Hostname: page-component-78c5997874-xbtfd Total loading time: 0 Render date: 2024-11-10T04:14:44.419Z Has data issue: false hasContentIssue false

Genome-wide meta-analysis of ascertainment and symptom structures of major depression in case-enriched and community cohorts

Published online by Cambridge University Press:  26 September 2024

Mark J. Adams*
Affiliation:
Division of Psychiatry, University of Edinburgh, Edinburgh, UK
Jackson G. Thorp
Affiliation:
Mental Health and Neuroscience, QIMR Berghofer Medical Research Institute, Brisbane, QLD, Australia
Bradley S. Jermy
Affiliation:
Institute for Molecular Medicine Finland, University of Helsinki, Helsinki, Finland
Alex S. F. Kwong
Affiliation:
Division of Psychiatry, University of Edinburgh, Edinburgh, UK MRC Integrative Epidemiology Unit, University of Bristol, Bristol, UK
Kadri Kõiv
Affiliation:
Estonian Genome Centre, Institute of Genomics, University of Tartu, Tartu, Estonia
Andrew D. Grotzinger
Affiliation:
Department of Psychology and Neuroscience, University of Colorado at Boulder, Boulder, CO, USA Institute for Behavioral Genetics, University of Colorado at Boulder, Boulder, CO, USA
Michel G. Nivard
Affiliation:
Department of Biological Psychology, Vrije Universiteit Amsterdam, Amsterdam, Netherlands
Sally Marshall
Affiliation:
Centre for Genomic & Experimental Medicine, Institute of Genetics and Cancer, University of Edinburgh, Edinburgh, UK
Yuri Milaneschi
Affiliation:
Department of Psychiatry, Amsterdam Public Health and Amsterdam Neuroscience, Amsterdam UMC, Vrije Universiteit Amsterdam, Amsterdam, Netherlands
Bernhard T. Baune
Affiliation:
Department of Psychiatry, University of Melbourne, Melbourne, VIC, Australia Florey Institute of Neuroscience and Mental Health, University of Melbourne, Melbourne, VIC, Australia Department of Psychiatry, University of Münster, Münster, NRW, Germany
Bertram Müller-Myhsok
Affiliation:
Department of Translational Research in Psychiatry, Max Planck Institute of Psychiatry, Munich, BY, Germany Munich Cluster for Systems Neurology (SyNergy), Munich, BY, Germany Institute of Population Health, University of Liverpool, Liverpool, UK
Brenda W. J. H. Penninx
Affiliation:
Department of Psychiatry, Amsterdam Public Health and Amsterdam Neuroscience, Amsterdam UMC, Vrije Universiteit Amsterdam, Amsterdam, Netherlands
Dorret I. Boomsma
Affiliation:
Department of Biological Psychology & Amsterdam Public Health Research Institute, Vrije Universiteit Amsterdam, Amsterdam, Netherlands
Douglas F. Levinson
Affiliation:
Department of Psychiatry & Behavioral Sciences, Stanford University, Stanford, CA, USA
Gerome Breen
Affiliation:
Social, Genetic and Developmental Psychiatry Centre, King's College London, London, UK NIHR Maudsley Biomedical Research Centre, King's College London, London, UK
Giorgio Pistis
Affiliation:
Department of Psychiatry, Lausanne University Hospital and University of Lausanne, Prilly, VD, Switzerland
Hans J. Grabe
Affiliation:
Department of Psychiatry and Psychotherapy, University Medicine Greifswald, Greifswald, MV, Germany
Henning Tiemeier
Affiliation:
Child and Adolescent Psychiatry, Erasmus University Medical Center Rotterdam, Rotterdam, Netherlands Social and Behavioral Science, Harvard T.H. Chan School of Public Health, Boston, MA, USA
Klaus Berger
Affiliation:
Institute of Epidemiology and Social Medicine, University of Münster, Münster, NRW, Germany
Marcella Rietschel
Affiliation:
Department of Genetic Epidemiology in Psychiatry, Central Institute of Mental Health, Medical Faculty Mannheim, Heidelberg University, Mannheim, BW, Germany
Patrik K. Magnusson
Affiliation:
Department of Medical Epidemiology and Biostatistics, Karolinska Institutet, Stockholm, Sweden
Rudolf Uher
Affiliation:
Psychiatry, Dalhousie University, Halifax, NS, Canada
Steven P. Hamilton
Affiliation:
Psychiatry, Kaiser Permanente Northern California, San Francisco, CA, USA
Susanne Lucae
Affiliation:
Max Planck Institute of Psychiatry, Munich, BY, Germany
Kelli Lehto
Affiliation:
Estonian Genome Centre, Institute of Genomics, University of Tartu, Tartu, Estonia
Qingqin S. Li
Affiliation:
Neuroscience Therapeutic Area, Janssen Research and Development, LLC, Titusville, NJ, USA
Enda M. Byrne
Affiliation:
Child Health Research Centre, University of Queensland, Brisbane, QLD, Australia
Ian B. Hickie
Affiliation:
Brain and Mind Centre, University of Sydney, Sydney, NSW, Australia
Nicholas G. Martin
Affiliation:
Mental Health and Neuroscience, QIMR Berghofer Medical Research Institute, Brisbane, QLD, Australia
Sarah E Medland
Affiliation:
Mental Health and Neuroscience, QIMR Berghofer Medical Research Institute, Brisbane, QLD, Australia
Naomi R. Wray
Affiliation:
Institute for Molecular Bioscience, University of Queensland, Brisbane, QLD, Australia Queensland Brain Institute, University of Queensland, Brisbane, QLD, Australia
Elliot M. Tucker-Drob
Affiliation:
Department of Psychology, University of Texas at Austin, Austin, TX, USA Population Research Center, University of Texas at Austin, Austin, TX, USA;
Cathryn M. Lewis
Affiliation:
Social, Genetic and Developmental Psychiatry Centre, King's College London, London, UK Department of Medical & Molecular Genetics, King's College London, London, UK
Andrew M McIntosh
Affiliation:
Division of Psychiatry, University of Edinburgh, Edinburgh, UK Institute for Genomics and Cancer, University of Edinburgh, Edinburgh, UK
Eske M. Derks
Affiliation:
Mental Health and Neuroscience, QIMR Berghofer Medical Research Institute, Brisbane, QLD, Australia
*
Corresponding author: Mark J. Adams; Email: mark.adams@ed.ac.uk
Rights & Permissions [Opens in a new window]

Abstract

Background

Diagnostic criteria for major depressive disorder allow for heterogeneous symptom profiles but genetic analysis of major depressive symptoms has the potential to identify clinical and etiological subtypes. There are several challenges to integrating symptom data from genetically informative cohorts, such as sample size differences between clinical and community cohorts and various patterns of missing data.

Methods

We conducted genome-wide association studies of major depressive symptoms in three cohorts that were enriched for participants with a diagnosis of depression (Psychiatric Genomics Consortium, Australian Genetics of Depression Study, Generation Scotland) and three community cohorts who were not recruited on the basis of diagnosis (Avon Longitudinal Study of Parents and Children, Estonian Biobank, and UK Biobank). We fit a series of confirmatory factor models with factors that accounted for how symptom data was sampled and then compared alternative models with different symptom factors.

Results

The best fitting model had a distinct factor for Appetite/Weight symptoms and an additional measurement factor that accounted for the skip-structure in community cohorts (use of Depression and Anhedonia as gating symptoms).

Conclusion

The results show the importance of assessing the directionality of symptoms (such as hypersomnia versus insomnia) and of accounting for study and measurement design when meta-analyzing genetic association data.

