Genetic Analysis of Psychosis Biotypes: Shared Ancestry-Adjusted Polygenic Risk and Unique Genomic Associations

Abstract

The Bipolar-Schizophrenia Network for Intermediate Phenotypes (B-SNIP) created psychosis Biotypes based on neurobiological measurements in a multi-ancestry sample. These Biotypes cut across DSM diagnoses of schizophrenia, schizoaffective disorder and bipolar disorder with psychosis. Two recently developed post hoc ancestry adjustment methods of Polygenic Risk Scores (PRSs) generate Ancestry-Adjusted PRSs (AAPRSs), which allow for PRS analysis of multi-ancestry samples. Applied to schizophrenia PRS, we found the Khera AAPRS method to show superior portability and comparable prediction accuracy as compared with the Ge method. The three Biotypes of psychosis disorders had similar AAPRSs across ancestries. In genomic analysis of Biotypes, 12 genes and isoforms showed significant genomic associations with specific Biotypes in Transcriptome-Wide Association Study (TWAS) of genetically regulated expression (GReX) in adult brain and fetal brain. TWAS inflation was addressed by inclusion of genotype principal components in the association analyses. Seven of these 12 genes/isoforms satisfied Mendelian Randomization (MR) criteria for putative causality, including four genes TMEM140, ARTN, C1orf115, CYREN, and three transcripts ENSG00000272941, ENSG00000257176, ENSG00000287733. These genes are enriched in the biological pathways of Rearranged during Transfection (RET) signaling, Neural Cell Adhesion Molecule 1 (NCAM1) interactions, and NCAM signaling for neurite out-growth. The specific associations with Biotypes suggest that pharmacological clinical trials and biological investigations might benefit from analyzing Biotypes separately.

Figure 1.. Overview of the study workflow.

GWAS = Genome Wide Association Study, PRS = Polygenic Risk Score for SCZ, AAPRS = Ancestry-Adjusted Polygenic Risk Score for SCZ, TWAS = Transcriptome Wide Association Study. PGC = The Psychiatric Genomics Consortium, B-SNIP = The Bipolar and SCZ Network for Intermediate Phenotypes consortium, GTEx = The Genotype-Tissue Expression project. SCZ = Schizophrenia, SAD = Schizoaffective Disorder, BD = BD with psychotic features, HC = Healthy Control. BT1 = Biotype 1, BT2 = Biotype 2, BT3 = Biotype 3. EUR = European, AFR = African, AMR = Admixed American, EAS = East Asian, SAS = South Asian. GReX = Genetically Regulated eXpression.

Figure 2.. Prediction accuracy of PRSs for case-control status before and after ancestry adjustment within and across 5 ancestries.

All the PRSs are calculated based on EUR, AFR and Asian SCZ GWAS summary statistics. (a) The area under the receiver operating characteristic (ROC) curve (AUC) of PRSs. (b) The proportion of the case-control variance (Nagelkerke’s pseudo-R²) explained by PRSs. Lines for Nagelkerke’s pseudo-R² in (b) correspond to 95% confidence intervals calculated via 1000 bootstrapping. The five ancestries were assigned by Random Forest inferred method based on 1KG reference. EUR = European, AFR = African, AMR = Admixed American, EAS = East Asian, SAS = South Asian, ALL = Combined multi-ancestry individuals of all the five ancestries. “Unadjusted” risk scores are the --meta option results from PRS-CSx prior to post hoc ancestry adjustment, and “Adjusted” refers to AAPRS (Ancestry-Adjusted Polygenic Risk Score) with post hoc ancestry adjustment of Khera or Ge. We find no overall advantage in prediction accuracy of case-control status for either adjustment method.

Figure 3.. Effects of post hoc ancestry adjustment on the overlap of PRS-CSx (meta option) distributions among 5 ancestries.

Figure shows the overlaps of density kernels of PRSs between different ancestries before and after ancestry adjustment. ‘estOV’ = estimated overlapping area of risk scores between different ancestries. Standard error of estOV is calculated by 1000 bootstrap draws (meaning 1000 iterations with bootstrapping), and the labelled error bar of upper and lower values is estOV +/− SE. Between the two PRS post hoc ancestry adjustment methods, Khera adjustment gave greater PRS overlap (97% vs. 96%) between different ancestries, both significantly higher than the overlap (91%) of unadjusted PRSs.

Figure 4.. Percentage of SCZ AAPRS variance explained by each factor in two-way Analysis of Variance (ANOVA).

“Unadjusted” is PRS-CSx meta PRS before post hoc ancestry adjustment. “Adjusted” refers to Ancestry-Adjusted Polygenic Risk Scores (AAPRSs) with post hoc ancestry adjustment of Khera or Ge. Minimal ancestry variance is desirable for AAPRS in a combined multi-ancestry sample. The residual variance would be the effect of SNPs on the PRS. Ideally, this would take up the largest share of the PRS variance, and the effects of the other variables would be minimized, as seen in the Khera AAPRS. With the Khera method, ancestry accounted for 1% of AAPRS variance (P = 4.37e-04) and Biotypes significantly accounted for 4% (P = 2.20e-17) vs. 14% (P = 3.03e-70) and 3% (P = 3.08e-17) for Ge. There were no significant interactions between Biotype and ancestry in either adjustment method (Table S6).

Figure 5.. Biotype differences on SCZ AAPRS.

AAPRS refers to Ancestry-Adjusted Polygenic Risk Score with post hoc ancestry adjustment of Khera. The differences of AAPRS among Biotypes in the combined multi-ancestry dataset are shown by violin plot. Wilcoxon tests were used for the comparison. Bonferroni-corrected significance threshold over 6 two-sample Wilcoxon tests is P-value < 8.33e-03. Only significant comparison results are labeled with asterisks. *** indicates P-value < 1.67e-04.

Figure 6.. Prediction accuracy of EUR SCZ and BD GWAS-summary-statistics-based PRSs for case-control status for different diagnostic groups within EUR ancestry before and after ancestry adjustment.

(a) The area under the receiver operating characteristic (ROC) curve (AUC) of PRSs. (b) The proportion of the case-control variance (Nagelkerke’s pseudo-R²) explained by PRSs. Lines for Nagelkerke’s pseudo-R² in (b) correspond to 95% confidence intervals calculated via 1000 bootstrapping. PGC EUR SCZ and EUR BD GWAS summary statistics were used for PRS construction. “based” refers to the diagnosis of the GWAS summary statistics used to generate the PRS.

Figure 7.. Histogram of significantly (P < 0.05) enriched biological pathways for the seven Biotype causal genes in Table 2.

Legend: RET = Rearranged during Transfection, NCAM = Neural Cell Adhesion Molecule.

Figure 8.. Representative Quantile-Quantile (Q-Q) plot of gene-level TWAS results in adult brain for Biotype 1 versus Healthy Control in the combined sample of the five ancestries.

(a) QQ plot of TWAS results without covariates. (b) QQ plot of TWAS results with the first five genotype principal components (PCs) as covariates. Each blue dot represents for a gene. Results show that the inflation was successfully addressed by including the genotype PCs in the logistic regression model in the association test.