A platform of genetically engineered bacteria as vehicles for localized delivery of therapeutics: Toward applications for Crohn's disease

Abstract For therapies targeting diseases of the gastrointestinal tract, we and others envision probiotic bacteria that synthesize and excrete biotherapeutics at disease sites. Toward this goal, we have engineered commensal E. coli that selectively synthesize and secrete a model biotherapeutic in the presence of nitric oxide (NO), an intestinal biomarker for Crohn's disease (CD). This is accomplished by co‐expressing the pore forming protein TolAIII with the biologic, granulocyte macrophage‐colony stimulating factor (GM‐CSF). We have additionally engineered these bacteria to accumulate at sites of elevated NO by engineering their motility circuits and controlling pseudotaxis. Importantly, because we have focused on in vitro test beds, motility and biotherapeutics production are spatiotemporally characterized. Together, the targeted recognition, synthesis, and biomolecule delivery comprises a “smart” probiotics platform that may have utility in the treatment of CD. Further, this platform could be modified to accommodate other pursuits by swapping the promoter and therapeutic gene to reflect other disease biomarkers and treatments, respectively.

consequently systemic side effects. [3][4][5][6][7] Given the ability to treat IBD with biologics and its localization to the gastrointestinal (GI) tract, it is possible to utilize bacteria as vehicles for both the synthesis and delivery of a therapy, referred to as "smart" or "turbo" probiotics. [8][9][10][11][12][13][14][15][16] The majority of such applications solely investigate the production of a biotherapeutic, yet few have addressed targeting mechanisms for localized delivery. We suggest this may be accomplished via a phenomenon known as pseudotaxis, whereby a cell's motility is regulated by a chemical to result in an accumulation of cells within a gradient of the signal; this targeting mechanism has been documented in bacteria with the goals of "smart probiotics" in mind. [17][18][19][20][21] Briefly, in this work pseudotaxis is achieved via the production of a motility regulator protein CheZ with a C-terminal degradation tag, YbaQ, in a cheZ-null mutant (ΔcheZ). 18,21 Cells lacking CheZ are unable to swim forward and they tumble continuously, 22,23 and when a chemical cue induces expression of CheZ-YbaQ, motility is restored and cells generally accumulate near sources of the chemical, referred to as a pseudoattractant. 21 Using a microfluidic device with controlled gradients, 24 we have demonstrated controlled pseudotaxis in response to a signaling molecule of the opportunistic pathogen, Pseudomonas aeruginosa. 21 In this work, motility was characterized via the run:tumble ratio of cells, and similarly their overall speed, as more runs correspond with faster cells. In order to achieve pseudotaxis in a CDrelevant translational setting, it is important to have a disease-specific chemical cue at the desired site (i.e., a biomarker) and it must be able to elicit a genetic response from the engineered bacteria. In patients with IBD, the chemical nitric oxide (NO) is a biomarker found in the intestinal fluid at levels approximately 100-fold higher than in healthy patients, 25,26 generated by intestinal epithelial cells that have been stimulated by pro-inflammatory cytokines produced by nearby T-cells. 25,[27][28][29][30][31] Nitric oxide is often produced by macrophages as an antimicrobial, [32][33][34] and various bacteria have evolved a natural genetically actuated defense mechanism for survival. In this work, we tap into this natural signal transduction mechanism using an NOresponsive promoter to produce CheZ-YbaQ and confer pseudotaxis. [35][36][37] In addition, we can analogously regulate expression of a biotherapeutic to treat disease, such as Crohn's disease.
In the context of a "smart" probiotic, the free radical NO is an ideal signaling molecule as its presence is localized due to its short half-life. Thus, engineered commensal bacteria should remain in an "off" state until they encounter the appropriate dose of NO, which would most likely occur at or near the site where a therapeutic is desired. Importantly, in our strategy, significant effort is made to minimally alter the engineered probiotics so that when they are in the "off" state, they function as they would normally function in native microenvironments. In recent articles, including from our group, with few exceptions, 8,18,[38][39][40] efforts to engineer motility or an ability to target have been lacking. 15,[41][42][43][44] In the present study, we have systematically examined the ability of engineered E. coli to respond to the NO cue and exhibit motility behaviors that suggest pseudotaxis toward NO, as indicated by observations parallel to those in our previous demonstration of pseudotaxis. 21 That is, we had previously shown that when (a) CheZ deficient cells were restored to swim at speeds near the wild-type controls (~15 μm/s) due to induced CheZ-YbaQ production, and (b) accompanying run:tumble ratios were similarly restored to the WT levels, and (c) when swimming was slowed due to CheZ degradation (via inclusion of a degradation tag), pseudotaxis was achieved. These design studies were validated using controlled chemical gradients within a microfluidic device. 21 Accordingly, they enabled subsequent de novo evaluation of alternative genetic circuits based on the engineered cells meeting these established motility metrics. This reduces dependence on complicated in vitro or in vivo analyses that involve signal cues that are unstable or where geometries are less defined.