Type
Original Article
Creative Commons
Creative Common License - CCCreative Common License - BY
This is an Open Access article, distributed under the terms of the Creative Commons Attribution licence (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted re-use, distribution and reproduction, provided the original article is properly cited.
Copyright
Copyright © The Author(s), 2024. Published by Cambridge University Press

Introduction

Major depressive disorder (MDD) is a mood disorder characterized by low mood, loss of interest or pleasure (anhedonia), irritable affect, biological symptoms (psychomotor agitation/slowing, altered sleep patterns, changes in appetite or weight), negative thought content, and associated loss of function. To qualify for a diagnosis of major depression, the standard diagnostic classification systems (American Psychiatric Association, 2000, 2013; World Health Organization, 1992) require one of two cardinal symptoms plus at least four other symptoms to be present. Although conceptualized as a single disorder, the diagnostic criteria for MDD can be met with any combination of these other symptoms, which entails the potential of hundreds or thousands of symptom profiles (Fried & Nesse, Reference Fried and Nesse2015a; Zimmerman, Ellison, Young, Chelminski, & Dalrymple, Reference Zimmerman, Ellison, Young, Chelminski and Dalrymple2015). A single categorical phenotype – that might mask a multitude of separate disorder types – stymies the testing of correlates and treatments. However, heterogeneity within the MDD diagnosis does have an upper bound: only around one quarter of the potential symptom profiles are actually observed (Fried & Nesse, Reference Fried and Nesse2015a; Zimmerman et al., Reference Zimmerman, Ellison, Young, Chelminski and Dalrymple2015).

Analyzing individual symptoms is one way to unwrap the heterogeneity of MDD (Cai, Choi, & Fried, Reference Cai, Choi and Fried2020; Fried & Nesse, Reference Fried and Nesse2015b). Phenotypic studies have derived and tested factor structures of MDD symptoms (Elhai et al., Reference Elhai, Contractor, Tamburrino, Fine, Prescott, Shirley and Calabrese2012; Krause, Bombardier, & Carter, Reference Krause, Bombardier and Carter2008; Krause, Reed, & McArdle, Reference Krause, Reed and McArdle2010) and twin models have been used to separate genetic from environmental sources of symptom covariance (Kendler, Aggen, & Neale, Reference Kendler, Aggen and Neale2013) and identify the low genetic concordance between symptoms assessed inside and outside of a depressive episode (Kendler & Aggen, Reference Kendler and Aggen2023). These models grouped symptoms together in two or three factors, which broadly contrast psychological v. somatic symptoms. Clinical subtypes are also part of diagnostic criteria and these have been used to classify depression profiles that are differentially associated with specific clinical, behavioral, and biological correlates (Milaneschi, Lamers, Berk, & Penninx, Reference Milaneschi, Lamers, Berk and Penninx2020; Penninx, Milaneschi, Lamers, & Vogelzangs, Reference Penninx, Milaneschi, Lamers and Vogelzangs2013). The context of symptom expression is an additional part of heterogeneity. For example, symptoms like sleep changes can have many causes unrelated to depression.

More recently, genetic studies of depressive symptoms have updated the findings from twin models using data from genome-wide association studies (GWAS). A confirmatory factor analysis of genetic covariance estimates obtained from GWAS results on current depressive symptoms showed that a psychological and somatic factor had the best fit to the data (Thorp et al., Reference Thorp, Marees, Ong, An, MacGregor and Derks2020). The detection of genetic correlates specific to each symptom implies that symptoms may have differing genetic causes and consequences, even if the symptoms themselves are highly genetically correlated.

Understanding the genetic architecture of MDD symptoms is complicated by symptom ascertainment. In clinically ascertained samples, symptom data is often only available on affected participants, and is thus conditioned on having been diagnosed with depression. Conditioning data presence on a diagnosis can induce downward bias in correlations amongst the symptoms comprising that diagnosis, removing any shared genetic component. In community and biobank cohorts, participants are typically screened for the presence of cardinal symptoms (depressed mood and anhedonia) and only participants who report at least one cardinal symptom are assessed for other symptoms of depression, which also leads to high levels of missing symptom data in these cohorts. Because community samples often contain symptom but not diagnostic information, many GWAS purporting to investigate MDD may actually be better characterized as investigating a broader dysphoria continuum rather than MDD specifically (Flint, Reference Flint2023). However, the use of cardinal symptom screening also potentially enhances the suitability of community cohorts to add to the understanding of non-cardinal symptom dimensions in the context of depression (Huang et al., Reference Huang, Tang, Rietkerk, Appadurai, Krebs, Schork and Cai2023).

In this study we sought to uncover the genetic structure of depression symptoms while accounting for how samples were recruited and how symptoms were assessed. We did this by conducting GWAS of individual symptoms of depression, testing factor models to investigate genetic heterogeneity as a function of sample ascertainment (Case v. Community cohorts) and measurement (with or without screening based on cardinal/gating symptoms). Finally, we assessed the validity of the identified latent factors of depression by estimating genetic correlations with external traits.

Specifically, we conducted GWAS of symptom data in six cohorts and meta-analyzed them in groups based on sample ascertainment. The first group (the ‘Case-enriched’ cohorts) consisted of clinical cases from the Psychiatric Genomics Consortium MDD cohorts, participants from the Australian Genetics of Depression study who were recruited based on depression diagnosis, and participants from Generation Scotland who met DSM criteria for depression. The second group (the ‘Community’ cohorts) consisted of the Avon Longitudinal Study of Parents and Children, Estonian Biobank, and UK Biobank, and thus contained data on participants who were not recruited with respect to depression status. Using the two sets of meta-analyzed symptom GWASs, we tested factor models that accounted for how the samples were ascertained (Case v. Community) and how symptoms were assessed (with or without skip structure based on cardinal symptoms). After understanding the measurements structure of the symptom GWASs, we then compared alternative factor models for the symptoms based on previous literature and diagnostic specifiers for depressive disorders. Using the best fitting overall models, we tested for shared and specific genetic correlates with other psychiatric, behavioral, and metabolic phenotypes that have known genetic links to MDD.

Methods

Samples and assessments of depression symptoms

We analyzed depression symptom data in six studies: the Psychiatric Genomics Consortium (PGC) (Major Depressive Disorder Working Group of the Psychiatric GWAS Consortium, 2013; Wray et al., Reference Wray, Ripke, Mattheisen, Trzaskowski, Byrne, Abdellaoui and Sullivan2018), the Australian Genetics of Depression Study (AGDS) (Byrne et al., Reference Byrne, Kirk, Medland, McGrath, Colodro-Conde, Parker and Martin2020; Mitchell et al., Reference Mitchell, Campos, Whiteman, Olsen, Gordon, Walker and Byrne2022), Generation Scotland: Scottish Family Health Study (GS:SFHS) (Smith et al., Reference Smith, Campbell, Linksted, Fitzpatrick, Jackson, Kerr and McGilchrist2012), the Avon Longitudinal Study of Parents and Children (ASLPAC) (Boyd et al., Reference Boyd, Golding, Macleod, Lawlor, Fraser, Henderson and Davey Smith2013; Fraser et al., Reference Fraser, Macdonald-Wallis, Tilling, Boyd, Golding, Davey Smith and Lawlor2013), Estonian Biobank (EstBB) (Leitsalu et al., Reference Leitsalu, Haller, Esko, Tammesoo, Alavere, Snieder and Metspalu2015), and UK Biobank (UKB) (Sudlow et al., Reference Sudlow, Gallacher, Allen, Beral, Burton, Danesh and Collins2015). We selected participants from the PGC and GS:SHFS cohorts who met DSM criteria for MDD based on structured diagnostic interviews or clinical assessments of their current or lifetime worst episode. Participants from AGDS were recruited based on history of receiving treatment for depression and were assessed for symptoms during their worst episode using an online questionnaire. The PGC, GS:SHFS, and AGDS samples were enriched for depression cases and were grouped together as ‘Case-enriched’ cohorts. In ALSPAC, current depressive symptoms were prospectively collected by interview in the original children sample. In EstBB and UKB, depression symptoms from worst episode were assessed retrospectively using online surveys. Symptom data in these two cohorts had a skip-structure, where all participants were asked about mood and anhedonia symptoms while only participants who endorsed at least one cardinal symptom were asked about the other DSM symptoms. In addition, in UKB we also used retrospective assessments of prolonged low mood and/or anhedonia from the Touchscreen questionnaire. Data from ALSPAC, EstBB, and UKB samples were included regardless of depression diagnosis and were grouped together as ‘Community cohorts’. Table 1 describes the effective sample size of number of participants with each symptom for each grouping of studies that were analyzed. Effective sample size was calculated within each study as N Eff = 4/(1/N Cases + 1/N Controls) and then summed to get total effective sample size for each meta-analysis (Grotzinger, de la Fuente, Privé, Nivard, & Tucker-Drob, Reference Grotzinger, de la Fuente, Privé, Nivard and Tucker-Drob2022). See Supplementary Material for additional information on study design, phenotyping, genotyping, and imputation.

