The role of dopaminergic medication and specific pathway alterations in idiopathic and PRKN/PINK1-mediated Parkinson’s disease
Metabolomics in idiopathic and PRKN/PINK1-linked PD
Institute of Neurogenetics, University of Lübeck, Lübeck, Germany.
Department of Neurology, University of Lübeck and University Hospital Schleswig-Holstein, Campus Lübeck, Lübeck, Germany.
Department of Clinical and Movement Neurosciences, UCL Queen Square Institute of Neurology, London, UK.
Digital Twin Center, School of Medicine, University of Galway, University Rd, Galway, Ireland.
Metabolomics and Analytics Centre, Leiden Academic Centre for Drug Research, Leiden University, Leiden, Netherlands.
Institute of Medical Biometry and Statistics, University of Lübeck, Lübeck, Germany.
*Corresponding author. Email: christine.klein@uni-luebeck.de (C.K.); silke.szymczak@uni-luebeck.de (S.S.)Abstract
Parkinson’s disease (PD) is the second most common neurodegenerative disease, with a rapidly increasing prevalence worldwide. Biomarkers monitoring state and progression are urgently needed, and metabolomics from easily accessible biofluids holds the potential to elucidate pathophysiological underpinnings in PD. Several studies suggested metabolomic differences between patients and controls, but findings are controversial, and independent replication is scarce. We thus applied state-of-the-art, large-scale metabolomics in patients with idiopathic and monogenic PD and controls from two independent samples, analyzed by a strict meta-analysis approach. Thereby, we (i) debunked that l-Dopa medication and not disease status causes the most substantial metabolomic differences and (ii) identified polyamine metabolism alterations, partly, but not entirely associated with l-Dopa treatment. Furthermore, we found explorative but robust evidence for alterations in endocannabinoid metabolites; detected lipid metabolism alterations, highlighting potential crosslinks with alpha-synuclein pathology; and provided evidence for a metabolomic signature for the role of oxidative damage in patients with PRKN- and PINK1-linked PD.
Teaser
L-Dopa medication substantially affects the Parkinson’s metabolome, while specific pathway alterations might be disease dependent.
INTRODUCTION
Parkinson’s disease (PD) is the second most common neurodegenerative disease with increasing prevalence worldwide (1). However, there are no blood-based biomarkers available in clinical practice (2), and metabolomics from easily accessible biofluids as blood hold the potential to assess disease state and progression. Studies investigating metabolites in the blood of patients with Parkinson’s disease (PD) have suggested alterations in a plethora of metabolites and biochemical pathways (3). However, only a small fraction of the suggested metabolites have been replicated in independent studies (4). Possible explanations lie in major technological and procedural pitfalls that can substantially influence the results and, therefore, must be accounted for in metabolomic studies. First, almost every patient with PD is treated with dopaminergic medication, which hampers the discrimination between disease- and drug-related effects; this is especially true of dopamine metabolism. Therefore, adjusting for treatment in statistical analyses may eliminate not only the effect of Levodopa (l-Dopa) but also the alterations caused by the disease itself when comparing only l-Dopa–treated PD patients to healthy controls (HCs). Second, as metabolite concentrations are very susceptible to handling, processing, and type of analysis, even slight differences in the procedures across laboratories can considerably affect the results (4). Third, sex, age, and lifestyle also substantially affect the metabolome (5).
Adding another level of complexity, PD is a heterogeneous disorder, including several forms of monogenic PD, such as PRKN- or PINK1-linked PD (6, 7). Most likely, differences in the underlying molecular course will involve different biochemical pathways, resulting in different levels of specific metabolites for each genetic subset of patients. For example, PRKN- and PINK1-linked PD involves mitochondrial dysfunction (8), making alterations in pathways of oxidative phosphorylation more likely.
An important role has been attributed to lipids in the pathogenesis of PD, as genetic alterations in several genes involved in lipid metabolism have been identified to cause (e.g., PINK1) PD or are involved in disease pathogenesis (e.g., VPS13C) (9). Strong evidence suggests that membrane lipids are highly important for alpha-synuclein (α-syn) metabolism, contributing to α-syn fibrilization and accumulation in laboratory models (10). Strikingly, α-syn–lipid interactions are likely an essential component in Lewy body formation and, possibly, for spreading pathology (9). Among other lipids, levels of phosphatidylcholines, lysophosphatidylcholines, and phosphatidylethanolamines (PEs) have been found to discriminate patients with idiopathic PD (IPD) (11, 12) and carriers of pathogenic LRRK2 (13) variants from HC. However, these lipids have yet to be investigated in detail in biallelic PRKN- or PINK1-linked PD. Furthermore, investigations of the lipid metabolism in patients with drug-naïve IPD are scarce. One study on patients with untreated IPD had suggested different free fatty acids, indolelactic acid, and phenylacetylglutamine to be able to distinguish patients from HC (14).
Our study addressed two aims: (i) To take the influence of dopaminergic treatment on blood-based metabolomics into account, hypothesizing that this medication has a substantial impact on metabolomics in patients with IPD and monogenic PD and (ii) to identify dopaminergic medication–independent metabolites with different intensities in idiopathic and PRKN/PINK1-linked PD compared to HC by large-scale targeted metabolomics, hypothesizing that distinct alterations in metabolites might be characteristic for (“mitochondrial”) PD.
