Article Text

Download PDFPDF

Genetic variation at the glycosaminoglycan metabolism pathway contributes to the risk of psoriatic arthritis but not psoriasis
  1. Adrià Aterido1,2,
  2. Juan D Cañete3,
  3. Jesús Tornero4,
  4. Carlos Ferrándiz5,
  5. José Antonio Pinto6,
  6. Jordi Gratacós7,
  7. Rubén Queiró8,
  8. Carlos Montilla9,
  9. Juan Carlos Torre-Alonso10,
  10. José J Pérez-Venegas11,
  11. Antonio Fernández Nebro12,
  12. Santiago Muñoz-Fernández13,
  13. Carlos M González14,
  14. Daniel Roig15,
  15. Pedro Zarco16,
  16. Alba Erra17,
  17. Jesús Rodríguez18,
  18. Santos Castañeda19,
  19. Esteban Rubio20,
  20. Georgina Salvador21,
  21. Cesar Díaz-Torné22,
  22. Ricardo Blanco23,
  23. Alfredo Willisch Domínguez24,
  24. José Antonio Mosquera25,
  25. Paloma Vela26,
  26. Simon Angel Sánchez-Fernández27,
  27. Héctor Corominas22,28,
  28. Julio Ramírez3,
  29. Pablo de la Cueva29,
  30. Eduardo Fonseca30,
  31. Emilia Fernández31,
  32. Lluis Puig32,
  33. Esteban Dauden33,
  34. José Luís Sánchez-Carazo34,
  35. José Luís López-Estebaranz35,
  36. David Moreno36,
  37. Francisco Vanaclocha37,
  38. Enrique Herrera38,
  39. Francisco Blanco39,
  40. Benjamín Fernández‐Gutiérrez40,
  41. Antonio González41,
  42. Carolina Pérez-García42,
  43. Mercedes Alperi‐López8,
  44. Alejandro Olivé Marques43,
  45. Víctor Martínez‐Taboada23,
  46. Isidoro González-Álvaro19,
  47. Raimon Sanmartí3,
  48. Carlos Tomás Roura44,
  49. Andrés C García-Montero45,
  50. Sílvia Bonàs-Guarch46,
  51. Josep Maria Mercader46,
  52. David Torrents46,47,
  53. Laia Codó48,
  54. Josep Lluís Gelpí48,
  55. Mireia López-Corbeto1,
  56. Andrea Pluma1,
  57. Maria López-Lasanta1,
  58. Raül Tortosa1,
  59. Nuria Palau1,
  60. Devin Absher49,
  61. Richard Myers49,
  62. Sara Marsal1,
  63. Antonio Julià1
  1. 1 Rheumatology Research Group, Vall d'Hebron Research Institute, Barcelona, Spain
  2. 2 Department of Experimental and Health Sciences, Universitat Pompeu Fabra, Barcelona, Spain
  3. 3 Rheumatology Department, Hospital Clínic de Barcelona and IDIBAPS, Barcelona, Spain
  4. 4 Rheumatology Department, Hospital Universitario Guadalajara, Guadalajara, Spain
  5. 5 Dermatology Department, Hospital Universitari Germans Trias i Pujol, Badalona, Spain
  6. 6 Rheumatology Department, Complejo Hospitalario Juan Canalejo, A Coruña, Spain
  7. 7 Rheumatology Department, Hospital Parc Taulí, Sabadell, Spain
  8. 8 Rheumatology Department, Hospital Universitario Central de Asturias, Oviedo, Spain
  9. 9 Rheumatology Department, Hospital Virgen de la Vega, Salamanca, Spain
  10. 10 Rheumatology Department, Hospital Monte Naranco, Oviedo, Spain
  11. 11 Rheumatology Department, Hospital de Jerez de la Frontera, Cádiz, Spain
  12. 12 Rheumatology Department, Instituto de Investigación Biomédica de Málaga, Hospital Regional Universitario de Málaga, Málaga, Spain
  13. 13 Rheumatology Department, Hospital Universitario Infanta Sofía, Universidad Europea, Madrid, Spain
  14. 14 Rheumatology Department, Hospital Universitario Gregorio Marañón, Madrid, Spain
  15. 15 Rheumatology Department, Hospital Moisès Broggi, Barcelona, Spain
  16. 16 Rheumatology Department, Hospital Universitario Fundación Alcorcón, Madrid, Spain
  17. 17 Rheumatology Department, Hospital Sant Rafael, Barcelona, Spain
  18. 18 Rheumatology Department, Hospital Universitari de Bellvitge, Barcelona, Spain
  19. 19 Rheumatology Department, Hospital Universitario La Princesa, IIS La Princesa, Madrid, Spain
  20. 20 Rheumatology Department, Centro de Salud Virgen de los Reyes, Sevilla, Spain
  21. 21 Rheumatology Department, Hospital Universitario Mútua de Terrassa, Terrassa, Spain
  22. 22 Rheumatology Department, Hospital de la Santa Creu i Sant Pau, Barcelona, Spain
  23. 23 Rheumatology Department, Hospital Universitario Marqués de Valdecilla, Santander, Spain
  24. 24 Rheumatology Department, Complexo Hospitalario de Ourense, Ourense, Spain
  25. 25 Rheumatology Department, Complejo Hospitalario Hospital Provincial de Pontevedra, Pontevedra, Spain
  26. 26 Rheumatology Department, Hospital General Universitario de Alicante, Alicante, Spain
  27. 27 Rheumatology Department, Hospital La Mancha Centro, Alcázar de San Juan, Spain
  28. 28 Rheumatology Department, Hospital Dos de Maig, Barcelona, Spain
  29. 29 Dermatology Department, Hospital Universitario Infanta Leonor, Madrid, Spain
  30. 30 Dermatology Department, Complejo Hospitalario Universitario de A Coruña, A Coruña, Spain
  31. 31 Dermatology Department, Hospital Universitario de Salamanca, Salamanca, Spain
  32. 32 Dermatology Department, Hospital de la Santa Creu i Sant Pau, Barcelona, Spain
  33. 33 Dermatology Department, Hospital Universitario La Princesa, IIS La Princesa, Madrid, Spain
  34. 34 Dermatology Department, Hospital General Universitario de Valencia, Valencia, Spain
  35. 35 Dermatology Department, Hospital Universitario Fundación Alcorcón, Madrid, Spain
  36. 36 Dermatology Department, Hospital Virgen Macarena, Sevilla, Spain
  37. 37 Dermatology Department, Hospital Universitario 12 de Octubre, Madrid, Spain
  38. 38 Dermatology Department, Hospital Universitario Virgen de la Victoria, Málaga, Spain
  39. 39 Rheumatology Department, INIBIC-Hospital Universitario A Coruña, A Coruña, Spain
  40. 40 Rheumatology Department, Hospital Clínico San Carlos, IDISSC, Madrid, Spain
  41. 41 Instituto de Investigación Sanitaria Hospital Clínico Universitario de Santiago, Santiago de Compostela, Spain
  42. 42 Rheumatology Department, Parc de Salut Mar Barcelona, Barcelona, Spain
  43. 43 Rheumatology Department, Hospital Universitari Germans Trias i Pujol, Barcelona, Spain
  44. 44 Rheumatology Department, Hospital Comarcal d’Amposta, Tarragona, Spain
  45. 45 Banco Nacional de ADN Carlos III, University of Salamanca, Salamanca, Spain
  46. 46 Barcelona Supercomputing Centre (BSC), Joint BSC-CRG-IRB Research Program in Computational Biology, Barcelona, Spain
  47. 47 Institució Catalana de Recerca i Estudis Avançats (ICREA), Barcelona, Spain
  48. 48 Life Sciences Department, Barcelona Supercomputing Centre, Barcelona, Spain
  49. 49 HudsonAlpha Institute for Biotechnology, Huntsville, Alabama, USA
  1. Correspondence to Antonio Julià, Rheumatology Research Group, Vall d’Hebron Research Institute, Barcelona 08035, Spain; toni.julia{at}; Sara Marsal, Rheumatology Research Group, Vall d’Hebron Research Institute, Barcelona 08035, Spain; sara.marsal{at}; Juan D Cañete, Rheumatology Department, Hospital Clínic de Barcelona and IDIBAPS, Barcelona 08036, Spain; jcanete{at}


