Cindy Boer

258 | Chapter 6 several advantages: such as minimizing phenotyping costs and reducing phenotypes to single or few data points, allowing for easy large scale data collection. However, min- imal phenotypes can introduce significant bias in the phenotype and reduce GWAS power[14]. For example individuals with the same hospital diagnosis of osteoarthritis (cases) may not have the same symptoms, pathology or underlying causal pathways, as osteoarthritis is a highly heterogenous disease [15, 16]. This heterogeneity also occurs for self-reported osteoarthritis, with an additional bias of uncertainty if the self-report- ed diagnosis is an accurate one. In Chapter 4 , not only different osteoarthritis case definitions were used in the different cohorts, with the majority being the minimal phenotype of self-reported/ hospital diagnosed osteoarthritis. This heterogeneity between phenotype definitions and bias from the minimal phenotypes significantly introduces noise into the GWAS results[17]. Specifically, some of the identified loci could be nonspecific for osteoarthri- tis[14], i.e., misdiagnosis in self-reported osteoarthritis, which would have significant implications for follow-up molecular characterization and pathway identification. Ad- ditionally, osteoarthritis was defined as a dichotomous phenotype: present or absent. However, this black and white presentation of osteoarthritis completely ignores the un- derlying genetic, biological and pathological complexity of this disease[13, 18, 19]. Thus these biases may explain the low genetic osteoarthritis heritability found in Chapter 4 . Also, they illustrate the need for phenotypic heterogeneity reduction methods and the use of phenotypes that do account for the genetic, biological and pathological complex- ity of osteoarthritis. Phenotyping: Quantitative phenotypes One way to increase GWAS power and use phenotypes that can more accurately rep- resent the range of osteoarthritis disease severity, is to use quantitative phenotypes [17]. In Chapter 2.2 and 3.1 , the radiographic summarized osteoarthritis phenotype of KLsum was used. The advantage of this phenotype is that it measures osteoarthritis severity across multiple joints at the same time. Creating a semi-quantitative measure, more accurately representing the severity of osteoarthritis in these joints than a dichot- omous trait could do. Thus improving GWAS power, as is evident by the thumb KLsum and osteoarthritis associated SNV rs10916199 in the WNT9A locus. This SNV was found to be genome-wide associated (p-value=2.*10 -12 ) with thumb KLsum in a modest total sample size of ~8,700 ( Chapter 3.2 ), however this variant was only found to be ge- nome-wide significant associated (3.5*10 -10 ) with thumb OA in the GO consortium with a sample size of 10,536 cases and 236,919 controls ( Chapter 4 ). Similar results were found for the osteoarthritis of the finger and hand KLsum associated SNV, rs4764133 located near MGP ( Chapter 3 and Chapter 4 )