For example, in vitro studies of NO rely on its release from NONOates, which in turn, rely on pH gradients for NO to dissociate from the parent compound. As a consequence, NO gradients are superimposed onto pH gradients and the former is a relatively unstable molecule, making it impractical to evaluate NO-targeted pseudotaxis within microfluidics. Instead, via the observation of metrics such as swimming speed with regards to the guidelines established for pseudotaxis from our earlier work, 21 we can characterize pseudotactic behavior without complex microfluidic assays.
The model biotherapeutic used in this study is granulocyte macrophage-colony stimulating factor (GM-CSF). This protein is chosen based on its reported therapeutic effects for individuals with CD, [45][46][47][48] and proven production in bacterial hosts. [49][50][51][52] Briefly, CD is believed to have numerous contributing causes, comprising of a dysfunctional innate immune system (i.e., neutrophils) and a compromised mucosal barrier in the intestines. 5,45,46,48,53,54 Additionally, pathogenic bacteria can penetrate the intestinal mucosa and induce a strong inflammatory response, which would otherwise be mitigated by functioning neutrophils. GM-CSF is reported to address each of these root causes, with capabilities to restore mucosal barrier functions via interactions with receptors on epithelial cells, a well-characterized ability to stimulate neutrophils, and a recently discovered ability to be an agonist against pathogenic bacteria. 5,46,[55][56][57][58] Thus, it has been investigated in clinical trials as a potential therapy, albeit via intravenous (i.v.) delivery. [45][46][47] In a trial conducted by Braat et al., whereby probiotics were used to deliver cytokines to treat CD, the authors suggest localized delivery may increase efficacy versus systemic administration, and thus may also promote GM-CSF's purported mucosal healing benefits. 5,15,46,59 In addition to engineering host motility, we created strains that express GM-CSF with an N-terminal ompA sequence for translocation into the periplasm, an environment often necessary for proper folding of recombinant proteins. [60][61][62] Expression of recombinant GM-CSF is also regulated by nitric oxide, but expression levels are amplified via a dual-plasmid system utilizing the T7lac promoter that expresses T7 Polymerase (T7Pol) based on NO and T7Pol then amplifies the overall GM-CSF synthesized. We have engineered this regulatory genetic circuit to minimize the off-target effects such as the associated metabolic burden. 37,63 To assist in GM-CSF export, the pore-forming protein TolAIII is co-expressed in the engineered cells with a signal sequence that guides it to the outer membrane. 64,65 Taken together, GM-CSF is first shuttled into the periplasm, and subsequently released into the extracellular space via the pores formed by TolAIII. The incorporation of a release mechanism for GM-CSF is important, as a biotherapeutic that remains intracellular is less likely to perform its desired function outside of the bacterial vehicle. Such approaches that have appeared thus far include inducing host cell lysis via colicin proteins, 8 or the synthesis of fusion proteins comprising the active therapeutic and a carrier such as the YebF protein. 18,66,67 We note that induced lysis requires finely tuned initiation of cell death: premature lysis results in insignificant accumulation of the biotherapeutic and delayed lysis could release the recombinant protein too late. Both scenarios could diminish the efficacy of the therapy-producing bacteria. Here, we opted to form pores in the outer membrane with goals being to: (a) increase the flux of periplasmic GM-CSF out of the cells (eliminating the YebF shuttling mechanism), and (b) ensure biologic activity and avoid unintended side effects by using a minimally modified GM-CSF protein.
Using these designed bacteria, the native promoter (hmp) facilitates the sensing of NO to regulate the synthesis and release of GM-CSF, as well as guide the cells to the desired sites via pseudotaxis. Our hypothesis is that through selective induction to produce the therapeutic at NO-rich locales (sites of inflammation), we may reduce systemic side effects and increase efficacy of the biopharmaceutical GM-CSF. As probiotics are designated safe for ingestion and occasionally even administered to patients with IBD, 3,68-73 we believe that the methodologies developed here provide robust preclinical designs for uniquely delivering biologics in the GI tract. That is, the experimental platform or "test track" efforts described in this work, in conjunction with our previous design studies, 21

