Ankylosing spondylitis macrophage production of higher levels of interleukin-23 in response to lipopolysaccharide without induction of a significant unfolded protein response




Previous studies of the HLA–B27–transgenic rat model of ankylosing spondylitis (AS) suggested that macrophages develop an intracellular stress response called the unfolded protein response (UPR) and, as a result, secrete increased amounts of cytokines in response to Toll-like receptor agonists such as lipopolysaccharide (LPS). Our objective was to determine whether macrophages from AS patients also undergo a UPR and secrete increased cytokines/chemokines in response to LPS.


Peripheral blood monocytes isolated from 10 AS patients and 10 healthy controls were differentiated in vitro with macrophage colony-stimulating factor. Select samples were treated with interferon-γ (IFNγ) to up-regulate class I major histocompatibility complex (HLA–B) expression prior to stimulation with LPS for either 3 hours (for RNA) or 8–24 hours (for supernatant). UPR induction was assessed by measuring the expression of messenger RNA for ERdj4, BiP, and CCAAT/enhancer binding protein homologous protein 10 (CHOP).


Although IFNγ treatment up-regulated HLA–B expression (2-fold; P < 0.0001), neither IFNγ nor LPS substantially enhanced BiP or CHOP expression (<1.3-fold). ERdj4 expression increased weakly, but not significantly, in AS samples treated with IFNγ plus LPS (2.2-fold; P = 0.31). In response to LPS, AS macrophages secreted more CXCL9, interleukin-10 (IL-10), IL-12p70, IL-23, and tumor necrosis factor α than did control macrophages (P ≤ 0.025). The most striking difference was observed for IL-23 (median 265 pg/ml in AS patients versus 9 pg/ml in controls; P = 0.0007). We did not detect significant differences in IL-6, IL-8, or IFNβ production.


The greater production of IL-23 by AS patient macrophages in response to LPS provides further support for the development of Th17/IL-23–directed therapy. Since significant UPR induction was not detected in AS patient macrophages, the relationship between UPR and inflammatory cytokine production remains unclear.

Ankylosing spondylitis (AS) is an insidious spinal inflammatory disease that affects young adults and incurs a high rate of disability (1, 2). Inflammation can also involve the eyes, skin, gut, peripheral joints, and tendinous insertions (entheses). AS is a complex genetic disease, with the presence of HLA–B27, the class I major histocompatibility complex (MHC) allele, conferring up to 40% of genetic risk; this allele is found in 90–95% of AS patients but only 8–9% of the Caucasian population (3–6). HLA–B27/human β2-microglobulin–expressing rats (B27-transgenic) were shown to develop a spondylarthritis-like disease involving gut inflammation and joint swelling that mimics human disease, highlighting the importance of B27 as a causative factor (7). However, almost 40 years after the strong linkage to B27 was discovered, it is still not clear how this allele predisposes to disease (8). Although the physiologic role of the class I MHC is to present intracellular peptides to CD8+ T cells, depletion of CD8+ T cells in the B27-transgenic rat does not affect disease (9).

One hypothesis for a B27-related pathogenesis of AS stems from the observation that HLA–B27 displays an unusual propensity to fold slowly and to misfold during its biosynthesis (10, 11). B27 forms aberrant dimers that can be recognized by leukocytes on the cell surface and that accumulate in the endoplasmic reticulum (ER) (12–14). Accumulation of misfolded protein in the ER triggers a conserved intracellular stress response, the unfolded protein response (UPR), that decreases protein production, enhances the size and folding capacity of the ER, and degrades excess ER protein. The UPR profoundly alters cellular metabolism, predominantly through the induction of gene expression (15).

Evidence of B27 induction of the UPR, including up-regulation of messenger RNA (mRNA) for binding immunoglobulin protein (BiP)/glucose-regulated protein 78-kd (Grp78), CCAAT/enhancer binding protein homologous protein 10 (CHOP), and UPR-dependent splicing of the X-box binding protein 1 (XBP-1) transcription factor, has been detected in cells transfected with B27, as well as in bone marrow macrophages from diseased B27-transgenic animals (16). In the animal model, the magnitude of UPR induction correlated strongly with B27 expression; in premorbid animals, visualization of the UPR required acute up-regulation of the class I MHC by inflammatory cytokines, such as tumor necrosis factor α (TNFα) and interferon (IFN) (16, 17). The clinical relevance of these observations is supported by BiP overexpression in synovial fluid mononuclear cells from the knee joints of B27+ spondylarthropathy patients (18). However, the relationship between B27-triggered UPR and disease remains unclear: in B27-transgenic rats with intermediate levels of B27 expression, additional human β2-microglobulin induces arthritis and spondylitis in the absence of intestinal inflammation, while simultaneously reducing B27 misfolding and BiP expression, as demonstrated in concanavalin A–treated spleen cells (19).

