Validation of Differentially Expressed Brain-Enriched microRNAs in the Plasma of Parkinson’s Disease Patients

Background There is a pressing need to identify and validate, minimally invasive, molecular biomarkers that will complement current practices and increase the diagnostic accuracy in Parkinson’s disease (PD). Brain-enriched miRNAs regulate all aspects of neuron development and function; importantly, they are secreted by neurons in amounts that can be readily detected in the plasma. Τhe aim of the present study was to validate a set of previously identied brain-enriched miRNAs with a diagnostic potential for idiopathic PD and recognize the molecular pathways affected by these deregulated miRNAs. Methods RT-qPCR was performed in the plasma of 92 healthy controls and 108 idiopathic PD subjects. Statistical and in silico analyses were used to validate deregulated miRNAs and pathways in PD, respectively. Results miR-22-3p, miR-124-3p, miR-136-3p, miR-154-5p and miR-323a-3p levels were found to be differentially expressed between healthy controls and PD patients. miR-330-5p, miR-433-3p and miR-495-3p levels were overall higher in male subjects. Most of these miRNAs are clustered at Chr14q32 displaying CREB1, CEBPB, and MAZ transcription factor binding sites. Gene Ontology (GO) annotation analysis of deregulated miRNA targets revealed that ‘Protein modication’, ‘Transcription factor activity’ and ‘Cell death’ terms were over-represented. Kyoto Encyclopedia of Genes and Genome (KEGG) analysis revealed that ‘Long-term depression’, ‘TGF-beta signaling’, and ‘FoxO signaling’ pathways were signicantly affected. We validated a panel of brain-enriched miRNAs that can be used along with other measures for the detection of PD, revealed molecular pathways targeted by these deregulated miRNAs, and identied upstream transcription factors that may be directly implicated in their regulation and thereafter PD pathogenesis. Fold change ± SEM towards the mean expression of healthy controls (equals to 1). Statistically signicant differences in comparison to healthy controls (Mann Whitney U test) are highlighted in bold.


Validation of Differentially Expressed Brain-Enriched microRNAs in the Plasma of Parkinson's Disease Patients Stylianos Ravanidis
Idryma Iatrobiologikon Ereunon tes Akademias Athenon available in most clinical laboratories. Moreover, clear-cut interpretations can be drawn, as mature miRNA levels generally correlate with miRNA activity, whereas post-translational modi cations of proteins confer a complex correlation between activity and expression levels. With respect to PD, the expression levels of many miRNAs are altered in different regions of PD brain or target PD-associated proteins including αsynuclein (SNCA, miR-7 and miR-153) [11] and β-glucocerebrosidase (GBA, miR-22) [12]. Previously, several studies, yet not of a critical mass, have identi ed deregulated miRNAs in the different bio uids of PD patients. Generally, there is limited overlap between ndings due to the source of bio uid used, sample size, methodology, and lack of appropriate controls [13].
Here, we validated the ndings of our previous study [14], that had a speci c focus on brain-derived miRNAs, using an independent idiopathic PD cohort, and in depth analyzed current and previous data together to reveal not only a panel of deregulated miRNAs but also the upstream and downstream mediators of their effect.

