## Introduction

Engineering vehicles for highly localized, targeted drug delivery has become a significant requirement of many new therapeutic treatments to target tumors or bacterial infection, due to the systemic toxicity, resistance development, and undesirable side effects that many effective therapeutics pose1,2. Antibiotic drug release in a localized environment can target a bacterial infection at an efficacious local concentration, above the minimum inhibitory concentration (MIC), while reducing risk of systemic toxicity and antibiotic resistance development3,4,5.

Recurrent caries, known also as secondary caries, at the margins of dental restorations is the result of acid production by caries-causing bacteria that reside in the restoration-tooth interface is a major reason for dental restorative materials’ (fillings) failure, and may affect over 100 million patients a year in the US, at a cost of \$34 billion6,7,8,9,10,11. In addition to acid production, caries-forming bacteria have enzymatic activity that degrades the materials and the restoration-tooth interface and may contribute to the formation and progression of recurrent caries9,10. To mitigate the above processes, there has been on-going interest in introducing local antimicrobial chemotherapy at the restoration to prevent bacterial proliferation and caries formation12,13 since it is impractical to treat the local infection with systemic antimicrobial delivery because of limited drug accessibility to the site of infections that require high systemic doses10,14. In particular, we anticipate that antimicrobial particles included into the adhesive layer of dental fillings, applied at the restoration-tooth margins, could be designed to release antimicrobial drug in this local environment to delay biofilm progress over a clinically relevant timescale of years15.

An important factor for drug encapsulation particles to achieve efficient targeted and controllable drug release is high drug loading (volume % concentration) to maintain an effective concentration of released drug over long timescales since when drug reservoir is limited, the mass of released drug will also be limited at any given time as it is directly proportional to the total reservoir of drug16. In a biological analogy, secretory vesicles such as pancreatic insulin granules or platelet alpha-granules feature dense, high concentration packing of enzyme or growth factors in highly-organized volumes, which are then triggered to release their contents17. Synthetically, the high concentration loading of drug molecules into nanomaterials remains a challenge because it requires some degree of self-organization18. Degradable drug storage and release vectors such as liposomes, nano-gels, micelles and layer-by-layer systems can achieve a wide range of drug loading (from <1% to over 50%)19,20,21,22,23 for targeted drug delivery, and programmed release for tumor targeting and insulin release, with their assembly driven by a range of intermolecular forces and chemical reactions24. Lysosome-inspired polymersomes, for example, have been shown to sequester anticancer drug molecules at up to a 1:1 ratio of polymer to drug25. In another study, granule-like colloids of the drug fulvestrant, stabilized with targeting proteins, acted as their own nanoparticle without the use of a third encapsulating material26.

Drug-loaded particles for orthopedic implant coatings or dental restorative materials should also ideally have an inert, nondegradable scaffold which can maintain structural integrity following release of the drug so not to compromise the implantable device or restorations, and have predictable drug release rates4. Progress has been made in addressing these challenges for bone scaffold materials in addition to being osteoconductive and angiogenic, but long-term, predictable, and engineered release of antimicrobial from these materials remains difficult27. Highly ordered nanoporous inorganic scaffolds, such as mesoporous silica nanoparticles (MSNs), are ideal because they have ordered, identical channels (2–3 nm diameter), are structurally and chemically stable, biocompatible, highly porous (50 volume %), and remain intact after drug release28,29,30,31. MSNs are better suited than conventional sol-gel silica (or other inorganic) drug reservoirs, where pores are frequently not interconnected and the scaffold structure often degrades upon release32.

Examples of MSNs loaded with drugs for diffusional release include antimicrobials such as chlorhexidine, peptides, and biosurfactants16,33,34. Ordered mesoporous silica, often with hexagonally-packed channels, is formed through the condensation of a silica precursor in co-assembly with an organic surfactant template to form a porous mesostructure following the removal of the template (Fig. 1a)28,31,35,36,37. MSNs have sufficient mechanical strength for antimicrobial applications as filler particles for composites38, or as coatings by ‘evaporation-induced self-assembly’ deposition16. The templating agent may be a molecule that forms micelles and presents a hydrophilic interface for silica to condense around, such as surfactants and block co-polymers39,40.