Our previous studies suggested a link between B27, the UPR, and innate immune responses. Macrophages stimulated with pharmacologic inducers of UPR (e.g., tunicamycin) responded to Toll-like receptor (TLR) agonists, such as lipopolysaccharide (LPS; endotoxin), with increased production of inflammatory cytokines, in particular IFNβ and interleukin-23 (IL-23) (20, 21). Bone marrow macrophages from diseased animals expressing UPR target genes also expressed type I IFN–regulated genes (16). B27+ macrophages from transgenic rats stimulated in vitro with IFNγ (to induce the UPR) produced increased levels of IFNβ and IL-23 in response to LPS. IL-23 has been shown to drive IL-17 production by pathogenic Th17 cells and is implicated in multiple autoimmune and inflammatory diseases (22). Involvement of the IL-23/Th-17 axis in the pathogenesis of AS is further supported by studies of the B27-transgenic rat colon as well as colon biopsy and peripheral blood mononuclear cell (PBMC) samples from AS patients (21, 23, 24).

Macrophages predominate in biopsy tissues obtained from early spondylarthritis inflammatory lesions (25). Also, in the B27-transgenic rat model, macrophages appear to be more susceptible than splenocytes to induction of the UPR (17). For this study, we examined peripheral blood–derived macrophages from AS patients to determine whether they undergo a UPR and if so, whether UPR induction correlates with increased cytokine/chemokine responses to LPS. Despite a 2-fold up-regulation of HLA–B by treatment with IFNγ, we did not observe significant UPR induction, as measured by the expression of mRNA for BiP, CHOP, or endoplasmic reticulum–localized DnaJ homolog 4 (ERdj4; an XBP-1 gene target). Except for IL-12 secretion, IFNγ treatment minimized differences between the AS patients and healthy controls. In response to LPS alone, even in the absence of obvious UPR induction, AS patient macrophages produced strikingly higher levels of IL-23 (P = 0.0007) than did control macrophages. These results have implications for AS pathogenesis and further support the development and application of anti–IL-23 therapeutic agents in AS.


Study subjects.

Nine patients meeting the modified New York criteria for a diagnosis of AS (26) were included in the study along with 1 B27+ patient with thoracic syndesmophytes and sacroiliac pain. None of the patients were related to each other. We were unable to collect culture supernatants from 1 AS patient sample. In another AS patient sample, we only had enough cells to collect RNA for the no stimulation and IFNγ conditions only (final numbers per group are indicated in the figures and tables). The healthy control subjects were unrelated to the AS patients and had no personal or family history of AS. Table 1 shows the characteristics of the AS patients and control subjects.

Table 1. Characteristics of the AS patients and healthy control subjects*
CharacteristicAS patients (n = 10)Healthy controls (n = 10)
  • *

    AS = ankylosing spondylitis; NA = not applicable; BASDAI = Bath Ankylosing Spondylitis Disease Activity Index; NSAIDs = nonsteroidal antiinflammatory drugs.

  • One control subject was taking 325 mg of aspirin daily because of a history of myocardial infarction. An additional 3 control subjects (not included here) were taking antiplatelet doses of aspirin (81 mg/day).

  • A total of 6 patients were taking disease-modifying antirheumatic drugs (DMARDs): methotrexate (MTX) and/or tumor necrosis factor (TNF) blockers.

Age, median (range) years47 (18–58)45 (28–57)
Sex, % male9090
Disease duration, median (range) years15 (2–32)NA
No. positive for HLA–B2790
BASDAI, median (range)3.2 (0–5.4)NA
Arthritis medications  
  TNF blockers50
 OtherSulfasalazine in 1; glucosamine in 1; alendronate in 1Allopurinol in 1; glucosamine in 1

All study subjects were recruited from within the University of Wisconsin Hospitals and Clinics system. Informed consent was obtained prior to participation, and the study protocol was approved by the University of Wisconsin Health Sciences Institutional Review Board.

Macrophage derivation and in vitro treatments.

Whole blood was obtained by venipuncture and collected into three 10-ml tubes containing acid citrate dextrose (BD Biosciences). Samples were processed within 18 hours of collection. Samples were diluted with phosphate buffered saline, underlaid with lymphocyte separation medium (Mediatech), and centrifuged to obtain PBMC buffy coats. For non-study control experiments (e.g., tunicamycin treatment), unidentified healthy PBMCs were obtained from a clinical research core facility. Cells were washed twice at 900 revolutions per minute to decrease the number of platelets. Monocytes were purified by negative selection using magnetic-activated cell sorter columns (Miltenyi Biotec) according to the manufacturer's instructions. Purity was 87–92% CD14+ cells by flow cytometry.

Monocytes were plated in serum-free RPMI 1640 in 6-well dishes for RNA analysis (350,000–500,000 cells/well) and in 24-well dishes (150,000–250,000 cells/well) for cytokine analysis. RNA and cytokine results were normalized (described below) to account for differences in cell numbers. For further purification, nonadherent cells were removed after 2 hours. The adherent cells were then cultured with RPMI 1640 containing 10% fetal bovine serum, penicillin/streptomycin, and L-glutamine (HyClone). Cultures were supplemented with 20 ng/ml of recombinant human macrophage colony-stimulating factor (M-CSF; R&D Systems), which was replaced on days 2 and 4.

