All experimental procedures were approved by and conducted in accordance with the Institutional Animal Care and Use Committee guidelines of The Johns Hopkins University. C57BL/6 J mice (stock no. 000664), CBA/J mice (stock no. 000656), and ROSA26^lsl-EYFP mice (stock no. 006148) were purchased from The Jackson Laboratory (Bar Harbor, Maine). STR/Ort mice were purchased from Harlan Laboratories (Frederick, MD). Cdkn2a^flox/flox mice were generated by Dr. Gloria H. Su’s laboratory from the Department of Pathology, Columbia University Medical Center (Qiu et al., 2011). Cdkn2a^tdTom reporter mice (C57BL/6 background) were generated by Dr. Norman E. Sharpless’s laboratory from University of North Carolina School of Medicine (Chapel Hill, NC) (Liu, 2019). The Tnfrsf11a^Cre/+ mouse strain was generously provided by Yasuhiro Kobayashi (Matsumoto Dental University, Japan) (Maeda et al., 2012). Tnfrsf11a^Cre/+; ROSA26^lsl-EYFP mice (RANK-EYFP mice) were generated by crossing Tnfrsf11a^Cre/+ with ROSA26^lsl-EYFP mice. We crossed the Tnfrsf11a^Cre/+ mice with Cdkn2a^flox/flox mice. The offspring were intercrossed to generate Tnfrsf11a^Cre/+; Cdkn2a^flox/flox (p16^cko) and Cdkn2a^flox/flox (WT) mice. The genotypes of the mice were determined by PCR analyses of genomic DNA using the following primers: Tnfrsf11a^Cre/+ allele forward, 5′-GCAATCCCCAGAAATGCCAGATTAC-3′and reverse,5′-GCAAGAACCTGATGATGGACATGTTCAG-3′; Cdkn2a^flox/flox allele (Note: Cdkn2a is the approved gene name for p16) P1, 5′-AGCAGCTTCTAATCCCAGCA-3′ P2, 5′-CCACTCCTGGAACTCAGCAT-3′ P3, 5′-AGGAGTCCTGGCCCTAGAAA-3′ and P4, 5′-CCAAAGGCAAACTTCTCAGC-3′; Cdkn2a^tdTom allele (Note: Cdkn2a is the approved gene name for p16) forward, 5′-ACCTCCCACAACGA-GGACTA-3′ and reverse, 5′-CTTGTACAGCTCGTCCATGC-3′; ROSA26^lsl-EYFP allele forward, 5′-AGGGCGAGGAGCTGTTCA-3′ and reverse,5′-TGAAGTCGATGCCCTTCAG-3′.

Mice were housed in a 12 hr light/12 hr dark cycle with ad libitum water and food access. At 10–12 weeks of age, mice were placed on a Western HFD (21% fat by weight) (TD 88137, Harlan Laboratories, Madison, WI) or a normal CHD for periods ranging from 2 weeks to 5 months. At the time of euthenasia, body weight was measured. The knee joints and serum were collected. For celecoxib treatment, mice were gavage-fed celecoxib at a dose of 16 mg/kg b.w for 2 months.

Whole-body fat and lean body mass were assessed by quantitative nuclear magnetic resonance (echo MRI), as previously described (Hu et al., 2020; Kim et al., 2019; Kim et al., 2017). Plasma triglycerides (Sigma-Aldrich, St. Louis, MO), cholesterol (BioAssay Systems, Hayward, CA), and glycerol (Sigma-Aldrich) were measured colorimetrically. Glucose levels were measured using a hand-held OneTouch Ultra glucose monitor (LifeScan Inc, Milpitas, CA).

MicroCT analysis of the tibial subchondral bone was performed as previously described (Su et al., 2020; Zhen et al., 2013). The knee joint was analyzed by μCT (voltage, 65 kVp; current, 153 μA; resolution, 9 μm/pixel) (Skyscan 1174, Bruker MicroCT, Kontich, Belgium). The parameters of the tibia subchondral bone image were analyzed using reconstruction software (NRecon v1.6, Bruker), data analysis software (CTAn v1.9, Bruker), and three-dimensional model visualization software (μCTVol v2.0, Bruker). Three-dimensional structural parameters analyzed were BV/TV, Tb.Pf, Tb.Th, Tb.N, and SBP.Th. Ten consecutive images from the whole subchondral bone medial compartment were used to do three-dimensional reconstruction and analysis.

Mouse knee joints were harvested after euthanasia. For frozen sections, the bones were fixed in 4% formaldehyde overnight, decalcified in 1.5 M EDTA (PH = 7.4) for 14 days, and embedded in optimal cutting temperature compound. We used 30-μm-thick sagittal-oriented sections for immunofluorescent staining using a standard protocol. We incubated the sections with primary antibodies to mouse VPP3 (1:100, Abcam, Cambridge, UK), HMGB1 (1:500, Novus Biologicals, Littleton, CO), lamin B1 (1:100, Santa Cruz Biotechnology, Dallas, TX), F4/80 (Abcam, 1:50), OCN (1:200, Takara Bio Inc, Shiga, Japan) overnight at 4 °C followed by corresponding fluorescence-linked secondary antibodies (Jackson ImmunoResearch Laboratories, West Grove, PA) for 1 hr while avoiding light. The sections were then co-stained with 4′,6-diamidino-2-phenylindole (H-1200, DAPI, Vector Laboratories, Burlingame, CA). The sample images were captured by a confocal microscope (Zeiss LSM 780). SA-βgal staining was conducted as previously described (Li et al., 2017; Liu et al., 2021), and the sample images were observed and captured by a microscope camera (DP71, Olympus BX51, Tokyo, Japan). For paraffin sections, bones were fixed in 4% formaldehyde overnight, decalcified in 1.5 M EDTA (PH = 7.4) for 21 days, and embedded in paraffin. We used 4-μm-thick sagittal-oriented sections for Safranin O (Sigma-Aldrich, S2255) and fast green staining (Sigma-Aldrich, 473 F7252). OARSI scores were calculated according to Safranin O–fast green staining. Osteoclasts were stained for TRAP. Osteoblasts were stained by OCN immunohistochemistry. Quantitative histomorphometry analyses were performed in a blinded fashion using OsteoMeasure Software (OsteoMetrics, Inc, Decatur, GA). Number of osteoblasts per bone marrow area (Ob.N/Bm. Ar.), osteocytes per bone area (Osteocytes N/B. Ar.), number of osteoclasts per bone perimeter (Oc.N/B. Pm) in the whole joint subchondral bone area per specimen. For each treatment group, 5–10 mice were used. For each sample, 3 tissue sections were used, and the whole joint subchondral bone area was analyzed.

