Systematic search
We initially identified 1398 studies, of which 1375 were excluded because they were either duplicates or irrelevant. After a full-text review of the remaining 23 articles, a further two were excluded because of missing data, leaving 21 studies for final analysis (Fig. 2 and Table 1) [48,49,50,51,52,53,54,55,56,57,58,59,60,61,62,63,64,65,66,67,68]. All studies investigated patients with schizophrenia diagnosed according to current diagnostic criteria, barring one study group of patients with schizoaffective disorder [66], and two studies of patients with first-episode psychosis [60, 62].
PRISMA 2020 flow diagram.
Individual metabolites
Arginine
Study characteristics
Twelve studies, including 16 study groups, reported serum arginine concentrations in 1116 patients with schizophrenia (mean age 35 years, 58% males) and 1110 healthy controls (mean age 34 years, 55% males) [48, 51, 53,54,55, 59, 60, 62, 64, 65, 67, 68]. Six studies were conducted in Europe [48, 51, 53, 54, 60, 64], and the remaining six in Asia [55, 59, 62, 65, 67, 68]. Liquid chromatography was used in all studies for the determination of arginine. Eight studies utilised mass spectrometry detection [51, 54, 59, 60, 62, 64, 67, 68], three fluorimetric detection [53, 55, 65], whilst the remaining one did not provide details regarding the detection method [48]. Patients received pharmacological treatment for schizophrenia in eight study groups [48, 53,54,55, 60, 62, 64, 67], and no treatment in five [48, 59, 60, 62, 64]. Both treated and untreated patients were included in two study groups [51, 68], whereas no treatment-specific information was provided in the remaining one [65].
Risk of bias
All studies had a low risk of bias (Supplementary Table 3) [48, 51, 53,54,55, 59, 60, 62, 64, 65, 67, 68].
Results of individual studies and syntheses
Random-effects models were used given the high heterogeneity observed (I2 = 94.4%, p < 0.001). Pooled results showed no significant differences in serum arginine concentrations between patients with schizophrenia and controls (SMD = 0.26, 95% CI −0.13 to 0.65, p = 0.19; Supplementary Fig. 1). Sensitivity analysis showed stability of the results, with an effect size ranging between 0.17 and 0.34 (Supplementary Fig. 2).
Publication bias
There was no evidence of publication bias according to either the Begg’s (p = 0.08) or the Egger’s (p = 0.59) test. Accordingly, the “trim-and-fill” method did not identify any missing study to be added to the funnel plot to ensure symmetry (Supplementary Fig. 3).
Meta-regression and sub-group analyses
No significant associations were observed in univariate meta-regression between the effect size and age (t = 0.64, p = 0.54), proportion of males (t = −0.78, p = 0.45), publication year (t = −0.28, p = 0.79), or sample size (t = −0.20, p = 0.84).
In subgroup analysis, a trend toward a significant difference (p = 0.066) was observed in the effect size between European (SMD = −0.12, 95% CI −0.31 to 0.07, p = 0.20; I2 = 53.3%, p = 0.03) and Asian studies (SMD = 0.74, 95% CI −0.05 to 1.53, p = 0.07; I2 = 96.7%, p < 0.001), with a lower between-study variance in the former subgroup (Supplementary Fig. 4). No significant differences (p = 0.33) in pooled SMD were observed between studies using mass spectrometry (SMD = 0.10, 95% CI −0.39 to 0.59, p = 0.69; I2 = 95.4%, p < 0.001) and fluorimetric detection (SMD = 0.75, 95% CI −0.04 to 1.54, p = 0.06; I2 = 89.8%, p < 0.001; Supplementary Fig. 5). Similarly, no significant differences (p = 0.69) in pooled SMD were observed between studies conducted in untreated (SMD = 0.45, 95% CI −0.24 to 1.13, p = 0.20; I2 = 93.4%, p < 0.001) and treated patients (SMD = 0.23, 95% CI −0.36 to 0.82, p = 0.45; I2 = 93.4%, p < 0.001; Supplementary Fig. 6).
Certainty of evidence
The initial low level of certainty, due to the cross-sectional nature of the selected studies (rating 2, ⊕⊕⊖⊖), was downgraded to very low (rating 1, ⊕⊖⊖⊖) after considering the relatively high imprecision (confidence intervals with threshold crossing).
