A-comprehensive-assessment-of-the-association-betw

Articles
A comprehensive assessment of the association between

common drugs and psychiatric disorders using Mendelian
randomization and real-world pharmacovigilance database
Zhuohui Chen,a,b,c,d,i Xiang Wang,b,i Ziwei Teng,e,f Jing Huang,e Jianzhong Mo,g Chunrun Qu,b Yinghua Wu,b Zhixiong Liu,b,d,j,∗∗∗
Fangkun Liu,a,b,d,h,j,∗∗ and Kun Xiaa,c,j,∗
a
MOE Key Laboratory of Pediatric Rare Diseases, Hengyang Medical School, University of South China, Hengyang, China
b
Department of Neurosurgery, Xiangya Hospital, Central South University, 87 Xiangya Road, Changsha, China
c
Center for Medical Genetics and Hunan Key Laboratory of Medical Genetics, School of Life Sciences, Central South University,
Changsha, China
d
Hypothalamic Pituitary Research Centre, Xiangya Hospital, Central South University, Changsha, China
e
National Clinical Research Center for Mental Disorders, Department of Psychiatry, China National Technology Institute on Mental
Disorders, The Second Xiangya Hospital of Central South University, Changsha, Hunan, China
f
Department of Psychiatry, Hunan Brain Hospital (Hunan Second People’s Hospital), Changsha, China
g
The Third Hospital of Changsha, Changsha, Hunan, China
h
National Clinical Research Center for Geriatric Disorders, Xiangya Hospital, Central South University, Changsha, China
Summary eBioMedicine
2024;107: 105314
Background Medications prescribed for chronic diseases can lead to short-term neuropsychiatric symptoms, but their
long-term effects on brain structures and psychiatric conditions remain unclear. Published Online 26
August 2024
https://doi.org/10.
Methods We comprehensively analyzed the FDA Adverse Event Reporting System database and conducted drug 1016/j.ebiom.2024.
target Mendelian Randomization (MR) studies on six categories of common drugs, 477 brain imaging-derived 105314
phenotypes (IDPs) and eight psychiatric disorders. Genetic instruments were extracted from expression
quantitative trait loci (eQTLs) in blood, brain, and other target tissues, protein quantitative trait loci (pQTLs) in
blood, and genome-wide association studies (GWAS) of hemoglobin and cholesterol. Summary statistics for brain
IDPs, psychiatric disorders, and gut microbiome were obtained from the BIG40, Psychiatric Genomics
Consortium, and MiBioGen. A two-step MR and mediation analysis were employed to screen possible mediators
of drug-IDP effects from 119 gut microbiota genera and identify their mediation proportions.
Findings Among 19 drug classes, six drugs were found to be associated with higher risks of psychiatric adverse events,
while 11 drugs were associated with higher risks of gastrointestinal adverse events in the FAERS analysis. We
identified ten drug-psychiatric disorder associations, 202 drug-IDP associations, 16 drug-microbiota associations, and
four drug-microbiota-IDP causal links. For example, PPARG activation mediated HbA1c reduction caused a higher
risk of bipolar disorder (BD) II. Genetically proxied GLP-1R agonists were significantly associated with an increase in
the volume of the CA3-head of the right hippocampus and the area of the left precuneus cortex, both of which have
been shown to correlate with cognition in previous studies.
Interpretation Common drugs may affect brain structure and risk of psychiatric disorder. Oral medications in
particular may exert some of these effects by influencing gut microbiota. This study calls for greater attention to be
paid to the neuropsychiatric adverse effects of drugs and encourages drug repurposing.
Funding National Natural Science Foundation of China (grant No. 82330035, 82130043, 82172685, and 82001223),
National Natural Science Foundation of Hunan Province (grant No. 2021SK1010), and the Science Foundation for
Distinguished Young Scholars of Changsha (grant No. kq2209006).
*Corresponding author. MOE Key Laboratory of Pediatric Rare Diseases, Hengyang Medical School, University of South China, Hengyang, China;
Center for Medical Genetics and Hunan Key Laboratory of Medical Genetics, School of Life Sciences, Central South University, Changsha, China.
**Corresponding author. MOE Key Laboratory of Pediatric Rare Diseases, Hengyang Medical School, University of South China, Hengyang, China;
Department of Neurosurgery, Xiangya Hospital, Central South University, 87 Xiangya Road, Changsha, China.
***Corresponding author. Department of Neurosurgery, Xiangya Hospital, Central South University, 87 Xiangya Road, Changsha, China.
E-mail addresses: xiakun@sklmg.edu.cn (K. Xia), liufangkun@csu.edu.cn (F. Liu), zhixiongliu@csu.edu.cn (Z. Liu).
i
These authors have contributed equally to this work and share co-first authorship.
j
These authors have contributed equally to this work and share co-corresponding authorship.
www.thelancet.com Vol 107 September, 2024 1

Articles
Copyright © 2024 The Author(s). Published by Elsevier B.V. This is an open access article under the CC BY-NC
license (http://creativecommons.org/licenses/by-nc/4.0/).
Keywords: Pharmacovigilance; Mendelian randomization; Common drugs; Psychiatric disorders; Brain structure;
Gut microbiome
Research in context
Evidence before this study pharmacovigilance database. This study allows a better
Previous studies and data from pharmacovigilance databases understanding of how medications can affect our brains and
suggested that some drugs used for chronic diseases were mental health in the long term. It also showed how the gut
associated with short-term neuropsychiatric symptoms. microbiome can play a role in this process, especially when
many common drugs are administrated orally.
Added value of this study
Through drug target Mendelian Randomization, we Implications of all the available evidence
discovered ten drug-psychiatric disorder associations, 202 This study is a great reminder to pay more attention to the
drug-IDP associations, 16 drug-microbiota associations, and effects of common drugs on our brains and hopefully
four drug-microbiota-IDP causal links. Some of the effects encourages the repurposing of old drugs for new psychiatric
were also observed in real-world data from FAERS, a conditions.
Introduction researchers to assess the potential effects of drugs on

Chronic diseases, such as diabetes, cardiovascular dis- various outcomes.
eases, and lung diseases, require prolonged medication. Further, how may oral drugs impact brain structures
Many drugs used for these diseases have diverse de- and diseases? In addition to direct ways such as crossing
grees of neurological adverse reactions,1,2 while a few the blood–brain barrier, affecting target gene expression,
medications may have potential protective effects for and activating signalling pathways in the neurons,12 the
neuropsychiatric conditions.3–9 For instance, treatment gut microbiota may serve as a key mediator linking oral
with liraglutide, a glucagon-like peptide-1 receptor medication use with brain structure and disease. Firstly,
(GLP-1R) agonist, enhanced associative learning and pharmacomicrobiomics has revealed that drugs can
memory in individuals with obesity.10 Nevertheless, pre- impact the composition of gut microbiota through direct
marketing clinical trials of drugs often evaluate only microbial killing, modulation of host immune reaction,
short-term neuropsychiatric symptoms, leaving the and alterations in intestinal pH.13–15 Secondly, gut micro-
long-term effects of drugs on brain structures and biota can influence brain function via the microbiota-gut-
associated psychiatric conditions unknown. Addressing brain axis, producing tryptophan metabolites and other
these gaps can significantly enhance clinical practice by neurotransmitters.16,17 Based on these established re-
improving patient care, facilitating early screening for lationships, we hypothesize that common drugs, most of
drug-related neuropsychiatric adverse effects, aiding in which are orally administrated, can affect brain structure
decision-making when prescribing medications, and partially by altering the gut microbiome, ultimately
increasing public awareness of the long-term impacts of resulting in altered risks of psychiatric disorders.
common drugs. The primary goal of this study is to investigate the
Mendelian randomization (MR) is an approach to causal effect of six categories of common drugs on 477
uncover causal relationships by employing outputs from brain imaging-derived phenotypes (IDPs) and eight
genome-wide association studies (GWAS) devoid of psychiatric disorders using MR. Pharmacovigilance
confounding variables. In essence, genetic variants, analysis of the FDA Adverse Event Reporting System
usually single nucleotide polymorphism (SNP), that are (FAERS) database was conducted to support the MR
strongly associated with the exposure are selected as findings. Through a two-step MR technique, 119 genera
instrumental variables (IVs), and the causal effect of the of gut microbiota were screened to identify possible
exposure on the outcome is thus estimated by the effect mediators, followed by mediation analysis to determine
of IVs on the outcome.11 By utilizing cis-expression the proportion mediated by a specific genus. A brief
quantitative trait loci (cis-eQTL) or SNPs linked with overview of this study is provided in Fig. 1. This study
the drug target effect as IVs, it is also feasible to evaluate can uncover long-term neuropsychiatric effects of
the impact of drugs on diseases or phenotypes, usually common medications, provide guidance on the appro-
referred to as drug target MR. Following the general priate usage of medication for patients afflicted by both
concepts of Mendelian randomization, drug target MR chronic disease and poor mental health, and explore
selected cis-variants that are associated with the drug repurposing of marketed drugs for the treatment of
target effect to proxy a certain drug class, allowing psychiatric disorders.
2 www.thelancet.com Vol 107 September, 2024

Articles
Fig. 1: Study workflow. Abbreviations: AN, anorexia nervosa; ANX, anxiety disorders; BD, bipolar disorder; COPD, chronic obstructive pulmonary
disease; eQTL, expression quantitative trait loci; FA, fractional anisotropy; IDP, imaging-derived phenotype; IV, instrumental variable; IVW,
inverse variance-weighted; LD, linkage disequilibrium; LDL, low-density lipoprotein; MAF, minor allele frequency; MD, mean diffusivity; MDD,
major depressive disorder; MR, Mendelian randomization; MVMR, multivariable MR; NSAIDs, nonsteroidal anti-inflammatory drugs; OCD,
obsessive-compulsive disorder; PD, panic disorder; PTSD, post-traumatic stress disorder; SCZ, schizophrenia; SMR, summary-data-based Men-
delian randomization.

