Mimicking Behçet’s disease: GM‐CSF gain of function mutation in a family suffering from a Behçet’s disease‐like disorder marked by extreme pathergy

Summary

Behçet’s disease (BD) is an inflammatory disease mainly affecting men along the ancient Silk Route. In the present study we describe a Dutch family suffering from BD‐like disease with extreme pathergic responses, but without systemic inflammation. Genetic assessment revealed a combination of the human leukocyte antigen (HLA)‐B*51 risk‐allele together with a rare heterozygous variant in the CSF2 gene (c.130A>C, p.N44H) encoding for granulocyte–macrophage colony‐stimulating factor (GM‐CSF) found by whole exome sequencing. We utilized an over‐expression vector system in a human hepatocyte cell line to produce the aberrant variant of GM‐CSF. Biological activity of the protein was measured by signal transducer and activator of transcription 5 (STAT‐5) phosphorylation, a downstream molecule of the GM‐CSF receptor, in wild‐type peripheral mononuclear cells (PBMCs) using flow cytometry. Increased STAT‐5 phosphorylation was observed in response to mutated GM‐CSF when compared to the wild‐type or recombinant protein. CSF2 p.N44H results in disruption of one of the protein’s two N‐glycosylation sites. Enzymatically deglycosylated wild‐type GM‐CSF also enhanced STAT‐5 phosphorylation. The patient responded well to anti‐tumor necrosis factor (TNF)‐α treatment, which may be linked to the capacity of TNF‐α to induce GM‐CSF in phorbol 12‐myristate 13‐acetate (PMA)‐treated PBMCs, while GM‐CSF itself only induced dose‐dependent interleukin (IL)‐1Ra production. The identified CSF2 pathway could provide novel insights into the pathergic response of BD‐like disease and offer new opportunities for personalized treatment.

Loss of N44 glycosylation in GM‐CSF leads to increased signaling activity resulting in a BD‐like phenotype marked by extreme pathergy. The GM‐CSF pathway may provide a novel target to reshape individual patient care in the pathogenesis of pathergy in BD.

Introduction

Since the Chapel Hill Consensus Conference in 2012, Behçet’s disease (BD) has been classified as systemic variable vessel vasculitis that can effect vessels of any size and type [1]. Therefore, clinical presentation is diverse and comprises unspecific symptoms. No universal diagnostic test is available to definitely confirm BD; therefore, patient identification relies upon clinical symptoms including: oral ulceration, aphthous or scarring genital ulcers, skin lesions (pseudofolliculitis, acneinform nodules or erythema nodosum), uveitis (anterior, posterior or retinal vasculitis), vascular complications (arterial/venous aneurysms, arterial/venous thrombosis) and a positive pathergy test [2]. Since the description of the first patients in 1937 by Hulusi Behçet, researchers were able to find various treatment options to reduce the inflammation and relieve the symptoms, but the true disease ontology of BD still remains an enigma, and only some disease risk factors were identified. The most well‐known genetic risk factor is carriage of the human leukocyte antigen (HLA)‐B*51 allele. Increased frequency of this allele was observed in populations located along the ancient Mediterranean Silk Route, where the highest prevalence of BD has also been reported [3]. More recently, other immune factors such as cytokines and cytokine receptors have been linked to BD pathogenesis [4], but how precisely they contribute to the development of BD has not yet been established. While genome‐wide association studies (GWAS) have hinted towards common genetic variants that are linked to disease development, genetic testing of rare variants identified monogenic mutations causing autoinflammatory diseases or primary immunodeficiencies that are able to mimic BD and result in atypical phenotypes, further complicating the search of the true disease origin [5]. In this study, we describe a novel genetic variant in CSF2 in a family with BD‐like symptoms that is associated with a BD‐like clinical phenotype characterized by extreme pathergy and no systemic inflammation. We describe an asparagine to histidine substitution at protein position 44 in CSF2, which leads to the loss of N‐glycosylation and thereby increased bioactivity of the mutated protein granulocyte–macrophage colony‐stimulating factor (GM‐CSF). This report suggests that local tumor necrosis factor (TNF)‐α production in the skin can drive GM‐CSF production in BD‐like disease [6]. Furthermore, GM‐CSF in turn could, in addition to inducing inflammation, contribute to significant production of the anti‐inflammatory cytokine interleukin (IL)‐1Ra. High levels of IL‐1Ra induced by hyperactive GM‐CSF might explain the undetectable C‐reactive protein (CRP) levels, even after explorative surgery in our patient. Increased GM‐CSF activity, therefore, might exacerbate local inflammatory responses without systemic responses and, in our patient combined with HLA*B51 carriage, lead to a BD‐like phenotype with extreme pathergy.

