Applications of Mendelian randomization in psychiatry: a comprehensive systematic review

Introduction

Mental disorders comprehend a wide variety of diseases with different presentations. They are generally characterized by a combination of abnormal thoughts, perceptions, emotions, behavior, and relationships with others. Mental disorders include major depression disorder (MDD), bipolar disorder (BD), schizophrenia (SCZ) and other psychoses, and developmental disorders including autism spectrum disorders (ASD) or attention-deficit and hyperactivity disorder (ADHD). It is estimated that 14.3% of deaths worldwide are caused by mental disorders (for a total of approximately 8 million), which lead to more than 125 million disability-adjusted life-years (Walker et al., 2015; GBD 2019 Mental Disorders Collaborators, 2019). Besides the heavy personal burden on patients and caregivers, psychiatric diseases also exact a heavy socioeconomic toll (Walker et al., 2015). It is particularly difficult to identify risk factors and causative mechanisms for psychiatric diseases due to the limited physiopathological and etiological insight we have, to the many confounding factors, and to the potential reverse causality between the risk factors and these diseases, which are likely multifactorial (i.e. with important genetic and environmental risk factors). This is one of the reasons that make Mendelian randomization (MR) an especially precious tool for studying these diseases. MR is an analytical method that employs genetic variants [or instrumental variables (IVs)] linked to a certain risk factor, to assess if an observational association between that risk factor and a health outcome is compatible with a causal relationship between the risk factor and the chosen outcome. More in detail, every person is inherently assigned a genetic variant that could influence a risk factor in different ways [e.g. a variant that regulates the blood level of LDL could influence the risk of coronary artery disease (CAD)]. In a Mendelian randomized study, one or more of these genetic variants are followed up to verify the development of a specific health outcome. As recapitulated in Fig. 1, MR relies on three main assumptions: a relevance assumption (that the variant is associated with the risk factor, or exposure, in study), an exclusion restriction assumption (that the variants are affecting the outcome only through the risk factor/exposure), and an independence assumption (that these variants have not a shared cause with the outcome) (Emdin et al., 2017).

F1Fig. 1:

Diagram of a typical Mendelian randomization design, with, as an example, the key assumptions of a possible association between higher plasmatic IL-6 and schizophrenia risk. IL-6, Interleukin 6; IV, instrumental variable; SNP, single-nucleotide polymorphism.

Thus, it is clear the importance of the concept of pleiotropy, which is the production by one single genetic trait [such as a single-nucleotide polymorphism (SNP)] of two or more apparently unrelated outcomes. If pleiotropy arises because an SNP influences one trait (the exposure), which then influences another (i.e. the outcome), then vertical pleiotropy exists and MR can be used to estimate the causal relationship between exposure and outcome. Prominent among the limitations of MR studies, is the nondemonstrable assumption that the exposure mediates the apparent pleiotropic associations (thus reflecting vertical pleiotropy), and that the selected SNP does not influence both the exposure and the outcomes through unrelated pathways (thus violating the exclusion restriction assumption due to ‘horizontal pleiotropy’) (Hemani et al., 2018). With these caveats, MR can use genetic associations obtained with genome-wide association studies (GWAS) to infer causality between exposures and outcomes.

Two main types of MR studies exist: the single sample randomization study and the double sample randomization study. The first one requires the measurement of the gene variant, the risk factor, and health outcome from the same sample of participants. The two-sample MR study requires two different study populations. In this approach, the gene variants and the health outcome are measured from one group, whereas the gene variants and risk factors are measured in another group (Pierce and Burgess, 2013). The main advantage of two-sample MR is that the outcome of interest and the risk factor do not need to be both measured in all participants, which is particularly important if they are expensive or difficult to measure. Thus, two-sample MR allows employing results from GWAS, which are usually precise and large studies (Davies et al., 2018).

