Dysregulated NADPH Oxidase Promotes Bone Damage in Murine Model of Autoinflammatory Osteomyelitis

This article is featured in In This Issue, p.1419

Autoinflammatory diseases represent a distinct class of disorders of the innate immune system. They are characterized by a pathological inflammation that typically arises spontaneously without detectable extrinsic cause and in the absence of autoantibodies or autoreactive T cells. The symptoms are rather diverse. The most characteristic include periodic fever attacks, skin rashes, arthralgia, myalgia, abdominal pain, arthritis, osteomyelitis, and other signs of systemic or organ specific inflammation (1–3). A number of autoinflammatory diseases are caused by a pathological hyperactivity of IL-1β pathway, either as a result of mutations in single genes affecting inflammasomes and other components of IL-1β activation machinery or from more complex causes in which the underlying genetic lesion is unknown (1, 3).

Bone damage or other types of bone involvement are common in IL-1β–driven autoinflammatory diseases (4). IL-1β promotes osteoclast activity by stimulating RANKL expression in osteoblasts and by direct binding to osteoclasts. This way it likely stimulates inflammatory bone resorption by these cells during the course of the disease (4, 5). Interestingly, different diseases of this group show different and distinct types of bone damage. Moreover, the bone damage is often observed only in a fraction of patients with a particular disease (4, 6, 7). These observations suggest that the character of genetic lesion, genetic modifiers, or other circumstances are critically affecting the outcome (4). However, the identity of these factors and the mechanisms of how they change the clinical picture are largely unknown.

One of the key activators of IL-1β pathway mutated in several autoinflammatory conditions is NLRP3 inflammasome. It is activated by aberrant ion fluxes; lysosomal damage by crystalline matter, such as silica or monosodium urate crystals; mitochondrial damage; presence of reactive oxygen species (ROS); and various microbial products and molecules associated with cellular damage (8–12). Several unifying mechanisms enabling recognition of such a variety of stress agents by a single type of inflammasome have been suggested, but none of them has yet gained universal acceptance (8, 9, 11–13). Production of ROS represents one such a mechanism that could connect cellular stress to NLRP3 inflammasome activation (for review, see 10, 14). In most of the cell types, there appear to be at least two main sources of ROS, NADPH oxidases, and mitochondria (15). In phagocytes, NOX2 NADPH oxidase is activated downstream of receptors for microbial products and other proinflammatory stimuli. It generates superoxide anion, which can be further converted to a number of additional ROS toxic to microorganisms. Various NADPH oxidases are also part of a broad array of signaling pathways in multiple cell types (16, 17). Mitochondrial ROS are produced mainly as a result of respiratory chain activity, and their generation can be enhanced by stress or mitochondrial damage (18–20). Although initial reports suggested that NADPH oxidase–derived ROS are critical for NLRP3 inflammasome triggering, more recent studies rather support the view that mitochondria are the essential source of ROS required for its activation (21, 22). However, whether NADPH oxidase ROS can contribute to NLRP3 inflammasome activation when deregulated as a result of neutrophil priming or during diseases that result in exaggerated NADPH oxidase-dependent ROS production has not been studied.

Several studies have shown increased production of ROS in monocytes from autoinflammatory disease patients (10, 23–28). In some of these works it has been proposed that these ROS are of mitochondrial origin (10, 23), but there are only limited options of how to study this aspect in patients. The effects of increased ROS production, whether of mitochondrial or NADPH oxidase origin, on the development and/or severity of autoinflammatory diseases is currently unknown.

There are relatively few mouse models of autoinflammatory bone diseases. One of the best characterized is Pstpip2^cmo mouse strain, which spontaneously develops severe bone and soft tissue inflammation mainly in hind paws and tail. In several aspects, the disease resembles a human condition known as chronic recurrent multifocal osteomyelitis (CRMO) and was thus termed chronic multifocal osteomyelitis (CMO). From there, the strain derives its name Pstpip2^cmo (29). The disease is caused by a point mutation in the gene coding for the adaptor protein PSTPIP2 (30). As a result, no PSTPIP2 is detectable in these mice at the protein level (31). The mechanism by which PSTPIP2 deficiency leads to CMO disease is only partially understood. It binds several inhibitory molecules, including PEST-family protein tyrosine phosphatases, phosphoinositide phosphatase SHIP1, and inhibitory kinase Csk, which likely mediate its negative regulatory effect on the inflammatory response (32, 33). In addition, it has been reported that osteomyelitis in Pstpip2^cmo mice is completely dependent on excessive IL-1β production by neutrophilic granulocytes (34–36). Genetic studies suggest a combined involvement of the NLRP3 inflammasome and a poorly characterized mechanism dependent on caspase-8. A relatively limited role of neutrophil proteases has also been demonstrated (36, 37). The involvement of NLRP3 inflammasome suggests that cellular stress and ROS might be involved in CMO disease pathology, especially, when we consider the fact that neutrophils are very potent producers of NADPH oxidase–derived ROS. Moreover, ROS are also activators of osteoclasts (38), a cell type likely responsible for inflammatory bone damage in CMO mice (39). In this study, we show that in Pstpip2^cmo neutrophils, superoxide generation by NADPH oxidase is profoundly dysregulated and these cells produce substantially increased amounts of superoxide in response to variety of stimuli. Strikingly, the dysregulated superoxide production by these neutrophils does not have a strong effect on IL-1β production and soft tissue inflammation, but rather on the bone inflammation and subsequent bone damage, suggesting that the role of NADPH-oxidase–derived ROS is not in triggering the CMO but rather in directing the damage accompanying this disease to the bones.