Articles
Methods the same CASEID; 3) retain the reports with the largest
Study overview and data sources PRIMARYID value for reports with the same CASEID
An overview of the study workflow is provided in Fig. 1. and FDA_DT. We deleted all the irrelevant events
The study was a two-sample drug target MR study sup- including product issues, medication errors (adminis-
plemented by pharmacovigilance analysis of the FDA tration, confusion, monitoring, etc.), and social circum-
Adverse Event Reporting System (FAERS) database. We stances, as listed in eTable 4. Demographic and clinical
incorporated six categories of common drugs used for information of all the analyzed reports was summarized
chronic diseases, including antidiabetic drugs, antihy- by each drug target. To compare the risk of AEs between
pertensive agents, lipid-lowering drugs, anti-stroke agents the drug of interest and other drugs in the database, a
(anticoagulants and antiplatelet agents), chronic obstruc- two-by-two contingency table was constructed (eTable 5)
tive pulmonary disease (COPD) medications, and and a disproportionality analysis was conducted using the
nonsteroidal anti-inflammatory drugs (NSAIDs). Detailed reporting odds ratio (ROR) and proportional reporting
drug target information was listed in eTable 1. This study ratio (PRR). ROR, PRR and their 95% confidence in-
is reported according to the Strengthening the reporting tervals (CIs) were calculated using the following formula:
of observational studies in epidemiology using mende-
lian randomisation (STROBE-MR) guidelines.18 a/c ad
ROR = =
Detailed information on all GWAS data sources is b/d bc
listed in eTable 2a. IDP GWAS summary statistics were
obtained from the Oxford Brain Imaging Genetics Server √̅̅̅̅̅̅̅̅̅̅̅̅
̅
(BIG40, https://open.win.ox.ac.uk/ukbiobank/big40/). 95% CI of ROR = eln(ROR) ± 1.96 a+b+c +d
1 1 1 1
The original study analyzed brain imaging data of 39,691

participants from UK biobank.19 We selected 477 IDPs
from the original set of 3935, including 68 cortical area a/(a + b)
metrics, 68 cortical thickness metrics, 66 cortical volume PRR =
c /(c + d)
metrics, 179 subcortical volume measures, and 96 white
matter tract measures, with detailed information pro-
vided in eTable 2b. GWAS summary data of eight psy- √̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅
̅
95% CI of PRR = eln(PRR) ± 1.96 a−a+b+c −c+d
1 1 1 1
chiatric disorders were acquired mainly through the

Psychiatric Genomics Consortium (PGC, https://pgc.
AE signals were regarded as valid when the number
unc.edu/), with sample sizes ranging from ∼10,000 to
of reports was no less than three and the lower threshold
∼500,000. Microbiome quantitative trait loci (mbQTLs)
of the 95% CI of the ROR exceeded one.22 Robust sig-
were obtained from MiBioGen (https://mibiogen.gcc.
nals were defined as valid signals meeting additional
rug.nl/). The original study involved an analysis of ge-
criteria of a PRR of at least 2 and a chi-squared value of
notypes and fecal 16 S microbiome data from 18,340
at least 4.23 For each drug target, we reported the ROR
participants across 24 cohorts.20 We only included 119
and PRR of reports with SOC (System Organ Classes)
known genera for MR analysis as listed in eTable 2c.
names of ‘Psychiatric disorders’ and ‘Gastrointestinal
GWAS data from other studies such as the Megastroke
disorders’. Additionally, we reported five PTs (Preferred
project21 were used for positive control analysis.
Terms) with the highest RORs in the SOC of ‘Psychi-
atric disorders’ and ‘Gastrointestinal disorders’.
Psychiatric, and digestive adverse events of Within the dataset analyzed, over 80% of cases had at
common drugs: a pharmacovigilance analysis of the least one missing value of clinical characteristics. Direct
FAERS database deletion of all these cases could potentially reduce statis-
We conducted a pharmacovigilance investigation on the
tical power and introduce new biases, such as reports from
psychiatric and digestive adverse effects of the drugs of
developed countries being more standardized and com-
interest based on the FAERS database, a large public
plete compared to those from developing countries. To
database of AEs reported by healthcare workers, phar-
address this, we performed sensitivity analyses by
macists and consumers worldwide. We selected 130
sequentially deleting cases with missing values for each
FDA-approved drugs with oral products available as
clinical characteristic. Specifically, sensitivity analyses were
drugs of interest, with detailed information listed in
conducted for sex, age, weight, and country to determine if
eTable 3a and b. We analyzed adverse events (AEs) that
the cases with missing values will greatly affect the results.
listed the drugs of interest as PS (Primary Suspect Drug)
and were reported by MD (Physician), PH (Pharmacist),
or OT (Other health-professional) from January 2004 to The effect of common drugs on IDP, psychiatric
March 2024. Duplicate reports were removed according disorders, and gut microbiome
to the following procedure: 1) sort the reports in the order Selection of IVs
of CASEID, FDA_DT, and PRIMARYID; 2) retain the MR analysis relies on three basic assumptions to make
reports with the largest FDA_DT value for reports with the causal inference: 1) the genetic variant is strongly

Articles
associated with the exposure; 2) the genetic variant is for antidiabetic drugs, we included glucose and type 2
not associated with the outcome via a confounding diabetes mellitus (T2D) as the outcome; for lipid-
pathway; and 3) the genetic variant does not directly lowering drugs, we incorporated coronary artery dis-
affect the outcome.24 Based on these assumptions, we ease (CAD) as the outcome; for anti-stroke agents, we
used two methods to select IVs to proxy the drug effect, included any stroke (AS), any ischemic stroke (AIS),
as described in previous MR studies.25 large artery stroke (LAS), cardioembolic stroke (CES),
For the first method, we selected SNPs that meet and small vessel stroke (SVS) as the outcome; for
the following requirements: 1) associated with the drug COPD medications, we included forced expiratory
effect: SNPs associated with glycated hemoglobin volume in 1 s/forced vital capacity (FEV1/FVC) ratio
(HbA1c) [false discovery rate (FDR) < 0.05] for and percentage of predicted forced expiratory volume
antidiabetic drugs; SNPs associated with low-density in the first second of expiration (FEV1% predicted) as
lipoprotein cholesterol (LDL-C) (P-value < 5e-8) for the outcome.
lipid-lowering drugs; protein quantitative trait loci For the second approach, we extracted common
(pQTLs), i.e., SNPs associated with plasma protein (MAF >0.01) expression quantitative trait loci (eQTLs)
level of F10 (P-value < 1e-5) and F2 (FDR < 0.05) for associated with the expression of drug target genes
anticoagulants; pQTLs of PDE4A (P-value < 1e-5) for (P-value < 5e-8) from eQTLGen Consortium, BrainMeta
COPD medications; 2) located within ±100 kb windows v2, and GTEx V8, with detailed information listed in
from the drug target gene region; 3) minor allele fre- eTable 2a. To reduce potential pleiotropy, we only
quency (MAF) > 0.01. As for the first two requirements, included cis-eQTLs as IVs, defined as eQTLs located
we intend to obtain SNPs that best mimic the effects of within ±1 Mb range from the encoded gene. The pri-
drugs on the human body as drug target IVs. There- oritization order of different database sources for
fore, we preferentially select SNPs that are associated selecting IVs is determined as follows. Generally, since
with the target effects (e.g., blood glucose reduction for we are interested in neuropsychiatric effects of common
antidiabetic drugs), and are located within certain drugs and gene expressions in the brain seems to more
range around the target gene region. No SNPs meeting directly affect psychiatric disorders and brain structures
the above criteria for antihypertensive agents, antico- than gene expressions in the blood or other tissues,
agulants, and NSAIDs were identified. However, using brain eQTLs as IVs may be more appropriate.
inspired by the work of Harshfield et al.,26 we realized However, the sample size of BrainMeta v2 is relatively
that coagulation Factor II and Factor Xa protein small (around 2400) compared with eQTLGen (around
expression could be used as a proxy for coagulation 31,000) and relevant studies have demonstrated a high
function, which is the target effect of thrombin in- correlation of cis-eQTLs across different tissues, espe-
hibitors and Factor Xa inhibitors. In other words, the cially between brain and blood.32 Therefore, we prefer-
effects of thrombin inhibitors and Factor Xa inhibitors, entially select blood eQTLs from eQTLGen. For other
are more direct without involving other mediating drug targets without cis-eQTLs that meet the re-
mechanisms. Under such circumstances, we used cis- quirements in the eQTLGen Consortium, we extract
pQTLs of F2, F10 and PDE4A as drug target IVs. As for eQTLs in brain (from BrainMeta v2), and subsequently
the P value threshold, we start from the most stringent in corresponding target tissues (from GTEx V8) if there
threshold (P < 5e-8) and gradually loosen the criteria to are no qualified cis-eQTLs in BrainMeta v2, since these
false discovery rate (FDR) < 0.05 or P < 1e-5, which eQTLs can reflect the major target effects of common
were both used in previous publications.27,28 Our pur- drugs. Specifically, we extracted eQTLs in skeletal
pose was to obtain an adequate number of IVs so that muscle for CACNG1, eQTLs in tibial artery for CAC-
the MR results will be more statistically powerful. NA1S, and eQTLs in subcutaneous adipose for NPC1L1.
Subsequent linkage disequilibrium (LD) clumping was Although these drugs have more than one target tissues,
conducted based on LD r2 < 0.3 and a clumping win- the other target tissues have no qualified eQTLs. We
dow of 500 kb, utilizing the 1000 genomes reference investigated genetic associations between selected IVs
panel (“EUR”). The clumping threshold was chosen to and drug indications or drug effects as positive control
balance the statistical power and the stability of drug analyses. Specifically, we incorporated Fasting glucose
target MR. Specifically, a loose LD r2 threshold reduces (FG), 2 h-glucose post–challenge (2hGlu), Glycated he-
the reliability of MR results, while a stringent LD r2 moglobin (HbA1c), and Fasting insulin (FI) as the
threshold reduces the statistical power.29 Therefore, we outcome for antidiabetic drugs; systolic blood pressure
adopted the clumping threshold of LD r2 < 0.3, a (SBP) and diastolic blood pressure (DBP) as the
widely used threshold in drug target MR.25,30,31 Expo- outcome for antihypertensive agents; five stroke types
sure and outcome data were harmonized and palin- (AS, AIS, LAS, CES, and SVS) as the outcome for anti-
dromic SNPs with intermediate allele frequencies were stroke agents; FEV1/FVC ratio as the outcome for
eliminated. To confirm the validity of IVs, positive COPD medications; pain in joint and low back pain as
control analyses were performed for each drug class: the outcome for NSAIDs.

Articles
To validate the effectiveness of the IVs, the F-statistic MR-Egger method remains efficacious even when all
was computed for each SNP using the provided for- IVs are invalid, but is based on a weaker assumption
mula33: termed InSIDE (Instrument Strength Independent of
Direct Effect).37 The slope coefficient and the intercept
R2 n−k−1 estimate obtained through the MR-Egger regression
F= ×
1−R2 k were utilized to estimate the causal effects and identify
potential pleiotropy, respectively. Results with an MR-
Egger intercept significantly differing from zero
(P-value <0.05) were deemed unreliable and biased due
β2 to pleiotropy. Additionally, we drew funnel plots to
R2 =
β +se2 ×n
2 assess heterogeneity across genetic variants visually and
calculated Cochran’s Q statistics from the IVW. I2 was
where n = sample size, k = number of IVs, calculated using the formula I2 = (Q-df)/Q × 100%,
R2 = explained variance of genetic instruments on where Q = Cochran’s Q statistics and df = degree of
exposure, β = effect size of SNPs, and se = standard error freedom; if I2 < 0, it was set to 0. To verify if the IVW
of effect size. results were affected by a genetic variant, a leave-one-out
We ensured that the F-statistics of all the IVs used analysis was performed. To identify potential outlier
for the MR analysis exceeded 10 to prevent any viola- SNPs, MR pleiotropy residual sum and outlier
tions of Assumption 1 of mendelian randomization (MR-PRESSO) were conducted.38 In cases where the
analysis.34 MR-PRESSO global test P-value was less than 0.05, IVW
was re-run after excluding outlier SNPs.
Primary MR analysis Additionally, several other approaches were lever-
The inverse variance-weighted (IVW) method with a aged to test the MR assumptions. MR Steiger test was
multiplicative random-effects model was applied for performed to determine whether the IVs were more
the first set of IVs. This method meta-analyzes the associated with the outcome than with the exposure.39
Wald ratios of each IVs to obtain an overall estimate, False directionality was considered present when ‘stei-
which has the highest statistical power compared with ger_dir’ is FALSE and ‘steiger_pval’ is less than 0.05.
other MR methods.35 Causal estimates were reported Under such circumstances, we re-performed IVW-MR
per 1% increase of HbA1c for antidiabetic drugs, per after removing IVs with false directionality. Moreover,
1 mmol/L increase of LDL cholesterol for lipid- to test the independence assumption, we searched for
lowering drugs, and per one standard deviation (SD) evidences that suggest the associations between the IVs
increase of protein expression for anti-stroke agents and potential confounders. Confounders are any factors
and COPD medications. that are linked with both the exposure and the outcome,
Summary-data-based MR (SMR) was employed for which may bias the causal inference. However, it is
the second set of IVs (eQTLs extracted from different difficult to identify all potential confounders, and hence
sources). This method first identifies the ‘top SNP’ that we chose four major confounders that have been proven
has the most significant associations with mRNA to be associated with psychiatric disorders based on
expression of exposure genes, and then uses the top established work, including smoking, alcohol con-
SNP to analyze the genetic associations between mRNA sumption, education attainment, and income.40–43
expressions and outcome traits. SMR is usually com- We queried three databases for associations regarding
bined with heterogeneity in dependent instruments all the IVs used in the IVW-MR analysis with a P value
(HEIDI) analysis to eliminate false positive results due threshold of < 5e-8, including the Genome-Wide Re-
to linkage disequilibrium (LD) between the Top SNP pository of Associations Between SNPs and Phenotypes
and other SNPs associated with the outcome, which will (GRASP)44 (https://grasp.nhlbi.nih.gov/Search.aspx),
be discussed later.36 Causal estimates were reported per the MRC IEU OpenGWAS database45,46 (https://gwas.
one SD increase of mRNA expression of target genes. mrcieu.ac.uk/), and the GWAS Catalog47 (https://www.
FDR correction was utilized to account for multiple ebi.ac.uk/gwas/). Queries in the GRASP were manu-
tests. Findings with FDR values less than 0.05 were ally done while searches in the other two databases were
viewed as significant, while those greater than 0.05 but through their REST API, with resources provided at the
with P-values less than 0.05 were considered nominally end of Methods. Subsequently, we re-performed IVW-
significant. MR after removing IVs associated with the four con-
founders. Multivariable MR (MVMR) is a MR method
Assessment of the MR assumptions and sensitivity analysis that can investigate the effects of multiple exposures on
For IVW-MR method, we computed the F-statistic as the outcome simultaneously, which can be used to
previously stated to assess the relevance assumption. To adjust for potential confounders, or calculating direct
examine the robustness of the IVW results, we utilized effects in mediation analysis.48–50 Eventually, multivari-
the MR-Egger method for sensitivity analysis. The able MR (MVMR) were conducted for significant

