Antibody-displaying extracellular vesicles for targeted cancer therapy

Abstract

Extracellular vesicles (EVs) function as natural delivery vectors and mediators of biological signals across tissues. Here, by leveraging these functionalities, we show that EVs decorated with an antibody-binding moiety specific for the fragment crystallizable (Fc) domain can be used as a modular delivery system for targeted cancer therapy. The Fc-EVs can be decorated with different types of immunoglobulin G antibody and thus be targeted to virtually any tissue of interest. Following optimization of the engineered EVs by screening Fc-binding and EV-sorting moieties, we show the targeting of EVs to cancer cells displaying the human epidermal receptor 2 or the programmed-death ligand 1, as well as lower tumour burden and extended survival of mice with subcutaneous melanoma tumours when systemically injected with EVs displaying an antibody for the programmed-death ligand 1 and loaded with the chemotherapeutic doxorubicin. EVs with Fc-binding domains may be adapted to display other Fc-fused proteins, bispecific antibodies and antibody–drug conjugates.

Main

One of the major improvements in cancer therapy during the past decades pertains to targeted therapies with antibodies, such as anti-HER2 (human epidermal receptor 2) treatment against breast cancer and, more recently, immunotherapy with immune checkpoint inhibition¹. One of the main targets is the programmed cell death protein 1 (PD-1) and its ligand PD-L1. Despite the overexpression of PD-L1 in various malignancies, only a subset of patients exhibit a durable response². Strategies to improve the response rate include combinational approaches with chemotherapies, multiple checkpoint inhibition and other immune stimulatory approaches. These strategies are, however, limited in the coordinated delivery of the disparate therapeutics to the target of interest. An interesting alternative approach is the use of extracellular vesicles (EVs), which are promising nanocarriers for drug delivery. EVs are a heterogeneous group of natural nanovesicles that are secreted by all cells. These nanovesicles range in size from 30 nm to 2,000 nm in diameter and can impact neighbouring cells or cells at a distance³. EVs contain lipids, proteins and nucleic acid species from the source cell and have the unique ability to convey these macromolecules via an advanced system of intercellular communication^4,5. Importantly, EVs benefit from the ability to cross biological barriers to reach distant organs^6,7 and can be engineered to display targeting moieties and loaded with a wide variety of therapeutic cargo molecules^8,9. In recent years, EVs have gained increasing attention, and there are currently numerous clinical trials being undertaken to evaluate the therapeutic potential of EVs⁴.

In this Article, we propose a highly modular technology for EV therapeutics. Using molecular engineering tools, we have developed EVs that can bind the fragment crystallizable (Fc) portion of antibodies, so that the variable regions are displayed for antigen recognition. The Fc-binding EVs (Fc-EVs) can be decorated with different types of antibody and thus be targeted to essentially any tissues of interest. Here, the Fc-EVs technology is designed as a targeted cancer therapy using tumour-specific therapeutic antibodies to guide the EVs to tumour cells and to deliver antitumoural drugs.

Results and discussion

Optimization of Fc-EVs

The purpose of the Fc-EV technology was to combine the success of monoclonal antibodies for tissue targeting and the immense therapeutic potential of EVs for drug delivery^4,10. Here, the aim was to develop a flexible system by displaying antibodies that can be interchanged depending on the intended target. To decorate the EVs with antibodies, an Fc-binding moiety was introduced to the EV surface. The EV-producing cells were engineered to express a fusion construct of an Fc-binding domain and an EV-sorting protein, to enable enrichment of the Fc domain on the EVs (Fig. 1). Given that there are many different available Fc-binding domains and various EV membrane-sorting domains, a systematic comparison of engineering strategies was conducted using similar strategies as recently reported by us⁹. The screening was conducted by producing fusion constructs combining each EV-sorting domain, which also expressed a fluorescent reporter, with each Fc-binding domain (Fig. 2a–c). The assessment was done by imaging flow cytometry (IFC), measuring the EV fluorescent reporter (mNG, mNeonGreen) and binding of fluorescently labelled (APC, allophycocyanin) immunoglobulin G (IgG) antibody (Supplementary Fig. 1a). The IFC used allowed detection of antibody binding to EVs at a single-vesicle level¹¹. The initial assessment was conducted on the EV-producing cells (Supplementary Fig. 1b). First, the expression of nine different EV-sorting domains, fused either C- or N-terminally to the reporter protein mNG, was assessed (Fig. 2a,b). The cells displayed the highest expression levels when the mNG was fused to the tetraspanins (cluster of differentiation 9 (CD9), CD63, CD81), annexin V or tumour necrosis factor receptor (TNFR) (Fig. 2b). The other EV-sorting domains showed lower mNG levels. Next, nine Fc binders (Fig. 2a) were chosen for the screen on cells to select the best candidates to bind antibodies using an APC⁺ IgG. Here, protein A and the derived z domains displayed higher antibody binding capacity compared to protein G and multidrug resistance protein 4, especially when fused to a tetraspanin or TNFR, whereas the other Fc displayed negligible binding (Supplementary Fig. 2).