Knee joints from the mice were dissected, and tibia subchondral bone tissue was homogenized in lysis buffer containing 1% Triton X-100 and protease inhibitors. Protein extracts were collected and protein concentration were measured. The antibody array for secreted factors was performed using Mouse XL Cytokine Array Kit (ARY028, R&D Systems, Minneapolis, MN) according to the manufacturer’s instructions.

Total RNA for qRT-PCR was extracted from the cultured cells using RNeasy Mini Kit (74104, QIAGEN, Hilden, Germany) according to the manufacturer’s protocol. For qRT-PCR, cDNA was prepared with random primers using the SuperScript First-Strand Synthesis System (Invitrogen, Waltham, MA). Then qRT-PCR was performed with SYBR Green Master Mix (QIAGEN) using C1000 Thermal Cycler (Bio-Rad Laboratories, Hercules, CA). Relative expression was calculated by the 2^−CT method with GAPDH for normalization.

For CFU-F assays, freshly isolated single-cell suspensions from the long bone of 12-week-old male mice were plated in 6-well plates with a density of 5×10⁵ cells per well. CFU-F colonies were counted after 10 days of culture with Crystal violet staining. We measured the colonies that had 50 cells or more. For CFU-Ob assays, freshly isolated bone marrow single-cell suspensions were seeded at a density of 5×10⁵ cells per well in six-well plates. After 21 days of culture with StemPro Osteogenesis Differentiation Kits (Invitrogen), osteogenic differentiation was detected by Alizarin red staining. The colony-forming efficiency was determined by counting the number of colonies per 5×10⁵ marrow cells plated.

Preosteoclasts were generated in vitro as previously described (Xie et al., 2014). Briefly, monocytes/macrophages were isolated from the bone marrow of 3-month-old WT male mice by flushing cells from the bone marrow of femora and tibiae. The flushed bone marrow cells were cultured overnight on Petri dishes in α-MEM containing 10% fetal bovine serum, 1% penicillin-streptomycin solution, 40 ng/mL M-CSF (Amizona Scientific LLC, AM10003-600). The cells were then incubated with osteoclastogenesis medium, containing αMEM with 10% fetal bovine serum, 1% penicillin-streptomycin solution, 40 ng/mL M-CSF and 60 ng/mL RANKL (Amizona Scientific LLC, AM10003-500) for 3 days, when most cells became mononuclear TRAP⁺ preosteoclasts (Xie et al., 2014). The formation of preosteoclasts was validated using TRAP staining (Sigma-Aldrich, 837) according to the manufacturer’s protocol. The cells were then challenged with 80 μg/mL oxLDL (Alfa Aesar, Cat# J64164) or vehicle (control) for 24 hr. Cellular senescence was confirmed using SA-βgal staining, Lamin B1 immunocytochemical staining, and qRT-PCR of Cdkn2a, Cdkn1a, Mki67 in the cells. CM was prepared by incubating the cells with serum-free medium for another 48 hr. After centrifugation (2500 rpm for 10 min at 4 °C), the CM was aliquoted and stored at –80 °C for different in vitro assays.

Preosteoclasts were challenged with H₂O₂ (200 µM for 2 hr, then 20 µM for 1 day) or vehicle (control). Total RNA of pre-osteoclasts was isolated using TRIzol (Life Technologies, USA). RNA-Seq library construction and RNA high-throughput sequencing were performed. We analyzed aging/senescence-induced genes (ASIGs) from publicly available mouse RNA-seq data (Aging Atlas database; KEGG pathway database; GO database; MSigD database). GO, KEGG pathway, GSEA analysis were analyzed based on the comparison of our data with ASIGs.

Data are presented as means ± standard errors of the mean. Unpaired, two-tailed Student t-tests were used for comparisons between 2 groups. For multiple comparisons, one-way analysis of variance with post hoc Tukey test was used. All data were normally distributed and had similar variation between groups. Statistical analysis was performed using SAS, version 9.3, software (SAS Institute Inc, Cary, NC). p<0.05 was deemed significant. All representative images of bones or cells were selected from at least three independent experiments with similar results unless indicated differently in the figure legend.

We selected OAI participants according to tailored stepwise criteria. (Supplementary file 1D) Knees with previous joint replacement surgery as shown in baseline radiographs were excluded (N=64, Exclusion #1, Supplementary Flowchart 1). From three OAI cohorts of knee osteoarthritis incidence, progression, and cohort non-exposed to osteoarthritis risk factors, we excluded participants from the non-exposed cohort of OAI (N=233, Exclusion #2, Supplementary file 1D) because of minimal risk of osteoarthritis incidence and progression. In order to assess OA-related subchondral BML damage, we included participants with available BML MOAKS scoring on the baseline and follow-up MRIs. We collected and pooled all previously conducted MRI-based measurements of participants from nested ancillary studies performed inside OAI. These studies' design and selection criteria are specially tailored to assess MRI-based OA structural damage worsening in a specific subset from all OAI participants (details of these studies are explained in the OAI online repository) (Overview and Description of Central Image Assessments, 2016). Following deletion of duplicate measurements (753 cases between different projects), MRI Osteoarthritis Knee Score (MOAKS) measurements for 1671 knees were included from the following OAI ancillary studies: (1) Foundation for the National Institute of Health (FNIH) Consortium Osteoarthritis Biomarkers Project (473 knees, project no. 22), (2) project no. 30 (125 knees) (3) projects no. 63 A-63F (328 knees) (4) Pivotal OAI MRI Analyses (POMA) study (751 knees). The same OAI team centrally performed all measurements according to the validated semi-quantitative MOAKS (Hunter et al., 2011). Participants without available MRI reads were excluded (N=7614, Exclusion #3, Supplementary file 1D). To further homogenize our selection criteria and delineate the role of subchondral bone damage worsening in the MetS-OA, we included participants with MetS-OA without a history of knee trauma (to exclude PTOA cases) versus participants with PTOA without MetS (to exclude MetS-OA cases). MetS-OA participants with a history of knee trauma and PTOA participants with MetS were excluded (N=1041, Exclusion #4, Supplementary file 1D). These selection criteria are specifically aimed to investigate the unique pathophysiology (e.g. prominent subchondral bone damage) in the MetS-OA phenotype compared to PTOA, an OA phenotype with the central role of cartilage degeneration.

