Abstract

   Parkinson’s disease (PD) is characterized by dopaminergic
   neurodegeneration and an abnormal accumulation of α-synuclein
   aggregates. A number of genetic factors have been shown to increase the
   risk of PD. Exploring the underlying molecular mechanisms that mediate
   PD’s transcriptomic diversity can help us understand neurodegenerative
   pathogenesis. In this study, we identified 9897 A-to-I RNA editing
   events associated with 6286 genes across 372 PD patients. Of them, 72
   RNA editing events altered miRNA binding sites and this may directly
   affect miRNA regulations of their host genes. However, RNA editing
   effects on the miRNA regulation of genes are more complex. They can (1)
   abolish existing miRNA binding sites, which allows miRNAs to regulate
   other genes; (2) create new miRNA binding sites that may sequester
   miRNAs from regulating other genes; or (3) occur in the miRNA seed
   regions and change their targets. The first two processes are also
   referred to as miRNA competitive binding. In our study, we found 8 RNA
   editing events that may alter the expression of 1146 other genes via
   miRNA competition. We also found one RNA editing event that modified a
   miRNA seed region, which was predicted to disturb the regulation of
   four genes. Considering the PD-related functions of the affected genes,
   25 A-to-I RNA editing biomarkers for PD are proposed, including the 3
   editing events in the EIF2AK2, APOL6, and miR-4477b seed regions. These
   biomarkers may alter the miRNA regulation of 133 PD-related genes. All
   these analyses reveal the potential mechanisms and regulations of RNA
   editing in PD pathogenesis.

   Keywords: Parkinson’s disease, A-to-I RNA editing, miRNA regulation,
   miRNA competition, gene expression

1. Introduction

   Parkinson’s disease (PD) is the second most common neurodegenerative
   disorder [[38]1,[39]2]. It is a complex progressive disease
   characterized by motor difficulties and non-motor issues, such as
   tremors, slowness, somnipathy, mood changes, and cognitive problems
   [[40]3,[41]4]. Its pathogenesis is modulated by a number of genetic
   factors [[42]5,[43]6,[44]7], as well as age and environmental exposures
   [[45]8,[46]9]. The first identified genetic factor for PD is the
   α-synuclein gene, which regulates synaptic vesicle trafficking and
   neurotransmitter release [[47]10]. Mutations in the α-synuclein gene
   are associated with PD risk [[48]11]. One regulator of α-synuclein, the
   highly conserved Argonaute 2 (AGO2), is abnormally expressed in PD
   patients [[49]12]. The inflammation-associated serine-threonine kinase
   (EIF2AK2) regulates α-synuclein by directly phosphorylating the Ser129
   residue, linking it to neurodegenerative disorders such as PD
   [[50]13,[51]14].

   Aside from α-synuclein pathology, the selective degeneration of
   dopaminergic neurons also modulates PD. A recent study has reported a
   decline of EEF1A in the PD-affected brain [[52]15]. This gene is
   involved in the prevention of dopaminergic neuronal cell death. Other
   genetic factors include MCL1 [[53]16,[54]17] and AIMP2 [[55]18], both
   of which are related to dopaminergic neuron cell death and α-synuclein
   aggregates.

   There is increasing evidence of blood biomarkers in PD, including
   circulating the serum neurofilament light chain [[56]19], small
   noncoding RNAs [[57]20], and cytokines [[58]21]. Studies on the
   identification of blood biomarkers have found altered
   cytosine-methylated regions and abnormal gene expression in PD patients
   compared to healthy controls [[59]22,[60]23]. Differentiating exosomal
   miRNAs, such as miR-10a-5p, miR-153, and mR-409-3p, can be detected in
   the bloodstream and can serve as diagnostic biomarkers of PD [[61]24].
   Alterations in blood-based RNA transcriptomes can provide valuable
   insights into PD diagnostics, progression, and mechanisms. Thus, in
   this study, 372 blood samples of PD patients were collected from the
   Parkinson’s Progression Markers Initiative consortium (PPMI)
   [[62]25,[63]26] for the discovery of novel PD biomarkers.

