For our “discovery” phase, 223 PS cases (132 cases with PS without arthritis and 91 PS cases with arthritis (PSA) were typed on the Illumina HumanHap300 arrays. We compared case data to publicly available genotype data of 519 European controls from the New York Cancer Project
collected with the same platform. The number of cases used for this scan is smaller than that used in many recently described genome wide association scans. However, the 91 cases of PSA had at least one first degree relative with PS and were expected to be enriched for genetic factors. Power calculations based on 223 controls and 519 controls indicated that using a threshold of P<5×10−5
, we had 70% power of detecting a locus with a genotype relative risk (GRR) of 2.0, and over 99% power to detect a locus with a GRR of 3.0 such as the MHC (see below). However, many replicated associations have small GRRs 
and we had only 10% power of detecting a locus with a GRR of 1.5.
Following the genotyping of samples, stringent quality control measures were implemented. We required that all samples used for the discovery phase passed a 93% genotyping call rate threshold, and that all SNPs passed a 95% call rate threshold. Justification for this threshold is based on the evaluation of empirical distributions (Figures S1
). With sample call rates ≤93%, there was an elevation in observed sample heterozygosity, i.e. deviation from Hardy-Weinberg equilibrium, suggesting possible genotyping errors (e.g., sample contamination or allele drop-out). Likewise, there was a significant discrepancy of missingness between case and control groups when the SNP success rate was <95%.
For the discovery phase, a total of 311,398 SNPs were pruned to 305,983 SNPs after filtering for low call rate, minor allele frequency <0.01 and deviation from Hardy-Weinberg equilibrium (P<0.001). Quality control also led to the removal of 29 samples, leaving 218 cases for further analysis (Table S1
). The average genotyping rate in the remaining individuals was 0.995.
To investigate other biases
that could be introduced with shared controls, we assessed the median distribution of test statistics using the genomic-control factor λGC
. With a set of 463 ancestry informative SNPs (AIMs), λGC
1 indicating no inflation). We also performed analysis on the same set of AIM SNPs with STRUCTURE software
. Under the assumption of two population clusters, there was no association between the most likely inferred cluster and case/control status and the average allele frequency difference between clusters was less than 2.5%. These results showed that population substructure is unlikely to be confounding our results. However, analysis of all markers used in the discovery study yielded λGC
1.101 before correction (Q-Q plot shown in Figure S3
). A similar value was obtained with EIGENSTRAT 
1.107. Examining stratified subsets of cases (PS without arthritis or PSA) with all markers also yielded similar λ values (PS without arthritis: λGC
1.07; PSA: λGC
1.05). Following adjustment of P values with the genomic control method, λ
1˜ The discovery P values adjusted by the genomic control method as implemented in PLINK
are presented in the tables and figures.
To detect associations, we first performed a preliminary analysis with a Cochran-Armitage trend test 
. illustrates negative logarithm of the P values obtained across the genome, considering all cases and all controls () and considering only the 91 PSA cases and all controls (). Results were then rank ordered on the basis of P values. 84 SNPs in 35 genomic regions were associated with P<5×10−5
; a level that we would informally expect to observe by chance roughly 15 times in this scan given the number of tests performed if all SNPs were independent. A subset of SNPs from 120 regions were investigated further. Criteria for selection included the strength of the discovery P value, particularly when several SNPs from a single region showed evidence for association, a possible biological role of a gene harboring a SNP with some evidence for association, or localization of SNPs with moderate evidence for association to a known psoriasis susceptibility locus (e.g. PSORS4). We also included the previously reported associated SNPs in IL23R and IL12B
Summary of genome-wide association scan results for all cases and the PSA subgroup.
An independent cohort of 577 PS cases from the U.S. and 737 U.S. controls were used for the replication stage; 94 of these cases had also been diagnosed with PSA. To examine the potential role of variants upon PSA susceptibility specifically, 576 PSA cases from the UK and 480 controls from the UK were also employed. An alternative genotyping technology (iPlex; Sequenom) was used for the replication phase. The platforms used for the discovery and replication phases gave very similar results: Concordance rates on the basis of 116 samples and 301 SNPs typed with both platforms was 98.74%.
