Polygenic Risk Scores Validated in Patient‐Derived Cells Stratify for Mitochondrial Subtypes of Parkinson's Disease

The aim of our study is to better understand the genetic architecture and pathological mechanisms underlying neurodegeneration in idiopathic Parkinson's disease (iPD). We hypothesized that a fraction of iPD patients may harbor a combination of common variants in nuclear‐encoded mitochondrial genes ultimately resulting in neurodegeneration.

p-value = 5.4e-04; COURAGE-PD OR = 1.23 [1.18-1.27],p-value = 1.5e-29).Functional analyses in fibroblasts and induced pluripotent stem cells-derived neuronal progenitors revealed significant differences in mitochondrial respiration between iPD patients with high or low OXPHOS-PRS (p-values < 0.05).Clinically, iPD patients with high OXPHOS-PRS have a significantly earlier age at disease onset compared to low-risk patients (false discovery rate [FDR]-adj pvalue = 0.015), similar to prototypic monogenic forms of PD.Finally, iPD patients with high OXPHOS-PRS responded more effectively to treatment with mitochondrially active ursodeoxycholic acid.Interpretation: OXPHOS-PRS may provide a precision medicine tool to stratify iPD patients into a pathogenic subgroup genetically defined by specific mitochondrial impairment, making these individuals eligible for future intelligent clinical trial designs.
ANN NEUROL 2024;96:133-149 P arkinson's disease (PD) is the fastest growing and as yet incurable neurodegenerative disease, affecting about 1-2% of the population over the age of 60. 1 PD patients typically display bradykinesia with rigidity and/or resting tremor accompanied by a heterogeneous panel of non-motor symptoms (e.g., sleep disturbance, mood disorders, cognitive changes, and autonomic dysfunction).Motor symptoms are mostly caused by disrupted dopamine signaling in the striatum due to loss of dopaminergic neurons (DANs) in the substantia nigra pars compacta of the brain.Typical proteinaceous inclusions (i.e., Lewy bodies), composed of α-synuclein, ubiquitin, and other aggregated proteins or organelles, are observed in the surviving DANs. 2 The etiology of PD is complex and influenced by both environmental and genetic factors.Monogenic familial forms, ascribable to single mutations in autosomal dominant (e.g., SNCA, LRRK2, VPS35) or recessive (PRKN, PINK1, PARK7) genes, account for approximately 5-10% of all PD cases. 3However, the contribution of genetics in the remaining 90-95% of patients with idiopathic PD (iPD) remains poorly understood, and disease susceptibility may be influenced by the synergistic effect of multiple common low-risk genetic variants. 4To unravel this "missing heritability", systematic approaches based on polygenic risk scores (PRSs) have been proposed in recent years, aimed at identifying individuals with a higher risk to develop PD. 5 PRSs are calculated as the sum of common risk single nucleotide variants (SNVs) weighted by their genome-wide associated studies (GWAS) effect sizes. 6Applying PRS to a limited number of genes that regulate biological functions known to be altered in PD could potentially reveal a relationship between the pathways involved and their impact in determining the clinical phenotype.Such an approach would allow the stratification of patients according to the underlying pathological mechanisms, thus enabling precision medicine therapeutic approaches. 5here is convincing and longstanding evidence that points to mitochondrial dysfunction as an early and causative event in PD pathogenesis.Indeed, both epidemiological studies on humans exposed to pesticides as well as toxin-induced PD models support a primary role for impaired mitochondrial electron transport chain (ETC) activity, as suggested by the selective degeneration of DANs following mitochondrial complex I disruption. 7,8Moreover, molecular genetic studies in PD families revealed that proteins mutated in early-onset forms (i.e., PINK1 and Parkin) regulate mitochondrial function or even localize to mitochondria. 91][12] Neurodegeneration in such iPD patients cannot be explained by ageing or environmental factors alone, 13 which possibly implies the existence of pathogenic variants in mitochondrial genes.In accordance with this possibility, a recent study on mitochondria-specific PRSs (mitoPRSs) demonstrated that combinations of small effect variants within genes regulating mitochondrial function were associated with higher PD risk. 14erein, we extended the mitoPRS concept to subsets of genes controlling distinct mitochondrial pathways and demonstrated, in independent case-control datasets from 2 large, deeply phenotyped PD cohorts, that common variants in genes regulating OXPHOS were significantly associated with an increased risk of developing PD.Importantly, we functionally validated individual OXPHOS-PRS profiles in the corresponding patient-based cellular models, showing significant differences in mitochondrial respiratory function between iPD patients with high or low OXPHOS-PRS.iPD patients with high OXPHOS-PRS displayed a significantly earlier age at disease onset compared to those with low OXPHOS-PRS.Finally, patients with high OXPHOS-PRS responded more effectively to treatment with the mitochondrially active agent ursodeoxycholic acid (UDCA), suggesting that the OXPHOS-PRS may indeed allow for the selection of more homogenous PD patient groups for precision medicine in mitochondriacentred clinical trials.

Study Design
Three studies were analyzed for this work.First, the exploratory dataset from the ongoing Luxembourg Parkinson's Study, a large longitudinal monocentric observational study in the framework of the NCER-PD (National Centre for Excellence in Research in Parkinson's Disease) program comprising, at the time of data export, 493 PD patients and 625 healthy controls (HCs). 15econd, the replication dataset from the COURAGE-PD (Comprehensive Unbiased Risk Factor Assessment for Genetics and Environment in Parkinson's Disease) consortium, including 21 sub-cohorts of European descent, but excluding samples from the Luxembourg Parkinson's Study and the International Parkinson's Disease Genomics Consortium (IPDGC) (7,422 PD patients and 6,904 HCs). 16,17Third, an independent UK-based cohort (23 HCs and 55 iPD) pooling participants from 2 published studies: an observational study of multimodal mechanistic stratification of PD according to bioenergetic dysfunction and a clinical trial of UDCA in PD (i.e., the UP study, trial registration EudraCT no.2018-001887-46). 18,19

