key: cord-0267659-n5laxjaa
authors: Gupta, Dhanu; Wiklander, Oscar P.B; Görgens, André; Conceição, Mariana; Corso, Giulia; Liang, Xiuming; Seow, Yiqi; Balsu, Sriram; Felldin, Ulrika; Bostancioglu, Beklem; Fiona Lee, Yi Xin; Hean, Justin; Mäger, Imre; Roberts, Thomas C.; Gustafsson, Manuela; Mohammad, Dara K; Sork, Helena; Bäcklund, Alexandra; Edvard Smith, C.I.; Wood, Matthew J.A.; Vandenbroucke, Roosmarijn; Nordin, Joel Z.; Andaloussi, Samir EL
title: Engineering of extracellular vesicles for display of protein biotherapeutics
date: 2020-06-15
journal: bioRxiv
DOI: 10.1101/2020.06.14.149823
sha: 792474c9d2290d9964a167f2ad0823ca4b5def93
doc_id: 267659
cord_uid: n5laxjaa

Extracellular vesicles (EVs) have recently emerged as a highly promising cell-free bio-therapeutics. While a range of engineering strategies have been developed to functionalize the EV surface, current approaches fail to address the limitations associated with endogenous surface display, pertaining to the heterogeneous display of commonly used EV-loading moieties among different EV subpopulations. Here we present a novel engineering platform to display multiple protein therapeutics simultaneously on the EV surface. As proof-of-concept, we screened multiple endogenous display strategies for decorating the EV surface with cytokine binding domains derived from tumor necrosis factor receptor 1 (TNFR1) and interleukin 6 signal transducer (IL6ST), which can act as decoys for the pro-inflammatory cytokines TNFα and IL6, respectively. Combining synthetic biology and systematic screening of loading moieties, resulted in a three-component system which increased the display and decoy activity of TNFR1 and IL6ST, respectively. Further, this system allowed for combinatorial functionalization of two different receptors on the same EV surface. These cytokine decoy EVs significantly ameliorated disease phenotypes in three different inflammatory mouse models for systemic inflammation, neuroinflammation, and intestinal inflammation. Importantly, significantly improved in vitro and in vivo efficacy of these engineered EVs was observed when compared directly to clinically approved biologics targeting the IL6 and TNFα pathways.

Extracellular vesicles (EVs) hold great potential as therapeutic agents with the ability to functionally deliver therapeutic cargos 1 . Our group and others have utilized the display of surface ligands to achieve targeted delivery of nucleic acid species in hard-to-reach tissues, such as the central nervous system (CNS) [2] [3] [4] [5] . While being a highly promising strategy, recent studies have highlighted the limitations associated with conventional endogenous surface display technologies, as they typically label only a fraction of the EV population and thus limit the targeting capabilities to a minor sub-set of EVs. Emerging evidence indicates that EVs have numerous subpopulations aside from the classical division into exosomes, microvesicles, and apoptotic bodies [6] [7] [8] [9] . This heterogeneity is critically important in EV engineering, especially when delivery of a therapeutic cargo is required in combination with a targeting ligand approach for successful therapy.

Here, we present a novel strategy to display different protein therapeutics simultaneously on the surface of EVs based on synthetic biology and a systematic screening of loading moieties.

As proof-of-concept, we targeted inhibition of IL6 and TNFα signaling pathway using an extracellular decoy strategy. Various studies have emphasized that both cytokines play a key role in stimulating inflammation and tissue damage 10, 11 . Hence, these pathways are correspondingly targeted by clinically used drugs, including blockers of TNF-receptor (TNFR) (Etanercept, Infliximab) and IL6 receptor (IL6R)(Tocilizumab), to alter the adaptive immune response in autoimmune and inflammatory diseases 12, 13 . The soluble TNFα homotrimers exert diverse biological functions, such as cell proliferation, differentiation, and apoptotic signaling, through binding to one of its two receptors, TNFR1 and TNFR2 14 . The cytokine IL6 has broad, pleiotropic biological activities and has been shown to exert both anti-inflammatory and proinflammatory signals in deregulated adaptive immune responses 15 . Studies have highlighted that the trans-signaling activation by IL6 complexed to soluble IL6R through IL6 signaling transducer (IL6ST), is linked to inflammation, whereas classical IL6 cis-signaling has been shown to be anti-inflammatory and involved in regenerative processes 12 . In this study, we thus aimed to express TNFR and IL6ST on EVs as a clinically relevant approach that enable us to assess the display of the therapeutic proteins on a functional level rather than the mere presence on EV surfaces. Furthermore, the therapeutic relevance of these two cytokines in various inflammatory diseases allowed us to investigate the potency of these receptor decoy systems in vivo. Here, a screen of multiple endogenous display strategies was conducted for the decoration of the EV surface with cytokine binding domains of TNFR1 and IL6ST, which can decoy the pro-inflammatory cytokines TNFα and IL6, respectively.

This approach allows us to display more than one receptor type simultaneously in multimeric form and subsequently enhance their inhibitory activity as compared to conventional therapeutics against the same cytokines. In addition, this platform elicits efficient anti- alphabody. This work shows great promise for developing engineered, combinatorial EVbased protein therapeutics, as the flexibility of this platform allows robust and efficient surface display of therapeutic proteins and potential targeting ligands.

