Preprint available; revised version under review at Communications Medicine

Genetic risk and psychiatric diagnoses leave different marks on the cortex

Rare genetic changes that raise the risk of autism, schizophrenia and other conditions are linked to the surface area of the brain’s cortex. The diagnoses themselves are linked mostly to its thickness.

The cortex as a folded sheet: thickness is its depth, surface area is its extent. Thickness Surface area

We compared three kinds of brain MRI results, each from its own data source:

Rare genetic variants
18 copy number variants and chromosome changes: 730 carriers of 11 CNVs or sex chromosome aneuploidies and 870 controls (individual-level MRI), plus 7 further variants from published meta-analyses.
Common genetic variants
Lead variants for each disorder from European-ancestry genetic studies, looked up in a cortical GWAS of 51,665 people (Grasby et al. 2020); risk scores tested in 31,413 UK Biobank participants of European ancestry who carry none of the studied CNVs.
Psychiatric diagnoses
8 disorders in 9 ENIGMA case-control groups (13,812 cases), from published ENIGMA studies.

Surface area tracks genetic risk; thickness tracks diagnosis

Genetic variants were linked to surface area differences about 11 times larger than those seen in psychiatric diagnoses. Diagnoses were linked mainly to thickness.

Variants differ in direction: a deletion and a duplication of the same region can shift the cortex in opposite ways. Start with “Direction and size” to see this; “Size only” then compares how strongly each variant or diagnosis is linked to each measure, whatever the direction, which is how the paper compares them.

Thickness differences in diagnoses come mostly from groups taking medication

Thickness differences were about 8 times larger in medicated than in unmedicated groups, while surface area did not differ. This suggests thickness partly reflects illness and treatment rather than genetic risk. Medication status may also stand for illness severity or duration.

Each dot is one genetic variant, or one diagnostic group compared with controls. Tap or hover over a dot to see its name, its effect size with sign, and its sample.

In “Direction and size”, dots left of zero mean smaller than controls, right of zero larger. In “Size only”, further right means a larger difference in either direction, on a log scale, and the dotted bar marks each group’s median.

Diagnoses are the 9 main case-control groups. Age-stratified groups (young and pediatric) were sensitivity analyses and are described in the paper.

Surface area. The area of the cerebral cortex, the folded outer sheet of the brain. Surface area is largely set during fetal development, when the pool of cortical progenitor cells expands.

Cortical thickness. The depth of the cortical sheet. Thickness keeps changing through adolescence and adulthood and can differ with age, illness and medication.

Effect size (Cohen’s d). The difference between a group (carriers or patients) and controls, divided by the spread of the measure. It lets studies of different sizes and scanners be compared on one scale. Here effects are adjusted for age, sex and scanning site. A group-level difference; it does not describe any single person.

Copy number variant (CNV). A rare genetic change in which a segment of a chromosome is missing (deletion) or present in an extra copy (duplication). The CNVs studied here are known to raise the risk of psychiatric and neurodevelopmental conditions. Sex chromosome aneuploidies (for example XXY) and trisomy 21 change the dose of whole chromosomes and are included with them.

From Figure 2 of the paper. Samples: 730 carriers of 11 CNVs or sex chromosome aneuploidies and 870 controls (individual-level MRI), plus 7 further variants from published meta-analyses; 8 disorders in 9 ENIGMA case-control groups (13,812 cases); medication subgroups: 9 medicated and 5 unmedicated adult groups.

Genetic variants: sensorimotor regions. Diagnoses: association regions.

Rare and common variants showed their largest surface area effects in regions for movement and the senses, the same regions where heritability is highest. Diagnoses showed theirs in regions for higher-order thinking.

The brain shows one shared regional pattern at a time. Orange marks regions with higher scores, purple lower. Drag to rotate; switch to the inner side to see the medial surface.

Choose Gradient to see the reference ordering of regions, from sensorimotor (blue) to association (yellow).