We conducted observational studies by analyzing the human OAI dataset, which consists of annual clinical and radiographic data for the 9,572 knees of 4796 participants’ (Peterfy et al., 2008) with a follow-up of 8 years at annual time points (2004–2015, clinicaltrials.gov identifier: NCT00080171, https://nda.nih.gov/oai/). We selected 1671 participants who had available MRI Osteoarthritis Knee Score (MOAKS) scoring from OAI ancillary studies. We defined MetS presence and its components according to the International Diabetes Federation (IDF) criteria (Alberti et al., 2006). A history of the knee was assessed with the answer to the question "have you ever injured your knee badly enough to limit your ability to walk for at least two days?" and participants with a positive history of knee injury were considered as having PTOA. We further used a detailed selection criteria and propensity score (PS)-matching method (Haj-Mirzaian et al., 2019; D’Agostino, 1998).

Knees of OAI participants who responded positive to the question "ever injured badly enough your knee to limit the ability to walk for at least two days?" were regarded as having PTOA phenotype. To further homogenize selection criteria, we excluded participants with PTOA who had criteria for MetS (Exclusion #3, Supplementary file 1D).

No unified criteria for the definition of MetS exist; therefore, we defined MetS presence and components at the baseline visit according to the most widely accepted criteria defined by the IDF: (1) Hypertension was defined as systolic blood pressure (BP) of ≥130 mm Hg or diastolic BP of ≥85 mm Hg at baseline physical examination or as the use of BP-lowering medication indicated in the participants’ medication inventory form. (2) Diabetes mellitus was indicated by self-report or the presence of oral or injectable anti-diabetic medications on the medication inventory form. (3) Dyslipidemia was defined as the use of lipid-lowering medications indicated in the participants’ medication inventory form at baseline. (4) Abdominal obesity was defined as a waist circumference of ≥94 cm in men and ≥80 cm in women on physical examination. Participants with abdominal obesity and at least two of the three other components (dyslipidemia, diabetes mellitus, hypertension) were regarded as having MetS. Participants with MetS were regarded as MetS-OA phenotype, whereas participants who did not meet the IDF criteria for MetS were not. Participants with missing medication data on medication inventory forms were excluded (4 knees).

At the baseline visit, participants who reported regular COX2 inhibitor use on the OAI medication inventory forms were categorized as COX2 inhibitor users, and the remaining participants were categorized as non-users. Among COX2 inhibitor users, participants with a≤1 year history of regular COX2 inhibitor use before the baseline time point were categorized as incident users (initiators). Participants with >1 year of regular COX2 inhibitor use were regarded as prevalent users and were excluded from the study (Johnson et al., 2013). Prevalent users of COX2 inhibitors were excluded from the study (150 knees), and only COX2 inhibitor incident users were included because the differences in duration of COX2 inhibitor use before participation in the study (1 month to 10 years) can lead to ‘Neyman’ bias (Danaei et al., 2012), in which very sick and very well participants are excluded from enrollment (Danaei et al., 2012). It is possible that chronic and long-lasting conditions associated with COX2 inhibitor use led to rapid deterioration in knee osteoarthritis, which resulted in exclusion from OAI enrollment. Use of cohort studies and new-exposure (i.e. ‘incident use’ of COX2 inhibitor) designs are recommended to avoid this bias (Johnson et al., 2013; Danaei et al., 2012; Ray, 2003).

To assess COX2 inhibitor use, at each visit, participants reporting regular use of COX2 inhibitor during the last 12 months were coded as COX2 inhibitor users for that year, whereas negative reports of COX2 inhibitor use at each annual visit were considered as non-users for that year. Included participants were followed annually for 8 years (median, 6 years; interquartile range, 5 years). The same approach was used to assess non-selective NSAID use. Using propensity-score- (PS) matching method, COX2 inhibitor user were matched with non-selective NSAID users considering potential confounders. Following a ‘per-protocol’ design in our longitudinal analysis, which is associated with less selection bias in observational designs than in ‘intention to treat’ design (Danaei et al., 2013), participants who did not adhere to their assigned treatment (COX2 inhibitor vs. non-selective NSAID use) in each visit were right-censored from that visit (Danaei et al., 2013). Details of participant selection, PS-matching, and per-protocol design is explained in detail in Materials and methods section.

The pattern of missing data in covariates used in PS matching was assessed using the test of missing completely at random (Little’s test), visual representation, and logistic regression models, which resulted in a ‘missing not at random’ pattern (Sterne et al., 2009) in the OAI dataset, with <1.5% of values missing for all matching variables. Despite the ‘missing not at random’ pattern of data, multiple imputation models were used according to previous studies to try to reduce possible bias (Resseguier et al., 2011).

We matched study groups according to baseline characteristics to minimize confounding by indication bias. We first used PS matching to assess the presence of MetS on knee osteoarthritis outcomes. Next, we separately used PS matching to select participants for the two assessments of knee osteoarthritis and subchondral BML worsening outcomes in this study (1) comparison between MetS-OA and PTOA (2) comparison between COX2 inhibitor users and non-selective NSAID users. Possible confounders were investigated using a directed acyclic graph to assess causal inference (Schisterman et al., 2009). The selected covariates for PS matching were age (years), sex (male/female), body mass index (categorized in pentiles for ease of matching of kg/m²), and race (Caucasian or non-Caucasian). For assessment of MetS presence, smoking (current smoker, yes/no), use of alcohol (<1 or≥1 unit/week), Physical Activity Score for the Elderly (PASE), baseline KL grade, and medial JSN OARSI grade were also included as covariates. Western Ontario and McMaster Universities Osteoarthritis Index (WOMAC) pain score and history of knee trauma were also included as covariates in the PS-matching model for COX2 inhibitor use. For every knee in the exposure group (presence of MetS or COX2 inhibitor use), one best-matched knee in the non-exposure group (PTOA or non-selective NSAID users, respectively) was selected according to the confounding above variables (1:1 match). We used the nearest-neighbor method with a caliper distance of 0.2 on the imputed dataset (using multiple imputation models, all variables had <1.5% missing). The best matches were defined as knees with the highest-level match on their PS, calculated using logistic regression. We calculated standardized mean difference (SMD) to confirm the balanced outcome of matching between groups. Variables that did not optimally match in the PS matching method (SMD ≥0.1, baseline KL and JSN grades in the PS-matching for MetS vs. PTOA participants) were included in all subsequent statistical models as covariates of adjustment.

