Mini-Review: The MSA transcriptome

Highlights

• RNA-sequencing and microarray studies assess coding and non-coding transcripts that may be involved in MSA pathogenesis.

• Reports indicate dysfunctions related to inflammation, protein modification, mitochondrial, and autophagy-related processes.

• Links have been found with prion disease and Alzheimer's disease-associated pathways.

• Several differential diagnostic biomarker candidates have been proposed.

• Cross-validation between studies is low; thus, clinical applicability and data reliability are challenging.

Abstract

Multiple system atrophy (MSA) is an atypical parkinsonism that rapidly affects motor ability and autonomic function, leaving patients wheelchair-bound and dependent for daily activities in 3–5 years. Differential diagnosis is challenging as cases may resemble Parkinson's disease or other ataxic syndromes depending on the clinical variant (MSA-P or MSA-C), especially in early stages. There are limited symptomatic treatments and no disease-modifying therapies. Pathologically, alpha-synuclein aggregates are found in glial cytoplasmic inclusions, among other proteins, as well as in neurons. The molecular pathogenesis of the disease, however, is widely unknown. Transcriptomic studies in MSA have tried to unravel the pathological mechanisms involved in the disease. Several biological and molecular processes have been described in the literature that associate disease pathogenesis with inflammation, mitochondrial, and autophagy related dysfunctions, as well as prion disease and Alzheimer disease associated pathways. These reports have also registered several differential diagnostic biomarker candidates. However, cross-validation between studies, in general, is poor, making clinical applicability and data reliability very challenging. This review will go over the main transcriptomic studies done in MSA, reporting on the most significant transcriptive and post-transcriptive changes described, and focusing on the main consensual findings.

Introduction

Multiple system atrophy (MSA) is a rare adult-onset neurodegenerative disease characterized by autonomic failure and motor symptoms that includes two main clinical and pathological variants. Clinically, variants characterize by either an early motor predominance of cerebellar symptoms such as ataxia (MSA-C) or an early poor levodopa responsive parkinsonism (MSA-P). Pathologically, MSA is considered an oligodendroglial synucleinopathy due to the presence of glial cytoplasmic inclusions, mostly containing aggregated α-synuclein, among other proteins [1]. There are two predominant pathological distributions: neuronal loss and gliosis in the cerebellum (known as olivopontocerebellar atrophy or OPCA) or basal ganglia (known as striatonigral degeneration or SND), which usually correlate with the clinical phenotype. Mixed clinical and pathological profiles are common, especially in late stages [2]. MSA-P and SND are more prevalent in North America, while MSA-C and OPCA in Asia. The reason for this different demographic distribution is still unknown, but environmental or genetic risk factors, as well as selection bias, have been proposed as possible explanations [3].

Diagnosis of the disease in vivo is clinical, following consensus guidelines [4]. In clinical practice, however, physicians may misdiagnose 20 % and up to 40 % of patients with a clinical diagnosis of MSA as Parkinson's disease (PD) or other neurodegenerative disorders such as Lewy body dementia (LBD) or progressive supranuclear palsy (PSP) [5,6]. Early diagnosis is especially difficult, and most cases fulfilling clinical criteria for probable MSA are already in late stages of the disease [7]. The need for objective biomarkers for an earlier and more accurate diagnosis, especially considering upcoming clinical trials for disease-modifying treatments, is critical.

A significant handicap is the uncertain molecular pathogenesis of the disease. In other words, why does α-synuclein accumulate in oligodendrocytes? In general, familial clustering is rare or absent with no clear causative gene; thus, it is considered a sporadic disease, though there may be genetic susceptibility in some cases [8]. Since α-synuclein aggregation seems to be a key player in the pathogenesis of the disease, its gene (SNCA) has been explored in several case-control specific and genome-wide association studies (GWAS). Though some reports have found evidence for a genetic association of SNCA variants with MSA [9,10], a large study with almost 1000 MSA cases did not, and instead found variants in the MAPT gene and others [11]. Other proposed genetic variants for increased MSA susceptibility include APOE, COQ2, LRRK2, SLC1A4, SQSTM1, EIF4EBP1, and GBA [[12], [13], [14], [15]]. However, results vary, are not always replicable, and in general, specific links to polymorphisms and monogenic mutations are either rare or inconsistent. Hence, other approaches have included studying mRNA α-synuclein levels in specific brain regions to assess if the cause for α-synuclein aggregation may be secondary to an overexpression of the gene. These results are also controversial, and most do not report apparent differences in SNCA expression levels [2,8,[16], [17], [18]].