Our 100 top ranked SNPs with any cohort (PS, PS without arthritis, or PSA) are listed in Table S2
, to facilitate future attempts to replicate our findings. A total of 289 SNPs, including SNPs from the MHC, and two previously reported associated SNPs within IL12B and IL23R
were genotyped in the replication analysis.
The MHC, and in particular, the HLA class I region, is the only region that has been shown to be consistently associated with PS. The first nine top-ranking SNPs were from the MHC and seven were significant, even when adjusted with the Bonferroni correction for multiple tests (Table S2
). Overall, 32 SNPs from the MHC had adjusted P values <5×10−5
(, Table S2
). The most significant association was with rs10484554 (adjusted P
, GWA scan; P
, replication; P
, combined) (, ). This SNP lies 34.7 kb upstream from the transcriptional start site of HLA-C. Strongest association with this region is consistent with previous results from our group and others 
. The rs10484554*T allele had frequencies of 0.325 in U.S. cases and 0.15 in U.S. controls (OR: 2.8 (95% CI: 2.4–3.3). To determine the relationship of this allele with the classical HLA-C allele strongly associated with psoriasis (HLA-Cw*0602), we investigated the transmission of this allele with classical HLA-C alleles in ~250 nuclear families with psoriasis that we have reported elsewhere 
. The rs10484554*T allele was detected on nearly all haplotypes with HLA-Cw*0602 or HLA-Cw*1203 alleles (results not shown), and was also strongly correlated with the previously described highly associated PSORS1 SNP n9*G
(rs10456057*G) allele. We have previously shown that SNPs upstream from HLA-C are more strongly correlated with PS than HLA-Cw*0602 is, and that these risk alleles are also correlated with HLA-Cw*1203 
. Hence, rs10484554*T may be a good proxy for the PSORS1 variant.
Association localization plots for the MHC following discovery and replication phases.
Summary of association with previously reported PS susceptibility loci (MHC, IL23R and IL12B) in U.S. PS cohort (810 cases, 1256 controls).
In the case of the U.K. PSA replication samples, rs10484554 was again highly significant (P
) (), although the frequency of the rs10484554*T allele exhibited population differences when frequencies in the U.K. and U.S. were compared. In the U.K. the rs10484554*T allele was found at a lower frequency in cases and controls (0.19 and 0.07 respectively; OR: 2.4 (95% CI: 1.8–3.1)).
Summary of association with previously reported PS susceptibility loci in U.K. PSA cohort (576 cases, 480 controls).
A second SNP from the HLA class I region lying between MICA and MICB (rs2395029) was highly associated with PS and PSA. This SNP results in the G2V polymorphism of the class I gene HCP5 (HLA complex P5) which encodes an endogenous retroviral element. For this SNP, PS was associated with a combined P
in the U.S. cohort and 1.86×10−10
in the U.K. PSA cohort (, ). The OR of the rs2395029*C allele with both PS and PSA was higher than with any other SNP tested (4.1 and 3.2 with PS and PSA respectively). This allele was found at a frequency of ~0.12 in cases and 0.04 in controls and did not exhibit the population frequency differences of rs10484554. The LD relationship between rs2395029 and rs10484554 is not strong (r2
0.33 in European CEPH HapMap samples and r2
0.23 in our U.S. case/control cohort). Conditioning upon rs10484554, the P value for rs2395029 was still significant (P
), hence effects from this SNP are likely to be independent.
HCP5 is expressed primarily in cells of the immune system such as spleen, blood and thymus (http://smd-www.stanford.edu/
), consistent with a potential role in autoimmunity. This allele was recently shown to explain 9.6% of the total variation in viral set point following HIV-1 infection
. This is of interest, since psoriasis can be triggered by infection with HIV and other viruses. Hence, it is possible that HCP5-C carriers mount a strong immune reaction to viral infection, but that in genetically susceptible individuals, this reaction leads to excessive inflammation in skin and joints. Overall, our observations indicate that MHC class I region SNPs are more highly associated with both PS and PSA than any other SNPs.