Abs to the following murine Ags were used: RAC1/2/3 (catalog no. 2465; Cell Signaling Technology, Danvers, MA); p47phox (D-10) and ERK2 (C-14) (Santa Cruz Biotechnology, Dallas, TX); B220-biotin, TER119-biotin, c-Kit-biotin, CD3ε-biotin, Ly6G-biotin, CD115-biotin, CD11b-allophycocyanin, CD11b-FITC, CD11b-PE, B220-FITC, Ly6C-PE-Cy7, Ly6C-FITC, Ly6G-FITC, Ly6G-allophycocyanin, c-Kit-PE, Sca-1-allophycocyanin, and CD16/32-PE/Cy7 (BioLegend, San Diego, CA); CD34-FITC, DX5-biotin, F4/80-biotin, and Thy1.2-FITC (eBioscience, ThermoFisher Scientific, Waltham, MA); Fc Block (2.4G2) (BD Biosciences, San Jose, CA); and HRP-conjugated goat anti-mouse IgG (Sigma-Aldrich) and HRP-goat anti-rabbit (Bio-Rad, Hercules, CA). The mouse mAb that recognizes murine PSTPIP2 has been described earlier (33). Heat-aggregated IgG was prepared as follows: IgG was purified from mouse serum (Sigma-Aldrich) on protein A–Sepharose (GE Healthcare, Uppsala, Sweden), transferred to PBS, and concentrated to 30 mg/ml on an Amicon Ultracel–30K unit (Millipore, Merck, Darmstadt, Germany). The aggregation was induced by heating to 63°C for 30 min.

In this study we also used luminol, HRP, LPSs from Escherichia coli O127:B8, fMLP, PMA, (all from Sigma-Aldrich), L-012 (Wako Chemicals), TNF-α, G-CSF (PeproTech, Rocky Hill, NJ), U0126 (Cell Signaling Technology), and Gö6976 (Calbiochem, Merck). Silica (silicon dioxide crystals) was obtained from Sigma-Aldrich. To enable fluorescent labeling for microscopy, 5 mg/ml silica particles were first coated with nonfat dry milk (2% in PBS, 1 h at room temperature) and then labeled with 5 μM Cell Proliferation Dye eFluor 670 (eBioscience), 30 min at 37°C.

Pstpip2^cmo mouse strain (C.Cg-Pstpip2^cmo/J) carrying the c.293T→C mutation in the Pstpip2 gene (on BALB/C genetic background) resulting in an L98P change in the PSTPIP2 protein (29, 30), B6.129S-Cybb^tm1Din/J lacking NADPH oxidase subunit gp91phox (40); MyD88 deficient mouse strain (B6.129P2(SJL)-MyD88^tm1.1Defr/J, derived from MyD88^fl mice (41); B6.SJL-Ptprca Pepcb/BoyJ (CD45.1⁺) congenic strain (42), B6.Cg-Tg(S100A8-cre,-EGFP)1Ilw/J with granulocyte-specific CRE expression (MRP8-Cre) (43); and Gt(ROSA)26Sor^tm1(DTA)Lky/J strain (44), in which diphtheria toxin expression can be induced by CRE recombinase, were obtained from The Jackson Laboratory (Bar Harbor, ME). Pstpip2^cmo mouse strain was backcrossed on C57BL/6J background for at least 10 generations and then used in the majority of experiments, with the exception of experiments in Figs. 1E, 4A, 4D–F and Supplemental Fig. 1. For these experiments, the original Pstpip2^cmo strain on BALB/c genetic background has been selected due to the higher number of neutrophils that could be obtained by the negative selection method and due to the better quality of immortalized granulocyte progenitors derived from this strain. Both genetic backgrounds showed similar disease symptoms and similar dysregulation in superoxide production. The BALB/c and C57BL/6J inbred strains were obtained from the animal facility of Institute of Molecular Genetics, Academy of Sciences of the Czech Republic (Prague, Czech Republic). Pstpip2^cmo-DTA-MRP8-Cre mouse strain was generated by breeding of the Pstpip2^cmo mice on C57BL/6J background with Gt(ROSA)26Sor^tm1(DTA)Lky/J strain mouse strain. Breeding this strain to B6.Cg-Tg(S100A8-cre,-EGFP)1Ilw/J mice carrying Cre transgene under the control of granulocyte-specific MRP8 promoter resulted in the generation of Pstpip2^cmo-DTA-MRP8-Cre strain lacking almost all granulocytes (Supplemental Fig. 3B, 3C). Mice were housed and bred in an accredited animal facility at the Institute of Molecular Genetics of the Czech Academy of Sciences (registration number CZ11760038). They were maintained under specific pathogen-free conditions. Animals were fed by standard breeding fortified diet (Altromin) cereal-based (soy, wheat, corn) fixed formula, which is free of alfalfa and fish/animal meal and deficient in nitrosamines, containing 22.6% crude protein, 5% crude fat, 4.5% crude fiber, 7.1% crude ash, autoclavable, and increased vitamin content. The drinking water was purified via reverse osmosis system and chlorinated by chlorine dioxide (ClO₂) as an alternative disinfectant to prevent secondary contamination. The final concentration of active chlorine was maintained between 0.6 and 1.0 ppm (in acid pH 4–5). Experimental cohorts, sex and age matched, were made from both genders, and standard randomization was applied. Unless indicated otherwise, age of animals ranged from 6 to 14 wk. Experiments in this work that were conducted on animals were approved by the Expert Committee on the Welfare of Experimental Animals of the Institute of Molecular Genetics and by the Academy of Sciences of the Czech Republic (registration numbers 69/2014, 62/2015, 50/2016, 66/2016, 45/2018) and were in agreement with local legal requirements and ethical guidelines.

All primary cells and cell lines were cultured at 37°C with 5% CO₂ in IMDM supplemented with 10% FCS and antibiotics. For bone marrow (BM) cell isolation, mice were sacrificed by cervical dislocation, BM was flushed with PBS supplemented with 2% FCS, and erythrocytes were lysed in an ACK buffer (150 mM NH₄Cl, 0.1 mM EDTA [disodium salt], 1 mM KHCO₃). Murine neutrophils were isolated from BM cells using anti-biotin MicroBeads and LS magnetic columns (Miltenyi Biotec, Bergisch Gladbach, Germany). For negative selection, cells were labeled with biotinylated Abs to B220, F4/80, DX5, c-Kit, CD3ε, CD115, and Ter119 prior to magnetic bead purification. For positive selection, only anti-Ly6G biotin was used. The purity of isolated cells was determined by flow cytometry. Primary murine monocytes were sorted from BM cells as Ly6G negative, Ly6C highly positive, and side scatter–low cells using BD Influx sorter (BD Biosciences). The following cell lines were used in this study: HEK293FT cells (Invitrogen), Platinum Eco cells (Plat-E cells; Cell Biolabs, San Diego, CA), and immortalized granulocyte progenitors. For preparation of immortalized granulocyte progenitors we used a modified version of the protocol for generation of immortalized macrophage progenitors (45). The progenitors were first enriched by the depletion of Mac-1⁺, B220⁺, and Thy1.2⁺ from mouse BM cells and cultured in the presence of IL-3, IL-6, and SCF (supplied as culture supernatants from HEK293FT cells transfected with constructs coding for respective cytokines) for 2 d. Next, progenitors were transduced with ER-HoxB8 construct. The transduced cells were enriched for the GMP progenitor population by FACS (Lin⁻, Sca-1⁻, c-Kit⁺, FcγR⁺, CD34⁺) and propagated in a medium containing 1 μM β-estradiol and 1% SCF-containing supernatant. Granulocyte differentiation was induced by β-estradiol withdrawal or by the β-estradiol withdrawal and replacement of SCF for G-CSF (50 ng/ml).

