Dimethyl Fumarate Disrupts Human Innate Immune Signaling by Targeting the IRAK4–MyD88 Complex

This article is featured in In This Issue, p.2515

Plasmacytoid dendritic cells (pDCs) are a minor DC subset that produce large amounts of type 1 IFN in response to viral and endogenous nucleic acids (1, 2). This cell type bridges the innate and adaptive immune systems, a duality reflected in its name. pDCs share a progenitor with the more common conventional DCs and constitutively express the FLT3 receptor; however, pDCs also display a similar morphology as plasma cells, hence the modifier plasmacytoid (3, 4). Other hallmarks of pDCs include low MHC class II expression and low (mouse) or negative (human) CD11c expression (1, 5). The pDC population comprises only 0.3–0.5% of human peripheral blood cells but is implicated in a variety of pathologic conditions (1) in which pDCs tend to accumulate in affected tissues to promote local inflammation.

Systemic lupus erythematosus patients display low levels of circulating pDCs but accumulate IFN-producing pDCs in the skin (6, 7). Similar results have been found in psoriasis patients, in which pDCs infiltrate the skin and become activated, driving the early stages of the disease (8). In a xenograft model of human psoriasis, blockade of IFN-α signaling or production prevented the development of disease (8), suggesting that inhibition of pDC function could have therapeutic value. pDCs have been identified in the cerebrospinal fluid of multiple sclerosis (MS) patients (9) and accumulate in demyelinated lesions of inflamed MS brains (10). It has been reported that depletion of pDCs in chronic and relapsing mouse models of MS results in the exacerbation of experimental autoimmune encephalomyelitis symptoms and increased CD4⁺ T cell activation in the CNS, suggesting a negative regulatory role for pDCs in modulating inflammatory T cell responses (11). Furthermore, pDCs harvested from IFN-β–treated MS patients display a lower expression of the activation markers CD83 and CD86 (12), and therefore IFN therapy may act at least in part through modulating pDC function.

Dimethyl fumarate (DMF) is an approved oral therapy for MS and has also been used for the treatment of psoriasis for several decades in an alternative formulation called Fumaderm, with DMF believed to be the active species (13–15). Despite its long history and use in multiple autoimmune conditions, the mechanism of the action of DMF remains poorly understood. DMF can promote an antioxidant response through activation of the NRF2–KEAP1 pathway (16–18). However, DMF has been found to suppress pathologic conditions in the experimental autoimmune encephalomyelitis model of MS independently of NRF2 (19), indicating that the therapeutic effect of DMF in MS and other autoimmune disorders may involve proteins and pathways beyond the KEAP1–NRF2 system.

DMF is an electrophilic small molecule that can react with nucleophilic cysteine residues in proteins. It is now appreciated that many immune signaling proteins harbor electrophile/oxidation-sensitive cysteines (20–22) and that adduction of reactive cysteines in the proteome by DMF and other electrophilic compounds may constitute a general mechanism for suppressing the activity of immune cells (20, 23, 24). We previously used a quantitative chemical proteomics platform to identify DMF-sensitive cysteines in primary human T cells, which included a CXXC motif in the C2 domain of PKCθ that was found to contribute to CD28 binding and T cell activation (20). However, the extent to which DMF impacts the function of other immune cells and whether such effects involve reactivity with specific cysteine residues in the proteome remains poorly understood. In this study, we have examined DMF activity in pDCs, an immune cell type that plays a central role in psoriasis. We first show that DMF, but not structurally related nonelectrophilic analogues, inhibit IFN-α and cytokine/chemokine production from primary human pDCs. DMF-mediated suppression of IFN-α production was independent of either NRF2 activation or global changes in the glutathione (GSH) content of pDCs. We then used the chemical proteomic method termed isotopic tandem orthogonal proteolysis (isoTOP)–activity-based protein profiling (ABPP) to map DMF-sensitive cysteines in the pDC line Cal-1. Among the >4000 total cysteines quantified by isoTOP-ABPP in Cal-1 cells, cysteine 13 (C13) of IL-1R–associated kinase IRAK4 stood out as being among the most sensitive to DMF. Noting that C13 resides proximal to the interface involved in MyD88 binding (25) and adjacent to a site of mutation in IL-1R–associated kinase (IRAK) 4 (R12C) implicated in human immunodeficiency (26), we proceeded to show that DMF disrupts both IRAK4–MyD88 interactions and IRAK4-mediated TNF-α production in a C13-dependent manner. Taken together, our data indicate that DMF modifies C13 of IRAK4 in human pDCs, and this reaction disrupts MyD88 binding and IRAK4 function. These studies potentially reveal a mechanistic basis for the efficacy of DMF observed in autoimmune syndromes like MS and psoriasis and point to a novel druggable protein–protein interaction (PPI) for controlling innate immune signaling.