Clinical Assessment
All participants in the Luxembourg Parkinson's Study underwent a comprehensive clinical assessment, as described elsewhere. 15D diagnosis was based on UKPDSBB criteria.20 Clinical data reported here include the Movement Disorder Society update of the Unified Parkinson's Disease Rating Scale I (MDS-UPDRS I), MDS-UPDRS II, MDS-UPDRS III, MDS-UPDRS-IV, the Quality of Life questionnaire (PDQ39), the L-dopa-equivalent daily dose (LEDD), the Scales for Outcomes in Parkinson's Disease-Autonomic Dysfunction (SCOPA-AUT) and the Montreal Cognitive Assessment (MoCA).Age at onset (AAO) was defined as the age at PD diagnosis. Diease duration was defined as the number of years since the official PD diagnosis up to the time of data collection, i.e., up to the baseline visits for all individuals analyzed.All participants in the UP Study supplied an ethylenediaminetetraacetic acid (EDTA) blood sample for genetic analysis.The UP Study methodology and results have been described elsewhere.18 As UDCA is a proposed mitochondrial rescue agent, phosphorus-31 magnetic resonance spectroscopy (31P-MRS) was used to assess target engagement within the midbrain.This allows the measurement of Gibbs free energy of ATP hydrolysis (ΔG ATP ), which reflects the useful energy released from the hydrolysis of ATP, therefore more negative values-as observed after 48 weeks treatment of UDCA compared to baseline-are possibly indicative of improved OXPHOS activity.18 Genotyping and Quality Control DNA samples from participants in all cohorts were genotyped using the Neurochip array (v.1.0and v1.1; Illumina, San Diego, CA) that was specifically designed to integrate neurodegenerative disease-related variants.21 Quality control (QC) of genotyping data was performed using PLINK v1.9, 22 as follows: samples with call rates <95% and whose genetically determined sex deviated from gender reported in the clinical data were excluded from the analysis, and the filtered variants were checked for cryptic relatedness and excess of heterozygosity.Samples exhibiting excess heterozygosity (F statistic > 0.2) and first-degree relatedness were excluded.After sample QC, SNVs with Hardy-Weinberg equilibrium p value < 1E-6, and missingness rates >5% were also excluded.QC was performed independently for each European cohort of the COURAGE-PD study, according to the standard procedures reported previously.17 To focus our analysis on iPD cases, for both cohorts we also excluded carriers of pathogenic PD-associated variants in 8 PD-related genes (ATP13A2, LRRK2, GBA, PARK7, PINK1, PRKN, VPS35, and SNCA) identified via genotyping data.23,24 For the Luxembourg Parkinson's Study, the presence of these mutations was confirmed by Sanger sequencing.To account for the population stratification, we calculated the first 3 principal components (PCs) using PLINK.The genotyping data were then imputed using the Haplotype Reference Consortium r1.1 2016 of the Michigan Imputation Server and filtered for imputation quality (R2 > 0.3).25

Mitochondrial Gene Sets and Pathway Resources
To assess the potential association between common variants in nuclear-encoded mitochondrial genes and PD risk, we first selected 3 different mitochondrial gene sets: Human MitoCarta3.0, which is a public inventory of 1,136 genes encoding proteins with putative mitochondrial localization, 26 and 2 gene sets previously reported by Billingsley and colleagues; the primary list (Billingsley I) contained 178 genes implicated in mitochondrial disorders, whereas the secondary list (Billingsley II) contained 1,327 genes regulating, more generally, mitochondrial function. 14oreover, we defined 6 groups of genes-obtained from the Molecular Signatures Database (MsigDB) v7.5.1-known to participate in the following mitochondrial pathways potentially related to PD pathogenesis: Mitochondrial DNA regulation, Mitophagy, Oxidative phosphorylation, TCA cycle, Mitochondrial protein import, and Mitochondrial proton transport (Table S1).

Calculation of PRSs
Genome-wide PD-PRSs were calculated using the PRSice2 R package with default settings. 27PRSs for each individual were generated by summing the weighted effects of the risk alleles (including all variants below the threshold for genome-wide significance) associated with PD-based on the largest PD GWAS summary statistics to date 4 -which are present in the 2 imputed target datasets (Luxembourg Parkinson's study and COURAGE-PD).PRSs for the general or pathway-specific mitochondrial gene sets (mitoPRSs) were generated using the PRSet function in PRSice2, using only risk alleles within gene regions outlined in the different gene lists.PRSice2 implement the clumping and thresholding (C + T) method.The criteria for linkage disequilibrium (LD) clumping of SNVs were pairwise LD r2 < 0.1 within a 250 kb window, ensuring that only independent genetic variants contributed to the PRS.PRSs were computed at different p-value thresholds ranging from 5e-08 to 0.5.PRSice2 identified the optimal p-value threshold for variant selection that explained the maximum variance in the target sample.PRSice2 was also used to determine the observed phenotypic variance (PRS model fit, R2) explained by the genetic contribution of each mitochondrial pathways-PRS.
[30] Fibroblast Selection and Culture Skin biopsies from the lower back region of study participants were collected in low-glucose Dulbecco's Modified Eagle Medium (DMEM, Thermo Fisher Scientific, #21885108) supplemented with 1% (v/v) penicillin-streptomycin (P/S, Thermo Fisher Scientific, #15140163).Each skin biopsy (5 mm diameter punch) was cut into multiple (up to 5) pieces and placed with the dermis facing down into a cell culture flask containing heat-inactivated Fetal Bovine Serum (HI-FBS, Thermo Fisher Scientific, #10270-106).After 10 minutes incubation at room temperature, DMEM containing 4.5 g/L glucose and 4 mM L-glutamine (Thermo Fisher Scientific, #41965039), supplemented with 10% (v/v) HI-FBS and 1% (v/v) P/S, was added to the flask, followed by incubation at 37 C in a 5% CO 2 humidified atmosphere.Cells were kept in the same medium for subsequent subculture, and subjected to Mycoplasma analysis on a regular basis to exclude potential contamination.Fibroblasts lines all had the same passage number at the time of the experiments, ranging from passage 5 to 10.
The smNPCs from 2 PD patients carrying the c.1366C > T mutation in the PINK1 gene (p.Q456X) were described previously. 32

