Abstract
Objective We examined the association between schizophrenia genetic susceptibility loci and neuropsychiatric systemic lupus erythematosus (NPSLE) features in childhood-onset SLE (cSLE) participants.
Methods Study participants from the Lupus Clinic at the Hospital for Sick Children, Toronto, met ≥ 4 of the American College of Rheumatology and/or SLE International Collaborating Clinics SLE classification criteria and were genotyped using the Illumina Multi-Ethnic Global Array or the Global Screening Array. Ungenotyped single-nucleotide polymorphisms (SNPs) were imputed, and ancestry was genetically inferred. We calculated 2 additive schizophrenia-weighted polygenic risk scores (PRS) using (1) genome-wide significant SNPs (P < 5 × 10–8), and (2) an expanded list of SNPs with significance at P < 0.05. We defined 2 outcomes compared to absence of NPSLE features: (1) any NPSLE feature, and (2) subtypes of NPSLE features (psychosis and nonpsychosis NPSLE). We completed logistic and multinomial regressions, first adjusted for inferred ancestry only and then added for variables significantly associated with NPSLE in our cohort (P < 0.05).
Results We included 513 participants with cSLE. Median age at diagnosis was 13.8 years (IQR 11.2–15.6), 83% were female, and 31% were of European ancestry. An increasing schizophrenia genome-wide association PRS was not associated with NPSLE (OR 1.04, 95% CI 0.87–1.26, P = 0.62), nor with the NPSLE subtypes, psychosis (OR 0.97, 95% CI 0.73–1.29, P = 0.84) and other nonpsychosis NPSLE (OR 1.08, 95% CI 0.88–1.34, P = 0.45), in ancestry-adjusted models. Results were similar for the model including covariates (ancestry, malar rash, oral/nasal ulcers, arthritis, lymphopenia, Coombs-positive hemolytic anemia, lupus anticoagulant, and anticardiolipin antibodies) and for the expanded PRS estimates.
Conclusion We did not observe an association between known risk loci for schizophrenia and NPSLE in a multiethnic cSLE cohort. This work warrants further validation.
Systemic lupus erythematosus (SLE) is a chronic autoimmune disease with varying clinical presentations that can affect virtually any organ system. Genetics plays an important role in SLE susceptibility, with heritability estimates of up to 66% and over 100 susceptibility single-nucleotide polymorphisms (SNPs).1 Genetics also contributes to the risk of developing certain SLE clinical features, including neuropsychiatric SLE (NPSLE).2
NPSLE features are heterogeneous; can involve the peripheral or central nervous system; and include psychosis, other nonpsychosis features (e.g., acute confusional state, cerebrovascular disease, seizure), and less SLE-specific symptoms such as headaches and mood or anxiety disorders.3 The prevalence of NPSLE in childhood-onset SLE (cSLE) is estimated to be up to 95% depending on the NPSLE case definition used.4 While there is no gold standard for NPSLE diagnosis, the American College of Rheumatology (ACR) published case definitions for NPSLE in 1999.3
Morbidity and mortality vary across cSLE cohorts. Some studies of cSLE participants show that when compared to cSLE without NPSLE, NPSLE is associated with disease-associated damage. Damage most frequently included cataracts as well as the musculoskeletal, neuropsychiatric, and renal systems.5,6,7
Schizophrenia is a severe, chronic psychiatric disorder with heterogeneous manifestations including hallucinations, delusions, cognitive impairment, and disorganized or abnormal motor behavior. It is strongly influenced by genetics, with heritability estimates of nearly 80% and over 100 risk loci identified by genome-wide association studies (GWAS).8 These prior GWAS demonstrate that schizophrenia is a polygenic disease, with many risk loci also linked with other disorders such as autism spectrum disorder, bipolar disorder, and major depressive disorder.8,9
Large-scale studies have also demonstrated significant comorbid and genetic associations between schizophrenia and several autoimmune diseases, including SLE.10,11 Schizophrenia GWAS have identified risk loci in the major histocompatibility complex (MHC), which encodes numerous immune-related genes involved in antigen presentation and inflammation. Other schizophrenia genetic risk loci associated with immunity outside of the MHC region have also been found, such as those encoding for CD19 and CD20 B lymphocytes. These suggest that inflammation and immune dysregulation may be important disease mechanisms for schizophrenia.8 No study to date has examined the relationship between genetic risk loci for schizophrenia and NPSLE risk. The purpose of this study was to examine the association between polygenic risk scores (PRS) for schizophrenia and both NPSLE and specific NPSLE features, in a multiethnic cohort of participants with cSLE. This approach enabled testing for pleiotropy (1 gene influencing ≥ 2 unrelated phenotypes) with schizophrenia and NPSLE, as well a link between schizophrenia genetic risk and NPSLE in children and adolescents with SLE.
