The bioreactor for biofilm mineralization adapts the design of ‘artificial mouth’ models used to simulate caries formation (Schwendicke et al., 2014) (Figure 1, bottom). Specimens made of a dental composite restoration material were first inoculated in a suspension of E. coli K-12 strain W3110 (Hayashi et al., 2006). This strain expresses periplasmic ALP (also known as PhoA) (Hayashi et al., 2006) and it enabled us to propose a mechanism for bacterial biofilm mineralization, which requires ALP activity (Zorzetto et al., 2023). Dental composite material was chosen both for availability and for the possibility of easily shaping and light curing. In the mouth, dental plaque grows on the surfaces or dental restorations, which are always covered with saliva. To simulate such growth conditions, dental composite specimens previously seeded with the bacteria were placed in a bioreactor chamber with continuous drop-wise supply of nutritious media for the duration of the biofilm growth and mineralization experiment from days to weeks (5 days in this proof of concept). The bioreactor chamber was designed using a computer-aided design (CAD) software and produced through an additive manufacturing technique (fused-deposition modeling) using a polymer that could withstand autoclaving (for more detailed description refer to the corresponding paragraph in material and methods). Two media were used: a control medium with nutritious substances and a mineralizing medium consisting in control medium supplemented with CaCl₂ and β-glycerophosphate, in concentrations sufficient to lead biofilm mineralization on agar substrates (Zorzetto et al., 2023). The spontaneous precipitation of mineral at these concentrations was excluded after incubating this liquid mineralizing medium alone for 3 days and detecting no sign of aggregate formation with calcein green (Zorzetto et al., 2023). Further details of the bioreactor assembly are reported in the Materials and Methods section.

Specimens were made of commercial dental composite shaped in a polydimethylsiloxane (PDMS) mold and, among other methods possible to shape the composite, we opted for a cost-effective and easily reproducible process. The PDMS 2-side-mold was fabricated from a master mold consisting in Teflon surrounded by a LEGO™ dam (Figure 2A). The Teflon and the LEGO™ were coated with vacuum pump grease to facilitate the silicone release. For the same reason, the dam was made with modular pieces that could be easily taken apart during de-molding. The chosen specimen geometry consisted in a square plate. In the center of the square, 9 pits were drilled in a 3 × 3 pattern (Figure 2B). This design is the result of an optimization process during which we started from composites disks with different surface finishes, but these specimens did not grant a predictable biofilm distribution. We discuss further their characteristics and limitations in the Supplementary Information (Supplementary Figure S1). To obtain the desired composite specimen geometry, the uncured dental composite was placed in the bottom part of the PDMS mold and flattened with a spatula (Figure 2C). Then, the mold was closed with its upper part and pressed to even out the paste. The dental composite paste was cured with a UV lamp, shining the blue light 30 s per side (Figure 2D). The pits were then drilled in the samples with a dental drill, using a metal guide to ensure repeatability in the pit size and position (Figure 2E). The guide, which had holes in the same position as the desired pits in the specimen, was placed on top of each composite specimen and the drill tip would go through the mask holes until a pre-set mark to create corresponding pits of controlled depth in the specimen. This shape had the double function of encouraging the accumulation of liquid nutrients (and thus of bacterial tissue) in a specific region, and of avoiding evaporation during the microscopy (Figure 2F). To visualize the distribution of biofilm and mineralization on the specimens’ geometrical features, we took pictures of the samples using fluorescence stereomicroscopy after 5 days of incubation in the bioreactor. From left to right, Figure 2G shows two representative pits in reflected light (bright field), and in green and red fluorescence, which respectively report the presence of mineral (calcein green) and of bacteria (mCherry). In this case, both bacteria and mineral were present mainly in the pits, indicating a success in confining bacteria accumulation and subsequent biofilm growth to a limited region. Thanks to the versatility of these composites more complex shapes can be produced and adapted to different scenarios.