Materials and methods

Index patient

Our patient’s clinical history starts at the age of 15 years (1988), when she underwent orthopedic surgery to correct a congenital leg length discrepancy. Her medical history at that time was unremarkable, besides a tonsillectomy and appendectomy. After surgery, wound healing was impaired and multiple additional surgical procedures were performed to debride the presumably infected tissue, eventually resulting in amputation of the lower right leg after 6 years in 1994. In 1996, she first presented to our outpatient clinic because of infection of a minor post‐traumatic skin lesion on her left cheek that persisted despite antibiotic treatment. At that point, based on normal peripheral blood counts and differentiation and her undistinguished medical history, an immune defect was deemed unlikely as a cause of patient’s infectious complications. In the following years, until 2012, she was frequently seen at the outpatient clinic or admitted because of several inflammatory and infectious conditions including, but not limited to, erythema annulare centrifugum thought to be triggered by tinea pedis (1997), sepsis (2005), erysipelas of the amputation stump (2009), fever and swollen tonsils followed by tonsillectomy (2010), pneumonia (2011) and skin and soft tissue infection of the left leg (2011/12). Therefore, in 2012 more detailed laboratory tests regarding immune cell function were carried out and revealed a low interferon (IFN)‐γ in response to in‐vitro stimulations. Based on these results, treatment with recombinant IFN‐γ was initiated shortly thereafter and proved beneficial. However, in 2013, our patient presented with a subcutaneous abscess (caused by a group C Streptococcus) of the right wrist, which was successfully treated with antibiotics and surgical drainage.

In 2015, she was admitted to our hospital with another skin and soft tissue infection of the left hand, initially diagnosed as necrotizing fasciitis. However, repeated surgical exploration consistently revealed vital tissue, although group B hemolytic streptococci and Gram‐positive cocci were cultured. Because of the extreme inflammatory reaction, prednisone and anakinra were eventually added to the broad‐spectrum antibiotic treatment. However, these had limited effect and the inflammatory lesion spread to her thorax and back, necessitating further surgical exploration and drainage, after which symptoms ameliorated. Later that year, the extreme pathergy and the constellation of other symptoms suggested Behçet’s disease. Therefore, anti‐TNF‐α treatment was given and the patient improved. Meanwhile, genetic assessment revealed a carrier‐state for the Behçet’s risk allele HLA‐B*51. In addition, whole exome sequencing revealed a variant in CSF2, which encodes for the proinflammatory cytokine GM‐CSF. In 2016/2017, the patient presented multiple times with flares of her Behçet’s, which led to additional anti‐inflammatory treatment with thalidomide, azathioprine and colchicine. Since then, the patient gradually improved and had less frequent flares. Notably, our patient demonstrated CRP values < 10 mg/l during her inflammatory episodes and her triple surgery. Hand‐written consent of the patient was obtained prior to material collection and conducting experiments.

Peripheral blood mononuclear cell (PBMC) isolation

Healthy volunteers donated blood for PBMC isolation. Blood was drawn into 10‐ml tubes [ethylenediamine tetraacetic acid (EDTA)] and cells were isolated using Ficoll gradient centrifugation, according to our established protocol. In short, PBMCs accumulated on top of the Ficoll were collected and washed three times with cold phosphate‐buffered saline (PBS). Then, cells were resuspended in RPMI‐1640 (gibco‐Invitrogen, Paisley, UK) supplemented with glutamax (10 mmol/l; Invitrogen, Carlsbad, CA, USA), pyruvate (10 mmol/; Invitrogen) and gentamycin (10 μg/ml; Centraform, Billerica, MA, USA). For fluorescence activated cell sorter (FACS) analysis, 0·5 × 10⁶ cells were stimulated per well in 96‐well round‐bottomed plates, without serum.

