Interaction Testing and Polygenic Risk Scoring to Estimate the Association of Common Genetic Variants With Treatment Resistance in Schizophrenia

Key Points Question Can common genetic variants be used to differentiate between treatment-resistant schizophrenia (TRS) and other forms of this disorder? Findings Data from this genome-wide association study including 85 490 participants were used to estimate genome-wide single-nucleotide variation effect size differences between individuals with and without TRS, which were compatible with a polygenic model of treatment resistance. Results were used to generate a polygenic risk score, which was significantly associated with TRS status in independent incidence and prevalence samples. Meaning Findings of this study based on common genetic variants indicate that TRS is heritable with a modest but significant single-nucleotide variation–based heritability.

level. Datasets were then merged into two batches by the similarity of their array content, and all nonoverlapping markers were discarded. Relatedness between individuals was assessed in a merged dataset consisting of 203,813 SNPs common between both batches, using the PC-Relate approach (Conomos et al., 2016). A random member of each related pair (φ>0.2) was selected for removal in further analyses, prioritising retaining TRS individuals.
Imputation was performed on each batch separately using the Minimac4 algorithm as provided by the Michigan Imputation Server  and HRC reference panel, which has similar accuracy for common variation as the 1000 Genomes Phase 3 panel used in CardiffCOGS (McCarthy et al., 2016).
After this process, imputed dosages were converted to best-guess genotype calls for use in polygenic scoring (Genotype probability > 90%; INFO > 0.8, MAF > 1%, HWE mid p-value > 10 -4 ). For their use in association testing, principal components were generated from the post-imputed data using markers in relative linkage equilibrium (r 2 <0.2) and the PC-AiR algorithm (Conomos et al., 2015).

Polygenic validation analyses
To avoid the known bias in PRS analysis when samples are present in both training and testing sets, we ensured that no samples from either CardiffCOGS or STRATA-G were included in any discovery GWAS analyses, and that any samples related to either of these cohorts (relatedness coefficient φ>0.2) were removed. All summary statistics were curated by retaining only non-ambiguous SNPs outside of long-range LD regions (Price et al., 2008) that had a MAF of 10% or higher and INFO ≥0.9.
In PRSice-2, p-value thresholds for the computed scores were set at 8 different intervals (p<10 -5 , p<10 -4 , p<0.001, p<0.05, p<0.01, p<0.1, p<0.5, p<1). Finally, a score was also generated via the PRS-CS software (Ge et al., 2019), which estimates PRS using all SNPs in a Bayesian framework. The UK-Biobank LD reference provided with PRS-CS was used for all computations. The global shrinkage parameter was set to φ=1x10 -4 for the TRS interaction and φ=1x10 -2 for the CLOZUK and PGC GWAS, reflecting the differential polygenicity of these analyses, as recommended by the software manual.
Association of the polygenic scores with the TR/non-TRS phenotype was calculated using logistic models with sex and the first 5 genotypic principal components (PCs) as covariates, in order to control for population stratification (Peloso and Lunetta, 2011). Given the presence of multiple ancestries in STRATA-G samples, we employed a previously described model to classify each sample into seven broadly defined biogeographical population groups using ancestry-informative markers (AIMs; Legge et al., 2019). We then used the underlying discriminant functions of this model to compute six © 2022 Pardiñas AF et al. JAMA Psychiatry.
variables reflecting the probability of each sample being classified as "European", "South West Asian", "East Asian", "Subsaharan African" or "North African"; and included these as covariates in all PRS analyses involving STRATA-G. To assess variance explained by the PRS metrics, Nagelkerke's R 2 values for each PRS logistic regression were calculated on the liability scale (Lee et al., 2012), assuming a 30% population prevalence for TRS in CardiffCOGS (Lally et al., 2016b) and 15% prevalence as a conservative estimate for STRATA-G (Kanahara et al., 2018;Siskind et al., 2021).
Finally, for the STRATA-G analyses, since this cohort was imputed in two independent batches, PRS association tests and their summary statistics were computed separately within each STRATA-G imputation batch (Supplementary  et al. (2017). All these computations were performed with functions from the R package "metafor" (Viechtbauer, 2010). An analogous meta-analysis was also carried out using the association summary statistics from CardiffCOGS and both STRATA-G batches (Supplementary Figure 3). Treated with or considered for clozapine during the follow up period. Never treated with or considered for clozapine.

PGC TRS rating system
EUGEI Istanbul (Turkey) Treated with or considered for clozapine during the follow up period. Never treated with or considered for clozapine.
EUGEI Paris (France) Treated with or considered for clozapine during the follow up period. Never treated with or considered for clozapine.

GAP (London, UK)
Treated with or considered for clozapine during the follow up period. Excluded those who were intolerant of antipsychotic medications or those who self-discontinued medication.
Never treated with or considered for clozapine.

NIFEPS & RGPI (Belfast, UK)
Treated with or considered for clozapine during the follow up period. Never treated with or considered for clozapine. PAFIP (Santander, Spain) Treated with or considered for clozapine during the follow up period. Never treated with or considered for clozapine.

TIPP (Switzerland)
Treated with or considered for clozapine during the follow up period. Excluded those with poor compliance and who met criteria for symptom severity.
Never treated with or considered for clozapine.
TOP (Oslo, Norway) Treated with or considered for clozapine during the follow up period. Never treated with or considered for clozapine. West London (London, UK) Treated with clozapine at the time of a follow up interview. Never treated with clozapine at the time of a follow up interview.