Multi-omic landscaping of human midbrains identifies disease-relevant molecular targets and pathways in advanced-stage Parkinson's disease

1 BACKGROUND

Parkinson's disease (PD) is the fastest-growing neurodegenerative disorder and affects up to 2% of individuals aged over 60 years.¹ While the chronic and progressive motor dysfunction is mostly due to degeneration of dopaminergic neurons in the nigrostriatal pathway, PD is now recognized to be a systemic disorder involving multiple other regions of the nervous system.² The sporadic form accounts for most cases of PD, whereas only up to 3% of cases comprise the autosomal forms.³ Several environmental and genetic factors are known to increase the disease risk.⁴ Recent studies point to a multifactorial pathogenesis that may differ between patients, suggesting that PD is not one homogeneous disease entity, but rather a syndrome with a unifying clinical phenotype and numerous molecular subgroups. The existence of non-motor symptoms that appear many years before the onset of motor manifestations suggests that molecular mechanisms may have a long lead-up period and result in chronic degeneration.⁵ One of the pathological hallmarks of PD is the presence of Lewy-bodies (LB), intracytoplasmic inclusions that are majorly composed of the protein alpha-synuclein (αSyn), but also contain ubiquitin and neurofilaments.⁶ Such proteinaceous inclusions occur in both familial and sporadic forms of PD, suggesting that defects in the protein handling machinery are directly related to the pathogenesis of the disease.⁷

Our current understanding of disease progression is still elusive, but considers the involvement of different disease mechanisms in different stages of the disease. Environmental factors are likely to represent triggers that start a pathological cascade of events that lead to the facilitation of molecular alterations, which are further aggravated as the disease progresses.⁵ Furthermore, mechanisms related to the regulation of gene expression have been extensively linked to the development and progression of a variety of brain diseases in recent years. For instance, alterations of miRNA expression have been linked to the development and progression of a variety of brain diseases, including PD,⁸ where deregulated miRNAs have been identified in nervous system tissues and in peripheral fluids.^{9, 10} Decreased levels of miR-133b were identified in the midbrain of patients with PD and in mouse models of PD.^{11, 12} Alterations in the levels of miR-34b/c were found in several regions of PD-affected brains. These miRNAs can mimic impairments in mitochondrial function and oxidative stress, disease mechanisms believed to be crucial for the development of PD.¹³ Two miRNAs (miR-7 and miR-153) were also shown to regulate the expression of αSyn. Interestingly, the former has been found to be altered in the striatum and substantia nigra of patients with PD, as well as in murine models of PD.^{14, 15}

Exploring individual profiles of the transcriptome, the microRNAome, or the proteome in PD-affected brains is a powerful strategy to understand neurodegenerative events that underlie the disease. For instance, defects in iron metabolism,¹⁶ in autophagy,¹⁷ mitochondrial dysfunctions,¹⁸ and dysfunctions in synaptic function^{16, 19} were all captured in previous transcriptomic and/or proteomics studies. Moreover, a number of promising targets were identified by such studies and have been regarded as potential therapeutic targets or disease biomarkers for PD (e.g. NR4A2, ULK1, OR51E2, NRF2, FTL, GGH and BSCL2).^16-21 Nevertheless, singular omic profiling also encompasses a number of limitations, as it provides only a snapshot of the pathological events that might be a part of a much bigger network of deregulation, and fails to capture changes occurring up-/down-stream to the selected omic level (e.g. miRNA regulation or post-translational modifications). Only a few studies have attempted to combine high-throughput profiling techniques to explore the molecular changes that take place in postmortem PD-affected brains, but the number of techniques employed in parallel was limited.^22-25 In order to present a more complete picture of the molecular events taking place in PD, here we present a comprehensive and integrative assessment of postmortem PD-affected and control (CTR) midbrains, capturing the multi-omic landscape of the disease, that is, genomic, miRNAomic, transcriptomic and proteomic levels. Each molecular layer was analysed individually and in an integrative fashion, aiming to depict deregulated pathways that permit the exploration of molecular alterations taking place in PD-affected brains. Overall, our findings point to putative molecular networks involved in the pathophysiology of PD, which might improve disease monitoring and delineate novel druggable targets for this devastating disease