   A-to-I RNA editing is an important transcriptional modification carried
   out by adenosine deaminases acting on RNA (ADAR) [[64]27]. This form of
   transcriptional modification can alter the protein-coding capacity,
   generate diverse protein isoforms, and regulate RNA and protein
   expression [[65]28]. Previous studies have reported A-to-I editing in
   the pathogenesis of neurological and neurodegenerative disorders. For
   example, deficient A-to-I editing of the Q/R site in GluA2 is
   correlated with synapse loss, neurodegeneration, and behavior
   impairments in Alzheimer’s disease [[66]29]. In brain tissue samples of
   PD patients, there are significant decreases in A-to-I RNA editing
   levels and the specificity of mRNA editing sites [[67]30]. Given the
   increasing evidence of A-to-I RNA editing in neurological and
   neurodegenerative disorders [[68]30,[69]31,[70]32] and its multiple
   roles in transcript diversity [[71]33,[72]34], it is important to
   investigate the potential mechanisms of A-to-I RNA editing biomarkers
   for PD.

   Here, we developed a bioinformatics pipeline to identify A-to-I RNA
   editing biomarkers and uncover their potential mechanisms and
   transcriptomic impact in PD. The pipeline is outlined in [73]Figure 1.
   First, A-to-I RNA editing events and gene expressions are identified
   from RNAseq data. Then, the editing–gene associations are analyzed
   using two statistical methods: quantitative trait loci (QTL) analysis
   and Pearson correlation. Next, potential mechanisms of RNA editing
   effects on genes are explored. Specifically, the impact of RNA editing
   on miRNA functions and regulations of genes are investigated
   [[74]33,[75]34]. On one hand, RNA editing can introduce new or abolish
   existing miRNA binding sites, which can directly affect miRNA–RNA
   binding and the regulation of the targeted gene. On the other hand, new
   miRNA binding sites can sequester miRNAs from regulating other genes.
   Similarly, the loss of miRNA binding sites can release miRNAs and allow
   for increased binding and regulation of other genes that share the same
   binding site [[76]35]. Finally, RNA editing in the seed regions of
   miRNAs may change their binding affinities and affect a number of
   targeted genes. Our pipeline thoroughly reveals the impact of RNA
   editing on miRNA regulation and gene expression in PD patients.

Figure 1.

   [77]Figure 1
   [78]Open in a new tab

   The flowchart of this study.

2. Materials and Methods

2.1. Samples Involved in This Study

   To study the effects of RNA editing on gene expression, whole-blood
   RNAseq data of 372 PD patients and 169 healthy controls were downloaded
   from the PPMI cohort [[79]25,[80]26]. Corresponding demographic and
   clinical information ([81]Table S1) were also downloaded from this
   consortium and used in this study. They include indexes describing the
   severity of the disease, such as Hoehn and Yahr staging scale (H&Y
   stages) [[82]36], α-synuclein observations, concentrations of amyloid
   β-protein [[83]37], tau protein levels [[84]38], and unified
   Parkinson’s disease rating scale [[85]39].

2.2. A-to-I RNA Editing Detection

   The sequencing reads from RNAseq data were first aligned to hg38
   reference genome (GENCODE v36) by STAR (v2.7.9a) [[86]40]. Then, RNA
   editing events were identified using REDItools [[87]41] with default
   settings (i.e., minimal read coverage, 10; minimal quality score, 30;
   and minimal mapping quality score, 255). To ensure confident
   identifications, (1) only known editing sites from REDIportal (December
   2020) [[88]42] were considered, (2) SNPs from dbSNP151 [[89]43] were
   removed, and (3) candidates with low reads (n < 3) or low editing
   frequencies (<0.1) were filtered out. On average, there were 15,104
   editing events per individual. The number of editing events was not
   associated with the disease state (p > 0.05). Furthermore, informative
   A-to-I RNA editing events that were found in more than 50 PD patients
   and had a standard deviation of higher than 0.05 were kept in our
   analysis. Their distributions across different genes, regions, and
   repeats were analyzed by ANNOVAR (Version: 2020-06-07) [[90]44].

2.3. Gene Expression Quantification

   Using the aligned reads as described in 2.2, gene expression was
   quantified (Transcripts per million, TPM) by the RSEM software
   (Version: 1.3.1) [[91]45]. Low-abundance genes with mean expressions of
   less than 1 TPM were excluded, leaving us with 6288 informative genes.
   To reveal the potential effects of RNA editing events on genes related
   to PD pathogenesis, 2303 PD-related genes ([92]Table S2) were curated
   from DisGeNET [[93]46], MalaCards [[94]47], phenopedia [[95]48], KEGG
   database [[96]49], GWAS catalog [[97]50], and one previous study
   [[98]13].