Asymmetric dimethylarginine
Study characteristics
Fifteen studies, including 21 study groups, reported circulating ADMA concentrations in 1050 patients with schizophrenia (mean age 36 years, 57% males) and 830 controls (mean age 33 years, 59% males) [49, 50, 52,53,54,55,56,57, 59,60,61, 63,64,65,66]. Nine studies were performed in Asia [50, 52, 55,56,57, 59, 61, 63, 65], and the remaining six in Europe [49, 53, 54, 60, 64, 66]. ADMA was measured using liquid chromatography in 14 studies [49, 50, 52,53,54,55,56,57, 59,60,61, 64,65,66], and enzyme-linked immunosorbent assay (ELISA) in the remaining one [63]. Among the liquid chromatography studies, four utilized mass spectrometry detection [54, 59, 60, 64], nine fluorometric detection [50, 52, 53, 55,56,57, 59, 61, 65, 66], whereas no relevant information was provided in the remaining one [49]. Patients received pharmacological treatment for schizophrenia in 13 study groups [53,54,55,56,57, 60, 61, 63,64,65,66], and no treatment in seven [50, 52, 59,60,61, 63, 64]. In one study group, the studied cohort included both treated and untreated patients [49].
Risk of bias
All studies had a low risk of bias (Supplementary Table 3) [49, 50, 52,53,54,55,56,57, 59,60,61, 63,64,65,66].
Results of individual studies and syntheses
Random-effects models were used because of the high heterogeneity observed (I2 = 92.5%, p˂0.001). Pooled results showed that serum ADMA concentrations were significantly higher in patients with schizophrenia compared to controls (SMD = 1.23, 95% CI 0.86–1.61, p < 0.001; Fig. 3). In sensitivity analysis, the corresponding pooled SMD values were stable, with an effect size ranging between 1.00 and 1.29 (Supplementary Fig. 7).
Forest plot of studies reporting serum ADMA concentrations in patients with schizophrenia and healthy controls.
Publication bias
There was a significant publication bias according to both the Begg’s (p = 0.012) and the Egger’s (p = 0.001) tests. Accordingly, the “trim-and-fill” method identified nine missing studies to be added to the left side of the funnel plot to ensure symmetry (Supplementary Fig. 8). The resulting effect size, albeit attenuated, remained significant (SMD = 0.52, 95% CI 0.11–0.92, p = 0.01).
Meta-regression and sub-group and analyses
No significant associations were observed in meta-regression between the effect size and age (t = −0.64, p = 0.53), proportion of males (t = 0.03, p = 0.98), publication year (t = −0.81, p = 0.43), or sample size (t = −0.85, p = 0.41).
In subgroup analysis, there were no significant (p = 0.16) differences in pooled SMD between European (SMD = 0.72, 95% CI 0.43–1.00, p < 0.001; I2 = 66.7%, p = 0.002) and Asian studies (SMD = 1.69, 95% CI 1.05–2.33, p < 0.001; I2 = 95.4%, p < 0.001; Supplementary Fig. 9), with a lower variance in the former subgroup. There were also no significant differences (p = 0.83) in pooled SMD between studies using liquid chromatography (SMD = 1.26, 95% CI 0.85–1.67 p < 0.001; I2 = 93.3%, p < 0.001) and ELISA (SMD = 1.11, 95% CI 0.15–2.06, p = 0.023; I2 = 82.4%, p < 0.001; Supplementary Fig. 10). Furthermore, no significant differences in pooled SMD (p = 0.20) were observed between liquid chromatography studies using fluorimetric detection (SMD = 1.73, 95% CI 1.05–2.40, p < 0.001; I2 = 94.5%, p < 0.001) and mass spectrometry detection (SMD = 0.60, 95% CI 0.23–0.98, p = 0.002; I2 = 83.4%, p < 0.001; Supplementary Fig. 11). Similarly, there were no significant differences (p = 0.44) in pooled SMD between studies investigating untreated (SMD = 1.68, 95% CI 0.71–2.64, p = 0.001; I2 = 96.6%, p < 0.001) and treated patients (SMD = 1.09, 95% CI 0.75–1.42, p˂0.001; I2 = 83.3%, p < 0.001; Supplementary Fig. 12).
