupleR package. Differential expression analysis was performed using PyDeseq2 package (v 0.4.12) with sex as a covariate and with the refit outlier option. All plots were generated with Numpy (v 1.26.4), Seaborn (v 0.13.2), Matplotlib (v 3.9.4), Scanpy (v 1.10.3) in Python. Pathway analysis and gene enrichment were computed using GSEApy package (v 1.1.4) using KEGG 2021 database.

RESULTS

Sample Characteristics

High-quality bulk ATAC-seq data were obtained from six brain regions: two isocortical - middle frontal gyrus (MDFG, Brodmann area 9) and superior and middle temporal gyri (SMTG, Brodmann areas 21 and 22); two striatal - caudate (CAUD) and putamen (PTMN); substantia nigra (SN); and hippocampus (HIPP) (Figure S2). Demographic and clinical characteristics of GBA+ (n=15) and GBA- (n=15) are described in Table S2. The GBA+ group was significantly younger at death than the GBA- group (M=74.5[5.9] vs. M=79.5[5.9], p <0.05). There was a lower proportion of females in both groups (20% GBA-, 40% GBA+), the difference between groups was not statistically significant.

GBA+ epigenomic profile localizes to substantia nigra and involves synaptic and calcium pathways

A similar number of chromatin accessibility regions (CARs) were found between GBA+ and GBA-across brain regions (Figure S3), with less CAR found in HIPP and SN than in isocortical and striatal regions. For downstream analysis, 131833 CARs were conserved accounting for 99% of the variance in the data (Figure S4). The primary known sources of variance in bulk samples were regional differences (corrected ANOVA p < 0.001) and sex differences (corrected ANOVA p < 0.001) (Figure S5).

We next investigated the region-specific epigenomic signature of GBA1 variants in PD by comparing GBA+ epigenomic profile with GBA-’s profile using univariate analysis. Most significant differences were found in the SN, with nineteen CARs significantly different between GBA+ and GBA- (n=30, FDR <0.05) (Figure 2A). PCA-based analyses on these CARs demonstrates that GBA- and GBA+ formed clearly separate clusters (Figure 2B). Among these CARs, we identified four promoters (essential for gene activation; EHMT1, CACNA1C, RAP1GAP2, and SLC25A48), and three 3’ UTR (important for mRNA regulation) (PIN1, MYZAP, and GOLGA3) down-regulated in GBA+ (Figure 2C). Pathway enrichment analysis associated these CARs with synaptic function, calcium signaling, ventricular cardiomyopathy, and aldosterone signaling (FDR<0.05) (Figure 2D).

Figure 2. The epigenomic impact of GBA1 variants in people with PD is predominantly localized in the SN and is notably associated with alterations in synaptic and calcium signaling pathways.

Figure 2.

Open in a new tab

(A) Nineteen significantly different accessible CARs were identified across brain regions when comparing GBA- and GBA+ groups using DESeq2 (FDR<0.05). (B) Mapping the patterns of differentially accessible CARs using PCA demonstrates that the GBA- and GBA+ groups form clearly separate clusters. (C) Volcano plot of differentially accessible CARs in SN, color-coded by CAR types. Most notable, three promoters—EHMT1, CACNA1C, and SLC25A48—were identified as differentially accessible between GBA- and GBA+ groups (DESeq2 adjusted P-value < 0.05 and |log2FC|>0.5). (D) Pathway enrichment analysis of the differentially accessible CARs in SN highlights associations with processes involved in synaptic function, calcium signaling, ventricular cardiomyopathy, and aldosterone signaling.

*p<0.05, **p<0.01, ***p<0.001. Abbreviations: GBA, glucocerebrosidase gene- associated PD; SN, substantia nigra

Cell type-specific divergence between GBA- and GBA+ during neuropathologic progression

We then applied the Cellformer model to investigate the cell type-specific effect of GBA1 variants in brain regions of people with PD. It was trained on single-nucleus ATAC-seq from various brain regions and comprehensively evaluated to deconvolute bulk ATAC-seq (to determine from which of five main brain cell types the DNA signal came): neuronal cells (NEU), astrocytes (AST), microglia (MIC), oligodendrocytes (OLD), and oligodendrocyte precursor cells (OPC) across different brain regions (Figure 1). Extensive validation demonstrates Cellformer’s ability to accurately predict cell type-specific epigenomic profiles (mean Pearson correlation > 0.82, Spearman correlation > 0.7 across cell types) and classify open chromatin accessible regions (mean AUROC > 0.85, AUPRC > 0.8 across cell types) (Figure S6). Additionally, high reproducibility was found between biological replicates (Figure S7A).

