Polygenic Risk for Schizophrenia, Brain Structure and Environmental Risk in UK Biobank

Schizophrenia is a heritable neurodevelopmental disorder characterized by neuroanatomical changes in the brain but exactly how increased genetic burden for schizophrenia influences brain structure is unknown. Similarly, the impact of environmental risk factors for schizophrenia on brain structure is not fully understood. We investigated how genetic burden for schizophrenia (indexed by a polygenic risk score, PRS-SCZ) was associated with cortical thickness (CT), cortical surface area (SA), cortical volume (CV) and multiple subcortical structures within 18,147 White British ancestry participants from UK Biobank. We also explored whether environmental risk factors for schizophrenia (cannabis use, childhood trauma, low birth weight and Townsend social deprivation index) exacerbated the impact of PRS-SCZ on brain structure. We found that PRS-SCZ was significantly associated with lower CT in the frontal lobe, insula lobe, lateral orbitofrontal cortex, medial orbitofrontal cortex, posterior cingulate cortex and inferior frontal cortex, as well as reduced SA and CV in the supramarginal cortex and superior temporal cortex, but not with differences in subcortical volumes. When models included environmental risk factors as covariates, PRS-SCZ was only associated with lower SA/CV within the supramarginal cortex, superior temporal cortex and inferior frontal cortex. Moreover, no interactions were observed between PRS-SCZ and each of the environmental risk factors on brain structure. Overall, we identified brain structural correlates of PRS-SCZ predominantly within frontal and temporal regions. Some of these associations were independent of environmental risk factors, suggesting that they may represent biomarkers of genetic risk for schizophrenia.


Brain imaging variables
All neuroimaging data were acquired, pre-processed, quality controlled and made available by UK Biobank (https://biobank.ctsu.ox.ac.uk/crystal/crystal/docs/brain_mri.pdf). Details on the acquisition parameters and the imaging protocol are documented online and are described within the protocol paper 42 . Derived phenotypes were used in this study. The main neuroimaging data consisted of global, lobar and regional values for CT, SA and CV of 33 regions (data of temporal pole are not available) in each hemisphere from the Desikan-Killiany cortical atlas 40 . Global and lobar values were calculated as per Neilson et al. 12 .
In addition, we analysed subcortical volumes including thalamus, caudate, putamen, pallidum, hippocampus, amygdala, accumbens processed by subcortical volumetric segmentation in Freesurfer. A more detailed description of these variables is provided within supplementary materials.

Environmental risk factors
Childhood trauma was calculated by summing five online questions on previous adversity (data field 20487 to 20491). These questions cover the self-reported frequency of feeling loved, being physically abused, feeling hated, being sexually molested and being taken to the doctor when needed as a child. Participants answered these questions by selecting "Prefer not to answer", "Never true", "Rarely true", "Sometimes true", "Often" or "Very often true" (supplementary materials).
Birth weight (https://biobank.ctsu.ox.ac.uk/crystal/field.cgi?id=20022) was reported by participants during a verbal interview. To maximise the number of participants included in analyses, missing values at the initial assessment visit were replaced by those at the first repeat assessment visit or the imaging visit.
The Townsend deprivation index (https://biobank.ctsu.ox.ac.uk/crystal/field.cgi?id=189) is based on postcode reported at baseline assessment in UK Biobank. A greater Townsend index score implies a higher level of neighbourhood social and economic deprivation.
All rights reserved. No reuse allowed without permission.
(which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

Statistical analysis
All analyses were carried out using Stata/MP 15.0. For associations between PRS-SCZ and brain structure, PRS-SCZ was set as the independent variable; each brain structural measure was set as an outcome. PRS-SCZ × hemisphere interactions were examined in a repeated measures format to determine whether analysis of left and right homologous structures separately was required, with sex, age, age 2 , hemisphere, ICV, scanner positions on the x, y and z axes, genotype array and the first fifteen genetic principal components included as covariates. For brain measures that did not show PRS-SCZ × hemisphere interaction, we repeated the linear mixed model with hemisphere as a fixed effect and included the same covariates as in analyses testing for interaction effects. If there was a significant interaction, analyses on both lateralised structures would be conducted separately. PRS-SCZ and each brain outcome were rescaled into zero mean and unitary standard deviation. Participants with PRS and each brain outcome beyond three standard deviations were excluded. False Discovery Rate (FDR) correction 43 , with a significance threshold of p < 0.05, was applied for each metric individually, by correcting over all eight possible lobar structures, twenty-six parcellations or seven subcortical volumes, using 'p.adjust' function in R.
Furthermore, considering the potential impact of environmental risk factors, we investigated whether associations between PRS-SCZ and brain measures were significant when childhood trauma, cannabis use, birth weight and Townsend deprivation index were included as additional covariates. In total, 8,707 individuals had complete data for all four environmental factors. Because the subsample was much smaller compared to that in the main analyses, we wished to clarify whether the reduction in sample size influenced the results. Therefore, we repeated previous analyses in this subsample (N = 8,707) without controlling for environmental risk factors. For further examination of the influence of environmental risk factors on the relationship between PRS-SCZ and brain structure, we excluded 21 individuals who definitely had used antipsychotic medication (self-reported at the imaging visit), considering that antipsychotic medication may also influence brain structure 44,45 . We also examined the interaction between PRS-SCZ and each of the environmental risk factors on brain structure. The interaction term, together with main effects of the PRS-SCZ and the environmental risk factor were included in the model, as well as same covariates in the analysis for PRS-SCZ and brain structure. Multiple comparisons correction for interaction effects was done using FDR correction (p < 0.05) in the same way. All rights reserved. No reuse allowed without permission.
(which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