METHODS
Study population. The study cohort included participants diagnosed with and followed for cSLE between 1983 and 2018 in the Lupus Clinic at the Hospital for Sick Children (SickKids), a tertiary care center in Toronto, Canada. Participants met ≥ 4 of the 1997 ACR revised criteria for SLE diagnosis and/or the 2012 SLE International Collaborating Clinics (SLICC) criteria.12,13
Genotyping and imputation. Participants were genotyped using the Illumina Multi-Ethnic Global Array (MEGA) or the Global Screening Array (GSA; Illumina Inc.).14,15 Genotyping was conducted following protocols specified by The Centre for Applied Genomics at SickKids (see quality control [QC] measures in the Supplementary Methods, available with the online version of this article). SNPs not genotyped were imputed using the Sanger Imputation Server and the Haplotype Reference Consortium (HRC) and 1000 Genomes Project (1KGP) phase 3 as references.16 For participants with inferred European ancestry, HLA allele dosages were imputed using SNP2HLA and Type 1 Diabetes Genetics Consortium as a reference.17,18
PRS calculations. We selected SNPs for inclusion in schizophrenia PRS based on the most comprehensive schizophrenia GWAS conducted to date by the Schizophrenia Working Group of the Psychiatric Genomics Consortium (SWG-PGC).8 Utilizing the SWG-PGC PRS calculation for schizophrenia, and QC procedures as guiding principles (Supplementary Methods, available with the online version of this article), we computed 2 PRS using (1) genome-wide significant SNPs (P < 5 × 10–8), and (2) an expanded set of SNPs showing significance at a threshold of P < 0.05. After excluding SNPs through QC procedures, the GWAS PRS included 76/128 SWG-PGC SNPs and the extended PRS included 15,305/24,850 SNPs.
We used these SNPs to calculate additive, allelic schizophrenia PRS for each participant by identifying the risk alleles, weighting the allele dosages by the log odds ratios (log(ORs)) from the SWG-PGC study, and summing up the values to derive the PRS.8
Outcome measures. Demographic, clinical, and laboratory features for all participants were extracted from the SickKids Lupus Database, with details supplemented by medical records when required. NPSLE features were defined using the 1999 ACR list of case definitions for NPSLE.3 Features were independently validated by 2 pediatric rheumatologists (RLC and TD). In the case of discrepancies, additional pediatric rheumatologists (LTH, AMK, DML) acted as tiebreakers. We defined 2 outcomes compared to absence of NPSLE features: (1) any NPSLE feature, and (2) subtypes of NPSLE features (psychosis and nonpsychosis NPSLE features).
Covariates. Ancestry was inferred using 1KGP and the HRC as referents (Supplementary Methods, available with the online version of this article). The software tool ADMIXTURE was used to estimate relative proportions of ancestral groups within discordant subjects.19 Participants with single ancestral proportions ≥ 80% were classified into 1 of 5 ancestral groups: African, Amerindian, East Asian, European, or South Asian. Participants displaying ancestral proportions < 80% were classified as “Admixed.”
We also calculated SLE non-HLA PRS in all participants and SLE HLA PRS in ancestral European participants, by weighting allele dosages using log(ORs) obtained from the largest multiethnic SLE GWAS study conducted to date.1
Statistical analyses. We calculated counts and proportions for categorical variables and medians and IQRs for continuous variables. We compared the characteristics of those with and without NPSLE using chi-square statistics for categorical values and Wilcoxon rank-sum test for continuous variables.
We tested the association between PRS for schizophrenia and NPSLE using logistic and multinomial logistic regression. We ran univariate models and multivariable models adjusted for (1) inferred ancestry categories (European, Admixed, African, East Asian, South Asian), and (2) variables significantly associated with NPSLE in our cohort (P < 0.05). We conducted the multivariable-adjusted analysis to account for NPSLE risk and to increase power to detect genetic effects in our cohort. We calculated ORs, 95% CIs, and P values.
