Design of PAwt-sANTXR-Ac2
Retargeting of PA to various cell surface receptors has previously been achieved by fusing a binding protein to the C-terminus of PA, and we have developed such strategy using DARPins [10]. Having fused an EpCAM-targeting DARPin (Ac2) with an affinity of 1.3 × 10−7 M [15] to the C-terminus of a mutated version of PA, ablating binding to its own receptors, capillary morphogenesis gene-2 (CMG2) and tumor endothelial marker-8 (TEM8) (Fig. 1a), we generated a highly efficient, cell-specific, retargeted delivery system. Even with low concentrations (20 nM) of the retargeting fusion construct PAm-Ac2, we could detect the cytosolic presence of cargo DARPins [10]. When increasing the concentration of PAm-Ac2, however, we observed that our delivery system was highly toxic for Flp-In 293-EpCAM-BirA cells stably overexpressing the targeted receptor, without any toxic cargo being delivered. Therefore, we performed an in-depth analysis of PAm-Ac2 to search for the possible cytotoxic mechanism and measures to overcome this.
When domain 4 of PAwt binds to the wild-type receptor, it forms a metal-ion-dependent structural bridge between domain 4 and the von Willebrand factor A (VWA) region of the anthrax toxin receptor (CMG2 or TEM8) (Fig. 1b). Especially two binding residues (N682, D683) within domain 4 are very important for PA binding [16]. Although receptor binding is mainly mediated by domain 4 of PA, parts of the VWA region also interact with domain 2. Binding to the 340-348 loop of PA prevents the rearrangement of the PA insertion loop and the contiguous 2β2 and 2β3 β-strands.
It has been shown that the prepore-to pore conversion of PAwt occurs at different pH, depending on it being incubated with or without its wild-type receptor [13, 17]. Using mutated PA (PAm), which is unable to bind its wild-type receptor, the stabilizing interactions between domain 2 and the VWA region are lost, which otherwise prevent the conformational change at neutral pH. Thus, merely fusing a retargeting molecule to PAm does not fully replicate the mechanism of PAwt, which limits the conformational changes to occur in late endosomes. Hence, we propose that the prepore-to-pore conversion of PAm-Ac2 can occur immediately upon oligomerization on the cellular surface, already at physiological pH, thus assembling an open pore allowing ions and other substances to freely pass in and out of the cell (Fig. 1c).
To prevent this premature prepore-to-pore conversion, we designed a domain-2/domain-4 interface-stabilized version of PA (Fig. 1d, e). To achieve this, we genetically fused the 19.5-kDa VWA domain of CMG2 (residues 40-217, C175A), which we termed sANTXR, to the C-terminus of PAwt. A long (G4S)5 linker between PAwt and sANTXR with an approximate length of 88 Å allows the correct orientation and functional interaction of the fusion partners. The covalent linker massively increases the local effective concentration of sANTXR, which in combination with the high affinity for the PA-binding domain is expected to effectively reduce off-target effects of PAwt binding to CMG2 or TEM8 on the cell surface [18]. This was deduced from the structure of the wild-type conformation of the PA prepore [13], PDB ID: 1TZN. C-terminally to the sANTXR receptor domain, we fused the EpCAM-targeting DARPin Ac2. We propose that the sANTXR domain impedes premature prepore-to-pore conversion by creating a very similar domain arrangement as in PAwt bound to its receptor CMG2. We thus expect that the pH where the prepore-to-pore conversion can occur shifts back to wild-type conditions (Fig. 1b–d), conditions that are present only in the (late) endosomes. The cytotoxicity of a premature prepore-to-pore conversion on the cell surface thus should get diminished.
To confirm that the stabilizing interaction is really due to the functional interaction of PA with the wild-type receptor domain, we designed a PA mutant construct, PAm-sANTXR-Ac2, with the mutations N682A and D683A (Additional file 1: Figure S1), which should prevent binding of PAm and sANTXR, thus having no stabilizing interaction. As another control, we also designed a variant with a very short linker between PAwt and the sANTXR domain, restraining the sANTXR domain to an orientation in which binding of PAwt to sANTXR is sterically prevented. Comparing these constructs, a functional dependency of the stabilizing interaction and prepore-to-pore conversion was tested.