Single-cell suspensions of BM cells were incubated with Fc block and fluorophore-conjugated Abs and analyzed on a BD LSR II flow cytometer. For calcium response measurement, single-cell suspensions of BM from 6- to 8-wk-old mice were loaded with 2 μM calcium indicator Fura Red (Invitrogen). Samples were analyzed using a BD LSR II flow cytometer for 30 s at rest and then another 210 s after activation (with fMLP, Silica, or E. coli with OD = 0.8). The relative calcium concentration was measured as a ratio of the Fura Red fluorescence intensity elicited by excitation wavelengths of 405 nm (emission measured at 635–720 nm) and 488 nm (emission measured at 655–695 nm). Data were acquired on a BD flow cytometer LSR II. Granulocytes were gated according to forward and side scatter properties. For F-actin detection, BM cells were fixed with 4% formaldehyde and labeled with fluorophore-conjugated Abs to CD11b, Ly6C, and Ly6G. Next, the cells were permeabilized with l-α-lysophosphatidylcholine (80 μg/ml; Sigma-Aldrich) and simultaneously stained with Alexa Fluor 488–conjugated phalloidin (Invitrogen). The cell fluorescence was measured on a BD LSR II flow cytometer. Granulocytes were gated as CD11b⁺, Ly6C^int, and Ly6G⁺. The data were analyzed with FlowJo software (Tree Star, Ashland, OR).

Superoxide production in vitro was assessed by luminol-based chemiluminescence assay as published previously (46, 47). BM cells or purified murine neutrophils in IMDM supplemented with 0.2% FCS were plated in a density of 10⁶ cells per well into a black 96-well plate (SPL Life Sciences, Naechon-Myeon, Korea). Cells were rested for 30 min at 37°C and 5% CO₂. Then, luminol at final concentration 100 μM and stimuli (100 ng/ml LPS, fMLP 1 μg/ml, TNF-α 10 ng/ml, E. coli OD₆₀₀ ∼0.8–5× diluted, silica 50 μg/cm², heat-aggregated murine IgG 300 μg/ml, PMA 100 ng/ml) were added. Luminescence was measured immediately on an EnVision plate reader (Perkin Elmer, Waltham, MA); each well was scanned every minute for 70 min. For fMLP-induced superoxide production, scanning every 10 s for 5 min was also used as indicated in figure legends. To measure superoxide production by cells in suspension, the cells were kept at 10⁷ per 0.9 ml IMDM with 0.2% FCS and 100 μM luminol in an Eppendorf tube at 37°C. After stimulation with 100 μl silica (1 mg/ml), every 5 min a 100-μl aliquot of cell suspension was gently transferred into an empty well of a black 96-well plate and the luminescence was immediately measured on an EnVision plate reader. When fMLP (1 μg/ml) was used as a stimulant, only a single aliquot of 10⁶ cells was measured in 10-s intervals in a single well immediately after cell transfer to the plate and activation. For exogenous peroxidase treatment, cells were incubated with HRP (10 μg/ml) for 30 min prior to stimulation with silica.

To assess ROS production in vivo, mice were i.p. injected with luminescence reporter L-012 in final concentration 75 mg/kg (1.8 mg/25 g mouse) dissolved in PBS as previously described (48). Luminescence signal was acquired by Xtreme whole body imager (Bruker, Billerica, MA), with the following settings: binning 8 × 8, exposure time: 5 min. The quantification of photon counts was performed in Molecular Imaging Software (Bruker).

Generation of MSCV-PSTPIP2-EGFP construct was as follows. The coding sequence of mouse PSTPIP2 was amplified from cDNA of mouse common myeloid progenitors and subcloned into pXJ41-EGFP cloning vector (Chum et al. 2016). IRES and Thy1.1 coding sequence was removed from MSCV-IRES-Thy1.1 retroviral vector (Clontech, Mountain View, CA) by digestion with EcoRI and ClaI followed by blunt ligation. PSTPIP2-EGFP coding sequence was then subcloned into modified MSCV vector using BglII and XhoI restriction sites to generate MSCV-PSTPIP2-EGFP.

Generation of MSCV-mPSTPIP2-TetOn inducible constructs. Wild-type (WT) and mutated sequences (W232A or 3YF) of mouse PSTPIP2 described earlier (33) were fused to C-terminal EGFP by PCR using P2A sequence as a linker. Fusion constructs were cloned into pLVX-Tet3G doxycycline inducible vector (Clontech) using AgeI and BamHI restriction sites. Resulting vectors were used as templates to amplify the Tet-On 3G, TRE3G, and PSTPIP2 sequences by PCR, and the resulting product was cloned into MSCV-IRES-EGFP vector using ClaI and BglII restriction sites.

For confocal microscopy, c-kit+ stem and progenitor cells were obtained from BM of Pstpip2^cmo (C57BL/6J) mice using magnetic purification (c-kit–biotin Ab, Anti-biotin microbeads). Cells were expanded in IL-3, IL-6, and SCF-supplemented media for 20 h, then infected with PSTPIP2-EGFP retroviral construct. For the production of replication incompetent retrovirus, ecotropic packaging cells (Plat-E) were plated in a 10-cm dish and transfected with 24 μg of plasmid DNA using Lipofectamine 2000 Reagent (Life Technologies) according to the manufacturer’s instructions. Virus-containing supernatant was collected, concentrated with Amicon Ultra centrifugal filters with molecular mass cut-off 100 kDa (Merck Millipore), and immediately used to infect the cytokine expanded c-kit+ BM cells. These cells were centrifuged with 150 μl of concentrated virus supernatant and 2.4 μl of Lipofectamine 2000 Reagent (Sigma-Aldrich) at 1250 × g for 90 min at 30°C and then incubated for another 4 h at 37°C in 5% CO₂ in a humidified incubator before the exchange of the media. Immortalized granulocyte progenitors were propagated in IMDM with 1 μM β-estradiol and 1% SCF-containing supernatant and then infected with PSTPIP2 mutant constructs using the same procedure described above.