We examined the impact of GBA1 variants on PD at the chromatin level by comparing cell type-specific epigenomic profiles of GBA+ with GBA- using ML. Cellformer significantly distinguished the two PD subgroups mainly in the SN, with mean cross-validated AUROC and AUPRC above the random baseline across all five cell types (AUROC>0.64, AUPRC>0.68) (Figure 3A). The two groups were also separated in NEU (AUROC=0.709 and AUPRC=0.706) and OLD (AUROC=0.615 and AUPRC=0.638) in the SMTG region. No significant differences were observed between model performances in males vs. females (Figure S7B). In the SN, cross-validated model prediction scores were significantly different between GBA+ and GBA- (p<0.05) (Figure 3B). NEU was the primary cell type most affected by GBA1 variants with the highest prediction scores (AUROC=0.8, AUPRC=0.8, p<0.001) followed by AST and OLD (p<0.001) outperforming model performance in MIC and OPC (p<0.01).

Figure 3. Cell-type specific profile of epigenomic impact of GBA1 variants in PD brain.

Figure 3.

Open in a new tab

(A) A machine learning (ML) model significantly separates GBA- from GBA+ in SN across all cell types, with the best performance obtained in NEU (AUROC=0.8, AUPRC=0.8). (B) Cross-validated predictions of ML model across cell type in SN showed significant differences between GBA- and GBA+ groups. (C) Inverse correlations between Cellformer’s model errors and total neuropathologic score were present across several brain cell types, including NEU, AST, MIC and OLD. (D) The number of differentially accessible CARs between GBA- and GBA+ across brain regions is dominated by the SN. (E) CARs distinguishing GBA- and GBA+ were seen largely in the intrionic and distal regions. (F) Pathway analysis highlights the significant downregulation of processes associated with dopaminergic and neuroactive ligand-receptor interactions in GBA+ (FDR<0.05). *p<0.05, **p<0.01, ***p<0.001, ***p<0.0001

Abbreviations: AST, astrocyte; CAR, chromatin accessible regions; CAUD, caudate; FDR, false discovery rate; GBA, glucoceribrosidase gene-associated PD; HIPP, hippocampus; MDFG, middle frontal gyrus; MIC, microglia; ML, machine learning; NEU, neuronal cell; OLD, oligodendrocyte; OPC: oligodendrocyte precursor cells; PD, Parkinson’s disease; PTMN, putamen; SMTG, superior and middle temporal gyri; SN, substantia nigra

We next investigated the effect of GBA1 variants in the SN on neuropathologic progression by comparing Cellformer error among SN samples and increasing ordinal rankings of total neuropathologic score as calculated by summing the ordinal NFT score (ranging from 0 to 3) and ordinal LB stage (ranging from 0 to 3) [12], and used this score as a surrogate marker of disease progression. Interestingly, we found that Cellformer inversely correlated with the neuropathologic score in GBA+ across all cell types, with Pearson correlation coefficients ranging from −0.23 in OPC to −0.64 in OLD. This correlation was statistically significant in AST, NEU, and OLD (p values <0.05) (Figure 3C). These results underscore the dynamic effects of GBA1 variants across cell types in disease progression and suggest that GBA+ diverges from GBA- as neuropathological load increases.

To interpret model prediction, we applied univariate differential accessibility analysis across the six brain regions. More than 88% of the differentially accessible CARs between GBA- and GBA+ are in the SN, highlighting the ability of the ML model to distinguish the two PD subgroups in the SN, while only a small percentage of differentially accessible CARs were found in MDFG (8%) and HIPP (2%) (Figure 3D). CARs distinguishing GBA+ from GBA- were mostly found in intronic (38%−53%) and distal (15–27%) followed by promoter and exonic regions (4%−9%) (Figure 3E). Pathway Gene Set Enrichment Analysis [15] highlighted significant enrichment downregulation of processes associated with dopaminergic (FDR<1×10−5) and neuroactive ligand-receptor interactions in NEU (FDR<0.005), including the promoter associated with DRD3 (Dopamine Receptor D3). No pathways were found to be enriched in other cell types (Figure 3F).

Independent validation with blood cell transcriptomic profile of GBA1 variants in PD

We sought to validate our findings by leveraging all available whole-transcriptome RNA-seq data from blood cells of PPMI participants categorized as GBA- (n=207) or GBA+ (n=44). A lower number of females were again noted in both groups. No age difference, was observed between GBA- and GBA+; however, the GBA- group had significantly shorter disease duration than the GBA+ group (Mann-Whitney p<0.0001) (Table S2).