On day 5, a total of 1,000 units/ml of IFNγ (PeproTech) was added to select cultures for 24 hours prior to stimulation (on day 6) with either 10 ng/ml or 100 ng/ml of LPS (Salmonella enteritidis; Sigma). IFNγ and M-CSF contained less than 0.01 endotoxin units of endotoxin, as detected by Limulus amebocyte assay (GenScript). For the tunicamycin (Sigma) assay, macrophages were stimulated with 0.1–1 μg/ml of tunicamycin for 4–16 hours starting on day 5. Cells obtained after 5 days of culture were relatively homogeneous, large, and had a “fried egg” appearance. Flow cytometry showed that the cells were >99% CD11b+, 94–99% HLA–DR+, and expressed macrophage markers CD163 and CD206 (80–87% and 86–88%, respectively).

IFNγ and LPS treatments affected viability (assessed by propidium iodide and trypan blue exclusion) by the end of day 7 as compared to no stimulation. The mean ± SEM frequencies of viable cells were as follows: 73 ± 7% for LPS at 10 ng/ml, 87 ± 7% for LPS at 100 ng/ml, 92 ± 11% for IFNγ, 60 ± 12% for IFNγ plus 10 ng/ml of LPS, and 55 ± 5% for IFNγ plus 100 ng/ml of LPS.

HLA–B, HLA–B27, and UPR gene expression.

On day 6, RNA was obtained following 24 hours of IFNγ treatment plus 3 hours of LPS stimulation (in select samples) by resuspending the cells in TRIzol (Invitrogen) and processing according to the manufacturer's instructions. RNA was treated with DNase (Invitrogen) and reverse transcribed to complementary DNA using reverse transcriptase (Promega) with random primers (Promega) according to standard protocols. The relative expression of HLA–B and the UPR-regulated genes BiP, CHOP, and ERdj4 was determined by real-time quantitative polymerase chain reaction using a MyiQ or iCycler (Bio-Rad) with normalization to 18S ribosomal RNA expression. HLA–B27–specific primers we used were previously described by Bon et al (27). Other primers were designed using Beacon Design 7.0 software (Premier Biosoft); primer sequences are available upon request from the corresponding author.

Cytokine/chemokine secretion.

Cell culture supernatants were collected at 8 hours (0.5 ml from the 1.5-ml culture) and at 24 hours: at 24 hours, all media were removed. Two time points were selected because of production kinetics for early- and late-peaking mediators of inflammation (e.g., IFNβ and CXCL9). Cells were lysed in 100 μl of sodium dodecyl sulfate lysis buffer, and total protein was quantified by bicinchoninic acid assay (Pierce) to account for differences in cell numbers per well. Supernatants were frozen at –80°C until they were analyzed by Luminex assay (Millipore), as measured on a Luminex 100 system. Multiplex analytes included IFNγ, IFNα2, IL-1β, IL-10, IL-12p70, IL-15, IL-1α, IL-6, IL-8, IFNγ-inducible 10-kd protein (IP-10), monocyte chemotactic protein 1, TNFα, and vascular endothelial growth factor (VEGF). Luminex assays for CXCL9 and IL-23 were run individually. IFNβ production was quantified by colorimetric enzyme-linked immunosorbent assay (PBL Biomedical) and read on a 354 Multiskan Ascent plate reader (Thermo Labsystems).

Initially, we screened 6 AS patient and 4 control samples for the 16 cytokines and chemokines listed above. For analytes that yielded a level of P < 0.15 by Student's t-test, comparing controls and AS patients within a treatment group (e.g., LPS 10 ng/ml, IFNγ plus LPS 10 ng/ml, etc.) we completed evaluation of the other 3 AS patient and 6 control samples. Duplicate samples were used to normalize values between Luminex runs. Thus, we measured CXCL9, IFNβ, IL-6, IL-8, IL-10, IL-12p70, IL-23, and TNFα in 9 AS patient and 10 control subjects. Only data from the complete analysis of 8 chemokines and cytokines in 19 subjects are presented below.

Statistical analysis.

UPR-regulated gene expression and up-regulation of HLA–B and HLA–B27 were analyzed using linear mixed effects (LME) models. These models included a random effect for subject and fixed effects for group (patient versus control), IFNγ, LPS 10 ng/ml, and LPS 100 ng/ml, plus 2-way interactions between group and all other terms and between IFNγ and the 2 LPS doses. An LME model (also called a repeated-measures analysis of variance [ANOVA]) is similar to ANOVA, except that at least 1 additional error term is included. Subject was included as a random effect in the LME model, to model within-subject correlations among all the measurements taken on a particular subject. The UPR responses were transformed to the log scale before analysis to obtain approximately normally distributed residuals. For BiP and HLA–B, the significant main effect of IFNγ in the complete model was further investigated by examining individual treatment comparisons. Correlations were performed using log-transformed data. Response to tunicamycin was also analyzed using an LME model with fixed effects for dose and time and a random effect for blood sample. Responses were transformed to the log scale before analysis. Wilcoxon's rank sum test was used to assess the differences in cytokine and chemokine production between patient and control groups for all the responses under each of the treatment conditions at both 8 hours and 24 hours.


Up-regulation of HLA–B.