Study population
The present study included 109 idiopathic PD patients and 92 healthy individuals in two separate cohorts. Patients were assessed with brain magnetic resonance imaging (MRI) or computed tomography (CT) and no relevant brain vascular lesions explaining the clinical phenotype were detected. The control group included spouses or unrelated companions of patients who had no known neurological disease, comorbidities or PD family history. Individuals with concurrent malignant tumors, psychiatric disorders, collagen diseases, endocrine and cardiovascular diseases, or infections were excluded from this study, since these conditions are expected to alter the expression pro le of circulating miRNAs. In addition, patients affected by atypical parkinsonism were also excluded. All patients and controls were recruited from the National and Kapodistrian University of Athens' neurological unit at Eginition hospital between 2018 and 2019. PD was diagnosed by two neurologists according to Postuma et al. criteria [4]. In all cases, essential demographic and clinical information, including study questionnaire for motor and nonmotor manifestations of the disease, rating scales [Hoehn & Yahr (H&Y) stage, mini-mental state examination (MMSE, cognitive impairment score < 26 [15]), and Uni ed Parkinson's Disease Rating Scale part III (UPDRS III) on or off state] were collected and documented. The demographic and clinical features of patients and controls are summarized in Table 1. Levodopa equivalent daily dose (LEDD) was calculated for the patient groups according to Tomlinson criteria [16]. miRNA isolation from plasma and RT-qPCR analysis Plasma samples were thawed at room temperature, centrifuged to pellet any cell debris, and spectrophotometrically analyzed for oxyhemoglobin absorbance at 414 nm. The cut-off level was set at 0.22 against water. Approximately 8.7% and 11.9% of healthy controls and PD plasma samples, respectively, were found hemolyzed and were discarded. miRNA extraction was performed using the NucleoSpin® miRNA plasma kit from Macherey-Nagel. 1 µg of MS2 (Roche) RNA was added to improve miRNA yield during the extraction method. Polyadenylation and reverse transcription reactions were performed in triplicate for every sample. Similarly, qPCR was performed in triplicate on the Roche Lightcycler® 96 using the SYBR FAST Universal 2X qPCR Master Mix from Kapa Biosystems. Stable reference (miR-103a-3p, miR-191-5p, miR-425-5p, miR-223-3p, miR-423-3p) and hemolysis (miR-23a, miR-451a) miRNA controls were included in the plasma analysis and were identical to previous study [14].
Primer sequences can be found in Table 2. The relative expression level of miRNAs was calculated using the 2 −ΔΔCt method. Samples with a cycle threshold difference Δ(Ct 23α− 3p -Ct 451α ) > 5.5, a PCR-based indicator of hemolysis, were excluded from the nal analysis.

Statistical analysis
Statistical analysis was performed using GraphPad PRISM v5.0 and R v3.5.3. All data underwent a normality test (Shapiro-Wilk), and were found to be non-normally distributed. The non-parametric Mann Whitney U test was used to observe differences between healthy controls and PD patients. Spearman's method, with Βonferroni correction for multiple comparisons, was used to correlate miRNA expression levels with participants' demographic and clinical characteristics. The threshold for signi cance was set to p-values less than 0.05.
To assess the possibility that sex is a confounding factor, a two-way ANOVA model was applied to the log-transformed data (normally distributed) with sex as an additional factor. No difference in the miRNAs that were statistically signi cant was found.
Receiver operating characteristic (ROC) curves were constructed and the area under the curve (AUC) was calculated to evaluate the predictive accuracy/power of plasma miRNAs for PD diagnosis. The cutoff for the ROC analysis was determined using the Youden Index. Data are presented as means ± SEM. miRNA selection was based on the stepwise removal approach. A logistic regression statistical model containing all available miRNAs as independent variables and PD status as the dependent variable was built. Then, the miRNAs with the least contribution in the model (as determined by an F test) were removed. This process continued until no further removals were possible.

Pathway analysis
The DIANA mirPath v.3 software suite was used to identify miRNA-regulated pathways. This software renders possible the functional annotation of one or more miRNAs using standard hypergeometric distributions, unbiased empirical distributions and meta-analysis statistics [17]. Here, predicted targets from DIANA microT-CDS algorithm with high quality experimentally supported interactions was used to identify KEGG molecular pathways, as well as GO terms targeted by each miRNA. The combinatorial effect of deregulated miRNAs was identi ed by simultaneously selecting multiple miRNAs in the software.

Chromosomal location of miRNAs
Human coordinates for each miRNA were obtained from miRbase release 22.1 website (http://www.mirbase.org/). To visualize data, the coordinates were uploaded on the PhenoGram software which was used to create the chromosomal ideogram [18].

Transcription factors regulating miRNAs
The TransmiR v2.0 literature-curated database of experimentally-validated transcription factors (TF) -miRNA regulations was searched to identify the TFs located 300 bp upstream and 100 bp downstream of each miRNA transcription start site (TSS) [19].