With the introduction of transcriptome-wide studies using genome-wide microarrays or RNA sequencing, studies have focused on the transcriptional regulatory pathways of α-synuclein and their relation with metabolic and inflammatory processes that may be involved in its pathological accumulation. In general, transcriptomic studies assess mRNA transcripts and non-coding RNA transcripts by comparing each transcript's expression between groups (disease vs. normal controls or between disorders). If there is a significant differential expression (DE), the transcripts can be assessed as candidate diagnostic biomarkers. Furthermore, a functional analysis (by gene ontology, enrichment analyses, or others) assessing the available in-silico and in-vitro data can predict what genes are targeted by these transcripts. Thus, giving us additional information on the disease's molecular pathogenesis, which may be useful for therapeutics. Hence, in MSA, RNA sequencing studies assessing brain tissue [19,20] have reported a series of DE genes that may be involved in the disease's pathological pathways [21]. In addition, alternative splicing studies have also found different SNCA isoforms that vary in the 5′ UTR regions and may interact with other regulatory proteins [22].

Moreover, non-coding RNAs that take part in post-transcriptive processes, such as microRNAs [23] and more recently circular RNAs [24], have also been studied for their use as potential diagnostic biomarkers in MSA and other neurodegenerative diseases. Yet, candidates often vary among studies making replicability and external validation for clinical applicability rather challenging. In this review, we will briefly go over the studies involving the human transcriptome's role in MSA, taking into account the principal transcription, post-transcription, and epigenetic studies that have been published.

In 2015, Mills et al. [19] performed the first RNA sequencing (RNA-Seq) study in human MSA brains. Frozen brain tissue was assessed from 6 MSA cases and 6 controls, specifically isolating gray (GM) and white matter (WM) from the frontal gyrus. The motivation behind selecting this relatively well preserved region was to avoid altered gene expression that could be secondary to the effects of cell loss rather than to more specific molecular pathology. The group performed two primary comparative analyses: one between both tissues (GM and WM) in MSA brains and another between MSA and healthy control (HC). There were four main findings in this study. The first was that gene clustering was dependent on tissue type rather than disease state, suggesting that the transcriptome is tissue specific and that the disease itself causes only moderate alterations in the transcriptome. Second, differential expression (DE) between MSA and healthy brains suggested a down-regulation of inflammatory processes and overexpression of all hemoglobin chains in MSA WM, along with a specific down-regulation of the gene TTR in MSA GM. Third, although not significant, the expression of SNCA was higher in both MSA tissues, with MSA GM having the highest overall expression levels of SNCA. In addition, the analysis of SNCA alternative splicing revealed the presence of three transcriptional isoforms. Finally, there was a greater number of long-intervening-noncoding RNAs (lincRNAs) in MSA brain tissue. These lincRNAs were specifically up-regulated in MSA WM when compared with MSA GM. These results led the group to further extend their research by including an even broader analysis of the transcriptome using strand-specific RNA-Seq (ssRNA-Seq) on the same cohort. This technique allows transcripts to be mapped back to the DNA strand of origin, thus identifying antisense transcripts. Hence, in 2016, they [20] reported that antisense transcription was prevalent throughout the control and MSA affected brain, though they could not identify any significantly DE antisense transcripts. An overexpression of lincRNAs in MSA brains was again reported, as well as overexpression of HBB (member of the hemoglobin protein complex) in MSA GM, and IL1RL1, TTR, and LOC389831, among others genes. A modest, non-significant, overexpression of SNCA was found as well, and the study of isoforms, in this case, revealed three different SNCA variants and one antisense transcript. Two of the SNCA splice variants had a different RNA secondary structure, which was postulated could impact the protein coding capabilities of SNCA, and in turn, its ectopic accumulation in oligodendrocytes. Considering that the non-coding transcriptome could have an essential role in MSA, the group published a new work [25] describing the differential expression of circular RNAs (circRNA), produced by backsplicing of precursor mRNAs, in MSA brains. In this case, they analyzed the previous transcriptome sequence data [20] using two specific circular RNA finders (circRNA_finder and CIRI packages), revealing 20 and 33 DE circRNAs, respectively. Five (IQCK, MAP4K3, EFCAB11, DTNA, and MCTP1 gene loci) were present in both DE lists and were later validated by RT-qPCR. The group also reported on circular hotspots (genes expressing more than ten distinct circRNA isoforms in a given tissue), revealing 21 genes in the MSA transcriptome, of which ten were specific to MSA. Again, there was a general increase in the non-coding circRNA transcriptome expression in MSA compared with controls. Of note, the linear counterparts of the MSA-specific circRNA did not show DE changes.

