Immune cell targeted fumaric esters support a role of GPR109A as a primary target of monomethyl fumarate in vivo

Dimethyl fumarate (DMF) is approved as a treatment for multiple sclerosis (MS), however, its mode of action remains unclear. One hypothesis proposes that Michael addition to thiols by DMF, notably glutathione is immunomodulatory. The alternative proposes that monomethyl fumarate (MMF), the hydrolysis product of DMF, is a ligand to the fatty acid receptor GPR109A found in the lysosomes of immune cells. We prepared esters of MMF and macrolides derived from azithromycin, which were tropic to immune cells by virtue of lysosomal trapping. We tested the effects of these substances in an assay of response to Lipopolysaccharide (LPS) in freshly isolated human peripheral blood mononuclear cells (PBMCs). In this system, we observed that the 4′′ ester of MMF (compound 2 and 3) reduced levels of Interleukins (IL)-1β, IL-12 and tumor necrosis factor alpha (TNFα) significantly at a concentration of 1 µM, while DMF required about 25 µM for the same effect. The 2′ esters of MMF (compound 1 and 2) were, like MMF itself, inactive in vitro. The 4′′ ester formed glutathione conjugates rapidly while the 2′ conjugates did not react with thiols but did hydrolyze slowly to release MMF in these cells. We then tested the substances in vivo using the imiquimod/isostearate model of psoriasis where the 2′ ester was the most active at 0.06–0.12 mg/kg (approximately 0.1 µmol/kg), improving skin score, body weight and cytokine levels (TNFα, IL-17A, IL-17F, IL-6, IL-1β, NLRP3 and IL-23A). In contrast, the thiol reactive 4′′ ester was less active than the 2′ ester while DMF was ca. 300-fold less active. The thiol reactive 4′′ ester was not easily recovered from either plasma or organs while the 2′ ester exhibited conventional uptake and elimination. The 2′ ester also reduced levels of IL-6 in acute monosodium urate (MSU) induced inflammation. These data suggest that mechanisms that are relevant in vivo center on the release of MMF. Given that GPR109A is localized to the lysosome, and that lysosomal trapping increases 2′ ester activity by > 300 fold, these data suggest that GPR109A may be the main target in vivo. In contrast, the effects associated with glutathione (GSH) conjugation in vitro are unlikely to be as effective in vivo due to the much lower dose in use which cannot titrate the more concentrated thiols. These data support the case for GPR109A modulation in autoimmune diseases.


Introduction
Fumaric esters are anti-inflammatory drugs used as therapeutic treatments against psoriasis and multiple sclerosis (MS). Fumaderm ® is a mixture of dimethyl fumarate (DMF) and monoethyl fumarate (MEF) in the mixed Ca 2+ , Mg 2+ and Zn 2+ (Gillard et al. 2015;Linker and Gold 2013) salt forms. It was first approved in 1994 and since then is solely licensed in Germany. DMF oral therapy was approved by the Food and Drug Administration (FDA) in the United States in 2004 as Tecfidera ® against MS (Linker and Gold 2013). A new form to be commercialized 1 3 as Skilarence ® contains DMF and XP23829 (prodrug for monomethyl fumarate (MMF)). These fumarates have anti-inflammatory, cytoprotective, and immunomodulatory properties, however the underlying mechanisms of action are still not fully understood. So far, it is known that fumarates are able to inhibit the up-regulation of the nuclear factor kappa B (NF-κB) pathway (McGuire et al. 2016). This impacts not only cell differentiation and apoptosis, but also lowers levels of inflammatory cytokines and adhesion molecules and leads to a shift from proinflammatory T helper cells (Th) 1 and Th17 to an anti-inflammatory Th2 response (Mills et al. 2018). DMF or MEF treatment also activates the nuclear factor erythroid 2-related factor 2 (Nrf2) pathway and thus regulation of cytoprotective mechanisms. Higher levels of Nrf2 are associated with reduced oxidative stress and cellular antioxidant responses (Brennan et al. 2015;Helwa et al. 2017;Liu et al. 2016a). The degree of oxidative cell stress is positively associated with the severity of psoriasis (Kadam et al. 2010;Pleńkowska et al. 2020). DMF also conjugates to GSH via Michael addition, causing intracellular depletion of GSH at higher doses under physiological conditions (Xu et al. 2018). The product is a 2-(S-glutathionyl)-succinic dimethyl ester and is specific to DMF but not MMF. Its formation is associated with anti-inflammatory, immunosuppressive and cytoprotective responses (Brennan et al. 2015;Schmidt et al. 2007;Sullivan et al. 2013;Ghoreschi et al. 2011;Zheng 2015). The reaction between GSH and fumaric esters can also take place with other thiols like cysteine or thiol containing proteins such as glyceraldehyde 3-phosphate dehydrogenase (GAPDH), an enzyme in the aerobic glycolytic pathway. With this, fumaric esters thus inhibit GAPDH activity (Kornberg et al. 2018;Angiari and O'Neill 2018;Park et al. 2019).
The G protein-coupled receptor GPR109A, a receptor for nicotinic acid, is also activated by MMF leading to reduced neutrophil adhesion, migration and recruitment (Gillard et al. 2015;Mrowietz et al. 2018). In neutrophils, nicotinic acid stimulation also leads to apoptosis via a cAMP driven mechanism (Kostylina et al. 2008). GPR109A is also implicated in nicotinic acid mediated skin proliferation and it is highly expressed in epidermal keratinocytes (Bermudez et al. 2011). Activation of GPR109A results in its transport to the endosome where it remains trapped while associated with its agonist ligand (Li et al. 2010). The endosome is often acidified in neutrophils, which may lead to an acidic ligand diffusing out of the endosome. Activation of GPR109A by nicotinic acid and beta-hydroxy butyrate leads to suppression of inflammatory cytokine production in retinal cells suggesting that anti-inflammatory effects of its activation may be more general than in neutrophils alone (Gambhir et al. 2012). Many of these observations are thought to support activity in MS.
In respect of GPR109A, DMF is the prodrug for the active metabolite and GPR109A ligand MMF. DMF, being neutral, is able to diffuse through cell membranes, whilst MMF, being a free acid, is hindered in passive diffusion (Mrowietz et al. 2018). After oral administration of DMF, a large proportion is rapidly hydrolyzed to MMF (Linker and Gold 2013;Mills et al. 2018;Helwa et al. 2017;Miglio et al. 2015) and the rest to the DMF-GSH-conjugate. It is not clear whether this form of DMF is further active in vivo. The reaction is reversible, but its metabolite the mercapturic acid of DMF is detected in urine (Wollina 2011). These data suggest that the dominant circulating product of DMF after oral application in vivo is MMF (Mrowietz et al. 2018).
Based on these data, we hypothesized that if MMF were preferentially directed to myeloid cells like neutrophils it should exert a more potent anti-inflammatory effect (Fig. 1B). To test this concept, we synthesized a set of new compounds (1-3 listed in Fig. 1A) designed to deliver the active metabolite MMF to immune cells/inflammation sites. This was done by coupling MMF to the macrolide Azithromycin. Azithromycin is known to be able to accumulate rapidly in tissue and immune cells (Bosnar et al. 2005;Carryn et al. 2003;Togami et al. 2013). It is proposed that the reason for the accumulation is acid trapping to the lysosomes, facilitated by the two amphiphilic amines. Bosnar et al. (2005) showed specific uptake of Azithromycin into immune cells which we confirmed in a previous study demonstrating beneficial distribution patterns (spleen, liver and lung) for anti-inflammatory effect after oral application (Straß et al. 2021). Although some previous studies showed a possible effect of Azithromycin in psoriasis, our focus was on understanding the efficacy of fumarates. While Azithromycin has robust general anti-inflammatory properties, these are mainly reported at much higher systemic doses (e.g. 5-10 mg/kg) whereas we seek more potent effects in the range of 0.1 mg/kg per oral (p.o.) in murine models (Huang and Shieh 2016;Saxena and Dogra 2010;Balloy et al. 2014). As a general rule, our previous data suggest that lipophilic amphiphiles like Azithromycin can have anti-inflammatory effects on the gut from 1 µmol/kg and systemic effects from 10 to 100 µmol/kg, which are doses sufficient to inhibit the lysosomal phospholipases (Burnet et al. 2015).
Our novel set of compounds and data provide a new means to investigate the mode of action of fumarates. In particular, we set out to answer, whether their effect is more related to interactions with GSH or possibly with the endosomal/lysosomal receptor GPR109A (Li et al. 2010).