A recent global association scan using a set of pooled PS samples and controls against a set of 25,215 genecentric SNPs confirmed a previously reported association with IL12B (rs3212227 in its 3′ UTR) 
and identified a second region of association 60 kb upstream from its mRNA start site (rs6887695) 
. An analysis of additional genes encoding components of the IL12B pathway lead to the identification of associations with Il23R (R381Q: rs11209026 and L310P: rs7530511) 
. These SNPs were proposed to mark a common psoriasis-associated haplotype. Rs11209026 is also the SNP within IL23R reported to be associated with Crohn's disease
In our discovery cohort, the most significant association in the IL23R interval was obtained with a different SNP from that described previously as being associated with PS (rs11209026). This SNP, where P
0.0039 in the discovery cohort (), has not previously been reported to be associated with PS. The LD relationship between rs12131065 and the previously associated rs11209026 SNP is low (r2=
0.031 in HapMap CEPH European samples; 0.009 in cases; 0.026 in controls). Conditioning upon rs11209026, the P value for rs11209026 was 0.013. Hence, effects from this SNP may be independent of rs11209026 and its association with PS should be investigated in other cohorts.
SNP rs12131065 lies downstream from IL23R (63 kb from rs11209026) and 4.041 kb upstream from the gene for interleukin 12 receptor B2 (IL12RB2) (). IL12RB2 is involved in IL12 dependent signaling, is upregulated by gamma interferon in Th
1 cells and plays a role in Th
. Association with a SNP closer to IL12RB2 than IL23R is of interest since animals where IL12RB2 is inactivated develop autoimmune disease
Association localization plots for the ILI23R region on chromosome 1.
Association with the previously reported IL23R associated SNP rs11209026 in the discovery cohort was not significant (adjusted P
0.081). Genotyping of rs11209026 and rs12131065 in the U.S. replication cohort yielded combined P values of 1.4×10−4
and 0.001 respectively () consistent with replication of this locus with respect to previous studies. In the case of these two SNPs, the protective T and A alleles were found at frequencies of ~0.04 and 0.2 in cases versus ~0.07 and ~0.24 of controls respectively. In the U.K. replication PSA cohort, association with rs11209026 was consistent with replication (P
), with the rs11209026*T allele being found at frequencies of ~0.04 in cases and ~0.08 in controls ().
Although the associated IL12B SNPs rs3212227 and rs6887695 were not interrogated by the Illumina screening panel of SNPs used here, typing of these SNPs in our replication U.S. case/control cohorts yielded P values of 0.021 and 5×10−5
and replicated previously reported associations (). In the U.K. PSA cohort, association with rs6887695 was also consistent with replication (P
0.0013) (OR: 0.69; 95% CI: 0.56–0.85)) ().
Novel Psoriasis Loci
In the discovery cohort, there were four SNPs from 13q13 where P<5×10−5
). These were: rs1186468, rs4514547, rs4569133 and rs7993214 (Table S2
). These SNPs lie within a region on chromosome13q13 that encodes the conserved oligomeric golgi complex component 6 (COG6) gene and a lipoma HMGIC fusion partner (LHFP)
. Three of the top associated SNPs were tested in the U.S. replication cohort, and all showed evidence of replication at P<0.05 (, ). Results were most significant with rs7993214 (adjusted P
, GWA scan; P
0.0033, replication; P
, combined). Rs3812888, (adjusted P
0.0017, GWA scan; P
, replication; P
, combined) was the only SNP where replication results would remain significant following the stringent Bonferroni correction for multiple tests (P
0.048). The OR of the rs3812888*C allele was 1.38 (95% CI: 1.15–1.66). The rs3812888*C allele was found at frequencies of 0.43 in cases and 0.35 in controls.
Potentially novel loci from GWA scan.
Association localization plots for novel replicated region on chromosome 13.
COG6 is a component of the conserved oligomeric golgi (COG) complex and is involved in intracellular transport and glycoprotein modification
. The glycosylation pathways in the golgi apparatus must be intact for protein secretion to continue unabated. In C. elegans
, a COG complex is required to glycosylate an ADAM protease (a disintegrin and metalloprotease)
. In humans, variants within some ADAM genes lead to inflammatory diseases. For example, ADAM33 is an asthma susceptibility gene whose catalytic domain undergoes glycosylation
. Recent genetic studies suggest that ADAM33 is a psoriasis susceptibility gene as well 
. Hence, COG6 could be involved in glycosylation of ADAM33 or other ADAM proteases. Staining of skin sections with a COG6 antibody revealed cytoplasmic staining in the epidermis as well as strong T-cell staining (). There was variable expression of the protein in non-lesional skin samples, but there was uniformly strong expression in all lesional sections. Very little is known about LHFP. It is a subset of the superfamily of tetraspan transmembrane protein encoding gene. Expression analysis from SOURCE (http://smd-www.stanford.edu/
) indicates that highest levels are found in the ear and spinal cord.