Given that long-read RNA-seq data can be analyzed at both the gene and transcript levels, we assessed the performance of ML models in distinguishing among PD subgroups using both data types. Our results demonstrated that transcript-level analysis provided superior predictive power, yielding high cross-validated performance metrics. Specifically, models trained on transcript-level data achieved very high cross-validation accuracy in separating GBA- from GBA+ (AUROC = 0.99; AUPRC = 0.92) (Figure S8). Additionally, these models consistently outperformed random baselines.

Model error analysis revealed that the number of years of education was the primary covariate inversely correlated with model error in both GBA- (Pearson correlation = −0.18, P-value = 1.2 × 10−6) and GBA+ (Pearson correlation = −0.18, P-value = 0.17) groups (Figure S9). Additionally, model error was significantly higher in GBA+ individuals with a first-degree family history of PD compared to those with no family history (Figure S10). Conversely, no significant differences in model predictions were observed when stratified by sex or dopamine replacement treatment status (Figure S11).

To interpret model predictions, we performed univariate differential expression analysis, identifying 5,216 transcripts that were significantly differentially expressed between GBA- and GBA+ (adjusted DESeq2 P-value < 0.05 and |log₂ fold change| > 0.5). Among the 5,216 signals, we identified two associated with CACNA1, four with EHMT1, one with SLC25A48, two with PIN1, and one with GOLGA3 downregulated in GBA+, supporting prior epigenomic findings in brain (Figure 4A). Pathway enrichment analysis of genes associated with these transcripts revealed significant enrichment for pathways linked to synaptic function, calcium signaling, aldosterone processes, and ventricular cardiomyopathy, validating our prior epigenomic signatures in brain regions (FDR < 0.05) (Figure 4B).

Figure 4: Transcriptomic differences between PD subgroups in human blood.

Figure 4:

Open in a new tab

(A) Based on univariate differential expression analysis, we identified targets that separated GBA- and GBA+ in the PPMI cohort (adjusted p-values shown). (B) Pathway enrichment analysis of differentially expressed transcripts between GBA- and GBA+ groups.

***p<0.001.

Abbreviations: GBA1; glucocerebrosidase gene-associated PD; PPMI, Parkinson’s Progression Markers Initiative

DISCUSSION

Our study investigated the brain regional, cell-type specific epigenomic changes in people with PD with and without GBA1 risk variants, using deep epigenomic profiling of human brain samples with verified LBD neuropathologic change. Here, we identified 19 CARs across six genes that separated GBA+ and GBA-, described GBA1 variant epigenomic impacts across multiple cell types in the context of PD, and validated our results by analyzing data from independent blood samples obtained and assayed by the PPMI. Importantly, the validation of CARs in the SN with blood cell RNA transcript levels of proximate genes importantly supports that at least some of the epigenetic changes observed mostly in the SN are not simply a consequence of ongoing neurodegeneration, which occurs in the SN but not in blood cells in PD. Moreover, these results suggest that these altered CARs may be a systemic feature in GBA1 variant carriers that, by as yet unclear mechanisms, manifest most prominently in SN among the six brain regions investigated. To our knowledge, this is the first study to compare GBA- and GBA+ groups in PD by analyzing bulk ATAC-seq data from multiple brain regions in an autopsy sample. Although a previous study did examine cell type-specific open chromatin accessibility in GBA+ individuals, this study was limited to two individuals per group and to frontal brain regions [16]. Here, we were able to extend this previous research, which found associations between OPC/OLD and PD risk loci), by demonstrating differences between GBA- and GBA+ across all five cell types, primarily in the SN. These novel results provide a springboard for future investigations into epigenomic variance and individualized treatment in PD.

Our work highlighted SN as the brain region most impacted epigenetically by GBA1 variants compared to the other five brain regions examined. Functional analysis of these CARs highlighted alterations in synaptic and calcium signaling pathways. Importantly, we deliberately designed our study to include participants who were clinically diagnosed with PD and verified to have LBD neuropathologic change, rather than including all GBA1 risk variant carriers regardless of diagnosis. As a result, it is less likely that the epigenetic differences are due to standard-of-care treatments or secondary changes of PD, such as decreased mobility. These results provide strong epigenetic evidence that GBA- and GBA+ fundamentally differ in chromatin state and regulation of gene expression, and thereby likely will require different approaches to development of precision medicines that target genetically defined subgroups of PD.

