Association of Genetic Risk for Rheumatoid Arthritis With Cognitive and Psychiatric Phenotypes Across Childhood and Adolescence

Key Points Question Is genetic liability for rheumatoid arthritis associated with cognitive and psychiatric phenotypes in children prior to the clinical onset of disease? Findings In this cohort study of 7977 children and adolescents, genetic liability for rheumatoid arthritis was associated with lower total, performance, and verbal IQ at age 8 years and symptoms of hyperactivity and inattention from ages 4 to 16 years. However, there was little evidence of association with other domains of psychopathology. Meaning These findings support a primary etiological association between genetic risk for rheumatoid arthritis and cognitive phenotypes that is not simply secondary to disease-related processes or reverse causation.


Hyperactive and inattentive symptoms
Hyperactive and inattentive symptoms were assessed using the parent-rated 5-item Strengths and Difficulties Questionnaire (SDQ) 17 at approximate ages 4-16 years. At each age individuals were given a hyperactive and inattentive symptoms score ranging from 0-10. Binary measures were created at each age using 7 as a cut-off, as previously used, 18,19 to indicate individuals with abnormal behavior.

Inflammatory markers
When the ALSPAC participants were 9 years of age, blood samples were collected from non-fasting participants and were immediately spun and frozen at -80°C. Inflammatory markers were assayed in 2008 after a median of 7.5 years in storage with no previous freeze-thaw cycles during this period. Interleukin-6 was measured by enzyme-linked immunosorbent assay (R&D Systems, UK), and high-sensitivity C-reactive protein was measured by automated particle-enhanced immunoturbidimetric assay (Roche, UK) as reported in Khandaker et al. 2014. 20 All inter-and intra-assay coefficients of variation for interleukin-6 and Creactive protein were less than 5%. C-reactive protein was also measured in blood samples were collected from fasting participants who gave consent for venipuncture during clinical assessment at age 16 years. Blood samples were immediately spun, frozen and stored at -80 °C, which were analyzed within 3 -9 months of blood sampling with no freeze-thaw cycles in between. High sensitivity C-reactive protein was measured by automated particle-enhanced immunoturbidimetric assay (Roche, UK).
Inflammatory marker measures were log transformed before use.

Gene-based analyses
Gene-based analyses were performed using MAGMA (Multi-marker Analysis of GenoMic Annotation). 21 Gene test-statistics were calculated using the "multi" flag option, which aggregates tests using the mean of all SNPs, and the most significant SNP per gene. MAGMA was also used to derive gene-set enrichment statistics. This uses a competitive test to assess whether genes within the gene-set are more associated with the phenotype than genes outside of the gene-set. MAGMA implements this using a specialized version of generalized linear modelling which controls for additional variables including gene-size, linkage disequilibrium and minor allele counts. We tested whether RA associations were enriched in 7 321 gene-ontology gene-sets. To account for multiple testing, we used q-value corrections as implemented in the R Package "qvalue". 22 For gene-sets surviving qvalue correction, we used hierarchical clustering analysis to identify gene-sets that were correlated based on shared gene overlap as implemented in the R package "stats". 23 Results of the hierarchical clustering analysis were plotted in as a dendrogram format and broader "cluster" definitions were derived via manual inspection (eFigure 3).

Sensitivity analyses
To assess the robustness of our findings, analyses were repeated using PRSs based on SNPs which were associated with RA, IBD and MS at a range of GWAS P value thresholds (PT; P ≤ 0.5 to P ≤ 1e -7 ) and PRSs that omitted all SNPs from the extended MHC region (chromosome 6: 25-34Mb) (see above). Analyses were also repeated after excluding 9 individuals (0.11% of participants with PRS data) who reported a doctor diagnosis of RA at age 22 years and 649 individuals (8.14% of participants with PRS data)