When biofilms grow in a wet environment, the water is often loaded with proteins and biopolymers that tend to adsorb on the substrate. For example, once saliva gets in contact with dental hard tissues a thin organic film, the dental pellicle covers the surfaces. This structure, called the dental pellicle mainly consists of proteins, glycoproteins, lipids and other organic compounds (Rasputnis et al., 2021). Despite the difference in biological function and molecular composition, about 30% of salivary proteins were observed in plasma and gene ontology analysis revealed overlapping in the distribution of plasma and saliva proteomes (Yan et al., 2009). Here, we opted for a pre-treatment of the composite specimens with sterile-filtered saliva to ensure the formation of a pellicle on their surface. To test the influence of this salivary pellicle on E. coli biofilm growth, we performed a separate experiment where we pre-conditioned three flat rectangular composite specimens in filtered saliva before incubating them with E. coli W3110 mCherry in nutritious medium. The samples were autoclaved and then completely submerged in saliva in a falcon. The dental composite did not undergo macroscopically visible changes after sterilization apart from the formation of superficial stains probably related to water evaporation. The falcon was tapped to ensure the removal of potential air bubbles. The residual saliva was not washed away from the samples before incubating them in the medium. Three control samples were directly incubated in the bacteria suspension without pre-conditioning (Figure 3A). Pictures were respectively taken with the stereomicroscope 24, 48 and 72 h after starting the incubation (Lagendijk et al., 2010; Zorzetto et al., 2023). The fluorescent bacteria were used to localize their presence; whereas thioflavin S was added to the medium to mark the production of extracellular matrix (ECM). Figures 3B, C show that this step did not lead to a significant change in ECM production, but it also did not interfere with the fluorescent dyes (Supplementary Figure S2). The decline in brightness at 72 h could be due to the bleaching of the fluorescent dye, because of the repeated pictures taken on each same sample. On the other hand, the bacteria signal tends to decline for the pre-treated samples compared to the non-treated samples (Figure 3D).

This could be related to the antibacterial properties of enzymes potentially present in the saliva (Hannig, Hannig, and Attin, 2005). Yet, this effect is not visible in the values of the optical density (OD600) of the culture medium (Figure 3E). Overall, the pellicle does not appear to affect dramatically the growth of E. coli, but it could be fundamental for further developments of the system employing oral bacteria.

To make this in vitro model for bacterial biofilm mineralization on rigid substrate amenable to screening approaches, it is essential to facilitate compositional and structural characterizations of the resulting samples. We thus established the possibilities to image the biofilms with two complementary imaging techniques: confocal laser scanning microscopy (CLSM) and scanning electron microscopy (SEM).

CLSM first provided an overview of the bacterial tissue with its different phases: bacteria, extracellular matrix and mineral. To identify them, we exploited the E. coli fluorescent strain producing mCherry fluorescent proteins in their cytoplasm and two fluorescent dyes provided in the liquid nutritious medium: calcein green and calcofluor white for the mineral and the ECM, respectively. In this case, calcofluor white was substituted to thioflavinS, aiming to avoid spectral overlapping with calcein green. Controls were made to check the possible crosstalk between the fluorescent dyes (Supplementary Figure S3). Samples were put upside-down on a glass coverslip and imaged without any treatment, to avoid possible artifacts from drying or fixation (Figure 4A). All the samples showed the presence of bacteria and extracellular matrix, but only samples grown in the mineralizing medium showed minerals (Figures 4B, C). Such results are in accordance with the ones obtained during sample geometry optimization, where we could show that the mineralizing medium leads to bacteria-induced mineral accumulation on the sample surface (Supplementary Figure S1).

SEM was then used to visualize the biofilms at the bacteria length-scale. Samples were observed after fixation in 70% ethanol and subsequent dehydration cycles until 100% ethanol. They underwent critical point drying (5 washing cycles) and gold coating. The pictures show the topology of the bottom of the sample pits (secondary electrons). The samples incubated in the control medium showed a mat of agglomerated bacteria forming a thin layer: indeed, the composite material on top which the biofilm grows was still visible under the bacteria (Figure 4D, arrow). For a direct comparison, the inset shows a region of bare dental composite. In contrast, the biofilms cultivated in mineralizing medium formed much thicker clusters of mineralized bacterial biofilm than their counterparts (Figure 4E, arrowheads). Additionally, there were also conglomerates of about 20 µm that hint towards mineralization (Figure 4E, circle).

The bioreactor chamber was designed to host four specimens in separate compartments to minimize the risks of cross-contamination. The entire bioreactor consisted in a bottom part where the specimens were placed and a cover through which the nutrients were deposited on the specimen’s surface (Supplementary Figure S4). To achieve this, we drew it using Fusion 360 (Autodesk, United States), a CAD software that is open source for non-commercial use. After exporting the two parts, we produced them with additive manufacturing: we used a fused deposition modeling printer with a 0.4 mm brass nozzle (Prusa MK3S from Prusa Research, Czech Republic) and a commercial filament resistant at high temperature (GreenTec Pro from Extrudr, Austria). The three-dimensional model was manufactured with a 0.1 mm layer height and the internal volume of the pieces were filled at 15% using gyroid shapes. The temperature resistance was necessary because the bioreactor chamber requires autoclaving before each experimental iteration.