Articles
associations between drugs and psychiatric disorder/ Assessment of the MR assumptions and sensitivity analysis
IDP/gut microbiota to adjust for the four confounders, Similar to the former process, we computed the F-sta-
with similar processes as univariate MR. We retrieved tistics and incorporated the weighted median, weighted
summary statistics of the four confounders using IVs mode, and MR-Egger methods for sensitivity analysis.
selected in the former univariate MR, with detailed The weighted median method can generate a reliable
GWAS information listed in eTable 2a. Then exposure estimate when less than half of the IVs are invalid.55 The
data were combined and harmonized with outcome weighted mode method is useful when over half of the
data, and MVMR was performed using the IVW IVs are invalid, and depends on the zero-modal pleiot-
method. Furthermore, since we used GWAS summary ropy assumption (ZEMPA). This assumption implies
statistics of HbA1c to proxy the target effects of antidi- that, among all the causal estimates obtained from each
abetic drugs, and HbA1c is also related to red blood cell instrument, the most frequent estimate is a robust es-
(RBC) count, RBC count is likely to be a confounding timate of the true causal effect.56 For forward MR, re-
factor due to its relationship with psychiatric disorders.51 sults with IVW P-value <0.05 and one of the additional
Therefore, for antidiabetic drugs we performed addi- three methods P-value <0.05 were considered signifi-
tional MVMR adjusting for RBC count. Although the cant. For reverse MR, results with IVW P-value <0.05
lipid-lowering drugs included in our study primarily were considered significant. Results with an Egger
reduce LDL-C, some of them may affect other lipid intercept P-value <0.05 were excluded from further
components like HDL-C and triglycerides (TG) as analysis due to pleiotropy bias. To assess heterogeneity,
well.52–54 To explore potential pathways other than we computed Cochran’s Q statistics and I2 from IVW
reducing LDL-C, we performed MVMR for formerly and used funnel and leave-one-out plots for visual ex-
identified significant associations between lipid- amination. MR-PRESSO was conducted and if the
lowering drugs and psychiatric disorder/IDP/gut global P-value of the MR-PRESSO test was less than
microbiota, incorporating LDL-C, HDL-C, and TG as 0.05, IVW was repeated after excluding outlier SNPs.
exposures. In MVMR analysis, we considered results Additional sensitivity analysis was performed by
with a P value of <0.05 as significant. removing IVs with false directionality identified by the
For the SMR method, the F-statistics of the top SNP Steiger test and subsequently removing IVs associated
were calculated using the above formula. HEIDI anal- with confounders, as described earlier.
ysis was performed, which incorporates significant
SNPs other than the top SNP in the cis-eQTL region to Mediation analysis
test whether genetic associations were due to linkage We then performed a mediation analysis to evaluate the
disequilibrium (LD). Results with HEIDI P-value <0.05 proportion of drug-IDP effect mediated by gut micro-
were considered due to linkage rather than functional biota. The direct effect was the IVW estimate of drug-
association. IDP associations. The indirect effect was calculated by
the product of drug-microbiome IVW causal estimate
Bidirectional MR testing the effect of gut and microbiome-IDP IVW causal estimate. The media-
microbiome on IDPs tion proportion was then derived by dividing the indirect
Selection of IVs effect by the direct effect. The point estimate and 95%
In the forward MR, gut microbiome was included as CI were obtained using the Monte Carlo method.
exposures and IDPs were included as outcomes, while
in the reverse MR, IDPs were seen as exposures and gut Replication analyses based on UK Biobank research
microbiome was seen as outcomes. SNPs that meet the We replicated the IVW-MR analysis of drugs on psy-
following requirements were selected as IVs: 1) associ- chiatric disorders using different GWAS data sources
ated with the gut microbiome for forward MR (P-value < listed in eTable 2a, applying a stringent P-value
1e-5) or IDPs for reverse MR (P-value < 5e-8); 2) threshold for IV selection. For antidiabetic drugs, we
mutually independent (LD r2 < 0.001) and clumping used a recent GWAS study of HbA1c involving 338,919
window of 10 Mb using the “EUR” panel; and 3) MAF UK Biobank participants.57 For anticoagulants, we uti-
>0.01. Exposure and outcome data were harmonized lized the pQTL data from the UK Biobank Pharma
and palindromic SNPs with intermediate allele fre- Proteomics Project (UKB-PPP) including 34,557 Euro-
quencies were eliminated. The F-statistic was computed pean participants.58 The P-value threshold was set to 5e-
for each SNP using the formula described above. 8. Other processes, including subsequent IV filtering,
positive control analysis, and IVW-MR analysis were
Primary MR analysis conducted as previously described.
The IVW method with a multiplicative random-effects
model was applied for the primary MR analysis. Software and packages
Causal estimates were reported per one SD increase in All the MR analyses, including sensitivity analysis, were
bacterial abundance for the forward MR, and per one performed via either ‘TwoSampleMR’ package (v0.5.6)
SD increase in the phenotype for the reverse MR. in R software 4.2.3 (https://www.r-project.org/) or SMR

Articles
software (v1.3.1)36 (https://yanglab.westlake.edu.cn/ For each drug class, five PTs (Preferred Terms)
software/smr/#Download). Queries for the associa- belonging to SOCs (System Organ Classes) of ‘Psychi-
tions of IVs with confounders were performed using the atric disorders’ with the highest ROR were listed in
‘tophits’ function implemented in the ‘ieugwasr’ pack- eTable 8. As shown in the ROR heatmap (Fig. 2a),
age (v1.0.0) and the ‘fromJSON’ function in the ‘jsonlite’ leukotriene receptor antagonists were associated with an
package (v1.8.8) to access the REST API of the IEU increased risk of multiple psychiatric PTs, with the
OpenGWAS and the GWAS Catalog, respectively. highest effects on ‘Separation anxiety disorder’
Cortical structures were visualized with the BrainNet (ROR = 1019.07 [432.75, 2399.79]).
Viewer59 (http://www.nitrc.org/projects/bnv/). Subcor- In sensitivity analyses, most results still remain
tical structures were drawn using the ‘ggseg’ package stable after deleting cases with missing age, country or
(v1.6.6). White matter tracts were visualized via the sex, respectively. For example, all of the valid signals in
‘ggsegICBM’ package (v1.0.1) and ‘ggseg3d’ package the SOC level still remained significant, and 76.1%,
(v1.6.3). FDR values were calculated using the 98.5%, and 89.6% of the top five PTs still remained in
Benjamini-Hochberg method via the ‘p.adjust’ function the top five list after deleting the cases missing age,
implemented in the ‘stats’ package (v4.2.3). The Monte country or sex, respectively. However, since the num-
Carlo method for the calculation of the point estimate ber of cases missing weight is relatively large, only 50%
and 95% CI in mediation analysis was implemented by of the valid signals in the SOC level and 38.8% of the
the ‘ci’ function of the ‘RMediation’ package (v1.2.2). top five PTs still remained after deleting the cases
missing weight.
Ethics
All MR analyses were conducted using publicly available The effect of common drugs on psychiatric
GWAS datasets, with detailed information provided in disorders
eTable 2a. All these GWAS studies have received ethical Since all drugs under investigation have specific target
approval and the participants have provided informed genes, it’s feasible to proxy the drug effects using cis-
consent. No individual-level data were utilized in this genetical variants. To further decipher the long-term
study, thus eliminating the need for new ethical review effect of common drugs on psychiatric disorders, we
board approval. conducted drug target MR using two different methods,
inverse variance-weighted MR (IVW-MR) and summary-
Role of funders data-based MR (SMR).
The funders had no role in the design and conduct of
the study; collection, management, analysis, and inter- IV selection and validation
pretation of the data; preparation, review, or approval of As shown in eTable 9, the minimum F-statistic of the
the manuscript; and decision to submit the manuscript IVs for IVW-MR was 13, suggesting a low probability of
for publication. weak instrument bias. In positive control analyses
(eTable 10a), most IVs yielded significant results, indi-
cating the reliability of the IVs. For instance, GLP1R-
Results mediated HbA1c increase (per one percent increase)
Discovery of psychiatric adverse effects of common was significantly associated with a higher risk of type 2
drugs through analysis of the FAERS database diabetes mellitus (T2D) (odds ratio [OR] = 78.18 [9.24,
We first carried out a retrospective analysis of the psy- 661.74], P = 6.33e-5, random-effect inverse variance-
chiatric adverse effects of common drugs using the weighted [RE-IVW]).
FAERS database. We searched for adverse event (AE) For the SMR method, the minimum F-statistic of the
records concerning the drugs under investigation and top SNP was 31, indicating strong associations between
found that most products documented in the database the IVs and gene expression (eTable 10b). However, in
were administered orally. Therefore, only orally the positive control analysis (eTable 10b), only 18.7%
administered medications were included to eliminate (17/91) results yielded significant results (p_SMR <0.05
adverse reactions caused by the modes of administra- and p_HEIDI >0.05), among which nine results yielded
tion, such as skin reactions from injections. Clinical estimates that were directionally correct.
characteristics of psychiatric AE reports, including sex,
age, weight, country, reporter type, and patient outcome Main findings
were summarized in eTable 6. Among 19 drug classes IVW-MR identified seven significant associations be-
analyzed, six drugs were significantly associated with tween four drugs and four psychiatric disorders (Fig. 2b
higher psychiatric AEs based on the reporting odds ratio and eTable 11). PPARG-mediated HbA1c increase was
(ROR) criteria, with a robust signal detected for leuko- associated with a significantly lower risk of bipolar dis-
triene receptor antagonists (Table 1). Leukotriene re- order II (BD II) (OR = 0.02 [0.00, 0.15], P = 1.75e-04, RE-
ceptor antagonists have the highest effects on IVW). HMGCR-mediated LDL cholesterol increase was
psychiatric AEs (ROR [95% CI] = 7.37 [7.16, 7.6]). associated with a lower risk of major depressive disorder