To recapitulate, MR is particularly relevant to study risk factors for diseases in which (a) it is difficult to ascertain causality between the risk factor and the disease, (b) reverse causality is possible, and (c) confounding factors are abundant and possibly hidden. Thus, thanks to its potentialities, MR is being used in different studies to evaluate the relationship between possible risk factors and the development of psychiatric diseases. As an example, MR may be employed to assess causality between BD risk and lifetime cannabis use (having ever used cannabis). In fact, if an association between cannabis use and BD were to be found with other study designs, it would be difficult to exclude that it is BD that leads to higher cannabis use, and not vice versa. Similarly, reverse causality is a substantial problem in SCZ psychoneuroimmunology, since associated psychological morbidities can lead to issues with personal hygiene or housing insecurity, which, together with medication side effects, can impact the immune system. Since MR uses genetic information (which generally existed before full-blown psychiatric symptoms) as risk factors, reverse causality is unlikely.

Important insights into MR limitations and strengths can emerge from literature reviews, as well as from the collaboration between clinicians, methodologists, and empirical researchers, from which have benefited also other areas of medical research (Davies et al., 2018). While the interest of MR in psychiatry is increasingly prominent (Wootton et al., 2022), most of the existing reviews focus on specific disorders (Belbasis et al., 2018; Köhler et al., 2018) or risk factors (Treur et al., 2021), or they do not systematically review MR studies in psychiatry (Wootton et al., 2022). In fact, to the best of our knowledge, there is no comprehensive overview of the applications of MR across different psychiatric diseases. Such a transdiagnostic approach would provide a unifying view of the limitations and potential of MR in psychiatry, as well as stimulating insight into potential common risk factors for different psychiatric disorders.

With the present systematic review, we thus aim at providing, for the first time, an unbiased and inclusive view of all existing applications of MR in psychiatric disorders, as defined by Diagnostic and Statistical Manual of Mental Disorders (DSM) or International Classification of Diseases (ICD) diagnoses. Such an approach aims at facilitating the identification of risk factors that may be common to different psychiatric diseases, and at paving the way to transdiagnostic MR studies in psychiatry (which are currently under-investigated), in agreement with current psychiatry research models advising to target aspects common to different pathologies, rather than traditionally defined diagnoses only (Insel, 2014).

Methods Search strategy and selection criteria

This systematic review was conducted following the Preferred Reporting Items for Systematic Reviews and Meta-analyses (PRISMA) guidelines (Supplementary eTable 1, Supplemental Digital Content 1, https://links.lww.com/PG/A287) (Moher et al., 2009), and is registered in the PROSPERO database (registration number CRD42021285647).

We used a two-step approach. First, we searched the Web of Science database by Thomas Reuters, Pubmed, and Scopus. The search strategy included terms related to MR and psychiatry [((Mendelian randomization) OR (Mendelian randomisation)) AND (psychiatry OR bipolar disorder OR borderline personality disorder OR schizophrenia OR depression OR ADHD OR anxiety OR PTSD OR panic))] for articles published until 3 May 2022. We excluded reviews or metanalysis through the search filters. Second, we performed a manual search of the lists of references of retrieved articles. Duplicate references were manually excluded. The remaining articles were screened by title and abstract, and the full texts identified were further inspected for eligibility against a priori-defined exclusion and inclusion criteria. We included original articles in English that met the following Participants, Interventions, Comparators, Outcomes, and Study design criteria. Participants of the study were psychiatric patients of any age, with any psychiatric diagnosis according to DSM or ICD criteria. The Intervention had to include the employment of MR in genetic analysis, applied to the identification of risk factors, to neuroimaging correlations, or to other settings relating to psychiatry (including studies that focused on how psychiatric diseases might be risk factors for other conditions). Comparators were the presence or absence of genetic traits that are determinants of exposure to a certain risk factor (which could be a psychiatric disorder or other risk factors). The Outcome was the risk for a certain condition (both psychiatric and not).

We included all study designs apart from case reports, case series, conference abstracts and presentations, pilot/feasibility studies, reviews, meta-analyses, and systematic reviews. Of note, we did not exclude studies that analyzed (considering different exposures/outcomes, or using different methods from each other) data stored in publicly available databases, or overlapping or partly overlapping populations. The selection process was documented in the PRISMA flow diagram (Fig. 2).