Assays were performed with the following reagents: DMF (242926; Sigma-Aldrich), monomethyl fumarate (MMF) (651419; Sigma-Aldrich), dimethyl succinate (DMS) (W239607; Sigma-Aldrich), and buthionine sulfoximine (BSO) (14484; Cayman Chemical). The Nrf2 inhibitor ML385 (SML1833; Sigma Aldrich), dimethyl maleate (238198; Sigma-Aldrich), diethyl fumarate (D95654; Sigma-Aldrich), diethyl maleate (D97703; Sigma-Aldrich), diisobutyl fumarate (7283-69-4; TCI Chemicals), and diisopropyl fumarate (7283-70-7; TCI Chemicals).

All studies with samples from human volunteers followed protocols approved by The Scripps Research Institute institutional review board. Blood from healthy donors (females aged 30–49) were obtained after informed consent. PBMCs were purified over Histopaque-1077 gradients (10771; Sigma-Aldrich) according to the manufacturer’s instructions. Briefly, blood (20 × 25 ml aliquots) was layered over Histopaque-1077 (12.5 ml), and the samples were then fractionated by centrifugation at 750 × g for 20 min at 20°C with no brake. PBMCs were harvested from the Histopaque–plasma interface and washed twice with PBS. After that time, pDCs were isolated with a CD304 (BDCA-4/Neuropilin-1) cell Isolation Kit (Miltenyi Biotec) according to the manufacturer’s instructions.

Synthesized oligodeoxynucleotides (ODNs) were purchased from InvivoGen. The sequences of the ODNs are as follows: CpG-A (ODN2216), 5′-ggGGGACGA:TCGTCgggggg-3′, and CpG-B (ODN1826), 5′-tccatgacgttcctgacgtt-3′. Bases shown in capital letters are phosphodiester, and those in lower case are phosphorothioate (nuclease resistant). Palindrome is underlined. Influenza virus was propagated and titrated in Madin-Darby canine kidney cells. Human pDC were stimulated with one multiplicity of infection of influenza virus overnight, and IFN-α levels were measured in the supernatants by ELISA. Isolated human pDCs were stimulated overnight with either 500 μM loxoribine (tlrl-lox; InvivoGen) or 100 nM R-848 (Sigma-Aldrich). Supernatant IFN-α levels were measured by ELISA. CAL-1 cells were obtained from S. Kamihira, Nagasaki University. CAL-1 cells were grown in RPMI medium (Invitrogen) containing 10% FBS, 2 mM l-glutamine, 100 mg/ml streptomycin, 100 U/ml penicillin at 37° C and 5% CO₂. Sendai virus, strain Cantell, was propagated at 37°C in 10-d-old embryonated chicken eggs.

ELISAs were performed using the VeriKine Human and Mouse IFN-α ELISA Kits (R&D Systems) according to the manufacturer’s instructions.

Cells were stained with the following anti-human Abs: PE-conjugated anti-CD123 Ab (clone 6H6; BioLegend) and allophycocyanin-conjugated anti–IFN-α Ab (clone 7N4-1; BD Pharmingen). Flow cytometry acquisition was performed with a BD FACSDiva-driven BD LSR II Flow Cytometer (Becton Dickinson). Data were then analyzed with FlowJo software (Tree Star).

Human IRAK4-deficient cells (patient identification: MB002334) were provided as a gift from the laboratory of Jean-Laurent Casanova (27). Cells were cultured in RPMI 1640 medium (Life Technologies Invitrogen) supplemented with heat-inactivated FBS and a penicillin–streptomycin solution (Life Technologies Invitrogen). Cells were maintained at a concentration of 1 million cells per 1 ml of medium with fresh medium added every 2 d.

To reconstitute IRAK4 expression within the IRAK4-deficient cells, the IRAK4 open reading frame was cloned from plasmid HsCD00330298 (Harvard Plasmid Information Database Repository) and inserted into the pLPC-myc-mCherry lentiviral expression system, which was provided as a gift from the Lazzerini Denchi laboratory. The C13 mutants were generated using the Q5 Site-Directed Mutagenesis Kit (New England BioLabs).

HEK293T cells were seeded 3E6 cells per dish in 10-cm dishes. The next day, to an Eppendorf tube was added either 1) 2.5 μg of IRAK4 construct and 2.5 μg of GFP construct or 2) 2.5 μg of IRAK4 construct and 2.5 μg of MyD88 construct. To each tube was added 500 μl of DMEM and 30 μl of polyethyleneimine (PEI). Twenty minutes later, the transfection solution was added dropwise to cells. Twenty hours later, DMF or DMSO vehicle was added to the cells as indicated.

