Overexpression of Transmembrane TNF Drives Development of Ectopic Lymphoid Structures in the Bone Marrow and B Cell Lineage Alterations in Experimental Spondyloarthritis

This article is featured in Top Reads, p.

Spondyloarthritis (SpA) is characterized by axial and peripheral joint inflammation, bone destruction, and bone formation. The disease can also have extra-articular manifestations, such as in the skin, intestine, or eye (1, 2). Inflammatory features in SpA include enthesitis, synovitis, and immune cell infiltration in the adjacent bone marrow (BM) also known as osteitis, which can be detected by MRI as BM edema in the spine (3, 4). Inflammation and bone formation are closely linked processes (5), also in SpA (3, 6). However, it is difficult to study this relationship in humans because of ethical and practical limitations. Experimental SpA models are therefore crucial to unravel the mechanisms underlying inflammation and bone formation in SpA pathology.

Chronic inflammation in inflammatory diseases such as SpA has been studied extensively, and recently, several effective treatments have been developed. Anti-TNF therapy in particular has improved SpA treatment, being the first biologic treatment showing efficacy in SpA (1). Synovial tissue studies have shown that synovial inflammation, including large lymphoid aggregates, in SpA diminish after anti-TNF treatment (7), underlining the important role of TNF in the pathogenesis of SpA. However, not all patients with SpA respond to this therapy and the underlying mechanisms via which TNF contributes to SpA inflammation remain largely unknown.

Most experimental models overexpressing TNF mimic rheumatoid arthritis (RA) more than SpA. A mouse model with human TNF (hTNF) overexpression (hTNF transgenic [tg] mice) for instance presents with erosive polyarthritis without spinal inflammation or enthesitis (8). Of note, transmembrane TNF (tmTNF) seems to contribute more to SpA pathology than soluble TNF (sTNF) (9). Overexpression of murine sTNF by deleting adenylate-uridylate–rich elements in the murine TNF gene (TNF^ΔARE mice) leads to polyarthritis, bone erosions, and enthesitis, but these mice lack spinal involvement and bone formation (10). We have recently shown that tmTNF tg mice display virtually all SpA features, including chronic inflammation, bone destruction, and ectopic bone formation (9). In this model, a murine TNF gene with a noncleavable region for the enzyme cleaving tmTNF into sTNF, ADAM17 (a disintegrin and metalloproteinase 17), also called TNF converting enzyme (TACE), is overexpressed. As a consequence, these mice have a generalized increase in tmTNF, resulting in chronic inflammation accompanied by an influx of lymphocytes leading to aggregates at the enthesis and BM of ankles and vertebrae (9, 11). SpA is generally regarded as autoinflammatory (12), but several studies have sparked interest in adaptive immunity and the cellular immune response in particular, including alterations in B and T cell subsets in SpA patients (13, 14). B cell aggregates are found both in SpA BM and in synovial tissue (7, 15, 16), of which the latter have been described to contain T and B cell segregation as well as plasma cells (PCs) (7, 16). However, it remains unclear if these are a secondary bystander phenomenon, if these alterations support inflammation, or if they represent an autoimmune phenomenon. Furthermore, autoantibodies have been reported in SpA (17), although their clinical relevance is still unclear, and elevated serum IgA levels linked to disease activity have been described (18). Of note, a clear gut–joint axis exists in this disease, as both inflammatory bowel disease (IBD) and subclinical gut inflammation are highly prevalent in patients with SpA (19). TNF-RI, a receptor for tmTNF, was shown to be crucial for intestinal B cell follicle maturation (20). We and others have shown previously that tmTNF tg mice on a TNF-RII^−/− background still developed clinical SpA-like symptoms, whereas tmTNF tg × TNF-RI^−/− mice were protected from clinical signs and symptoms (9, 11).