Tissue culture

Wild‐type and mutant human GM‐CSF expression plasmids

The p.N44H variant was introduced into the mutant sequence by replacing the adenine into a cytosine nucleotide (c.130A>T) (see Supporting information). Wild‐type and mutant GM‐CSF gene sequences were chemically synthesized (BaseClear, Leiden, the Netherlands) and cloned by the manufacture in a pUC‐57‐Amp plasmid. HindIII and XbaI restriction sites were flanking the start and stop position of both wild‐type and mutant GM‐CSF, respectively. Subsequently, wild‐type and mutant GM‐CSF were excised from the pUC‐57‐Amp vector by double digestion with HindIII and XbaI (New England Biolabs, Ipswich, MA, USA) restriction endonucleases and ligated in a HindIII/XbaI linearized pCDNA3‐Amp expression vector. Finally, the expression vectors were sequenced (BaseClear, Leiden, the Netherlands) and contained the correct sequences, either wild‐type or mutant GM‐CSF.

Over‐expression of wild‐type and mutant human GM‐CSF

Human HepG2 cells were maintained in Dulbecco’s modified Eagle’s medium (DMEM)–Glutamax medium (gibco‐Invitrogen, Paisley, UK) complemented with 10% fetal calf serum (FCS), pyruvate (gibco‐Invitrogen) and gentamicin (Gibco‐Invitrogen). Before transfection, cells were trypsinized and seeded in 24‐well plates (Corning Incorporated, Corning, NY, USA) in medium without antibiotics. The following day, DNA‐liposome complexes were prepared using Fugene HD (Promega, Madison, WI, USA), according to the manufacturer’s protocol. Subsequently, medium was replaced with fresh medium without antibiotics and DNA‐liposome complexes were added to the cells. Twenty‐four h after transfection, RNA and supernatants were isolated. As well as GM‐CSF wild‐type and mutant over‐expression, eGFP was over‐expressed as control.

RNA isolation, cDNA synthesis and GM‐CSF gene expression

To verify over‐expression of GM‐CSF, RNA was isolated with the Trizol (Sigma‐Aldrich, St Louis, MO, USA) method and transcribed into cDNA (iScript kit; BioRad, Hercules, CA, USA). Relative GM‐CSF expression was determined by qPCR (StepOnePlus; Applied Biosystems, Foster City, CA, USA) using GM‐CSF‐specific primers (forward: 5‐CAGCCACTACAAGCAGCACT‐3; reverse: 5‐GGGGATGACAAGCAGAAAGT‐3) and normalized against GAPDH (forward: 5‐AGGGGAGATTCAGTGTGGTG‐3; reverse: 5‐ CGACCACTTTGTCAAGCTCA‐3).

Enzyme‐linked immunosorbent assay (ELISA)

Supernatants of HepG2 cells carrying either the wild‐type, mutant or empty vector plasmid were collected after 24 h. Levels of GM‐CSF were determined using the R&D ELISA (R&D Systems, Minneapolis, MN, USA) kit following the instructions provided in the manufacturers’ protocol. Concentrations measured in cell culture supernatants were used to adjust GM‐CSF to the desired concentrations for functional assays.

Signal transducer and activator of transcription 5 (STAT‐5) FACS analysis

Freshly isolated 0·5 × 10⁶ PBMCs were stimulated with 10 pg/ml of either recombinant glycosylated GM‐CSF (R&D Systems) or supernatant obtained from HepG2 cells carrying the vector containing wither the wild‐type, mutant or empty vector plasmid. The volume of supernatant added was adjusted according to GM‐CSF concentration measured by ELISA. In short, cells were stimulated for 15 min, washed with PBS containing 1% bovine serum albumin (BSA). After fixation and permeabilized using BD Fix&Perm solution, according to the protocol provided by the manufacturer, cells were washed two more times in permeabilization buffer (BD) before 1 ml of 100% methanol was added per sample and stored at −20°C for at least 12 h. To remove all methanol cells were washed two times followed by staining with phosphorylated (p)STAT‐5‐phycoerythrin (PE) (eBioscience, San Jose, CA, USA).