Table 1. Effective sample size of number of participants with each symptom and symptom prevalences of genome-wide association studies

Case-enriched (PGC, AGDS, GenScot) and Community (ALSPAC, EstBB, UKB-MHQ) cohorts meta-analyses and UKB Touchscreen.

Genome-wide association symptom meta-analysis

Genome-wide association study (GWAS) analyses were conducted on each symptom separately in the cohorts (PGC, AGDS, GS:SFHS, ALSPAC, EstBB, UKB-Mental Health Questionnaire) on participants who had genetic similarity with each other and with the 1000 Genomes European reference. Participants in UKB who clustered with other reference populations were not analyzed because sample sizes did not meet the threshold for LD score estimation (N > 5000). We also included GWAS of separate measures of depressed mood and anhedonia symptoms from the UK Biobank (UKB-Touchscreen). See Supplementary Material for more information on the individual study GWASs. We meta-analyzed the GWAS summary statistics for each symptom into ‘Case-enriched’ (PGC, AGDS, and GS:SFHS) and ‘Community’ (ALSPAC, EstBB, and UKB-MHQ) groups. We performed the meta-analyses using Ricopili (Lam et al., Reference Lam, Awasthi, Watson, Goldstein, Panagiotaropoulou, Trubetskoy and Ripke2020) and calculated SNP-based heritability using LD Score Regression (LDSC) (Bulik-Sullivan et al., Reference Bulik-Sullivan, Loh, Finucane, Ripke, Yang and Neale2015). We assessed significant associations in the meta-analyzed summary statistics at p < 5 × 10−8/22 (the number of meta-analyses conducted) or at p < 5 × 10−8 with prior association or biological evidence at the locus.

Confirmatory factor analysis of genetic covariance structure

We fit confirmatory genetic factor analysis models to the meta-analyzed cohort (i.e. Case-enriched and Community) and UKB Touchscreen summary statistics for each symptom using Genomic SEM (Grotzinger et al., Reference Grotzinger, Rhemtulla, de Vlaming, Ritchie, Mallard, Hill and Tucker-Drob2019). This method uses LDSC to estimate genetic variances of and covariances among all the summary statistics. It then uses this matrix to condition structural equation models fit in lavaan (Rosseel, Reference Rosseel2012). We first fit a common factor model, where all symptoms load on a single factor as a baseline, using symptoms with a non-negative LDSC heritability (Model ‘Depr’). To explore how sample ascertainment influenced the genetic correlations among the symptoms, we fit a series of models that captured various aspects of the sampling, measurement, and missing data processes. We then used these results to inform the construction of models that grouped the symptoms based on previous findings and diagnostic criteria. We assessed relative model fit using Akaike Information Criterion (AIC) to pick the best model and absolute model fit with Standardized Root Mean Square Residual (SRMR) to determine how well the model was capturing the genetic correlations among symptoms. We examined residual correlations to understand what aspects of symptom structure were not being captured. Factor structures are listed in online Supplementary Table S4 and illustrated in Fig. 2 and online Supplementary Figure S1.

Ascertainment/measurement models

The most pertinent measurement difference among the symptoms was based on the type of recruitment, so we created a two-factor model where all symptoms from the same cohorts (Case-enriched or Community) loaded together (Model ‘Case-Comm’). The next model considered the effect of the cardinal symptoms as gating items responsible for missing data patterns in UK Biobank and posited a general MDD factor that all the symptoms loaded on alongside an uncorrelated Gating factor with loadings from just the Community and UKB Touchscreen low mood and anhedonia symptoms (Model ‘Depr-Gate’). The Gating factor would therefore isolate variation associated with differences across the full non-clinical to clinical (dysphoria) spectrum. Symptoms not loading on the gating factor (i.e. those for which data are conditional on the presence of the two gating symptoms) represent variation within the more severe region of the spectrum and are thus more directly comparable to analyses of data from cases only. We then combined the Case-Community and Gating models to create a three-factor model (Model ‘Case-Comm-Gate’).

Symptom models

Based off the best measurement model, we then fit models that grouped symptoms into two or three factors based on previous findings from phenotypic, twin, and Genomic SEM models; and from diagnostic criteria. The first set of models grouped symptoms into Psychological and Somatic (Model ‘Psyc-Soma’); Psychological and Neurovegetative (Model ‘Psyc-Neur’); or Affective and Neurovegetative (Model ‘Affc-Neur’) factors (Elhai et al., Reference Elhai, Contractor, Tamburrino, Fine, Prescott, Shirley and Calabrese2012; Krause et al., Reference Krause, Bombardier and Carter2008, Reference Krause, Reed and McArdle2010; Thorp et al., Reference Thorp, Marees, Ong, An, MacGregor and Derks2020). We further tested a model (Model ‘Cog-Mood-Neur’) with cognitive, mood, and neurovegetative symptom factors (Kendler et al., Reference Kendler, Aggen and Neale2013).

We also fit factor models that disaggregated symptoms that involved an increasing or decreasing change (appetite/weight, sleep, or psychomotor). One such model (Model ‘CogMood-App-Leth’) was based on previous findings that identified factors for Cognitive/Mood, Appetite, and Lethargy symptoms (van Loo, Aggen, & Kendler, Reference van Loo, Aggen and Kendler2022). Finally, we considered a three-factor model (Model ‘AffCog-Melc-Atyp’) based on diagnostic criteria of melancholic and atypical depression, with the remaining symptoms loading on an Affective/Cognitive factor.

Genetic multivariable regression

Using the best fitting symptom model, we tested how the factors were related to correlates of depression. We selected phenotypes that are known to genetically correlate with depression, including psychiatric disorders (anxiety disorder, bipolar disorder, PTSD, schizophrenia); depression defined through clinical ascertainment (MDD) and through broader or more minimal definitions (major depression); and other health, behavioral, and social phenotypes (see Supplementary Materials for list of studies). We tested whether the other phenotypes genetic correlations with each symptom factor changed after adjusting for the other factors. We did this by first fitting single regressions of a phenotype on each symptom factor. We then compared this to a multivariable regression of the phenotype on all symptom factors simultaneously. We used Benjamini–Yekutieli FDR adjustment to correct for multiple testing (Benjamini & Yekutieli, Reference Benjamini and Yekutieli2001).

Results

Genome-wide association and meta-analyses

We conducted GWAS for each symptom separately in all cohorts and meta-analyzed within sample ascertainment groups (Case-enriched cohorts: PGC, AGDS, GS:SFHS; Community cohorts: ALSPAC, EstBB, UKB-MHQ). We also supplemented the symptoms data with GWAS of cardinal symptoms collected at baseline in UKB (UKB-Touchscreen). (Table 1 and online Supplementary Table S1). Two associations met the stringent multiple testing burden (p < 5 × 10−8/22 = 2.27 × 10−9). One (rs1421085, p = 1.97 × 10−16) was an intron in FTO (ENSG00000140718, alpha-ketoglutarate dependent dioxygenase, a gene involved in food intake) associated with Weight gain in the Community cohorts. The other (rs30266, p = 1.94 × 10−9) was associated with Anhedonia in the Community cohorts and was an intron variant in an uncharacterized non-coding RNA gene (LOC105379109/ENSG00000251574) and previously associated with depression (Howard et al., Reference Howard, Adams, Clarke, Hafferty, Gibson and Shirali2019), and loneliness (Day, Ong, & Perry, Reference Day, Ong and Perry2018) (online Supplementary Table S2).

