HLA-DR risk variants in rheumatoid arthritis: what we know and still do not know

Abstract

Rheumatoid arthritis (RA) susceptibility is strongly influenced by genetic variations within the major histocompatibility complex, with HLA-DRB1 risk alleles representing the largest contributors to heritability, particularly in seropositive (anti-citrullinated protein antibody positive) disease. The shared epitope (SE) hypothesis, first proposed more than three decades ago, explained the clustering of susceptibility within HLA-DRB1*04 subtypes. Subsequent fine-mapping expanded HLA-DRB1 disease associations beyond the SE motif, identifying key amino acid residues at positions 11/13, in addition to 71 and 74, that account for much of the genetic contribution to susceptibility in seropositive RA. Recent studies demonstrate that these residues not only determine risk of developing seropositive RA but also stratify disease outcome (ie, radiographic progression or mortality), with valine at position 11 or histidine at position 13 as central determinants. However, their role in response to treatment remains unclear. Structural work has partially elucidated the molecular basis of these associations, showing that SE-bearing HLA-DR molecules preferentially bind nonpolar or negatively charged amino acids and directly engage T cell receptors, while genetic data support a role in shaping thymic repertoire selection. However, only a few RA autoantigens have been identified so far. Beyond professional antigen-presenting cells, HLA-DR expression is found on fibroblasts in the inflamed synovium and activated lymphocytes, where its function remains unknown. Importantly, HLA-DRB1 association to RA is not unique; HLA-DRB1 alleles contribute pleiotropically to the susceptibility of a wide range of autoimmune diseases, reflecting fundamental roles in immune regulation. This review synthesises established findings and current uncertainties in HLA-DRB1 research, identifying outstanding challenges and opportunities for translation into precision medicine.

What is already known on this topic

• HLA-DRB1 remains the strongest and most reproducible genetic association with rheumatoid arthritis (RA), particularly in seropositive (anti-citrullinated protein antibody positive) disease, with risk concentrated in alleles carrying the shared epitope motif.

What this study adds

• This review summarises recent developments on amino acid-level fine-mapping, which pinpoint specific residues in the DRB1 binding pocket, particularly 11/13, 71 and 74, refining classical allele associations.

• We attempt to put these genetic association results in their structural and functional context. Studies demonstrated that high-risk HLA-DRB1 variants encode a positively charged P4 peptide binding pocket preferentially accommodating non-positive amino acid side chains, with antigen-presentation influencing T cell selection and activation.

How this study might affect research, practice or policy

• We highlight areas of unmet needs for future research in RA and beyond, as HLA-DRB1 risk architecture extends beyond RA, contributing to susceptibility in other autoimmune diseases such as multiple sclerosis and systemic lupus erythematosus.

History

Rheumatoid arthritis (RA) is a chronic systemic autoimmune disease with a strong genetic component, estimated to account for approximately 65% of disease heritability based on twin studies.¹ Immunological studies in the late 1960s showed that lymphocytes from patients with RA displayed altered reactivity in mixed lymphocyte cultures, failing to stimulate one another compared with healthy controls, suggesting an underlying immunogenetic abnormality.² In the following decade, the first HLA association with RA was reported when the B-cell alloantigen DRw4 (a serologically defined subset of HLA-DR4, or HLA-DRB1*04, according to the new nomenclature³) was shown to be markedly enriched among patients with RA.⁴ Linkage and family studies conducted in the next two decades firmly implicated the HLA-DR region of the major histocompatibility complex (MHC) in RA susceptibility, accounting for approximately one-third of the overall heritable risk.^{1 5 6}

Molecular studies advanced the field further. The HLA-DR molecule is expressed on the cell surface as a heterodimer comprising an alpha chain (HLA-DRα, encoded by the HLA-DRA gene) and a beta chain (HLA-DRβ, encoded by HLA-DRB1, HLA-DRB3, HLA-DRB4 or HLA-DRB5).⁷ The idea that shared amino acid motifs within HLA class II molecules could influence disease susceptibility had been articulated earlier by Silver and Goyert.⁸ This concept was subsequently formalised in the context of RA by Gregersen et al through the shared epitope (SE) hypothesis, whereby RA risk is conferred not by whole HLA alleles but by a conserved amino acid sequence motif, called the SE, at positions 70–74 in the third hypervariable region of the HLA-DRβ1 chain⁹ (QKRAA, QRRAA and RRRAA, with Q for glutamine, K for lysine, R for arginine and A for alanine¹⁰). At that time, DR4 was subdivided serologically into Dw subtypes, and it soon became clear that only some of these, such as Dw4 and Dw14, carried the SE and conferred strong risk, while others like Dw10 did not.¹¹ These findings underscored the functional importance of specific amino acid differences within DRB1.¹²

The subsequent introduction of DNA-based typing and the adoption of the WHO four-digit allele nomenclature in the early 1990s allowed these Dw specificities to be precisely mapped to individual DRB1 alleles (eg, Dw4 ≈ DRB1*04:01, Dw10 ≈ DRB1*04:02).¹³ This higher resolution confirmed that susceptibility was driven by particular four-digit DRB1 alleles encoding the SE motif, while others within the same serotype group were neutral, providing a unifying molecular explanation for earlier serological observations.

Despite decades of progress, important gaps remain. Not all individuals carrying HLA-DR risk variants develop RA, indicating incomplete penetrance and gene–environment interplay.¹ Moreover, the strength and nature of HLA associations vary across populations, with DRB1*01 and other alleles contributing in some ancestries.⁹ Thus, while HLA-DRB1 risk variants explain a substantial proportion of genetic susceptibility, they are neither necessary nor sufficient for disease development. Subsequent studies demonstrated that the association between HLA-DRB1 SE alleles and RA is largely driven by the presence of anti-citrullinated protein antibodies (ACPA), defining seropositive RA, with a substantially weaker effect in seronegative RA.¹⁴

In this review, we summarise the history of HLA-DR research in RA, from early immunogenetic observations to molecular insights, and discuss what is well established and what remains uncertain in understanding HLA-DR risk variants in RA.

HLA-DRB1 across populations

High-resolution (four-digit) typing of the HLA-DRB1 locus has revealed remarkable allelic diversity across populations.³ Of the currently known DRB1 alleles,³ at least 53 carry the conserved amino acid sequence at positions 70–74 defining the SE.¹⁵ The SE association has been demonstrated repeatedly in anti-citrullinated protein/peptide antibody (ACPA)-positive disease and across diverse ancestral groups16,18 (table 1).

Table 1. 19 haplotypes of HLA-DRB1 formed by the amino acids at positions 11, 13, 71 and 74: comparison with the SE model.

