CED Clinic
Cannabis Use and Cannabis Use Disorder Are Not Genetically Interchangeable
A large 2026 imaging-genetics study found that cannabis use and cannabis use disorder had partly different genetic relationships with selected measures of brain structure and function. The work helps separate two commonly conflated research categories, but it does not prove distinct clinical brain states, explain how individuals progress to disorder, or establish a special safety category for medical cannabis.
What This Study Teaches Us About Cannabis Use and Cannabis Use Disorder
The study’s clearest contribution is conceptual. Genetic liability associated with ever using cannabis was not related to exactly the same imaging-derived phenotypes as genetic liability associated with cannabis use disorder. This supports analyzing cannabis exposure and clinically problematic use as related but noninterchangeable research phenotypes.
Its main limitation is equally important. These were statistical relationships among separate genome-wide datasets, not direct clinical comparisons of occasional users, medical patients, and people who developed severe CUD. The analysis can identify patterns of shared genetic architecture, but it cannot tell us what happened in any individual brain or whether cannabis caused the observed brain differences.
Why This Matters
A history of cannabis use is not itself equivalent to addiction, impairment, or cannabis use disorder. This study reinforces the importance of asking about function, control, distress, consequences, frequency, context, and vulnerability rather than treating exposure alone as a diagnosis. It does not provide personal reassurance that disorder cannot develop, nor does it estimate an individual’s future risk.
The paper supports resisting diagnostic shorthand. Cannabis exposure, physiological adaptation, heavy use, loss of control, craving, functional impairment, and severe addiction are not interchangeable concepts. At the same time, the study offers no clinical test, prescribing rule, product threshold, treatment recommendation, or validated way to distinguish low-risk use from future CUD in an individual patient.
Research and policy discussions often collapse all cannabis involvement into a single exposure category. This paper shows why that can obscure meaningful differences. Still, the authors compared inherited statistical liabilities defined in heterogeneous cohorts. The findings should not be converted into claims that regulated use, medical use, or occasional use has now been proven biologically separate from problematic use.
Study Snapshot
| Study Type | Secondary genome-wide association analysis using genetic correlation, latent causal variable, Mendelian-randomization, local-correlation, colocalization, gene-set, and drug-repurposing methods |
| Cannabis Use Population | 162,082 participants of genetically inferred European ancestry from the International Cannabis Consortium and UK Biobank |
| Cannabis Use Definition | Self-reported lifetime cannabis use, essentially whether a participant had ever used cannabis |
| CUD Population | 886,025 participants of genetically inferred European ancestry, including 42,281 CUD cases and 843,744 controls |
| CUD Definition | Cases identified in prior cohorts through standardized screening questionnaires or electronic health records |
| Brain-Imaging Dataset | Genome-wide statistics for multimodal imaging-derived phenotypes from 33,224 UK Biobank participants |
| Comparator | CU and CUD genetic associations were analyzed separately and compared across imaging-derived phenotypes; this was not a direct patient-level comparison |
| Primary Outcomes | Global and local genetic correlations between CU, CUD, and brain-imaging phenotypes, followed by genetically informed causal-inference and pathway analyses |
| Journal | Neuropsychopharmacology |
| Year and DOI | 2026; 10.1038/s41386-026-02506-y |
| Funding | Pilot award from the Yale Center for the Science of Cannabis and Cannabinoids; additional named support for two investigators from the American Foundation for Suicide Prevention and MQ Transforming Mental Health |
| Competing Interests | Renato Polimanti reported payment for editorial work by Complex Psychiatry. All other authors reported no conflicts. |
| Source Boundary | The complete 10-page main article was reviewed. The online supplementary tables and methods were not included in the supplied PDF. |
This study supports treating cannabis use and cannabis use disorder as related but noninterchangeable research phenotypes. It does not establish separate clinical brain states, determine whether cannabis caused the associated imaging variation, predict who will develop CUD, or provide evidence that medically supervised cannabis belongs in a proven lower-risk biological category.
What This Paper Looked At
The researchers did not recruit one group of cannabis consumers and scan their brains. Instead, they brought together previously generated genome-wide association statistics from several large research programs. One dataset represented lifetime cannabis use, another represented cannabis use disorder, and a third contained genetic associations with thousands of brain-imaging measurements.
The cannabis-use phenotype was broad. Participants were generally classified according to whether they had ever used cannabis during their lifetime. This definition mixes people with very different histories, including one-time experimentation, occasional use, sustained recreational use, and potentially medical use. It did not include reliable information about dose, potency, cannabinoid composition, route, frequency, duration, clinical supervision, or reason for use.