Synthesis
We synthesized three new fumarate macrolide compounds (1-3) and to avoid/control for non-specific effects of the carrier construct, we prepared a reference substance based on Azithromycin esterified with methyl succinate (4) to be used as a comparator to the active compounds in in vitro studies. This comparator substance (4) replicates many of the properties of the fumarate esters but does not possess the double bond that can react with GSH and is not reported to interact with GPR109A. All chemicals were purchased from commercial sources and used as received. Reaction monitoring was performed via mass spectrometry (Finnigan LCQ Deca XP MAX, Software Xcalibur 2.0.7 SP1) and thin layer chromatography (TLC; Merck TLC Silica gel 60 F254). TLC spots were detected with Hanessian's stain, based on a Cerium Molybdate solution and heat (through heat gun for approximately 30 s). Nuclear magnetic resonance (NMR) spectra were recorded with a Bruker Avance 400 (400 MHz) or Bruker Avance III (300 MHz). Substances were dissolved in CDCl 3 and chemical shifts (ppm) were referenced to CHCl 3 /Tetramethylsilane. Coupling constants (J) are given in Hz. After reaction steps solvents were evaporated with rotary evaporator (RV8 IKA, KNF SC 920) under vacuum (at, if not stated differently, 35-40 °C). To purify substances flash chromatography was performed (Interchim puriFlash 5.020 with Interchim PF-15SIHP-F0040 or PF-50SIHP-F0040 columns). Purity of reaction products was determined via high performance liquid chromatography (HPLC) Varian (ProStar) and evaporative light scattering (ELS) detection (Sedere Sedex 80). Mobile phases contained water (0.05% formic acid) and methanol (0.05% formic acid) as gradients. Stationary phase was ReproSil-Pur 120 C 18 -AQ, 5 µm, 75 × 3 mm (Dr. Maisch). High resolution mass spectra were recorded with a Bruker maXis 4G ESI-TOF from Daltonik [JL1], using ESI + mode with following settings: Capillary voltage 4.5 kV, source temperature 200 °C, gas flow 6 L/ min, nebulizer gas pressure 1.2 bar, end plate offset -0.5 kV and an m/z range of 100-1350.

Formation of Azi-Fum-GSH derivatives
Formation of glutathione fumarate adducts in mechanistic experiments were tested in situ and ex vivo. For in situ tests, compounds 1-3 were dissolved (diluted from 10 mM dimethyl sulfoxide (DMSO) stock to 10 µM final concentration) in water with 0.5% triethylamine. Glutathione (diluted from 100 mM DMSO stock to 100 µM final concentration) was added to the solution and reaction was monitored via MS. Formed adducts (see table 2 supporting information (SI)) were used to tune HPLC-MS/MS for screening assays. For ex vivo experiments whole blood was taken from tail vein of healthy male wistar rat. Compound 1-3 were added to rat blood (diluted from 10 mM DMSO stock to 10 µM final concentration) and incubated for 1 h at 37 °C and analyzed via MS/MS. Additionally, three C57BL/6 female mice were euthanized, livers explanted, immersed in ice-cold phosphate buffered saline pH 7.4 (PBS), minced and washed with the same solution. The mince was homogenized with Fastprep (FastPrep-24 5G, MP Biomedicals) using double amount of PBS to the liver. The homogenate was centrifuged at 700 × g for 10 min at 4 °C. The pellet was discarded and the supernatant was used as liver homogenate. 20 µM of each compound was added separately from 10 mM DMSO Stock (1 µL in 500 µL homogenate), vortexed and incubated at 37 °C. Samples were after 4 h and analyzed via MS/MS.