Raychaudhuri’s classification			2010 classification of HLA-DRB1 alleles*	Gregersen-Silver-Winchester’s classification
HLA-DRB1 amino acid	Haplotype name	Risk stratification (OR)		SE positive	70–74 motif	Effect in RA
Position 11–13–71–74
Valine–phenylalanine–arginine–alanine	VFRA haplotype	4.65	*10:01	Yes	RRRAA	Risk
Valine–histidine–lysine–alanine	VHKA haplotype	4.03	*04:01, *04:09	Yes	QKRAA	Risk
Valine–histidine–arginine–alanine	VHRA haplotype	3.63	*04:04, *04:05, *04:08, *04:10	Yes	QRRAA	Risk, generally strong for *04, moderate for *01
Leucine–phenylalanine–arginine–alanine	LFRA haplotype	2.11	*01:01, *01:02
Aspartate–phenylalanine–arginine–glutamate	DFRE haplotype	1.82	*09:01	No	RRRAE
Proline–arginine–arginine–alanine	PRRA haplotype	1.58	*16:01, *16:02	No	DRRAA
Valine–histidine–arginine–glutamate	VHRE haplotype	1.29	*04:03, *04:06, *04:07, *04:11	No	QRRAE
Serine–glycine–arginine–alanine	SGRA haplotype	1.04	*12:01, *12:02	No	DRRAA
Valine–histidine–glutamate–alanine	VHEA haplotype	1.03	*04:02, *04:37	No	DERAA†
Proline–arginine–alanine–alanine	PRAA haplotype	1.00	*15:01, *15:02, *15:03	No	QARAA
Glycine–tyrosine–arginine–glutamine	GYRQ haplotype	0.92	*07:01	No	DRRGQ
Serine–serine–lysine–alanine	SSKA haplotype	0.87	*13:03	No	DKRAA
Serine–glycine–arginine–leucine	SGRL haplotype	0.83	*08:01, *08:02, *08:03, *08:04, *08:06, *14:15	No	DRAAL
Serine–serine–arginine–glutamate	SSRE haplotype	0.77	*14:01, *14:05, *14:07	No	RRRAE
Serine–serine–arginine–alanine	SSRA haplotype	0.76	*11:01, *11:04, *11:06, *11:08, *13:05, *14:06; *14:02	No/yes	DRRAA/QRRAA for *14:02	Risk for *14:02
Leucine–phenylalanine–glutamate–alanine	LFEA haplotype	0.71	*01:03	No	DERAA†
Serine–serine–lysine–arginine	SSKR haplotype	0.67	*03:01, *03:02, *03:04	No	QKRGR
Serine–serine–glutamate–alanine	SSEA haplotype	0.6	*11:02, *11:03, *13:01, *13:02, *13:04	No	DERAA†
Serine–glycine–arginine–glutamate	SGRE haplotype	0.49	*14:04	No	RRRAE

OR is the odds ratio for susceptibility to RA, where 1.00 (PRAA haplotype) is the reference, anything above 1.00 represents an increased risk and lower represents a decreased risk (=protective effect) of developing RA. This table is modified from Table S8 of Han et al.26 Shared Epitope motif’s effects across RA populations are also shown. The motifs are represented with the following amino acids: Q: glutamine; K: lysine; R: arginine; D: aspartic acid; E: glutamic acid; A: alanine.

Highlighted cells are discussed in the main text body.

^*2010 WHO nomenclature, see Marsh et al.3

^†DERAA is a protective motif according to Balsa et al.16

RA, rheumatoid arthritis; SE, shared epitope.

Validation has come from multiple global cohorts. In Spain, SE alleles were associated with increased risk and higher ACPA titres.¹⁶ In Israel, DRB1*01:01 and *10:01 were over-represented among patients with RA compared with controls, confirming the role of SE beyond European populations.¹⁹ Smaller scale studies in Latin America involving about 700 patients with RA further established that SE alleles are enriched in admixed populations.²⁰ These findings, together with evidence from European and Asian fine-mapping¹⁷ and replication in Koreans,¹⁸ demonstrate the consistency of the SE association worldwide.

Despite reproducibility, the strength of association (effect size) and the frequency of SE alleles vary between populations. In black South Africans, more than 90% of patients with RA carry at least one SE allele, while in Cameroonian patients with RA the frequency is closer to 30%.^{21 22} Among African Americans, SE enrichment is observed largely through European admixture.²³ In East Asians, SE alleles confer risk, but additional alleles such as DRB1*09:01 contribute independently, reflecting ancestry-specific architecture.¹⁷ Fine-mapping studies further showed that in these populations, amino acid position 13 and 57 of HLA-DRB1 also have strong and independent effects beyond the classical SE motifs, underscoring that risk in Asian cohorts cannot be explained by SE alone¹⁷ (table 1).

Protective effects have also been described. Alleles encoding the DERAA motif at positions 70–74, including DRB1*13:01 and *13:02, are associated with reduced risk and with lower ACPA titres.¹⁶ The frequency of these protective alleles differs across ancestries, which may help explain variation in disease prevalence and severity. These protective effects are inferred from statistical models, defined through relative risk comparisons against other HLA-DRB1 alleles, and their mechanistic basis remains unknown. If ‘protective’ alleles are ‘truly protective’ or result from the concomitant absence of SE alleles (non-predisposition), it has not been determined yet in RA, but examples of tolerogenic effects mediated by protective alleles have been described in other diseases. In Goodpasture disease, an immune-mediated disorder with a strong HLA association, mechanistic studies have demonstrated that protective HLA alleles present antigens and can actively enforce tolerance by skewing self-reactive CD4⁺ T cells towards a regulatory phenotype rather than pathogenic effector responses.²⁴

Taken together, these studies show that the SE has been robustly validated across ancestries, with genuine but heterogeneous effect across populations of Western ancestry, across Africa, Asia and the Americas, including Indigenous populations. However, other genetic predisposing factors within HLA-DRB1, but outside the SE, play important ancestry-specific roles in RA aetiology. Large-scale studies with dense genetic resolution in underrepresented populations are still needed to clarify the strength and spectrum of SE or other HLA associations globally.

Advances in genetic approaches

The turn of the century has witnessed major advances in genetic approaches that have transformed our understanding of HLA-DRB1 risk in RA. Early association studies relied on classical serotyping, which lacked resolution and was limited by the extensive linkage disequilibrium (LD) of the MHC. More recent methods such as high-density single-nucleotide polymorphisms genotyping, imputation of classical HLA alleles from genotyping array data and fine-mapping of amino acid polymorphisms have provided sharper insights into the specific residues conferring risk. Important breakthroughs came from imputation-based fine mapping.

Moving beyond the SE

Raychaudhuri and colleagues demonstrated that five amino acid positions across three HLA proteins (HLA-DRβ1, HLA-DPβ1 and HLA-B) almost completely explain the MHC association to seropositive RA, with positions 11/13, 71 and 74 of HLA-DRB1 representing the strongest associations. Positions 11 and 13 have been grouped together as they are in high LD (inherited together) in populations of European ancestry, making it hard to distinguish which one of these two positions is driving the association. The amino acids that confer risk to RA at these positions are valine and leucine for position 11; histidine and phenylalanine for position 13. These positions are of high interest as they are more strongly associated with seropositive RA susceptibility than any other amino acid position, despite being located outside the SE. The combination of amino acids at these 4 positions defines 16 haplotypes,²⁵ stratifying the general population into 16 different risk categories.

Han et al further refined risk signals to 19 haplotypes (Raychaudhuri’s classification—table 1 and figure 1) by increasing the total sample size of the study further and analysing both ACPA-positive and ACPA-negative RA (figure 2).²⁶ They also identified a sixth independent effect within the HLA (protective effect of asparagine at HLA-A position 77 with OR=0.85). They pinpointed two amino acid positions that independently accounted for risk in both ACPA-positive and ACPA-negative subsets, HLA-DRB1 position 11 and HLA-B position 9, each located within the peptide-binding groove (figure 3). Serine and leucine at HLA-DRB1 position 11 and aspartate at HLA-B position 9 have both OR around 1.3–1.4, highlighting the contribution of both class II and class I molecules to disease susceptibility.

Figure 1. The 19 haplotype groups defined by amino acid positions 11, 13, 71 and 74 of HLA-DRB1 and their effect on rheumatoid arthritis (RA) susceptibility/prevalence in populations of European ancestry. The size of the dot was determined by the case allele frequencies for each haplotype in populations of European ancestry. The y-axis shows the OR for susceptibility to RA for the four HLA-DRB1 amino acid motifs shown on the x-axis (positions 11, 13, 71 and 74). Amino acids on the x-axis, Q: glutamine; K: lysine; R: arginine; D: aspartic acid; E: glutamic acid; A: alanine. The data used for the plot were taken from Table S8 of Han et al.²⁶.

