Article Text

Pathogenic antibody response to glucose-6-phosphate isomerase targets a modified epitope uniquely exposed on joint cartilage
  1. Taotao Li1,
  2. Changrong Ge1,
  3. Alexander Krämer1,
  4. Outi Sareila1,2,
  5. Monica Leu Agelii2,
  6. Linda Johansson3,
  7. Kristina Forslind4,5,
  8. Erik Lönnblom1,
  9. Min Yang1,
  10. Bingze Xu1,
  11. Qixing Li6,
  12. Lei Cheng1,
  13. Göran Bergström7,
  14. Gonzalo Fernandez1,
  15. Alf Kastbom8,
  16. Solbritt Rantapää-Dahlqvist3,
  17. Inger Gjertsson2,9,
  18. Rikard Holmdahl1
  1. 1 Section of Medical Inflammation Research, Department of Medical Biochemistry and Biophysics, Karolinska Institute, Stockholm, Sweden
  2. 2 Department of Rheumatology and Inflammation Research, University of Gothenburg Sahlgrenska Academy, Göteborg, Sweden
  3. 3 Department of Public Health and Clinical Medicine/Rheumatology, Umeå Universitet, Umeå, Sweden
  4. 4 Department of Clinical Sciences, Section of Rheumatology, Faculty of Medicine, Lund University, Lund, Sweden
  5. 5 Spenshult Research and Development Center, Halmstad, Sweden
  6. 6 Center for Medical Immunopharmacology Research, Southern Medical University, Guangzhou, China
  7. 7 Department of Clinical Physiology, Sahlgrenska University Hospital, Göteborg, Sweden
  8. 8 Department of Rheumatology and Department of Biochemical and Clinical Sciences, Linköping University, Linköping, Sweden
  9. 9 Rheumatology, Sahlgrenska University Hospital, Göteborg, Sweden
  1. Correspondence to Professor Rikard Holmdahl, Medical Inflammation Research, Karolinska Institute, Stockholm 171 77, Sweden; rikard.holmdahl{at}


Objectives To identify the arthritogenic B cell epitopes of glucose-6-phosphate isomerase (GPI) and their association with rheumatoid arthritis (RA).

Methods IgG response towards a library of GPI peptides in patients with early RA, pre-symptomatic individuals and population controls, as well as in mice, were tested by bead-based multiplex immunoassays and ELISA. Monoclonal IgG were generated, and the binding specificity and affinity were determined by ELISA, gel size exclusion chromatography, surface plasma resonance and X-ray crystallography. Arthritogenicity was investigated by passive transfer experiments. Antigen-specific B cells were identified by peptide tetramer staining.

Results Peptide GPI293-307 was the dominant B cell epitope in K/BxN and GPI-immunised mice. We could detect B cells and low levels of IgM antibodies binding the GPI293-307 epitopes, and high affinity anti-GPI293-307 IgG antibodies already 7 days after GPI immunisation, immediately before arthritis onset. Transfer of anti-GPI293-307 IgG antibodies induced arthritis in mice. Moreover, anti-GPI293-307 IgG antibodies were more frequent in individuals prior to RA onset (19%) than in controls (7.5%). GPI293-307-specific antibodies were associated with radiographic joint damage. Crystal structures of the Fab–peptide complex revealed that this epitope is not exposed in native GPI but requires conformational change of the protein in inflamed joint for effective recognition by anti-GPI293-307 antibodies.

Conclusions We have identified the major pathogenic B cell epitope of the RA-associated autoantigen GPI, at position 293–307, exposed only on structurally modified GPI on the cartilage surface. B cells to this neo-epitope escape tolerance and could potentially play a role in the pathogenesis of RA.

  • arthritis, experimental
  • arthritis, rheumatoid
  • autoimmunity
  • autoantibodies
  • inflammation

Data availability statement

All data relevant to the mouse study are included in the article or uploaded as supplementary information. Data from several clinical studies (BARFOT, TIRA-2, Umeå, TIRA-1 and WINGA,) cannot be made publicly available due to ethical restrictions. The crystallographic coordinates and structure factors elucidated in this study have been deposited in the Protein Data Bank with the accession code 8BBH.

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.


  • Serum from mice with circulating antibodies to glucose-6-phosphate isomerase (GPI) protein mediate arthritis.

  • Controversy exists regarding the specificity and sensitivity of serum levels of anti-GPI antibodies in patients with rheumatoid arthritis (RA).


  • A dominating neo-epitope on GPI protein is defined and shown to be exposed on the inflamed cartilage surface.

  • Antibodies to the new neo-epitope are arthritogenic in mice.

  • Antibodies to the new neo-epitope are increased in early RA.


  • The identification of specific pathogenic antibodies and their functions offers the possibility of subtyping RA and a better understanding of its aetiology.


