Published on 15 September 2026
Do suicidal ideation and suicidal behaviour share the same genetics?
In brief
Fifty-four cohorts, up to nearly 1.6 million participants for the largest phenotype, four phenotypes of the suicidality spectrum analysed separately. Seventy-seven regions of the genome cross the significance threshold, fifty-nine of them never reported for suicidality before. The count is not what matters for the consulting room. What matters is the genetic correlation between suicidal ideation and suicidal behaviour, estimated at 0.88 and significantly below 1. The two overlap widely without merging. SNP-based heritability stays low in the European-ancestry analyses, 3.0% for ideation and 6.7% for behaviour, and the best polygenic score, the one for suicidal behaviour, explains only 1.16% of the variance in that phenotype. No individual tool comes out of this, and the authors say so. The document analysed is a preprint, it has not yet been peer reviewed.
The context
The question asked in the consulting room is nearly always the same. We ask about dark thoughts, and from the answer we infer a risk of acting. The reasoning assumes a continuum: the thought first, the act next, the same thing further along. Clinical scales are built on that assumption.
The work analysed here tests the hypothesis by another route. If suicidal ideation and suicidal behaviour were one phenomenon at two intensities, the genetic architecture they share should be complete. It is not. That is a biological argument for a practice many clinicians already follow, without always being able to justify it other than by experience: asking two distinct questions rather than one.
The method
A genome-wide association study compares the frequency of millions of common variants between people who carry a phenotype and people who do not. It measures an association at the scale of a population. It measures neither causality nor individual risk, and it says nothing about rare variants, which genotyping arrays miss.
A genetic correlation between two phenotypes quantifies the overlap of their architectures. A value of 1 would mean that the same variants act in the same way on both. A value below 1, if the gap is statistically established, means that some of the determinants differ. That measure, and not the number of loci, carries the claim of the paper.
The study at a glance
| Population | |
| Participants from 54 cohorts worldwide, brought together by the Suicide Working Group of the Psychiatric Genomics Consortium, among them the Million Veteran Program and UK Biobank. Four phenotypes analysed separately. Suicidal ideation, 259,747 cases against 1,309,943 controls, across 37 cohorts. Suicide attempt, 64,993 against 1,269,037, across 46 cohorts. Suicidal behaviour, attempt or death, 75,300 against 1,311,895, across 49 cohorts. Suicide death, 9,197 against 668,162, across 7 cohorts. Ancestries predominantly European, 176,147 of the 259,747 ideation cases and 53,919 of the 64,993 attempt cases. | |
| Analysis | |
| Inverse-variance-weighted fixed-effects meta-analyses run with METAL inside the standardised pipeline of the consortium, RICOPILI. Then linkage disequilibrium score regression for heritability and genetic correlations, Bayesian fine-mapping with SuSiE, PolyFun and SuSiEx, gene prioritisation by summary data-based Mendelian randomisation using brain quantitative trait loci for expression, splicing, methylation and protein abundance, then by chromatin interaction data. Gene-set enrichment with MAGMA and GSA-MiXeR, cell-type enrichment by stratified linkage disequilibrium score regression from single-cell transcriptomic data, polygenic scores with PRS-CSx in 25 target cohorts, each excluded in turn from the discovery meta-analysis, and a query of two drug target databases. | |
| Comparator | |
| Participants without the corresponding suicidality phenotype in each cohort, following a phenotyping protocol published by the working group. Psychiatric diagnoses were used as an exclusion criterion neither for cases nor for controls, a deliberate choice made to avoid bias related to comorbidity. | |
| Outcomes | |
| Number of genome-wide significant loci, SNP-based heritability, genetic correlations between phenotypes, prioritised genes, cell-type enrichments, variance explained by a polygenic score. | |
| Design | |
| Meta-analysis of genome-wide association studies, observational design, not experimental. Preprint posted on medRxiv on 23 October 2025, version 1, not peer reviewed at the date of verification. |
Quality control
| Item checked | Verdict |
|---|---|
| Sample size and number of cohorts | Unmatched |
| FindingFifty-four cohorts, and up to 1,569,690 participants for the ideation analysis, 259,747 cases and 1,309,943 controls. Samples overlap from one phenotype to another and the authors give no single total. Statistical power is not the limiting factor of this work, which is rare in psychiatric genomics and makes the identification of 59 loci not previously reported for suicidality credible. | |
| Inflation of test statistics | Controlled |
| FindingThe intercepts of the linkage disequilibrium score regression lie between 0.99 and 1.06. A value close to 1 indicates that the observed signal comes from association and not from poorly corrected population stratification. | |
