Regulation of protein secretion through chemical regulation of endoplasmic reticulum retention signal cleavage

Design of membER and lumER secretion system

Two types of ER-localized and proteolysis-based protein secretion systems were designed, each featuring a different C-terminal retention signal; the first type, termed membER, was designed for membrane-bound proteins retained in the ER by the addition of a transmembrane domain and a C-terminal KKXX sequence exposed to the cytosol (Fig. 1a). This system is modeled after naturally occurring membrane-bound proteins localized in the ER. Based on previous reports, we opted to use the Lys-Lys-Met-Pro (KKMP) and Lys-Lys-Tyr-Leu (KKYL) sequences in the membER system, which were shown to be particularly potent Golgi- to-ER retrieval signals^18,19. A transmembrane domain, from the B-cell antigen receptor complex-associated protein beta chain (CD79B) was used to anchor the protein of interest to the ER membrane. A potyviral protease cleavage site was inserted between the transmembrane domain and the ER retention signal, allowing for the control of secretion by the removal of the retention signal through proteolytic cleavage by a cytosolic protease.

The second type of an inducible protein secretion system, termed lumER, was designed for soluble proteins retained inside the ER lumen through the addition of a KDEL retention signal at their C-terminus (Fig. 1a). Binding of the KDEL sequence by the KDELR (and thus recycling the POI back to the ER) is pH sensitive, with maximal binding at pH below 6²⁰. The acidic pH 6.2 inside the cis-Golgi lumen initiates binding, while the more neutral pH 7.2–7.4 inside ER lumen^21,22 releases the interaction between KDEL and KDELR. KDELR itself features a KKXX retention signal at its cytosol-facing C-terminus, which interacts with the COPI vesicle machinery, responsible for the retro-transport of COPI vesicles from cis-Golgi to the ER²³.

Both membER and lumER constructs were equipped with a signal peptide from the human immunoglobulin kappa²⁴ at the N-terminus of the secretion constructs to ensure translocation of the protein into the ER during translation (sequences in Supplementary Tables 1–3). However, because the two retention systems work through different pathways, each of them required additional unique design considerations. For the membER system, removal of the retention signal enables the membrane-bound protein to bypass the cis-Golgi retro-transport, but once the protein passes through the GA, the protein remains anchored into the cell membrane due to the transmembrane domain. To achieve the release of the POI from the plasma membrane once it had passed the GA and before reaching the plasma membrane, an additional furin protease (FURp) cleavage site (amino acid motif R-X-X-R²⁵) was inserted in the ER lumen-facing side of the target construct between the POI and the transmembrane domain. Since FURp is enriched in the GA²⁶, it cleaves the POI off the membrane, enabling its release upon the fusion of the secretory vesicles with the plasma membrane^{27,28,29,30,31}. The proteins presenting solvent-exposed R-X-X-R motif would be therefore cleaved on the way towards the plasma membrane.

On the other hand, the lumER secretion design relies on retention in the ER by the KDELR, which recognizes and binds the ER lumen-exposed C-terminal KDEL signal. For this reason, the lumER constructs do not need to be anchored to the ER-membrane for them to be retained in the ER and thus only require the N-terminal signal peptide and the C-terminal KDEL sequence, separated from the POI by a selected protease cleavage site (Abstract figure, Fig. 1a). However, the proteolytic processing in the lumER system occurs inside the ER lumen, thus requiring a protease which retains its location and activity within the secretory pathway. Additionally, this protease would need to be equipped with a signal sequence, for the transport into the ER, and a retention signal to halt its progression to the cell surface. We found a previously described ER-localized TEVp, termed erTEVp, appropriate for the execution of the proteolytic processing inside the ER lumen³² that could be regulated by a chemical signal. To expand the repertoire of functional proteases used in our secretion systems, we replaced the TEVs in the membER system with a cleavage site of other orthogonal viral proteases from the Potyviridae family of viruses, namely plum pox virus protease (PPVp)³³, soybean mosaic virus protease (SbMVp)³⁴, and sunflower mild mosaic virus protease (SuMMVp)³⁵, which have been previously utilized in the SPOC system, the synthetic protease and coiled-coil-based signal-processing system¹⁵. In a similar fashion to the described erTEVp, an ER-localized SbMVp (erSbMVp) and PPVp  (erPPVp) were designed to also feature an N-terminal signal sequence and C-terminal KDEL sequence for use with the lumER system.