Certainty of evidence
The initial level of certainty was low because of the cross-sectional nature of the selected studies (rating 2, ⊕⊕⊖⊖). This was upgraded to moderate (rating 3, ⊕⊕⊕⊖) after taking into account the relatively large effect size (SMD = 1.26) [37].
Symmetric dimethylarginine
Study characteristics
Six studies, including eight study groups, assessed circulating SDMA concentrations in 383 patients with schizophrenia (mean age 34 years, 56% males) and 326 healthy controls (mean age 31 years, 53% males) [49, 54, 55, 60, 64, 65]. Four studies were conducted in Europe [49, 54, 60, 64], and the remaining two in Asia [55, 65]. All studies used liquid chromatography, three with mass spectrometry detection [54, 60, 64], two with fluorometric detection [55, 65], whereas no relevant information was provided in the remaining one [49]. Patients received pharmacological treatment in five study groups [54, 55, 60, 64, 65], and no treatment in two study groups [60, 64]. One study group investigated both treated and untreated patients [49].
Risk of bias
The risk of bias was considered low in all studies (Supplementary Table 3) [49, 54, 55, 60, 64, 65].
Results of individual studies and syntheses
The moderate between-study heterogeneity (I2 = 72.4%, p = 0.001) warranted the use of random-effects models. Pooled results showed no significant differences in serum SDMA concentrations different between patients with schizophrenia and healthy controls (SMD = 0.21, 95% CI −0.09 to 0.51, p = 0.17; Supplementary Fig. 13). Sensitivity analysis showed that the corresponding pooled SMD values were stable, with an effect size ranging between 0.11 and 0.30 (Supplementary Fig. 14).
Publication bias
Assessment of publication bias could not be performed because of the small number of studies.
Meta-regression and sub-group and analysis
The small number of studies prevented the conduct of meta-regression analysis.
In sub-group analysis, there were no significant differences (p = 0.10) in pooled SMD between Asian (SMD = 0.79, 95% CI −0.17 to 0.29, p = 0.61; I2 = 35.6%, p = 0.17) and European studies (SMD = 0.06, 95% CI −0.17 to 0.29, p = 0.13; I2 = 90.0%, p < 0.001; Supplementary Fig. 15), with a reduced between-study variance in the former subgroup. No significant differences (p = 0.11) in pooled SMD were also observed between studies using mass spectrometry (SMD = 0.14, 95% CI −0.05 to 0.32, p = 0.14; I2 = 0.0%, p = 0.85) and fluorimetric detection (SMD = 0.79, 95% CI −0.23 to 1.80, p = 0.13; I2 = 90.0%, p = 0.002; Supplementary Fig. 16), with a virtual absence of between-study variance in the former subgroup. Similarly, there were no significant differences (p = 0.75) in pooled SMD between studies performed in treated (SMD = 0.34, 95% CI −0.03 to 0.71, p = 0.07; I2 = 75.7%, p = 0.002) and untreated patients (SMD = 0.20, 95% CI −0.11 to 0.51, p = 0.20; I2 = 0.0%, p = 0.47; Supplementary Fig. 17).
Certainty of evidence
The initial level of certainty was considered low because of the cross-sectional nature of the selected studies (rating 2, ⊕⊕⊖⊖). This was downgraded to extremely low (rating 0, ⊖⊖⊖⊖) after considering the relatively high imprecision (confidence intervals with threshold crossing) and the lack of assessment of publication bias.
Arginine/asymmetric dimethylarginine ratio
Study characteristics
Three studies, including four study groups, reported serum arginine/ADMA ratios in 202 patients with schizophrenia (mean age 29 years, 58% males) and 187 controls (mean age 27 years, 52% males) [53, 54, 64]. All studies were performed in Europe and used liquid chromatography, two with mass spectrometry detection [54, 64], and one with fluorometric detection [53]. Patients were treated in three study groups [53, 54, 64], and untreated in one [64].
Risk of bias
The risk of bias was considered low in all studies (Supplementary Table 3) [53, 54, 64].