Results
miRNAs are differentially expressed in the plasma of idiopathic PD patients The demographic and clinical characteristics of 92 healthy controls and 109 idiopathic PD (iPD) patients are summarized in Table 1. RT-qPCR was used to analyze the differential expression of 11 brain-enriched miRNAs [14] and the ubiquitous miR-22-3p that targets GBA mRNA [12] in the plasma of control and PD cohorts. Eight of these miRNAs have been previously identi ed as signi cantly deregulated in PD, while four approached statistical signi cance [14].
Data revealed that 4 out of 12 miRNAs were signi cantly altered in the plasma obtained from PD patients compared to healthy controls (Fig. 1). These were miR-22-3p, miR-139-5p, miR-154-5p, and miR-330-5p. Table 3 compares the results obtained from previous [14] and current analyses as well as the data when both studies were pooled together. With the exception of miR-139-5p and miR-433-3p that showed reverse expression, the remaining ten miRNAs displayed a very similar expression pattern between the two studies. Pooling the data together, to increase accuracy and statistical power, it was revealed that miR-22-3p, miR-124-3p, miR-136-3p, miR-154-5p and miR-323a-3p are differentially expressed between healthy controls and PD patients.  [14].
b Pooled data from previous [14] and current studies.

Association between miRNA levels and clinical features
Spearman correlation test was used to relate miRNA levels to idiopathic PD patients' clinical features. We found no correlation between age-at-onset, disease duration, UPDRS III, MMSE, LEDD and miRNA levels, or patients' on/off state and miRNA levels (Table 4). Furthermore, no difference in the results was observed when data from both studies were pooled together; there was only a weak positive correlation of miR-22-3p with age-at-onset, but this was not signi cant (p = 0.001) (Supplemental Fig. 1). Finally, correcting clinical scores with LEDD did not reveal any more associations (data not shown).

Association between miRNA levels and age
There was no signi cant correlation between miRNA expression and age in either healthy controls or PD patients. When data from both studies were pooled together, miR-22-3p and miR-124-3p showed a positive correlation with age in PD, while miR-136-3p showed a negative correlation with age in healthy controls ( Fig. 2A to C). Since both miR-22-3p and miR-124-3p were found upregulated in patients with PD, we performed an analysis of covariance (ANCOVA) with the PD group as a factor and age as a covariate to explore whether these alterations were associated with aging. No differences in these miRNAs between the ANCOVA and the ordinary analysis of variance (ANOVA) were found. Hence, even in the case where age correlates with miRNA levels, PD pathology has an additional effect on miRNA expression.

Association between miRNA levels and sex
We found no signi cant differences between miRNA relative expression and sex in the present cohort. However, when data from both studies were merged, miR-330-5p, miR-433-3p and miR-495-3p showed higher expression in male subjects (Fig. 3A to C).

Discriminant Analysis
To evaluate the utility of plasma miRNA levels in discriminating subjects with PD from healthy controls, ROC curve analysis was performed after merging the data from both studies. The diagnostic sensitivity and speci city of a three-miRNA panel (miR-7-5p, miR-136-3p, miR-409-3p), when age and sex are taken into account, were 72% and 67%, respectively, and the AUC was 0.736 (Fig. 4).
To identify the transcription factors regulating the differentially-expressed miRNAs, the manually curated TransmiR v2.0 experimental database was used with highly stringent parameters to locate TF sites downstream of each miRNA transcription start site. The analysis revealed that TF binding sites for cAMP responsive element binding protein 1 (CREB1), CCAAT enhancer binding protein beta (CEBPB) and MYC associated zinc nger protein (MAZ) were found in over half of the miRNAs, including all ve miRNAs located on Chr14 (Fig. 5B).

