Early downregulation of hsa-miR-144-3p in serum from drug-naïve Parkinson’s disease patients

Advanced age represents one of the major risk factors for Parkinson’s Disease. Recent biomedical studies posit a role for microRNAs, also known to be remodelled during ageing. However, the relationship between microRNA remodelling and ageing in Parkinson’s Disease, has not been fully elucidated. Therefore, the aim of the present study is to unravel the relevance of microRNAs as biomarkers of Parkinson’s Disease within the ageing framework. We employed Next Generation Sequencing to profile serum microRNAs from samples informative for Parkinson’s Disease (recently diagnosed, drug-naïve) and healthy ageing (centenarians) plus healthy controls, age-matched with Parkinson’s Disease patients. Potential microRNA candidates markers, emerging from the combination of differential expression and network analyses, were further validated in an independent cohort including both drug-naïve and advanced Parkinson’s Disease patients, and healthy siblings of Parkinson’s Disease patients at higher genetic risk for developing the disease. While we did not find evidences of microRNAs co-regulated in Parkinson’s Disease and ageing, we report that hsa-miR-144-3p is consistently down-regulated in early Parkinson’s Disease patients. Moreover, interestingly, functional analysis revealed that hsa-miR-144-3p is involved in the regulation of coagulation, a process known to be altered in Parkinson’s Disease. Our results consistently show the down-regulation of hsa-mir144-3p in early Parkinson’s Disease, robustly confirmed across a variety of analytical and experimental analyses. These promising results ask for further research to unveil the functional details of the involvement of hsa-mir144-3p in Parkinson’s Disease.


Methods
Materials: serum samples. Serum samples from centenarians (CENT), de novo patients (dnPD), controls (CTR), advanced patients (adPD) and patients' siblings (PDsibs) were collected at three different Collection Sites: the University of Bologna (UNIBO; 27 CENT and 19 CTR), the University of Goettingen (GOE; 205 dnPD, 12 adPD, 58 CTR) and the Servicio Andaluz de Salud (SAS; 48 PDsibs). Samples were used in two distinct phases (see details below): discovery to identify potential biomarkers and validation to assess the robustness of the findings. Samples from UNIBO were recruited in the framework of a study approved by the local Ethical Committee (S. Orsola Hospital-University of Bologna; Prot. n. 2006061707, amendment 08/11/2011). Informed consent was obtained from all subjects involved in the study. The samples from GOE used in the discovery phase (61 dnPD and 58 CTR) were part of the DeNoPa cohort, a longitudinal study previously described 19,20 retaining only baseline samples. The samples from GOE used in the biological validation were part of the independent, crosssectional Kassel cohort more recently collected 21 . Patients enrolled in this study were clinically phenotyped and diagnosed as PD according to UK Brain Bank Criteria 22 before collection of samples. Phenotyping included Unified PD Rating Scale, 1.5 Tesla magnetic resonance imaging, quantitative levodopa testing as published 23  www.nature.com/scientificreports/ identification test, Mini Mental Status Examination (MMSE) and further cognitive testing, as well as videosupported polysomnography to determine REM sleep behaviour disorder in a subset of patients. Subjects with marked vascular lesions in MRI or with normal pressure hydrocephalus by MRI were excluded. Siblings of patients with a diagnosis of sporadic idiopathic PD (PDsibs) were recruited over 20 months between September 2016 and January 2019 by Servicio Andaluz de Salud (Spain). The study was approved by the local Ethics Committee (ethical committee no. of approval 2014/PI173 of September 2016).
Serum miRNA were extracted using the miRNeasy Serum/Plasma Advanced kit (Qiagen) according to the manufacturer's instructions. All experiments were performed in accordance with relevant guidelines and regulations.
Library preparation and miRNA sequencing. Libraries for miRNA sequencing (NGS data) were prepared using QIAseq miRNA Library kit (Qiagen) following manufacturer's instructions. Labchip DNA High Sensitivity assay (Perkin Elmer) was used to analyse miRNA libraries, further quantified using the Qubit 2.0 Fluorometer (ThermoFisher). Libraries were sequenced with read length of 75 bp on a NextSeq 500 sequencer (Illumina). The total sequencing reads obtained per sample were 10.83 million, on average. The data have been deposited in NCBI's Gene Expression Omnibus 24 and are accessible through GEO Series accession number GSE180193 (https:// www. ncbi. nlm. nih. gov/ geo/ query/ acc. cgi? acc= GSE18 0193).