We also tested the association between each of the following and NPSLE: each individual genome-wide significant SNP (using a Bonferroni-corrected P value threshold of 6.58 × 10–4 [0.05/76]) and 2 SNPs located in the MHC region (rs115329265 and rs114541829) that were significant at a threshold of P < 0.05 in the SWG-PGC study but not included in their PRS calculations. To test the association between known SLE susceptibility loci and NPSLE, we regressed SLE non-HLA PRS and NPSLE, and SLE HLA PRS and NPSLE among European participants.
We completed 3 sensitivity analyses for less NPSLE-specific features of headache, anxiety, and/or mood disorders. First, we censored these participants from analyses. Second, we included them as controls. Last, we included them in their own subcategory of NPSLE features. All analyses were conducted in R version 3.6.3.20 This project was approved by the SickKids Institutional Research Ethics Board (REB #1000058324).
RESULTS
Our study cohort consisted of 513 participants, of which 424 (83%) were female. The median age at SLE diagnosis was 13.8 (IQR 11.2–15.6) years, and the median follow-up duration was 4.6 (IQR 2.8–7.5) years. Genetically inferred ancestry indicated that most participants were European (n = 157, 31%) or East Asian (n = 143, 28%). Four-hundred seventy-one participants were genotyped on the MEGA array, and 42 were genotyped on the GSA array. A total of 201 had any NPSLE feature (39%; Table 1). Inferred ancestry categories, malar rash, oral or nasal ulcers, arthritis, lymphopenia, Coombs-positive hemolytic anemia, and lupus anticoagulant and/or anticardiolipin antibodies were all significantly associated with NPSLE in our cohort (P < 0.05; Table 1). Of the 201 participants with NPSLE, subtype classification resulted in 60 (30%) participants with psychosis as a feature and 141 (70%) with nonpsychosis features (Table 2).
Demographic characteristics and clinical and laboratory features of cohort (n = 513).
Characteristics of NPSLE features among study cases by category (n = 201).
An increase in the GWAS PRS for schizophrenia was not significantly associated with increased odds of having any NPSLE feature vs no features (OR 1.04, 95% CI 0.87–1.26, P = 0.62) in ancestry-adjusted models (Table 3). Similarly, an increase in the GWAS PRS for schizophrenia was not significantly associated with increased odds of having psychosis (OR 0.97, 95% CI 0.73–1.29, P = 0.84) or other nonpsychosis NPSLE features (OR 1.08, 95% CI 0.88–1.34, P = 0.45) compared to no NPSLE. Results did not differ significantly in the univariate or full multivariable-adjusted models (Table 3), nor for the expanded schizophrenia PRS, which remained nonsignificant (Supplementary Table 1, available with the online version of this article).
Association between NPSLE, NPSLE subtypes, and schizophrenia GWAS PRS.a
Analyses of individual schizophrenia GWAS SNPs with NPSLE risk in ancestry-adjusted models indicated no significant associations with NPSLE (Supplementary Table 2, available with the online version of this article). Additional analyses of 2 schizophrenia risk SNPs in the MHC region that were excluded from the PRS calculations demonstrated a marginally significant association with NPSLE for rs115329265 (allele G, OR 0.69, 95% CI 0.48–0.97, P = 0.04) and no significant association for rs114541829 (allele G, OR 1.00, 95% CI 0.58–1.73, P > 0.99). We repeated these analyses in the European subset of our cohort (n = 157); however, results were not significant (results not shown).
SLE non-HLA PRS were normally distributed across our study population (P = 0.86). There was no significant association between SLE non-HLA PRS and NPSLE in the total cohort in ancestry-adjusted models (OR 1.21, 95% CI 0.98–1.49, P = 0.07). Results were similar when adjusted for all significant covariates (results not shown).
Of the 157 participants of inferred European ancestry, 76 (48%) had at least 1 NPSLE feature. SLE HLA PRS were not normally distributed across this population (P = 1.04 × 10–5). The distribution of SLE HLA PRS was not significantly different between participants with and without NPSLE (P = 0.64). There was no significant association between SLE HLA PRS and NPSLE (OR 1.17, 95% CI 0.83–1.67, P = 0.37) in the multivariable-adjusted model.
In sensitivity analyses, in (1) censoring participants with headaches, anxiety, and/or mood disorders as their only NPSLE features; (2) including them as controls; or (3) including them in their own subcategory of NPSLE features, we did not find an association between the schizophrenia PRS and NPSLE (results not shown).
DISCUSSION
Our study found no significant association between schizophrenia PRS and overall NPSLE risk, or with specific NPSLE features. We also did not find a significant association between SLE susceptibility loci and NPSLE. In this multiethnic cohort of patients with cSLE, there was no evidence for shared risk between schizophrenia genetic susceptibility loci and NPSLE.