RNA was purified with Zymo Research Quick-RNA Miniprep Plus Kit from neutrophils isolated by positive selection (see above). The reverse transcription was performed with RevertAid First Strand cDNA Synthesis Kit (ThermoFisher Scientific). Real-time quantitative PCR was carried out using LightCycler 480 SYBR Green I Master mix (Roche) and the following primers: 5′-TGTAATGAAAGACGGCACACC-3′ + 5′-TCTTCTTTGGGTATTGCTTGG-3′ for IL-1β and 5′-GATCTGGCACCACACCTTCT-3′ + 5′-GGGGTGTTGAAGGTCTCAAA-3′ for β-actin on Roche LightCycler 480 Instrument II. The primer functionality was verified on LPS-stimulated BM-derived dendritic cells.

One day postinfection, EGFP-positive cells were sorted on an Influx sorter and injected into sublethally irradiated (6 Gy in a single dose) CD45.1⁺ recipient mice. After 2 wk, mice were sacrificed, and BM cells isolated and resuspended in IMDM with 0.1% FCS. The cells were activated by 50 μg/cm² fluorescent silica (see above) in a 96-well plate for 10 min. The cells were transferred to 4% paraformaldehyde in PBS and fixed at room temperature for 20 min. Cell nuclei were stained with 10 μg/ml Hoechst 33258 (Sigma-Aldrich) for 15 min. Cells were than washed two times with PBS, resuspended in 150 μl of ddH2O, and centrifuged on glass slide at 300 × g for 5 min using Centurion Scientific K3 cytospin centrifuge (Centurion Scientific, Stoughton, U.K.). Cell samples were then mounted in 10 μl of DABCO mounting reagent (Sigma-Aldrich) and covered with glass coverslip (Zeiss, Oberkochen, Germany). The microscope setup was as follows: sequential two-color imaging was performed using a Leica TCS SP8 laser scanning confocal microscope (Leica, Wetzlar, Germany) with a 63 × 1.4 numerical aperture oil-immersion objective. Acquired images were manually thresholded to remove signal noise detected outside of the cell using ImageJ software.

For Western blotting, the cells were washed and resuspended in IMDM with 0.1% FCS at a concentration of 1–4 × 10⁷ cells/ml. Subsequently, the cells were stimulated as indicated at 37°C. The activation of cells was stopped by the addition of an equal volume of a 2× concentrated SDS-PAGE sample buffer (128 mM Tris [pH 6.8], 10% glycerol, 4% SDS, 2% DTT), followed by the sonication and heating of the samples (99°C for 2 min). The samples were analyzed by SDS-PAGE followed by Western blotting. For detection of p47phox phosphorylation, phosphoproteins were isolated from BM cells using PhosphoProtein purification Kit (Qiagen, Hilden, Germany) according to manufacturer’s instructions, followed by detection of p47phox by immunoblotting.

A total of 2 × 10⁷ neutrophils (silica activated or not) were lysed in 1 ml lysis buffer (25 mM HEPES [pH 7.2], 150 mM NaCl, 10 mM MgCl₂, 1 mM EDTA, 1% NP-40, 10% glycerol, 100 × diluted Protease Inhibitor Cocktail Set III [Calbiochem, Merck]) containing 5 μg PAK-RBD-GST (RAC-binding domain from PAK1 fused to GST, isolated from E. coli strain BL21 transformed with corresponding expression plasmid). After preclearing the lysate by centrifugation, the complexes of active RAC and PAK-RBD-GST were isolated on Glutathione Sepharose (GE Healthcare). RAC was then detected by immunoblotting.

Mice for in vivo imaging were anesthetized by i.m. injection of Zoletil (20 mg/ml)–Xylazine (1 mg/ml) solution with Zoletil dose 100 mg/kg and Xylazine dose 1 mg/kg.

Hind paws were scanned in in vivo x-ray microcomputed tomography (μCT) Skyscan 1176 (Bruker). Scanning parameters were voltage, 50 kV; current, 250 μA; filter, 0.5 mm aluminum; voxel size, 8.67 μm; exposure time, 2 s; rotation step, 0.3° for 180° total; object to source distance, 119.271 mm; and camera to source distance, 171.987 mm; with time of scanning, 30 min. Reconstruction of virtual slices was performed in NRecon software 1.6.10 (Bruker) with setup for smoothing = 3, ring artifact correction = 4, and beam hardening correction = 36%. Intensities of interest for reconstruction were in the range of 0.0045 to 0.0900. For reorientation of virtual slices to the same orientation, the DataViewer 1.5.2 software (Bruker) was used.

For μCT data analysis, CT Analyzer 1.16.4.1 (Bruker) was used. The volume of interest was chosen there containing the distal part of hind paw starting from the half of metatarsus. Based on differences of x-ray absorption, three parts were analyzed separately: the whole volume of interest, the newly formed bone connected mostly with arthritis, and the area inhabited by the original bone of phalanges and metatarsi. The total volume was recorded for all three parts. For original and new bone, other parameters from two-dimensional and three-dimensional analysis were recoded to describe changes in the structure, namely, surface of the bone, surface/volume ratio, number of objects, closed porosity, mean fractal dimension, mean number of objects per slice, mean closed porosity per slice, and mean fractal dimension per slice. Scans with technical artifacts caused by spontaneous movements of animals were excluded from the analysis. Raw data are available upon request.