HEK293T cells were harvested, lysed by sonication, and diluted to a concentration of 2 mg protein/ml. Protein concentrations were measured with the Bio-Rad Laboratories DC protein assay reagents A and B (5000113, 5000114; Bio-Rad Laboratories). The proteome sample (500 μl) was treated with 100 μM iodoacetamide (IA)-alkyne probe by adding 5 μl of a 10 mM probe stock (in DMSO). The labeling reactions were incubated at room temperature for 1 h, after which the samples were conjugated to isotopically labeled, tobacco etch virus (TEV)–cleavable tags (TEV tags) by copper-catalyzed azide-alkyne cycloaddition (click chemistry). Heavy click chemistry reaction mixture (60 μl) was added to the DMSO-treated control sample, and 60 μl of the light reaction mixture was added to the compound-treated sample. The click reaction mixture consisted of TEV tags (10 μl of a 5 mM stock; light [compound-treated] or heavy [DMSO treated]), CuSO₄ (10 μl of a 50 mM stock in water), and TBTA (30 μl of a 1.7 mM stock in 4:1 tBuOH:DMSO). To this was added TCEP (10 μl of a 50 mM stock). The reaction was performed for 1 h at room temperature. The light- and heavy-labeled samples were then centrifuged at 16,000 × g for 5 min at 4°C to harvest the precipitated proteins. The resulting pellets were resuspended in 500 μl of cold methanol by sonication, and the heavy and light samples were combined pairwise. Combined pellets were then washed with cold methanol, after which the pellet was solubilized by sonication in PBS and 1.2% SDS. The samples were heated at 90°C for 5 min and subjected to streptavidin enrichment of probe-labeled proteins, sequential on-bead trypsin and TEV digestion, and liquid chromatography–tandem mass spectrometry, according to the published isoTOP-ABPP protocols (28–30).

RAW Xtractor (version 1.9.9.2; available at http://fields.scripps.edu/downloads.php) was used to extract the MS2 spectra data from the raw files (MS2 spectra data correspond to fragments analyzed during the second stage of mass spectrometry). MS2 data were searched against a reverse concatenated, nonredundant variant of the Human UniProt database (release-2012_11) with the ProLuCID algorithm (publicly available at http://fields.scripps.edu/downloads.php) (31). Cysteine residues were searched with a static modification for carboxyamidomethylation (+57.02146) and up to one differential modification for either the light or heavy TEV tags (+464.28595 or +470.29976, respectively). Peptides were required to have at least one tryptic terminus and to contain the TEV modification. ProLuCID data were filtered through DTASelect (version 2.0) to achieve a peptide false-positive rate below 1% (32).

The quantification of heavy/light ratios (isoTOP-ABPP ratios, R values) was performed by in-house CIMAGE software (30) using default parameters (three MS1s per peak and a signal-to-noise threshold of 2.5). Site-specific engagement of electrophilic compounds was assessed by blockade of IA-alkyne probe labeling. For peptides that showed a ≥95% reduction in MS1 peak area from the compound-treated proteome (light TEV tag) when compared with the DMSO-treated proteome (heavy TEV tag), a maximal ratio of 20 was assigned. Overlapping peptides with the same labeled cysteine (for example, they had the same local sequence around the labeled cysteines but had different charge states, Multidimensional Protein Identification Technology segment numbers, or tryptic termini) were grouped together, and the median ratio from each group was recorded as the R value of the peptide for that run.