Objective Psoriatic arthritis (PsA) is a chronic inflammatory arthritis affecting up to 30% of patients with psoriasis (Ps). To date, most of the known risk loci for PsA are shared with Ps, and identifying disease-specific variation has proven very challenging. The objective of the present study was to identify genetic variation specific for PsA.

Methods We performed a genome-wide association study in a cohort of 835 patients with PsA and 1558 controls from Spain. Genetic association was tested at the single marker level and at the pathway level. Meta-analysis was performed with a case–control cohort of 2847 individuals from North America. To confirm the specificity of the genetic associations with PsA, we tested the associated variation using a purely cutaneous psoriasis cohort (PsC, n=614) and a rheumatoid arthritis cohort (RA, n=1191). Using network and drug-repurposing analyses, we further investigated the potential of the PsA-specific associations to guide the development of new drugs in PsA.

Results We identified a new PsA risk single-nucleotide polymorphism at B3GNT2 locus (p=1.10e-08). At the pathway level, we found 14 genetic pathways significantly associated with PsA (pFDR<0.05). From these, the glycosaminoglycan (GAG) metabolism pathway was confirmed to be disease-specific after comparing the PsA cohort with the cohorts of patients with PsC and RA. Finally, we identified candidate drug targets in the GAG metabolism pathway as well as new PsA indications for approved drugs.

Conclusion These findings provide insights into the biological mechanisms that are specific for PsA and could contribute to develop more effective therapies.

  • psoriatic arthritis
  • genetics
  • genome-wide association study
  • glycosaminoglycan
  • drug repurposing

Statistics from

Request Permissions

If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.

