Feasibility of engineered Bacillus subtilis for use as a microbiome‐based topical drug delivery platform

Abstract Non‐adherence to medication is a major challenge in healthcare that results in worsened treatment outcomes for patients. Reducing the frequency of required administrations could improve adherence but is challenging for topical drug delivery due to the generally short residence time of topical formulations on the skin. In this study, we sought to determine the feasibility of developing a microbiome‐based, long‐acting, topical delivery platform using Bacillus subtilis for drug production and delivery on the skin, which was assessed using green fluorescent protein as a model heterologous protein for delivery. We developed a computational model of bacteria population dynamics on the skin and used its qualitative predictions to guide experimental design choices. Using an ex vivo pig skin model and a human skin tissue culture model, we saw persistence of delivered bacteria for multiple days and observed little evidence of cytotoxicity to human keratinocyte cells in vitro. Finally, using an in vivo mouse model, we found that the delivered bacteria persisted on the skin for at least 1 day during every‐other‐day application and did not appear to present safety concerns. Taken together, our results support the feasibility of using engineered B. subtilis for topical drug delivery.


| INTRODUCTION
Improving patient adherence to existing medications is a major challenge in healthcare that, due to its wide-reaching nature, could yield greater health benefits than improvements to any specific medication. 1 In fact, nonadherence accounts for an estimated 10% of drug-related hospital admissions, and proper adherence to medication is <50% in countries with advanced economies; it is expected to be much lower in other parts of the world. 2 Improved drug delivery platforms have the potential to make medication adherence easier for patients by minimizing side effects, allowing more comfortable administration routes, and reducing the number of required administrations. 3In the case of dermatology, topical drug delivery platforms are attractive because the therapeutic can be delivered directly to the site of action (i.e., the skin) in a noninvasive way, typically by patients themselves.This approach avoids the use of needles and other barriers to obtaining treatment, which can help minimize hospital visits and unwanted side effects.
Despite the advantages of topical drug delivery, it is difficult to deliver drugs topically for an extended time.Gels, ointments, and lotions often require daily or more frequent application because of their limited residence time on the skin due to removal by clothing, sweating, rubbing and other effects. 4The need for frequent administration of topical delivery formulations can negatively impact patient adherence, especially in the case of chronic diseases such as psoriasis and atopic dermatitis. 5 exciting area emerging in the field of drug delivery is the use of bacteria as drug delivery vehicles. 6,7The symbiotic relationship between bacteria and the humans they colonize combined with existing technology to engineer bacteria to produce heterologous molecules creates an opportunity to use engineered bacteria as a form of in situ drug production and delivery.[8] In the context of dermatology, there are preclinical and clinicalstage companies currently working to develop engineered bacteria as topical drug delivery agents.Notable examples include Azitra, which has multiple preclinical products using engineered Staphylococcus epidermidis to treat skin disorders 9 ; Ilya Pharma, which has developed a clinical-stage strain of Limosilactobacillus reuteri expressing the chemokine CXCL12 to promote wound healing 8,10 ; and Xycrobe Therapeutics, which is developing a preclinical strain of Cutibacterium acnes engineered to deliver IL-10 for acne vulgaris. 8 this study, we investigate the potential use of engineered Bacillus subtilis as a topical drug delivery platform.B. subtilis may have advantages over other candidate bacteria as a platform for drug delivery to the skin because of its safety profile and genetic tractability.It is found in the skin microflora and is metabolically active on the skin. 11,124][15] B. subtilis has generally regarded as safe status from the FDA, and multiple B. subtilis probiotic products as well as a genetically modified strain of B. subtilis are currently commercially available. 7,16,17 important characteristic of B. subtilis is that it is commonly used in biotechnology for the production of proteins, vitamins, and antibiotics because of its efficient protein secretion system, ease of cultivation, and high genetic tractability. 18As a result, many tools and libraries have been developed for its genetic manipulation, making the design and implementation of engineered B. subtilis-based biotherapeutics more straightforward than the similar use of other, non-model organisms. 18,19tivated by these characteristics of B. subtilis, we sought to assess the feasibility of using B. subtilis as a topical drug delivery platform in terms of the duration of drug production on skin and safety.
To facilitate characterization, we used an engineered strain of B. subtilis producing green fluorescent protein (GFP) as an easily detectable, model heterologous protein.Through a combination of experimental studies supported by computational modeling, we identified that microbiome-based drug delivery via B. subtilis is feasible, with certain constraints.

