Published on 17 September 2026
Fear disorders and generalised anxiety disorder: how far does the genetic background overlap?
ter Kuile AR et al. Molecular Psychiatry, 2026, advance online publication of 29 June 2026, volume, issue and pagination not yet assigned.
DOI 10.1038/s41380-026-03677-2
PMID 42374129
Scientific 75
Editorial 62
The essentials
Classifications separate the so-called fear disorders (panic disorder, agoraphobia, specific phobia, social phobia) from generalised anxiety disorder, which is placed alongside depression on the distress side of the spectrum. This genome-wide association meta-analysis combines three datasets, two British and one Australian, with up to 35,523 cases for fear disorders and up to 60,879 cases for generalised anxiety disorder, compared against 99,960 broadly screened controls, or against 157,447 and 117,064 specifically screened controls. The genetic correlation between the two sets runs from 0.87 to 0.98. Only one of the four comparisons departs significantly from unity (0.87; p = 9.32 × 10-3), the one that excludes the dataset enriched for depression and uses specifically screened controls. Heritability attributable to common variants reaches 12.9% (standard error 0.9%) for fear disorders and 12.2% (standard error 0.7%) for generalised anxiety disorder in the primary analysis. The authors do not conclude that the two sets are identical: they describe a shared background of internalising vulnerability, with residual specificity that shows up mainly in relationships with other traits. The study covers only participants of European genetic ancestry. Nothing changes in consultation.
Context
The question is not academic. It decides how a clinician reads a patient who describes a fear of driving, constant apprehension, and rumination about their health. Should this count as two disorders, or one expressed differently? Classifications answer by separating them, clinical experience shows constant overlap, and the treatments barely differ. Twin studies already argued for two distinct but correlated sub-factors within the anxiety spectrum. Molecular genetics brings a measurable argument to this debate here, without settling it on its own.
How to read a genetic correlation
What the figure says, and what it does not say
A genetic correlation estimates how far the same common variants, with the same direction of effect, contribute to the risk of two phenotypes. A value close to 1 means that the share of risk carried by these variants is practically shared. It does not mean that the two disorders are identical: it says nothing about the environment, nothing about rare variants, nothing about clinical trajectory, nothing about prognosis, and nothing about treatment response. Nor does it mean that the clinical distinction is pointless, since a shared vulnerability can produce presentations that are not managed the same way. Finally, a high correlation tends to run higher the more broadly the phenotypes were defined and the less specifically the controls were screened, and this is a central key to reading this study: it is precisely by tightening control screening and excluding a sample recruited on depression that the authors see the correlation drop to 0.87 and depart from unity.
The study at a glance
| Population | |
| Adult participants of European genetic ancestry, drawn from three datasets: GLAD+ in the United Kingdom (the GLAD, EDGI UK and COPING studies), UK Biobank, and QIMR Berghofer in Australia (the Australian Genetics of Depression Study and the QSkin Sun and Health Study). Up to 35,523 cases for fear disorders, up to 60,879 cases for generalised anxiety disorder | |
| Phenotypes | |
| Fear disorders group at least one diagnosis among panic disorder, agoraphobia, specific phobia and social phobia. The second phenotype is generalised anxiety disorder. Diagnoses rest either on detailed symptom measures, DSM-5 criteria or the short form of the Composite International Diagnostic Interview, or on a single self-report item | |
| Method | |
| Four genome-wide association studies, meta-analysed by inverse-variance weighting, followed by heritability and genetic correlation estimation through linkage disequilibrium score regression. Gene- and gene-set-level analyses, local genetic correlations, genomic structural equation modelling | |
| Comparison | |
| Two control definitions were crossed with two meta-analysis compositions. Controls broadly screened for any anxiety or depressive disorder, or controls screened only for the disorder under study. A full meta-analysis of all three datasets, or a restricted meta-analysis excluding QIMR Berghofer, whose cases are enriched for depression | |
| Reported endpoints | |
| Discovery of loci reaching genome-wide significance, heritability attributable to common variants, genetic correlation between the two phenotypes and with 106 external traits |
Quality control
| Point checked | Judgement |
|---|---|
| Sample size | Sufficient |
| Finding The sample sizes exceed those of earlier work devoted specifically to fear disorders, thanks to the combination of detailed and brief measures. They remain smaller than those of meta-analyses covering any anxiety diagnosis, and the authors themselves state they are insufficiently powered to analyse each of the four fear disorders separately | |
| Statistical methods | Appropriate |
| Finding Genomic principal components as covariates, tests suited to case-control imbalance, linkage disequilibrium score regression intercepts between 1.00 and 1.01, inflation attributable to polygenicity for 93% to 100%. Explicit Bonferroni corrections and sensitivity analyses excluding the dataset enriched for depression | |
| Phenotype definition | Heterogeneous |
| Finding The UK Biobank mental health questionnaire provides only brief measures for fear disorders, and QIMR Berghofer has no detailed measure of fear at all. The broad term panic was used to define cases, for lack of a specific measure of panic disorder in every dataset. A broad definition on both sides mechanically increases the observed correlation | |