Key messages

What is already known about this subject?

  • Psoriatic arthritis (PsA) has a higher heritability than psoriasis, indicating the presence of a specific genetic risk component.

  • So far, very little genetic variation has been specifically associated with the development of PsA.

What does this study add?

  • Using a pathway-based genome-wide association study on two case–control cohorts, we have identified multiple pathways associated with PsA risk.

  • The glycosaminoglycan (GAG) metabolism pathway is specifically associated with PsA risk and is not associated with purely cutaneous psoriasis or rheumatoid arthritis.

  • Network-based analysis of the GAG pathway suggests two new drug candidates for PsA.

How might this impact on clinical practice or future developments?

  • The integration of biological knowledge on to genetic analysis can improve the ability to identify more effective therapies for PsA.


Psoriatic arthritis (PsA) is an inflammatory arthritis affecting up to 0.5% of the population and ~30% of patients with psoriasis (Ps).1 2 Compared with patients having skin-only affectation (ie, purely cutaneous psoriasis, PsC), patients with PsA have a substantially worse quality of life.3 Most of the drugs currently used to treat Ps are also indicated for PsA,4 but the efficacy can differ significantly. Therefore, there is a need to better understand the biological mechanisms underlying PsA in order to develop more effective therapies.

Familiar aggregation studies have demonstrated that PsA has a larger sibling recurrence rate (λs) than Ps (PsA λs~37 vs Ps λs~7),5–8 indicating the presence of a specific genetic risk. To date, more than 15 genome-wide association studies (GWAS) have been performed in Ps,9–14 identifying more than 50 susceptibility loci. Conversely, only a few GWAS have been performed exclusively in PsA. These studies have allowed the identification of 15 PsA risk loci.15–22 However, most of these risk variants are shared with Ps, indicating that the biological mechanisms that cause autoimmunity to the skin are also central for PsA. Identifying disease-specific loci has proven elusive, and to date only PTPN22, CSF2-P4HA2 and ADAMTS9-MAGI1 have shown a significant association with PsA. Taking into account the effects of all known risk loci, less than 50% of the heritability for PsA is currently explained.11 13 23 Therefore, new biological mechanisms could still be discovered that are relevant for disease aetiology, resulting in more effective therapies than the present ones.

A major challenge in the genetics of complex diseases is the identification of genes with small effects.24 To overcome this problem, the predominant approach has been to recruit increasingly larger patient cohorts.25 While this can help to identify new risk variation, this is extremely costly and time-consuming. Most of these GWAS have been performed at the single marker level, and therefore the statistical power to detect new risk variation soon becomes insufficient. To address this issue, different strategies have been developed. One of the most successful approaches has been to leverage the biological information underlying DNA variation like biological pathway annotation.26 Genome-wide pathway analysis (GWPA) efficiently integrates the risk variation from multiple, functionally related genes into a unique statistic.27 Additionally, using well-curated biological information in GWPA significantly accelerates the translation of the genetic association results.28 With this strategy, new genetic variation has been identified in different common diseases, including autoimmune diseases like Ps.29 30

To identify new genetic variation specifically associated with PsA, we have performed a GWAS at the single marker and pathway levels. We genotyped 835 patients with PsA and 1558 controls from Spain and performed a meta-analysis with a previous GWAS of PsA consisting of 1430 cases and 1417 controls from North America. Using this approach, we identified a new association at B3GNT2 locus and 14 genetic pathways associated with PsA risk. We next tested these genetic associations in GWAS cohorts of patients with PsC and rheumatoid arthritis (RA), and we found the glycosaminoglycan (GAG) metabolism pathway to be specific for PsA. Based on this evidence, we propose the GAG metabolism as a new source for drug discovery in PsA. Using network analysis and knowledge on drug action, we find evidence that the GAG pathway could be a useful target to treat PsA. These findings confirm the utility of GWAS to identify specific biological mechanisms and suggest repositioning of existing drugs for PsA.


Study population

Patients with PsA were selected from the rheumatology departments of 16 Spanish hospitals belonging to the Immune-Mediated Inflammatory Disease Consortium.31 All patients with PsA were diagnosed according to the Classification Criteria for Psoriatic Arthritis.32 Controls were recruited from healthy blood donors from Spanish hospitals in collaboration with the Spanish DNA Bank. A case–control cohort of 835 patients with PsA and 1588 controls were finally recruited and used for GWAS.

Meta-analysis was performed with a previous GWAS performed on 1430 patients with PsA and 1417 controls collected from USA and Canada.15 To identify PsA-specific variation, we used GWAS data from a cohort of 614 patients with PsC and 1191 patients with RA from Spain. Patients with PsC were defined as patients with plaque-type Ps for >10 years and free of any inflammatory disease in the joints. The main features of these cohorts are shown in online supplementary material 1 and online supplementary table S1.