Articles
Drug category Drug target Number of Total number Percentage of ROR (95% CI) P_value_ROR PRR (95% CI) χ2 P_value_PRR
psychiatric of AEs (a+b) psychiatric AEs
AEs (a) [a/(a+b)]
Antidiabetic drugs ABCC8+KCNJ11* 522 6940 7.52% 1.51 (1.38, 1.65) 1.56E-19 1.47 (1.39, 1.56) 83.8 3.81E-39
Antidiabetic drugs DPP4 647 26,378 2.45% 0.47 (0.43, 0.51) 2.11E-67 0.48 (0.4, 0.56) 383.3 1.22E-17
Antidiabetic drugs GLP1R 1218 50,432 2.42% 0.46 (0.43, 0.49) 4.02E-120 0.47 (0.42, 0.53) 754.7 4.40E-37
Antidiabetic drugs PPARG 197 53,836 0.37% 0.07 (0.06, 0.08) 1.68E-287 0.07 (0, 0.21) 2502.0 1
Antidiabetic drugs SLC5A2 1057 67,339 1.57% 0.3 (0.28, 0.31) 0 0.31 (0.25, 0.37) 1742.5 1.13E-31
Antihypertensive agents ACE 3457 87,128 3.97% 0.77 (0.74, 0.79) 2.42E-55 0.78 (0.74, 0.81) 233.2 4.47E-27
Antihypertensive agents ADRB1* 7455 105,052 7.10% 1.42 (1.39, 1.46) 3.11E-172 1.39 (1.37, 1.42) 865.9 5.54E-284
Antihypertensive agents AGTR1 2578 79,569 3.24% 0.62 (0.6, 0.65) 3.28E-121 0.63 (0.6, 0.67) 573.1 1.54E-60
Antihypertensive agents CCB* 9489 100,768 9.42% 1.94 (1.9, 1.98) 0 1.85 (1.83, 1.87) 3889.9 0
Antihypertensive agents SLC12A3 796 18,701 4.26% 0.83 (0.77, 0.89) 4.58E-07 0.83 (0.77, 0.9) 27.8 2.84E-06
Lipid-lowering drugs HMGCR 5766 137,391 4.20% 0.81 (0.79, 0.84) 2.69E-41 0.82 (0.8, 0.85) 235.0 1.09E-37
Lipid-lowering drugs NPC1L1 271 8643 3.14% 0.6 (0.53, 0.68) 9.36E-16 0.61 (0.5, 0.73) 69.2 3.05E-07
Anti-stroke agents F10 2062 170,257 1.21% 0.23 (0.22, 0.24) 0 0.24 (0.19, 0.28) 5373.3 3.50E-47
Anti-stroke agents F2 473 48,321 0.98% 0.18 (0.17, 0.2) 0 0.19 (0.1, 0.28) 1701.8 2.57E-10
Anti-stroke agents P2RY12 1687 106,795 1.58% 0.3 (0.28, 0.31) 0 0.31 (0.26, 0.36) 2753.5 3.39E-45
COPD medications ALOX5 6 145 4.14% 0.8 (0.35, 1.82) 5.96E-01 0.81 (0.03, 1.59) 0.3 8.35E-01
COPD medications CYSLTR1** 6007 21,230 28.29% 7.37 (7.16, 7.6) 0 5.57 (5.55, 5.59) 23600.1 0
COPD medications PDE4* 599 6018 9.95% 2.06 (1.89, 2.24) 2.00E-62 1.95 (1.87, 2.03) 292.3 3.96E-223
NSAIDs PTGS1+PTGS2* 16,523 305,619 5.41% 1.06 (1.05, 1.08) 5.14E-16 1.06 (1.05, 1.08) 58.7 5.14E-16
Disproportionality analysis were performed by calculating the ROR and PRR. Significant adverse reaction signals (the number of reports was not less than three and the lower limit of the 95% CI of the ROR
exceeds one) were marked with an asterisk (*). Robust signals (meeting additional criteria of a PRR of at least 2 and a chi-squared value of at least 4) were marked with two asterisks (**).
Table 1: ROR and PRR of psychiatric AEs among 19 drugs.
(MDD) (OR = 0.92 [0.86, 0.97], P = 3.98e-03, RE-IVW), a of PPARG on BD II by drinking and income, 77.8% of
lower risk of obsessive-compulsive disorder (OCD) direct effects remained significant after adjustment.
(OR = 0.44 [0.25, 0.77], P = 4.02e-03, RE-IVW), and a Among the three antidiabetic drugs, only PPARG ago-
higher risk of panic disorder (PD) (OR = 3.05 [1.73, nists have more than one SNP as drug target IVs, hence
5.35], P = 1.06e-04, RE-IVW). PCSK9-mediated LDL we performed MVMR for PPARG agonists adjusting for
cholesterol increase was associated with a lower risk of red blood cell (RBC) count. As shown in eTable 14b,
MDD (OR = 0.89 [0.85, 0.94], P = 5.42e-06, RE-IVW). there are no significant changes after adjusting for RBC
Genetically determined F10 protein expression was count, suggesting the robustness of our results. To
associated with a higher risk of BD I (OR = 1.14 [1.05, explore potential mechanisms behind these associa-
1.25], P = 2.57e-03, RE-IVW) and BD II (OR = 1.37 [1.18, tions, we performed MVMR for lipid-lowering drugs
1.59], P = 4.95e-05, RE-IVW). incorporating LDL-C, HDL-C, and triglycerides (TG). As
In sensitivity analyses, MR Egger produced one sig- demonstrated in eTable 14c, the effect of HMGCR
nificant result (P < 0.05, MR Egger) out of the seven mediated LDL decrease on OCD became insignificant
associations identified by IVW, which was in a consis- after adjustment while the effect of HDL on OCD ob-
tent direction as that from IVW-MR. Other sensitivity tained significant results, suggesting that the effect of
analyses revealed no abnormalities (eTable 11). The HMGCR inhibitors on OCD may be mediated through
scatter plots of the seven significant results are shown in TG reduction rather than LDL reduction.
eFig. 1. We did not observe significant asymmetry in SMR analyses revealed three significant associations
funnel plots (eFig. 2). Leave-one-out analysis revealed between three drugs and schizophrenia (SCZ), as shown
that the combined effect estimate was not driven by any in Fig. 2c and eTable 15. Genetically determined ACE
single SNP (eFig. 3). MR Steiger test did not detect any expression was associated with a lower risk of SCZ
IVs with wrong directionality. After removing IVs (OR = 0.57 [0.44, 0.75], P = 4.41e-05, SMR). Genetically
associated with the four confounders (i.e., smoking, determined CACNA2D1 expression was associated with
drinking, education, and income) as listed in eTable 12, a lower risk of SCZ (OR = 0.90 [0.85, 0.95], P = 3.22e-04,
the IVW-MR results did not change significantly, as SMR). Genetically determined F2 expression was asso-
shown in eTable 13. In the MVMR analysis adjusting for ciated with a lower risk of SCZ (OR = 0.87 [0.80, 0.93],
the four confounders as shown in eTable 14a, although P = 1.84e-04, SMR). HEIDI tests did not show any sig-
relatively large mediation effects were observed in the nificant findings that could be attributed to linkage
effects of HMGCR on OCD by smoking heaviness, and disequilibrium.

Articles
Fig. 2: Psychiatric effects of common drugs. a, ROR heatmap for psychiatric PTs of common drugs from the disproportionality analysis on the
FAERS database. The data are expressed as log-transformed ROR values with unadjusted P values (disproportionality analysis). *P < 0.05;
**P < 0.01; ***P < 0.001. b, Significant results (FDR <0.05, RE-IVW) from IVW-MR on the effects of common drugs on psychiatric disorders. Data
were analyzed using IVW as the primary method, and other three methods (weighted median, weighted mode, and MR Egger) as sensitivity
analysis. The causal estimates are expressed as odds ratio values with 95% CI values. The width of the lines extending from the midpoint
represent the 95% CI (the scale has been log-transformed). Both unadjusted P values (RE-IVW, weighted median, weighted mode, and MR
Egger) and FDR values (RE-IVW) are given. c, Significant results (FDR <0.05, SMR) from SMR on the effects of common drugs on psychiatric
disorders. The causal estimates are expressed as odds ratio values with 95% CI values. The width of the lines extending from the midpoint
represent the 95% CI. Both unadjusted P values (SMR) and FDR values (SMR) are given.