Previous examination of granulocyte–macrophage colony-stimulating factor (GM-CSF)–derived macrophages from AS patients by microarray techniques did not show evidence of a UPR (28). However, HLA–B up-regulation was also not detected in that study. In the B27-transgenic rat model, up-regulation of HLA–B27 appears to play a critical role in visualizing the UPR; in premorbid animals, acute up-regulation of B27 expression by inflammatory cytokines was essential, and the magnitude of UPR induction strongly correlated with the fold increase in B27 expression (16, 17). To further examine the role of the UPR in AS, we derived macrophages from PBMCs with either GM-CSF or M-CSF and stimulated them with 100–1,000 units/ml of TNFα, IFNα, or IFNγ for 24–48 hours prior to analyzing surface expression of HLA–B by flow cytometry (data not shown). Derivation with M-CSF and 24 hours of stimulation with 1,000 units/ml of IFNγ yielded the greatest increase in HLA–B (mean ± SEM 1.82 ± 0.12–fold) with the least impact on viability, as determined by propidium iodide exclusion (mean ± SEM 11 ± 30%). These conditions were used for the following studies.

HLA–B expression in macrophages from AS patients and controls was analyzed by real-time quantitative polymerase chain reaction (Table 2). AS patient and control macrophages responded similarly to treatment, and only IFNγ had a significant impact on the mean expression (P < 0.0001). In pairwise tests combining patients and controls, IFNγ, IFNγ plus 10 ng/ml of LPS, and IFNγ plus 100 ng/ml of LPS all showed significantly increased expression of HLA–B over no stimulation (P ≤ 0.02). In the AS patients, HLA–B expression increased by an average of 2.21-fold (range 1.09–5.95-fold; P = 0.0063) from no stimulation to IFNγ stimulation, by 1.83-fold (range 0.34–4.98-fold; P = 0.02) from no stimulation to IFNγ plus 10 ng/ml of LPS stimulation, and by 2.44 (range 1.3–6.75-fold; P = 0.0027) from no stimulation to IFNγ plus 100 ng/ml of LPS stimulation. In this last group, HLA–B expression was up-regulated in all AS patients (Figure 1B).

Table 2. HLA–B expression following in vitro treatment with IFNγ and/or LPS*
TreatmentAS patientsHealthy controls
  • *

    There was no significant difference in treatment response between cells from ankylosing spondylitis (AS) patients and cells from healthy controls (P = 0.119). The only significant effect was with interferon-γ (IFNγ) treatment, which increased the response of cells from both patients and controls to all treatments (IFNγ only or IFNγ plus lipopolysaccharide [LPS]) (P < 0.0001). Samples from 7–8 AS patients and healthy controls were analyzed under each condition. Values are the mean (95% confidence interval). The mean expression was derived from log-transformed data as described in Patients and Methods.

No stimulation559 (356–875)770 (492–1,204)
 10 ng/ml664 (413–1,065)1,551 (969–2,479)
 100 ng/ml596 (380–933)855 (546–1,336)
IFNγ1,249 (799–1,949)1,239 (793–1,933)
 Plus LPS 10 ng/ml1,071 (668–1,713)1,911 (1,195–3,052)
 Plus LPS 100 ng/ml1,375 (881–2,145)1,577 (1,010–2,459)
Figure 1.

HLA–B and unfolded protein response (UPR)–regulated gene expression. A, Expression of binding immunoglobulin protein (BiP), CCAAT/enhancer binding protein homologous protein 10 (CHOP), and endoplasmic reticulum–localized DnaJ homolog (ERdj4) following tunicamycin (Tu) stimulation. Human macrophages were treated with the indicated doses of tunicamycin for 4, 8, or 16 hours. Gene expression was normalized to the values at 4 hours in untreated controls to yield the fold change values. Values are the geometric mean ± SEM of 4 independent experiments. The geometric mean best represents the typical fold change, but large fold changes are downweighted. B, Expression of HLA–B, CHOP, BiP, and ERdj4 genes by macrophages from ankylosing spondylitis (AS) patients and healthy controls (C). Cells were left unstimulated or were stimulated with interferon-γ (IFNγ) for 24 hours followed by 100 ng/ml of lipopolysaccharide (LPS) for 3 hours. Relative gene expression was determined by real-time quantitative polymerase chain reaction with normalization to 18S ribosomal RNA. Individual data points for unstimulated and stimulated samples (connected by lines) are shown at the left. Box plots showing the 25th and 75th percentiles (boxes), median (lines within the boxes), means (▪), and the 10th and 90th percentiles (bars) are shown at the right. The numbers of subjects were as follows: for HLA–B, 8 patients and 8 controls (P = 0.0006 for up-regulation of HLA–B in combined patient and control samples); for CHOP and BiP, 9 patients and 10 controls, and for ERdj4, 9 patients and 7 controls.

Other investigators have demonstrated increased B27 expression on PBMCs from AS patients as compared to those from healthy B27+ controls, in the absence of overall increased HLA class I expression (29). Thus, we specifically examined B27 expression in AS patient samples. Using an LME model, HLA–B27 behaved similarly to HLA–B, in that only IFNγ treatment was significant (P = 0.0005). IFNγ induced a mean increase of 2.24-fold (range 0.86–7.8-fold; P = 0.016), IFNγ plus 10 ng/ml of LPS induced a mean increase of 3.74-fold (range 0.61–10.4-fold; P = 0.063), and IFNγ plus 100 ng/ml of LPS induced a mean increase of 1.96-fold (range 1.32–11.23-fold; P = 0.016). These ranges were slightly larger than those for HLA–B, which is consistent with previously published data (29).

UPR gene expression.