| Ancestry diversity | European only |
| Finding The analyses are restricted to clusters of European genetic ancestry defined by principal component analysis. Transposability to other ancestries has not been assessed, and this limits the general reach of the result | |
| Independence | To be qualified |
| Finding Institutional and charitable funding: NIHR BioResource and NIHR Biomedical Research Centres, the Medical Research Council, the Health and Social Care R and D Division of Northern Ireland, the National Centre for Mental Health of Wales, the Australian NHMRC, the Canadian SSHRC and FRQSC councils, the Maudsley Charity, Guy’s and St Thomas’ Charity, and the European Research Council under Horizon 2020. Two authors declare links with industry: honoraria, research or conference funding, and consulting work with Illumina, Otsuka and COMPASS Pathfinder for one, and Eli Lilly, Janssen, Illumina and the Sackler Trust for the other. The remaining authors declare no conflicts of interest | |
Results
| Result | Reported value |
|---|---|
| Genetic correlation between the two phenotypes | 0.87 to 0.98 |
| Reading All four estimates differ significantly from zero. Only one differs from unity: the one combining specifically screened controls with the restricted meta-analysis, at 0.87 with p = 9.32 × 10-3. The other three are not statistically distinguishable from complete sharing. The publication plots these estimates with error bars corresponding to standard errors; it does not give their numerical value in the text or in the supplementary tables, and no confidence interval is published for these four correlations | |
| Heritability attributable to common variants, fear disorders | 12.9%, standard error 0.9% |
| Reading Value from the primary analysis, full meta-analysis with broadly screened controls, on the liability scale and for an assumed prevalence of 8.8%. Depending on the meta-analysis and the control screening, the estimate ranges from 7.5% to 12.9%. This is the share of variance attributable to the common variants analysed, not the total heritability estimated by twin studies, which the authors place between 20% and 60% for anxiety disorders | |
| Heritability attributable to common variants, generalised anxiety disorder | 12.2%, standard error 0.7% |
| Reading Same primary analysis, for an assumed prevalence of 12.9%. Depending on the meta-analysis and the control screening, the estimate ranges from 9.1% to 12.2%. These modest values are a reminder that common genetics explains only a fraction of the risk | |
| Loci reaching genome-wide significance | 1 for fear, 4 for generalised anxiety |
| Reading The locus associated with fear disorders is at 14q24.3, lead variant rs10047892, odds ratio 1.07 with a standard error of 0.01 and p = 3.52 × 10-8, and it reaches this threshold only with broadly screened controls. Three genes are associated with fear (SNX29, PAX6, ERI3) and three with generalised anxiety disorder (TMEM106B, YLPM1, SORCS3). An odds ratio of this size has no individual predictive value | |
| Correlations with external traits | 106 traits tested, 12 significant differences |
| Reading For most traits, the two phenotypes behave the same way, including with respect to depression. Twelve traits are exceptions, with a median correlation difference of 0.20. Educational attainment, cognitive ability and coronary artery disease correlate more strongly with fear disorders; neuroticism, bipolar I disorder and anorexia nervosa correlate more strongly with generalised anxiety disorder. These differences are genetic associations, not cause-and-effect relationships | |
Critical appraisal
| Domain | Judgement |
|---|---|
| Internal validity | Sound |
| Finding The methods used are those of the field, and the expected controls are present, including the exclusion of participants present in two cohorts and the projection of principal components onto a common reference space | |
| Dependence on control choice | Decisive |
| Finding The central result changes status depending on how controls are screened and on whether a sample recruited on depression is included. This is not a flaw in the study, which discloses this dependence, but it rules out settling on a single correlation value | |
| Risk of phenotypic circularity | Present |
| Finding When two phenotypes are broadly identified from the same instruments, part of the overlap may reflect measurement rather than biology. The authors themselves note that in GLAD+, only 5% of cases met criteria for an isolated anxiety disorder with no depressive comorbidity | |
| Residual specificity | Partial but real |
| Finding Presenting this work as abolishing the boundary between fear and generalised anxiety would overstate it. The structural equation models leave generalised anxiety disorder with essentially no residual genetic variance once the worry factor is accounted for, whereas fear disorders retain 30% to 36% | |
| Scope of the claim | Contained |
| Finding The authors write that their results partially support a genetic distinction between distress and fear, but argue more strongly for a broad genetic vulnerability to internalising disorders. They conclude neither that the disorders are identical nor that classifications need to change, and this restraint should be kept | |
| Therapeutic consequence | None at this stage |
| Finding No tool, no marker, no treatment direction follows from these estimates, which describe populations, not individuals | |
Level of evidence
Oxford level of evidence 3a, the study meta-analysing case-control genetic association comparisons, with no longitudinal follow-up or randomisation. Confidence is good in the existence of a high genetic overlap between these two sets of phenotypes in participants of European ancestry. It is low on the exact value of the correlation, which depends on how cases, and especially controls, were defined. It is nil on any immediate practical consequence.