Analysis of Mitochondrial Respiration
Oxygen consumption rates (OCRs) were determined by using the Seahorse XFe96 FluxPak (Agilent, #102416-100) and XF Cell Mito Stress Test (Agilent, #103015-100) kits, in accordance with the manufacturer's instructions.For primary skin fibroblasts, 5,000 cells per well were seeded in a 96-well Seahorse cell culture plate (at least 5 technical replicates per line) and incubated overnight.The following day, fibroblast medium in each well was replaced with 175 μl Seahorse XF DMEM medium pH 7.4 (Agilent, #103575-100), supplemented with 25 mM glucose, 2 mM L-glutamine and 1 mM sodium pyruvate, followed by 1 h incubation at 37 C without CO 2 .In the meantime, the sensor cartridge-previously equilibrated overnight in the XF Calibrant solution at 37 C without CO 2 -was loaded with standard mitochondrial toxins, namely Oligomycin (1 μM final concentration in the assay well), carbonyl cyanide p-trifluoromethoxyphenylhydrazone (FCCP, 0.6 μM final concentration in the assay well) and a mixture of Rotenone and Antimycin A (both at the concentration of 0.5 μM in the assay well).After an additional incubation at 37 C without CO 2 for 30 minutes, the sensor cartridge, as well as the cell culture plate, were loaded on the XFe96 Extracellular flux Analyzer (Agilent).Three OCR measurements were performed for both basal respiration and after each automated injection of mitochondrial toxins.OCR data were analyzed using the XFe Wave software (Agilent) and exported using the Seahorse XF Cell Mito Stress Test Report Generator.Initial OCR values as well as OCR measurements after each injection step were used to calculate different parameters related to oxidative phosphorylation, including basal respiration, proton leakage across the inner mitochondrial membrane, respiration coupled to ATP synthesis, maximal respiration, and spare respiratory capacity. 33eahorse experiments in neuronal progenitor cells were performed in Neurobasal medium containing either glucose or galactose as carbon source. 34Briefly, 35,000 cells per well were first seeded in a Geltrex-coated 96-well Seahorse cell culture plate (at least 10 replicates per line), in 100 μl N2B27 medium containing CHIR, PMA and AA.The day after, N2B27 medium was removed and replaced by Neurobasal-A without glucose and sodium pyruvate (Thermo Fisher Scientific, #A2477501), then supplemented with 0.727 mM sodium pyruvate and either 25 mM glucose or 25 mM galactose.Similar to the standard N2B27 used for maintenance of neuronal progenitor cells, this medium also contained N2, B27 and glutamax supplements, 1% (v/v) P/S, as well as CHIR, PMA, and AA.Neuronal progenitor cells remained in glucose-or galactose-based media for 48 hours before proceeding with the Seahorse protocol described above.Either glucose or galactose (always 25 mM) were also added to the Seahorse assay medium (XF DMEM pH 7.4) used during the entire procedure.
OCR data were normalized against protein (fibroblasts) or DNA (smNPCs) content, as defined by the PIERCE™ BCA protein assay kit (Thermo Fisher Scientific, #23225) or the CyQUANT cell proliferation assay kit (Thermo Fisher Scientific, #C7026), respectively.

Statistical Analysis
The predictive accuracy of the PRS model was determined using the area under the receiver operating curve (AUC, pROC R package).A higher AUC indicates a better PRS model able to discriminate between PD cases and controls.We compared the PRS distribution in HC vs iPD subjects using the nonparametric Wilcoxon rank-sum test, as the data did not display a normal distribution.To further investigate whether the PRSs could predict PD risk, a logistic regression model was used to calculate the odds ratio (OR).OR is always given with confidence interval.Gender, age at assessment, and the first 3 PCs from the population stratification were included as covariates.
Individuals were stratified into 3 groups based on OXPHOS-PRS percentiles: low (<10%), intermediate (10-90%), and high (>90%).Comparison of PD-specific clinical outcomes, AAO, and disease duration between individuals in the high and low OXPHOS-PRS groups was performed using the Welch's t-test.Statistical analyses were done in R (v4.0.4).To account for multiple hypothesis testing and reduce the risk of type I errors, we applied the false discovery rate (FDR) correction using the Benjamini-Hochberg method, with an adjusted p-value threshold of <0.05 deemed to be significant.For the functional assays (all involving more than 2 groups), statistical significance was assessed using 1-way ANOVA, after confirming normality of the data using a Shapiro-Wilk test.For experiments with 2 independent variables, 2-way ANOVA was used considering interaction effects.In both cases, Tukey's post hoc correction for multiple comparisons was applied after the ANOVA tests.Statistical analyses of functional data were performed using the GraphPad Prism software (v9.4.0).Data with error bars are represented as mean AE SEM and are representative of at least 3 independent biological replicates.Two-way ANOVA combining treatment group and PRS risk groups with an interaction effect was used to assess for differential treatment responses between these groups in clinical parameters from The UP Study.Any significant ANOVA underwent post-hoc analysis using Tukey's test.