There were three additional associations that were supported by prior association studies and met the genome-wide significance threshold (p < 5 × 10−8). Two of the associations were with Depressed mood in the Community cohorts: rs55780333 (p = 1.78 × 10−8), an intron in COMP (ENSG00000105664, cartilage oligomeric matrix protein) also near CRTC1 (ENSG00000105662, CREB regulated transcription coactivator 1), a gene that regulates metabolism and results in social withdrawal behaviors when knocked out in a mouse model (Breuillaud et al., Reference Breuillaud, Rossetti, Meylan, Mérinat, Halfon, Magistretti and Cardinaux2012); and rs28665026 (p = 2.13 × 10−8) in an intron in an uncharacterized gene (LOC107986777) and associated with schizophrenia (Trubetskoy et al., Reference Trubetskoy, Pardiñas, Qi, Panagiotaropoulou, Awasthi, Bigdeli and Bertolino2022). An upstream variant (rs6884321, p = 4.27 × 10−8) for an uncharacterized long intergenic non-protein coding RNA (LINC01938) was associated with Community Anhedonia while this region was previously associated with neuroticism and MDD (Turley et al., Reference Turley, Walters, Maghzian, Okbay, Lee, Fontana and Benjamin2018).

LDSC-estimated heritabilities were primarily in the 0.025–0.1 range. Many of the symptoms in the Case-enriched cohorts had negative heritabilities which is potentially an indication of inadequate power due to low variation. The psychomotor symptoms from the Community cohorts did not meet the sample size inclusion criteria (NEff > 5000). Out of the 12 total symptoms (taking into account directionality), 8/12 from Case-enriched meta-analysis and 10/12 from the Community meta-analysis were taken forward. The two additional cardinal symptoms from the UKB Touchscreen sample also met inclusion criteria.

Confirmatory factor analysis

We brought forward symptoms from the Case-enriched and Community cohorts’ meta-analyses and the UKB Touchscreen assessment that had a $h_{SNP}^2$ greater than 0 and sample sizes >5000 (Fig. 1, online Supplementary Table S3) for confirmatory factor analysis (Fig. 2, online Supplementary Tables S4-6, Figures S1a–n).

Figure 1. LDSC-estimated heritabilities.

Heritably ($h_{SNP}^2$) calculated on the liability scale for summary statistics that met inclusion criteria (NEff > 5000, $h_{SNP}^2$ > 0). Depression symptoms abbreviations are listed in Table 1. Case-enriched = PGC + AGDS + GS:SFHS meta-analysis, Community = ALSPAC + EstBB + UKB-MHQ meta-analysis, UK Biobank = UKB-Touchscreen GWAS.

Figure 2. Structure and loadings of confirmatory factor models.

Points representing loadings of each symptom (columns) onto each factor (rows) for confirmatory models and for the multivariate meta-analysis of well-powered GWASs to illustrate model structure, for Case-enriched (red), Community (green), and UKB Touchscreen (blue) GWASs. Size of points scaled to absolute value of factor loadings. Symptoms arranged in order so that symptoms (Affective/cognitive: Sui, Dep, :Anh, Guilt, Conc; typical somatic: MotoInc, SleDec, AppDec; and atypical somatic: AppInc, MotoDec, Fatig, SleInc) that tend to load onto the same factor are listed next to each other.

A common factor model (‘Depr’) of the symptoms showed poor fit (CI, 0.932, SMR = 0.169, AIC = 5355). A model (‘Case-Comm’) with separate factors for Case-enriched and Community cohort symptoms had slightly poorer fit (AIC = 5369) and yielded a genetic correlation between the two factors of rg = 0.63 ± 0.14, p = 1.3 × 10−5. An alternative model (‘Depr-Gate’) that only split off the Community and UKB-Touchscreen Mood and Anhedonia symptoms into an orthogonal factor, capturing these symptoms use as gating items in EstBB and UKB-MHQ showed substantially improved fit (AIC = 3317). A model (‘Case-Comm-Gate’) combining the sample factors with the orthogonal Gating factor showed slightly poorer fit (AIC = 3375) compared with ‘Case-Comm’ model. Therefore, we investigated the factor structure of MDD symptoms and included a gating factor accounting for symptom skip-structure in subsequent analyses.

We tested whether models that grouped symptoms together across cohorts fit better than the factor models based on sampling methodology. The best fitting of the symptom models was the ‘CogMood-App-Leth’ model which included factors capturing Cognitive/Mood (Depressed mood, Anhedonia, Feelings of guilt, Insomnia, Psychomotor agitation, and Suicidality), Appetite (Appetite/Weight increase and decrease), and Lethargy (Psychomotor slowing, Hypersomnia, Fatigue) symptoms. Because of a high correlation (rg = 0.91) between the Cognitive/Mood and Lethargy factors, we made a model that merged these two factors (model ‘CogMoodLeth-App’; CFI = 0.968; SRMR = 0.147). The correlation between the Cognitive/Mood/Lethargy and the Appetite factors was rg = 0.53 and we brought this model forward for subsequent analysis (Fig. 3).

Figure 3. Model structural diagram.

Standardized loadings (standard errors) of factors on symptoms and genetic correlations among factors for the model (CogMoodLeth-App) used for further analysis. Symptom abbreviations are listed in Table 1.

An inspection of the residual genetic correlations (online Supplementary Figure S3) indicated correlations between the same symptoms across the two cohorts (e.g. Case cohorts Appetite decrease with Community cohorts Appetite decrease) were not fully represented by the factor structure. We thus tested how adding residual correlations between symptoms that were well-powered enough to have been included from both cohorts (Appetite decrease, Appetite increase, Insomnia, Hypersomnia, and Suicidality) improved absolute model fit (model ‘CogMood-App-Leth [Res]’). The addition of these residual correlations lowered SRMR to 0.139 (online Supplementary Figure S1N).

Genetic multivariable regression

We used genetic multivariable regression to test the genetic correlations of each MDD symptom factor with twelve clinically relevant phenotypes, using genome-wide summary statistics. For each external phenotype, we used Genomic SEM to fit single regressions of the phenotype onto each MDD symptom factor separately. We then fitted a multiple regression of each phenotype onto all factors to test whether a phenotype's association with each factor changed after adjusting for the other factors. We tested phenotypes against the Cognitive/Mood/Lethargy, Appetite, and Gating factors (model ‘CogMoodLeth-App’).

In the single regression (unadjusted) analysis, the genetic relationship of each phenotype with all the factors were significant and in the same direction, apart from educational attainment which had a negative relationship with most of the factors (at p < 0.0005) but a positive yet non-significant relationship with the Gating factor (Fig. 4, online Supplementary Table S7). When adjusting for the Cognitive/Mood/Lethargy and Gating factors, Appetite symptoms factor had a larger magnitude genetic correlation with BMI and educational attainment and an unchanged correlation with smoking. After adjustment, the genetic correlation of the Appetite factor with the other phenotypes was close to 0, with the exception of pain, which decreased only slightly. Adjusting for the Cognitive/Mood/Lethargy factor for the other factors did not change its genetic correlation with alcohol dependence, anxiety, bipolar disorder, major depression and MDD, neuroticism, PTSD, or long sleep duration. Genetic correlations for the Gating factor were mostly attenuated (decreasing substantially or going to 0), except that it increased for educational attainment and flipped sign for BMI.

Figure 4. Genetic multivariable regression.

(a) Model diagrams for single regressions and (b) multiple regressions of a phenotype Y on Appetite/Weight, Cognitive/Mood/Lethargy, and Gating symptom factors (symptom indicator variables omitted for clarity). (c) Single genetic regression standardized beta coefficients (green triangles) and multiple genetic regression (red circles) coefficients (point estimates plotted with 95% confidence intervals). FDR correction indicated for significant (darker shading) and non-significant (lighter shading) coefficients. Multiple regression models adjust for the other factors. AlcDep, alcohol dependence; Anxiety, anxiety disorder; BIP, bipolar disorder; BMI, body-mass index; EA, educational attainment; MD, major depression; MDD, major depressive disorder; Neu, neuroticism; Pain, chronic pain; PTSD, post-traumatic stress disorder; Sleep, long sleep duration; Smoking, cigarettes per day.