The CUD phenotype was assembled from multiple large cohorts using screening instruments or electronic health records. Those methods can capture clinically important disorder, but the paper’s main analysis does not separate mild, moderate, and severe cases or show whether the same genetic relationships apply across different symptom combinations.
The investigators then asked whether genetic influences associated with CU and CUD overlapped with the same imaging-derived traits. This is a study of genetic pleiotropy, meaning that some inherited variants may contribute to more than one trait. It is not direct evidence that cannabis exposure changed the measured brain systems.
What the Paper Found
The reported genetic correlation between cannabis use and cannabis use disorder was 0.48. That indicates a meaningful positive relationship and incomplete overlap. It should not be translated into the claim that exactly 48 percent of their biology is shared or that half of the genetic architecture is the same.
Cannabis use was associated with selected functional-network traits
At the global genetic-correlation level, CU was associated with resting-state activity related to the default mode network and connectivity between default mode and central executive networks. These are statistical genetic relationships with imaging-derived phenotypes, not proof that cannabis use produced a particular pattern of brain activity.
CUD was associated with different selected traits
CUD was genetically correlated with connectivity linking default mode and salience networks and with a measure of white-matter microstructure in the right superior thalamic radiation. The study therefore identified different patterns at the selected-significance threshold, although this does not establish mutually exclusive biological systems.
The causal analyses did not agree
Latent causal variable analyses suggested partial genetic causality for several imaging traits, with different findings for CU and CUD. The follow-up Mendelian-randomization analyses did not confirm the proposed effects, and some models showed evidence of directional pleiotropy. The paper therefore raises causal possibilities without resolving them.
Local genetic correlations suggested different loci
The investigators identified several genomic regions where CU or CUD shared local genetic influences with prioritized brain-imaging phenotypes. These locus-level findings implicated genes involved in neurodevelopment, signaling, chromatin regulation, synaptic biology, and white-matter processes. They remain mechanistic leads, not clinical biomarkers.
Pathway analyses generated biological hypotheses
Gene-set analyses highlighted inflammatory-response and cell-activation pathways for CU, and apoptotic-signaling and immune-regulation pathways for CUD. These results indicate enrichment among genetic signals. They do not show that cannabis exposure caused inflammation, apoptosis, immune dysfunction, or a measurable clinical syndrome.
The drug-repurposing results are computational
The gene2drug analysis identified nine compounds whose transcriptomic signatures overlapped with enriched pathways. Raloxifene and albendazole received particular attention in the discussion. The analysis does not establish that any identified compound prevents, treats, or reverses CUD.
How Strong Is This Evidence?
The study draws strength from very large cannabis-related GWAS datasets, a substantial imaging-genetics resource, multiple complementary analytic approaches, correction for multiple testing, and explicit comparison of CU and CUD within one analytical framework.
Its evidentiary ceiling remains limited. GWAS summary statistics describe population-level genetic associations. They do not reveal deterministic genes, direct brain effects of cannabis, diagnostic categories, treatment response, or individual prognosis. The divergence is scientifically credible as a pattern, while the clinical meaning remains uncertain.
Where This Paper Deserves Skepticism
1. Lifetime ever-use is an extremely broad exposure
The CU phenotype does not distinguish one experiment from years of frequent use. It does not identify medical intent, THC exposure, CBD exposure, potency, dose, route, age at initiation, duration, or supervision. That breadth makes the phenotype useful for population genetics but clinically blunt.
2. CUD cases came from heterogeneous sources
Questionnaire-defined cases and electronic-record diagnoses may represent different clinical thresholds, populations, and severity distributions. Some of the observed divergence may reflect how the CU and CUD groups were constructed rather than a clean biological division.
3. The analysis was restricted to European ancestry
The authors appropriately disclose that the available datasets limited the analysis to participants of genetically inferred European descent. Genetic architecture, linkage patterns, environmental context, diagnosis, and exposure distributions may differ in other populations.
4. The main global-correlation screen used an exploratory threshold
Among 2,505 eligible imaging-derived phenotypes, two were identified for CU and two for CUD after applying a false-discovery-rate threshold of 10 percent. That can be reasonable for exploratory work, but the small number of selected findings and permissive threshold argue for replication rather than categorical interpretation.