Part of our hypothesis predicted that IFNγ would induce the UPR by acutely up-regulating HLA–B expression in AS patient–derived macrophages. We chose to study the expression of BiP, CHOP, and ERdj4 since these gene targets are robustly up-regulated by the UPR and are relatively specific for the 3 major identified signaling pathways of the UPR initiated by activation of the resident proteins in the ER: RNA-dependent protein kinase–like ER kinase (PERK), inositol-requiring enzyme 1 (IRE-1), and activating transcription factor 6 (ATF-6), respectively (15, 30, 31). In previous studies in the transgenic rat model, 20–24 hours of IFNγ stimulation had been sufficient to observe UPR induction and a further 3 hours of LPS stimulation for additional XBP-1 splicing (16, 20). As seen in Figure 1A, these gene products were sensitive to change following treatment with very low doses of tunicamycin. All tunicamycin doses showed significant elevation over no stimulation (P values ranging from P = 0.048 to P < 0.001) with the exception of ERdj4 at 0.1 μg/ml (P = 0.094). The increase between 4 hours and 16 hours of stimulation was significant for BiP (P = 0.0044). All other effects of increased time were not significant (P values ranging from P = 0.057 to P = 0.88).

Contrary to our prediction, in vitro stimulation with IFNγ and LPS had minimal impact on mean UPR-regulated gene expression, even under conditions of the greatest HLA–B up-regulation (IFNγ plus 100 ng/ml of LPS) (Figure 1). The mean gene expression in AS patients was not significantly elevated compared to that in controls in any of the treatment groups. There was no statistical difference between the responses of patients and controls to the various treatments (Table 3). Combining patients and controls, the only significant difference compared to no stimulation was seen in response to IFN plus 10 ng/ml of LPS, where BiP and CHOP decreased (changes of 0.68-fold and 0.7-fold, respectively; P < 0.02). ERdj4 showed a trend toward differential regulation in AS patient and control macrophages: in the IFN plus LPS treatment groups, control macrophages had unchanged or slightly decreased mean expression of ERdj4, whereas AS patient macrophages tended to up-regulate ERdj4 expression (mean 2.2-fold in response to IFNγ plus 100 ng/ml of LPS [P = 0.31], a significant patient–IFN interaction at P = 0.03). However, the mean expression of ERdj4 in AS patient-derived macrophages remained below the control values. In AS patient macrophages, there was no significant correlation between the fold up-regulation of HLA–B and UPR-regulated gene expression.

Table 3. Levels of unfolded protein response–regulated genes following in vitro treatment with IFNγ and/or LPS*
Gene, treatmentAS patientsHealthy controls
  • *

    The total numbers of samples evaluated under each condition were as follows: 10 ankylosing spondylitis (AS) patients and healthy controls for no stimulation and interferon-γ (IFNγ) stimulation; 7 patients and controls for lipopolysaccharide (LPS) 10 ng/ml and for IFNγ plus LPS 10 ng/ml; and 9 patients and 9–10 controls for LPS 100 ng/ml and for IFNγ plus LPS 100 ng/ml. Values are the mean (95% confidence interval). The mean expression was derived from log-transformed data as described in Patients and Methods.

 No stimulation262 (136–499)150 (76–299)
 LPS 10 ng/ml197 (95–400)205 (99–414)
 LPS 100 ng/ml259 (131–501)185 (93–360)
 IFNγ206 (106–399)115 (57–223)
 IFNγ + LPS 10 ng/ml204 (98–414)69 (30–147)
 IFNγ + LPS 100 ng/ml261 (133–505)133 (67–258)
 No stimulation15.4 (9–23.9)11.1 (5.8–18.2)
 LPS 10 ng/ml8.1 (3–15.1)10.5 (4.7–18.4)
 LPS 100 ng/ml10.9 (5.5–18.2)7.0 (2.6–12.9)
 IFNγ11.7 (6.3–19)7.3 (3–13.1)
 IFNγ + LPS 10 ng/ml7.9 (2.8–14.9)4.5 (0.4–10.1)
 IFNγ + LPS 100 ng/ml10.9 (5.5–18.2)9.2 (4.4–15.7)
 No stimulation185 (83–397)593 (268–1,295)
 LPS 10 ng/ml110 (40–275)644 (248–1,651)
 LPS 100 ng/ml214 (94–476)651 (282–1,486)
 IFNγ151 (67–327)284 (131–604)
 IFNγ + LPS 10 ng/ml213 (83–520)237 (94–579)
 IFNγ + LPS 100 ng/ml391 (175–860)486 (219–1,064)

We further analyzed the AS patients according to whether they were treated with DMARDs (6 had received DMARDs and 4 had not). Few cases were significant: in response to IFNγ, BiP expression changed by 0.64-fold in the DMARD group and 1.16-fold in no DMARD group (P = 0.038). In response to IFNγ plus 100 ng/ml of LPS, CHOP expression changed 0.71-fold in the DMARD group versus 0.93-fold in the no DMARD group (P = 0.016), and ERdj4 changed 1.31-fold in the DMARD group versus 4.83-fold in the no DMARD group (P = 0.032). Thus, in select cases, patients who were not taking DMARDs tended to show a greater increase in gene induction or less of a decrease as compared to those who were taking DMARDs. Our study was most likely underpowered to detect other differences between DMARD groups.