As stated above, we have demonstrated clear pathological bone features in tmTNF tg mice that resemble human SpA and characterized the local inflammatory processes in tmTNF tg mice in relation to human SpA inflammatory features (9). However, the underlying mechanisms and importance of lymphoid aggregate formation remain unknown. In the current study, we investigated this in more detail and show that aggregates present in the BM precede the onset of arthritis and/or enthesitis and exhibit characteristics of ectopic lymphoid structures (ELS). This is accompanied by increased IgA⁺ PCs in the BM and increased circulating IgA concentrations. In addition, we demonstrate that ELS formation is predominantly dependent on TNF-RI.

TgA86 (tmTNF tg) mice have a murine TNF_Δ1–12-human β-globin hybrid gene construct containing the murine TNF-α gene promoter and were provided by Prof. Dr. G. Kollias (11). Breeding and housing were in specific pathogen–free conditions in the Academic Medical Center animal facility in accordance with national guidelines and legislation for animal breeding and housing. Breeding, housing, and experiments were approved by the Ethical Committee of Animal Welfare of the Academic Medical Center and the University of Amsterdam in accordance with the central authority for scientific procedures on animals (Centrale Commissie Dierproeven). tmTNF tg mice were on a C57BL/6J background. Heterozygous tmTNF tg and non-tg littermates were used for the experiments. Heterozygous tmTNF tg mice were crossed with TNF-RI^−/− (Tnfrsf1a^tm1Imx) or TNF-RII^−/− (Tnfrsf1b^tm1Mwm). TNF-RI^−/− and TNF-RII^−/− were purchased from The Jackson Laboratory. Both male and female mice were used in these experiments. All mice were genotyped by PCR (11).

Paraffin-embedded vertebrae were cut in 5-µm sections after overnight 4% formalin fixation and 4 wk of decalcification (osteosoft; Merck). Gut tissue was embedded in paraffin and cut after formalin fixation. Vertebral sections were stained with H&E (Sigma-Aldrich), whereas gut sections were stained with IgA (1:12.000, NB7506; Novus) and hematoxylin. Tissue preparation for immunofluorescence (IF) imaging was modified from the protocol previously described (21). In brief, freshly dissected murine paws and vertebrae from tmTNF tg mice and non-tg same sex littermates were fixed in ice-cold 2% paraformaldehyde for 2–6 h, depending on the age of the mice. The bones were decalcified with 0.5 M EDTA in PBS (pH 8) at 4°C for 24–48 h. The tissues were cryoprotected in a 20% sucrose and 2% polyvinylpyrrolidone PBS solution overnight. The tissues were embedded in a 20% sucrose, 10% gelatin, and 2% polyvinylpyrrolidone PBS solution at room temperature for up to 1 h. The solidified molds were stored at −80°C. Cryosections of 60 µm thick for IF and morphological stainings were cut on a CM1950 Leica cryostat. Specimen and chamber temperature were −23°C. Thawed sections were air-dried, followed by permeabilization with 0.2% Triton X-100 in PBS for 20 min and blocking in 5% donkey serum (Jackson ImmunoResearch Laboratories) and 0.2% Triton X-100 in PBS for 30 min, all at room temperature. Primary Ab incubation in 5% donkey serum in PBS was performed overnight at 4°C with the following: Alexa Fluor 594–conjugated B220 (1:200, RA3-6B2; BioLegend), CD138 (1:100, 142501; BioLegend), Alexa Fluor 488–conjugated CD3 (1:200, 17A2; BioLegend), CD31 (1:100, AF3628; R&D Systems), Alexa Fluor 647–conjugated IgA (1:400, 1040-31; SouthernBiotech), DyLight 633–conjugated MECA-79 (1:400, kindly provided by Prof. Dr. E. Butcher), and biotin-conjugated peanut agglutinin (PNA; 1:1000, B-1075; Vector Laboratories). The next day, sections were washed three times with PBS for 5 min and counterstained with appropriate Alexa Fluor–conjugated secondary Abs (1:500; Invitrogen) or streptavidin (1:1000; Invitrogen) for 2 h at room temperature. Nuclei were counterstained with Hoechst (1:1000; Invitrogen). Sections were washed three times with PBS for 5 min and mounted with Fluoromount-G (Thermo Fisher Scientific). Sections were analyzed with a Leica TCS SP8 × mounted on a Leica DMI6000 inverted microscope. Z-stacks of the images were acquired with LAS-X software (version 4.9.0). Adobe Illustrator was used for image processing according to The Journal of Immunology’s guide for digital images.