Articles
The effect of common drugs on IDPs the four confounders as shown in eTable 18a, although
IVs selection and validation were already described relatively large mediation effects were observed in some
above. IVW-MR analyses identified 174 significant (false cases such as the effects of PPARG on Volume of 5th-
discovery rate [FDR] < 0.05, RE-IVW) drug-IDP associ- Ventricle by the four confounders, 78.2% of direct ef-
ations between eight drugs (except PDE4A) and 143 fects remained significant after adjustment. As shown
IDPs (including 22 cortical volume metrics, 16 cortical in eTable 18b, 78.2% of the results remained significant
area metrics, 28 cortical thickness metrics, 11 subcor- or insignificant after adjusting for RBC count. As
tical volume metrics, 37 subregions of Amygdala Nuclei, demonstrated in eTable 18c, 47.3% of the direct effects
Hippocampal Subfields, and Thalamic Nuclei, 29 white remained significant after adjusting for HDL and TG,
matter tract measures), as shown in eTable 16. The top suggesting that half of associations between lipid-
20 significant results, which also passed Bonferroni lowering drugs and IDPs may be mediated through
correction, are shown in Fig. 3a, with the schematic other mechanisms other than LDL.
diagram of the relevant brain regions presented in SMR analyses revealed 26 significant gene-IDP as-
Fig. 3b and c. Interestingly, both adverse effects and sociations (FDR <0.05, SMR) between three drug target
positive effects of common drugs on brain structures genes (F2, ABCC8, and CACNA2D2) and 25 IDPs
were found. Among all the drugs, genetically proxied (including 3 cortical measures, 15 subcortical measures,
HMGCR inhibitors had the most extensive impact on and 7 white matter tract measures), as shown in Fig. 4
brain structures, which were linked with 92 IDPs with and eTable 19. Only positive drug-IDP associations
the majority being positive associations. It was followed were identified for genetically determined F2 and
by PCSK9 inhibitors and F2 inhibitors, which were ABCC8 expressions, while genetically determined
associated with 27 and 21 IDPs respectively, also with CACNA2D2 expression only had negative drug-IDP as-
the majority of them being positive associations. sociations. Specifically, genetically proxied CACNA2D2
Notably, PPARG agonists, SGLT2 inhibitors and F10 inhibition led to a reduction in the area and volume of
inhibitors mainly had adverse effects on brain struc- the middle temporal gyrus, and a decrease in mean FA
tures. Additionally, genetically proxied GLP1R agonists in the left corticospinal tract, right superior longitudinal
were significantly associated with a substantial increase fasciculus, external capsule, posterior limb of internal
in the volume of the CA3-head of the right hippocampus capsule, and left superior corona radiata. HEIDI tests
(|beta| = 4.64) and the area of the left precuneus cortex revealed no significant results that were due to linkage
(|beta| = 4.40). disequilibrium.
In sensitivity analyses, MR Egger yielded 29 signifi-
cant results (P < 0.05, MR Egger) among 174 associa- The effect of common drugs on gut microbiota and
tions identified by IVW, which were all in the same gastrointestinal AEs of common drugs
direction as those from IVW-MR. MR-Egger intercept IVs selection and validation were already described
analysis indicated no significant horizontal pleiotropy. above. IVW-MR identified 15 drug-microbiota associa-
Cochran’s Q test revealed significant heterogeneity be- tions between five drugs (PPARG agonists, HMGCR
tween the genetic variants for F2-IDP.328 (P = 0.034, inhibitors, NPC1L1 inhibitors, PCSK9 inhibitors, and
Cochran’s Q test, I2 = 58%). However, this heteroge- PDE4 inhibitors) and 13 genera, as shown in Fig. 5a and
neity will not significantly affect our results since we eTable 20.
used the IVW method with a multiplicative random- In sensitivity analyses, MR Egger produced one sig-
effects model. MR-PRESSO revealed no significant nificant result (P < 0.05, MR Egger) out of the 15 asso-
outlier SNPs. All the above results are displayed in ciations identified by IVW, which was in a consistent
eTable 16. The scatter plots of the top 20 significant direction as that from IVW-MR. Other sensitivity ana-
results are shown in eFig. 4. Asymmetry in funnel plots lyses revealed no abnormalities (eTable 20). The scatter
was observed for PCSK9 inhibitors on IDP.396 (eFig. 5). plots of the 15 significant results are shown in eFig. 7.
Leave-one-out analysis revealed that the combined effect Asymmetry was observed in some funnel plots, prob-
estimate was not affected by the removal of any single ably due to the insufficient number of SNPs (eFig. 8).
SNP (eFig. 6). As presented in eTable 17a, MR Steiger Leave-one-out analysis revealed that the combined effect
test detected five SNPs with wrong directionality, but estimate was not driven by any single SNP (eFig. 9). As
after removing those SNPs, the MR results did not presented in eTable 21a, MR Steiger test detected ten
change significantly. After removing IVs associated with SNPs with wrong directionality, but after removing
the four confounders as listed in eTable 12, 97.2% of the those SNPs, the MR results did not change significantly.
IVW-MR results did not change significantly, as shown After removing IVs associated with the four con-
in eTable 17b. However, four associations previously founders as listed in eTable 12, 99.1% of the IVW-MR
identified as significant became insignificant, and nine results did not change significantly, as shown in
associations previously identified as insignificant eTable 21b. The effect of F2 inhibitors on genus.-
became significant after removing confounder- Romboutsia was previously insignificant but became
associated SNPs. In the MVMR analysis adjusting for significant after removing confounder-associated SNPs.

Articles
b c
Fig. 3: Effects of common drugs on IDPs identified by IVW-MR. a, Top 20 significant results, also those passing Bonferroni correction. Data were
analyzed using IVW as the primary method and MR Egger as sensitivity analysis. The causal estimates are expressed as β values with 95% CI values. The
width of the lines extending from the midpoint represent the 95% CI. Both unadjusted P values (RE-IVW and MR Egger) and FDR values (RE-IVW) are
given. b, Schematic diagram of the brain regions in lateral, medial, and dorsal views. c, Schematic diagram of the white matter tract in right lateral view.

Articles
a
Drug class Target gene Tissue Outcome Top SNP Beta (95% CI) P value FDR
Volume of the left Thalamus-Proper
antidiabetic ABCC8 brain (IDP.195) rs2074312 -0.043 (-0.068, -0.018) 6.68E-04 2.65E-02
Volume of the left Ventral
antidiabetic ABCC8 brain Diencephalon (IDP.202)
rs2074312 -0.049 (-0.074, -0.024) 1.08E-04 8.95E-03
Volume of the left VPL Thalamic
antidiabetic ABCC8 brain Nuclei (IDP.292) rs2074312 -0.046 (-0.071, -0.021) 2.92E-04 1.59E-02
Volume of the left CM Thalamic Nuclei
Volume of the left VAmc Thalamic
antidiabetic ABCC8 brain Nuclei (IDP.298)
rs2074312 -0.041 (-0.065, -0.016) 1.20E-03 3.58E-02
Volume of the left MDl Thalamic Nuclei
Volume of the left VM Thalamic Nuclei
Volume of the left VLp Thalamic Nuclei
antidiabetic ABCC8 brain (IDP.309)
rs2074312 -0.042 (-0.066, -0.017) 9.52E-04 3.24E-02
Volume of the right VPL Thalamic
Volume of the right VM Thalamic
Volume of the left Whole-thalamus
rs2074312 -0.052 (-0.077, -0.027) 4.11E-05 8.95E-03
Volume of medulla oblongata
rs2074312 -0.045 (-0.069, -0.020) 4.22E-04 1.83E-02
antidiabetic ABCC8 brain Volume of Whole-brainstem (IDP.343) rs2074312 -0.040 (-0.064, -0.015) 1.63E-03 4.21E-02
Volume of the right middle temporal
antihypertensive CACNA2D2 blood gyrus (IDP.390)
rs62260815 0.324 (0.142, 0.506) 4.78E-04 3.26E-02
Area of the left middle temporal gyrus
antihypertensive CACNA2D2 blood (IDP.662) rs62260815 0.307 (0.127, 0.488) 8.63E-04 4.21E-02
Area of the right middle temporal
antihypertensive CACNA2D2 blood gyrus (IDP.696) rs62260815 0.400 (0.213, 0.586) 2.61E-05 9.40E-03
Mean FA in the left corticospinal tract
Mean FA in posterior limb of the right
antihypertensive CACNA2D2 blood internal capsule (IDP.1470) rs62260815 0.341 (0.157, 0.525) 2.83E-04 2.63E-02
Mean FA in posterior limb of the left
antihypertensive CACNA2D2 blood internal capsule (IDP.1471) rs62260815 0.355 (0.170, 0.540) 1.69E-04 2.50E-02
Mean FA in the left superior corona
antihypertensive CACNA2D2 blood radiata (IDP.1477)
rs62260815 0.393 (0.206, 0.580) 3.94E-05 9.40E-03
Mean FA in the right external capsule
antihypertensive CACNA2D2 blood (IDP.1484)
rs62260815 0.349 (0.165, 0.534) 2.10E-04 2.50E-02
Mean FA in the left external capsule
Mean FA in the right superior
antihypertensive CACNA2D2 blood longitudinal fasciculus (IDP.1492) rs62260815 0.312 (0.129, 0.494) 8.20E-04 4.21E-02
Volume of the left VLa Thalamic Nuclei
stroke F2 brain (IDP.294)
rs11606709 -0.144 (-0.214, -0.073) 6.86E-05 1.64E-02
Volume of the left VLp Thalamic Nuclei
stroke F2 brain (IDP.309) rs11606709 -0.127 (-0.195, -0.059) 2.65E-04 4.22E-02
Volume of the right PuL Thalamic
stroke F2 brain Nuclei (IDP.328) rs11606709 -0.145 (-0.216, -0.074) 6.24E-05 1.64E-02
-0.2 0.0 0.2 0.4 0.6

Effect estimate (beta) and 95% CIs
b c
Brainstem
Thalamus (left)
Thalamus (right)
Ventral diencephalon
d
L R
Middle temporal gyrus (right) Corticospinal tract

Posterior limb of the internal capsule (right)
Middle temporal gyrus (left)
Posterior limb of the internal capsule (left)
Superior corona radiata
External capsule (right)
External capsule (left)
Superior longitudinal fasciculus

Articles
In the MVMR analysis adjusting for the four con- as shown in Fig. 6a and b and eTable 28. In sensitivity
founders as shown in eTable 22a, 36.9% of direct effects analyses, weighted median also produced significant
remained significant after adjustment. As shown in results (P < 0.05, weighted median), which was in a
eTable 22b, 64.7% of the results remained significant or consistent direction as that from IVW-MR, suggesting
insignificant after adjusting for RBC count. As demon- that the results are reliable when less than half of the
strated in eTable 22c, 83.3% of the direct effects of lipid- IVs are invalid. Other sensitivity analyses revealed no
lowering drugs on gut microbiota remained significant abnormalities (eTable 28). The scatter plots of the five
after adjusting for HDL and TG. significant results are shown in eFig. 10. Asymmetry
SMR analyses did not produce significant drug- was not observed in funnel plots (eFig. 11). The leave-
microbiota associations (eTable 23). one-out analysis did not yield favorable results
Gut microbiota is a critical component of the intesti- (eFig. 12). MR Steiger test did not detect any SNPs with
nal ecosystem and its disturbance can contribute to the wrong directionality. In the reverse MR, IVW did not
development of many gastrointestinal diseases.60,61 identify significant reverse causations for the five
Hence, we next investigated the gastrointestinal adverse microbiota-IDP pairs (eTable 29). In the MR of IDPs on
effects of common drugs using the FAERS database to gut microbiota, 96.7% of the results remained to be
verify the MR findings. Clinical characteristics of significant or insignificant after removing IVs associated
gastrointestinal adverse event (AE) reports were sum- with the confounders (eTable 30).
marized in eTable 24. Among 19 drug classes analyzed,
11 drugs were significantly associated with higher Mediation analysis
gastrointestinal AEs based on the ROR criteria, with two Combining the findings above, we identify five possible
robust signals detected for GLP-1R agonists and angio- drug→microbiota→IDP causal links (eTable 31). We
tensin receptor blockers (eTable 25). GLP-1R agonists conducted a mediation analysis to determine the extent
have the highest effects on gastrointestinal AEs to which the gut microbiota mediates the effect of com-
(ROR = 4.51 [4.43, 4.6]). For each drug class, five PTs mon drugs on IDPs (Fig. 6c–f and eTable 31). The pro-
(Preferred Terms) belonging to SOCs (System Organ portions mediated by gut microbiota vary between 16.5%
Classes) of ‘Gastrointestinal disorders’ with the highest and 27.6%. We did not calculate the mediation proportion
ROR were listed in eTable 26. As shown in Fig. 5b, DPP- of genus Intestinimonas since the total effect and the prod-
IV inhibitors, GLP-1R agonists, and angiotensin receptor uct of drug-microbiota causal estimate and microbiota-IDP
blockers were associated with increased risk of multiple causal estimate were in opposite directions.
gastrointestinal PTs, with angiotensin receptor blockers
having the greatest effect on ‘Sprue-like enteropathy’ Replication analyses based on UK Biobank research
(ROR = 802.17 [733.72, 877.01]). We further performed replication analysis for antidiabetic
In sensitivity analyses, most results still remain stable drugs and anticoagulants on psychiatric disorders using
after deleting cases with missing age, country or sex, UK Biobank data as the source of IV selection, applying a
respectively. For example, 90.9%, 90.9%, 100%, and 90.9% P-value threshold of 5e-8. As shown in eTable 32, the
of the valid signals in the SOC level still remained signif- minimum F-statistic of the IVs for IVW-MR was 31,
icant after deleting the cases missing age, country, sex or which was larger than that in the previous analysis. In
weight, respectively. Additionally, 73.5%, 96.4%, and positive control analyses (eTable 33), most IVs yielded
81.9% of the top five PTs still remained in the top five list significant results. As presented in eTable 34, 66.7% (2/3)
after excluding the cases missing age, country or sex, of the original significant results remained significant in
respectively. However, since the number of cases missing the replication analysis. Besides, we identified five new
weight is relatively large, only 50.6% of the top five PTs still associations that were not discovered in the original
remained after deleting the cases missing weight. analysis. For instance, SLC5A2-mediated HbA1c increase
was associated with higher risk of OCD (OR = 0.02
Causal associations between gut microbiota and [0.003, 0.20], P = 3.29e-4, RE-IVW).
IDPs
IVs used for forward and reverse MR were given in
eTable 27a and b, respectively. The minimum F-statistic Discussion
was 19, suggesting a low risk of weak instrumental bias. We explore the long-term effects of common medications
We identified five results with IVW P value < 0.05 for chronic conditions on the central nervous system
plus one of the additional three methods P value < 0.05, (CNS) through genetic methods and retrospective
Fig. 4: Effects of common drugs on IDPs identified by SMR. a, Significant results with FDR <0.05 (SMR). The causal estimates are expressed as β
values with 95% CI values. The width of the lines extending from the midpoint represent the 95% CI. Both unadjusted P values (SMR) and FDR
(SMR) values are given. b, Schematic diagram of the brain cortical structures in lateral and medial views. c, Schematic diagram of the subcortical
structures in coronal and sagittal section views. d, Schematic diagram of the white matter tracts in lateral and rear views.

