Mosaic-CLSM Assessment of Bacterial Spatial Distribution in Cosmetic Matrices According to Matrix Viscosity and Bacterial Hydrophobicity

: The reliability of the challenge test depends, among other parameters, on the spatial distribution of microorganisms in the matrix. The present study aims to quickly identify factors that are susceptible to impair a uniform distribution of inoculated bacteria in cosmetic matrices in this context. We used mosaic confocal laser scanning microscopy (M-CLSM) to obtain rapid assessment of the impact of the composition and viscosity of cosmetic matrices on S. aureus spatial distribution. Several models of cosmetic matrices were formulated with di ﬀ erent concentrations of two thickeners and were inoculated with three S. aureus strains having di ﬀ erent levels of hydrophobicity. The spatial distribution of S. aureus in each matrix was evaluated according to the frequency distribution of the ﬂuorescence values of at least 1350 CLSM images. We showed that, whatever the thickener used, an increasingly concentration of thickener results in increasingly bacterial clustered distribution. Moreover, higher bacterial hydrophobicity also resulted in a more clustered spatial distribution. In conclusion, CLSM-based method allows a rapid characterization of bacterial spatial distribution in complex emulsiﬁed systems. Both matrix viscosity and bacterial surface hydrophobicity a ﬀ ect the bacterial spatial distribution which can have an impact on the reliability of bacterial enumeration during challenge test. Bacterial cells were harvested by centrifugation at 1575 × g for 10 min, washed twice in 150 mM NaCl, and resuspended in several phosphate bu ﬀ ers with a pH from 3 to 8. Each measurement recorded the movements of approximately 100 individual bacterial cells by automated video analysis. Each experiment was performed at least six times with two independent cultures.


Introduction
Controlling the contamination of cosmetic products is essential to avoid alterations of their organoleptic characteristics (color, odor, composition, etc.) and maintain their innocuity for consumers. These products can be contaminated during the processing or by consumer's repetitive usage. Among other pathogens, Staphylococcus aureus has been found in various cosmetic products [1,2]. This Gram-positive bacterium is present on the skin and mucous membranes of 30% of the population [3] and causes a wide range of skin infections such as abscesses, furuncles, impetigo, etc. [4,5].
Each cosmetic product is characterized by a level of microbiological risk according to the standard ISO 29621:2017, which depends on several parameters such as the formula composition (preservative, ethanol, water activity, pH) or the type of packaging (unidose, airless pump, pots) [6,7].
The preservation efficiency of a cosmetic product is evaluated by a challenge test, as defined in the European standard EN ISO 11930:2019. This procedure consists in artificially inoculating a product with a target microorganism at a defined concentration that is between 10 5 and 10 6 CFU mL −1

Bacterial Strains and Culture Conditions
The strains used in this study were S. aureus CIP 4.83, S. aureus ATCC 29213 and S. aureus ATCC 27217. S. aureus CIP 4.83 is a strain recommended by the EN ISO 11930:2019 standard for cosmetic-product challenge tests. Each strain, stored in cryovials at −80 • C, was resuscitated by two successive subcultures in tryptic soy broth (TSB) before each experiment. Cultures were grown at 30 • C until the end of the exponential growth phase.
The bacterial inoculum was calibrated to 10 10 to 10 11 CFU.mL −1 in 150 mM NaCl in order to observe at least 10-100 bacteria per CLSM image (290.62 × 290.62 × 1.6 µm 3 ). S. aureus cells were labeled with 2 µL Syto 9 (3.34 mM in DMSO, Life Technologies) and incubated in the dark for 1 h at room temperature. Syto 9 is a green cell-permeable nucleic-acid marker that stains live and dead bacteria. One gram of matrix in a 50 mL centrifuge tube was inoculated with the stained inoculum and vortexed for 30 s.
When necessary, 200 µL of latex green fluorescent beads latex (L4655, carboxylate-modified polystyrene, 1 µm) from Sigma-Aldrich were added to the model matrices at a concentration of 4.75 × 10 10 beads mL −1 .

