Whole transcriptome analysis of peripheral blood mononuclear cells from de novo and drug-naive Parkinson's disease patients

Abstract

Blood-based biomarkers represent a highly convenient and minimally invasive method to improve the diagnosis of Parkinson's disease (PD), particularly at early stages. In this study, we present a comprehensive analysis of transcriptional changes in peripheral blood mononuclear cells (PBMCs) from de novo and drug-naive, recently diagnosed PD patients (n ​= ​23) and healthy sex- and age-matched controls (n ​= ​16). Whole-transcriptome analysis revealed differentially expressed genes (DEGs) in PBMCs from PD patients, including genes with immune-related functions such as CHI3L1, FAM198B, ID3, MDX1, and PROK2. Gene set enrichment analysis (GSEA) further identified alterations in immune pathways, including the IL-6/JAK/STAT3 signaling pathway and the complement cascade, associated with PD. We performed cross-validation using two independent whole-blood transcriptomic datasets from PD patients to assess the reproducibility and biological relevance of our findings. Both comparisons showed moderate but highly significant correlations in gene expression patterns. Overall, our results reveal robust and reproducible transcriptional alterations in PBMCs from early-stage PD patients, underscoring the contribution of immune dysregulation to PD pathogenesis. These findings support the potential of PBMC transcriptomics as a valuable platform for biomarker discovery in Parkinson's disease.

Introduction

Parkinson's disease (PD) is a clinical motor syndrome associated with neurodegeneration of the substantia nigra pars compacta and deposition of alpha-synuclein [1,2]. In recent decades, PD's incidence and associated social and clinical impact have increased dramatically [3]. PD is the second most common neurodegenerative disease and one of the leading causes of neurological disability [4,5]. According to the latest systematic analysis of the global disease burden of the nervous system in 2021, the prevalence of PD was 11.8 million patients, 2.3 million more than the previous estimate in 2019. It was responsible for 388,000 deaths and 7.47 million disability-adjusted life years (DALYs).

The diagnosis of PD has relied mainly on the history of motor and non-motor clinical features typical of patients with this condition. However, it can be mistaken for other neurological diseases. A definitive and accurate diagnosis of PD can only be made by analyzing post-mortem brain tissue [6]. In addition, clinical scales and questionnaires do not provide a definitive diagnosis from the earliest stages nor allow us to accurately determine the prognosis of the clinical course of patients with PD or the most appropriate and successful therapeutic approach [[7], [8], [9]].

There has been increasing reliance on biomarkers to improve clinical diagnosis and quantify the stage of PD from a biological perspective, particularly in the early stages of the disease when clinical manifestations are subtle or mild [10]. A preclinical phase of PD has been proposed, characterized by disease-specific pathology without clinical symptoms or detectable biomarkers. There is consensus that the pathological process leading to clinically manifest PD begins long before it can be identified using current diagnostic criteria. However, no biomarkers have been established to reliably detect these preclinical stages with high sensitivity and specificity [6]. Early detection of individuals in the preclinical phase of PD, particularly those with genetic predispositions or early non-motor symptoms, remains a significant challenge. The identification of sensitive and specific biomarkers could transform how patients are classified at the time of initial diagnosis, offering the possibility of personalizing treatment and predicting disease progression to take appropriate action and, hopefully, soon, to implement disease-modifying therapies [11].

Recently, the idea of establishing a biological classification for PD has gained importance based on motor and non-motor clinical features, including a neuropathological component mainly related to the accumulation of alpha-synuclein in fluids such as cerebrospinal fluid (CSF), a neurodegenerative phenomenon demonstrated by neuroimaging studies, and the identification of pathogenic genetic variants related to the etiology or risk of the disease [12]. In this context, using minimally invasive techniques, such as simple blood collection combined with omics-based transcriptome analysis, offers a valuable approach to identifying blood-based biomarkers for PD [13,14]. The potential is even more significant when studying newly diagnosed (de novo) patients who have not yet begun pharmacological treatment (drug-naive), as this minimizes bias from the effects of dopaminergic medications, thereby enhancing the identification of early diagnostic biomarkers [15].

Thus, the primary goal of this study was to evaluate the transcriptome in peripheral blood mononuclear cells (PBMCs) from de novo and drug-naive PD patients, paired with control subjects, to identify potential gene expression alterations that could reflect the primary targets, systems, and phenomena encompassing the first stages of PD. The results may provide valuable information regarding identifying biomarkers that could improve the diagnosis, prognosis, monitoring, and treatment of patients with PD.

Materials and methods

Subjects

Blood samples from 23 de novo and drug-naive, recently diagnosed PD patients and 16 healthy sex- and age-matched controls (CT) were recruited from the neurology services of the 12 de Octubre University Hospital of the Community of Madrid (Spain) and the General University Hospital of Alicante (Spain). All patients fulfilled the diagnostic criteria for PD disease according to the Hoehn and Yahr scale (stages I-III), which evaluates functional disability associated with PD [16]. Although no genetic testing was performed, a family history of PD was excluded through self-reporting and review of the clinical history to avoid including genetic forms of the disease. CT patients were included in the study only if they had no personal or family history of neurological or psychiatric disorders. The main demographic and clinical characteristics of the CT and PD patients are shown in Table 1, Table 2. This study was approved by the Ethics Committees of the 12 de Octubre University Hospital, the General University Hospital of Alicante, and Miguel Hernández University. Written informed consent was obtained from all study participants following the Declaration of Helsinki.