To develop an efficient EV surface display technology, which can be adapted for targeting domains or therapeutic proteins, we designed numerous surface display designs using luminal EV proteins found to be highly enriched in EV proteomic data sets published by us and others [16] [17] [18] . As a proof-of-concept model for the display of therapeutic proteins on EVs, we fused these EV domains to the cytokine binding domains of either TNFR1 or to IL6ST, for decoy sequestration of TNFα or IL6/IL6R heterodimeric complexes respectively, and further engineered these receptors to be signaling incompetent. This enabled evaluation of various surface display designs in a semi-high throughput workflow by assessing the ability of engineered EVs to decoy their respective cytokines (see schematic illustration of decoy EVs in Figure 1A ). An array of genetic constructs were designed using different exosomal sorting proteins, or their respective domains annotated for EV sorting ( Figure 1B -E).

HEK293T cells were transiently transfected with plasmids encoding the different display constructs and engineered EVs were purified by ultracentrifugation and quantified by nanoparticle tracking analysis (NTA) (Supp. Figure 1A and 1B). The potency of the purified EVs was assessed using an in vitro reporter system for the respective cytokine, either by detecting luciferase activity driven by a NF-κB minimal promoter (TNFα, Figure 1B and Supp. Figure 1C ) or secreted alkaline phosphatase (SEAP) driven by a STAT3 minimal promoter (IL6/sIL6R, Figure 1C and Supp. Figure 1D ). We observed inhibitory activity of engineered decoy EVs with various designs in a dose-dependent manner ( Figure 1F , G). Constructs with inclusion of the N-terminal sorting domain derived from Syntenin (TNFR1-N term Syntenin and IL6ST-N term Syntenin), a protein implicated in sorting of protein cargo into EVs, significantly and reproducibly exhibited the best inhibitory activity for both IL6ST and TNFR1 signaling incompetent receptor constructs ( Figure 1D -G and Supp. Figure 2A) . Furthermore, the functionality of the cytokine decoy EVs was corroborated by quantitative assessment of EVs by western blot (WB) probing for the respective decoy receptor (TNFR1 or IL6ST) or the fused His-Tag on C terminus of each construct. Notably, TNFR1-N term Syntenin (Supp. Figure   6A -B).

To further increase efficiency, multimerization domains were introduced in different positions within the constructs to increase the number of decoy receptors per EV and to mimic the natural receptor state in situ 12, 13 . A trimerization domain 'Foldon' derived from T4 fibritin protein of T4 bacteriophage 20 was introduced to the lead TNFR1 design either in the extracellular-or intracellular region (Figure 2A ). The addition of a multimerization domain to the TNFR1-N term Syntenin construct further increased the efficiency of the decoy EVs to sequester TNFα ( Figure 2B -C). Similarly, we introduced a dimerization domain 'GCN4 L.Z', derived from yeast 21 , and a tetramerization domain 'Fragment X', derived from Phosphoprotein P of human metapneumovirus 22 , to the IL6ST-N term Syntenin construct in the intracellular domain ( Figure   2D ). Both designs showed a significant enhancement over their predecessors ( Figure To reduce the variability associated with cellular transfection and to further scale up the production of therapeutic EVs for in vivo applications, stable engineered HEK293T producer cells for production of IL6ST∆-LZ-NST-and TNFR1∆∆-FDN-NST-decoy EVs were established using lentiviral transduction. HEK293T IL6ST∆-LZ-NST decoy EVs inhibited IL6/IL6R induced trans-signaling with reduced STAT3 activation and TNFR1∆∆-FDN-NSTdecoy EVs could inhibit TNFα stimulated NF-kB activation in a dose dependent manner (Supp. Figure 7A and 7B). To further validate whether our EV engineering strategy is applicable to other cell types, we validated the approach in a more therapeutically relevant cell source, mesenchymal stem cells (MSC), which were engineered to stably produce the respective decoy EVs (Supp. Figure 7C ). MSC TNFR1∆∆-FDN-NST and IL6ST∆-LZ-NST EVs displayed a dose response to decoy the cytokines similar to what was observed using HEK293T cellderived decoy EVs ( Figure 3A -B). The engineered MSC-derived EVs were characterized using NTA, showing that the majority of the EVs were in the size range of exosomes with a peak of around 100 nm (Supp. Figure 7D ). Characterization by WB of isolated EVs from the respective cell source confirmed expression of both common EV markers ALIX and TSG101, absence of apoptotic body marker calnexin and presence of the respective decoy proteins ( Figure 3C and Supp. Fig 8) 24 . In addition, the presence of decoy receptors (TNFR1 or IL6ST) on EVs was confirmed by immunogold electron microscopy, using primary antibodies against the respective decoy receptor ( Figure 3D ).

