Serum MicroRNAs Predict Isolated Rapid Eye Movement Sleep Behavior Disorder and Lewy Body Diseases

ABSTRACT Background Isolated rapid eye movement sleep behavior disorder (IRBD) is a well‐established clinical risk factor for Lewy body diseases (LBDs), such as Parkinson's disease (PD) and dementia with Lewy bodies (DLB). Objective To elucidate whether serum microRNA (miRNA) deregulation in IRBD can antedate the diagnosis of LBD by performing a longitudinal study in different progression stages of IRBD before and after LBD diagnosis and assessing the predictive performance of differentially expressed miRNAs by machine learning–based modeling. Methods Using genome‐wide miRNA analysis and real‐time quantitative polymerase chain reaction validation, we assessed serum miRNA profiles from patients with IRBD stratified by dopamine transporter (DaT) single‐photon emission computed tomography into DaT‐negative IRBD (n = 17) and DaT‐positive IRBD (n = 21), IRBD phenoconverted into LBD (n = 13), and controls (n = 20). Longitudinally, we followed up the IRBD cohort by studying three time point serum samples over 26 months. Results We found sustained cross‐sectional and longitudinal deregulation of 12 miRNAs across the RBD continuum, including DaT‐negative IRBD, DaT‐positive IRBD, and LBD phenoconverted IRBD (let‐7c‐5p, miR‐19b‐3p, miR‐140, miR‐22‐3p, miR‐221‐3p, miR‐24‐3p, miR‐25‐3p, miR‐29c‐3p, miR‐361‐5p, miR‐425‐5p, miR‐4505, and miR‐451a) (false discovery rate P < 0.05). Age‐ and sex‐adjusted predictive modeling based on the 12 differentially expressed miRNA biosignatures discriminated IRBD and PD or DLB from controls with an area under the curve of 98% (95% confidence interval: 89–99%). Conclusions Besides clinical diagnosis of IRBD or imaging markers such as DaT single‐photon emission computed tomography, specific miRNA biosignatures alone hold promise as progression biomarkers for patients with IRBD for predicting PD and DLB clinical outcomes. Further miRNA studies in other PD at‐risk populations, such as LRRK2 mutation asymptomatic carriers or hyposmic subjects, are warranted. © 2022 The Authors. Movement Disorders published by Wiley Periodicals LLC on behalf of International Parkinson and Movement Disorder Society

A BS TRACT: Background: Isolated rapid eye movement sleep behavior disorder (IRBD) is a well-established clinical risk factor for Lewy body diseases (LBDs), such as Parkinson's disease (PD) and dementia with Lewy bodies (DLB). Objective: To elucidate whether serum microRNA (miRNA) deregulation in IRBD can antedate the diagnosis of LBD by performing a longitudinal study in different progression stages of IRBD before and after LBD diagnosis and assessing the predictive performance of differentially expressed miRNAs by machine learning-based modeling. Methods: Using genome-wide miRNA analysis and realtime quantitative polymerase chain reaction validation, we assessed serum miRNA profiles from patients with IRBD stratified by dopamine transporter (DaT) singlephoton emission computed tomography into DaTnegative IRBD (n = 17) and DaT-positive IRBD (n = 21), IRBD phenoconverted into LBD (n = 13), and controls (n = 20). Longitudinally, we followed up the IRBD cohort by studying three time point serum samples over 26 months. Results: We found sustained cross-sectional and longitudinal deregulation of 12 miRNAs across the RBD continuum, including DaT-negative IRBD, DaT-positive IRBD, and LBD phenoconverted IRBD (let-7c-5p, miR-19b-3p, miR-140, miR-22-3p, miR-221-3p, miR-24-3p, miR-25-3p, miR-29c-3p, miR-361-5p, miR-425-5p, miR-4505, and miR-451a) (false discovery rate P < 0.05). Age-and sex-adjusted predictive modeling based on the 12 differentially expressed miRNA biosignatures discriminated IRBD and PD or DLB from controls with an area under the curve of 98% (95% confidence interval: 89-99%). Conclusions: Besides clinical diagnosis of IRBD or imaging markers such as DaT single-photon emission computed tomography, specific miRNA biosignatures alone hold promise as progression biomarkers for patients with IRBD for predicting PD and DLB clinical outcomes. Further miRNA studies in other PD at-risk populations, such as LRRK2 mutation asymptomatic carriers or hyposmic subjects, are warranted. © 2022 The Authors.