| Editorial status | Preprint |
| FindingThe document has not been peer reviewed. The preprint does carry a code availability statement, the polygenic score analysis scripts being deposited in a public repository, and a data availability statement, the summary statistics being accessible on application through the consortium portal, with the exception of the Utah cohorts for two phenotypes. The institutional standing of the consortium does not replace peer review. | |
| Phenotype definitions | Heterogeneous |
| FindingIdeation and attempt data come from structured psychiatric interviews, self-report questionnaires, codes of the International Classification of Diseases, or combinations of these sources. Deaths are identified from cause of death codes, coroners’ reports or medical examiners’ reports. A shared phenotyping protocol frames these definitions without making them uniform. A noisy measure mechanically lowers an estimated genetic correlation, so this heterogeneity cannot be dismissed as a partial explanation of the gap below 1. That reading is the reviewer’s, the authors do not put it in those terms. | |
| Ancestry diversity | Predominantly European |
| FindingThe authors present this as the most diverse study to date in the field, and African, South Asian, East Asian and Latin American cohorts are represented. But heritability does not reach significance in the African, East Asian and Latin American analyses, and polygenic scores perform less well there than in European cohorts. Limited transferability is a shortcoming of the whole field, not of this study in particular. | |
| Funding and interests | Public funding declared |
| FindingFunding declared in the acknowledgements: a National Science Foundation graduate research fellowship, a National Institute of Mental Health grant, consortium grants to seven universities, a biomedical research centre of the British National Institute for Health and Care Research, academic computing resources. No declaration of interests appears in the document consulted, which counts neither for nor against the existence of such interests. | |
The findings
| Result | What the data show |
|---|---|
| Significant loci | 77 in all, 59 of them not previously reported for suicidality |
| ReadingThe figure reflects first of all an unprecedented statistical power. A significant locus is a region of the genome where an association signal passes the conventional threshold, not a gene for suicidality. | |
| SNP-based heritability | European ancestry: attempt 6.6%, behaviour 6.7%, death 5.1%, ideation 3.0% |
| ReadingThese values are low, and they are expressed on the liability scale assuming lifetime prevalences set by the authors, 9% for ideation, 2% for attempt and behaviour, 0.2% for death. The authors note that the values may therefore be over or underestimated depending on the population. The common variants measured here explain a small fraction of the variability of the phenotype in the population. The rest lies in what the method does not capture, rare variants, environment, interactions, and in measurement error of the phenotype itself. | |
| Genetic correlations between phenotypes | Ideation and behaviour 0.88, death and behaviour 0.73, death and ideation 0.70 |
| ReadingAll three are significantly below 1, with probabilities under 8.01 times ten to the minus five for the tests of being below 1. This is the central result of the paper. It holds from one pair of phenotypes to the next, and the value falls as one moves further from ideation. Conversely, suicide attempt and suicidal behaviour are near indistinguishable genetically, a correlation of 1.02, which follows from sample overlap and led the authors to favour suicidal behaviour for the secondary analyses. | |
| Polygenic score | Suicidal behaviour: 1.16% of variance explained, odds ratio 1.43 |
| ReadingOne per cent of variance explained authorises no individual decision. The odds ratio of 1.43, 95% CI 1.29 to 1.62, compares the top quintile of the score with the middle quintile, in research cohorts, and not two extremes nor one patient against another. The corresponding area under the curve is 0.57. The ideation score does less well still, 0.49% of variance explained and an odds ratio of 1.23. | |
| Fine-mapping and prioritised genes | 27 candidate causal variants, 20 genes retained, 7 for ideation and 13 for behaviour |
| ReadingThe authors call these variants putative causal, which is the right wording. Bayesian fine-mapping ranks candidates within an interval, it does not demonstrate a mechanism. | |
| Biological enrichments | Behaviour: synaptic and dopaminergic pathways. Cell types: amygdala excitatory, medium spiny neurons, hippocampal CA1 to CA3 |
| ReadingThese enrichments sketch a plausible biology. They remain statistical associations between a genetic signal and annotations built from other datasets, with the uncertainty that implies. Two reservations come from the authors themselves. No gene set reaches significance for ideation or for death, so gene-set enrichment concerns suicidal behaviour alone. And the sets brought forward by the second method are almost all driven by a single, highly associated gene, which argues against concluding that a whole pathway is involved. | |
| Drug targets | Signals on genes already targeted by existing molecules |