Association of Common Variants in Mitochondrial Genes with PD Risk
To gain essential knowledge about the mitochondriarelated "missing heritability" in iPD, we analyzed the distribution of mitoPRSs in 2 distinct case-control datasets, namely the Luxembourg Parkinson's Study 15 and COURAGE-PD 16,17 cohorts.After filtering and QC, the final dataset for the Luxembourg Parkinson's Study comprised 412 iPD patients and 576 HC. iPD patients were older than the HC (67.5 AE 10.9 vs 59.1 AE 12.2 years, p-value < 0.01) with a mean AAO of 62.2 AE 11.8 years and a mean disease duration of 5.4 AE 5.0 years (Table ).Next, we applied the same approach to different mitochondrial pathways, aimed at obtaining PRSs related to specific mitochondrial alterations possibly involved in PD pathogenesis.We, therefore, selected 6 potentially relevant mitochondrial pathways, containing groups of genes annotated in the corresponding Molecular Signatures Database (MsigDB) sub-collections (Table S1), and calculated mitochondria-specific PRSs  .Of note, using common risk alleles associated with other diseases (i.e., type 2 diabetes, schizophrenia, and AD GWAS summary statistics) [28][29][30] for the calculation of mitoPRSs in the imputed Luxembourg Parkinson's Study cohort, we found that none of them were significantly associated with an increased PD risk (Fig S3), thus confirming the specificity of our findings based on common PD GWAS SNVs.Next, we used the 21 sub-cohorts of the COURAGE-PD consortium (Table S2) 16,17  We also assessed the phenotypic variance and the predictive accuracy of each mitoPRS model, using the R2 metric and AUC analysis, respectively.Strikingly, among the tested mitochondrial pathways, only the OXPHOS-PRS model significantly predicted PD status in both, the Luxembourg Parkinson's study (R2 = 0.015, FDR-adj p-value = 0.003; AUC = 0.56) and COURAGE-PD (R2 = 0.010, FDR-adj p-value = 3.4e-33; AUC = 0.56) cohorts (Fig S4 and Table S3).Finally, to correct for imbalance in the case-control ratio, we have also measured the phenotypic variance after adjusting for an estimated PD prevalence of 0.005. 36The resulting adjusted R2 (R2*) for the OXPHOS-PRS model was still the highest score among the tested pathways in both, the Luxembourg Parkinson's study (R2* = 0.0058) and COURAGE-PD (R2* = 0.00366) cohorts (Table S3).
The To determine whether the OXPHOS-PRS could represent a pathophysiology relevant tool to stratify iPD patients based on mitochondrial dysfunction, we sought to functionally validate the predicted phenotypes in patient-derived cellular models.Participants were grouped into low-, intermediate-, or high-risk groups according to the PRS distribution in the lower and higher deciles. 37,38ecile cutoffs of 10% and 90%, corresponding to the lowest and highest PD ORs, were chosen for the subsequent functional studies (Fig S5).Thus, iPD patients with the highest or lowest OXPHOS-PRSs (10th and 90th percentile, respectively) were identified among the Luxembourg Parkinson's Study participants and available primary skin fibroblasts from these individuals (at least n = 4 from each group) were used in functional experiments.In parallel, fibroblasts from HCs with extreme OXPHOS-PRS values (10th and 90th percentile, at least n = 4 for each group) were selected as reference for the comprehensive mitochondrial phenotyping performed in cells from iPD patients (Fig 1E and Table S4).

Mitochondrial Oxygen Consumption Is Significantly Elevated in Primary Skin Fibroblasts from iPD Patients with High OXPHOS-PRS
To functionally validate the association between OXPHOS-PRS and PD risk in patient-derived cells, we first assessed mitochondrial respiratory chain performance using the Seahorse technology, which provides a comprehensive overview of mitochondrial bioenergetics by measuring oxygen consumption rates (OCRs) under basal conditions and after targeted inhibition of specific respiratory chain complexes. 33trikingly, basal and ATP-linked respiration were significantly enhanced in fibroblasts from iPD patients with high OXPHOS-PRS compared to low OXPHOS-PRS cells (Fig 2).In this setting, maximal respiration was not significantly different between high and low OXPHOS-PRS iPD fibroblasts, indicating a reduced spare respiratory capacity in the high OXPHOS-PRS iPD group.Of note, all respiratory parameters analyzed were not significantly different between HCs with high or low OXPHOS-PRS (Fig 2).Similarly, biogenesis of respiratory chain complexes (RCC) did not differ significantly between the high and low OXPHOS-PRS groups-both in HCs and in iPD patients-as revealed by immunoblot analysis of RCC protein subunits (Fig S6 ).Taken together, these results suggest that fibroblasts from iPD patients with high OXPHOS-PRS display features of mitochondrial hyperactivity even under steady-state conditions, a phenotype that appears to be independent of RCC protein expression.

Other Functional Readouts of Mitochondrial Activity Are Not Significantly Altered in Fibroblasts from iPD Patients Stratified Based on OXPHOS-PRS
To assess whether changes in mitochondrial respiration were accompanied by differences in reactive oxygen species (ROS) production, fibroblasts from both iPD patients and HCs stratified according to their individual OXPHOS-PRS were subjected to CellROX staining.Surprisingly, increased ROS levels were detected in HCs from the high OXPHOS-PRS group compared to both low-risk HC and iPD patients, arguing against a causal link between elevated OXPHOS and ROS accumulation in these cells (Fig 3A).Given the important role of the electrochemical gradient across the inner mitochondrial membrane in generating the proton motive force used by the OXPHOS machinery to synthesize ATP, 39 we also measured mitochondrial membrane potential (ΔΨ m ) in primary skin fibroblasts from each OXPHOS-PRS group.To this end, cell lines were stained with tetramethylrhodamine ethyl ester (TMRE), a fluorescent dye that specifically accumulates in active, polarized mitochondria, followed by high-throughput automated confocal microscopy analysis.Treatment with the OXPHOS uncoupler carbonyl cyanide 3-chlorophenylhydrazone (CCCP), known to induce mitochondrial depolarization, was used as a positive control for decreased ΔΨ m .As expected, CCCP-treated cells all failed to accumulate the dye within mitochondria and, therefore, displayed a robust reduction of TMRE fluorescence.However, no significant differences in ΔΨ m were observed between untreated fibroblasts from both HCs and iPD patients previously stratified according to their low or high OXPHOS-PRS (Fig 3B).OXPHOS can also be affected by abnormal mitochondrial fission/fusion processes, as revealed by changes in ETC activity and respiratory function in response to dynamic transitions of organelle morphology-from fragmented to elongated, and vice versa. 40,41Again, we observed no difference in morphological parameters such as mitochondrial aspect ratio and form factor in fibroblasts derived from HCs and iPD patients with high or low OXPHOS-PRS (Fig 3C).mtDNA copy number, transcription/replication rates and deletions, which could impinge on ETC activity by altering the expression and stoichiometric assembly of mtDNA-encoded RCC subunits, 12,42,43 were also not significantly different between the high and low OXPHOS-PRS groups (Fig S7).Finally, based on the emerging link between mitochondrial dysfunction and inflammation in PD, 44,45 we measured interleukin-6 (IL-6) levels in blood plasma samples from the Luxembourg Parkinson's Study participants (both HCs and iPD patients) with the highest or lowest OXPHOS-PRS.Indeed, we recently found increased IL-6 levels in the serum of PD patients carrying biallelic rare variants in PINK1 or PRKN, 44 a phenotype mainly associated with impaired mitophagy and dependent on activation of the pro-inflammatory cGAS-STING pathway. 45Here, we did not observe significant differences in plasma IL-6 levels between the high and low OXPHOS-PRS groups (Fig 3D ).