Movement Disorders published by Wiley Periodicals LLC on behalf of International Parkinson and Movement Disorder Society
Rapid eye movement (REM) sleep behavior disorder (RBD) is characterized by the lack of REM sleeprelated muscle atonia, disinhibited motor activity, and dream-enacting behaviors. The isolated form of RBD (IRBD) represents the prodromal state of the synucleinopathies because, after 14 years of follow-up, 91% of the patients with IRBD are diagnosed with Parkinson's disease (PD) and dementia with Lewy bodies (DLB), collectively known as Lewy body disorders (LBDs), and rarely with multiple system atrophy (MSA). 1 Yet, predicting disease progression in IRBD is challenging given that time to phenoconversion is variable and the resulting clinicopathological phenotype is heterogeneous, being DLB and MSA rapidly progressive diseases compared with PD. Thus, investigating IRBD can help assess the earliest molecular changes occurring at prodromal stages of α-synucleinopathies and identify potential diagnostic or prognostic biomarkers for these diseases.
MicroRNAs (miRNAs) are small noncoding RNAs that regulate gene expression by mRNA cleavage and translation repression. 2 Cumulative evidence showed serum or plasma miRNA deregulation in patients with PD and related α-synucleinopathies, 3-8 including its prodromal phases such as IRBD. 9,10 Yet, miRNA studies in premotor cohorts are still scarce, and longitudinal follow-ups in seriated IRBD progression stages are needed. In this study, we characterized the serum miRNA expression profiles of a Spanish IRBD cohort without motor and cognitive impairment and segregated by dopamine transporter single-photon emission computed tomography (DaT-SPECT) into DaT-negative IRBD, DaT-positive IRBD, IRBD phenoconverted into PD and DLB, and controls. We used this method to stratify IRBD into successive disease progression stages, given that abnormal DaT-SPECT represents a short-term marker for developing PD and DLB in less than 5 years. [11][12][13] Cross-sectionally, we identified novel differentially expressed miRNAs (DEmiRs) associated with IRBD and α-synucleinopathy, which we subsequently validated using three longitudinal follow-up serum samples from DaT-negative and DaT-positive IRBD. Lastly, by applying machine learning modeling algorithms, we assessed the disease prediction ability of the identified DEmiR to discriminate IRBD and PD or LBD patients from controls alone. The goal of the study was to investigate potential miRNAs as prognostic biomarkers for α-synucleinopathies.

Subjects
All subjects provided written informed consent, and the Ethics Committee of Institut d'Investigacions Biomèdiques August Pi i Sunyer (IDIBAPS)-Hospital Clínic de Barcelona approved the study. The IRBD diagnosis required a history of dream-enacting behaviors, audiovisual-polysomnographic confirmation of excessive electromyographic activity in REM sleep, and absence of motor and cognitive impairment. 1 All the participating IRBD subjects were characterized by dopamine transporter single-photon emission computed tomography (DaT-SPECT) imaging using 123 I-2βcarbomethoxy-3β-(4-iodophenyl)-N-(3-fluoropropyl)nortropane as previously described. 13 Overall, the cross-sectional cohort consisted of 51 patients with IRBD recruited and diagnosed at the Sleep Centre of the Neurology Service at the Hospital Clínic de Barcelona, Spain. Of these, 38 patients with IRBD were free of neurodegenerative disease, 17 of which had normal DaT-SPECT (DaT-negative IRBD) and 21 abnormal DaT-SPECT (DaT-positive IRBD). The remaining 13 patients were initially diagnosed with IRBD, and clinical follow-up showed the phenoconversion into LBD, PD (n = 8), and DLB (n = 5), fulfilling accepted criteria. 14,15 The study included 20 healthy control patients without evidence of neurological or sleep disorders from the Parkinson's Disease and Movement Disorders Unit at the Hospital Clínic de Barcelona (Table 1). Basal RBD and LBD medication is summarized in Table S1. All patients with IRBD were clinically followed up every 3-12 months, and serum samples were collected for at least three time points over 26 months (Table S2). During this period, three DaTpositive IRBD subjects (14%) were diagnosed with PD, one between the first and the second time point and two between the second and the third.