Table 1.

Demographic data, origin and RIN value from samples of CT subjects. 12OUH: 12 Octubre University Hospital, GUHA: General University Hospital of Alicante, RIN: RNA integrity number, SD: standard deviation.

Subject ID	Age	Gender	Origin	RIN value
CT 1	66	Female	12OUH	7.3
CT 2	60	Female	12OUH	8.6
CT 3	61	Female	GUHA	7.4
CT 4	58	Male	GUHA	8.4
CT 5	66	Female	12OUH	8.7
CT 6	69	Male	12OUH	8.7
CT 7	68	Male	12OUH	8.7
CT 8	49	Male	12OUH	7.8
CT 9	47	Female	12OUH	8.8
CT 10	49	Female	12OUH	6.4
CT 11	62	Female	12OUH	6.5
CT 12	50	Female	12OUH	6.3
CT 13	55	Male	12OUH	6.3
CT 14	45	Male	12OUH	6.4
CT 15	52	Male	12OUH	6.3
CT 16	66	Male	12OUH	7.5
n ​= ​16	Mean: 57.69 SD: 8.17	Females, n ​= ​8 Males, n ​= ​8	12OUH, n ​= ​14 GUHA, n ​= ​2	Mean: 7.51 SD: 1.03

Table 2.

Demographic and clinical data, origin and RIN value from samples of de novo and drug-naive PD subjects. 12OUH: 12 Octubre University Hospital, GUHA: General University Hospital of Alicante, RIN: RNA integrity number, SD: standard deviation.

Subject ID	Age	Gender	Hoehn and Yahr stage	Origin	RIN value
PD 1	71	Male	I	12OUH	7.6
PD 2	72	Female	II	12OUH	7.7
PD 3	62	Female	I	12OUH	7.4
PD 4	65	Male	I	12OUH	6.7
PD 5	82	Male	II	12OUH	8.2
PD 6	75	Male	I	12OUH	7.8
PD 7	106	Male	II	12OUH	8.2
PD 8	81	Female	II	12OUH	8.4
PD 9	74	Male	I	12OUH	8.1
PD 10	71	Male	I	GUHA	9.7
PD 11	69	Male	II	GUHA	9.6
PD 12	70	Female	I	GUHA	9.7
PD 13	77	Male	II	12OUH	9.3
PD 14	63	Female	I	12OUH	5.1
PD 15	71	Female	III	12OUH	7.3
PD 16	84	Female	II	12OUH	8.1
PD 17	82	Female	II	12OUH	8.9
PD 18	78	Female	I	12OUH	6.5
PD 19	42	Female	I	GUHA	9.4
PD 20	55	Female	I	GUHA	9.1
PD 21	56	Male	II	GUHA	9.1
PD 22	47	Male	I	12OUH	8.0
PD 23	71	Male	II	12OUH	6.4
n ​= ​23	Mean: 70.61 SD: 13.36	Females, n ​= ​11 Males, n ​= ​12	H&Y I, n ​= ​12 H&Y II, n ​= ​10 H&Y III, n ​= ​1	12OUH, n ​= ​17 GUHA, n ​= ​6	Mean: 8.10 SD: 1.19

PBMCs isolation

PBMCs, including lymphocytes, monocytes and dendritic cells, were isolated from blood samples of de novo and drug-naive PD and CT patients (10 ​ml) using BD Vacutainer® CPT™ Cell Preparation Tubes with Sodium Heparin, following the manufacturer's protocol (BD Biosciences, Spain), and then frozen and kept at −80 ​°C.

RNA extraction, processing, and hybridization

Total RNA from PBMCs was extracted using TRI Reagent extraction reagent (Applied Biosystems, Madrid). The RNA integrity number (RIN) was determined using the 2100 Bioanalyzer (Agilent, Spain) before performing whole transcriptome analyses. RNA samples meeting the quality criteria were analyzed for gene expression using the Clariom S Array (Human, Affymetrix), according to the manufacturer's guidelines. This array quantifies gene-level expression for 21448 gene features. Hybridization was performed in-house by the Multigenic Analysis Section at the Central Research Unit, Faculty of Medicine, Valencia.

Data processing

Data analysis was conducted in the R computing environment (version 4.4.2) using the Bioconductor (version 3.20) suite to investigate gene expression differences in PD. Raw.CEL files were imported, and initial quality control checks were performed using the oligo package [17] and the Transcriptome Analysis Console (TAC) software (version 4.0.3, ThermoFisher Scientific). Quality control metrics were evaluated, resulting in the exclusion of one of the 39 initial samples due to failure of labeling control thresholds.

Robust Multi-array Average (RMA) normalization was applied to correct for background noise, normalize signal intensities, and summarize expression values in a modular workflow (Fig. 1). Probes were annotated using the Clariom_S_Human.r1.na36.hg38.a1.transcript.csv reference file, and probes that did not map to annotated genes were excluded. Following Klaus and Reisenauer's (2016) [16] guidelines, lowly expressed genes were filtered out to enhance data reliability. The final dataset consisted of 38 samples, characterized by 18,091 probes.

Fig. 1.

Quality control of microarray data before (A, B) and after (C, D) RMA normalization. The panels show boxplots of observed intensities (A, C) and Kernel density estimators of the microarrays' intensity distributions (C, D). The normalization process improved the consistency and comparability of intensity measurements between arrays.