Functional Validation of OXPHOS-PRS in iPSC-Derived Neuronal Progenitors
To validate our OXPHOS-PRS approach in more disease-relevant patient-derived models, we reprogrammed fibroblast lines from both HCs and iPD patients with high or low OXPHOS-PRS into induced pluripotent stem cells (iPSCs).We successfully reprogrammed 7 HC (3 low-risk and 4 high-risk) and 11 iPD (5 low-risk and 6 high-risk) lines.The newly generated iPSCs all showed no chromosomal aberrations and an identical genetic background compared to the original skin fibroblasts (Fig S8A ,B).Different from Next, iPSC lines were differentiated into smNPCs, 31 and then subjected to an assessment of mitochondrial respiration.Seahorse-based bioenergetics analyses in smNPCs did not reveal major differences in OCRs between the low and high OXPHOS-PRS groups (Fig 4 ), likely due to their high glycolytic activity and low reliance on OXPHOS metabolism. 46In accordance with this hypothesis, shifting smNPCs from glucose to galactose medium as carbon source dramatically increased basal and ATP-linked respiration, indicative of a metabolic switch from glycolysis to OXPHOS (Fig 4).In this setting, smNPCs derived from iPD patients with high OXPHOS-PRSs displayed a significant decrease of basal respiration, proton leak, and ATP-linked respiration compared to the low OXPHOS-PRSs group, comparable to that observed in smNPCs from patients with autosomal recessive PD caused by rare variants in the PINK1 gene.Notably, OCR values did not significantly vary between HCs with high or low OXPHOS-PRS, both in glucose and in galactose media (Fig 4).

Association of OXPHOS-PRSs with PD-Specific Clinical Outcomes
We sought to investigate whether our functionally validated OXPHOS-PRS approach could be used to define distinct phenotypic patterns among iPD patients.To this end, we first analyzed 8 PD-specific motor and non-motor clinical scores in iPD patients from the Luxembourg Parkinson's Study stratified according to their OXPHOS-PRS, namely the Movement Disorder Society update of the Unified Parkinson's Disease Rating Scale I (MDS-UPDRS I), MDS-UPDRS II, MDS-UPDRS III, MDS-UPDRS-IV, the Quality of Life questionnaire (PDQ39), the L-dopa-equivalent daily dose (LEDD), the Scales for Outcomes in Parkinson's Disease-Autonomic Dysfunction (SCOPA-AUT), and the Montreal Cognitive Assessment (MoCA).The whole-genome PRS was used as reference.In general, the trend of all clinical outcomes analyzed was similar to that observed for the wholegenome PRS, and none of them was significantly different between iPD patients with high or low OXPHOS-PRS (Fig S9).However, iPD patients with high OXPHOS-PRS had an earlier AAO compared to low-risk patients, a phenotype that was particularly evident in the larger Finally, as an exploratory analysis we wanted to investigate if our approach using OXPHOS-PRS for patient stratification could allow the identification of PD patients that are more responsive to drugs targeting mitochondria.To this end, participants of the "UP Study" with genetic data available were retrospectively stratified according to the same OXPHOS-PRS strategy, applied to the entire UK cohort separately.Risk classification was based on quartiles; within each group (iPD or HCs), individuals with PRS values within the highest quartile were assigned to 'high' risk, those in the lowest quartile to 'low' and the remainder being assigned 'intermediate'.
Although less stringent, quartile-based classification was necessary to ensure sufficient numbers of participants within each risk group to enable this exploratory analysis.As previously shown, 18 the UDCA treatment group had an overall improvement in the Gibbs free energy of ATP hydrolysis compared to the placebo group (ΔG ATP = À2.82,95% CI À4.78, À0.86, p-value = 0.0074).There was a significant interaction between UDCA treatment and the OXPHOS-PRS risk group (p-value = 0.04824).This treatment effect was significantly greater within the high OXPHOS-PRS groups taking UDCA compared to the high OXPHOS-PRS group on placebo (ΔG ATP = À6.47,95% CI À12.02, À0.93, p-value = 0.0172, Fig 5C).Of note, no significant differences were observed upon patients' stratification according to the general MitoCarta-PRS apart from the expected global treatment effect (Fig 5D).