2 METHODS

3 RESULTS

In the present analysis, we leveraged a comprehensive multi-omics dataset of a cohort of midbrain samples including those of patients with PD and those of individuals without any indication of neurodegeneration (CTR). We obtained human midbrain tissue samples from 19 PD and 12 CTR individuals from the Parkinson's UK Brain Bank in two batches. The largest batch—comprising 13 PD and 10 CTR cases—was used as the discovery cohort and processed to obtain protein, RNA and DNA lysates, which were further subjected to a multi-omic analysis (Figure 1A). Our discovery cohort dataset comprised 57 992 total RNA, 31 186 small RNA (of which 4383 are miRNAs) and 2257 proteins (Figure 1B). The second, independent batch of samples, 6 PD and 2 CTR cases—here referenced as replication cohort—was used for the verification of expression levels of selected candidates identified by the DE analyses in the discovery cohort.

3.3 Expression profiles of small and total RNA and integrative approaches

After isolation of total RNA, we analysed the transcriptomic profiles and miRNA expression patterns in our cohorts using small and total RNA sequencing (Figures 2 and 3A; Figure S2A). For the DE analysis, we leveraged two bioinformatics frameworks, which we call “A” and “B” (Figure 2A). These frameworks differed in the pre-processing stage: “A” was data-driven and performed filtering using the distribution quantile information, while “B” took a more supervised approach and removed the negative CTR omics, that is, omics with zero effect on the variance (Methods in the Supporting Information). Following small RNA sequencing, mapping results showed that most sequencing counts were composed of miRNAs in both PD and CTR conditions (90.39% and 92.61% of all mapped counts for different small RNA species, respectively) (Figure 3B). Using the same RNA source, mRNA libraries were prepared using a strand-specific, massive-parallel cDNA library preparation protocol. Sequencing read counts were mapped and assigned to the reference genome, accounting for a total of 46 500 genes with valid read values before filtering.

For both frameworks, data quality assessments were conducted using unsupervised learning methods (Figures S3B–F and S4A–E). Principal component analysis and t-distributed stochastic neighbour embedding of the small RNA seq data revealed that the control (CTR12) did not cluster with the rest of the samples (Figures S3D and S4C). The retrieved neuropathological report for this sample indicated the presence of AD-type Tau pathology-Braak-stage II,³¹ and multiple small demyelinating plaques, which also suggested that it was a potential asymptomatic multiple sclerosis case.³² Following an outlier comparative analysis via the Grubbs’ test (p-value = .033), this sample was regarded as an outlier and removed from further analyses (Figures S3E and S4D).

For all DE analyses presented here, we defined significantly different genes and miRNAs for both frameworks based on p-adjusted < .1. For framework “A”, DE analyses revealed 641 deregulated genes (512 up- and 129 down-regulated in PD) but no DE miRNAs (Figures 2B and 3B,D; Dataset 1 in the Supporting Information). Application of framework “B” resulted in 126 DE genes (105 up- and 21 down-regulated in PD) and 4 significantly DE miRNAs (miR-539-3p; miR-376a-5p; miR-218-5p; miR-369-3p) all of which were up-regulated in PD (Figure 2C and 3B,E; Dataset 2 in the Supporting Information). In addition, the frameworks shared 101 common genes, all with the same regulation directionality regarding PD and CTR. Our results highlight the usage of two distinct analysis pipelines for datasets encompassing multiple characteristics: framework “A” was suitable for diverse data with high variability between biological groups, such as our gene expression data (Figure S5A), whereas framework “B” was more sensitive to omic data with low expression and less dispersion, which is true for our miRNA expression data (Figure S5B).