Genome-wide genotyping and imputation

GWAS genotyping of the 2393 individuals from Spain was performed using the Illumina Quad610 array (Illumina, USA). Genotype calling and quality control (QC) were performed using GenomeStudio V.2011.1 (Illumina) and PLINK software, respectively (online supplementary material 1 and online supplementary figure S1). After QC, 506 926 single-nucleotide polymorphisms (SNPs) from 744 patients with PsA and 1454 controls were available for analysis.

We conducted genotype imputation to facilitate meta-analysis with the North America GWAS data. Only high-quality and directly genotyped SNPs (n=506 926 SNPs) were used for this analysis. After prephasing the haplotypes of each loci using SHAPEIT V.2-644 software, imputation was conducted with the IMPUTE V.2 software.33 We used the phase 1 release of the 1000 Genomes Project V.3 as reference panel.34 Only SNPs showing an minor allele frequency (MAF) >0.05 and an imputation quality >0.8 were selected for the GWPA. After filtering, 1 387 382 variants were available for analysis.

GWAS genotyping of the independent PsA case–control cohort was performed using the Illumina HumanOmni1-Quad array (Illumina) as previously described.15 After QC, 791 217 SNPs from 1430 patients with PsA and 1417 controls were used for the single marker and pathway meta-analyses.

The disease specificity of the validated loci and pathways was tested using GWAS data from two cohorts of patients with PsC and RA. These data sets were generated using the Illumina Quad610 array (Illumina) as previously described.29 35

In order to investigate the existence of PsA-specific variation across the human leukocyte antigen (HLA) region, we performed imputation of the classical alleles and amino acid polymorphisms from the HLA class I (HLA-A, HLA-B, HLA-C) and class II (HLA- DPA1, H LA -DPB1, H LA - DQA1, HLA - DQB1 and HLA -DRB1) loci in the PsA and PsC cohorts from Spain. HLA imputation was conducted using the SNP2HLA V.1.0.3 software.36

Association analysis of single genetic markers

The SNPs previously associated with Ps risk (p<5e-08; online supplementary table S2) were tested for association using a logistic regression model. The same analytical procedure was followed to analyse the association between whole genome variation and PsA risk in the Spain cohort using the SNPTEST V.2 software.37

The imputed HLA alleles and amino acid polymorphisms from the HLA class I and class II loci were tested for association using a stepwise logistic regression model. In this analysis, the most strongly associated marker was included as a covariate by addition to the null model until no markers reached the significance threshold determined by the false discovery rate (FDR) method. Since HLA haplotypes HLA-C*06:02 and HLA-B-27 have been previously shown to be differentially associated in PsA compared with PsC,38 39 the stepwise association analysis was started conditioning on these two established disease risk markers.

Genome-wide meta-analysis in PsA

The two independent PsA case–control cohorts were genotyped using different Illumina arrays (n Quad610 -QC =1 387 382 SNPs; n HumanOmni1 -QC =791 217 SNPs). Before GWAS meta-analysis, we identified the genetic variants that were commonly genotyped in both cohorts (n=720 582 SNPs; online supplementary table S3). We subsequently performed the GWAS meta-analysis using the METAL software.40 The association statistics were weighted by the sample size of the two cohorts and adjusted for the genomic inflation factor (λSpain Spain 1.08, λNorthAmerica North America 1.26).

Genome-wide pathway analysis

A total of 1077 reference pathways from the Molecular Signatures Database were included in the study. The SNP-gene mapping was performed using proximity-based criteria, which is the predominant approach in GWPA.26 41 According to reference studies in GWPA,26 28 42 we used an SNP-gene distance window of 20 Kb. The National Center for Biotechnology Information (NCBI) RefSeq Database Release 63 was used for SNP annotation (online supplementary table S4). Given the high linkage disequilibrium and gene density of the HLA region, the proximity-based criteria could yield false-positive results for pathways including genes within this locus. Similar to previous studies, the HLA region genes were excluded from the GWPA (online supplementary material 1).29 43 The statistical association between genetic pathways and disease risk was analysed using the set-based method implemented in PLINK, as described in online supplementary material 1.29 30

Gene expression analysis of the GAG metabolism pathway

To investigate the specificity of the GAG metabolism pathway in PsA at the functional level, we used whole blood transcriptomic data obtained from a previous study on patients with PsA, patients with PsC and healthy controls (Gene Expression Omnibus data set: GSE61281).44 In this previous study, the whole genome expression profile was evaluated aiming to identify differentially expressed genes. Here, based on our previous evidence at the genetic level, we hypothesised that the GAG pathway as a whole would be differentially expressed in PsA. To do this, after QC and quantile normalisation of the gene expression data, we tested for differential expression of the GAG metabolism pathway genes (t-test, nominal significance p=0.05) between patients with PsA (n=20) and PsC (n=20), and between each disease and healthy controls (n=12). We then used the binomial test to assess whether the observed number of differentially expressed genes in the pathway is greater than expected by chance. Furthermore, we also studied the changes in the coexpression of the GAG metabolism pathway between diseases. For this objective, we calculated the intramodular connectivity (IMC) measure. IMC is a network measure that efficiently captures gene coexpression information and is computed as the average of the gene connectivity within the pathway genes. The IMC is implemented in the WGCNA software.45 Student’s t-test was used to compare the IMC values of the GAG metabolism pathway between diseases.