Bacterial Surface Hydrophobicity
S. aureus surface hydrophobicity was measured by bacterial adherence to hydrocarbons (BATH) [30]. Bacterial cells were harvested by centrifugation at 1575× g for 10 min and resuspended to an OD 400 of 0.8 in 150 mM NaCl, precisely measured and noted as A 0 . Hexadecane (0.25 mL) was added to 1.5 mL of bacterial suspension. The two-phase system was vortexed for 2 min and allowed to separate for 15 min at room temperature before the OD 400 was measured in the aqueous phase (A 1 ). The percentage of solvent affinity was calculated as: the percentage of solvent affinity = (1 − A 1 /A 0 ) × 100. Each experiment was performed in triplicate with independently grown cultures.

Bacterial Electrophoretic Mobility
The bacterial electrophoretic mobility (EM) was measured by determining the rate at which bacteria migrate in a constant electric field (120 mV). EM was assessed using a Zetameter ZetaCOMPACT TM from CAD instrumentation (Les Essarts le Roi, France). Bacterial cells were harvested by centrifugation at 1575× g for 10 min, washed twice in 150 mM NaCl, and resuspended in several phosphate buffers with a pH from 3 to 8. Each measurement recorded the movements of approximately 100 individual bacterial cells by automated video analysis. Each experiment was performed at least six times with two independent cultures.

Commercial Matrices
Six commercial matrices (Laboratoires CLARINS, Pontoise, France), named from 1 to 6, were selected to represent the main emulsified cosmetic products, such as body lotions or creams.

Preparation of the Emulsified Model Matrix (35/65 O/W%)
The aqueous phase was first prepared with 0.25% carbomer in water, with the addition of acrylate copolymer (0.4% to 1.6%) when necessary and heated to 75 • C before glycerin (moisturizer, 9%) was added. The oil phase, composed of 28.8% cetearyl isononanoate (emollient), 3.5% cetearyl glucoside (emulsifier), 0.2% tocopheryl acetate (antioxidant) and 2.5% glyceryl stearate (co-emulsifier), was heated to 75 to 80 • C before being blended with the aqueous phase at 1800 rpm using a rotor-stator homogenizer (Rayneri 33/300P, Group VMI) to obtain an emulsion. Tromethamine (base, 0.15%) was finally added. Guar gum was pre-mixed with glycerin and added to a final concentration of 2.5% when necessary.

Viscosity Measurements
The viscosity of industrial or model matrices was measured by penetrometry using a penetrometer (PNR10, PetroMesures). A specific cone was released in 300 g of matrix and the penetration depth measured (in mm ± 0.1 mm) after 5 s: the deeper the cone sinks into the matrix, the softer the matrix. Three replicates were performed for each measurement.

Determination of Lipid Droplet Electrophoretic Mobility
Emulsified matrices, with or without acrylate copolymer, were diluted 1500 times in appropriate phosphate buffer prior to analysis. Electrophoretic mobility (EM) was measured at pH 4 to 6 using a ZetaCOMPACT TM Zetameter (CAD instrumentation; Les Essarts le Roi, France).

Enumeration of the Bacterial Population
Just after bacterial inoculation, 15 samples were harvested from different locations of the contaminated matrix and enumerated by serial dilution. Enumeration was performed on tryptone soya agar (TSA, BioMérieux) by the drop-plate method [31]. Plates were incubated at 30 • C for 24 to 48 h before counting. Each experiment was performed in triplicate. The ratio of recovery is the log CFU/g divided by the log inoculated CFU/g. The enumerated bacterial population was similar to the level of inoculation if the ratio of recovery is equal to 1 ( Figure 3A).