Cell type deconvolution analysis

To estimate the relative abundance of immune cell subtypes in PBMC samples, we performed cell type deconvolution using CIBERSORTx [18], a digital cytometry tool based on linear v-support vector regression. The normalized and filtered gene expression matrix described above was uploaded to the CIBERSORTx platform (https://cibersortx.stanford.edu). Deconvolution was performed using the LM22 signature matrix, which profiles 22 distinct human immune cell types [19]. The immune cell composition analyses were performed with 1000 permutations using the default parameters. Batch correction was disabled, as all arrays were processed under consistent conditions.

The estimated cell type proportions were then compared between the PD and CT groups to explore potential differences in immune cell composition associated with PD.

Data analysis

Differential expression analysis was performed using the limma package [20] to compare PD and CT samples. Linear models were fitted for each gene to estimate group-specific effects, and an empirical Bayes approach was employed to stabilize variance estimates. In light of the principal component analysis (PCA), which revealed that part of the variability in the data was explained by sex, the linear models were adjusted to include sex as a covariate. This adjustment allowed us to assess the differences between PD and CT samples while controlling for potential confounding effects due to sex. Genes were ranked according to their adjusted p-values, which were calculated using the False Discovery Rate (FDR) method to control for multiple testing [21]. Genes with an adjusted p-value <0.1 and an absolute log₂ fold change >0.5 were defined as differentially expressed genes (DEGs) and selected for further biological interpretation.

GSEA [22] was conducted to evaluate microarray data at the level of predefined gene sets. This analysis aimed to identify groups of genes exhibiting altered expression patterns between the PD and CT groups. Gene sets from the Hallmark database encapsulated specific, well-defined biological states or processes and demonstrated coherent expression patterns. Significantly enriched gene sets were identified using an FDR threshold of ≤0.25, thereby highlighting pathways that may be involved in the pathology of PD. GSEA plots were generated using the ReplotGSEA function by Thomas Kuilman (https://github.com/PeeperLab/Rtoolbox/blob/master/R/ReplotGSEA.R).

Validation of transcriptional changes in PD

To evaluate the reproducibility and biological relevance of our differential expression findings, we compared our PBMC-derived transcriptomic data with findings from two independent studies. Specifically, we examined overlaps and consistency with results from Craig et al. (2021) [23] and Calligaris et al. (2015) [24].

To compare with Craig et al., which utilized RNA sequencing of whole-blood samples from PD patients, we aligned gene symbols between the two datasets and generated scatter plots of log₂ fold changes. The correlation between our data and Craig et al.’s data was quantified using Pearson's correlation coefficient. Furthermore, we visualized concordance using a volcano plot from the Craig et al. dataset, which examined differential expression of RNA species between individuals from the Parkinson's Progression Marker Initiative (PPMI) cohort with and without a PD diagnosis. Genes identified as DEGs in our analysis were highlighted and color-coded to indicate the direction of expression changes: upregulated in both studies, downregulated in both studies or showing opposite (anticorrelated) expression patterns.

For the comparison with Calligaris et al., which profiled whole-blood transcriptomes of early-stage, drug-naive sporadic PD patients using Affymetrix microarrays, we first filtered DEGs reported by those authors using a log₂ fold change threshold (|log₂FC| ​> ​1), in line with the original study's criteria. We then matched gene symbols across datasets and compared log₂ fold changes obtained in both studies. Pearson's correlation coefficient was computed to quantify the concordance between expression changes.

Results

RNA samples from PBMCs were analyzed using transcriptome-wide gene expression microarrays, encompassing 21448 probes, to explore transcriptional changes associated with PD. This comprehensive approach enabled the capture of gene expression differences between PD patients and CT subjects. Differential expression analysis was performed to identify key transcriptional alterations. Additionally, GSEA was conducted to assess coordinated gene expression across functionally related gene sets in PD.

Cell type deconvolution analysis

To explore the cellular composition of PBMC samples from PD patients and CT, we performed a deconvolution analysis using CIBERSORTx with the LM22 signature matrix, a reference gene expression matrix explicitly designed for immune cell deconvolution [18]. This approach enabled profiling of the immune cell landscape in both groups (Fig. 2, Supplementary Table 1). The model demonstrated strong robustness, with all comparisons reaching high statistical significance. Correlation coefficients ranging from about 0.38 to 0.62 indicated a moderate positive agreement between the predicted and observed cell-type proportions. Furthermore, root mean square error (RMSE) values between 0.78 and 0.93 suggest a reasonably low prediction error, reinforcing the accuracy of the estimated immune cell fractions derived from bulk data. The deconvolution analysis revealed distinct patterns in immune cell composition between the CT and PD groups. Both cohorts exhibited a predominance of CD4⁺ memory T cells, CD8⁺ T cells, NK cells, and monocytes. However, PD samples showed a trend toward increased CD8 T cells, eosinophils, and elevated neutrophils (p ​= ​0.030, t ​= ​2.28, 95 ​% CI [0.018, 0.001]). At the same time, controls presented higher levels of CD4 naive T cells (p ​= ​0.044, t ​= ​−2.10, 95 ​% CI [−0.0015, −0.097]) and slightly greater plasma cell counts (p ​= ​0.006, t ​= ​−2.94, 95 ​% CI [−0.00096, −0.00527]) (Fig. 2, Supplementary Table 1). These differences suggest altered immune cell distributions in PD, consistent with an inflammatory and immune-activated state.