Longitudinal risk of knee osteoarthritis outcomes was assessed. Radiographic osteoarthritis incidence was defined as a Kellgren-Lawrence (KL) grade of ≥2 in follow-up assessment of knees with KL grades <2 at baseline (Felson et al., 2011). Progression was defined as an increase of 1 grade or more in medial Osteoarthritis Research Society International (OARSI) medial joint space narrowing (JSN) score ≥1 during follow-up assessments (Felson et al., 2011; Reijman et al., 2007). Knee radiographs were read centrally, and all KL grading and medial JSN OARSI scorings at baseline and follow-up points (1–8 years) are publicly available. For symptom assessment, participants’ self-reported pain was assessed using Western Ontario and McMaster Universities Osteoarthritis Index (WOMAC) scores for pain and disability at baseline and follow-up points (1–9 years). WOMAC pain and disability scores were standardized to a range of 0–100 and were summed, resulting in a standardized combined WOMAC score. A standardized combined WOMAC pain/disability score of ≥80 in 2 consecutive years was used as the definition for non-acceptable symptom state (Tubach et al., 2005; Gandek, 2015; Angst et al., 2005) incidence. Time of standard osteoarthritis outcomes in the study was defined as the earliest year with each outcome (Christensen, 1987). For participants with no event, the last available follow-up was used as the time in the study (Christensen, 1987). We adjusted all analyses for participants’ PS to minimize ‘confounding by indication’.

Three Tesla MRI systems (Trio, Siemens Healthcare) were used for OAI MRI acquisition. Parameters and pulse sequence protocol of OAI MRIs have been previously reported (Hunter et al., 2011). For comparison of MetS-OA and PTOA participants (i.e. MetS-OA⁺(PTOA^–) versus PTOA⁺(MetS-OA^–)), we used available data on MOAKS scorings of the OAI ancillary studies. These include the available 1671 MOAKS measurements for the knees that we pooled from all previously conducted from nested ancillary studies performed inside OAI to assess OA-related subchondral bone damage (Hunter et al., 2011) [explained in detail in Materials and methods section and the OAI online repository (Overview and Description of Central Image Assessments, 2016)]. Next, to compare BML worsening between COX-2 inhibitor users and non-selective NSAID users, MOAKS measurements of the included participants were not available on OAI ancillary studies. Therefore, a musculoskeletal radiologist with 12 years of experience (S.D.) read and scored baseline and 24-month follow-up T2 knee MRIs, using the same validated MOAKS scoring method (Hunter et al., 2011). The reader was blind to the participants' group (COX2 inhibitor users versus non-selective NSAID users). A 24-month BML score worsening was defined as a whole- or within-grade change, where within-grade was defined as a definite visual change while not fulfilling a whole-grade change definition. Previously validated MOAKS BML measures (Collins et al., 2016) were used: (1) worsening in the number of affected subregions with BML (improvement, no change, worsening in 1 subregion, and worsening in ≥2 subregions), (2) maximum worsening in BML score (no change, within-grade worsening, worsening by 1 grade, and by ≥2 grades), (3) improvement in the number of affected subregions (yes/no), and (4) improvement in subregions BML score (yes/no) (Collins et al., 2016).

We used Cox proportional hazards regression to determine associations between either MetS presence or COX2 inhibitor use with outcomes. HRs and 95% CIs are reported. All modeling was conducted using complex sample analysis, in which matched users/non-users, and same-participant knees were included in the specific clusters with equal weight. Logistic mixed-effect regression models were used to assess 24 month worsening in the BML MOAKS scores while considering random intercept for each cluster of matched COX2 inhibitor: non-selective NSAID user and within-subject similarities (due to the inclusion of both knees of included knees). Analyses were performed using the R platform (version 4.3.3). Two-tailed p-values <0.05 were considered significant.

The animal protocol (MO20M127) was reviewed and approved by the Institutional Animal Care and Use Committee of The Johns Hopkins University. We used data from the longitudinal multi-center OAI study (2004–2015 clinicaltrials.gov identifier: NCT00080171). All 4796 enrolled patients gave written informed consent. Institutional review boards of four OAI collaborating centers have approved the OAI study’s Health Insurance Portability and Accountability Act-compliant protocol (approval number: FWA00000068).

To delineate the specific pathophysiology of MetS-OA and further similarize our study sample with non-traumatic animal model models of MetS-OA, we excluded participants with MetS-OA who had a history of knee trauma (MetS⁺ PTOA⁻), and we selected participants with PTOA and without MetS as the control group (PTOA⁺ MetS⁻). A total of 630 knees met the inclusion criteria (explained in detail in Materials and methods section). After 1:1 PS matching, 338 matched knees were included, of which 169 were MetS-OA and 169 were PTOA. MetS-OA participants had a mean ± standard deviation (SD) age of 63.7±8.0 years and were 59.8% female. Similarly, PTOA participants had a mean ± SD age of 63.4±8.3 years and were 56.2% female. Supplementary file 1A shows the baseline characteristics of MetS⁺ and PTOA participants before and after matching. There was no imbalance in either of the potential confounding variables included in the matching between MetS-OA and PTOA groups (standardized mean difference or SMD <0.1) except for their baseline KL and JSN grades (SMDs of 0.159 and 0.132) which both were included as covariates of adjustment in all further statistical analyses. The mean follow-up duration for standard knee OA outcomes assessment was 6.9 years (Median and 1st and 3rd quartiles of 8 years). Participants with MetS-OA, had similar risk of radiographic osteoarthritis incidence (radiographic hazard ratio [HR], 0.89; 95% CI, 0.56–1.41, and symptomatic HR, 0.95; 95% CI, 0.54–1.67) and progression (assessed by OARSI JSN, HR, 1.04; 95% confidence interval (95% CI), 0.73–1.49) compared to matched PTOA participants (Table 1). However, despite the similar risk of knee OA incidence and radiographic progression, participants with MetS-OA had increased odds of 24 month worsening in subchondral bone marrow lesions (BML) damage compared to participants with PTOA. This finding was evident as the higher odds of increased knee joint subregions with subchondral BMLs (odds ratio [OR], 1.37; 95% CI, 1.06–1.77). Despite MetS-OA participants display a similar risk of knee OA incidence and progression compared to PTOA participants, they have more subchondral bone damage worsening compared to the matched PTOA participants. Therefore, MetS-OA participants have unique subchondral bone damage worsening, distinguishable from that of PTOA.