Pre‐treatment PNGaseF

Supernatants collected form HepG2 cells were treated for 24–36 h with glycopeptidase F (PNGaseF), according to protocol provided with the reagents. In summary, the volume of the supernatant was based on ELISA results to equal the GM‐CSF concentration of 10 pg/ml; ×10 reaction buffer was added at a final concentration of ×1 together with 1 μl PNGaseF or water. Mixtures were incubated at 37°C. After incubation, PBMCs were freshly isolated from healthy donors and the pSTAT‐5 staining protocol was followed exactly as stated above.

Genetic analysis by whole‐exome sequencing

Genomic DNA from the index patient was isolated from whole blood using standard procedures. The experimental work‐flow of exome sequencing was performed at BGI Europe (Beijing Genome Institute Europe, Copenhagen, Denmark) as described previously [7]. Exonic regions were enriched with the SureSelect version 4 (50 Mb) exome kit (Agilent Technologies, Santa Clara, CA, USA) and sequenced on Illumina Hiseq sequencer (Illumina, San Diego, CA, USA) with 101 base pairs (bp) paired‐end reads. Sequence reads were mapped to the reference human genome (hg19) using the BWA mapping algorithm (version 0.7.12) and variants were called by the GATK unified genotyper (version 3.4). Variants were annotated using an in‐house pipeline for exome analysis [8] containing gene and variant specific information (including variant population frequencies from > 24 000 in‐house exomes, dbSNP144, ExAC and GnomAD). Variant filtering was applied as previously reported [8]. Briefly, only the variants affecting coding exons and canonical splice sites were selected. Subsequently, synonymous variants were filtered out, and only rare variants (frequency < 0·1% in all the used databases) with high quality remained (Supporting information, Table S1). All the genes with rare non‐synonymous variants were systematically checked for involvement in immunological pathways or phenotypes (Gene Ontology term, Online Mendelian Inheritance in Man, mouse knock‐out phenotype, Kyoto Encyclopedia of Genes and Genomes). Furthermore, variants with immune‐related functions were manually inspected with an integrative genomics viewer to exclude any remaining false‐positive variants. Nucleotide conservation score (vertebrate PhyloP) and potential functional effects of the variants were further evaluated by in‐silico prediction tools (PolyPhen, SIFT, MutationTaster, CADD) (Supporting information, Tables S1 and S2).

Co‐segregation analysis by Sanger sequencing

DNA from all the available family members was isolated from whole blood using standard procedures. The CSF2 variant was validated by Sanger sequencing in the index patient and subsequently tested in family members.

Statistics

ELISA results were analyzed using Prism GraphPad software version 5.0. Significance was accepted when *P < 0·05, **P < 0·01 and ***P < 0·001.

Results

Identification of a CSF2 substitution variant

By the time our patient presented to our hospital she had been treated for many years, due to atypical skin complications and seemingly unexplained localized inflammatory outbursts in response to injury. Her clinical manifestations are characteristic for BD with aphthous oral lesions and genital ulcera, while her skin condition appeared to be extremely unusual, as she displayed extreme pathergic responses (Fig. 1a). The patient was identified to be a carrier of the HLA‐B*51 risk allele during routine diagnostic work‐up. Based on the remarkable intensity of her skin involvement, genetic analysis was included to further look for rare genetic variants that could explain her phenotype. Whole‐exome sequencing of DNA obtained from the index patient provided at least 20‐fold coverage for 98·05% of the target and resulted in 217 rare non‐synonymous and canonical splice site variants (Supporting information, Table S1). Subsequently we selected the candidate gene variants by filtering for gene function, in‐silico predictions and population frequency. This analysis, focusing on immune‐related genes, resulted in nine genetic variants that were either conserved and/or were predicted to have a deleterious effect to the protein function (Supporting information, Table S2). From this list, an extremely rare heterozygous variant in CSF2 (c.130A>C, p.N44H) was selected as the best candidate that could explain the patient’s phenotype, because GM‐CSF has been indicated as a major cause of severe tissue inflammation [9]. This variant was not detected in any other individual in our in‐house database (> 24 000 exomes) and was observed only in three individuals in the GnomAD database (> 140 000 individuals).

