Literature Search
Brain GSH literature findings
The search returned 218 unique records (Figure 1). Of the records screened, 46 studies were excluded as they were non-clinical studies (including reviews, editorials, and or conference abstracts), 121 studies were excluded because those studies involved non-human subjects, 28 studies were excluded as they were not conducted in AD or MCI patients, 1 study was excluded as it did not have a healthy control group, and 14 studies were excluded as they did not measure GSH in the brain. One paper was excluded as it was an erratum clarification that was not relevant to the results. One additional study was excluded from quantitative analysis as full results could not be obtained. A total of 4 studies were included in the AD brain GSH meta-analysis (28-31), and 4 studies were included in the MCI analysis (29, 32-34) (Table 2). Studies reporting multiple brain locations were analyzed as sub-studies, and when bilateral measures were available, the left and right voxels were averaged. A total of 7 studies and sub-studies were included for AD brain GSH analysis and 8 studies and sub-studies were included for MCI analysis. Assessment of included studies showed a consistently low risk of bias in the brain GSH literature (Table 3).
Blood GSH literature findings
The search returned 299 unique records (Figure 2). Of the records screened, 40 studies were excluded as they were non-clinical studies (including reviews, editorials, and or conference abstracts), 70 studies were excluded because these studies involved non-human subjects, 81 studies were excluded as they were not conducted in AD or MCI patients, 23 studies were excluded as they were post-mortem studies, 9 studies were excluded as they did not include a healthy control group, 47 studies were excluded as they did not measure GSH in whole blood, plasma, or serum, and 2 were excluded as full results could not be obtained. A total of 27 studies qualified, with 26 of these studies being included in the AD blood GSH meta-analysis(35-60), and 7 of these studies being included in the MCI analysis (38, 39, 41, 42, 45, 51, 61). Studies reporting GSH levels in different blood components (plasma, serum, blood cells) were analyzed as sub-studies, with a total of 31 studies/sub-studies used for AD blood GSH analysis and 8 studies/sub-studies used for MCI blood GSH analysis. The risk of bias was variable in the AD blood GSH literature but consistently low in MCI blood GSH literature (Table 3).
Diagnostic criteria used in AD and MCI
AD patients were identified primarily using the Diagnostic and Statistical Manual of Mental Disorders (DSM) (62) and/or the National Institute of Neurological and Communicative Diseases and Stroke/Alzheimer's Disease and Related Disorders Association (63). The National Institute on Aging-Alzheimer's Association diagnostic guidelines (64), The Dementia Rating Scale-2 (65), International Classification of Diseases 10th Revision (66), and the Consortium to Establish a Registry for Alzheimer's Disease (67) neuropsychological battery were used in 5 GSH studies respectively (30, 36, 47, 49, 58). In studies examining blood GSH in AD, 7 studies used the Hachiniski Ischaemic Score (HIS ≤ 4) to differentiate those with AD from those with potential vascular causes (40, 45, 54, 57, 59, 60, 68), and 5 of those studies further used neuroimaging to support diagnosis (54, 57, 59, 60, 68) (Table 2).
For MCI patient samples, the Petersen criteria (69) were commonly used to diagnose MCI, though other studies used revised Petersen criteria (70), or a combination of the Montreal Cognitive Assessment, DSM-IV, Clinical Dementia Rating, and Mini Mental State Examination (33, 61). While amnestic-type MCI patients were specifically selected in 2 blood GSH studies (41, 42), most of the studies measuring blood GSH and all the studies measuring brain GSH either did not specify, or included both amnestic and non-amnestic patients (Table 2).
Brain GSH concentrations and investigating heterogeneity
Brain GSH did not differ in AD (pooled SMD [95%CI] = 0.07 [-1.29, 1.43], p=0.6) and MCI (pooled SMD [95%CI] = -0.43 [-1.19, 0.33], p=0.26) compared to healthy controls. Significant heterogeneity was found in both AD (I2=96.5%, p<0.001) and MCI (I2=92.4%, p<0.001) and supported the use of random effect models. Subgroup analysis evaluating the use of MRS acquisition methods found that Meshcher-Garwood Point Resolved Spectroscopy (MEGA-PRESS) studies had reduced heterogeneity (AD: I2=22.5%, p=0.28, MCI: I2=67.1%, p=0.03), and non-MEGA-PRESS studies remained heterogeneous (I2=94.7%, p<0.001). In the MEGA-PRESS subgroup, brain GSH was lower in both AD (SMD [95%CI] = -1.45 [-1.83, -1.06], z=7.41, p<0.001) (Figure 3) and MCI (-1.15 [-1.71, -0.59], z=4.0, p<0.001) groups (Figure 4). Subgroup analyses of different brain regions and use of creatine or water as the reference molecule did not significantly reduce heterogeneity in brain GSH measurements (data not shown), with the exception of the study by Marjanska et al. 2019, use of water as a reference molecule overlapped with MEGA-PRESS studies in AD and MCI (Table 1).
Blood GSH concentrations and investigating heterogeneity
Blood GSH was lower in AD (SMD [95%CI] = -1.10 [-1.58, -0.62], z=4.46, p<0.001) but not in MCI groups compared to controls (SMD [95%CI] = -0.70 [-1.84, 0.44], z=1.12, p=0.23). Significant heterogeneity was observed for both analyses (AD: I2=95.7%, p<0.001, MCI: I2=97.8%, p<0.001). In AD, both intracellular and extracellular blood GSH were lower (intracellular SMD [95%CI] = -1.65 [-2.37, -0.93], p<0.001; extracellular SMD [95%CI] = -0.86 [-1.49, -0.24], p=0.001) without reduced heterogeneity (AD intracellular: I2=93.9%, p<0.001; extracellular: I2=96.7%, p<0.001) (Figure 5). Intracellular GSH was lower in MCI (SMD [95%CI] = -0.66 [-1.11, -0.21], p=0.025) with reduced but still significant heterogeneity (MCI intracellular: I2=67.8%, p<0.025) (Figure 6). Subgroup analysis of GSH assay type did not significantly reduce heterogeneity in blood GSH measurements. Meta-regression showed that studies having a higher proportion of male participants reported greater decreases in GSH levels in AD compared to controls (p = 0.001, I2res = 96.2%, Radj2 = 39.4%) (Figure 7). Meta-regressions with mean age and MMSE scores did not significantly reduce heterogeneity (data not shown).
Effect of study bias, publication bias, and small-study effects
In all analyses, the pooled estimated SMDs for the subgroups of studies deemed to have low bias was within the 95% CI of the overall (Table 4), suggesting the impact of studies with higher bias was small. Publication bias was not detected by funnel plots, Egger’s or trim and fill tests in AD brain GSH literature and MCI blood GSH literature. However, Egger’s test detected significant risk of publication bias in MCI brain GSH literature (bias [95%CI] = -11.28 [-20.6, -1.95], p=0.03) and AD blood GSH literature (bias [95%CI] = -7.04 [-11.49, -2.95], p=0.003). Blood GSH remained lower in AD compared to controls after adjusting for potential publication bias using trim and fill (estimated SMD [95%CI] = -1.17 [-1.65, -0.71], p<0.001).