Discussion

We used genome-wide association data to analyze the genetic relationships among symptoms of depression based on cohort sampling and symptom content and to estimate whether the genetic factors had specific correlates with other phenotypes. We analyzed data from two sets of cohorts: Clinical cohorts that were ascertained to have depression through clinical or interview assessments or were recruited preferentially on a history of treatment for depression; and Community cohorts that were not recruited based on disease status (but for which symptom data was typically conditioned based on endorsement of cardinal gating symptoms). We conducted GWAS of major depression symptoms in each cohort then meta-analyzed within the Clinical and Community groups.

We identified loci associated with individual major depression symptoms and with a multivariate meta-analysis of a subset of well-powered symptom GWASs. Several associations from the individual symptoms meta-analysis (rs7515828, rs30266, s6884321) have been identified previously in GWAS of or unipolar depression (EFO |ID EFO_0003761) (Sollis et al., Reference Sollis, Mosaku, Abid, Buniello, Cerezo, Gil and Harris2023) or in meta-analyses of MDD (Als et al., Reference Als, Kurki, Grove, Voloudakis, Therrien, Tasanko and Børglum2023; Howard et al., Reference Howard, Adams, Clarke, Hafferty, Gibson and Shirali2019; Levey et al., Reference Levey, Stein, Wendt, Pathak, Zhou, Aslan and Gelernter2021; Wray et al., Reference Wray, Ripke, Mattheisen, Trzaskowski, Byrne, Abdellaoui and Sullivan2018). SNPs associated with Appetite / weight increase have primarily come up in GWAS of body mass index and related traits (Elsworth et al., Reference Elsworth, Lyon, Alexander, Liu, Matthews, Hallett and Hemani2020; Hoffmann et al., Reference Hoffmann, Choquet, Yin, Banda, Kvale, Glymour and Jorgenson2018; Howe et al., Reference Howe, Nivard, Morris, Hansen, Rasheed, Cho and Davies2022; Yengo et al., Reference Yengo, Sidorenko, Kemper, Zheng, Wood and Weedon2018) but another SNP in the FTO gene has also been associated with atypical subtypes (Milaneschi et al., Reference Milaneschi, Lamers, Mbarek, Hottenga, Boomsma and Penninx2014).

While the low heritabilities of symptoms from the Case-enriched cohorts limited the comprehensiveness of alternative factor models that could be tested, the best fitting model did not have a separation between Case-enriched and Community cohort symptoms. The lower power in some of the symptom GWASs also do not allow for their inclusion in a multivariate meta-analysis of the ascertainment or symptom factors, as only the psychological symptoms and data from the Community cohorts were sufficiently powered for such an analysis. We also showed that model fit was substantially improved by modeling the use of cardinal symptoms (Low mood and Anhedonia) as gating items for surveys of depression symptoms. Among the models that grouped symptoms together without consideration for symptom direction, such as a split between psychological and somatic symptoms identified in previous phenotypic (Elhai et al., Reference Elhai, Contractor, Tamburrino, Fine, Prescott, Shirley and Calabrese2012) and genetic (Thorp et al., Reference Thorp, Marees, Ong, An, MacGregor and Derks2020) analyses, had worse fit than Case-enriched/Community factor models. When directional symptoms were portioned out based on diagnostic specifiers, we found that a model capturing Cognitive/Mood, Appetite, and Lethargy symptoms (van Loo et al., Reference van Loo, Aggen and Kendler2022) had the best fit among all models considered. The correlations among the factors indicated that the Cognitive/Mood and Lethargy symptoms should be grouped together, with only the Appetite symptoms making up a possibly different dimension of depression.

For the symptoms suitable for inclusion in the models that were available from both sets of cohorts (Appetite, Sleep, Feelings of guilt, and Suicidality), the Case-enriched cohorts contributed between 10 and 35% of the total effective sample size. However, the Case-enriched cohort symptoms had low loadings in both the sample-based and symptom-based models (except for the Case-enriched Appetite/Weight and Suicidality symptoms), and thus the model fit was driven primarily by capturing the structure among the Community cohort symptoms. This observation is consistent with the fact that the Clinical cohorts are more selected than the community cohorts, and that conditioning data presence on a diagnosis can induce downward bias in correlations amongst the symptoms that aggregate to form the diagnosis. Similar attenuation, albeit to a lesser degree, may be expected for items in community samples whose presence was conditioned on endorsement of cardinal symptoms. Like many recent genetic studies of depression, there is thus a need to increase the proportion of severely affected participants included in the analysis and, more specifically for understanding heterogeneity, to score symptoms in such a way as to capture more variation in severity.

A multivariable genetic regression analysis showed discriminative validity between the symptom factors, with the Appetite factor still being genetically correlated with BMI and smoking after adjusting for the other factors, and a similar pattern being observed for the Cognitive/Mood/Lethargy factor with other psychiatric phenotypes. The increase in magnitude of the genetic correlation of BMI and educational attainment with Appetite symptoms combined with the sign flip for the Gating symptom could be a part of study participation bias. However, a positive genetic correlation between increase in appetite/weight with BMI has previously been shown with PGC cohorts (Milaneschi et al., Reference Milaneschi, Lamers, Peyrot, Baune, Breen and Dehghan2017) and in UKB (Badini et al., Reference Badini, Coleman, Hagenaars, Hotopf, Breen, Lewis and Fabbri2022), and our findings show that this result holds even when adjusting for genetic overlap with other symptoms. Yet the reliability of these findings is limited by the poor absolute fit of the models considered, which can be attributed to the proposed models all missing some aspect of the genetic structure as well as to small sample size in some of the contributing GWAS, particularly from the Case-enriched cohorts.

Our results demonstrate the challenges and insights associated with considering symptoms of depression separately. Substantial care must be taken to consider how samples are ascertained (clinical v. community recruitment), how symptoms are measured (the use of gating items in symptom inventories), and whether assessments of item direction (e.g. insomnia v. hypersomnia) are included when modeling the genetic structure of depression symptoms. However, the evaluation of direction was limited to a small subset of symptoms and did not include distinctions such as low v. irritable mood, or included only partial assessments, such as weight but not appetite changes being assessed in UKB. The symptoms also did not cover all diagnostic features of the atypical specifier or other sources of heterogeneity such as onset, life event exposure, or treatment outcomes (Harald & Gordon, Reference Harald and Gordon2012) which may have a differential biological and genetic basis (Beijers, Wardenaar, van Loo, & Schoevers, Reference Beijers, Wardenaar, van Loo and Schoevers2019; Milaneschi et al., Reference Milaneschi, Lamers, Berk and Penninx2020; Nguyen et al., Reference Nguyen, Harder, Xiong, Kowalec, Hägg, Cai and Lu2022). Even the best fitting model that we tested had poor absolute fit, and thus the search for alternative models, girded by more complete data, will continue. The strongest genetic associations were between symptoms of weight/appetite change and genes linked to satiety and metabolism. This highlights the need to phenotype somatic symptoms (weight or sleep changes and fatigue) outside of the context of mental health assessments, so that their specific role in depression can be better isolated, and mirrors the larger need consider how symptoms are expressed inside and outside of a depressive episode (Kendler & Aggen, Reference Kendler and Aggen2023). Likewise, the use of gating symptoms makes it difficult to fully capture the range of genetic risk between everyday dysphoria and differences among affected individuals. While the results support the idea that depression is heterogeneous, the genetic liability for symptom profiles and comorbidities can be captured in relatively few dimensions.

Supplementary material

The supplementary material for this article can be found at https://doi.org/10.1017/S0033291724001880.