Rheumatoid arthritis (RA) is a chronic, inflammatory joint disease characterised by leucocyte invasion, synovial inflammation, cartilage and bone destruction. The disease affects populations worldwide with a prevalence of 0.5%–1%. Autoantibodies are hallmarks for RA and often appear years before the clinical onset.1 2 Anti-citrullinated protein antibodies (ACPAs) and rheumatoid factors have been widely used as biomarkers for clinical diagnosis of RA,2 3 but other autoantibodies such as antibodies to glucose-6-phosphate isomerase (GPI)4–6 and type II collagen (COL2) are also present in the serum.7

GPI is a ubiquitously expressed protein with multiple functions. Patients with RA harbour elevated serum and synovial fluid levels of free and immune-complexed GPI.5 8 9 The arthritogenic effects of GPI have been demonstrated in the K/BxN murine model, a spontaneous arthritis model of GPI T cell receptor transgenic mice,10 and GPI protein-induced arthritis (GPIA) models.11 The recognition of peptide GPI282–294 by transgenic T cells10 and subsequent production of high titers of anti-GPI antibodies cause severe joint inflammation.12 Passive transfer of K/BxN serum or purified anti-GPI immunoglobulin G (IgG) into naïve mice induce arthritis.12 In the GPIA model, the major MHCII Aq restricted T cell epitope is peptide GPI325-339, which induces arthritis in a T and B cell dependent manner.13–15 Arthritis can be induced by a mixture of multiple anti-GPI monoclonal antibodies (mAbs), but not by a single antibody, and the main causative epitope is not known.16

Regarding a potential role of GPI in RA, it is still unclear whether RA is associated with serum levels of GPI protein and anti-GPI antibodies. Some studies claim to detect increased concentrations of anti-GPI antibodies, as well as increased soluble GPI protein concentrations in serum and synovial fluid in 50%–64% of patients with RA.4 5 8 17 18 However, other studies argue that GPI is not an autoantigen specific for RA,19 or that anti-GPI antibodies could only be detected in a minority of patients with RA.6 20 21

This study aims at identifying the pathogenic B cell epitopes on GPI and to understand how a response to a systemically expressed protein can be directed to cartilaginous joints, to help understand the initiation of RA.



B6NQ.Ncf1m1j/m1j , BQ.Cia9i, BQ.FcγR2b−/ and DBA/1 mice were used in this study. Detailed information of the mouse strains can be found in previous studies.22 23 All mice were kept and bred in a climate-controlled specific pathogen-free environment with 12-hour light-dark cycles, housed in polystyrene cages containing wood shavings and provided with standard rodent chow and water ad libitum in the animal facility KM-A, Karolinska Institute (Stockholm, Sweden). All animal procedures were approved by animal welfare authorities (Stockholm, Sweden) with permit number N35/16, 2660-2021. All the mice were age-matched, randomised to an experimental group, 3–10 mice per group and 3–5 mice per cage were used. All experiments were performed in 8–10-week-old mice following Animal Research: Reporting of In Vivo Experiments (ARRIVE) guidelines.24

Bead-based multiplex flow immunoassay

Mouse and human serum or plasma IgG antibodies or mAbs were detected by bead-based flow immunoassay as described previously.25 Briefly, samples were incubated with beads immobilised with different biotinylated peptides. R-phycoerythrin-goat-anti-human IgG or -anti-mouse IgG/IgM was used as secondary antibody. Results were obtained as median fluorescence intensity. Data normalisation is as same as study.26 Normalised results are expressed as arbitrary units. Limit of detection (LoD) was determined by applying the principles of the Clinical & Laboratory Standards Institute (CLSI) guideline.27

Patients and population-based donors

Pre-symptomatic individuals (n=518), population controls (n=530) and patients with early RA (n=241) were recruited in northern Sweden (designated here as ‘Umeå study’).2 Samples (plasma) from pre-symptomatic individuals were collected before the onset of any symptoms of joint disease. Matched population controls were selected from the same Biobank cohorts as the pre-symptomatic individuals. Samples from patients with early RA, who also had a pre-dating sample, were collected at the time of being diagnosed with RA.

Serum samples were collected from patients with early RA who were included in the Swedish BARFOT28 and TIRA-229 cohorts (n=1493 and 502, respectively). Serum samples from population controls (n=935, age-matched and sex-matched with respect to BARFOT and TIRA-2) were obtained from the population-based health surveys MFM ÅUS (Malmö Förebyggande Medicin; Återundersökning) study. Patients with RA undergoing prosthetic hip or knee replacement surgery and patients with fractured femur head were enrolled in this study to collect human articular cartilage.30

This study was performed in compliance with the Declaration of Helsinki. All participants gave their written informed consent when donating samples. Participants are further described in online supplemental materials.

Supplemental material

Radiographic assessments

In the BARFOT study, posterior–anterior radiographs of the hands and feet at 12 months were assessed according to the van der Heijde modification of the Sharp Score (SHS)31 based on 32 joints in the hands and 12 in the feet, calculating total SHS (range 0–448), Erosion Score (ES) (range 0–280) and Joint Space Narrowing Score (range 0–168). The radiographs were read by one of two experienced readers. Double readings of a fraction of films showed good agreement between the two readers. The intraclass correlation coefficient for SHS was excellent (0.940–0.998). Furthermore, the agreement between two observers in identifying the presence or absence of erosions was calculated. Kappa proved to be 0.80 implying substantial agreement.32

Statistical analyses

GraphPad Prism V.8.3.0 was used to analyse the results of most mouse experiments. Quantitative data from the animal experiments are expressed as the mean±SEM. The heatmap which showed the antibody responses in multiplex flow immunoassays was performed by ComplexHeatmap package developed in the R programme (R Core Team, 2017)

Continuous data were compared (pre-symptomatic individuals vs controls or vs patients) by non-parametric analyses using Mann-Whitney’s test for two groups. Relationships between categorical data (positive vs negative) were compared using the χ2 analysis or Fisher’s exact test as appropriate. Receiver operating characteristics (ROC) curves were generated for Umeå and BARFOT+TIRA2 studies, and a cut-off point corresponding to a specificity of >90% was determined according to Youden’s index. The IgG response among BARFOT patients with RA was defined as highly positive when the response exceeded median+5 median absolute deviation of that in population controls. The risk ratio for associations with RA was calculated with ORs and 95% CIs.