F2Fig. 2:

PRISMA 2009 flow diagram. PRISMA, Preferred Reporting Items for Systematic Reviews and Meta-analyses.

Data extraction

Data extraction was performed by independent researchers (L.F.S. and S.G.). Any discrepancy was discussed until a consensus was reached. Disagreements were resolved by a third reviewer (G.R.).

The following variables were extracted from each article: authors, year of publication, sample size, genetic information analyzed, main MR method, main findings, presence of pleiotropy analysis, and psychiatric disorder considered as an outcome. If more than one method were employed and results were homogeneous across methods, we reported the first one described. If results obtained through different methods of analysis differed, we reported each method in the corresponding column, and we indicated between parathesis which method was employed to obtain each result in the ‘Outcome and finding’ column. If variables were not available, and no reply was obtained from the corresponding author of the article concerned (at least a 1-month delay for each query), we wrote ‘NG’, not given. If crucial information was missing, we contacted the corresponding author and excluded the study if no answer was received (at least 1-month delay for each query). In the case of articles reporting on more than one psychiatric disease, we included only results on diseases that had been diagnosed according to ICD or DSM criteria, in agreement with our inclusion criteria.

Quality assessment

The quality of the selected studies was assessed independently by two reviewers (S.G. and L.F.S.) with the Newcastle-Ottawa Scale (NOS) (Lo et al., 2014). Any discrepancy was discussed until a consensus was reached. Disagreements were resolved by a third reviewer (G.R.). The NOS is a risk of bias assessment tool that has been used also to evaluate MR studies (Cheng et al., 2020; Spiga et al., 2021). It consists of two sections: one for case-report studies and the other for cohort studies. Studies were evaluated using NOS considering three aspects: patient selection, comparability, and exposure (Supplementary eTable 2, Supplemental Digital Content 1, https://links.lww.com/PG/A287). The risk of bias and concerns regarding applicability were analyzed for each domain.

Results

As described in Fig. 2, 1039 nonduplicates studies were selected through database searching. Two hundred and eighty-seven of these were excluded based on the title, as they were not relevant. A total of 752 articles were screened based on the abstract to exclude articles that were not in English (n = 3), that were not pertinent (i.e. not about MR in psychiatry) (n = 497) or not original articles (e.g. reviews, case reports, metanalyses, and opinion articles) (n = 74). Upon eligibility screening, 128 studies were not eligible as 71 articles included at least some psychiatric patients with diagnoses not based on ICD or DSM criteria, two studies had missing information (and no reply was obtained from the corresponding author of the article concerned after at least 1-month delay for each query), and 55 articles were not pertinent.

The final sample, thus, included 50 articles. Although we searched for studies published from 1966 until 2022, the oldest included study dates back to 2014 (Hung et al., 2014). This confirms the novelty of this research topic. Thirty studies focused on SCZ, 15 on MDD, four on ADHD or ASD, and nine on other disorders. Of note, the sum of the articles included in each group does not equal 50 because studies presenting distinct results concerning multiple diseases are reported in each of the relevant tables for each disease. Details of each study are reported in Tables 1–4.

The NOS scores were fair and homogenous among the studies. More in detail, all studies scored high in the ‘selections of patients’ and ‘comparability’. Further details are provided in Supplementary eTable 2, Supplemental Digital Content 1, https://links.lww.com/PG/A287.

Mendelian randomization in schizophrenia

The most numerous group of articles that we included is on SCZ, comprising 30 articles (Table 1) (Wium-Andersen et al., 2015; Prins et al., 2016; Taylor et al., 2016; Inoshita et al., 2016; Gage et al., 2017a; Gage et al., 2017b; Hartwig et al., 2017; Arafat and Minica, 2018; Li et al., 2018; Pasman et al., 2018; Polimanti et al., 2018; Tomioka et al., 2018; Byrne et al., 2019; Gao et al., 2019; Jang et al., 2020; Jones et al., 2020; Kim et al., 2020, 2021; Luo et al., 2020; Peters et al., 2020; Wootton et al., 2020; Yang et al., 2020; Zhuang et al., 2020; Jin et al., 2021; Johnson et al., 2021; Perry et al., 2021b; Song et al., 2021; Chen et al., 2021,2022; Andreu-Bernabeu et al., 2022; Ni et al., 2022). Most studies focused on inflammation, metabolic traits, or substance use and their associations with SCZ.