Fig. 1.

Index patient description. (a) Documentation of an extreme pathergic response with days from admission after a small wound on her left cheek showed no evidence for infection. Arrows indicate skin lesions. (b) Family pedigree and clinical picture of the index patient. (c) Top: ribbon structure overview of wild‐type and mutant granulocyte–macrophage colony‐stimulating factor (GM‐CSF) (http://www.cmbi.ru.nl/hope). The protein is shown in gray, the side chain of the mutated residue is colored magenta and shown as small balls. Bottom: close‐up of the mutated residue. The wild‐type is shown in green, the mutant in red, the rest of the protein is shown in gray.

DNA testing of both parents revealed the presence of the HLA‐B*51 risk allele in her father’s DNA, whereas the N44H variant in CSF2 was detected on the mother’s side (Fig. 1b). Her father was known to have aphthous lesions orally but did not have BD. Her mother only reported to have exaggerated responses in the skin when bitten by insects, plaques that could be as large as 7 cm, but the rest of her history was unremarkable. While the medical status of her older brother is currently unknown, the patient’s twin brother did not exhibit any inflammatory complications or clinical signs matching BD. In line with these observations, genetic analysis revealed the absence of both the HLA‐B*51 risk allele and CSF2 variant in the twin brother’s DNA (Fig. 1b). The patient has three children: two daughters and one son. Her youngest daughter was known with oral and genital ulcera at the age of 9 years and frequently had unexplained episodes of abdominal pain lasting a week. Her son had the same abnormal skin reaction to mosquito bites as his grandmother. The daughter carried both the HLA‐B*51 risk allele and CSF2 variant, while her son only had the CSF2 variant. Online prediction models of protein structure and function indicate that the substituted residue is located at the surface of the protein. Moreover, the mutated amino acid N44 is one of the protein’s N‐glycosylation sites, which is thereby disrupted (Fig. 1c).

Patient GM‐CSF demonstrates strongly enhanced signaling capacity due to loss of glycosylation

The post‐translational processing of GM‐CSF by means of glycosylation has previously been shown to be crucial for protein–receptor interactions [10]. Therefore, we aimed to develop an assay to assess signaling capacities of wild‐type and mutant GM‐CSF. When GM‐CSF binds its receptor, Janus kinase 2 (JAK2) will phosphorylate the intracellular domain of the receptor, creating a docking site for STAT‐5 which, in turn, will become phosphorylated. We used freshly isolated PBMCs from healthy volunteers to measure the amount of STAT‐5 phosphorylation upon exposure to commercially available GM‐CSF‐ or HepG2‐expressed wild‐type and mutant GM‐CSF proteins. At high cytokine concentrations (100 pg/ml), all three forms of GM‐CSF were clearly able to induce STAT‐5 phosphorylation to a similar extent, suggesting that loss of glycosylation at N44 does not result in a loss of protein function (Fig. 2a). However, at lower concentrations (10 pg/ml), recombinant GM‐CSF‐ and HepG2‐expressed wild‐type GM‐CSF did not induce measurable STAT‐5 phosphorylation, whereas mutant GM‐CSF induced STAT‐5 phosphorylation (Fig. 2b), suggesting an increase of function. In order to investigate whether the observed effect was indeed caused by loss of glycosylation, we treated wild‐type and mutant GM‐CSF with PNGaseF (Fig. 1c), an enzyme specifically removing N‐glycans. While enzyme treatment did not impact STAT‐5 phosphorylation of the mutated GM‐CSF, we measured an increase in intracellular signaling by PNGaseF treated wild‐type GM‐CSF (Fig. 1d). These results support the hypothesis that loss of N‐glycosylation causes increased signaling capacity of GM‐CSF.

Fig. 2.