Protease-dependent release of proteins from mammalian cells

For visualization of trafficking of the membER and lumER constructs within the cell, TagRFP³⁶ was fused to secretion constructs and expressed in HEK293T cells. The addition of a signaling sequence and a retention signal localized the constructs mostly in the ER (Fig. 1 b, c, Supplementary Figs. 1, 2a, b, c). For detection of proteins released into the cell media upon coexpression of an appropriate protease, reporter proteins secreted alkaline phosphatase (SEAP) and Gaussia luciferase (Gluc) were incorporated into the lumER and membER systems. The expression of both lumER-SEAP and membER-SEAP was verified by Western blot (Supplementary Fig. 2d, e). Since the implementation of KKYL retention signal produced less leakage compared to the KKMP (Fig. 1d, Supplementary Fig. 3a), we opted for the use of the former retention signal in the membER system. We found that all four tested potyviral proteases were capable of inducing secretion when co-transfected with a membER construct containing the appropriate substrate. However, TEVp provided the best fold increase of the secreted reporter into the media in the presence of the protease (Fig. 1d). Since secretion of the membER system depends on FURp processing to release the protein from the plasma membrane, cells were co-transfected with a plasmid expressing the native FURp. A further increase in the amount of a secreted protein was observed when FURp was added, without increasing protein secretion in the absence of a protease (Supplementary Fig. 3b). In summary, implementation of the membER secretion system with 4 orthogonal and highly specific proteases was demonstrated.

lumER-based SEAP secretion is controlled by the proteolytic processing by an ER-localized protease. Reporter secretion was minimal in the absence of a potyviral protease, while the addition of erTEVp led to the 20-fold increased release of SEAP into the media (Fig. 1e). Similarly, we tested PPVp and SbMVp on the lumER system, by replacing the TEVs with PPVs and SbMVs, and directing the two proteases into the ER lumen through inclusion of a signal peptide and a KDEL sequence. Coexpression of an ER-targeted PPVp however did not lead to the release of SEAP into the cell medium (Supplementary Fig. 3c). A large proportion of proteins traversing through ER and GA is processed by a diverse array of post-translational modifications, most notably glycosylation by ER- and GA-resident glycosyltransferases, which can modify proteins³⁷. Glycosylation might decrease or altogether abolish the activity of a protein, depending on its position within the protein. Indeed, TEVp required mutations at two sites to prevent glycosylation and maintain its activity inside the ER lumen³². An N-glycosylation site prediction tool NetNGlyc 1.0 Server (http://www.cbs.dtu.dk/services/NetNGlyc/) revealed two potential glycosylation sites in PPVp (N23, N173), potentially explaining the observed lack of activity of this protease in the ER. After changing the two amino acid residues (N23Q, T173G) at positions corresponding to erTEVp, small increase of secretion was observed (Supplementary Fig. 3d). ER-targeted SbMVp, on the other hand, was able to induce strong secretion of SEAP (Fig. 1e), demonstrating here for the first time the activity and retention of SbMVp in the ER, therefore providing an additional orthogonal regulator of protein secretion.

The addition of several TEVs between the retention signal and transmembrane domain increased the secretion of SEAP after the addition of TEVp (Supplementary Fig. 3e), likely due to an increased accessibility of the substrate and increased distance between the ER membrane and the retention signal³⁸. Increasing the amount of TEVp in combination with the membER constructs led to the increased secretion of SEAP (Supplementary Fig. 3f). On the other hand, increasing the amount of the ER-localized protease in the lumER system above a certain threshold led to a decreased secretion of SEAP (Supplementary Fig. 3g). We reasoned that this might be due to the competitive binding of the erTEVp to the KDELR, which also features a KDEL retention sequence on its C-terminus, and could lower the capacity of the retrograde transport. This was indeed confirmed by rescuing reporter secretion by the ectopic expression of the KDELR1, which led to an almost 2-fold increase of the amount of secreted SEAP triggered by the erTEVp (Supplementary Fig. 3h).

Kinetics of the membER and lumER secretion systems

We next aimed to regulate secretion using chemically inducible proteases. For this purpose, the FKBP/FRB and ABI/PYL1 heterodimerization domain pairs were used as chemically induced dimerization (CID) systems regulated by rapamycin³⁹ and abscisic acid (ABA)⁴⁰, respectively. Both CID systems were used to reconstitute split fragments of the four aforementioned potyviral proteases, which regained their activity upon heterodimerization. For use in the ER lumen, we designed pairs of split protease fragments of erTEVp, a signal sequence and a KDEL retention signal, fused to CID domains. The resulting pairs were termed er-rapa-TEVp and er-aba-TEVp (Fig. 2a). HEK293T cells expressing the membER secretion system with SEAP as the POI and FKBP-cTEVp/FRB-nTEVp (rapa-TEVp) or ABI-cTEVp/PYL1-nTEVp (aba-TEVp) as the inducible split protease, secreted SEAP when stimulated with rapamycin or ABA, respectively (Fig. 2b). As with single-chain proteases, the KKYL retention signal resulted in higher fold of induction than KKMP (Fig. 2b, Supplementary Fig. 4).