Finally, the most extensive RNA-Seq study to date was published by Piras et al. [26]. They used post-mortem cerebellar tissue from two cohorts, analyzing a total of 66 MSA samples and 57 healthy controls. Each cohort underwent RNA-Seq. DE genes were obtained by combining the results from both cohorts using a meta-analysis approach, revealing a set of DE genes that ranged from 1 (MSA-P) to 747 (MSA-C) depending on the MSA clinical subtype. Functional analysis of these genes curiously revealed amyloid-β metabolism as the most enriched network module with a central role for the APP gene. All 747 DE genes were then classified using an external database to estimate specific cell-type expression. The highest significance was found for oligodendrocyte (down-regulated in MSA) and neuronal genes (up-regulated in MSA). The group subsequently conducted a Weighted Correlation Network Analysis (WGCNA) in the MSA-C cohorts to identify disease-specific modules or clusters of co-expressed genes and networks. The gene ontology (GO) enrichment analysis of these modules also found an up-regulation for synaptic functional classes (hub genes included TIAM1 and SYNGAP1) and a down-regulation for myelination and oligodendrocyte classes (top hub gene was QKI). In parallel, the group specifically studied transcription in oligodendrocytes by laser capture microdissection, reporting a total of 187 DE genes in oligodendrocytes. In this case, the functional analysis detected a network including four modules enriched for telomere maintenance, non-coding RNA processing, immune processes, and cell growth. Myelination processes were also identified, mostly due to down-regulated genes when analyzing the complete gene list in oligodendrocytes (Table 1a).

In conclusion, all these studies used brain tissue (though different regions). There was an active MSA transcriptome with important roles for non-coding transcripts (lincRNAs, circRNA, antisense transcripts). Reports suggest a down-regulation of oligodendrocyte genes specialized in myelination processes and up-regulation of synaptic genes, as well as essential links to inflammation, iron homeostasis, cytoskeleton/collagen related processes, and possibly common transcription processes related to amyloid aggregation

Although next-generation sequencing platforms such as RNA-Seq are the gold standard for transcriptomic studies, microarrays still offer a high throughput analysis at a more affordable price and less detailed computational analysis. Two genome-wide transcriptomic studies have been published using microarray platforms in MSA. Table 1

Langerveld et al. [27] published the first genome-wide transcriptomic study in MSA rostral pons tissue in 2007. They selected pons tissue since it is specifically affected in both MSA variants and less so in other neurodegenerative diseases. An Affymetrix high throughput microarray identified 282 significantly DE transcripts (198 were down-regulated and 84 were up-regulated). Eighty-four of the down-regulated genes and 13 of the up-regulated genes were statistically associated with the amount of aggregated α-synuclein present in pons tissue. Gene ontology analysis found enrichment of genes related to mitochondrial function, protein modification, metabolism/glycolysis, ion transport, and proteasome structure. On the other hand, most up-regulated genes were related to transcription/RNA modification, signal transduction, and inflammation/response to stress.