Bioinformatic pre-processing. The sequencing data were processed using a bioinformatic pipeline produced in house across three main steps: (i) pre-processing, (ii) differential expression analysis, (iii) functional annotation. The trimming of miRNA sequencing adapters and selection of sequences corresponding to mature miRNAs, was performed using Cutadapt v1.18 25 . The filtered reads were then aligned on the miRbase human hairpin database release 22.1 26 and on Exiseq NGS Spike-In sequences using SHRiMP v2.2.3 27 . Alignment was performed from fastq files interpreting quality values in fastq input as PHRED + (parameter "-qv-offset 33"), mode "mirna", and "-strata" option, to output the highest scoring mappings for any given read. The BEDTools suite v2.27.0 28 was used to select overlapping mappings on miRNA hairpins with mature miRNAs. qPCR duplicates filtering and UMI (Unique Molecular Identifiers) counts summarization was performed using a custom script to process same sequence and same UMI (Unique Molecular Identifier) sets of reads and to return a single read for each set.
The average number of reads mapped to mature miRNAs was between 2 and 3 million. The average number of miRNAs detected per sample was between 524 and 602. The final raw counts matrix had 2578 expressed miRNAs (UMI count > 0 in at least one sample) and 165 samples. Differential expression. The miRNAs expression counts were normalized using R package DESeq2 v1.22.1 29 . Principal component analysis (PCA) was performed using as input the VST (Variance Stabilizing Transformation) counts and plots were examined for quality check. Differential expression analysis was performed using the Wald test, and miRNAs with adjusted p-value below 0.05 were selected as differentially expressed (DEMs).
Validation and biomarkers selection. Technical validation of NGS results was achieved by qPCR. To this purpose, a subset of DEMs was selected by ranking adjusted p-values and filtering out average normalized expression below 5 counts (according to the qPCR plates supplier recommendations). Additional filtering was done considering the most significant overlaps among the already performed comparisons (Fig. 1). MiRNAs were selected among contrasts with higher statistical significance resulting from the DEMs intersection analysis: candidates were selected either when (i) shared between the most statistically significant contrasts or (ii) being the most significant and unique to a contrast. The final set of miRNA targets included also two miRNAs selected from the literature named "inflamma-miR" 30,31 . All details can be found in Supplementary Table S2.
Quantitative RT-PCR analysis. Technical validation on NGS data was performed using 12 plates for measuring the expression of 28 miRNA targets, 3 endogenous controls and 1 exogenous control (cel-miR-39-3p). For each plate 6 samples were processed: 1 CENT sample (UNIBO), 1 CTR sample (UNIBO), 2 CTR samples (GOE), 2 dnPD samples (GOE), to avoid batch effects. The cDNA was synthesized using TaqMan Advanced miRNA cDNA Synthesis according to the manufacturer's instructions. Further, 1 µl of spike-in cel-miR-39-3p 5 pM was added to poly(A) tailing reaction to check for reverse transcription efficiency. The qPCR validation was performed using pre-plated TaqMan Advanced miRNA Assays (ThermoFisher) and Taqman Fast ADVANCED Master MIX in a QuantStudio5 instrument. Initially, three miRNA normalizers were selected for validation by analysing four NGS samples (2 dnPD samples, 1 CTR from GOE and 1 CTR from UNIBO) on two custom-made ready-to-use plates with 30 miRNAs as endogenous controls (ThermoFisher). The selection of the three endogenous controls was made both on the Ct range and on the NGS analysis results (stable across conditions, as any variation in their expression may obscure real changes and produce artefacts in expression). After data review, miR-16-5p, miR-93-5p and miR-186-5p were selected as endogenous controls for the normalization. Real-time qPCR was performed in duplicate.