We have developed a bio-inspired alternative to post-synthesis MSN drug loading, to self-assemble an antimicrobial drug (octenidine dihydrochloride, OCT) directly with silica to form an ordered drug/silica nanocomposite (mesostructure) (Fig. 1b). OCT is a cationic surfactant antiseptic that shows broad efficacy against gram positive and negative bacteria52. OCT has high biocompatibility, no known bacterial resistance, and is used currently as a mouth rinse, wound cleansing agent, topical antiseptic, and in other applications52,53,54. The self-assembly and micellar properties of OCT have not previously been investigated.

While there have been attempts to modify model drugs to be used as amphiphilic templates55, as well as using non-drug agents for templating mesostructured particles that release upon degradation of the particles34, this work represents the first example of a structurally-stable drug-MSNs for an existing, commercially available drug. Through incorporation with dental restorative adhesives or applied as an implant coating, these OCT/silica MSNs can target bacteria in the confined volumes of degraded restoration-tooth interface, where cariogenic bacteria reside10 and over implant surfaces. We hypothesize that OCT-templated silica mesostructured particles will contain significantly higher drug loading when compared with traditionally synthesized particles, and therefore could provide efficient, long term antimicrobial release from composite polymer materials and implant surfaces.

## Results

### OCT-MSN Physical analysis

We measured the critical micelle concentration (CMC) of OCT to be 3.79 mM by monitoring conductivity change as drug concentration was lowered. For comparison, the CMC of cetyltrimethylammonium is 0.90 mM56. A precipitation-based synthesis of drug-templated particles (OCT-MSNs) was carried out at varying concentrations of OCT in a basic aqueous solution using tetraethyl orthosilicate (TEOS) as a silica source. Traditional MCM-41 MSNs were also synthesised as a control, using cetyltrimethylammonium bromide (CTAB) as a pore templating agent, which was subsequently removed via 3 post-synthesis washes and calcination37. Loading of MCM-41 control was carried out through a solvent-evaporation process for a theoretical 40% wt. OCT in the particles. Scanning electron microscopy (SEM) showed the OCT-MSNs to be spherical particles with relatively high size monodispersity and average diameter of 424 ± 75 nm (high magnification of single particle in Fig. 2(a), example overview of population of particles used for measurement shown in Supplementary Figure S2). The as-prepared drug-loaded particles appear to have a slightly dimpled texture with no visible porosity. Transmission electron microscopy (TEM) showed the OCT-MSNs to appear solid, with little internal contrast immediately post-synthesis (Fig. 2b).

In Fig. 2(c), OCT-MSNs are shown after sonication in ethanol for 2 hours, with a rough, porous structure exposed, suggesting drug removal from the particles exposes the underlying porous silica. This drug-removal mechanism is corroborated by drug release detection in the ethanol using UV-vis and TGA of OCT-MSNs post-ethanol-sonication. In Fig. 2(d), a disordered porous structure is visible in all particles in samples that have had drug removed by ethanol sonication, due to increased contrast between silica and vacuum compared to silica and drug. These post-drug-release particles also appear identical in size and shape to as-synthesized particles, indicating that drug release is not dependent on degradation of the silica scaffold as for other sol-gel systems57, but rather by diffusion from pores.

SEM and TEM of OCT-loaded MCM-41 (Fig. 2e and f respectively) shows 71 ± 9 nm diameter particles with a highly ordered hexagonally packed porous structure. SEM images appear to show particles fused together by material. TEM images show a layer of material deposited on the exterior of these particles beyond the hexagonally packed pore structure. Since particles were washed and calcined prior to evaporative drug loading, it is presumed that this external material is drug crystalized on the surface of particles. This is in contrast with the uncoagulated OCT-MSNs that do not appear to possess this external layer.