HEK293T cells (2.5 × 10⁶) were plated in a 10-cm plate 24 h prior to transfection. For transfection, DNA (5 μg, myc-tagged IRAK4 kinase-dead C13C, R12C, or other C13 mutants; FLAG-tagged MyD88; or FLAG-tagged MetAP2), serum-free DMEM (500 μl; Corning), and PEI (30 μl) were mixed together and allowed to incubate for 30 min at room temperature. The mixture was then added dropwise to plated cells and cells rested for 48 h. Media from cells transfected with FLAG-tagged protein was aspirated, and the cells were washed with PBS (5 ml). Cells were transferred to a 1.5 ml Eppendorf tube and washed with PBS 1×. Freshly prepared ice-cold lysis buffer was added to the cell pellet (400 μl; 50 mM HEPES [pH 7.4], 150 mM NaCl, 1% Triton-X 100, and 1 mM DTT with protease [Mini EDTA-free; Roche] and phosphatase [Phos-STOP; Roche] inhibitors). Samples were incubated for 30 min on ice and centrifuged (10 min at 10,000 × g at 4°C). Protein concentration was normalized to 2 mg/ml. FLAG-tagged protein input samples were generated (100 μg protein in 50 μl sample). FLAG beads (22.5 μl per 1.5 mg protein; Thermo Fisher Scientific) were prewashed with lysis buffer and pelleted by centrifugation (2 min at 1600 × g at 4°C). This bead wash protocol was repeated 1×. Prewashed FLAG beads were then added to 1.5 mg lysate (750 μl) and rotated (3 h at 4°C). Meanwhile, media from cells transfected with myc-tagged protein was aspirated, and cells were washed with PBS. Cells were then scraped and transferred to a 1.5 ml Eppendorf tube and subjected to an additional wash with PBS 1×. Freshly prepared ice-cold lysis buffer was added to each cell pellet (400 μl; 50 mM HEPES [pH 7.4], 150 mM NaCl, 1% Triton-X 100, 1 mM DTT with protease [Mini EDTA-free; Roche] and phosphatase [Phos-STOP; Roche] inhibitors). Cell lysates were incubated on ice for 30 min and centrifuged (10 min at 10,000 × g at 4°C). Protein concentration was normalized to 1 mg/ml. Myc-tagged protein input samples were generated (100 μg protein in 50 μl sample). Upon completion of FLAG enrichment, beads were pelleted by centrifugation (2 min at 1600 × g at 4°C). Flow through was aspirated away, and ice-cold lysis buffer was added. Tubes were inverted gently to wash beads (500 μl). These washes were repeated 2×. FLAG enrichment was resuspended in lysis buffer (100 μl) and transferred to myc-tagged protein lysate (750 μg, 750 μl). The sample was then incubated on a rotator (2 h at 4°C). Upon completion, beads were pelleted by centrifugation (2 min at 1600 × g at 4°C), and unbound lysate was aspirated away. Fresh ice-cold lysis buffer was added, and the tubes were inverted gently to wash beads (500 μl). This bead wash was repeated 2×. Beads were resuspended in 1× loading buffer (45 μl lysis buffer and 15 μl 4× loading buffer [40% glycerol, 0.4% bromophenol blue, 2.8% 2-ME (pH 6.8)]), and samples were boiled to denature proteins and cleave disulfides (5 min, 95°C). Samples were then loaded onto a 10% Tris-glycine gel for SDS-PAGE separation (30 μl immunoprecipitation, 15 μl input).

HEK293T cells (2.5 × 10⁶) were plated in a 10-cm plate 24 h prior to transfection. For transfection, DNA (5 μg, myc-tagged IRAK4 kinase-dead C13C, R12C, or other C13 mutants; FLAG-tagged MyD88; or FLAG-tagged MetAP2), serum-free DMEM (500 μl; Corning), and PEI (30 μl) were mixed together and allowed to incubate for 30 min at room temperature. The mixture was then added dropwise to plated cells, and cells were rested for 48 h. Cells expressing myc-tagged proteins were treated with DMSO, DMF, MMF, or DMS (1000× stock in DMSO, final concentration 100 μM) and incubated for 4 h. Media from cells transfected with FLAG-tagged protein was aspirated, and the cells were washed with PBS (5 ml). Cells were transferred to a 1.5 ml Eppendorf tube and washed with PBS 1×. Freshly prepared ice-cold lysis buffer was added to the cell pellet (400 μl; 50 mM HEPES [pH 7.4], 150 mM NaCl, 1% Triton-X 100, 1 mM DTT with protease [Mini EDTA-free; Roche] and phosphatase [Phos-STOP; Roche] inhibitors). Samples were incubated for 30 min on ice and centrifuged (10 min at 10,000 × g at 4°C). Protein concentration was normalized to 2 mg/ml. FLAG-tagged protein input samples were generated (100 μg protein in 50 μl sample). FLAG beads (22.5 μl per 1.5 mg protein; Thermo Fisher Scientific) were prewashed with lysis buffer and pelleted by centrifugation (2 min at 1600 × g at 4°C). This bead wash protocol was repeated 1×. Prewashed FLAG beads were then added to 1.5 mg lysate (750 μl) and rotated (3 h at 4°C). Meanwhile, media from cells transfected with myc-tagged protein was aspirated, and cells were washed with PBS. Cells were then scraped and transferred to a 1.5 ml Eppendorf tube and subjected to an additional wash with PBS 1×. Freshly prepared ice-cold lysis buffer was added to each cell pellet (400 μl; 50 mM HEPES [pH 7.4], 150 mM NaCl, 1% Triton-X 100, 1 mM DTT with protease [Mini EDTA-free; Roche] and phosphatase [Phos-STOP; Roche] inhibitors). Cell lysates were incubated on ice for 30 min and centrifuged (10 min at 10,000 × g at 4°C). Protein concentration was normalized to 1 mg/ml. Myc-tagged protein input samples were generated (100 μg protein in 50 μl sample). Upon completion of FLAG enrichment, beads were pelleted by centrifugation (2 min at 1600 × g at 4°C). Flow through was aspirated away, and ice-cold lysis buffer was added. Tubes were inverted gently to wash beads (500 μl). These washes were repeated 2×. FLAG enrichment was resuspended in lysis buffer (100 μl) and transferred to myc-tagged protein lysate (750 μg, 750 μl). The sample was then incubated on a rotator (2 h at 4°C). Upon completion, beads were pelleted by centrifugation (2 min at 1600 × g at 4°C), and unbound lysate was aspirated away. Fresh ice-cold lysis buffer was added, and the tubes were inverted gently to wash beads (500 μl). This bead wash was repeated 2×. Beads were resuspended in 1× loading buffer (45 μl lysis buffer and 15 μl 4× loading buffer [40% glycerol, 0.4% bromophenol blue, 2.8% 2-ME (pH 6.8)]), and samples were boiled to denature proteins and cleave disulfides (5 min, 95°C). Samples were then loaded onto a 10% Tris-glycine gel for SDS-PAGE separation (30 μl immunoprecipitation, 15 μl input).