eFigure 4. Associations between polygenic risk scores (PRSs) for rheumatoid arthritis based on single-nucleotide polymorphisms (SNPs) within identified gene-set clusters, and hyperactive and inattentive symptoms at age 13 years.
PRSs were generated using SNPs within the gene-set sub-clusters of cluster 1 (C1): C1a, lymphocyte activation; C1b, lymphocyte differentiation; and SNPs within the gene-set sub-clusters of cluster 2 (C2): C2a, activated lymphocyte homing; C2b, lymphocyte effector functions; C2c, immune activation; C2d, Th2 effector characteristics; C2e, Th1 and Th17 effector characteristics; C2f, lymphokine activities; C2g, immune effector functions. PRSs were also generated using all SNPs within cluster 1 (All C1), cluster 2 (All C2) and clusters 1 and 2 combined (All C1 and C2). Odds ratios (OR) per standard deviation (sd) change in PRS are shown (data markers), with upper and lower error bars indicating 95% confidence intervals (CIs). ORs are shown for PRSs generated using lists of SNPs meeting a 0.05 P value threshold. Red dashed line indicates the null value (1).   IQ measures are standard IQ measures as measured by the WISC-III. The working memory score is the raw score (i.e. number of correct answers) from the Backward Digit Span task of the WISC-III. The verbal learning score is the raw score (i.e. number of correct answers) from the Non-word Repetition task of the WISC-III. The processing speed score is the raw score (i.e. number of correct answers) from the Coding task of the WISC-III. The problem solving score is the raw score (i.e. number of correct answers) from the Block Design task of the WISC-III. The selective attention score is measured as the average time taken (in seconds) to identify pairs of identical spaceships (adjusted for motor speed) within the Sky Search task of the TEA-Ch. The attentional control is measured as the average time taken (in seconds) to complete the Opposite Worlds task of the TEA-Ch. All measures were standardised to have a mean of zero and a standard deviation of 1 before analysis. b sample sizes for each association analysis eTable 2. Mean age, number of individuals and proportion of sample with psychopathology for each psychopathology outcome measure for the total sample and sample with genetic data. Note: PRS, polygenic risk score; N, analysis sample size; β, linear regression coefficients representing a standard deviation change in IQ per standard deviation change in the rheumatoid arthritis (RA) PRS; 95% CI, 95% confidence interval; P, linear regression p-value; R 2 , proportion of variance in IQ which can be explained by the RA PRS a PRSs were generated using single-nucleotide polymorphisms (P-value threshold ≤ 0.05) within the gene-set sub-clusters of cluster 1 (C1): C1a, lymphocyte activation; C1b, lymphocyte differentiation; and SNPs within the gene-set sub-clusters of cluster 2 (C2): C2a, activated lymphocyte homing; C2b, lymphocyte effector functions; C2c, immune activation; C2d, Th2 effector characteristics; C2e, Th1 and Th17 effector characteristics; C2f, lymphokine activities; C2g, immune effector functions. PRSs were also generated using all SNPs within cluster 1 (All C1), cluster 2 (All C2) and clusters 1 and 2 combined (All C1 and C2) a see eTable 1 for exact ages in months eTable 8. Associations between polygenic risk scores for rheumatoid arthritis based on single-nucleotide polymorphisms meeting a P-value threshold of 0.05 within identified gene-set clusters, and hyperactive and inattentive symptoms at age 13 years. Note: PRS, polygenic risk score; N, analysis sample size; OR, logistic regression odds ratio representing a change in odds of hyperactive and inattentive symptoms compared to baseline per standard deviation change in the rheumatoid arthritis (RA) PRS; 95% CI, 95% confidence interval; P, logistic regression p-value; pR 2 , pseudo R 2 used to evaluate the goodness-of-fit of logistic models a PRSs were generated using single-nucleotide polymorphisms (P-value threshold ≤ 0.05) within the gene-set sub-clusters of cluster 1 (C1): C1a, lymphocyte activation; C1b, lymphocyte differentiation; and SNPs within the gene-set sub-clusters of cluster 2 (C2): C2a, activated lymphocyte homing; C2b, lymphocyte effector functions; C2c, immune activation; C2d, Th2 effector characteristics; C2e, Th1 and Th17 effector characteristics; C2f, lymphokine activities; C2g, immune effector functions. PRSs were also generated using all SNPs within cluster 1 (All C1), cluster 2 (All C2) and clusters 1 and 2 combined (All C1 and C2) a see eTable 2 for exact ages in months eTable 9. Sensitivity analysis for associations between rheumatoid arthritis polygenic risk scores and cognitive phenotypes at age 8 years a . Note: N, analysis sample size; OR, logistic regression odds ratio representing a change in odds of hyperactive and inattentive symptoms compared to baseline per standard deviation change in the rheumatoid arthritis (RA) PRS; 95% CI, 95% confidence interval; P, linear regression p-value; pR 2 , pseudo R 2 used to evaluate the goodness-of-fit of logistic models a see eTable 2 for exact ages in months eTable 11. Associations between polygenic risk scores for rheumatoid arthritis, inflammatory bowel disease and multiple sclerosis, and IQ phenotypes at age 8 years a across a range of polygenic risk score P value thresholds.