Demographics
A total of 18,147 participants (ages 45-80 years, 8,651 males) were included in the analysis.
(which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.  Regions were mapped on the left hemisphere.
In the subsample with complete data on environmental risk factors (childhood trauma, cannabis use, birth weight and Townsend deprivation index) we found modest negative associations between PRS-SCZ and CT in the above regions (Table S4), although none of these associations were significant after FDR correction in this smaller sample. Similarly, after controlling for the environmental risk factors and excluding participants who had used antipsychotic medication, there were no significant associations, although a similar negative pattern of association was found (Table 3 shows significant associations only; Table S5 lists all the results; Figure S1).

Associations between PRS-SCZ and surface area
No interactions were observed between PRS-SCZ and hemisphere for global or regional SA (Table S2). Although PRS-SCZ was not associated with global SA (β = -0.002, p corrected = 0.672; Table S3; Figure 1B), we found an association with reduced SA in the supramarginal cortex (β = -0.016, p corrected = 0.026; Table 2) and the superior temporal cortex (β = -0.015, p corrected = 0.026). Analyses in the subsample also found significant associations with these two regions, in addition to inferior frontal cortex (Table S4).

Associations between PRS-SCZ and cortical volume
Analyses for the interaction between PRS-SCZ and hemisphere on global or regional CV did not demonstrate any significant effects (Table S2). Similar to the results for surface area, PRS-SCZ was not associated with global CV (β = -0.004, p corrected = 0.449; Table S3; Figure   1C), but there were significant negative associations with supramarginal cortex (β = -0.016, p corrected = 0.013; Table 2) and superior temporal cortex (β = -0.018, p corrected < 0.001).
Analyses in the subsample with complete environmental risk factor data found significant negative associations with the supramarginal cortex and inferior frontal cortex (Table S4).
All rights reserved. No reuse allowed without permission.
(which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

Associations between PRS-SCZ and subcortical volumes
Analyses for the interaction between PRS-SCZ and hemisphere found a significant interaction effect for the hippocampus (β = 0.011, p corrected = 0.025; Table S2) and nucleus accumbens (β = -0.015, p corrected = 0.025). Therefore, we also examined the left and right hippocampus and nucleus accumbens separately in following analyses.
PRS-SCZ had a negative association with volume in the right accumbens before (β = -0.013, p uncorrected = 0.043; Table S3), but not after FDR correction. Similarly, when environmental risk factors were taken into account, no subcortical structure survived correction for multiple comparisons correction (Table S5).
For the interaction between PRS-SCZ and environmental risk factors, we found no statistically significant results, albeit some modest associations before FDR correction. For example, PRS-SCZ×childhood trauma on SA in the medial orbitofrontal cortex (β = -0.006, p uncorrected = 0.013; Table S6): there was a negative association of PRS-SCZ with SA in the medial orbitofrontal cortex in participants exposed to more childhood trauma and a positive association in those who experienced less childhood trauma. More detailed results for the interaction effect are shown in Table S6-S9. Statistics for main effects of the environmental factors are also given (table S10-S13).
All rights reserved. No reuse allowed without permission.
(which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.
The copyright holder for this preprint this version posted April 17, 2021. ; 1 0

Discussion
In this study we have observed associations between an increased genetic liability for developing schizophrenia and wide-spread cortical differences in generally population, including lower CT in frontal lobe, insula lobe, lateral orbitofrontal cortex, medial orbitofrontal cortex, inferior frontal cortex and posterior cingulate cortex, as well as reduced SA and CV in the supramarginal cortex and superior temporal cortex. In addition, the associations with reduced SA and CV remained after adjustment for environmental risk factors, indicating that some cortical alterations might be driven predominantly by genetic liability for schizophrenia rather than environmental risk factors. Finally, our results suggest that PRS-SCZ and these environmental risk factors may contribute independently to brain abnormalities in schizophrenia. Overall, our findings have expanded on previous findings by making use of data within a much larger population-based sample.