Immunostaining of normal, non-lesional and lesional skin for IL2, IL21, COG6 and SPPL2A proteins.
When the PSA discovery cohort (n
91) was analyzed separately, four SNPs from a region on chromosome 15q21 between ubiquitin specific protease 8 (USP8) and tumor necrosis factor, alpha-induced protein 8-like 3 (TNFAIP8L3) were associated with P<5×10−5
). In the case of the most highly associated SNP (rs4775919), adjusted P
. Following replication genotyping, this and rs3803369 were associated with PS with P values consistent with replication (for rs3803369, adjusted P
, GWA scan; P
0.013, replication; P
combined; Table S2
, , ). The rs3803369*A allele was found at a frequencies of ~0.2 in cases and 0.15 in controls (OR 1.43, 95% CI: 1.21–1.69). Other genes in this region include the transient receptor potential melastatin 7 (TRPM7)
, signal peptide peptidase like 2a (SPPL2A)
and AP4E1, a member of the heterotetrameric adaptor protein (AP) complexes (). TNFAIP8L3 is a novel protein. It harbors a domain (DUF758) that is found in several proteins induced by tumor necrosis factor alpha (TNFA), but its function is unknown. However, the most plausible candidate is SPPL2A that catalyzes the intramembrane cleavage of TNFA, triggering the expression of IL12 by activated human dendritic cells
. Staining of skin sections with an SPPL2A antibody () revealed profound staining of the epidermis, and staining of some dermal cells in both lesional and non-lesional skin.
Association localization plots for novel replicated region on chromosome 15.
This region is also of interest however, because a processed pseudogene for one of the genes in this region (USP8) is found upstream from HLA-C
. As discussed earlier, this region is one that is most likely to harbor the PSORS1 variant. To ensure that our observations were not due to cross-hybridization of chromosome 15q21 SNPs with PSORS1 SNPs, we investigated alignment of genes from this region of chromosome 15q21 with the remainder of genome. We did not detect any significant identity with any other region, including the MHC. This, and the fact that chromosome 15 associated SNPs are in Hardy-Weinberg equilibrium indicate that our results are unlikely to be artifactual and due to amplification of PSORS1 sequences. The similarity between the PSORS1 and 15q21 variants and their biological consequences need to be investigated further since it may provide important insights into the nature of the PSORS1 variant. However, it is worth noting that our PSA cases which provided strongest evidence for association with 15q21 all had at least one first degree relative with PS, and association with this locus may be harder to detect in “sporadic” cases.
We also observed association of PS with a region of the Epidermal Differentiation Complex (EDC), which harbors a previously established psoriasis locus (PSORS4)
. In this instance, rs6701216 yielded a combined P
(OR 1.45) (, Figure S4
). This SNP lies within the late cornified envelope 1C gene (LCE1C)
, and is one of a family of genes that are transcribed very late in epidermal differentiation. Our localization of PSORS4 to the LCE region is of interest, and its refinement will be described in detail elsewhere (Robarge et al., in preparation).
There were two other regions selected for follow-up where P values were <0.05 in the PS replication cohort, and where evidence for association increased in the combined cohort (). One was an intergenic region located between granulysin (GNLY) and atonal (ATOH) on chromosome 2p11. The most highly associated SNP was rs2164807 (adjusted P
0.0015, GWA scan; P
0.0039, replication; P
combined). For this SNP, the G allele was found at a frequency of ~0.47 in cases and ~0.39 in controls (OR: 1.35, 95% CI: 1.18–1.54). GNLY (Protein NKG5, Lymphokine LAG-2) is of considerable interest with respect to psoriasis. It is present in cytotoxic granules of cytotoxic T lymphocytes and natural killer cells, and is released upon antigen stimulation
. It has been shown to have antimicrobial activity against a broad range of microbes including Gram-positive and Gram-negative bacteria, fungi, and parasites including M. tuberculosis
and other organisms. Priming of granulysin in CD4 is dysregulated in the CD4+ T cells of HIV-infected patients
Other genes that should be evaluated in additional PS cohorts on the basis of replication P values <0.05, and increased significance in combined cohorts () are calponin-like transmembrane domain protein (CLMN)
, the gene for the catenin member, CTNNA2
and a gene desert on chromosome 3q13.