Cytokine/chemokine production.

The evaluation of inflammatory mediator production was based upon previously described differences between AS patients and controls as well as studies examining the regulation of inflammation by the UPR. We had previously described a marked synergy between the UPR and LPS induction of both IFNβ and IL-23 in the context of both pharmacologic UPR induction and B27+ transgenic rat macrophages undergoing a UPR (20, 21). We hypothesized that IFNγ-treated AS patient macrophages would produce excess IFNβ and IL-23 in response to LPS stimulation. In addition to IFNβ and IL-23, we also examined macrophage production of CXCL9, IL-6, IL-8, IL-10, IL-12p70, and TNFα in 9 AS patients and 10 controls (Table 4).

Table 4. Cytokine and chemokine production following in vitro treatment with IFNγ and/or LPS*
Mediator, treatment conditionsAS patientsHealthy controlsP
  • *

    The total numbers of samples evaluated under each condition were as follows: 4–5 ankylosing spondylitis (AS) patients and healthy controls for no stimulation and interferon-γ (IFNγ) stimulation; 7 patients and 7–9 controls for lipopolysaccharide (LPS) 10 ng/ml and for IFNγ plus LPS 10 ng/ml; and 9 patients and 9–10 controls for LPS 100 ng/ml and for IFNγ plus LPS 100 ng/ml. Values are the median (25th, 75th percentiles). P values less than 0.05 were considered significant. No significant differences were detected for interleukin-8 (IL-8) production (data not shown). NA = not applicable; TNFα = tumor necrosis factor α.

  • Values were at the threshold of detection for the assay.

CXCL9, ng/ml   
 24 hours   
  No stimulation0.06 (0, 0.14)0.47 (0.29, 0.64)0.1832
  LPS 10 ng/ml40.6 (22.3, 154.5)12.6 (9.4, 13.4)0.0041
  LPS 100 ng/ml27.7 (23.4, 71.3)31.9 (16.5, 35.1)0.5457
  IFNγ1,796 (1,661, 2,026)1,091 (1.4, 2,269)0.885
  IFNγ + LPS 101,593 (1,447, 1,884)832 (731, 1,104)0.0175
  IFNγ + LPS 1001,650 (1,263, 2,031)833 (603, 1,677)0.1615
IFNβ, pg/ml   
 8 hours   
  No stimulation00NA
  LPS 10 ng/ml76 (0, 132)0 (0, 65)0.3507
  LPS 100 ng/ml51 (0, 148)0 (0, 78)0.4530
  IFNγ + LPS 10522 (302, 1,087)362 (274, 460)0.7577
  IFNγ + LPS 100627 (225, 1,181)301 (260, 377)0.7802
IL-6, pg/ml   
 8 hours   
  No stimulation6 (3, 8)7 (6, 22)0.4568
  LPS 10 ng/ml2,413 (1,211, 3,569)3,012 (2,811, 3,771)0.6065
  LPS 100 ng/ml2,305 (1,616, 3,440)2,972 (2,192, 3,713)0.6965
  IFNγ19 (14, 26)7 (3, 12)0.2000
  IFNγ + LPS 108,904 (6,456, 12,302)5,600 (4,698, 7,336)0.1142
  IFNγ + LPS 1008,037 (5,606, 13,858)5,161 (3,228, 8,601)0.0831
IL-10, pg/ml   
 24 hours   
  No stimulation34 (20, 47)8 (8, 37)0.2000
  LPS 10 ng/ml2,240 (1,523, 4,869)486 (383, 1,153)0.0229
  LPS 100 ng/ml2,126 (1,438, 2,749)682 (471, 1,725)0.0653
IL-12p70, pg/ml   
 8 hours   
  No stimulation00NA
  LPS 10 ng/ml3 (1, 4)0 (0, 0)0.0203
  LPS 100 ng/ml2 (0, 5)0 (0, 1)0.0929
  IFNγ + LPS 10634 (411, 908)165 (150, 250)0.0110
  IFNγ + LPS 100516 (369, 931)158 (118, 194)0.0101
IL-23, pg/ml   
 24 hours   
  No stimulation00NA
  LPS 10 ng/ml265 (188, 393)9 (6, 92)0.0007
  LPS 100 ng/ml357 (149, 405)15 (3, 122)0.0022
TNFα, pg/ml   
 8 hours   
  No stimulation3 (2, 4)3 (2, 13)0.8571
  LPS 10 ng/ml2,667 (2,260, 2,719)1,393 (1,094, 2,397)0.0562
  LPS 100 ng/ml2,732 (2,431, 3,284)1,410 (1,128, 2,053)0.0016
  IFNγ10 (7, 16)4 (2, 13)0.8571
  IFNγ + LPS 10 ng/ml3,077 (2,952, 3,663)2,806 (2,625, 3,193)0.0556
  IFNγ + LPS 100 ng/ml3,545 (3,156, 3,604)2,671 (2,212, 2,949)0.0408
 24 hours   
  No stimulation7 (5, 7)3 (2, 6)0.2938
  LPS 10 ng/ml2,430 (2,352, 2,550)1,617 (1,224, 1,739)0.0068
  LPS 100 ng/ml2,620 (1,964, 2,957)1,536 (1,157, 2,213)0.0177

We observed statistically significant increases in CXCL9, IL-10, IL-12p70, IL-23, and TNFα production by AS patient macrophages as compared to control macrophages (Table 4 and Figure 2). We did not detect significant differences in production of IL-6, IL-8, or IFNβ. Apart from IL-12p70, these differences were primarily found in the LPS-only treatment groups, where there was no HLA–B up-regulation or UPR gene induction. Contrary to our prediction, IFNγ pretreatment minimized differences for all inflammatory mediators tested except IL-12p70 (Table 4 and data not shown). Median differences between AS patients and controls for IL-10, IL-12p70, and CXCL9 ranged from 1.9-fold to 4.6-fold. Statistical differences in IL-12 production were detected only at 8 hours, but not at 24 hours (data not shown).