Serum miRNA Isolation
For each subject, we collected 5 mL peripheral blood in tubes with a clot activator (#366468; BD Vacutainer), preserved it for 30 minutes at room temperature, and centrifuged at 1500g for 10 minutes at 4 C. Serum volumes of 2 mL were removed from the supernatant, aliquoted in polypropylene CryoTubes (#121263; Greiner Bio-One), and stored at À80 C. We mixed 200 μL serum from each sample with 2 μL of yeast transfer RNA (tRNA) (#AM7119; Invitrogen) at a final concentration of 10 μg/mL as a carrier for nucleic acid precipitation. Following the manufacturer's instructions, total RNA enriched in miRNAs was isolated using the miRNeasy Serum/Plasma Kit (#217184; QIAGEN). The RNA concentration was determined on a NanoDrop ND-3300 fluorospectrometer (Thermo Scientific). We performed an electropherogram analysis using an Agilent Small RNA Kit in n = 22 random samples as quality control for 6-to 40-bp miRNAenriched fractions and observed high-quality miRNA enrichment for all studied samples (Fig. S1).

Genome-Wide miRNA Profiling
We explored the genome-wide miRNA expression profiles in serum using the Affymetrix GeneChip miRNA 4.0 Array, which interrogates the levels of 4603 human miRNAs, including 2578 mature miRNAs and 2025 pre-miRNAs (#902413; Thermo Fisher) (Product datasheet: https://www.thermofisher.com/). Serum miRNAs from individual subjects were hybridized onto the probe set separate arrays for 42 hours and scanned using an Affymetrix GeneChip Scanner 3000 7G following the manufacturer's instructions. Data files generated by the Affymetrix GeneChip Command Console were processed with the Expression Console software to determine the data quality. Expression raw data were analyzed using the Partek Genomic Suite v7.0 software (Partek Inc.) applying the robust multiarray average background correction model, which permits the relative comparison of miRNA abundance in different arrays. Only miRNAs with detection values greater than 2.4 arbitrary luminescence units in at least 50% of all samples, patients, and controls and a significant expression value above the background signal (P < 0.05) were considered as expressed in serum. In line with previous reports, 16 we observed that approximately 10% of the mature human screened miRNAs were expressed in serum from our cohort. Adjusting by sex and age, we used the global mean of all expressed serum miRNAs as a reference to define candidate differentially expressed miRNAs (DEmiRs) and the sign of the fold-change difference. Under a 2-tailed Student t test, we applied the criteria of a combination of a fold-change > j1.5j and P < 0.05 to select candidate miRNAs from the array for subsequent realtime quantitative polymerase chain reaction (RT-qPCR) validation (Table S3).

Predictive Modeling Through Machine Learning Analysis
To predict each patient's probability of having the disease (ie, IRBD, DLB, or PD) versus being a control, we computed binary classification analyses using the caret framework in R, specifying age, sex, and RT-qPCR expression values of the 12 validated serum DEmiRs as predictors. Multifold imputations of missing data were performed by the predictive mean matching method implemented in the mouse R package. Numerical variables were centered and normalized, resulting in a standard distribution of mean 0 and variance 1. The gradient boosting machine (GBM) algorithm in the gbm package (v.2.1.8) was employed to classify patients in the two groups (disease vs control), using leave-one-out cross-validation to cut for potential overfitting. Variable selection was applied to each training dataset using the least absolute shrinkage and selection operator (Lasso) method in the glmnet package (v.2.0-18) (Table S7) so that only those features with a regression Beta greater than 0 were subsequently downstreamed to the GBM classifier. To address imbalanced class distributions, we performed random upsampling of the minority class on each training data fold so that the two classes acquired the same frequency using the upSample function from the caret. To evaluate the predictive performance of this model classifier, we inferred the accuracy, sensitivity, specificity, and positive and negative predictive values using the crossval package. We assessed discriminative ability by calculating the area under the curve (AUC) on a receiver operating characteristic (ROC) curve using the pROC package.