Vertebrae and femora were dissected, and surrounding soft tissue was removed. Bones were crushed in PBS containing 0.5% BSA and 2 mmol EDTA with a mortar and pestle. Single-cell suspensions were acquired using 100-µm cell strainers. Fluorescently labeling of 2 × 10⁶ cells with mAbs (Supplemental Table I) was performed as published previously (22). To stain intracellular Igs in PCs, suspensions were first fixed with BD Biosciences Cytofix/Cytoperm Buffer and permeabilized with Perm/Wash Buffer. Stainings were measured with a BD Biosciences LSR-II flow cytometer. Analysis was performed using FlowJo software (version 10; BD Biosciences).

Serum Ig levels from tmTNF tg, tmTNF tg × TNF-RI^−/−, tmTNF tg × TNF-RI^−/−, and non-tg mice were measured using sandwich ELISA. IgA (BD Biosciences) and IgM, IgG1, IgG2a, IgG2b, and IgG3 (all from SouthernBiotech) assays were performed as previously described (23).

The data are shown as mean ± SEM. Significant differences of mean values were calculated with Mann–Whitney U test; p < 0.05 with a 95% confidence interval was considered as statistically significant. Statistical analysis was performed with GraphPad Prism (version 8).

We have recently shown that tmTNF tg mice develop arthritis accompanied with bone remodeling in the axial and peripheral skeleton, which is highly reminiscent of human SpA (9). Histological signs of inflammation include immune cell infiltration and cellular aggregates, mostly adjacent to the cortex of the bone, in affected vertebrae (Fig. 1A). As a sign of enthesitis in the spine, infiltrating B220⁺ B cells and CD3⁺ T cells can be found along the enthesis in connective tissue (Fig. 1B). The aggregates located in the BM, which are absent in non-tg mice, consist of CD3⁺ T cells and in particular B220⁺ B cells (Fig. 1C). The lymphoid aggregates can already be found at the age of 3 wk, before onset of disease symptoms. During aging, the lymphoid aggregates increase in size and exhibit more B and T cell areas (Fig. 1D). The aggregates also increase in number to ∼10 per vertebra at the age of 8 mo (Fig. 1E). Interestingly, BM lymphoid aggregate formation appears to be site specific, as they were only present near inflamed sites, including the vertebrae (Fig. 1B–E) and the ankle joints (Fig. 1F), but were absent at other sites such as the femur (Supplemental Fig. 1A). Therefore, tmTNF overexpression causes lymphoid aggregates that predominantly consist of B cells and are linked to sites of inflammation in joints and spine.

In human SpA, we previously reported that aggregates can show germinal center (GC) and ELS characteristics (7, 16). Therefore, we assessed the expression of markers characteristic of GC and ELS at the age of 12 wk. The BM lymphoid aggregates in tmTNF tg mice are close to blood vessels. The CD31⁺ vessels show colocalization with MECA79 (PNAd), a marker for high endothelial venules (HEVs) (Fig. 2A). HEVs are associated with extensive naive lymphocyte recruitment and migration. HEVs are normally not present in BM but typically found in secondary lymphoid organs and ELS that can be found in chronic inflammatory diseases, including arthritis, and cancer (24, 25). Of note, the HEVs were already present in 3-wk-old mice and were more apparent around the aggregates than in the 8-mo-old mice (six out of six animals analyzed; data not shown). The lymphoid aggregates clearly displayed staining for the GC marker PNA (Fig. 2B). In addition, CD138⁺ PCs were found around the aggregates in the tmTNF tg BM (Fig. 2C). These results show that tmTNF overexpression induces formation of lymphoid aggregates that resemble ELS.