Mosaic-Confocal Laser Scanning Microscopy (M-CLSM) and Image Analysis
The inoculated matrix was dropped into the wells of polystyrene 96-well microtiter plates (Greiner Bio-One, France). Image acquisition was performed using a Leica SP8 AOBS Confocal Laser Scanning Microscope (Leica Microsystems, France) at the INRAE MIMA2 microscopy platform (https://doi.org/10.15454/1.5572348210007727E12). Syto 9 and Nile red were excited at 488 nm and 561 nm and the emitted fluorescence collected from 498 to 560 nm and 630 to 750 nm, respectively.
Each image was acquired using a 40× air objective (N.A. = 0.85) and its size was 290.62 × 290.62 × 1.6 µm 3 (512 × 512 pixels). The CLSM control software was programmed to take in each well a mosaic of 15 × 15 images, each image corresponding to an observed volume of 1.3 × 10 −7 mL. Acquiring mosaics instead of single images allows to visualize bigger volumes of matrix and to decrease the detection threshold per gram. For commercial matrices, a mosaic was acquired in three well for each experiment and experiments were performed in duplicate (1350 images in total for each condition). For model matrices, a mosaic was acquired in 5 wells for each experiment and experiments were performed in triplicate (3375 images in total for each condition).
The percentage of surface fluorescence (SF) for each image was calculated by binarizing the images using the MaxEntropy function in an automatic macro executed in the ImageJ software, (National Institutes of Health, Bethesda, Md., USA) [32]. In our experimental conditions, the range of SF that can be visualized on one image is between 0.0005% (one bacteria) to 100% (over 2 × 10 5 bacteria).
The equivalent CFU mL −1 for each image acquired by CLSM was calculated first by dividing the SF of the image by the SF of one bacterium leading to a number of bacteria per image. Bacterial concentration was then calculated in bacteria per g taking into account the volume observed for one image and the matrix density (1.15 g/mL). The ratio of recovery is the equivalent log bacteria/g divided by log inoculated CFU/g.

Statistical Analysis
The bacterial spatial distribution was quantified by calculating the frequency distribution, the mean and the variance of the fluorescence values. A variance superior, equal or inferior to its mean indicates a clustered, uniform or regular spatial distribution, respectively [15].
All statistical analyses were performed using a free version of XLSTAT (Addinsoft). Factors were considered to have a statistically significant effect for a critical probability associated with the Fisher test p value lower than 0.05.

Bacterial Surface Hydrophobicity
The three S. aureus strains were chosen according to their hydrophobicity to evaluate the impact of bacterial surface properties on bacterial distribution in matrices. S. aureus CIP 4.83 was the most hydrophobic strain (95% affinity for hexadecane), whereas S. aureus ATCC 292312 and ATCC 27217 had 76% and 42% affinity, respectively. We have also tested the behavior of fluorescent polystyrene microspheres. These inert beads have similar morphology to that of S. aureus (sphere of 1 µm in diameter) but are more hydrophilic than the tested bacteria (5% affinity for hexadecane).

Bacterial Surface Charge
The electrophoretic mobility of the three strains was measured at a pH from 3 to 8 ( Figure 1A). All strains were negatively charged at an ionic strength of 1 mM. Their electrophoretic mobility was near −3 µm s −1 V −1 cm −1 at pH 3, increased to approximatively −1.5 µm s −1 V −1 cm −1 at pH 6, and remained constant from pH 6 to 8. There was no significant difference in electrophoretic mobility between the three strains.

Bacterial Surface Hydrophobicity
The three S. aureus strains were chosen according to their hydrophobicity to evaluate the impact of bacterial surface properties on bacterial distribution in matrices. S. aureus CIP 4.83 was the most hydrophobic strain (95% affinity for hexadecane), whereas S. aureus ATCC 292312 and ATCC 27217 had 76% and 42% affinity, respectively. We have also tested the behavior of fluorescent polystyrene microspheres. These inert beads have similar morphology to that of S. aureus (sphere of 1 µm in diameter) but are more hydrophilic than the tested bacteria (5% affinity for hexadecane).

Bacterial Surface Charge
The electrophoretic mobility of the three strains was measured at a pH from 3 to 8 ( Figure 1A). All strains were negatively charged at an ionic strength of 1 mM. Their electrophoretic mobility was near −3 µm s −1 V −1 cm −1 at pH 3, increased to approximatively −1.5 µm s −1 V −1 cm −1 at pH 6, and remained constant from pH 6 to 8. There was no significant difference in electrophoretic mobility between the three strains.  Model matrices formulated with 0%, 0.4%, 0.8% or 1.6% acrylate copolymer had penetrometry values of 31.5 ± 0.4 mm, 28.5 ± 0.3 mm, 27.7 ± 0.4 mm and 26.6 ± 0.3 mm, respectively. Those with 2.5% guar gum had a penetrometry value of 25.5 ± 0.4 mm.