Discussion
We previously reported that brain-enriched miRNAs can differentiate healthy controls from genetic and idiopathic PD subjects [14]. Here, we report results of an equally large validation study conducted on twelve miRNAs that were signi cantly or marginally non-signi cantly altered in our original study. With the exception of two miRNAs, miR-139-5p and miR-433-3p, all other miRNAs displayed very similar expression patterns and differences between healthy control and PD subjects in both studies. Combining the data, it was revealed that miR-22-3p, miR-124-3p, miR-136-3p, miR-154-5p and miR-323a-3p are differentially expressed between healthy controls and PD subjects. Of these, miR-22-3p [20][21][22], miR-124-3p [23], and miR-136-3p [21,24] have been identi ed as deregulated in other PD biomarker studies, further validating current ndings and their replicability at different neurological centers.
There is considerable information regarding the biological role of some of these deregulated miRNAs. miR-22-3p targets GBA, which encodes for β-glucocerebrosidase, a lysosomal enzyme involved in sphingolipid degradation; mutations in GBA are the single largest risk factor for PD [25]. In addition, miR-22 is neuroprotective in multiple neurodegeneration and traumatic brain injury models, has regenerative capabilities, and is involved in several aspects of neuronal development including cell cycle length, polarization of migrating neurons, and long-term synaptic plasticity [26][27][28]. miR-124-3p is the most abundant neuronal miRNA in the nervous system, and is considered indispensable for neuronal fate determination, differentiation, and plasticity [29,30]. It is anti-in ammatory and protects dopaminergic neurons from MPTP-and 6-OHDA -induced toxicity via multiple pathways [31]. Less is known about miR-136-3p, miR-154-5p, and miR-323a-3p. miR-136-3p expression is upregulated in synaptoneurosomes at preclinical stage of prion disease and its overexpression protects cells from neuroin ammation following ischemic insults [32,33]. miR-154-5p is differentially regulated during morphine self-administration and is predicted to have an important role in dopaminergic neuron differentiation and mu-opioid receptor regulation [34]. miR-323a-3p is upregulated in mild cognitive impairment, and differentially regulated by 6-OHDA and ischemia/reperfusion injury [35,36]. Importantly, we found that all ve deregulated miRNAs are signi cantly increased in PD; the fact that most of these miRNAs appear to have strong neuroprotective properties at multiple settings, may suggest that they are regulated as a compensatory response to brain impairment.
Men have fty percent higher incidence of PD [37] and, interestingly, we found that the expression of three brain-enriched miRNAs, miR-330-5p, miR-433-3p and miR-495-3p was signi cantly higher in male subjects. Considering that the ratio of males to women was different in our two studies, we think that this partly explains the inconsistency in miR-433-3p expression between them. Little is known of their biological roles. miR-330-5p targets mRNAs involved in activity-dependent synaptic plasticity in the hippocampus [38]. miR-433-3p targets follicle-stimulating hormone (FSH) expression in the anterior pituitary, and HIF1α levels in the brain during hypoxia inhibiting neuron proliferation and migration [39,40]. miR-495-3p targets are enriched for addiction and pro-survival genes, including BDNF, CAMK2 and ARC [41]. Furthermore, miR-495-3p expression is upregulated following deep brain stimulation [42].
Collectively, it is documented that these three miRNAs have rather negative impact on neuronal processes affected in PD, however, further work is required to better delineate their roles, and if/how their higher expression in males affects vulnerability to PD.
Examining the chromosomal coordinates of the differentially-expressed brain miRNAs, we found that miR-136-3p, miR-154-5p, miR-323a-3p, miR-433-3p, and miR-495-3p are located on Chr14q32, suggesting that they are co-deregulated by transcription factors or methylation. This area is inherently unstable and a 1.1 Mb microdeletion (14q32.2.q32.3, coordinates Chr14: ~100,400,000 -101,500,000) has been reported in a number of patients displaying motor delay, hypotonia, and feeding problems [43,44]. This genomic rearrangement is thought to be generated after the 500 bp expanded repeats anking the deletion boundaries undergo either non-allelic homologous recombination (NAHR) or form secondary structures that interfere with normal DNA replication and chromosome condensation [43]. It should be noted that the particular location also hosts fteen protein coding genes, some of which are imprinted [43,44]. More recently, miRNAs of this cluster were found to be among the most longitudinally stable miRNAs, indicating that they are ideal biomarkers to monitor the progression pathophysiological states including PD, reiterating the importance of the current ndings [45]. Moreover, we have used the TransmiR v2 experimental, manually curated, database to probe the TFs that regulate the expression of these deregulated brain miRNAs. Transcription binding sites for CREB1, CEBPB and MAZ were found in all miRNAs located at Chr14q32. In addition, a CREB1 site was found at miR-22 gene locus. CREB1 is essential for neuronal survival and axonal growth via both transcription of neurotrophins and neurotrophin-dependent CREB1-mediated transcription of pro-survival genes [46]. CREB1 is also required for adult neurogenesis, synaptic plasticity and memory formation [47,48]. Similarly to CREB1, CEBPB has been implicated in the control of neuronal development and survival, including processes such as cell fate determination, synthesis and response to trophic factors, learning, and memory [49,50]. MAZ on the other hand, has been linked to neural stem cell differentiation towards the glial cell lineage [51] and the induction of the NMDA receptor subunit type 1 (NR1) gene after neuronal differentiation [52]. Overall, these data indicate that epigenetic deregulation at Chr14q32 locus maybe responsible or contribute to neuronal miRNA differential expression in PD.
To explore the molecular pathways controlled by the deregulated miRNAs, in silico analysis of KEGG pathways and GO terms was performed. KEGG categories such as 'Long-term depression', 'TGF-beta signaling pathway', 'FoxO signaling pathway', 'Estrogen signaling pathway', 'ErbB signaling pathway', and 'Neurotrophin signaling pathway', that are implicated in PD-associated processes, particularly dopaminergic neuron development and survival, were over-represented [53][54][55][56][57][58][59][60]. 'Cellular protein modi cation', 'Nucleic acid binding transcription factor activity', 'Cell death', 'Catabolic process', were overrepresented among the biological processes affected. Cellular protein modi cations such as phosphorylation, ubiquitination, truncation, acetylation, nitration and sumoylation of PD-linked proteins have emerged as important modulators of pathogenic mechanisms in PD [61,62]. Transcription factor changes are also of particular interest, as they indicate that there is not only misexpression at the mRNA translation level by miRNA deregulation, but that there exists a second wave of en masse deregulation involving transcription-wide changes. Recently, we have integrated data across twenty-four PD biomarker studies and identi ed 25 miRNAs reported deregulated in at least two studies [13].