Murine paws were homogenized in 1 ml RIPA lysis buffer (20 mM TRIS [pH 7.5], 150 mM NaCl, 1% Nonidet P-40, 1% sodium deoxycholate, 0.1% SDS) containing 5 mM iodoacetamide (Sigma-Aldrich) and 100 × diluted Protease Inhibitor Cocktail Set III (Calbiochem, Merck) using Avans AHM1 Homogenizer (30 s, speed 25). Any insoluble material was removed by centrifugation (20,000 × g, 5 min, 2°C), and concentration of the proteins in the samples were normalized to the same level using Bradford solution (AppliChem). Concentrations of IL-1β in the samples were determined by Ready-SET-Go! ELISA kits from eBioscience according to the instructions of the manufacturer.

The paws were fixed in 10% formol solution for 24 h and decalcified in Osteosoft (Merck) solution for 1 wk, followed by paraffin embedding and histological cutting. The slides were stained in automatic system Ventana Symphony (Roche), and slides were scanned in Axio Scan.Z1 (Zeiss). The image postprocessing and analysis was done in Zen software (Zeiss).

The p values were calculated in GraphPad Prism software (GraphPad Software, La Jolla, CA) using unpaired t test (two-tailed) for data in Figs. 2B–G, 2I, 3C, 5E; one-way ANOVA with Tukey–Kramer or Bonferroni multiple comparison posttest for data in Figs. 1E, 4A, 6E, 7B, and 7D; Mann–Whitney U test for Fig. 5C; and Gehan–Breslow–Wilcoxon test for disease-free curve in Fig. 6C. Symbol meanings are as follows: n.s. p > 0.05, *p ≤ 0.05, **p ≤ 0.01, ***p ≤ 0.001, ****p ≤ 0.0001. Unless stated otherwise, “n” in figure legends represent number of experiment repeats, points in column scatter plots represent biological replicates (in most cases animals), error bars in these plots show median with interquartile range, and points and error bars in superoxide production time course curves (arbitrary luminescence intensity curves) represent mean ± SEM values obtained from two to eight technical replicates.

Disease development in Pstpip2^cmo mice is, in part, dependent on NLRP3 inflammasome (37). Because ROS are involved in the NLRP3 inflammasome regulation, we tested if their production was dysregulated in Pstpip2^cmo BM cells. We isolated these cells from WT C57BL/6 and Pstpip2^cmo mice (backcrossed to the same genetic background) and stimulated these cells with silica particles, a well-established activator of NLRP3 inflammasome (49) employed in previous studies of Pstpip2^cmo mice (33–35). Strikingly, this stimulation led to a substantially higher superoxide production by Pstpip2^cmo cells, when compared with their WT counterparts (Fig. 1A). This assay was performed in a 96-well plate, where the response could be affected by adhesion of the cells to plastic. Therefore, we carried out a similar measurement on the cells maintained strictly in suspension. This resulted in signals of lower intensity. However, the dysregulation of superoxide production by Pstpip2^cmo cells could still be observed (Fig. 1B).

Because superoxide generation is a characteristic feature of neutrophils, which form a large fraction of BM leukocytes (Fig. 1C), we have isolated these cells for further testing. Silica stimulation of adherent neutrophils, isolated by negative selection, led to even higher production of superoxide when compared with the full BM. Moreover, the difference between WT and Pstpip2^cmo cells was still preserved (Fig. 1D). In our experiments, negatively selected neutrophils were typically more than 90% pure. However, a large fraction of the contaminating cells were monocytes. Because these cells are also known to respond by ROS production to a variety of stimuli, we have analyzed superoxide generation by purified monocytes. To determine the impact of these cells on our results, we have adjusted the quantity of monocytes to 10% of the neutrophil numbers, which is similar to the amount of monocytes contaminating our neutrophil samples prepared by negative selection. We compared the response of these monocytes with the superoxide production by neutrophils isolated by positive selection on Ly6G. This purification resulted in virtually pure (more than 99%) neutrophils. The response of purified monocytes was almost two orders of magnitude lower than that of purified neutrophils, and there was no significant difference between WT and Pstpip2^cmo cells (Fig. 1E). Addition of exogenous peroxidase (HRP) to compensate for lower myeloperoxidase expression in monocytes (50) did not substantially alter the results (Fig. 1E). These data demonstrated that in the neutrophil samples prepared by negative selection, the monocyte contribution to the measured superoxide production is negligible. They also lead to the conclusion that even in nonseparated BM, the vast majority of superoxide originated from neutrophils and that neutrophils are responsible for enhanced superoxide production by Pstpip2^cmo BM cells.

To find out how universal the superoxide overproduction in Pstpip2^cmo neutrophils is, we treated either BM cells or purified neutrophils with silica, PMA, live E. coli bacteria, heat-aggregated mouse IgG (as a model of immunocomplexes), TNF-α, LPS, or fMLP. All these experiments demonstrated dysregulated superoxide production in Pstpip2^cmo BM cells (Fig. 2A–D, 2F) and purified neutrophils (Fig. 2B, 2E). The same dysregulation was also observed in the BM cells with Pstpip2^cmo mutation on BALB/c genetic background (Supplemental Fig. 1A). These results show that PSTPIP2 deficiency renders neutrophils more sensitive and prone to produce more ROS than WT cells. Interestingly, unstimulated BM cells from Pstpip2^cmo mice produced low but detectable levels of superoxide even in the absence of any stimulus. This constitutive production has not been observed in WT BM (Fig. 2A, 2G). It also was not observed in both WT and Pstpip2^cmo cells maintained in suspension (data not shown). In addition, nonadherent cells also did not show any superoxide production in response to treatment with LPS, TNF-α, and fMLP when measured in 5-min intervals (data not shown). However, in the case of fMLP, rapid transient response was detectable within the first 5 min of the measurement (Fig. 2H). Maximum superoxide production was not significantly altered in Pstpip2^cmo BM cells. However, it was more sustained, and total superoxide production was thus higher than in WT cells (Fig. 2I).