ES at 12 months in relation to categorical data (peptide positive/negative) was analysed in a zero-inflated negative binomial model (SAS proc GENMOD) that can accommodate for the excess of zero scores and overdispersion. The model is composed of two parts—one that models the probability of a zero-ES and one that models the non-zero part. We have observed that the anti-cyclic citrullinated peptide (CCP)-negative patients were more likely to have a zero-ES at 12 months (p value=0.002), so anti-CCP status was introduced in the first part of the model to account for anti-CCP antibodies (positive/negative). The anti-CCP status was not significant for the non-zero part of the model.

Statistical calculations were performed using the SPSS programme for Windows, V.28.0 (SPSS) or SAS V.9.4 (SAS Institute). Considering the explorative nature of the study, a p value ≤0.05 was defined as statistically significant in all analyses.


GPI293-307 is the dominant B cell epitope in K/BxN mice and in GPI-immunised mice

To map the dominant B cell epitopes in K/BxN and GPIA mice, we analysed serum responses to 28 GPI peptides using a multiplex assay. The response in the K/BxN mouse sera was limited to only three peptides, GPI293-307, GPI301-325 and GPI340-354, and the strongest antibody response was to GPI293-307 (figure 1A). Peptide GPI293-307 shares eight amino acids with GPI301-325, located at positions 301–307. The signal to peptide GPI301-325 was abolished and the signal to peptide GPI293-307 was substantially reduced by pre-incubating K/BxN mouse sera with peptide GPI293-307 (online supplemental figure S1A). This indicates that the dominant epitope is located at the overlapping peptide GPI301-307. Moreover, sera from GPI-immunised DBA/1 mice displayed a similar pattern to K/BxN serum, showing the strongest IgG response towards peptide GPI293-307 (figure 1B).

Figure 1

Serum antibody response to GPI peptides in mice. (A) Detection of IgG responses to 28 GPI peptides in sera of K/BxN mice by bead-based flow immunoassay. (B) DBA/1 mice were immunised with 100 µg of human GPI protein on day 0 and boosted on day 12 with 50 µg of human GPI protein. Sera were collected at DPI15. Sera IgG responses to 28 GPI peptides were detected by bead-based flow immunoassay. (C) B6NQ.Ncf1m1j/m1j mice were immunised with 100 µg of human GPI protein on day 0 and boosted on day 12 with 50 µg of human GPI. Sera were collected weekly postimmunisation until DPI 35, then at DPIs 56, 70 and 80. Time course of sera antibody response to GPI was detected by bead-based flow immunoassay. (D) Time course of sera IgG response to peptide GPI293-307 determined by bead-based flow immunoassay. (E, F) Time course of sera IgM, total IgG (IgG), IgG1, IgG2a and IgG2b responses to peptide GPI293-307 and GPI76-90 determined by ELISA. DPI, day post-immunisation; GPI, glucose-6-phosphate isomerase; MFI, median fluorescence intensity.

To further study the epitope GPI293-307, we followed the dynamics of the antibody responses after GPI immunisation and throughout the course of arthritis. We used B6NQ mice with a mutation of Ncf1 (B6NQ.Ncf1m1j/m1j ), which are highly sensitive to arthritis.22 Immunisation with human GPI protein led to 100% arthritis with disease severity peaking at day post-immunisation (DPI) 14 (online supplemental figure S1B,C). Mice were bled weekly. Serum IgG responses showed similar specificity to GPI peptides as K/BxN and GPI protein-immunised DBA/1 mice. Surprisingly, the highest levels of the IgG antibody response to GPI and GPI293-307 peptide were detected already at the earliest time point, DPI 7 (figure 1C,D, online supplemental figure S1D), which was also the earliest time point when arthritis symptoms appear (online supplemental figure S1C). The response to denatured GPI protein was higher than the response to native GPI protein. In the following experiments, we tracked the responses daily, not only total IgG, but also IgM, IgG1, IgG2a and IgG2b responses to peptides GPI293-307 and responses to the control peptide GPI76-90 (figure 1E,F, online supplemental figure S1E,F). Unexpectedly, GPI293-307-specific IgM was clearly detectable above background levels in naïve mice, and gradually decreased after immunisation. GPI293-307-specific IgG appeared at DPI 7, with IgG1 and IgG2b being the major IgG subclasses (figure 1F). Antibody reactivity to the negative control peptide GPI76-90 was not detected at any of the time points, and from DPI 14 onwards, antibodies reactive against other GPI peptides emerged (figure 1C). Therefore, we conclude that GPI293-307 is the dominant B cell epitope of GPI, and that the consistent early appearance and high levels of GPI293-307-specific antibodies in K/BxN and GPIA mouse models suggest that GPI293-307-specific antibodies are produced by naturally selected B cells and are associated with the development of arthritis.