Exploratory drug-repurposing analysis of the GAG pathway

To investigate the GAG metabolism as a new source for drug discovery in PsA, we combined network and drug-repurposing analyses. First, we performed a network analysis on the GAG metabolism pathway to identify those genes that are central to the network and, therefore, more likely to be key for the pathway functionality. Second, we screened the drugs approved by the Food and Drug Administration (FDA) to identify drugs that target central genes in the GAG metabolism pathway. Third, we defined a topology-based measure to evaluate the functional impact of these drugs on the GAG metabolism. Fourth, we compared the topology-based measure between the GAG metabolism and the rest of human biological processes for each of the candidate drugs. The details of this exploratory analysis are described in online supplementary material 1 and online supplementary tables S5, S6, S7.


Replication of established Ps risk variation

We found that 17 out of the 77 SNPs previously associated with Ps risk were also associated with PsA susceptibility in the Spain cohort (p<0.05; table 1). The ‘DNA repair’ pathway, a gene set previously associated with Ps risk, was also associated with PsA in our cohort (p=0.01; online supplementary table S8).

Table 1

Established Ps risk variants associated with PsA susceptibility in the Spanish cohort

Identification of new genetic loci associated with PsA

In the GWAS meta-analysis, we identified five loci associated with PsA at the genome-wide scale (p<5e-08; table 2). From these, the B3GNT2 locus (rs10865331, p=1.10e-08) has not been previously associated with PsA risk (figure 1). The complete list of associated markers is shown in online supplementary table S9.

Figure 1

Regional association plot of the new genetic variant rs10865331 associated with psoriatic arthritis (PsA) risk. Each circle represents a genetic variant that is plotted according to its association with PsA risk in the negative logarithmic scale. Circles are coloured according to the linkage disequilibrium with the SNP rs10865331 (ie, violet circle). The blue line shows the recombination rate across the plotted region (data source: 1000 Genomes Project; build GRCh37/hg19). In the bottom line, the genes mapping to the PsA-associated locus are shown.

Table 2

Genetic variants associated with PsA risk at the genome-wide scale

In the association analysis of the HLA markers comparing patients with PsA and PsC, we confirmed the genome-wide significant association between the HLA-C*06:02 allele and PsA risk (p=6.96e-11). The amino acid residues HLA-B Leu95 and HLA-C Ala305 were found to be also strongly associated with the risk of developing PsA (p<5e-8). In the stepwise conditional analysis, only the amino acid residue HLA-A Ala77 remained significant (pFDR<0.05). The complete list of associated HLA markers is shown in online supplementary table S10 (p<0.05).

Identification of genetic pathways associated with PsA

In the Spain cohort, we identified 76 genetic pathways significantly associated with PsA risk (p FDR <0.05; online supplementary table S11). Fifty out of these pathways (65.8%) were found to include genes from the HLA region and/or the IL12B gene (online supplementary table S12), which are the strongest genetic risk loci for both Ps and PsA. To confirm that the observed pathway associations are not only due to the strong signals present at these loci, we retested the pathways after removing these regions (HLA region: 25.6–33.3 Mb in chromosome 6, n=4021 SNPs; IL12B: 158 741 791–158 757 481 bp in chromosome 5, n=81 SNPs). After excluding these regions, we found that 9 genetic pathways were still significantly associated with PsA, giving a total of 35 pathways for validation in the independent cohort. Using the North America case–control cohort, we replicated the association of 16 genetic pathways with PsA risk (45.7%, p FDR <0.05; table 3).

Table 3

Genetic pathways associated with PsA risk and validated in the replication stage

Biological pathways can share a varying number of genes, and therefore redundancy in pathway association can occur. To filter out highly redundant results, we computed the gene overlap between the PsA-associated pathways. We found a marked gene overlap (>95% shared genes) between the ‘Costimulation by the CD28 family’ (n=63 genes) and ‘CD28 costimulation’ pathways (n=32 genes), as well as between the ‘Metabolism of carbohydrates’ (n=247 genes) and ‘GAG metabolism’ pathways (n=111 genes; online supplementary figure S2). In these two cases, we selected the pathway most significantly associated with PsA for downstream analyses (ie, ‘Costimulation by the CD28 family’ and ‘GAG metabolism’ pathways).

GAG metabolism pathway is specifically associated with PsA

In order to test for disease-specific risk, we tested the new PsA locus and pathways in the PsC and RA cohorts. We found that B3GNT2 risk allele is significantly associated with PsA in both comparisons (p [PsAvsPsC]=0.029, OR[95% CI]=1.16 [1.02 to 1.36]; p [PsAvsRA]=2.41e-04, OR[95% CI]=1.26 [1.08 to 1.44]). When comparing each disease with the control cohort, we found a significant association between PsC and B3GNT2 (p=6.11e-3, OR[95% CI]=1.21 [1.05 to 1.38]) but not with RA (p=0.07, OR[95% CI]=1.11 [0.99 to 1.24]).