Results
We first profiled overall miRNA levels in serum samples from patients with IRBD and healthy control patients as an approach to screen global miRNA expression variation and identify specific candidate miRNAs for validation. Compared with control patients, by genome-wide miRNA expression analysis, we found 116 candidate miRNAs in DaT-negative IRBD patients and 50 miRNAs in DaT-positive IRBD patients, which surpassed the fold-change threshold greater than j1.5j and a P value <0.05 (Table S3, Fig. S2). Given the higher sensitivity of RT-qPCR, 24,25 we defined DEmiRs based only on the stringent criteria of RT-qPCR validation with a fold-change > j1.5j and an FDR adjusted P < 0.05. To this end, we selected a total of 10 miRNAs among the topmost deregulated miRNAs identified at the genomewide profiling, including seven miRNAs from DaTpositive IRBD patients (let-7c-5p, miR-140-3p, miR-24-3p, miR-25-3p, hsa-miR-361-5p, miR-425-5p, and miR-451a) and three from DaT-negative IRBD patients (miR-1207-5p, miR-1227-5p, and miR-3613-5p).
To further test the miRNA biosignature discriminative power for unbiased subject classification, we performed an unsupervised hierarchical clustering analysis using R. Consistent with the clinical status, we observed clear segregation into three main subclusters, including controls, patients with IRBD, and PD/DLB phenoconverted patients (Fig. 2B). Driven by these results, we next interrogated specific miRNA differences occurring within the IRBD continuum. We found no miRNA level differences between DaTnegative and DaT-positive IRBD, indicating no main relation between miRNA levels and DaT-SPECT status. However, we observed statistically significant expression differences for all 12 biosignature miRNAs between DaT-positive IRBD and PD/DLB patients, thus suggesting an association between the identified 12 miRNAs and the clinical phenoconversion into α-synucleinopathy (Table S9). These findings indicate that IRBD exhibits differential serum miRNA profiles compared with controls and PD/DLB phenoconverted patients, supporting that the identified serum miRNA biosignature can be informative as a disease progression biomarker for IRBD-initiated α-synucleinopathies.
Lastly, we conducted a biological enrichment analysis to explore the functionality of the genes targeted by the miRNAs differentially expressed at all longitudinal time points. We found substantial deregulation at biological terms previously involved in clinically manifested PD, such as ubiquitin-dependent protein degradation, inositol triphosphate signaling, apoptosis, 26 myelin sheaths, synaptic transmission, postsynaptic assembly, synaptic maturation, and synaptic vesicles, among others. Regarding localization, we found exosomes, microvesicles, and extracellular vesicles among the top terms (Table 3).
To the best of our knowledge, this is the first miRNA study in patients with IRBD characterized by DaT-SPECT with longitudinal follow-up, but two previous studies explored miRNAs in IRBD. A first study using candidate miRNAs reported the deregulation of miR-19b-3p in IRBD serum (n = 104) up to 5 years before α-synucleinopathy diagnosis. 9 Another deep sequencing study using whole blood from a mixed cohort of IRBD and hyposmic subjects (n = 223) reported 500 miRNAs. 10 Four of the 12 biosignature miRNAs here were topmost deregulated in the earlier reports (miR-19b-3p, miR-29c-3p, miR-221-5p, and miR-140-5p). 9,10 Complementing these studies, we found a miRNA biosignature informative for all IRBD progression stages, including IRBD and PD/DLB phenoconverted patients. Our findings align with studies showing early molecular deregulation in prodromal stages of α-synucleinopathies. Thus, recent works in IRBD observed misfolded α-synuclein in the cerebrospinal fluid (CSF) before developing motor and cognitive impairment. 28 Another study in IRBD showed the presence of phosphorylated α-synuclein aggregates in the minor labial salivary glands. 29 Akin to these studies, our findings support that early molecular changes associated with α-synucleinopathies occur at early IRBD prodromal stages and can antedate the clinical manifestations of PD and DLB.