To investigate subchondral bone changes during the progression of MetS-OA, we studied two MetS mouse models: high-fat diet (HFD)-challenged and STR/Ort mice. Mice fed a HFD for 3 months had higher body weight, body fat mass, and serum glucose relative to CHD mice (Figure 1—figure supplement 1A-D). Serum lipid oxidation products assessed by malondialdehyde (MDA) level were also markedly higher in the HFD mice than in the CHD mice (Figure 1—figure supplement 1E). Therefore, HFD mice developed key components of MetS. We investigated the changes in osteoarthritic joints of HFD mice. Three-month-old mice were placed on a CHD or HFD for various periods, and subchondral bone changes were assessed. Mice fed a HFD for 3 and 5 months had different degrees of cartilage degeneration, indicated by proteoglycan loss and OARSI scoring (Figure 1A and B). Cartilage degeneration was not found in mice fed a HFD for 0.5 month or 1 month. Notably, three-dimensional micro-computed tomography (μCT) analysis showed a high subchondral bone mass phenotype in HFD-challenged male mice (Figure 1C and Figure 1—figure supplement 2). Tibial subchondral bone volume (BV)/tissue volume (TV) ratio, subchondral bone plate thickness (SBP.Th), and subchondral trabecular bone thickness (Tb.Th) increased dramatically in mice fed a HFD (vs. CHD) for all time periods tested (Figure 1D-F), indicating an early and severe thickening of the subchondral plate and trabecular bone. Subchondral trabecular number (Tb.N) was reduced in mice fed a HFD for 1, 3, or 5 months (Figure 1G), likely because of the fusion of the trabeculae. HFD-challenged female mice had a same high subchondral bone mass phenotype (Figure 1—figure supplement 3A), with increased BV/TV, SBP.Th, and Tb.Th but decreased Tb.N (Figure 1—figure supplement 3B-E). Moreover, osteophyte formation, another characteristic of osteoarthritis (Dieppe and Lohmander, 2005), was also identified in mice fed a HFD for 5 months (Figure 1—figure supplement 2). However, unlike PTOA mice that have apparent trabecular bone irregularity, as indicated by an increase in trabecular pattern factor (Tb.Pf) (Su et al., 2020; Zhen et al., 2013), HFD mice had similar subchondral Tb.Pf relative to CHD mice (Figure 1H). Thus, HFD-challenged mice exhibited a unique subchondral bone phenotype distinguishable from that of PTOA mice.

We also investigated the joint changes in STR/Ort mice, which have been shown to develop osteoarthritis spontaneously early in life with concomitant hypercholesterolemia and hyperlipidemia (Staines et al., 2017; Mason et al., 2001). Male STR/Ort and CBA control mice were used because male mice are known to have a higher incidence of OA than female mice (Staines et al., 2017). Higher body weight (Figure 1—figure supplement 4A) and greatly increased serum triglyceride and cholesterol levels were detected in STR/Ort mice relative to those of CBA control mice (Figure 1—figure supplement 4B, C). Significant proteoglycan loss in joint cartilage (Figure 1—figure supplement 4D) and increased OARSI scores (Figure 1—figure supplement 4E) were detected only in 4-month-old mice but not in 2-month-old STR/Ort mice. However, an increase in subchondral bone mass was detected in STR/Ort mice as early as 2 months of age, with higher BV/TV ratio (Figure 1—figure supplement 4F), SBP.Th (Figure 1—figure supplement 4G), and Tb.Th (Figure 1—figure supplement 4), as well as unchanged Tb.N (Figure 1—figure supplement 4J). Collectively, the MetS-OA mice, similar to our findings in human MetS-OA participants, developed progressive subchondral bone damage and, importantly, exhibited rapid increases in subchondral bone plate and trabecular thickness before the occurrence of cartilage degeneration.

We then attempted to elucidate the cellular changes that underlie subchondral bone thickening in MetS. Immunostaining analysis showed that osteocalcin (OCN)⁺ osteoblasts were localized primarily on the subchondral bone surface in mice fed a CHD (Figure 2A). Although OCN⁺ cell number per bone surface was unchanged in HFD mice relative to CHD mice (Figure 2B), OCN⁺ cell number per bone marrow area markedly increased and formed clusters (Figure 2A and C) in mice at 0.5 month and 1 month after a HFD challenge. The results suggest that osteoblasts are located on subchondral bone surface in normal healthy joints but aberrantly accumulate in bone marrow cavity in the joints of MetS-OA mice. A higher number of osteocytes in subchondral bone was also detected in HFD mice relative to CHD mice (Figure 2D). Tartrate-resistant acid phosphatase (TRAP) staining showed reduced bone surface TRAP⁺ osteoclasts in HFD mice relative to CHD mice (Figure 2E and F). Consistently, HFD challenge induced a significant decrease in VPP3⁺ osteoclasts on the bone surface (Figure 2G and H). We studied Tnfrsf11a^Cre/+; ROSA26^lsl-EYFP (RANK-EYFP) mice, in which RANK⁺ cells and their progeny are labeled with EYFP. Although the EYFP⁺ cells on the bone surface were reduced (Figure 2I and J), the EYFP⁺ cell number within the bone marrow cavity were elevated in HFD mice compared to those in CHD mice (Figure 2I and K). These results suggest that bone marrow RANK⁺ preosteoclasts may be unable to mature into bone surface osteoclasts.