| Model simulation
In our overall approach to assess the viability of B. subtilis-based skin drug delivery, we relied on a hybrid computational and experimental approach where initial model predictions helped to guide experiments.
We first developed an agent-based model for bacterial population dynamics on skin based on the Gutlogo model of population dynamics in the gut. 20This model predicted that the populations representing Corynebacteria, Staphylococci, and Acinetobacter would reach a steady state comprising about 84% Corynebacteria, 15% Staphylococci, and 1% Acinetobacter.When B. subtilis was introduced at timestep 4500 (3.125 days), it survived for about half a day before quickly dying off, after which the other species largely returned to their previous steady-state levels (Figure 1).

| Experimental findings
Before assessing survival time of B. subtilis on skin experimentally, we first wanted to account for the possibility that in situ production of a therapeutic for delivery could have a substantial growth burden on the cells, providing a negative selection pressure.We sought to identify the impact on cell growth of expressing a heterologous reporter protein from a plasmid using GFP as a simple model protein, which could represent either a protein directly serving as a therapeutic or an enzyme used to synthesize a small-molecule therapeutic.
Using reporter plasmids transformed into B. subtilis, we tested six different ribosome binding site (RBS) candidates expected to yield a variety of expression levels to determine if there were expressiondependent effects on cell fitness.Using doubling time, fluorescence in culture, and fluorescence in co-culture with the parent (non-fluorescent) strain to monitor cell fitness, we identified that GFP expression had small (albeit significant) effects on cell fitness (Figure S1).Different drug products could result in larger growth defects than what we observed with GFP, and this will likely vary significantly between different drugs.To assess the possible effects that a larger growth defect might have on survival time, we used the computational model to explore the survival of B. subtilis with a 10%, 25%, 50%, and 100% growth defect and found no significant difference to survival time (Figure S2).
To make later measurements of B. subtilis survival and protein expression as sensitive as possible, we used RBS5 in strain BS-GFP for experiments moving forward, as it yielded almost an order of magnitude more GFP expression than any other RBS with minimal additional impact on cell growth.We validated that for the timescale of our proof-of-concept experiments, the stability of pGFP-RBS5 in B. subtilis in the absence of antibiotics in liquid culture was such that the plasmid was maintained for at least 24 h (Figure S3).We used the resulting strain as the basis for experiments in an ex vivo pig skin model.On pig skin that had microorganisms cleaned from the surface (i.e., cleansed, but not sterile), the BS-GFP strain apparently survived on the skin and produced GFP for about 4 days, after which fluorescence reached a plateau (Figure 2).This duration of survival was different from that seen in the model simulation, probably because there was only limited competition with other bacteria on the cleansed pig skin.Comparing B. subtilis with Escherichia coli, a gram-negative species with limited survival on human skin 21 but similar or faster growth rates in liquid culture, we found that BS-GFP produced more GFP than EC-GFP in the pig skin model (two-way ANOVA, p < 0.0001) even though EC-GFP had higher fluorescence expression than BS-GFP in liquid culture (Table S1).Taken together, these results support the potential viability of B. subtilis as a skin drug delivery platform.

| Effect of added antibiotics
A way to extend the lifetime of added B. subtilis in a skin microbiome may be to add a selective pressure that provides an advantage for B. subtilis compared to the native microbiome.We investigated one way of forcing a selective advantage: adding an antibiotic to the environment to which B. subtilis is resistant but the rest of the microbiome is sensitive.