Figure 2.

Increased production of interleukin-23 (IL-23), CXCL9, IL-12p70, and tumor necrosis factor α (TNFα) by macrophages from patients with ankylosing spondylitis (AS). Individual data points in AS patients and controls (C) are shown at the left. Box plots showing the 25th and 75th percentiles (boxes), median (lines within the boxes), and the 10th and 90th percentiles (bars) are shown at the right. See Table 3 for associated P values and numbers of samples per experimental condition. A, For IL-23, cells were stimulated for 24 hours with 10 ng/ml or 100 ng/ml of lipopolysaccharide (LPS). B, For CXCL9, cells were left unstimulated or were stimulated for 24 hours with interferon-γ (IFNγ) and then for another 24 hours with 10 ng/ml of LPS. C, For IL-12p70, cells were pretreated for 24 hours with IFNγ and then stimulated for another 8 hours with either 10 ng/ml or 100 ng/ml of LPS. The outlier data points for IL-12p70 production in response to both doses of LPS were for cells from the same control subject. D, For TNFα, cells were left unstimulated or were stimulated for 24 hours with IFNγ and then for another 8 hours with 100 ng/ml of LPS.

Several in vitro treatments revealed differences in TNFα production between AS patient and control macrophages, although the P values reflected relatively less variability rather than profound differences in median production (<2-fold). The most striking AS patient–control differences were observed for the median IL-23 production following 24 hours of stimulation with LPS (Table 4 and Figure 2): AS patient macrophages produced greater than a log-fold more IL-23 compared to control macrophages (median production 265–357 pg/ml in AS patients versus 9–15 pg/ml in controls; P = 0.0022 to P = 0.0007). In our initial screening of 4 controls and 6 patients, no differences in IL-23 production were observed for IFNγ-pretreated cells, and production after 8 hours of LPS stimulation was comparable to that after 24 hours of stimulation (median production 213–341 pg/ml in AS patients versus 24–64 pg/ml in controls).


Based on the B27-transgenic rat spondylarthritis model, we hypothesized that increased HLA–B27 expression and misfolding during inflammation induces a UPR that renders innate immune cells (macrophages and dendritic cells) more proinflammatory and hyperresponsive to bacterial TLR agonists. UPR-driven cells secrete excessive levels of IL-23 (Th17 activation) and type I IFN (20, 21). However, the transgenic rat highly overexpresses B27, and thus, B27-misfolding effects would most likely be exaggerated. It is important to assess the UPR/TLR inflammatory model in human disease, where there are, at most, 2 copies of the B27 gene.

In this study, we did not detect significantly increased UPR-regulated gene expression in AS patient macrophages relative to those from controls at baseline, nor did we observe significant up-regulation of classic UPR target genes following in vitro stimulation with IFNγ and LPS. Recently, LPS was shown to activate IRE-1–dependent XBP-1 splicing via the induction of an NADPH oxidase while suppressing activation of the PERK and ATF-6 arms of the UPR (32). In this study, 10 ng/ml of LPS also tended to decrease the mean UPR gene expression in AS patients. However, we did not detect increased UPR gene expression following IFNγ stimulation, even in the absence of the additional LPS treatment. The significance of the increase in ERdj4 mRNA following IFNγ plus LPS treatment in AS patients is unclear, since mean expression remained below that observed in control macrophages.

The simplest explanation for the differences in our findings and the B27-transgenic animal studies lies in the magnitude of HLA–B up-regulation. In the report describing the correlation of UPR magnitude and B27 expression in transgenic rat macrophages, IFNγ treatment was reported to up-regulate B27 4–10-fold; although even a 2-fold up-regulation of B27 by TNFα treatment yielded a 2-fold induction of BiP expression (17). The mean 2–3-fold B27 induction we observed may not have been sufficient to detect UPR induction in patient samples, particularly given patient-to-patient variability. Our small sample sizes may have been underpowered to detect subtle up-regulation of gene expression. In addition, the representative UPR target genes we chose (BiP, CHOP, and ERdj4) may not be the most sensitive indicators of a B27 UPR, although they are widely used to detect the UPR (15, 32). Alternatively, the UPR observed in the B27-transgenic rats may depend upon dramatic HLA–B27 overexpression related to multiple transgene copy numbers and, thus, may be less relevant to human disease. However, the increased BiP expression found in the synovial cells from spondylarthritis patients would provide evidence against this possibility (18). Another possibility is that factors in addition to IFNγ activation and B27 up-regulation are required to induce a UPR in AS macrophages. Our in vitro system may have been too simplistic compared to the in vivo inflammatory milieu. Future examination of inflamed tissues from AS patients may be more revealing. Related to all these factors, our results do not rule out a role of the UPR in the pathogenesis of AS.

