Skip to main content
NCBI home page
JAMA Network logoLink to JAMA Network
. 2022 Jan 12;79(3):1–11. doi: 10.1001/jamapsychiatry.2021.3799

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

Antonio F Pardiñas 1,✉, Sophie E Smart 1,2, Isabella R Willcocks 1, Peter A Holmans 1, Charlotte A Dennison 1, Amy J Lynham 1, Sophie E Legge 1, Bernhard T Baune 3,4,5, Tim B Bigdeli 6,7,8, Murray J Cairns 9,10,11, Aiden Corvin 12, Ayman H Fanous 6,7, Josef Frank 13, Brian Kelly 14, Andrew McQuillin 15, Ingrid Melle 16,17, Preben B Mortensen 18,19, Bryan J Mowry 20,21, Carlos N Pato 22,23,24, Sathish Periyasamy 20,21, Marcella Rietschel 13, Dan Rujescu 25,26, Carmen Simonsen 16,27, David St Clair 28, Paul Tooney 9,11, Jing Qin Wu 29, Ole A Andreassen 16,17, Kaarina Kowalec 30,31, Patrick F Sullivan 31,32,33, Robin M Murray 2, Michael J Owen 1, James H MacCabe 2, Michael C O’Donovan 1, James T R Walters 1,✉; and the Genetics Workstream of the Schizophrenia Treatment Resistance and Therapeutic Advances (STRATA) Consortium and the Schizophrenia Working Group of the Psychiatric Genomics Consortium (PGC), Olesya Ajnakina 34,35, Luis Alameda 36,37,38,39, Thomas R E Barnes 40, Domenico Berardi 41, Elena Bonora 42, Sara Camporesi 39,43, Martine Cleusix 43, Philippe Conus 39, Benedicto Crespo-Facorro 37,38, Giuseppe D’Andrea 41, Arsime Demjaha 36, Kim Q Do 43, Gillian A Doody 44, Chin B Eap 45,46,47,48, Aziz Ferchiou 49, Marta Di Forti 50,51, Lorenzo Guidi 42, Lina Homman 52,53, Raoul Jenni 43, Eileen M Joyce 54, Laura Kassoumeri 36, Inès Khadimallah 43, Ornella Lastrina 41, Roberto Muratori 42, Handan Noyan 55, Francis A O’Neill 53, Baptiste Pignon 49,56, Romeo Restellini 39,43, Jean-Romain Richard 49, Franck Schürhoff 49,56, Filip Španiel 57,58, Andrei Szöke 49,56, Ilaria Tarricone 42, Andrea Tortelli 49,59, Alp Üçok 60, Javier Vázquez-Bourgon 61,62,63
PMCID: PMC8756361  PMID: 35019943

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.

This genome-wide association study examines the genetic architecture of treatment-resistant schizophrenia using genetic data from schizophrenia studies and generates a polygenic risk score for TRS status.

Abstract

Importance

About 20% to 30% of people with schizophrenia have psychotic symptoms that do not respond adequately to first-line antipsychotic treatment. This clinical presentation, chronic and highly disabling, is known as treatment-resistant schizophrenia (TRS). The causes of treatment resistance and their relationships with causes underlying schizophrenia are largely unknown. Adequately powered genetic studies of TRS are scarce because of the difficulty in collecting data from well-characterized TRS cohorts.

Objective

To examine the genetic architecture of TRS through the reassessment of genetic data from schizophrenia studies and its validation in carefully ascertained clinical samples.

Design, Setting, and Participants

Two case-control genome-wide association studies (GWASs) of schizophrenia were performed in which the case samples were defined as individuals with TRS (n = 10 501) and individuals with non-TRS (n = 20 325). The differences in effect sizes for allelic associations were then determined between both studies, the reasoning being such differences reflect treatment resistance instead of schizophrenia. Genotype data were retrieved from the CLOZUK and Psychiatric Genomics Consortium (PGC) schizophrenia studies. The output was validated using polygenic risk score (PRS) profiling of 2 independent schizophrenia cohorts with TRS and non-TRS: a prevalence sample with 817 individuals (Cardiff Cognition in Schizophrenia [CardiffCOGS]) and an incidence sample with 563 individuals (Genetics Workstream of the Schizophrenia Treatment Resistance and Therapeutic Advances [STRATA-G]).

Main Outcomes and Measures

GWAS of treatment resistance in schizophrenia. The results of the GWAS were compared with complex polygenic traits through a genetic correlation approach and were used for PRS analysis on the independent validation cohorts using the same TRS definition.

Results

The study included a total of 85 490 participants (48 635 [56.9%] male) in its GWAS stage and 1380 participants (859 [62.2%] male) in its PRS validation stage. Treatment resistance in schizophrenia emerged as a polygenic trait with detectable heritability (1% to 4%), and several traits related to intelligence and cognition were found to be genetically correlated with it (genetic correlation, 0.41-0.69). PRS analysis in the CardiffCOGS prevalence sample showed a positive association between TRS and a history of taking clozapine (r2 = 2.03%; P = .001), which was replicated in the STRATA-G incidence sample (r2 = 1.09%; P = .04).

Conclusions and Relevance

In this GWAS, common genetic variants were differentially associated with TRS, and these associations may have been obscured through the amalgamation of large GWAS samples in previous studies of broadly defined schizophrenia. Findings of this study suggest the validity of meta-analytic approaches for studies on patient outcomes, including treatment resistance.