Figure 2. Distinct effect sizes of amino acid residues at HLA-DRB1 position 11 for ACPA-positive and ACPA-negative RA. Although the carriage of a valine increases the risk of developing seropositive RA, it decreases the risk of developing seronegative RA. The opposite relationship is observed for serine. However, the carriage of a glycine, proline or leucine influences the risk of seropositive and seronegative RA in the same direction. Reprinted from Han et al,²⁶ with permission from Elsevier. Licence Number 6137680993837. ACPA, anti-citrullinated protein antibodies; Asp, aspartate; Gly, glycine; Leu, leucine; Pro, proline; RA, rheumatoid arthritis; Ser, serine; Val, valine.

Figure 3. Three-dimensional structures of amino acid positions in HLA class I and class II molecules that affect susceptibility to seropositive and seronegative rheumatoid arthritis (RA). The side chains of susceptibility amino acids all point towards the inside of the peptide binding groove, implicating antigen-binding and presentation in the aetiology of RA. Used with permission of Annual Reviews, from ‘Recent advances in defining the genetic basis of rheumatoid arthritis’, Annual Review of Genomics and Human Genetics, Volume 17, Issue 1, 2016; permission conveyed through Copyright Clearance Center. Licence number 1667078-1.

Importantly, Raychaudhuri’s classification was established using an additive genetic model, explicitly capturing allelic dosage effects, whereby risk increases with the number of risk-conferring amino acids or haplotypes carried. The OR for the carriage of a single copy of each HLA-DRB1 allele is reported in table 1 (eg, 4.65 for HLA-DRB1*10:01), but these ORs must be combined to obtain the risk for homozygote or heterozygote patient groups. The total OR for a pair of haplotypes will be the product of the two haplotypic additive ORs. For example, the OR for the carriage of two copies of HLA-DRB1*10:01 will be 21.6 (4.65×4.65); it will be 3.1 for one copy of HLA-DRB1*11:01 and one copy of HLA-DRB1*04:01 (0.76×4.03). We explain below how to calculate the total OR for haplotype combinations with reported interactions. These considerations are important to compute polygenic risk scores (PRSs) for precision medicine.

Amino acid-level association analyses alone do not establish causality; in addition, Val11 and Phe13 are in strong LD with the SE in humans, which makes it harder to disentangle their independent effects on disease aetiology in functional experiments. Structural studies using X-ray crystallography of putative RA autoantigenic peptides in complex with SE-bearing HLA-DR molecules have allowed the visualisation of the biophysical interactions between SE amino acids, defining the P4 pocket of the HLA-DR molecule, and the autoantigenic peptide (see next section on mechanism). However, these studies have not investigated the structural impact of amino acids at position 13, also contributing to the P4 pocket, or the impact of amino acids at position 11, the only variable amino acid in the P6 pocket,²⁷ on peptide binding.

Nonetheless, insight into the possible independent effect of position 11/13 in humans has been provided from a non-human primate study, where the strong LD with the SE is absent: in macaques immunised with citrullinated peptides, Val11 and Phe13, rather than the 70–74 SE, were the main determinants of citrullinated peptide-specific T cell responses, with effects that differed from those inferred from human genetic risk.²⁸ Together, these findings support a hypothesis in which positions 11/13 contribute to antigen-specific T cell responses, integrating signals from human genetic association and non-human primate functional studies. However, additional structural and functional evidence on the independent effect of positions 11/13 is required in humans; this could be gained by comparing crystallographic structures and CD4⁺ T cell responses of DRB1*01 and DRB1*04 individuals sharing a common SE motif (QRRAA) but differing only in the amino acids at positions 11/13 (table 1).

Classification of HLA-DRB1 susceptibility alleles

Several different classification systems have been proposed in addition to the Gregersen-Silver-Winchester’s classification (the SE model), such as the Tezenas du Montcel, the de Vries or the Mattey classification systems and they have been reviewed elsewhere.²⁹ Amino acid-level studies highlighted distinct genetic architectures for seropositive and seronegative RA, consistent with different immunopathogenic mechanisms. They also resolved some of the controversy around the SE hypothesis by demonstrating that risk extends beyond the canonical motif and involves additional structural positions within the peptide-binding groove. While the Gregersen-Silver-Winchester’s classification stratifies the population into two risk categories (SE positive and SE negative), the classification by Raychaudhuri introduces 19 categories, stratifying the population on a risk continuum, ranging from lowest risk (protective effects) to highest risk, with the top four categories corresponding to the three SE epitope motifs. HLA-DRB1*09:01 ranks just below (table 1), providing a clear rationale for the association of this allele with RA susceptibility, despite the lack of carriage of the SE motif, demonstrating that amino acids at positions 11 and/or 13 are major contributors to RA susceptibility. Indeed, they are necessary to explain why some HLA-DRB1*14 alleles, sharing the non-SE motif RRRAE with HLA-DRB1*09:01, are not associated with an increased risk of disease (table 1). Similarly, the 19-haplotype classification system provides more granularity to the classification of protective alleles, stratifying DERAA alleles into three groups (table 1).

Although Raychaudhuri’s classification aligns overall with the SE classification by identifying additional risk categories within SE-positive and SE-negative patients, it also has its own limitations. This is illustrated by the misclassification of HLA-DRB1*14:02 into a low-risk category (table 1) based on the carriage of amino acids at positions 11, 13, 71 and 74, which are identical to some low-risk alleles of HLA-DRB1*11. However, HLA-DRB1*14:02, which carries the SE motif, has been consistently associated with a strong increase in RA risk in Indigenous North Americans³⁰ (table 1). This misclassification highlights the limitations of studies performed in cases and controls of White European ancestries, where the LD structure and prevalence of each amino acid at each position does not allow sufficient statistical power to capture all effects. Likely, the ranking of first, second and third independent HLA positions on RA risk is population-specific and likely position 70 (and possibly also 6731,33) represents an additional independent position to RA risk, not captured in the classification by Raychaudhuri. Adding position 70 and 67 in table 1 might split the SSRA-haplotype into two haplotypes, one increasing the risk (eg, HLA-DRB1*14:02), the other one decreasing the risk (eg, HLA-DRB1*11:01) of developing RA (relative to the neutral PRAA haplotype), something that the SE classification already does. *14:02 and *14:06 have the following motif for position 67 to 74 LLEQRRAA while *11:01, *11:04, *11:06, *13:05 all have FLEDRRAA.

Other HLA-DRβ chains

Extended HLA-DR haplotypes carrying RA-associated HLA-DRB1*04 (or DR4) alleles frequently include HLA-DRB4 alleles (or DRB4, which is distinct to DR4), which have been suggested to be associated with RA, with functional evidence.²⁹ From a statistical genetic point of view, despite the strong LD between HLA-DRB1 and HLA-DRB4, conditional analyses in a Korean cohort have shown that the class II association with RA is overwhelmingly driven by HLA-DRB1, with no clear independent effect attributable to DRB4 after accounting for DRB1 variations.³⁴ Previous large-scale genome-wide association studies (GWAS) in patients from diverse ancestries from the Raychaudhuri lab^{17 25} had not been able to address this question, due to the lack of a reference panel for imputing the genetic variants at HLA-DRB4. Therefore, additional genetic evidence (eg, from full sequencing efforts of the HLA locus in well powered studies across different ancestries) is required to better disentangle the effects of DRB1 and DRB4.