| Cell lysate preparation
For Western Blotting: Cells were grown and resuspended in supplemented M9 minimal media as described above and induced as desired.
Samples (~5 ml volume) were spun down at chosen timepoints at 10,000 g for 10 min. The supernatant was discarded, and the remain-

| Supernatant preparation
Cells were grown and resuspended in supplemented M9 media to an OD 600 of 0.

| Western blotting
Approximately 13.5 μg or 20 μg total protein per sample (for 15 or 10 well gel, respectively) was loaded in a 12.5% SDS-PAGE gel and transferred to a nitrocellulose membrane using the semi-dry Trans-Blot SD cell (Bio-Rad, Hercules, CA). Blots were blocked overnight at 4 C with Tris-buffered saline with 0.1% Tween-20 (TBST) and 10% nonfat milk. Blots against the CheZ protein are described previously. 21

| Motility videos and analysis
Cells were grown and resuspended in supplemented M9 media to an OD 600 of 0.

| Protein quantification
A competitive enzyme linked immunosorbent assay (ELISA) was performed using a His-tag detection kit, following manufacturer's instructions (Genscript, Piscataway, NJ). Dilutions were first optimized with a trial assay as per the manufacturer's recommendation.
Generally, a 100-fold dilution of samples was the optimal dilution used in the assay. Plotted values are corrected based on the concentration factor (33.3-fold using the Amicon spin-filter) and normalized between samples using the total cellular protein as determined by the BCA assay.     (WT). These findings are corroborated in Figure 1c which shows the CheZ protein levels within these cells. In Figure 1c-e, RM21 and RM22 cells are RM11 or RM12 cells respectively, that were transformed with the pT5G plasmid to confer fluorescence (eGFP production) for imaging purposes. The effect of pT5G on CheZ production was negligible (Supporting Information Figure S1c). Importantly, uninduced cells showed no detectable CheZ on the Western blots. We   indicating that with our NO-mediated production of CheY or CheY**, pseudotaxis will not be achievable. While both RM31 and RM32 quickly become less motile upon induction, over the course of 90 min a full restoration of motility was observed. Likely, the generation of genomic CheZ was eventually able to balance the increase in its substrate, CheY, after induced. Surprisingly, no difference was observed between RM31 and RM32 cells, and thus CheY** may be able to be