Matrix Droplet Electrophoretic Mobility
Model matrices had a pH between 5 and 6. The mobility of droplets was near zero at pH 4 and slightly decreased from pH 5 to 6 for matrices with or without acrylate copolymer ( Figure 1B). There was no significant difference between the electrophoretic mobility of droplets in either matrix.

Bacterial Spatial Distribution in Commercial Emulsions
We assessed the spatial distribution of S. aureus CIP 4.83 in six commercial cosmetic matrices of various viscosities by CLSM imaging to identify the formulas in which uniform homogenization is not complete. Figure 2A,B shows the frequency distribution of surface fluorescence (SF), the mean and the variance for two of the six matrices taken as an example. For matrix 1, the variance of the SF was inferior to the mean, which indicates a bacterial regular distribution. Figure 2C illustrates the regular distribution in matrix 1 showing one series of 225 images. Matrices 2 to 5 showed a similar regular distribution with SF means between 0.03% and 0.45% (data not shown). The matrices 1 to 5 had low viscosities, with penetration values superior to 32 mm. In contrast, matrix 6 showed a bacterial clustered distribution with a variance of SF superior to the mean ( Figure 2B

Bacterial Spatial Distribution and Enumeration in Matrices with Thickener
From the composition of matrix 6, several assays were conducted to identify the component that has impaired the regular distribution of S. aureus in the matrix. It appears to be the acrylate copolymer (data not shown). To confirm the impact of its presence, we formulated two emulsified model matrices, without or with acrylate copolymer at 1.6% and evaluated the ratio of recovery between the initial inoculum and by plate count (Figure 3A) or by M-CLSM enumeration ( Figure 3B). Bacterial enumeration for both methods was significantly more variable in matrices with 1.6% acrylate copolymer than in those without acrylate copolymer (p < 0.05). We can note a higher variability for enumeration by CLSM imaging than by plate count. This is due to the substantially smaller volume analyzed by CLSM (1.3 × 10 5 µm 3 ) which is equivalent to 1.5 × 10 −7 g in comparison of 1 g by plate count.

Bacterial Spatial Distribution and Enumeration in Matrices with Thickener
From the composition of matrix 6, several assays were conducted to identify the component that has impaired the regular distribution of S. aureus in the matrix. It appears to be the acrylate copolymer (data not shown). To confirm the impact of its presence, we formulated two emulsified model matrices, without or with acrylate copolymer at 1.6% and evaluated the ratio of recovery between the initial inoculum and by plate count (Figure 3A) or by M-CLSM enumeration ( Figure 3B). Bacterial enumeration for both methods was significantly more variable in matrices with 1.6% acrylate copolymer than in those without acrylate copolymer (p < 0.05). We can note a higher variability for enumeration by CLSM imaging than by plate count. This is due to the substantially smaller volume analyzed by CLSM (1.3 × 10 5 µm 3 ) which is equivalent to 1.5 × 10 −7 g in comparison of 1 g by plate count. We further investigated the impact of the concentration of thickener by formulating four model matrices with 0%, 0.4%, 0.8% or 1.6% acrylate copolymer and inoculating them with three different S. aureus strains or fluorescent beads. The frequency distribution of surface fluorescence was more dispersed with increasing acrylate copolymer concentration, regardless of the strain. The variance was superior to the mean for the 0.8% and far more for the 1.6% acrylate copolymer resulting in a bacterial clustered distribution (Table 1). Without acrylate copolymer or with 0.4%, the variance was inferior to the mean resulting in a regular distribution. The results were similar for hydrophilic beads which were regularly distributed in matrices with 0% to 0.4% acrylate copolymer, in contrast to 0.8% and 1.6% (Table 1). The Figure 4 illustrates the frequency distribution according to the concentration of acrylate copolymer or guar gum for S. aureus CIP 4.83.  We further investigated the impact of the concentration of thickener by formulating four model matrices with 0%, 0.4%, 0.8% or 1.6% acrylate copolymer and inoculating them with three different S. aureus strains or fluorescent beads. The frequency distribution of surface fluorescence was more dispersed with increasing acrylate copolymer concentration, regardless of the strain. The variance was superior to the mean for the 0.8% and far more for the 1.6% acrylate copolymer resulting in a bacterial clustered distribution (Table 1). Without acrylate copolymer or with 0.4%, the variance was inferior to the mean resulting in a regular distribution. The results were similar for hydrophilic beads which were regularly distributed in matrices with 0% to 0.4% acrylate copolymer, in contrast to 0.8% and 1.6% ( Table 1). The Figure 4 illustrates the frequency distribution according to the concentration of acrylate copolymer or guar gum for S. aureus CIP 4.83.  We intend next to determine if the bacterial clustered distribution is due to the viscosity which increases when the concentration of acrylate copolymer increases or if it is due to the thickener itself. We thus formulated model matrices with guar gum, another thickener. The viscosity obtained with 2.5% guar gum was similar to that obtained with 1.6% acrylate copolymer (25.5 ± 0.4 mm versus 26.6 ± 0.3 mm). The spatial distribution of S. aureus was clustered with the guar gum (Table 1), but not without, regardless of the strain. The results were similar for hydrophilic beads (Table 1).