To further study the BM ELS caused by tmTNF overexpression, we first defined the B cell subsets of tmTNF tg and non-tg control mice in vertebrae and femora. Flow cytometric analysis confirmed the IF results and demonstrated that the proportion of B cells was specifically increased in vertebral, but not femoral BM of tmTNF tg mice (Fig. 3A, gating strategy in Supplemental Fig. 2A). Consistent with the presence of ELS and therein expressed GC markers (Fig. 2), a significantly increased proportion of IgD⁻CD95⁺ GC B cells was found in the tmTNF tg vertebrae in comparison with non-tg vertebrae (Fig. 3B). The increased expression of the costimulatory markers CD80 and CD86 on tmTNF tg GC B cells confirms that these cells are activated (Fig. 3C). Because memory B cells can also express CD95, we gated for CD80 and PD-L2 within the IgD⁻CD95⁺ B cell fraction. Only a small proportion of IgD⁻CD95⁺ B cells (<10%) had high CD80 and PD-L2 expression, compatible with memory B cells (Supplemental Fig. 2B). There was also a clear segregation of CD86⁺CXCR4⁻ light zone GC B cells and CD86⁻CXCR4⁺ dark zone B cells in the tmTNF tg mice (Fig. 3D). Meanwhile, early B cell development was unaffected in tmTNF tg mice, both in femora and in vertebrae (Supplemental Fig. 1B, 1C). All together, these findings show that ELS formation in tmTNF tg BM coincides with the expansion and activation of GC B cells.

Because we identified GC B cells in tmTNF tg vertebrae, we also examined the presence of follicular T cells in the vertebral ELS. Flow cytometric analyses demonstrated that the proportion of CD4⁺ T cells of total BM cells was similar between tmTNF tg and non-tg mice for both vertebrae and femoral BM (Fig. 4A). However, within the T cell population of tmTNF tg femur and vertebrae, a significant increase in the proportion of CD4⁺ T cells was observed (Fig. 4A). This tmTNF tg–induced CD4⁺ T cell skewing was more prominent in vertebrae (56%) in comparison with femoral BM (24%) (Fig. 4A). Upon examining the vertebral CD4⁺ T cell compartment of tmTNF tg mice, we identified an increase in both CXCR5⁺PD-1^highFoxp3⁻CTLA4⁻ T follicular helper (Tfh) cells and CXCR5⁺PD-1^highFoxp3⁺CTLA4⁺ T follicular regulatory (Tfr) cells (Fig. 4B). ICOS expression by PD-1^high CD4⁺ T cells was increased in the tmTNF tg vertebrae (Fig. 4C), which could imply an enhanced Th response, Tfh cell development, and GC reaction in general (26). CXCR5⁻PD-1⁻Foxp3⁺ T regulatory cells (Tregs) were also increased compared with the non-tg littermates (Fig. 4D). Collectively, CD4⁺ T cell skewing and increased presence of Tfh and Tfr cells support the notion that these BM aggregates are true ELS.

The vertebral tmTNF tg ELS appeared to be functionally active, as they contain GC B and T cells. To study their output, next we examined the specific PC subsets present in the vertebrae of tmTNF tg mice. In line with the presence of PCs in areas surrounding the ELS (Fig. 2C), flow cytometry data revealed a trend toward increased PCs (CD11b⁻TACI⁺CD138⁺) in the vertebrae of tmTNF tg mice in comparison with non-tg mice (Fig. 5A). To examine Ig class switching, we investigated the proportions of IgA-, IgG-, and IgM-expressing PCs. Interestingly, there was a significantly increased proportion of IgA⁺ PCs in vertebrae of tmTNF tg mice, indicating an altered class switching in favor of IgA (Fig. 5A, gating strategy Supplemental Fig. 2C). The same IgA-directed shift was present in the tmTNF tg femur (Fig. 5B). The increase in IgA⁺ cells in vertebral tmTNF tg BM was further demonstrated by IF microscopy (Fig. 5C). We did not find a significant increased number of IgA⁺ cells in the gut (Fig. 5D). The increase of IgA⁺ cells in these tissues was associated with systemic (5.3-fold) increase of IgA serum levels in tmTNF tg compared with non-tg mice. In addition, IgG1 was also increased, whereas IgG2b, IgG2c, IgG3, and IgM levels were not different between tmTNF tg and non-tg mice (Fig. 5E). When taken together, these data clearly show IgA-skewed class switching and expansion of PCs in the BM of tmTNF tg mice.