Genetic interactions

Gene–gene interactions between HLA-DRB1 SE alleles and genetic susceptibility polymorphisms outside the HLA (PTPN22) have been reported, particularly in seropositive disease.³⁵ Recent large-scale studies have uncovered widespread non-additive and interaction effects within HLA loci. Lenz et al showed that heterozygous combinations of HLA alleles confer higher RA risk than expected under additivity, with ~1% increase in phenotypic variance explained by seven HLA-DRB1 allele–allele interactions.³⁶ The strongest interaction for disease risk was found between HLA-DRB1*01:01 and HLA-DRB1*15:01 with an interaction OR of 2.06. Calculation of the total OR for a given haplotype pair with reported significant interaction term should therefore be carried out by multiplying the product of the two haplotypic additive ORs by the interaction OR (see legend of figure 3 of the paper by Lenz et al³⁶). The total OR for the carriage of one copy of HLA-DRB1*01:01 and one copy of HLA-DRB1*15:01 will be 4.3 (2.11×1.00×2.06), the reference being the homozygote HLA-DRB1*15:01 haplotype with OR=1.00. A complementary approach with genotypic variability-based GWAS has been developed to capture non-additive signatures, which revealed extensive epistasis within the MHC.³⁷

Outside the HLA, large multiancestry studies have expanded the catalogue of RA loci.³⁸ A transethnic meta-analysis with over 100 000 samples identified dozens of new loci and implicated key immune pathways.³⁹ An updated multiancestry GWAS of over 270 000 individuals discovered additional signals and improved fine-mapping, while also showing that PRSs perform better when trained across ancestries.⁴⁰ These advances, although extending beyond HLA-DR, demonstrate the value of inclusive study designs for refining causal variants and informing translation.

Together, these methodological innovations, from amino acid fine mapping to non-additive modelling and multiancestry integration, have refined our understanding of HLA-DRB1 risk in RA. They illustrate how increasingly sophisticated genetic approaches are opening avenues for personalised risk prediction and mechanistic insights.

Association of SE with RA outcome and implications of the discovery of valine 11

SE was initially described as a major determinant of susceptibility, but subsequent studies demonstrated that it also associates with various disease-related outcomes. SE-positive patients experience a more severe disease course, including increased radiographic progression and higher mortality, although effect sizes vary across cohorts.⁴¹ Refinements provided by fine-mapping at the amino acid level have since established that position 11 of HLA-DRB1, and in particular valine at this site, plays a pivotal role in the heterogeneity of disease outcome.

Radiographic outcome

The first study to compare the effect of SE and non-SE susceptibility amino acid positions on disease outcome showed a strong correlation between all susceptibility and severity amino acids in RA.⁴¹ This study was conducted using generalised linear latent and mixed models to include multiple X-ray measurements over time (longitudinal modelling, instead of a cross-sectional setting), taking into account different disease trajectories (latent classes), therefore demonstrating an association between HLA-DRB1 variants and a more severe disease course. In particular, the effect of valine or leucine at position 11 on radiographic progression was subsequently replicated in a large meta-analysis.⁴² Carriers of valine or leucine at position 11 had significantly greater radiographic progression compared with non-carriers.^{41 42} This effect was independent of SE status, but not independent of ACPA, indicating that the pathogenic impact of these residues is mediated within the seropositive subset⁴² (table 2). Further analysis of positions 11/13, 71 and 74 identified an effect of these genetic variations on multiple laboratory and clinical measures of inflammation, disease activity and disability, modelled longitudinally, suggesting an association with a more severe disease course, in general.⁴³ The effect of valine 11 on longitudinal measures of swollen joint counts appeared to be mainly mediated through ACPA and C-reactive protein (CRP), but a direct genetic effect was also observed, independent of CRP. These observations suggest that the effect of genetic variations is mediated through multiple independent pathways.

Table 2. Associations of HLA-DRB1 position 11 and 11–71–74 haplotypes with RA outcomes.

HLA-DRB1 allele	Outcome association	Context	Reference
Position 11: valine or leucine	Increased radiographic progression	Independent of SE; dependent on ACPA	van Steenbergen et al42
Position 11: valine	Higher inflammation and disease activity	Observed/replicated in two independent cohorts; independent of CRP levels, but mediated by ACPA status	Ling et al43
Position 11: valine	Greater radiographic severity and higher all-cause mortality	Observed/replicated in two independent cohorts; independent of clinical risk factors	Viatte et al41
11–71–74 SEA haplotype (serine–glutamic acid–alanine)	Stratifies cardiovascular mortality	SEA haplotype protective; independent of CRP and anti-CCP	Sharma et al44
Position 11 valine and SE-related alleles	Associated with improved response to abatacept, not TNFi	Observational cohorts in Japan (DRB1*04:05), Korea (Val11), US registry (SE+ and seropositive patients); SE dose–response also reported	Hirose et al;90 Inoue et al;47 Cha et al;46 Harrold et al48
Position 11 valine	No association with response to TNFi	Observational cohorts in Sweden	Jiang et al39
Position 11 valine and SE status	Not associated with drug response in randomised or observational studies	NORD-STAR RCT (no Val11/SE interaction across abatacept, TNFi, tocilizumab); BRAGGSS abatacept versus adalimumab (no clinically useful stratification)	Lend et al;49 Yap et al15
SE and DRB1*04 alleles (often encoding Val11)	Reduced risk of antidrug antibody development to adalimumab	Protective effect against immunogenicity; additive with methotrexate co-therapy	Yap et al50

A few selected examples of studies investigating the relationship between HLA-DRB1 susceptibility positions and outcome, including response to treatment (pharmacogenetics). SEA is haplotype is the HLA-DRB1 haplotype defined by the carriage of a serine at position 11, a glutamic acid at position 71 and an alanine at position 74.

ACPA, anti-citrullinated protein antibodies; BRAGGSS, Biologics in Rheumatoid Arthritis Genetics and Genomics Study Syndicate; CRP, C-reactive protein; RCT, randomised clinical trial; SE, shared epitope; TNFi, TNF inhibitors; TNFα, tumour necrosis factor α.

Mortality

The same hierarchy defined by amino acid positions 11/13, 71 and 74 also extends to mortality. In the Norfolk Arthritis Register, valine at position 11 was strongly associated with both radiographic severity and increased all-cause mortality, with carriers showing higher death rates than non-carriers. This association persisted after adjustment for known clinical confounders⁴¹ (table 2). Further analysis confirmed that haplotypes defined by these positions also predict cardiovascular mortality, with effects independent of traditional cardiovascular risk factors, baseline CRP and anti-CCP status.⁴⁴ Protective haplotypes such as the SEA group, characterised by different residues at these positions (serine (S) at position 11, glutamic acid (E) at 71 and alanine (A) at 74), were consistently associated with reduced cardiovascular risk.

