Plasma Lipid Alterations Track Multidimensional Psychosis Severity Across Diagnostic Boundaries

This study demonstrates that plasma lipidomic alterations, particularly in specific lipid classes like sphingolipids and phosphatidylcholines, robustly track multidimensional psychosis severity across schizophrenia and bipolar disorder, suggesting these lipid changes reflect transdiagnostic biological processes linked to illness burden that are not fully captured by categorical diagnoses or polygenic risk scores.

The Gist

The Big Picture: Looking for the "True" Measure of Illness

Imagine mental health disorders like Schizophrenia and Bipolar Disorder as two different brands of cars that sometimes break down in similar ways. Traditionally, doctors have tried to fix them by looking at the "brand name" (the diagnosis) or by checking a single gauge on the dashboard (like a symptom score).

But this study suggests that looking at the brand name or a single gauge isn't enough to understand how "broken" the car really is. Instead, the researchers wanted to see if they could find a universal dashboard warning light that tells them exactly how severe the problem is, regardless of whether it's a "Schizophrenia car" or a "Bipolar car."

They decided to look at the fuel and oil of the body (specifically, fats in the blood called lipids) to see if these chemical signals could reveal the true severity of the illness.

---

The Cast of Characters

1. The Patients: A group of over 1,300 people with serious mental health conditions.
2. The "Severity Score": Instead of just asking "Are you sick?", the researchers used a smart computer program (Machine Learning) to look at dozens of factors at once (hallucinations, thinking speed, ability to work, etc.). This gave them a "Severity Probability Score"—a number from 0 to 1 that says, "How much does this person look like they are in the most severe category of illness?"
3. The Genetic Blueprint (PRS): Think of this as the "family inheritance" of risk. It's a score based on a person's DNA that predicts how likely they are to get these illnesses.
4. The Blood Fats (Lipidomics): This is the main focus. The researchers took blood samples and measured 361 different types of fats. Think of these fats as the oil, grease, and fuel running through the body's engine.

---

What They Found: The "Oil Change" Analogy

1. The Genetic Blueprint Didn't Tell the Whole Story

The researchers checked if the "Genetic Blueprint" (DNA risk) matched the "Severity Score."

• The Result: It was a weak match. Knowing someone's family history of mental illness didn't perfectly predict how severe their current symptoms were.
• The Analogy: Imagine two cars with the same engine design (DNA). One is driving smoothly, and the other is smoking heavily. The engine design alone doesn't explain why one is broken down and the other isn't. Something else is happening right now.

2. The Blood Fats Were the Real Storytellers

This was the big discovery. The researchers found that specific types of fats in the blood changed in a very predictable way as the illness got worse.

• The Result: They found nine specific "families" of fats that were linked to severity.
  • Some fats (like certain "plasmalogens" and "sphingolipids") went up when the illness was severe.
  • One specific type of fat went down.
• The Analogy: It's like checking the oil in a car. When the engine is running hot and under extreme stress, the oil changes color and consistency. The researchers found that the blood fats of people with severe psychosis looked like "overheated engine oil," regardless of whether they had a Schizophrenia diagnosis or a Bipolar diagnosis.
• Key Insight: These fat changes happened even after accounting for the specific diagnosis or the medication the patients were taking. This suggests the fats are reacting to the stress of the illness itself, not just the label on the bottle.

3. The "Middleman" Mystery (Mediation)

The researchers asked: Does the DNA cause the fat changes, which then cause the severe symptoms? (Like: Bad Engine Design $\rightarrow$ Bad Oil $\rightarrow$ Car Breakdown).

• The Result: No. They couldn't prove that the DNA was directly causing the fat changes that led to the severe symptoms.
• The Analogy: It turns out the "Bad Oil" isn't necessarily caused by the "Engine Design." Instead, the oil might be getting dirty because the car has been driven too hard for too long (illness duration, inflammation, stress, or lifestyle). The fat changes are more like a symptom of the current state (a "state marker") rather than a permanent genetic flaw (a "trait marker").

4. The "Teamwork" Test (Canonical Correlation)

Finally, they tried to see if the DNA and the Blood Fats worked together as a team to predict the illness.