Quantification and monitoring of analytes using HPLC-MS/MS & MS/MS
Quantification of analytes was performed with an Agilent 1290 Infinity system coupled to a triple quadrupole Sciex API 4000 MS/MS detector. An Agilent C 18 Poroshell 120 column (4.6 × 50 mm, 2.7 µm) was used for separation. The mobile phase was composed of water containing 0.1% formic acid (eluent A) and acetonitrile containing 0.1% formic acid (eluent B). Gradient used was: 10% B for 1 min, to 100% B in 2 min, 100% B for 3 min, to 5% B in 1 min, 5% for 3 min. MS detection parameters can be found in SI table 2. Reaction monitoring and fragmentation of synthesis products and of possible adducts was performed via MS/MS.

In vitro assays
Human blood products used in the in vitro assays (used for cell stimulation, viability, stability and uptake assays) were obtained from the center for transfusion medicine in Tübingen, Germany (Zentrum für Klinische Transfusionmedizin Tübingen GmbH, (ethical approval number ZKT-FoPro202106-2305-01 and ZKT-FoPro202012-2211)).

Stability assay
1 mL of either, human whole blood or plasma was mixed with 1 µL of a 10 mM DMSO stock of tested compound and incubated at 37 °C. At indicated time points 50 µL of blood or plasma was transferred to 150 µL of acetonitrile. Mixture was vortexed, sonicated for 5 min and centrifuged for 7 min at 20.000 g at 4 °C. Supernatant was transferred to HPLC vials and analyzed as describe in the Quantification of analytes using HPLC-MS/MS & MS/MS section.

Uptake assay
Human buffy coat was prepared to contain 5 × 10 6 cells/ mL. 1 mL was mixed with 1 µL of a 10 mM DMSO stock of tested compounds. At time points 0 and 30 min sample was centrifuged for 5 min at 400 × g at 4 °C. 50 µL of supernatant was added to 150 µL of acetonitrile, remaining supernatant was discarded. Pellet was resuspended in 50 µL of ice cooled PBS and centrifuged again (5 min, 400 × g, 4 °C). Supernatant was discarded and pellet resuspended in another 50 µL of PBS. Cell suspension was transferred to 150 µL of acetonitrile. Mixture was vortexed, sonicated for 5 min and centrifuged for 7 min at 20.000 g at 4 °C. Supernatant was transferred to HPLC vials and analyzed as describe in the Quantification of analytes using HPLC-MS/ MS & MS/MS section.

Kinetic assay of the Michael addition to GSH
For kinetic measurements 100 mM stocks of compounds 1-4 were prepared in DMSO, and stocks of DMF, MMF and GSH in water. For each compound (1-4, DMF or MMF) 1 µL were mixed with 10 µL of GSH in 989 µL of buffered solution. Buffers were prepared at pH 5 (250 mM sodium acetate/acetic acid buffer), pH 7.3 (250 mM NaH 2 PO 4 / Na 2 HPO 4 ) and pH 8 (250 mM NaH 2 PO 4 /Na 2 HPO 4 ). Absorption was measured with NanoPhotometer NP80 Implen at 226 nm every 90 s for 15 min.

In vivo studies
All study protocols were approved by the local Animal Care and Ethics Committee (Federal government ethics committee, Tübingen, Germany under the licenses 35/9185.81-7/ SYN 06/20; 35/9185.81-7/SYN 07/18; and 35/9185.81-7/ SYN 11/19). Mice used in these studies (BALB/c and C57BL/6, all female and 8 weeks old) were purchased from Janvier Labs (Le Genest-Saint-Isle). Animals were acclimatized for at least 1 week with standard chow (Mouse Maintenance, V1534-000, Ssniff Spezialdiäten GmbH, Germany) and drinking water ad libitum before the start of the experiment. Animals were housed in type IV cages (4 animals per cage), with bedding and enrichment material and kept at 22 °C (± 1 °C), 45-65% humidity, with 12/12 h dark/light cycles period. Animals were monitored daily and at the end of the experiment, terminated painlessly with overflow of CO 2 .

Pharmacokinetic studies
For this study, 18 female BALB/c mice were utilized and each experimental group consisted of 3 animals. Animals were treated orally once with the testing compounds dissolved in 0.1% citric acid (12 µmol/kg). Tail blood was collected after 5, 15, 30, 45, 60, 90 min. Animals were euthanized with CO 2 . and organs collected 2 and 4 h after treatment. Collected samples stored at -25 °C until further processing, as published (Straß et al. 2021).

Cream-induced mouse model
In this study, 56 female BALB/c mice were examined and each experimental group consisted of 8 animals (4 animals per cage). On day 0, prior to psoriasis induction, dorsal skin was shaved using a clipper, followed by application of depilatory cream to complete fur removal. Psoriasis was induced by daily application of an induction cream. Briefly, the cream was formulated with IMQ (5%), DMSO (3.8%) and isostearic acid (25%; supplier TCI I0184), mixed with a spatula till homogenous, then hydroxypropyl methylcellulose (HPMC, 2%, dry) was added. At the end, base cream (Deutscher Arzneimittel Codex) DAC was added and homogenized. Cream was stored at 4 °C. From day 1 to day 6, 50 mg of induction cream was weighed and applied using a spreader to achieve homogeneity on 4 cm 2 of depilated skin. Treatments (compounds 1-3, DMF and vehicles) were administered orally once daily. For oral administration the vehicle was a mixture of 0.5% citric acid and 0.5% hydroxypropyl methylcellulose (HPMC, dry). Parameters measured were: daily body weight (BW), skin thickness, skin redness and scaling (0 = no change, 1 = marginal effects, 2 = moderate effects, 3 = strong effects, 4 = maximum). Animals were euthanized on day 7, 27 h after last exposure to cream and 4 h after last treatment. At termination, pictures of back skin were taken (ante mortem) and skin samples (for histology and for gene expression) were collected and the weights of spleen and liver were recorded.