PAwt-sANTXR-Ac2 reduces cytotoxicity and is dependent on functional interaction of PAwt with its wild-type receptor domain
We tested the cytotoxicity of our previously developed construct, PAm-Ac2, in comparison to the new construct PAwt-sANTXR-Ac2. Upon incubation of Flp-In 293-EpCAM-BirA cells, which have been made to stably overexpress EpCAM, with increasing concentrations of PAwt-sANTXR-Ac2, no change in cellular viability was observed up to 500 nM, the maximal concentration tested (Fig. 2a). Flp-In 293-EpCAM-BirA cells, when incubated with PAm-Ac2 (not containing the receptor domain fusion), however, showed a decrease in viability of ~ 50% already at 19 nM, and even down to only 10% viability at a concentration of 167 nM PAm-Ac2. PAm-sANTXR-Ac2 (which comprises the mutated, non-interacting domains) showed a similar reduction of viability for concentrations ≥ 56 nM, confirming the necessity of a functional interaction between the VWA domain and PAwt. PAm-sANTXR-Ac2 appears to be less toxic than PAm-Ac2, presumably due to steric hindrance of the slightly larger fusion construct, impeding pore formation. We observed in time-lapse imaging video microscopy that the cytotoxicity with this construct occurs at a later timepoint, as described below (Additional file 1: Figure S2, showing the analysis of the videos of Additional files 2 and 3). To ensure that the toxicity was not due to the mutations associated with PAm, we used PAwt-Ac2 as a further control, which, in addition to binding CMG2, will bind EpCAM via Ac2. We expected a comparable toxicity of PAwt-Ac2 and PAm-Ac2 on Flp-In 293-EpCAM-BirA, since binding will be mostly via the highly overexpressed EpCAM without prepore stabilization, and only to a limited extent via CMG2 and TEM8. Indeed, PAwt-Ac2 shows a similar toxicity as PAm-Ac2. A non-targeted control, without the EpCAM binding DARPin Ac2, had no effect on the cells.
To confirm the receptor-specific cytotoxicity of the PA prepore, we incubated Flp-In 293-EpCAM-BirA cells with 100 nM PAm-Ac2, which showed clear toxic effects (Fig. 2a), and titrated the DARPin Ac2 (Ac2-FLAG) as a binding competitor. With increasing concentrations of competitor, the cytotoxicity was reduced, and with a ~ 3-fold excess of Ac2 DARPin over PAm-Ac2, 100% viability was restored, indicating the cytotoxicity is due to the interaction of PAm-Ac2 with EpCAM and not due to a non-specific cytotoxic effect (Fig. 2b).
In addition to the cell proliferation assay, we performed time-lapse imaging over 18 h. Flp-In 293-EpCAM-BirA cells were treated with 100 nM PAwt-sANTXR-Ac2 or PAm-Ac2, propidium iodide (PI), a marker of cell death, and eGFP fused to the C-terminus of LFN, LFN-eGFP. Cells were imaged over time with an automated LionHeart FX microscope. We measured the increase in PI staining for PAwt-sANTXR-Ac2 and PAm-Ac2 (Fig. 2c and Additional files 4 and 5). Up to 250 cells are PI positive in wells incubated with PAm-Ac2 in a time-dependent manner, while PAwt-sANTXR-Ac2 remained constant at the initial number of ~ 50 PI-positive cells. The lag in response time immediately after addition of PA variants can be attributed to the binding and pore formation on the cell surface, as well as the tolerance of cells to a certain number of pores formed on the plasma membrane (Additional file 1: Figure S3). We also confirmed cell death by PI staining with confocal microscopy. Cells were treated with 100 nM of the respective constructs and incubated for 3 h before confocal imaging. PAwt-sANTXR-Ac2 shows no cytotoxicity and is thus indistinguishable from untreated control cells, while cells treated with PAm-Ac2 detach and stain highly positive for PI (Fig. 2d).
With the control PAm-sANTXR-Ac2 (without functional interface between these components), we observed a slight delay in cytotoxicity in initial time-lapse imaging compared to PAm-Ac2 (Additional file 1: Figure S2). We propose that the slightly larger receptor fusion construct PAm-sANTXR-Ac2 sterically hinders rapid prepore-to-pore conversion on the cell surface.