Among the key 19 CARs identified in SN afflicted by PD, EHMT1, CACNA1C, SLC25A48, PIN1, and GOLGA3 were strongly validated by an orthogonal technique applied to blood cell samples from different patients who were diagnosed by different physicians and that were independently collected and assayed. EHMT1 is a histone methyltransferase that regulates gene expression through methylation of histone H3 at lysine 9 (H3K9me2), and has been found to play a key role in synaptic dysfunctions in AD and PD mouse models [17, 18]. Our data add to the body of evidence supporting development of histone deacetylase inhibitors for the treatment of neurodegenerative disorders [19]. CACNA1C encodes a calcium channel subunit essential for proper neuronal function [20]. CACNA1C-associated SNPs have been linked to an increased risk of PD in vitamin D-deficient patients [21]. SLC25A48 (Solute Carrier Family 25 Member 48) was among the top hits in GWAS study involving PD in an Ashkenazi Jewish population [22], and encodes an inner mitochondrial transporter expressed widely throughout the body and brain that influences plasma choline levels [23]. Interestingly, the observed altered expression of PIN1 is supported by evidence showing its up-regulation in the midbrain of PD patients [24], which has been implicated with increased α-synuclein aggregation [25] and proapoptotic dopaminergic neurotoxicity [24]. GOLGA3 gene product is involved in vesicular transportation at the Golgi apparatus and increased expression might be the result of compensatory mechanisms taking place in GBA1 variant cases [26].

Deconvolution offers an efficient tool to study cell type-specific effects while benefiting from the cost-effective and technologically robust advantages of bulk sequencing. This enables a reliable investigation of the cell type vulnerability in disease, using a larger cohort size that more comprehensively reflects the heterogeneity among individuals and reduces false positive discoveries. To handle the large number of features, we used a machine learning framework more adapted to high-dimensional data analysis [27]. We found that GBA1 variants affect the five main brain cell types in people with PD but most extensively neuronal cells. Pathway analysis clearly pinpointed reduced chromatin accessibility associated with genes involved in dopaminergic synapse function and neuroactive ligand-receptor interactions in GBA+. Our results align with previous findings demonstrating the reduction of synaptic activity in dopaminergic neurons derived from induced pluripotent stem cells (iPSCs) of E326K-GBA1 carriers with PD compared to individuals without a GBA risk variant [28].

Longitudinal data from GBA1 variant carriers with PD show average faster disease progression and cognitive decline than in people with sporadic PD [29]. However, directly investigating the epigenomic changes driving the rate of PD progression is not possible as it would require longitudinal brain autopsy samples. Thus, we used ordinal rankings of neuropathologic severity as a surrogate marker for disease progression. Interestingly, we found that our model significantly separated GBA+ from GBA-, especially at a later stage in neuropathologic progression marked by more advanced LBD and NFT. These results at the molecular level support previous findings from longitudinal studies that showed the accelerating effect of GBA1 variants on progression of motor impairment and cognitive decline [30].

Our study has some limitations. First, the autopsy sample size is small due to the overall lower prevalence of GBA1 variant carriers. Within this small sample, the number of individual samples available for each brain region differs, which may in part influence the results. A further limitation of the small sample size was that we were unable to adequately explore the epigenomic profile of other genes known to be associated with cognition in PD, including LRRK2, a causal PD gene. Second, bulk data obtained from samples originating from different brain banks may also affect our findings due to potential differences in collection and processing. A third limitation of this study is the potential confounding presence of other known PD risk factors that were not screened in these cases. Fourth, the inclusion of both pathogenic and nonpathogenic variants of GBA1 may introduce heterogeneity into the study, as these are associated with different levels of disease penetrance and severity. GBA1 risk variants are commonly grouped in PD research due to relatively small sample sizes, particularly in autopsy cohorts. However, the limited number of GBA1 pathogenic variant carrier samples prevents in-depth analysis and robust identification of the impact of individual variants or of heterozygous vs. biallelic variant carriers on brain regional epigenomic profile in PD. Nevertheless, our inference on CARs in SN in the combined group of GBA1 variants demonstrates a notable separation between GBA- and GBA+ groups. Future work as larger sample sets accrue will stratify analyses to understand the effect of specific GBA1 variants. Finally, the resolution of Cellformer restricts the depth of the analysis to the five main cell types of the brain, preventing a deep understanding of the neuronal and glial heterogeneity; this partially derives from the lack of available high-quality and deeply annotated single-nuclei ATAC-seq data from human brain to train Cellformer.