For integrative assessment of the RNA sequencing datasets, we conducted a miRNA-target prediction analysis of the four significantly DE miRNAs (miR-539-3p; miR-376a-5p; miR-218-5p; miR-369-3p) using databases for validated and predicted targets (Methods in the Supporting Information). We identified six predicted and four validated gene targets for these miRNAs that were also present at deregulated levels in our total RNA sequencing data (Table S6). Among the identified pairs, miR-369-3p/general transcription factor IIH subunit 3 (GTF2H3) and miR-218-5p/Ras-related protein Rab-6C (RAB6C) presented a discordant expression with an up-regulation of the miRNAs and down-regulation of transcripts in PD, representing a valid interaction between miRNAs and their target genes. Additionally, we performed an exploratory integrative analysis that considered significant genes and all of the mapped miRNAs (independent of their significance level) for both frameworks. Here, most of the regulated genes had a corresponding targeting miRNA as a valid interactor (discordant expression) (Figure 3I–L). The expression levels of the most variant miRNAs and genes with valid miRNA pairs for both discovery and replication cohorts were visualized using boxplots (Figure S16). Framework “A” yielded 4795 unique pairs of miRNAs and gene symbols (2459 with concordant levels of expression, and 2336 discordant ones), whereas framework “B” yielded 800 pairs (346 with concordant levels of expression and 454 discordant ones) (Figure 3B,I–L). Based on DE and integrative results, candidate miRNAs and genes were selected for validation by quantitative real-time polymerase chain reaction (qRT-PCR). This included the four miRNAs identified by framework B, which we found consistently and significantly upregulated in PD (Figure S7). In addition, we validated two of the miRNA target-genes captured by our analyses (RAB6C and GTF2H3), and seven additional genes which were either highly deregulated or established PD pathophysiological drivers (Figure S8C). Overall, seven of those nine candidate genes presented significant regulation (p-value < .05), whilst all nine genes displayed concordant gene regulation patterns in validation qRT-PCR and RNA sequencing experiments (Figure S8A–C).

3.6 Protein expression profiling, functional annotation, integration with RNA sequencing data

In order to profile the proteomics changes in our samples, we subjected midbrain tissues of our discovery cohort to SWATH-MS (Figure S2B). Following preparation of an annotated peptide spectral library, we were able to detect and quantitate a total of 2257 proteins across all samples at 1% FDR.

DE analyses after normalization for the discovery cohort revealed 22 significantly deregulated proteins between the PD and CTR groups. In the PD group, 17 proteins were up-regulated while five proteins were down-regulated (Figure 4A; Dataset 3). Analysis of PPI networks including all significantly deregulated proteins revealed several hub proteins and interactors, including TH, selenium-binding protein 1 (SELENBP1), fumarylacetoacetase (FAH), fatty acid-binding protein 5 (FABP5), HSPA1B, MAP2K2, FNIP2 and peroxiredoxin-1 (PRDX1)(Figure 4B). The significantly deregulated proteins were matched to the transcriptomics results (Figure 4C,D), and the four most deregulated genes (up- and down-regulated in PD) with their respective significant protein products were highlighted (up-regulated: chitinase-3-like-protein 1 (CHI3L1), DNAJB1, C1QC and SERPINA1; down-regulated: ALDH1A1, ACTA2, TAGLN and DES; (|log₂FC| > 1.4). Only CHI3L1 appeared significantly up-regulated in the PD group, in both proteomics and total RNA sequencing. Similar to the integration of transcripts and miRNA, all mapped proteins were now related to the transcriptomic analysis and miRNA data, and this revealed several links with previous datasets (Figure 4E–H). Functional enrichment of proteins up-regulated in PD showed the involvement of pathways related to inflammatory and immune responses (12 of the top 20 most significantly enriched categories) (Figure 4I). Despite the small number of down-regulated proteins in the PD group, high-level GO term grouping yielded a functional enrichment to the category's responses to stress, cellular localization and regulation of molecular function (Figure 4J). Consecutively, the expression of selected candidates was verified in an independent replication cohort (Figures S16 and S17).

Finally, protein candidates were selected for validation by Western Blotting based on proteomics DE results and integrative approaches within multi-omic datasets. Six proteins were selected based both on the deregulation shown in proteomics experiments and on the relevance to PD pathology (namely, CHI3L1, HSPA1B, USP12, TH, FNIP2 and ALDH1A1). The down-regulation of TH and FNIP2 was confirmed by Western Blotting (Figure S9E–G). In addition, we also validated the up-regulation of CHI3L1 and HSPA1B (Figure S9B–D). We observed concordant trends that ALDH1A1 levels were lower and USP12 levels were increased in PD patients compared to CTR subjects, on average (although not statistically significant; p-value_ALDH1A1 = .09; p-value_USP12 = .5) (Figure S9B–E).