To further investigate the structure-function relationship, we designed a construct with a very short linker (SL) between PAwt and the wild-type receptor domain, preventing the correct orientation and binding of PAwt to the VWA domain. With this construct, PAwt-SL-sANTXR-Ac2 (Additional file 1: Figure S1), we performed a viability assay and could observe a reduced cell viability to 63% at 580 nM (Additional file 1: Figure S4a). The higher concentrations where a cytotoxic effect is observed compared to PAm-Ac2 could have a similar cause as PAm-sANTXR-Ac2: steric hindrance with respect to form functional intramolecular complexes. To test this hypothesis, we performed a delivery assay to see if it would be still capable of prepore assembly, prepore-to-pore conversion, and delivery (see next section) as discussed below. Even though PAwt-SL-sANTXR-Ac2 was provided as a fusion with N-terminal His6-MBP, we want to point out that His6-MBP will be cleaved off by furin and the fusion construct, His6-MBP-PAwt-SL-sANTXR-Ac2, has previously been shown to demonstrate equivalent delivery to PAm-Ac2 [19].
PAwt-sANTXR-Ac2 reduces cytotoxicity in a receptor expression level-dependent manner
In order to understand to what extent the cytotoxic effects of premature prepore-to-pore conversion is a function of the receptor expression level, we tested our constructs on a panel of EpCAM-positive cells, having different levels of receptor expression: HT29, MCF7, SKBR3, with EpCAM-negative RD cells as control.
First, we assessed the EpCAM expression levels via flow cytometry using an Alexa Fluor 488-labeled anti-EpCAM mouse mAb (Fig. 3a, Additional file 1: Figure S5). EpCAM has the highest expression levels in the constructed Flp-In 293-EpCAM-BirA cells stably expressing EpCAM, followed by HT29, MCF7, SKBR3, and the EpCAM-negative RD cell line with no detectable surface EpCAM. Since Chernyavska et al. [20] recently estimated EpCAM levels of MCF7 cells at about 5.3 × 105 receptors/cell, we can assume that levels of the high-expressing Flp-In 293-EpCAM-BirA cells are around 2 million receptors/cell, even though these numbers have considerable uncertainty.
We then assessed whether the receptor expression level correlates with the oligomerization and prepore formation of PAwt-sANTXR-Ac2. It is possible to visualize PA oligomers by saturating available binding sites with LFN-eGFP, which is not transported (Additional file 1: Figure S6). Using confocal microscopy, we found that a higher receptor density resulted in more prepore formation, reflecting successful PA oligomerization (Fig. 3b). The signal was highest for Flp-In 293-EpCAM-BirA cells, followed by HT29 cells. For MCF7 and especially SKBR3, however, very little signal can be detected, although the receptor expression levels are in similar ranges as for the HT29 cell line. No signal for RD cells was observed, the EpCAM-negative control cell line. For cells expressing EpCAM, we detected a membrane-like staining pattern when incubated with LFN-eGFP and PAwt-sANTXR-Ac2. For Flp-In 293-EpCAM-BirA cells and HT29 cells, we further detected a dotted staining within cellular compartments, showing endo-/lysosomal localization. Endosomal entrapment of LFN-eGFP has been confirmed with the BirA assay (Additional file 1: Figure S6). The detection of an endosomal-like staining for LFN-eGFP in MCF7 and SKBR3 cells is not evident due to the detection threshold of the microscope in combination with the limited numbers of receptors.
We propose that the non-linear dependency of PA prepore formation on receptor density is due to a receptor-level threshold below which pore formation becomes less efficient. Additionally, varying mobilities of the receptors or different internalization and degradation rates of EpCAM in the different cell lines as well as different efficiency of furin activation may also contribute to these differences [21].
We then performed a viability assay with the panel of cell lines with PAwt-sANTXR-Ac2, PAm-sANTXR-Ac2, PAm-Ac2, PAwt-Ac2, and the non-targeted control, PAwt (Fig. 3c). A reduced cell viability can be observed for HT29 cells (Fig. 3c) with concentrations of 167 nM of PAwt-Ac2 and 500 nM of PAm-Ac2, leading to a viability of 46% and 33%, respectively. For MCF7, SKBR3, and RD cells, no cytotoxicity could be observed, which is in agreement with the lower expression levels of the receptor and it correlates to the expected lower levels of prepore formation on these cells.