| ReadingThe pairings reported link, for ideation, a gene targeted by a serotonin and noradrenaline reuptake inhibitor, and genes targeted by cancer or anti-inflammatory treatments, and for behaviour, three main targets including a tyrosine kinase receptor and an oestrogen receptor. A target identified by querying pharmacological databases is a hypothesis to be tested, not an indication. No clinical efficacy data accompany this result, and nothing here justifies changing a prescription. | |
Critical appraisal
| Domain | Judgement |
|---|---|
| Design and level of evidence | Observational |
| FindingA genome-wide association study is an observational design. The natural randomisation of alleles limits some confounding, it does not make the work a trial. Nothing in this paper establishes that acting on a target would change a risk. | |
| Phenotype measurement | Weak point |
| FindingThis is the main limitation, and it bears directly on the conclusion. A noisy measure lowers the estimated genetic correlation. The gap below 1 is therefore real in the data, but its size could be overstated. The authors acknowledge the complexity and heterogeneity of the ideation phenotype, and the heterogeneity introduced by combining many cohorts, but they do not present that point as an explanation of the gap below 1. The reservation is the reviewer’s. | |
| Statistical precision | High |
| FindingConventional threshold of 5 times ten to the minus eight maintained, inflation controlled through the intercepts of the linkage disequilibrium score regression, quality control of the summary statistics before meta-analysis, Benjamini and Hochberg false discovery rate correction for the genetic correlations, block jackknife testing to compare two correlations, resampling confidence intervals for the variance explained, posterior inclusion probability above 0.5 for fine-mapping. The statistical rigour calls for no reservation. The dependence of the heritabilities on assumed prevalences remains a declared limitation. | |
| Fit between claim and evidence | Calibrated |
| FindingLoci are presented as associations, variants as putative causal, targets as potential. The paper promises no clinical application. The title announces a count, and the count is established. | |
| External validity | Limited |
| FindingEuropean ancestries dominate the sample, and heritability does not reach significance in the African, East Asian and Latin American groups, for want of power. Polygenic scores also perform less well there, which restricts the reach of the predictive results far more than that of the biological ones. | |
| Independence and transparency | To be confirmed |
| FindingPublic funding announced, sources named. The polygenic score analysis code and the terms of access to the summary statistics are already declared in the preprint. No declaration of interests, however, appears in the document consulted, and that piece remains to be checked against the published version. | |
Level of evidence
Confidence is high that the genetic architecture of suicidal ideation and that of suicidal behaviour do not overlap completely. Sample size, the standardisation of the pipeline and the consistency across pairs of phenotypes all converge.
It is moderate on the exact size of the gap. Heterogeneity in the phenotype definitions pushes the estimated correlation downwards, and nobody today can separate the share of biology from the share of measurement noise.
It is low, indeed nil, on everything to do with individual use. A polygenic score explaining 1.16% of the variance predicts nothing for a given person, and the authors write themselves that these scores alone “are not adequate for individual-level clinical prediction or risk stratification”. They add that clinical utility remains conceivable in future, combined with other sources of data, which is yet to be shown. The preprint status finally imposes a general reservation: these values may move on publication.
The colleague test
What an experienced colleague would say if you put this study to them in two minutes, between two consultations.
“ It confirms what I already do. I do not ask about dark thoughts and stop there, I also ask whether the person has prepared anything. Genetics now says these are not quite the same questions. Beyond that, 6% heritability and 1% of variance explained, I am not going to change my prescribing tomorrow morning. And it is a preprint. ”
What this means in practice: the paper changes no therapeutic decision. It gives a biological argument for an interviewing practice, the one that explores suicidal thinking and the prepared act separately. On prediction it closes the door for the present, the authors judging these scores unfit for individual use, and leaves it ajar for the future, on condition that they are combined with other data.
What you can do with this on Monday morning.
- Ask about suicidal ideation and about suicidal behaviour separately in the interview, rather than treating the second as an expected worsening of the first. This work gives that practice a biological footing it was missing.
- Be ready to answer a patient or a family who asks whether a genetic test of suicide risk exists. The answer is no, and the figure to remember is that the best score from this work explains 1.16% of the variability of suicidal behaviour.