4 DISCUSSION

The lack of disease-modifying therapies in PD is a constant reminder of the need to better understand the molecular mechanisms underlying its pathology. A better characterization of the molecular changes in PD could allow more effective/etiology-driven therapies for this neurodegenerative disorder. Profiling the molecular landscape of PD-affected brains may reveal pathological events taking place in the course of the disease, both at the cellular and systemic levels, pinpointing deregulated molecular pathways and novel druggable targets. In this study, we leveraged the potential of multi-omic tissue analysis to characterize pathology-related molecular alterations in postmortem midbrain tissue of PD patients and age-correlated controls.

As our analysis aimed to identify pathways that characterize sporadic PD, we first excluded the presence of genetic alterations (known to cause monogenic forms of PD) in our samples. Comprehensive mutational screening using gene panel sequencing and by MLPA indicated the absence of genetic abnormalities for known PD genes in the discovery PD cohort (Figure S1, Table S1). A single-nucleotide VUS in the POLG gene was identified in one PD patient who belonged to the independent replication cohort; therefore, it was not included in the main analysis. Similar nucleotide variances in the POLG gene have been linked to alterations in PD predisposition in Finnish and Chinese populations, but mechanistic evidence of its contribution to the pathophysiology of PD is lacking.^{36, 37} Our findings confirmed that the discovery cohort is composed of idiopathic PD cases and excludes a major influence of genetic alterations in our profiling results. In addition, we found no significant differences in the cohort composition in terms of sex (Table 1), and no significant influence when controlling for sex in the DE analysis (Figure S11). Most importantly, almost all of the candidates that we identified by our integrative analyses (20 out of 23) were present when correcting or not for sex, while most of the sex-specific genes did not present significant regulation in the analysed cohort (Figure S11). Overall, these findings confirm that sex differences did not exert major influences in the analyses presented here.

Next, unbiased hierarchical clustering analyses demonstrated a high heterogeneity and did not permit clustering of subjects according to disease entity in the top most variant samples (Figure 1C,E,G; Figure S6A–C) and identified DE transcripts and small RNAs (Figure 1D,F; Figure S6D–F). In contrast, DE proteins revealed a strong clustering of biological subgroups in the discovery cohort (Figure 1H). In the clustering analysis of the PD cohort, the transcriptomics analysis showed the highest homogeneity among subjects, whereas small RNAs and proteins were more heterogeneous (Figure 1I–K), arguing for changes induced by post-translational modifications. The number of post-mortem samples in this project was too small to identify distinct patient subgroups; however, we recently analysed the diversity of miRNA in cerebrospinal fluid (CSF) of patients with PD, which revealed distinct molecular subgroups that were independent of the clinical phenotype.³⁸ A future analysis of a larger number of samples could be more suited to identify distinct subgroups and will represent an important prerequisite for the development of personalized therapeutic approaches based on the molecular phenotype in PD.

DE analysis of RNA sequencing data was conducted using two different frameworks for the pre-processing stage (Figure 2A), each one having a different focus, for a broader and more complete investigation of different RNA sequencing datasets with distinct count distributions (Figure S5). While framework “A” considered a distribution-driven approach appropriate for sparse data, framework “B” had a variance-based supervised procedure. A considerable number of genes were found to be significantly deregulated in framework “A”, while no statistically significant miRNAs could be detected. In contrast, application of framework “B” resulted in multiple deregulated genes and miRNAs (Figure 2D). Analysis of the RNA sequencing data using two different frameworks provided a holistic view of the data sets and revealed subtle levels of deregulation that might not have been captured if only one or the other system had been used. Because our data spanned multiple omics layers, we examined all deregulated candidates (independent of log₂FC) for integration analysis and then focused on the top hits for individual omics analyses (results with other log₂FC thresholds can be found in Table S7). Overall, this depicts the importance of choosing the most suitable bioinformatic paradigm covering the variety of human expression data (Figure S5).