Introduction

Precision psychiatry provides a potential pathway to improve psychiatric classification and develop treatments that are better tailored to specific patients. Major advances in determining the role of genetic variation in the risk of developing psychiatric disorders are helping realize this potential,1 but the relevance of these findings to patient outcomes and response to treatment is unclear. Evidence from nonpsychiatric disorders points to distinct genetic bases for disorder susceptibility and prognosis,2,3 although few sufficiently powered psychiatric genetic studies have been conducted investigating disorder progression or outcome.4 Such study designs could offer valuable insights as to the feasibility of precision psychiatry approaches and their potential impact for patients who have poor outcomes because of the lack of effectiveness of current treatments.5

Some of the most disadvantaged individuals in this respect are those with a diagnosis of schizophrenia whose symptoms do not respond adequately to conventional antipsychotic medication.6,7,8 This clinical picture is known as treatment-resistant schizophrenia (TRS), which affects approximately 20% to 30% of people with this disorder. The only licensed treatment for TRS is clozapine, which is effective in approximately 60% of cases9 and improves most indicators of morbidity and mortality.10,11 While the mechanism of action of clozapine is still not fully understood,12 it has been proposed that its efficacy might be related to the biological underpinnings of treatment resistance, suggesting a distinct neurobiological etiology to that of non-TRS.13 Additionally, several studies have shown that delay in clozapine prescription is associated with resistance even to clozapine.14,15 This makes early identification of TRS critical and the ascertainment of correlates of TRS a priority for the field of schizophrenia research.

To date, there is considerable heterogeneity in the findings of genetic studies related to TRS. A family history of schizophrenia is likely associated with developing TRS,16 but a recent systematic review did not find any individual genes robustly and specifically associated with this condition.17 Researchers have also investigated the long-standing hypothesis that TRS is a more severe form of schizophrenia resulting from a large burden of schizophrenia risk alleles.18 While 2 studies reported that aggregating risk variants into a polygenic risk score (PRS) revealed small differences between TRS and non-TRS samples,19,20 other analyses did not replicate this result.16,21,22,23 However, because each study used slightly different recruitment criteria and definitions of TRS, heterogeneity in the results is expected,24 making their collective interpretation difficult. This limitation likely reflects the previously discussed problem of individuals with treatment-resistant psychiatric symptoms being disproportionately underrepresented in research studies owing to poor health, limited capacity to consent, and other causes of attrition, including therapeutic pessimism,25 which have precluded the execution of well-powered genome-wide association studies (GWASs).26

In this study, we aim to characterize the contribution of common genetic variants to treatment resistance in schizophrenia by exploiting data generated by large consortium-based GWASs of schizophrenia,27,28 in which individuals with TRS can be formally defined. Our main hypotheses are that schizophrenia risk alleles will show different genetic associations in the analysis of individuals with and without TRS and that these differences reflect the underlying genetics of treatment resistance. When assessed at a genome-wide level, such differences could reveal commonalities with other complex traits or be validated in clinical cohorts using the PRS approach,29 helping us to better understand the biological and epidemiological characteristics of treatment resistance in schizophrenia.17

Methods

For our main analysis, we used samples from large genomic studies of schizophrenia.27,28 Some of the largest cohorts in these research initiatives were recruited based on clozapine prescription (a proxy of TRS status), and forming a case-case data set from them would require avoiding confounding factors, such as GWAS batch effects30 or population stratification,31 which are difficult to control in a multiple-cohort design.32 As a safeguard against these, we have used a meta-analytic procedure to assess the differences between GWAS in which individuals with TRS and non-TRS have been compared with matched sets of healthy controls, before comparing the allelic association effect sizes of these 2 GWASs on a genome-wide basis to create a GWAS specific to treatment resistance.

Genetic Samples and Analysis

We used the CLOZUK1 and CLOZUK2 cohorts as our primary source of individuals with treatment-resistant symptoms, with a total sample size of 10 501 individuals with TRS and 24 542 controls. These cohorts have been described in previous studies.28,33 All the individuals with TRS in these samples were prescribed clozapine in the UK after failure of at least 2 trials of antipsychotics, following National Institute for Health and Care Excellence guidelines for TRS.34 The use of a history of taking clozapine as equivalent to a research diagnosis of TRS has been validated in these samples,28 as well as in independent studies.35,36 Control individuals were collected from public databases or through collaboration with population sequencing projects in the UK.

To identify individuals with non-TRS, we used 34 studies included in the meta-analysis by the Schizophrenia Working Group of the Psychiatric Genomics Consortium (PGC),27 with a total sample size of 20 325 individuals with schizophrenia and 30 122 controls. In 14 of these studies, available clinical records allowed us to identify and remove all individuals with TRS (eMethods in Supplement 1). The remaining 20 studies were not screened as comparable data were not available, and thus, we conservatively included these samples in the analysis as non–treatment-resistant cases (eTable 1 in Supplement 1). Healthy control individuals in this analysis were a mixture of publicly available samples and clinically ascertained (nonpsychiatric) individuals, and extensive analyses to discard population outliers and assess stratification were also carried out.27

GWASs of individuals with TRS vs controls and individuals with non-TRS vs controls were carried out separately using the logistic regression model implemented in PLINK version 2.37 Further details on the imputation, quality control, and association testing procedures are described in the eMethods in Supplement 1.

All studies used in the GWAS were reviewed and approved by their local ethical committee. Written informed consent (or legal guardian consent and participant assent) was obtained for all study participants except for those in the CLOZUK cohort, as this study used anonymized blood samples as approved by the UK Multicenter Research Ethics Committee.

Statistical Analysis

Comparing the TRS and Non-TRS Association Studies

To generate association statistics that reflect TRS vs non-TRS differences, we used the test for interaction proposed by Altman and Bland,38 which is analogous to a fixed-effect test for moderators in the meta-analytic setting.39 This test allows us to calculate an estimate of the difference between 2 odds ratios (ORs), which in our case were those of individuals with TRS vs controls (OR1) and those with non-TRS vs controls (OR2). The difference, in the scale of the regression β coefficient, equates to difference = log(OR1) − log(OR2), with standard error SE(d) = √(SE[OR1]2 + SE[OR2]2). We transformed this effect size into a z score and calculated its associated P value at each overlapping single-nucleotide variation (SNV; formerly single-nucleotide polymorphism) between the TRS and non-TRS GWAS, excluding 146 SNVs with a minor allele frequency difference greater than 20% between both data sets to mirror preimputation quality control. For consistency, we also used the same definition of OR1 and OR2 in all our tests to preserve directions of effect, and thus, positive z scores in the interaction analysis reflect stronger SNV associations in individuals with TRS compared with non-TRS.