Conclusions
We performed an RT-qPCR analysis of brain-derived miRNAs in the plasma of a relatively large cohort of patients with PD and matched controls to replicate and validate previous ndings. The identi ed miRNAs form a robust set of deregulated brain-associated miRNAs in PD, that can now be further evaluated, along with other measures, as diagnostic and therapeutic tools for PD. Importantly, existing information of their neurological functions provides a clue of the processes they regulate during PD. In silico analysis provided a comprehensive guide of the pathways and processes they control, improving current understanding of their biological role. Further, Chr14q32 turns out to be a hotspot of these deregulated miRNAs. Importantly, they are co-transcribed by speci c transcription factors that regulate neuronal survival, memory and plasticity in the adult brain. The impact of the latter ndings will now await further exploration.   Scatter plots of miRNA relative expression in the plasma of healthy control and idiopathic PD groups.
Graphs show mean levels +/-SEM. Non-parametric Mann Whitney U test was used to determine the signi cance of differences between the two groups (p-values in the box next to each corresponding graph). *p<0.05, **p<0.01, ***p<0.001.

Figure 1
Scatter plots of miRNA relative expression in the plasma of healthy control and idiopathic PD groups. Graphs show mean levels +/-SEM. Non-parametric Mann Whitney U test was used to determine the signi cance of differences between the two groups (p-values in the box next to each corresponding graph). *p<0.05, **p<0.01, ***p<0.001.     The receiver operating characteristic (ROC) curve analysis for discriminating idiopathic PD from healthy control subjects. ROC curve of miR-7-5p, miR-136-3p, and miR-409-3p when age and sex are taken into account differentiates iPD from HC cases. AUC, area under the curve. The receiver operating characteristic (ROC) curve analysis for discriminating idiopathic PD from healthy control subjects. ROC curve of miR-7-5p, miR-136-3p, and miR-409-3p when age and sex are taken into account differentiates iPD from HC cases. AUC, area under the curve.