Proteins were transferred to nitrocellulose membrane in a wet transfer cell (50 V, 2.5 h). Membrane was blocked in 5% milk for 1 h followed by primary incubation overnight at 4°C (1:2500 anti-myc [Cell Signaling Technology] or 1:2500 anti-DYKDDDDK [Cell Signaling Technology]). Membranes were washed with TBST (3×, 10 min) and incubated with secondary anti-rabbit Ab (LI-COR Biosciences) for 1 h at room temperature.

IRAK4 kinase activity was evaluated by the IRAK4 Kinase Enzyme System (V9421; Promega) according to the manufacturer’s instructions. Per manufacturer’s instructions, staurosporine (S6942; Sigma-Aldrich) was administered to generate a kinase inhibitor dose-response curve.

Significance in Figs. 1–3 were determined by two-tailed unpaired t test from at least two independent experiments of at least four replicates per experiment using Prism 7 (GraphPad). Significance in Fig. 4 was determined by a paired t test from three independent experimental replicates using Prism 7 (GraphPad).

Primary human pDCs isolated from normal human donors were purified, stimulated with CpG-A, and treated concomitantly with DMF, the major DMF metabolite MMF, or the saturated unreactive analogue DMS (50 μM each compound; Fig. 1A). DMF, but not MMF or DMS, produced a concentration-dependent inhibition of CpG-A–induced IFN-α release, displaying an IC₅₀ value of ∼7 μM (Fig. 1B, 1C). The inhibitory action of DMF on IFN-α production occurred without an effect on pDC viability (Supplemental Fig. 1A) and was accompanied by the suppression of multiple additional cytokines and chemokines (Supplemental Fig. 1B). We further found that intracellular IFN-α was also decreased by DMF, but not MMF or DMS (Fig. 1D–F), which pointed to a mechanism of action in which DMF blocks the expression, rather than secretion, of IFN-α (Fig. 1D).

We next evaluated additional pDC stimuli and found that, as observed with CpG-A, DMF also blocked IFN-α production following the treatment with CpG-B, another TLR9 agonist, or influenza (TLR7 agonist) (Fig. 2A). Interestingly, however, the impact of DMF on other cytokines varied depending on the stimulus, with a broader suppressive effect being observed for DMF with the TLR9 agonists CpG-A and CpG-B but not following stimulation with influenza virus (Supplemental Fig. 2). To further investigate the potential role of DMF in blocking TLR7-induced cytokine secretion independent of a viral recognition event, we treated pDCs with the additional TLR7/8 stimulus R848 and the TLR7-specific agonist loxoribine. As we found for influenza virus, DMF treatment impaired cytokine production in response to these additional TLR7 stimuli (Fig. 2A). These data suggest that DMF differentially impairs the activation of innate signaling pathways acting through multiple TLRs.