Higher superoxide production under resting conditions and after activation with a wide range of stimuli demonstrates general dysregulation of pathways leading to superoxide production in Pstpip2^cmo neutrophils. This dysregulation could be cell intrinsic because of PSTPIP2 deficiency or a side effect of ongoing bone inflammation, which could prime BM neutrophils located in the proximity of the inflamed tissue. It has previously been reported that autoinflammation in Pstpip2^cmo mice is completely dependent on IL-1β and its receptor (34, 35). Signaling through this receptor is critically dependent on MyD88 adaptor protein (51). To determine whether the observed overproduction of ROS in Pstpip2^cmo neutrophils is not just the effect of ongoing inflammation, we crossed Pstpip2^cmo mice with MyD88-deficient strain to block IL-1β signaling. As expected, Pstpip2^cmo mice were in the absence of MyD88 completely protected from the disease development as determined by visual inspection (Fig. 3A) and x-ray μCT of hind paws (Fig. 3B). MyD88-deficient Pstpip2^cmo BM cells displayed the same dysregulation in superoxide production triggered by a variety of stimuli as Pstpip2^cmo cells. Treatments with LPS or E. coli were the only exceptions where the response was substantially lower in both Pstpip2^cmo/MyD88^−/− and MyD88^−/− cells, probably due to the higher dependence of signaling triggered by these activators on the TLR/MyD88 pathway. Nevertheless, even in MyD88-deficient cells, the cmo mutation gave rise to a stronger response to these two stimuli when compared with MyD88-deficient cells without the cmo mutation (Fig. 3C, Supplemental Fig. 1B). These results demonstrate the cell-intrinsic dysregulation of NADPH oxidase machinery that is not caused by chronic exposure to the inflammatory environment.

Another potential explanation for the observed dysregulation of superoxide generation is that Pstpip2^cmo neutrophils are in a constitutively primed state. Increased superoxide production is one of the hallmarks of neutrophil priming. However, number of other changes also characterize primed neutrophils, including increased CD11b surface expression, actin cytoskeleton reorganization, and increase in IL-1β promoter activity (52–58). We have analyzed all these parameters, but we could not detect any alterations in these features (Fig. 3D–F).

To elucidate the mechanism of how PSTPIP2 suppresses superoxide production, we employed conditionally immortalized Pstpip2^cmo granulocyte progenitors (45) we had established previously (33) and reconstituted these cells with doxycycline-inducible retroviral constructs coding for WT PSTPIP2 and its mutated versions unable to bind PEST-family phosphatases (W232A) and SHIP1 (3YF) (32, 33). After maturation of these progenitors into neutrophils and induction of PSTPIP2 expression with doxycycline, we treated these cells with silica and measured superoxide production. We observed around 50% reduction of superoxide generation in cells expressing WT PSTPIP2. In contrast, both mutated versions of PSTPIP2 were unable to substantially inhibit silica-induced superoxide production (Fig. 4A), despite similar expression levels of these constructs (Fig. 4B).

To analyze subcellular localization of PSTPIP2 during silica stimulation, we isolated BM progenitors from Pstpip2^cmo mice and transduced these cells with retroviral construct coding for PSTPIP2 fused to EGFP. Next, we transplanted these cells into lethally irradiated mice, and after 2 wk, we collected neutrophils expressing PSTPIP2-EGFP for microscopy analysis. In neutrophils, PSTPIP2 showed diffuse distribution throughout the cytoplasm, with occasional formation of speckles in a small fraction of cells (Supplemental Fig. 2, left panel, see an arrowhead). After addition of fluorescently labeled silica particles, neutrophils interacted with these particles and phagocytosed some of them (Supplemental Fig. 2, right panel, see an arrowhead). However, we did not observe any changes in PSTPIP2 subcellular localization during this process (Supplemental Fig. 2). This result suggests that large-scale redistribution of PSTPIP2 inside the cells is not part of the mechanism of how PSTPIP2 controls neutrophil activity during the treatment with silica.

To identify the dysregulated process leading to superoxide overproduction at the biochemical level, we measured the calcium response in WT and Pstpip2^cmo BM cells. Cells were loaded with Fura Red dye and stimulated with silica particles. We observed the same calcium response in both WT and Pstpip2^cmo cells (Fig. 4C), indicating that proximal signaling steps leading to calcium response are not responsible for increased ROS production in Pstpip2^cmo cells.

One of the major events further downstream is phosphorylation of NADPH oxidase cytosolic subunits by members of protein kinase C (PKC) family, including phosphorylation of p47phox, which then serves as an assembly hub for building the active NADPH oxidase complex (16, 59). To detect p47phox phosphorylation, we isolated phosphoproteins from untreated and silica-treated BM cells or purified neutrophils and detected p47phox in the isolated material by immunoblotting. In both Pstpip2^cmo BM cells and neutrophils, we found substantially stronger phosphorylation of p47phox when compared with WT cells (Fig. 4D). This difference was observed as early as 5 min after stimulation and was maintained for at least 30 min (Fig. 4E).

Although PKCs are critical for p47phox activation, costimulatory effects of other kinases have also been demonstrated (59, 60). Of these, ERK MAP kinase was shown to be dysregulated in silica-treated CMO neutrophils (33). To assess the roles of PKCs and ERK pathway in p47phox dysregulation, we have treated Pstpip2^cmo neutrophils either with PKC inhibitor Gö6976 or MEK1/MEK2 inhibitor U0126 prior to activation with silica. Only the treatment with PKC inhibitor led to specific block of p47phox phosphorylation (Fig. 4F).

Small G-protein RAC is another critical component of active NADPH oxidase. We have tested the activation status of RAC after silica treatment of BM cells, but no difference between WT and Pstpip2^cmo cells has been observed (Fig. 4G).

Collectively, these data suggest that PSTPIP2 via its binding partners suppresses pathways leading to PKC-mediated p47phox phosphorylation and that this is the mechanism by which PSTPIP2 attenuates NADPH oxidase activity and superoxide production.

To analyze the ROS production in vivo during the disease development, we used luminol derivative L-012 to visualize ROS generation in living anesthetized mice. Very interestingly, we observed a strong luminescent signal already in freshly weaned 3-wk-old mice that were otherwise asymptomatic (Supplemental Fig. 3A). The signal was mostly localized along the tail and with weaker intensity in the hind paws. Visualization at later time points revealed that at 4 wk of age, the ROS production was equally intensive in the tail and paws (Fig. 5A, 5B) and gradually moved to the hind paws during the weeks 6–8. At this age, ROS production became predominant in hind paws with more restricted focal localization (Fig. 5B, Supplemental Fig. 3A).