PRS-SCZ and CT
Previous studies of the relationship between PRS-SCZ and global CT have been inconsistent, with global cortical thinning 12,29 or no association 46 reported. Here, we found no association between PRS-SCZ and global CT, which is unsurprising because schizophrenia-associated genetic variants showed no significant enrichment in mean CT 9 . For regional measures, we

PRS-SCZ and SA
We found that PRS-SCZ was not associated with global SA. Regarding this, previous studies have reported inconsistent findings 12,46,54 . Analyses for regional SA found negative associations within temporal cortices: the supramarginal cortex and superior temporal cortex.
This is in line with the significant enrichment of schizophrenia GWAS loci in SA of temporal regions 9 . Previous studies support structural differences in superior temporal gyrus and inferior frontal cortex as possible biomarkers in schizophrenia 55,56 . Baseline grey matter All rights reserved. No reuse allowed without permission.
(which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.
Similarly, the involvement of supramarginal and inferior frontal cortex in auditory hallucinations has been extensively documented 60 and the supramarginal gyrus is associated with delusions of reference and persecutory delusions 47 . Alterations in these regions may represent general susceptibility to psychotic symptoms and vulnerability to developing schizophrenia.

PRS-SCZ and CV
For the relationship between PRS-SCZ and brain volume, findings were also contradictory.
Van Scheltinga et al. 25 found PRS-SCZ was significantly associated with smaller total brain volume regardless of disease status, but later studies found different results even within larger samples 12,26,61,62 . We found no such association. However, we found significant volume reduction in supramarginal cortex and superior temporal cortex. When environmental risk factors were taken into account, an association with inferior frontal cortex volume was significant, and both supramarginal cortex and superior temporal cortex were nominally significant. As indicated earlier, this is in line with putative roles for these structures in processing auditory inputs, and their relation to auditory hallucinations in schizophrenia 57,63,64 .

PRS-SCZ and subcortical volumes
For subcortical volumes, although some studies have reported associations of PRS-SCZ with smaller thalamus and pallidum volume 13,26,54,65 , others have found no association 5,46,66 . Here, we observed no association with thalamus or pallidum volumes and a modest association with the right nucleus accumbens. Smaller nucleus accumbens has been reported in individuals with schizophrenia 4,5 , as well as in their first-degree relatives 22,48 . The nucleus accumbens is implicated in numerous neurological and psychiatric disorders, and is a main target of antipsychotic drugs and neurosurgical intervention in schizophrenia 67,68 . However, it should be noted that the association with the right nucleus accumbens was no longer significant after FDR correction. Overall, our data suggest that subcortical brain volumes are not strongly associated with genetic mechanisms of schizophrenia.

Influence of environmental risk factors
When environmental risk factors were included in the model as covariates and participants on antipsychotic medication were excluded, we observed a similar pattern of grey matter reduction; the effects for the supramarginal cortex, superior temporal cortex and inferior frontal All rights reserved. No reuse allowed without permission.
(which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

Strengths and Limitations
This study represents the largest neuroimaging PRS-SCZ study to date, identifying associations with multiple brain measures. Nevertheless, it is unclear to what extent these structural differences are causes, consequences or confounders of schizophrenia. Importantly, we also considered the influence of environmental risk factors for schizophrenia and explored gene-by-environment interactions on structural brain measures. Considering the modest significant associations, further studies which take these variables into account will need to be undertaken with larger samples. Furthermore, our study was conducted in White British participants aged from 45 to 80, which may restrict generalization of our findings to the other populations with different ancestry or age range. Finally, the environmental factors in this study were limited to childhood trauma, cannabis use, Townsend deprivation index and birth weight.
To develop a fuller picture of gene-by-environment interactions, additional studies on other factors such as migration background 91 are required.

Conclusion
In summary, this study suggests that grey matter reductions in multiple regions, predominantly in frontal and temporal regions, are neuroanatomical correlates of increased genetic liability for schizophrenia. Further, environmental risk factors such as childhood trauma, cannabis use, Townsend deprivation index and birth weight, appear to contribute very little to these associations, plus genetic liability for schizophrenia and these environmental factors contribute independently to brain abnormalities in schizophrenia. Our findings are meaningful in terms of identifying neuroimaging-based biomarkers for schizophrenia and elucidating the complex pathophysiology of schizophrenia.
All rights reserved. No reuse allowed without permission.
(which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.
The copyright holder for this preprint this version posted April 17, 2021. (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.
The copyright holder for this preprint this version posted April 17, 2021.  (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.  (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.  (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.  (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.