Low angle powder X-ray diffraction (XRD) in Fig. 3(a) of OCT-MSNs reveals a broad diffraction peak corresponding to a d-spacing of 2.64 ± 0.11 nm, indicating a semi-ordered porous structure16,35,58,59,60. Interestingly, the diffraction peak shifts little as OCT concentration changes in the synthesis solution, but disappears completely when the concentration is brought below its CMC. These results clearly demonstrate that OCT micelle self-assembly is contributing to a mesophase formation. The XRD peak also remains after drug removal via sample calcination (550 °C in air), with d-spacing decreasing 0.23 ± 0.03 nm, further indicating that diffraction is due to a templated silica mesostructure and not due to crystalized drug. Shrinking d-spacing of about 10% is the result of increased silica condensation and densification at higher temperature and is expected for mesoporous silica40,61. The lack of a sharp XRD peak indicates the absence of a well-ordered mesophase. OCT, as a di-surfactant, may not favor assembly into a rigid hexagonally packed assembly of rods but rather a rapidly changing worm-like array of extended micelles62. The XRD spectra is consistent with other worm-like structures where disordered pores run continuously throughout the MSNs19,56.

Brunauer-Emmett-Teller (BET) analysis by N2 adsorption/desorption was used to determine pore size, volume, and total surface area39,40,56. Surface area for calcined OCT templated MSNs was found to be 856 m2 g−1 which is lower but comparable to that of the MCM-41 control (1098 m2 g−1) and other mesoporous materials (the theoretical value for smooth spheres of an equivalent size and density is just 13.1 m2 g−1). The pore volume was 0.47 cm3 g−1, or approximately 50.8% porosity, compared to 1.084 cm3 g−1 (70.4% porosity) for the MCM-41. Lower pore volume may be attributed to the less ordered mesophase of OCT-MSNs when compared to hexagonally closed-packed MCM-41 pores. The high open porosity of as-synthesized MCM-41 supports the notion that externally deposited material is drug condensed on the particle surface upon loading, and not material previously deposited during synthesis or template removal.

Density functional theory (DFT) analysis of BET isotherms in Fig. 3(b) shows that OCT-MSNs have a trimodal pore size distribution of approximately 1.4, 1.7 and 2.0 nm diameter, suggesting the presence of multiple mesophases within particles, helping account for prior “disordered” XRD results. The “tail-head-tail-head-tail” arrangement of OCT, whose corresponding theoretical straight tail lengths are approximately 1 nm, may allow for a broad range of conformations62 to maximize hydrophobic/hydrophilic and electrostatic interactions under dilute conditions63,64, and thus a distribution of pore diameters may be expected. The d-spacing is approximately twice the average pore size, suggesting an approximate 1:1 ratio of drug-filled pore and silica structure when viewed in cross-section, which corroborates the 50% porosity by BET analysis.

Energy dispersive x-ray (EDX) mapping was performed for Si and N to determine the spatial distribution of silica and the OCT phases (Fig. 3c,d). Very strong N signal was obtained from particles that were otherwise free of external debris or visible coating layer, providing evidence that the OCT micelles interpenetrate the silica network as a uniform mesostructure.

### Drug Release Kinetics of OCT-MSNs

Short-term drug release from OCT-MSNs or MCM-41 control at 0.02 mg mL−1 of particles suspended in phosphate buffered saline (PBS) (pH = 7.2, 22 °C) was monitored by UV-Vis absorption at 281 nm and compared with release from the MCM-41 control (Fig. 4a). Due to the difference in size between OCT-MSNs and MCM-41 control, the rate of the release alone is not a sufficient means of comparison, and we therefore must investigate the mechanism and model of release. Release from MSN pores has been previously found to follow the kinetics described by Higuchi for a spherical particle with a granular matrix containing a soluble compound, and may be approximated by the formula for release from a granular matrix plane for release below 50% of total drug loading45,49,55. In this model, cumulative release (Q) over time (t) is given by

$$Q=A\ast f({\rm{D}},\varepsilon ,\tau ,{\rm{S}},{{\rm{C}}}_{{\rm{s}}})\ast {t}^{0.5}$$
(1)

where A is the fixed spherical outer surface area of the silica particles, D is the diffusivity of drug in the solvent through matrix pores, ε is the porosity of the matrix (silica without templating drug), τ is a tortuosity factor to account for an increased diffusion path in non-linear pores, S is the solubility of the drug in the release media, and CS is the mass of drug per unit volume of matrix. These parameters may be simplified to k (units of µg t−0.5). Release from OCT-MSNs fits well to a t0.5 profile (R2 = 0.94), while release from OCT-MCM-41 deviates (R2 = 0.88) (supplemental material). The release of drug from MCM-41 has a release profile inconsistent with the Higuchi model during initial exposure to solvent which we suggest is due to the dissolution of the drug crystallized externally on and between clusters of particles, seen previously in SEM (Fig. 2e,f). Because OCT is confined to the pores of the OCT-MSNs, its release is limited by diffusion of solvent into pores and subsequent diffusion of drug out of the particle. This internal confinement avoids the initial uncontrolled burst of drug release typically seen with highly-loaded traditional mesoporous materials65.