Estimating Heritability

SNV-based heritability was estimated from the TRS interaction summary statistics using LD SCore version 1.01,40 after restricting to markers present in the HapMap3 study41 and a precomputed linkage disequilibrium reference panel based on the 1000 Genomes phase 3 samples. As point estimates of SNV-based heritability can differ based on model assumptions, we also estimated this quantity via the SumHer framework implemented in LDAK5,42 using a default parameter to define the relationship between minor allele frequency and effect size (α = −.25) and multiplicative inflation correction (genomic control). The LD reference for LDAK was estimated from the European samples of the public Haplotype Reference Consortium panel.43

Polygenic Validation

We sought to investigate whether a TRS GWAS PRS was associated with treatment resistance status in 2 independent samples: one from a schizophrenia prevalence cohort (Cardiff Cognition in Schizophrenia [CardiffCOGS]) and the other from a first-episode incidence cohort (Genetics Workstream of the Schizophrenia Treatment Resistance and Therapeutic Advances [STRATA-G]). The CardiffCOGS prevalence cohort is a sample of individuals with schizophrenia that included both individuals with TRS and treatment-responsive schizophrenia, defined here again from a history of taking or not taking clozapine (n = 817; 315 with TRS and 502 with non-TRS). This is a cross-sectional study with detailed clinical ratings based on research diagnostic interviews and contemporaneous records, as previously described.44 We also used a new multiancestry incidence cohort of people with first-episode psychosis, the STRATA-G consortium, in which the participants have been followed up for at least 1 year after initial presentation to ascertain diagnosis and treatment response (eMethods in Supplement 1). As both the first-episode and broad psychosis nature of this sample likely make it diagnostically heterogeneous, we first restricted our analyses to individuals with a diagnosis of schizophrenia at the last time of follow-up. TRS was rated as the presence or absence of a history of taking clozapine (n = 562; 71 with TRS and 492 with non-TRS) to best mirror our TRS GWAS and the definition used in the prevalence cohort. Further details about the recruitment and phenotyping of the STRATA-G sample as well as details of the genotyping and imputation procedure for both data sets are given in the eMethods and eTable 2 in Supplement 1.

The CardiffCOGS study had the relevant UK National Health Service ethical approval, and studies included in STRATA-G were reviewed and approved by their local ethical committees. Written informed consent was obtained for all participants.

PRS Analysis

Our primary PRS analysis used the TRS interaction GWAS summary statistics as training set, and the prevalence and incidence cohorts as testing sets where their treatment-resistance phenotype was tested for association. We also investigated the association of PRS and treatment resistance status based on (1) the CLOZUK TRS vs healthy control GWAS and (2) the PGC non-TRS GWAS. Polygenic scores were estimated using PRSice-2 version 2.3545 and PRS-CS version June 4, 2021,46 as detailed in the eMethods in Supplement 1. Within each pairing of training and testing set, PRS association P values were corrected for multiple testing using the Benjamini and Hochberg false discovery rate (FDR) method47 as implemented in the R framework stats package.

Estimating Genetic Correlation and Colocalization

Correlations between treatment resistance and other complex traits were computed using LD-Hub version 1.93,48 again based on the LD-Score regression40 framework. We only performed the analysis for GWAS categories that had significant phenome-wide associations with psychiatric disorders, based on previous research.49 The categories chosen were education, cognitive, personality, psychiatric, and smoking, for a total of 28 summary statistics. Multiple testing correction of these results was also carried out using FDR, since most of the summary statistics tested have overlapping samples and thus yield partially dependent P values.

For all associations, statistical significance was set at P < .05. All P values were 2-tailed.

Results

TRS/Non-TRS GWAS

The study included a total of 85 490 participants (48 635 [56.9%] male) in its GWAS stage and 1380 participants (859 [62.2%] male) in its PRS validation stage. Discovery GWAS of individuals with TRS and non-TRS against independent controls showed highly consistent genome-wide significant signals (eFigure 1 in Supplement 1), all reported in previous schizophrenia studies. This consistency extended to most of the polygenic architecture, as shown by a high genetic correlation estimate between the data sets (r = 0.966; SE, 0.037; P = 1.43 × 10−147) in which the 95% CI includes 1. No genome-wide SNVs were detected by the TRS interaction analysis (Figure 1), although a slight departure from normality was detected in the genome-wide Q-Q plot (λ = 1.062; λ1000 = 1.002; LD-Score intercept, 1.032; SE, 0.007; LDAK scaling estimate, 1.017; SE, 0.010; eFigure 2 in Supplement 1). This is compatible with a polygenic signal,50 as the 2 discovery GWASs did not show evidence of inflation caused by other confounding factors.27,28 Indeed, the 2 methods we used to estimate the common variant contribution to the observed-scale heritability of TRS returned small values of similar magnitude (SNV-based heritability: LD-Score, 0.013; SE, 0.006; SNV-based heritability: LDAK, 0.040; SE, 0.014).

Figure 1. Manhattan Plot of the Treatment-Resistant Schizophrenia (TRS) vs Non-TRS Interaction Genome-Wide Association Study.

Figure 1.