Fig. 2.

Inferred immune cell-type composition per sample using CIBERSORTx.

Differential expression analysis

Transcriptome-wide gene expression profiling revealed transcriptional differences between PD patients and CT subjects. PCA of the top 100 probes demonstrated that the first principal component accounted for 46 ​% of the variance in the data and partially separated samples by pathological condition, highlighting differences between the PD and CT groups (Fig. 3). Notably, the second principal component explained 14.7 ​% of the variance and distinctly segregated samples by sex, indicating strong sex-specific gene expression patterns. Although separation based on disease status was not complete — likely due to high inter-individual variability — the observed patterns suggest distinct gene expression profiles associated with PD.

Fig. 3.

PCA of PD and CT samples. PCA was performed to visualize the significant sources of variance in the dataset. Each point represents an individual sample, plotted according to the first two principal components (PC1 and PC2), which explain 46 ​% and 14.7 ​% of the total variance, respectively. Samples are colored based on different metadata categories: A, Disease condition (Control vs. Parkinson's disease); B, Sex (Female vs. Male); C, Hospital of origin (12OUH vs. GUHA); D, Age (gradient scale); E, RNA Integrity Number (RIN; gradient scale); F, Hoehn and Yahr stages (I–III, N.D.: not determined). The separation along PC1 highlights a disease-associated signature, while other variables contribute only a limited amount to this major axis of variance.

Next, we performed differential expression analysis using the limma package, which applies linear models and an empirical Bayes method, providing a robust framework for assessing differential expression. To avoid confounding effects, sex was included as a covariate in the model, following observations from PCA analysis. This analysis identified 23 differentially expressed probes, corresponding to 22 DEGs between PD patients and controls, with an adjusted p-value <0.1 and an absolute log₂ fold change >0.5 (Fig. 4; Table 3; Supplementary Table 2). Two probes mapped to the same gene (Table 3). A heatmap was generated to visualize the expression profiles of DEGs, illustrating distinct patterns between PD and control PBMC samples and highlighting clusters of genes with coordinated regulation (Fig. 4A). Among the 22 DEGs, 18 genes were upregulated, while 4 were downregulated in the PBMCs of PD patients. These findings are further illustrated in the volcano plot (Fig. 4B), where DEGs are color-coded based on significance and fold change: upregulated genes with an adjusted p-value <0.05 are shown in red, while those with an adjusted p-value between 0.05 and 0.1 are shown in orange; downregulated genes with an adjusted p-value <0.05 are shown in blue, and those with an adjusted p-value between 0.05 and 0.1 are shown in green.

Fig. 4.

Differential gene expression in PD. (A) Heatmap illustrating the expression profiles of DEGs with an adjusted p-value <0.1 and absolute log2 fold change (|log₂FC|) ​> ​0.5 across individual samples. The Z-score color scale denotes relative expression (red: upregulated, blue: downregulated). Genes with p-adj <0.05 are bolded; those with 0.05 ​≤ ​p-adj <0.1 are in regular font. (B) Volcano plot of the differential expression results. The x-axis represents the log₂ fold change, and the y-axis represents the –log₁₀ adjusted p-value (FDR). Significantly upregulated genes with p-adj <0.05 are shown in red; those with p-adj between 0.05 and 0.1 are shown in orange. Downregulated genes with p-adj <0.05 are in blue; those with p-adj between 0.05 and 0.1 are in green. Non-significant genes (p-adj ≥0.1) are gray.

Table 3.

Differentially expressed genes between PD patients and controls identified by limma analysis.

Probe ID	Log FC	Average Expression	p-value	Adjusted p-value	Gene symbol
TC0900009860.hg.1	−0,54493	7,35466	1,88E-09	3,41E-05	AQP3
TC0800009437.hg.1	2,03373	7,95867	1,28E-06	7,71E-03	DEFA1B
TC0800012381.hg.1	2,03373	7,95867	1,28E-06	7,71E-03	DEFA1
TC0800012382.hg.1	1,91536	7,94868	1,86E-06	8,42E-03	DEFA3
TC0100013293.hg.1	−0,59221	5,76598	4,44E-06	1,61E-02	ID3
TC0100016983.hg.1	0,79444	4,91516	6,22E-06	1,88E-02	CHI3L1
TC1300008388.hg.1	0,53128	6,24302	1,19E-05	3,04E-02	MTMR6
TC0200012977.hg.1	1,40624	5,80115	1,34E-05	3,04E-02	MXD1
TC2100008510.hg.1	0,90592	5,25215	1,99E-05	3,47E-02	KCNJ15
TC0300011517.hg.1	0,75234	7,28603	2,86E-05	4,24E-02	PROK2
TC0200016471.hg.1	0,60044	8,71197	3,70E-05	4,46E-02	MXD1
TC0400012245.hg.1	0,90643	8,31725	3,62E-05	4,46E-02	FAM198B
TC0500008785.hg.1	1,80286	7,97630	4,15E-05	4,69E-02	EGR1
TC0200016752.hg.1	−0,71346	6,78416	1,07E-04	7,00E-02	TTN
TC0100018262.hg.1	−0,55797	7,48395	9,11E-05	7,00E-02	GSTM2
TC1200006721.hg.1	0,84159	6,55486	9,74E-05	7,00E-02	CLEC4D
TC0300010949.hg.1	1,44762	7,19659	1,16E-04	7,00E-02	LTF
TC1500007355.hg.1	1,12160	8,15259	1,22E-04	7,12E-02	AQP9
TC1800009268.hg.1	0,86678	5,49724	1,34E-04	7,34E-02	DSC2
TC0800007007.hg.1	0,61851	8,06346	1,40E-04	7,44E-02	TNFRSF10C
TC0500008794.hg.1	0,50014	7,80606	1,86E-04	8,41E-02	CTNNA1
TC1200011385.hg.1	0,82397	6,13034	2,08E-04	8,94E-02	LIN7A
TC0100014349.hg.1	1,80611	7,28563	2,32E-04	9,64E-02	JUN