The immunodominant B cell epitope of GPI is hidden in the native protein but exposed with unfolding

To study anti-GPI antibodies, 10 mAbs with different epitope specificity and IgG isotypes were generated by B cell hybridoma technology (table 1). mAbs TL1 and TL2 shared the same immunodominant epitope GPI293-307. The 10 clones were tested for their ability to bind native and denatured human and mouse GPI proteins (figure 2A,B). To determine the potential cross-reactivity to other joint-related proteins, COL2, the major cartilage protein, was included. The mAbs UL1 and M2139 were used as positive controls for the binding to COL2.33 None of the 10 mAbs cross-reacted with native or denatured COL2. mAbs TL1 and TL2 did not bind the native form of GPI protein but showed strong binding after protein denaturation (figure 2A,B). Other mAbs (eg, TL3 and TL7) bound strongly to both native and denatured GPI. To investigate antibody–antigen interaction in solution, we applied size exclusion chromatography and found that TL1 and TL2 could bind only to the denatured variant of GPI protein (figure 2C,D). In contrast, antibody TL7 complexed with both native and denatured GPI proteins. Consequently, mAbs TL3 and TL7 could capture circulating serum GPI protein whereas mAbs TL1 and TL2 could not (figure 2E).

Table 1

The characteristics of anti-GPI mAbs

Figure 2

Antigen specificities of anti-GPI mAbs. (A) Binding to native and denatured human and mouse GPI protein and COL2 of 10 mAbs were detected by ELISA. (B) The binding of mAbs TL1, TL2, TL3 and TL7 to native and denatured human and mouse GPI protein and COL2 was titrated and detected by bead-based flow immunoassay. (C, D) Human GPI protein was mixed with different antibodies in a 2:1 molar ratio in PBS followed by end-to-end rotation at 4°C overnight. The mixture was analysed by SEC. The single chromatograms were overlapped on one plot. (C) mAbs TL1, TL2 and TL7 complexed with denatured GPI eluted at a retention volume of 49 mL. TL1 alone eluted at 69 mL. (D) mAbs TL7 complexed with native GPI eluted at a retention volume of 59 mL. TL1 alone eluted at 69 mL. Non-complexed TL1 and TL2 (mixed with native GPI) eluted at 69 mL. Native GPI was eluted at a retention volume of 80 mL. (E) ELISA plate was coated with mAb TL7, followed by the addition of serum from healthy donor. Biotinylated anti-GPI mAbs (TL1/TL2/TL3/TL7) were used to test binding to the native GPI protein circulating in human serum. EU corresponds to OD620 nm measured by DELFIA. COL2, type II collagen; GPI, glucose-6-phosphate isomerase; mAbs, monoclonal antibodies; PBS, phosphate-buffered saline; SEC, size exclusion chromatography.

Structure of the TL1 Fab fragment in complex with GPI293-307

To bring light to the question why the GPI293-307 epitope is unique, we determined its molecular interaction with the TL1 antibody. The TL1 mAb bound GPI293-307 with high affinity (K d<6.23×10−13 M) (figure 3A,B). At 1.62 Å resolution, the crystals belong to space group P22121 and contain one Fab–peptide complex per asymmetric unit. Clear and continuous electron density is visible for residues of the light chain and the residues 1–132 and 136–216 of heavy chain of the TL1 Fab fragments. It consists of four canonical immunoglobulin domains (figure 3C), two formed by the heavy chain and two by the light chain, each stabilised by an intramolecular disulfide bond (H:22–H:96, H:144–H:199, L:23–L:93 and L:139–L:199).

Figure 3

Crystal structures of the TL1 Fab–GPI peptide complexes. (A, B) The binding affinity of mAb TL1 with peptide GPI293-307 was measured by SPR. (C) The TL1 Fab heavy and light chains are shown in teal blue and pink, respectively. The bound GPI293-307 peptide (shown as green sticks) is in the paratope formed by CDR loops from each heavy and light chain of the TL1 Fab. (D) The intra hydrogen bonds formed by GPI peptide. The dashed line indicates the formed hydrogen bonds. Carbon atoms are depicted in green; oxygen atoms are depicted in red and nitrogen atoms are depicted in blue. (E) The inter hydrogen bonds formed between complementarity-determining region (CDR) residues of TL1 Fab and peptide GPI293-307. The dashed line indicates the formed hydrogen bonds. Carbon atoms are depicted in green; oxygen atoms are depicted in red and nitrogen atoms are depicted in blue. (F) The orientation of GPI peptide at the binding site. The blue region indicates the positively charged potential while the negatively charged potential was shown in red. (G) The location of epitope GPI293-307 (depicted in magenta) on the GPI protein. GPI, glucose-6-phosphate isomerase; mAbs, monoclonal antibodies; SPR, surface plasmon resonance.

