Genetic Risk Score Prediction in Ankylosing Spondylitis

Abstract

Background/Purpose:

The diagnosis of ankylosing spondylitis (AS) is delayed by on average 8–11 years after the onset of symptoms, and there is increasing evidence that early intervention in the disease can lead to better outcomes for patients. Early diagnosis of AS is challenging and the performance is still imperfect, even with magnetic resonance imaging (MRI). Whilst genetic risk scores (GRS) utilising only genomewide-significant associated SNPs capture only a small proportion of total heritability and consequently have modest discriminatory capacity, polygenic risk scores involving hundreds to many thousand SNPs capture a higher proportion of overall disease heritability and have greater discriminatory capacity and accuracy in disease-risk prediction.

Methods:

In this study, we developed two polygenic GRS, one for AS based on European-descent samples (7,742 AS patients and 14,542 controls), and a second based on East Asian-descent samples (6,001 AS patients and 4,943 controls). AS was defined according to the Modified New York Criteria and all the samples were genotyped using Illumina CoreExome microarrays involving ~270,000 SNPs after quality control. The European-GRS involved 3,947 SNPs and East Asian-GRS 8,659 SNPs. Validation was performed using 10-fold cross-validation in the originator population, and also tested in Turkish (873 patients and 961 controls) and Iranian case-control cohorts (430 AS patients and 761 controls). Genetic risk scores were calculated using the adaptive MultiBLUP algorithm (Speed and Balding, Genome Res, 2014). Discriminatory capacity was tested by receiver operating characteristic analysis, and reported as area under the curve (AUC), with sensitivity and specificity reported at the best Matthews correlation coefficient.

Results:

For the European GRS the AUC was 0.92 (83% sensitivity, 92% specificity), compared with 0.87 by using imputed HLA-B27 status alone (P=2.4×10-10 vs GRS). In the East Asian cohort, the AUC is 0.95 (91% sensitivity, 95% specificity). The risk score in AS patients is significantly higher than the score in controls in both cohort (p<2.2×10-15). Both models showed moderate discriminant capacity in other ethnics groups (European model AUC 0.85, 0.85 and 0.78 in the Turkish, Iranian and East Asian cohorts respectively; East Asian model AUC 0.84, 0.87 and 0.88 in the Turkish cohort, Iranian and European cohorts respectively).

Conclusion:

Our results indicate that in AS GRS has high discriminatory capacity and could be of clinical utility in early diagnosis. Given that the AUC for MRI imaging is currently thought to be 0.90, this data suggests that genetic test scoring performs at least as well as the current gold standard imaging procedure. Given the cost of a SNP microarray is <$US50, this test is potentially far cheaper than MRI and less expensive than traditional HLA-B27 testing.