Recently, our group [28] assessed the MSA transcriptome using whole blood from 20 MSA cases (10 MSA-P and 10 MSA-C), compared to 10 Parkinson's disease (PD) cases and 10 healthy controls in an attempt to find differential transcripts for biomarker use. By a resampling analysis, differential expression was dependent on the clinical variant, with more DE transcripts when assessing MSA-P cases against controls (113) than MSA-C vs. controls(8). In contrast, when comparing MSA variants to PD, more DE transcripts were found for MSA-C vs. PD. Functionally, we also found that MSA-P and MSA-C seemed to have different enriched biological pathways, with MSA-P cases appearing to share more common gene sets with PD than MSA-C cases. Biological processes were extracted by gene set enrichment analysis (GSEA), mainly finding an up-regulation of inflammation and down-regulation of protein translation, targeting, and modification processes in MSA cases. In addition, both MSA-P and PD groups revealed a down-regulation of gene sets related to cellular differentiation and development of the nervous system.

Similarly, microarray studies in PD have identified several differential transcriptome bio-signatures; however, when assessing these studies altogether, there is little agreement between individual gene lists, even when using the same tissue. Alternatively, when lists of differentially expressed genes are grouped into functional pathways, the agreement between different studies is improved [29]. Following this principle, in MSA, we can see that inflammation and protein modification processes in both microarray studies seemed to be dysregulated and are probably key players in the disease's molecular pathogenesis.

Micro RNAs (MiRNAs) are small non-coding RNAs with post-transcriptional regulatory functions. MiRNAs bind to complementary mRNA sequences causing a translational inhibition or degradation of the mRNA target, thus changing the expression and translation of proteins in a specific situation. These regulatory RNAs have been associated with several developmental and physiological processes, as well as with several diseases, aging, and neurodegeneration [23]. In neurodegenerative diseases such as PD and MSA, miRNAs may modulate proteins, autophagy/lysosome pathways, and mitochondrial function [30]. Hence there is a particular interest in assessing specific disease-related miRNAs for use as biomarkers or therapeutic targets. In MSA, there have been several studies examining the differential expression of miRNAs. Most studies have analyzed the human brain (mainly frontal lobe and cerebellum) but also blood and CSF. Yet, similar to the above transcriptomic studies, there is little consensus on the individual miRNA lists. Table 1.

The first to report miRNA changes in MSA brains were Lee et al. [31]. They analyzed human cerebellum comparing MSA to healthy controls and identified 9 down-regulated miRNAs and 2 up-regulated miRNAs. The most up-regulated miRNA was miR-202, and one of its mRNA targets (Oct1) was found to be decreased in cerebellar tissue. Furthermore, in vitro studies found a correlation between reduced expression of Oct1 and greater oxidative damage in neuronal cells.

Subsequently, Ubhi et al. [32] published a cross-species (human and mouse) cross-disease approach analyzing miRNA profiles in MSA, LBD, Alzheimer's disease (AD), PSP, and corticobasal degeneration (CBD). MSA was the disease that showed most DE miRNAs (all up-regulated, followed by AD and PSP). Moreover, the cross-disease analysis indicated that many of these miRNAs were altered in more than one disorder. The miRNA processing machinery (mRNA levels of Drosha, DGCR8, Exportin5, Dicer1, and Ago2) was also studied, but the group did not find any alterations, thus suggesting that the dysregulation was probably disease-related. Specifically for MSA, an up-regulation of miR-96 was found across the human and mouse samples. Potential gene targets were identified, including FOXO, SOX, FYN, neuregulin, and SLC proteins (SLC1A1 and SLC6A6), with various putative roles that could be involved in neurodegeneration such as neuronal antioxidant function, zinc homeostasis, and excitotoxicity.

Instead, Wakabayashi et al. [33] used formalin-fixed paraffin-embedded samples from the cerebellum and pons of MSA and control cases. None of their up-regulated miRNAs were up-regulated in the study by Ubhi et al. However, 6 of 33 miRNAs down-regulated in the pons, as well as 5 of 18 miRNAs down-regulated in the cerebellum, were up-regulated in Ubhi et al.'s study [32]. In comparison with Lee et al. [31], 3 miRNAs (miR-129-2-3p, miR-129-5p, and miR-132-3p) were found to be down-regulated in the cerebellum of both studies and 2 of the miRNAs (miR-129-2-3p and miR-129-5p) found to be down-regulated in the pons, were down-regulated in the cerebellum as reported by Lee et al. [31].