Association with Chromosome 4q27: A Recently Identified Autoimmune Locus
In the discovery cohort there were 3 SNPs from chromosome 4q27 with P<5×10−5
). These were rs13151961, rs6822844 and rs6840978 (Table S2
, Figure S6). The most significant of these was rs13151961, where adjusted P
(). Association of this region with PSA was confirmed in an independent cohort of patients from the UK (for rs13151961, P
0.003, ), where the frequency of the associated T allele was ~0.25 in cases and ~0.31 in controls (OR: 0.64; 95% CI: 0.49–0.84). Association with additional SNPs in high LD with rs13151961 (rs6840978 and rs6822844), was also replicated in this cohort. Although association could not be replicated in our U.S. cohort with a case/control approach, the trend in allele frequencies in cases versus controls was similar to that seen in the U.K. cohort (for rs13151961 the frequency of the T allele was ~0.26 in U.S. cases and ~0.29 in our U.S. controls, ).
Association results at chromosome 4q27 in PSA and PS cohorts (U.S.: 810 PS cases, 1256 U.S. controls; U.K.: 576 PSA cases, 480 controls).
A recent study from the Wellcome Trust Case Control Consortium 
identified this 4q27 region in a search for risk factors for type 1 diabetes (T1D). In a follow-up study 
, some support for this association with T1D was provided. In the latter study, association of this region with Grave's disease (GD) was also tested, and some evidence for association with the complementary allele of rs17388568 to that seen in T1D was obtained. The same region was also reported to be associated with Celiac disease (CeD) 
. Recent evidence is also provided for a role for this region in rheumatoid arthritis and T1D from the Netherlands
. In that study the rs6822844*A allele was reported to be a perfect proxy for a haplotype that is highly associated with autoimmune diseases
with frequencies of 0.13 in cases versus 0.19 in N. European controls.
The risk alleles of rs13151961, rs13119723, rs6822844 and rs6840978 associated with PS and PSA in the current study are similar to those reported for CeD 
. Overall, the risk alleles/haplotype of GD, CeD, PS and PSA appear to be the same, and of similar frequency. For example the frequency of the rs6822844*A allele is 0.14 in U.S. PS patients, ~0.16 in U.K. PSA patients, 0.14 in RA patients 
, ~0.13 in T1D patients from the Netherlands
, 0.12 in Celiac Disease patients from the Netherlands
, and ~0.14 in Irish CD patients
. This contrasts with frequencies of 0.19–0.20 in European control populations
. Although the frequency of the rs6822844*A allele was 0.14 in our PS cases and hence similar to frequencies seen in CeD, RA and T1D cases, its frequency in our combined cohort of U.S. controls was 0.16 (). This is lower than that reported for European controls. However, it has been previously reported that geographic variability exists at this locus across Europe 
. Our U.S. “European” controls are likely to be more disparate in origin, and are likely to account for our inability to obtain significant evidence for association with PS and 4q27.
To explore association of this region in our cohort of U.S. PS cases, without the possible confounding influence of subtle geographic variability at this locus in Europeans, we performed family based association tests in our 242 psoriasis nuclear families which are described elsewhere
. This approach provided evidence for replication of association of PS with rs6822844 and rs6840978 (PDT P
0.029 and 0.007 respectively). For these SNPs, the over-transmitted rs6822844*G and rs6840978*C alleles were also the risk allele from case/control studies. Haplotype studies in families also revealed association with the rs6822844*G/rs6840978*C haplotype (multipoint TDT P
0.006). Hence, our findings support chromosome 4q27 as harboring a variant/haplotype for PSA and PS.