2.4. Correlation Analysis between A-to-I RNA Editing Events and Genes

   To study the associations between A-to-I RNA editing events and genes,
   two different analysis methods were used: quantitative traits locus
   (QTL) with MatrixeQTL [[99]51] (FDR < 0.05) and Pearson correlation (p
   < 0.05). Age and sex were considered as covariates in the QTL analysis,
   which improves the identification sensitivity and reduces non-genetic
   phenotypic variance. For these gene-associated RNA editing events,
   their locations in genomic regions were analyzed. In total, 1203 RNA
   editing events were located in the 3′-UTRs of pre-mRNAs or lncRNAs and
   one RNA editing event was in the miRNA seed region. These events may
   alter miRNA binding and regulation of genes, especially for PD-related
   genes.

2.5. Analysis of RNA Editing Effects on Genes via Potential miRNA Regulation
Mechanisms

   RNA editing events in miRNA-binding targets, such as 3′-UTRs of
   pre-mRNAs or lncRNAs, may directly disturb the expression of the edited
   genes. To study this, miRNA-binding regions in wild-type and RNA-edited
   sequences were first detected by TargetScan (v7.2) [[100]52] and
   miRanda (v3.3) [[101]53]. The gain of the miRNA-binding target was
   determined when the miRNA–RNA interactions occurred in the RNA-edited
   sequences but not in the wild-type sequences, as detected by both
   tools, and vice versa for the loss of the miRNA-binding target. RNA
   editing events may alter original miRNA binding sites and directly
   affect the expressions of edited genes. To further elucidate the
   biological impact and functions of the RNA editing events, the
   interactions between edited genes and RNAs were extracted from starBase
   [[102]54] and their enriched pathways were determined by Metascape
   (Version: 3.5) [[103]55], DAVID (Version: 6.9) [[104]56], and Enrichr
   [[105]57].

   As previously mentioned, RNA editing events in miRNA-binding regions
   may not only affect the edited genes but also indirectly alter other
   genes due to miRNA competition. To investigate this, TargetScan and
   miRanda were also used to detect miRNA-binding regions in genes
   competing with the edited genes. To clarify, a competing gene is a gene
   that shares miRNA binding sites with the ones altered in the RNA-edited
   gene. This includes both lost (abolished) and gained (acquired) binding
   sites from the A-to-I editing. Similarly, the functional roles of the
   competing gene-associated RNA editing events were annotated using
   starBase and pathway enrichment analysis.

   Last but not least, RNA editing events can also be located in the seed
   regions of miRNAs (2–8 nt). The edited seed region can bind to and
   regulate a different set of genes than the intended genes. Similar
   procedures were applied to study the RNA editing events in miRNA seed
   regions, including the identification of downstream RNAs regulated by
   the edited miRNA and the analysis of their enriched pathways.

3. Results

3.1. A-to-I RNA Editing Events Are Involved in Parkinson’s Disease via Their
Effects on Gene Expressions

   Our RNA editing detection pipeline identifies 25,799 informative RNA
   editing events ([106]Table S3). Most of these editing events occur in
   protein-coding genes (93.49%), non-exonic/splicing regions (99.77%),
   and Alu repeats (91.76%), as shown in [107]Figure 2A. Their
   distributions are consistent with the preferred locations of RNA
   editing events reported in previous studies [[108]33,[109]58].
   Moreover, there is a lower overall editing frequency in PD patients
   compared to healthy controls (p = 5.92 × 10^−3), which is consistent
   with the decreased editing levels found in the brains of PD patients
   [[110]30]. These analyses reveal the reliability of the identified RNA
   editing events.

Figure 2.

   [111]Figure 2
   [112]Open in a new tab

   The overview of RNA editing effects on gene expressions. (A) The
   distributions of informative RNA editing events in genes, regions, and
   repeats. (B) Manhattan plots show the effects of 9897 RNA editing
   events on 6286 genes by QTL (FDR < 0.05) and Pearson correlation
   analyses (p < 0.05). The top and bottom panels present 137,563 positive
   and 483,277 negative associations, respectively. (C) The PD-related
   enrichment pathways of the editing-associated genes by DAVID. (D) The
   density distributions of editing-associated genes on chromosomes. The
   color bar shows the number of editing-associated genes in a 10 M bps
   (base pairs) region. (E) The types of editing–gene associations. Most
   RNA editing events are associated with distal genes (dist > 1 × 10^6,
   trans-associations) and genes in different chromosomes.