Differential transcriptome analyses revealed that there are more up- than down-regulated genes in PD (Figure 2B,C), which is in line with a recent meta-analysis on substantia nigra transcriptome.³⁹ Transcriptomic overexpression was on average more common in the PD groups both at the single gene level and also when analysing chromosomal segments with expressed sequencing tags. Interestingly, despite the different transcripts obtained with each framework, both the regulation directionality and the enrichment results were very similar in both settings. Up-regulated genes in PD were largely enriched to inflammatory and immune responses, but also to terms related to stress and apoptotic responses, cytoskeleton organization, differentiation and development and metabolic-related processes (Figure 2E; Figure S12A,B). Processes related to immune cell proliferation and defence response were more enriched in framework “A” than they were in “B”, which might be due to the smaller number of significantly deregulated genes yielded by the latter (Figure 2D). Similarly, KEGG pathway enrichment results are related either to inflammation/immune responses, or to a variety of infectious diseases (Figure 2G), likely because pathways of infectious diseases contain genes related to immune cell activation and inflammation. One particular pathway—chemokine signalling pathway—which is directly related to inflammation/leukocyte recruitment,⁴⁰ was shared by both frameworks (Figure 2G; Figure S12).

Moreover, transcriptomics analysis through framework “A” also yielded functional enrichment for stress/apoptosis and metabolic/biosynthetic processes (Figure S12A,B), reflecting features of neurodegeneration occurring in PD-affected midbrains.² Metabolic dysfunction has been frequently linked to the pathogenesis of PD⁴¹ and dopaminergic neurons are known to be especially sensitive to oxidative stress and mitochondrial dysfunction.⁴² Strikingly, PD down-regulated genes also showed enrichment for numerous mitochondrial processes, known to be markedly impaired in autosomal recessive forms of PD⁴³ (Figure 2F). Framework “B” led to enrichment for differentiation and development, and a similar enrichment was observed for the deregulated miRNAs. Cell cycle-related proteins play an important role in the survival of mature neurons, and also in neuronal apoptotic processes.⁴⁴ Overall, those findings support the connection between the miRNA and gene expression datasets, encouraging the integrative efforts employed in this study.

For the small RNA data, we were able to identify four differentially expressed miRNAs using framework “B”, all of which were up-regulated in PD (Figure 3G). Similar findings also suggested an overall up-regulation of circulating miRNAs in PD.^{45, 46} Remarkably, all candidates depicted here have already been linked to PD and other neurodegenerative diseases. miR-218-5p was found up-regulated in peripheral blood mononuclear cells (PBMCs) of PD patients and its levels were reversed after deep brain stimulation therapy.⁴⁷ In the CSF of Alzheimer's disease (AD) patients, miR-218-5p was found to be decreased.⁴⁸ miR-376a-5p was found increased both in PD PBMCs and in vitro. Interestingly, it regulates genes that are involved in mitochondrial dysfunction and were previously linked to PD pathogenesis (TFAM; PGC1α),⁴⁶ but also SIRT2,⁴⁹ a protein implicated in PD for its effects in α-syn aggregation.⁵⁰ It has also been postulated as a biomarker for PD, since its levels in PD PBMCs also seem to correlate with disease severity.⁵¹ One study showed elevated levels of miR-369-3p in PD postmortem substantia nigra,⁵² while alterations in the striatal levels of miR-539-3p were reported in a rodent model of PD.⁵³ Functionally, these miRNAs were linked to neuron-related gene ontology terms, indicating the neuronal origin of these miRNAs (Figure 3J–L). Additionally, terms related to cell-cycle, proliferation, differentiation and development were highly enriched. These are known to also play a vital role in the survival and maintenance of mature neurons⁵⁴ and are directly influenced by miRNA levels during neurodegeneration.⁵³ In concordance with the DE analyses, WGCNA results also revealed a strong enrichment of up-regulated genes in inflammation-related modules (Figure 5), as well as modules related to nervous system development, cell metabolism and transcriptional regulation (Figure 5 and Figure S13). Highlighting the importance of these pathways, the results of the WGCNA also underscore the role of the four deregulated miRNAs identified in the DE analysis.

