New Genetics Study Splits "Cannabis Use" From "Cannabis Use Disorder"

A new study published in Neuropsychopharmacology, led by researchers at Yale and supported by the Yale Center for the Science of Cannabis and Cannabinoids, compares the brain biology underlying cannabis use (CU) with the brain biology underlying cannabis use disorder (CUD) directly — treating them, from the outset, as two related but distinct phenomena rather than points on a single scale. It is the first genetically informed study to do so.

What the study found

Researchers led by Yale's Renato Polimanti drew on genetic data from hundreds of thousands of participants across the UK Biobank, the International Cannabis Consortium, the Psychiatric Genomics Consortium, and the US Million Veteran Program. Rather than asking which genes are linked to cannabis broadly, the team tested whether the genetic signatures associated with using cannabis line up with the same brain networks as the genetic signatures associated with a diagnosed disorder.

Largely, they did not. CU was genetically linked to activity and connectivity within the brain's default mode network — the system active during rest, self-reflection, and mind-wandering — and its coupling with regions involved in executive control. CUD was linked to a different set of connections, between the default mode and salience networks, and to white matter changes, alongside genetic pathways related to immune and stress-response activity.

Genetically, CU and CUD overlap only partially — roughly half their genetic architecture is shared, and half is not. The study's own framing is that this divergence "may reflect low levels of adaptive risk-taking associated with CU," while CUD looks more tied to processes involved in tolerance, dependence, and self-medication.

Senior author Renato Polimanti, of Yale's Department of Psychiatry, put it plainly in the university's own coverage of the study: the shift from use to disorder isn't just "more of the same" escalating quantitatively, but a qualitatively different neurobiological process. As cannabis becomes more accessible globally, he argues, it's important to keep the biology of use and the biology of disorder analytically separate, since collapsing them risks obscuring the mechanisms researchers most need to understand.

Reading the finding

The practical significance of the finding is that using cannabis — including under medical supervision — is not simply disorder in waiting. The biology looks different, not just the severity. That runs against a lot of public discourse and clinical shorthand, where any cannabis use is treated as a step on the same ladder as dependence, and it gives prescribing doctors a firmer evidence base for treating medical cannabis use as its own clinical category, separate from addiction pathways. It also gives some empirical weight to an argument the legal supply sector has made on other grounds: that regulated, monitored use sits in a different risk category from the unsupervised, high-frequency use patterns that dominate most cannabis research to date.

Where the study doesn't reach

The dataset is not a medical-cannabis population. It is a mix of general-population cohorts, many measuring lifetime "ever used" cannabis, and disorder cohorts drawn from health records and screening questionnaires — not people using cannabis under prescription, at controlled doses, for defined conditions. The study speaks to genetic architecture in cannabis-using populations broadly, not to the safety profile of medical use specifically. It is also restricted to participants of European ancestry, limiting how far the findings generalise. And several of the study's own causal-inference methods, which look for cause-and-effect rather than just correlation, disagreed with each other — a signal the authors are appropriately cautious about in their own discussion.

A question the study doesn't ask: what is "CUD," exactly?

There's a deeper issue sitting underneath this study, and underneath most cannabis genetics research, that deserves more space than can be given here but bears directly on how findings like this should be read.

"Cannabis use disorder" is not a single, stable entity even within the clinical literature. Under DSM-5, a diagnosis requires only two of eleven possible criteria, met in any combination — meaning a person with mild tolerance and one unsuccessful attempt to cut down sits in the same diagnostic category as someone experiencing severe craving and serious functional impairment. Genetic studies typically pool CUD "cases" from a mix of self-report screening and clinical or health-record diagnoses, which likely capture different severity bands and different populations altogether.

It gets more complicated again outside the DSM. Australia's formal diagnostic and coding framework runs on the ICD system, not DSM-5, and research comparing the two shows they don't agree well with each other: population-level studies find that DSM-5's use-disorder category captures a different aspect of problematic use and selects a different group of people than ICD-10 or ICD-11 do, with concordance between the systems generally described as moderate rather than close. One study of adolescents in addiction treatment found something worth noting specifically: among those who met the newer ICD-11 dependence criteria, few reported the loss of control over their use that most people would consider central to what "addiction" means — most had simply reported tolerance or heavy time spent using. For medical patients specifically, it is also worth stating plainly that tolerance or withdrawal arising from prescribed cannabinoid treatment does not, on its own, indicate cannabis use disorder; impairment or distress has to be present too.

The Yale-led study gives this critique some quiet support. In its own limitations section, the research team acknowledges that differences between the cohorts pooled into their CU and CUD genetic phenotypes — different countries, different screening methods, different eras — may have shaped the distinct brain patterns they report. Some of what looks like a clean biological split between "use" and "disorder" could partly reflect how differently those two groups were assembled in the underlying data, not only how differently their brains are wired.

None of this means CUD isn't real, or that it can't be severe. It clearly can be. But it does mean that when a genetic study reports genetic causality or immune-pathway involvement in CUD, the underlying category being studied is less settled than the confident, disease-style language suggests. Even the study's own university coverage illustrates the point in passing: reporting on researchers explicitly arguing that use and disorder are distinct, it still slips into using "addiction" and "cannabis use disorder" as interchangeable terms, the kind of conflation that makes a mild, two-criteria diagnosis and a severe, functionally disabling one sound like the same thing. A follow-up piece will examine what this diagnostic instability means for how cannabis genetics research should be interpreted, and how the label gets used in clinical, insurance, and public conversations about cannabis patients.

The bottom line

This study adds real weight to the case that cannabis use and cannabis use disorder are biologically distinct, evidence that runs counter to the assumption that any use sits on a path toward addiction. But its evidence base doesn't extend to medical cannabis populations specifically, and the "disorder" side of its comparison rests on a diagnostic category that is, by the field's own measure, still unsettled. Both points are worth holding onto as this area of research develops — and neither should be read as medical advice; anyone considering medical cannabis treatment should discuss it with a registered health practitioner.

Source: Chen, Q., He, J., Qiu, D., et al. Genetic pleiotropy differentially linking brain variation to cannabis use and cannabis use disorder. Neuropsychopharmacology (2026). DOI: 10.1038/s41386-026-02506-y

Additional reporting drawn from Yale School of Medicine's coverage of the study, including comment from senior author Renato Polimanti.

This article was researched and drafted with AI assistance and fact-checked against primary sources.