### Micrographic Tracking of Drug Release

TEM micrographs of particles at various stages of an accelerated release, performed by sonication in ethanol, are shown in Fig. 5 (left). The particles part-way through release are visually dissimilar, with gradual changes of appearance that resemble those differences observed between the fully-loaded particle in Fig. 2(b) and the empty particle following the drug’s release in Fig. 2(d). This contrast is difficult to discern visually, but is most stark between 100%, 27%, and 0% loaded particles. An image analysis process was sought to quantitatively describe these differences and compare populations of particles.

Although drug release from a spherical particle may be modeled as release from a plane up to 50% (as before), the more accurate Higuchi modeling for release from a sphere is described by the relation between r′ and R, the unextracted radius of the sphere and the initial unchanging matrix radius, respectively49. The relationship between these values and the residual drug fraction in the sphere (Re) is

$${\rm{Re}}={(\frac{r^{\prime} }{R})}^{3}$$
(2)

### Retention of Activity of Released Antimicrobial

OCT is a well established effective broad-spectrum antimicrobial with a high degree of biocompatibility and no known formation of bacterial resistance52,66. However, analysis was performed to ensure that the stability and efficacy of OCT was preserved through the OCT-MSN synthesis process, and that released drug still functioned as an antimicrobial agent. Mass spectrometry was used to compare as-received OCT purity with OCT released from OCT-MSNs over 21 months under ambient conditions, with no difference in drug compound seen. Minimum inhibitory concentration (MIC) of as-received OCT and OCT-MSN-released OCT taken from the supernatant of the 8-day experiment above was performed at multiple concentrations against the cariogenic oral bacteria Streptococcus mutans UA15967. MIC was measured to be 2 µg mL−1 in both cases, in agreement with previous studies (between 1 and 32 µg mL−1 against a variety of pathogenic species in planktonic form) and further confirming that OCT was unchanged by the MSN synthesis process and will remain an effective antimicrobial agent66,68,69.

## Discussion

In these co-assembled materials, release rates are easily modeled due to the idealized drug loading within pores and lack of silica matrix breakdown during and after release, differentiating this process from a more typical amorphous sol-gel approach32. Other drugs exist with self-assembling and self-aggregating properties which could now be considered for mesoscale co-assembly to design a range of new drug encapsulation vehicles through bottom-up synthesis70,71,72.

Researchers have taken advantage of self-assembly of therapeutics and supporting carrier molecules, mimicking biological processes, to increase drug loading in carriers for some time, especially in the field of anticancer drug delivery18,20,73. These systems utilize intermolecular forces that cause components to arrange themselves in a predictable and repeatable manor, letting the drug delivery properties of the material be tuned to a high degree. We have taken this approach and expanded it to organic/inorganic co-assembly, introducing the benefits of inorganic systems unavailable from a strictly organic molecule approach. Previous work has shown the potential benefits of releasing the templating molecule in terms of kinetics and loading34,43, and that some drugs may be modified to self-assemble55. We believe our work builds on these studies by demonstrating this co-assembly for a drug that is unmodified and approved for clinical use. MSNs are an ideal reservoir material for drug release applications, especially when physical strength, consistent and predictable long-term release, and complete access to drug reservoir is desired. Work continues with the material developed here to incorporate OCT-templated and other drug-templated silica nanoparticles into antimicrobial polymeric restorative and implant coating systems, as well as continued investigation of the synthesis kinetics to optimize the co-assembly process. These underlying improvements in the drug-carrier particles could translate into improvements in therapeutic longevity in the prevention of secondary caries over previous antimicrobial MSN-resin composites and implant coatings25,33,74.