Spatial distribution of three S. aureus strains according to their level of hydrophobicity in matrices with thickener.
Spatial distribution was not affected by the level of hydrophobicity for the three bacterial strains in matrices with 0% or 0.4% acrylate copolymer (Table 1). Conversely, in the matrix with 1.6% acrylate copolymer, the variance of SF was highly superior to the mean and have reached 176.26 for S. aureus CIP 4.83, 11.32 for S. aureus ATCC 29213 and 8.43 for S. aureus ATCC 27217 ( Table 1). The most hydrophobic strain has also shown the most clustered spatial distribution in the matrix with 2.5% guar gum. Hence, the spatial distribution was increasingly clustered with increasing hydrophobicity of the strain in high viscosity matrices.

Discussion
The study of the spatial distribution for three S. aureus strains with different levels of hydrophobicity in several matrices with specific characteristics allowed us to understand the determinants of their distribution. First, we succeeded to obtain accurate M-CLSM images that allow rapid characterization of the bacterial spatial distribution in emulsified matrices. Despite the light diffraction by high lipid-content droplets, we optimized the settings to obtain high resolution images with a low fluorescence background signal. The mosaic mode allowed lowering the threshold of bacterial enumeration and each mosaic image (225 images) was quickly acquired in only eight minutes. An ImageJ macro was developed to batch, accelerate and standardize image quantification. Furthermore, we have used the method of Jongenburger et al. [15], highly relevant to statistically analyze the dispersion of bacterial spatial distribution in matrices. The frequency distribution was evaluated and the comparison of the variance and the mean of SF values (1350 to 3375 images per condition) allowed classifying the bacterial spatial distribution as clustered, We intend next to determine if the bacterial clustered distribution is due to the viscosity which increases when the concentration of acrylate copolymer increases or if it is due to the thickener itself. We thus formulated model matrices with guar gum, another thickener. The viscosity obtained with 2.5% guar gum was similar to that obtained with 1.6% acrylate copolymer (25.5 ± 0.4 mm versus 26.6 ± 0.3 mm). The spatial distribution of S. aureus was clustered with the guar gum (Table 1), but not without, regardless of the strain. The results were similar for hydrophilic beads (Table 1).

Spatial Distribution of Three S. aureus Strains According to Their Level of Hydrophobicity in Matrices with Thickener
Spatial distribution was not affected by the level of hydrophobicity for the three bacterial strains in matrices with 0% or 0.4% acrylate copolymer (Table 1). Conversely, in the matrix with 1.6% acrylate copolymer, the variance of SF was highly superior to the mean and have reached 176.26 for S. aureus CIP 4.83, 11.32 for S. aureus ATCC 29213 and 8.43 for S. aureus ATCC 27217 ( Table 1). The most hydrophobic strain has also shown the most clustered spatial distribution in the matrix with 2.5% guar gum. Hence, the spatial distribution was increasingly clustered with increasing hydrophobicity of the strain in high viscosity matrices.