We further explored the structure–activity relationship for the blockade of IFN-α production by treating pDCs with a set of fumarate ester analogues with varied stereoelectronic properties (Fig. 2B). Although several compounds with increased steric bulk (e.g., diethyl fumarate, diisopropyl fumarate, diisobutyl fumarate, dibenzyl fumarate) suppressed IFN-α production, none exhibited comparable activity to DMF (50 μM of each compound; Fig. 2C). We also tested the impact of double bond geometry and found that dimethyl maleate and diethyl maleate partially blocked IFN-α production, but neither compound displayed equivalent activity to DMF (Fig. 2C). To the extent that the Z-olefin arrangement in dimethyl maleate and diethyl maleate would be expected to increase electrophilicity, these data suggest a more complex, nonlinear relationship between the intrinsic reactivity and immunosuppressive effects of fumarate esters.

The pharmacological effects of DMF have been proposed to occur, at least in part, through activation of the KEAP1–NRF2 pathway (16), which is a major mediator of cellular responses to electrophilic/oxidative stress (33). We found, however, that DMF fully suppressed IFN-α production in pDCs cotreated with the NRF2 inhibitor ML385 (34) (Fig. 2D). ML385 treatment alone produced a partial reduction in IFN-α, but this effect was much less dramatic than the complete blockade of IFN-α production by DMF (Fig. 2D). DMF has also been shown to deplete intracellular concentrations of GSH (14, 17, 18, 35), and we found that the GSH synthesis inhibitor BSO modestly decreased IFN-α release from pDCs (Fig. 2E). Again, this change was much lower in magnitude than the IFN-α suppression caused by DMF (Fig. 2E). These data, taken together, indicate that the primary mechanism by which DMF suppresses IFN-α production is independent of NRF2 and involves biochemical pathways beyond the reduction of GSH content in pDCs.

We next sought to identify candidate protein targets that might contribute to the IFN-α suppressing activity of DMF in pDCs. We previously used the chemical proteomic method isoTOP-ABPP to globally map DMF-sensitive cysteines in primary human T cells (20). The sample requirements for such chemical proteomic experiments are, however, quite high (milligrams of protein) and beyond the scale that could be prepared from primary human pDCs. We therefore turned to Cal-1 cells as a pDC model cell line (36) for the discovery of DMF-sensitive cysteines.

We first confirmed that DMF maintains IFN-α–suppressing activity in Cal-1 cells (Fig. 3A) and then applied isoTOP-ABPP to map DMF-sensitive cysteines in these cells. In brief, Cal-1 cells were treated with DMF (50 μM, 1 or 4 h) or DMSO control, lysed, and exposed to the general cysteine-reactive probe IA-alkyne, followed by click chemistry–mediated conjugation (37) to isotopically differentiated azide-biotin tags containing a TEV protease–cleavable linker. Samples were then subject to streptavidin enrichment and protease cleavage, after which heavy and light samples were combined and analyzed by liquid chromatography–tandem mass spectrometry to quantify IA-alkyne–labeled cysteines using an LTQ-Orbitrap Velos instrument (Fig. 3B) (20, 21, 30).

Among the 4408 cysteines quantified by isoTOP-ABPP in Cal-1 cells, ∼170 residues showed substantial sensitivity to DMF as reflected in >4-fold changes in IA-alkyne reactivity in DMF-treated cells (Fig. 3C). Most of these DMF-sensitive cysteines showed a greater change in IA reactivity at 4 h (Fig. 3C), likely reflecting a time-dependent increase in modification by DMF. A representative example of such time-dependent change is shown for C75 of adenosine deaminase in Fig. 3D. A select subset of cysteines, in contrast, showed a near-complete loss of IA-alkyne reactivity following only 1 h of DMF treatment. These DMF-hypersensitive cysteines included C13 of IRAK4 (Fig. 3C, 3D), a protein kinase that plays a central role in TLR-mediated innate immune cell signaling pathways that produce IFN-α (38). The immunological functions of IRAK4 have inspired the development of several ATP-competitive inhibitors of this kinase as potential drugs to treat autoimmune and autoinflammatory disorders (39, 40). Interestingly, however, C13 is not found in the IRAK4 active site but rather located at the interface of IRAK4 that binds to the adaptor protein MyD88 (Fig. 3E). We next sought to understand if DMF modification of C13 in IRAK4 affects interactions with MyD88 and downstream signaling.

A primary function of pDCs is to produce IFN-α in response to viral or endogenous nucleic acids (1, 2). The receptors for these nucleic acids are TLR7/8 and TLR9 (5), which in turn signal through the Myddosome complex comprising MyD88, IRAK2, and IRAK4 (41). The location of the DMF-hypersensitive cysteine in IRAK4, C13, is at the interface of the IRAK4–MyD88 PPI (Fig. 3E), suggesting a potential functional role for this residue. This hypothesis is also supported by human genetics as a mutation of the adjacent residue Arg12 (R12C) produces an immunodeficiency syndrome that manifests as an increased susceptibility to pyogenic infections in children (27, 42, 43). The R12C mutation has been shown to disrupt binding between IRAK4 and MyD88 (44), presumably hampering pDC responses to exogenous nucleic acids. We found that recombinantly expressed IRAK4 (produced by transient transfection in HEK293T cells) maintained site-specific sensitivity to DMF at C13 (Fig. 3F), and this interaction was disrupted by coexpression of MyD88 in HEK293T cells (Fig. 3G).