Treatment response and pharmacogenetics

The clinical relevance of SE and haplotypes with valine at position 11 has been investigated in relation to biologic drug response and led to conflicting results, with studies showing an association with response to tumour necrosis factor α inhibitors (TNFi).^{41 45} Several observational studies in Japanese and Korean cohorts reported that SE positivity and valine at position 11 were associated with improved response to abatacept, while such effects were weaker or absent with TNFi. For example, in a prospective Korean study, valine at position 11 conferred a sixfold higher odd of good response to abatacept, but no effect on TNFi response.⁴⁶ A Japanese study further confirmed that carriers of HLA-DRB1*04:05, an SE allele encoding valine at position 11, had significantly better outcomes with abatacept but not with other biologics.⁴⁷ Real-world registry data from the USA suggested similar trends, with abatacept outperforming TNFi in SE-positive, ACPA-positive patients.⁴⁸ However, large randomised controlled trials have not supported a clear or clinically actionable association between Val11 or SE status and differential response to treatment. In the NORD-STAR trial, SE and valine at position 11 did not show significant heterogeneity of treatment effect across abatacept, TNFi or tocilizumab arms.⁴⁹ Likewise, a UK study comparing abatacept with adalimumab found no evidence that SE or valine at position 11 status could guide therapy choice, although there was weak evidence for an independent association of valine at position 11 with European Alliance of Associations for Rheumatology (EULAR) response.¹⁵ Taken together, these findings indicate that any effect of Val11 on treatment response is likely modest or absent, or context dependent. Beyond efficacy, Val11-related haplotypes also appear relevant to drug immunogenicity. A study of adalimumab-treated patients demonstrated that carriers of HLA-DRB1*04 alleles, which often encode valine at position 11, had reduced risk of developing anti-drug antibodies⁵⁰ (table 2).

Implications for clinical translation

Taken together, these studies show that valine at position 11 provides a unifying marker that explains much of the heterogeneity in outcome previously attributed solely to the SE. Haplotypes defined by positions 11/13, 71 and 74 form a hierarchy that stratifies risk across radiographic progression, mortality and, possibly, but to a lesser extent, treatment response. Because associations between positions 11/13 and RA susceptibility or outcome are independent of the SE, the Raychaudhuri classification of HLA-DRB1 alleles based on positions 11/13 should be seen as an extension to the SE model, rather than an alternative. Therefore, the discovery of Val11 has refined prognostic modelling, rather than revolutionised it, and offers a potential molecular hypothesis for further mechanistic investigations and the computing of PRSs for precision medicine in RA. Nevertheless, inconsistencies across populations and trial designs, and modest effects of genetic markers on their own, mean that clinical implementation of genetic stratification is not yet justified. Replication in larger and more diverse cohorts will be required before SE and Val11 haplotypes can reliably guide therapeutic decisions.

Mechanisms

Despite important progress in our understanding of genetic associations with RA susceptibility, there is currently no strong scientific consensus on the mechanism underlying these associations. After decades of research, several postulates remain open and supported by different levels of evidence. The ‘arthritogenic peptide hypothesis’⁵¹ proposes that HLA-DR molecules, comprising a β chain encoded by HLA-DRB1 susceptibility alleles, will present an autoantigenic peptide to T cells, which will induce, maintain and/or worsen arthritis. Different suggestions for the origin of this peptide are on the table, including non-citrullinated peptides derived from cartilage proteins, citrullinated peptides from various sources or peptides directly derived from peptidylarginine deiminase (PAD) enzymes.

Alternative hypotheses to the ‘arthritogenic peptide hypothesis’ suggest that RA is not primarily antigen driven. Some theories advocate the complex interplay of cytokine networks and other factors,⁵² while ‘the MHC Cusp theory’⁵³ by Holoshitz postulates that HLA molecules, including the SE, encode amino acid sequences in their hypervariable regions, named ‘cusps’ based on their conformation. These cusps would act as ligands to yet unidentified receptors, leading to pathological signalling and, ultimately, in combination with additional factors, to autoimmunity.

These different hypotheses are not mutually exclusive, and it is imaginable that different mechanisms are at play in different patients, or even within the same patient, explaining heterogeneity in clinical presentation and response to treatment.

In this review, we will focus on antigen presentation and recognition (the ‘arthritogenic peptide hypothesis’) and attempt to understand the relationship between HLA-DRB1 polymorphisms, a putative arthritogenic peptide and the T cell receptor (TCR).

Antigen presentation

Since the 1970s, autoimmunity against collagen type II, expressed in the cartilage, has been suggested as a potential aetiological mechanism in RA.⁵⁴ Studies conducted by Feldmann around the time of the discovery of the SE identified persistent collagen type II-specific T cell clones in the synovial membrane of patients with RA,⁵⁵ adding to the suggestive evidence that collagen type II-derived peptides are presented on HLA class II molecules to CD4⁺ T cells in RA. Further cartilage proteins (eg, human cartilage glycoprotein-39 or aggrecan) were also identified as potential CD4⁺ T cell antigens around that time.^{52 56} Interestingly, despite the presence of humoral autoimmunity against citrullinated proteins in RA, these early T cell antigens were not citrullinated.

The study of the peptide binding specificity of HLA-DR molecules associated with RA⁵⁷ and the X-ray crystal structure of HLA-DR4 (DRα*01:01, DRβ1*04:01) complexed with a peptide from human collagen type II⁵⁸ in the 90s has provided crucial mechanistic insights into our understanding of the role of HLA-DRB1 susceptibility alleles in antigen presentation.

The peptide binding register for HLA-DRβ1*04:01 shows amino acid preferences at six positions within the peptide (P1, P2, P3, P4, P6 and P7), the majority of which are also anchor residues to binding pockets located within the HLA-DR molecule: the peptide side chains at P1, P4, P6, P7 and P9 stretch into pockets in the peptide-binding groove. The SE forms part of the α helix at one edge of the peptide-binding groove of HLA-DR molecules and SE residues shape the biophysical properties of the P4 pocket; in particular, the positively charged amino acid at position 71 (Lys-β71 for *04:01 or Arg-β71 for other SE alleles) will repulse positively charged peptide side chains (such as Arg or Lys), therefore creating a biophysical environment preferentially accommodating nonpolar (Met, Ala, Val and Leu) or negative amino acids (Glu or Asp).^{57 58}

The crystal structure of the complex DRB1*04:01/collagen type II_1168–1180 peptide⁵⁸ (QYMRADQAAGGLR; M=P1) illustrates this mechanism, with the negatively charged P4 Asp (D) side chain (figure 4A) deep in the P4 pocket, forming a salt-bridged hydrogen bond with the ɛ-amino group of the positively charged Lys-β71 of the SE, which is pointing directly into the peptide-binding site. However, this structural analysis has also shown how chemical bonds between the antigenic peptide and the HLA-DR molecule occur at positions which have not been associated with genetic susceptibility. For example, the arginine at CII_1168–1180 P2 hydrogen-bonds to the carbonyl oxygen of β chain Thr-77; there are also several hydrogen bonds between the peptide and conserved HLA-DR residues.

Figure 4. Structural basis of peptide selectivity of HLA-DR molecules with β chains encoded by HLA-DRB1 alleles associated with RA susceptibility. (A) Three-dimensional x-ray crystal structure of the human collagen II_1168–1180 peptide as it lies bound in the cleft of HLA-DRα*01:01/HLA-DRβ1*04:01. The bound collagen peptide (QYMRADQAAGGLR; with the Methionine M at the P1 position) is in an extended conformation. Nitrogen atoms (N) are represented in blue, while oxygen atoms (O) are in red. The side chain of the P4 residue, aspartic acid (Asp), extends its negatively charged carboxylate group (-COO⁻) deep into the P4 pocket (not shown) to interact with the positively charged Lys-β71 encoded by the SE. The side chains of Q (Gln) at P−2 and P5, and L (Leu) at P10 are not shown (these amino acids were modelled as alanine). Reproduced with permission of Elsevier, from Dessen et al⁵⁸; permission conveyed through Copyright Clearance Center. Licence number 6221990983121. (B) Interactions with citrulline in the P4 pocket of HLA-DRβ1*04:01 and HLA-DRβ1*04:04. The amino acid residues of HLA-DRβ1 associated with RA and the binding of antigenic peptides are highlighted in grey, specifically at positions 11, 13 and 70–74. In the peptide binding groove, nitrogens are coloured in blue, and oxygens in red and hydrogen bonds are shown in dashed lines. For peptide naming convention, the numbers preceding ‘Cit’ indicate the vimentin residues that are citrullinated, and the subscripted numbers following ‘Cit’ indicate the peptide span (starting and end position within the protein sequence). (a) Vimentin-71Cit_66–78 coloured in yellow, (b) vimentin-64Cit_59–71 coloured in pink, (c) vimentin–64–69–71Cit_59–71 coloured in green and (d) aggrecan–93–95Cit_89–103 (coloured in blue, all four peptides bound to HLA-DRβ1*04:01). Residues from the β chain important for contacts with the P4 citrulline are represented as sticks. (e) Vimentin–71Cit_66–78 coloured in teal bound to HLA-DRB1*04:04. Used with permission of Rockefeller University Press, from Scally et al⁶⁵; permission conveyed through Copyright Clearance Center. Licence number 1667078-2.

