Taiwan Study Identifies Four Gene Hubs Linked to Schizophrenia Before Age 18

A 58 gene network points to earlier illness

Researchers at Taiwan's National Health Research Institutes have identified four potential gene hubs associated with schizophrenia that begins before age 18, using genetic data from 1,190 patients across two independent family cohorts. The genes, COPS6, MRPS7, RUVBL2 and PSMD2, emerged from a network of 58 genes whose predicted activity was associated with earlier symptoms.

The research, published online in Molecular Psychiatry on July 3, 2026, connects inherited DNA differences with predicted gene activity in brain tissue. It offers leads for investigating why some people develop schizophrenia during adolescence, when the brain is still maturing and illness can disrupt education, cognition and family life.

The findings are not a diagnostic test or a new treatment. They identify candidate biological mechanisms that need further testing. There is also a numerical difference between the institute's explanation and the journal abstract: the institute gives an adjusted odds ratio of 1.32 for the gene network score, while the abstract reports 1.22 for a measure labeled M10sub-GeRS.

The team includes distinguished research fellow Wei J. Chen, associate investigator and attending physician Yen Feng Lin, and postdoctoral researcher Ya Wen Jen at the institute's Center for Neuropsychiatric Research. Their work draws on Taiwanese family studies established decades before the present analysis.

Why focus on symptoms before adulthood?

Schizophrenia affects an estimated 6.35 people per 1,000 over a lifetime. About 8% of patients develop symptoms before age 18. These can include delusions, hallucinations and cognitive difficulties, with earlier onset generally associated with more severe illness and poorer response to antipsychotic medication.

Jen explained that earlier onset is associated with a higher proportion of affected family members. Studying this subgroup may therefore help researchers detect genetic signals that become harder to identify when patients with very different illness histories are analyzed together.

Twin and family studies estimate schizophrenia's heritability at about 80%. That figure describes how much variation in susceptibility across a population is associated with genetic differences. It does not mean that genes determine 80% of an individual's illness or that a child in an affected family has an 80% chance of developing schizophrenia.

The clinical burden extends beyond psychotic symptoms. Illness during adolescence can interrupt schooling while cognitive abilities are still developing. The original account also describes academic pressure and social stigma as burdens for patients and families. Although about 30% of schizophrenia patients respond poorly to medication, that figure concerns schizophrenia broadly, not specifically the adolescent subgroup examined here.

Two family cohorts provide a comparison

The institute's research account describes two Taiwanese cohorts with contrasting family histories. The Taiwan Schizophrenia Linkage Study recruited families with at least two affected siblings between 1998 and 2002. The Schizophrenia Trio Genomic Research in Taiwan recruited patients from families with one affected member, together with their unaffected parents, between 2009 and 2014.

The analysis included 595 patients from families with multiple affected members: 223 with earlier onset and 372 with later onset. Researchers selected another 595 patients from families with one affected member, matched for sex and age at onset. This allowed them to examine whether the association also appeared in a group with a lower family burden of illness.

The 223 earlier onset patients represented about 37.5% of the first cohort, calculated as 223 divided by 595. That is much higher than the estimated 8% among schizophrenia patients generally. These selected research cohorts should therefore not be treated as a population survey of how commonly schizophrenia begins in childhood or adolescence.

The paper's methods define earlier onset as psychotic symptoms before age 18, but its results section labels the group as age 18 or younger. That wording difference matters for patients whose first symptoms occurred at exactly 18 and should be clarified when interpreting the study's age boundary.

How DNA was used to predict brain activity

The researchers did not directly measure gene activity in the brains of the Taiwanese participants. They used DNA variants measured with the PsychChip array and prediction models built from international brain tissue datasets to estimate the component of gene expression influenced by inherited variation.

Gene expression is the process through which a cell uses instructions encoded in DNA. A transcriptome records RNA activity, providing information about which genes are being used and at what levels. By connecting DNA differences with expression patterns in reference tissue, researchers can estimate genetically regulated expression without collecting brain tissue from living patients.

The reference panels came from the CommonMind Consortium's dorsolateral prefrontal cortex dataset and the GTEx brain cortex dataset. The CommonMind panel included 254 people with schizophrenia, 52 with bipolar disorder and 279 controls. GTEx contributed data from 205 general population donors.

The team also used schizophrenia genetic association results from the Psychiatric Genomics Consortium, covering 76,755 cases and 243,648 controls, to weight the predicted expression signals. Those weighted estimates formed gene expression risk scores, or GeRS.

This differs from a conventional polygenic risk score, or PRS, which combines the effects of many DNA variants. GeRS attempts to connect inherited variation with the activity of particular genes in relevant tissue. It remains a prediction, however, rather than a direct measurement of every biological influence on a person's brain.

What Module 10 shows, and what the odds mean

Rather than examining genes only one at a time, the researchers grouped them into 13 established coexpression modules. A coexpression module is a collection of genes whose activity tends to vary together, suggesting shared biological functions or regulation.

Module 10 produced the most consistent association with earlier onset across the two family cohorts. The institute describes a 58 gene network and reports that each increase of one standard deviation in its expression risk score was associated with an adjusted odds ratio of 1.32, even after accounting for the conventional polygenic risk score.