Acute peritonitis
Acute peritonitis was induced with the intraperitoneal injection of MSU crystals, which were prepared following protocols published earlier (Roberge et al. 1994). The in-house prepared crystals were characterized by examination under microscope and in in vitro assays versus commercially available MSU crystals (Invivogen, data not shown). C57BL/6 mice (male, 8 weeks) were treated per oral with either compound 3 (n = 15) or PBS (n = 5) and were injected with MSU crystals (200 µL of a 15 mg/mL suspension of MSU crystals in sterile PBS) intraperitoneal15 min later. Animals were euthanized after 4 h. At termination heart plasma and peritoneal lavage (sampled with 1-2 mL of ice-cold PBS) were collected and samples analyzed via ELISA for cytokine concentrations of IL-1β, IL-6 and IL-10 following protocols described in section cytokine studies.

Real-time polymerase chain reaction (qRT-PCR)
Flash frozen skin samples were homogenized in lysis buffer using a standard fast prep procedure. Subsequently, RNA was isolated using the Qiagen RNeasy ® MiniKit. 3.6 µg of RNA was used for cDNA synthesis, using PerfeCTa DNase I (Quanta Bioscience) and 5 × PrimeScript RT Mastermix (Takara). Gene-specific primers used for qPCR are listed in Table 4 SI. The qPCR reaction was performed in duplicates with Blue S'Green qPCR Mix Sepa-rate ROX (Biozym) in 96-well plates. Amplification was carried out in the QuantStudio ® 3 qPCR system (ThermoFisher Scientific/ Quantstudio ™ Design & Analysis Software v.1.4.3). Cycle conditions were set as 95 °C for 2 min followed by 40 cycles of 95 °C for 5 s and 60 °C for 17 s; a melt-curve analysis was performed for each qPCR reaction. Primer efficiency was determined for each primer pair to ensure linear standard curve and high amplification efficiency. Threshold cycle (Ct) values were used for the calculation of the relative gene expression levels. For this, all values were first normalized to respective Ct values of the (hypoxanthine-guanine phosphoribosyltransferase) HPRT housekeeping gene. Then, normalized ΔCt values of the different treatment groups were compared to the ΔCt levels of the vehicle treated control group which was set to 1. Normalized expression ratios (2(−∆∆Ct)) are given for all treatment groups relative to the vehicle group. Ct values of all samples were normalized to the respective Ct values obtained for HPRT housekeeping gene, resulting in the ∆Ct value for each sample: ∆Ct (gene x) = Ct (gene x)-Ct (HPRT). To calculate the ∆∆Ct value, the ∆Ct average of the control condition (vehicle) was used. This analysis was also performed for the individual control condition samples to determine variance within the control group. Subsequently, ∆∆Ct values of each sample were transformed into normalized expression ratio which gives the fold increase (or decrease) of the target gene expression compared to the control condition (vehicle) and normalized to the reference gene (HPRT). Normalized expression ratio (fold change) = 2 −∆∆Ct . Data were plotted as scatter plots using GraphPad Prism 9.1.0.

Statistics
All experimental results were first tested for normal distribution using Shapiro-Wilk for normality. Statistical analysis between multiple groups was done using two-way ANOVA followed by Bonferroni correction. Kruskal-Wallis test was used for non-parametric data. Non-marked bars are considered as not significant by which statistical significance was considered at the level of p-value < 0.05. Calculations of e.g., half-life were made with GraphPad Prism 9.3.1.

Design of targeted fumarates-from synthesis to mechanistic insights
Structures and substitution patterns were confirmed by NMR (Straß et al. 2021). Specific shifts in 1 H spectra of compounds 1 to 4 at proposed positions 2′ and/or 4″ were observed. NMR spectra of given compounds can be found in supplemental information (SI). Additional, MS/MS experiments showed distinct fragmentation patterns of synthesized compounds (Table 1 SI). Purity was tested via HPLC-UV/ ELSD. A substance was declared pure at greater than 95%.

GSH conjugation
Based on previous studies from Schmidt et al. we incubated DMF, MMF and compounds 1-4 at different pH values (5, 7.3 and 8) with GSH (1/10 the fumarate concentration) and measured the absorbance at 226 nm (maximum of double bond) over time. At pH 8 (SI Fig. 1) compound 1 hydrolyzed and there was no change in absorbance. Compounds 2, 3 and DMF show similar initial kinetics of GSH conjugation, but 3 stabilizes earlier, suggesting that the reaction stops before the available double bonds are depleted. LC-MS/MS confirmed the conversion to conjugates (data not shown). MMF and compound 4 do not show any reaction to GSH in the given time. MMF is believed to not react or only over a long period (Schmidt et al. 2007). Compound 4 does not carry the necessary double bond and is therefore used as a negative control. At pH 5 (SI Fig. 1) hydrolysis is slowed down for all fumarate diesters providing more diester for the GSH conjugation to occur. At pH 5, DMF conjugates to GSH more slowly, than the macrolide fumarate esters. At the physiological pH 7.3 ( Fig. 2A), we observed an increased conjugation rate for the macrolide fumarate esters 1-3 compared to DMF. Compounds 1-3 react within seconds to minutes (T 1/2 of 1 25 s; 2 103 s and 3 153 s), whilst DMF has the longest half-life (T 1/2 = 17 min). Although, conversion for 1 can be detected, absorption at 226 nm is higher compared to compounds 2 and 3. This might be due to high hydrolysis rate and therefore conversion to MMF. Overall, these results lead to the conclusion that the new fumarate diesters have, compared to DMF, a faster conjugation rate at physiological pH towards possible Michael donors, like GSH.
Addition of GSH to our fumarate esters can be shown not only via change in absorbance of the double bond, but also as a stable Azithromycin-fumarate-GSH complex via MS (Fig. 2B). These peaks were not only found in situ, but also after incubation in rat whole blood and mouse liver homogenates (SI Fig. 2 to 7). Adducts of fumarates with hydrolysis of the macrolide core system were not identified. At higher pH values hydrolysis of the methyl ester at the fumarate side was observed by MS (SI Fig. 4). The GSH conjugates were relatively stable and could be detected over a longer time course.