Discussion
The study of the spatial distribution for three S. aureus strains with different levels of hydrophobicity in several matrices with specific characteristics allowed us to understand the determinants of their distribution. First, we succeeded to obtain accurate M-CLSM images that allow rapid characterization of the bacterial spatial distribution in emulsified matrices. Despite the light diffraction by high lipid-content droplets, we optimized the settings to obtain high resolution images with a low fluorescence background signal. The mosaic mode allowed lowering the threshold of bacterial enumeration and each mosaic image (225 images) was quickly acquired in only eight minutes. An ImageJ macro was developed to batch, accelerate and standardize image quantification. Furthermore, we have used the method of Jongenburger et al. [15], highly relevant to statistically analyze the dispersion of bacterial spatial distribution in matrices. The frequency distribution was evaluated and the comparison of the variance and the mean of SF values (1350 to 3375 images per condition) allowed classifying the bacterial spatial distribution as clustered, uniform or regular. M-CLSM appears to be a quick and relevant in situ approach for both bacterial enumeration and investigation of spatial distribution in complex matrices.
The results demonstrated that an increasingly viscous matrix, due to increase in the thickener concentration, results in a bacterial increasingly clustered distribution. By consequence, the heterogeneity of the enumeration results, affected by the clustered bacterial spatial distribution, also appeared to be driven by matrix viscosity (Figure 2, Table 1). In the field of metal production, in which metallic particles are added before solidification, the distribution of such particles is critical and can affect the quality of the final product. It was shown that viscosity, among other factors, influences the metal particle distribution in the matrices [33]. In an agar-based food model system, Wimpenny et al. [34] also stated that the distribution of bacteria is affected by agar concentration. The structure of emulsions, especially droplet size, may also play a role in the bacterial spatial distribution. Brocklehurst et al. [35] have studied the spatial distribution of Listeria monocytogenes and Yersinia enterocolitica in a hexadecane/water emulsion. They showed that droplet size correlates with viscosity: the smaller the droplet size, the higher the emulsion viscosity, and thus, the greater the bacterial immobilization. Furthermore, Jongenburger et al. [9] simulated a homogeneous or heterogeneous distribution of bacterial contaminants in batches of powdered infant formula and they showed that a low level of contamination and a heterogeneous bacterial distribution lead to a higher variability of bacterial enumeration.
Besides viscosity of the matrix, our results have also shown that an increase in S. aureus surface hydrophobicity induces an increase in the clustered spatial distribution (Table 1). These observations are in accordance with those of Ly et al. [36], who demonstrated the existence of physicochemical interactions between bacteria and the constituents of food matrices. They highlighted the role of the surface hydrophobicity of lactic-acid bacteria showing that the LLD18 strain, with a highly hydrophobic surface, interacts more with lipids and hydrophobic compounds than a hydrophilic strain (LLD16). Another study showed that hydrophobic bacteria can stabilize oil-in-water emulsions by adhering to the interface and preventing droplets from coalescing, whereas hydrophilic bacteria cannot [21]. The emulsified matrix used in our study consisted of a 65% hydrophilic phase and a 35% lipophilic phase. Our hypothesis was that the hydrophobic S. aureus strain was more highly trapped at the surface of the hydrophobic droplets than the two other strains, which led to a more heterogeneous distribution of the bacteria. This shows that the choice of the bacterial strain can have a strong impact on the bacterial spatial distribution and thus on the results of enumerations after sampling.
Interactions between bacteria and emulsions could also be affected by the surface charge of the bacteria cell wall. Li et al. [22] have studied interactions between E. coli and emulsion droplets (hexadecane/water). E. coli and these droplets having opposite surface charges, the bacteria were attracted to the surface of the droplets and destabilized the emulsion. However, in our study, the S. aureus strains showed no difference in electrophoretic mobility at an ionic strength of one millimole. The negative charge was the smallest at a pH of five to six. Moreover, at these pH levels, the negative charge of the droplets of the emulsified matrices, with or without acrylate copolymer, was very low. By consequence, in this study, the bacterial surface charge was unable to explain the difference of the bacterial distribution between strains.
In conclusion, M-CLSM was shown to be a rapid and relevant technique to characterize the bacterial spatial organization in complex emulsified systems. Both matrix viscosity and bacterial surface hydrophobicity were shown to affect the S. aureus spatial distribution in cosmetic matrices. The method could be used to quickly determine the ability of a formula to allow or not a bacterial regular distribution and thus to classify cosmetic matrices upon this parameter. Hence, this method will be helpful to increase the reliability of challenge tests by adapting the sampling plans to the class of matrices. The proposed M-CLSM methodology could also be used to analyze spatial patterns of microbial contamination in food matrices or medical functional biogels.