A meta-analysis of polygenic risk scores for mood disorders, neuroticism, and schizophrenia in
antidepressant response
Giuseppe Fanelli
1,2
, Katharina Domschke
3
, Alessandra Minelli
4,5
, Massimo Gennarelli
4,5
, Paolo
Martini
4
, Marco Bortolomasi
6
, Eduard Maron
7,8,9
, Alessio Squassina
10
, Siegfried Kasper
11
, Joseph
Zohar
12
, Daniel Souery
13
, Stuart Montgomery
14
, Diego Albani
15
, Gianluigi Forloni
15
, Panagiotis
Ferentinos
16
, Dan Rujescu
11
, Julien Mendlewicz
17
, Diana De Ronchi
1
, European College of
Neuropsychopharmacology (ECNP) Pharmacogenomics & Transcriptomics Thematic Working
Group, Bernhard T Baune
18,19,20
, Alessandro Serretti
1
*, Chiara Fabbri
1,21
1
Department of Biomedical and Neuromot or Sciences, University of Bologna, Bol o gna, Italy
2
Department of Human Genetics, Radboud University Medical Center, Donder s Inst itute for Br ain, Cognition and
Behaviour, Nijmegen, The Netherlands
3
Department of Psy chiatry and Psychothera py, Medical Center – University of Freiburg, Faculty of Medicine, Un iversity
of Freiburg, Freiburg, Germany
4
Department of Molecular and Translational Medicine, Univers ity of Br es cia, Bresci a, Ital y
5
Genetics Unit, IR C CS Istit uto Centro San Giovanni di Dio Fate benefrate lli, Brescia, Ital y
6
Ps ychiatr ic Hos pital "Villa Santa Chiara" , Verona, Ital y
7
Department of Psych iatry, University of Tartu, Tartu, Estonia
8
Psychiatric Clinic, Wes t Tallinn Central Hospital, Ta llinn, Estonia
9
Centre for N europs ychopharmacology, Division of Brain Scie nces, Imperial College London, London, UK
10
Department of Biomedical Sciences, University of Cagliari, Cagliari, Ital y
11
Department of Psychiatry and Psychotherapy, Medical University Vienna, Vienna, Austri a
12
Department of Psychia try, Sheba Medical Center, Tel Hashomer, a nd Sac kler School of Med icine, Tel Aviv University,
Tel Hashomer, Isra el
13
Laboratoire de Ps ycho logie M édicale, U niversit é Libre de Br uxelles and Ps y Pluriel, C en tre Européen de Psychol ogie
Médicale, Brussels, Belgium
14
Imperial College School of Medicine, London, UK
15
Laboratory of Biology of Neurodegen erative Disorders, Department of Neuroscience, Istituto di Ricerche
Farmacologic he Mario Negri IR C CS, Milan, Italy
16
Department of Psych iatry, Athens University Medical School, Athens, Greece
17
Unive rsité Libre de Bruxelles, Brussels, Belgium
18
Department of Psychiatry and Psychot herapy, Universit y of Münster, Müns ter, Germany
19
Department of Ps y chiatry, Melbour ne Medical School, University of Melbourn e, Parkville, VIC, Aus tralia
20
The Florey Institute of Neuroscience and M ental Health, The University of Melbo urne, Parkville, VIC, Australia
21
Social, Genetic & Development al Ps ychiatry Centre, Institute of Psychiatry, Psycho logy & Neuroscience, King’s
College London, London, UK
*
Corresponding author:
Alessandro Serretti, MD, PhD
Departm ent of Biom edical and Neuromotor Sciences
Universit y of Bologna
Viale Carlo Pepol i 5, 40123 Bologna, Ital y
Tel +39 051 6584233
Fax +39 051 521030
Mobile +39 320 4269332 +39 347 3024020
E-mail alessandro.serretti@ unibo.it
. CC-BY-NC-ND 4.0 International licenseIt is made available under a
is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)
The copyright holder for this preprint this version posted June 1, 2021. ; https://doi.org/10.1101/2021.05.28.21257812doi: medRxiv preprint
NOTE: This preprint reports new research that has not been certified by peer review and should not be used to guide clinical practice.
Abstract
About two-thirds of patients with major depressive disorder (MDD) fail to achieve symptom
remission after the initial antidepressant treatment. Despite a role of genetic factors was
proven, the specific underpinnings are not fully understood yet. Polygenic risk scores (PRSs),
which summarise the additive effect of multiple risk variants across the genome, might
provide insights into the underlying genetics. This study aims to investigate the poss ible
association of PRSs for bipolar disorder, MDD, neuroticism, and schizophrenia (SCZ) with
antidepressant non-response or non-remission in patients with MDD. PRSs were calculated
at eight genome-wide P-thresholds based on publicly available summary statistics of the
largest genome-wide association studies. Logistic regressions were performed between
PRSs and non-response or non-remission in six European clinical samples, adjusting for age,
sex, baseline symptom severity, recruitment sites, and population stratification. Res ults
were meta-analysed across samples, including up to 3,637 individuals. Bonferroni correction
was applied. In the meta-analysis, no result was significant after Bonferroni correction. The
top result was found for MDD-PRS and non-remission (p=0.004), with patients in the highest
vs. lowest PRS quintile being more likely not to achieve remission (OR=1.5, 95% CI=1.11-
1.98, p=0.007). Nominal associations were also found between MDD-PRS and non-response
(p=0.013), as well as between SCZ-PRS and non-remission (p=0.035). Although PRSs are still
not able to predict non-response or non-remission, our results are in line with previous
works; methodological improvements in PRSs calculation may improve their predictive
performance and have a meaningful role in precision psychiatry.
