This was a prospective study on 107 consecutive patients with RA meeting the 2010 ACR/EULAR criteria (15) and 113 consecutive individuals without RA (non-RA) referred for dual-energy X-ray absorptiometry (DXA) scan. These non-RA subjects were subsequently classified either “healthy” or having “primary osteoporosis/osteopenia” according to the WHO criteria (osteoporosis in case of T-score of less or equal to −2.5 and osteopenia if T-score was between −2.5 and −1.0) (16). Patients were recruited at the outpatient clinics of the Rheumatology and Endocrinology Departments of the Medical University Graz, respectively. Detailed family and medical history including disease duration, prior and current treatments, as well as fracture risk assessment tool (FRAX) (17) data identifying clinical risk factors for osteoporosis and osteoporotic fractures were obtained from each individual. Clinical visits of RA patients were performed at baseline and subsequently for every 6 months up to 2 years. Non-RA controls underwent a baseline visit only. Synovial fluid samples were obtained from RA patients undergoing routine joint aspiration.

This study was approved by the Institutional Review Board of the Medical University Graz, and written informed consent was obtained from each individual.

Measurement of BMD by DXA was performed according to routine protocols. All RA patients underwent the scans at baseline, after 1 year and after 2 years, whereas controls underwent baseline scans only. BMD was assessed at the lumbar spine (L1–L4) and the left (or the right, if not possible otherwise) hip and femoral neck. Vertebral bodies with significant degenerative changes or compressive fractures (as judged by the investigating physician) were excluded from analysis, and the mean BMD was calculated using the remaining vertebral bodies (applicable only in case of at least two adjacent analyzable vertebrae). BMD was expressed in g/cm² and has a T-score, which is derived from the SD of bone-healthy comparators.

Laboratory parameters at baseline and every 6 months during follow-up were determined (detailed in Figure S1 in Supplementary Material).

Peripheral venous blood or synovial fluid was drawn from each individual, and PBMCs were isolated by Histopaque density-gradient centrifugation. The total cell number was determined by a Beckmann Coulter. Cells were cultured at 1 × 10⁶ cells/ml in RPMI 1640 containing 10% fetal calf serum, 2 mM l-glutamine, 100 U/ml penicillin, and 100 μg/ml streptomycin in the presence of 20 U/ml human-recombinant IL-2 (SIGMA, Vienna, Austria) and initial stimulation with 10 μg/ml plate-bound anti-CD3 Ab (eBioscience).

For stimulation assays, PBMCs were stimulated with 100 ng/ml of IL-15, IL-6, and, TNF-α (all Sigma-Aldrich) for 3 days or with 10 μg/ml plate-bound anti-CD3 Ab overnight. Unstimulated cells were served as a negative control.

Surface and intracellular staining of freshly isolated PBMCs was performed according to routine protocols and using appropriate combinations of antibodies for the detection of CD3, CD4, CD8, CD28, CD45RA, CD45RO, CD57 (all Becton Dickinson, San Diego, CA, USA), and RANKL (eBioscience). For intracellular staining of cytokines and intracellular RANKL, Golgi transport was inhibited by Brefeldin A (10 ng/ml) and Monensin (10 ng/ml) for 4 h prior to intracellular staining. Appropriate isotype controls were used. Stained cells were measured using a FACS Canto II (Becton Dickinson), and data analysis was conducted with DIVA software and FlowJo.

For functional assays, CD4⁺ T-cells were isolated by the positive selection of PBMCs labeled with magnetic-bead-conjugated anti-human CD4 mAbs using MACS MultiSort Kit and autoMACSPro according to manufacturer’s instructions (Miltenyi Biotech, Amsterdam, The Netherlands). Purified CD4⁺ T-cells were then separated into the CD28⁺ and CD28⁻ fractions by another sorting step using FACS technology (FACS Aria). For validation, flow cytometry was performed to determine the purity (>95%) of selected cells.

Separation of the monocyte fraction from PBMCs was carried out by the plastic adherence technique using 96-well plates (0.3 × 10⁶ cells/well) and serum-free α-MEM supplemented with 1% penicillin/streptomycin, 10 ng/ml M-CSF, and 1 ng/ml TGF-β (Peprotech, Hamburg, Germany). Monocytes were obtained from healthy donors. After 5–7 days, monocyte-enriched cells were washed with PBS and treated with trypsin/EDTA for detachment. Monocytes of 5 × 10⁴ were then cultivated for 13 days in the absence or presence of 1 × 10⁵ CD4⁺CD28⁺ and CD4⁺CD28⁻ T-cells from RA patients or 1 ng/ml RANKL (Peprotech) as a positive control. Medium was replaced twice weekly. Visual inspection confirmed the continued adherence of cells with each change of medium.