Stability and uptake
Stability assays in human whole blood and plasma (SI Fig. 10) showed that the 2′ position hydrolyses almost immediately. Compounds carrying fumaric esters in this position (found in compounds 1 and 2) form their more stable metabolites very rapidly. 4″ esters (found in compounds 2 and 3) were slow to hydrolyze, with low levels of the free macrolide detected. However, the total amount of these substances decreases over time. This is due to the fact that stable GSH adducts form with this class. In general, it was observed that hydrolysis of the 2′ position is slower in whole blood compared to plasma. This is due to accumulation to acidic cell compartments and stabilizing effects of the lower pH value. The stabilization was already shown in previous publications by our group using SCFA esters of azithromycin (Straß et al. 2021).
In addition, we have performed cellular uptake experiments to show possible accumulation into immune cells. Human buffy coat was therefore incubated with compounds 1 to 3 at a concentration of 10 µM for a total of 30 min and compared to concentrations at t 0 . We measured total macrolidic concentration and found an intra-vs. extracellular factor of 14 × for compound 1, 6 × for compound 2 and 8 × for compound 3 (SI table 5). All compounds partition to immune cells and accumulate over time. Factors would be even higher, but compounds bound to GSH or other cysteine carrying peptides were not quantified.

Effects on inflammatory cytokines in vitro
Cytokine concentrations were measured 24 h after stimulation of PBMCs with LPS/INFy and treatment with DMF or compounds 1-4 at different concentrations (Fig. 2D). Experiments were performed according to protocols published by Lehmann et al. (Lehmann et al. 2007). The results indicate that the mechanism of DMF and the set of new compounds is likely to be similar. The negative control 4 (2′-succinate) and compound 1 (2′-fumarate) don't exert a major influence on the measured cytokines. Compound 1 hydrolyses to the inactive MMF too quickly to influence cytokine levels (Straß et al. 2021). Compounds 2 and 3 carry a fumaric ester in the 4″ position. This position hydrolyzes more slowly than the 2′ and seems to have major influence on cytokine production.
Results from PBMCs for 2, 3 and DMF show significant reduction in released cytokines IL-1β and TNFα, but not for IL-12. This reduction can be seen for 2 and 3 from 1 µM for IL-1β and 5 µM for TNFα. DMF reduces IL-1β and TNFα from 25 µM suggesting that DMF is 5 to 25-fold less active in vitro. The results for the control 4 suggest that this cytokine effect is not due to the macrolide itself.
To further investigate the possible underlying mechanism and assess the effects of inflammatory cytokines in cream induced psoriasis we stimulated human PBMCs, according to Hoyle et al. (2022) first with LPS (and co-treatment with substances) and in a second stimulus with IMQ. This stimulus induced a fast increase of proinflammatory cytokines in these cells. DMF and the compounds 2 and 3 (Fig. 2E) were able to significantly reduce the levels of IL-1β and TNFα. The reduction of IL-1β and TNFα can be seen in the range between 1 and 5 µM for compound 3 and 10 to 25 µM for compound 2 and DMF. The negative control compound 4 and the 2′ fumaric ester compound 1 were inactive in this assay.   In a next step tested the compounds' possible effects in the inflammasome pathway. Our set of compounds is similar to DMF and expected act in a similar way. DMF is known to play a role in the inflammasome response via binding to cysteine residues on gasdermin D (Humphries et al. 2020). NLRP3 inhibition has been shown to be more potent for DMF than MMF (Miglio et al. 2015;Hoyle et al. 2022). Inflammasome activation was done by priming PBMCs with LPS (and co-treatment with substances) and nigericin as second stimulus. This was done for compound 2 and 3 at 5 µM. Both substances reduced IL-1β levels significantly, with compound 2 being slightly more effective (Fig. 2F).
Taken together, compounds 2 and 3 showed anti-inflammatory effects in different stimulation settings and that they are typically at least an order of magnitude more potent than DMF in vitro.

Viability assay
The Viability assays (live/dead staining and MTT) performed in peripheral blood mononuclear cells (PBMCs), (Fig. 2C) under the same conditions as the cytokine assays (macrolides are solubility limited) suggest, that the observed cytotoxicity is not related to the cytokine effects. For compounds 2 and 3, the highest concentration (25 µM) induced apoptosis mainly in monocytes. These effects are known for DMF starting from 10 µM (Michell-Robinson et al. 2016). Our data (SI Fig. 8) shows similar effect for DMF, 2 and 3 at concentrations starting at 25 µM. However, due to high cell uptake, the effective concentration of the macrolides in these cells is much higher (see stability and uptake) (Bosnar et al. 2005;Straß et al. 2021;Burnet et al. 2018).
Given that fumaric esters (MMF as well as DMF) are capable of influencing cell metabolism via inhibition of glycolysis (modification of GAPDH) in the tricarboxylic acid (TCA) cycle (Kornberg et al. 2018;Angiari and O'Neill 2018;Park et al. 2019;Ocana et al. 2021) we assessed metabolism via the MTT assay. Compounds 2 and 3 (SI Fig. 9) reduced MTT more than DMF with a 20-fold difference in potency. At 5 µM, however, the difference was reduced to twofold difference in the observed effects. Compounds 1 and 4 did not show any influence on the metabolic activity.