To test if neutrophils are the source of dysregulated ROS observed in vivo, we have generated Pstpip2^cmo mouse strain where the majority of neutrophils were deleted via MRP8-CRE-dependent expression of Diphtheria toxin (Pstpip2^cmo-DTA-MRP8-CRE, Supplemental Fig. 3B, 3C). In vivo ROS imaging revealed that ROS production in the tails and hind paws of these mice was almost completely abolished (Fig. 5C). In addition, these mice also did not show any symptoms of autoinflammatory disease, whether determined by visual inspection (Fig. 5D) or by x-ray μCT analysis (Fig. 5E). These data strongly support the idea that increased ROS production preceding the onset of the disease originates in neutrophils and, at the same time, confirm that neutrophils are critical for the development of disease symptoms (36).

Strong unprovoked production of ROS in very young mice preceding visible symptoms weeks before their demonstration suggested that ROS may act upstream of IL-1β in osteomyelitis development. To determine the contribution of high in vivo ROS generation to disease development, we crossed Pstpip2^cmo mice to gp91phox-deficient mouse strain. In the absence of gp91phox, we were unable to detect any superoxide production upon silica, E. coli, or aggregated IgG stimulation even in Pstpip2^cmo cells (Fig. 6A, data not shown). These data confirm that NADPH oxidase was responsible for the dysregulated ROS production in Pstpip2^cmo neutrophils. Surprisingly, Pstpip2^cmo mice lacking gp91phox developed similar disease symptoms as Pstpip2^cmo mice (Fig. 6B) and with similar, only slightly delayed, kinetics (Fig. 6C). Blind scoring of the disease severity by visual inspection of hind paw photographs collected throughout various experiments revealed that the symptoms of the disease are only partially alleviated in gp91phox-deficient animals, by approximately one to two points on 8-point scale (Fig. 6D). Moreover, ELISA analysis detected comparable amount of IL-1β in hind paw extracts from Pstpip2^cmo and Pstpip2^cmo/gp91phox^−/− animals (Fig. 6E), and similar amount of processed IL-1β p17 was found in the lysates of silica-stimulated BM cells by immunoblot (Fig. 6F).

These data demonstrated that the phagocyte NADPH oxidase is dispensable for autoinflammatory disease initiation, but it affects the severity of the disease. We also noticed that the character of the hind paw edema was somewhat different in Pstpip2^cmo/gp91phox^−/− mice. Typically, the swelling was most serious at the distal part of phalanges and only rarely affected metatarsal area in Pstpip2^cmo/gp91phox^−/− animals, whereas in Pstpip2^cmo mice, metatarsi were frequently enlarged and the phalanges were often most seriously affected in their central parts (Fig. 6B).

To find out if these differences were caused by different character of bone inflammation, we performed x-ray μCT analysis of Pstpip2^cmo and Pstpip2^cmo/gp91phox^−/− mice. Very surprisingly, bone destruction in Pstpip2^cmo/gp91phox^−/− animals was almost entirely missing, whereas in Pstpip2^cmo mice substantial bone damage could be observed (Fig. 7A). To support this observation with a quantitative analysis, we calculated bone surface to volume ratio and bone fragmentation from the x-ray μCT data. Pstpip2^cmo/gp91phox^−/− mice showed similar values to WT, whereas values for Pstpip2^cmo mice were substantially higher (Fig. 7B). Timeline x-ray μCT scans of hind paws revealed progressive bone lesion formation in Pstpip2^cmo mice whereas Pstpip2^cmo gp91phox^−/− littermates remained largely protected (Supplemental Fig. 4).

The lack of bone damage can also be demonstrated on tissue sections from tarsal area of hind paws (Fig. 7C). The CMO mice show very high level of osteolysis of tarsal bones, with almost missing joint cartilages due to arthritic changes accompanied with robust granulomatous infiltration. The WT mice have normally developed and structurally well-defined tarsal bones with undamaged joint cartilages with no infiltration of immune cells and no adverse changes in the BM. The Pstpip2^cmo/gp91phox^−/− mice show a rescue effect in ossified parts of tarsal bones with no or minimal signs of bone damage by immune cells, and the soft tissue infiltration in the metatarsal area is minimal compared with CMO mice. The cartilages are also well shaped and are covering the joint areas comparably with WT animals. The difference from WT animals is in hypercellular structure of BM resulting in decreased volume of ossified tissue.

To address protection potential of gp91phox deficiency in old Pstpip2^cmo mice, we performed x-ray μCT scans on 7-mo-old mice. Old Pstpip2^cmo mice suffered from strong bone destruction and remodeling but Pstpip2^cmo/gp91phox^−/− mice were still protected from adverse effects of osteomyelitis (Supplemental Fig. 4). To gain a quantitative insight into the level of soft tissue inflammation, we have performed computational reconstruction of soft tissues from x-ray μCT scans described above in Fig. 7B and calculated soft tissue volume. This measurement revealed that soft tissues in Pstpip2^cmo/gp91phox^−/− hind paws were significantly enlarged, albeit not to the same extent as in Pstpip2^cmo mice (Fig. 7D). Collectively, these data demonstrate that despite significant swelling that can be detected in the hind paw soft tissues of Pstpip2^cmo/gp91phox^−/− mice, bones remain largely protected in the absence of NADPH oxidase activity.

Monogenic autoinflammatory diseases develop as a result of dysregulation of the innate immune system. Although the specificity of this branch of the immune system is relatively limited, these diseases still show tissue and organ selectivity. The mechanisms of this selectivity are often poorly understood (3). Pstpip2^cmo mice represent an important model of tissue-selective IL-1β–driven autoinflammatory disease that affects mainly bones and surrounding tissue in hind paws and tails (29–31). Our current studies demonstrate that IL-1β pathway is not the only pathway dysregulated in these animals. Superoxide production by neutrophil NADPH oxidase is also substantially enhanced, independently of IL-1β activity. Moreover, our data suggest that dysregulated ROS production is a critical part of the selectivity mechanisms directing inflammatory damage to the bones.

Increased superoxide generation by neutrophil NADPH oxidase is one of the major consequences of neutrophil priming. It is a state of enhanced responsiveness attained after an exposure to priming agents such as LPS or TNF-α. These agents typically do not elicit superoxide production by themselves but rather increase its generation triggered by other substances (53, 54). To find out if Pstpip2^cmo neutrophils are spontaneously constitutively primed, we have tested several additional parameters known to be associated with neutrophil priming, including increased expression of integrin subunit CD11b, changes in cytoskeleton organization, and IL-1β promoter activity (52–58). However, none of these traits were altered in Pstpip2^cmo neutrophils.