We reasoned that the reserved preosteoclasts in subchondral bone may undergo cellular senescence, causing an inability to migrate to the bone surface and fuse together into mature osteoclasts. We investigated SnCs in the joints of a senescence reporter transgenic mouse line, p16^tdTom (Cdkn2a is the approved gene name for p16 protein, and Cdkn2a^tdTom is used hereafter). Intriguingly, 1-month and 3-month HFD challenge led to the accumulation of tdTom⁺ SnCs exclusively in the subchondral bone (Figure 3A and C). tdTom⁺ SnCs were not seen in articular cartilage until later 5 months after HFD challenge (Figure 3A and B) when significant cartilage degeneration occurs (Figure 1A, B). Detection of other cellular senescence markers showed consistently increased SA-βGal⁺ cell number at the subchondral bone/bone marrow (Figure 3D and E) but not in articular cartilage in mice fed a HFD for 1 month or 3 months (Figure 3D and F). Conversely, fewer SA-βGal⁺ cells were found in the same region of CHD mice. We also performed immunostaining for HMGB1, the redistribution of which from nucleus to extracellular as a secretory protein initiates the process of cellular senescence (Davalos et al., 2013). Although most cells had strong nuclear staining of HMGB1 in CHD mice, many cells lost nuclear HMGB1 in the subchondral bone of the mice fed a HFD for 1 month or 3 months (Figure 3—figure supplement 1A B). Downregulation/loss of Lamin B1 has been recognized as a biomarker and crucial step for the development of cellular senescence (Freund et al., 2012). Unlike the well-preserved Lamin B1 expression in the cell membrane of CHD mice, many bone marrow cells lost Lamin B1 expression in the subchondral bone of HFD mice (Figure 3—figure supplement 1C D). We also investigated cellular senescence in the subchondral bone of STR/Ort mice. Similar to the HFD-challenged mice, the number of SA-βGal⁺ cells in the subchondral bone marrow was higher in STR/Ort mice relative to CBA control mice at 2 months of age, when joint cartilage degeneration was not yet developed (Figure 3—figure supplement 2A B). Consistently, more subchondral bone marrow cells lost HMGB1 expression in STR/Ort mice relative to CBA mice (Figure 3—figure supplement 2C D), indicating that cellular senescence occurs at the pre- or early-osteoarthritis stage.

We then investigated whether osteoclast lineage cells undergo cellular senescence in HFD-challenged Cdkn2a^tdTom mice. Consistent with Figure 2G–2H, fewer VPP3⁺ osteoclasts at the bone surface (Figure 4A and B) and more bone marrow tdTom⁺ SnCs (Figure 4A and C) were detected in HFD mice than in CHD mice. However, the percentage of tdTom⁺ cells out of the bone surface VPP3⁺ cell population was not different in HFD mice compared with CHD mice (Figure 4A and D), indicating that mature osteoclasts did not undergo senescence in the subchondral bone of HFD-challenged mice. On the contrary, a markedly higher percentage of tdTom⁺ SnCs in RANK⁺ preosteoclasts was found in the subchondral bone marrow of HFD mice relative to CHD mice (Figure 4E and F). Approximately 49.2% and 56.5% of bone marrow RANK⁺ cells expressed tdTom after HFD challenge of 0.5 and 1 month, respectively. These results suggest that bone marrow preosteoclasts, but not bone surface mature osteoclasts, exhibit senescence-like changes in mice with HFD challenge.

To determine whether cellular senescence plays a role in osteoarthritis development, we generated conditional Tnfrsf11a^Cre/+; Cdkn2a^flox/flox mice, in which the senescence gene Cdkn2a is deleted in RANK⁺ osteoclast lineage cells (Maeda et al., 2012; Zou et al., 2016). Consistent with Figure 3D, SA-βGal⁺ cells accumulated in subchondral bone marrow in WT mice with 1 month HFD treatment, and importantly this increase was greatly dampened in the p16^cKO mice relative to WT mice (Figure 5—figure supplement 1A B). The results suggest that deletion of Cdkn2a in RANK⁺ cells efficiently prevents/blocks subchondral cellular senescence. We then investigated subchondral bone changes in the mice. Consistent with Figure 1, μCT analysis showed that tibial subchondral BV/TV ratio, SBP.Th, and Tb.Th were all higher in wild-type (WT) mice fed a HFD (vs. CHD) (Figure 5A-D). However, these subchondral bone alterations induced by 1 month HFD treatment were not significant in the p16^cKO mice. Moreover, the reduction in the number of bone surface osteoclasts (Figure 5E and F) and the increase in the osteoblast clusters in the bone marrow cavity (Figure 5G and H) induced by HFD treatment were both alleviated after Cdkn2a deletion in preosteoclasts. Of note, the μCT measurements (Figure 5A-D) and the histological parameters (Figure 5E-H) were not significantly different in the p16^cKO mice compared with the WT mice fed a normal CHD, suggesting that subchondral bone remodeling, as well as the activity of osteoclast and osteoblast lineage were not affected by Cdkn2a deletion in osteoclast lineage cells at baseline. Finally, we investigated joint cartilage changes and found that articular cartilage was well preserved in WT and p16^cKO mice fed a CHD. WT mice treated with 5 months of HFD exhibited obvious proteoglycan loss in the joint cartilage and an increased OARSI score, which were not observed in p16^cKO mice (Figure 5I and J). Consistently, HFD-induced increase in the percentage of MMP13⁺ chondrocytes, another feature of articular cartilage degeneration, was dramatically reduced in p16^cKO mice relative to WT mice (Figure 5K and L). Therefore, deletion of Cdkn2a in osteoclast lineage cells attenuated HFD-induced osteoarthritis progression.

One of the characteristics of SnCs is the SASP. We investigated whether subchondral bone preosteoclasts in HFD mice acquire SASP by performing proteomic profiling. We assessed the differentially expressed proteins in tibial plateau subchondral bone from p16 ^flox/flox (WT) and p16^cKO mice with HFD challenge (Figure 6A). Overall, 12 of 111 cytokines/growth factors were increased in the subchondral bone of HFD-challenged WT mice relative to that of those fed a normal CHD (Figure 6B). Of note, the levels of 9 of these 12 factors were restored in the subchondral bone of p16^cKO mice, suggesting that these factors are secreted from the p16 +senescent preosteoclasts.