Psoriasis mouse model
Compounds 1-3 were tested in a dose response setting in a cream-induced psoriasis model, to find an optimal dose range for each compound. All studies consisted of cream application on the skin for seven consecutive days to induce psoriasis. Doses were based on pharmacokinetic results ( Fig. 4), in vitro results (Fig. 2) and previous studies with similar compounds (e.g. (Straß et al. 2021;Burnet et al. 2018)). DMF was used in one study as positive control with a dose of 30 mg/kg based on reported activity in the literature (Lehmann et al. 2002;Chen et al. 2014;Liu et al. 2016b).
The 2′ monoester of MMF, compound 1, reduced score at all doses (0.03, 0.06, 0.12 and 0.24 mg/kg) (Fig. 3A). This substance hydrolyses relatively quickly, showing the lowest interaction with GSH in vitro. However, since the macrolidic substances partition to acidic compartments in vivo, they tend to be more stable than in cell-based assays. The 0.12 mg/kg dose was ideal for most parameters for this substance, with most prominent effects on erythema (redness) (SI Fig. 12). At termination, the induction cream group (labelled as IMQ) had an average score (consisting of three parameters redness, scaling and thickness of the skin) of 6.5, while the 0.12 mg/kg group was score 3.3. At a dose of 0.12 mg/kg the compound 1 tended to from protect body weight change. Loss of BW is an effect of IMQ and was maximal at day 3. The area under the score-time curve is also optimal at 0.12 mg/kg. Skin thickness was most reduced at doses of 0.06 and 0.12 mg/kg compared to induction cream. Spleen and liver sizes showed no significant differences between all groups (SI Fig. 11).
The 2′, 4′′ macrolide diester, compound 2, showed beneficial effects vs. the induction cream in terms of skin score and area under the curve at 0.03 and 0.06 mg/kg (about half the dose of the monoester 1) (Fig. 3B). This can be explained by the effectively two moles of fumarate per mole at the same dose in mg/kg for this compound. On the final day, the nontreated group had score 5 vs. score 4.3, 3.1, 2.9 and 4.4 for 0.015, 0.03, 0.06 and 0.12 mg/kg respectively. These data suggest, as seen in other substances, that effects on other immune functions (e.g. gut homeostasis) may develop as dose increases. Liver and spleen weights did not show significant differences (SI Fig. 11). Skin thickness on termination was improved in the 0.015 mg/kg and 0.03 mg/kg doses (SI Fig. 11). Compound 2 performed better than compound 1 in GSH conjugation in vitro. It shares the 2′ ester with compound 1 and the 4′′ ester with compound 3.
The 4′′ monoester, compound 3 was given at the same doses as compound 1 (0.03, 0.06, 0.12 and 0.24 mg/kg) so that the molar dose of MMF was the same (Fig. 3C). DMF was included as a comparison at 30 mg/kg-being between 125 and 1000 times higher than the doses used for Compound 3. DMF effects were slow in onset and first visible at day six, as were the effects of compound 3 at 0.24 mg/kg. At termination, total scores for induction cream (labelled as IMQ) were at 6.8, vehicle 5.8 and DMF was at 5.1. Compound 3 was 5.8, 5.1, 5.4 and 4.5 at 0.03, 0.06, 0.12 and 0.24 mg/kg respectively. Significantly lower area under the curve was only found for the positive control DMF at 30 mg/ kg (Fig. 3C). There was no difference in BW, skin thickness, liver or spleen weight.
Compound 3 differs from compounds 1 and 2 in that it is most GSH bound and that the 4′′ site is the most stable of the esters. Thus, it is most likely to react with GSH and the least likely to release free MMF. To investigate distribution and fate, a subgroup of mice received 10 mg/kg compound 3 by oral gavage, 2 h before termination. Analysis of blood, skin, spleen, lung, liver, colon, thymus (via HPLC-MS/MS) showed no major amounts of 3 (m/z = 861), its GSH adduct (m/z = 1168), nor, its major metabolite detected in stability assays, Azithromycin (m/z = 749) (SI table 3). Since elimination of macrolides is known to be slower and accumulation and distribution of similar molecules has been published already (Straß et al. 2021) the most likely possibility is that the compounds bind to other cysteine carrying molecules or proteins as reported for DMF (McGuire et al. 2016) and this may be relevant to 3.
Inflammatory processes in psoriatic skin lesions are initially driven by the proinflammatory cytokines IL-23 and IL-12. These cytokines stimulate Th17 cells to induce the release of IL-17, IL-22 and TNFα. Analysis of gene expression of induced back skin in our performed studies showed significant differences in most inflammatory markers (Fig. 4). Lowered TNFα levels compared to NTC animals were detected for treatment groups and one IMQ group. Overall, TNFα mRNA levels are low compared to NTC. Compound 1 even lowered TNFα significantly. The IL-17 subtypes IL-17A and F were upregulated (50-to-300-fold change) in skins of mice induced but not treated, whilst mice All studies were conducted in 6-10 weeks old female BALB/c mice n = 8 (only exception is IMQ control group for compound 2 (C) where n = 4). Animals treated with induction cream (50 mg) for seven consecutive days. Back skin was scored daily for parameters redness, thickness and scaling (scores each between 0 and 4, with 4 being most severe). Body weight was recorded daily. Induction cream consisted of 5% IMQ solved in 3.6% DMSO and 25% isostearic acid in DAC base cream. 0.5% citric acid and 0.5% HPMC in water served as vehicle. First graph in each row shows accumulated skin score of compounds 1-3, DMF, IMQ and p.o. vehicle scored over 7 days. Second graph in each row shows change of body weight of groups treated with compounds 1-3, DMF, IMQ or p.o. vehicle. Daily measured, in comparison to day 0. Third graph in each row shows total area under the curve for skin score of each treatment. P values ≤ 0.0001 are indicated with ****p ≤ 0.001 are indicated as ***p ≤ 0.01 are indicated as **p ≤ 0.05 are indicated as*. Groups were statistically analyzed against IMQ treatment Fig. 4 Cytokine induction measured in skin samples collected from 6-10 weeks old female BALB/c mice. Samples were collected on final day of studies (day 7) and analyzed via qPCR. Analyzed were best performing (in regards of total skin score; see Fig. 3) treatment groups vs. induction cream receiving groups from this study (IMQ) vs. not induced and not treated animals (NTC). A NTC vs. IMQ vs. compound 1 at 0.12 mg/kg. B NTC vs. IMQ vs. compound 2 at 0.06 mg/kg. C NTC vs. IMQ vs. compound 3 at 0.24 mg/kg vs. DMF at 30 mg/kg. P values ≤ 0.0001 are indicated with ****p ≤ 0.001 are indicated as ***p ≤ 0.01 are indicated as **p ≤ 0.05 are indicated as*. Groups were statistically analyzed against NTC. Kruskal-Wallis test with Dunn's comparison was used after Shapiro-Wilk test for normality. Data represented as means with error bars as SD (n = 8) treated with compound 1 and 2 (Fig. 4A, B) showed only minor changes in comparison to NTC mice. For compound 3, there were no major changes in cytokine expression level compared to IMQ induced reference group. DMF in contrast showed higher levels of IL-17A and F then IMQ and compound 3. Gene expression levels of IL-23A, important for initial activation of Th17 cells, is significantly upregulated in IMQ groups and DMF and compound 3 receiving groups.
Compound 1 and 2 showed to have beneficial effects for reduction of IL-23A (almost same levels as NTC). The same pattern is seen for IL-6 induction. Inflammasome activation is displayed by RNA levels of IL-1β and NLRP3. Both parameters are increased with induction cream in all studies. Reduction from these levels can be seen for compounds 1 and 2.