We next established an in vitro binding assay for measuring the IRAK4–MyD88 interaction. We immobilized FLAG-tagged MyD88 on anti-FLAG beads for incubation with lysate from HEK293T cells expressing various R12/C13 mutants in the context of a kinase-dead variant of IRAK4 (K213A/K214A; termed IRAK4-K2D). We chose to use the kinase-dead IRAK4 because this form of IRAK4 has been shown to interact more stably with MyD88, likely indicating that IRAK4 and MyD88 engage in a dynamic signaling complex that can be disassembled by autophosphorylation of IRAK4 (45–47). Consistent with expectations based on past work (44), we found that the R12C mutant of IRAK4-K2D blocked coprecipitation of this protein with FLAG-tagged MyD88 (Supplemental Fig. 3A). We next treated IRAK4-K2D–transfected cells with DMF (100 μM, 4 h) and found that this compound, but not structurally related analogues that lack IFN-α–suppressing activity (MMF, DMS), substantially blocked the interaction of IRAK4-K2D with MyD88 in vitro (Fig. 4A). This antagonistic effect of DMF on the IRAK4-K2D–MYD88 interaction was not observed for a C13A-mutant of IRAK4-K2D (Fig. 4B), supporting that DMF reactivity with C13 is responsible for disrupting IRAK4-K2D binding to MyD88. Interestingly, we found that mutation of C13 to negatively charged residues glutamate or aspartate, but not other amino acids, including glutamine, mimicked the inhibitory effect of DMF on the IRAK4-K2D–MYD88 interaction (Fig. 4C). One interpretation of these findings is that DMF, once reacted with C13 of IRAK4-K2D, may undergo esterolysis to present a carboxylate group that, like the C13D and C13E mutants, impairs binding to MyD88.

Our studies pointed to a kinase activity-independent mechanism by which covalent DMF modification of IRAK4 disrupts Myddosome function (i.e., disruption of IRAK4–MyD88 interactions). To more directly assess whether DMF impacts IRAK4 kinase activity, we treated purified IRAK4 (50 ng/reaction) with DMF (4 mM–2 nM) or the pan-kinase inhibitor staurosporine (40 μM–0.02 nM) and measured residual IRAK4 activity using a substrate assay. DMF did not alter IRAK4 kinase activity in this assay, in contrast to staurosporine, which showed clear concentration-dependent inhibition of IRAK4 (Fig. 4D).

One of the downstream consequences of Myddosome complex formation is the activation of the NF-κB signaling pathway. DMF has been reported to inhibit NF-κB signaling in a variety of cell types (48–52). Consistent with these past findings and with the disruption of the Myddosome complex, we found that DMF blocks phosphorylation of the NF-κB pathway member p65 in pDCs stimulated with CpG-B (Supplemental Fig. 3B).

We next examined the impact of DMF on IRAK4-induced cytokine production in human immune cells. We employed EBV immortalized B cells (B-EBV) that were generated from PBMCs of a patient with a large deletion in the IRAK4 gene that renders it nonfunctional (27). We used the B-EBV cells as a source of IRAK4-null cells amenable to genetic complementation studies. IRAK4-deficient B-EBV cells were then reconstituted with either wild-type (WT) or the C13A mutant of IRAK4 using lentiviral transduction. Reconstitution of IRAK4-deficient B-EBV cells with WT-IRAK4 resulted in significantly greater TNF-α production following LPS and CpG stimulation compared with IRAK4-deficient B-EBV cells, whereas the TLR3/RIG-I/MDA5 agonist polyinosinic-polycytidylic acid (Poly I:C), which signals independent of MyD88/IRAK4, induced TNF-α production in both WT-IRAK4 and IRAK4-deficient cells (Supplemental Fig. 3C). We next discovered that DMF, but not MMF or DMS, suppressed TNF-α production following stimulation with LPS and CpG in the B-EBV cells reconstituted with WT-IRAK4 (Fig. 4E). In contrast, DMF did not affect TNF-α production following Poly I:C stimulation (Fig. 4E), suggesting that DMF acts to suppress pDC cytokine production predominantly through IRAK4/MyD88 signaling. Finally, we confirmed that DMF does not decrease CpG-induced TNF-α production in IRAK4-deficient cells reconstituted with the C13A-IRAK4 mutant (Fig. 4F). We should note that B-EBV cells expressing the C13A-IRAK4 mutant showed a lower TNF-α induction than WT-IRAK4–reconstituted cells, but regardless, the TNF-α induction in C13A-IRAK4 mutant cells was still much larger than that of IRAK4-deficient cells (which showed a negligible response; see Supplemental Fig. 3C), and DMF treatment did not further reduce TNF-α production in the C13A-IRAK4 cells.