In the pathway analysis we found that the GAG metabolism pathway was significantly associated with PsA when compared with PsC (p=0.018; online supplementary table S13) and RA (p=0.0018; online supplementary table S13). Subsequent testing of these two autoimmune diseases against the control cohort showed no evidence of association (p=0.71 and p=0.58 for PsC and RA pathway analyses, respectively).

GAG metabolism is associated with PsA at the transcriptomic level

Using whole blood transcriptomic data from patients with PsA and PsC,46 we found that 14 out of the 111 genes included in the GAG metabolism pathway were differentially expressed between the two diseases (online supplementary table S14). This difference is higher than expected by chance (p Binomial ≤0.005; figure 2). When comparing each disease with the control group (online supplementary table S14), the number of differentially expressed genes was also found to be significant in PsA (p Binomial ≤0.005), but not in PsC (p Binomial =0.27).

Figure 2

GAG metabolism pathway genes are differentially expressed in patients with PsA and PsC. Network representation of the GAG metabolism pathway. Genes, represented as nodes, are connected by edges according to the evidence of functional association between their encoded proteins. Differentially (p<0.05) and non-differentially (p≥0.05) expressed genes are represented by rhombus and grey circles, respectively. Differentially expressed genes that are upregulated and downregulated in PsA are coloured in red and orange, respectively. Gene diameter is proportional to the significance of differential expression in the negative logarithmic scale. GAG, glycosaminoglycan; PsA, psoriatic arthritis.

In the coexpression analysis of the GAG metabolism pathway, we detected that the pathway genes have a higher coexpression in PsA (IMC=1.43) than in PsC (IMC=1.38). Of relevance, when comparing the coexpression of the pathway between the PsA cohort and the mixed cohort of patients with PsA and PsC (IMC=0.54), the coexpression of the pathway was found to significantly drop when both diseases are analysed together as a single disease entity (p=8.37e-10).

GAG metabolism is a new source for drug discovery in PsA

In this exploratory analysis, we identified six FDA-approved drugs that target proteins encoded by GAG pathway genes (figure 3A). The analysis of the network topology of the GAG pathway showed that NCAN and VCAN genes, both targeted by the hyaluronic acid drug (HAd), and the DCN gene, targeted by the tromethamine drug (TMd), have the highest centrality properties (figure 3B). Since a higher network centrality is indicative of a predominant regulatory role in the pathway,47 we next evaluated the impact of HAd and TMd in the regulation of the GAG pathway. We found that both HAd and TMd significantly modulate the GAG pathway functionality (p HAd =0.0097 and p TMd =0.017), and therefore these two drugs could be repurposed as candidates for PsA treatment.

Figure 3

Drug-repurposing and network analyses in the GAG metabolism pathway. (A) Six FDA-approved drugs (amiloride, hyaluronic acid, N-acetyl-D-glucosamine, palifermin, sargramostim and tromethamine) have target genes in the GAG metabolism pathway. (B) Identification of the most central genes in the functional-based network of the GAG metabolism pathway. Gene diameter is proportional to its degree centrality value and is coloured according to its betweenness centrality value, ranging from white (lowest) to red (highest). FDA, Food and Drug Administration; GAG, glycosaminoglycan.


The identification of genetic variation that is associated with PsA and not with PsC has proven very challenging. Here, we have analysed two large PsA case–control cohorts from independent populations to identify disease-specific variation both at the single marker and at the pathway levels. Using this approach, we have identified a significant association of B3GNT2 locus as well as 14 genetic pathways with PsA risk. From these, we have found that genetic variation at the GAG metabolism pathway is specifically associated with PsA. Investigating the GAG metabolism pathway with drug-repurposing and network analyses, we have identified potentially new PsA indications for common drugs as well as new candidate drug targets for PsA.

The SNP rs10865331 associated with PsA risk is located on chromosome 2p15 at 99.6 Kb downstream of the B3GNT2 gene, which encodes for a transmembrane enzyme that synthesises the carbohydrate structure of polylactosamine onto glycoproteins.48 This polymorphism maps to a genomic region enriched in promoter and enhancer histone marks from blood, which has a marked immune cell burden.49 According to the Genotype-Tissue Expression, there is evidence of strong cis-regulation between this SNP and B3GNT2 expression in whole blood (p=5.1e-13; online supplementary figure S3).50 Previous studies in other arthritic diseases have shown that the SNP rs10865331 is associated with ankylosing spondylitis and the B3GNT2 locus with RA.51 52 Consistently, B3gnt2 knockout mice have shown hyperactivation of T and B lymphocytes as well as enhanced macrophage activation.48 53 In a GWAS meta-analysis, B3GNT2 has also been associated with Ps risk.11 In this previous study, however, a stratified analysis with patients affected with PsA was not performed, and consequently it remains unclear the contribution of each disease to the observed association. Here we show, for the first time, that the SNP rs10865331 at B3GNT2 locus is associated with PsA at the genome-wide level and that the frequency of the risk allele is significantly higher in PsA than in PsC.