The reported CU-CUD genetic correlation of 0.48 cannot be restated as exactly 48 percent shared architecture or half-shared biology. A genetic correlation measures correspondence between estimated additive genetic effects across traits. Its square, magnitude, and interpretation should not be converted casually into a biological pie chart.
6. LCV and Mendelian randomization disagreed
The latent causal variable models suggested partial genetic causality, but Mendelian-randomization analyses did not confirm the proposed effects. Some models also showed directional pleiotropy. This disagreement weakens any confident account of which trait causes which.
7. Separate datasets are not a longitudinal progression study
The study did not observe cannabis users over time and determine which people developed CUD. Language about progression should therefore be understood as a broad research framing, not a measured transition within this dataset.
8. Imaging-derived phenotypes are not clinical brain diagnoses
The reported network and white-matter relationships are sophisticated statistical phenotypes. They do not establish that a clinician can scan an individual and determine whether the person has used cannabis, has CUD, or will develop CUD.
9. “Adaptive risk-taking” was not directly established
The authors suggest that the CU pattern may reflect low levels of adaptive risk-taking. That is an interpretive possibility, not a direct measurement or demonstrated benefit. Public summaries should label it as hypothesis rather than finding.
10. The supplement was not part of the supplied review package
The main article cites extensive supplementary tables for the full correlation, causal-inference, local-locus, colocalization, enrichment, and drug-repurposing results. The main paper provides enough information for the central interpretation, but this review cannot independently audit every supplementary estimate or analytic decision.
What This Paper Does Not Show
- It does not prove that cannabis use and cannabis use disorder are completely separate biological entities.
- It does not prove that CUD is unrelated to heavier exposure, cumulative exposure, or progression from earlier use.
- It does not show that cannabis changed the reported brain networks or white-matter traits.
- It does not show that the selected brain traits caused cannabis use or CUD.
- It does not predict which cannabis users will develop CUD.
- It does not estimate the risk associated with a particular THC dose, CBD dose, product, potency, route, or frequency.
- It does not study a defined population of medical-cannabis patients.
- It does not compare supervised and unsupervised use, or regulated and unregulated supply.
- It does not validate a brain scan, genetic test, immune marker, or pathway signature for diagnosing CUD.
- It does not establish that any drug identified computationally can prevent or treat CUD.
How This Fits With the Broader Clinical Conversation
Cannabis research has often treated exposure as though it were a single, stable variable. In practice, “cannabis use” can describe a few lifetime experiments, daily high-THC inhalation, low-dose nighttime use, clinician-guided treatment, self-treatment, compulsive use, or a severe pattern marked by loss of control and functional harm. A binary ever-use variable cannot carry all of those distinctions.
This paper adds evidence that CU and CUD should not be assumed to have identical genetic relationships with brain variation. That is consistent with prior genome-wide work cited by the authors showing that CUD has stronger genetic relationships with several psychiatric disorders than broad lifetime cannabis use. The paper’s novelty lies in directly comparing their relationships with a large set of imaging-derived phenotypes.
The diagnostic conversation remains more difficult. CUD is clinically real and can be severe, but the category includes considerable heterogeneity. People may meet different combinations of criteria, carry different severity levels, and arrive in datasets through different screening or coding systems. This paper acknowledges cohort heterogeneity but does not test whether mild and severe CUD have the same genetic architecture.
For medical cannabis, the responsible conclusion is narrow. The paper makes it harder to defend language that treats every cannabis-exposed patient as already occupying the same biological category as a patient with clinically impairing CUD. It does not give supervised medical use a demonstrated genetic exemption from risk. That question requires studies designed around medical indications, formulations, dosing, monitoring, symptom outcomes, and prospective development of problematic use.
What interests me most is that the researchers did not begin with the assumption that cannabis use and cannabis use disorder were simply two positions on one biological ruler. They tested them separately and found partly different genetic relationships with selected brain-imaging traits. That is useful. Medicine becomes careless when exposure, adaptation, heavy use, dependence, loss of control, and disability are allowed to collapse into one word.
I also would not push the paper farther than it can go. The cannabis-use phenotype is lifetime ever-use, which may be one of the least clinically informative definitions imaginable. The CUD phenotype comes from several systems and severity bands. Finding different genetic patterns across those broad categories is scientifically interesting, but it is not the same as discovering two clean, naturally bounded brain states.