   To associate the frequencies of RNA editing events to gene expressions,
   quantitative traits loci analysis (QTL, FDR < 0.05) and Pearson
   correlation (p < 0.05) were performed. In total, there are 9897 RNA
   editing events associated with 6286 genes ([113]Table S3), as shown in
   the Manhattan plots in [114]Figure 2B. To understand the functions and
   impact, we conducted pathway enrichment analysis on these
   editing-associated genes and found that they are enriched in
   neurodegeneration-related functions ([115]Figure 2C, [116]Table S4).
   Thus, RNA editing events may affect PD pathogenesis via their effects
   on these genes.

   Next, we analyzed the distributions of the editing–gene associations.
   The densities of editing-associated genes ([117]Figure 2D) and editing
   events ([118]Figure S1A) are different. There are more affected genes
   in the gene-rich regions of chr19 (Chr19:10M−20M) and chr11
   (Chr11:60M-70M) ([119]Figure 2D and [120]Figure S1B). In addition, the
   majority of these associations belong to the trans- and
   cross-chromosomal RNA editing regulations of genes ([121]Figure 2E).
   The preference of RNA editing events for regulating distal genes urges
   us to explore the potential mechanism underlying these associations.
   The whole analysis process can be divided into three parts. First, RNA
   editing events may abolish or create miRNA-binding sites within the
   genes and directly alter their expressions ([122]Section 3.2). Second,
   RNA editing events may release miRNAs to or sequester miRNAs from other
   genes, indirectly affecting the expressions of these miRNA-competing
   genes ([123]Section 3.3). Third, RNA editing events may affect miRNA
   seed regions and then alter their downstream targets ([124]Section
   3.4). All these mechanisms are covered in the next sections.

3.2. A-to-I RNA Editing Events May Affect miRNA Regulations of Their Host
Genes

   A-to-I RNA editing can modulate gene expression by altering the miRNA
   binding sites of the host gene. In total, there are 581 RNA editing
   events located in their associated genes ([125]Figure 3A). Of them, 72
   RNA editing events were predicted to create 11,370 new miRNA-binding
   sites and eliminate 815 original ones. The QTL and Pearson correlation
   results for these editing events and gene expressions are presented in
   [126]Table S5.

Figure 3.

   [127]Figure 3
   [128]Open in a new tab

   A-to-I RNA editing events may alter miRNA regulations of their host
   genes. (A) Analysis procedures to uncover one mechanism of the
   associations between RNA editing events and their host genes. (B) The
   scatter plot shows the RNA editing events, which may alter miRNA
   regulations of their host genes. The editing example in EIF2AK2 is
   highlighted. (C) This RNA editing event is negatively associated with
   EIF2AK2 expression. EIF2AK2 is overexpressed in PD and in more severe
   PD samples. The RNA editing event shows lower editing frequencies in PD
   samples and negative associations with tau protein levels. (D) The 16
   EIF2AK2-interacted genes are significantly associated with EIF2AK2 and
   the RNA editing event. Genes known to be implicated in PD are in bold.
   (E) The 16 genes are enriched in neurodegeneration-related processes by
   Metascape. (F) Of them, MCL1 is overexpressed in PD and in more severe
   PD samples. (G) The analyses above may uncover the potential effects of
   this RNA editing event on five PD-related genes.

   Interestingly, there is a noteworthy RNA editing event in the 3′-UTR of
   EIF2AK2 (Chr2:37104057), as shown in [129]Figure 3B. This RNA editing
   event creates new binding sites for miR-3622a-3p and miR-3622b-3p,
   which may increase the degradation of this gene and result in the
   observed decrease in gene expression ([130]Figure 3C). In fact, the
   frequency of this RNA editing event is negatively correlated to the
   expression of EIF2AK2 (QTL: FDR = 1.39 × 10^−4, β = −12.60; Pearson
   correlation: p = 4.94 × 10^−4, R = −0.42). The associated EIF2AK2 gene
   is known for its function in promoting PD progression via the
   phosphorylation of α-synuclein protein [[131]13]. It is also
   significantly highly expressed in PD (p = 0.01) and in more severe PD
   samples (p = 0.03). The Chr2:37104057 RNA editing event is
   significantly less frequent in PD samples compared to healthy controls
   (p = 0.02) and is negatively associated with tau protein levels (p =
   0.06, R = −0.24). Thus, this RNA editing event may be the reason for
   the observed downregulation of EIF2AK2 and may be considered a
   potential biomarker for PD.