The identification of disease-specific genetic variation is highly useful to discover relevant pathogenic mechanisms in complex diseases.54 In PsA, the existence of disease-specific variation is supported by the evidence of a larger familial aggregation compared with Ps.5–8 Using the GWAS data from this study, we find that, while PsA and PsC show a significant genetic correlation (r2=0.73, SE=0.12, p=3.76e-9), the SNP-based heritability of PsA (46%, SE=12%) is significantly higher than PsC (34%, SE=2%, p<0.05) (online supplementary material 1). In line with familial aggregation studies, our findings also support the existence of PsA-specific genetic variation. To this regard, when directly comparing the PsA and PsC cohorts, we have not only replicated the association of the HLA-C*06:02 and HLA-B*27 haplotypes,38 39 but also we have identified a new association between the HLA-A Asp77 and PsA risk. At the pathway level, we have identified a specific association between PsA and the GAG metabolism pathway. GAGs are linear, negatively charged oligosaccharides that include hyaluronic acid (HA), chondroitin sulfate and keratan sulfate (KS).55 Of relevance, GAGs are crucial components of proteoglycans and the major component of the cartilage, which is the main target tissue of PsA inflammatory destruction.56 57 Our results suggest that cartilage degradation in PsA could derive from an altered GAG metabolism that is not perturbed in other arthritis like RA.

GAG metabolism has been shown to be altered in complex diseases including autoimmune diseases.58 59 In in vitro models, uncontrolled proteolysis of aggrecan (ie, cartilage-specific proteoglycan) in response to proinflammatory cytokines promotes cartilage damage in the articular joint.60 After aggrecan destruction, GAGs are released from the extracellular matrix (ECM) to the synovial fluid.61 Accordingly, the levels of GAGs like HA have been found increased in patients with PsA compared with control subjects, both in serum and synovial fluid.62 63 Previous experimental studies have demonstrated the existence of a GAG-mediated mechanism for cartilage destruction that is driven by the degradation of HA by chondrocytes and that is independent from aggrecanases.64 Consistent with this evidence, genetic variation at the GAG metabolism pathway could diminish the cartilage-specific biosynthesis of HA, and consequently reduce its availability for both aggrecan and cartilage formation in patients with PsA.

In this study we also validated the association between 13 genetic pathways and PsA risk. From these, JAK-STAT signalling, type 1 diabetes mellitus, costimulation by CD28 family, Th1/Th2 differentiation and ErbB signalling pathways had been previously associated with Ps risk.29 Crucial proinflammatory cytokines for the development of Ps like interleukin (IL)-12 and IL-23 rely on the JAK-STAT signalling pathway. Importantly, small molecule inhibitors of JAK proteins (eg, ruxolitinib and tofacitinib) have been recently proven successful for the treatment of the disease.65 Our results provide additional genetic evidence supporting the functional role of this group of biological pathways in the aetiology of Ps. The eight remaining pathways (ie, G alpha signalling, purine metabolism, KS biosynthesis, extracellular matrix, mTOR signalling, IL-7 signal transduction, Rac-1 cell motility signalling and Met signalling) were found to be only significantly associated with PsA. Recent studies have shown that the mTOR signalling pathway is responsible for inducing the proliferation of a synovial T cell subpopulation (ie, Th9 cells) that enhances the immune response in PsA.66 There is also previous evidence supporting the implication of the IL-7 signal transduction pathway in the development of PsA. In in vitro models, lymphocytes and synovial fluid fibroblasts have shown to produce IL-7 cytokine and promote the formation of osteoclasts,67 which are the main mediators of bone matrix degradation in PsA. Additional studies on independent PsA and PsC cohorts will be needed to confirm the PsA-specific nature of these additional pathway associations.

Current drug discovery research is shifting from targeting single genes towards the modulation of specific biological pathways.68 Here, we show that the GAG metabolism could be a druggable pathway for PsA treatment. Our analyses suggest that FDA-approved drugs HAd and TMd are good candidates for repurposing for PsA, since they target central genes in the GAG metabolism network and have a significant impact on its functionality. We, like others, show the power of genetics to identify potential new drug targets and opportunities for drug repurposing in autoimmune diseases.69 In all these studies, however, downstream validation of the in silico findings in adequate in vitro and in vivo studies is still an indispensable step. Therefore, future experimental and clinical studies will be necessary to corroborate the utility of these two new drug targets to treat PsA.