Osteoclast differentiation was evaluated by staining cells for tartrate-resistant acid phosphatase (TRAP) using the Leukocyte Acid Phosphatase Kit (Sigma-Aldrich, Vienna, Austria) according to the manufacturer’s instructions. Differentiated osteoclasts were defined as TRAP positive and multinucleated cells with ≥3 nuclei.

All statistical analyses were performed using the SPSS program, version 23 (Chicago, IL, USA). The Kolmogorov–Smirnov test was used to analyze the distribution of the variables. In case of a parametric distribution of continuous variables, data were reported as mean and SD, and we applied the two-sided Student’s t-test (comparison of two groups) for comparisons. In case of a non-parametric distribution, the results were described as median and range, and we conducted the Mann–Whitney U- and the Kruskal–Wallis tests. Paired data were compared with the Wilcoxon test. Correlation between variables was evaluated by Spearman’s rank correlation coefficient.

In total, 220 (107 RA and 113 non-RA) patients were included in the investigation. Clinical baseline characteristics are depicted in Table 1. FRAX data, full laboratory parameters, and results of genetic testing for a predisposition of lactose intolerance are shown in Tables S1–S3 in Supplementary Material. In the RA cohort, 12 (12.5%), 39 (40.6), and 26 (27.1) out of the 96 patients with available SDAI values had high-, moderate-, and low-disease activity, respectively, and 19 (19.8) patients were in remission (18). Six (6%) RA patients had early disease (≤2 years of duration). Clinical characteristics of RA patients at follow-up visits are shown in Table S4 in Supplementary Material.

Only minor changes of BMD could be observed in RA patients for 2 years in this study (for details, see Table 2).

A progression of bone loss as indicated by a loss of 5% of BMD over time (24 months) was observed in 17 (15.9%) (lumbar spine 13/total hip 4/femoral neck 6) RA patients. These 17 patients did not significantly differ from patients with stable BMD concerning medication and FRAX scores and lymphocyte subsets (data not shown). Baseline disease activity parameters correlated weakly with progressive bone loss at the lumbar spine (corr_coeff with DAS28 = 0.319, p = 0.033; corr_coeff with SDAI = 0.319, p = 0.029). In addition, a correlation of baseline osteocalcin levels with osteoporosis progression of the femoral neck (corr_coeff = 0.410, p = 0.002) was observed.

Rheumatoid arthritis is associated with premature senescence of T-lymphocytes (11, 13), and the accumulation of senescent CD4⁺CD28⁻ T-cells has been linked with disease severity (9, 14). The possible involvement of senescent T-cells in RA-associated and primary bone loss, however, has not been studied so far.

Rheumatoid arthritis patients with reduced bone mass showed elevated frequencies of CD4⁺CD28⁻ T-cells compared to patients with normal bone mass [2.2% (0–41.2) vs. 0.5% (0.1–8.6), p = 0.013, Figure 1A], whereas the prevalence of CD8⁺CD28⁻ T-cells was not different between the groups [45.2% (3.4–89.3) vs. 37.4% (10–61), p = 0.197, Figure 1B]. RA patients with osteoporosis showed the highest levels of senescent CD4⁺ T-cells [4.4% (0.2–39.7)] as compared to RA patients with osteopenia [1.4% (0–41.2)] and patients with normal bone mass [0.5% (0.1–8.6), Figure S1A in Supplementary Material].

In the non-RA cohort, we also noted increased frequencies of CD4⁺CD28⁻ T-cells [2.9% (0.1–34) vs. 0.7% (0.1–7.3), p = 0.010, Figure 1A] as well as CD8⁺CD28⁻ T-cells [38.3% (4.1–82.1) vs. 28.7% (6.6–59.8), p = 0.035, Figure 1B] in patients with reduced bone mass compared to healthy controls. Patients with osteopenia showed the highest levels of CD4⁺CD28⁻ T-cells [3.7% (0.1–34)] followed by patients with osteoporosis [2.4% (0.1–19.2)] and healthy controls [0.7% (0.1–7.3), Figure S1B in Supplementary Material].

In the RA cohort, frequencies of senescent CD4⁺CD28⁻ and CD8⁺CD28⁻ T-cells correlated significantly with BMD of the lumbar spine (CD4: corr_coeff = −0.361, p = 0.001, Figure 1C; CD8: corr_coeff = −0.248, p = 0.016), the hip (CD4: corr_coeff = −0.310, p = 0.003, Figure 1D; CD8: corr_coeff = −0.204, p = 0.045), and the femoral neck (CD4: corr_coeff = −0.275, p = 0.009; CD8: corr_coeff = −0.175, p = 0.088). Such correlations, however, were not observed in the non-RA group.

During follow-up, the frequencies of T-cell subsets remained stable in the RA cohort. None of the T-cell subsets at baseline predicted any change of BMD over time (data not shown).