   To further explore the biological impact and downstream effects of this
   RNA editing event, we examined the role of the edited gene. EIF2AK2
   encodes for a kinase that binds to dsRNA (RNA binding protein) and
   plays a major role in the regulation of gene transcription [[132]59].
   According to EIF2AK2–RNA interactions from StarBase [[133]54], RNAInter
   [[134]60], and RNAct [[135]61], there are 2158 potential EIF2AK2
   targets and 4129 potential non-EIF2AK2 targets. All these genes are
   sufficiently expressed in blood with mean expressions of more than one
   TPM. Of 2158 potential EIF2AK2 targets, 31.37% have associations with
   both this editing event and EIF2AK2 ([136]Table S6). Of 4129 potential
   non-EIF2AK2 targets, 26.45% have associations with both this editing
   event and EIF2AK2 ([137]Table S6). There is an enrichment of
   EIF2AK2-interacted genes associated with the editing event (Fisher
   test: p = 4.54 × 10^−5). This suggests that EIF2AK2 editing may
   regulate these genes via its RNA binding functions. Taking a closer
   look at the experimentally validated EIF2AK2–RNA interactions from
   starBase, there are 16 protein-coding RNAs significantly correlated to
   the editing event and EIF2AK2 ([138]Figure 3D). The 16
   editing-associated genes are involved in the neurodegenerative
   NIK/NF-KB signaling and immune processes ([139]Figure 3E, [140]Table
   S7). For example, the NF-KB gene plays a prominent role in dopaminergic
   neurodegeneration [[141]62]. The upregulation of inflammatory cytokines
   can initiate a cascade of pro-inflammatory signaling that ultimately
   results in neurotoxicity related to PD [[142]6].

   Among the 16 affected genes, at least 4 genes (MCL1 [[143]16], P4HB
   [[144]63], DNAJB6 [[145]64], and APP [[146]65]) are known to be
   implicated in PD. Specifically, MCL1 has reported its associations with
   the loss of dopamine neurons, which is one of the reasons for the motor
   symptoms in PD [[147]16]. In our study, MCL1 is overexpressed in PD (p
   = 1.87 × 10^−4) and in more severe PD patients (p = 0.02) ([148]Figure
   3F). Its coexpressions with the edited gene (p = 3.21 × 10^−36, R =
   0.59) and its negative correlations with the editing event (p = 1.73 ×
   10^−3, R = −0.38) further support their relationships. Therefore, the
   decrease of the RNA editing event (Chr2:37104057) may upregulate
   EIF2AK2 [[149]13], which then affects the expression of at least four
   PD-related genes ([150]Figure 3G). Based on our findings, RNA editing
   events in miRNA targets may perturb miRNA degradation functions,
   leading to abnormal gene expression and the subsequent altered
   biological pathways.

3.3. A-to-I RNA Editing Events May Alter miRNA Competitions between Their
Host Genes and Other Genes

   RNA editing events in one gene may indirectly affect other genes.
   Editing events that create new miRNA binding sites may sequester miRNAs
   from regulating other RNAs. Similarly, editing events that abolish
   miRNA binding sites may release the miRNAs to regulate other RNAs. The
   process is also known as editing-mediated miRNA competition. In our
   study, eight RNA editing events that can alter miRNA competitions
   between their host genes and 1146 other genes were identified
   ([151]Figure 4A, [152]Table S8). These affected genes are enriched in
   the biological process related to expression regulation and PD, as
   shown in [153]Figure 4B and [154]Table S9. Among them, 127 have
   reported their roles in PD previously ([155]Table S2). These PD-related
   genes may be affected by miRNA dysregulations from four RNA editing
   events.

Figure 4.

   [156]Figure 4
   [157]Open in a new tab

   A-to-I RNA editing events may alter miRNA competitions between their
   host genes and other genes. (A) Analysis procedures to uncover one
   possible mechanism of the trans-associations between RNA editing events
   and genes. (B) The enrichment results of the genes affected by
   editing-mediated miRNA competitions. (C) One RNA editing event in the
   3′-UTR of APOL6 causes the binding loss of miRNAs and releases the
   miRNAs to regulate EEF1A1. (D) The frequency of this editing event is
   positively associated with the expression of APOL6 and negatively
   correlated to EEF1A1. APOL6 is highly expressed in severe PD samples.
   The editing event has higher frequencies in the samples with increased
   H&Y stages or tau protein levels. (E) The editing event may directly
   result in the overexpressions of APOL6 and indirectly cause the
   downregulations of 47 genes including EEF1A1 and 8 ribosome proteins.
   (F) The potential roles of this RNA editing event related to PD
   pathogenesis.