Biological validation on the independent cohorts was performed using in total 17 plates for measuring expression of 3 miRNA targets, 1 endogenous control (miR-186-5p) and 1 exogenous control (cel-miR-39-3p). The cDNA was synthesized using TaqMan Advanced miRNA cDNA Synthesis according to the manufacturer's instructions. 1 µl of spike-in cel-miR-39-3p 5 pM was added to poly(A) tailing reaction. The qPCR reaction was performed for miR-144-3p, miR-150-5p and miR-215-5p using Advanced miRNA Assay with Taqman Fast ADVANCED Master MIX in a QuantStudio3 instrument, again, qPCR was performed in duplicate. www.nature.com/scientificreports/ The qPCR differential expression analysis was performed using the Applied Biosystems™ Analysis Software web application (apps.thermofisher.com), Relative Quantitation Analysis Module (2020.2.1-Q2-20-build4 of 2020-04-27) in Singleplex mode, Version 1.1 and the CT method. The endogenous controls were used for normalization and the T-test statistic for computing p-values, adjusted with Benjamin-Hochberg statistical method. By default, significant differential expression is reported by the software when the observed RQ (Fold Change) is > 2 or < 0.5 and the corrected p-value is < 0.05. ARACNE and network reconstruction. The miRNAs interaction matrix was reconstructed using the ARACNE network imputation tool 32 . The network reconstruction approach is part of a wider set of gene network inference tools, based on information-theory related methods 33 . First, the mutual information M ij is computed between samples, using a Gaussian kernel methodology [32][33][34] . Second, the mutual information matrix is thresholded, and all edges for which the condition of genes being mutually independent cannot be ruled out, are discarded from the network. Then, a Data Processing Inequality (DPI) is applied, discarding the weakest edges. In particular we used the Parmigene R (https:// cran.r-proje ct. org/ web/ packa ges/ parmi gene/ index. html) implementation of the ARACNE reconstruction, with the multiplicative factor. More specifically, given three nodes i, j and k , the regulatory network reconstruction removes the weakest edge according to the relationship: and where 0 < 1 − τ < 1 is a multiplicative factor. In our experiments the value of τ was set tot the default value of 0.15. The matrix of mutual interaction reconstructed using the ARACNE procedure was then used as the adjacency matrix of our network. Finally, communities of miRNAs were identified, through the Louvain Community detection method, described in the following section.
Louvain community detection. To perform miRNAs networks clustering, we applied the Louvain Community detection method. The algorithm was first published in 2008 35 , and it was designed to work with large-scale MicroRNA-target enrichment, functional enrichment and network-based analysis have been performed using MIENTURNET 42 . For the miRNA-target enrichment analysis, the threshold of minimum number of miRNA-target interactions was set to 2 and the threshold for the adjusted p-value (FDR) was set to 0.05. The miRNA-target interactions database used was miRTarBase website Release 7.0, using only interactions with Strong evidence. The plots of miRNAs augmented KEGG pathways from miRNA-targets interactions (miRTarBase v7.0) were generated using the integrate_mir function of R package mirIntegrator 43 . Ethics approval and consent to participate. The human samples derive from existing multi-centre cohorts collected by the PROPAG-AGEING Consortium partners before the PROPAG-AGEING project, in the framework of other national and international projects.

Results
Design of the study. To uncover the complex biology shared by ageing and PD, we first performed miRNAseq analyses on serum samples from recently diagnosed, drug-naive PD patients (dnPD) and age-and sexmatched controls (CTR) 19,20 and centenarians (CENT) ( Table 1, discovery cohort). Standard analyses to identify differentially expressed miRNAs [44][45][46][47] were followed by network modelling 32 . Network reconstruction has, in fact, been proven to be a valid tool to model interactions between miRNAs, and several studies have shown the importance of analysing miRNA-miRNA networks to deepen the understanding of phenomena, beyond the identification of differential signatures. Further, miRNAs emerging from these analyses as biomarkers of PD and/or ageing were validated by qPCR, comparing the CTRs from the discovery phase with an independent validation cohort including additional dnPD, advanced PD (adPD) and siblings of PD patients (PDsibs) ( Table 3).
Finally, to complete the study, advanced functional analyses were further run with an array of state-of-theart tools for the study in silico of miRNAs-mRNA target interactions and functional enrichment-miRWalk 39  The comparisons (a.k.a. contrasts) were run to identify differentially expressed miRNAs (DEMs, adjusted p-values < 0.05 and average expression levels > 5). In this analysis, both dnPD and CENT were compared with CTR, considering UNIBO and GOE samples independently (CTR_UNIBO, CTR_GOE, respectively) and combined (CTR_ALL). Centenarians were analysed, as super-controls, in comparison with dnPD samples and CTR. The total number of non-redundant DEMs identified from all comparisons is 169 (Supplementary Table S1), with the largest intersection being 45 DEMs in contrasts CENT vs CTR and dnPD vs CENT, and the most statistically significant ones being dnPD vs (CTR + CENT) and dnPD vs CTR (UNIBO), sharing 17 DEMs (Fig. 1). The analysis then focused on DEMs that were either shared across the maximum number of contrasts or on DEMs that were specific to one contrast only (i.e. specific to a phenotype).