In real care, I do not diagnose a disorder because a patient has used cannabis, developed tolerance, or takes a cannabinoid regularly. I ask whether control has been lost, whether use continues despite harm, whether life has narrowed around the substance, whether the original therapeutic purpose is still being served, and whether the patient can adjust behavior when circumstances require it. This paper supports the need for that distinction, even though it does not validate any particular clinical threshold.
The medical-cannabis implication should remain disciplined. A supervised patient using a defined product for a defined purpose should not be presumed to have CUD merely because cannabis is involved. Yet this study did not examine that patient, that product, or that care model. The paper gives us a better reason to separate the questions. It does not answer the medical-use question for us.
What a Careful Reader Should Take Away
The objective finding is that broad lifetime cannabis use and cannabis use disorder showed partly different patterns of genetic relationship with selected imaging-derived phenotypes. That finding is credible enough to challenge the habit of treating CU and CUD as interchangeable variables.
The subjective interpretation requires restraint. The study does not establish two separate brain conditions, prove that occasional use cannot progress to disorder, or demonstrate that medical and regulated cannabis occupy a lower-risk biological pathway. The paper sharpens the categories. It does not settle their clinical boundaries.
Read This Study Through Eight Different Lenses
The same genetic finding can be interpreted differently by patients, clinicians, statisticians, researchers, and public commentators. These views separate the measured result from the conclusions people may be tempted to attach to it.
How to use this: Select a lens to reveal a focused interpretation.
Patient Takeaway
A person who has used cannabis is not automatically a person with cannabis use disorder. The study supports keeping those categories separate rather than assuming that every exposure represents an early stage of addiction. That distinction matters because disorder is fundamentally about a clinically problematic pattern, not the mere presence of cannabis in someone’s history.
The paper cannot tell an individual whether their current pattern is safe, whether they will develop CUD, or whether a medical pattern of use carries a lower risk. Its cannabis-use category included anyone who reported ever using cannabis. Personal risk still depends on information the study did not measure, including frequency, THC exposure, age, psychiatric vulnerability, control, consequences, reason for use, and functional impact.
Clinician’s POV
The exam-room value is conceptual rather than diagnostic. The paper supports distinguishing exposure from disorder and asking about function, distress, craving, impaired control, competing obligations, persistent use despite harm, and the patient’s ability to change course. It does not support diagnosing or excluding CUD through a genetic profile, imaging pattern, tolerance history, or simple report of cannabis use.
For patients using cannabis therapeutically, the study provides no direct product-level or treatment-level guidance. A clinician can reasonably say that cannabis involvement exists along a heterogeneous spectrum and that problematic use should be evaluated on its own clinical features. The study does not demonstrate that supervision, regulation, or medical intent eliminates the possibility of disorder.
A Skeptical Read
A serious skeptic would begin with the phenotype definitions. Lifetime ever-use is a very loose category, while CUD cases came from several cohorts, countries, screening methods, and electronic-record systems. The contrast may therefore contain genuine biology, measurement differences, or both.
The finding that CU and CUD selected different imaging-derived traits is intriguing, but it should not be romanticized into ordinary use having one benign biology and disorder having another pathological biology. The analysis found population-level genetic relationships at selected thresholds. It did not observe behavior becoming disorder, establish biological boundaries, or test whether the categories remain distinct when dose, frequency, age, psychiatric illness, and polysubstance exposure are measured precisely.
Study Critic
The strongest technical caution is the gap between genetic association and causal explanation. The latent causal variable models suggested partial causal directions for several imaging phenotypes, but Mendelian-randomization analyses did not confirm those effects. Directional pleiotropy also appeared in some models, indicating that the instruments may violate assumptions needed for clean causal inference.
The global screen examined 2,505 eligible imaging phenotypes and identified two for CU and two for CUD using an exploratory false-discovery-rate threshold of 10 percent. The results are appropriately hypothesis-generating. They are less secure as evidence that the two conditions have been divided into stable neurobiological systems. The reported genetic correlation of 0.48 also should not be translated into a literal percentage of shared biology.
Compared to Past Research
The paper builds on prior genome-wide studies cited in its introduction and discussion. Those studies had already reported that broad cannabis use and CUD show different genetic relationships with psychiatric traits, with CUD generally carrying stronger overlap with several psychiatric disorders. The present work extends that comparison into imaging-derived phenotypes and analyzes CU and CUD within one framework.
This Lens Card is grounded primarily in the supplied paper and its discussion of earlier work. It should not be read as an independent systematic review of all prior cannabis genetics or neuroimaging studies. The paper adds a new layer to an existing argument: use and disorder should not be treated as genetically identical. It does not resolve whether the distinction is categorical, dimensional, developmental, or partly produced by phenotype definition.