RESULTS
Study population
The study sample at both study sites consists of a total of 223 participants. The 140 patients with IPD [55 females (39%)] had a median age at examination of 62 years [interquartile range (IQR) of 54 to 70 years; range, 28 to 84 years], a median age at onset of 54 years (IQR of 46 to 63 years; range, 25 to 81 years], and a median disease duration of 6.3 years (IQR of 4.1 to 10.9 years; range, 0.9 to 22.9 years). The 19 PRKN/PINK1-linked PD patients [9 females (47%)] had a median age at examination of 52 years (IQR of 47 to 59 years; range, 24 to 72 years), a median age of onset of 31 years (IQR of 23 to 37 years; range, 13 to 66 years), and a median disease duration of 16 years (IQR of 9.6 to 30.2; range, 4.0 to 48.3 years). The HC group comprised 64 individuals with a median age at examination of 62 years (IQR of 33 to 67 years; range, 25 to 80 years). Clinical scores, l-Dopa equivalent daily dose (LEDD), and daily l-Dopa dose are shown in Table 1.
| All IPD | l-Dopanegative | l-Dopapositive | MitoPD | HC | |
|---|---|---|---|---|---|
| n | 140 | 30 | 110 | 19 | 64 |
| AOO (years) | 54 (46–63; 25–81) | 49 (42–58; 25–69) | 55 (48–65; 31–81) | 31 (23–37; 13–66) | NA |
| AAE (years) | 62 (54–70; 28–84) | 54 (49–61; 28–78) | 65 (58–71; 37–84) | 52 (47–59; 24–72) | 62 (33–67; 25–80) |
| Disease duration (years) | 6.3 (4.1–10.9; 0.9–22.9) | 4.3 (2.9–6.5; 0.9–22.9) | 6.5 (4.6–11.4; 0.9–22.8) | 16.3 (9.6–30.2; 4.0–48.3) | NA |
| LEDD (mg) | 499.5 (300–804; 0–1782) | 147.5 (85–209; 0–500) | 595 (409–954; 127–1782) | 350 (255–694; 0–1288) | 0 |
| Daily l-Dopa dose (mg) | 300 (150–506; 0–1225) | 0 | 375 (300–600; 37.5–1225) | 225 (112.5–312.5; 0–750) | 0 |
| MDS UPDRS III | 25 (18–37; 4–70) | 25 (18–35; 6–69) | 25.5 (18–38; 4–70) | 25 (16–34; 6–64) | 3 (2–6; 0–14) |
| H/Y | 2 (1–2; 1–4) | 2 2 (1–2; 1–3) | 2 (1–2; 1–4) | 2 (2–3; 1–5) | 0 |
| Sex | F: 55 (39%); M: 85 (61%) | F: 15 (50%); M: 15 (50%) | F: 40 (36%); M: 70 (64%) | F: 9 (47%); M: 10 (53%) | F: 34 (53%); M: 30 (47%) |
Analyses
Because of the site effects described in Materials and Methods, the final analyses reported in the text refer to a meta-analysis. The figures show the individual-level data for the specific metabolites; forest plots summarizing the site specific and combined effect estimates are shown in fig. S1. For comparison purposes, we also conducted a discovery/replication stage analysis using the University College London (UCL) cohort as the discovery cohort.
l-Dopa treatment but not dopamine agonist treatment increases l-Dopa–dependent metabolites in IPD
First, we explored differences in metabolites comparing all patients with IPD (n = 140) and HC. Here, we identified five metabolites, i.e., 3-methoxytyramine, methyldopa, putrescine, (+/−)-16-hydroxydocosahexaenoic acid (16-HDoHE), and ornithine, as the only significant metabolites after correction for multiple testing (Table 2 and table S4). We observed a heterogeneous distribution within the IPD group, particularly for 3-methoxytyramine and methyldopa, as these metabolites were predominantly increased in l-Dopa–treated individuals (in the following referred to as l-Dopapositive), but not similarly in untreated patients with IPD and individuals only treated with dopamine agonists (Fig. 1). Concluding that l-Dopa treatment is responsible for these notable differences between patients with IPD and HC, we continued analyzing the l-Dopapositive and the l-Dopanegative group separately to elucidate differences in the metabolome not caused by dopaminergic treatment. The demographical and clinical data of the two IPD subgroups are likewise displayed in Table 1.