Clearly defined electron density is observed only for the C-terminal eight amino acids of the GPI293-307 peptide (300AHWMDQHF307), confirming the solid phase assay predictions, which adopt a β-hairpin-like structure in the groove (figure 3C). The β-hairpin conformation of the peptide is stabilised by a total of four intrapeptide hydrogen bonds (figure 3D). D304 forms three of these bonds, two via its carboxyl group with amide group of A300 and H301, and the other one to H301 via its amide group. In addition, H301 forms side chain atom-mediated hydrogen bonds with M303. The interaction between GPI293-307 and TL1 Fab is mainly mediated by several hydrogen bonds (figure 3E). Y31 and T96 from the light chain formed hydrogen bonds with A300 and W302, respectively. T30 and T53 from the heavy chain formed hydrogen bonds with Q305 in the peptide. In addition, the shape complementarity also drives the interaction, exemplified by the two pockets formed by paratope of TL1 for accommodating the W302 and M303 residues (figure 3F). Our in silico docking of TL1 to GPI around the peptide region suggested no feasible physical contact at the native conformation of GPI due to strong steric clash (figure 3G). In contrast, antibodies bound to native proteins (TL3–TL10) interacted with epitopes exposed on the surface of GPI proteins, as shown in online supplemental figure S2.

Thus, the binding structure explains why the antibody can only interact with a denatured or an unfolded protein, which will restrict or change its accessibility in vivo.

GPI293-307-specific B cells are not deleted in the bone marrow and are present in naïve mice

The isotype switch of serum anti-GPI293-307 antibodies from IgM to IgG within 1 week postimmunisation led us to hypothesise that GPI293-307-specific B cells are present in naïve mice. To confirm this, we used tetramers consisting of a fluorochrome-labelled streptavidin core and four identical biotinylated GPI peptides to identify the antigen-specific B cells in naïve mice, see gating strategy in online supplemental figure S3A. To exclusively identify the GPI293-307-specific B cells, we used GPI76-90-tetramer as a control. Dual-stained (APC/BV421+) GPI293-307-tetramer+ B cells were more frequent in bone marrow (BM), spleen (SP) and inguinal lymph nodes (iLN) of naïve mice compared with those detected by the control tetramer (figure 4A–C, online supplemental figure S3B–D). The frequency of this population was increased in the iLN of GPI protein-immunised mice compared with naïve mice, but not in BM or SP. However, the total cell number of this population was increased in all organs analysed (figure 4D,E). Most dual-stained GPI293-307-tetramer+ B cells were IgM-positive in naïve mice and GPI-protein immunisation reduced IgM expression on GPI293-307-specific B cells (figure 4F,G), indicating class switching. The specificity of the antigen-specific B cell staining was confirmed by successful inhibition by Strep-GPI293-37-PE tetramer (figure 4H,I). We conclude that GPI293-307-specific B cells escape negative selection, and we assume that they can be positively selected by an unfolded or modified form of GPI.

Figure 4

GPI293-307-specific B cells are detectable in naïve mice. (A) Representative flow cytometric analysis of dual tetramer staining (APC-and BV421-labelled) of GPI293-307-specific B cells and GPI76-90-specific B cells in the bone marrow (BM), spleen (SP) and inguinal lymph nodes (iLN) of naïve and GPI protein-immunised mice (DPI15). (B, C) Frequency and total cell number of dual tetramer positive peptide-specific B cells in the BM, SP and iLN of naïve mice. (D, E) Frequency and total cell number of dual tetramer positive GPI293-307-tetramer+ B cells in the BM, SP and iLN of naïve and immunised mice. (F) Representative flow cytometric analysis showing surface IgM expression on GPI293-307-tetramer+ populations in the iLN of naïve and immunised mice. (G) Frequency of IgM+ GPI293-307-tetramer+ B cells in the BM, SP and iLN of naïve mice and GPI protein immunised mice. (H–, I) Representative plots of GPI293-307-specific B cell staining in the BM and SP by dual tetramer after incubation with Strep-GPI293-307-PE or Strep-GPI76-90-PE tetramers (blocking). Results are expressed as the mean±SEM Statistical significance was assessed by Mann-Whitney test. In total, three independent experiments were performed, each experiment contained at least three mice per group. GPI, glucose-6-phosphate isomerase.

GPI293-307-specific IgG induce arthritis in mice and uniquely target disrupted cartilage

To assess the arthritogenicity of the 10 mAbs, we injected IgG antibodies with different GPI peptide specificity into naïve BQ.Cia9i mice, single or together with M2139. Single TL1 or TL2 induced mild arthritis, and when each of the antibodies were co-injected with M2139, the disease severity was enhanced compared with M2139 plus isotype controls (figure 5A,B, and online supplemental figure S4A–D). In contrast, none of the other antibodies which bind native GPI protein induced or enhanced arthritis (online supplemental figure S4E–H). On co-injection of TL1 with M2139 into more susceptible BQ.FcγR2b−/ mice,23 arthritis was already provoked on day 2, even without lipopolysaccharide (LPS) boost (figure 5C,D). To explain the arthritogenicity of the GPI293-307-specific mAbs, we next determined whether they target joint cartilage. The M2139 and TL7 groups were used as positive and negative staining reference, respectively. M2139 bound to cartilage from arthritic and naïve mice, as well as cartilage from patients with RA and controls (figure 5E, I–IV). The GPI293-307-specific mAb TL1 bound specifically to the cartilage surface of cryo-sectioned joint tissue from arthritic but not naïve mice (figure 5E, V–VI). Similarly, TL1 stained the disrupted cartilage surface of joint tissue from patients with RA, but not to joints from controls (figure 5E, VII–VIII) or osteoarthritis (OA) patients (online supplemental figure S5). mAb TL7 did not bind any cartilage (figure 5E, IX–XII). A summary of the number of samples used for IF staining and the percentage of positive staining for each group can be seen in online supplemental table S6. The non-pathogenic antibodies binding native GPI proteins (TL3, TL5, TL6, TL7, TL10) failed to stain cartilage from healthy or arthritic mice (online supplemental figure S6). To investigate whether TL1 exclusively bound cartilage, we stained other organ tissues (liver, skin, lung, kidney, oesophagus) from naïve and diseased mice, no positive staining could be found (online supplemental figure S7). Therefore, we conclude that the GPI293-307 epitope is pathogenic and uniquely exposed on the cartilage surface in arthritic joints.