A follow-up study [34] for Ubhi et al.' s [32] findings analyzed the expression levels of 6 specific autophagy-regulating miRNAs in the striatum (caudate nucleus and putamen) of MSA-P cases. They reported a significant increase of 2 miRNAs (let-7b and miR-101) and a decrease of miR-34c. In this case, the dysregulation of miR-96, found in their previous study, was not significant [32]. Subsequently, the paper focused its work on miR-101, demonstrating that miR-101 indeed inhibited autophagy in oligodendrocytes. Simultaneously, the expression of an antimiR-101 agent improved some of the deficits in-vitro and in an in-vivo mouse model of MSA. Conversely, a recent study by the group [35] (also analyzing the striatum of MSA-P cases) only replicated DE for let-7b. Of interest, they found that gene expression was increased in glial-type cells and decreased in neurons, and analysis of neuronal, glial, and endothelial markers in the array showed significant changes in oligodendroglial markers. Their functional studies reported links with extracellular matrix-receptor interactions, neuroinflammation, prion disease, as well as with Alzheimer's disease. Their findings replicated some of Wakabayashi's findings as well: down-regulation of miR-129-2-3p, miR-123-3p, miR-128-3p, miR-149-5p, miR-124-3p, and miR-379-5p and up-regulation of miR-1290 and miR-23a.

In blood, miRNA studies have also found variable results. In 2014, Vallelunga et al. [36] were the first to complete a whole serum miRNA transcriptome profiling in MSA. They compared MSA to controls and PD cases, identifying 12 and 5 DE miRNAs, respectively. By RT-qPCR, using a new validation set (n = 75), they confirmed up-regulation of miR-24, miR-148b, miR-223*, miR-324-3p, down-regulation of miR-339-5p in MSA vs. controls, and up-regulation of miR-24, miR-34b, and miR-148b in MSA vs. PD. Only mir-339-5p was specifically down-regulated in MSA, but not in PD, compared with controls. In contrast, a later study by this group [37], evaluating two specific miRNAs (miR-30c-5p and miR-148b) as differential diagnostic biomarkers between MSA and PD, only found significant differences between MSA and PD for miR-30c-5p (previously not identified in the MSA group). The sensitivity of miR-30c-5p to discriminate between MSA and PD was 82 %; however, specificity was low (54 %).

Likewise, using a microarray for serum samples, Kume et al. [38] found 67 dysregulated miRNAs; among these, miR‑24 and miR‑223 were found up-regulated as described previously [36]. The most up-regulated miRNA was miR‑16, and the group also found an up-regulation of the let‑7 family. MSA-P cases were compared to MSA-C as well, identifying 22 up-regulated miRNAs and 17 down-regulated miRNAs in MSA-P serum. Recently, our group [39] confirmed DE of miR-24, miR-16, and let-7, as well as miR103a-3p, miR‐106a‐5p, miR‐107, and miR‐25‐3p (common to Kume et al. [38]). We also replicated five miRNAs that had been initially reported by Kim et al. [35] to be DE in MSA striatum (miR‐24‐3p, miR‐93‐5p, miR‐25‐3p, miR‐181a‐5p, and let‐7b‐5p). Similarly, the functional analysis reported links with prion disease, fatty acid metabolism, and NOTCH signaling (implicated in oligodendrocyte demyelination [40]). MSA cases were also compared to PD cases, and miR‐7641 and miR‐191‐5p were the most differential miRNAs in two different sample sets.