## Methods

### Chemicals

All chemicals were purchased from commercial sources and used without further purification. Type 1 ultra-pure water was used at 18.2 MΩ cm (Millipore Direct-Q system). Octenidine dihydrochloride (OCT) was purchased from TCI America (Portland, OR, USA). TEOS, ammonium hydroxide (29% weight in water) and CTAB were purchased from Sigma-Aldrich (Oakville, ON, Canada). Sodium hydroxide (10.0 N), hydrochloric acid (6.0 N), sodium chloride, potassium chloride, disodium phosphate, and monopotassium phosphate were purchased from Bioshop Canada Inc. (Burlington, ON, Canada). Anhydrous ethanol was purchased through the University of Toronto MedStore (house brand, Toronto, ON, Canada).

### Mesoporous Silica Nanoparticle preparation

Synthesis of OCT-MSNs was carried out in a 15 mL total volume with the molar ratios 150 H2O: 0.052 NaOH: 0.03 OCT: 1 TEOS. OCT was dissolved in water before adding NaOH. Solutions were stirred at approximately 750 RPM using a magnetic stir bar, while TEOS was added drop-wise over 30 s. Concentration of OCT was varied as per the experiment. Solutions were stirred for 30 min before being allowed to age for an additional 23.5 h. Solutions were centrifuged for 1 h at 10 kRPM and supernatant was removed, before adding water, vortexing for 30 s and centrifuging twice more to remove excess reactants. The recovered OCT-MSNs were dried for 24 h at 65 °C before being ground using a mortar and pestle. MCM-41 was produced following an identical procedure replacing OCT with CTAB and using the molar ratios 4000 H2O: 30.2 NH3: 0.125 CTAB: 1 TEOS. Recovered particles were washed in water using the same procedure as OCT-MSNs, calcined at 550 °C for 6 h, and suspended in OCT dissolved in ethanol for loading. The solutions were mixed and ethanol was allowed to evaporate at 37 °C.

### Critical Micelle Concentration Estimation of Octenidine Dihydrochloride by Conductivity Measurement

OCT was dissolved in ultra-pure water at 5 mM and lowered to 2 mM while monitoring conductivity (Thermo Scientific Orion VERSA STAR with conductivity module and Orion DuraProbe 4-Electrode Conductivity Cell). CMC was estimated as the point of inflection when conductivity was plotted against octenidine concentration.

### MSN Analysis

Electron microscopy was performed using field emission SEM, at 1 kV accelerating voltage (Hitachi SU8230), TEM at 300 kV accelerating voltage with EDX mapping and imaging (Hitachi HF3300), low magnification SEM for overview at 2 kV accelerating voltage (Hitachi SU3500), and TEM for imaging particles for micrographic release tracking study at 200 kV accelerating voltage (FEI Tecnai 20). TEM samples were prepared using ultrathin carbon on holy carbon copper grids. For all electron microscopy imaging techniques, particles were imaged as-is, with no further preparation. XRD was carried out from 1 to 6° using a beam energy of 30 kV and current of 10 mA from a Cu K α radiation source (Rigaku MiniFlex 600). BET analysis was carried out with density functional theory data analysis (Autosorb-1-C and Quantachrome software version 2.11). TGA was carried out using a 10 °C min−1 ramp to 550 °C (TA Instruments Q50 TGA). MS identification and confirmation of OCT purity was carried out using ultra high-performance liquid chromatography combined with mass spectrometry (Waters Xevo G2-XS QTof). OCT was dissolved in MS-grade methanol (Thermo Fisher Scientific, Mississauga, ON, Canada) for the pre-synthesis control. OCT-MSNs stored in ambient temperature and humidity for 21 months post synthesis and washing were suspended in MS-grade methanol, allowed to settle, and a fraction was taken and diluted for analysis.

### OCT Release Analysis

MSNs were suspended in PBS prepared via the Cold Spring Harbor protocol and adjusted to a pH of 7.2 through the addition of HCl. OCT concentration in the media was monitored either continuously by fiber-optic probe, or cuvette measurements taken after centrifuging for 1 h at 10 kRPM a 2 mL aliquot from the sample, at λmax = 281 nm (Agilent Cary 60 UV-Vis spectrophotometer). Aliquots were re-suspended and returned to the release sample after reading.