Following the screen of cells, a comparison of engineered EVs with Fc binders was conducted. By comparing the conditioned media of the different EV-producing cells expressing the nine different EV-sorting domains, the TNFR and tetraspanins, especially CD63 C-terminal fusions, outcompeted the other EV-sorting domains in terms of mNG-positive (mNG⁺) events per millilitre, indicating a greater production of engineered EVs (Fig. 2c). The EVs were thus designed to display either protein A, z domains or 4z domains, by fusion with the tetraspanins or TNFR. All constructs generated engineered EVs with Fc binders, and the combination of CD63–z generated the highest level of expression (Fig. 2d). All EVs showed antibody binding capacity, with similar high binding efficiency among the tetraspanin constructs when using the z domain, but it was significantly lower using TNFR as an expression scaffold (Fig. 2d). A similar tendency was observed when using 4z or protein A as Fc binder. As CD63–z fusions also resulted in engineered EVs with the highest level of expression (Fig. 2e), the CD63–z domain combination was selected as the best candidate and was thus used in the subsequent experiments, henceforth denoted Fc-EVs.

Characterization of Fc-EVs and confirmation of specific antibody binding

There is an intrinsic complexity of characterizing EVs, and several complementary characterization methods are thus needed¹². When analysed by nanoparticle tracking analysis (NTA), the Fc-EVs displayed a peak size of approximately 100 nm, which is within the range of small EVs (Fig. 3a). The size of approximately 100 nm was further confirmed by immune electron microscopy, which showed binding of nanogold particle-labelled antibodies to Fc-EVs but not to the Fc-negative control EVs (ctrl-EVs) (Supplementary Fig. 3a). To further confirm the binding of antibody to Fc-EVs, size exclusion chromatography (SEC)^13,14,15 was conducted following incubation of mNG-labelled Fc-EVs or ctrl-EVs with APC⁺ IgG. The majority of mNG⁺ Fc-EVs and ctrl-EVs was observed in the expected fractions 3–6 as detected by mNG fluorescence (Fig. 3b, left y-axis). However, the APC antibodies co-localized only in the same fractions when incubated with Fc-EVs but not ctrl-EVs (Fig. 3b, right y-axis), confirming Fc-EV association with antibody. Western blot was implemented on the Fc-EVs, which confirmed the presence of the classical EV markers TSG101, CD63¹² and the CD63-fused luminescent reporter nano-luciferase (nLuc, used as Fc-EV reporter for in vivo testing below), which confirms successful engineering (Supplementary Fig. 3b). The Fc-EVs were subsequently analysed using bead-based multiplex flow cytometry^16,17. Following incubation of fluorescently labelled human Fc fragments with Fc-EVs, a clear shift in fluorescence intensity was observed for all beads, whereas this shift was not seen with ctrl-EVs or without EVs (Supplementary Fig. 3c), thus indicating efficient binding of Fc fragments and capture beads to Fc-EVs. Next, IFC was again used to further assess antibody binding to EVs. Fc-negative ctrl-EVs did not bind the APC⁺ IgG, whereas Fc-EVs showed a clear dose-dependent binding when incubated with increasing IgG amounts (Fig. 3c and Supplementary Fig. 3d,e). IFC was further used throughout the study to control for the antibody-binding capacity of the Fc-EVs. Next, the affinity of different IgG subtypes to Fc-EVs was assessed. In line with previous reports of protein A and z domain^18,19, human IgG1 (hIgG1) had the greatest affinity to Fc-EVs as assessed by incubating increasing doses of different IgG subtypes with the EVs (Fig. 3d). This is also an important consideration as most clinically approved therapeutic Abs are hIgG1 based²⁰.