The crystal structure also shows how peptide side chains at multiple positions contribute to the epitope recognised by the TCR; P −1 (P minus 1), P2 and P5 project towards the TCR, with P −2, P3, P4, P6, P7, P8 and P9 also exposed, possibly contributing to the surface recognised by T cells. Interestingly, a direct recognition of the SE by the T cell is also suggested, with Gln-β70 pointing out of the binding site towards the TCR.

These observations are compatible with a molecular model where the selective binding of peptides by the P4 pocket is a crucial part of the mechanism explaining the association of SE alleles with RA.

Citrullinated CD4⁺ T cell autoantigens

In 2003, Hill et al demonstrated that the conversion (or deimination) of the positively charged P4 arginine to the uncharged citrulline of a vimentin autoantigenic peptide importantly increased both the affinity for HLA-DRβ1*04:01 and the immunogenicity of the peptide; that is, the capacity to elicit, in this context, a CD4⁺ T cell response.⁵⁹ Citrullination can occur as a post-translational modification catalysed by PAD enzymes. Using a wide range of techniques (eg, peptide binding assays, peptide-loaded fluorochrome-labelled HLA-DRβ1*04:01 tetramers, crystallography, transgenic mice, cloning of antigen-specific CD4⁺ T cells from human subjects, TCR cloning and re-expression in cell lines for in vitro characterisation), several studies from independent groups confirmed that citrullination of selected antigenic peptides derived from different proteins (eg, vimentin, fibrinogen, cartilage intermediate-layer protein, α-enolase, aggrecan, tenascin) increased their affinity for SE alleles and their immunogenicity.60,65 Some peptides would only bind to HLA-DR in their citrullinated form⁶⁶ and some are uniquely immunogenic when citrullinated.⁶³

A crystallographic study by Scally et al in 2013 demonstrated that HLA-DRβ1*04:01 and *04:04 molecules fit citrulline much better than arginine in the P4 pocket⁶⁵ (figure 4B). These findings provided an extension of the early structural framework presented above and linked humoral immunity (ACPA) to cellular immunity (CD4⁺ T cells) against citrulline. These mechanistic studies of the effect of HLA-DRB1 in RA are complementing epidemiologic and genetic evidence showing that SE alleles are more consistently associated with ACPA/anti-CCP positivity than with disease susceptibility per se.¹²

However, many reports on the effect of citrullination60,65 also observed that citrullination of other peptides, or at other positions, than those primarily investigated, would not increase binding affinity to SE alleles, with arginine and citrulline containing peptides binding equally well, or even decrease binding affinity and/or immunogenicity, in some instances.^{61 63}

The interpretation of the functional impact of citrullination is further complicated by the fact that increased affinity does not correlate with increased immunogenicity, as low affinity peptides have been reported to elicit functional T cell responses in RA.⁶⁷ Moreover, citrullination also increases binding affinity to non-HLA SE molecules, such as HLA-DQ.⁶⁸

Therefore, the question was raised of whether citrullination systematically contributes to autoantigen presentation by SE alleles in RA. The study by the lab of Sette in 2017⁶⁹ and another by the lab of Rossjohn in 2018⁶⁶ provided crucial answers to this question.

The study by the Sette’s lab investigated the binding affinity of four previously reported peptides derived from collagen type II, aggrecan, vimentin and fibrinogen to 28 common HLA class II molecules.⁶⁹ In a second experiment, they tested over 200 citrullinated peptides derived from vimentin and collagen II to determine their binding affinity to the SE alleles DRB1*01:01 and DRB1*04:01.

The authors confirmed previous observations that citrullination of selected peptides at P4 increases binding affinity to SE alleles. They also observed peptide/HLA combinations where binding of the arginine and citrullinated peptides is identical, and situations where citrullination has a deleterious impact on binding, depending on the position where citrullination happened. They also demonstrated the crucial impact of modification of non-anchor residues on binding. Considering all combinations of peptide/HLA/anchor positions analysed in their study, the authors concluded that there is no consistent impact of citrullination on binding. Citrullination of any randomly selected arginine residue, without taking the structural context into account, like surrounding amino acids within the peptide and the HLA, will not systematically generate CD4⁺ T epitopes with increased binding affinity to HLA class II molecules.

The Rossjohn lab also systematically tested the effect of citrullination on affinity and antigen presentation. They compared the binding register of three different SE-containing HLA class II molecules based on sets of up to 3000 peptides eluted from each HLA molecule (naturally selected ligands). In general, SE peptide motifs disfavoured the positively charged lysine and arginine at P4, but also favoured specific amino acids outside P4 (eg, at P1, P6 and P9). Interestingly, the amino acid preferences at some positions outside P4 were different between different SE alleles (in particular at P1 and P9). In a second step, they used a peptide binding assay to determine the binding affinity of 34 peptides to DRB1*04:01, *04:04 and *04:05 and generated multiple peptide–HLA crystal structures.⁶⁶ Their findings showed that, for the peptides examined, citrullination at P4 enhanced binding across these alleles, with the P4 citrulline adopting a convergent orientation within the binding groove. At the same time, differences at other polymorphic sites outside the SE, including the P1 and P9 pockets, shaped peptide hierarchies, meaning that not all peptides were equally stabilised by citrullination. These data refined the model by demonstrating that SE alleles share a common structural basis for accommodating citrullinated epitopes yet retain allele-specific preferences that influence which autoantigens are preferentially presented.

In summary, the study by the Rossjohn lab,⁶⁶ in agreement with previous studies, established independent determinants of peptide binding affinity to HLA class II molecules carrying the SE motif: (a) presence/absence of a citrulline; (b) the position of citrullination (P4 increasing affinity, with no impact on affinity in cases of citrullination outside the P4 anchor⁶⁶); (c) the number of citrullinated peptide residues; (d) the biophysical properties of amino acid side chains at positions outside P4 (P1 and P9 anchor residues); (e) structural differences of the HLA outside the P4 pocket (for example HLA-DRβ1 residues at position 57 or 86, defining the P9 or P1 pockets, respectively). These observations explain why citrullination cannot be seen as the sole pathogenic factor linking the HLA with clinical symptoms.

Noteworthy, neither position 57 nor 86 was associated with RA susceptibility independently of positions 11/13, 71 and 74 in the large study by Raychaudhuri presented above.²⁵ Also, in agreement with early observations from the Wiley lab, crystal structures of the Rossjohn lab identified a number of chemical bonds between antigenic peptides and HLA residues with no significant genetic associations with RA.