Results of individual studies and syntheses
Random-effects models were used because of the moderate heterogeneity observed (I2 = 54.3%, p = 0.087). Pooled results showed that the arginine/ADMA ratio was significantly lower in patients with schizophrenia compared to controls (SMD = −0.35, 95% CI −0.65 to −0.04; p = 0.026; Fig. 4). In sensitivity analysis, the corresponding pooled SMD values were stable, with an effect size ranging between −0.46 and −0.25 (Supplementary Fig. 18).
Forest plot of studies reporting arginine/ADMA ratios in patients with schizophrenia and healthy controls.
Publication bias
The assessment of publication bias could not be performed because of the small number of studies.
Meta-regression and sub-group and analysis
Meta-regression and sub-group analyses could not be performed because of the small number of studies.
Certainty of evidence
The initial level of certainty for cross-sectional studies (rating 2, ⊕⊕⊖⊖) was downgraded to very low (rating 1, ⊕⊖⊖⊖) after considering the lack of assessment of publication bias.
Citrulline
Study characteristics
Eight studies, including 12 study groups, reported circulating citrulline concentrations in 599 patients with schizophrenia (mean age 31 years, 59% males) and 697 healthy controls (mean age 30 years, 58% males) [48, 54, 58,59,60, 62, 64, 67]. Four studies were conducted in Europe [48, 54, 60, 64], and the remaining four in Asia [58, 59, 62, 67]. All studies used liquid chromatography, six with mass spectrometry detection [54, 59, 60, 62, 64, 67], and one with ultraviolet detection [58], whereas no relevant information was provided in the remaining one [48]. Patients received pharmacological treatment in six study groups [48, 58,59,60, 62, 64], and no treatment in the remaining six [48, 54, 60, 62, 64, 67].
Risk of bias
The risk of bias was considered low in all studies (Supplementary Table 3) [48, 54, 58,59,60, 62, 64, 67].
Results of individual studies and syntheses
The high between-study heterogeneity observed (I2 = 91%, p < 0.001) warranted the use of random-effects models. Pooled results showed no significant differences in serum citrulline concentrations between patients with schizophrenia and controls (SMD = 0.32, 95% CI −0.10 to 0.74 p = 0.13; Supplementary Fig. 19). Sensitivity analysis showed stability of the corresponding pooled SMD values, with an effect size ranging between 0.15 and 0.42 (Supplementary Fig. 20).
Publication bias
There was no evidence of publication bias according to either the Begg’s (p = 0.15) or the Egger’s (p = 0.45) test. Accordingly, the “trim-and-fill” method did not identify any missing study to be added to the funnel plot to ensure symmetry (Supplementary Fig. 21).
Meta-regression and sub-group and analyses
There were no significant associations between the effect size and age (t = −0.81, p = 0.44), proportion of males (t = −2.12, p = 0.08), publication year (t = 0.00, p = 0.99), or sample size (t = −0.30, p = 0.77) in univariate meta-regression analysis.
In subgroup analysis, the pooled SMD was significantly different in Asian (SMD = 0.95, 95% CI 0.19 to 1.72, p = 0.015; I2 = 93.4%, p < 0.001) but not European studies (SMD = −0.15, 95% CI −0.51 to 0.21, p = 0.41; I2 = 75.1%, p = 0.001; Supplementary Fig. 22). No significant differences (p = 0.26) were also observed in pooled SMD values between studies in untreated (SMD = 0.03, 95% CI −0.54 to 0.60, p = 0.41; I2 = 90.3%, p < 0.001) and treated patients (SMD = 0.61, 95% CI −0.08 to 1.30, p = 0.08; I2 = 92.2%, p < 0.001; Supplementary Fig. 23).
Certainty of evidence
The initial level of certainty was considered low because of the cross-sectional nature of the selected studies (rating 2, ⊕⊕⊖⊖). This was downgraded to extremely low (rating 0, ⊖⊖⊖⊖) after considering the high, unexplained, heterogeneity and the relatively high imprecision (confidence intervals with threshold crossing).