Signal transducer and activator of transcription 5 (STAT‐5) phosphorylation of wild‐type and mutant granulocyte–macrophage colony‐stimulating factor (GM‐CSF). (a,b) Freshly isolated peripheral blood mononuclear cells (PBMCs) of healthy volunteers stimulated with wild‐type or mutant GM‐CSF at high (100 pg/ml) or low (10 pg/ml) concentrations. Intracellular phosphorylated (p)STAT‐5 was measured by flow cytometry (one representative donor shown). At high concentrations, no differences in pSTAT‐5 induction could be observed between wild‐type and mutant GM‐CSF. At low concentrations only the mutant GM‐CSF is able to induce phosphorylated STAT‐5. (c) Schematic overview of predicted O‐glycosylation (green) and N‐glycosylation (blue) sites of GM‐CSF. At position 44 in the mutant protein, the asparagine residue has been substituted by histidine, resulting in loss of the glycosylation site. (d) Mutant and wild‐type GM‐CSF have been treated with glycopeptidase F (PNGaseF). The amount of pSTAT‐5 phosphorylation in PBMCs was measured by flow cytometry (one representative donor shown).

GM‐CSF modulates inflammatory responses

In general, BD is recognized as an autoinflammatory illness caused by hyper‐reactivity of mainly the innate immune system (elevated IL‐1β, TNF‐α, enhanced neutrophil activation) [11]. The extreme pathergic reactions of the patient responded well to the treatment with TNF‐α inhibitor infliximab. Therefore, we wanted to explore whether TNF‐α could directly induce GM‐CSF in human PBMCs. Here, we show that TNF‐α is able to modulate and increase GM‐CSF expression and release by prestimulated PBMCs (Fig. 3a). Moreover, we wanted to understand why the patient did not have detectable CRP levels despite several invasive procedures and clinical inflamed tissue sites. CRP is produced in the liver and is induced by IL‐6 which, in turn, is stimulated by IL‐1β. To assess the effect of GM‐CSF on the expression of anti‐inflammatory mediators, we stimulated freshly isolated PBMCs for 24 h with GM‐CSF and measured IL‐1Ra release by ELISA. GM‐CSF is able to induce IL‐1Ra production in a dose‐dependent manner (Fig. 3b), which might explain the undetectable CRP levels in our patient with hyperactivated GM‐CSF.

Fig. 3.

Granulocyte–macrophage colony‐stimulating factor (GM‐CSF) modulates the inflammatory network. (a) Peripheral blood mononuclear cells (PBMCs) of healthy volunteers were stimulated with tumor necrosis factor (TNF)‐α in combination with phorbol 12‐myristate 13‐acetate (PMA). Results show that increasing amounts of TNF‐α have a synergistic effect on PMA‐induced GM‐CSF induction (two independent experiments, combined n = 8). (b) The effect of GM‐CSF on interleukin (IL)‐1Ra production of healthy PBMCs has been tested. Increasing amounts of GM‐CSF induce IL‐1Ra release (n = 4).

Discussion

In this study we identify a novel, very rare, missense variant in CSF2 in a single family which results in hyperactive GM‐CSF and that might contribute to the development of a BD‐like clinical phenotype characterized by extreme pathergic reactions. Functional testing revealed increased signaling activity of the mutant protein, which is likely to be caused by structural changes due to loss of N‐glycosylation at the mutated residue. We provide evidence that the newly identified N44H variant may contribute as a disease modifier in the background of HLA‐B51 and can mimic several aspects of BD, and therefore may possibly serve as a potential screening candidate to assist in BD‐like diagnosis, especially when the disease is associated with striking pathergic reactions and the patient is HLA‐B51‐positive. Additionally, GM‐CSF signaling could present a novel therapeutic target in BD‐like disease, especially for skin pathology, although more data are needed.