| Name | z value | Nominal P | Adjusted P | β (CI) |
|---|---|---|---|---|
| 3-Methoxytyrosine | 13.376 | 8.38 × 10−41 | 2.55 × 10−38 | 1.514 (1.292; 1.736) |
| Methyldopa | 8.624 | 6.45 × 10−18 | 9.8 × 10−16 | 1.179 (0.911; 1.447) |
| Putrescine | 4.321 | 0.0000155 | 0.00157 | 0.5 (0.273; 0.726) |
| (+/−) 16-HDoHE | 3.906 | 0.000094 | 0.00714 | 0.629 (0.313; 0.945) |
| Ornithine | 3.6 | 0.000318 | 0.0194 | 0.516 (0.235; 0.797) |
| -Threonine | 2.897 | 0.00377 | 0.191 | 0.464 (0.15; 0.779) |
| Triglyceride (54:7) | −2.701 | 0.00692 | 0.301 | −0.466 (−0.805; −0.128) |
| 1-Arachidonoyl Glycerol / 2-Arachidonoyl Glycerol | −2.589 | 0.00963 | 0.366 | −0.413 (−0.726; −0.1) |
| Triglyceride (54:1) | −2.476 | 0.0133 | 0.378 | −0.427 (−0.764; −0.089) |
| N6,N6,N6-Trimethyl-l-lysine | 2.472 | 0.0134 | 0.378 | 0.378 (0.078; 0.677) |
| Tiglylcarnitine | −2.465 | 0.0137 | 0.378 | −0,406 (−0,728; −0.083) |
| Homocysteine | 2.372 | 0.0177 | 0.399 | 0.353 (0.061; 0.644) |
| Triglyceride (50:4) | −2.368 | 0.0179 | 0.399 | −0.4 (−0.73; −0.069) |
| 2-ketoglutaric acid | −2.317 | 0,0205 | 0.399 | −0.362 (−0.669; −0.056) |
| Phosphatidylcholine (38:4) | −2.287 | 0.0222 | 0.399 | −0.405 (−0.751; −0.058) |
| Phosphatidylcholine (O-36:5) | −2.252 | 0.0243 | 0.399 | −0.395 (−0,739; −0.051) |
| Methylmalonylcarnitine | −2.227 | 0.026 | 0.399 | −0.341 (−0.641; −0.041) |
| Gamma-Glutamylglutamine | 2.211 | 0.027 | 0.399 | 0.35 (0.04; 0.661) |
| Triglyceride (54:5) | −2.208 | 0.0273 | 0.399 | −0.386 (−0.729; −0.043) |
| Phosphatidylethanolamine (O-38:5) | −2.173 | 0.0298 | 0.399 | −0.396 (−0.753; −0.039) |
| Phosphatidylcholine (38:7) | −2.171 | 0.0299 | 0.399 | −0.378 (−0.719; −0.037) |
| Cystathionine | 2.131 | 0.0331 | 0.399 | 0.286 (0.023; 0.549) |
| Triglyceride (52:4) | −2.124 | 0.0337 | 0.399 | −0.376 (−0.724; −0.029) |
| Phosphatidylcholine (36:4) | −2.112 | 0.0347 | 0.399 | −0.369 (−0.712; −0.027) |
| Triglyceride (48:3) | −2.088 | 0.0368 | 0.399 | −0.334 (−0.648; −0.02) |
| Glycoursodeoxycholic acid | −2.084 | 0.0371 | 0,399 | −0.281 (−0.545; −0.017) |
| Phosphatidylethanolamine (O-36:5) | −2.063 | 0.0391 | 0.399 | −0.372 (−0.726; −0.019) |
| Triglyceride (51:4) | −2.063 | 0,0391 | 0.399 | −0.346 (−0.674; −0.017) |
| Triglyceride (56:6) | −2.061 | 0.0393 | 0.399 | −0.362 (−0.706; −0.018) |
| Glycochenodeoxycholic acid | −2.06 | 0.0394 | 0.399 | −0.308 (−0.6; −0.015) |
| Ceramide (d18:0/22:0) | −2.047 | 0.0407 | 0.399 | −0.352 (−0.69; −0.015) |
Considering further factors potentially influencing these five metabolites in all patients with IPD, we investigated a possible association between 3-methoxytyramine, methyldopa, putrescine, 16-HDoHE, and ornithine and disease duration. Here, we only found a weak association between 16-HDoHE levels and disease duration {regression coefficient = 0.057 [confidence interval] (CI): 0.023 to 0.091, adjusted P = 0.028}, which was not present in the other four metabolites.
Investigating the l-Dopapositive group separately, we identified the same five metabolites already mentioned above, i.e., methoxytyramine, methyldopa, 16-HDoHE, ornithine, and putrescine, to be significantly elevated in patients compared to HC after adjusting for multiple testing (Fig. 2, Table 3, and table S5). As expected, the l-Dopa downstream metabolites (15) methoxytyramine and methyldopa (adjusted P < 0.0001) were highly elevated. Despite also accounting for dopaminergic treatment other than l-Dopa, the LEDD was associated with the levels of both metabolites in the l-Dopapositive group [methoxythyramine: regression coefficient = 0.55 (CI: 0.42 to 0.66, adjusted P < 0.0001); methyldopa: regression coefficient = 0.39 (CI: 0.24 to 0.52, adjusted P < 0.0001)]. As expected, the daily l-Dopa dose in l-Dopapositive individuals was also strongly associated with methoxytyramine and methyldopa levels [methoxythyramine: regression coefficient = 4 × 10−04 (CI: 0.00021 to 0.00060, adjusted P = 0.0002); methyldopa: regression coefficient = 0.00127 (CI: 0.00060 to 0.00195, adjusted P = 0.0004)].