We analyzed bulk ATAC-seq and applied Cellformer to obtain insight into chromatin accessibility in five different cell types in six regions of the brain from people with PD with and without known genetic risk for PD and cognitive impairment. Our results show that the epigenomic landscape in brain of people with PD varies with underlying genetic risk, highlighting specific cell types and brain regions. These findings, independently validated using long-read RNA-seq data, provide further evidence that genetic risk stratification and precision therapeutics likely will be needed to treat motor and non-motor features of PD.

Supplementary Material

2

HIGHLIGHTS.

  • Nineteen chromatin-accessible regions were associated with GBA1 Parkinson’s disease

  • These regions are largely associated with calcium-dependent signaling pathways

  • The effects in GBA1 Parkinson’s disease spanned the main brain cell types

  • Chromatin differences in GBA1 Parkinson’s disease increased with disease severity

  • Validation in blood supports that these altered regions may be systemic in GBA1

ACKNOWLEDGMENTS

We are grateful to the Banner Sun Health Research Institute Brain and Body Donation Program of Sun City, Arizona for the provision of human biological materials. The Brain and Body Donation Program has been supported by the National Institute of Neurological Disorders and Stroke (U24 NS072026), National Brain and Tissue Resource for Parkinson’s Disease and Related Disorders, the National Institute on Aging (P30 AG019610 and P30AG072980, Arizona Alzheimer’s Disease Center), the Arizona Department of Health Services (contract 211002, Arizona Alzheimer’s Research Center), the Arizona Biomedical Research Commission (contracts 4001, 0011, 05–901 and 1001 to the Arizona Parkinson’s Disease Consortium) and the Michael J. Fox Foundation for Parkinson’s Research. Schematics were created with BioRender.com.

FUNDING

This work was supported by the National Institutes of Health: GM138353 (N.A and T.J.M), NIH RF1 AG077443 (T.J.M), U01 U01AG072573 (T.J.M, M.R.C), P01AG073082 (M.R.C). RM1-HG007735 (H.Y.C.). E.B. is funded by the Phil & Penny Knight Initiative for Brain Resilience at the Wu Tsai Neurosciences Institute, Stanford University. H.Y.C. is an Investigator of the Howard Hughes Medical Institute.

Footnotes

Declaration of interests

The authors declare the following financial interests/personal relationships which may be considered as potential competing interests:

Eloise Berson reports financial support was provided by National Institutes of Health. Nima Aghaeepour reports financial support was provided by National Institutes of Health. Thomas J. Montine reports financial support was provided by National Institutes of Health. M. Ryan Corces reports financial support was provided by National Institutes of Health. Howard Y. Chang reports financial support was provided by National Institutes of Health. Eloise Berson reports financial support was provided by Phil & Penny Knight Initiative for Brain Resilience. Howard Y. Chang reports financial support was provided by Howard Hughes Medical Institute. Nima Aghaeepour reports a relationship with MaraBio Systems that includes: consulting or advisory. Nima Aghaeepour reports a relationship with Takeoff AI that includes: board membership. Nima Aghaeepour reports a relationship with January AI that includes: board membership. Nima Aghaeepour reports a relationship with Parallel Bio that includes: board membership. Nima Aghaeepour reports a relationship with Celine Therapeutics that includes: board membership. Nima Aghaeepour reports a relationship with WellSim Biomedical Technologies that includes: board membership. If there are other authors, they declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

COMPETING INTEREST STATEMENT

N.A. is a cofounder of Takeoff AI, a member of the Scientific Advisory Boards of January AI, Parallel Bio, Celine Therapeutics, and WellSim Biomedical Technologies and is a paid consultant for MaraBio Systems.

Publisher's Disclaimer: This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.

DATA STATEMENT

The raw and processed data and the code developed for this study are available through GEO accession (GSE273511). Code and trained Cellformer model are available at https://github.com/elonsrb/Cellformer/tree/main/cellformer_PD. The code used to reproduce this analysis and generate the figures is publicly available at https://github.com/elo-nsrb/PD-epigenomic-analysis. Additionally, human blood bulk data and all the information of the individuals used in the preparation of this article was obtained on 2024–12-05 from the Parkinson’s Progression Markers Initiative (PPMI) database (www.ppmi-info.org/access-dataspecimens/download-data), RRID:SCR_006431. For up-to-date information on the study, visit www.ppmi-info.org.