sTNF, but also tmTNF, can activate several intracellular signaling pathways by binding to TNF-RI or TNF-RII (27). To determine the contribution of TNF-RI and TNF-RII to BM ELS formation in tmTNF tg mice, we crossed tmTNF tg mice on either a TNF-RI– or TNF-RII–deficient background. ELS were absent in tmTNF tg × TNF-RI^−/− vertebrae, whereas B220⁺ aggregates and HEVs were still present in vertebrae of tmTNF tg × TNF-RII^−/− mice (Fig. 6A, 6B). However, the number of aggregates per vertebra was significantly lower in tmTNF tg × TNF-RII^−/− mice, suggesting that both receptors contribute to ELS formation through tmTNF overexpression (Fig. 6C). In line with this, the proportion of GC B cells in the vertebral BM in these mice was reduced in comparison with tmTNF tg mice (Fig. 6D). Meanwhile, all tmTNF tg mice show increased CD4⁺ T cell skewing within the T cell population, although this is not significant for tmTNF tg × TNF-RII^−/− mice (Fig. 6E). Analogous with the phenotype and the absence of lymphoid aggregates, vertebral PCs in tmTNF tg × TNF-RI^−/− mice were not increased or skewed to the IgA subclass (Fig. 6F). Although ELS formation was reduced in tmTNF tg × TNF-RII^−/− mice, IgA-skewed class switching remained present (Fig. 6F). Serum IgA levels were not increased in tmTNF tg mice on either the TNF-RI– or TNF-RII–deficient background (Fig. 6F). These data demonstrate that TNF-RI, and to a lesser extent TNF-RII, signaling contribute to ELS formation in tmTNF tg mice.

In the current study, we demonstrate that overexpression of tmTNF in mice results in the induction of ELS in vertebral BM adjacent to sites of inflammation and increased IgA⁺ PC differentiation, which is mediated by TNF-RI–induced signaling events. TNF is a well-known initiator of ELS formation (24), and TNF inhibition can reduce ELS in SpA (7). Several studies in TNF-based, immune-mediated inflammatory disease mouse models and human immune-mediated inflammatory diseases show that leukocyte infiltration and GC formation are present at sites of inflammation and associated with disease severity (7, 24, 28, 29). However, the mechanism by which TNF contributes to the pathogenesis of inflammatory diseases, including SpA, is not fully understood.

Under physiological conditions, ELS formation does not occur in BM (24). In this study, we show that tmTNF can induce ELS in the BM that are PNA⁺ and contain HEVs. Studies in human SpA focusing on inflammation in BM are limited (15, 30–32). However, inflamed SpA and RA synovial tissue can also contain lymphoid aggregates that show great similarity with the ELS found in the tmTNF tg mice, including B cell dominance and CD138⁺ PCs surrounding the aggregates (7). We observed a gradual increase in separate T and B cell zones with age, conforming to the idea that ELS are organized structures that develop and mature over time (24). Interestingly, aggregates and HEVs are already visible in young tmTNF tg mice, even before onset of arthritis. These HEVs occupy a larger proportion of the BM compared with older mice, in line with the fact that HEVs are important in ELS formation (25). Stromal tmTNF overexpression is sufficient to cause the inflammatory phenotype in the tmTNF tg model (9). However, the relative contribution of tmTNF overexpression on specific stromal cells and lymphocytes to the observed pathological features is incompletely understood. Whether the ELS possess the ability to promote recruitment, proliferation, and activation of lymphoid cells also merits further investigation.