Table 1 - Mendelian randomization in schizophrenia References Genetic instrument and exposure (P-value) Main statistic test Exclusion of pleiotropy Outcome and findings Outcome sample size Andreu-Bernabeu et al., 2022 116 SNPs associated with SCZ (P < 5 × 10−8) WM Yes Bidirectional causal effect between polygenic scores for loneliness/isolation and SCZ risk (WM−β (SE) = 1.37 (0.40); P = 6.14 × 10−4) 35 476 SCZ cases 19 SNPs associated with loneliness/social isolation (P < 5 × 10−6) 46 839 controls (PGC) Chen et al., 2022 2422 SNPs associated with 41 systemic inflammatory regulators, 2788 SNPs associated with SCZ (P < 1 × 10−6) IVW Yes Genetic predisposition to increased HGF, IL-17, IL-1ra, MCP3, TRAIL causally associated with higher SCZ risk (P < 5 × 10−2) 36 989 SCZ cases Genetic liability to SCZ causally associated with CTAK and RANTES (P < 3 × 10−2) 11 3075 controls (PGC) Ni et al., 2022 >700 SNPs associated with gut microbiota and 28 SNPs associated with SCZ (P < 1 × 10−5) IVW Yes Genetic predisposition to increased class Actinobacteria causally associated with higher SCZ risk (P = 1.33 × 10−3) 152 805 SCZ cases, 18 473 controls (MiBioGen) No causal association between genetic instruments for SCZ and risk of different gut microbiota composition Perry et al., 2021b 53 SNP associated with insulin resistance phenotype (P < 5 × 10−3) IVW Yes No causal association between genetic instruments for insulin resistance and SCZ (IVW) 40 675 SCZ cases 58 SNP associated with CRP (P < 5 × 10−8) MVMR No significant association between inflammation-related SNP and SCZ after adjusting for genetic instruments for CRP (MVMR) 64 643 controls Kim et al., 2021 4 SNP associated with PD risk (P < 5 × 10−6) IVW Yes Genetic predisposition to PD causally associated with higher SCZ risk (OR, 1.10; 95% CI, 1.05−1.15; P = 3.49 × 10−5) 35 476 SCZ cases 46 839 controls (PGC) Jin et al., 2021 93 SNP associated with PCOS (P < 5 × 10−8) IVW, Yes No causal association between genetic instruments for PCOS predisposition and SCZ risk 33 640 SCZ cases 43 456 controls Chen et al., 2021 31 SNP associated with cigarette smoking (P < 1 × 10−5) IVW, MR-Egger Yes No causal association between genetic instruments for smoking and SCZ 7711 SCZ cases 18 327 controls Yang et al., 2020 530 SNP associated with 2-methoxyacetaminophen sulfate levels (P < 1 × 10−5) IVW Yes Genetic predisposition to higher 2-methoxyacetaminophen sulfate level causally associated with higher SCZ risk (P = 1.73 × 10−5) 36 989 SCZ cases and 113 075 controls Pasman et al., 2018 74 SNP associated with cannabis use (P < 1 × 10−5) IVW Yes Genetic predisposition to cannabis use causally associated with higher SCZ risk (OR, 1.10; 95% CI, 0.99–1.21; P = 0.074) 130 072 CZ cases 109 with SCZ (P < 5 × 10−8) Genetic liability to SCZ causally associated with cannabis use (OR, 1.16; 95% CI, 1.06–1.27; P = 0.001) 180 934 cases of cannabis use Zhuang et al., 2020 30 SNP associated with Enterobacteriaceae family and order (P = 3.72 × 10−11) IVW Yes Genetic predisposition to increased: 21 246 SCZ cases Enterobacteriaceae family and order causally associated with higher SCZ risk (OR, 1.09; CI, 1.00–1.18; P = 0.048) Gammaproteobacteria class (OR, 0.90; CI, 0.83–0.98; P = 0.011) causally associated with lower SCZ risk 15 with Gammaproteobacteria class (P < 1.4 × 10−8) Gut production of serotonin causally associated with higher SCZ risk (OR, 1.07; CI, 1.00–1.15; P = 0.047) 32 with increased gut production of serotonin (P < 8.96 × 10−6) No causal associations with SCZ risk for other types of gut microbiota 38 072 controls Wootton et al., 2020 114 SNP associated with SCZ IVW Yes Genetic predisposition to lifetime smoking causally associated with higher SCZ risk (OR, 2.27; 95% CI, 1.67–3.08; P < 0.001) 36 989 SCZ cases 126 with lifetime smoking (P < 5 × 10−8) Genetic liability for SCZ causally associated with higher risk of lifetime smoking (β= 0.022; 95% CI, 0.005–0.038; P = 0.009) 113 075 controls (PGC) 213 372 ever-smokers and 249 318 never-smokers (UK Biobank) Peters et al., 2020 319 SNP associated with WHRadjBMI IVW, MR-Egger Yes No causal association between genetic instruments for WHRadjBMI and SCZ risk 40 675 SCZ cases 64 643 controls (PGC) Kim et al., 2020 70 SNP associated with SCZ (P < 5 × 10−8) IVW Yes Genetic liability to SCZ causally associated with higher breast cancer risk (OR, per log odds increase in SCZ risk, 1.069; 95% CI, 1.028–1.112; P < 0.001) 82 315 SCZ cases (PGC) 117, 146, 204 SNP associated with breast cancer (5 × 10−8, 5 × 10−7, 5 × 10−6respectively) No causal association between genetic instruments for breast cancer and SCZ 228 951 breast cancer cases (BCAC) Jones et al., 2020 5 SNP associated with anxiety IVW Yes Genetic liability to neuroticism (OR, 1.33; 95% CI, 1.12–1.59) and anxiety (OR, 1.10; 95% CI, 1.01–1.19) causally associated with higher SCZ risk 33 640 SCZ cases 116 with neuroticism (P < 5 × 10−8) 43 456 controls (PGC) Jang et al., 2020 341 SNP associated with substance use phenotypes (smoking, alcohol, or cannabis use) IVW, WM, MR-Egger; Latent Causal Variable model Yes No causal association between genetic instruments for substance use phenotypes and the risk SCZ 105 318 SCZ cases Song et al., 2021 1 SNP associated with pars triangularis volume (P < 5 × 10−8) IVW No Genetic predisposition to lower pars triangularis volume causally associated with SCZ risk (OR, 0.48; 95% CI, 0.34-0.69; P = 5.9 × 10−5) 40 675 SCZ cases 64 643 controls (PGC) Luo et al., 2020 45 SNP associated with SCZ (P < 5 × 10−8) IVW Yes Genetic liability to SCZ causally associated with higher serum uric acid risk (per 10-s% increase in SCZ risk: beta: −0.039, SE: 0.013; P = 0.003) 35 476 SCZ cases 26 SNP associated with serum uric acid (P < 5 × 10−8) No causal association between genetic instruments for higher serum uric acid and SCZ 46 839 controls (PGC) Gao et al., 2019 110 SNP associated with insomnia (P < 5 × 10−8) MR-Egger, GSMR Yes No causal association between genetic instruments for insomnia and SCZ 33 426 SCZ cases 32 541 controls Byrne et al., 2019 142 SNP associated with SCZ GSMR Yes Genetic liability to SCZ causally associated with higher breast cancer risk (SE = 0.008; P = 2.2 × 10−4) 40 675 SCZ cases and 64 643 controls (PGC) 191 with breast cancer (P < 5 × 10−8) No causal association between genetic instruments for breast cancer and SCZ 122 977 breast cancer cases and 105 974 controls Tomioka et al., 2018 1 SNP associated with serum pyridoxal levels (P = 0.006) IVW No No causal association between genetic instruments for serum pyridoxal level and SCZ 365 SCZ cases 911 controls Polimanti et al. 2018 10 SNP associated with T2D (P < 5 × 10−8) Instrumental variable analysis Yes No bidirectional causal associations between SCZ and T2D 34 840 T2D cases and 114 981 controls (DIAGRAM consortium) 108 SNP associated with SCZ (P < 5 × 10−8) 34 241 SCZ cases and 45 604 controls (PGC) Arafat and Minica, 2018 60 SNP associated with birth weight (P < 5 × 10−8) IVW Yes No causal association between genetic instruments for birth weight and SCZ 34 241 SCZ cases 45 604 controls (PGC) Li et al., 2018 13 SNP associated with fasting insulin levels IVW Yes No causal effect of any of the genetic instruments on SCZ risk, nor in Europeans nor in East Asians, in the BMI-adjusted analysis 84 514 SCZ cases 30 with fasting glucose in Europeans, 14 in East Asians 126 949 controls (PGC2, BIOX) 36 with HbA1c in Europeans, 27 in East Asians 140 with T2D (for Europeans, East Asians and trans-ancestry groups) (P < 5 × 10−8) Hartwig et al., 2017 18 SNP associated with CRP serum levels IVW Yes Genetic predisposition to higher serum CRP causally associated with lower SCZ risk (OR, 0.90; 95% CI, 0.84-0.97; P = 0.005) 36 989 SCZ cases 1 associated with higher sIL-6R serum levels Genetic predisposition to higher serum sIL-6R causally associated with higher SCZ risk (OR, 1.06; 95% CI, 1.01-1.12; P = 0.02) 2 associated with higher IL-1Ra serum levels (P < 5 × 10−5) No causal association between genetic instruments for IL-1Ra and SCZ risk 113 075 controls (PGC) Gage et al., 2017a 128 SNP associated with SCZ risk (P < 5 × 10−8) IVW Yes Genetic liability to SCZ causally associated with risk for cannabis initiation (OR, 1.10 per doubling of the odds of SCZ; 95% CI, 1.05–1.14; P = 2.64 × 10−5) 36 989 SCZ cases 21 with cannabis initiation (P < 10−5) Genetic predisposition to cannabis initiation causally associated with SCZ risk (OR, 1.04 per doubling odds of cannabis initiation; 95% CI, 1.01–1.07; P = 0.019) 113 075 controls (PGC) Gage et al., 2017b 21 SNP associated with smoking initiation (P < 10−6) IVW Yes Genetic predisposition to smoking initiation causally associated with SCZ risk (OR, 1.73; 95% CI, 1.30–2.25, P < 0.001) 36 989 SCZ cases 94 with SCZ risk (P < 5 × 10−8) No causal association between genetic instruments for SCZ and risk for smoking initiation (OR, 1.01; 95% CI, 0.98–1.04; P = 0.32). 113 075 controls (PGC) 35 845 smokers (TAG consortium) Taylor et al., 2016 4 SNP associated with serum vitamin D levels (P < 6 × 10−10) IVW Yes No causal association between genetic instruments for higher serum vitamin D and SCZ risk. 36 989 SCZ cases 113 075 controls (PGC) Prins et al., 2016 18 SNP associated with CRP levels (P < 1 × 10−4) GRS IV Yes Genetic predisposition to higher serum CRP causally associated with reduced SCZ risk (per 10-s% increase in CRP level; OR, 0.86; 95% CI, 0.79–0.94; P < 0.0010) 34 241 SCZ cases 45 604 controls (PGC) Inoshita et al., 2016 15 SNP associated with CRP serum levels (P < 5 × 10−8) IVW Yes Genetic predisposition to higher serum CRP causally associated with SCZ risk (OR, 1.10; 95% CI, 1.02–1.19; P = 0.015) 418 SCZ cases 1365 controls Wium-Andersen et al., 2015a 1 SNP associated with smoking intensity IVW No Genetic predisposition to higher smoking intensity causally associated with SCZ risk (OR, 1.06; 95% CI, 1.04–1.08) in ever- and never-smokers combined, but not in each group alone. 67 SCZ cases 40 014 ever-smokers and 23 282 never-smokers