• The Result: They didn't really work together. The DNA and the fats seemed to be telling two different stories.
• The Analogy: It's like having a mechanic look at the blueprints (DNA) and another look at the oil (Fats). They aren't agreeing on what's wrong. This reinforces the idea that the fat changes are happening independently of the genetic risk, likely due to the current state of the body.

---

Why Does This Matter?

1. We need better maps.
Currently, we treat mental illness like we treat a broken lightbulb: if it's a "Schizophrenia bulb," we use one tool; if it's a "Bipolar bulb," we use another. This study says, "Wait, the bulb is actually just very dim." The severity of the dimness is what matters, and the blood fats are a great way to measure that dimness across all types of bulbs.

2. New targets for treatment.
If these fat changes are caused by the body being under stress (inflammation, oxidative stress), maybe we can treat the illness by fixing the "oil." Instead of just trying to fix the "genetic blueprint" (which is hard to change), doctors might be able to use drugs or lifestyle changes to normalize these fats and reduce the severity of the symptoms.

3. A new way to measure success.
Right now, doctors ask, "Do you hear voices?" This study suggests we should also check the "blood oil." If the fats look healthier, the patient might be doing better, even if they still have a diagnosis.

The Bottom Line

This study found that blood fats act like a universal "severity meter" for psychosis. They change when the illness is severe, regardless of the specific diagnosis. However, these changes aren't directly caused by your genes; they seem to be the body's reaction to the illness itself. This opens the door to new ways of measuring and treating mental health based on the body's current state, rather than just the label on the chart.

Technical Summary

1. Problem Statement

Psychotic disorders, specifically Schizophrenia (SCZ) and Bipolar Disorder (BD), share significant symptom overlap (e.g., hallucinations, delusions) and genetic liability, yet current diagnostic categories often fail to capture the underlying biological heterogeneity.

• Limitations of Current Biomarkers: Previous lipidomic studies have yielded heterogeneous findings with low diagnostic specificity. Associations between lipid profiles and illness severity have been inconsistent when using unidimensional symptom scales (e.g., PANSS), suggesting these scales may fail to capture biologically relevant variation.
• Genetic vs. Clinical Disconnect: While Polygenic Risk Scores (PRS) effectively predict categorical diagnoses, they show weak associations with symptom severity and functional outcomes in diagnosed individuals.
• Knowledge Gap: It remains unclear whether peripheral plasma lipidomic alterations track multidimensional psychosis severity across diagnostic boundaries, and whether these lipid changes act as mediators between genetic liability (PRS) and clinical severity, or if they represent independent state-dependent processes.

2. Methodology

The study utilized data from the PsyCourse Study, a longitudinal deep-phenotype cohort in Germany/Austria.

• Cohort:
  • Total Sample: $N=1,320$ individuals with SCZ, schizoaffective disorder, brief psychotic disorder, or BD.
  • Lipid Subset: $N=428$ participants with comprehensive plasma lipidomics data.
• Key Variables:
  • Outcome Measure: A continuous "Severe Psychosis Subtype Probability" score (0–1). This was derived using a Support Vector Machine (SVM) classifier trained on multidimensional clinical variables (positive symptoms, cognitive impairment, functional disability) to identify a specific severe psychosis subtype.
  • Genetic Data: 13 Polygenic Risk Scores (PRS) covering psychiatric traits (SCZ, BD, MDD), cognitive traits (Educational Attainment, Alzheimer's), and personality/sleep traits. Lipid-metabolism-enriched PRS were also constructed by restricting SNPs to genes within the REACTOME lipid metabolism pathway.
  • Lipidomics: Non-fasting plasma samples analyzed via LC-MS. 361 lipid species across 16 classes were retained after quality control (excluding fasting-sensitive species and outliers).
• Statistical Approaches:
  • Regression Analysis: Ordinary Least Squares (OLS) with robust standard errors to test associations between PRS/lipids and severity probability, adjusting for age, sex, BMI, smoking, illness duration, diagnosis, and medication.
  • Lipid Class Enrichment: A Gene Set Enrichment Analysis (GSEA) adapted for lipidomics, ranking species by regression coefficients to identify coordinated class-level associations.
  • Mediation Analysis: Bootstrap-based mediation models to test if lipid classes transmit the effect of PRS on severity.
  • Canonical Correlation Analysis (CCA): To identify shared multivariate structure between the 13 PRS and 16 lipid classes and test their joint association with severity.