To obtain a more holistic view, the RNA sequencing datasets comprising transcriptome and microRNAome were analysed in an integrative fashion (Figure 3E–I). PD up-regulated genes included transcription factor E2F2, implicated in hemisphere-dependent PD pathology⁵⁵ and neuroinflammation after spinal cord injury,⁵⁶ Glutathione-specific gamma-glutamylcyclotransferase-1 (CHAC1), reported to regulate unfolded protein response⁵⁷ and linked to paraquat-induced neurotoxicity,⁵⁸ and CXCR4, a chemokine receptor linked to neuroinflammation which is expressed in dopaminergic neurons.⁵⁹ The latter was experimentally validated and showed strong up-regulation in the PD samples analysed here (Figure S8B). DRD2 and OR51E2 presented marked down-regulation in PD. The former has been found in altered levels in PD substantia nigra and has been regarded as a susceptibility locus for PD,^{39, 60} whereas the latter is postulated as a therapeutic target for PD. It is involved in neuromelanin pigmentation in dopaminergic neurons and found down-regulated in PD-affected brains.^{21, 61}

When considering only significantly differentially expressed miRNAs and transcripts, the list was reduced to two valid miRNA-target gene pairs: miR-218-5p/RAB6C and miR-369-3p/GTF2H3 (Table S4). Importantly, we have validated the expression of those four candidates by qRT-PCR, adding further evidence to the relevance of these pairs for the pathophysiology of the analysed PD brains. The RAB6C is a member of Rab GTPases, which are pivotal regulators of intracellular protein transport and vesicle trafficking.⁶² RAB6C has been implicated in autophagy and proteostasis and increased levels of Rab6 isoforms have been linked to protein aggregation in vitro⁶³ Importantly, Rab GTPases are substrates of LRRK2, the most commonly mutated gene in familial cases of PD. Disruption of Rab phosphorylation in the LRRK2 site leads to neurotoxicity in vitro and dopaminergic neurodegeneration in vitro.⁶⁴ The GTF2H3 is directly involved in RNA transcription and nucleotide excision repair.⁶⁵ Although GTF2H3 has not been directly implicated in PD, dysfunctions in nucleotide excision repair play an important role in chronic neurodegenerative disorders.⁶⁶ Moreover, members of its family (TFIIH) have been shown to play a role in oxidative stress and mitochondrial DNA alterations in PD.⁶⁷ Altogether, these candidates underline pathways that are involved in the pathogenesis of PD and support the involvement of gene expression regulation by miRNAs as a disease mechanism. The regulation of these targets is supported by the independent replication cohort (Figure S9), and the experimental validations of selected genes (Figures S7 and S8).

Finally, proteomic profiling ensured a more complete characterization of disease-relevant post-translational mechanisms. In line with transcriptomics, most differentially regulated proteins were increased in PD (Figure 4A). As expected, we found a marked down-regulation in tyrosine hydroxylase (TH), the rate-limiting enzyme for dopamine synthesis and a hallmark of dopaminergic neuron depletion in PD.^{2, 68} PPI networks revealed other important hub-proteins (Figure 4B), such as the SELENBP1, FAH, FABP5 and PRDX1. Remarkably, we previously identified selenium as part of a bioelement-signature in the CSF that allowed for differentiation between PD and CTR samples.⁶⁹ FAH and PRDX1 were reported to play a role in oxidative stress damage in PD, and the latter has also been postulated as a new therapeutic target for PD because of its role in the generation of reactive oxygen species.^{70, 71} FABP5 has been linked to postnatal neurogenesis,⁷² and also to neuronal oxidative damage in vitro.²⁰ Another important PD up-regulated protein that was experimentally validated was heat shock 70 kDa protein 1B (HSPA1B), a chaperone known for its role in protein folding and degradation as well as neuronal apoptosis (Figure S9B,C)⁷³. Chaperones are known to interact with αSyn,⁷⁴ and defects in that system contribute to αSyn misfolding and the formation of inclusions/protein aggregates in PD.⁷⁵

Significantly deregulated proteins were substantially fewer than deregulated transcripts, most likely because only the most abundant/uniquely identified proteins within all samples were considered for quantification. Furthermore, proteomics and RNA sequencing results are fundamentally very different, starting with the nature of the signals and the coverage of those techniques limiting the overlap between such datasets.⁷⁶ However, their integration resulted in a better understanding of the regulation of identified candidate molecules, for example by comparison of individual protein expression levels to respective RNA sequencing results (Figure 4C,D) and by analysis of expression levels in the independent replication cohort (Figure S16).