Articles
Fig. 5: Causal effects of drugs on gut microbiota and ROR heatmap for gastrointestinal AEs of common drugs. a, Significant results (FDR <0.05,
RE-IVW) from IVW-MR on the effects of common drugs on gut microbiota. Data were analyzed using IVW as the primary method and MR Egger
as sensitivity analysis. The causal estimates are expressed as β values with 95% CI values. The width of the lines extending from the midpoint
represent the 95% CI. Both unadjusted P values (RE-IVW and MR Egger) and FDR values (RE-IVW) are given. b, ROR heatmap for gastrointestinal
PTs of common drugs from the disproportionality analysis on the FAERS database. The data are expressed as log-transformed ROR values with
unadjusted P values (disproportionality analysis). *P < 0.05; **P < 0.01; ***P < 0.001.
analysis of public database. A variety of positive and Among all categories of common drugs, genetically
negative genetic associations between common drugs proxied HMGCR inhibition and its representative drug
and brain structures or psychiatric disorders were iden- statins showed concordant results between MR and
tified. The results from the retrospective analyses of the FAERS analysis. Statins are lipid-lowering agents that
adverse event records in the FAERS database generally function by inhibiting 3-hydroxy-3-methyglutaryl coen-
support the findings from MR. zyme A (HMG-CoA) reductase (HMGCR), a rate-

Articles
c d
e f
Fig. 6: Effects of gut microbiota on IDPs identified by IVW-MR and mediation analysis. a, Significant results with IVW P-value <0.05 (RE-IVW)
and one of the additional three methods P-value <0.05 (weighted median, weighted mode, and MR Egger). Data were analyzed using IVW as
the primary method and other three methods (weighted median, weighted mode, and MR Egger) as sensitivity analysis. The causal estimates
are expressed as β values with 95% CI values. The width of the lines extending from the midpoint represent the 95% CI. Unadjusted P values
(RE-IVW, weighted median, weighted mode, and MR Egger) are given. b, Schematic diagram of the brain regions in lateral, medial, and dorsal
views. c-f, Significant results from the mediation analysis.

Articles
limiting enzyme in the cholesterol synthesis pathway. GLP-1Rs, further indicating the significance of the CNS
We identified that genetically proxied HMGCR inhibi- effects of this drug class.77
tion was associated with higher risks of MDD through Moreover, combining our MR results with previous
MR. Our analysis of FAERS database revealed that sta- MR studies can help discover potential causal links.
tins were associated with higher risks of ‘Self esteem Thiazolidinediones are a class of glucose-lowering
decreased’ (ROR = 18.92 [14.09, 25.42]), ‘Agitated agents that act by activating the peroxisome
depression’ (ROR = 9.76 [4.54, 20.99]) compared with proliferator-activated receptor-γ and thus improving
other drugs (eTable 8), which is consistent with the MR tissue sensitivity to insulin. We have identified that it
results. Likewise, a recent cross-sectional study using was significantly associated with a higher risk of BD II.
medical claims data (n = 7,481,168) showed that the Combining our findings with previous MR of IDPs on
association between statin use and depression may be psychiatric disorders,33 we identified that thiazolidine-
dose-dependent.62 However, previous meta-analyses on diones may increase the risk of BD by decreasing the
observational studies proposed that statins were not volume of the left accumbens. An RCT also discovered
associated with risks of MDD or severity of depressive that pioglitazone treatment was not superior, or even
symptoms,63–65 emphasizing the necessity of rigorous worse than placebo, in terms of antidepressant
randomized controlled trials to solve the problems. efficacy.78
Furthermore, we discovered that statins were associated Combining the idea of pharmacomicrobiomics13 and
with significant changes in various brain structures, the microbiota-gut-brain axis,16,17,79 we speculate that gut
suggesting the wide impact of statins on the CNS. microbiota may function as an important mediator in
Emerging evidence supports that statins exert active the causal effects of common drugs on brain structures.
functions in the CNS mainly through inhibiting iso- Through step-by-step screening, we successfully identi-
prenylation of small GTPases and play a neuroprotective fied five significant drug-microbiota-IDP causal links
role in neurodegenerative disorders, including AD and and calculated the proportion mediated by the micro-
Parkinson’s disease, and neurodevelopmental disorders biota. Previous studies have proved that statin therapy
like Rett syndrome, fragile X syndrome (FXS), and tu- can modulate the gut microbiome composition and
berous sclerosis.66,67 Notably, the three lipid-lowering reduce the risk of microbiota dysbiosis in obese partic-
drugs have opposite effects on several IDPs, as identi- ipants80 and patients with acute coronary syndrome.81
fied by the IVW-MR (eTable 16), including the area of An animal study also suggested that long-term atorvas-
the left inferior parietal cortex, the area of the right tatin treatment led to improved cognitive function via
middle temporal gyrus, and the volume of the right modulating the gut microbiome, affecting retinoic acid
middle temporal gyrus. This can be explained by metabolism and altering immune cells in naturally ag-
different mediation effects of other lipids, as shown in ing rats.82 Zhang et al. reported that atorvastatin treat-
eTable 18c, and possible pleiotropic effects.68–70 ment reduced neuroinflammation and improved gut
Besides, GLP-1R agonists, a highly recommended barrier function via restoring changed gut microbiome
class of glucose-lowering agents for the treatment of composition in mice with middle cerebral artery occlu-
T2D, have accumulated evidence of improving cognitive sion (MCAO). Fecal microbiota transplantation (FMT)
functions, which was indicated by the significant in- from atorvastatin-treated mice significantly improved
crease in the volume of the CA3-head of the right hip- cognitive function and inhibited neuroinflammation in
pocampus and the area of the left precuneus cortex in the MCAO group, indicating that altered microbiota in
our investigation. Proximal CA3 interacts with the atorvastatin-treated mice played a crucial role in the
dentate gyrus (DG) to facilitate pattern separation, the neuroprotective effect.83 Research findings on the effects
ability to make two similar inputs more distinct, while of PPARG agonists (thiazolidinediones) on gut micro-
distal CA3 forms an auto-associative network to perform biota composition are inconsistent. Madsen et al. found
pattern completion, the ability to retrieve stored full no significant effect of rosiglitazone treatment on gut
pattern from incomplete inputs.71 The precuneus is an microbiota composition in diabetic db/db mice.84 In
important region involved in complex cognitive func- contrast, another study revealed that pioglitazone treat-
tions including episodic memory, theory of mind and ment resulted in significant alterations in the gut
self-referential processes, etc.72 Therefore, GLP-1R ago- microbiome of high-fat diet (HFD)-fed obese mice,
nists may improve cognitive functions like episodic including an increase in the abundance of genus Lacto-
memory, which has also been proved by clinical bacillus, which was also identified in our MR.85 Further
studies.3,10,73–75 For example, a 16-week randomized studies are warranted to confirm the mediation model
controlled trial (RCT) demonstrated that liraglutide identified in this study.
improved cognitive functions of delayed memory, One interesting point is that IVW-MR appears to
attention, and executive function, in part through a yield more positive results than SMR. This may be
direct effect on left hippocampal activation.76 A recent attributed to the disparity in the two methodologies;
study revealed that GLP-1R agonists exert anti- IVW meta-analyzes effect estimates from multiple ge-
inflammatory effects depending on central neuronal netic variants, while SMR only leverages the top SNP to