To determine the impact of this engineering strategy on the EV surface proteomic profile, a multiplex bead-based flow cytometry assay was applied for simultaneous flowcytometric detection of 37 surface proteins on CD63/CD81/CD9 positive vesicles 25 ( Figure 3E ). MSCderived TNFR1∆∆-FDN-NST and IL6ST∆-LZ-NST decoy EVs exhibited a highly similar surface protein profile as compared to MSC Ctrl EVs for 37 different surface markers on tetraspanin positive vesicles ( Figure 3F and Supp. Figure 9A -C). Furthermore, to determine whether the decoy receptors are present on the tetraspanin positive subpopulation of EVs, we modified the bead-based assay by using decoy receptor-based detection instead of tetraspaninbased detection of the 37 different capture beads. Upon using hTNFR1 antibody as a detection antibody for assessing the 37 different antigens, CD63 and CD81 were observed to be the most enriched surface markers on TNFR1∆∆-FDN-NST positive vesicles, whereas Ctrl EVs and IL6ST∆-LZ-NST EVs were negative for all markers ( Figure 3G and Supp. Figure 9D -F). A similar trend was observed for IL6ST∆-LZ-NST EVs upon using mIL6ST antibody as a detection antibody for the capture beads ( Figure 3H and Supp. Figure 9G -I). These results clearly indicate that our engineering strategy allows for highly efficient engineering of EVs with negligible disruption of their endogenous surface protein profile. Producer cells are genetically modified to express cytokine receptors without the signalling domain fused to an EV sorting domain for efficient display of cytokine receptors on the surface of the secreted EVs (decoy EVs), which can decoy cytokines specifically. B) and C) Schematic illustrations of the evolution of cytokine receptors to facilitate EV surface display and assessment of various designs in a cytokine induced reporter cell system for high throughput screening of TNFR1 and IL6ST decoy EVs. D) and E) List of various TNFR1 or IL6ST sorting domain fusions assessed in the initial screen. F) Engineered decoy EVs displaying TNFR1 purified from HEK293T cells transfected with the constructs encoding the different display constructs (Figure1D) evaluated for TNFα decoy in an in vitro cell assay responsive to TNFα induced NF-κB activation. Data were normalized to control cells treated with TNFα only (5 ng/ml). G) Engineered EVs displaying IL6ST purified from HEK293T cells transfected with constructs encoding the different display constructs (Figure1G) evaluated for IL6/sIL6R decoy in an in vitro cell assay respondent to IL6/sIL6R induced STAT3 activation. Data were normalized to control cells treated with IL6/sIL6R (5 ng/ml). F, G, Error bars, s.d. (n = 3), **** P < 0.0001, *** P < 0.001, ** P < 0.01, statistical significance calculated by two-way ANOVA with Dunnett's post-test compared with response of Ctrl EVs at the respective dose.

Schematic illustration showing the evolution of TNFR1 design by addition of trimerization domains to enhance loading and binding efficiency of the EV-displayed decoy receptors. B) and C) Systematic comparison of various TNFR1 designs with multimerization domains. Engineered EVs displaying TNFR1 purified from HEK293T cells transfected with constructs encoding TNFR1 multimerization sorting domain fusion proteins (as listed in Figure 2A ) evaluated for TNFα decoy in an in vitro cell assay responsive to TNFα induced NF-κB activation. Data were normalized to control cells treated with TNFα only (5 ng/ml). D) Schematic illustration showing the evolution of IL6ST designs by addition of different multimerization domain to enhance loading and binding efficiency of displayed decoy receptors on EVs. E) and F) Engineered EVs displaying IL6ST purified from HEK293T cells transfected with constructs encoding IL6ST multimerization sorting domain fusion constructs (as listed in Figure 2D ) respectively evaluated for IL6/sIL6R decoy in an in vitro cell assay respondent to IL6/sIL6R induced STAT3 activation. Data were normalized to control cells treated with IL6/sIL6R (5 ng/ml). B, C, E, F, Error bars, s.d. (n = 3), **** P < 0.0001, *** P < 0.001, ** P < 0.01, * P < 0.05, statistical significance calculated by two-way ANOVA with Dunnett's post-test compared with response of Ctrl EVs at the respective dose.

Next, we sought to investigate the efficacy of engineered EVs and compare them to a clinically approved biologic against TNFα, Etanercept (a dimeric TNFR2 protein). Using the aforementioned TNFα reporter model, HEK293T TNFR1∆∆-FDN-NST decoy EVs showed a 10-fold lower IC50 value compared to Etanercept 13 ( Figure 4A This was further corroborated with MSC-derived TNFR1∆∆-FDN-NST decoy EVs (1×10 11 ), which showed improved survival (100% up to 60 hours) compared to 160 µg Etanercept (25% at 60 hours) and 1×10 11 Ctrl EVs (50% at 60 hours) ( Figure 4E ). In a separate experiment, mice treated with 6.5×10 11 MSC IL6ST∆-LZ-NST and/or 6.5×10 11 TNFR1∆∆-FDN-NST decoy EVs displayed reduced weight loss ( Figure 4F ). The protective effect in mice with LPS-induced inflammation was further improved by combinatorial treatment with both decoy EVs, as compared to 6.5×10 11 unmodified MSC-EVs. These results collectively underpin the therapeutic potential of decoy EVs and led us to continue assessing the therapeutic antiinflammatory potential of decoy EVs in an autoimmune MS disease model.