Pharmacokinetics
Pharmacokinetic studies aimed to assess the compounds distribution and hydrolysis effects further. Initial pharmacokinetics for compound 1 (Fig. 5A and SI Fig. 13) suggested that the macrolide esters of MMF had a high volume of distribution and adequate exposure to circulating immune cells as well as immune organs like the spleen and to skin. These properties, combined with the much lower dose should result in a more selective exposure to compartments relevant immune-mediated diseases, as psoriasis. Conclusions on the distribution of compounds 2 and 3 (Fig. 5B, C) can only be made to a limited extent, since a large proportion of the substance quantities were not detected. This is a further indication of binding to thiol-bearing groups in proteins. However, compounds 2 and 3 show sufficient concentrations (> 100 nM) in skin tissue compared with liver for some activity to be expected. The more metabolically active liver may hydrolyze the fumaric esters faster than the skin.

Acute peritonitis mouse model (MSU crystal-induced)
The MSU crystal induced peritonitis model is a murine model for human acute gout. We treated C57BL/6 mice (male, 8 weeks, 15 animals per group and 5 animals PBS vehicle) orally with 0.1 mg/kg Compound 1 and injected MSU crystals into the peritoneum 15 min after treatment. Animals were terminated 4 h post-MSU induction to determine cytokine concentration in plasma and peritoneal lavage. Animals treated with compound 3 (0.1 mg/kg p.o.) (Fig. 6) showed a significant reduction of the proinflammatory cytokine IL-6 in both, peritoneal lavage and heart plasma. However, IL-1β and IL-10 were not significantly reduced in peritoneal lavage and plasma by the treatment (SI Fig. 14).

Design of targeted fumarates-from synthesis to mode of action
Monomethyl fumarate (MMF) has risen to prominence as a treatment for MS where it is commercialized as Tecfidera, a DMF preparation in which DMF is considered the prodrug of MMF. DMF was commercialized in the past as a treatment for psoriasis in a mixed ester salt formulation. The mode of action of DMF/MMF is not fully understood, however, recent reports of its use in MS suggest that it tends to drive immune cell subsets towards anti-inflammatory response, notably Th1 and Th17 cells (Mills et al. 2018). It is thought to be proapoptotic, and this may explain the reduction of specific lymphocyte sub-sets in treatment. At the molecular level, these effects are proposed to be related to: (i) interactions with NF-kB signaling, (ii) binding to GPR109A, (iii) conjugation to glutathione, (iv) binding to cysteines in GAPDH and (v) interactions with the Nrf2 pathway.
As a therapeutical use in human, the typical dose of DMF is about 500 mg daily, which approximates to 7-8 mg/kg or 75 µmol/kg. The majority of DMF is converted to MMF on uptake and circulates as MMF in plasma. MMF appears not to be thiol reactive, however, DMF entering the gut mucosa is thiol reactive prior to hydrolysis of the first ester. If thiol reactivity was important to the mode of action in the autoimmune setting, more stable esters, such as longer chain alcohol groups, would be favored, which appears not to be the case.
Here, we have developed novel fumarate diesters to differentiate effects on immune cell activation and thiol conjugation. Our hypothesis was that GPR109A is the primary target and that the most relevant form is in immune cells. To test this hypothesis, we synthesized substances that concentrate MMF into the endosomes of immune cells where it could maintain GPR109A in its activated or bound form in the endosome. These compounds are approximately 300-900fold more potent than DMF in vivo on a molar basis in a murine psoriasis model. Given their high uptake into target cells, and their concentration in endosomes, a target located in the endosome could explain the potency gain exhibited by these substances.
However, the substances also exhibit much faster reaction with GSH than that of DMF (initial rates 10-200-fold faster) which implies that this mode of action should also be considered. GSH is required for oxidative burst reactions in myeloid cells (Brennan et al. 2015), and its depletion can limit activation reactions. Similarly, GSH is located in the endo-and lysosomes, where the conjugates concentrate (Xu et al. 2018). Nonetheless, the abundance of GSH, cysteines and protein thiols, is such that they can titrate low doses (e.g. 0.03 mg/kg) of substance. Furthermore, in both clinical and preclinical studies (Mrowietz et al. 2018;Dibbert et al. 2013;Linker and Haghikia 2016), MMF is the dominant product detected in plasma and MMF conjugates with GSH at much lower rate (see Fig. 1) (Schmidt et al. 2007). It is difficult to estimate to what extent DMF or the conjugate diesters may be bound to thiols in large molecules, however, comparing recovery of thiol binding and non-binding compounds, the difference suggests that 90% is reacting with thiols in large molecules. Thus, the balance of data suggests that the thiol binding aspect is most likely a secondary effect that is more relevant in vitro, where the total volume of medium relative to cells is higher (> 1000x) and where the total substance available can titrate a higher proportion of thiols. In areas where the substance is more abundant relative to cells, such as the gut, there may also be thiol binding effects and these may interfere with innate immune reactions. These in turn may impact innate control of gut bacteria in the epithelium and be the cause of tolerability issues in the gut.
The effects of DMF in vitro reported here reflect data reported by others. McGuire et al. (2016) showed reduction of cytokines in bone marrow-derived macrophages (BMDMs) after stimulation with LPS at DMF concentrations starting at 25 µM. Lehmann et al. (2007)  Cytokine concentrations of IL-6 in peritoneal lavage and heart plasma sampled from mice treated with PBS (vehicle and MSU group) or Compound 3 (0.1 mg/kg) and terminated 4 h after administration of MSU crystals. P values ≤ 0.0001 are indicated with ****p ≤ 0.001 are indicated as ***p ≤ 0.01 are indicated as **p ≤ 0.05 are indicated as*. Groups were statistically analyzed against MSU treatment. Kruskal-Wallis test with Dunn's comparison (for heart plasma) or Brown-Forsythe and Welch test with Dunnett's comparison (for lavage) was used after Shapiro-Wilk test for normality. Data represented as means with error bars as SD (n = 15) stimulated with phytohemagglutinin (PHA) with different fumaric esters, with best effects using diesters at concentrations higher than 30 µM. Our assays confirm these values for DMF. In contrast, the endosomal directed compounds (2 and 3) show efficacy from 1 µM on TNFα and IL-1β, suggesting at least a 20-fold increase in potency in vitro, when using substances with strong cellular uptake.
Striking in these studies was the relatively low ratio in the concentrations between effective and potentially toxic effects of fumarate esters. We monitored cell viability using the MTT assay which reports on mitochondrial metabolic activity. McGuire et al. (2016) showed a loss in cell viability over time at 50 µM DMF in BMDMs. Michell-Robinson et al. (2016) showed specific toxicity on monocytes and Sebök et al. (1994) showed cytotoxic effects on keratinocytes. In all these previous studies and ours (see Fig. 2) a margin between the lowest effective dose and lowest toxic dose of DMF was observed in the range of 2-4.
Taken together, the results from MTT assay (hence reduction of metabolic activity) and GSH binding (hence Michael addition to cysteines) might suggest a covalent binding to GAPDH and thus reducing its enzymatic activity by blocking the binding of NAD + as a mechanism of action. This could lead to a lower concentration of cytokines simply through the inhibition of cell activity and energy supply. While potentially valid in vitro, it strikes one as unlikely in vivo except under conditions of very high substance to cell ratio. As noted above, these conditions may be found in segments of the gut after oral administration, but are unlikely in systemic circulation.