Articles
derive the effect estimate. Prior studies have illustrated weak instruments.89–91 To further validate our results, we
that SMR utilizing summary statistics of the top SNP conducted a replication analysis using the large-scale
yielded less power than multi-SNP prediction when UKB data, employing a stringent P-value threshold
identifying gene expression-trait relationships.86 (<5e-8). This replication was only employed in simu-
Considering the undesirable performance in the posi- lating drug effects on psychiatric disorders, since
tive control analysis, the results of SMR need to be exploring other mediation effects using IVs from UKB
interpreted with caution. data will violate the two-sample MR assumption. The
These findings have extensive implications for the outcomes of this replication were largely consistent with
appraisal of drug side effects, the clinical use of medi- our original findings, thereby reinforcing the robustness
cations and drug repurposing. For example, genetically of our results.
proxied HMGCR inhibition has wide impacts on mul- In conclusion, this study unraveled the long-term
tiple IDPs and is linked with higher risks of MDD and effects of common drugs on brain IDPs and psychiat-
OCD. Therefore, long-term statin users may require ric disorders through drug target MR. Both positive and
regular mental status assessments and patients with negative effects were found, implicating clinical drug
MDD or OCD should use statins with caution. Since use and drug repurposing. Further MR analysis identi-
genus Senegalimassilia mediates the effect of statins on fied four drug-IDP associations that were partially
the area of the right supramarginal gyrus and the left mediated by gut microbiota with mediation proportions
inferior parietal cortex, proper supplementation of this of around 20%. Retrospective analysis of the FAERS
genus may reduce the neuropsychiatric side effects of database identified psychiatric and gastrointestinal side
statins. Additionally, genetically proxied Factor Xa in- effects of common drugs, supporting the MR findings.
hibitors were associated with lower risks of BD I and BD These findings have wide applications and may hope-
II, suggesting that this drug class may be repurposed for fully stimulate the utilization of genetic methods in drug
the treatment of affective disorders. We have only effect prediction.
touched on some of the potential applications of our
Contributors
study, with the intention of encouraging further thought
All authors read and approved the final version of the manuscript.
and discussion. Conceptualization, Z.C, X.W, Z.L, F.L, and K.X; methodology, Z.C,
Our study has the following strengths. Firstly, in X.W, Z.T, and J.H; investigation, Z.C, X.W, Z.T, J.H, J.M, C.Q, and
terms of research content, we incorporated multiple Y.W; formal analysis, Z.C, X.W, and Z.T; visualization, Z.C and X.W;
common drugs and systematically investigated their writing—original draft, Z.C and X.W; writing—review & editing, Z.T,
J.H, J.M, C.Q, and Y.W; supervision, Z.L, F.L, and K.X; resources, Z.L,
effects on various brain structures and psychiatric F.L, and K.X; project administration, Z.L, F.L, and K.X; funding acqui-
disorders. Secondly, the implementation of two MR sition, J.H, Z.L, F.L, and K.X.
methodologies, IVW and SMR, allows us to assess the Z.L, F.L, and K.X verified the underlying data.
long-term neuropsychiatric effects of common drugs
without interference from confounders. Supplemen- Data sharing statement
All GWAS summary statistics involved in this study are publicly avail-
tation with retrospective analysis of the FAERS data-
able on the corresponding websites listed in eTable 2a. Adverse event
base also enriched this study, making the findings reports of drugs are publicly available and can be retrieved from the
more realistic. Thirdly, we hypothesized that the gut FAERS database (https://fis.fda.gov/extensions/FPD-QDE-FAERS/FPD-
microbiome acts as a mediator in drug-IDP associa- QDE-FAERS.html). R codes for relevant analysis in this study will be
tions and identified four drug-microbiota-IDP causal shared upon reasonable request to the corresponding authors.
links.
Declaration of interests
However, several limitations also exist. Firstly, this The authors declare no competing interests.
study leverages GWAS summary statistics of European
populations, and thus further analyses remain needed Acknowledgements
for other populations. Secondly, the FAERS database is a Data on glycaemic traits have been contributed by MAGIC investigators
self-reporting system with some inherent downsides. and have been downloaded from www.magicinvestigators.org. Data on
coronary artery disease have been contributed by CARDIo-
For example, most reports were from the United States GRAMplusC4D investigators and have been downloaded from http://
with the majority of ethnicity being European and not all www.cardiogramplusc4d.org/. The MEGASTROKE project received
AEs of drugs that happened were recorded. Thirdly, our funding from sources specified at http://www.megastroke.org/
study faced limitations in accessing GWAS data with acknowledgments.html. The OCD summary data was supported by
grants from the Judah Foundation, the Tourette Association of America,
large sample sizes for enough genome-wide significant
NIH grants MH079489 and MH073250, American Recovery and Re-
IV selection. To achieve better statistical efficacy, the IVs investment Act (ARRA) awards NS40024-07S1, NS16648-29S1,
selected for simulating antidiabetic drugs, anticoagu- MH071507, MH079489, MH079487, MH079488 and MH079494. Other
lants, and COPD medications did not meet the genome- data sources are listed and described in eTable 2a.
wide significance threshold (P < 5e-8). Despite this This work was supported by the National Natural Science Founda-
tion of China (grant No. 82330035, 82130043, 82172685, and 82001223),
limitation, the minimum F-statistic for the drug target National Natural Science Foundation of Hunan Province (grant No.
IVs in our study was 13, exceeding the conventional 2021SK1010), and the Science Foundation for Distinguished Young
threshold of F > 10,87,88 which indicates a low risk of Scholars of Changsha (grant No. kq2209006).

Articles
Appendix A. Supplementary data 24 Didelez V, Sheehan N. Mendelian randomization as an instru-

Supplementary data related to this article can be found at https://doi. mental variable approach to causal inference. Stat Methods Med Res.
org/10.1016/j.ebiom.2024.105314. 2007;16(4):309–330.
25 Huang W, Xiao J, Ji J, Chen L. Association of lipid-lowering drugs
with COVID-19 outcomes from a Mendelian randomization study.
Elife. 2021;10.
References 26 Harshfield EL, Sims MC, Traylor M, Ouwehand WH, Markus HS.
1 Cham S, Koslik HJ, Golomb BA. Mood, personality, and behavior The role of haematological traits in risk of ischaemic stroke and its
changes during treatment with statins: a case series. Drug Saf Case subtypes. Brain. 2020;143(1):210–221.
Rep. 2016;3(1):1. 27 Yang J, He X, Qian L, et al. Association between plasma proteome
2 Feng Z, Zhao Q, Wu J, et al. Nonselective beta-adrenoceptor and childhood neurodevelopmental disorders: a two-sample Men-
blocker use and risk of Parkinson’s disease: from multiple real- delian randomization analysis. eBioMedicine. 2022;78:103948.
world evidence. BMC Med. 2023;21(1):437. 28 Tang B, Wang Y, Jiang X, et al. Genetic variation in targets of
3 Horska K, Ruda-Kucerova J, Skrede S. GLP-1 agonists: superior for antidiabetic drugs and alzheimer disease risk: a mendelian
mind and body in antipsychotic-treated patients? Trends Endocrinol randomization study. Neurology. 2022;99(7):e650–e659.
Metab. 2022;33(9):628–638. 29 Woolf B, Rajasundaram S, Cronjé HT, Yarmolinsky J, Burgess S,
4 Nowell J, Blunt E, Gupta D, Edison P. Antidiabetic agents as a novel Gill D. A drug target for erectile dysfunction to help improve
treatment for Alzheimer’s and Parkinson’s disease. Ageing Res Rev. fertility, sexual activity, and wellbeing: mendelian randomisation
2023;89:101979. study. BMJ. 2023;383:e076197.
5 Kosowski M, Smolarczyk-Kosowska J, Hachuła M, et al. The effects 30 Ference BA, Ray KK, Catapano AL, et al. Mendelian randomization
of statins on neurotransmission and their neuroprotective role in study of ACLY and cardiovascular disease. N Engl J Med.
neurological and psychiatric disorders. Molecules. 2021;26(10). 2019;380(11):1033–1042.
6 Adesuyan M, Jani YH, Alsugeir D, et al. Antihypertensive agents 31 Rosoff DB, Bell AS, Jung J, Wagner J, Mavromatis LA, Lohoff FW.
and incident Alzheimer’s disease: a systematic review and meta- Mendelian randomization study of PCSK9 and HMG-CoA reduc-
analysis of observational studies. J Prev Alzheimers Dis. tase inhibition and cognitive function. J Am Coll Cardiol.
2022;9(4):715–724. 2022;80(7):653–662.
7 Fenger-Grøn M, Vestergaard CH, Ribe AR, et al. Association be- 32 Qi T, Wu Y, Zeng J, et al. Identifying gene targets for brain-related
tween bipolar disorder or schizophrenia and oral anticoagulation traits using transcriptomic and methylomic data from blood. Nat
use in Danish adults with incident or prevalent atrial fibrillation. Commun. 2018;9(1):2282.
JAMA Netw Open. 2021;4(5):e2110096. 33 Guo J, Yu K, Dong S-S, et al. Mendelian randomization analyses
8 Delhaye S, Bardoni B. Role of phosphodiesterases in the patho- support causal relationships between brain imaging-derived phe-
physiology of neurodevelopmental disorders. Mol Psychiatry. notypes and risk of psychiatric disorders. Nat Neurosci.
2021;26(9):4570–4582. 2022;25(11):1519–1527.
9 Kaduševičius E. Novel applications of NSAIDs: insight and future 34 Burgess S, Thompson SG. Avoiding bias from weak instruments in
perspectives in cardiovascular, neurodegenerative, diabetes and Mendelian randomization studies. Int J Epidemiol. 2011;40(3):755–764.
cancer disease therapy. Int J Mol Sci. 2021;22(12). 35 Burgess S, Butterworth A, Thompson SG. Mendelian randomiza-
10 Vadini F, Simeone PG, Boccatonda A, et al. Liraglutide improves tion analysis with multiple genetic variants using summarized data.
memory in obese patients with prediabetes or early type 2 diabetes: Genet Epidemiol. 2013;37(7):658–665.
a randomized, controlled study. Int J Obes. 2020;44(6):1254–1263. 36 Zhu Z, Zhang F, Hu H, et al. Integration of summary data from
11 Smith GD, Ebrahim S. ’Mendelian randomization’: can genetic GWAS and eQTL studies predicts complex trait gene targets. Nat
epidemiology contribute to understanding environmental de- Genet. 2016;48(5):481–487.
terminants of disease? Int J Epidemiol. 2003;32(1):1–22. 37 Bowden J, Davey Smith G, Burgess S. Mendelian randomization
12 Dehnavi S, Kiani A, Sadeghi M, et al. Targeting AMPK by statins: a with invalid instruments: effect estimation and bias detection
potential therapeutic approach. Drugs. 2021;81(8):923–933. through Egger regression. Int J Epidemiol. 2015;44(2):512–525.
13 Zhao Q, Chen Y, Huang W, Zhou H, Zhang W. Drug-microbiota 38 Verbanck M, Chen CY, Neale B, Do R. Detection of widespread
interactions: an emerging priority for precision medicine. Signal horizontal pleiotropy in causal relationships inferred from Men-
Transduct Target Ther. 2023;8(1):386. delian randomization between complex traits and diseases. Nat
14 Jackson MA, Verdi S, Maxan ME, et al. Gut microbiota associations Genet. 2018;50(5):693–698.
with common diseases and prescription medications in a 39 Hemani G, Tilling K, Davey Smith G. Orienting the causal rela-
population-based cohort. Nat Commun. 2018;9(1):2655. tionship between imprecisely measured traits using GWAS sum-
15 Jackson MA, Goodrich JK, Maxan ME, et al. Proton pump in- mary data. PLoS Genet. 2017;13(11):e1007081.
hibitors alter the composition of the gut microbiota. Gut. 40 Prochaska JJ, Das S, Young-Wolff KC. Smoking, mental illness, and
2016;65(5):749–756. public health. Annu Rev Public Health. 2017;38:165–185.
16 Socała K, Doboszewska U, Szopa A, et al. The role of microbiota- 41 Hjorthøj C, Østergaard ML, Benros ME, et al. Association between
gut-brain axis in neuropsychiatric and neurological disorders. alcohol and substance use disorders and all-cause and cause-
Pharmacol Res. 2021;172:105840. specific mortality in schizophrenia, bipolar disorder, and unipolar
17 Mayer EA, Nance K, Chen S. The gut-brain axis. Annu Rev Med. depression: a nationwide, prospective, register-based study. Lancet
2022;73:439–453. Psychiatr. 2015;2(9):801–808.
18 Skrivankova VW, Richmond RC, Woolf BAR, et al. Strengthening 42 McCloud T, Kamenov S, Callender C, Lewis G, Lewis G. The as-
the reporting of observational studies in epidemiology using sociation between higher education attendance and common
mendelian randomization: the STROBE-MR statement. JAMA. mental health problems among young people in England: evidence
2021;326(16):1614–1621. from two population-based cohorts. Lancet Public Health.
19 Smith SM, Douaud G, Chen W, et al. An expanded set of genome- 2023;8(10):e811–e819.
wide association studies of brain imaging phenotypes in UK Bio- 43 Ridley M, Rao G, Schilbach F, Patel V. Poverty, depression, and
bank. Nat Neurosci. 2021;24(5):737–745. anxiety: causal evidence and mechanisms. Science. 2020;370(6522).
20 Kurilshikov A, Medina-Gomez C, Bacigalupe R, et al. Large-scale 44 Leslie R, O’Donnell CJ, Johnson AD. GRASP: analysis of genotype-
association analyses identify host factors influencing human gut phenotype results from 1390 genome-wide association studies and
microbiome composition. Nat Genet. 2021;53(2):156–165. corresponding open access database. Bioinformatics.
21 Malik R, Chauhan G, Traylor M, et al. Multiancestry genome-wide 2014;30(12):i185–i194.
association study of 520,000 subjects identifies 32 loci associated 45 Elsworth B, Lyon M, Alexander T, et al. The MRC IEU OpenGWAS
with stroke and stroke subtypes. Nat Genet. 2018;50(4):524–537. data infrastructure. bioRxiv. 2020. https://doi.org/10.1101/2020.08.
22 Zhou C, Peng S, Lin A, et al. Psychiatric disorders associated with 10.244293.
immune checkpoint inhibitors: a pharmacovigilance analysis of the 46 Hemani G, Zheng J, Elsworth B, et al. The MR-Base platform
FDA Adverse Event Reporting System (FAERS) database. eClini- supports systematic causal inference across the human phenome.
calMedicine. 2023;59:101967. Elife. 2018;7.
23 Evans SJW, Waller PC, Davis S. Use of proportional reporting ra- 47 Sollis E, Mosaku A, Abid A, et al. The NHGRI-EBI GWAS Catalog:
tios (PRRs) for signal generation from spontaneous adverse drug knowledgebase and deposition resource. Nucleic Acids Res.
reaction reports. Pharmacoepidemiol Drug Saf. 2001;10(6):483–486. 2023;51(D1):D977–D985.