Chemical stimulation of cells expressing er-rapa-TEV and the lumER secretion system resulted in an inducible secretion of SEAP (Fig. 2b). Similar to the single chain protease, higher amount of er-rapa-TEV resulted in a decreased amount of secreted SEAP after stimulation with rapamycin (Supplementary Fig. 5a), which can be attributed to the competition with the KDEL receptor. In the case of er-aba-TEV heterodimerization domains within the ER, no secretion was detected for ABA stimulation (Supplementary Fig. 5b). The N-glycosylation site prediction tool indicated several potential glycosylation sites on both ABI and PYL1 heterodimerization domains (ABI: N292; PYL1: N66, N90), which likely hindered its function in the ER.

Rapamycin, on the other hand, was able to induce secretion already at picomolar concentrations and reached a plateau at approximately 25 nM with the membER system and 10 nM with the lumER system, while ABA was able to induce secretion in the membER system at nanomolar concentrations and reached a plateau at approximately 0.5 mM concentration (Supplementary Fig. 6). Taken together, both systems were able to elicit a concentration-dependent secretion of POI, however ABA-inducible secretion functioned only for the membER system.

To monitor the secretion kinetics, cells harboring constructs for secretion systems and split chemically inducible TEVp were stimulated either with rapamycin or ABA and cell medium was harvested at designated time points. To directly compare the kinetics of a chemically regulated secretion system with a chemically regulated transcriptional activation system, we placed the expression of SEAP under a minimal promoter that was regulated by an inducible CRISPR-based transcription factor. A dCas9 protein fused to FKBP or ABI (dCas9:FKBP and dCas9:ABI) was brought into the proximity of the promoter by an sgRNA targeting the minimal promoter sequence. Transcription was initiated by the addition of rapamycin- or ABA-inducing heterodimerization of the dCas9 fusion with the transcription activator domain VPR fused to the complementary CID protein (FRB:VPR or PYL1:VPR, Fig. 2a). Secretion kinetics for membER, lumER and the transcriptional activation-based systems were measured and compared at varying concentrations of the chemical inducers (Supplementary Fig. 7). While the proteolysis-based secretion systems were able to induce secretion in less than 1 h, the transcriptional activation-based secretion was only observed after several hours. The lumER system was able to induce secretion with faster kinetics than the membER system; in the former a statistically significant increased amount of SEAP in the cell media compared to non-stimulated cells was observed at 45 min (Fig. 2c). While the additional co-transfection of FURp with the membER system demonstrated increased secretion after the extended time (Supplementary Fig. 3b), it did not significantly increase the secretion in the first 4 h (Supplementary Fig. 8).

We next tested the ability of both secretion systems for multiple stimulations. After induction with either rapamycin or ABA, cell media were sampled for up to 3-4 h, whereafter the cells were washed and fresh media were supplemented. SEAP concentrations continued to remain low in the media up until 4 h after washing, after which cells were restimulated and a repeated secretion of SEAP was observed in stimulated cells, with minor leakage from the unstimulated cells even after 20 h (Fig. 2d), suggesting that both systems could be activated with multiple pulses. We cannot however exclude the possibility that the first rapamycin stimulus did not deplete all synthesized protein from the ER leaving the second pulse with no more protein to release. Therefore, we next determined how long it takes to release the total amount of the ER-localized protein pool from the cell after the induction of secretion. Since the constitutive expression of our constructs is continuously replenishing the ER with the new biosynthesis of POI, we aimed to halt the transport of newly formed protein into the ER lumen, so that only the pre-synthesized ER protein pool could be secreted. Cells were treated with Eeyarestatin I (Sigma-Aldrich, E1286), a potent inhibitor of Sec61-mediated protein translocation into the endoplasmic reticulum, which has been shown to inhibit secretion of cellular proteins at the ER-resident ribosomal biosynthesis step⁴¹. Media were harvested after induction of secretion with rapamycin. We observed that the majority of the available SEAP was secreted within 6-7 h after stimulation for both membER and lumER system (Supplementary Fig. 9).