REFERENCES

  • [1].Bock MA, Tanner CM, The epidemiology of cognitive function in Parkinson’s disease, Prog Brain Res 269(1) (2022) 3–37. 10.1016/bs.pbr.2022.01.004 [DOI] [PubMed] [Google Scholar]
  • [2].Skrahin A, Horowitz M, Istaiti M, Skrahina V, Lukas J, Yahalom G, Cohen ME, Revel-Vilk S, Goker-Alpan O, Becker-Cohen M, Hassin-Baer S, Svenningsson P, Rolfs A, Zimran A, GBA1-Associated Parkinson’s Disease Is a Distinct Entity, Int J Mol Sci 25(13) (2024). 10.3390/ijms25137102 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • [3].Nishioka K, Ross OA, Vilarino-Guell C, Cobb SA, Kachergus JM, Mann DM, Snowden J, Richardson AM, Neary D, Robinson CA, Rajput A, Papapetropoulos S, Mash DC, Pahwa R, Lyons KE, Wszolek ZK, Dickson DW, Farrer MJ, Glucocerebrosidase mutations in diffuse Lewy body disease, Parkinsonism Relat Disord 17(1) (2011) 55–7. 10.1016/j.parkreldis.2010.09.009 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • [4].Gegg ME, Burke D, Heales SJ, Cooper JM, Hardy J, Wood NW, Schapira AH, Glucocerebrosidase deficiency in substantia nigra of parkinson disease brains, Ann Neurol 72(3) (2012) 455–63. 10.1002/ana.23614 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • [5].Moors TE, Paciotti S, Ingrassia A, Quadri M, Breedveld G, Tasegian A, Chiasserini D, Eusebi P, Duran-Pacheco G, Kremer T, Calabresi P, Bonifati V, Parnetti L, Beccari T, van de Berg WDJ, Characterization of Brain Lysosomal Activities in GBA-Related and Sporadic Parkinson’s Disease and Dementia with Lewy Bodies, Mol Neurobiol 56(2) (2019) 1344–1355. 10.1007/s12035-018-1090-0 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • [6].Smith AR, Richards DM, Lunnon K, Schapira AHV, Migdalska-Richards A, DNA Methylation of alpha-Synuclein Intron 1 Is Significantly Decreased in the Frontal Cortex of Parkinson’s Individuals with GBA1 Mutations, Int J Mol Sci 24(3) (2023). 10.3390/ijms24032687 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • [7].Kim J, Kim J, Park M, Synthetic interventions in epigenome: Unraveling chromatin’s potential for therapeutic applications, Current Opinion in Systems Biology 37 (2024) 100504. 10.1016/j.coisb.2023.100504 [DOI] [Google Scholar]
  • [8].Buenrostro JD, Giresi PG, Zaba LC, Chang HY, Greenleaf WJ, Transposition of native chromatin for fast and sensitive epigenomic profiling of open chromatin, DNA-binding proteins and nucleosome position, Nat Methods 10(12) (2013) 1213–8. 10.1038/nmeth.2688 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • [9].Corces MR, Trevino AE, Hamilton EG, Greenside PG, Sinnott-Armstrong NA, Vesuna S, Satpathy AT, Rubin AJ, Montine KS, Wu B, Kathiria A, Cho SW, Mumbach MR, Carter AC, Kasowski M, Orloff LA, Risca VI, Kundaje A, Khavari PA, Montine TJ, Greenleaf WJ, Chang HY, An improved ATAC-seq protocol reduces background and enables interrogation of frozen tissues, Nat Methods 14(10) (2017) 959–962. 10.1038/nmeth.4396 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • [10].Berson E, Sreenivas A, Phongpreecha T, Perna A, Grandi FC, Xue L, Ravindra NG, Payrovnaziri N, Mataraso S, Kim Y, Espinosa C, Chang AL, Becker M, Montine KS, Fox EJ, Chang HY, Corces MR, Aghaeepour N, Montine TJ, Whole genome deconvolution unveils Alzheimer’s resilient epigenetic signature, Nat Commun 14(1) (2023) 4947. 10.1038/s41467-023-40611-4 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • [11].Litvan I, Goldman JG, Troster AI, Schmand BA, Weintraub D, Petersen RC, Mollenhauer B, Adler CH, Marder K, Williams-Gray CH, Aarsland D, Kulisevsky J, Rodriguez-Oroz MC, Burn DJ, Barker RA, Emre M, Diagnostic criteria for mild cognitive impairment in Parkinson’s disease: Movement Disorder Society Task Force guidelines, Mov Disord 27(3) (2012) 349–56. 