Gene set enrichment analysis

GSEA was applied to evaluate whether predefined gene sets are disproportionately represented at the extremes of a ranked gene list, correlating with phenotypic differences. GSEA identified key pathways associated with PD (Table 4), including the PI3K/AKT/mTOR signaling pathway (cell survival and metabolism), the IL6/JAK/STAT3 signaling pathway (inflammation and immune regulation), apoptosis, and complement pathways (innate immunity) (Fig. 5). Additional pathways, such as hypoxia and fatty acid metabolism, suggest broader impacts on cellular stress and metabolic processes.

Table 4.

Results of GSEA analysis showing significant enrichment of hallmark gene sets. Key metrics include ES (Enrichment Score), NES (Normalized Enrichment Score), NOM p-val (Nominal p-value), FDR q-val (False Discovery Rate q-value), FWER p-val (Family-Wise Error Rate p-value), and details of the leading edge subset.

Name	Size	ES	NES	NOM p-val	FDR q-val	FWER p-val	Rank at max	Leading edge
PI3K AKT MTOR signaling	98	0.418	1.700	0.006	0.148	73	3972	tags ​= ​43 ​%, list ​= ​24 ​%, signal ​= ​56 ​%
Androgen response	94	0.514	1.636	0.012	0.142	125	4392	tags ​= ​52 ​%, list ​= ​26 ​%, signal ​= ​70 ​%
Protein secretion	93	0.555	1.621	0.019	0.118	145	4305	tags ​= ​59 ​%, list ​= ​26 ​%, signal ​= ​79 ​%
G2M checkpoint	171	0.425	1.597	0.004	0.117	176	4648	tags ​= ​45 ​%, list ​= ​28 ​%, signal ​= ​62 ​%
IL6 JAK STAT3 signaling	80	0.607	1.552	0.018	0.148	244	2532	tags ​= ​41 ​%, list ​= ​15 ​%, signal ​= ​48 ​%
Apoptosis	153	0.546	1.552	0.033	0.124	245	3038	tags ​= ​39 ​%, list ​= ​18 ​%, signal ​= ​48 ​%
Complement	184	0.492	1.548	0.024	0.109	251	2602	tags ​= ​33 ​%, list ​= ​16 ​%, signal ​= ​38 ​%
Mitotic spindle	182	0.384	1.531	0.021	0.111	283	4186	tags ​= ​40 ​%, list ​= ​25 ​%, signal ​= ​53 ​%
Hypoxia	183	0.475	1.521	0.035	0.107	298	1718	tags ​= ​23 ​%, list ​= ​10 ​%, signal ​= ​26 ​%
MTORC1 signaling	189	0.494	1.516	0.072	0.101	308	3384	tags ​= ​42 ​%, list ​= ​20 ​%, signal ​= ​52 ​%
KRAS signaling up	177	0.483	1.466	0.042	0.140	406	1846	tags ​= ​25 ​%, list ​= ​11 ​%, signal ​= ​28 ​%
TGF BETA signaling	52	0.480	1.452	0.050	0.144	426	4470	tags ​= ​54 ​%, list ​= ​27 ​%, signal ​= ​73 ​%
Fatty acid metabolism	140	0.421	1.444	0.045	0.145	455	3016	tags ​= ​27 ​%, list ​= ​18 ​%, signal ​= ​33 ​%
Inflammatory response	172	0.613	1.418	0.093	0.163	502	2218	tags ​= ​40 ​%, list ​= ​13 ​%, signal ​= ​45 ​%
Cholesterol homeostasis	67	0.529	1.412	0.094	0.158	513	2993	tags ​= ​39 ​%, list ​= ​18 ​%, signal ​= ​47 ​%
Estrogen response late	166	0.386	1.391	0.044	0.171	545	2753	tags ​= ​23 ​%, list ​= ​17 ​%, signal ​= ​27 ​%
TNFA signaling via NFKB	189	0.720	1.379	0.133	0.174	566	1712	tags ​= ​44 ​%, list ​= ​10 ​%, signal ​= ​48 ​%
Angiogenesis	32	0.499	1.364	0.100	0.183	597	1779	tags ​= ​25 ​%, list ​= ​11 ​%, signal ​= ​28 ​%
Glycolysis	187	0.343	1.348	0.051	0.188	615	2211	tags ​= ​19 ​%, list ​= ​13 ​%, signal ​= ​22 ​%
P53 pathway	181	0.379	1.319	0.151	0.212	665	3317	tags ​= ​27 ​%, list ​= ​20 ​%, signal ​= ​33 ​%

Fig. 5.