Discussion
Despite recent advances in the generation and analysis of large-scale genotyping data, unravelling the complex genetic architecture of PD remains still a major challenge.Low-penetrance genetic variants that significantly increase the risk of developing PD have been identified in sporadic cases (e.g., GBA L444P and LRRK2 G2019S), but typical hallmarks of PD heritability can also be observed in families with unknown genetic causes, 47 which suggests a potential involvement of genetic factors in up to 30% of identified PD cases. 48issing heritability in PD and the clear association between mitochondrial dyshomeostasis and neurodegeneration led us to investigate the role of diseaseassociated SNVs in genes regulating mitochondrial function.Thus, we performed association analyses targeting common variants in nuclear-encoded mitochondrial genes and analyzed the mitoPRS distribution in 2 distinct iPD case-control studies, namely the Luxembourg Parkinson's Study and the COURAGE-PD consortium.After testing several common mitochondrial gene-sets, we found that mitoPRSs were always significantly associated with increased PD risk in both cohorts.Of note, PRSs can also be applied to subsets of genes that control distinct, disease-related cellular activities (e.g., mitochondrial quality control or lysosomal function in PD), aimed at identifying specific pathways implicated in the pathological phenotype. 36Using this approach, Paliwal and colleagues recently demonstrated that certain mitochondrial pathways-including OXPHOS among others-were significantly associated with increased AD risk. 49In a similar manner, here, we used gene lists for 6 mitochondrial pathways relevant to PD and calculated pathway-specific mitoPRSs.Strikingly, only common variants in genes regulating OXPHOS were significantly associated with increased PD risk in the Luxembourg Parkinson's Study and COURAGE-PD cohorts, reinforcing the notion of impaired mitochondrial respiration as a relevant aspect in PD pathogenesis.
][52] However, to date, only 3 studies used PRS-based approaches to assess mitochondrial risk in PD. 14,36,53 Two of them demonstrated that a combination of small effectsize common variants in nuclear-encoded mitochondrial genes was significantly associated with higher PD risk, 14,54 while the other one failed to confirm this association but showed functional enrichment of mitochondria-related pathways-including OXPHOS. 36One study also demonstrated an association between mitoPRSs and clinical outcomes. 53However, none of these studies considered the need to functionally validate PRSs in patient-based cellular models, which would legitimize the effectiveness of genetically predicted cellular phenotypes thus paving the way for precision medicine therapeutic approaches. 54n addition, there have been attempts to stratify iPD patients exclusively based on mitochondrial phenotyping assays. 55However, this type of stratification is timeconsuming, labor-intensive, and error-prone; in fact, since most functional assays rely on relative measures, the definition of cutoffs between "mitochondrial" and "non-mitochondrial" PD subtypes is extremely challenging.In this context, the translation of a purely mitochondrial phenotyping approach into clinical trials for the stratification of large numbers of patients seems impractical.
Our study follows a different idea: the stratification of patients is entirely based on genetics and therefore defined by 'absolute' measures.This makes our approach robust, cost-effective and high-throughput compatible, as genetics can be easily assessed and provides 'definitive' data.Here, for the first time in the PD field, we functionally validated individual OXPHOS-PRS profiles in primary skin fibroblasts and the corresponding iPSC-derived neuronal progenitor cells established from genetically stratified iPD patients, showing significant differences in mitochondrial respiration between high-and low-risk groups (Fig 6).In primary skin fibroblasts, OCRs were significantly elevated in the high OXPHOS-PRS groupespecially basal and ATP-linked respiration.It has been postulated that a similar phenotype, indicative of steadystate mitochondrial hyperactivity, may be an early event in PD pathogenesis, as suggested by increased neuronal viability and improved motor function in worms treated with the mitochondrial complex I inhibitor metformin. 56,57ased on this model, hyperactive mitochondria may not only reduce the ability of the OXPHOS machinery to cope with a sudden increase in energy demand (e.g., in response to acute cellular stress), but also cause an excessive accumulation of ROS, which in turn increases oxidative damage and ultimately induces neurodegeneration.In this scenario, reduced ETC complex activity and impaired mitochondrial respiration-commonly observed in peripheral tissues and post-mortem brains of PD patients-would be late endpoint phenotypes, caused by ROS-mediated damage of mitochondrial complex I. 58 However, in the current study, increased mitochondrial respiration in fibroblasts with high OXPHOS-PRS was not accompanied by enhanced ROS production, likely due to the activation of compensatory mechanisms resulting in increased scavenging capacity. 59High OCRs may also reflect the specific metabolism of the cells or tissues studied, including their ability to cope with mitochondrial dysfunction and to ensure ATP production.Similar to our findings, mitochondrial respiration was found to be significantly higher in Parkin-mutant fibroblasts compared to controls, 60,61 and mitochondrial hyperactivity was also observed in lymphoblasts from iPD patients. 62In these studies, high OCRs were not accompanied by alterations in mitochondrial membrane potential or mtDNA copy number, 60,62 which is in line with our observations and further supports our stratification approach based on nuclear-encoded mitochondrial genes.
The results obtained in dermal fibroblasts are certainly informative and help to identify biological processes and molecular alterations also playing a relevant role in the neuronal environment, but their gene expression profile, metabolic adaptation to stress and signaling pathways strongly differ from that of neurons. 63Moreover, the epigenetic changes that occur during somatic cell reprogramming to iPSCs and the subsequent differentiation into neuronal precursors could lead to opposing results between primary skin fibroblasts and smNPCs. 64n accordance with this, mitochondrial respiration in iPSC-derived neuronal progenitor cells established from the corresponding reprogrammed fibroblasts was very similar between low and high OXPHOS-PRS groups.However, when forced to use OXPHOS instead of glycolysis to produce ATP (i.e., in galactose medium), smNPCs derived from iPD patients with high OXPHOS-PRSs displayed a significant reduction in their OCRs compared to the low OXPHOS-PRS group.Importantly, the degree of mitochondrial dysfunction in smNPCs from iPD patients with high OXPHOS-PRSs was comparable to that observed in cells from autosomal recessive PD patients carrying PINK1 loss-of-function variants (see Fig 4).Of note, PINK1-PD is typically characterized by early onset and long disease duration, a phenotype that-according to our findings-is also observed in iPD patients with high OXPHOS-PRSs.Interestingly, smNPCs from sporadic PD patients with high OXPHOS-PRS displayed no significant differences in mitochondrial respiration compared to the control groups (either with low or high OXPHOS-PRS).This further highlights the capacity of our stratification approach based on OXPHOS-PRS to identify, among the heterogenous group of sporadic PD patients, those individuals with altered mitochondrial function who will benefit the most from therapies targeting mitochondria.Indeed, based on insights into the pathophysiology of neurodegeneration in PD, mitochondria early emerged as a potential therapeutic target for disease-modifying treatments (DMTs). 65However, all clinical trials focusing on mitochondrially active compounds in PD (e.g., coenzyme Q10) were inconclusive, raising doubts about the validity of the concept of applying DMTs to unselected patient groups given the complex and heterogeneous nature of PD. 66 Therefore, more recently, first clinical trials have been initiated, using genetic stratification strategies to define subgroups of PD patients, for example, carriers of rare variants in genes associated with mitochondrial dyshomeostasis and early-onset forms of monogenic PD. 67 If such approaches were successful, it would be still a very limited subgroup of PD patients that could benefit from these efforts.Thus, our robust and high-throughput compatible OXPHOS-PRS strategy may address the current needs for more focused DMT trials directed to sporadic iPD patients with an underlying, genetically defined, mitochondrial phenotype, who may benefit from drugs specifically targeting the respiratory chain activity.Such an approach, coupled with compounds that have shown ample preclinical and clinical evidence of targeting mitochondrial OXPHOS defects, 68 may be effective even if a given drug has previously failed in trials involving unselected PD patient groups.In accordance with this, retrospectively applying our OXPHOS-PRS approach to a double-blind, randomized trial in PD using UDCA, a naturally occurring bile acid shown to rescue mitochondrial function, 18 we found that UDCA-treated iPD patients with high OXPHOS-PRS had the greatest improvement in brain bioenergetics as quantified by 31P-MRS.In contrast, those patients in the placebo group who had the highest OXPHOS-PRS had the most marked worsening in their Gibbs free energy (reflecting ATP hydrolysis).Due to the limited number of study participants in some PRS groups and the retrospective exploratory nature of the analysis, we must acknowledge that these findings need to be confirmed in larger datasets, which would then also need to be adequately powered to allow correlation with a possible clinical effect.However, these results further corroborate the potential of our functionally validated genetic stratification approach and highlight the possibility to enrich future clinical trials of mitochondrially active drugs with iPD patients who have a high OXPHOS-PRS.Finally, as increased mitochondrial genetic risk is often associated with early-onset PD forms, individuals with very high OXPHOS-PRS could also represent a suitable target for preventative therapies.
Although the association between OXPHOS-PRS and PD risk is weaker than that observed with the wholegenome PD-PRS (see Fig 1), narrowing specific pathways-with fewer genes, but still significant effectmay shed new light on potential molecular mechanisms underlying PD pathogenesis.Thus, unlike the wholegenome PD-PRS, we believe that our genetic stratification approach based on OXPHOS-PRS represents a "precision tool" that may have a considerable impact on personalized medicine in PD.Another limitation of this study is the relatively small size of the Luxembourg Parkinson's Study cohort, which decreased the statistical power of our analyses; therefore, one could speculate that individuals with extreme OXPHOS-PRS values may not be in the top/bottom ranks of a larger cohort.This would explain why the inverse association between OXPHOS-PRS and AAO was only statistically significant in iPD patients from the large COURAGE-PD dataset, whereas the differences, although consistent (and corroborated by the correlation analysis), did not reach statistical relevance in the Luxembourg Parkinson's Study.It is worth mentioning that also whole-genome PRSs were previously found to be associated with earlier AAO, 69,70 whereas another study reported a positive correlation between mitochondrial PRSs and PD AAO. 14 While such conflicting results may simply be attributed to the different sizes of gene-sets used, it should be noted that monogenic PD forms with predominant mitochondrial pathogenesis (e.g., PINK1-PD) are also typically characterized by an early AAO.Importantly, our findings indicate that iPD patients with high OXPHOS-PRS are not only clinically, but also functionally similar to PINK1-PD patients.Functional validation of OXPHOS-PRSs in a larger number of iPD patients from additional large and deep-phenotyped PD cohorts are warranted to further explore the relationship between genetic stratification and the development of PD-associated clinical outcomes.On the other hand, available genetic information, clinical data and, at least in part, biospecimens from the subjects longitudinally followed in the Luxembourg Parkinson's Study, make this monocentric cohort quite unique, and the concomitant assessment of PD phenotypes, environmental risk factors and omics data could be integrated with OXPHOS-PRS to improve stratification accuracy of iPD patients.Combining genetic analyses, experimental validation studies and clinical assessment, all from the same individuals, certainly goes far beyond existing work, and provides a novel concept for precision medicine that is relevant for physicians and researchers engaged in translating fundamental research into therapeutic intervention for the treatment of neurodegenerative diseases.
. The degree of overlap between the different mitochondrial gene-sets is presented in Fig S1.As shown in Fig 1C, only variants in the OXPHOS gene set were significantly associated with a higher PD risk in the Luxembourg Parkinson's Study (OR = 1.31[1.14-1.50],FDR-adj p-value = 5.4e-04).To assess genetic risk without considering the potential contribution of variants in known PD loci included in the OXPHOS gene list (i.e., PINK1, SNCA and PARK7, as defined by the MDS gene classification list for PD, https://www.mdsgene.org),we performed the PRS analysis for the significant OXPHOS pathway after the exclusion of these genes.In this setting, the OR was slightly reduced (OR = 1.24[1.08-1.43],FDR-adj p-value = 5.3e-03), but the association with PD risk remained significant (Fig 1D).To take into account SNVs in potential gene regulatory elements, we have also calculated pathwayspecific mitoPRSs after extending boundaries upstream (35 kb) and downstream (10 kb) of each gene. 35Similar results as before were observed, except for the Billingsley I gene set that was no longer significant (OR = 1.15[1.00-1.32],FDR-adj p-value = 0.08).The other mitochondrial gene-sets were still significantly associated with PD: Billingsley II (OR = 1.21[1.05-1.39],FDR-adj p-value = 1.4e-02),MitoCarta3.0 (OR = 1.27[1.10-1.46],