Finally, ranked DEMs were assessed against the literature (see "Methods"), leading to the identification of 28 candidates (Supplementary Table S2 www.nature.com/scientificreports/ From this, hsa-miR-122-5p and hsa-miR-150-5p result to be significantly down-regulated (corrected p-value < 0.05 and log2(FoldChange) < 1) in CENT with respect to CTR while hsa-miR-215-5p was significantly down-regulated both in PD and CENT with respect to CTR. Weaker trends were also observed towards downregulation in PD with respect to CTR for hsa-miR-150-5p, and, in dnPDs with respect to CTR and to CENT for hsa-miR-144-3p. In a conservative perspective all four miRNAs were retained for the following network-based in silico analysis.
In silico validation by network reconstruction. In the effort to further assess the role of these findings, network reconstruction was performed to analyse communities of interacting or co-expressing miRNAs in the two phenotypically relevant cohorts of dnPD and CENT. To reconstruct the networks, two separate adjacency matrices (double entry tables listing a distance between each possible couple of miRNAs, see "Methods" for details) for the dnPD and CENT cohorts were produced using the ARACNE network reconstruction approach 32 . All miRNAs identified in each cohort (i.e. the DESeq2 normalized counts for all samples within a cohort) were used to construct the adjacency matrices, as depicted in Fig. 2. The total number of nodes (miRNAs) and edges (connections between miRNAs, based on their similarity as described in the equation reported in Materials and Methods) obtained for each network is summarized in Supplementary Table S7. The two networks were then partitioned into smaller communities (i.e. clusters) by applying the Louvain community detection algorithm 35 . This enables the extraction of smaller, and putatively more relevant, groups of miRNAs characterized by tighter interactions i.e. similar expression behaviours within each cohort. A total of 31 communities for CENT and 49 communities for dnPDs were obtained (Fig. 2). Additional statistics emerging from the topological analysis of the network were computed to obtain more insight into the relevance of specific nodes: in particular the node degree, i.e. the number of edges departing from a node, is reported in Table 2. The node degree represents a miRNA's level of connection with other miRNAs, and is therefore a relevant proxy of the molecule's involvement in a variety of pathways and biological functions 49,50 . The degree can be weighted, or not, by the similarity to the connected miRNA (see "Methods" section for more details). While the unweighted node degree measures the number of connections, the weighted node degree reports the importance of these connections proportionally to the edge number (connectivity of the node) and therefore both are relevant to establish network properties. A full list of the distribution of weighted and unweighted node degrees, can be found in Supplementary Table S8.
Details on the topological distribution of the four miRNAs of interest in the phenotypes under study were achieved when community and differential expression analysis results were integrated. The lists of differentially expressed miRNAs were matched against the communities to identify the maximum overlap between the lists of DEMs and the lists of miRNAs found in the communities.
de novo PD. This integration lead, interestingly, to the identification of only one dnPD community containing the four DEMs identified in the NGS analysis and technically validated. Hsa-miR-150-5p, hsa-miR-215-5p, hsa-miR-144-3p and hsa-miR-122-5p cluster together in community 24, which in total includes 419 miRNAs. Hsa-miR-144-3p and hsa-miR-122-5p present the highest weighted and unweighted node degrees, respectively, in the community ( Table 2). This indicates that these two miRNAs hold the highest amount of pairwise connec- Centenarians. In this case the four miRNAs appear to be assigned to different communities and have low node degrees, with respect to the distribution of all the node degrees in the respective communities as we show in Table 2. The same analysis indicates that the miRNA having the highest weighted and unweighted node degree is hsa-miR-122-5p (with both weights falling above the upper quartile of the distribution of the unweighted node degrees and between the median and the upper quartile of the unweighted degrees distribution of the miRNAs in the community).

Validation in an independent cohort of dnPD, adPD patients and PD siblings.
We then focused on the three miRNAs that are statistically significantly differentially expressed in dnPD vs CTR, i.e. hsa-miR-215-5p, hsa-miR-150-5p and hsa-miR-144-3p. A much larger, independent data set including 204 samples (Table 3) was tested via qPCR. Importantly, the validation cohort included, in addition to dnPD samples, also adPDs and PDsibs, absent in the discovery cohort. We used as reference the CTR from UNIBO and GOE already included in the discovery cohort, both independently (CTR_GOE, CTR_BO) and combined (CTR_ALL). As shown in Fig. 3 and Table 4, hsa-miR-144-3p appears to be down-regulated in dnPDs with respect to CTR, confirming NGS results and the trend observed in the technical validation (Supplementary Table S9). Interestingly, hsa-miR-144-3p showed a down-regulation trend in adPDs and PDsibs as well, with respect to CTR (Supplementary Tables S10 and S11).