| Name | z value | Nominal p | Adjusted p | Beta (CI) |
|---|---|---|---|---|
| 3-Methoxytyrosine | 1.95 | 0 | 0 | 1.95 (1.873; 2.026) |
| Methyldopa | 1.571 | 6.36 × 10−40 | 9.67 × 10−38 | 1.571 (1.339; 1.804) |
| (+/−) 16-HDoHE | 0.809 | 2.24 × 10−06 | 0.000227 | 0.809 (0.474; 1.144) |
| Putrescine | 0.547 | 6.39 × 10−06 | 0.000485 | 0.547 (0.309; 0.784) |
| Ornithine | 0.582 | 6.36 × 10−05 | 0.00387 | 0.582 (0.297; 0,868) |
| N6,N6,N6-trimethyl-l-lysine | 0.493 | 0.00265 | 0.135 | 0.493 (0.171; 0.814) |
| l-Threonine | 0.486 | 0.00366 | 0.159 | 0.486 (0.158; 0.813) |
| Homocysteine | 0.403 | 0.00949 | 0.309 | 0.403 (0.099; 0.708) |
| l-Tyrosine | 0.416 | 0.0102 | 0.309 | 0.416 (0.099; 0.734) |
| 2-ketoglutaric acid | −0.424 | 0.00907 | 0.309 | −0.424 (−0.742; −0.105) |
| Triglyceride (54:7) | −0.446 | 0.0144 | 0.398 | −0.446 (−0.803; −0.089) |
| Triglyceride (54:1) | −0.39 | 0.0161 | 0.407 | −0.39 (−0.708; −0.073) |
| Methylmalonylcarnitine | −0.367 | 0.0176 | 0.41 | −0.367 (−0.67; −0.064) |
| Cystathionine | 0.324 | 0.0196 | 0.425 | 0.324 (0.052; 0.597) |
| Gamma-glutamylglutamine | 0.377 | 0.0232 | 0.47 | 0.377 (0.052; 0.703) |
| Glycocholic acid | −0.321 | 0.0301 | 0.571 | −0.321 (−0.611; −0.031) |
| Glycochenodeoxycholic acid | −0.345 | 0.0319 | 0.571 | -0.345 (−0.66; -0.03) |
| Deoxycarnitine | −0.266 | 0.0346 | 0.584 | -0.266 (−0.513; -0.019) |
| Saccharopine | 0.354 | 0.0403 | 0.625 | 0.354 (0.016; 0.692) |
| 2-hydroxybutyric acid | −0.361 | 0.0425 | 0.625 | -0.361 (−0.711; -0.012) |
| Glycoursodeoxycholic acid | −0.295 | 0.0473 | 0.625 | -0.295 (−0.587; -0.004) |
| Cer(d18:0/22:0) | −0.345 | 0.0444 | 0.625 | -0.345 (−0.68; -0.009) |
| Phosphatidylcholine (38:4) | −0.375 | 0.0456 | 0.625 | -0.375 (−0.742; -0.007) |
We confirmed these findings by applying a discovery/replication design analysis using the UCL cohort as the discovery cohort because it contains more patients with IPD and controls. Fifteen metabolites were selected for the replication stage (table S6), including 3-methoxytyramine, methyldopa, ornithine, and putrescine, but not (+/−)-16-HDoHE bceause the nominal P value was 0.125. Testing these metabolites in the UL cohort, 3-methoxytyramine, methyldopa, ornithine, and putrescine could be replicated (tables S7 and S8).
Differences between l-Dopanegative IPD patients and HCs
When comparing l-Dopanegative IPD patients to HC, there were no significant differences in metabolite concentration if only adjusted P values were considered. Notably, the most strikingly elevated metabolites in l-Dopapositive individuals, i.e., methoxytyramine and methyldopa, were not at all increased in l-Dopanegative IPD patients compared to controls (Fig. 1). Regarding 16-HDoHE, ornithine, and putrescine, there was no difference in l-Dopanegative IPD patients compared to HC after adjusting for multiple testing. However, putrescine displayed a nominal P value of 0.026 [adjusted P = 0.56, regression coefficient = 0.386 (CI: 0.046 to 0.725)] (Fig. 2).
Among the top hits with nominal P values < 0.05 in l-Dopanegative IPD patients, which did not show up in l-Dopapositive individuals, we found several membrane lipids such as lysophosphatidic acids (16:0 and 14:0), phosphatidylcholines (36:4, 38:7, and 0 to 36:5), and phosphoethanolamines (34:2 and 36:4). Furthermore, we found the endocannabinoids arachidonoyl glycerol (1-AG/2-AG) and linoleoyl glycerol (1-LG/2-LG) to be lowered in l-Dopanegative IPD patients. Notably our methods did not allow us to distinguish between 1-AG and 2-AG as well as 1-LG and 2-LG, respectively. In addition, we found triglycerides (50:1, 52:1, 54:7, and 56:6), glutathione, citric acid, and thromboxane B2 to be elevated. Three acetyl-carnitines were elevated (hexadecenoylcarnitine, oleoylcarnitine, and tetradecenoylcarnitine), whereas one was lowered (tiglylcarnitine) (Table 4 and table S9).
| Name | z value | Nominal P | Adjusted P | β (CI) |
|---|---|---|---|---|
| Lysophosphatidic acid (16:0) | −3.418 | 0.000631 | 0.0959 | −0.685 (−1.078; −0.292) |
| 1-Arachidonoyl glycerol/2-arachidonoyl glycerol | −3.589 | 0.000332 | 0.0959 | −0.762 (−1.178; −0.346) |
| Glutathione | 2.751 | 0.00594 | 0.301 | 0.432 (0.124; 0.74) |
| 1-Linoleoyl glycerol (18:2)/2-linoleoyl glycerol (18:2) | −2.766 | 0.00567 | 0.301 | −0.573 (−0.978; −0.167) |
| Phosphatidylcholine (36:4) | −2.858 | 0.00426 | 0.301 | −0.634 (−1.069; −0.199) |
| Phosphatidylcholine (38:7) | −2.764 | 0.0057 | 0.301 | −0.571 (−0.975; −0.166) |
| Citric acid | 2.618 | 0.00885 | 0.384 | 0.57 (0.143; 0.997) |
| Hexadecenoylcarnitine | 2.532 | 0.0114 | 0.432 | 0.533 (0.12; 0.946) |
| Phosphatidylethanolamine (34:2) | −2.458 | 0.014 | 0.449 | −0.606 (−1.089; −0.123) |
| Phosphatidylethanolamine (36:4) | −2.438 | 0.0148 | 0.449 | −0.58 (−1.046; −0.114) |
| Oleoylcarnitine | 2.383 | 0.0172 | 0.474 | 0.505 (0.09; 0.92) |
| Tiglylcarnitine | −2.321 | 0.0203 | 0.514 | −0.61 (−1.126; −0.095) |
| Putrescine | 2.227 | 0.026 | 0.56 | 0.386 (0.046; 0.725) |
| Lysophosphatidic acid (14:0) | −2.107 | 0.0351 | 0.56 | −0.398 (−0.768; −0.028) |
| Glycoursodeoxycholic acid | −2.117 | 0.0342 | 0.56 | −0.449 (−0.865; −0.033) |
| Thromboxane B2 | −2.132 | 0.033 | 0.56 | −0.374 (−0.718; −0.03) |
| Phosphatidylcholine (O-36:5) | −2.088 | 0.0368 | 0.56 | −0.505 (−0.978; −0.031) |
| Triglyceride (50:1) | −2.195 | 0.0282 | 0.56 | −0.49 (−0.928; −0.052) |
| Triglyceride (52:1) | −2.176 | 0.0296 | 0.56 | −0.49 (−0.932; −0.049) |
| Triglyceride (54:7) | −2.154 | 0.0313 | 0.56 | −0.515 (−0.984; −0.046) |
| Oleic acid | 2.024 | 0.043 | 0.623 | 0.486 (0.015; 0.956) |
| Tetradecenoylcarnitine | 1.963 | 0.0497 | 0.653 | 0.445 (0.001; 0.89) |
| Triglyceride (56:6) | −1.968 | 0.049 | 0.653 | −0.478 (−0.955; −0.002) |
Disease duration and disease severity in l-Dopanegative IPD patients are associated with alterations in different metabolite classes
We conducted explorative regression analyses to investigate whether there was a nominal association between the discovered altered metabolites and disease duration and disease severity in the l-Dopanegative group. Regarding carnitines, we observed an exploratory association between disease duration and hexadoconylcarnitine (nominal P value < 0.05) (Fig. 3A), but no association between carnitines and Hoehn and Yahr (H/Y) scores.