Finally, the study of peptides carrying two citrullinated residues⁶⁶ has shown that, in some instances, one citrulline is used to anchor the peptide while the other points towards the TCR. The complexity of the effect of citrullination is further exemplified by the complexity of peptide binding to HLA-DRβ1*14:02, which can accommodate both arginine and citrulline, but in opposite direction.³⁰

Therefore, citrullination, or more generally post-translational peptide modification, can generate novel T cell epitopes by either increasing binding affinity to HLA through modification of anchor residues, or by altering non-anchor residues pointing towards the TCR. The genetic and molecular conclusions on the binding and presentation of citrullinated and non-citrullinated epitopes by SE-containing HLA-DR molecules are corroborated by the presence of antigen-experienced CD4⁺ T cells, specific for these epitopes, in patients with SE⁺ RA.60,63 However, the SE hypothesis and the arthritogenic peptide hypothesis remain hypotheses; there is insufficient evidence to support some of the pathogenic steps from genetics to clinical manifestation. In particular, we still do not know if ‘autoreactive’ CD4⁺ T cells, including citrulline-specific CD4⁺ T cells, are pathogenic or if they arise as a consequence of the disease. More work is required to characterise autoreactive T lymphocytes and understand if and how peptide selection by HLA-DRβ1 polymorphisms induces, alters or modifies the quality and intensity of T cell responses (Th1, Th2, Th17, peripheral helper T cells (Tph)).

Direct binding to the TCR and shaping the TCR repertoire

More recent studies confirmed earlier findings that RA susceptibility alleles contribute not only to peptide binding, but also directly to TCR engagement. Structures of TCR–HLA-DRB1*04:01 complexes bound to citrullinated fibrinogen peptides showed that the SE motif itself formed a major TCR contact site.⁷⁰

In addition, while citrullination at P4 stabilised peptide binding, citrullination at other positions such as P2 could directly interact with the TCR, altering repertoire selection. Functional analyses confirmed biased TCR usage against citrullinated fibrinogen epitopes, both in DRB1*04:01 transgenic mice and in patients with RA. Together, these findings support a model that the SE can exert dual functionality, stabilising citrullinated peptides within the binding groove and simultaneously acting as a recognition determinant for TCR engagement, thereby shaping autoreactive T cell responses that are closely linked to ACPA development.

Complementary evidence comes from genetic and structural analyses of thymic selection. HLA-DRB1 position 13, together with SE residues 71 and 74, strongly influences the TCR β-chain CDR3 repertoire in healthy individuals.⁷¹ This indicates that RA risk alleles bias thymic selection towards autoreactive TCRs prior to disease onset, supporting the central hypothesis of HLA-mediated autoimmunity.

Alternative mechanistic hypotheses

While these structural insights clarify how SE-bearing HLA-DR molecules present citrullinated and non-citrullinated epitopes and shape TCR engagement, alternative immunological models have also been proposed. One such model is the ‘hapten–carrier’ hypothesis developed by Roudier,⁷² in which citrulline-specific B cells would recognise citrullinated antigens (haptens) still bound to PAD enzymes, which would function as carriers. The hapten–carrier complexes would be taken up by antigen-presenting cells (like eg, B cells), processed and PAD-derived peptides (but not citrullinated peptides) would be presented by SE-containing HLA-DR molecules.

This model is supported by convergent genetic, biochemical and immunological evidence. In mouse models, PAD immunisation can elicit both anti-PAD cellular and humoral responses (anti-PAD-specific CD4⁺ T cell responses and anti-PAD autoantibodies) and ACPA responses, without inducing anti-citrulline-specific CD4⁺ T cell responses (assessed by the absence of CD4⁺ T cell proliferation after in vitro stimulation with citrullinated fibrinogen).⁷³ In humans, PAD4-directed CD4⁺ T cell responses and anti-PAD4 antibodies co-occur in some patients with RA and show enrichment in SE-positive individuals, consistent with PAD4 functioning as a T cell antigen that can provide help to B cells recognising citrullinated substrates.⁷⁴ From a genetic point of view, HLA-DRB1 genotype-associated risk has been reported to correlate with the capacity of the encoded HLA-DR molecules to bind PAD4-derived peptides, providing a mechanistic link between genetic risk and antigen presentation within this framework.⁷⁵

Therefore, in the hapten–carrier hypothesis, the autoimmune response to PAD enzymes would extend beyond PAD themselves to multiple citrullinated proteins, in a model where PAD-driven T cell help supports subsequent epitope spreading to citrullinated proteins, without the need of SE-bearing HLA-DR molecules to present citrullinated peptides.⁷³

The hapten–carrier model would help explain the breadth of ACPA responses in RA, consistent with genetic evidence that HLA-DRB1 primarily influences autoantibody status. It would also conveniently identify PAD enzymes as unique therapeutic targets in patients with ACPA⁺ RA, with emerging evidence of the feasibility of such tolerogenic approaches in mice,⁷⁶ without the need to induce tolerance against each individual citrullinated antigen.⁷²

However, despite supporting evidence, the PAD hapten–carrier model remains to be fully established.⁷⁴ Many mechanistic aspects of this postulate have not been investigated yet; for example, PAD4-specific CD4⁺ T cells have not been deeply characterised to the depth and breadth that citrulline-specific CD4⁺ T cells have been, and they were not directly demonstrated to help B cells specific for PAD4 and/or citrullinated antigens.⁷⁴ In addition, only some patients with RA seem to develop immune responses to PAD4,⁷⁴ leaving space for alternative aetiologic mechanisms in the other patients.

Together, the findings discussed in this review suggest that HLA-DRB1 risk in RA arises from multiple, potentially overlapping immunological mechanisms, many of which appear to act primarily through the modulation of autoantibody responses, including peptide-specific effects of citrullination on HLA-DR presentation, SE-mediated TCR recognition, biases in thymic selection and PAD-centred immune responses that may facilitate epitope spreading. Structural and functional studies have clarified the molecular basis underlying some of these mechanisms. However, the relative contribution, temporal ordering and interplay of these mechanisms in vivo remain incompletely resolved. It is therefore plausible that distinct but complementary pathways operate in parallel or in different immunological contexts, or in different patients, collectively contributing to disease initiation and progression. Therefore, the exact mechanism explaining how the HLA-DRB1 susceptibility alleles contribute to disease aetiology remains unknown.

Cell type-specific roles of HLA-DRB1 in RA

HLA-DRB1 risk alleles contribute to seropositive RA pathogenesis through expression on multiple cell types. Professional antigen presenting cells (APCs) such as dendritic cells (DCs), macrophages and B cells are central, but other cells such as fibroblasts, endothelial cells and activated lymphocytes can also upregulate HLA-DR in the inflamed synovium.

Antigen presenting cells

Conventional DCs are the primary initiators of HLA class II-restricted immune responses. DCs constitutively express high levels of HLA class II molecules, including HLA-DR, enabling them to efficiently interact with various CD4⁺ T cell subsets (eg, Th1, Th17, Tph)^{77 78} (table 3). DCs present peptide antigens loaded on HLA-DR, prime, reactivate or inhibit CD4⁺ T cells and are therefore key players in synovial immune homeostasis. Single-cell mapping has defined different DC subsets and functional states (eg, DC1, DC2, DC3, iDC3) in the synovium, including a tolerogenic AXL⁺ DC2s in the lining layer.⁷⁸ The pathogenic relevance of these pathways is underscored by direct ex vivo evidence of autoreactive CD4⁺ T cells recognising citrullinated peptides such as vimentin, fibrinogen, aggrecan and collagen II, confirming HLA-DRB1 as the restricting element.^{60 79 80} Translational studies have sought to harness this biology for therapy: autologous DCs pulsed with citrullinated peptides or incubated with autologous synovial fluid were safely administered to patients with RA, demonstrating the feasibility of tolerogenic vaccination.^{81 82}

Table 3. Some cell types expressing HLA-DR in RA and the proposed roles of HLA-DR in this cellular context.