We found that lymphoid aggregates in the BM of tmTNF tg mice are rich in B cells and that the BM infiltrates are linked to site of inflammation and disease severity. Of interest, hTNF tg mice have lymphoid aggregates in the BM adjacent to cortical bone that contribute to inflammatory joint pathology. These aggregates were located in the ankles and were rich in B cells (28). Both TNF tg mouse models exhibit similar histology, although ELS characteristics have not been shown in hTNF tg mice. The models are representative of two different clinical diseases, SpA and RA, respectively. hTNF tg mice have polyarthritis without involvement of the spine, whereas this site is severely affected in tmTNF tg mice (28). Intriguingly, the earlier mentioned TNF^ΔARE mice lack lymphoid aggregates in BM (10). Therefore, tmTNF and sTNF exert unique functions in the pathogenesis of these diseases, including differential effects on the process of ELS formation.

SpA is generally considered an autoinflammatory disease with dysfunctional innate immune reactions (12). However, adaptive immune system alterations have been reported as well (7, 13, 14). In this study, we show that the tmTNF tg mouse model of SpA also has adaptive immune system abnormalities. The presence of GC B cells and Tfh cells in vertebral BM suggests functional ELS activity. The increase in both Tregs and Tfr cells in our model should also be noted, although it is unknown if these regulatory cells are functional and whether they truly regulate inflammation, as impaired function of Tregs has been described in human SpA (33). Interestingly, an elevated number of Tfr and Tfh cells have also been found in the peripheral blood of SpA patients (34).

Elevated serum IgA has been linked to disease activity in SpA patients (18). In this study, we show that tmTNF tg mice also have increased BM PC formation and IgA-skewed class switching and an increase in serum IgA levels. The increase in serum IgA was tmTNF specific because hTNF tg mice did not have elevated IgA compared with non-tg mice (data not shown). The increase of IgA⁺ PCs in the tmTNF tg BM may (in part) be due to recirculation (35), as we also found increased IgA⁺ PCs in femoral BM. Moreover, we did not observe a significant increase in the number of IgA⁺ cells in the gut, suggesting that the BM is the primary source, and the increase in serum IgA may therefore be a reflection of osteitis rather than of gut inflammation. Tfh cells and Tregs are also associated with increased IgA⁺ PCs through different mechanisms (36, 37). In addition, IgA class switching in vivo could be due to a combination of many factors, such as CD40L, TGF-β, IL-21, BAFF, APRIL, and TBK1 (38–40). A previous study on SpA and RA synovial fluid reported a link between IL-17 and IgA that might be induced through TGF-β and BAFF (41). In the current study, we evaluated IL-17 expression in femoral and vertebral B and T cells after PMA–ionomycin stimulation but did not find an increased IL-17 expression in tmTNF tg compared with non-tg mice (data not shown). The role of IgA in SpA pathogenesis remains unclear, as IgA can both passively protect against and actively induce inflammation. It can, for instance, promote recruitment and/or activation of neutrophils (42), which are involved in SpA pathogenesis (12) and found abundantly in inflamed tmTNF tg tissues (43). In addition, recent studies have reported IgA anti-CD74 autoantibodies (17, 44) and a strong IgA antioxidative collagen type II response (45) in SpA. Patients with psoriatic arthritis, a subset of SpA, have elevated serum IgA, a decrease in serum IgM, and increased IgA Abs against the gut-associated gliadin. Those Abs seem to be associated with active disease (46). Nevertheless, we were not able to detect autoantibodies in the tmTNF tg serum by using HEp-2 IF staining for IgA, IgG, and IgM or by incubating Rag1^−/− BM sections with tmTNF serum and non-tg serum (data not shown).

IgA is best studied for its role in mucosal tissues, and an excess is associated with several diseases, such as IBD and IgA nephropathy (2, 42, 47). Both IBD and IgA nephropathy are linked to SpA (19, 47). TNF blockade is effective in treating diseases such as IBD and SpA, and the cytokine is regarded as important in the pathogenesis of IgA nephropathy as well (48). In the current study, we also observed a moderate increase in the number of IgA⁺ cells in the gut of tmTNF tg mice compared with non-tg mice, but this was not significant and much less pronounced than in BM. It would still be of interest to further examine the gut–joint axis in these mice (19, 49). Whether IgA autoantibodies contribute to immune complexes and mediate inflammation in SpA still remains unknown. B cell–depleting treatments such as anti-CD20 (rituximab) seem to have only a modest effect in SpA, although no controlled studies have been conducted thus far, and these therapeutic strategies do not deplete PCs (50). However, B cells may be involved primarily in the early stages of SpA and could be an interesting therapeutic target if disease is diagnosed early or perhaps in (future) preventive settings.