10.1002/mds.24893 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • [12].Hyman BT, Phelps CH, Beach TG, Bigio EH, Cairns NJ, Carrillo MC, Dickson DW, Duyckaerts C, Frosch MP, Masliah E, Mirra SS, Nelson PT, Schneider JA, Thal DR, Thies B, Trojanowski JQ, Vinters HV, Montine TJ, National Institute on Aging-Alzheimer’s Association guidelines for the neuropathologic assessment of Alzheimer’s disease, Alzheimers Dement 8(1) (2012) 1–13. 10.1016/j.jalz.2011.10.007 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • [13].Adler CH, Beach TG, Shill HA, Caviness JN, Driver-Dunckley E, Sabbagh MN, Patel A, Sue LI, Serrano G, Jacobson SA, Davis K, Belden CM, Dugger BN, Paciga SA, Winslow AR, Hirst WD, Hentz JG, GBA mutations in Parkinson disease: earlier death but similar neuropathological features, Eur J Neurol 24(11) (2017) 1363–1368. 10.1111/ene.13395 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • [14].Marek K, Chowdhury S, Siderowf A, Lasch S, Coffey CS, Caspell-Garcia C, Simuni T, Jennings D, Tanner CM, Trojanowski JQ, Shaw LM, Seibyl J, Schuff N, Singleton A, Kieburtz K, Toga AW, Mollenhauer B, Galasko D, Chahine LM, Weintraub D, Foroud T, Tosun-Turgut D, Poston K, Arnedo V, Frasier M, Sherer T, Parkinson’s Progression Markers I, The Parkinson’s progression markers initiative (PPMI) - establishing a PD biomarker cohort, Ann Clin Transl Neurol 5(12) (2018) 1460–1477. 10.1002/acn3.644 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • [15].Mootha VK, Lindgren CM, Eriksson KF, Subramanian A, Sihag S, Lehar J, Puigserver P, Carlsson E, Ridderstrale M, Laurila E, Houstis N, Daly MJ, Patterson N, Mesirov JP, Golub TR, Tamayo P, Spiegelman B, Lander ES, Hirschhorn JN, Altshuler D, Groop LC, PGC-1alpha-responsive genes involved in oxidative phosphorylation are coordinately downregulated in human diabetes, Nat Genet 34(3) (2003) 267–73. 10.1038/ng1180 [DOI] [PubMed] [Google Scholar]
  • [16].Dehestani M, Kozareva V, Blauwendraat C, Fraenkel E, Gasser T, Bansal V, Transcriptomic changes in oligodendrocytes and precursor cells associate with clinical outcomes of Parkinson’s disease, Mol Brain 17(1) (2024) 56. 10.1186/s13041-024-01128-z [DOI] [PMC free article] [PubMed] [Google Scholar]
  • [17].Zhang Z, Wang R, Zhou H, Wu D, Cao Y, Zhang C, Sun H, Mu C, Hao Z, Ren H, Wang N, Yu S, Zhang J, Tao M, Wang C, Liu Y, Liu L, Liu Y, Zang J, Wang G, Inhibition of EHMT1/2 rescues synaptic damage and motor impairment in a PD mouse model, Cell Mol Life Sci 81(1) (2024) 128. 10.1007/s00018-024-05176-5 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • [18].Zheng Y, Liu A, Wang ZJ, Cao Q, Wang W, Lin L, Ma K, Zhang F, Wei J, Matas E, Cheng J, Chen GJ, Wang X, Yan Z, Inhibition of EHMT1/2 rescues synaptic and cognitive functions for Alzheimer’s disease, Brain 142(3) (2019) 787–807. 10.1093/brain/awy354 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • [19].Kumar V, Kundu S, Singh A, Singh S, Understanding the Role of Histone Deacetylase and their Inhibitors in Neurodegenerative Disorders: Current Targets and Future Perspective, Curr Neuropharmacol 20(1) (2022) 158–178. 10.2174/1570159X19666210609160017 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • [20].Smedler E, Louhivuori L, Romanov RA, Masini D, Dehnisch Ellstrom I, Wang C, Caramia M, West Z, Zhang S, Rebellato P, Malmersjo S, Brusini I, Kanatani S, Fisone G, Harkany T, Uhlen P, Disrupted Cacna1c gene expression perturbs spontaneous Ca(2+) activity causing abnormal brain development and increased anxiety, Proc Natl Acad Sci U S A 119(7) (2022). 10.1073/pnas.2108768119 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • [21].Wang L, Maldonado L, Beecham GW, Martin ER, Evatt ML, Ritchie JC, Haines JL, Zabetian CP, Payami H, Pericak-Vance MA, Vance JM, Scott WK, DNA variants in CACNA1C modify Parkinson disease risk only when vitamin D level is deficient, Neurol Genet 2(3) (2016) e72. 