Data availability statement

Primary code is available from the PGC GitHub Repository (https://github.com/psychiatric-genomics-consortium/mdd-symptom-gwas/) and meta-analyzed summary statistics are available for download from the PGC website (https://www.med.unc.edu/pgc/download-results/). Individual-level PGC data is available by application to the PGC Data Access Committee (https://www.med.unc.edu/pgc/shared-methods/). Data from Estonian Biobank (https://genomics.ut.ee/en/content/estonian-biobank), UK Biobank (https://www.ukbiobank.ac.uk), and ALSPAC (http://www.bristol.ac.uk/alspac/) are available to bona fide researchers upon application. Data from AGDS is available for collaboration by contacting NGM ().

Acknowledgements

We are extremely grateful to all the families who took part in this study, the midwives for their help in recruiting them, and the whole ALSPAC team, which includes interviewers, computer and laboratory technicians, clerical workers, research scientists, volunteers, managers, receptionists, and nurses. UKB analysis conducted under project 4844. This work made use of the NL Genetic Cluster Computer (http://www.geneticcluster.org) hosted by SURFsara and resources provided by the Edinburgh Compute and Data Facility (ECDF) (http://www.ecdf.ed.ac.uk/). This publication is the work of the authors and MJA will serve as guarantor for the contents of this paper. For the purposes of open access, the author has applied a Creative Commons Attribution 4.0 International Public License (CC BY 4.0) to any Accepted Author Manuscript version arising from this submission.

Funding statement

AGDS supported by National Health and Medical Research Council (NHMRC) (1086683, 1145645, 1078901, 1087889, 1173790). The UK Medical Research Council and Wellcome Trust (217065/Z/19/Z) and the University of Bristol provide core support for ALSPAC; GWAS data was generated by Sample Logistics and Genotyping Facilities at Wellcome Sanger Institute and LabCorp (Laboratory Corporation of America) using support from 23andMe. Estonian Biobank supported by the European Union through the European Regional Development Fund (Project No. 2014-2020.4.01.15-0012). NTR-NESDA supported by Biobanking and Biomolecular Resources Research Infrastructure (BBMRI-NL; 184.021.007 and 184.033.111), National Institutes of Health (NIH, R01D0042157-01A, MH081802, Grand Opportunity grants 1RC2 MH089951 and 1RC2 MH089995). Part of the genotyping and analyses were funded by the Genetic Association Information Network (GAIN) of the Foundation for the National Institutes of Health. Funding for NTR is acknowledged from NWO-MW 904-61-193; NWO 985-10-002; NWO 904-61-090; Royal Netherlands Academy of Science Professor Award (PAH/6635) to DIB; European Research Council (ERC-230374). Funding for the infrastructure of the NESDA study (www.nesda.n) was obtained from the Netherlands Organization for Scientific Research (Geestkracht program grant 10-000-1002); the Center for Medical Systems Biology (CSMB, NWO Genomics), VU University Medical Center, GGZ inGeest, Leiden University Medical Center, Leiden University, GGZ Rivierduinen, University Medical Center Groningen, University of Groningen, Lentis, GGZ Friesland, GGZ Drenthe, Rob Giel Onderzoekscentrum. MJA, ASFK, and AMMc are supported by the Wellcome Trust (104036/Z/14/Z, 220857/Z/20/Z). SEM is supported by NHMRC APP1172917, APP1138514, and MRF1200644. ADG, MGN, and EMTD were supported by National Institute of Mental Health grant R01MH120219. KL and KK were supported by the Estonian Research Council grant PSG615. The PGC has received major funding from the National Institute of Mental Health and the National Institute on Drug Abuse (U01 MH109528, U01 MH109532, U01 MH094421, U01 MH085520). This paper represents independent research part-funded by the NIHR Maudsley Biomedical Research Centre at South London and Maudsley NHS Foundation Trust and King's College London.

Declarations

CL sits on the SAB for Myriad Neuroscience, has received consultancy fees from UCB, and speaker fees from SYNLAB. HJG has received travel grants and speaker's honoraria from Fresenius Medical Care, Neuraxpharm, Servier and Janssen Cilag as well as research funding from Fresenius Medical Care.

The authors assert that all procedures contributing to this work comply with the ethical standards of the relevant national and institutional committees on human experimentation and with the Helsinki Declaration of 1975, as revised in 2008.

Footnotes

*

Andres Metspalu, Lili Milani, Tõnu Esko, Reedik Mägi, Mari Nelis & Georgi Hudjashov