| Model simulation
We first used the computational model to simulate the potential impact of adding such an antibiotic.According to the model, antibiotic concentrations <1000 times the defined MIC for the native microbiome species had almost no effect on the survival of B. subtilis, with the other populations recovering shortly after the B. subtilis population died.At the highest concentration tested in silico (1000 times the defined MIC for the native microbiome species), the B. subtilis population appeared to achieve long-term survival, at the expense of the Acinetobacter population dying completely (Figure 3).

| Experimental findings
We next tested the effects of antibiotics in the ex vivo pig skin model using the antibiotic kanamycin, to which the BS-GFP strain has resistance due to the selection marker on the plasmid.We found that there was a dose-dependent effect of antibiotic on total GFP production, as well as an increase in the length of time over which GFP was produced from 4 to 5 days when antibiotic was added (Figure 4).Consistent with the model simulation, the dosedependent antibiotic effect may be due to decreased competition with residual microorganisms on the cleansed skin.Coupled with nutrient limitations, this may have led to increased growth of the B. subtilis cells and thus higher overall fluorescence.It is possible that the effect of the antibiotic was also due to decreased competition with B. subtilis cells that spontaneously lost plasmid and would otherwise have had a growth advantage by virtue of not producing heterologous protein.Again, coupling this with nutrient limitations would lead to more plasmid-bearing B. subtilis and thus higher GFP expression levels.
We characterized the GFP production curves (Figure 4a) with a rate constant (Figure 4b).There was a significant increase in the rate constant of the fluorescence curve for the lower antibiotic concentration and a trend toward an increased rate constant for the higher antibiotic concentration that was not significant (Figure 4b).Taken together, these results suggest that using an antibiotic adjuvant for

| Effect of added carbon source
Another way to extend the lifetime of added B. subtilis in a skin microbiome may be to supplement a nutrient source that B. subtilis would have a distinct advantage in using compared to the native microbiome.It has previously been shown that creation of an exclusive metabolic niche in the gut can lead to long term colonization of a probiotic in mice. 22We hypothesized that creating a similar niche on the skin may yield similar results.

| Model simulation
We thus used the computational model to evaluate the potential effect of adding a carbon source to specifically benefit B. subtilis.
We chose to use malate as the additional carbon source; while malate can be consumed by other species, it is a preferred carbon source for B. subtilis (in addition to glucose), so we would expect malate addition to most benefit B. subtilis.Since the model does not differentiate between preferred and non-preferred carbon sources, we modeled this supplementation as the creation of an exclusive niche where no other species could use malate as a carbon source.Addition of increasing concentrations of malate led to an increase in predicted B. subtilis survival time from about half a day to about 1 day (Figure 5).We also tested the effect of adding additional, non-specific, nutrients such as LB at the same time as B. subtilis and observed a slight increase in survival time (Figures S4 and S5).

| Experimental findings
Malate was added to the ex vivo pig skin model at two doses (0.1 and 1 M, which correspond to $3.6 and 36 μmol/cm 2 , respectively).We observed a dose-dependent effect of the added malate on total GFP production.We also observed an increase in the length of time over which GFP was produced to about 5 days for the 1 M malate group, as well as significant increases in the rate constants for both malate concentrations (Figure 6).Thus, malate could serve as a useful added nutrient source for B. subtilis on skin, enabling increased protein production and a longer duration of protein production.MatTek) to assess the ability of B. subtilis to survive and produce GFP on human skin that had been precolonized with microorganisms from a swab of human forearm skin (Figure S6).Using fluorescence measurements to track GFP production, we found that BS-GFP survived and produced GFP for 2.1 ± 0.7 days.(Figure 7a), which is generally consistent with the findings from our pig skin model.