To investigate the prevalence of CD4⁺CD28⁻ T-cells at the site of inflammation, we analyzed the samples of RA synovial fluid. The frequency of CD4⁺CD28⁻ T-cells was elevated compared to peripheral blood [4.1% (1.7–24.3) vs. 1.4% (0–41.2), p = 0.047, Figure 1E]. By contrast, the prevalence of CD8⁺CD28⁻ T-cells was diminished [26.6% (15.4–34.5) vs. 43.1% (3.4–89.3), p = 0.013, Figure 1F].

We next investigated whether senescent T-cells express RANKL, which is a key mediator of bone loss (19).

In the RA cohort, surface RANKL was significantly higher in CD4⁺CD28⁻ T-cells compared to CD28⁺ T-cells [3.7% (0.2–57.9) vs. 2.4% (0.2–35.4), p = 0.001, Figure 2A]. Within the CD28⁺ T-cell subset, memory CD45RO⁺ T-cells produced more surface RANKL than naïve CD45RA⁺CD4⁺ T-cells [2.7% (0.2–38) vs. 2.2% (0.2–30.5), p < 0.001, Figure S2A in Supplementary Material]. In CD4⁺CD28⁻ T-cells isolated from synovial fluid, RANKL expression was comparable to that of CD4⁺CD28⁻ T-cells isolated from peripheral blood [3.7% (0.2–57.9) vs. 2.8% (0.2–24.1), p = 0.893; Figure 2B]. Intracellular production of RANKL was higher in CD4⁺CD28⁻ T-cells compared to that in CD4⁺CD28⁺ T-cells [MFI: 901.5 (630–1069) vs. 660.5 (443–796), p = 0.028, Figure 2C] in patients with RA. Within the CD4⁺CD28⁺ population, memory CD45RO⁺ T-cells produced more intracellular RANKL than naïve CD45RA⁺ T-cells [781.5 (544–880) vs. 589.5 (373–669), p = 0.027, Figure S2C in Supplementary Material].

Results in the non-RA cohort were similar: surface RANKL production was significantly higher in CD4⁺CD28⁻ T-cells compared to that in CD28⁺ T-cells [2.4% (0–32.1) vs. 1.3 (0–34.8), p = 0.023, Figure 2A]. Again, memory CD4⁺ T-cells produced more surface RANKL than naïve CD4⁺ T-cells [1.4% (0–30.8) vs. 1.2 (0–45.3), p < 0.001, Figure S2B in Supplementary Material].

Each T-cell subset (i.e., naïve, memory, and senescent) in the RA cohort expressed more surface RANKL compared to the respective subset in the non-RA group (naïve: p < 0.001, memory: p = 0.001, senescent: p = 0.017).

TNF-α, IL-6, and IL-15 are cytokines that are strongly associated with the pathogenesis of RA and were also reported to promote osteoclastogenesis (20–22). We therefore tested the ability of these cytokines to alter RANKL expression in different T-cell subsets.

As depicted in Figure 3A, IL-15 significantly enhanced surface RANKL expression on both the CD28⁺ and the CD28⁻ CD4⁺ T-cell populations. The effect, however, was more prominent in CD28⁻ (5.4-fold higher expression compared to unstimulated cells) compared to CD28⁺ T-cells (3.5-fold). Within the CD28⁺ subset, IL-15 upregulated RANKL more strongly on memory (4.5-fold) compared to naïve T-cells (1.5-fold).

TNF-α as well as IL-6 failed to alter RANKL expression in all T-cell subsets (Figures 3B,C). Stimulation with anti-CD3 led to an increase of surface RANKL expression on CD28⁺ T-cells (7.9-fold), whereas senescent T-cells remained almost unaffected (Figure 3D).

Intracellular production of RANKL was modestly enhanced by anti-CD3 in CD28⁺ (1.5-fold) and CD28⁻ T-cells (1.3-fold), whereas stimulation with IL-15 did not further enhance RANKL production on either cell subset.

To evaluate whether the increased RANKL expression on CD28⁻ as compared to CD28⁺CD4⁺ T-cells results in a higher capacity to induce osteoclastogenesis in vitro, osteoclast differentiation assays were conducted. A higher number of TRAP + multinuclear cells and larger areas of bone resorbtion were observed in co-cultures of monocytes and senescent T-cells as compared to cultures involving CD28⁺ T-cells [34 osteoclasts/well (12–51) vs. 23 (7–29), p = 0.046, Figure 4 and Figure S3 in Supplementary Material]. Osteoclasts generated in the presence of either CD28⁻ or CD28⁺ T-cells were smaller and less intensively stained for TRAP than monocytes differentiated in cultures containing recombinant-soluble RANKL.