Neutrophil adhesion is also often considered a priming stimulus, because it results in similar phenotypic changes and is able to elicit enhanced superoxide production, even in response to solely priming agents such as TNF-α (54, 61). Accordingly, whereas we observed TNF-α– and LPS-induced superoxide production by adherent neutrophils, cells in suspension did not show any response, regardless of the genotype. The reaction to silica and fMLP was also substantially attenuated in nonadherent cells. However, the deregulation of responses in Pstpip2^cmo neutrophils could still be detected under these conditions. These data support the conclusion that dysregulation of NADPH oxidase activity in Pstpip2^cmo neutrophils is independent of adhesion. They also show that Pstpip2^cmo neutrophils are not fully primed and can still undergo priming.

Collectively, our data do not support the idea of spontaneous priming of Pstpip2^cmo neutrophils. They rather suggest dysregulation of pathways controlling NADPH oxidase activity. However, it should be noted that different priming agents elicit varying sets of phenotypic changes in neutrophils (53), and some specific form of constitutive priming cannot be completely disregarded.

We found that increased ROS production in Pstpip2^cmo mice was associated with p47phox hyperphosphorylation, leading to the conclusion that one or more pathways governing p47phox phosphorylation are negatively regulated by PSTPIP2. MAP kinase ERK was shown to have a supporting role in p47phox activation by phosphorylating Ser 345 and 348 (62, 63). We have shown previously that ERK is also hyperactive in silica-treated CMO neutrophils (33). However, in this study we found that ERK pathway inhibition did not have any substantial effect on p47phox phosphorylation, whereas this phosphorylation could be almost completely abolished by inhibition of PKC (Fig. 4E). These data suggest that PSTPIP2 deficiency mainly affects the phosphorylation of p47phox that is PKC mediated. However, the precise mechanism is still unclear.

To our knowledge, genetically determined hyperactivation of NADPH oxidase has not yet been described or studied in the context of IL-1β activation or in autoinflammatory disease. Our observations show that elevated NADPH oxidase activity does not affect IL-1β pathway but rather the inflammatory bone damage in CMO. These data are in agreement with several other reports that disprove the role of NADPH oxidase–generated ROS in IL-1β processing by inflammasome. They are mainly based on analyses of monocytes and macrophages from NADPH oxidase–deficient patients and mice, where IL-1β production is not altered or it is even enhanced (64–67). In a single study on human neutrophils, NADPH oxidase deficiency also did not lead to reduction of NLRP3 inflammasome activity (68). In contrast, inhibition of mitochondrial ROS production in monocytes/macrophages results in an impairment of IL-1β production in these cells (19, 69), showing that the majority of ROS supporting inflammasome activation in these cell types is generated by mitochondria.

The roles of IL-1β and ROS in CMO pathophysiology appear to be different form each other. Whereas dysregulated IL-1β production is a critical trigger of the disease development, enhanced ROS production modifies the outcome. However, ROS are not able to initiate the disease on their own in the absence of IL-1β signaling. It is documented by our experiments with MyD88-deficient Pstpip2^cmo mice, which displayed the same ROS dysregulation as Pstpip2^cmo mice and yet they did not develop any symptoms of autoinflammation. These data also demonstrate that enhanced ROS production is not downstream of IL-1β because MyD88 is critical for signaling by IL-1R (70).

ROS are known to play a key role in differentiation and activity of osteoclasts. These cells are responsible for physiological bone resorption during bone remodeling processes. They are also involved in pathological bone damage in a number of disease states (71). Pstpip2^cmo mice exhibited increased osteoclastogenesis and osteoclast hyperactivity, suggesting that osteoclasts are responsible for inflammatory bone damage in these mice (39). Our data show that the bone damage can be almost completely abolished when phagocyte NADPH oxidase is inactivated by deletion of its gp91phox subunit. One possibility is that deficiency in osteoclast gp91phox results in defects in their differentiation and activity and reduced bone damage. However, in gp91phox^−/− mice, no bone abnormalities have been observed. In addition, gp91phox-deficient osteoclasts differentiate normally and have normal bone resorption activity (72–74). These results show that gp91phox expressed in osteoclasts is dispensable for differentiation and activity of these cells. In fact, other NADPH oxidases were shown to be more important for their function (72, 75). In contrast, exogenous ROS generated in culture media after addition of xanthine oxidase were shown to upregulate osteoclast numbers and activity in bone cultures in vitro (76). Our data together with published results thus favor the explanation that exogenous ROS originating from hyperactive neutrophils, ample production of which we observed in Pstpip2^cmo mice in vivo, lead either directly or indirectly to increased differentiation and/or activity of osteoclasts and resulting bone damage.

PSTPIP2 mutations in humans have not yet been described. However, PSTPIP2 gene has been sequenced only in a limited number of CRMO patients and patients with closely related SAPHO syndrome (77–79). CRMO and SAPHO form a rather heterogeneous disease spectrum, which may in fact represent a number of distinct disorders in which various defects at the molecular level may lead to similar outcome, and so PSTPIP2 mutations in some of these patients may still be discovered in the future. The data on ROS production in these diseases are also largely missing. We are aware of only a single study in which the superoxide production by neutrophils was analyzed in two SAPHO patients from a single family without any mutations in PSTPIP2 gene. These data showed reduced superoxide generation after activation with multiple activators, including PMA, fMLP, and TNF-α, when compared with healthy controls (77). However, from the information provided, it was unclear whether the patients were undergoing anti-inflammatory treatment that could suppress the response at the time of analysis. Further studies are needed to fully understand the role of PSTPIP2 and ROS in CRMO, SAPHO, and other inflammatory bone diseases in humans.

Inflammatory bone damage is a serious problem accompanying a number of human disorders. Full understanding of possible mechanisms that can govern its development is critical for designing successful therapeutic interventions. Our data reveal how dysregulated ROS production results in bone damage in the specific case of CMO. However, these findings may represent a more general mechanism with broader validity for other syndromes where inflammatory bone damage is involved, and analysis of ROS production in other instances of inflammatory bone damage may prove beneficial.

We thank the staff of Institute of Molecular Genetics core facilities for excellent support and help.

The authors have no financial conflicts of interest.