| Experimental findings
For B. subtilis to be a viable vehicle for therapeutic delivery to the skin, it must not be toxic to the human cells that it will encounter.The  The LD 50 of LB and supernatants from 168 and 15841 cells were 64.0%(95% CI: 49.8-90.50),33.5% (95% CI: 31.6-35.5)and 28.6% (95% CI: 26.2-31.4),respectively, after 3 days of exposure (Figure S7).The LD 50 values for both strains 168 and 15841 were significantly different from that of LB media (p < 0.0001), which means that the supernatants of both strains had greater toxicity than the media alone but were of the same order of magnitude.
2.6 | Safety and survival of B. subtilis on mouse skin in vivo

| Experimental findings
Given that the in vitro, ex vivo, and computational model results suggest the viability and safety of B. subtilis on skin, we used a mouse model in vivo for further validation.BS-GFP, the parent strain B. subtilis 168, and LB medium were inoculated onto the backs of hairless mice every other day, with daily swabbing to track whether BS-GFP was still present on mouse skin by counting green colonies after plating on selective medium.We found that BS-GFP was consistently detected on the day following application and sometimes on the following day (Figure 7b).
After the last application of bacteria, BS-GFP was detectable by swab for 3 days and then returned to negligible levels.Mice were then sacrificed, and biopsy samples of skin exposed to bacteria or media were taken for histological analysis.No evidence of tissue damage or infiltrating immune cells were observed for the mice receiving B. subtilis treatment (BS and BS-GFP) compared to mice treated with just medium (Figure 7c-e).