10.1212/NXG.0000000000000072 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • [22].Liu X, Cheng R, Verbitsky M, Kisselev S, Browne A, Mejia-Sanatana H, Louis ED, Cote LJ, Andrews H, Waters C, Ford B, Frucht S, Fahn S, Marder K, Clark LN, Lee JH, Genome-wide association study identifies candidate genes for Parkinson’s disease in an Ashkenazi Jewish population, BMC Med Genet 12 (2011) 104. 10.1186/1471-2350-12-104 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • [23].Bernard DJ, Pangilinan F, Mendina C, Desporte T, Wincovitch SM, Walsh DJ, Porter RK, Molloy AM, Shane B, Brody LC, SLC25A48 influences plasma levels of choline and localizes to the inner mitochondrial membrane, Mol Genet Metab 143(1–2) (2024) 108518. 10.1016/j.ymgme.2024.108518 [DOI] [PubMed] [Google Scholar]
  • [24].Ghosh A, Saminathan H, Kanthasamy A, Anantharam V, Jin H, Sondarva G, Harischandra DS, Qian Z, Rana A, Kanthasamy AG, The peptidyl-prolyl isomerase Pin1 up-regulation and proapoptotic function in dopaminergic neurons: relevance to the pathogenesis of Parkinson disease, J Biol Chem 288(30) (2013) 21955–71. 10.1074/jbc.M112.444224 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • [25].Jeong J, Usman M, Li Y, Zhou XZ, Lu KP, Pin1-Catalyzed Conformation Changes Regulate Protein Ubiquitination and Degradation, Cells 13(9) (2024). 10.3390/cells13090731 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • [26].Wei Y, Awan MUN, Bai L, Bai J, The function of Golgi apparatus in LRRK2-associated Parkinson’s disease, Front Mol Neurosci 16 (2023) 1097633. 10.3389/fnmol.2023.1097633 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • [27].Dunkler D, Sanchez-Cabo F, Heinze G, Statistical analysis principles for Omics data, Methods Mol Biol 719 (2011) 113–31. 10.1007/978-1-61779-027-0_5 [DOI] [PubMed] [Google Scholar]
  • [28].Rosh I, Tripathi U, Hussein Y, Rike WA, Djamus J, Shklyar B, Manole A, Houlden H, Winkler J, Gage FH, Stern S, Synaptic dysfunction and extracellular matrix dysregulation in dopaminergic neurons from sporadic and E326K-GBA1 Parkinson’s disease patients, NPJ Parkinsons Dis 10(1) (2024) 38. 10.1038/s41531-024-00653-x [DOI] [PMC free article] [PubMed] [Google Scholar]
  • [29].Cilia R, Tunesi S, Marotta G, Cereda E, Siri C, Tesei S, Zecchinelli AL, Canesi M, Mariani CB, Meucci N, Sacilotto G, Zini M, Barichella M, Magnani C, Duga S, Asselta R, Solda G, Seresini A, Seia M, Pezzoli G, Goldwurm S, Survival and dementia in GBA-associated Parkinson’s disease: The mutation matters, Ann Neurol 80(5) (2016) 662–673. 10.1002/ana.24777 [DOI] [PubMed] [Google Scholar]
  • [30].Szwedo AA, Dalen I, Pedersen KF, Camacho M, Backstrom D, Forsgren L, Tzoulis C, Winder-Rhodes S, Hudson G, Liu G, Scherzer CR, Lawson RA, Yarnall AJ, Williams-Gray CH, Macleod AD, Counsell CE, Tysnes OB, Alves G, Maple-Grodem J, C. Parkinson’s Incidence Cohorts, GBA and APOE Impact Cognitive Decline in Parkinson’s Disease: A 10-Year Population-Based Study, Mov Disord 37(5) (2022) 1016–1027. 10.1002/mds.28932 [DOI] [PMC free article] [PubMed] [Google Scholar]

Associated Data

This section collects any data citations, data availability statements, or supplementary materials included in this article.

Supplementary Materials

2
NIHMS2116230-supplement-2.docx (6.5MB, docx)

Data Availability Statement

The raw and processed data and the code developed for this study are available through GEO accession (GSE273511). Code and trained Cellformer model are available at https://github.com/elonsrb/Cellformer/tree/main/cellformer_PD. The code used to reproduce this analysis and generate the figures is publicly available at https://github.com/elo-nsrb/PD-epigenomic-analysis. Additionally, human blood bulk data and all the information of the individuals used in the preparation of this article was obtained on 2024–12-05 from the Parkinson’s Progression Markers Initiative (PPMI) database (www.ppmi-info.org/access-dataspecimens/download-data), RRID:SCR_006431. For up-to-date information on the study, visit www.ppmi-info.org.

RESOURCES