Compared with previous GWAS, the use of PsC and RA cohorts to differentiate the genetic pathways that are PsA-specific from those that are not is a distinctive strength of the present study. The pathway-based analysis methodology used here has limitations, nonetheless. One limitation is the SNP annotation to the genes within each pathway. SNPs that are located far from the genes or in other chromosomes and that could regulate gene expression through cis-expression Quantitative Trait Loci (eQTL) and trans-eQTL mechanisms were not included in the present GWPA. To our knowledge, there is yet no pathway-based method that integrates this information and that has been able to identify disease risk variation. One of the major problems for this approach is the context-dependent nature of many eQTLs. There is growing evidence that many eQTLs are cell type-dependent and also vary in relation to many contextual aspects like the level and type of stimulation.70–72 The integration of this regulatory information is therefore still a challenge for GWPA analysis methods. With the increasing regulatory information that is currently being derived from single-cell expression studies,73 a more profound ascertainment of the impact of SNP variation on gene expression levels will be obtained, and eventually more comprehensive GWPA methods will be developed.

In conclusion, we have identified variation at B3GNT2 locus and 14 pathways significantly associated with the risk of PsA. From these, the GAG metabolism pathway showed a specific association with PsA when contrasted to PsC and RA. Using network and drug-repurposing analyses, we provide evidence that the GAG pathway could be a new source for drug discovery in PsA. This study represents an important step towards the characterisation of biological mechanisms that are specific for PsA and the finding of more effective drugs in PsA treatment.


We thank the patients and clinical specialists collaborating in the IMID Consortium for participation. Genotype and phenotype data from the North American PsA case–control cohort were provided by Dr JT Elder, University of Michigan, with collaborators Dr D Gladman, University of Toronto, and Dr P Rahman, Memorial University of Newfoundland, providing samples.


  1. 1.
  2. 2.
  3. 3.
  4. 4.
  5. 5.
  6. 6.
  7. 7.
  8. 8.
  9. 9.
  10. 10.
  11. 11.
  12. 12.
  13. 13.
  14. 14.
  15. 15.
  16. 16.
  17. 17.
  18. 18.
  19. 19.
  20. 20.
  21. 21.
  22. 22.
  23. 23.
  24. 24.
  25. 25.
  26. 26.
  27. 27.
  28. 28.
  29. 29.
  30. 30.
  31. 31.
  32. 32.
  33. 33.
  34. 34.
  35. 35.
  36. 36.
  37. 37.
  38. 38.
  39. 39.
  40. 40.
  41. 41.
  42. 42.
  43. 43.
  44. 44.
  45. 45.
  46. 46.
  47. 47.
  48. 48.
  49. 49.
  50. 50.
  51. 51.
  52. 52.
  53. 53.
  54. 54.
  55. 55.
  56. 56.
  57. 57.
  58. 58.
  59. 59.
  60. 60.
  61. 61.
  62. 62.
  63. 63.
  64. 64.
  65. 65.
  66. 66.
  67. 67.
  68. 68.
  69. 69.
  70. 70.
  71. 71.
  72. 72.
  73. 73.


  • Handling editor Josef S Smolen

  • Contributors AA, JDC, SM and AJ conducted the study design and data interpretation. JT, CF, JAP, JG, RQ, CM, JCT-A, JJP-V, AFN, SM-F, CMG, DR, PZ, AE, JR, SC, ER, GS, CD-T, RB, AWD, JAM, PV, SAS-F, HC, JR, PdlC, EdF, EmF, LP, ED, JLS-C, JLL-E, DM, FV, EH, FB, BF-G, AG, CP-G, MA-L, AOM, VM-T, IG-A, RS, CTR, ML-C, AP, ML-L, RT, NP and SM were involved in sample collection. AA, AJ, SB-G, JMM, DT, LC, HLG, DA and RM performed genetic analyses. AA and AJ performed functional and drug-repurposing analyses. AA, JDC, SM and AJ wrote the manuscript. All authors revised the manuscript and gave final approval for its submission.

  • Funding This study was funded by the Spanish Ministry of Economy and Competitiveness (grant numbers: PSE-010000-2006-6 and IPT-010000-2010-36, cofunded by the European Regional Development Fund). This work was also sponsored by the 'Agència de Gestió d’Ajuts Universitaris i de Recerca' (AGAUR, FI-DGR2016, grant number: 00587), which is supported by the 'Secretaria d’Universitats i Recerca' (Economy and Knowledge Department, Generalitat de Catalunya) and cofunded by the European Social Fund. The obtention of the GWAS data from the PsA case-control cohort from North American population was supported by grants from the NIH, the Canadian Institutes of Health Research, the Krembil Foundation, the Babcock Memorial Trust, the Barbara and Neal Henschel Charitable Foundation and the Ann Arbor Veterans Affairs Hospital. The study sponsors had no role in the collection, analysis or interpretation of the data

  • Competing interests None declared.

  • Patient consent Obtained.

  • Ethics approval The study was approved by Hospital Universitari Vall d'Hebron Clinical Research Ethics Committee. This study was conducted according to the principles of the Declaration of Helsinki. Protocols were reviewed and approved by the local institutional review board of each participating centre.

  • Provenance and peer review Not commissioned; externally peer reviewed.