| DISCUSSION
For topical drug delivery, medications are commonly applied to the skin via gels, ointments, lotions, creams, and sprays, which are advantageous for their ease of application over the affected area of skin. 23wever, topical formulations are also easily removed from the skin after a short amount of time and often require daily or more-frequent application to be effective. 4Continuous production of a drug or other molecule of interest on skin throughout the day could support enhanced adherence and better outcomes with these forms of treatment.Skin-resident bacteria like B. subtilis could be engineered to continuously produce drugs on the skin for lasting therapy.
Across our various models, we found that B. subtilis survived on the skin while producing GFP for at least 1 day and appeared to be safe.Although B. subtilis is found on healthy skin, we wanted to confirm that doses of interest for therapeutic applications would not be harmful to the skin, even if at levels much higher than typically found on human skin. 12While we found that the supernatants of the two B. subtilis strains tested were more toxic to human keratinocyte cells than just LB medium alone, the LD 50 values were similar, leading us to conclude that the strains would likely be safe for use on skin, but more work is needed to fully answer this question.Future studies incorporating real drugs would need to test the cytotoxicity of the engineered strain in addition to the parent strain to account for any drug-related toxicity.Furthermore, no signs of negative skin reaction were observed on mouse skin in vivo after 1 week of every-other-day application of B. subtilis.Altogether, our data suggest that B. subtilis may be well tolerated by skin, at least at the levels used in this study.
A safety component which was not experimentally investigated in this study was the effect of B. subtilis treatment on the existing microbiome.The microbial community present on the skin is associated with skin health, so it would be important to also ensure that any changes to the existing community are not harmful. 24It is worth noting that the computational model predicted that, without antibiotic addition, after the delivered species dies off the existing community would return to its previous levels.This suggests that perturbations to the skin microbiome due to the use of B. subtilis as a delivery vehicle may just be transient.In agreement with this observation, other work analyzing the shift in mouse ear microbiota composition during and after B. subtilis administration found that there were shifts in the microbiota composition during administration of the bacteria, but the microbiota composition was similar to that of the untreated control group within a week after stopping administration of bacteria. 25arting with equal numbers of the representative skin microbiome species in this model, the relative steady state percentages predicted by the computational model qualitatively matched relative percentages of their genera identified from swabs of human forearm skin. 26The computational model also predicted that a species with doubling time and carbon source utilization characteristics representative of B. subtilis could survive within an established community representing the skin microbiome for a little over half a day.It predicted that adding a carbon source to specifically benefit the added species could lead to a modest increase in survival time and that application of an antibiotic to which the added species has resistance could lead to long-term survival, although with long-term consequences to the existing community.These predictions generally aligned with experimental results using an ex vivo pig skin model, where we found that the engineered strain of B. subtilis survived on skin while producing GFP for multiple days, and the addition of either kanamycin, to which the engineered strain had resistance, or malate, a preferred carbon source for B. subtilis, led to significantly increased GFP production.
While addition of antibiotics increased B. subtilis survival in both models, the FDA has recommended against the use of antibiotic selection markers in clinical stage products. 27We used antibiotics in this study to develop a proof-of-concept understanding of B. subtilis survival on skin in the presence of selective survival pressures, but in a future human study or commercial product using antibiotics as a selective pressure would not be suitable.Alternative selective pressures, such as added nutrients that favor B. subtilis growth, would be a preferred approach.
We used a full thickness human skin tissue culture model to evaluate the survival of B. subtilis on human skin in the presence of microorganisms obtained from human forearm skin and found that BS-GFP survived while producing GFP for about 2 days.While this model provided further evidence that B. subtilis can survive on human skin, we applied a relatively low density of human skin microorganisms to the tissue surface, so further studies would be required to further determine the ability of B. subtilis to survive in the context of the skin microbiome.Moreover, our inclusion of LB in the delivery vehicle may have provided nutrients for B. subtilis survival not typically present on the skin and thus affecting cell growth; however, this approach was useful to maintain consistency across experimental models, and additives could be used in a final formulation if the additional nutrients prove critical to enabling this approach.
Finally, when testing survival of B. subtilis on mouse skin in vivo, we found that BS-GFP was repeatedly detectable by swab 1 day after application, which was generally consistent with the computational model predictions.After applications ceased, BS-GFP was still detectable by swab for 2 days but returned to baseline within 3 days of the last application.A study in which B. subtilis was applied to mouse ears for seven consecutive days found that B. subtilis levels returned to baseline levels within 4 days after the last application, which is in general agreement with our results. 25ken together, these findings led us to conclude that B. subtilis could be suitable as a drug delivery platform for a daily or every-other-day application but may not be suitable for longer-term delivery.
Wild-type B. subtilis may be a more transient member of the skin microbiome compared to other skin-resident bacteria, 28 such that longer-resident species, such as S. epidermidis, may be more suited to multiday delivery scenarios. 29For example, the company, Azitra, has multiple clinical and preclinical stage products incorporating engineered S. epidermidis to treat various skin disorders. 9However, variable colonization efficacy has remained a challenge in the field of topical probiotics utilizing resident skin commensal species including S. epidermidis, possibly due to competition with the existing microbiota or variability in the skin sites within and between individuals. 30Furthermore, due to biocontainment concerns, a common strategy to prevent transfer of the engineered strain to undesired areas is to render the strain an auxotroph requiring regular application of an additional substance, such as D-ala- nine, to maintain viability. 31We detected B. subtilis using fluorescent measurements and growth on selective agar plates, but it is possible that residual cells could persist on the skin that have lost the ability to produce recombinant protein, which would result in an underestimation of B. subtilis viability on the skin.This possibility warrants further investigation, and biocontainment would need to be considered for B. subtilis as well, likely employing a similar strategy to the engineered auxotrophy strategy currently used in industry.With these considerations in mind, the genetic tractability of B. subtilis makes it an attractive option for situations in which daily or every-other-day application is appropriate.
A limitation of this study is that GFP production represents a base case scenario in which recombinant expression of a model protein had little effect on cellular fitness, whereas this may not be the case for in situ biosynthesized therapeutics.Small-molecule drugs that require complex pathways for production could require expression of multiple heterologous genes and/or the biosynthetic reactions could affect cellular metabolism, 32 which would likely affect cell growth and ability to survive on skin.Even protein therapeutics may still have deleterious effects on cellular fitness, especially when produced at high levels.3][34] Nonetheless, for implementation of in situ biosynthesis of any therapeutic, the specific toxicity and defects associated with making that specific product, as well as the impacts of that product on the host's natural microbiome, would need to be studied and accounted for in the final design of the organism or delivery vehicle.Also, logistical considerations of human routines such as hand and body washing and transfer of bacteria from skin to other surfaces could limit the potential applications of using engineered bacteria as a topical drug delivery platform.