The prevalence of NPSLE in our cohort was consistent with previous reports of NPSLE among cSLE cohorts at 39%, with a high prevalence of headaches (31%), psychosis (12%), and cognitive dysfunction (10%).21 While several studies utilize the 1999 ACR classification criteria to define NPSLE features, there remains a lack of case definition standardization.21 Depending on the manifestations and diagnostic methods, the prevalence of NPSLE reported varies between 22–95% among cSLE cohorts.21 Recognizing these discrepancies, we examined not only overall NPSLE risk, but also NPSLE risk by subtype, given our large cohort and detailed clinical and laboratory data to classify participants into specific subtypes. We defined 2 NPSLE subphenotypes: psychosis and nonpsychosis NPSLE, focusing on the clinical similarity between psychosis in NPSLE and schizophrenia to define subtypes.22 Our sensitivity analyses accounted for potential misclassifications of NPSLE status. With these groupings, we assumed similar relationships between genetic risk and each of the features within each group.
The SWG-PGC study from which we selected SNPs for inclusion in our study represents the largest schizophrenia GWAS conducted to date, including almost 37,000 cases and over 113,000 controls, and over 9 million SNPs tested.8 More recently, GWAS have looked at endophenotypes of schizophrenia, and cross-disorder analyses have looked at the genetic association between schizophrenia and other psychiatric or immune-mediated disorders. However, the SWG-PGC study remains the largest schizophrenia GWAS to date.9,10,11 Some of these studies have found shared genetic risks between SLE and schizophrenia to be mediated primarily by HLA alleles.10,23 We found nonsignificant and marginally significant associations between 2 individual schizophrenia SNPs in the MHC region and overall NPSLE risk in our cohort, suggesting a distinct pathobiology for schizophrenia and NPSLE in cSLE. Our findings warrant further validation.
Our study findings should be considered in light of some limitations. Our cSLE cohort was multiethnic, yet susceptibility loci for schizophrenia were derived from a primarily European GWAS.8 As a result, we may have excluded important non-European susceptibility loci from our PRS. In recent years, there has been an effort to improve representation of ancestrally diverse world populations in genetic studies. However, until the sample sizes for these studies are sufficient, our ability to draw inferences from non-European populations are limited.24,25 A multiethnic schizophrenia meta-GWAS in East Asian and European populations found 53 novel loci.24 Yet, the total number of cases included in this meta-GWAS was still 15,000 less than the SWG-PGC GWAS in Europeans that we used to select SNPs for our PRS. By using the largest GWAS to date, we are including the most robust risk loci for schizophrenia. Due to the highly polymorphic nature of the MHC region, and lack of reliable non-European HLA reference populations, we could only examine HLA alleles in Europeans.10,23 Neuropsychological evaluations were not routinely completed in all patients and neurocognitive function was determined by physician diagnosis. Efforts were made to avoid diagnostic biases by having 2 pediatric rheumatologists independently validate NPSLE features, as described in the methods. Finally, our analyses may have lacked sufficient power to detect significant associations, particularly in the NPSLE subtype analyses and in analyses stratified by ancestry.
Our study has several strengths. To our knowledge, it is the first to examine the association between genetic risk loci for schizophrenia and NPSLE specifically. The detailed clinical, laboratory, and genetic data for this large multiethnic cSLE cohort allowed us to look not only at overall NPSLE risk but also at specific subtypes, and to control for covariates associated with NPSLE in our cohort including ancestry and the presence of arthritis, lymphopenia, and antiphospholipid antibodies.
We did not observe a significant association between genetic risk loci for schizophrenia and NPSLE in a multiethnic cohort of children and adolescents with SLE. To our knowledge, our study is the first of its kind and requires additional validation in adult-onset SLE cohorts and independent childhood-onset SLE cohorts. Our observed lack of association between schizophrenia risk loci and NPSLE does not obviate a role for genetics for NPSLE risk. A genome-wide interrogation of NPSLE-specific manifestations is warranted. Knowledge of the genetics of NPSLE will provide a better understanding of the molecular mechanisms driving this complex disease, ultimately improving therapy and outcomes for people with SLE.
Footnotes
LTH is supported by The Arthritis Society Stars Career Award.
The authors declare no conflicts of interest relevant to this article.
- Accepted for publication September 23, 2021.
- © 2022 by the Journal of Rheumatology