FIGURE 1 :
FIGURE 1: Common variants in mitochondrial genes are associated with higher PD risk.(A-D) Forest plots of the odds ratio (OR) and 95% confidence interval for the whole genome (A), 3 different mitochondrial gene sets (B), and 6 selected mitochondrial pathways (C) polygenic risk scores (PRSs) regressed with PD diagnosis in the Luxembourg Parkinson's Study (in red) and COURAGE-PD (in dark blue) cohorts.(D) OR for the significant OXPHOS pathway was also plotted upon removal of PD-related genes (i.e., PINK1, SNCA, and PARK7) from the OXPHOS gene list.Only significant FDR-adjusted p-values are shown on top of each point (* p-value < 0.05).The number of SNVs used to calculate each PRS is shown on the right side.(E, F) Distribution of standardized OXPHOS-PRS in iPD patients (iPD, red) and healthy controls (HCs, light blue) from the Luxembourg Parkinson's Study (E) and COURAGE-PD (F) cohorts.The PRS distribution between iPD and HC was significantly different in both datasets (** Wilcoxon test p-value < 0.01; **** Wilcoxon test p-value < 0.0001).Colored dots in E indicate PRS values of iPD cases and HC in the highest and lowest OXPHOS-PRS range (10 th and 90 th percentile, n = 4 for each group) whose primary skin fibroblasts were available for functional studies.[Color figure can be viewed at www.annalsofneurology.org] as a larger replication dataset to validate the results obtained in the exploratory Luxembourg Parkinson's Study.After filtering and QC, the COURAGE-PD dataset comprised 7,270 iPD cases and 6,819 HCs.The demographic features of COURAGE-PD are shown in Table.The genome-wide PRS as well as all tested mitoPRSs, including OXPHOS-PRS (OR = 1.23[1.18-1.27],FDR-adj p-value = 1.5e-29), were significantly associated with PD risk in COURAGE-PD (Fig 1A-D).Again, OXPHOS-PRS remained significantly associated with PD risk when the PD-related genes PINK1, SNCA, and PARK7 were excluded from the OXPHOS gene list (OR = 1.06[1.02-1.10],FDR-adj p-value = 5.6e-04) (Fig 1D).
PRS distribution between iPD and HCs was significantly different both in the Luxembourg Parkinson's Study (Wilcoxon test p-value = 1.4e-03,Fig 1E) and in the COURAGE-PD cohorts (Wilcoxon test p-value = 2.2e-16, Fig 1F).