Practical Considerations
The paper’s practical contribution is to improve the questions clinicians, patients, and researchers ask. “Has this person ever used cannabis?” is not an adequate substitute for “What is the pattern, what is the purpose, what is the dose, what has changed, and is there meaningful harm or loss of control?” A binary exposure variable is convenient for large datasets and weak for individualized care.
Nothing in the analysis provides a dosing threshold, preferred formulation, screening schedule, treatment algorithm, or medical-cannabis exception. In real practice, interpretation still depends on the patient’s symptoms, goals, product consistency, THC sensitivity, psychiatric history, concurrent substances, adherence, function, and capacity to reduce or stop use when appropriate.
Future Directions (Expected)
The next honest step is not merely a larger GWAS. The field needs prospective studies that begin with cannabis users and follow patterns of exposure, product chemistry, dose, route, age at initiation, indication, mental-health status, and functional outcomes over time. That design could examine which people develop clinically significant CUD and whether medical, recreational, and self-treatment pathways differ after comparable exposure is considered.
Genetic findings also require replication across ancestries and better harmonization of CUD severity. Future imaging-genetics studies should test whether the reported network relationships reproduce under stricter correction and whether they add meaningful prediction beyond clinical variables. A clinically useful model would need external validation, calibration, and evidence that it improves decisions rather than merely classifying datasets.
Misreadings & Bad-Faith Takes
Distortion: “Scientists proved cannabis use is biologically unrelated to addiction.” The paper did not show that. CU and CUD had a positive genetic correlation and therefore meaningful shared liability, alongside partly different imaging relationships.
Distortion: “Medical cannabis is now proven to be outside addiction pathways.” The study did not identify medical users, prescribed products, clinical supervision, or treatment outcomes.
Distortion: “CUD is just a labeling artifact.” The paper does not invalidate CUD. It supports investigating heterogeneity within the category while recognizing that severe CUD can involve substantial impairment.
Distortion: “Brain scans can distinguish users from addicted patients.” No diagnostic imaging test was developed or validated.
Distortion: “Immune and apoptosis pathways prove cannabis damages the brain.” Gene-set enrichment identifies overlapping genetic signals. It does not establish cannabis-induced immune injury, cell death, or clinical damage.
Join the Conversation
Where should clinicians draw the line between expected physiological adaptation, problematic use, dependence, and cannabis use disorder? Thoughtful disagreement is welcome, especially when the diagnostic language remains tied to function and evidence.
Frequently Asked Questions
Does this study prove cannabis use and CUD are biologically separate?
No. It found partly different genetic relationships with selected imaging-derived traits. That supports noninterchangeability but does not establish two categorically separate biological conditions.
What does a genetic correlation of 0.48 mean?
It means the estimated additive genetic effects associated with CU and CUD were moderately positively correlated. It does not mean exactly 48 percent of their biology is shared.
Did the researchers scan cannabis users and people with CUD?
Not directly. The study compared GWAS summary statistics for cannabis phenotypes with separate GWAS statistics for brain-imaging phenotypes.
Does this prove cannabis changes the brain?
No. Shared genetic relationships can reflect inherited vulnerability, pleiotropy, correlated traits, or other pathways. The causal-inference methods in this study did not agree.
Does cannabis use always progress to CUD?
No. The paper cites prior estimates suggesting that only a minority of people who use cannabis develop CUD. This study itself did not measure individual progression over time.
Did the study examine medical cannabis?
No defined medical-cannabis cohort was studied. The CU phenotype was based primarily on lifetime ever-use and did not identify indication, clinician supervision, product, dose, or therapeutic outcome.
Does tolerance alone mean someone has CUD?
Tolerance may occur with repeated exposure and is only one possible diagnostic feature. A clinical diagnosis requires a qualifying pattern of symptoms and clinically significant impairment or distress, interpreted in context.
Can these findings diagnose CUD?
No. The study did not develop or validate a genetic test, imaging test, pathway marker, or individual prediction model for CUD.
Do the drug-repurposing findings identify treatments?
No. They are computational transcriptomic matches that may help generate laboratory or clinical research questions. They are not evidence of treatment effectiveness.
What is the safest overall interpretation?
Cannabis exposure and cannabis use disorder should not be treated as identical research or clinical concepts. Their relationship remains partly shared, heterogeneous, and incompletely understood.