EVs displaying IL6ST purified from MSC cells stably expressing the optimised IL6ST∆-LZ-NST display construct, evaluated for IL6/sIL6R decoy in an in vitro cell assay respondent to IL6/sIL6R induced STAT3 activation. EVs purified from MSC stably expressing Ctrl construct were used as control. Data were normalized to control cells treated with IL6/sIL6R (5 ng/ml). B) Engineered EVs displaying TNFR1 purified from MSC cells stably expressing the optimized TNFR1∆∆-FDN-NST display construct, evaluated for TNFα decoy in an in vitro cell assay responsive to TNFα induced NF-κB activation. EVs purified from MSC stably expressing Ctrl construct were used as control.

Data were normalized to control cells treated with TNFα ( 

We have previously shown that EVs can be used to treat hard-to-reach tissues, including the ability to cross the blood-brain-barrier and exhibit therapeutic effects in the Central nervous system(CNS) 4 . To explore the potential effect of decoy EVs in neuroinflammation, the experimental autoimmune EAE mouse model, mimicking MS in humans, was used. To evaluate the effect of treatment in these mice, clinical scores on a scale of 0-5 are used that reflect the disease severity (EAE-score, Table 1 ). Upon repeat subcutaneous (SC) administration of control MSC EVs, as depicted in Figure 5A , no effect was observed on disease progression compared to mock treatment. However, as hypothesized, TNFR1∆∆-FDN-NST decoy EVs (4×10 10 ) Subcutaneous (SC) treatment significantly reduced disease progression over time ( Figure 5B ). At the endpoint (day 16), mice treated with decoy EVs could still move freely (Supp. Movie 1), with only minor tail and/or hind limb weakness, as compared to mock treated mice that were hind limb paralyzed. Additionally, significantly lower EAE-scores were observed in mice treated with TNFR1∆∆-FDN-NST decoy EVs (EAEscore 1.7/5) compared to control EVs (EAE score 3.0), which was similar to mock treatment (EAE score 2.9/5) ( Figure 5C ). In addition, mock treated EAE mice displayed a gradual decrease in body weight after symptom onset, reflecting disease progression (Supp. Figure   11A ). Mice treated with TNFR1∆∆-FDN-NST decoy EVs displayed sustained bodyweight over time, with increased weight at the endpoint (4.2%) compared to mock treated mice (Supp. Figure 11B ). In addition, treatment with TNFR1∆∆-FDN-NST decoy EVs reduced levels of the pro-inflammatory cytokines TNFα and IL6 in spinal cord compared to mock treatment (Supp. Figure 12A and 12B) .

In a similar set-up, depicted in Figure 5D , we next tested the therapeutic potential of blocking IL6 signaling in neuroinflammation, but instead of a sustained release route we opted for a rapid release route (IV, intravenous) for an immediate effect. Repeated injection of IL6ST∆-LZ-NST EVs (5×10 9 ) in mice induced with EAE until onset of symptoms, showed significant reduction in clinical score at day 16 (EAE score 1.33/5) as compared to mock treatment (EAE Figure 12C ). Importantly, administration of a single dose of IL23 decoy EVs (6×10 10 ) in EAE mice after onset of symptoms reduced the clinical score compared to mock treatment ( Figure 5I ). Taken together, these data clearly reflect the adaptability of the engineering platform and the potential of using EVs to display therapeutic receptors in inflammatory diseases including hard-to-treat CNS inflammation. 3,5,7,8,9 & 11) . G) Schematic description of treatment protocol for IL23 decoy EVs in EAE. H) Clinical score of disease progression over time and I) EAE-score at endpoint (day 16) in mice induced with EAE using MOG peptide and treated I.V with either 1×10 10 HEK293TIL23B-LZ-NST EVs pre symptomatic (n=5) (on day 5, 7 & 10), 6×10 10 HEK293TIL23B-LZ-NST EVs post symptomatic (n=5) (on day 13), or saline (n=6). C, F, I, Error bars, SEM *** P < 0.001, ** P < 0.01, * P < 0.05 statistical significance calculated by two-way ANOVA with Dunnett's post-test compared with response to mock treated animal.

After successful application of engineered EVs displaying single biologics, we next sought to generate combinatorially engineered EVs displaying two different surface proteins simultaneously. To this end, MSC cells stably expressing both TNFR1∆∆-FDN-NST and IL6ST∆-LZ-NST were generated (Supp. Figure 13A ). The optimized engineered EVs isolated from conditioned medium were characterized using NTA, showing that the majority of the EVs were in the size range of exosomes with a peak of around 100 nm (Supp. Figure   13B ). Characterization of isolated EVs from the respective cell source confirmed surface expression of both common EV markers ALIX and TSG101, absence of the apoptotic body marker calnexin, and co-expression of both decoy proteins by WB (Supp. Figure 14) . To validate the extent of engineering and to identify populations of EVs displaying both, or at least a single version, of the decoy receptors, single vesicle imaging flow cytometry 28 was performed after labelling of EVs with anti-TNFR and anti-IL6ST antibodies ( Figure 6A ). As expected, we observed a heterogenous pool of engineered EVs in our analysis, where 23% of the population of engineered EVs were found to carry both decoy receptors simultaneously, whereas 37% and 40% of engineered EVs were determined to carry either TNFR1 or IL6ST respectively ( Figure 6B and Supp. Figure 15A ). This was further validated by immuno-gold EM, where double positive EVs could be detected ( Figure 6C ). Furthermore, we observed a similar trend as with single engineered EVs in the multiplex bead-based flow cytometry assay, where the surface protein profile was similar to MSC Ctrl EVs, and the decoy receptors could be detected on several sub-populations ( Figure 6D and Supp. Figure 15 B-E). Decoy EVs purified from these double stable cells showed similar potency to decoy both cytokines in the in vitro cell assays as compared to their single decoy EV counterparts ( Figure 6E -F).