Open in a new tab

Polygenic Score–Based Prediction of TRS in Independent Samples

A PRS generated from the TRS interaction GWAS was positively associated with treatment resistance in our validation prevalence cohort (CardiffCOGS), explaining up to 2.03% of the variance on the liability scale of treatment resistance (P = .001; FDR P = .01; Figure 2; eTable 3 in Supplement 1). Analyzing the 2 GWASs used in the interaction analysis, we found that a PRS based on CLOZUK GWAS results (TRS vs healthy controls) also explained a significant proportion of the variance in the TRS phenotype (maximum r2 = 1.63%; P = .004; FDR P = .03), very similar to that obtained by testing the larger PGC non-TRS sample (maximum r2 = 1.21%; P = .01; FDR P = .10). However, the effect sizes of these PRSs were generally in opposite directions (Figure 2; eTable 3 in Supplement 1); ie, those taking clozapine in CardiffCOGS had higher PRS derived from the TRS GWAS (OR, 1.22; 95% CI, 1.05-1.41) but a lower PRS from the non-TRS GWAS (OR, 0.83; 95% CI, 0.72-0.96). Focusing on the polygenic prediction via Bayesian regression and continuous shrinkage priors estimates, which do not require LD clumping or P value thresholding,51 we also observed a significant and positive difference between the effect sizes of the TRS interaction and CLOZUK PRS, while the PGC PRS had a null effect (Figure 2). This pattern of results in an independent case-case comparison is consistent with our interaction analysis detecting a true polygenic signal for treatment resistance, with a greater burden of risk alleles from both this and the CLOZUK TRS GWAS associating with a history of clozapine prescription. Alternatively, the PRS derived from non-TRS samples is not as clearly associated with treatment resistance, and some results of these analyses are even compatible with its enrichment in individuals with non-TRS.

Figure 2. Polygenic Score Analysis of the Cardiff Cognition in Schizophrenia (CardiffCOGS) Cohort and Genetics Workstream of the Schizophrenia Treatment Resistance and Therapeutic Advances (STRATA-G) Cohort Using 3 Different Schizophrenia-Related Polygenic Risk Scores (PRS).

Figure 2.

Open in a new tab

Associations between PRS and treatment resistance in schizophrenia (TRS) were defined as a history of taking clozapine in people with a schizophrenia diagnosis. There were 315 individuals with TRS and 502 individuals with non-TRS in the CardiffCOGS cohort and 71 individuals with TRS and 492 control individuals in the STRATA-G cohort. PGC indicates Psychiatric Genomics Consortium; PRS-CS, polygenic prediction via Bayesian regression and continuous shrinkage priors. Whiskers indicate 95% CIs.

aP < .05.

Generation of PRS and their analysis within the first-episode incidence sample (STRATA-G) showed that a greater interaction PRS burden was also associated with having treatment resistance in this setting, explaining up to 1.09% of the variance (P = .04; FDR P = .21). In contrast, neither the CLOZUK nor PGC PRS were significantly associated with treatment resistance across any of the tested P value thresholds or using polygenic prediction via Bayesian regression and continuous shrinkage priors (Figure 2; eTable 3 in Supplement 1). However, a meta-analysis of both CardiffCOGS and STRATA-G resulted in narrower confidence intervals around all the PRS effect size estimates previously estimated from CardiffCOGS, supporting the overall consistency in directions of association between these samples (eFigure 3 and eTable 3 in Supplement 1).

To interpret the results from the prevalence and incidence cohorts alongside each other, we merged the genotyped SNVs from both samples (to a total of 199 074 SNVs) and estimated polygenic scores using all LD-independent SNVs (P < 1). As a reference point for this analysis, we retained a small subset of 242 screened unaffected control individuals that had been genotyped as part of the Genetics and Psychosis project,26 one of the STRATA-G cohorts. Splitting the combined sample by TRS status showed that the polygenic profile of treatment resistance is largely consistent between CardiffCOGS and STRATA-G, with the TRS interaction PRS burden showing no difference between incident and prevalent TRS individuals (Figure 3). However, we noted a distinction in terms of general schizophrenia risk alleles, as indexed by the CLOZUK and PGC PRS, which were enriched within the individuals with first-episode TRS compared with the individuals with TRS from our cross-sectional sample (CLOZUK: OR, 1.52; SE, 0.21; P = .047; PGC: OR, 3.48; SE, 0.39; P = 7.69 × 10−4). None of the scores we tested were significantly different between the 2 groups of individuals with non-TRS, and all individuals with schizophrenia showed larger mean scores than the unaffected controls.

Figure 3. Polygenic Profile Derived From Combined Genotype Data From the Cardiff Cognition in Schizophrenia (CardiffCOGS) and Genetics Workstream of the Schizophrenia Treatment Resistance and Therapeutic Advances (STRATA-G) Cohorts Using Scores Calculated With Linkage Disequilibrium–Independent (P < 1) Single-Nucleotide Variants.

Figure 3.

Open in a new tab

Dots indicate the mean values of the standardized polygenic risk score (PRS; corrected for sex, relevant principal components, and ancestry-informative marker–based ancestry estimates). Whiskers delimit the 95% CIs of the PRS means. PGC indicates Psychiatric Genomics Consortium; TRS, treatment-resistant schizophrenia.

Genetic Correlation Between TRS and Other Traits

We used the results of the TRS interaction GWAS to examine the genetic associations between the treatment-resistant phenotype and other disorders and traits. In these analyses, 8 of 28 publicly available GWAS summary statistics displayed nominally significant (P < .05) genetic correlations with TRS, of which 5 survived multiple testing correction (FDER p < 0.05; Figure 4; eTable 4 in Supplement 1). All 5 genetically correlated phenotypes were associated with cognitive measures and educational attainment (genetic r, 0.41-0.69), showing our GWAS of TRS to be genetically correlated with lower cognitive ability and lower educational attainment.

Figure 4. Linkage Disequilibrium Score Regression Genetic Correlation Results of LD-HUB Phenotypes With the Treatment-Resistant Schizophrenia Interaction Genome-Wide Association Study.

Figure 4.

Open in a new tab

Top, Genetic correlation coefficients shown with standard errors. Bottom, Genetic correlation P values. The dashed line indicates a false discovery rate–corrected significant level of P = .05.