Keywords
: Polygenic risk scores; Major depressive disorder; Antidepressants;
Pharmacogenomics; Remission; Treatment response.
. CC-BY-NC-ND 4.0 International licenseIt is made available under a
is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)
The copyright holder for this preprint this version posted June 1, 2021. ; https://doi.org/10.1101/2021.05.28.21257812doi: medRxiv preprint
1.
Introduction
Major depressive disorder (MDD) is a common psychiatric condition that is among the
leading causes of disability worldwide, accounting for a 61.1% increase in the number of
disability-adjusted life years (DALYs) over the past two decades (Diseases and Injuries,
2020). Up to 60% of patients with MDD do not respond adequately to the first prescribed
antidepressant, requiring either a dose increase, switching to another antidepressant, or
augmentation with a different pharmacological agent (De Carlo et al., 2016). Remission is
achieved by only 37.5% of patients after six weeks of treatment with first-line
antidepressants, and non-responding patients undergoing several consecutive treatment
steps achieve lower remission and higher relapse rates (Rush et al., 2006). Therefore, early
identification of each patient’s most appropriate treatment might help to reduce the burden
of the disease and the related cost to society.
Several socio-demographic and clinical predictors of non-response or non-remission have
been identified, including older age, longer duration of the depressive episode, greater
severity at baseline, and the presence of anxiety symptoms (Kautzky et al., 2018). Genetic
variability may also contribute, as indicated by a single-nucleotide polymorphism (SNP)-
based heritability of 13.2% for remission, though the specific loci involved were not
identified (Pain et al., 2020).
Polygenic risk scores (PRSs) summarise the additive effect of common genetic risk variants
across the genome, and they have shown promising clinical utility in other fields of medicine
(Natarajan et al., 2017). Interestingly, people with a PRS for coronary artery disease above
the 80
th
percentile were shown to benefit most from statin treatment in terms of
preventing acute cardiac events, with a relative risk reduction of 44% compared to 26%
observed in patients with a lower PRS (Natarajan et al., 2017). Using the same approach, we
found that higher PRSs for schizophrenia (SCZ) in patients with MDD may be associated with
worse response to the first antidepressant treatment, and that individuals having a lower
SCZ-PRS showed higher chances of response when antidepressants were not augmented
with antipsychotics (Fanelli et al., 2021). A positive genetic correlation was also identified
between non-response to antidepressants and neuroticism (NEU), schizotypy, and mood
disorders, suggesting the existence of underlying shared genetics (Wigmore et al., 2020).
In light of these findings, we aimed to extend our previous results to other samples (Fanelli
et al., 2021), through a large meta-analysis of relevant PRSs (MDD, bipolar disorder (BP),
SCZ, NEU) and antidepressant non-response and non-remission in MDD. PRSs could indeed
help to better stratify patients with respect to their chances of response/remission and lead
to the early implementation of second-line treatment strategies.
2.
Methods
2.1.
Target samples
. CC-BY-NC-ND 4.0 International licenseIt is made available under a
is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)
The copyright holder for this preprint this version posted June 1, 2021. ; https://doi.org/10.1101/2021.05.28.21257812doi: medRxiv preprint
2.1.1.
Brescia
This sample included a total of 501 subjects with MDD (Diagnostic and Statistical Manual of
Mental Disorders-IV (DSM-IV) criteria) who had been referred to the “Villa Santa Chiara”
Psychiatric Hospital in Verona, Italy. Diagnosis of unipolar depression was confirmed using
the Structured Clinical Interview for DSM-IV Axis I Disorders (SCID-I). Patients were excluded
if they had met criteria for another primary neuropsychiatric disorder or comorbid eating
disorder, substance/alcohol abuse or dependency. Treatment non-response was defined as
the failure to respond to at least one adequate trial of antidepressants. Genotyping was
performed using the Infinium PsychArray-24 BeadChip or the Infinium Multi-Ethnic
Genotyping Array (N=215 and 286, respectively). Additional information is a vailable
elsewhere (Minelli et al., 2015).
2.1.2.