Articles
48 Burgess S, Thompson SG. Multivariable Mendelian randomization: 71 Lee H, GoodSmith D, Knierim JJ. Parallel processing streams in
the use of pleiotropic genetic variants to estimate causal effects. Am the hippocampus. Curr Opin Neurobiol. 2020;64:127–134.
J Epidemiol. 2015;181(4):251–260. 72 Dadario NB, Sughrue ME. The functional role of the precuneus.
49 Carter AR, Sanderson E, Hammerton G, et al. Mendelian ran- Brain. 2023;146(9):3598–3607.
domisation for mediation analysis: current methods and chal- 73 Mansur RB, Ahmed J, Cha DS, et al. Liraglutide promotes im-
lenges for implementation. Eur J Epidemiol. 2021;36(5): provements in objective measures of cognitive dysfunction in in-
465–478. dividuals with mood disorders: a pilot, open-label study. J Affect
50 Sanderson E, Davey Smith G, Windmeijer F, Bowden J. An ex- Disord. 2017;207:114–120.
amination of multivariable Mendelian randomization in the single- 74 Yaribeygi H, Rashidy-Pour A, Atkin SL, Jamialahmadi T,
sample and two-sample summary data settings. Int J Epidemiol. Sahebkar A. GLP-1 mimetics and cognition. Life Sci. 2021;264:
2019;48(3):713–727. 118645.
51 Lindqvist D, Mellon SH, Dhabhar FS, et al. Increased circulating 75 Monney M, Jornayvaz FR, Gariani K. GLP-1 receptor agonists effect
blood cell counts in combat-related PTSD: associations with on cognitive function in patients with and without type 2 diabetes.
inflammation and PTSD severity. Psychiatry Res. 2017;258:330–336. Diabetes Metabol. 2023;49(5).
52 Fruchart JC, Duriez P. HDL and triglyceride as therapeutic targets. 76 Cheng H, Zhang Z, Zhang B, et al. Enhancement of Impaired
Curr Opin Lipidol. 2002;13(6):605–616. Olfactory neural activation and cognitive capacity by liraglutide, but
53 Arsenault BJ, Boekholdt SM. Clinical and biological relevance of not Dapagliflozin or Acarbose, in patients with type 2 diabetes: a 16-
statin-mediated changes in HDL metabolism. Curr Atheroscler Rep. week randomized parallel comparative study. Diabetes Care.
2014;16(1):379. 2022;45(5):1201–1210.
54 Schoch L, Sutelman P, Suades R, et al. Hypercholesterolemia- 77 Wong CK, McLean BA, Baggio LL, et al. Central glucagon-like
induced HDL dysfunction can Be reversed: the impact of diet and peptide 1 receptor activation inhibits Toll-like receptor agonist-
statin treatment in a preclinical animal model. Int J Mol Sci. induced inflammation. Cell Metab. 2023;36:130.
2022;23(15). 78 Aftab A, Kemp DE, Ganocy SJ, et al. Double-blind, placebo-
55 Bowden J, Davey Smith G, Haycock PC, Burgess S. Consistent controlled trial of pioglitazone for bipolar depression. J Affect Dis-
estimation in mendelian randomization with some invalid in- ord. 2019;245:957–964.
struments using a weighted median estimator. Genet Epidemiol. 79 Góralczyk-Bińkowska A, Szmajda-Krygier D, Kozłowska E. The
2016;40(4):304–314. microbiota–gut–brain Axis in psychiatric disorders. Int J Mol Sci.
56 Hartwig FP, Davey Smith G, Bowden J. Robust inference in sum- 2022;23(19):11245.
mary data Mendelian randomization via the zero modal pleiotropy 80 Vieira-Silva S, Falony G, Belda E, et al. Statin therapy is associated
assumption. Int J Epidemiol. 2017;46(6):1985–1998. with lower prevalence of gut microbiota dysbiosis. Nature.
57 Sinnott-Armstrong N, Tanigawa Y, Amar D, et al. Genetics of 35 2020;581(7808):310–315.
blood and urine biomarkers in the UK Biobank. Nat Genet. 81 Hu X, Li H, Zhao X, et al. Multi-omics study reveals that statin
2021;53(2):185–194. therapy is associated with restoration of gut microbiota homeosta-
58 Sun BB, Chiou J, Traylor M, et al. Plasma proteomic associations sis and improvement in outcomes in patients with acute coronary
with genetics and health in the UK Biobank. Nature. 2023;622 syndrome. Theranostics. 2021;11(12):5778–5793.
(7982):329–338. 82 Xu T-C, Lv Y, Liu Q-Y, Chen H-S. Long-term atorvastatin improves
59 Xia M, Wang J, He Y. BrainNet viewer: a network visualization tool cognitive decline by regulating gut function in naturally ageing rats.
for human brain connectomics. PLoS One. 2013;8(7):e68910. Immun Ageing. 2022;19(1).
60 Jones RM, Neish AS. Gut microbiota in intestinal and liver disease. 83 Zhang P, Zhang X, Huang Y, et al. Atorvastatin alleviates microglia-
Annu Rev Pathol. 2021;16:251–275. mediated neuroinflammation via modulating the microbial
61 Cruz MS, Tintelnot J, Gagliani N. Roles of microbiota in pancreatic composition and the intestinal barrier function in ischemic stroke
cancer development and treatment. Gut Microb. 2024;16(1): mice. Free Radic Biol Med. 2021;162:104–117.
2320280. 84 Madsen MSA, Grønlund RV, Eid J, et al. Characterization of local
62 Leutner M, Matzhold C, Kautzky A, et al. Major depressive disorder gut microbiome and intestinal transcriptome responses to rosigli-
(MDD) and antidepressant medication are overrepresented in high- tazone treatment in diabetic db/db mice. Biomed Pharmacother.
dose statin treatment. Front Med. 2021;8:608083. 2021;133.
63 Köhler-Forsberg O, Otte C, Gold SM, Østergaard SD. Statins in the 85 Wang D, Liu J, Zhong L, et al. Potential benefits of metformin and
treatment of depression: Hype or hope? Pharmacol Ther. 2020;215: pioglitazone combination therapy via gut microbiota and metabo-
107625. lites in high-fat diet-fed mice. Front Pharmacol. 2022;13.
64 Lee MC, Peng TR, Lee CH, et al. Statin use and depression risk: a 86 Veturi Y, Ritchie MD. How powerful are summary-based methods
systematic review and meta-analysis. J Affect Disord. 2021;282: for identifying expression-trait associations under different genetic
308–315. architectures? Pac Symp Biocomput. 2018;23:228–239.
65 De Giorgi R, Waters S, Pesci NR, Rosso G, Cowen PJ, Harmer CJ. 87 Skrivankova VW, Richmond RC, Woolf BAR, et al. Strengthening
The effects of statin monotherapy on depressive symptoms: a sys- the reporting of observational studies in epidemiology using
tematic review and meta-analysis. J Affect Disord. 2022;311:336–343. mendelian randomisation (STROBE-MR): explanation and elabo-
66 Ling Q, Tejada-Simon MV. Statins and the brain: new perspective ration. BMJ. 2021;375:n2233.
for old drugs. Prog Neuro Psychopharmacol Biol Psychiatr. 88 Davies NM, Holmes MV, Davey Smith G. Reading Mendelian
2016;66:80–86. randomisation studies: a guide, glossary, and checklist for clini-
67 Fracassi A, Marangoni M, Rosso P, et al. Statins and the brain: cians. BMJ. 2018;362:k601.
more than lipid lowering agents? Curr Neuropharmacol. 2018;17 89 Lawlor DA, Harbord RM, Sterne JA, Timpson N, Davey Smith G.
(1):59–83. Mendelian randomization: using genes as instruments for making
68 Basiak M, Kosowski M, Cyrnek M, et al. Pleiotropic effects of causal inferences in epidemiology. Stat Med. 2008;27(8):1133–1163.
PCSK-9 inhibitors. Int J Mol Sci. 2021;22(6). 90 Staiger D, Stock JH. Instrumental variables regression with weak
69 German CA, Liao JK. Understanding the molecular mechanisms of instruments. Econometrica. 1997;65(3):557–586.
statin pleiotropic effects. Arch Toxicol. 2023;97(6):1529–1545. 91 Sekula P, Del Greco MF, Pattaro C, Köttgen A. Mendelian
70 Oesterle A, Laufs U, Liao JK. Pleiotropic effects of statins on the randomization as an approach to assess Causality using observa-
cardiovascular system. Circ Res. 2017;120(1):229–243. tional data. J Am Soc Nephrol. 2016;27(11):3253–3265.

A-comprehensive-assessment-of-the-association-betw

Uploaded by

Copyright:

Available Formats

A-comprehensive-assessment-of-the-association-betw

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

A-comprehensive-assessment-of-the-association-betw

Uploaded by

Copyright:

Available Formats

Articles

A comprehensive assessment of the association between

www.thelancet.com Vol 107 September, 2024 1

Introduction researchers to assess the potential effects of drugs on

2 www.thelancet.com Vol 107 September, 2024

www.thelancet.com Vol 107 September, 2024 3

The original study analyzed brain imaging data of 39,691

chiatric disorders were acquired mainly through the

4 www.thelancet.com Vol 107 September, 2024

www.thelancet.com Vol 107 September, 2024 5

6 www.thelancet.com Vol 107 September, 2024

www.thelancet.com Vol 107 September, 2024 7

8 www.thelancet.com Vol 107 September, 2024

Table 1: ROR and PRR of psychiatric AEs among 19 drugs.

www.thelancet.com Vol 107 September, 2024 9

10 www.thelancet.com Vol 107 September, 2024

www.thelancet.com Vol 107 September, 2024 11

12 www.thelancet.com Vol 107 September, 2024

-0.2 0.0 0.2 0.4 0.6

Middle temporal gyrus (right) Corticospinal tract

www.thelancet.com Vol 107 September, 2024 13

14 www.thelancet.com Vol 107 September, 2024

www.thelancet.com Vol 107 September, 2024 15

16 www.thelancet.com Vol 107 September, 2024

www.thelancet.com Vol 107 September, 2024 17

18 www.thelancet.com Vol 107 September, 2024

Appendix A. Supplementary data 24 Didelez V, Sheehan N. Mendelian randomization as an instru-

www.thelancet.com Vol 107 September, 2024 19

20 www.thelancet.com Vol 107 September, 2024

You might also like