Discussion

We report a large study of treatment resistance in schizophrenia, showing that common genetic variants were associated with this presentation, that these variants were also associated with general risk of schizophrenia, and that the difference in their associations with the 2 phenotypes may have influenced the polygenic score analyses. Furthermore, our results, particularly the large genetic correlation between the TRS and non-TRS GWASs, are consistent with the risk conferred by common SNVs being similar regardless of whether individuals respond to first-line treatments. This suggests that treatment resistance, at least to the extent that we have been able to define it and explore it within this study, is not likely to index a cluster of individuals with a range of etiologies and pathophysiologies fundamentally different from those that contribute to schizophrenia more widely. However, by also assessing differences at the level of individual allelic associations, our results also support the existence of a polygenic contribution associated with treatment resistance that seems largely distinct from liability to schizophrenia. While this approach has not identified specific SNVs or genes that could be followed up as potential drivers of treatment outcomes, we find genome-wide correlations that recapitulate previous epidemiological findings, such as the association of treatment resistance with poorer cognitive performance.52,53 Such polygenic overlaps validate our indirect approach for carrying out the GWAS of TRS in the absence of a single large-scale harmonized case-case sample. However, they should not be interpreted as proof of causality in either direction and do not account for confounders, such as sociodemographic indicators, medication adherence, or antipsychotic effects in treatment response.

Implications

In revealing the existence of a common, heritable genetic signal for TRS, our study adds a new layer of evidence to the ongoing discussion of whether TRS is categorically distinct from treatment-responsive schizophrenia.17 Despite a large genetic correlation between both conditions, we show that PRS derived from them perform differently when attempting out-of-sample prediction of TRS, which suggests that the heterogeneous results previously obtained in PRS analyses might have been influenced by the presence of individuals with TRS in the generic schizophrenia training sample.19 While follow-up studies would likely require rich clinical data on large numbers of individuals with TRS and non-TRS to fully explore the implications of these results, they pose an important consideration for research seeking to understand whether the genetics of disorder susceptibility are the same as disease course in schizophrenia2 and whether treatment response can be influenced by the accumulation of genetic and clinical factors.54,55 Finally, while our results show that TRS is associated with a polygenic signal, the variance in TRS explained remains modest, and associated area under the curve values are small (eTable 3 in Supplement 1). Thus, polygenic scores tapping into this signal are thus unlikely to be of clinical utility in predicting treatment resistance, although their contribution to multifactorial predictive models (integrating rare variants, neuroimaging biomarkers, environmental exposures, demographic factors, and clinical measures) is an interesting avenue for future research.

Limitations

This study has limitations. TRS is an underreported diagnosis, and while our research definition for this phenotype aligns with international criteria,56 we acknowledge that some individuals with treatment-resistant symptoms might still be present in the non-TRS data set, particularly in those samples where no ascertainment of TRS could be carried out. The effect of such misclassification would be akin to a reduction of effective sample size, reducing SNV discovery power and adding noise to GWAS effect size estimates.57 This further limits our power to detect differences in allelic associations through the interaction analysis, which is already only sensitive to relatively large differences in effect size between studies.58 Additionally, the interaction test does not model or addresses potential between-study heterogeneity. This precludes the use of more sophisticated statistical methods for downstream analyses, such as tissue-specific enrichment analyses or transcriptome-wide association studies.59

Imperfect phenotyping and potential misclassification might also have attenuated our results, and given the small total heritability detected for our TRS GWAS, we note that the magnitude of genetic correlations with other traits should be interpreted cautiously. Also, similar to what has been argued in the study of etiologically heterogeneous phenotypes through GWAS,50,60 the existence of significant genetic correlations and consistent polygenic association in independent samples reassures us that the common genetic variants we detect have (en masse) an association with TRS liability, even if our results cannot quantify the degree of contribution from nongenetic causes. In this regard, given that most of our analyses are based on European-based and UK-based samples, our conclusions might not be generalizable to non-European countries, where the diagnosis and treatment pathways of TRS might be influenced by racial and ethnic or cultural backgrounds.61

Conclusions

In this GWAS, TRS had a small but detectable heritability associated with common risk alleles. Validation work, via the PRS method, showed that the contribution of these alleles were similar across incidence and prevalence samples. This is despite differences in cohort characteristics, and the use of drug prescription data as a proxy for TRS instead of quantitative metrics. Altogether, these results highlight the usefulness of well-controlled clinical phenotype data in psychiatric genetics to explore beyond diagnostic classifications and into treatment outcome and response to aid precision psychiatry.

Supplement 1.

eMethods.

eFigure 1. Mirrored Manhattan plot of the 2 GWAS analysed with the TRS interaction procedure.

eFigure 2. Q-Q plot of the TRS interaction GWAS.

eFigure 3. PRS meta-analysis of CardiffCOGS and STRATA-G cohorts.

eTable 1. Data sets included in the PGC non-TRS GWAS sample.

eTable 2. Data sets included in the STRATA-G sample.

eTable 3. Polygenic risk score analysis results.

eTable 4. LD-Score and LD-Hub analyses of the TRS GWAS summary statistics.

eReferences.

Click here for additional data file. (1MB, pdf)
Supplement 2.

Nonauthor collaborators. Collaborators of the Genetics Workstream of the Schizophrenia Treatment Resistance and Therapeutic Advances (STRATA) Consortium and the Schizophrenia Working Group of the Psychiatric Genomics Consortium (PGC).

Click here for additional data file. (147.5KB, pdf)