In addition, to assess whether Fc-EVs would retain the displayed antibodies when exposed to blood, a stability assay was performed with human or mouse plasma. mNG⁺ Fc-EVs were incubated with APC-labelled IgG and then exposed to either phosphate-buffered saline (PBS) or human or mouse plasma, with no loss of APC labelling among mNG⁺ Fc-EVs as shown by unchanged double positivity (of mNG and APC) in IFC, at 1 to 30 min exposure (Supplementary Fig. 3f). Similarly, the double positivity was unaffected by additional incubation with different non-APC-labelled IgG subtypes (Supplementary Fig. 3g). If needed, alternative approaches to further functionalize the antibody display could have been considered, as previously explored by others utilizing recombinant phosphatidylserine binding of nanobodies²¹ or more recently by EV display of avidin that can be combined with biotinylated antibodies²². However, as the Fc-EVs showed robust antibody display, further functionalization was not required for the current study. Moreover, cellular uptake of APC⁺ IgG co-localized with mNG⁺ Fc-EVs in HeLa cells, whereas no antibody could be detected when cells were treated with naked IgG or in combination with mNG⁺ ctrl-EVs (Supplementary Fig. 3h). Furthermore, imaging by nanoimager²³, was utilized to further assess the association of Fc-EV and antibody at a detailed level. Fractions from the SEC of Fc-EVs with or without antibodies confirmed the association of green Fc-EVs and red antibody, rendering yellow particulates, (Fig. 3e,f). Finally, to enumerate the number of antibodies displayed per Fc-EV, single-particle profiler (SPP) was used. SPP is a recent method based on fluorescence fluctuations, which has been developed for biophysical profiling and measurement of nano-sized bioparticles²⁴. Using this technique, mNG⁺ Fc-EVs were shown to display an average of 105 APC-labelled hIgG per vesicle, whereas Fc-EVs did not display mouse IgG (≤1 Ab/EV), and ctrl-EVs with human or mouse IgG all presented negligible amounts of IgG with ≤1 Ab/EV (Fig. 3g–i and Supplementary Fig. 3i–k). Taken together, the different characterization techniques all showed efficient and stable binding of antibodies to Fc-EVs, which encouraged us to further investigate targeted delivery of Fc-EVs guided by antibodies in vitro and in vivo.

Antibody-guided Fc-EVs as a versatile targeting technology

To demonstrate the multifaceted potential of the Fc-EV technology, two different relevant therapeutic targets were assessed in vitro. HER2 is an oncogene that is amplified and used as a biomarker and therapeutic target in several cancer types²⁵. Here, the clinically used HER2 antibody trastuzumab was incubated with Fc-EVs and assessed in HER2 positive breast cancer cells (SKBR-3). A 339-fold increase of EV uptake, as measured by mean fluorescent intensity (MFI), was observed in the cells when the mNG⁺ Fc-EVs were guided by trastuzumab compared to Fc-EVs without antibody (Fig. 4a). Ctrl-EVs displayed no increased uptake with any of the antibodies, and the Fc-EV uptake was unaffected by control-Ab. As expected, pre-treatment of the SKBR-3 cells with naked trastuzumab significantly decreased the uptake of Fc-EVs displaying trastuzumab (Supplementary Fig. 4a). However, the decrease was quite modest, most probably due to the rapid recycling of HER2²⁶. Taken together, this shows that the system is highly specific and that the antibody indeed drives the enhanced uptake…