GSEA enrichment plots for Hallmark gene sets. These plots illustrate the enrichment of genes within selected Hallmark pathways in the analyzed dataset: (A) PI3K/AKT/mTOR signaling, (B) IL-6/JAK/STAT3 signaling, (C) complement pathway, and (D) inflammatory response. The x-axis represents the gene rank order based on their differential expression between PD and control samples, while the y-axis displays the running enrichment score (ES). The pink curve traces the ES profile for each pathway, and vertical black lines mark the positions of genes from the pathway within the ranked gene list. These pathways are significantly enriched, underscoring their potential involvement in PD pathophysiology.

Validation of DEGs in independent cohorts

To assess the robustness and biological relevance of our transcriptomic findings in PD, we compared our PBMC-derived data with two independent whole-blood datasets: Craig et al. (2021), which utilized RNA sequencing, and Calligaris et al. (2015), which applied microarray profiling in drug-naive PD patients.

Craig et al. profiled whole-blood RNA-seq data from PD patients. A volcano plot from their dataset (Fig. 6A) was overlaid with DEGs from our study, with genes consistently upregulated shown in red and those consistently downregulated in blue. Genes with opposite trends were highlighted in green. To quantify overall concordance, we generated a scatter plot of log₂ fold changes for all genes common to both datasets (Fig. 6B). Genes identified as DEGs in our analysis were highlighted. A moderate yet highly statistically significant positive correlation was observed (Pearson's r ​= ​0.323; t ​= ​39.17, df ​= ​13,160, p ​< ​2 ​× ​10⁻¹⁶), supporting the reproducibility of our findings across distinct cohorts and platforms. Several immune-related genes showed consistent differential expression across studies, including MXD1, AQP9, PROK2, and CLEC4D (upregulated), as well as TTN, ID3, and AQP3 (downregulated), highlighting shared biological signatures relevant to PD.

Fig. 6.

Cross-study comparison of DEGs between our PBMC dataset and the whole-blood transcriptomic data from Craig et al. (2021). (A) Volcano plot displaying the differential expression results from Craig et al., with an overlay of DEGs identified in our dataset. Genes consistently upregulated in both studies are highlighted in red, consistently downregulated genes in blue and those with discordant expression directionality in green (FAM198B). Grey points represent genes that are not significantly differentially expressed in our dataset. (B) Scatter plot of log₂ fold changes for genes shared between our study and Craig et al. Genes identified as DEGs in our dataset are red (upregulated) and blue (downregulated). A moderate but highly significant positive correlation was observed between the datasets (Pearson's r ​= ​0.323; t ​= ​39.174, df ​= ​13,160; p ​< ​2 ​× ​10⁻¹⁶), supporting the reproducibility of gene expression changes across platforms and cohorts.

We further assessed the reproducibility of our findings by comparing our results to those from Calligaris et al. (2015), who analyzed whole-blood transcriptomes from early-stage, drug-naive PD patients using Affymetrix microarrays. That study reported 282 DEGs based on PUMA-derived significance estimates. Of these, 263 genes were matched to our dataset based on gene symbols. Log₂ fold changes showed a statistically significant correlation between studies (Pearson's r ​= ​0.357; t ​= ​6.17, df ​= ​261, p ​= ​2.53 ​× ​10⁻⁹), indicating moderate agreement (Fig. 7). To align with the criteria used in Calligaris et al., we applied a |log₂FC| > 0.1 threshold, identifying 120 genes in our dataset. Among these, 108 genes (90 ​%) showed concordant directionality. Notably, AQP3 was significantly downregulated in both datasets.

Fig. 7.

Scatter plot of DEGs from Calligaris et al. (2015) and our study. The scatter plot compares the log₂FC values of DEGs identified in the Calligaris et al. dataset (analyzed using PUMA) with their corresponding logFC values in our dataset (analyzed using limma). Each point represents a gene positioned according to its log₂FC in both studies. Genes with |log₂FC| < 0.1 in our analysis are shown in gray, while genes with |log₂FC| ≥ 0.1 and concordant directionality are coloured (upregulated genes are shown in red, and downregulated genes in blue). The gene AQP3 is significantly downregulated in both datasets, as indicated by the dark red label. The plot highlights variations and reproducibility in gene expression between the two analyses, providing a visual assessment of consistency across datasets. Pearson's r = 0.355; t = 6.1373, df = 261, p = 3,1 x 10^-9.

The consistent differential expression patterns observed across our PBMC-derived data and two independent whole-blood cohorts validate the robustness and biological relevance of our findings. The convergence of differentially expressed immune-related genes across independent cohorts and platforms —including RNA-seq and microarrays — highlights systemic transcriptional alterations in early PD. These results underscore the potential utility of peripheral blood biomarkers for the early detection and monitoring of PD.

Discussion

Blood-based biomarkers for PD present a highly convenient and minimally invasive alternative to current diagnostic methods, which often rely on imaging techniques that require specialized expertise and costly equipment or invasive procedures like lumbar puncture [12]. Beyond their ease of access, PBMCs are central mediators of immune and immunopathological processes, providing valuable insights into health status, disease onset, progression, and prognosis [25]. They are also biologically relevant in PD, given the accumulating evidence of immune dysregulation [26]. Moreover, transcriptomic studies have demonstrated significant differences in gene expression profiles in the peripheral blood of PD patients compared to healthy controls, indicating the potential for a blood-based diagnostic test [23,24,[27], [28], [29], [30]]. However, despite encouraging findings, no gene expression-based blood test for PD has been independently validated or clinically implemented, highlighting the need for further research to establish reliable biomarkers for early and accurate PD detection.