References

  • 1.Sullivan PF, Agrawal A, Bulik CM, et al. ; Psychiatric Genomics Consortium . Psychiatric genomics: an update and an agenda. Am J Psychiatry. 2018;175(1):15-27. doi: 10.1176/appi.ajp.2017.17030283 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 2.Paternoster L, Tilling K, Davey Smith G. Genetic epidemiology and mendelian randomization for informing disease therapeutics: conceptual and methodological challenges. PLoS Genet. 2017;13(10):e1006944. doi: 10.1371/journal.pgen.1006944 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 3.Lee JC, Biasci D, Roberts R, et al. ; UK IBD Genetics Consortium . Genome-wide association study identifies distinct genetic contributions to prognosis and susceptibility in Crohn’s disease. Nat Genet. 2017;49(2):262-268. doi: 10.1038/ng.3755 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 4.Rees E, Owen MJ. Translating insights from neuropsychiatric genetics and genomics for precision psychiatry. Genome Med. 2020;12(1):43. doi: 10.1186/s13073-020-00734-5 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 5.Birnbaum R, Weinberger DR. Special article: translational science update. pharmacological implications of emerging schizophrenia genetics: can the bridge from ‘genomics’ to ‘therapeutics’ be defined and traversed? J Clin Psychopharmacol. 2020;40(4):323-329. doi: 10.1097/JCP.0000000000001215 [DOI] [PubMed] [Google Scholar]
  • 6.Lally J, Gaughran F, Timms P, Curran SR. Treatment-resistant schizophrenia: current insights on the pharmacogenomics of antipsychotics. Pharmgenomics Pers Med. 2016;9:117-129. doi: 10.2147/PGPM.S115741 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 7.Mørup MF, Kymes SM, Oudin Åström D. A modelling approach to estimate the prevalence of treatment-resistant schizophrenia in the United States. PLoS One. 2020;15(6):e0234121. doi: 10.1371/journal.pone.0234121 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 8.Siskind D, Orr S, Sinha S, et al. Rates of treatment-resistant schizophrenia from first-episode cohorts: systematic review and meta-analysis. Br J Psychiatry. 2021:1-6. doi: 10.1192/bjp.2021.61 [DOI] [PubMed] [Google Scholar]
  • 9.Kumra S, Kranzler H, Gerbino-Rosen G, et al. Clozapine and “high-dose” olanzapine in refractory early-onset schizophrenia: a 12-week randomized and double-blind comparison. Biol Psychiatry. 2008;63(5):524-529. doi: 10.1016/j.biopsych.2007.04.043 [DOI] [PubMed] [Google Scholar]
  • 10.Conley RR, Kelly DL. Management of treatment resistance in schizophrenia. Biol Psychiatry. 2001;50(11):898-911. doi: 10.1016/S0006-3223(01)01271-9 [DOI] [PubMed] [Google Scholar]
  • 11.Tiihonen J, Mittendorfer-Rutz E, Majak M, et al. Real-world effectiveness of antipsychotic treatments in a nationwide cohort of 29 823 patients with schizophrenia. JAMA Psychiatry. 2017;74(7):686-693. doi: 10.1001/jamapsychiatry.2017.1322 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 12.Pardiñas AF, Owen MJ, Walters JTR. Pharmacogenomics: a road ahead for precision medicine in psychiatry. Neuron. 2021;S0896-6273(21)00683-8. [DOI] [PubMed] [Google Scholar]
  • 13.Krivoy A, Hochman E, Sendt K-V, et al. Association between serum levels of glutamate and neurotrophic factors and response to clozapine treatment. Schizophr Res. 2018;192:226-231. doi: 10.1016/j.schres.2017.05.040 [DOI] [PubMed] [Google Scholar]
  • 14.Shah P, Iwata Y, Brown EE, et al. Clozapine response trajectories and predictors of non-response in treatment-resistant schizophrenia: a chart review study. Eur Arch Psychiatry Clin Neurosci. 2020;270(1):11-22. doi: 10.1007/s00406-019-01053-6 [DOI] [PubMed] [Google Scholar]
  • 15.Shah P, Iwata Y, Plitman E, et al. The impact of delay in clozapine initiation on treatment outcomes in patients with treatment-resistant schizophrenia: a systematic review. Psychiatry Res. 2018;268:114-122. doi: 10.1016/j.psychres.2018.06.070 [DOI] [PubMed] [Google Scholar]
  • 16.Kowalec K, Lu Y, Sariaslan A, et al. Increased schizophrenia family history burden and reduced premorbid IQ in treatment-resistant schizophrenia: a Swedish National Register and Genomic Study. Mol Psychiatry. 2021;26(8):4487-4495. doi: 10.1038/s41380-019-0575-1 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 17.Gillespie AL, Samanaite R, Mill J, Egerton A, MacCabe JH. Is treatment-resistant schizophrenia categorically distinct from treatment-responsive schizophrenia? a systematic review. BMC Psychiatry. 2017;17(1):12. doi: 10.1186/s12888-016-1177-y [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 18.Sobell JL, Mikesell MJ, McMurray CT. Genetics and etiopathophysiology of schizophrenia. Mayo Clin Proc. 2002;77(10):1068-1082. doi: 10.4065/77.10.1068 [DOI] [PubMed] [Google Scholar]
  • 19.Frank J, Lang M, Witt SH, et al. Identification of increased genetic risk scores for schizophrenia in treatment-resistant patients. Mol Psychiatry. 2015;20(2):150-151. doi: 10.1038/mp.2014.56 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 20.Gasse C, Wimberley T, Wang Y, et al. Schizophrenia polygenic risk scores, urbanicity and treatment-resistant schizophrenia. Schizophr Res. 2019;212:79-85. doi: 10.1016/j.schres.2019.08.008 [DOI] [PubMed] [Google Scholar]
  • 21.Martin AK, Mowry B. Increased rare duplication burden genomewide in patients with treatment-resistant schizophrenia. Psychol Med. 2016;46(3):469-476. doi: 10.1017/S0033291715001701 [DOI] [PubMed] [Google Scholar]
  • 22.Legge SE, Dennison CA, Pardiñas AF, et al. Clinical indicators of treatment-resistant psychosis. Br J Psychiatry. 2020;216(5):259-266. doi: 10.1192/bjp.2019.120 [DOI] [PubMed] [Google Scholar]