With respect to lipids, higher levels of two phosphatidylcholines (PC 36.4 and PC 38.7) were associated with longer disease duration (nominal P value < 0.05) (Fig. 3, B and C). Last, we found a negative association between putrescine and glutathione and disease severity (nominal P value < 0.05) (Fig. 3, D and E), but no association with disease duration.
None of the other metabolites showed any significant association with H/Y stage or disease duration upon explorative regression analysis. The association between all investigated metabolites and disease duration and H/Y stage is shown in tables S10 and S11.
In the mitoPD group, hydroxyeicosatetraenoic acids (HETEs) are elevated, which were not found to be altered in patients with IPD besides likewise increased l-Dopa–dependent metabolites
Most of the patients with mitoPD (27 of 30) were treated with l-Dopa, thus investigating a l-Dopanegative group was not possible. Again, the l-Dopa–dependent metabolites methyldopa [regression coefficient = 1.95 (CI: 1873 to 2026); adjusted P < 0.0001] and methoxytyramine [regression coefficient = 1571 (1339 to 1804); adjusted P < 0.0001] were elevated in the mitoPD group compared to HC after adjusting for multiple testing. Within metabolites with nominal P values < 0.05, we found among the top hits not only 16-HDoHE, l-tyrosine, putrescine, and ornithine but also 5-HETE, 8-HETE, 11-HETE, and 15(S)-HETrE (table S12). Association analyses comparing metabolite levels and disease duration and severity in patients with mitoPD were not performed due to the small sample size.
DISCUSSION
Our study demonstrates (i) that l-Dopa treatment substantially affects the metabolomic profile in PD and (ii) that levels of specific lipids and endocannabinoids were different between patients with PD and controls even after confounding effects of l-Dopa therapy were excluded. Our approach of differentiating between l-Dopa and agonist treatment allows new insights into the PD metabolome, implicating previous studies’ interpretation and future metabolomics study design in PD.
Although preanalytical sample handling was strictly harmonized as described above, we still observed variability in metabolite levels between the two study sites. The distinct reasons for this variability are not entirely clear. However, variability might be influenced by differences in diet (16); lifestyle factors, including physical activity and comorbidities (17, 18); and the gut microbiome (19) between participants. We also cannot fully exclude variations during temperature-controlled shipment from the different study sites to the institution where metabolomic measurements were performed. This variability must be considered in multicenter metabolomic studies in general. However, it can be addressed by appropriate statistical methods, as demonstrated in our study by the performed meta-analysis approach.
l-Dopa treatment
Many previous PD metabolomic studies have highlighted several metabolites as promising candidates for differentiating HC from patients with PD. However, by differentiating between l-Dopa and agonist treatment/no dopaminergic medication, we found several of these metabolites were increased by l-Dopa treatment itself, but not by agonist treatment.
Thus, we have addressed a major challenge faced in studies of the PD metabolism, namely, that most of the patients included are already undergoing treatment with l-Dopa. Expectedly, methyldopa and methoxytyramine, both breakdown products of l-Dopa, were highly elevated in patients with l-Dopa–treated IPD and mitoPD. They were neither elevated in patients with IPD without l-Dopa treatment nor in patients with l-Dopa agonist treatment only. As expected, the LEDD and the daily l-Dopa dose correlated with levels of methyldopa and methoxytyramine in the blood.
In addition, we found 16-HDoHE, a very long-chain fatty acid, to be significantly elevated in l-Dopa–treated patients. Higher levels of 16-HDoHE were also weakly associated with longer disease duration. This suggests that this metabolite is partly influenced by l-Dopa intake and, at the same time, affected by disease duration. 16-HDoHE is a type of hydroxyeicosatetraenoic acid (HETE) that is derived from arachidonic acid (20). Both arachidonic acid and docosahexaenoic acids are polyunsaturated fatty acids present in phospholipids of membranes of the body’s cells and are abundant in the brain. HETEs are markers of oxidative damage and are increased in IPD (21).
α-Syn has been reported to immediately change its structure in the presence of both arachidonic acid and docosahexaenoic acids to its α helical conformation. Upon prolonged exposure to docosahexaenoic acids, α-syn gradually assembles into amyloid-like fibrils, with the docosahexaenoic acid being part of the aggregate (10, 22).
However, 16-HDoHE did not cluster with its downstream products or other arachidonic acid derivatives such as other hydroxydocosahexaenoic acids in our study. It had similar levels in the plasma of non–l-Dopa–treated participants and HC. Therefore, the role of 16-HDoHE in the PD metabolome is unclear based on our data. Otherwise, HETEs were among the top hits of altered metabolites investigating patients with mitoPD, which are, however, entirely treated with l-Dopa, potentially supporting the role of l-Dopa treatment on this specific metabolite pathway.
MitoPD
Only one study investigated the metabolome in PRKN-linked PD (49), and no investigations addressing patients with PINK1-linked PD have been available so far. We combined these two autosomal recessively inherited forms of PD, as the proteins encoded by these genes act in a common biochemical pathway, mainly involved in the degradation of damaged mitochondria (8), but links to changes in the innate immune system have also been described to be associated with PRKN and PINK1 dysfunction (50). We were able to partly reproduce the differences described in the only other metabolomics study investigating patients with PRKN-linked PD, where an elevation of fatty acids and oxidized lipids and a decrease of antioxidants, caffeine, and benzoate-related metabolites were reported. We found several hydroxyeicosatetraenoic acids [5-HETE, 8-HETE, 11-HETE, and 15(S)-HETrE] elevated at nominal P values. These metabolites were not found to be different in the comparison between IPD and HC. As explained above, this might reflect different pathways of oxidative dysfunction in these subsets of patients with PD known to display mitochondrial alterations. This idea of different pathophysiological pathways in monogenic and IPD (51), as well as different mechanisms such as mitochondrial versus lysosomal dysfunction within idiopathic PD (52), is well-established and should be considered in potential disease-modifying trials. However, as almost all patients with PRKN/PINK1-linked PD in this study were treated with l-Dopa, conclusions drawn from these analyses are limited due to the substantial effect of l-Dopa treatment on the metabolome. Methoxytyrosine and methyldopa were the only metabolites that were significantly different between the mitoPD group and HC. Furthermore, most metabolites with nominal P values < 0.05—such as (+/−) 16-HDoHE, l-tyrosine, putrescine, and ornithine—were identified as l-Dopa–dependent within our study. Thus, investigating the mitoPD group did not allow us to draw definite conclusions, as almost all patients were treated with l-Dopa, and the sample size was too small.
Strengths and limitations
The strengths of our study consist of considering the substantial effect of l-Dopa medication and investigating two independent samples of probands recruited with the same detailed protocol to ensure identical preanalytical handling. Nonetheless, we observed site-dependent effects. Thus, we applied a strict statistical approach to correct for this bias and for multiple testing. Furthermore, we used state-of-the-art large-scale comprehensive metabolomic platforms to measure more than 300 metabolites and identify differences in single metabolites rather than in clusters.
Although our statistical approach includes replication of metabolites at both study sites, some of our findings must be considered preliminary, as they formally only reach nominal significance and, therefore, require replication in independent studies. Moreover, relevant effects in the l-Dopanegative group may have been missed due to the sample size of n = 30. Further limitations include the cross-sectional study design and the relatively small sample size regarding patients with PRKN-/PINK1-linked disease, making association analysis with disease severity impossible. Moreover, no group of patients with pathogenic biallelic variants in PRKN and PINK1 without l-Dopa treatment was available, preventing deeper insight into the effects of dopaminergic medication on the metabolome in this group of patients with PD. In addition, to investigate the detected alterations as potential biomarkers for PD, a comparison with atypical PD forms, such as progressive supranuclear palsy and multiple system atrophy, would be required in future studies.
In conclusion, we investigated a broad range of more than 300 metabolites that covered all major biochemical pathways. Adjusting for multiple testing, only metabolites affected by l-Dopa treatment significantly differed, highlighting the dominant effect of l-Dopa on the PD metabolome. However, several metabolites had nominal P values < 0.05 comparing HCs and non–l-Dopa–treated PD patients, which we consider relevant as (i) applying strict statistical correction for multiple testing might mitigate all but significant effects if such a magnitude of tests are performed; (ii) several of these metabolites have been consistently reproduced in other studies as we recently reviewed (4); (iii) several of these metabolites belong to the same biological pathways, making a role in PD metabolism more plausible; and (iv) some of the relevant metabolites were associated with disease state and progression, further highlighting their biological impact. Of special importance, our statistical approach directly includes the replication of our findings in two independent cohorts, representing a crucial step that was lacking in most of the PD metabolomics studies performed so far (4).
Overall, our study underscores the large impact of l-Dopa treatment on metabolomic studies in PD. Thus, we highlight the risk of misinterpreting differences between patients with IPD and HCs as disease specific, when in fact they may simply relate to l-Dopa treatment. Thus, the impact of dopaminergic treatment should be considered in all future metabolomics studies. This is also relevant for biomarker studies dealing with PD in general, as with recent findings implicate that l-Dopa treatment even affects PD imaging biomarkers (53). Furthermore, our association analyses revealed the evidence of different metabolite trajectories throughout the disease; however, longitudinal metabolomic studies are warranted to assess relevant metabolite changes over the disease course. Last, we found evidence for the role of specific glycerophospholipids and endocannabinoids in disease progression and severity, supporting recent findings that lipid pathway alterations influence PD pathogenesis.
MATERIALS AND METHODS
The SysMedPD study
The Systems Medicine of Mitochondrial PD (SysMedPD) study (www.sysmedpd.eu) was initiated by an international consortium, including basic and clinician scientists, to contribute to the elucidation of mitochondrial underpinnings of PD. The consortium consisted of eight partners from five European countries. At the same time, recruitment of patients with idiopathic and different forms of monogenic PD, as well as healthy controls, took place at the Institute of Neurogenetics at the University of Lübeck (UL) and the Department of Neurology at the UCL. Before recruitment, inclusion and exclusion criteria, the clinical work-up, and the sampling of biomaterials were harmonized between both centers.
Ethics statement
The study was approved by the Ethics Committee of the University of Lübeck (study number 16-195) and London (REC reference number 17/LO/1166). Written informed consent was obtained from all participants. The study was performed in accordance with the Declaration of Helsinki.
Proband recruitment
The total sample at the two recruitment sites comprised 289 participants. Inclusion criteria for affected individuals consisted of a definite diagnosis of PD according to the MDS clinical diagnostic criteria for PD (54), eligibility according to the harmonized genetic criteria (see below), and material availability to perform metabolomics analyses. Otherwise, patients with clinical suspicion of atypical PD, coexisting neurodegenerative disorders, diagnosed dementia, or a history of severe infection (defined by the need for antibiotics) or surgery in 2 weeks before participation were excluded. A flowchart providing the selection process for inclusion in the study is depicted in Fig. 4. All participants underwent an in-depth clinical assessment for motor symptoms and clinical severity using the Movement Disorder Society Unified Parkinson’s Disease Rating Scale III (MDS-UPDRS III) and the H/Y scale. However, because of superior robustness, we only used H/Y scores for exploratory association analyses. Moreover, the LEDD was calculated as published (55).
Genetic testing
We recruited patients with known pathogenic variants in PRKN and PINK1 and uncovered priorly unknown genetic alterations by genotyping performed within the study. All participants underwent conclusive genetic testing within the study, which differed slightly between both sides. At UL, all samples underwent gene panel sequencing at CENTOGENE (Rostock, Germany). The panel covered the genes PRKN, PINK1, DJ-1, SNCA, LRRK2, GBA1, VPS35, GCH1, and others irrelevant to the present study (56). We confirmed potentially disease-related variants by Sanger sequencing. At UCL, the Illumina NeuroChip array, a customized version of the Infinium HumanCore-24 v1.0 backbone, was used, covering nearly 200,000 variants associated with neurodegenerative diseases, including but not limited to PD (57). Furthermore, we performed multiplex ligation-dependent probe amplification (MLPA) to investigate copy number variations (i.e., deletions or duplications) in PD-associated genes (SNCA, PRKN, PINK1, DJ-1, LRRK2, and GCH1). We performed sequence analyses and MLPA in patients and control participants at both sites. The pathogenicity of detected variants was assessed according to the ACMG (American College of Medical Genetics and Genomics) (58) and the MDSGene (www.mdsgene.org) criteria.
Genetic study inclusion criteria were likewise harmonized between both study sites: (i) We excluded patients with PD and controls with pathogenic variants in known PD genes besides pathogenic biallelic PRKN and PINK1 variants. (ii) We excluded individuals with heterozygous pathogenic variants in PRKN and PINK1 due to conflicting evidence regarding their role in PD (59, 60).
Study groups
After the exclusion of screened participants due to either clinical criteria (n = 4), genetic criteria (n = 46), or lack of sufficient biomaterial to perform the experiments (n = 13), 223 participants were included. This comprised 140 patients with IPD, 19 patients with PRKN/PINK1-linked PD (PRKN: n = 16 individuals; PINK1: n = 3) and 64 healthy controls undergoing metabolomic analyses (Fig. 4). Carriers of biallelic pathogenic variants in PRKN and PINK1 formed the “mitoPD” group for all analyses.
Preanalytic sample processing
Blood samples for metabolomic and genetic analysis were collected from each participant at 8 a.m. in the morning after a 12-hour fasting period, where participants were only allowed to take medication and drink water. Following the venous blood collection, the samples were placed on ice and promptly conveyed to the laboratory within 15 min. The specimens underwent centrifugation at a force of 2000g for 10 min at a controlled temperature of 4°C (centrifuges: Beckman Coulter Allegra X-15R and Eppendorf centrifuge 5417R).
Subsequently, the resulting plasma supernatants were transferred into a fresh 15-ml Falcon tube, prechilled on ice, and subjected to thorough mixing through inversion. Following this step, aliquots of 300-ml each were prepared, rapidly cryopreserved by immersion in liquid nitrogen, and ultimately stored at a temperature of −80°C. Specimens from UL and UCL were shipped on dry ice to the University of Leiden under strictly temperature-controlled conditions.
Statistics
Metabolite data were further preprocessed and quality controlled. If a metabolite was available on several platforms, then measurements from the platform with the lowest relative SD based on quality-control samples were kept. Missing values were replaced by a constant metabolic-specific value defined as 0.5 × minimally observed value. A log transformation (base2) was applied to make the distribution of the intensities of each metabolite more symmetric. Following this transformation, the z score was calculated for each metabolite to achieve a distribution with a mean of 0 and an SD of 1 using the mean value of intensities of each metabolite divided by the SD, facilitating the comparison of metabolites.
Additional quality control steps included missing frequencies per metabolite and proband, comparison of intensity distributions across subjects, principal components analysis (PCA), and checks for duplicate samples as implemented in the R package QCnormSE (version 0.99.4.9000) available on Zenodo (https://zenodo.org/records/14728213). No metabolites or samples were excluded due to quality issues.
However, we observed a strong batch effect of recruitment sites on the overall distribution of the metabolite intensities in the PCA plots, especially for amines and lipid metabolites. Thus, we first applied a linear regression analysis for each metabolite separately, with transformed metabolite intensity as the dependent variable and recruitment site as the variable of interest and adjusted for age at examination, sex, and diagnosis group (IPD or control) as independent variables. P values for the effect of the recruitment site were adjusted for multiple testing using the Benjamini-Hochberg procedure (64). Because many metabolites showed a significant association, we conducted all further association analyses of metabolite intensities with clinical variables stratified by recruitment site. Site-specific results containing the mean metabolite intensities in each group and the results of the linear regression analysis with metabolite intensity as the dependent variable and the group, sex, and age as independent variables (regression coefficient for group, SE, and nominal and adjusted P values) are shown in tables S7 and S8.
Again, we used a linear regression model for each metabolite separately, with transformed metabolite intensity as the dependent variable and the variable of interest (categorical or continuous) as the independent variable and included additional covariates. Details on the participants and covariates included in each analysis and the corresponding tables and figures can be found in table S13. In particular, most of the analyses were adjusted for age at examination. However, we did not adjust for age at examination when analyzing disease severity measured as H/Y stage because of the substantial overlap between increasing age and disease severity in patients with PD that is similar to the overlap of PD status and levodopa treatment. Therefore, metabolites associated with age at examination and disease severity will be separately discussed.
We used a meta-analysis approach to combine the results of the two recruitment sites, which has several substantial advantages compared to a two-stage design with a discovery and replication cohort. It provides a more comprehensive and objective assessment of the evidence, greater statistical power, and more precise estimates (65) without the necessity to specify the criteria for selecting metabolites for replication and declaring successful replication in advance (66). In particular, the risk of false positive findings is substantially decreased because similar effects in both studies must be observed for each metabolite to be called statistically significant. In addition, the risk of false-negative results is also reduced because all individuals are used for identification compared to the smaller sample size in the discovery cohort.
Regression coefficients (β) and SEs of the variable of interest estimated for the two recruitment sites were combined in a meta-analysis using the R package metafor (version 3.8-1) (67). We used a fixed-effects meta-analysis because a random-effects meta-analysis is not recommended with two studies (68), and the same protocols for recruitment, clinical characterization, and sample collection were used at the two sites. P values were adjusted for multiple testing using the Benjamini-Hochberg procedure (64). In the results section, we clearly distinguish between significant findings (adjusted P value < 0.05) and nominally significant results (nominal = unadjusted P value < 0.05) for some explorative analyses. In both cases, we report the meta-analysis’s regression coefficients (β) with corresponding 95% CIs in Results. In addition, we generated forest plots for the main findings (fig. S1), which show the fixed-effects regression coefficients together with information on the heterogeneity of effects between studies. In particular, we report the results of Cochran’s Q test (69) and the heterogeneity statistic I2 using the DerSimonian-Laird estimator of τ2 (70).
We conducted a discovery/replication study to compare it to our preferred meta-analysis approach. We selected the UCL dataset as the discovery cohort because it contained more patients with IPD and controls. We used the comparison of patients with l-Dopapositive versus controls. Metabolites were selected for the replication stage if the nominal P value was <0.05 in the discovery cohort. The P values in the replication cohort were adjusted for multiple testing using the Benjamini-Hochberg procedure (tables S6 to S8) (64).
Acknowledgments
We thank the study participants at both clinical recruitment sites for their willingness to support our research.
Untitled section
Funding: The project was funded by the EU Horizon 2020 programme, grant agreement 668738 (SysMedPD). The German Research Foundation (DFG) funded A.B.’s research stay at Leiden, the Netherlands. X.L. is funded by a Chinese Scholarship Council (CSC) scholarship (grant number 202006370070).
Untitled section
Author contributions: A.B.: Writing—original draft, conceptualization, investigation, methodology, resources, data curation, validation, formal analysis, project administration, and visualization. M.B.: Writing—original draft, conceptualization, investigation, methodology, resources, project administration, and visualization. P.C.: Conceptualization, writing—review and editing, methodology, and resources. X.L.: Writing—review and editing and validation. J.H.: Writing—review and editing and resources. T.B.: Investigation, Writing—review and editing, methodology, resources, data curation, validation, formal analysis, and project administration. C.L.: Investigation, writing—review and editing, methodology, and resources. A.H.: Investigation, writing—review and editing, and methodology. K.L.: Writing—review and editing, resources, and validation. E.B.: Writing – review and editing, resources, data curation, and formal analysis. H.R.M.: Conceptualization, investigation, writing—review and editing, methodology, funding acquisition, supervision, and project administration. A.H.S.: Conceptualization, investigation, writing—review and editing, methodology, resources, funding acquisition, validation, and project administration. T.H.: Writing—review and editing, methodology, funding acquisition, supervision. R.F.: Writing—review and editing, methodology, and supervision. S.S.: Conceptualization, writing—review and editing, data curation, formal analysis, and visualization. C.K.: Conceptualization, investigation, writing—review and editing, methodology, resources, funding acquisition, data curation, validation, supervision, and project administration.
Untitled section
Competing interests: The authors declare that they have no competing interests.
Untitled section
Data and materials availability: The raw data supporting this study’s findings were uploaded to Zenodo (https://zenodo.org/records/15124574). In addition, the data were submitted to the Leipzig Health Atlas (www.health-atlas.de/projects/72). To request access to the data, please fill out and sign the Data Access Agreement document (www.health-atlas.de/documents/58/download?version=1) and send it to the SysMedPD Metabolomics Data Access Committee (christine.klein@uni-luebeck.de).