The colleague test
What an experienced colleague would say if you put this study to them in two minutes, between two consultations.
“ Nothing to change on Monday morning. The shared background is very broad, broader than the manual’s boundary. But I wouldn’t say the boundary is purely clinical: as soon as you tighten the control screening, the correlation drops to 0.87 and departs from unity, and the two phenotypes don’t relate the same way to neuroticism, bipolar I disorder, or cognitive ability. In other words, a broad shared foundation, plus a piece of its own that still needs characterising. ”
What this means in practice: the result supports a dimensional reading of anxiety, useful for explaining to a patient why they carry several labels at once. It does not remove the need to characterise the presentation precisely, since it is the presentation, not the genetic background, that guides treatment.
What you can take from this on Monday morning
- Explain to a patient carrying several anxiety diagnoses that these labels largely describe different expressions of a single vulnerability, rather than independent diseases that add up.
- Do not conclude from this work that the categories are useless: they remain the basis for treatment indication and for communication between clinicians, and the study retains a partial genetic specificity.
- Keep the teaching value of the notion of genetic correlation, distinct from heritability, itself distinct from individual risk.
- Also keep in mind that the result depends on how controls were defined, a useful argument when discussing the robustness of any genetic correlation.
- If asked, remind patients that no genetic test currently has any use in diagnosing or treating anxiety disorders.
- The course of action is set out in the NICE decision tree for generalised anxiety and panic disorder.
Frequently asked questions
Does this mean fear disorders and generalised anxiety disorder are the same disorder?
No, and the authors do not say so. The sharing concerns the share of risk explained by common variants, and a partial specificity remains, visible in the correlations with other traits and in the genetic variance specific to fear disorders. Two presentations arising from a similar vulnerability can still differ in their course, their impact and their management.
Does this result hold for all patients?
The data come exclusively from participants of European genetic ancestry, in the United Kingdom and Australia. No extrapolation to other ancestries is possible as things stand, for lack of analysis in those populations.
Why does the heritability look so low?
Because it is the share explained by the common variants studied alone, about 12% here, not the total heritability. Estimates from twin studies, 20% to 60% for anxiety disorders, do not measure the same thing.
Can a genetic test help in practice?
No. These results concern populations, not individuals. The single locus associated with fear disorders corresponds to an odds ratio of 1.07, with no individual predictive value at all.
Annotated bibliography
Source study. ter Kuile AR, Mitchell BL, Lee SH, Morneau-Vaillancourt G, Skelton M, Coleman JRI, et al. Genome-wide genetic overlap between fear-based disorders and generalised anxiety disorder. Molecular Psychiatry. 2026, advance online publication of 29 June 2026, volume, issue and pagination not yet assigned. DOI 10.1038/s41380-026-03677-2, PMID 42374129. Open access article under a Creative Commons Attribution 4.0 licence. Received 18 March 2025, revised 5 May 2026, accepted 3 June 2026. What the work contributes: the first direct genome-wide comparison of fear disorders and generalised anxiety disorder using detailed phenotyping, and the demonstration that the genetic overlap depends on control screening. What it does not contribute: no analysis by individual disorder, no non-European ancestry, no longitudinal data, no element of therapeutic reach. Summary statistics from the two association studies are deposited on Zenodo, DOI 10.5281/zenodo.17228387, access to individual-level data remaining subject to approval from each cohort’s committees.
What was consulted. Verification carried out on 13 August 2026 against the full text of the published version, the article’s fourteen pages, abstract, introduction, methods, results, tables 1 to 3, figures 1 to 3, discussion, data and code availability, bibliography, acknowledgements, contributions, funding, conflict-of-interest statement and ethical approval. The supplementary material was consulted in full: the supplementary methods, results and discussion document, with its figures 1 to 11 and its references, as well as the workbook of 26 supplementary tables, including table 4 for the assumed prevalences, table 5 for the heritabilities and quality-control statistics by meta-analysis, table 11 for the comparison between detailed and brief phenotyping, tables 12 and 16 for the genetic correlations with external traits and neuroticism components, and tables 13 to 15 for the local genetic correlations. The figures in the supplementary material could not be read as images, only their captions were. The numerical values of the standard errors for the four genetic correlations between the two phenotypes are published neither in the text nor in the supplementary tables: they appear only as error bars in figure 2A and are therefore not reproduced here. The digital object identifier and the PubMed identifier were verified against the publisher and PubMed on 13 August 2026; the volume, issue and pagination had not yet been assigned as of that date. Article submitted to an independent double reading.