| Computational model
We adapted the code for Gutlogo, an agent-based model made in NetLogo to simulate population dynamics of the gut microbiota, to represent the skin environment. 20The basic rules of Gutlogo were preserved in our skin model, but with the flow component removed, species characteristics adjusted to reflect a set of common skin microbiome constituents, and metabolite concentrations adjusted to reflect expected conditions on the skin.Malate was considered a carbon source for only B. subtilis.Additional model details are available in the Data S1.Code for our model, Skinlogo, is publicly available at https:// github.com/gtStyLab/skinlogo.git.

| Bacterial culture
Strains and plasmids used in this study are listed in Table 1.B. subtilis 168 and E. coli DH5α were cultured in lysogeny broth (LB), consisting of 0.5% w/v yeast extract (Life Technologies, Carlsbad, CA), 1% w/v NaCl, and 1% w/v tryptone (Life Technologies), at 37 C with shaking at 200 rpm.Growth was measured by absorbance at 600 nm.
Fluorescence and absorbance were measured with a Biotek Synergy H4 Hybrid Microplate Reader (Agilent, Santa Clara, CA).For fitness comparisons, each strain containing a different ribosome binding site (RBS) was inoculated into a co-culture with equal volume of the parent strain B. subtilis 168 in LB media without antibiotics, and fluorescence was compared with that of the engineered strain in monoculture.Fluorescence was normalized by absorbance at 600 nm, and the normalized fluorescence values after 24 h in culture were averaged for three replicates.

| Plasmid assembly
DNA primers used in this study are listed in Table S5.All plasmids were assembled in E. coli DH5α and then transformed into B. subtilis 168.E. coli cells were transformed by heat shock following the NEB high efficiency transformation protocol (New England Biolabs, Ipswich, MA).B. subtilis cells were transformed by electroporation using an ECM 600 electroporator (BTX, Holliston, MA) set to 2.1 kV, 129 Ω, and 50 μF with 1 mm gap cuvettes.Superfolder GFP was synthesized by Eurofins (Louisville, KY).GFP was inserted into pRB374 by Gibson assembly, resulting in plasmid pGFP-RBS0. 35For RBS analysis, each RBS was incorporated into the forward primer for amplification of GFP, which was inserted into pRB374 by Gibson assembly, resulting in the plasmids pGFP-RBSX (X = 1-5).To assess plasmid stability, BS-GFP was grown in LB media with or without selecting antibiotic (10 μg/mL kanamycin).Fluorescence over time was compared between the two growth conditions and was normalized by absorbance at 600 nm.The average value of three replicates was reported.
The plasmid pGFP-RBS5 was used to express GFP in E. coli and B. subtilis for ex vivo and in vivo experiments (EC-GFP and BS-GFP, respectively).During strain and plasmid construction, E. coli and B. subtilis were grown in LB media with 100 μg/mL carbenicillin and 10 μg/mL kanamycin, respectively, at 37 C.

| LD 50 cytotoxicity assay
HaCaT cells were seeded in 96-well plates at a density of 1 Â 10 3 cells per well and incubated for 3 days at 37 C and 5% CO 2 in a humidified incubator.After 3 days, the medium in the wells was replaced with keratinocyte media containing either LB medium or sterile-filtered supernatant from overnight B. subtilis 168 or 15841 cultures at concentrations ranging from 0.00019% to 50% v/v, with eight replicates per condition.Bacterial supernatants were adjusted to pH 7 with 1 M NaOH before being added to mammalian cells.Plates were incubated another 3 days, after which the medium was replaced with sterile phosphate-buffered saline (PBS), 10 μL alamarBlue (ThermoFisher) was added to the wells, and plates were incubated for 1-4 h at 37 C. Wells were read using a Biotek Synergy H4 Hybrid Microplate Reader according to the assay protocol.For the positive toxicity control, 2% Tween 20 was added to the cell culture media to kill the cells.
For the negative toxicity control, no supernatant was added to the cell culture media.Additional details about the alamarBlue assay are available in Data S1.