GM‐CSF is a heavily glycosylated cytokine encoded by the CSF2 gene, which has pleiotropic functions during homeostasis and inflammation [12]. Low circulating GM‐CSF levels are commonly found, which will increase in response to cytokines such as IL‐1β and TNF‐α signals present upon tissue injury or during inflammation. Locally produced GM‐CSF promotes tissue infiltration of granulocytes and macrophages at the side of infection and enhances their survival. In line with these findings, blockage of GM‐CSF in animal studies has been proven beneficial and alleviates signs of autoimmune/inflammatory diseases [13, 14]. Increased levels of GM‐CSF have been shown to promote severe inflammation [15], which is in line with our observations of the hyperactive function of GM‐CSF in a patient presenting with uncontrolled local skin inflammation. GM‐CSF with loss of N‐glycosylation has been shown to increase its biological activity probably by increasing the binding affinity to its receptor [10, 16, 17]. We demonstrate that by enzymatic removal of glycoslyation of the wild‐type GM‐CSF protein there was increased STAT‐5 phosphorylation. Moreover, the enzyme‐treated wild‐type GM‐CSF behaves very similarly to the patients GM‐CSF, while the mutated GM‐CSF was not affected by treatment, further strengthening our hypothesis that the loss of the N44 glycosylation site is the reason for the aberrantly increased protein function in our patient. Our results are in line with Moonen et al., who demonstrated the importance of post‐translational N‐glycosylation with regard to GM‐CSF function: loss of N‐glycosylation facilitates ligand–receptor interaction [10]. This might also have implications for the clinical use of recombinant GM‐CSF that is produced by pharmaceutical companies. A more potent GM‐CSF that can be produced by adding just one AA change could be a cost‐effective effort and patients would have to have lower dosages to have the same effect, which has also been suggested by another study [18].

The CSF2 variant was found in the mother of the patient, who was not diagnosed with BD and did not carry the HLA‐B*51 allele. The father carried the HLA‐B*51 allele, which is a risk factor for BD, but besides having aphthous lesions in his mouth he was never diagnosed with BD or BD‐like disease and he did not carry the CSF2 variant. These observations, together with notice that only the index patient and her daughter that carry both the HLA‐B*51 and the CSF2 variant were diagnosed with BD‐like disease and none of the other family members, suggests that CSF2 might be a disease modifier responsible for the development of symptoms of BD‐like disease in the background of HLA‐B51. Other studies identified single nucleotide polymorphisms (SNPs) in IL‐10, IL‐23R and IL‐12RB2 genes that were associated with BD [4]. BD pathology not only resembles autoimmune diseases, but also presents with strong autoinflammatory features. Therefore, genes driving innate immunity, especially within the IL‐1 family, have been screened for BD associated genetic variants. SNPs in IL‐37 and IL‐18RAP have been found to correlate with sterile uveitis, a clinical sign frequently observed in BD [19]. Many genetic association studies have been performed, and increasing numbers of genes contributing to BD development are described; however, no specific signaling pathway driving BD pathology has been identified.

Recent publications pinpoint nuclear factor kappa B (NF‐κB)‐mediated signaling as the major driver of BD‐like pathology. On one hand, several loss of function mutations in TNFAIP3 (A20) were identified in patients suffering from early‐onset autoinflammatory syndrome with characteristics of BD [20, 21]. In healthy cells, A20 mediates proteasomal degradation of the active TNF‐α receptor unit, thereby dampening NF‐κB signaling. Additionally, A20 regulates inflammasome [NLR family pyrin domain containing 3 (NLRP3)] activity. Loss of function in A20 will increase IL‐1β and IL‐18 release and thus aggravate the inflammatory responses [21]. On the other hand, a mutation in nuclear factor kappa B essential modulator (NEMO) (p.D406V) was identified to cause BD‐like disease in females. NEMO is essential for proper activation of NF‐κB and thus is tightly intertwined with regulating inflammatory responses. The D406V substitution mutation in NEMO is located in the zinc finger domain of the protein that is essential for its ubiquitination and proteasomal degradation. Loss of function of NEMO eventually leads to a BD‐like phenotype [22]. Finally, a dominant negative mutation in NF‐κB1 (p.H67R) was found in a family presenting with respiratory infections and BD‐like lesions. Interestingly, the mutant protein showed increased affinity to NEMO, therefore delaying transcription factor entry into the nucleus and causing decreased NF‐κB activity and increased inflammasome activation [23]. All three mutations, A20, NEMO and NF‐κB1, have been reported as monogenic causes of BD‐like diseases by interfering with NF‐κB activity and causing inflammation. Interestingly, a retrospective study from 2019 looking at patients that have originally been classified as atypical BD cases revealed that, in fact, the patients did not suffer from true BD but carried different monogenic mutations, including A20, that caused diseases mimicking BD [5]. This study indicates that careful genetic assessment of rare disease‐causing variants is crucial, particularly in cases of atypical disease presentation, in order to accurately diagnose BD. This will be an important step not only to guarantee optimal treatment strategy for the patient but also to gain further insights into the underlying BD mechanism.