The unique ability of EVs to achieve body-wide distribution, including hard-to-reach tissues, makes them a versatile delivery vector. We and others have previously shown that upon systemic administration, EVs distribute to the gastro-intestinal tract and deliver therapeutic cargo 4, 29 . Therefore, in order to validate the therapeutic effect of double decoy EVs in inhibiting intestinal inflammation in vivo, we used a chemically induced (TNBS) mouse model, which mimics inflammatory bowls disease (IBD). A single injection of double decoy EVs 24 hours post symptom onset, showed a significant dose-dependent reduction of weight loss at 96 hours (-1.34% for the highest dose (3×10 11 ) and -3.2% for the lowest dose (3×10 10 )) compared to untreated mice (-6.8%) ( Figure 6E ). Importantly, the survival was also improved by double decoy EV treatment (92.3% survival) compared to untreated mice (66.7% survival) ( Figure   6F ). Furthermore, double decoy EVs outperformed equivalent doses (in terms of IC50 of mouse counterparts) of clinically used soluble TNFR1 protein (Etanercept) and anti-IL6R Figure 16A) . We also observed similar protective effect of double decoy EVs in a separate experiment with TNBS-induced colitis with improved survival (50% at 120 h) and weight change (+1.1%) compared to PBS treatment (20% survival, -1.3% weight) (Supp. Figure 16B 

With growing evidence on the critical role of EVs in a multitude of physiological processes, the potential of using EVs as a therapeutic modality has been increasingly explored. Our group and others have used various engineering strategies to achieve efficient loading of therapeutic cargos, such as nucleic acids and proteins, into EVs as well as to decorate them with various targeting ligands 1, 2, 5, [30] [31] [32] [33] . The approaches developed thus far rely on multi-modular engineering strategies, where a set of several proteins are used for achieving drug loading and imparting targeting moieties. Recent work by our group shows that the majority of these proteins, especially Lamp2b and CD63, fail to co-localize in the same vesicle subpopulation upon overexpression in producer cells 31 . Furthermore, Lamp2b, one of the most widely used EV engineering proteins for displaying targeting peptides, labeled only a fraction of EVs in our hands 31 . As a result of EV heterogeneity, there is a risk that for multi-modular strategies that the various modules (e.g. therapeutic and targeting) are distributed between mutually exclusive EV subpopulations, thereby negatively affecting the therapeutic efficacy of the EVs.

Here, we explored various EV engineering approaches for devising an efficient strategy to surface display therapeutic proteins. To assess the efficacy of the surface decoration and as a therapeutic application, we used the cytokine receptors TNFR1 or IL6ST, lacking their respective signaling domains. These engineered EVs were able inhibit TNFα or IL6/sIL6R complexes and hence decrease the activation of NF-κB and STAT3, respectively. This strategy allowed us to decoy pro-inflammatory cytokines in order to further treat inflammatory disorders models.

The focus of this study was to optimize the loading of the decoy receptors onto the surface of EVs by genetically modifying their producer cells. In the initial screenings, we observed efficient loading of decoy receptors onto EVs with the N-terminal fragment of Syntenin.

Syntenin is a cytoplasmic adaptor of Syndecan proteoglycans and aids in the interaction of Syndecan to ALIX, a key component of the ESCRT machinery, which induces membrane budding and abscission 34 . To further enhance the efficiency of the decoy EVs, we introduced oligomerization, dimerization and trimerization domains for IL6ST and TNFR1, respectively.

These were hypothesized to increase receptor decoy EV potency in two ways. First, oligomerization of exosomal sorting domains enhances the active shuttling of the cargo into EVs 35 and second, it mimics the natural state of the receptors during ligand binding on the cell surface 12, 13 . This addition of an oligomerization domain increased the inhibitory activity of TNFR1 decoy EVs by 10-fold and hence showed the importance of rigorous engineering to increase the efficacy of the EV drug, hence resulting into lower effective dosing of the treatment, thereby reducing manufacturing and production costs of this platform 36 .

To determine the therapeutic utility of these engineered EVs to suppress inflammation, we assessed the efficacy in three different inflammation models in vivo; LPS induced systemic inflammation, EAE, and TNBS induced colitis, which mimic sepsis, MS and IBD, respectively. 