The chart below places each pattern on that gradient, as in Figure 5 of the paper. Left of zero means larger effects in sensorimotor regions, right of zero in association regions. Regional heritability, from twin and genetic studies, sits on the same side as the genetic variants. Filled symbols passed the spin test (p < 0.05); open symbols did not. Common-variant patterns were available for surface area only.

The same regional preference held with other ways of summarising the maps and when leaving out one variant or diagnosis at a time.

Shared pattern (first principal component). For each set of maps (for example, all CNV surface area maps), principal component analysis finds the pattern that best summarises how regions vary together. The map shows each region’s score on that first pattern. Units are arbitrary; only the relative pattern matters. The shared pattern summarises a modest part of the variation; individual variants and diagnoses differ, and some go in opposite directions.

Sensorimotor-to-association gradient. A well-established ordering of cortical regions, from primary sensory and motor areas (vision, touch, movement), which develop early, to association areas (planning, language, social thinking), which keep developing into adulthood.

Spin test. Neighbouring brain regions tend to have similar values, which can make two maps look alike by chance. The spin test rotates one map many times on the cortical sphere and asks how often a random rotation matches as well as the real map.

Desikan atlas. A widely used atlas that divides each hemisphere of the cortex into 34 named regions. Values were averaged across the two hemispheres and are drawn on one hemisphere.

From Figures 3 to 5 of the paper. Samples: 11 variants with individual-level MRI; disorder-linked common variants in the cortical GWAS; ENIGMA diagnostic groups.

Effects in both directions offset each other

Disorder-linked variants were tied to larger or smaller surface area in similar numbers. Added together into a risk score, these effects largely offset each other.

Illustration: variant effects pointing up and down add to almost nothing; the same effects all pointing up add to a large total. As observed = All one way =

Polygenic risk scores for bipolar disorder and schizophrenia showed no association with total surface area in 31,413 UK Biobank participants. Simulations using the measured variant effects (bipolar disorder) show why: with the observed mix of directions, a risk score explains more than 100 times less surface area than if the same effects all pointed one way.

Disorder-linked variants also varied more in their surface area effects than random variants matched for frequency: 1.23 times the spread for bipolar disorder (p = 0.089, not significant).

The arrows are an illustration, not data. The chart shows the simulations: each dot is the median of 10,000 simulated risk scores, and the line is its 95% interval. The vertical axis is logarithmic; each step is ten times larger.

The sizes of the variant effects are the same in both scenarios; only their directions differ.

For bipolar disorder and schizophrenia, 53 to 57% of risk variants were tied to smaller surface area and the rest to larger. These analyses cover these two disorders only, using European-ancestry genetic studies and participants. Rare variants showed the same balance: about half of the CNVs scored positively and half negatively on the shared surface area pattern.

Common variant. A single-letter difference in DNA carried by many people. Large genome-wide studies have linked hundreds of such variants to each psychiatric disorder; each one has a very small effect.

Polygenic risk score. A score that adds up a person’s common variants linked to a disorder, weighting each by its effect on disorder risk. It summarises genetic liability but does not diagnose anyone.

Variance ratio. The spread of surface area effects for disorder-linked common variants, divided by the spread for randomly chosen variants with matched allele frequency. Above 1 means the disorder variants have larger effects, in both directions.

Simulated risk-score association (R²). In simulations, risk scores were built from the measured variant effects, either with their observed mix of directions or with all directions made the same, keeping the sizes unchanged. R² is the share of surface area differences between people that the score explains.

From the paper (Results). Samples: 234 bipolar disorder lead variants after allele matching; 31,413 UK Biobank participants of European ancestry who carry none of the studied CNVs.

What this means

  • Brain differences seen in people with a psychiatric diagnosis are not a direct readout of their genetic risk.
  • Surface area, especially in sensorimotor regions, may be a clearer window onto genetic effects; thickness may reflect factors linked to illness, such as medication.
  • Averaging many genetic variants together can hide real effects that point in opposite directions.

These are group-level associations from cross-sectional data. They do not predict or diagnose anything in an individual. How the study was done