We then used an in vitro osteoclast-based cell culture system (Xie et al., 2014), in which isolated bone marrow monocytes/macrophages were treated with M-CSF and RANKL for different durations to obtain mononuclear preosteoclasts and multinuclear mature osteoclasts. Cellular senescence of the preosteoclasts was induced by challenging the cells with oxidized low-density lipoprotein (oxLDL) (Figure 6C), a key mediator of MetS-associated abnormalities in multiple tissues (Holvoet et al., 2008; Hurtado-Roca et al., 2017). The mRNA expression of senescence genes Cdkn2a and Cdkn1a was upregulated, and proliferation marker Mki67 was downregulated in oxLDL-treated cells compared to those of vehicle-treated cells (Figure 6D). Moreover, number of SA-β-gal⁺ cells increased (Figure 6E and F) and the number of Lamin-B1⁺ cells decreased (Figure 6G and H) in response to oxLDL treatment. Therefore, oxLDL induces in vitro preosteoclast senescence efficiently. We then used quantitative real-time polymerase chain reaction (qRT-PCR) to evaluate whether the same secreted factors identified in the subchondral bone of HFD mice were also elevated in oxLDL-challenged preosteoclasts. oxLDL-treated cells had much higher expression of interleukin-1β (IL-1β) and IL-6, which are two common SASP factors. Moreover, seven factors identified in the subchondral bone of HFD mice, including osteopontin (OPN), Lipocalin-2, Cystatin C, IL-33, vascular endothelial growth factor (VEGF), cellular communication network factor 4 (CCN4), and platelet-derived growth factor BB (PDGF-BB), were also markedly increased in oxLDL-treated cells relative to vehicle-treated cells (Figure 6I), indicating that these are important SASP factors involved in osteoarthritis pathogenesis.

We tested whether the factors secreted by senescent preosteoclasts act on surrounding cells to regulate their activity in a paracrine manner. To do this, we prepared conditioned medium (CM) from control non-senescent preosteoclasts (Con-CM) and senescent preosteoclasts (SnC-CM) (Figure 6C). We first examined the functional changes of osteoclast lineage cells by incubating bone marrow Mo/Mac, osteoclast precursor cells, with the SnC-CM. SnC-CM inhibited the osteoclastogenesis ability of the non-adherent bone marrow Mo/Mac compared with Con-CM (Figure 6—figure supplement 1), indicating an inhibitory effect of preosteoclast SASP on osteoclastogenesis. We then test whether preosteoclasts lose their capacity to further differentiate into mature osteoclasts when become senescent. Cellular senescence of in vitro differentiated mononuclear preosteoclasts was induced by challenging the cells with hydrogen peroxide (H₂O₂) (Figure 6—figure supplement 2A), which has been widely used to achieve oxidative stress-induced cellular senescence (Toussaint et al., 2000). Senescence of preosteoclasts was successfully induced as indicated by upregulated expression of senescence genes Cdkn2a (Figure 6—figure supplement 2B) and nuclear loss of HMGB1 (Hernandez-Segura et al., 2018; Figure 6—figure supplement 2C) relative to vehicle-treated control cells. Moreover, bulk RNA-seq analysis revealed 4,056 differentially expressed genes in the senescent vs. control preosteoclasts (p<0.05). Comparison of our data with previously defined aging/senescence-induced genes (ASIGs) from publicly available mouse RNA-seq data sets (Aging Atlas database; KEGG pathway database; GO database; MSigD database) identified a total of 150 ASIGs in the senescent preosteoclasts (vs. control non-senescent preosteoclasts) (Figure 6—figure supplement 3A). Among these ASIGs, 31 genes were upregulated, and 119 genes were downregulated in the senescent preosteoclasts relative to control cells (Figure 6—figure supplement 3B). In the main biological process and molecular function genes, alterations in those involved in ‘Aging’, ‘Damaged DNA binding’, ‘chromatin DNA binding’, and ‘NF-kappaB binding’ were notable for their known links to senescence and SASP triggering (Tilstra et al., 2012; Di Micco et al., 2021; Figure 6—figure supplement 3C). Of note, osteoclast differentiation- and bone resorption-associated genes are among the most significantly downregulated genes (Figure 6—figure supplement 4), indicating a diminished osteoclast differentiation capacity of the preosteoclasts after becoming senescent. Therefore, the senescent preosteoclasts, on one hand, have declined differentiation capacity toward mature osteoclasts; on the other hand, secrete SASP factors that inhibit the differentiation of non-senescent osteoclast precursors to osteoclasts in a paracrine manner.

We then tested whether SnC-CM affects the activity of osteoblast lineage cells. Human bone marrow stromal cells (BMSCs), precursors of osteoblasts, incubated with SnC-CM, had greater osteoblast differentiation capacity than Con-CM, as detected by colony forming unit–osteoblasts (CFU-OB) (Figure 7C and D). SnC-CM did not affect the colony-forming capacity of the BMSCs (Figure 7A and B). Consistently, the expression of osteoblast differentiation markers alkaline phosphatase (Alpl), Bglap, and collagen type 1 A (Col1a1) were all upregulated in BMSCs incubated with SnC-CM (vs. Con-CM) (Figure 7E). The expression of RUNX2, the master regulator for the commitment of undifferentiated mesenchymal stem cells toward the osteoblast lineage, was unchanged in the cells treated with SnC-CM compared with Con-CM. These results suggest that the SASP factors produced by preosteoclasts promote late-stage osteoblast differentiation without affecting the lineage commitment of BMSCs.