  • 23.Wimberley T, Gasse C, Meier SM, Agerbo E, MacCabe JH, Horsdal HT. Polygenic risk score for schizophrenia and treatment-resistant schizophrenia. Schizophr Bull. 2017;43(5):1064-1069. doi: 10.1093/schbul/sbx007 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 24.Hodgson K, McGuffin P, Lewis CM. Advancing psychiatric genetics through dissecting heterogeneity. Hum Mol Genet. 2017;26(R2):R160-R165. doi: 10.1093/hmg/ddx241 [DOI] [PubMed] [Google Scholar]
  • 25.Hofer A, Radner V, Edlinger M, Kemmler G, Rettenbacher MA, Fleischhacker WW. Why do individuals with schizophrenia drop out of observational clinical trials? Psychiatry Res. 2017;256:1-5. doi: 10.1016/j.psychres.2017.06.010 [DOI] [PubMed] [Google Scholar]
  • 26.Lally J, Ajnakina O, Di Forti M, et al. Two distinct patterns of treatment resistance: clinical predictors of treatment resistance in first-episode schizophrenia spectrum psychoses. Psychol Med. 2016;46(15):3231-3240. doi: 10.1017/S0033291716002014 [DOI] [PubMed] [Google Scholar]
  • 27.Schizophrenia Working Group of the Psychiatric Genomics Consortium . Biological insights from 108 schizophrenia-associated genetic loci. Nature. 2014;511(7510):421-427. doi: 10.1038/nature13595 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 28.Pardiñas AF, Holmans P, Pocklington AJ, et al. ; GERAD1 Consortium; CRESTAR Consortium . Common schizophrenia alleles are enriched in mutation-intolerant genes and in regions under strong background selection. Nat Genet. 2018;50(3):381-389. doi: 10.1038/s41588-018-0059-2 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 29.Wray NR, Lee SH, Mehta D, Vinkhuyzen AAE, Dudbridge F, Middeldorp CM. Research review: polygenic methods and their application to psychiatric traits. J Child Psychol Psychiatry. 2014;55(10):1068-1087. doi: 10.1111/jcpp.12295 [DOI] [PubMed] [Google Scholar]
  • 30.Sinnott JA, Kraft P. Artifact due to differential error when cases and controls are imputed from different platforms. Hum Genet. 2012;131(1):111-119. doi: 10.1007/s00439-011-1054-1 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 31.Peterson RE, Edwards AC, Bacanu S-A, Dick DM, Kendler KS, Webb BT. The utility of empirically assigning ancestry groups in cross-population genetic studies of addiction. Am J Addict. 2017;26(5):494-501. doi: 10.1111/ajad.12586 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 32.Yuan J, Xing H, Lamy AL, Lencz T, Pe’er I; Schizophrenia Working Group of the Psychiatric Genomics Consortium . Leveraging correlations between variants in polygenic risk scores to detect heterogeneity in GWAS cohorts. PLoS Genet. 2020;16(9):e1009015. doi: 10.1371/journal.pgen.1009015 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 33.Hamshere ML, Walters JTR, Smith R, et al. ; Schizophrenia Psychiatric Genome-wide Association Study Consortium; Wellcome Trust Case Control Consortium+; Wellcome Trust Case Control Consortium 2 . Genome-wide significant associations in schizophrenia to ITIH3/4, CACNA1C and SDCCAG8, and extensive replication of associations reported by the Schizophrenia PGC. Mol Psychiatry. 2013;18(6):708-712. doi: 10.1038/mp.2012.67 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 34.Mortimer AM, Singh P, Shepherd CJ, Puthiryackal J. Clozapine for treatment-resistant schizophrenia: National Institute of Clinical Excellence (NICE) guidance in the real world. Clin Schizophr Relat Psychoses. 2010;4(1):49-55. doi: 10.3371/CSRP.4.1.4 [DOI] [PubMed] [Google Scholar]
  • 35.Suzuki T, Remington G, Mulsant BH, et al. Defining treatment-resistant schizophrenia and response to antipsychotics: a review and recommendation. Psychiatry Res. 2012;197(1-2):1-6. doi: 10.1016/j.psychres.2012.02.013 [DOI] [PubMed] [Google Scholar]
  • 36.Üçok A, Çıkrıkçılı U, Ergül C, et al. Correlates of clozapine use after a first episode of schizophrenia: results from a long-term prospective study. CNS Drugs. 2016;30(10):997-1006. doi: 10.1007/s40263-016-0358-z [DOI] [PubMed] [Google Scholar]
  • 37.Chang CC, Chow CC, Tellier LC, Vattikuti S, Purcell SM, Lee JJ. Second-generation PLINK: rising to the challenge of larger and richer datasets. Gigascience. 2015;4(7):7. doi: 10.1186/s13742-015-0047-8 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 38.Altman DG, Bland JM. Interaction revisited: the difference between two estimates. BMJ. 2003;326(7382):219. doi: 10.1136/bmj.326.7382.219 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 39.Kontopantelis E, Sperrin M, Mamas MA, Buchan IE. Investigating heterogeneity of effects and associations using interaction terms. J Clin Epidemiol. 2018;93:79-83. doi: 10.1016/j.jclinepi.2017.09.012 [DOI] [PubMed] [Google Scholar]
  • 40.Bulik-Sullivan BK, Loh P-R, Finucane HK, et al. ; Schizophrenia Working Group of the Psychiatric Genomics Consortium . LD score regression distinguishes confounding from polygenicity in genome-wide association studies. Nat Genet. 2015;47(3):291-295. doi: 10.1038/ng.3211 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 41.Altshuler DM, Gibbs RA, Peltonen L, et al. ; International HapMap 3 Consortium . Integrating common and rare genetic variation in diverse human populations. Nature. 2010;467(7311):52-58. doi: 10.1038/nature09298 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 42.Speed D, Balding DJ. SumHer better estimates the SNP heritability of complex traits from summary statistics. Nat Genet. 2019;51(2):277-284. doi: 10.1038/s41588-018-0279-5 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 43.McCarthy S, Das S, Kretzschmar W, et al. ; Haplotype Reference Consortium . A reference panel of 64,976 haplotypes for genotype imputation. Nat Genet. 2016;48(10):1279-1283. doi: 10.1038/ng.3643 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 44.Rees E, Kirov G, Sanders A, et al. ; Wellcome Trust Case Control Consortium . Evidence that duplications of 22q11.2 protect against schizophrenia. Mol Psychiatry. 2014;19(1):37-40. doi: 10.1038/mp.2013.156 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 45.Choi SW, O’Reilly PF. PRSice-2: polygenic risk score software for biobank-scale data. Gigascience. 2019;8(7):giz082. doi: 10.1093/gigascience/giz082 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 46.Ge T, Chen C-Y, Ni Y, Feng YA, Smoller JW. Polygenic prediction via Bayesian regression and continuous shrinkage priors. Nat Commun. 2019;10(1):1776. doi: 10.1038/s41467-019-09718-5 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 47.Benjamini Y, Hochberg Y. Controlling the false discovery rate: a practical and powerful approach to multiple testing. J R Stat Soc Series B Stat Methodol. 1995;57(1):289-300. doi: 10.1111/j.2517-6161.1995.tb02031.x [DOI] [Google Scholar]