References

Als, T. D., Kurki, M. I., Grove, J., Voloudakis, G., Therrien, K., Tasanko, E., … Børglum, A. D. (2023). Depression pathophysiology, risk prediction of recurrence and comorbid psychiatric disorders using genome-wide analyses. Nature Medicine, 29(7), 18321844. doi:10.1038/s41591-023-02352-1CrossRefGoogle ScholarPubMed
American Psychiatric Association. (2000). Diagnostic and statistical manual of mental disorders: DSM-IV-TR (4th ed., text revision). Washington, DC: American Psychiatric Association.Google Scholar
American Psychiatric Association (2013). Diagnostic and statistical manual of mental disorders: DSM-5 (5th ed.) Washington, D.C: American Psychiatric Association.Google Scholar
Badini, I., Coleman, J. R. I., Hagenaars, S. P., Hotopf, M., Breen, G., Lewis, C. M., & Fabbri, C. (2022). Depression with atypical neurovegetative symptoms shares genetic predisposition with immuno-metabolic traits and alcohol consumption. Psychological Medicine, 52(4), 726736. doi:10.1017/S0033291720002342CrossRefGoogle ScholarPubMed
Beijers, L., Wardenaar, K. J., van Loo, H. M., & Schoevers, R. A. (2019). Data-driven biological subtypes of depression: Systematic review of biological approaches to depression subtyping. Molecular Psychiatry, 24(6), 888900. doi:10.1038/s41380-019-0385-5CrossRefGoogle ScholarPubMed
Benjamini, Y., & Yekutieli, D. (2001). The control of the false discovery rate in multiple testing under dependency. The Annals of Statistics, 29(4), 11651188. doi:10.1214/aos/1013699998CrossRefGoogle Scholar
Boyd, A., Golding, J., Macleod, J., Lawlor, D. A., Fraser, A., Henderson, J., … Davey Smith, G. (2013). Cohort profile: The ‘children of the 90s’—the index offspring of the Avon longitudinal study of parents and children. International Journal of Epidemiology, 42(1), 111127. doi:10.1093/ije/dys064CrossRefGoogle Scholar
Breuillaud, L., Rossetti, C., Meylan, E. M., Mérinat, C., Halfon, O., Magistretti, P. J., & Cardinaux, J.-R. (2012). Deletion of CREB-regulated transcription coactivator 1 induces pathological aggression, depression-related behaviors, and neuroplasticity genes dysregulation in mice. Biological Psychiatry, 72(7), 528536. doi:10.1016/j.biopsych.2012.04.011CrossRefGoogle ScholarPubMed
Bulik-Sullivan, B. K., Loh, P.-R., Finucane, H. K., Ripke, S., Yang, J., Schizophrenia Working Group of the Psychiatric Genomics Consortium, … Neale, B. M. (2015). LD score regression distinguishes confounding from polygenicity in genome-wide association studies. Nature Genetics, 47(3), 291295. doi:10.1038/ng.3211CrossRefGoogle ScholarPubMed
Byrne, E. M., Kirk, K. M., Medland, S. E., McGrath, J. J., Colodro-Conde, L., Parker, R., … Martin, N. G. (2020). Cohort profile: The Australian genetics of depression study. BMJ Open, 10(5), e032580. doi:10.1136/bmjopen-2019-032580CrossRefGoogle ScholarPubMed
Cai, N., Choi, K. W., & Fried, E. I. (2020). Reviewing the genetics of heterogeneity in depression: Operationalizations, manifestations and etiologies. Human Molecular Genetics, 29(R1), R10R18. doi:10.1093/hmg/ddaa115CrossRefGoogle ScholarPubMed
Day, F. R., Ong, K. K., & Perry, J. R. B. (2018). Elucidating the genetic basis of social interaction and isolation. Nature Communications, 9(1), 2457. doi:10.1038/s41467-018-04930-1CrossRefGoogle ScholarPubMed
Elhai, J. D., Contractor, A. A., Tamburrino, M., Fine, T. H., Prescott, M. R., Shirley, E., … Calabrese, J. R. (2012). The factor structure of major depression symptoms: A test of four competing models using the Patient Health Questionnaire-9. Psychiatry Research, 199(3), 169173. doi:10.1016/j.psychres.2012.05.018CrossRefGoogle ScholarPubMed
Elsworth, B., Lyon, M., Alexander, T., Liu, Y., Matthews, P., Hallett, J., … Hemani, G. (2020). The MRC IEU OpenGWAS data infrastructure [preprint]. bioRxiv, 2020.08.10.244293. doi:10.1101/2020.08.10.244293Google Scholar
Flint, J. (2023). The genetic basis of major depressive disorder. Molecular Psychiatry, 28, 22542265. doi:10.1038/s41380-023-01957-9CrossRefGoogle ScholarPubMed
Fraser, A., Macdonald-Wallis, C., Tilling, K., Boyd, A., Golding, J., Davey Smith, G., … Lawlor, D. A. (2013). Cohort profile: The Avon longitudinal study of parents and children: ALSPAC mothers cohort. International Journal of Epidemiology, 42(1), 97110. doi:10.1093/ije/dys066CrossRefGoogle ScholarPubMed
Fried, E. I., & Nesse, R. M. (2015a). Depression is not a consistent syndrome: An investigation of unique symptom patterns in the STAR*D study. Journal of Affective Disorders, 172, 96102. doi:10.1016/j.jad.2014.10.010CrossRefGoogle ScholarPubMed
Fried, E. I., & Nesse, R. M. (2015b). Depression sum-scores don't add up: Why analyzing specific depression symptoms is essential. BMC Medicine, 13(1), 72. doi:10.1186/s12916-015-0325-4CrossRefGoogle ScholarPubMed
Grotzinger, A. D., Rhemtulla, M., de Vlaming, R., Ritchie, S. J., Mallard, T. T., Hill, W. D., … Tucker-Drob, E. M. (2019). Genomic structural equation modelling provides insights into the multivariate genetic architecture of complex traits. Nature Human Behaviour, 3(5), 513525. doi:10.1038/s41562-019-0566-xCrossRefGoogle ScholarPubMed
Grotzinger, A. D., de la Fuente, J., Privé, F., Nivard, M. G., & Tucker-Drob, E. M. (2022). Pervasive downward bias in estimates of liability-scale heritability in genome-wide association study meta-analysis: A simple solution. Biological Psychiatry, 93(1), 2936. doi:10.1016/j.biopsych.2022.05.029CrossRefGoogle ScholarPubMed
Harald, B., & Gordon, P. (2012). Meta-review of depressive subtyping models. Journal of Affective Disorders, 139(2), 126140. doi:10.1016/j.jad.2011.07.015CrossRefGoogle ScholarPubMed
Hoffmann, T. J., Choquet, H., Yin, J., Banda, Y., Kvale, M. N., Glymour, M., … Jorgenson, E. (2018). A large multiethnic genome-wide association study of adult body mass index identifies novel loci. Genetics, 210(2), 499515. doi:10.1534/genetics.118.301479CrossRefGoogle ScholarPubMed
Howe, L. J., Nivard, M. G., Morris, T. T., Hansen, A. F., Rasheed, H., Cho, Y., … Davies, N. M. (2022). Within-sibship genome-wide association analyses decrease bias in estimates of direct genetic effects. Nature Genetics, 54(5), 581592. doi:10.1038/s41588-022-01062-7CrossRefGoogle ScholarPubMed
Huang, L., Tang, S., Rietkerk, J., Appadurai, V., Krebs, M. D., Schork, A. J., … Cai, N. (2023). Polygenic analyses show important differences between MDD symptoms collected using PHQ9 and CIDI-SF. Genetic and Genomic Medicine, 95(12), 11101121. doi:10.1101/2023.02.27.23286527Google Scholar
Kendler, K. S., & Aggen, S. H. (2023). A population-based twin study of the symptomatic diagnostic criteria for major depression that occur within versus outside of major depressive episodes. Psychological Medicine, 53(15), 74587465. doi:10.1017/S0033291723001241CrossRefGoogle ScholarPubMed
Kendler, K. S., Aggen, S. H., & Neale, M. C. (2013). Evidence for multiple genetic factors underlying DSM-IV criteria for major depression. JAMA Psychiatry, 70(6), 599. doi:10.1001/jamapsychiatry.2013.751CrossRefGoogle ScholarPubMed
Krause, J. S., Bombardier, C., & Carter, R. E. (2008). Assessment of depressive symptoms during inpatient rehabilitation for spinal cord injury: Is there an underlying somatic factor when using the PHQ? Rehabilitation Psychology, 53(4), 513520. doi:10.1037/a0013354CrossRefGoogle Scholar
Krause, J. S., Reed, K. S., & McArdle, J. J. (2010). Factor structure and predictive validity of somatic and nonsomatic symptoms from the patient health questionnaire-9: A longitudinal study after spinal cord injury. Archives of Physical Medicine and Rehabilitation, 91(8), 12181224. doi:10.1016/j.apmr.2010.04.015CrossRefGoogle ScholarPubMed
Lam, M., Awasthi, S., Watson, H. J., Goldstein, J., Panagiotaropoulou, G., Trubetskoy, V., … Ripke, S. (2020). RICOPILI: Rapid imputation for COnsortias PIpeLIne. Bioinformatics (Oxford, England), 36(3), 930933. doi:10.1093/bioinformatics/btz633Google ScholarPubMed
Leitsalu, L., Haller, T., Esko, T., Tammesoo, M.-L., Alavere, H., Snieder, H., … Metspalu, A. (2015). Cohort profile: Estonian Biobank of the Estonian Genome Center, University of Tartu. International Journal of Epidemiology, 44(4), 11371147. doi:10.1093/ije/dyt268CrossRefGoogle ScholarPubMed
Levey, D. F., Stein, M. B., Wendt, F. R., Pathak, G. A., Zhou, H., Aslan, M., … Gelernter, J. (2021). Bi-ancestral depression GWAS in the million veteran program and meta-analysis in >1.2 million individuals highlight new therapeutic directions. Nature Neuroscience, 24(7), 954963. doi:10.1038/s41593-021-00860-2CrossRefGoogle Scholar
Howard, D. M., Adams, M. J., Clarke, T.-K., Hafferty, J. D., Gibson, J., Shirali, M., … Major Depressive Disorder Working Group of the Psychiatric Genomics, C. (2019). Genome-wide meta-analysis of depression identifies 102 independent variants and highlights the importance of the prefrontal brain regions. Nature Neuroscience, 22, 343352. doi:10.1038/s41593-018-0326-7CrossRefGoogle ScholarPubMed
Major Depressive Disorder Working Group of the Psychiatric GWAS Consortium (2013). A mega-analysis of genome-wide association studies for major depressive disorder. Molecular Psychiatry, 18, 497511. doi:10.1038/mp.2012.21CrossRefGoogle Scholar
Milaneschi, Y., Lamers, F., Mbarek, H., Hottenga, J.-J., Boomsma, D. I., & Penninx, B. W. J. H. (2014). The effect of FTO rs9939609 on major depression differs across MDD subtypes. Molecular Psychiatry, 19(9), 960962. doi:10.1038/mp.2014.4CrossRefGoogle ScholarPubMed
Milaneschi, Y., Lamers, F., Berk, M., & Penninx, B. W. J. H. (2020). Depression heterogeneity and its biological underpinnings: Toward immunometabolic depression. Biological Psychiatry, 88(5), 369380. doi:10.1016/j.biopsych.2020.01.014CrossRefGoogle ScholarPubMed
Mitchell, B. L., Campos, A. I., Whiteman, D. C., Olsen, C. M., Gordon, S. D., Walker, A. J., … Byrne, E. M. (2022). The Australian genetics of depression study: New risk loci and dissecting heterogeneity between subtypes. Biological Psychiatry, 92(3), 227235. doi:10.1016/j.biopsych.2021.10.021CrossRefGoogle ScholarPubMed
Nguyen, T.-D., Harder, A., Xiong, Y., Kowalec, K., Hägg, S., Cai, N., … Lu, Y. (2022). Genetic heterogeneity and subtypes of major depression. Molecular Psychiatry, 27(3), 16671675. doi:10.1038/s41380-021-01413-6CrossRefGoogle ScholarPubMed
Penninx, B. W. J. H., Milaneschi, Y., Lamers, F., & Vogelzangs, N. (2013). Understanding the somatic consequences of depression: Biological mechanisms and the role of depression symptom profile. BMC Medicine, 11, 129. doi:10.1186/1741-7015-11-129CrossRefGoogle ScholarPubMed
Rosseel, Y. (2012). Lavaan: An R package for structural equation modeling. Journal of Statistical Software, 48(2), 136. doi:10.18637/jss.v048.i02CrossRefGoogle Scholar
Smith, B. H., Campbell, A., Linksted, P., Fitzpatrick, B., Jackson, C., Kerr, S. M., … McGilchrist, M. (2012). Cohort profile: Generation Scotland: Scottish Family Health Study (GS: SFHS). The study, its participants and their potential for genetic research on health and illness. International Journal of Epidemiology, 42(2), 689700.CrossRefGoogle Scholar
Sollis, E., Mosaku, A., Abid, A., Buniello, A., Cerezo, M., Gil, L., … Harris, L. W. (2023). The NHGRI-EBI GWAS Catalog: Knowledgebase and deposition resource. Nucleic Acids Research, 51(D1), D977D985. doi:10.1093/nar/gkac1010CrossRefGoogle ScholarPubMed
Sudlow, C., Gallacher, J., Allen, N., Beral, V., Burton, P., Danesh, J., … Collins, R. (2015). UK biobank: An open access resource for identifying the causes of a wide range of complex diseases of middle and old age. PLoS Medicine, 12(3), e1001779. doi:10.1371/journal.pmed.1001779CrossRefGoogle ScholarPubMed
Yengo, L., Sidorenko, J., Kemper, K. E., Zheng, Z., Wood, A. R., Weedon, M. N., … the GIANT Consortium. (2018). Meta-analysis of genome-wide association studies for height and body mass index in ~700000 individuals of European ancestry. Human Molecular Genetics, 27(20), 36413649. doi:10.1093/hmg/ddy271CrossRefGoogle ScholarPubMed
Milaneschi, Y., Lamers, F., Peyrot, W. J., Baune, B. T., Breen, G., Dehghan, A., … the Major Depressive Disorder Working Group of the Psychiatric Genomics Consortium. (2017). Genetic association of major depression with atypical features and obesity-related immunometabolic dysregulations. JAMA Psychiatry, 74, 12141225. doi:10.1001/jamapsychiatry.2017.3016CrossRefGoogle ScholarPubMed
Thorp, J. G., Marees, A. T., Ong, J.-S., An, J., MacGregor, S., & Derks, E. M. (2020). Genetic heterogeneity in self-reported depressive symptoms identified through genetic analyses of the PHQ-9. Psychological Medicine, 50(14), 23852396. doi:10.1017/S0033291719002526CrossRefGoogle ScholarPubMed
Trubetskoy, V., Pardiñas, A. F., Qi, T., Panagiotaropoulou, G., Awasthi, S., Bigdeli, T. B., … Bertolino, A. (2022). Mapping genomic loci implicates genes and synaptic biology in schizophrenia. Nature, 604(7906), 502508. doi:10.1038/s41586-022-04434-5CrossRefGoogle ScholarPubMed
Turley, P., Walters, R. K., Maghzian, O., Okbay, A., Lee, J. J., Fontana, M. A., … Benjamin, D. J. (2018). Multi-trait analysis of genome-wide association summary statistics using MTAG. Nature Genetics, 50(2), 229237. doi:10.1038/s41588-017-0009-4CrossRefGoogle ScholarPubMed
van Loo, H. M., Aggen, S. H., & Kendler, K. S. (2022). The structure of the symptoms of major depression: Factor analysis of a lifetime worst episode of depressive symptoms in a large general population sample. Journal of Affective Disorders, 307, 115124. doi:10.1016/j.jad.2022.03.064CrossRefGoogle Scholar
World Health Organization (Ed.). (1992). The ICD-10 classification of mental and behavioural disorders: Clinical descriptions and diagnostic guidelines. Geneva: World Health Organization.Google Scholar
Wray, N. R., Ripke, S., Mattheisen, M., Trzaskowski, M., Byrne, E. M., Abdellaoui, A., … Sullivan, P. F. (2018). Genome-wide association analyses identify 44 risk variants and refine the genetic architecture of major depression. Nature Genetics, 50(5), 668681. doi:10.1038/s41588-018-0090-3CrossRefGoogle ScholarPubMed
Zimmerman, M., Ellison, W., Young, D., Chelminski, I., & Dalrymple, K. (2015). How many different ways do patients meet the diagnostic criteria for major depressive disorder? Comprehensive Psychiatry, 56, 2934. doi:10.1016/j.comppsych.2014.09.007CrossRefGoogle ScholarPubMed
Figure 0