For the TNFR1 display constructs, cDNA was amplified by PCR and cloned downstream of CMV promoter into a pEGFP-C1 vector backbone using NheI and BamHI. For IL6ST display constructs, codon optimized designs were synthesized (Gen9) and cloned downstream of CAG promoter into a pLEX vector backbone using EcoRI and NotI. The different constructs were assessed by transient transfection using branched polyethylenimine (PEI: total pDNA µg ratio 1.5:1). Next, the complete CDS of the different display constructs was cloned into the lentiviral p2CL9IPwo5 backbone downstream of the SFFV promoter using EcoRI and NotI, and upstream of an internal ribosomal entry site-Puromycin or Neomycin resistance cDNA cassette. All expression cassettes were confirmed by sequencing and the sequences are listed in Supplementary table1.

motion of nanometer-sized particles (Brownian motion) and commonly used for quantifying the concentration and size distribution of submicron-sized particles. For all our recordings, we used a camera level of 10-13 and automatic functions for all post-acquisition settings except for the detection threshold which we fixed at 6-7. Samples were diluted in PBS between 1:500 to 1:5,000 to achieve a particle count of between 2 × 10 8 and 2 × 10 9 per ml. The camera focus was adjusted to make the particles appear as sharp dots. Using the script control function, five 30 seconds videos for each sample were recorded, incorporating a sample advance and a 5 seconds delay between each recording.

Samples were treated with RIPA buffer and vortexed every 5 minutes for 30 minutes to lyse the EVs, subsequently the sample was spun at 12,000 g for 12 minutes to remove any lipids and the supernatant was collected. 30 μl of lysed sample was mixed with a sample buffer containing 0.5 M ditiothreitol (DTT), 0.4 M sodium carbonate (Na 2 CO 3 ), 8% SDS and 10% glycerol, and heated at 65 °C for 5 minutes. Samples were then loaded on a NuPAGE® Novex® 4-12% Bis-Tris Gel and ran at 120 V in MOPES running buffer (Invitrogen). The proteins on the gel were transferred to an iBlot nitrocellulose membrane (Invitrogen) for 7 minutes with the iBlot system. The membranes were blocked with Odyssey blocking buffer (LiCor) diluted 1:1 in PBS for 60 minutes at room temperature with gentle shaking.

After the blocking step, the membrane was incubated with freshly prepared primary antibody solution minutes each and visualized by scanning both 700 nm and 800 nm channels on the LI-COR Odyssey CLx infrared imaging system.

Purified TNFR1∆∆-FDN-NST EVs or double decoy EVs were incubated with 1 µl of 1% BSA diluted in PBS, for 5 minutes. 2 µl of primary antibodies (1 mg/ml, anti-hTNFR, Abcam, ab19139) were added and incubated for 45 minutes. For the immuno-gold labeling, 2 µl of protein A conjugated 10 nm gold nanoparticles (BBI Solutions) were added and incubated for 45 minutes.

Purified IL6ST∆-LZ-NST EVs or double decoy EVs were incubated with 1 µl of 1% Rabbit Serum (Sigma) diluted in PBS, for 5 minutes. 2 µl of primary antibody (0.2 mg/ml, anti-mGp130 from R&D, #AF468) were added and incubated for 45 minutes. For the immuno-gold labeling, 2 µl of rabbit anti-goat conjugated 5 nm gold nanoparticles (BBI Solutions) were added and incubated for 45 minutes.

Finally, 3 µl of labeled EVs were added onto glow-discharged formvar-carbon type B coated electron microscopy grids (Ted Pella Inc) for 3 minutes. The grid was dried with filter paper, washed twice with distilled water and blotted dry with filter paper. After the wash, the grid was stained with 2% uranyl acetate in double distilled H 2 O (Sigma) for 10 seconds and filter paper dried. The grid was air-dried and visualized on a transmission electron microscope (Tencai 10).

Surface expression of decoy constructs on engineered MSC lines was assessed by using either APC-conjugated rat-anti-mouse gp130 (IL6ST) antibodies (clone FAB4681A, R&D Systems) or AlexaFluor647-conjugated mouse-anti-human CD120a (TNFR1) antibodies (clone H398, Bio-Rad). DAPI was used for dead cell exclusion.

Multiplex bead-based flow cytometry analysis (MACSPlex Exosome Kit, human, Miltenyi Biotec) was implemented to characterize general surface protein composition of decoy EVs and specific surface proteins co-expressed on engineered decoy receptor EVs. Assays were performed based on an optimized protocol described previously 25 . In brief, EVs were used at an input dose of 1×10 9 NTA-based particles per assay, diluted with MACSPlex buffer to a total 

Re-Engineering Extracellular Vesicles as Smart Nanoscale Therapeutics

Delivery of SiRNA to the Mouse Brain by Systemic Injection of Targeted Exosomes

Exosome-Mediated Delivery of SiRNA in Vitro and in Vivo

Extracellular Vesicle in Vivo Biodistribution Is Determined by Cell Source

Vesicles

Systemic Exosomal SiRNA Delivery Reduced Alpha-Synuclein Aggregates in Brains of Transgenic Mice

Reassessment of Exosome Composition

Protein Composition Reflects Extracellular Vesicle Heterogeneity

Subpopulations of Extracellular Vesicles from Human Metastatic Melanoma Tissue Identified by Quantitative Proteomics after Optimized Isolation

Immunomodulation of Autoimmune Arthritis by Pro-Inflammatory Cytokines

Cytokines in Autoimmunity: Role in Induction, Regulation, and Treatment

Interleukin-6: Designing Specific Therapeutics for a Complex Cytokine

Pathogenic Mechanisms and Emerging Therapeutic Strategies

From Mediators of Cell Death and Inflammation to Therapeutic Giants -Past, Present and Future. Cytokine and Growth Factor Reviews