On the contrary, hsa-miR-150-5p was significantly up-regulated in dnPD, adPD and PDsibs with respect to CTR, and hsa-miR-215-5p showed a similar trend (Supplementary Tables S9, S10 and S11). While confirming the co-regulation of these two miRNAs, this observation did not return the results obtained in the discovery cohort that, therefore, were not considered in the following analyses, focusing exclusively on hsa-miR-144-3p.

Functional analysis.
To gain insights into the potential role of hsa-miR-144-3p in PD, a wide variety of tools and databases for functional inference were exploited as described in the following. miRNA targets: miRTarBASE. Target genes were obtained from a database of experimentally validated miRNA-mRNA target interactions (MTI). MTIs in miRTarBase are annotated as non-functional, strong functional and weak functional based on the technique used for the assay [36][37][38] . Most of the 211 targets of hsa-miR-144-3p were supported by weak evidence, while 23 genes had a strong functional MTI evidence (3 targets with non-func- Table 2. Node degrees statistics of miRNA communities. Columns 1 and 2 show the weighted and unweighted node degrees for each miRNA of interest in the given community. Column 3 reports the community ID, a consequent number, only needed to identify the location of a given miRNA with no other meaning. Column 4 shows the dimension (number of nodes) of each of the community described in Column 3.   Table S12).
In addition, hsa-miR-144-3p has been functionally annotated using HMDD3.2 40 , a database of miRNAdisease association and causation obtained using MDCAP (MiRNA-Disease Causal Association Predictor) 40 . Hsa-miR-144 has 112 annotated disease associations, 36 of which with causality. Most of these were related to cancer and some instances include also neurodegenerative diseases (Supplementary Table S13). 42 ) is a web-tool recently developed to analyse miRNA-gene targets interaction networks. In a first step the tool creates a list of significantly enriched interacting target-genes from a user-defined list of miRNAs, prioritizing the target genes based on a  www.nature.com/scientificreports/ user-defined FDR cut-off. Following this, a network of miRNA-target interactions genes is constructed, to enable the exploration of the network properties of the miRNA-genes complex interactions. The list of 419 miRNAs belonging to community 24 of dnPD, that includes hsa-miR-144-3p, was fed into the tool: Supplementary Tables S14 and S15 report all summary statistics concerning the corresponding pathways enrichment. Results, offering enrichment from KEGG and Reactome databases, present again a majority of cancer-related pathways, as well as, interestingly, platelets associated functions.

Functional context: MIENTURNET (MicroRNA Enrichment TURned NETwork
To gain additional insight in the above findings, we exploited MiRIntegrator 67 an additional tool designed to visualize KEGG enriched pathways in the form of miRNA-augmented pathways, that is, the graphical integration of miRNAs of interest into signalling pathways, to highlight a pictorial representation of the miRNA-mRNA targets interactions. This integration in particular, highlights post-transcriptional gene repression mediated by the miRNAs of interest, giving additional causal insight into the phenotype under study (Fig. 4).

Discussion
In the present work we explored the possibility to identify candidate-circulating blood biomarkers for PD within the (healthy) ageing framework. An overview of the obtained results, together with the methods applied, is summarized in Fig. 5.
First, we characterized by miRNA-seq the sera from a discovery cohort that includes dnPD, CTR and CENT. De novo PDs are a valuable model to study the disease as they represent the early stage of the malady and measurements are not affected by drug treatments.
Differential expression analysis led to the identification of a subset of modulated miRNAs selected for their relevance across different contrasts and from the literature. Among them, the technical validation by qPCR confirmed the differential expression of four miRNAs: hsa-miR-122-5p, hsa-miR-144-3p, hsa-miR-150-5p and miR-215-5p. While hsa-miR-122-5p was specific for the contrast CENT vs CTR and hsa-miR-144-3p was specific for the contrast dnPD vs CTR, hsa-miR-150-5p and miR-215-5p were similarly down-regulated in dnPD and CENT with respect to CTR. This observation was of potential interest because it supported our original hypothesis of a link between ageing and PD.