   The editing event of most interest is the one in the Chr22:35661178
   position of APOL6. It is located in the 3′-UTR and results in the loss
   of miR-19 and miR-4659 binding sites, as shown in [158]Figure 4C. The
   lost miRNA regulations may directly increase the expression of APOL6
   (QTL: FDR = 1.84 × 10^−4, β = 8.55; Pearson: p = 7.13 × 10^−8, R =
   0.29). This releases the two miRNAs from APOL6, allowing them to
   regulate other genes and potentially decrease the transcription of 47
   genes, including EEF1A1 (QTL: FDR = 2.29 × 10^−3, β = −609.93; Pearson:
   p = 6.37 × 10^−7, R = −0.27; [159]Figure 4D) and 8 ribosomal proteins
   ([160]Figure 4E, [161]Table S3). The edited gene, APOL6, induces a
   neurodegeneration-related apoptosis process [[162]66,[163]67,[164]68]
   and is highly expressed in more severe PD samples (p = 0.02). One
   PD-related gene competing with the edited gene is EEF1A1. It is known
   for its pro-survival function in the protection of dopaminergic neurons
   [[165]15]. The eight ribosomal proteins and the EEF1A1 translation
   factor are responsible for the synthesis of proteins, especially
   PD-related peptides. The dysregulation of these PD-related genes
   suggests that this editing event at Chr22:35661178 is a potential
   pathogenic biomarker for PD ([166]Figure 4F). It is also supported by
   the highly edited frequencies in more severe PD samples characterized
   by H&Y stages (p = 0.04) and tau protein levels (p = 0.09, R = 0.09).
   The analyses in this section provide one explanation for the
   trans-associations of A-to-I RNA editing events and genes.

3.4. A-to-I RNA Editing Events May Modify miRNA Seed Regions to Disturb Their
Regulations

   Given that miRNA seed regions are highly conserved, a single nucleotide
   change (e.g., A-to-I RNA editing) can dramatically alter the targeted
   genes of the miRNAs. In total, we discovered one RNA editing event in
   the seed region of miR-4477b (Chr9:63819627). This change may
   dysregulate four genes ([167]Table S10), including one PD-related gene,
   AGO2 ([168]Figure 5A).

Figure 5.

   [169]Figure 5
   [170]Open in a new tab

   A-to-I RNA editing events may modify miRNA seed regions to disturb
   their regulations. (A) Analysis procedures to uncover the possible
   mechanisms for the associations between RNA editing events in miRNA
   seed regions and their targets. The editing example in miR-4477b is
   highlighted. (B) This RNA editing event alters the miRNA seed region,
   thus leading to the loss of miRNA binding on AGO2. (C) The frequencies
   of this RNA editing event are positively associated with the
   expressions of AGO2. The editing event shows higher frequencies in PD
   samples and positive associations with the levels of amyloid β-protein
   and α-synuclein. (D) The 46 AGO2-interacted RNAs are significantly
   associated with AGO2 and the editing event. The red and green circles
   describe the positive and negative associations, respectively. (E) The
   associated genes are enriched in the regulations of miRNAs and RNA
   binding proteins by Enrichr. (F) One PD-related gene, AIMP2, is
   positively associated with AGO2 and the editing event. (G) The analyses
   above may uncover the potential effects of this RNA editing event in
   miRNA seed region on five PD-related genes.

   This RNA editing event alters the nucleotide from A to I in the fourth
   position of the miRNA seed region and results in the loss of binding on
   three sites in AGO2 ([171]Figure 5B). The loss of miR-4477b-AGO2 is
   associated with an increased expression of AGO2 (QTL: FDR = 0.02, β =
   7.21; Pearson: p = 1.63 × 10^−4, R = 0.26; [172]Figure 5C). The
   associated gene, AGO2, is involved in the production of α-synuclein
   [[173]12]. The RNA editing event shows a higher frequency in PD samples
   compared to the healthy controls (p = 0.06). In addition, RNA editing
   in miR-4477b is positively correlated with the levels of amyloid
   β-protein (p = 0.05, R = 0.14) and α-synuclein (p = 0.09, R = 0.12).
   This suggests that the RNA editing event at Chr9:63819627 leads to the
   loss of miRNA regulation on AGO2 related to PD.