Our study provides a comprehensive analysis of transcriptional changes in PBMCs from PD patients, revealing significant insights into the molecular underpinnings of the disease. We identified 22 DEGs between PD and controls through differential expression analysis by employing transcriptome-wide gene expression microarrays. GSEA revealed enrichment in key pathways implicated in PD, including PI3K/AKT/mTOR signaling (cell survival), IL6/JAK/STAT3 signaling (inflammation), apoptosis, and complement pathways (innate immunity). Additional pathways related to hypoxia and metabolism suggest broader impacts on cellular stress.

Our analyses consistently identified immune dysregulation as a prominent feature in the peripheral blood of PD patients, both at the level of differentially expressed genes and enriched gene sets. These immune-related alterations were among the most consistent signals in the dataset. Inflammatory responses involving the central and peripheral immune compartments have been widely reported in PD and are believed to play a key role in its pathophysiology [26,[31], [32], [33], [34], [35], [36], [37]].

Using CIBERSORTx to infer immune cell composition from bulk gene expression data, we observed a significant decrease in naive CD4⁺ T cells and plasma cells in PD samples compared to controls. A reduction in absolute lymphocyte counts, particularly a deficiency within the CD4⁺ T cell compartment, has been consistently reported [26,38]. In addition, this analysis revealed an enrichment of neutrophils. It is likely that this results in neutrophil contamination in some PD samples. However, previous studies have consistently reported elevated neutrophil counts or increased neutrophil-to-lymphocyte ratios in PD patients [23,[38], [39], [40]], indicating that these shifts may reflect genuine disease-associated immune alterations rather than technical artifacts. Thus, although the inferred changes in cell populations observed in our study could be influenced by sample purity, they may also capture biologically relevant immune signatures intrinsic to PD.

Among the DEGs identified, DEFA1B, DEFA1, and DEFA3 were found to be significantly upregulated. DEFA1 and DEFA3 encode α-defensins, small cationic peptides essential for innate immunity [41]. Although defensins are abundantly expressed in granulocytes, their expression is not exclusive to these cells; other PBMC-resident populations such as macrophages, NK cells, B cells, and specific T cell subsets have also been shown to produce defensins under certain conditions [42,43]. Additionally, we observed an upregulation of LTF, which encodes lactotransferrin (LTF), an iron-binding protein involved in iron homeostasis [[44], [45], [46]]. Increased LTF levels and its receptor have been reported in nigral neurons of PD patients, reinforcing the link between iron dysregulation and neurodegeneration [46,47]. Iron accumulation is known to induce neurotoxicity, highlighting the potential role of iron homeostasis in the progression of PD [47].

A particularly noteworthy finding in our study is the dysregulated expression of two aquaporins, AQP3 and AQP9. Abnormal aquaporin expression has been investigated in various neurodegenerative diseases [48,49]. In the context of PD, increased free water levels in the substantia nigra —detectable via diffusion magnetic resonance imaging— have been reported in patients [50,51], suggesting that aquaporin-mediated water homeostasis may play a role in neurodegeneration [52].

While the role of AQP3 in PD remains unexplored, it has been shown that PD is associated with decreased AQP4 mRNA levels in blood serum, indicating reduced expression [53]. Moreover, AQP9 has been speculatively linked to PD pathogenesis [54]. Mitochondrial AQP9 expression in dopaminergic neurons has been hypothesized to contribute to their selective vulnerability in PD [55]. Experimental evidence further supports this notion. Zahl et al. (2023) demonstrated that the targeted deletion of Aqp9 in mice significantly suppressed inflammatory responses induced by the parkinsonian toxin 1-methyl-4-phenylpyridinium (MPP+), a known inducer of neuroinflammation [56]. Stahl et al. (2018) also provided mechanistic insights by showing that AQP9 facilitates MPP ​+ ​transport in Xenopus oocytes and HEK cells, thereby increasing cellular susceptibility to MPP+ and arsenite, another parkinsonogenic toxin [57]. Conversely, targeted deletion of Aqp9 in mice conferred neuroprotection against MPP ​+ ​toxicity in both organotypic midbrain slice cultures and in vivo models.

We also identified an upregulation of PROK2, which encodes Prokineticin 2 [58]. Prokineticins are chemokines that play a crucial role in the immune system and inflammatory diseases [59] and have been involved in the pathogenesis of neurological disorders [[59], [60], [61]]. The involvement of PROK2 in dopaminergic neurodegeneration has been demonstrated through its early upregulation in response to neurotoxic stress. PROK2 expression increases in dopaminergic neurons exposed to TNFα and the parkinsonian neurotoxin MPP+ in vitro. It is also upregulated in preclinical PD models before the onset of neuronal degeneration and motor deficits [62]. Elevated PROK2 levels have also been reported in nigral tissue and serum from PD patients, correlating with metabolic markers of mitochondrial stress [63]. Interestingly, overexpression of PROK2 may also confer neuroprotection in dopaminergic neurons [62].

MXD1, another upregulated gene, encodes a transcriptional repressor involved in cell differentiation and immune regulation [[64], [65], [66]]. Its function in dendritic cell maturation suggests a role in modulating immune responses in PD [67]. MXD1 has recently been proposed as a candidate diagnostic biomarker for PD [64], warranting further investigation.