Table 1. Effective sample size of number of participants with each symptom and symptom prevalences of genome-wide association studies

Figure 1

Figure 1. LDSC-estimated heritabilities.Heritably () calculated on the liability scale for summary statistics that met inclusion criteria (NEff > 5000,  > 0). Depression symptoms abbreviations are listed in Table 1. Case-enriched = PGC + AGDS + GS:SFHS meta-analysis, Community = ALSPAC + EstBB + UKB-MHQ meta-analysis, UK Biobank = UKB-Touchscreen GWAS.

Figure 2

Figure 2. Structure and loadings of confirmatory factor models.Points representing loadings of each symptom (columns) onto each factor (rows) for confirmatory models and for the multivariate meta-analysis of well-powered GWASs to illustrate model structure, for Case-enriched (red), Community (green), and UKB Touchscreen (blue) GWASs. Size of points scaled to absolute value of factor loadings. Symptoms arranged in order so that symptoms (Affective/cognitive: Sui, Dep, :Anh, Guilt, Conc; typical somatic: MotoInc, SleDec, AppDec; and atypical somatic: AppInc, MotoDec, Fatig, SleInc) that tend to load onto the same factor are listed next to each other.

Figure 3

Figure 3. Model structural diagram.Standardized loadings (standard errors) of factors on symptoms and genetic correlations among factors for the model (CogMoodLeth-App) used for further analysis. Symptom abbreviations are listed in Table 1.

Figure 4

Figure 4. Genetic multivariable regression.(a) Model diagrams for single regressions and (b) multiple regressions of a phenotype Y on Appetite/Weight, Cognitive/Mood/Lethargy, and Gating symptom factors (symptom indicator variables omitted for clarity). (c) Single genetic regression standardized beta coefficients (green triangles) and multiple genetic regression (red circles) coefficients (point estimates plotted with 95% confidence intervals). FDR correction indicated for significant (darker shading) and non-significant (lighter shading) coefficients. Multiple regression models adjust for the other factors. AlcDep, alcohol dependence; Anxiety, anxiety disorder; BIP, bipolar disorder; BMI, body-mass index; EA, educational attainment; MD, major depression; MDD, major depressive disorder; Neu, neuroticism; Pain, chronic pain; PTSD, post-traumatic stress disorder; Sleep, long sleep duration; Smoking, cigarettes per day.

Supplementary material: File

Adams et al. supplementary material

Adams et al. supplementary material
Download Adams et al. supplementary material(File)
File 3.2 MB