For each iteration, four specimens made from the commercial dental composite Tetric Ceram (Ivoclar Vivadent, Liechtenstein) were manufactured as illustrated in Figure 2. They were shaped in a PDMS (Sylgard 184 silicone, Dow Europe GMBH c/o Dow Silicones, Germany) mold (monomer to crosslinking agent ration of 10:1) and cured with a dental blue light lamp (Valo Cordless; ULtradent, Köln, Germany). PDMS was chosen because it can be easily cleaned and it is transparent and thus ensured the polymerization of the dental composite through blue light. After fabrication, the specimens were autoclaved and subsequently incubated in filtered saliva for 1 h at 37°C shaking at 250 rpm. Then, the specimens were separately transferred without washing in 50 mL falcon tubes filled with 5 mL of LB medium (Luria/Miller) (Roth X968) and each vial was inoculated with a different single E. coli microcolony. The transfer was made under sterile hood using sterilized tweezers. The E.coli K-12 strain W3110 was chosen as a well-characterized model strain periplasmic alkaline phosphatase (Hayashi et al., 2006). To visualize the bacterial cells using fluorescence microscopy, the bacteria were transformed with the plasmid pMP7604 (TetR; obtained from Guido V. Bloemberg, University of Zurich) that carries the gene for the fluorescent protein mCherry (Lagendijk et al., 2010).

The specimens in the inoculated LB medium were left overnight at 37°C shaking at 250 rpm. Subsequently each seeded specimen was inserted in the bottom part of the bioreactor chamber using sterilized tweezers. To assess the potential viability difference between the different colonies, we measured the optical density (OD600) of the leftover nutritious medium used for the overnight culture and it was consistently around 0.5 after a 1:10 dilution in culture medium with a maximum 10% variability. E. coli biofilms were cultivated at the solid-air interface on the surface of dental composite specimens in conditions of 100% humidity. The nutrients necessary for microbial tissue growth were provided by continuously dropping liquid medium on the specimens from silicone tubing passing through the top part of the bioreactor chamber. Excess medium was removed from the bottom of the chamber to a waste container. Both tubes coming in and out the chamber had been autoclaved and were connected to peristaltic pumps to allow the nutrient to flow from the reservoirs to the chamber (0.25 mL/min) and from the chamber to the waste (0.5 mL/min) for 5 days at 37°C in aerobic conditions. The double flow rate of the waste collecting pump avoids fluid accumulation in the chamber. The length of the tubes was determined beforehand, placing empty reservoir bottles, pumps, empty bioreactor and waste. This prevented to have tubes that are too short to connect the components, or too long and risking entanglement. The tube connection was tested with sterile water in the days before the actual use to ensure that there was no leakage in the various connections. Both pumps were calibrated the day before the bioreactor assembly using the procedure recommended by the manufacturer to ensure that the desired flow was achieved and maintained during the experiment. The bioreactor CAD drawings are available upon request to the corresponding author (“Artificial_mouth_4samples.f3d” file).

Salt-free LB medium (Luria/Miller) reservoirs (control medium) were prepared with 1% w/v tryptone (Roth 8952) and 0.5% w/v yeast extract (Roth 2363) dissolved in double distilled water. For the mineralizing media, calcium ions were added in the form of CaCl₂ (Sigma Aldrich 223506). Sodium ß-glycerophosphate (Sigma Aldrich 35675) was added to the LB medium as an organic source of phosphates. CaCl₂ and sodium ß-glycerophosphate stock solutions were sterile-filtered (0.22 µm) and added to the autoclaved salt-free LB medium to reach a final concentration of 10 mM ß-glycerophosphate and 10 mM CaCl₂. These CaCl₂ and ß-glycerophosphate concentrations had lead previously to mineralization in E. coli biofilms cultivated on LB agar (Zorzetto et al., 2023).