The intracellular trafficking of membER and lumER proteins within the cell was monitored under the fluorescence microscope, by the inclusion of a super-folder GFP (sfGFP), split into two segments between the 10^th (GFP1-10) and 11th β-strand (GFP11), which have been previously shown to allow for a high resolution imaging within cells⁴². GFP1-10 was fused to the membER (SS-GFP1-10-FURs-TD-TEVs-KKYL) and lumER (SS-GFP1-10-TEVs-KDEL) secretion systems, while seven tandem repeats of GFP11 (GFP11x7) were fused to the human Fas receptor transmembrane domain. The trafficking of the ER-localized construct to GA and further to the plasma membrane could be visualized by the reconstitution of the split sfGFP when they were localized in the same cellular compartment (Supplementary Figs. 10a, 11a). Since the GFP11x7 fragment, once produced, needs to pass the ER to reach GA and the plasma membrane, the fluorescence could be already observable in the ER (Supplementary Figs. 10b, 11b). Both the lumER and membER systems retained the reconstituted protein within the ER when no protease was added. Coexpression of TEVp and erTEVp led to the appearance of a fluorescent signal at the plasma membrane (Supplementary Figs. 10b, 11b). The reconstitution of a split TEVp and split erTEVp similarly led to the accumulation of the split GFP fragments at the plasma membrane after 3 h (Supplementary Figs. 10c, 11c). When visualizing a group of cells, a noticeable increase in fluorescence was observed in the GA in less than 15 min after stimulation with rapamycin, while the signal at the plasma membrane was observed after 1 h (Fig. 3).

Since the two systems are based on differently localized proteases and different recycling signals it is likely that they could be regulated independently. Orthogonality of the two secretion systems was tested by co-transfecting both the lumER and membER systems simultaneously. For the membER system a SEAP reporter was used, while the lumER was fused to a Gluc reporter (Fig. 4a). Cells expressing both systems were co-transfected with TEVp, erTEVp or both proteases and measured Gluc and SEAP activity in the media. Each system elicited secretion only in the presence of the appropriate protease (Fig. 4b). Similarly, we managed to elicit the selected secretion by co-transfecting a cytosolic aba-TEVp and ER-localized er-rapa-TEVp (Fig. 4c) and inducing secretion by the addition of either ABA or rapamycin (Fig. 4d).

Construction of an input signal-processing module to regulate protein secretion via the membER and lumER systems

The ability to regulate the removal of the retention sequence with several chemical or biological signals would allow for a stringent control over protein secretion, as well as allow the system to be controllable by different combinations of external but also internal signals. While KKYL or KDEL is required at the C-terminus of the protein to be retained within the ER, only a single retention signal can be present on each protein, making it difficult to design a construct responding to a combination of two inputs, with the exception of an OR gate. We therefore employed the previously described SPOC system¹⁵, based on split proteases and combinations of coiled-coil peptides, which is able to process two-input signals according to Boolean logic operations and produce an output in the form of a reconstituted pottyviral protease activity¹⁵. The reconstituted pottyviral protease could therefore be further used to regulate protein secretion. The principle of SPOC logical circuits used in this study is described in Supplementary Fig. 12. We employed the constructs to demonstrate an A nimply B function, activated by the presence of one signal and the absence of the other one (Fig. 5a), as well as an AND function, activated by the presence of both signals using membER secretion system (Fig. 5b), both with excellent difference between the active and inactive output state. However, any two- or multiple input Boolean logic could be implemented in principle.

Optimization of each protease to retain its activity in the ER for lumER system would likely need to be accomplished by the introduction of appropriate mutations in the proteases, as demonstrated in this study. More importantly however, transferring the components of SPOC into the ER might increase leakage of the system due to the saturation of the recycling machinery. We therefore employed an alternative strategy to adapt SPOC to lumER and designed a system that allows retention of proteins in the ER and their release upon signal processing of two-input signals. We appended the KDEL sequence to an ER-localized designed coiled-coil forming P3 peptide via a TEVp cleavage site (P3-TEVs-KDEL), which interacts with its complementary peptide P4, located on our POI (SEAP-P4). The interaction of P3/P4 coiled-coil pair was capable of retaining SEAP within the ER when the KDEL sequence was present only on one of the two interacting constructs (Supplementary Fig. 13a). In this system, cleavage with TEVp resulted in the release of a POI. To perform an AND function in the lumER system, an additional KDEL sequence was fused to the protein of interest with SbMVs inserted between P4 and KDEL (SEAP-P4-SbMVs-KDEL) (Supplementary Fig. 13b). This resulted in induction of secretion only when both erTEVp and erSbMVp are present (Fig. 5c).