Interleukin-6 and Its Receptors: A Highly Regulated and Dynamic System

Exocarta as a Resource for Exosomal Research

Proteomic Profiling of NCI-60 Extracellular Vesicles Uncovers Common Protein Cargo and Cancer Type-Specific Biomarkers

Tumor Necrosis Factor (TNF) Receptor Shedding Controls Thresholds of Innate Immune Activation That Balance Opposing TNF Functions in Infectious and Inflammatory Diseases

the Natural Trimerization Domain of T4 Fibritin, Dissociates into a Monomeric A-State Form Containing a Stable β-Hairpin: Atomic Details of Trimer Dissociation and Local β-Hairpin Stability from Residual Dipolar Couplings

The GCN4 Basic Region Leucine Zipper Binds DNA as a Dimer of Uninterrupted α Helices: Crystal Structure of the Protein-DNA Complex

STRUCTURAL DESCRIPTION OF THE NIPAH VIRUS PHOSPHOPROTEIN AND ITS INTERACTION WITH STAT1

Immunosilencing a Highly Immunogenic Protein Trimerization Domain

Minimal Information for Studies of Extracellular Vesicles 2018 (MISEV2018): A Position Statement of the International Society for Extracellular Vesicles and Update of the MISEV2014 Guidelines

Systematic Methodological Evaluation of a Multiplex Bead-Based Flow Cytometry Assay for Detection of Extracellular Vesicle Surface Signatures

Structural Basis of IL23 Antagonism by an Alphabody Protein Scaffold

Role of IL-12/IL23 in the Pathogenesis of Multiple Sclerosis

Optimisation of Imaging Flow Cytometry for the Analysis of Single Extracellular Vesicles by Using Fluorescence-Tagged Vesicles as Biological Reference Material

Reduction of the Therapeutic Dose of Silencing RNA by Packaging It in Extracellular Vesicles via a Pre-MicroRNA Backbone

Exosomes Facilitate Therapeutic Targeting of Oncogenic KRAS in Pancreatic Cancer

Systematic Characterization of Extracellular Vesicle Sorting Domains and Quantification at the Single Molecule -Single Vesicle Level by Fluorescence Correlation Spectroscopy and Single Particle Imaging

Anchor Peptide Captures, Targets, and Loads Exosomes of Diverse Origins for Diagnostics and Therapy

Advances in Therapeutic Applications of Extracellular Vesicles

Higher-Order Oligomerization Targets Plasma Membrane Proteins and HIV Gag to Exosomes

Dynamic Biodistribution of Extracellular Vesicles in Vivo Using a Multimodal Imaging Reporter

Golden Exosomes Selectively Target Brain Pathologies in Neurodegenerative and Neurodevelopmental Disorders

Decoy Exosomes Provide Protection against Bacterial Toxins

Nef Neutralizes the Ability of Exosomes from CD4+ T Cells to Act as Decoys during HIV-1 Infection

Reproducible and Scalable Purification of Extracellular Vesicles Using Combined Bind-Elute and Size Exclusion Chromatography

Resolution Imaging Flow Cytometry Reveals Impact of Incubation Temperature on Labelling of Extracellular Vesicles with Antibodies. Cytom. Part

Amyloid β Oligomers Disrupt Blood-CSF Barrier Integrity by Activating Matrix Metalloproteinases

Induction of TNBS Colitis in Mice

cells (MSCs) were grown at 37°C, 5% CO 2 atmosphere. HEK293T cells were cultured in Dulbecco's Modified Eagle's Medium (DMEM) (Invitrogen), supplemented with 10% Fetal Bovine Serum (FBS) (Invitrogen)

MSCs were cultured in Roswell Park Memorial Institute (RPMI-1640) (Invitrogen) medium supplemented with 10% FBS, 10 -6 mol/l hydrocortisone, and 1% P/S. 48 hours prior to harvest of conditioned medium (CM) for EV isolation, the cells were washed with PBS and media was changed to OptiMem

RAW264.7 cells were grown in DMEM supplemented with 10% FBS and 1× Antibiotic-Antimycotic at 37°C in 5% CO 2

000 cells per well. The next day, cells were treated with 100 ng/ml of lipopolysaccharide (LPS) (L-5886, Sigma), in the presence or absence of EVs. The supernatant was collected 6 hours and 24 hours after treatment, and TNFα levels were evaluated by ELISA (BioLegend

HEK293T cells were co-transfected with p2CL9IPw5 plasmids containing CD63 fused to luminescent proteins, the helper plasmid pCD/NL-BH, and the human codon-optimized foamy virus envelope plasmid pcoPE using the transfection reagent JetPEI (Polyplus, Illkrich Cedex). induced with 10 mM sodium butyrate (Sigma-Aldrich) for 6 hours before fresh media was added to the cells, and the supernatant was collected 22 hours later. Viral particles were pelleted at 25,000 × g for 90 min at 4°C. The supernatant was discarded, and the pellet was resuspended in 2 ml of Iscove's Modified Dulbecco's Media supplemented with 20% FBS and 1% P/S. Aliquots were stored at −80°C until usage. To generate stable cell lines

Briefly, conditioned media (CM) was harvested and spun first at 500 g for 5 minutes to remove cells, followed by 2,000 g for 10 minutes to remove cell debris and thereafter filtrated through an 0.22 μm filter to remove any larger particles