In PD and DLB, developing disease prediction strategies ahead of the motor and cognitive manifestations is a challenge for the early neuroprotective intervention. 30,31 By machine learning, we evaluated the disease prediction capacity of the combined expression levels from the 12 miRNA biosignatures adjusted by age and sex to estimate each proband's probability of disease, IRBD and PD/DLB versus being control and without any further clinical or imaging input. The miRNA biosignature achieved a high discriminative capacity with an AUC of 98% (95% CI: 89-99%) and distinguished prodromal and manifest PD or DLB from controls more accurately than other previously proposed candidate biomarkers, including α-synuclein. 8,[32][33][34][35] The predictive model also accurately classified both positive and negative individuals (94% accuracy), specifically identifying true positives (98% sensitivity). However, while the fraction of false negatives was only 2%, the false-positive rate was 14% (86% specificity). Therefore, it is important to emphasize the need to develop more specific strategies to capture true negatives. In this sense, regarding the most clinically relevant miRNAs detected by the Lasso algorithm using a feature importance selection, only four of the six miRNAs selected in more than 90% of training folds displayed AUC scores greater than 75% at the individual level (miR-1227-5p, miR-425-5p, miR-451a, and miR-221-3p). 32,33 Altogether, these findings support the proof-of-principle that, beyond clinical or imaging markers such as DaT-SPECT, specific serum miRNA deregulation alone can be informative to predict the presence of IRBD and PD or DLB.
Functionally, the target genes from the 12 longitudinal DEmiRs are involved in ubiquitin-dependent protein degradation, inositol triphosphate signaling, apoptosis, 26 and neural terms such as myelin sheaths, synaptic transmission, postsynaptic assembly, synaptic maturation, or synaptic vesicle. Although the function and origin of serum miRNAs are not clear, 50 the terms found collectively support a plausible role in the CNS. Conceptually, brainderived exosomes crossing the brain-blood barrier represent a potential biomarker source for neurodegenerative diseases, 51 which is in line with our finding of exosome as the top localization term. In this context, the biological enrichment analysis in our study pinpointed a plausible brain or CSF origin of at least some of the biosignature miRNAs, thus supporting the concept that minimally invasive serum miRNAs could mirror pathophysiological processes occurring at the CNS in α-synucleinopathies. Future neuronal exosome studies in PD serum should address this question, especially if interrogating PD prodromal stages. Illustratively, a recent study reported increased α-synuclein at neuronal exosomes from IRBD serum preceding the diagnosis of PD, persisting during disease progression, and predicting PD clinical outcomes versus atypical parkinsonisms. 52 Despite the promising identification of a novel miRNA biosignature that is informative of different disease progression stages in α-synucleinopathies, our study has limitations. First, we performed a restrictive miRNA expression analysis filtering in only miRNAs validated by RT-qPCR, but additional candidates could also be nominated for validation. Second, our miRNA analyses have not been corrected for IRBD medication, such as melatonin, which has been shown to interact with specific miRNAs in mice. 53,54 Third, ours is the first serum miRNA longitudinal study in IRBD characterized by DaT-SPECT, but the follow-up (26 months) was limited. Fourth, the power to discriminate between subtypes of α-synucleinopathy in our predictive modeling requires larger LBD sample sizes. Indeed, given the retrospective character of this study, further validation analyses in independent, larger cohorts of patients are warranted to ensure the predictive ability of this serum miRNA biosignature. Lastly, it has to be mentioned that the reproducibility of PD miRNA to date has been limited because of ancestry differences, lack of gold standard normalization methods, and reduced cohort sizes. In the absence of gold standard normalizers across populations, 18 we applied both endogenous and exogenous controls 17,55 as an improved strategy that can increase cross-laboratory miRNA reproducibility.
In summary, we identified a serum miRNA biosignature that is informative in discriminating healthy controls, patients with IRBD, and patients with PD/DLB and holds potential as a disease progression biomarker. If validated, our findings may have implications for disease prediction strategies and early detection of α-synucleinopathies such as PD and DLB. Further studies in other prodromal PD/DLB or PD/DLB at-risk cohorts are warranted.