An important finding was the identification of CHI3L1 which was up-regulated in both transcriptomics and proteomics datasets, and was validated by both qRT-PCR and Western Blot. It also showed the highest levels of up-regulation in both analyses after integration (Figure 4C). CHI3L1, also known as YKL-40, is widely expressed in immune cells and regulates inflammatory responses, tissue injury and repair.⁷⁷ Remarkably, it has been previously identified in CSF studies as a potential circulating biomarker for several neurodegenerative diseases, including AD, amyotrophic lateral sclerosis and PD.^78-80 It was also considered as a pivotal marker for immune/inflammatory changes in tauopathies.⁸¹ Functional analysis highlighted the GO-BP immune effector process as the most enriched term for the up-regulated proteins in PD (Figure 4I), matching the massive immune/inflammatory activation reported for RNA sequencing results and the deregulation of several immune mediators in the proteomics dataset. Remarkably, functional alterations related to neuroinflammation/immune response activation were captured across all omics results presented in our study and were independent of data handling paradigms. Neuroinflammation has already been considered to be a predictive feature for the appearance of non-motor symptoms and cognitive decline in patients with PD,⁸² and inflammation-related mechanisms seem to accompany the pathophysiological events of PD even from earlier stages.^{5, 83} A recent study showed that neuroinflammatory mechanisms are triggered with the release of αSyn from apoptotic neurons, aggravating the disease through a number of pathways that include microglial activation, mitochondrial damage and inflammasome formation.⁸⁴ Overall, the integrative results from diverse omics layers depicted in this study provide further evidence for the role of neuroinflammation in PD pathology and suggest its exploitation as a therapeutic target.

It is important to stress that factors modulating protein metabolism (e.g., synthesis and degradation rates of proteins) were out of the scope of this study and were therefore not considered for the integrative analyses presented here, although they likely contribute to the overall picture. The analysis of the phospho-proteome and the metabolome was also not included because of the postmortem nature of the source material. Limitations also apply to the number of analysed subjects and the nature of the material analysed, because high-quality post-mortem tissues with excellent clinical characterization are scarce, particularly for CTR subjects without neurodegenerative pathology. Nevertheless, we demonstrated that quality CTR measures, such as RIN and PMI, showed no correlation to data quality (Table S4), similar to what has been reported previously.^85-87 Furthermore, clinical parameters in PD had no major effect on the differential results (Table S5), and there was no difference in cell type composition between samples (Figure S10), excluding a major effect of these factors on the results. While information about medication for the analysed PD cohort was incomplete, the clinical data presented here does not indicate the effects of medication on disease progression. Moreover, Levodopa—the most common symptomatic antiparkinsonian treatment—has been shown to not alter the course of PD in terms of progression.^{88, 89}

A caveat of our data is that we analyse a snapshot of advanced PD stages, that is, patients with a disease duration of 7–25 years. As all analyses strongly underscored the role of neuroinflammation, we do not expect that an increase in sample number would change this overall picture. Nevertheless, an analysis of larger cohorts will be much more powerful for subgroup identification and may identify additional molecular targets characterizing subtypes of PD pathology.

5 CONCLUSIONS

In summary, our integrative multi-omics analyses identified multiple levels of deregulation in human midbrains affected by PD, several of them overlapping across the different datasets. We identified and experimentally validated four deregulated miRNAs: miR-539-3p, miR-376a-5p, miR-218-5p and miR-369-3p. These miRNAs contribute to the regulation of transcription and inflammatory pathways, highlighting these mechanisms in the context of PD pathology. We also identified and validated two miRNA-target gene interacting pairs (miR-218-5p/RAB6C and miR-369-3p/GTF2H3). The deregulation of these players underlines the involvement of vesicle trafficking, proteostasis and oxidative stress in the pathogenesis of PD. Established PD-related proteins, such as TH, were also captured in our study, along with less evident candidates, such as CHI3L1, FNIP2 and HSPA1B.

The integrative view of diverse multi-omics layers confirmed an important enrichment of neuroinflammation-related molecules, underscoring the pivotal role for neuroinflammation in the pathogenesis of advanced PD and delineating a yet unaddressed drug target in this disease stage.⁹⁰ Further mechanistic and clinical studies would be required to substantiate the pathological importance of our findings in the context of earlier disease stages.

CONFLICT OF INTERESTS

The authors declare no conflict of interest.