Fluorescent stains were added to both media in order to identify the mineral accumulations and the extracellular matrix (ECM). For the mineral, a calcein green (Merck KGaa, 102315) stock solution (1 mM in 10 mM NaOH) was supplemented to the liquid LB media to reach a final concentration of 4 µM. Note that this type of calcein reports extracellular calcium. For the ECM, a calcofluor white (Sigma-Aldrich, 18909) stock solution (calcofluor white M2R, 1 g/L and Evans blue, 0.5 g/L) was added to the liquid media reservoirs to reach a final concentration of 3 mg/L of calcofluor white M2R and 1.5 mg/L Evans blue. Both stain stock solutions were sterile-filtered (0.22 µm) before adding them to the media reservoirs.

Apart from the samples used for confocal microscopy, all biofilms were fixed with 4% paraformaldehyde (PFA, Bioster Ar106) in phosphate-buffered saline (PBS, Sigma-Aldrich P4417). The samples were left in contact with the fixing solution for 2 h at 37°C at 60 rpm. The samples were rinsed with the aforementioned PBS solution to remove residual PFA. The experiment was repeated three times with two specimens per conditions, for a total of six specimens per condition.

After autoclaving, three composite specimens were incubated fully submerged in filtered saliva for 1 h at 37°C shaking at 100 rpm and subsequently put in contact with nutritious medium (20 mL) inoculated with a single colony of E. coli mCherry. As control, three additional autoclaved specimens were put directly in the inoculated medium. The medium consisted in 1% trypton, 0.5% yeast extract and Thioflavin S (Merck, T1892; 2 mg/mL in 70% ethanol), a fluorescent marker for biofilm extracellular matrix (ECM) used at a final concentration of 40 μg/mL. Both sample groups were cultivated for a total of 72 h and image at 24, 48 and 72 h of growth with an AxioZoomV.16 fluorescence stereomicroscope (Zeiss, Germany). At each time-step, optical density of the cultivated medium was measured. The fluorescence signal from thioflavin S was collected with the 38 HE green fluorescent protein filter set (excitation: 470/40 nm; beam splitter: FT 495 nm; emission 525/50 nm; exposure time: 225 ms). The fluorescence from mCherry was detected using the 63 He red fluorescent protein filter set (excitation: 572/25 nm; beam splitter: FT 590 nm; emission: 629/62 nm; exposure time: 500 ms).

For each channel, sample average intensity was estimated using custom-written MATLAB codes (Matlab 9.7.0 R2019b, MathWorks, Natick, MA). Fluorescence images were imported and a threshold was used to define a region of interest (ROI) corresponding to the composite surface using the function “im2bw”. This was possible thanks to the high contrast between the composites and the background. Average intensity was calculated per each channel as the mean of the intensity values of the gray scale of the original pictures in the ROI. The values are reported between 0 (black) and 1 (white).

To test the different specimen geometries, pictures were taken from air dried samples cultivated for 5 days in the bioreactor with an AxioZoomV.16 fluorescence microscope (Zeiss, Germany). Images were taken in reflected light mode with an exposure time of 15 ms. The fluorescence signal from Calcein Green was collected with the 38 high efficiency (HE) green fluorescent protein filter set (excitation: 470/40 nm; beam splitter: 495 nm; emission 525/50 nm). The exposure time was 225 ms. mCherry fluorescence was detected using the 63 He red fluorescent protein filter set (excitation: 572/25 nm; beam splitter: 590 nm; emission: 629/62 nm). The exposure time was 500 ms.

Confocal LSM700 microscope equipped with ×10, ×20 and ×50 objectives (Zeiss, Germany) was used to image untreated samples after 5 days cultivation in the bioreactor. Calcofluor white, calcein green and mCherry gene were excited using 408 nm, 488 nm and 555 nm laser lines, respectively.

Phenom Pro Desktop SEM (Thermofisher, Waltham, Massachusetts, United States) was used to image dried and gold-coated samples. Samples underwent ethanol dehydration with an increasing gradient of ethanol (70%, 80%, 90% and 100%), leaving the samples for 24 h in each ethanol solution. After the ethanol dehydration, samples were transferred in 100% acetone overnight before undergoing 5 cycles (10 min each) of critical point drying (CPD). Eventually, the dried samples were sputtered with gold. SEM parameters were: voltage intensity of 15 kV and pressure of 60 Pa.

To evaluate the pre-treatment with saliva, we carried on the experiment in triplicates and we measured the average fluorescent signal for the extracellular matrix and the bacteria, as well as the optical density of the liquid cultivation medium for each sample, at each measuring time point (24 h, 48 h and 72 h after the first inoculation). To verify the statistical significance, we performed a one-way ANOVA test under the hypothesis of normal distribution considering the significance level equal to 1%.