- Keep giving family history of suicide its place in the assessment. A heritability of the order of 6% for common variants is not the total heritability of the phenotype: the authors recall that twin and family studies put the heritability of suicidality between 30% and 55%.
- Treat the drug repurposing leads as research hypotheses. None translates into a prescription, and none has been tested against a clinical outcome.
- Wait for the peer reviewed version before citing the figures of this study in a formal setting.
Frequently asked questions
Does this mean that suicide is genetic?
No. The study measures the share of the variability of a phenotype, in a population, attributable to common genetic variants. That share is low, from 3.0% to 6.7% depending on the phenotype in the European-ancestry analyses. Most of the variability lies elsewhere, and nothing here says what will happen to one person.
Is there a test that can be used in the consulting room?
No. A score explaining 1.16% of the variance, with an area under the curve of 0.57, does not discriminate between two patients. The authors write that these scores, taken alone, “are not adequate for individual-level clinical prediction or risk stratification”, and that their use today is in research, not in the clinic.
Should anything change in prescribing?
No. The molecules named are named because a gene flagged by the analysis appears in a pharmacological target database. That is a research lead, with no clinical efficacy data on a suicidality outcome. Nothing justifies starting or changing a treatment on that basis.
What does a correlation of 0.88 actually change?
It supports an interviewing practice rather than a therapeutic decision. If ideation and behaviour shared exactly the same determinants, exploring one would inform you about the other. That is not quite the case, so the two questions are asked separately.
Is a preprint reliable?
A preprint is work made public before peer review. Results can be revised, and figures sometimes move. Its consortium origin makes a substantive revision less likely, without ruling it out. The data cited here will need rechecking against the published version.
Annotated bibliography
Source study. Colbert SMC, the Psychiatric Genomics Consortium Suicide Working Group, Ruderfer D, Docherty AR, Mullins N. Genome-wide association studies identify 77 loci for suicidality and provide novel biological insights. medRxiv, 23 October 2025: 2025.10.22.25338076, version 1. DOI 10.1101/2025.10.22.25338076 · PMID 41282879.
Status of the document. Preprint posted on medRxiv, not peer reviewed at the date of verification. PMID 41282879, PMCID PMC12633560. No supplementary material accompanied the document consulted: the supplementary figures and tables of this publication were not consulted, and no data appearing only there is reproduced here.
Background references. The works below are those the source publication explicitly cites and for which it gives bibliographic details. They were not consulted in full text, and what is attributed to them here is what the source publication reports of them.
Voracek M, Loibl LM. Genetics of suicide: a systematic review of twin studies. Wien Klin Wochenschr. 2007;119:463-475. Contribution: this is the source of the heritability estimate of 30% to 55% that serves as the yardstick for judging how low the SNP-based heritability is. Limitation: a review of twin studies, whose estimates rest on strong assumptions about shared environment.
Mullins N, et al. Dissecting the shared genetic architecture of suicide attempt, psychiatric disorders, and known risk factors. Biol Psychiatry. 2022;91:313-327. Contribution: earlier work from the same consortium, from which the source publication takes the definition of the aggregated suicidal behaviour phenotype. Limitation: sample sizes far below those of the present meta-analysis.
Docherty AR, et al. GWAS meta-analysis of suicide attempt: identification of 12 genome-wide significant loci and implication of genetic risks for specific health factors. Am J Psychiatry. 2023;180:723-738. Contribution: a direct point of comparison, 12 loci for suicidal behaviour and a maximum variance explained of about 1.1% by polygenic score. Limitation: lower power, which explains part of the apparent gain of the present work.
Docherty AR, et al. Genome-wide association study of suicide death and polygenic prediction of clinical antecedents. Am J Psychiatry. 2020;177:917-927. Contribution: an earlier study of suicide death, which reported six variants across two loci on chromosomes 13 and 15. Limitation: those loci are not recovered in the present analysis, with probabilities above 0.01, which illustrates how unstable results are at small sample sizes.
Ashley-Koch AE, et al. Genome-wide association study identifies four pan-ancestry loci for suicidal ideation in the Million Veteran Program. PLoS Genet. 2023;19:e1010623. Contribution: the earlier reference for suicidal ideation, whose polygenic score explained only 0.2% of the variance. Limitation: a population of United States veterans, whose representativeness is open to question.