5MM, Spectrum Laboratories) using a tangential flow filtration (TFF) system (KR2i TFF System, Spectrum Laboratories) at a flow rate of 100 ml/min

psi and shear rate at 3700 sec -1 ) and concentrated down to approx. 40-50 ml after diafiltration of PBS. The pre-concentrated CM was subsequently loaded onto BE-SEC columns (HiScreen Capto Core 700 column, GE Healthcare Life Sciences) and connected to an ÄKTAprime plus or ÄKTA Pure 25 chromatography system (GE Healthcare Life Sciences). Flow rate settings for column equilibration, sample loading and column cleaning in place (CIP) procedure were chosen according to the manufacturer's instructions. The EV sample was collected according to the 280 nm UV absorbance chromatogram and concentrated using an Amicon Ultra-15 10 kDa molecular

NTA is based on the captured EVs, either a mixture of APC-conjugated anti-CD9, anti-CD63 and anti-CD81 detection antibodies (supplied in the MACSPlex kit, 5 µl each) or anti-decoy receptor antibodies (AlexaFluor647-labelled anti-human TNFR1, Bio-Rad, cat #MCA1340A647, clone H398, lot 0410; or APC-labelled anti-mouse gp130, R&D Systems, cat #FAB4681A, clone 125623, lot AAOK0114071; 200 ng, respectively) were added to each well in a total volume of 135 µl and the plate was incubated at 450 g for 1 hours at room temperature. Next, the samples were washed twice, resuspended in MACSPlex buffer and analyzed by flow cytometry with a MACSQuant Analyzer 10 flow cytometer (Miltenyi Biotec). FlowJo software (version 10.6.2, FlowJo, LLC) was used to analyze flow cytometric data. Median fluorescence intensities (MFI) for all 39 capture bead subsets were background-corrected by subtracting respective MFI values from matched non-EV containing buffer controls

Single EV analysis by Imaging Flow Cytometry TNFR-decoy EVs and double decoy EVs were analyzed by single EV Imaging Flow Cytometry (IFCM) to confirm decoy receptor

Bio-Rad, cat #MCA1340PE, clone H398, lot 0407) and APC-labelled anti-mouse gp130 antibodies (R&D Systems, cat #FAB4681A, clone 125623, lot AAOK0114071; final concentration during staining 10 nM) at . Samples (and buffer controls without EVs, i.e. PBS + antibodies) were diluted 200

All analyses were performed by using the 60× objective and deactivated Remove Beads option. All lasers were set to maximum powers, and all data was acquired with a 7 μm core size and low flow rate (~0.38 μL/min). Data was recorded for 5 min and pre-gated on SSC (low) events as described previously. Dulbecco's PBS pH 7.4 (Gibco) was used as sheath fluid and for all dilution steps. Data was analyzed with optimized masking settings and by excluding coincidence events as described before using Amnis IDEAS software

Reverse Transcription-quantitative PCR (RT-qPCR)

Total RNA was isolated using TRIzol reagent (Invitrogen) and Aureum Total RNA Isolation Mini Kit (Bio-Rad), according to manufacturer's instructions. cDNA synthesis was performed using iScript cDNA synthesis kit (Bio-Rad Laboratories), according to manufacturer's instructions. RT-qPCR was performed with the Light Cycler 480 system

Expression levels in the spinal cord were normalized to the expression of the two most stable housekeeping genes, which were determined using geNorm 43 : ubiquitin-C (Ubc) and hypoxanthine-guanine phosphoribosyltransferase (Hprt)

Fw 5'-AGTGTTGGATACAGGCCAGAC-3', Rev 5

Il6 (Fw 5'-TAGTCCTTCCTACCCCAATTTCC-3', Rev 5

Fw 5'-ACCCTGGTATGAGCCCATATAC-3', Rev 5

Cxcl1 (Fw 5'-CTGGGATTCACCTCAAGAACATC-3', Rev 5

EVs were I.V injected via the tail vein subsequent to LPS induction and the animals

±5 g) female C57BL/6 mice were immunized by subcutaneous S.C injection of 100 µl of the MOG35-55-CFA emulsion subcutaneously, distributed to 3 different locations. I.P injections of 400 ng pertussis toxin were given on the day of and two days following immunization to induce disease. Mice were subsequently monitored for change in body weight and assessed using EAE-scoring, see Table 1. EVs were injected either S.C on day 7, 9 and 13 or given as single I.V injection

Trinitrobenzene sulfonic acid (TNBS) induced Colitis was induced as described previously 44

±5 g) female BALB/c mice were pre-sensitized with peritoneum skin application of 60

:1) mix per mouse. One week later, Colitis was induced by intra-rectal administration of 30 µl TNBS + 42.1 µl 95% ethanol + 27

Statistical analyses of the data were performed using Prism 6.0 (GraphPad Software Inc.) by one-way ANOVA or two-way ANOVA for all P-values. All results are expressed as mean ±SEM. All graphs were made in Prism 6

We would like to acknowledge Bavo Vanneste and Griet Van Imschoot for their indispensable assistance. AG is an International Society for Advancement of Cytometry Marylou Ingram Scholar 2019-2023.