  • 48.Zheng J, Erzurumluoglu AM, Elsworth BL, et al. ; Early Genetics and Lifecourse Epidemiology (EAGLE) Eczema Consortium . LD Hub: a centralized database and web interface to perform LD score regression that maximizes the potential of summary level GWAS data for SNP heritability and genetic correlation analysis. Bioinformatics. 2017;33(2):272-279. doi: 10.1093/bioinformatics/btw613 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 49.Krapohl E, Euesden J, Zabaneh D, et al. Phenome-wide analysis of genome-wide polygenic scores. Mol Psychiatry. 2016;21(9):1188-1193. doi: 10.1038/mp.2015.126 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 50.Niemi MEK, Martin HC, Rice DL, et al. Common genetic variants contribute to risk of rare severe neurodevelopmental disorders. Nature. 2018;562(7726):268-271. doi: 10.1038/s41586-018-0566-4 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 51.Pain O, Glanville KP, Hagenaars SP, et al. Evaluation of polygenic prediction methodology within a reference-standardized framework. PLoS Genet. 2021;17(5):e1009021. doi: 10.1371/journal.pgen.1009021 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 52.de Bartolomeis A, Balletta R, Giordano S, Buonaguro EF, Latte G, Iasevoli F. Differential cognitive performances between schizophrenic responders and non-responders to antipsychotics: correlation with course of the illness, psychopathology, attitude to the treatment and antipsychotics doses. Psychiatry Res. 2013;210(2):387-395. doi: 10.1016/j.psychres.2013.06.042 [DOI] [PubMed] [Google Scholar]
  • 53.Frydecka D, Beszłej JA, Gościmski P, Kiejna A, Misiak B. Profiling cognitive impairment in treatment-resistant schizophrenia patients. Psychiatry Res. 2016;235:133-138. doi: 10.1016/j.psychres.2015.11.028 [DOI] [PubMed] [Google Scholar]
  • 54.Zhang JP, Robinson D, Yu J, et al. Schizophrenia polygenic risk score as a predictor of antipsychotic efficacy in first-episode psychosis. Am J Psychiatry. 2019;176(1):21-28. doi: 10.1176/appi.ajp.2018.17121363 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 55.Dwyer DB, Kalman JL, Budde M, et al. An investigation of psychosis subgroups with prognostic validation and exploration of genetic underpinnings: the Psycourse study. JAMA Psychiatry. 2020;77(5):523-533. doi: 10.1001/jamapsychiatry.2019.4910 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 56.Howes OD, McCutcheon R, Agid O, et al. Treatment-resistant schizophrenia: Treatment Response and Resistance in Psychosis (TRRIP) Working Group consensus guidelines on diagnosis and terminology. Am J Psychiatry. 2017;174(3):216-229. doi: 10.1176/appi.ajp.2016.16050503 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 57.Wray NR, Lee SH, Kendler KS. Impact of diagnostic misclassification on estimation of genetic correlations using genome-wide genotypes. Eur J Hum Genet. 2012;20(6):668-674. doi: 10.1038/ejhg.2011.257 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 58.Stone-Romero EF, Anderson LE. Relative power of moderated multiple regression and the comparison of subgroup correlation coefficients for detecting moderating effects. J Appl Psychol. 1994;79(3):354-359. doi: 10.1037/0021-9010.79.3.354 [DOI] [Google Scholar]
  • 59.Gusev A, Ko A, Shi H, et al. Integrative approaches for large-scale transcriptome-wide association studies. Nat Genet. 2016;48(3):245-252. doi: 10.1038/ng.3506 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 60.Zilhão NR, Olthof MC, Smit DJA, et al. Heritability of tic disorders: a twin-family study. Psychol Med. 2017;47(6):1085-1096. doi: 10.1017/S0033291716002981 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 61.Wheeler AJ. Treatment pathway and patterns of clozapine prescribing for schizophrenia in New Zealand. Ann Pharmacother. 2008;42(6):852-860. doi: 10.1345/aph.1K662 [DOI] [PubMed] [Google Scholar]

Associated Data

This section collects any data citations, data availability statements, or supplementary materials included in this article.

Supplementary Materials

Supplement 1.

eMethods.

eFigure 1. Mirrored Manhattan plot of the 2 GWAS analysed with the TRS interaction procedure.

eFigure 2. Q-Q plot of the TRS interaction GWAS.

eFigure 3. PRS meta-analysis of CardiffCOGS and STRATA-G cohorts.

eTable 1. Data sets included in the PGC non-TRS GWAS sample.

eTable 2. Data sets included in the STRATA-G sample.

eTable 3. Polygenic risk score analysis results.

eTable 4. LD-Score and LD-Hub analyses of the TRS GWAS summary statistics.

eReferences.

Click here for additional data file. (1MB, pdf)
Supplement 2.

Nonauthor collaborators. Collaborators of the Genetics Workstream of the Schizophrenia Treatment Resistance and Therapeutic Advances (STRATA) Consortium and the Schizophrenia Working Group of the Psychiatric Genomics Consortium (PGC).

Click here for additional data file. (147.5KB, pdf)

Articles from JAMA Psychiatry are provided here courtesy of American Medical Association

RESOURCES