pDCs were first definitively characterized in 1999 and have since been implicated in the initiation of psoriasis (8, 53) and the progression of MS (9, 10, 12). Despite their low abundance in the blood, pDCs are the primary producers of IFN-α and can thus drive local inflammation (1). The immunomodulatory drug DMF is a widely prescribed and effective treatment for MS and psoriasis, both of which are inflammatory diseases. More recently, DMF has also been reported as a potential therapy for cutaneous T cell lymphoma and is at times prescribed off label to bone marrow transplant patients because of its immunomodulatory properties (54). The mechanism of DMF is not well understood, but this drug has been shown to impair the activation of T cells (19, 20) and conventional DCs (55). However, little is known about the effect of DMF on pDCs. Our studies indicate that pDCs are sensitive to DMF, and that this pharmacological effect is largely independent of NRF2 and, at least in part, through the NF-κB signaling pathway, with covalent adduction of C13 of IRAK4 residing at the top of the signaling cascade. Given that DMF is relatively well tolerated and capable of suppressing proinflammatory signaling in both activated T cells and now pDCs, a more detailed understanding of its mechanisms of action may reveal potential new therapeutic applications in hematological malignancies and autoimmune disorders.

We have shown in this study that DMF and more elaborated electrophilic analogues, but not the poorly reactive hydrolytic metabolite MMF or DMS, inhibit IFN-α secretion from pDCs. This modulation depends primarily on proteins and pathways beyond the conventional electrophile-sensing NRF2–KEAP1 pathway or GSH depletion. Additionally, DMF inhibits multiple modes of pDC stimulation, consistent with the drug targeting a central signaling pathway underlying different types of immune cell activation. TLR7 and TLR9 are the primary innate immune receptors on pDCs (3). Their activation promotes the assembly of the Myddosome complex, which is comprised of MyD88, IRAK2, and IRAK4. Using a global chemical proteomic method that quantified >4000 cysteine residues in the human pDC line Cal-1, we identified C13 of IRAK4 as a “hot spot” for modification of DMF. IRAK4 signaling underlies each of the different modes of pDC activation shown to be inhibited by DMF. That not only DMF but also the mutation of C13 to aspartate or glutamate disrupted IRAK4–MyD88 interactions suggests a potential mechanism where DMF reacts with C13 on IRAK4 and is subsequently hydrolyzed to unveil a negatively charged appendage that disrupts MyD88 binding. Importantly, but not entirely surprising, given the distal location of the modified residue to the active site, DMF adduction of C13 does not affect IRAK4 kinase activity, suggesting that disruption of the PPI of IRAK4 and MyD88 is the primary mechanism by which DMF impairs Myddosome function in pDCs.

Although PPIs play numerous and critical roles in cell signaling, targeting this class of interactions with small-molecule drugs has historically proven challenging because many PPI interfaces are flat and cover large surface areas (56). Our discovery that DMF can disrupt the IRAK4–MyD88 interaction through cysteine reactivity demonstrates the potential of covalent small molecules as chemical probes and therapeutics to target disease-relevant PPIs. Further elaboration of the methyl fumarate ester still suppressed IFN-α production but to a much lesser degree than DMF. This suggests that the small size of DMF is critical to its pharmacological activity. It is possible that some of this effect reflects a greater vulnerability of the methyl-succinylated cysteine on DMF-adducted proteins to undergo hydrolysis (enzymatic or solvolytic) to furnish a negatively charged group.

It has also been reported that endogenous fumarate production through the citric acid cycle is associated with oncogenesis (57, 58). We have demonstrated that other methyl fumarate-bearing small molecules (59) are vulnerable to enzymatic hydrolysis by carboxylesterases, but it is unclear whether DMF can be hydrolyzed twice to reveal the free-acid fumarate oncometabolite in appreciable concentrations. We have previously reported that the singly hydrolyzed metabolic product of DMF, MMF, does not appreciably react with proteinaceous cysteines at treatment-relevant concentrations or suppress immune cell function (20). Nonetheless, we cannot rule out the possibility of an increase of free fumarate or a pharmacological effect of this possible outcome as a result of DMF treatment.

The authors have no financial conflicts of interest.