The gene encoding TNF-RI has been associated with SpA (51), and TNF-RI signaling seems key for inflammation in TNF mouse models (9–11), whereas tmTNF can activate both TNF-RI and TNF-RII (27). TNF-RI signaling is crucial for intestinal isolated lymphoid follicle maturation (20). In line with these findings, we did not observe ELS or IgA class switching in tmTNF tg × TNF-RI^−/− mice, whereas tmTNF tg × TNF-RII^−/− did develop ELS and IgA⁺ PCs. This suggests that TNFR-I, but not TNFR-II, signaling is required for the observed skewed IgA class switching in tmTNF tg mice. However, the absence of vertebral BM ELS formation in tmTNF tg × TNF-RI^−/− hampers the study of IgA skewing in these mice. Furthermore, both tmTNF tg × TNF-RI^−/− and tmTNF tg × TNF-RII^−/− have a similar reduction on IgA secretion in comparison with tmTNF tg mice. It might be that this correlates with the lower amount of vertebral BM ELS in tmTNF tg × TNF-RII^−/− in comparison with tmTNF tg mice. However, additional TNF-RI/RII effects leading to reduced IgA serum levels may be involved as well. Together, this demonstrates a crucial role for TNF-RI and, to a lesser extent, TNF-RII in ELS formation and pathology in this SpA mouse model.

Importantly, both BM edema and increased entheseal vascularization is present in SpA, in line with ELS and HEVs in the tmTNF tg model (3, 9). Both in SpA-like mice and human SpA patients, BM abnormalities and inflammation occur at locations of biomechanical stress, such as in the spine and at the enthesis (3, 9, 52). Although the vertebral inflammation in human SpA has not been fully characterized, T and B cell aggregates have been reported (15, 32). Meanwhile, in tmTNF tg mice, the alterations in the femur are very limited. It is intriguing that not all (axial) joints show arthritis, enthesitis, and/or BM infiltration, which warrants further research into the role of location-specific triggers and biomechanical stress in ELS formation. BM abnormalities may also promote pathological bone formation in SpA, whereas this does not occur at unaffected sites (4, 28, 32). Although the sequence of the inflammatory and bone-forming events is not entirely clear in human SpA, abnormalities in the BM are also associated with, and may even promote, bone formation and other pathological features in human SpA (4, 6).

This study shows that the tmTNF tg mouse model of SpA strongly resembles human disease, including BM inflammation and increased IgA levels. tmTNF overexpression leads to formation of ELS containing activated GC B cells and increased numbers of Tfr and Tfh cells in the BM at sites of inflammation. Furthermore, tmTNF tg mice exhibit enhanced IgA⁺ class switching of PCs in the BM that may account for the increased IgA serum levels rather than gut inflammation. These tmTNF-induced pathological findings are primarily dependent on TNF-RI signaling, which may help to elucidate SpA pathogenesis and could serve as a potential target for novel therapies.

We thank Daisy I. Picavet (Amsterdam University Medical Center), Susanne Adam (Friedrich-Alexander University Erlangen-Nürnberg and Universitätsklinikum Erlangen Friedrich-Alexander University Erlangen-Nürnberg), Marjolein J. W. de Bruijn, Jennifer A. C. van Hulst and Stefan F.H. Neys (Erasmus University Medical Center, Rotterdam) for expert technical assistance. We also thank George Kollias for providing the tmTNF tg mice.

D.L.P.B became an employee at UCB Pharma at the time these experiments were conducted. The other authors have no financial conflicts of interest.