3. Key Contributions

• Transdiagnostic Severity Phenotyping: The study moves beyond categorical diagnoses by utilizing a machine-learning-derived continuous probability score for a "severe psychosis" subtype, allowing for the detection of biological signals that cut across SCZ and BD boundaries.
• Lipid Class vs. Species Resolution: The study demonstrates that while individual lipid species associations are often weak or confounded by diagnosis/medication, lipid class enrichment analysis reveals robust, reproducible signatures of severity.
• Integration of Multi-Omics: It is the first study to apply Canonical Correlation Analysis (CCA) to integrate PRS and peripheral lipidomics in a psychotic disorder cohort to test for shared genetic-metabolic architecture.
• Differentiation of Trait vs. State: By testing mediation and adjusting for medication, the study provides evidence distinguishing between genetic trait markers and state-dependent biological processes.

4. Key Results

A. Polygenic Risk Scores (PRS)

• Associations: Three PRS showed modest but significant associations with severe psychosis probability:
  • SCZ-PRS: Positive association ($\beta=0.02$).
  • BD-PRS: Negative association ($\beta=-0.02$), suggesting bipolar genetic liability may be orthogonal to severe psychotic features.
  • Educational Attainment (EA) PRS: Negative association ($\beta=-0.02$), aligning with cognitive impairment in severe cases.
• Limitation: When adjusted for diagnosis, these associations became non-significant, indicating PRS primarily differentiate diagnostic groups rather than severity within a transdiagnostic cohort.
• Lipid-Enriched PRS: No significant associations were found for PRS restricted to lipid metabolism genes.

B. Lipidomic Alterations

• Individual Species: Only three species (all PE-P plasmalogens) were significant before adjustment; none survived correction for diagnosis and medication.
• Lipid Class Enrichment: Nine lipid classes showed significant enrichment with severe psychosis probability ($q < 0.05$). These associations were largely robust to adjustment for diagnosis and medication:
  • Positively Associated (Increased in severity): Sphingomyelins (dSM), Ceramides (dCer), Phosphatidylcholines (PC), Triacylglycerides (TAG), Phosphatidylethanolamine plasmalogens (PE-P), Cholesteryl Esters (CE), and Plasmanyl-PC (PC-O).
  • Negatively Associated (Decreased in severity): Phosphatidylcholine plasmalogens (PC-P).
• Specific Patterns: A distinct pattern of PE-P increase and PC-P decrease suggests headgroup-specific ether-lipid remodeling. Sphingolipid enrichment (dSM, dCer) remained significant even after medication adjustment (though Ceramide attenuation was noted).

C. Mediation and Shared Variation

• Mediation: No significant indirect effects were found. Lipid alterations did not mediate the relationship between PRS and severity.
• Canonical Correlation (CCA): The first canonical correlation between PRS and lipid classes was modest ($r=0.33$) and not statistically significant ($p=0.604$). Neither the PRS nor lipid canonical scores predicted severity.

5. Significance and Conclusions

• Lipids Track Severity, Not Just Diagnosis: Plasma lipidomic variation tracks multidimensional psychosis severity across diagnostic boundaries. The findings suggest that lipid alterations reflect transdiagnostic biological processes linked to illness burden (state-dependent) rather than diagnosis-specific mechanisms or common-variant genetic liability.
• State vs. Trait: The lack of mediation and the robustness of lipid associations to medication/diagnosis adjustments suggest these lipid changes are likely driven by state-dependent processes (e.g., neuroinflammation, oxidative stress, metabolic dysfunction) rather than being direct downstream effects of inherited polygenic risk.
• Methodological Insight: The study highlights that unidimensional symptom scales may lack the sensitivity to detect lipid-severity relationships, whereas machine-learning-derived multidimensional phenotypes can reveal robust biological correlates.
• Future Directions: The findings support the potential of lipidomics as a biomarker for illness burden and treatment stratification. Future research requires larger samples, longitudinal designs, and integration with neuroimaging to clarify the temporal dynamics and causal mechanisms of these lipid alterations.

Limitations Noted: Peripheral plasma as a proxy for brain lipids, non-fasting samples, potential residual confounding from medication, and limited statistical power for mediation/CCA analyses.