| Preparation of bacteria for skin model experiments
Overnight cultures were diluted in LB and grown with shaking at 200 rpm and 37 C with appropriate antibiotics (100 μg/mL carbenicillin for E. coli and 10 μg/mL kanamycin for B. subtilis) until they reached an absorbance at 600 nm wavelength of $0.6.Cells were then diluted 1:10 in fresh LB with appropriate antibiotics to be inoculated onto skin models.

| Pig skin model
Six millimeter biopsy punches were taken from ex vivo pig ear inner pinna skin (Pel-freez, Rogers, AR).The skin pieces were rinsed in 0.1% v/v peracetic acid (Pfaltz & Bauer, Waterbury, CT) diluted in sterile PBS, adjusted to pH 7.0-7.4with 1 M NaOH, for 3 h to clean microorganisms from the skin surface. 36The skin was then rinsed in two washes of sterile PBS for a total of 30 min to remove residual peracetic acid.Skin pieces were loaded into an HTS Transwell 96-well Permeable Support plate with a black receiver plate (8 μm pore size, Corning), with sterile PBS in the bottom wells refilled daily to maintain moisture in the skin.
Bacteria were inoculated onto the surface of the skin pieces as 10 μL of BS-GFP or EC-GFP in LB, with eight replicates per experimental group.For the malate and kanamycin experiments, the added bacteria were supplemented with L-malic acid (10 μL of either 0.1 M or 1 M in sterile PBS, Eastman, Kingsport, TN), which was neutralized to pH 7.0 with 5 M NaOH before application to skin, or kanamycin (10 μL of 50 μg/mL or 100 μg/mL in sterile PBS).B. subtilis 168 in LB and sterile PBS were used as negative controls, and 100 μg/mL rGFP (VWR, Radnor, PA) in sterile PBS was used as a positive control.Plates were stored at 32 C, with lids raised slightly up from the plate to prevent condensation on the skin pieces.
Fluorescence was measured at 475 nm excitation and 510 nm emission wavelengths using a fluorescence area scan with matrix size T A B L E 1 DNA plasmids and bacterial strains.

| CONCLUSION
In this study, we assessed the feasibility of using B. subtilis as a topical drug delivery platform.Using GFP as a model heterologous protein, we found that engineered B. subtilis could survive on multiple skin models for at least 1 day and appeared to be safe for skin.Due to its wide commercial use and relative ease of genetic modification, we believe that B. subtilis could be an attractive candidate to make the development of bacteria-based topical delivery platforms more straightforward and accessible.

F I G U R E 1
Simulation of skin bacteria population dynamics using an agent-based model.Population representing Bacillus subtilis was added on day 3.125.Lines represent mean values and error bars represent standard deviation of three simulations.In most cases, the error bars are smaller than the line thickness.
Fluorescence of GFP expressed by Bacillus subtilis over time in an ex vivo pig skin model.rGFP, recombinant GFP added to skin surface at time 0; BS, B. subtilis strain 168; BS-GFP, B. subtilis strain 168 harboring pGFP-RBS5 plasmid; EC-GFP, Escherichia coli strain DH5α harboring pGFP-RBS5 plasmid.Data show mean +/À standard deviation of eight replicates.bacterial-based drug delivery to skin could have positive impacts on total dosage and persistence of delivery vehicle viability.

2. 4 |
Survival of B. subtilis on human skin culture in vitro 2.4.1 | Experimental findings Guided by evidence of B. subtilis survival on pig skin, we next used a full thickness human skin tissue culture model (Epiderm FT, from F I G U R E 3 Simulation of skin bacteria population dynamics when adding antibiotics concurrently with Bacillus subtilis.(a) 0 mg/cm 2 ; (b) 4 mg/cm 2 ; (c) 8 mg/cm 2 ; (d) 10 mg/cm 2 of antibiotic.(e) Survival time of B. subtilis with different concentrations of antibiotic.At 10 mg/cm 2 , B. subtilis appeared to survive long-term, but is shown as 3 days, because the simulation ended at that point.Lines and bars show mean +/À standard deviation of three replicate simulations.