To further control secretion, we aimed at establishing a simplified version of an A nimply B logic function to function as an OFF switch, which was intended to halt secretion, at some selected delay after it had been initiated. For this purpose, we inserted PPVs between the heterodimerization domains and the split TEVp fragments (FKBP-PPVs-cTEVs, FRB-PPVs-nTEVp). The addition of PPVp cleaves off the split protease fragments from the heterodimerization domains, preventing their reconstitution and, consequently, secretion (Supplementary Fig. 14). To make the OFF switch inducible, ABA CID heterodimerization domains were used to control the assembly of PPVp (ABI-cPPVp/PYL1-nPPVp) (Fig. 5d). Cells which had been stimulated with rapamycin and were subsequently stimulated with ABA, demonstrated a decreased secretion of SEAP into the media compared to cells only stimulated with rapamycin (Fig. 5d).

Control of therapeutically relevant protein trafficking using lumER and membER systems

Finally, we applied our secretion system to therapeutically relevant proteins, where fast release upon stimulation is important. Diabetes mellitus affects millions of people worldwide and is characterized by depletion or exhaustion of insulin-producing β-cells in the pancreas⁴³. Insulin is normally secreted from specialized granules inside of β-cells and is tuned to be released within minutes to hours after the rise of blood glucose level⁴⁴. Engineered systems have been previously designed to trigger the release of insulin or GLP-1, which however exhibited a response time of several hours⁶. A system for the release of pre-synthesized proteins relied previously on protein aggregation inside the ER lumen³¹. This platform may however lead to increased ER stress, which could result in apoptosis⁴⁵. In comparison to previously described systems based on conditional protein aggregates in the ER³¹, the lumER and membER system prevented activation of ER stress-characteristic XBP1 and decreased cell viability, even when overexpressed (Supplementary Fig. 15), demonstrating its advantage to ER stress-sensitive cells. We used a design of human insulin adapted for protease processing and secretion through the conventional secretion pathway⁴⁶. This was achieved by modification of the insulin sequence to feature two FURs surrounding the C-peptide, which could be processed to a mature insulin by FURp. The preproinsulin sequence was appended to the lumER secretion system (Fig. 6a) and combined with the er-rapa-TEV. Human insulin levels secreted by HEK293T cells were quantified by measuring the amount of secreted C-peptide upon rapamycin stimulation. Alternatively, a reporter system was designed, which substituted the C-peptide of a preproinsulin with a short secretory Gaussia luciferase, which has been previously used as a surrogate reporter of insulin secretion^47,48 (Supplementary Fig. 16). Elevated C-peptide levels were detected in cell media as early as 30 min after the addition of rapamycin (Fig. 6b). The presence of a C-peptide in cell media in the case of the lumER system was detected within 45 min, in contrast to several hours, in the case of a system based on the transcriptional induction of insulin expression by the addition of rapamycin (Fig. 6b). Using a similar strategy, the membER system was demonstrated to control the release of the anti-inflammatory cytokine interleukin 10 (IL-10), a potential therapeutic protein for several therapeutic indications⁴⁹. Increased levels of IL-10 were observed rapidly after induction of secretion with rapamycin (Fig. 6c).

In addition to secretion of soluble proteins, translocation of membrane-associated proteins to the plasma membrane may be useful for a therapeutic application. Chimeric antigen receptors are used to armor T cells against cancer cells presenting specific membrane markers, such as e.g. CD19 for B-cell leukemias and lymphomas. Regulation of CAR T cell responsiveness is desired, as the excessive activation of engineered T cells may lead to often fatal cytokine release syndrome. Apart from the strategy using a kill switch^50,51, which permanently eliminates therapeutic cells, regulation of the activity of CARs² represents a better alternative but its response may be slow. We proposed here to regulate translocation of CARs to the plasma membrane using a chemically regulated translocation system. In this case CARs are restricted within the cell until the signal triggers their rapid translocation to the plasma membrane, where they can engage target cells and trigger cell activation. CARs were engineered to comprise the ER retention signal that could be cleaved off by the chemically regulated protease. The system was demonstrated to regulate translocation of CARs to the plasma membrane, resulting in chemical regulation of the response of CAR T cells to target B-cells presenting CD19 (Fig. 6d, e; Supplementary Fig. 17), which could be used to license the therapeutic response to a level comparable to the clinically used CAR construct design.