Six mental health conditions long treated as distinct from one another may share more biological ground than psychiatry assumed — with three genes emerging as a molecular bridge between two of them

Schizophrenia and insomnia, two conditions that live in very different corners of the diagnostic manual, appear to share a molecular bridge — three genes running through the same regulatory pathway. That’s the finding at the center of a new paper in Nature Mental Health, and it sits inside a larger project: mapping the biological overlap between six conditions psychiatry has spent decades treating as unrelated — ADHD, autism spectrum disorder, bipolar disorder, major depressive disorder, schizophrenia and insomnia.

The researchers went looking for the molecular threads running under different combinations of those six, using blood drawn from more than a thousand healthy teenagers who had never been diagnosed with anything at all. Two of those threads converged in a striking way I’ll get to below.

I spend my working days building something adjacent to what this study just did with genes, only from the outside: objective, measurable signals of emotional states that currently exist mostly as things people report about themselves, imperfectly, after the fact. So when a paper like this shows up, I read it twice. Once as someone with a personal stake in the idea that mental health has a body. And once as a researcher who wants to know exactly how solid the ground is before standing on it.

How the study actually worked

The team, led by Luheng Qian and Runye Shi, drew on the Nature Mental Health journal to publish results built on the IMAGEN project, a long-running European study of adolescent brain development. They analyzed blood-derived data from 1,274 teenagers with no mental health diagnoses, looking for genetic variants tied to DNA methylation and gene expression, then used a statistical technique called Mendelian randomization to test which of those molecular changes plausibly cause disease risk rather than simply appearing alongside it. That process pointed to three genes, MAD1L1, MRPL2 and HLA-DRB1, as mediators of the DNA methylation changes linked to two of the six conditions, schizophrenia and insomnia, in a shared regulatory pathway. Other genes and methylation sites, tracked separately, turned up across different, overlapping combinations of the other four conditions.

Qian, Shi and their colleagues were direct about why they went looking for this overlap in their paper:

“The biological mechanisms underlying major neuropsychiatric disorders remain largely elusive.”

What “causal” is doing here, and what it isn’t

It’s worth sitting with what Mendelian randomization can and can’t claim. The method uses genetic variants as natural experiments to estimate whether a biological change plausibly causes an outcome, which is a meaningfully stronger claim than correlation. But “putatively causal,” the study’s own phrase for its findings, is a hedge for a reason. These genes were identified in blood, not brain tissue, and in adolescents with no diagnosis at all, which means the paper is describing risk architecture, not confirmed disease mechanisms in people who are actually unwell. The researchers frame the genes as promising targets for future research and treatment development, not as an explanation that has settled the matter.

The three genes also turned up in pathways linked to autoimmune disease, which the authors read as evidence of shared genomic architecture between the immune system and neuropsychiatric conditions, a connection researchers have suspected for years without molecular evidence behind it.

Why I’m reading this as someone building the other half

The research I’m part of approaches the same basic problem from the opposite direction. Instead of blood and DNA, we work with ecological momentary assessment, the practice of asking people to report their emotional state repeatedly throughout ordinary days rather than once in a clinician’s office, alongside trait questionnaires and what we call digital behavioral markers, patterns pulled from things like facial expression in video, that might track emotional states more reliably than self-report alone. The goal, blood-based or behavior-based, is the same: find something about a mental state that can be measured rather than only described. Psychiatry has run for its entire history on symptoms and self-report, which are real data, but noisy data, shaped by memory, insight, culture, and whatever mood someone happens to be in on the day they see a doctor.

I say this as a researcher working on measurement, not as a clinician, and nothing in this piece is meant as a diagnostic tool for anyone reading it at home.

The condition that isn’t on this list

Anxiety’s absence from this particular six is mostly an artifact of which conditions this specific team chose to study, not evidence that anxiety stands apart biologically. A separate, larger analysis published in Nature in late 2025, co-led by Kenneth Kendler at Virginia Commonwealth University, mapped genetic overlap across 14 psychiatric disorders in more than 6 million people and found that major depression, anxiety disorders and PTSD share around 90 percent of their genetic risk.

Kendler put the underlying problem plainly: “Psychiatry is the only medical specialty with no definitive laboratory tests.” Diagnosis, as he’s put it elsewhere, still comes down to symptoms and signs rather than anything a lab can confirm.

I’ve spent a long time being told, in various indirect ways, that my anxiety was a personality trait to manage rather than a biological state that could ever be measured. Findings like these don’t change what anxiety feels like on a given morning. But they chip away at the idea that mental health conditions are separate rooms with no doors between them, which is closer to how they’ve actually felt from the inside than any diagnostic manual ever gave them credit for.

What this means for the people building the tools

For health tech founders in Europe, the practical implication of transdiagnostic biomarker research is structural. Most mental health apps and screening tools are still built around single-diagnosis categories, a depression tracker here, an anxiety tool there, largely because that’s how insurers, regulators and clinical trials are organized. Research like this argues for tools built around shared underlying mechanisms instead, screening or monitoring that doesn’t assume a person’s symptoms will stay neatly inside one diagnostic box. That’s a genuinely harder product to build than a single-condition app. It’s also a more honest one.

I don’t know yet whether MAD1L1, MRPL2 and HLA-DRB1 will matter to anyone outside a lab in five years. Science this early rarely survives contact with replication in the exact shape it arrives in. What I keep coming back to is smaller than that: the quiet relief of research treating the body as a place where an answer might actually be hiding, instead of asking people to keep describing their own minds and hoping the description is enough.

If any of this lands closer to home than it does interesting, a conversation with an actual clinician will tell you more than any study I can link to here.