F
I G U R E 5 Simulation of skin bacteria population dynamics when adding supplemental carbon source concurrently with Bacillus subtilis.(a) 0 mol/cm 2 malate added; (b) 37 nmol/cm 2 malate; (c) 3.7 μmol/cm 2 malate; (d) 0.37 mmol/cm 2 malate.Lines represent mean values and error bars show standard deviation across three simulations.In most cases, the error bars are smaller than the line thickness.(e) survival time of B. subtilis with different concentrations of carbon source.Data show mean +/À standard deviation of three replicate simulations.F I G U R E 4 Green fluorescent protein (GFP) production in ex vivo pig skin model with antibiotic addition.(a) Fluorescence measurements over time.BS: Bacillus subtilis strain 168; BS-GFP: B. subtilis strain 168 harboring pGFP-RBS5 plasmid; BS-GFP + 50 μg/mL kan: BS-GFP with 50 μg/mL kanamycin added immediately before bacteria; BS-GFP + 100 μg/mL kan: BS-GFP with 100 μg/mL kanamycin added immediately before bacteria.** p < 0.01, **** p < 0.0001, two-way ANOVA compared to BS-GFP.(b) Rate constant of fluorescence curves shown in part (a).* p < 0.05, oneway ANOVA compared to BS-GFP.Data show mean +/À standard deviation of eight replicates.potential toxicity of B. subtilis was thus tested against HaCaT human keratinocyte cells by exposing the keratinocytes to B. subtilis culture supernatants as part of the keratinocyte growth medium.We tested two strains of B. subtilis, the common laboratory strain 168 and the antifungal lipopeptide-producing strain 15841, using exposure to LB media alone as a negative control.We expressed the LD 50 in terms of the percentage of the supplemented keratinocyte medium composed of LB or supernatant.

F I G U R E 6
Green fluorescent protein (GFP) production in ex vivo pig skin model with malate addition.(a) Fluorescence measurements over time.BS: Bacillus subtilis strain 168; BS-GFP: B. subtilis strain 168 harboring pGFP-RBS5 plasmid; BS-GFP + 0.1 M malate: BS-GFP with 0.1 M malate added immediately before bacteria; BS-GFP + 1 M malate: BS-GFP with 1 M malate added immediately before bacteria.**** p < 0.0001, two-way ANOVA compared to BS-GFP.(b) Rate constant of fluorescence curves shown in part (a).* p < 0.05, *** p < 0.001, one-way ANOVA compared to 0 M. Data show mean +/À standard deviation of eight replicates.F I G U R E 7 Safety and survival of BS-GFP on human skin in vitro and mouse skin in vivo.(a) GFP production in in vitro human skin tissue culture model by BS-GFP.Data show mean +/À standard deviation of six replicates.(b) Persistence of BS-GFP on mouse skin in vivo.Bacterial or medium solutions were added every other day (indicated by arrows), and mice were swabbed daily.BS: B. subtilis strain 168; BS-GFP: B. subtilis strain 168 harboring pGFP-RBS5 plasmid; Media: LB medium.**** p < 0.0001, two-way ANOVA comparing BS-GFP to media.Data show mean +/À standard deviation of three mice per treatment and three technical replicates of swab plating for each mouse.(c-e) Representative images of mouse skin stained with hematoxylin and eosin.(c) treated with B. subtilis strain 168; (d) treated with B. subtilis strain BS-GFP; (e) treated with LB medium.Mouse skin in vivo treated with B. subtilis or media as shown in (b) was biopsied on day 13, sectioned, stained, and imaged by brightfield microscopy.
American Type Culture Collection (Manassas, VA); NEB, New England Biolabs (Ipswich, MA).compare rate constants, LD 50 values, fluorescence in liquid culture, and doubling times.Doubling times were calculated by fitting the midexponential phase of the growth curves to an exponential growth equation.Statistical analyses and linear regressions were performed using GraphPad Prism version 9.4.1 for Windows (San Diego, California, www.graphpad.com).The p-values <0.05 were considered significant.