   Furthermore, as an RNA binding protein, AGO2 interacts with 168
   protein-coding RNAs, according to starBase. Of them, 46 RNAs are
   significantly associated with the editing frequency of miR-4477b and
   the expression of AGO2 ([174]Figure 5D). These associated genes include
   the highly conserved Argonaut family of proteins, which includes AGO1
   and AGO2 and plays a role in short-interfering-RNA-mediated gene
   silencing ([175]Figure 5E). They also include four other PD-related
   genes: AIMP2 ([176]Figure 5F), PSMC3IP, RMDN3, and NPC1. These four
   genes have been reported to correlate with neurodegenerative diseases
   [[177]69,[178]70], have activation roles in age-dependent dopaminergic
   neuronal loss [[179]18], and contain regulatory functions in PD-related
   autophagy [[180]71]. All the above analyses support that RNA editing
   events in miRNA seed regions affect their regulatory functions on
   multiple PD-related genes ([181]Figure 5G).

3.5. A-to-I RNA Editing Effects on Genes via Disturbing miRNA Regulations in
Other Datasets

   In our analysis, we discovered a list of RNA binding events that may
   alter the miRNA regulations of genes ([182]Tables S5, S8 and S10). To
   further support our findings, we expand our analyses to other cohorts.
   Three datasets were analyzed: (1) a small PD blood dataset
   ([183]GSE165082), (2) a healthy blood dataset (PPMI), and (3) a large
   Alzheimer’s disease (AD) dataset. The small PD blood dataset did not
   contain the three RNA editing events introduced in detail in this
   study; however, it provides the validation of one RNA editing event
   (Chr1:160998027; [184]Figure 6A,B and [185]Table S11). In the healthy
   blood dataset and AD dataset, there are 1135 significant associations
   between RNA editing events and genes ([186]Table S11), which are
   consistent with the 1300 associations found in our study ([187]Tables
   S5, S8 and S10). The 1135 significant associations include the three
   RNA editing events in the EIF2AK2, APOL6, and miR-4477b seed regions
   described in detail ([188]Figure 6C–H). All these analyses support the
   mechanism that A-to-I RNA editing events alter miRNA binding and
   dysregulate genes related to the pathogenesis of PD and AD [[189]33].
   In conclusion, this study expands the knowledge of RNA editing effects
   on neurodegeneration.

Figure 6.

   [190]Figure 6
   [191]Open in a new tab

   The effects of RNA editing events on genes in other datasets. (A,B) In
   blood samples of PD patients from PPMI and GEO ([192]GSE165082), the
   RNA editing event at Chr1:160998027 shows significant associations with
   F11R. (C–E) In healthy blood samples, the three RNA editing events
   discussed in detail also show significant associations with
   corresponding genes. (F) In temporal cortex (TCX) regions of AD
   patients, the RNA editing event at Chr2:37104057 is significantly
   associated with EIF2AK2 and MCL1. (G,H) In cerebellum (CER) and
   dorsolateral prefrontal cortex (DLPFC) regions of AD patients, the RNA
   editing event (Chr9:63819627) in the seed region of miR-4477b may
   dysregulate AGO2 and AIMP2.

4. Discussion

   In recent years, A-to-I RNA editing in PD has attracted the interest of
   researchers from all over the world [[193]30,[194]72]. Previous studies
   focused on the differential RNA editing events in coding regions and
   miRNAs. However, their mechanisms and the potential functions of the
   editing events in lncRNAs or 3′-UTRs of mRNAs remain largely unclear.
   In this study, we analyzed 372 PD patients and identified the
   associations between 9897 A-to-I RNA editing events and 6286 genes. Of
   the gene-associated RNA editing events, more than 26% were
   significantly correlated to PD or PD severity (p < 0.05; [195]Figure
   S2A and Table S12). A total of 1224 PD-specific RNA editing events were
   also significantly associated with 612 PD genes ([196]Figure S2B).
   These RNA editing events may affect the expression of proximal genes
   (dist ≤ 1 × 10^6), also known as cis-regulatory effects. They can also
   regulate the expression of distant genes (dist > 1 × 10^6), which are
   considered trans-regulatory effects. Most of the associations were
   found to be trans-regulatory ones, in which the RNA editing event was
   significantly associated with a distal gene ([197]Figure 2C).