BCAC, Breast Cancer Association Consortium; BIOX, Bio-X Institutes; CI, confidence interval; CRP, C-reactive protein; CUD, cannabis use disorder; HGF, hepatocyte growth factor; CTACK, cutaneous T-cell attracting chemokine; GRS IV, Genetic risk score instrumental variable; (GS)MR, (Generalized Summary-based) Mendelian randomization; (s)IL-R, (soluble) interleukin receptor; IVW, inverse variance–weighted; MCP3, monocyte-specific chemokine 3; MVMR, multivariable Mendelian randomization; OR, odds ration; PGC, psychiatric genomic consortium; RANTES, regulated on Activation; SCZ, schizophrenia; SE, standard error; SNP, single-nucleotide polymorphisms; s%, symmetric percentage; TAG, Tobacco and Genetics Consortium; TRAIL, TNF-related apoptosis-inducing ligand; T2D, type 2 diabetes; WHRadjBMI, waist-to-hip ratio adjusted for BMI; WM, weighted median.

Contradictory findings exist on the association between smoking and risk for SCZ: while Chen et al. (2021) did not find any causal association between their genetic instrument for smoking and SCZ risk, two other studies with more numerous sample sizes did show that genetic predisposition to smoking was causally associated with higher SCZ risk, but not vice versa (Gage et al., 2017b; Wootton et al., 2020). On the other hand, there may be a bidirectional causal association between cannabis use and SCZ that is genetic risk for SCZ is causally associated with cannabis use and vice versa (Gage et al., 2017a; Pasman et al., 2018).