As reported elsewhere the 4q27 locus that contains these associated SNPs corresponds to two closely correlated ~439 kb and ~40 kb haplotype blocks 
. This extensive LD makes it very difficult to determine the predisposing variant. Chromosome 4q is also the location of PSORS3, which is generally placed slightly more distally 
. However, the locus identified here may contribute in part to the previous observations of linkage. The long region of LD at chromosome 4q27 contains several genes 
: Testis nuclear RNA-binding protein (TENR), a gene encoding a protein of unknown function (KIAA1109), and genes encoding the interleukin-2 (IL2) and interleukin-21 (IL21) cytokines. TENR is expressed primarily in testis and KIAA1109 transcripts are ubiquitous, hence their roles in autoimmunity are not particularly compelling. However, IL2 and IL21 are of particular interest with respect to PS. IL2 is considered to be a pathogenic cytokine for PS
, and blockade of the IL2 receptor with therapeutic antibodies has induced disease resolution in some cases
. IL2 is a survival factor for T cells and promotes the differentiation of cytotoxic T-lymphocytes and NK cells. Both of these cell types are present in psoriasis lesions. Moreover, many IL2 receptor (IL2R) positive T-cells that fit the phenotypic definition of regulatory T cells (Tregs
) are also present in psoriasis lesions. IL-2 may influence how a common precursor T-cell differentiates into either a Treg
or a Th
17 T-cell, since addition of IL-2 has been shown to suppress the differentiation of Th
17 T-cells in mice
. IL-2 antibodies stain normal epidermis and psoriatic epidermis, with generally lower staining in the dermis (). The pattern of staining appears to be to dendritic cells (DCs) which are likely to be epidermal Langerhans cells. This pattern of IL-2 staining is probably due to DC activation and upregulation of IL2R. Cells with IL2 receptors include T-cells, B-cells, NK-cells, and dendritic cells.
The epidermal staining for IL-21 is much lower than for IL-2 () and appears to be mainly on dendritic cells in the superficial dermis. IL-21 is a product of activated T cells (under conditions of Th
17 polarization). It then acts in an autocrine or paracrine fashion on T-cells to up-regulate expression of the IL23 receptor which has already been implicated in psoriasis pathogenesis. IL23R sensitizes cells to IL-23 which stimulates IL17 synthesis and/or prolongs the survival of Th
17 cells 
. Blocking IL21 reduces the progression of lupus erythematosus 
which is one of the autoimmune diseases that is now being considered as a “Th
17” mediated disease. Therefore IL21 may play a role in Th
17 formation in this and other autoimmune diseases where these cells are pathogenic.
Extensive resequencing of IL2 coding and flanking regions has already been performed in T1D samples and no coding or obvious regulatory/splice variants were identified 
. As stated previously, this region needs to be resequenced thoroughly followed by comprehensive genotyping in larger numbers of samples to identify the autoimmune associated variant/s 
The observed associations in the current study are of interest for several reasons. It is noteworthy that the strongest association is with the MHC. Even in PSA, where associations are reportedly less than with PS (without PSA), associations with the class I region appear to be more significant than with any other region. We were also able to replicate previously reported associations with IL12B and IL23R and detected a potentially novel association upstream from IL12RB2. Novel associations within COG6 and the region on chromosome 15q21 harboring USP8 and SPPL2A are of interest. These and other regions reported here are worthy of follow up in other cohorts. Moreover, the association with chromosome 4q27 provides further evidence that this region is a common locus for multiple forms of autoimmune disease.
A recent study reported the IL13/IL4 region from chromosome 5q31 as being associated with PS
. Overall, the risk contributed by the MHC, the IL13/IL4 region and the IL23R and IL12B variants was estimated to be 3.83. With the COG6 and chromosome 15q21 loci described here, the risk would be increased. However, PS is a complex disease, and overlapping subsets of risk factors may be sufficient for susceptibility, so that risk effects cannot be computed in an additive manner.
The ability to identify low risk variants for common diseases such as PS and PSA will be limited by the cohort size, and larger numbers of cases and controls will be necessary to identify the majority of genetic factors for these diseases. Moreover, some of the SNPs with borderline discovery P values are also likely to be genetic risk factors for disease. It is worth noting that our discovery P value for the associated IL23R R381Q SNP did not reach significance (P
0.057, adjusted P
0.081) although allele frequencies in the discovery dataset revealed a M.A.F. of 0.041 in cases and 0.065 in controls, which is similar to what has been reported in other studies
. Additional genome-wide association scans and replication studies are required to identify additional variants and to confirm some of those found in the current study. Such studies include a genome-wide scan for psoriasis variants from the Genome Wide Association Network (GAIN) consortium 
. Genes such as these are important for determining the pathogenesis of PS and PSA and in identifying novel drug targets for these inflammatory diseases of the skin and joints.