Figure 5

Antibodies specific for the dominant GPI293-307 epitope specifically bind cartilage and induce arthritis. (A–D) Mean clinical score of arthritis severity (A, C) and incidence (B, D), 4 mg TL1/TL7 was injected into BQ.Cia9i mice (A–B) and BQ.FcγR2b-/- mice (C, D) alone or together with another 4 mg anti-COL2 mAb M2139 intravenously. BQ.Cia9i mice were boosted with LPS intraperitoneally 5 days after antibody injection. In total three independent experiments were performed, each experiment contained at least three mice per group respectively. (E) Representative images of immunofluorescence staining of sections from arthritic mouse ankle joints (I, V, IX), naïve mouse ankle joints (II, VI, X), cartilage explants of patients with RA (III, VII, XI) and controls (IV, VIII, XII) with FITC-conjugated anti-COL2 mAb M2139 or anti-GPI mAbs (TL1/TL7) (green); nuclei were counterstained with DAPI (blue). Scale bar=50 µm. COL2, type II collagen; GPI, glucose-6-phosphate isomerase; mAbs, monoclonal antibodies; RA, rheumatoid arthritis; *, p<0.05; ***, p<0.001; ****, p<0.0001.

Figure 6

(A, B) ROC curves of the response to GPI293-307 in early rheumatoid arthritis (eRA) patients and population controls in the Umeå study (AUC of 0.669 (asymptotic 95% CI 0.628–0.710)) (A) and in BARFOT+TIRA-2 studies (AUC of 0.705 (asymptotic 95% CI 0.683 to 0.727)) (B). (C) ROC curve of the response to an irrelevant GPI peptide (GPI325-339) in BARFOT+TIRA-2 studies. AUC and the cut-offs with corresponding specificity and sensitivity in percentages are shown. (D) IgG antibody response against GPI293-307 in patients with early RA, pre-symptomatic individuals and population controls in the Umeå study. The response (AU) and frequencies (%) in pre-symptomatic individuals are stratified for pre-dating time back to 6 years, one sample per individual closest to symptom onset (n=330). Significant differences of the frequencies of IgG antibodies between groups, calculated with the χ2 test, are presented in the figure. Cut-off level (10.23) for positivity is marked with a red line. n=number of cases at each time point. GPI, glucose-6-phosphate isomerase; RA, rheumatoid arthritis; ROC, receiver operating characteristics.

Figure 7

The erosion scores at 12 months in patients with early RA in the BARFOT cohort. The patients were grouped based on their serum IgG antibody response against GPI293-307. Patients were defined as ‘high’ (n=74) for being highly positive if their serum IgG antibody response against GPI293-307 was above the cut-off (11.0), and ‘low’ (n=1050) if their response was below the cut-off. Erosion scores are presented in boxplot, with box limits representing the 25–75th data percentiles; median represented as the black line inside the box. GPI, glucose-6-phosphate isomerase; RA, rheumatoid arthritis.

Increased IgG specific for GPI293-307 in pre-symptomatic individuals and patients with early RA

To determine if IgG antibodies with the same specificity as in the mouse also develop in human RA, we analysed cohorts of pre-symptomatic (Umeå) individuals and patients with early RA (BARFOT and TIRA2).28 29 Comparison of circulating GPI293-307-specific IgG levels in patients with RA and controls in the Umeå study (figure 6A) and in BARFOT and TIRA-2 cohorts combined (figure 6B, online supplemental figure S8) showed an increased response in patients, illustrated with ROC curves. In contrast, there was no increased IgG response in the BARFOT/TIRA-2 cohorts to an irrelevant GPI325-339 peptide (figure 6C).

The frequency of patients with antibodies to GPI293-307 among the patients with early RA in the Umeå study was 24.5%, and higher compared with controls (7.9%; OR=3.77 (95%CI 2.45 to 5.80)). Similar frequencies were observed in the combined BARFOT and TIRA-2 cohorts (24.5%), compared with the population controls (11.5%; OR=1.23 (95%CI 0.93 to 1.64). For confirmation, an additional cohort in patients with early RA (TIRA-1) and a control population cohort (WINGA) was assayed with similar results (online supplemental Table S7). In pre-symptomatic individuals the frequency of positivity was 10.2% when calculated for one sample per individual (n=518) and non-significant compared with controls. However, when the pre-symptomatic samples (one sample per individual closest to symptom onset) were stratified according to the pre-dating time, the frequency of positivity for GPI293-307 antibodies in pre-symptomatic individuals, just prior to RA onset (<−1 year), was 19.0 %, like patients with early RA and significantly different from controls (OR=2.72 (95%CI 1.31 to 5.63) (figure 6D).