FIGURE 2 :
FIGURE 2: Analysis of mitochondrial respiration in primary skin fibroblasts from iPD patients and HCs stratified based on OXPHOS-PRS.Oxygen consumption rates (OCRs) were measured under basal conditions and after targeted inhibition of specific respiratory chain complexes by using a standard Seahorse Mito Stress test.Histobars represent the mean AE SEM of 7 fibroblast lines per group established from the Luxembourg Parkinson's Study participants (HCs vs iPD patients) with high (HC_H; PD_H) or low (HC_L; PD_L) OXPHOS-PRS.For each cell line, at least 3 independent experiments were performed.At least 5 technical replicates per group (i.e., distinct Seahorse wells) were analyzed in each experiment.One-way ANOVA correcting for multiple comparisons using the Tukey's post hoc test.*p-value < 0.05.[Color figure can be viewed at www.annalsofneurology.org]

FIGURE 3 :
FIGURE 3: Assessment of additional readouts of mitochondrial activity in primary skin fibroblasts and plasma samples from iPD patients and HCs stratified based on OXPHOS-PRS.(A-C) High-throughput confocal microscopy analyses in primary skin fibroblasts derived from Luxembourg Parkinson's Study participants with high (HC_H; PD_H) or low (HC_L; PD_L) OXPHOS-PRS.(A) ROS levels were quantified by normalizing the CellRox mean fluorescence intensity against the nuclear area, as defined by the Hoechst staining.Histobars represent the mean AE SEM of 4 distinct fibroblast lines per group.Three independent experiments were performed for each line.One-way ANOVA correcting for multiple comparisons using the Tukey's post hoc test.**p-value < 0.01; *** p-value < 0.001.ns = not significant.(B) Mitochondrial membrane potential (ΔΨ m ) was measured after normalization of TMRE mean fluorescence intensity by the nuclear area.Treatment with the OXPHOS uncoupler carbonyl cyanide 3-chlorophenylhydrazone (CCCP), known to induce mitochondrial depolarization, was used as positive control for decreased ΔΨ m .Dimethyl sulfoxide (DMSO) was used as a vehicle.Histobars represent the mean AE SEM of 4 distinct fibroblast lines per group.At least 3 independent experiments were performed for each line.Two-way ANOVA correcting for multiple comparisons using the Tukey's post hoc test.****p-value < 0.0001.ns = not significant.(C) Morphometric analysis of the mitochondrial network.Form factor and aspect ratio were quantified as described previously. 71Histobars represent the mean AE SEM of 4 distinct fibroblast lines for each group.At least 3 independent experiments were performed for each line.One-way ANOVA correcting for multiple comparisons using the Tukey's post hoc test.ns = not significant.(D) Measurement of IL-6 levels in plasma samples obtained from Luxembourg Parkinson's Study participants (HCs vs iPD patients) with high (HC_H, n = 7; PD_H, n = 10) or low (HC_L, n = 7; PD_L, n = 9) OXPHOS-PRS.Brown-Forsythe and Welch ANOVA correcting for multiple comparisons using Dunnett's T3 test.ns = not significant.[Color figure can be viewed at www.annalsofneurology.org]

FIGURE 5 :
FIGURE 5: Association of OXPHOS-PRSs with age at PD onset and disease duration.(A, B) Comparison of mean age at PD onset (AAO, A) and mean disease duration (B) in iPD patients from the Luxembourg Parkinson's Study and COURAGE-PD cohorts with high or low OXPHOS-PRS.*p-value < 0.05, ns = not significant.(C, D) Changes in Gibbs free energy of ATP hydrolysis (ΔG ATP ) in UP Study PD patients treated with UDCA or placebo for 48 weeks.Mean (diamond) and standard deviation (error bars) are shown for high-(red), intermediate-(orange), or low-(blue) PRS groups stratified according to OXPHOS-PRS (C) or MitoCarta-PRS (D).Two-way ANOVA considering treatment effect and interaction between treatment and PRS groups, followed by posthoc pairwise t-test between significant comparison groups using Tukey's correction.*p-value < 0.05.[Color figure can be viewed at www.annalsofneurology.org]

FIGURE 6 :
FIGURE 6: Study design and workflow.Created with BioRender.com.[Color figure can be viewed at www.annalsofneurology.org] Genotyping of the UP study cohort was undertaken at the Laboratory of Neurogenetics, NIH, Bethesda, Maryland, USA (Dr Sonja W. Scholz).All participants signed a written informed consent according to the Declaration of Helsinki.Ethical approval was granted by the National Committee for Ethics in Research, Luxembourg (Comité National d'Ethique de Recherche; CNER #201411/05).For UK-based studies, ethics were granted by the local Research Ethics Committees; REC 18/NW/0328 and REC 18/EE/0280.

TABLE . Demographic
Data of iPD Patients and Healthy Controls (HCs) from the Luxembourg Parkinson's Study and COURAGE-PD cohorts Note: Means and percentages were calculated after filtering and quality controls of genotyping data.