| Generating cells to secrete GM-CSF, regulated by nitric oxide
With the goal of overproducing GM-CSF under the hmp promoter, an initial concept would be to utilize the high-production Δhmp cells.
However, as discussed above, cells lacking genomic hmp are immobile and thus the notion of a bacterium that can overproduce GM-CSF and at the same time migrate toward NO sources is a significant challenge based on the use of the same promoter. Fortunately, a dualplasmid system employing the T7lac promoter to amplify a response to nitric oxide has been developed by our group. 37 Using this genetic circuit, the T7 RNA Polymerase is generated under the hmp promoter on one plasmid (the relay plasmid), which leads to significant expression under the T7lac promoter on a second plasmid (the production plasmid). In this way, the GM-CSF can be overexpressed (T7lac promoter), while the CheZ is expressed at natural levels (hmp promoter).
Further, we have developed a range of relay plasmids that result in varying levels of expression via the production plasmid.
That is, to investigate the ability of this system to express GM-CSF, we created the pRM101 production plasmid, which expresses gmcsf under the T7lac promoter. From the selection of relay plasmids, we chose pRM44 and pRM52, which created RM74 and RM80 cells, respectively. To compare the ability of these cells to produce GM-CSF versus cells that produce GM-CSF under the hmp promoter alone (RM60), we performed qPCR and Western Blot analyses ( Figure 4).
We believe this is attributed to a "fatal" combination of high expression seen in pRM52-based systems, where the presence of the ompA tag on GM-CSF resulted in destructive clogging of the secretion pathway. 37,[87][88][89] Together, we believe that expression of this gene has to be tightly regulated in order to maintain a stable metabolic burden of the host bacteria. Despite this, both RM60 and RM74 cells could stably produce GM-CSF upon induction, and exhibit no detectable protein without the presence of NO. Figure 4b corroborates the ability of RM74 cells to express higher levels of mRNA (and consequently protein in Figure 4a) than RM60 cells, confirming that the amplification circuit does indeed outperform the hmp promoter alone. pRM44. An illustration of RM85 is shown in Figure 7a.
Keeping in mind our goal of a probiotic host for an engineered system of GM-CSF delivery, we implemented the pRM44 relay plasmid into E. coli Nissle 1917 cells, a commensal strain already used as a therapy for patients with IBD. 70,71,90 The production plasmids pRM101 and pRM102 were also transformed, generating RM94 and RM95 cells, respectively. The capability of the aforementioned W3110 cells along with the Nissle strains to produce and secrete GM-CSF into the extracellular space is presented in Figure 5; GM-CSF was indeed produced in all examined cells, regardless of host strain or the presence of (a) the TolAIII pore, (b) the T7-mediated amplification, or (c) expression of cheZ (Figure 5a). A blot on the supernatants of these bacteria reveals that cells harboring a relay plasmid for  Figure S2d). Considering these findings, we believe the primary cause of GM-CSF leakage in tolAIII − cells is a consequence of high levels of periplasmic GM-CSF. Further discussion of observed leakiness can be found in the Supporting Information.
We next sought to quantify the levels of protein produced by the cells via ELISA (Figure 5d) to complement the qualitative analysis in dose-dependent response. As in Figure 5a,b, the host genome (WT vs. ΔcheZ) and subsequent NO-induced expression of cheZ did not dramatically alter the levels of GM-CSF produced. It was also discerned that despite the effects of the pore, significant amounts of GM-CSF

| Secreted GM-CSF is biologically active
Previous reports indicate that recombinant GM-CSF produced in  Figure S4). Additionally, the supernatants underwent at least three freeze-thaw cycles during the course of these experiments, suggesting a modest level of stability of the recombinant protein.

| Combining pseudotaxis capabilities and GM-CSF secretion
In order to achieve a singular cell to serve as a delivery vehicle for GM-CSF with a targeting mechanism, we incorporated both genetic circuits into one host cell. As discussed previously, ΔcheZ host cells are preferred for pseudotaxis, and thus these E. coli are transformed with plasmids pRM45 (expressing T7Pol and cheZ-YbaQ) and pRM101 or pRM102 (expressing gmcsf AE tolAIII) to produce RM84 and RM85 cells, respectively.  and −30 timepoints). As noted previously, a rapid loss of motility is beneficial for pseudotaxis purposes, as it encourages the engineered cells to slow down when exiting pseudoattractant (i.e., NO) rich areas. 21 The trends observed in Figures 2 and 7 where RM24 and RM85 cells, respectively, have their motility restored in the presence of nitric oxide but quickly become tumbly in the absence of the inducer, mirror the behavior of pseudotactic cells previously reported. 21 As discussed earlier, we therefore believe by extension that the ΔcheZ cells engineered in this study to produce CheZ-YbaQ in response to NO should exhibit pseudotaxis and accumulate near sources of nitric oxide. Their ultimate ability to be retained near NOrich locales naturally will depend on many factors, however.

| Using an in vitro
Crohn's disease model to stimulate GM-CSF production Using the pseudotaxis-capable delivery vehicle cells RM85, and the probiotic proof-of-concept RM95, we sought to confirm their response to a simple CD in vitro model consisting of inflamed intestinal epithelial cells using the Caco-2 cell line. Briefly, a confluent layer of Caco-2 cells secrete a basal level of NO, and we stimulate a subpopulation with pro-inflammatory cytokines to significantly increase their NO production to mimic inflammation. 95-97 We then exposed our engineered bacteria to the supernatant of the Caco-2 cells (i.e., the "lumen") for 90 min, and analyzed their gmcsf expression (Supporting Information Figure S5). The results indicate a small increase in gmcsf expression in response to the basal NO secretion, and, importantly an elevated response (up to 20-fold) when exposed to the inflamed model. We estimate that the NO level in these cells may be upwards of 20 μM. 97