We explored the mechanisms by which preosteoclast-secreted factors affect osteoblast differentiation. Most of these SASP factors, such as IL-1β, IL-6, VEGF, PDGF-BB, OPN, Lipocalin-2, myeloperoxidase (MPO), Cystatin C, resistin, and IL-33, are direct COX2 gene-activating factors (Hurtado-Roca et al., 2017; Chien et al., 2009; Zhang et al., 2010; Su et al., 2017; Li et al., 2018; Hamzic et al., 2013; Panagopoulos et al., 2017; Samad et al., 2001). COX2 is the inducible enzyme in the production of PGE2, which acts on osteoblastic precursors to stimulate osteoblast differentiation (Pilbeam, 2020). We detected the mRNA expression of COX2 in BMSCs using various culture media and measured PGE2 produced by the cells. COX2 gene expression was upregulated in the cells incubated in osteoblast differentiation medium (DM) compared with those incubated in growth medium (GM). Importantly, the addition of SnC-CM, but not Con-CM, stimulated much higher COX2 expression in the cells compared to those incubated with DM alone (Figure 7F). PGE2 protein level was also higher in the culture medium from the cells incubated with SnC-CM relative to that of the cells incubated with DM alone (Figure 7G). Further, treating the cells with selective COX2 inhibitor celecoxib significantly downregulated all 3 osteoblast differentiation marker genes stimulated by SnC-CM (Figure 7H), suggesting a requirement of the COX2-PGE2 pathway in the osteoblast differentiation of the BMSCs. Our results suggest that preosteoclast-derived SASP factors have a paracrine effect on surrounding osteoblast lineage cells for COX2-PGE2 activation, which consequently promotes osteoblast differentiation in a cell-autonomous fashion (Figure 7I).

To further define the role of COX2-PGE2 pathway in stimulating subchondral bone formation in vivo, we assessed COX2 expression in the subchondral bone of HFD-challenged mice using immunostaining analysis. Our results showed that there were very few COX2⁺ cells at cartilage in mice fed HFD for shorter periods (0.5, 1, 3, and 4 months) relative to mice fed CHD, whereas there was a much higher number of COX2⁺ cell number on bone surface (osteoblasts) and in mineralized bone (osteocytes) at these time points (Figure 8A, C and D). A higher number of COX2⁺ chondrocytes in cartilage was found in mice only after 5 months of HFD treatment (Figure 8A and B), during which severe cartilage degeneration occurs. Therefore, the majority of COX2⁺ cells were primarily subchondral osteoblasts and osteocytes during the progression of MetS-OA.

To investigate whether inhibiting COX2-PGE2 signaling could alleviate subchondral bone changes, such as subchondral plate thickening, and thus slow osteoarthritis progression in MetS-OA mice, we gavage-fed a selective COX2 inhibitor, celecoxib, to HFD mice at a dose of 16  mg/kg^–1 daily for 2 months. Subchondral bone alterations induced by the HFD were alleviated by celecoxib treatment relative to vehicle treatment (Figure 9A), as evidenced by significantly reduced subchondral BV/TV, SBP.Th, and Tb.Th. (Figure 9B-D). The reduction of Tb.N induced by a HFD was also normalized by celecoxib treatment (Figure 9E). Importantly, the degeneration of articular cartilage was also ameliorated after celecoxib treatment (Figure 9F and G). Celecoxib treatment in CHD mice did not change the architecture of the subchondral bone or articular cartilage. This result, consistent with the in vitro data, suggests that the elevation of COX2 levels in subchondral bone is a key mediator of HFD-induced subchondral bone alteration and osteoarthritis progression.

To determine whether selective COX2 inhibitor also alleviates joint structural alterations in humans with MetS-OA, we conducted a longitudinal comparison of standard knee osteoarthritis outcomes between PS-matched COX2 inhibitor users and non-selective nonsteroidal anti-inflammatory drug (NSAID) users. Of 315 knees of COX2 inhibitor users and 488 knees of non-selective NSAID users, we selected 239 COX2 inhibitor users along with pair-matched 239 non-selective NSAID users using a 1:1 PS-matching method for potential confounders (Supplementary file 1E). The mean follow-up duration for standard knee OA outcomes was 4.4 years (median: 4 years, 1st and 3rd quartiles of 1 and 8 years, respectively). All analyses were further stratified according to the presence or absence of the MetS-OA phenotype (i.e., MetS-OA and no MetS-OA). Supplementary file 1B summarizes the baseline characteristics of included participants before and after applying PS matching. The mean (± standard deviation) age of the matched participants was 61 years±9 (61.3 years±8.8 in COX2 inhibitor users and 60.9 years±9.2 in non-selective NSAID users) and both groups were 73% female. The hazard of medial JSN OARSI osteoarthritis progression was significantly lower in COX2 inhibitor users compared with non-selective NSAID users, in participants with the MetS-OA phenotype (HR, 0.18; 95% CI, 0.04–0.85) but not in participants without MetS-OA (HR, 0.96; 95% CI, 0.53–1.74) (Table 2). No significant difference was found between COX2-inhibitor users and non-selective NSAID users in knee osteoarthritis incidence or symptoms of knee osteoarthritis in either study strata. (Table 2). We further compared 24 month worsening in subchondral BML structural damage between COX2 inhibitor users and non-selective NSAID users in participants with the MetS-OA phenotype. Our results showed that MetS-OA COX2 Inhibitor users, compared to non-selective NSIAD users, had lower odds of subchondral damage worsening. This was evident as lower odds of worsening in the number of affected knee subregions with BMLs (OR, 0.35; 95% CI, 0.13–0.93) and lower odds of worsening in the BML scores (OR, 0.45; 95% CI, 0.20–0.99). (Table 3).

A representative knee radiograph showed grade 2 medial JSN (using the OARSI grading method) with 3.5 mm of joint space width in a 64-year-old woman with MetS-OA phenotype (MetS⁺) at baseline (Figure 9Hi). After 24 months of COX2 inhibitor use, follow-up radiographs of the same knee showed minimal progression of medial JSN, as evidenced by no interval change in OARSI grading and minimal narrowing of the joint space width to 2.9 mm (Figure 9Hii). Conversely, a knee radiograph of a 62-year-old MetS⁺ woman showed grade 2 JSN with minimal (3.7 mm) joint space width at baseline (Figure 9Hv); however, a follow-up radiograph of the same knee showed progression of JSN, as evidenced by interval change in OARSI grading to grade 3 with a joint space width of 1.0 mm (Figure 9Hvi). Consistently, magnetic resonance imaging (MRI) showed a significant increase in the size and number of subchondral bone marrow lesions in the knee of a non-selective NSAID user (Figure 9Hviii vs. vii) but unchanged subchondral bone marrow lesion size and number in the knee of a selective COX2 inhibitor user (Figure 9Hiv vs. iii). The decreased risk of medial JSN OARSI osteoarthritis progression in COX2 inhibitor users with MetS-OA phenotype is consistent with our findings in animal models of MetS-OA.