   To further explore the underlying mechanisms of these associations, we
   analyzed the perturbed miRNA regulations of genes caused by A-to-I RNA
   editing events. Of all the 624,325 editing–gene pairs, 77.81% are
   negatively associated. One reason for the reduced expressions is that
   RNA editing events caused more gained miRNA binding sites than lost
   sites (Second 3.2) [[198]33,[199]73]. In total, there are 72 RNA
   editing events that created or eliminated miRNA binding sites and
   directly affected miRNA regulations of their host genes ([200]Table
   S5). Additionally, there are 8 RNA editing events that released or
   sequestered miRNAs and indirectly resulted in the dysregulation of 1146
   other genes ([201]Table S8). Finally, there is one RNA editing event
   that modified a miRNA seed region, which was predicted to potentially
   disturb the regulations of four genes ([202]Table S10). Of the 73 RNA
   editing events that altered miRNA binding ([203]Tables S5, S8 and S10),
   37 events had differential frequencies in PD and in more severe PD
   samples (p < 0.1, [204]Table S12). Of the 1179 genes affected by
   editing-mediated miRNA regulatory alterations ([205]Tables S5, S8 and
   S10), 598 genes displayed abnormal expressions in PD (p < 0.05;
   [206]Table S13, Figure S3) based on PPMI and four GEO datasets
   ([207]GSE165082, [208]GSE99039, [209]GSE72267, and [210]GSE18838).
   According to these analyses and the PD-related functions of the
   affected genes, 25 A-to-I RNA editing biomarkers for PD are proposed,
   including the 3 editing events in the EIF2AK2, APOL6, and miR-4477b
   seed regions. These biomarkers may alter the miRNA regulation of 133
   PD-related genes.

   To increase the reliability of our study, both QTL and Pearson
   correlation tests were used to identify the associations between A-to-I
   RNA editing events and genes. The associations were only determined
   when they all passed the statistical significance thresholds from the
   two tests. Among the 627,064 significant pairs identified by QTL
   analysis, 620,840 also passed Pearson correlation tests (p < 0.05). Of
   them, 620,616 editing–gene associations were still significantly
   correlated after Benjamini–Hochberg corrections (p.adjust < 0.05). To
   understand the mechanisms underlying these associations, two widely
   used miRNA-binding prediction tools, TargetScan and miRanda, were
   applied to identify the effects of RNA editing on the alterations of
   miRNA binding. The altered miRNA–RNA interactions were only kept when
   they were identified by both tools. All these criteria ensured the
   accuracy of the editing–gene associations and the miRNA–RNA
   interactions.

   Our study identified potential RNA editing biomarkers for PD and
   elucidated their effects on downstream genes and pathways. We presented
   three editing events associated with dysregulated PD-related genes.
   These A-to-I RNA editing events may also be affected by other factors,
   such as the main editing enzyme (ADAR) and genetic variants [[211]74]
   ([212]Figure S4). Collectively, DNA mutations, ADAR activities, A-to-I
   RNA editing events, miRNA regulations, and gene expressions can explain
   the regulatory processes in neurodegenerative pathogenesis.

   Additional research can be conducted to further understand RNA editing
   effects on PD pathogenesis. For example, multiple RNA editing events
   can affect multiple miRNA–RNA regulations and produce complex outcomes
   on PD-related genes ([213]Figure S5). A network to analyze the effects
   of editing modules on PD hub genes is needed. Furthermore, studies are
   also needed to explain the many editing–gene pairs identified in our
   study. Of the 624,325 editing–gene pairs, only 1838 associations can be
   explained by the miRNA–RNA mechanisms presented in this study. A number
   of gene-associated RNA editing events are located in RNA-binding
   proteins or their targets ([214]Figure S6). These can affect the
   regulations of the expression of multiple genes. Future studies can be
   conducted to fully capture the regulatory effects of the RNA-binding
   proteins that are affected by RNA editing. These two studies will help
   further understand RNA editing functions in PD pathogenesis.

5. Conclusions

   In sum, we systematically analyzed the effect of A-to-I RNA editing
   events on the miRNA regulations of PD-related genes. In total, we
   discovered 25 A-to-I RNA editing biomarkers for PD. These biomarkers
   may alter the miRNA regulation of 133 PD-related genes. Three RNA
   editing events in the EIF2AK2, APOL6, and miR-4477b seed regions were
   discussed in detail. Their effects on the downstream genes and pathways
   related to PD were revealed. Our work helps researchers understand the
   mechanisms of the abnormal expressions of PD-related genes and the
   impact of A-to-I editing in neurodegeneration.

Acknowledgments