To test the specificity of anti-GPI293-307 antibody response, we performed an inhibition assay by incubating RA serum with phosphate-buffered saline (PBS), GPI293-307, GPI76-90 peptides and hGPI protein before performing the bead-based flow immunoassay. GPI293-307-specific IgG response was specifically blocked by peptide GPI293-307 (online supplemental figure S9).

The prevalence of anti-GPI293-307 antibodies in other chronic inflammatory diseases (psoriatic arthritis, osteoarthritis and ankylosing spondylitis) was assayed as well, the detection rates in established RA was 30% (95% CI 20% to 29%) and 15.7% (95% CI 10% to 21%) in control diseases (online supplemental table S8), the results point to that these antibodies could be more important in RA but studies with larger sample size are needed to give a solid conclusion. As an overview, the results are summarised in online supplemental table S9.

Increased levels of IgG against GPI293-307 are associated with radiographic joint damage

Increased levels of anti-GPI293-307 antibodies before the onset of arthritis in our rodent and human studies suggested that high levels of anti-GPI293-307 antibodies may also be associated with joint destruction in this group of patients with RA. To address this question, we analysed bone destruction using patients from the well described cohort for early RA, the BARFOT cohort (n=1493). In total, n=1124 of these patients had recorded ESs at 12 months and of these, 43% had erosions. The patients highly positive to GPI293-307 (above the cut-off 11.0) had on average 1.78 units (p=0.009), 95% CI (Wald)=(1.15 to 2.74) higher ES compared with GPI293-307 negatives 12 months after inclusion, after accounting for anti-CCP antibodies at baseline (figure 7). Anti-CCP antibody negative patients were more likely not to have any erosion. Covariates including gender, smoking, age at inclusion, symptom duration, erosion at baseline and treatment change between 3 months and 6 months had no influence on the outcome.

To validate the association of increased levels of IgG against GPI293-307 with radiographic joint damage, we analysed the results using a lower cut-off (10.6; figure 6B), and the results point to the same direction: those with IgG response above the cut-off have 1.42 (95% CI 1.06 to 1.89; p=0.02) higher ES at 12 months compared with the ones below the cut-off.


A subset of patients with RA develops antibodies to the epitope GPI293-307, and we can now show that antibodies to this epitope induce arthritis in mice, by targeting a modified protein on the cartilage surface. The antibody response to this modified epitope is dominating in mouse models with an immune response to GPI, the K/BxN and GPIA models.

The question remains how the conformationally changed GPI is targeted on the joint articular surface. It has been described earlier that glycosaminoglycans (GAGs), which are abundantly expressed in the cartilage, is a substrate for GPI.34 Antibody binding to cartilage induces release of GAGs,35 which could optimise GPI binding and modification. Combining our cartilage staining results, we hypothesise that GPI293-307-specific antibodies bind to the conformationally changed GPI protein which accumulates on the joint articular surface. The deposition of anti-GPI antibody-immunocomplexes could trigger or enhance a local inflammatory attack. Interestingly, in contrast to B cells reactive with epitopes on native GPI, i.e., with conformational structures on the protein surface, B cells reactive to the hidden GPI293-307 epitope are not negatively selected. Rather, GPI293-307-specific B cells can be identified in naïve mice, and it is possible that they escape negative selection as they do not interact with the systemically occurring native GPI proteins. Instead, these B cells could be positively selected on modified GPI expressed in the bone marrow, in similarity with B cells to COL2.36

Previous studies with mAbs to GPI have been restricted to antibodies against the native protein and the studied antibodies were heavily somatically mutated and of high affinity.16 Expression of their BCR in a VDJ knock in mouse clearly shows that they undergo negative selection involving receptor editing.37

The role of the autoreactive GPI293-307-specific B cells is unknown, but if they get strong T cell help, such as in the K/BxN or GPIA mice, they differentiate to plasma cells38 and produce large amounts of pathogenic antibodies. This is supported by the observed rapid rise in antibody levels within only 1 week after immunisation. These antibodies are not neutralised by common circulating GPI protein but react with modified variants of GPI precipitating on the cartilage surface.

The GPI protein is highly conserved, and it is likely that a similar series of events could occur in RA. The frequency of GPI293-307-specific IgG antibodies in pre-symptomatic individuals just prior to RA onset was significantly higher than in controls, consistent with our mouse study in which GPI293-307-specific IgG antibodies emerged prior to the onset of arthritis. Approximately 20%–30% of the human participants had serum anti-GPI293-307 responses above the LoD, which constrains interpretations of the ROC curves generated including all observations. However, the cut-offs used for determination of antibody positivity were confirmed to be above the LoD, and thus the frequencies reflect reliably detectable levels. High levels of GPI293-307-specific IgG antibodies were associated with radiological bone ES, however, the association with radiological cartilage damage remains to be investigated.