To assess the reproducibility of our findings, we compared our DEGs with those of Craig et al. (2021), who profiled whole-blood RNA-seq data from PD patients. Despite differences in sample type and platform, we observed a moderate but highly significant correlation in fold changes, with several immune-related genes — MXD1, AQP9, PROK2, CLEC4D (upregulated), and TTN, ID3, AQP3 (downregulated) — showing consistent differential expression. This concordance supports the robustness of our findings across independent datasets and highlights conserved molecular mechanisms underlying PD.

We also compared our results with Calligaris et al. (2015), who analyzed whole-blood transcriptomes from early-stage, drug-naive PD patients. Of the 263 overlapping genes, a moderate and significant correlation in fold changes was observed. Applying a |log₂FC| > 0.1 threshold, 90 ​% of genes showed concordant directionality, including consistent significant downregulation of AQP3, reinforcing its potential as a reliable peripheral biomarker. Overall, our findings provide a strong foundation for identifying novel blood-based biomarkers that could enhance the early diagnosis and monitoring of PD. By analyzing transcriptional changes in PBMCs, we identified DEGs with potential diagnostic relevance.

A key consideration is that validation was performed using independent cohorts that differed in both profiling technology (microarray vs. RNA sequencing) and biospecimen type (PBMCs vs. whole blood). While such heterogeneity may limit direct replication at the single-gene level, it also offers an opportunity to test the robustness of disease-associated signatures under diverse methodological conditions. The observation of correlations across datasets supports the presence of reproducible peripheral transcriptional alterations in PD, despite differences in sample type and platform. Notably, several genes and immune-related pathways identified in our PBMC cohort were also observed in Craig et al.’s large-scale RNA-seq dataset (n ​= ​1599), underscoring the biological validity of our findings. Thus, the convergence of results across platforms and biospecimens strengthens the translational potential of PBMC transcriptomics as a clinically accessible biomarker source in PD. Given the established link between immune dysregulation and neurodegeneration, further investigation into these genes and their associated pathways is critical for developing reliable biomarkers to detect PD at its earliest stages, to elucidate the precise mechanisms contributing to disease progression, and to explore their potential as therapeutic targets.

A potential limitation of our study is that PBMCs were cryopreserved before transcriptomic profiling, which can influence PBMC gene expression [68]. However, all samples in our study (PD and controls) were processed identically, minimizing systematic bias. Furthermore, RNA integrity was assessed before analysis, and key findings were validated against independent whole-blood transcriptomic datasets, supporting the robustness of the observed disease-associated signatures.

From a translational perspective, identifying blood-based gene expression changes associated with PD underscores the potential for developing minimally invasive diagnostic tools. Further validation in more extensive, independent cohorts is essential to establish their clinical utility. Our study's PD and control groups differed significantly in age, with a mean age of 70 in the PD group and 57.7 in the controls (Welch Two Sample t-test, p ​= ​0.0012). This age disparity is an important confounding factor, as age can influence immune cell composition and gene expression profiles in PBMCs. Although we attempted to adjust for age as a covariate in our differential expression analysis, the limited sample size reduced statistical power, resulting in very few significant findings. Consequently, we opted to present the exploratory analysis adjusted exclusively for sex as a covariate to capture broader disease-associated transcriptional signals. We validated key gene expression changes using data from the larger, age-matched PPMI cohort to reinforce our conclusions and mitigate potential age-related bias. Nonetheless, we acknowledge that age remains a limitation of our study, and future research with larger, better age-matched cohorts will be essential to disentangle the effects of age and disease.

Ultimately, functional studies using in vitro and in vivo models will be crucial in determining whether these molecular markers serve as diagnostic tools and represent viable therapeutic targets. Integrating transcriptomic biomarkers into clinical practice could ultimately transform PD diagnosis, enabling earlier intervention and improving patient outcomes.

In conclusion, our study highlights significant transcriptional alterations in PBMCs from de novo and drug-naive patients with PD, emphasizing the role of immune dysregulation in disease pathogenesis. Identifying DEGs and enriched pathways related to immune responses and Inflammation provides a foundation for future research to unravel the complex molecular mechanisms underlying PD. Further studies are needed to validate these findings as potential biomarkers for PD.

Data availability

The data sets generated in the current study are available at the Gene Expression Omnibus (GEO; http://www.ncbi.nlm.nih.gov/projects/geo/) under the accession number GSE290333. In addition, we used the microarray data set GSE72267 from Caligaris et al. (2015) for specific comparisons.

Code availability

To facilitate the replication and expansion of our work, we have made the pipeline and analysis scripts publicly available on GitHub at https://github.com/mguillotfer/parkinson-microarray-analysis.git. The repository includes all code, figures, and supplementary materials used in this study. Specifically, it contains R scripts and data tables detailing each step of the preprocessing and statistical analysis workflow.

Author contributions

F.N. and L.M.H. performed PBMCs isolation from blood samples, RNA extraction, and collected patients’ information. M.G.F. designed, performed, and validated the bioinformatic analyses of transcriptome data under the supervision of J.P.L-A. J.A.M. recruited participants and performed clinical evaluations. M.G.F. and F.N. wrote the manuscript, and J.P.L-A. and J.M. reviewed the contents. F.N. and J.M. conceived and designed the overall study. F.N., M.G.F., L.M.H., D.N., J.A.M., J.P.L.-A. and J.M. read, edited, and approved the manuscript.

Declaration of Competing Interest

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Appendix A. Supplementary data

The following are the Supplementary data to this article: