To identify relevant studies for inclusion in this meta-analysis, we searched PubMed, Google Scholar, and the Cochrane Library from inception through July 30, 2017. To avoid omitting relevant trials, we also searched conference summaries and reference lists from general reviews on mGluR5 treatment in advanced PD. Two reviewers (XRZ and ZRZ) independently screened the titles, abstracts, and references from all identified reports.

The Medline (PubMed) search strategy was as follows:

#1. (Parkinson's disease [mh]) OR (Idiopathic Parkinson's Disease) OR (Lewy Body Parkinson Disease) OR (Lewy Body Parkinson's Disease) OR (Primary Parkinsonism) OR (Parkinsonism, Primary) OR (Paralysis Agitans)

We included clinical trials that met the following criteria: (1) randomized controlled design trials comparing mGluR5 antagonists with placebo to treat idiopathic advanced PD, (2) trials with participants aged 30 years or older and masked assessment of outcomes, (3) trials describing advanced PD patients with L-dopa-induced dyskinesia, and (4) reports that were published in English. mGluR5 antagonists are defined as a drug with reported antagonism on mGluR5 irrespective of actions at other receptors. Trials were excluded for the following reasons: (1) trials that were not RCTs, including case reports, abstracts, comments, reviews, and editorials; (2) trials conducted in non-human subjects; (3) trials that did not conduct tests to evaluate the effects of mGluR5 antagonists on LID patients; and (4) reports that were duplicate publications.

For each trial, detailed information was carefully extracted from all the eligible trials, including the first author, year of publication, sample size, and basic sample characteristics, study population, study design, mGluR5 antagonist protocols, and outcome measures. Outcome measures included modified Abnormal Involuntary Movements Scale (mAIMS) (Marconi et al., 1994), Lang-Fahn Activities of Daily Living Dyskinesia Scale (LFADLDS) (Parkinson Study Group, 2001), Unified Parkinson's Disease Rating Scale (UPDRS), and adverse events. Finally, we also extracted the main results of the included trials.

The risk of bias for included trials was assessed independently using the Cochrane Handbook for Systematic Reviews of Interventions (Higgins et al., 2011). The details of risk of bias analysis are consistent with our previous paper (Xie et al., 2016).

All statistical analyses were done in accordance with the intention-to-treat (ITT) principle. In this study, primary outcome measures (such as LFADLDS, mAIMS and UPDRS) were treated as continuous data. Weighted mean differences (WMD), standard statistics that measure the absolute difference between the mean values in two groups, were given in this meta-analysis. The mean effect was expressed as WMD with 95% confidence intervals (Vesterinen et al., 2014). We adapted a random effect model rather than a fixed effect model to measure pooled effect sizes. We chose the random model because it takes into account the potential clinical or methodologic heterogeneity between multiple studies and also yields a more conservative estimate of the pooled effect. We used the Q statistic and the I² index to assess the statistical heterogeneity. A probability value of P ≤ 0.1 and an I² value >50% are indicative of heterogeneity between included studies, as the values exceed what is expected by chance (Higgins et al., 2003). If outcomes were presented at different time points, we extracted data from the last time point of analysis. If means and standard deviations were not provided, we calculated them from standard errors, CI, or other statistical indices. We performed a sensitivity analysis to examine whether our results would have differed by removing each individual study from the total and reanalyzing the remainder. Finally, we used the Grading of Recommendations, Assessment, Development and Evaluation (GRADE) approach to assess the quality of evidence. There are four levels for rating quality of evidence: high, moderate, low, and very low (Atkins et al., 2004). Publication bias was not assessed due to the limited trials. All analyses were performed with RevMan version 5.1. Probability value of P < 0.05 was considered significant.

Our initial search of electronic databases yielded 320 publications. After screening through titles, abstracts and full texts, 9 trial reports were identified for the final meta-analysis (Berg et al., 2011; Stocchi et al., 2013, 2014; Trenkwalder et al., 2014, 2016; Kumar et al., 2016; Tison et al., 2016). No further studies were identified by manual searches of the three journals from which most eligible studies were identified electronically (Movement Disorders, Parkinsonism & Related Disorders and Neurology). Of the resulting 9 reports that were reviewed in full text, 8 studies used mavoglurant as the mGluR5 antagonist and the 1 remaining study used dipraglurant (Figure 1).

Nine RCTs, with a total of 776 LID patients, met the inclusion criteria and were included in this review. Among the patients, 507 were randomized to mGluR5 antagonist groups, and 269 were randomized to placebo groups. The number of subjects included in this meta-analysis ranged from 14 to 197. Meanwhile, the time of follow-up ranged from 16 days to 12 weeks. The mGluR5 antagonists used in this study include mavoglurant (n = 8) and dipraglurant (n = 1). For outcome measures, LFADLDS was recorded in 6 trials, mAIMS in 8 trials, and UPDRS Parts III and IV in 7 and 6 trials, respectively. One paper by Daniela et al. (Berg et al., 2011) contained two trials (NCT00582673 and NCT00888004); similarly, one paper by Claudia et al. (Trenkwalder et al., 2016) included two phase 2 randomized, double-blind trials (NCT01385592 and NCT01491529). Two trials have not yet been published (Higgins et al., 2003; Atkins et al., 2004), and the information of two other trials with mavoglurant are due to be presented at the 18th International Congress of Parkinson's disease and Movement Disorders. The basic characteristics of the 9 trials are summarized in Table 1.

Figure 2 shows the risk of bias in the included trials. Eight trials described the method of randomization used. Seven trials assessed whether an adequate concealment of allocation procedure was used, and eight trials reported methods for blinding participants. Seven trials described intention-to-treat analyses (ITT) and reported follow-up data. Selective reporting was found in one trial. Therefore, all of the included trials were determined to have a low risk of bias.

Antidyskinetic: In this meta-analysis, mAIMS was available from 6 trials of mGluR5 antagonists compared with placebo and showed a significant difference between the two groups (p = 0.010, WMD = −2.20, 95% CI: −3.88 to −0.53, Figure 3A). Meanwhile, there was obvious heterogeneity in the analysis of mAIMS between trials (Chi² = 112.55, p < 0.00001, I² = 96%, Figure 3A). Two trials failed to pool analysis due to the data being deficient, and neither reported significant differences observed between the two groups (p = 0.2095 and p = 0.8640, respectively). For LFADLDS, no significant difference was obtained between mGluR5 antagonists and placebo based on four trials (n = 234, WMD = 0.06, 95% CI: −0.78 to 0.89, P = 0.90; heterogeneity: Chi² = 14.34, p = 0.002, I² = 79%, Figure 3B). The remaining two trials were not included in the meta-analysis due to the lack of detailed data reported. Analogously, neither of them reported significant effects of mavoglurant in reducing LFADLDS scores compared with placebo group (p > 0.05). Moreover, we merged all the data and discovered no significant difference between mGluR5 antagonists and placebo according to UPDRS Part IV (n = 234, WMD = −0.11, 95% CI: −0.35 to 0.13, P = 0.36; heterogeneity: Chi² = 24.78, p < 0.0001, I² = 88%, Figure 3C). Antiparkinsonian: Four trials reported no significant effect of mGluR5 antagonists for reducing UPDRS Part III scores compared with placebo (n = 310, p = 0.26, WMD = −0.42, 95% CI: −1.17 to 0.32, Figure 3D). Meanwhile, all the results showed heterogeneity when using the heterogeneity test (Chi² = 35.28, p < 0.00001, I² = 89%, Figure 3D).

In the nine studies, the incidence of AEs was higher in patients in the mGluR5 antagonist groups than in the placebo groups (Table 2). AEs were typically mild to moderate in severity and included nervous system, psychiatric, and gastrointestinal disorders. Among them, the most common AEs were dizziness (16.3% in mGluR5 antagonist groups vs. 4.3% in placebo groups), visual hallucination (10.1 vs. 1.1%), fatigue (10.1 vs. 4.8%), insomnia (6.1 vs. 1.6%), nasopharyngitis (6.1 vs. 1.6%) and diarrhea (5.1 vs. 2.1%). Overall, 10.1% of patients were reported to have dyskinesia in the mGluR5 antagonist groups, which was lower than 21.9% in the placebo groups. No deaths were reported in either of the treatment groups in any of the trials.

Using the GRADE criteria, we characterized the quality of evidence presented in this meta-analysis as low to moderate. LFADLDS and UPDRS Part IV had a low level of evidence across all domains, and mAIMS and UPDRS Part III had a moderate level of evidence. The basic information for all core comparisons of the outcomes of the included trials is summarized in Table 3.

Several conclusions can been draw based on the findings of this meta-analysis. First, our meta-analysis found that mGluR5 antagonists had a greater treatment effect on mAIMS scores (p = 0.01) in LID patients. However, there were no significant improvements in anti-dyskinesia in LFADLDS (p = 0.42) and UPDRS Part IV (p = 0.20) compared with placebo, as well as no effects on anti-parkinsonism (p = 0.25). Second, 9 trials evaluated the incidence of adverse events and none were reported. There were no major safety concerns associated with mavoglurant (100–200 mg) or dipraglurant (50–300 mg). The incidence of AEs was higher in patients in the mGluR5 antagonist groups than in the placebo group, especially increased dizziness, visual hallucination and fatigue. Finally, using the GRADE system, we characterized the quality of evidence in our meta-analysis as low to moderate. To our best knowledge, the meta-analysis was unable to determine if mGluR5 antagonists were beneficial in the treatment of PD or LID patients.

The development of LID is associated with increased glutamatergic signaling in the basal ganglia, particularly in its major input system (Chase and Oh, 2000). Inhibition of glutamatergic transmission signals by the selective mGluR5 antagonists have been shown to alleviate LID in rodent and monkey models of PD (Levandis et al., 2008; Morin et al., 2010). Moreover, the NMDA-receptor antagonist amantadine has shown effectiveness in ameliorating LID in patients, although its transitory effects and multiple neurological side effects in some patients limit its clinical use (Bibbiani et al., 2005). Consistently, our meta-analysis found that mGluR5 antagonists had a greater treatment effect compared with placebo on mAIMS scores in LID patients. Nevertheless, based on the combined analysis, we did not find other secondary end points with consistent improvements in dyskinesia scores (LFADLDS and UPDRS Part IV) for mGluR5 antagonists compared to placebo. Namely, the results on clinician-rated measures of dyskinesia were conflicting; a slight improvement was observed in the mAIMS scores, while the LFADLDS and UPDRS Part IV scores were not different between mGluR5 antagonists and placebo. The reasons for this discrepancy in the study outcomes were not apparent, but there were differences in the design of these trials and the trial sample was limited. Meanwhile, the LFADLDS and UPDRS Part IV scales may not have been sensitive enough to detect the observed antidyskinetic effects, or the magnitude of therapeutic effect may be too small to generate a change on these measures (Goetz et al., 2013). Therefore, these results should be interpreted with caution and additional large multi-center RCTs are required to assess the full potential of mGluR5 antagonists in patients with LID. Moreover, mGluR5 antagonists were not related to any significant effects on the UPDRS Part III scores in either trial, suggesting mGluR5 antagonists did not alter the antiparkinsonian effect of L-dopa therapy.

Several limitations of this meta-analysis should be mentioned. First, a relatively small sample size of LID patients was included from seven papers with nine trials. Our results could be constrained by the unclear risk of bias due to incomplete data in a few trials. Second, the longest duration of treatment was 12 weeks; therefore, the long-term antidyskinetic effects and cost-effectiveness of mGluR5 antagonists could not be assessed here. Additionally, a few uncontrolled variables, such as disease stage, medication use, and types of mGluR5 antagonists, could confound the results. Third, mGluR5 antagonists in this study include only mavoglurant (n = 8) and dipraglurant (n = 1) because the other antagonists trials were not identical and may have influenced our results. Moreover, only one trial used dipraglurant. Finally, two trials have not yet been published, but will be presented at the 18th International Congress of Parkinson's Disease and Movement Disorders. These trials lacked detailed data on all scales.

The findings from this meta-analysis using mGluR5 antagonists showed a greater treatment effect on the mAIMS scores but no significant improvements on LFADLDS and UPDRS Part IV scores. Therefore, more rigorously designed RCTs with longer follow-up periods are necessary in future. A number of adverse events occurred more frequently with mGluR5 antagonist treatments than with placebo, and they were mainly central nervous system symptoms. Therefore, the balance between the efficacy and side effects requires future studies. In general, there were no reports of life-threatening events in these trials. In addition, optimal dosages of compounds need to be refined to define the therapeutic window between efficacy and unacceptable side effects in future. Furthermore, combination studies of mGluR5 antagonists with other non-glutamatergic antidyskinetic drugs would be interesting. Finally, once the efficacy of mGluR5 antagonists is established, new lines of research should focus on establishing predictors of the intervention outcomes. These predictors could include age of disease onset, types of mGluR5 antagonists and treatment regimens, and genetic influences on the response to mGluR5 antagonists. All these issues should be researched by future RCTs.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.