In this study, we developed a psoriasis genetic risk score (GRS) which included 10 highly replicated loci from recent GWAS and large cohort studies. We found that a weighted GRS, which accounts for the odds ratio of each allele, is a better discriminator of cases and controls than a simple count GRS. This agrees with the results of Piccolo et al 
who also found that a weighted GRS could capture considerably more genetic risk compared to a cGRS. Our weighted GRS was more associated with risk of psoriasis than any single SNP alone, with persons in the highest wGRS quartile having a greater than 10-fold increased risk of psoriasis compared to persons in the lowest quartile.
Of the 10 SNPs evaluated in this study, the strongest signal was found at the HLA-C
locus at rs10484554, for which there was a 206% elevated risk of psoriasis with each risk allele. Notably, the predictive capability for this single SNP was as good as that of the other 9 non-MHC loci combined (). Furthermore, our study may actually underestimate the true effect of HLA-Cw6
because this allele has been shown to be more strongly associated with psoriasis than its tag SNP rs10484554 used in this study 
. The strong effect of this single locus on disease susceptibility stands in contrast to other common diseases such as Type 2 diabetes, coronary artery disease and prostate cancer, in which the risk is more evenly distributed amongst multiple alleles 
. Thus, the genetic architecture underlying common complex diseases may vary according to the disease. The 9 additional non-MHC SNPs had a more modest effect on the risk of psoriasis, with each allele conferring an elevated risk of 94% or lower. While these alleles have little predictive power individually 
, their combined addition to the HLA-C
allele resulted in a composite wGRS that displayed a slightly enhanced ability to discriminate between people with and without psoriasis.
With regard to sub-phenotypes, our data show that the weighted genetic risk score was higher amongst cases with an earlier-onset of psoriasis and with a family history of disease. This may largely be explained by the observation that the HLA-C
locus has a large impact on the wGRS, and several studies have previously demonstrated that HLA-Cw6
-positive patients show an earlier disease onset and greater family aggregation than those without it
. In addition, since a positive family history is a surrogate for genetic predisposition, it is not surprising that a higher GRS was seen in cases with affected first degree relatives. A marginal significance (p
0.072) was observed between the wGRS and guttate psoriasis. This could be explained by the stronger association of HLA-Cw6 with guttate psoriasis compared to plaque psoriasis 
. We did not detect an association between the wGRS and PsA. This may be in part due to the small sample sizes of the subgroup analyses. In addition, since the GRS SNPs used in this study were identified based on genetic studies of psoriasis and not PsA per se, lack of association with PsA may simply indicate that the present SNPs do not account for genetic susceptibility to the arthritic component.
The estimated proportion of genetic variation explained by the 10 included SNPs in this study was11.6%, including 6.7% due to HLA-C
. This suggests that many additional susceptibility loci for psoriasis remain to be discovered. Similar results have been reported in other complex diseases. For example, the proportion of heritability explained by known common SNPs for Crohn's disease, systemic lupus erythematosus and type 2 diabetes was estimated to be around 20%, 15% and 6%, respectively 
. One hypothesis is that rare variants with higher penetrance, not covered by the present genotyping, might be able to explain the remaining heritability 
. Searching for such genetic factors are underway thanks to advances in next generation sequencing technologies 
. Furthermore, although the wGRS is highly significantly associated with psoriasis susceptibility, the predictive capability is seemingly modest (AUC for ROC curve 72.0%). It will be interesting to see whether the discriminatory accuracy can be improved as additional rare variants in psoriasis are included.
Several limitations to this study need to be acknowledged. First, although we captured the most significant psoriasis loci known to date for calculation of the GRS, we did not include all known psoriasis loci. For example, human β defensin (HBD
) was not evaluated in this study since we were unable to identify a proxy SNP for HBD copy number
. Additionally, during the preparation this manuscript, several novel psoriasis loci were identified by very recent genome-wide association studies but not included in this study 
. Among them, TRAF3IP2
was shown to be the most reproducibly associated locus and showed a psoriasis odds ratio of 1.70. However, none of the newly identified loci is estimated to explain more than 1% of the heritability. Inclusion of these additional loci to the SNPs described in our model results in an estimated proportion of heritability explained of 15%. Second, to increase the robustness of our control data set, we included a number of controls from the public iControlDB database. However, we used a panel of European AIMs to correct for potential population stratification. Moreover, when we conducted the association analysis removing the iControlDB subjects, we obtained similar results with respect to the direction of association and magnitude of the odds ratio for each SNP (data not shown). Third, two of the GRS SNPs in the iControlDB dataset were imputed rather than directly genotyped and in two instances we used proxy SNPs for the originally reported variants. Although using imputed or proxy SNPs might lead to less accurate results, we ensured that only SNPs with high imputation confidence >98% and proxy SNPs with r2
>0.9 were included in final analysis. Fourth, only Caucasians were included in this study. Thus, the characteristics of the GRS calculated here may not applicable to other ethnic populations. Fifth, our prediction model is based on the same data used to construct the model, which may lead to overfitting and overestimation of the AUC. Finally, this was not a prospectively-designed study, and thus we were unable to truly examine the discriminatory power of these SNPs.
In summary, we found that a GRS combining 10 psoriasis risk loci was significantly associated with an increased risk of psoriasis. The weighted GRS approach increased the power in discriminatory accuracy, compared to the counted GRS and any of the risk alleles considered alone. A higher GRS was associated with early-onset of disease and positive family history. The 10 risk loci account for only 11.6% of the genetic variance in psoriasis, suggesting that additional susceptibility alleles remain to be identified.