Cell type	HLA-DR expression features	Proposed functional role in RA	References
Dendritic cells	Constitutive HLA-DR; elevated in inflamed synovium; peptide presentation capacity	Prime autoreactive CD4+ T cells with citrullinated autoantigenic peptides, promote inflammatory Th1/Th17 responses, potential tolerogenic roles	Wehr et al;77 MacDonald et al;78 Law et al;79 Benham et al81
Macrophages	HLA-DR upregulated in inflamed synovium	Amplify local antigen presentation, sustain inflammation	Tu et al83
Fibroblast-like synoviocytes	IFNγ-inducible HLA-DR on CD90+ fibroblasts	Possibly present antigens locally, interact with NK cells, contribute to joint inflammation	Zhao et al84
T cells (Tph, Treg, CD8)	‘Aberrant’ HLA-DR expression in at-risk, pre-RA and early RA	Unknown	Takada et al85
B cells	Constitutive HLA-DR expression	Antigen presentation and antibody production, known contributors to RA	Wehr et al77

IFN, interferon; NK, natural killer; RA, rheumatoid arthritis.

Macrophages also contribute to antigen presentation in the synovium. Macrophages also express HLA-DR and contribute to antigen presentation mainly within inflamed tissues. Distinct subsets include tissue-resident macrophages, which may have regulatory functions, and monocyte-derived macrophages, which accumulate during inflammation and express high levels of HLA-DR.⁸³ While less efficient than DCs in priming naïve T cells, HLA-DR⁺ macrophages act as amplifiers of inflammation and help sustain local antigen presentation.

Fibroblast-like synoviocytes, traditionally considered structural cells, can be induced to express HLA-DR in the presence of interferon-γ. A subset of CD90⁺ fibroblasts in the RA synovium has been shown to acquire HLA-DR expression and to interact with CD4⁺ T cells, possibly through antigen presentation.⁸⁴ This demonstrates that fibroblasts may contribute directly to the pool of antigen-presenting cells in inflamed joints.

Despite these recent progresses, the exact sequence of events, cell-cell interactions and their location (which DC subset interacts with which CD4⁺ T cell subset, when and where) have not been characterised yet. Therefore, the spatiotemporal context in which HLA-DRB1 susceptibility polymorphisms contribute to the initiation and maintenance of RA remains unknown.

Non-APCs

Beyond APCs and stromal cells, lymphoid cells can also upregulate HLA-DR under pathological conditions. Activated T cell subsets express HLA-DR, but its function in this context remains unknown. Tph, ICOS^hi memory regulatory T cells and subsets of CD8⁺ T cells with high HLA-DR expression are expanded in individuals at risk of RA, as well as in preclinical and early disease stages⁸⁵ (table 3). In established patients with RA, mass cytometry analysis showed expansion of HLA-DR+T-bet⁺ effector CD4⁺ T cells (Th1 cells) and a reduction of FOXP3⁺ HLA-DR⁺ regulatory T cells, indicating that HLA-DR expression persists in certain T cell subsets in chronic disease.^{86 87} Although the functional significance of HLA-DR expression on these lymphocytes remains uncertain, their expansion suggests an additional layer of abnormal antigen presentation or immune modulation.

HLA-DR expression in RA thus extends beyond classical APCs to fibroblasts and lymphocytes, broadening the network of potential antigen presentation in the joint. The relative importance of these non-classical roles remains uncertain, but they highlight potential targets for tolerising therapies.

HLA-DRB1 disease pleiotropy and utility in prediction

HLA-DRB1 alleles are not exclusively associated with RA susceptibility or outcome. They act as pleiotropic risk and protective factors across a wide spectrum of immune-mediated diseases. This breadth of associations reflects the central role of HLA-DR molecules in antigen presentation and immune regulation, with HLA-DRB1 strongly implicated across systemic autoimmunity. Across conditions such as systemic lupus erythematosus, type 1 diabetes, inflammatory bowel disease and multiple sclerosis, overlapping but non-identical HLA-DRB1 associations have been reported (table 4), often with allele effects that differ by ancestry and may mirror or invert those observed in RA. These patterns are well documented in the literature and have been comprehensively reviewed elsewhere.⁸⁸

Table 4. HLA-DR alleles association with some autoimmune diseases.

Disease	HLA-DRB1 associations (risk/protective)	Reference (file)
Type 1 diabetes	Risk: DRB1*03:01, DRB1*04:01, DRB1*04:05 and DRB1*04:02 Protective: DRB1*15:01, DRB1*14:01, DRB1*07:01, DRB1*04:03	Noble91
Multiple sclerosis	Risk: DRB1*15:01, DRB1*03:01, DRB1*13:03 and DRB1*08:01	Moutsianas et al92
Systemic lupus erythematosus	Risk: DRB1*03:01, DRB1*08:01 and DQA1*01:02 (Europeans), DRB1*09:01, DRB1*15:01 (Asians)	Kwon et al93
Inflammatory bowel disease	Risk: DRB1*01:03	Cleynen et al94

Importantly, such cross-disease associations highlight shared immunological pathways rather than disease-specific mechanisms and do not imply equivalent pathogenic roles for individual alleles across conditions. In RA, for example, several HLA-DRB1 alleles that confer risk in other autoimmune diseases are neutral or protective, underscoring the context-dependence of HLA effects.

The potential clinical utility of HLA-DRB1 variation has been explored primarily in the context of disease differentiation rather than disease prediction. Because HLA-DRB1 risk alleles are common in the general population and RA prevalence is relatively low, their positive predictive value for incident disease is inherently limited. Probabilistic models such as G-PROB, with demonstrated reproducibility, therefore aim to improve differential diagnosis between inflammatory arthritides rather than to predict disease onset in unselected populations.⁸⁹

Although HLA-informed tools are promising, broader ancestry representation and functional studies are needed before these insights can be fully translated into clinical practice. In addition, several uncertainties regarding disease mechanisms remain. It is not clear why the same allele predisposes to one autoimmune disease yet protects against another, nor how structural peptide-binding preferences map to distinct disease pathways. The contribution of ancestry, environmental exposures and gene–gene interactions to these effects is also incompletely understood.

Conclusion

Decades of research have established specific HLA-DRB1 alleles as the dominant genetic risk factors for RA, with the strongest and most consistent effects observed in seropositive (ACPA/anti-CCP positive) disease, yet their contribution extends far beyond susceptibility. The shift from serological typing to amino acid fine-mapping has provided a unifying explanation for risk, outcomes and heterogenous effects across populations. Mechanistic studies now link HLA-DRB1 risk residues to altered peptide binding, direct TCR engagement and thymic repertoire bias, while synovial studies highlight that HLA-DR is expressed by both professional APCs and other immune and non-immune cell types. Associations across autoimmunity underscore the central role of HLA-DRB1 polymorphisms in human immune regulation. Despite this progress, several gaps remain: the relative importance of different cellular contexts, the dominant mechanism by which risk is realised in vivo, and the feasibility of translating genetic knowledge into personalised therapy. Addressing these challenges will be critical in order to translate knowledge from genetic association studies to clinical application, positioning HLA-DRB1 not only as a marker of risk but as a gateway to new tolerising and stratified approaches in the clinical management of RA.