We propose the designation of this type of antibodies as a new autoantibody class found in RA, in addition to ACPAs, the hallmark autoantibodies in RA. Whereas a pathogenic role of ACPAs so far is unclear,39 40 we provide evidence for pathogenicity and association with human RA of antibodies to a modified form of GPI.

We suggest that on exposure to unique genetic and/or environmental triggers, these modified GPI specific B cells interact with autoreactive T cells, and thereby activated and maturing into plasma cells, mounting an IgG antibody response, eventually leading to arthritis. It is likely that there are also other RA autoantigens with conformationally exposed neo-epitopes that have not been discovered yet, which could together explain the pathogenesis towards RA in a larger number of patients.

In summary, the characterisation of this novel autoantibody class in RA provides new unique insights about anti-GPI antibodies and unravels their relevance in RA.

Data availability statement

All data relevant to the mouse study are included in the article or uploaded as supplementary information. Data from several clinical studies (BARFOT, TIRA-2, Umeå, TIRA-1 and WINGA,) cannot be made publicly available due to ethical restrictions. The crystallographic coordinates and structure factors elucidated in this study have been deposited in the Protein Data Bank with the accession code 8BBH.

Ethics statements

Patient consent for publication

Ethics approval

This study involves human participants and was approved by The Regional Ethics Committee at the University Hospital, Umeå, Sweden approved Umeå study with reference number : Dnr 2013-347-31. The Regional Ethical Review Board, Stockholm, Sweden approved BARFOT study with reference number: Dnr T2016/297-31/1. The Regional Ethical Review Board, Linköping, Sweden approved TIRA-2 study with reference number: Dnr M168-05 and amendment numbers T35-07, T77-07 and T84-09. The Regional Ethical Review Board, Lund, Sweden approved MFM ÅUS study with reference number: Dnr 2017-758, 2017-09-29. JointID (Dnr 334-15 by the Regional Ethical Review Board in Gothenburg, Sweden). TIRA-1 (Dnr 96035 by the Regional Ethical Review Board Linköping, Sweden), WINGA (Dnr 676-08, T953-15 and 2016-03-23 by the Regional Ethical Review Board in Gothenburg, Sweden). Healthy donors (Dnr 2020-05001 by the Swedish Ethical Review Authority, Sweden). Collection of cartilage explant samples from healthy subjects or patients with RA was approved by the Regional Ethical Review Board Göteborg and by the Swedish ethical review board (Etikprövningsmyndigheten), Sweden (Dnr 334-15, 2015-05-18, T1075-17, 2017-12-18, 2019-04373, 2019-09-11,2020-07116, 2022-01977-02). Participants gave informed consent to participate in the study before taking part.


We thank Dr Fredrik Wermeling for generously providing K/BxN sera. We thank the Protein Science Facility at Karolinska Institutet for providing the crystallisation infrastructure and for recombinant protein production. We thank Payam Emami and Pär Engström at the National Bioinformatics Infrastructure Sweden at Science for Life Labratory for normalisation of the data for the TIRA-2, BARFOT and Umeå cohorts. We thank Dr Kristina Albertsson for reading radiographs in BARFOT. We thank Huqiao Luo and Beyza Betül Toker for technical help.


Supplementary materials

  • Supplementary Data

    This web only file has been produced by the BMJ Publishing Group from an electronic file supplied by the author(s) and has not been edited for content.


  • Handling editor Josef S Smolen

  • Twitter @taoli

  • Contributors TL, CG, RH contributed to the study design. TL, CG, AKr, KF, EL, AKa, SR-D, IG contributed to the data acquisition, data analysis and data interpretation. OS, MLA, LJ, MY contributed to the data analysis and data interpretation. TL, RH, OS, CG, AKr, MLA, LJ, KF, SR-D contributed to draft the work. BX, QL, LC, GB, GF contributed to the data acquisition. All authors revised the draft critically for intellectual content and approved the final version of the manuscript. RH has full responsibility for the work and conduct of the study, has access to the data and controlled the decision to publish.

  • Funding RH was supported by grants from the Knut and Alice Wallenberg Foundation (KAW 2019.0059), the Swedish Association against Rheumatism (R-757331) and the Swedish Research Council (2019-01209). KF was supported by grants from the Swedish Association against Rheumatism (R-931878) and the Foundation for Assistance to Disabled People in Skåne, Sweden. GB was supported by Heart and Lung foundation (20210383), the Swedish Research Council (2019-01140) and LUA/ALF: ALFGBG-718851.

  • Competing interests RH is the founder, Outi Sareila is an employee and Erik Lönnblom is a consultant of Vacara AB.

  • Patient and public involvement Patients and/or the public were not involved in the design, or conduct, or reporting, or dissemination plans of this research.

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

  • Supplemental material This content has been supplied by the author(s). It has not been vetted by BMJ Publishing Group Limited (BMJ) and may not have been peer-reviewed. Any opinions or recommendations discussed are solely those of the author(s) and are not endorsed by BMJ. BMJ disclaims all liability and responsibility arising from any reliance placed on the content. Where the content includes any translated material, BMJ does not warrant the accuracy and reliability of the translations (including but not limited to local regulations, clinical guidelines, terminology, drug names and drug dosages), and is not responsible for any error and/or omissions arising from translation and adaptation or otherwise.