Accordingly, we observe that hsa-miR-150-5p and hsa-miR-215-5p, and noteworthy hsa-miR-122-5p and hsa-miR-144-3p, belong to the same miRNAs' community identified via complex network modelling. Communities offer a first indication of the potential synergies or similarities shared by the nodes of a network and produce therefore an independent measure of the relevance of these miRNAs in the corresponding cohort, giving Complement and coagulation cascades augmented pathway.    51 the authors proposed a pipeline to identify clusters of miRNAs cooperatively involved in triple negative breast cancer. Other works in the area also identify miRNA-target networks 52 or miRNA-diseases networks 53 , to detect groups of miRNAs playing an important role in pathologies and diseases. We then focused on the three miRNAs differentially expressed in PD (hsa-miR-144-3p, hsa-miR-150-5p and miR-215-5p) and assessed them in an independent cohort comprising dnPD, adPD and PDsibs. In this experimental context, hsa-miR-144-3p differential expression was confirmed for dnPD, offering a robust result for potential translation in preventive medicine, a fundamental tool in the fight against the disease. Additionally, hsa-miR-144-3p was also observed to be down-regulated in adPD and in a risk population such as PDsibs, in a statistically significant manner, but lacking of appropriate magnitude (Corrected p-value < 0.05 but log2(FoldChange) <|1|), granting further research in this direction to assess the relevance of this marker not only in prevention but also in the progression of the disease. It is worth noting that hsa-miR-144-3p did not change in centenarians that, despite the advanced age and the associated nigral neural loss 54 , did not develop clinically overt PD.
In our experimental settings, hsa-miR-144-3p did not change with age. Similarly, we did not find evidences in the literature supporting age-related differential expression in serum, with a single exception reporting an increase in mir144 (precursor of hsa-mir-144-3p) expression in cerebellum during ageing 55 . The same study also found an increase of mir144 in the frontal cortex of Alzheimer's disease (AD) patients. One of the targets of hsa-mir144-3p is the Amyloid Precursor Protein (APP), a pivotal gene in AD, which also plays a role in regulating mitochondrial activity 56 . Accordingly, the authors demonstrated that mir144-3p was down-regulated in vitro and this resulted in induced mitochondrial dysfunction. The expression of mir144-3p was assessed also in brain tissue from PD patients, showing down-regulation in the pre-frontal cortex 57,58 , and upregulation in the cingulated gyrus 59 . Down-regulation is also confirmed in a recent comprehensive analysis of sncRNAs in blood 60 .
Finally, we used a variety of functional analysis tools to get insights into the role of hsa-mir144-3p in PD. Among the pathways that emerged as significant, those related to platelets functions are of particular interest.
Indeed, alterations in coagulation are among the known hallmarks of systemic inflammation and changes in the normal blood clotting have been described in PD 61 . Several studies demonstrated that platelets share molecular pathways with neurons and neuronal disorders 62 and morphological and biochemical alterations were found in platelets from PD patients 63,64 . Furthermore, Fibrinogen Gamma Gene (FGG) is among hsa-mir144-3p targets and multiple reports show changes in FGG levels in PD patients' plasma 65,66 .
Recently, Patil et al. 18 evaluated miRNAs in serum from drug naïve PD samples and controls and identified three differentially expressed miRNAs (miR-335-5p, miR-3613-3p, miR-6865-3p). In our dataset, miR-6865-3p and miR-3613-3p had very low to no expression, while miR-335-5p was down-regulated both in CTR and PD versus centenarians. The expression level of this miRNA was not significantly different in dnPD with respect to CTR, therefore it was not selected for technical validation.

Conclusions
Our results consistently show the down-regulation of hsa-mir144-3p in early PD, robustly confirmed across a variety of analytical and experimental analyses. Additionally, hsa-miR-144-3p was also observed to have a down-regulated trend (statistically significant, but lacking of appropriate magnitude) in adPD and in PDsibs, offering an important starting point for the investigation of this marker's role in the progression of the disease. Functional analysis additionally revealed that hsa-miR-144-3p is involved in the regulation of coagulation, a process known to be altered in Parkinson's Disease. These results suggest that hsa-miR-144-3p may be an early marker of PD. Further research is needed to unveil the functional details of its relevance and involvement to confirm these promising results.

Data availability
The dataset supporting the conclusions of this article is available in the GEO repository (GSE180193). www.nature.com/scientificreports/