Blood metabolomic profiling reveals new targets in the management of psychological symptoms associated with alcohol use disorder

Reviewer #1 (Public Review):

Summary:

This work by Leclercq and colleagues performed metabolomics on biospecimens collected from 96 patients diagnosed with several types of alcohol use disorders (AUD). The authors discovered strong alterations in circulating glycerophospholipids, bile acids, and some gut microbe-derived metabolites in AUD patients compared to controls. An exciting part of this work is that metabolomics was also performed in frontal cortex of post-mortem brains and cerebrospinal fluid of heavy alcohol users, and some of the same metabolites were seen to be altered in the central nervous system. This is an important study that will form the basis for hypothesis generation around diet-microbe-host interactions in alcohol use disorder. The work is done in a highly rigorous manner, and the rigorously collected human samples are a clear strength of this work. Overall, many new insights may be gained by this work, and it is poised to have a high impact on the field.

Strengths:

(1) The rigorously collected patient-derived samples.

(2) There is high rigor in the metabolomics investigation.

(3) Statistical analyses are well-described and strong.

(4) An evident strength is the careful control of taking blood samples at the same time of the day to avoid alterations in meal- and circadian-related fluctuations in metabolites.

Weaknesses:

(1) Some validation in animal models of ethanol exposure compared to pair-fed controls would help strengthen causal relationships between metabolites and alterations in the CNS.

(2) The classification of "heavy alcohol users" based on autopsy reports may not be that accurate.

(3) The fact that most people with alcohol use disorder choose to drink over eating food, there needs to be some more discussion around how dietary intake (secondary to heavy drinking) most likely has a significant impact on the metabolome.

Reviewer #2 (Public Review):

The authors carried out the current studies with the justification that the biochemical mechanisms that lead to alcohol addiction are incompletely understood. The topic and question addressed here are impactful and indeed deserve further research. To this end, a metabolomics approach toward investigating the metabolic effects of alcohol use disorder and the effect of alcohol withdrawal in AUD subjects is valuable. However, it is primarily descriptive in nature, and these data alone do not meet the stated goal of investigating biochemical mechanisms of alcohol addiction. The current work's most significant limitation is the cross-sectional study design, though inadequate description and citation of the underlying methodological approaches also hampers interest.

Most of the data are cross-sectional in the study design, i.e., alcohol use disorder vs controls. However, it is well established that there is a high degree of interpersonal variation with metabolism, and further, there is somewhat high intra-personal variation in metabolism over time. This means that the relatively small cohort of subjects is unlikely to reflect the broader condition of interest (AUD/withdrawal). The authors report a comparison of a later time-point after alcohol withdrawal (T2) vs. the AUD condition. However, without replicative time points from the control subjects it is difficult to assess how much of these changes are due to withdrawal vs the intra-personal variation described above. Overall, there is not enough experimental context to interpret these findings into a biological understanding. For example, while several metabolites are linked with AUD and associated with microbiome or host metabolism based on existing literature, it's unclear from the current study what function these changes have concerning AUD, if any. The authors also argue that alcohol withdrawal shifts the AUD plasma metabolic fingerprint towards healthy controls (line 153). However, this is hard to assess based on the plots provided since the change in the direction of the orange data subset is considers AUD T2 vs T1. In contrast, AUD T2 vs Control would represent the claimed shift. To support these claims, the authors would better support their argument by showing this comparison as well as showing all experimental groups (including control subjects) in their multi-dimensional model (e.g., PCA). The authors attempt to extend the significance of their findings by assessing post-mortem brain tissues from AUD subjects; however, the finding that many of the metabolites changed in T2/T1 are also present in AUD brain tissues is interesting; however, not strongly supporting of the authors' claims that these metabolites are markers of AUD (line 173). Concerning the plasma cohort itself, it is unclear how the authors assessed for compliance with alcohol withdrawal or whether the subjects' blood-alcohol levels were independently verified.

The second area of concern is the need for more description of the analytical methodology, the lack of metabolite identification validation evidence, and related statistical questions. The authors cite reference #59 regarding the general methodology. However, this reference from their group is a tutorial/review/protocol-focused resource paper, and it is needs to be clarified how specific critical steps were actually applied to the current plasma study samples given the range of descriptions provided in the citations. The authors report a variety of interesting metabolites, including their primary fragment intensities, which are appreciated (Supplementary Table 3), but no MS2 matching scores are provided for level 2 or 3 hits. Further, level 1 hits under their definition are validated by an in-house standard, but no supporting data are provided besides this categorization. Finally, a common risk in such descriptive studies is finding spurious associations, especially considering many factors described in the current work. These include AUD, depression, anxiety, craving, withdrawal, etc. The authors describe the use of BH correction for multiple-hypothesis testing. However, this approach only accounts for the many possible metabolite association tests within each comparison (such as metabolites vs depression). It does not account for the multi-variate comparisons to the many behavior/clinical factors described above. The authors should employ one of several common strategies, such as linear mixed effects models, for these types of multi-variate assessments.