It has been suggested that GM‐CSF production is induced by activation of the NF‐κB pathway by TNF‐α signaling [24], and anti‐TNF‐α treatment is currently used as a treatment for patients with BD. We further explored whether TNF‐α drives GM‐CSF production in human immune cells. While phorbol 12‐myristate 13‐acetate (PMA)‐activated monocytes did not produce GM‐CSF, the addition of TNF‐α resulted in the release of GM‐CSF. These data suggest that a primed immune system will respond to TNF‐α by producing GM‐CSF and that blocking TNF‐α might lead to a reduction in GM‐CSF release. Notably, a report published in 2004 states that recombinant GM‐CSF given to patients can induce Sweet’s syndrome, which is an acute febrile neutrophilic skin disease [25]. BD is also a neutrophilic skin disease, and an overlap in clinical features with Sweet’s syndrome has been suggested [26]. These reports further support our hypothesis that increased GM‐CSF activation contributes to the pathogenesis of skin lesions in BD. Our patient received anti‐TNF‐α and her clinical signs, especially pathergic reaction, improved, possibly due to reducing GM‐CSF production and thus dampening the local inflammatory response. She also had much less flares and signs of mucocutaneous BD when she was on anti‐TNF‐α treatment. Targeting GM‐CSF directly might be a promising future approach to treat our patient and might even be a therapeutic strategy for other patients suffering from BD without extreme pathergy (Fig. 4).

Fig. 4.

Schematic overview of mutations in the nuclear factor kappa B (NF‐κB) and granulocyte–macrophage colony‐stimulating factor (GM‐CSF) signaling pathways and their effect on inflammation involved in Behçet’s disease (BD) pathology. In the present study we describe the N44H substitution variant in the CSF2 gene, which may cause neutrophil recruitment and increased production of IL‐1Ra, therefore possibly explaining skin lesions as well as the absence of systemic inflammation in our patient.

One other striking feature of our patient was the lack of systemic inflammation while she underwent explorative surgery three times for a suspicion of fasciitis necroticans (Fig. 1). Her CRP levels remained undetectable and she had no fever during this episode. Although we could not pinpoint the exact cause, we observed that GM‐CSF in vitro was only able to induce IL‐1Ra in PBMCs. This effect was dose‐dependent, and therefore it is tempting to speculate that her hyperactive GM‐CSF is able to induce large amounts of the anti‐inflammatory cytokine IL‐1Ra, which is able to suppress CRP production by the liver during inflammatory conditions. In addition, we observed that several earlier episodes of disease of our index patient started with the suspicion of an infection. It is tempting to speculate that the patient suffers from an aberrant innate pattern‐recognition (PRR) signaling; however, functional studies where PBMCs of the patient were exposed to fungal, Gram‐negative or Gram‐positive bacteria did not show differences in the induction of innate cytokines compared to healthy controls (data not shown). Also, we did not find rare mutations in genes encoding the proteins involved in PRR‐signaling pathways.

Taken together, the present study identifies a single family in which a GM‐CSF as a new monogenic mutation may support the development of BD‐like disease. Loss of glycosylation leads to hyperactivity of GM‐CSF and might eventually contribute to pathergic reactions without systemic inflammation. These findings allow a rationale to re‐evaluate existing genetic data of atypical BD patients and check for GM‐CSF pathway‐related mutations. In particular, patients presenting with conspicuous cutaneous involvement might benefit from blocking TNF‐α or anti‐GM‐CSF treatment in the future.

Disclosures

The authors declare no competing interests.

Supporting information

Table S1. Whole exome sequencing statistics and the filter settings applied to exclude non‐coding, synonymous, common and low quality variants.

Table S2. The rare genetic variants identified in the index patient, selected for their function in the immune system and in silico predictions.

Data Availability Statement

Data are available upon request.