Lower toxicity of PAwt-sANTXR-Ac2 enables greater cytosolic protein delivery
Previously, we have shown that PAm-Ac2 can efficiently deliver various cargoes to the cytosol of Flp-In 293-EpCAM-BirA cells stably overexpressing EpCAM [10]. Our goal in this study was to increase the amount of cytosolically delivered cargo molecules, which previously was not possible, since concentrations higher than 20 nM of the pore-forming PAm-Ac2 drastically reduced cellular viability even within the short 4-h incubation time (Additional file 2). Our newly designed, prepore-stabilizing PAwt-sANTXR-Ac2 was therefore next tested for efficient protein delivery with the biotin ligase assay [22].
We incubated Flp-In 293-EpCAM-BirA cells with PAwt-sANTXR-Ac2, PAm-sANTXR-Ac2, and PAm-Ac2 for 4 h in the presence of the proteasome inhibitor MG-132. MG-132 was included to assess the delivery systems independently of proteasomal degradation. As cargo proteins, we tested three different DARPins, varying in size and thermostability, which have previously shown to be effectively translocated [10]. These cargo molecules contain the biotin-acceptor avi-tag and an HA-tag at their C-terminus and are fused with their N-terminus to LFN. Cytosolically localized cargo proteins are biotinylated by a cytoplasmically encoded BirA of Flp-In 293-EpCAM-BirA cells [22]. Cargo molecules which are trapped within the endosome, not reaching the cytosol, are not biotinylated. The HA-tag is used to determine total cellular uptake, located in the cytosol and in any other cellular compartment, allowing the determination of the intracellular localization of a cargo molecule. After cell harvest and western blotting, biotinylated cargoes were detected with streptavidin IRDye 680LT and total cellular uptake was measured via an HA-tag antibody [22]. For quantification of cytosolically present cargo molecules, the protein(s) detected at around 70 kDa, which we hypothesized earlier to be endogenous heat shock protein 70 (HSP70), were chosen as a loading control [10].
With increasing concentrations of PAwt-sANTXR-Ac2, total cellular uptake (Fig. 4b, d) and cytosolic delivery (Fig. 4a, c) of the smallest DARPin NI1C increase. An increase in cytosolically present cargo can be seen up to an external concentration of 200 nM. Further increases in the concentration of PAwt-sANTXR-Ac2 did not yield higher amounts of delivered DARPin (Fig. 4c, d), presumably due to a saturation of the receptors exploited for delivery.
At 20 nM, similar delivery efficiencies can be observed for PAwt-sANTXR-Ac2, PAm-sANTXR-Ac2, and PAm-Ac2, but an increase to 100 nM does not lead to an increase in cytosolically present cargo for PAm-sANTXR-Ac2 and PAm-Ac2, as it does for PAwt-sANTXR-Ac2, likely due to the premature prepore-to-pore conversion of PAm-Ac2 and PAm-sANTXR-Ac2 on the cell surface. This lack of functional pores renders the cells unable to unfold and translocate LFN-cargo proteins (Fig. 4c, d). Slightly higher total cellular uptake of LFN-cargo can be observed with PAm-sANTXR-Ac2 than for PAm-Ac2, probably due to the delayed cytotoxicity compared to PAm-Ac2.
Similar results have been observed for LFN-NI2Cdest. as well as LFN-NI3Cdest., constructs that have been slightly destabilized to facilitate their unfolding and refolding during transport through the pore [10, 19] (Fig. 4e, f). With increasing concentrations (20 nM and 100 nM), an increase in cytosolic cargo delivery can be observed.
The BirA assay for His6-MBP-PAwt-SL-sANTXR-Ac2 (Additional file 1: Figure S4b) showed a reduced amount of total cellular uptake, suggesting a steric inhibition effect already at the start of the internalization process. The results for this construct are in line with the results for PAm-sANTXR-Ac2 and confirm the functional dependency of PA on interactions with the sANTXR domain.