European Group for the Study of Resistant Depression (GSRD
)
The sample included 1,346 genotyped patients (Infinium PsychArray-24 BeadChip) w ith
MDD recruited from the European Group for the Study of Resistant Depression (GSRD) as
part of a multicentric study. MDD was diagnosed using the Mini International
Neuropsychiatric Interview (MINI). Patients were excluded if they had met criteria for
another primary psychiatric disorder in the six months prior to enrolment. Treatment
response/remission were determined using the Montgomery-Åsberg Depression Rating
Scale (MADRS) (50% improvement and MADRS ≤10, respectively). Further details are
available elsewhere (Souery et al., 2007). This sample was previously included in a similar
PRS study focused on non-response to the last antidepressant and treatment-resistant
depression (TRD) (Fanelli et al., 2021).
2.1.3.
Münster
It is a naturalistic study of 621 participants aged 18 – 85 years with MDD, as assessed by the
SCID-I. Participants were recruited at the Department of Psychiatry, University of Münster,
Germany (Baune et al., 2010). Patients with SCZ spectrum disorders, BD, current alcohol or
drug dependence, neurological or neurodegenerative illnesses were excluded. Response
and remission at week six were measured using the 21-item Hamilton Depression Rating
Scale (HAMD
21
) (50% improvement and HAMD
21
≤7, respectively). Genotyping was
performed using the Infinium PsychArray-24 BeadChip.
2.1.4.
Sequenced Treatment Alternatives to Relieve Depression (STAR*D)
The STAR*D study was conducted to compare tolerability and efficacy of antidepressants
throughout four sequential treatment levels in patients with MDD of at least moderate
severity. Symptom severity was assessed using the Quick Inventory of Depressive
Symptomatology Clinician-rated scale (QIDS-C
16
) every two weeks; level 1 exit data were
considered for this study. Response/remission were defined as a 50% decrease in symptom
severity and QIDS-C
16
≤5 at week 12, respectively. A total of 1,948 participants were
. CC-BY-NC-ND 4.0 International licenseIt is made available under a
is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)
The copyright holder for this preprint this version posted June 1, 2021. ; https://doi.org/10.1101/2021.05.28.21257812doi: medRxiv preprint
genotyped (Affymetrix GeneChip Human Mapping 500K Array Set or Affymetrix Genome-
Wide Human SNP Array 5.0). The study is described in depth elsewhere (Howland, 2008).
2.1.5.
Tartu
This sample included 83 outpatients with MDD recruited at the Psychiatric Clinic of the
University Hospital of Tartu, Estonia. The diagnosis was made using the MINI 5.0.0,
psychiatric history, and medical records. Patients with other primary neurological or
psychiatric disorders were excluded from the study. Treatment response/remission were
measured using the MADRS, in line with the previous samples. Further information is
available elsewhere (Aluoja et al., 2018). The samples sequenced were genotyped using the
Illumina 370CNV array. Further information is available elsewhere (Tammiste et al., 2013).
2.2.
Quality control of the target datasets
Quality control (QC) and population principal component analysis (PCA) were performed
through the Ricopili pipeline in each of the six target samples separately (Lam et al., 2020).
Single-nucleotide polymorphisms (SNPs) were retained if they had a call rate ≥0.95,
differences in call rates between cases and controls (missing difference) ≤0.02, minor allele
frequency (MAF) ≥0.01, and Hardy-Weinberg equilibrium p-value ≥1e-6. Individuals were
retained if they had an autosomal heterozygosity deviation within ±0.2, call rate ≥0.98, and
no genetic/pedigree sex mismatch.
To assess between-subjects relatedness and population stratification, all pairs of individuals
with identity-by-descent proportion >0.2 were identified using linkage disequilibrium-
pruned data (r
2
<0.2), and one individual from each pair was removed. PCA was used to
determine population stratification (Eigenstrat); population outliers were removed
according to the mean ±6 standard deviations of the first 20 principal components.
Genotype imputation was carried out on the Michigan Imputation Server (Das et al., 2016)
using Minimac4 and the Haplotype Reference Consortium (HRC) r1.1 2016 (GRCh37/hg19).
Post-imputation QC was performed by filtering out variants having a poor imputation
quality score (i.e., R
2
<0.3) and MAF <0.05.
2.3.
Statistical analyses
Polygenic risk scores for BP, MDD, NEU, and SCZ were calculated in each target sample after
hard-calling with a genotype probability threshold of 0.9, using PRSice v2.3.3
(https://prsice.info). Summary statistics of the largest genome-wide association studies
(GWASs) on BP, MDD, NEU, and SCZ available at the time of conducting our analyses were
used as base datasets (Table S1). Clumping was performed to remove SNPs in linkage
disequilibrium (r
2
> 0.1, 250 kb window). Eight
a priori
GWAS P-value thresholds (P
T
) (1e-4,
0.001, 0.05, 0.1, 0.2, 0.3, 0.4, 0.5) were used to select SNPs to be included in each PRS (Choi
et al., 2020).
Logistic regressions between each scaled PRS and the two clinical outcomes (non-response
and non-remission) were conducted in R v4.0.2, adjusting for age, sex, baseline symptom
. CC-BY-NC-ND 4.0 International licenseIt is made available under a
is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)
The copyright holder for this preprint this version posted June 1, 2021. ; https://doi.org/10.1101/2021.05.28.21257812doi: medRxiv preprint