The published study reports an adjusted odds ratio of 1.22 for M10sub-GeRS, with a 95% confidence interval of 1.07 to 1.40. The institute's account and the journal abstract therefore give different estimates under different score labels. They should not be presented as a single interchangeable result. The journal page notes a correction dated August 3, 2026, but the information presented here does not establish whether that correction explains the difference.

Both estimates concern odds of earlier rather than later onset among people already diagnosed with schizophrenia. An odds ratio of 1.32 means 32% higher odds per standard deviation of the score in that analysis, not a 32 percentage point increase in the probability of developing schizophrenia. Neither figure establishes a child's absolute risk of illness.

Four shared hubs, rather than four proven causes

Within the 58 gene network, the researchers used two complementary approaches to identify genes occupying central positions. Ingenuity Pathway Analysis examines molecular relationships, while STRING examines protein interaction networks. COPS6, MRPS7, RUVBL2 and PSMD2 were prioritized by both.

A hub gene is a gene with a central position in the network being analyzed. That designation makes it a useful research candidate, but does not prove that changing its activity causes schizophrenia or that targeting it would safely treat the disease.

The paper reports nine genes prioritized through the Ingenuity approach and ten through STRING. The four highlighted genes are shared candidates, not the complete set of hubs identified by either method.

Functional analyses connected the hub networks with protein breakdown, DNA repair, chromatin remodeling, mitochondrial translation and aminoacyl tRNA synthesis. These processes help cells maintain proteins, manage genetic instructions and produce components needed for normal operation. Chromatin remodeling changes how DNA is packaged and accessed, while mitochondrial translation produces proteins inside the cell structures responsible for much of its energy supply.

These are fundamental cellular functions, rather than a single schizophrenia specific pathway. Their appearance in the analysis points toward coordinated regulation and cellular maintenance as subjects for further investigation.

Developmental timing adds a biological clue

The institute reports that the highlighted genes were mainly enriched in excitatory neurons, cells that generally increase the likelihood that other neurons will transmit signals. Their expression mostly began increasing during late fetal development and continued through childhood and adulthood.

This pattern is consistent with a possible role in brain development. It does not establish when a disease process begins in an individual patient, nor show that the expression pattern inevitably produces symptoms during adolescence.

Single cell analyses also connected the hub genes with oligodendrocyte functions and B lymphocyte immune regulation networks. Oligodendrocytes produce the insulating material around nerve fibers; B lymphocytes are immune cells. These findings broaden the biological picture beyond neurons, without proving that either cell type independently causes earlier onset.

The authors of the study, a team investigating psychiatric genetics and gene regulation, describe the findings in their abstract as follows:

These findings highlight gene regulatory networks linking genetic susceptibility to molecular mechanisms in early-onset SZ and may inform risk stratification and targeted interventions.

The conditional wording matters. The study supports further investigation of these networks, rather than demonstrating that they can already guide treatment decisions.

How the researchers tested the signal

The paper describes 1,000 simulations using random gene sets matched in size to the identified module. The empirical p values were 0.006 and 0.001 in the two samples, supporting the finding that the module association was stronger than typically seen with comparable random sets.

The authors also tested a method designed to account for statistical inflation that can arise when a trait is influenced by many genetic variants. This sensitivity analysis weakened individual gene signals but did not materially change the module association. That distinction supports the network result more directly than a claim that each highlighted gene has been independently established as a cause.

There are further limits to individual prediction. In the CommonMind reference model, average prediction performance across genes was 8%, with substantial variation. The paper also reports that no individual predicted gene expression association with earlier onset reached its stringent significance threshold in either cohort in that analysis.

The evidence is therefore strongest as a lead about coordinated gene networks. It does not establish a reliable screening threshold, an individual prognosis or a clinically validated biomarker.

What happens before a treatment can follow?

Current antipsychotic treatment largely works through neurotransmitter pathways, including dopamine. Chen described the four candidates as possible starting points for new therapies if their actual roles in disease formation and neural development can be confirmed.

That requires moving beyond statistical association. Further work must establish which changes in gene activity matter, how they affect developing brain cells, and whether intervening in those processes would be effective and safe. The institute also identifies the remaining genes in the 58 gene module as candidates for further investigation.

The chronology separates the long collection period from publication and subsequent public discussion:

• 1998 to 2002: recruitment for the Taiwan Schizophrenia Linkage Study.
• 2009 to 2014: recruitment for the Taiwan schizophrenia trio study.
• July 3, 2026: online publication in Molecular Psychiatry.
• August 3, 2026: correction listed on the journal page.
• October 12, 2026: public reporting of the institute's research announcement.

No clinical trial start, regulatory submission or date for a diagnostic test is announced. The immediate contribution is a more specific set of molecular candidates for explaining earlier illness, not a change to established patient care.

Key Points

• Researchers analyzed 1,190 schizophrenia patients across two Taiwanese family cohorts.
• A 58 gene network, Module 10, was consistently associated with earlier onset.
• COPS6, MRPS7, RUVBL2 and PSMD2 were prioritized by both network analysis methods.
• The institute reports an adjusted odds ratio of 1.32; the journal abstract reports 1.22 for M10sub-GeRS.
• The findings concern predicted gene expression and do not prove that the four genes cause schizophrenia.
• Further validation is needed before any diagnostic or treatment application.