Uwatoko et al.'s study [41]; however, analyzing plasma instead of serum, reported opposing results. In this case, after microarray analysis and RT-qPCR validation (of the top 11 DE miRNAs), miR-671-5p, miR-19b-3p, and miR-24-3p were found to be DE in MSA and PD. Conversely, miR-24-3p was found to be down-regulated in MSA instead of up-regulated. This group also found differences among MSA variants, with MSA-P cases showing more similar results to PD (miR-671 down-regulated in both cases). Similar results were found by Marques et al. [42], in this case, using CSF. They selected ten miRNAs from literature and evaluated their expression levels in PD, MSA, and control subjects, trying to find differential miRNAs between diseases. Only four miRNAs were found to be DE in MSA CSF (miR- 19a, miR-19b, miR-24, and miR-34c) compared to controls, and in this case, they were also found to be down-regulated. No individual miRNA was found to significantly differentiate PD from MSA. However, using the combination of miR-133b and miR-148b, the ROC analysis showed an AUC of 0.77, proposing that combination panels may be more useful. In this sense, another CSF study [43], assessing a subset of miRNAs taken from a previous pilot study and from other miRNAs described in the literature, found that miR-7-5p, miR-34c-3p, and let-7b-5p (AUC = 0.88) were the best combination for discriminating MSA from controls, and miR-9-3p and miR-106b-5p were the best subset to differentiate between PD and MSA (AUC = 0.87). The group also studied miRNAs in plasma, though they found no similarity with their CSF miRNA findings. In plasma, the best differentiating subset of miRNAs between PD and MSA were miR-92-3p, miR-10a-5p, and miR-1-3p (AUC = 0.73).

As seen, there is little consensus between miRNA studies, even when using the same tissues, although mir-24 and let-7 seem to come up in several reports. When aiming to replicate results from previous literature, there is an evident lack of reliable reproducibility, possibly due to methodological differences, including differences in sample storage procedures, arrays or panels used, endogenous normalizers for validation, analytical software, statistical significance cut-offs, etc., as well as due to the intrinsic clinical heterogeneity of the disease or polyvalent nature of miRNAs. Perhaps, a "gold standard" procedure should be fixed to harmonize results better, searching for more reliable combinations of miRNAs if we wish to use these techniques as biomarker tools.

The first epigenome-wide study evaluating post-mortem brain tissue from MSA cases was published by Bettencourt et al. [44]. This group studied DNA methylation in white matter from three brain regions (cerebellum, frontal lobe, and occipital lobe) in MSA mixed subtypes. Subsequently, in a follow-up phase, they replicated their results in another cohort of patients that also included SND and OPCA variants, studying in this case, cerebellum only. Results pointed towards several genes that are mainly expressed in oligodendrocytes, such as HIP1 and MOBP. Eleven CpGs were consistently identified across MSA mixed-subtype cohorts reaching nominal significance, which mapped to HIP1 (cg15769835), an intergenic CpG on chromosome 15 (cg20123217), and LMAN2 (cg23483530). Stronger effects were dependent on the pathological variant. Co-methylation networks identified 10 out of 45 modules that seemed disease-specific, with enriched pathways that included cell death signaling via NRAGE, NRIF, and NADE; axon guidance, cerebellar white matter atrophy, infection, and drug metabolism-related pathways, among others.

A recent study by Rydbirk et al. [45] also investigated DNA methylation but complemented their findings with hydroxymethylation levels (separate 5mC and 5hmC levels, while Bettencourt et al. investigated total levels). This group identified five differentially methylated probes in the 5mC fraction, two of which mapped to gene bodies in AREL1 or KTN1 genes, with no significant MSA subtype-specific methylation differences. The group also validated this finding by RT-qPCR, showing increased AREL1 expression levels in MSA brains and increased MHC class I HLA gene expression. To directly compare their results with Bettencourt et al.'s study, they investigated total methylation levels; however, the most significant probes on HIP1 and LMAN2 genes were only nominally significant, and no probes on MOBP were significant. Furthermore, using blood samples from a new MSA patient cohort, they investigated the composition of peripheral blood and identified a decrease in the fraction of non-classical CD14 + CD16++ monocytes, suggesting an active neuroimmune response with increased antigen presentation and reduction of non-classical monocytes. Bettencourt et al. later complemented this work by publishing a letter [46] reporting that an additional loci-specific analysis had been done, showing no differential expression for AREL1 and only a nominally significant down-regulation of KTN1. Furthermore, only HLA-A showed a nominally significant up-regulation in one cohort group. These conflicting results may be due to tissue and cell-type differences, as Rydbrick et al. included gray matter, and AREL1 is highly expressed in neurons. Table 1