The study by Tran et al (19), showing worse arthropathy in B27-transgenic rats where the UPR had been modulated by β2-microglobulin, calls into question the relationship between B27-related UPR and joint disease. Multiple non-UPR hypotheses have sought to explain the striking contribution of B27 to genetic risk (8). For example, expanded numbers of natural killer cells and CD4 T cells that recognize cell surface B27 dimers have been identified in the circulation of AS patients (14). Other efforts have focused on specific antigen presentation by HLA–B27 (33). With some variability across studies, HLA–B27 has also been shown to alter the survival and persistence of intracellular organisms through unclear mechanisms (8, 34). Ultimately, our data suggest that non-UPR, cytokine-modulatory mechanisms may contribute to a proinflammatory diathesis and that non-UPR hypotheses bear greater exploration.

The tendency of IFNγ to minimize differences in inflammatory mediator production is consistent with our previous data showing a global decrease in IFNγ-regulated gene expression in AS patient macrophages that could be recovered with exogenous IFNγ treatment (28). This study and others have highlighted a relative deficit in the production of Th1 cytokines such as IFNγ by AS patient cells (28, 35, 36). However, the IL-12p70 data (Table 4) suggest that under certain infectious or inflammatory conditions in which IFNγ is produced, AS patients could potentially mount an even more robust Th1 response than controls. Our findings of increased CXCL9, TNFα, and IL-10 in AS patient macrophages corroborate previous studies of patient samples (37, 38). Increased CXCL9, CXCL10/IP-10, and IL-12/IL-23p40 are present in synovial fluid from patients with spondylarthropathy (37, 39, 40). Studies examining serum and PBMC samples from AS patients have demonstrated variable increases in TNFα, IL-1β, IL-6, IL-8, VEGF, and IL-10 (38). What is unique to this study is removal of monocytes from the context of patient inflammation and medication. The in vitro derivation and stimulation conditions are controlled; thus, the increased cytokine/chemokine production most likely reflects an intrinsic, genetically determined property of AS macrophages.

The most striking finding from this study is that even in the absence of an obvious UPR, AS macrophages produced significantly more IL-23 in response to LPS alone. The 1 patient known to be B27 negative produced IL-23 at levels close to the median in AS patients (335 pg/ml in response to 10 ng/ml of LPS and 405 with 100 ng/ml of LPS). The increased IL-23 was much more robust than the differences observed for TNFα. ER stress–induced CHOP has recently been shown to directly regulate IL-23 expression (41). Although UPR and IL-23 production have also been closely associated in the B27-transgenic rat model, our results suggest the potential for excess IL-23 production by patient macrophages in the absence of an overt UPR. The mean CHOP expression actually decreased following IFNγ treatment as well as early during the LPS stimulation period (21). In the recent report by Martinon et al (32), LPS stimulation decreased pharmacologically induced UPR target gene expression (BiP, CHOP, and ERdj4) out to 9 hours (32). In the present study, we did not collect RNA after 24 hours of LPS stimulation and therefore cannot absolutely rule out late occurrence of an LPS-induced UPR.

Although this study was underpowered to assess the impact of HLA–B27 positivity in AS patients, our results support the hypothesis that AS macrophages produce an overabundance of IL-23 in response to TLR-4 agonists in the infectious environment, thus predisposing to the development of inflammatory lesions. The IL-23/Th17 axis has become increasingly recognized as a key component of antibacterial immunity, accounting for the genetic pressures to maintain risk alleles for a debilitating arthritic condition over the centuries (42). Given the responses we observed to LPS, it will be important to determine whether macrophages from AS patients produce excess IL-23 in response to colonic and infectious organisms. The finding of robust IL-23 expression in subclinical gut inflammation from AS patients suggests that this may be the case; however, more defined studies will be helpful in elucidating pathogenesis (24).

Our IL-23 results extend and support other recent studies examining the IL-23/Th17 axis in AS. Polymorphisms in the IL-23 receptor were identified in genome-wide association studies as susceptibility alleles for AS (43). Increased IL-23 and IL-17 levels have been observed in the serum and cultured PBMC supernatants from AS patients with active disease (23, 44). AS patients also have a higher proportion of IL-17–producing CD4+ T cells in their circulation (23).

Anti–IL-12p40 agents (e.g., ustekinumab) that block both IL-12– and IL-23–mediated effects are highly efficacious in psoriasis, more so than etanercept (a TNF blocker) (45). In this study, the overproduction of IL-23 by AS patient macrophages was much more impressive than that observed for TNFα, even though TNF blockers are currently the therapy of choice for AS (46). The evidence from this study and others implicating IL-23 expression in AS pathogenesis provides a strong rationale for the development of IL-23–blocking agents as treatment for AS.


All authors were involved in drafting the article or revising it critically for important intellectual content, and all authors approved the final version to be published. Dr. Smith had full access to all of the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.

Study conception and design. Zeng, Smith.

Acquisition of data. Zeng, Smith.

Analysis and interpretation of data. Zeng, Lindstrom, Smith.