Ornithine
Study characteristics
Eight studies, including 12 study groups, reported circulating ornithine concentrations in 872 patients with schizophrenia (mean age 33 years, 59% males) and 904 healthy controls (mean age 34 years, 55% males) [48, 51, 59, 60, 62, 64, 67, 68]. Four studies were conducted in Europe [48, 51, 60, 64], and the remaining four in Asia [59, 62, 67, 68]. Liquid chromatography was used in all studies, seven with mass spectrometry detection [51, 59, 60, 62, 64, 67, 68], whereas no relevant information was provided in the remaining one [48]. Patients received pharmacological treatment in five study groups [48, 59, 60, 62, 64], an no treatment in other five [48, 60, 62, 64, 67]. Two study groups investigated both treated and untreated patients [51, 68].
Risk of bias
The risk of bias was considered low in all studies (Supplementary Table 3) [48, 51, 59, 60, 62, 64, 67, 68].
Results of individual studies and syntheses
Random-effects models were used given the moderate heterogeneity observed (I2 = 59.9%, p = 0.004). Pooled results showed that serum ornithine concentrations were significantly higher in patients with schizophrenia compared to controls (SMD = 0.32, 95% CI 0.16–0.49, p < 0.001; Fig. 5). The corresponding pooled SMD values remained stable, with an effect size ranging between 0.28 and 0.36 (Supplementary Fig. 24).
Forest plot of studies reporting serum ornithine concentrations in patients with schizophrenia and healthy controls.
Publication bias
There was no significant publication bias according to either the Begg’s (p = 1.00) or the Egger’s (p = 0.56) test. Accordingly, the “trim-and-fill” method did not identify any missing study to be added to the funnel plot to ensure symmetry (Supplementary Fig. 25).
Meta-regression and sub-group and analysis
In meta-regression analysis, there were no significant associations between the effect size and age (t = 1.28, p = 0.24), proportion of males (t = 1.14, p = 0.30), publication year (t = −0.28, p = 0.79), or sample size (t = −0.28, p = 0.78).
In subgroup analysis, there were no significant differences (p = 0.94) in effect size between European (SMD = 0.33, 95% CI 0.04–0.62, p = 0.026; I2 = 73.2%, p = 0.001) and Asian studies (SMD = 0.34, 95% CI 0.17–0.51, p < 0.001; I2 = 21.1%, p = 0.28; Supplementary Fig. 26), with a substantial reduction in between-study variance in the latter subgroup. Notably, the pooled SMD was significantly different in studies of untreated (SMD = 0.26, 95% CI 0.02–0.50, p = 0.034; I2 = 49.0%, p = 0.098) but not treated patients (SMD = 0.41, 95% CI −0.04 to 0.87, p = 0.077; I2 = 78.9%, p = 0.001; Supplementary Fig. 27).
Certainty of evidence
The initial level of certainty was low because of the cross-sectional nature of the selected studies (rating 2, ⊕⊕⊝⊝). This remained low after considering the low risk of bias in all studies, the moderate but partially explained heterogeneity, the lack of indirectness, the relatively low imprecision, the relatively small effect size, and the lack of publication bias.
Dimethylamine
Study characteristics
Two studies, including three study groups, reported circulating dimethylamine concentrations in 147 patients (mean age 26 years, 52% males) and 147 healthy controls (mean age 24 years, 53% males) [54, 60]. Both studies were performed in Europe and used liquid chromatography with mass spectrometry detection [54, 60]. Patients received pharmacological treatment in two study groups [54, 60], and no treatment in the remaining one [60].
Risk of bias
The risk of bias was considered low in both studies (Supplementary Table 3) [54, 60].
Results of individual studies and syntheses
Fixed-effects models were used given the virtual absence of between-study heterogeneity (I2 = 0.0%, p = 0.68). Pooled results showed that dimethylamine serum concentrations were significantly higher in patients with schizophrenia compared to controls (SMD = 0.47, 95% CI 0.24 to 0.70, p < 0.001; Fig. 6).
Forest plot of studies reporting serum dimethylamine concentrations in patients with schizophrenia and healthy controls.
Publication bias
Assessment of publication bias could not be performed because of the small number of studies.
Meta-regression and sub-group and analysis
Meta-regression and sub-group analyses could not be performed because of the small number of studies.
Certainty of evidence
The initial level of certainty for cross-sectional studies (rating 2, ⊕⊕⊝⊝) was downgraded to very low (rating 1, ⊕⊝⊝⊝) after considering the lack of assessment of publication bias.