Metabolic syndrome is a well-known comorbidity of patients with SCZ, but this is likely due to antipsychotic use. In fact, no causal relationship has been highlighted between genetic instruments for metabolic traits [e.g. insulin resistance (Li et al., 2018; Perry et al., 2021b), BMI (Peters et al., 2020), or diabetes (Polimanti et al., 2018)] and SCZ risk.

Findings on the link between the immune system and SCZ risk are contrasting (Prins et al., 2016; Hartwig et al., 2017; Chen et al., 2022). Two papers studying partially overlapping samples from the Psychiatric Genomics Consortium (PGC) highlighted that a genetic predisposition for a higher serum C-reactive protein (CRP) was causally associated with reduced SCZ risk (Prins et al., 2016; Hartwig et al., 2017), whereas a study on a smaller sample size reported opposite findings (Inoshita et al., 2016). Studying systemic inflammation in SCZ is particularly relevant considering that two recent studies showed causal links between genetic predisposition to altered gut microbiota composition and SCZ risk (Zhuang et al., 2020; Ni et al., 2022), and it is known that microbiota composition might influence systemic immunity (Lo et al., 2021). However, the classes of gut bacteria identified as risk factors for SCZ differ between these two studies [i.e. actinobacteria by Ni et al. (2022) and gammaproteobacteria by Zhuang et al. (2020)].

Interestingly, genetic liability to other psychiatric disorders (i.e. anxiety and neuroticism) was causally associated with higher SCZ risk (Jones et al., 2020), whereas this was not true for insomnia (Gao et al., 2019).

Finally, two works (Byrne et al., 2019; Kim et al., 2020) found that genetic liability to SCZ was causally associated with an increased risk for developing breast cancer (but not vice versa), in partially overlapping populations.

Mendelian randomization in major depressive disorder

We included 15 studies on MR in MDD (Table 2) including data from populations ranging between 500 and 143 265 MDD patients (Wium-Andersen et al., 2014; Hung et al., 2014; Wium-Andersen et al., 2015a; Wium-Andersen et al., 2015b; Kwok et al., 2016; Sequeira et al., 2017; Clarke et al., 2017

留言 (0)

沒有登入
gif