New fumarates and their in vivo action
Dose response studies in the psoriasis model were conducted to estimate the relative potency increase generated via the macrolide esters. The rationale was that most DMF provided by the oral route is preferentially delivered to the gut where it appears to be associated with occasional signs of discomfort. We hypothesized that by using a carrier, we could deliver MMF to the immune cell compartment and, potentially more importantly, the endosome where GPR109A may be most amenable to interactions with its ligand and the stabilization of signaling.
A further argument for extensive dose response studies was the observation that the high potency of short chain fatty acid analogs can impact bacterial processing and thus, a secondary goal of our studies was to identify doses at which we could analyze effects on innate immune function such as gut homeostasis and bacterial killing (Straß et al. 2021).
Differences between in vitro and in vivo data are especially prominent for the novel conjugates. In vitro, conjugation to the 4″ position was most effective in reducing inflammatory cytokines, whilst in vivo, in the psoriasis models, the 2′-esters appeared to be more effective. The 4′′ conjugates react at least 10-times faster with glutathione while the 2′ esters readily hydrolyze to MMF. In vitro, MMF was essentially inactive, suggesting that either the free carboxy group impedes cellular uptake, or, that interaction with glutathione is the more important aspect of in vitro activity (see medium volume to cellular mass concept above). Given that compound 1 has significant uptake into cells, the issue of access to the cytoplasm appears less relevant for this compound. MMF released in cells from Compound 1 after uptake appears not to interact with GSH and may be subject to efflux. In vivo, Compound 1 had the more standard pharmacokinetics but exhibited steady loss of the MMF 2′ ester. It was also the most active in the psoriasis model. These data suggest that MMF is the more relevant agent in vivo.
If MMF is the effective agent in vivo, hydrolyzed MMF can then ligate to the target receptor GPR109A. The pharmacokinetics of Compound 1 show that it is preferentially taken up to immune cells and steadily hydrolyzed. From similar compounds, the site of uptake is the lysosomal compartment where GPR109A is located. The concentration at the site of action is consistent with the more potent effect of this substance in vivo and an interaction with GPR109A and not thiols.
In the murine model for human acute gout the MSU crystals act as putative ligands for CD14, TLR2, and TLR4 and can, within the peritoneum of mice, activate macrophages which release proinflammatory cytokines via the activation of NLRP3 inflammasome complex (Mariotte et al. 2020;Khameneh et al. 2017). Effects seen here on IL-6 may be useful given that it is one of the central cytokines involved in the acute phase of MSU induced gout (Guerne et al. 1989). This might be explained by the direct effect of fumaric diesters on inflammasome activity. That DMF is in this model more effective compared to MMF was published earlier (Miglio et al. 2015;Hoyle et al. 2022) and is consistent with our results in this model.

Conclusion
Here we investigated the mode of action of DMF/MMF using novel conjugates that differentiate thiol binding activities while targeting the immune cell compartment. Based on the observations of these conjugates, we propose a mode of action based on MMF binding to GPR109A in the lysosomal membrane. Effective doses in mice of 0.03-0.12 mg/kg correspond to human doses in the range of 1-2 mg (given the higher body weight, but lower metabolic rate). These very low amounts provide for both dramatic economies in terms of bulk ingredient and pill burden, but they also provide opportunities to avoid tolerance issues and adverse effects. More generally, the substances described here demonstrate that delivery to the lysosome can dramatically increase or modulate efficacy for substances where the target is lysosomal. Given that lysosomes are probably less than 1% of overall cell volume, targeting the compartment provides for efficiency and dose reductions. Finally, these data provide insight into the role of GSH in immune functions. In vitro data from this study suggests that titrating GSH with a Michael acceptor dampens inflammatory reactions in cultured cells and primary immune cells.