2.1.

Fluorescent Probe and Skin Autofluorescence

Nile Red (Sigma-Aldrich, Germany) was used as a lipophilic fluorescent dye in the prepared submicron emulsion. The fluorescence spectrum of Nile Red remains stable over a wide pH range while influenced by the polarity of its solvents.²⁵ Figure 1 shows the emissions spectra of the dye, solved in 70% ethanol, in comparison to natural porcine skin measured by the confocal laser-scanning microscope. Nile Red has an emission peak at 580 nm excited with the 476 nm line of an argon laser. The broad autofluorescence spectrum of skin exists due to the numerous intrinsic fluorophores.

Fig. 1

Emission spectra of untreated porcine skin and the fluorescent dye Nile Red (Excitation: 476 nm). The spectral data was measured using the confocal microscope. The normalized presentation of the data shows the distinguishable emission peaks and the spectral overlap.

2.2.

Drug Delivery System Preparation

Submicron emulsions were chosen as a delivery system because they have shown to enhance the penetration of substances into skin due to their particle size and surfactants.²⁶ Submicron emulsions are oil-in-water emulsions with a droplet size below one micrometer. It is a kinetic stable emulsion in which the droplet size is achieved by an incorporation of emulsifiers.²⁷ The oil phase consisted of oleth-3, -10, ethyloleat (Croda, Germany), coco-caprylat/caprat, cetearyl isononanoat (Cognis, Germany) and 0.002% Nile Red, while the aqueous phase consisted of sodium chloride, citric acid, potassium sorbate (Fagron, Germany), magnesium sulfate and glycerol (Caelo, Germany). The surfactant concentration was adjusted to achieve a hydrophilic-lipophilic balance value of 11. All ingredients obtained were of pharmaceutical grade. The physico-chemical properties were recorded to characterize the delivery system. The dynamic light scattering analyzing method (Zetasizer Nano ZS 90, Malvern Instruments, UK) showed a z-average size of $258\pm 3\mathrm{nm}$ with a polydispersity index of $0.16\pm 0.01$. The viscosity of $2.4\pm 0.2\mathrm{mPa}·\mathrm{s}$ was determined by RheoStress 300 Rheometer (ThermoHaake, Germany). Due to the preservation with potassium sorbate, the pH value was adjusted to $4.8\pm 0.2$.

2.3.

Porcine Skin Samples

Porcine ear skin is quite similar to human skin and was therefore used in this study.²⁸ Fresh ears of domestic pigs were obtained from a local abattoir and kept on ice during transport to the laboratory. After slaughtering, porcine ears were cleaned with water and shaved to remove the bristles. The full thickness skin was displaced with a scalpel and stored at $-20°\mathrm{C}$ for less than three months.

Two preparation methods were used to obtain reference skin samples for validation measurements and penetration skin samples. The latter were used for the penetration study that evaluates delivery system performance.

2.4.

Reference Skin Samples

The frozen skin specimens were cut into 10-μm thick slices using a cryo-microtom (Leica Microsystems, Germany) and then transferred into 70% ethanol with different Nile Red concentrations for 10 min. The dye was soaked up and excessive dye was removed from the skin under flowing water. The dye concentrations (0.2, 1, and $2\mathrm{\mu g}/\mathrm{mL}$) were chosen to cover a common recovery range of agents in the specimens of drug penetration studies. These reference skin samples were used to derive the sensitivity of the imaging scheme.

2.5.

Penetration Skin Samples

The penetration of Nile Red into skin was performed on a Franz diffusion cell setup. On the day of experiment the frozen skin was thawed and fixed between the donor and receptor chamber of the Franz diffusion cell (Gauer Glas, Puettlingen, Germany). The penetration area had a diameter of $1.767{\mathrm{cm}}^2$ and was in contact to 12 mL phosphate buffered saline (PBS) temperatured to 32.5 °C on the dermis side. The submicron emulsion was applied as infinite-dose (300 µL) for 3 h, 6 h, 8 h and 24 h to three porcine skin samples each to get a total of 12 experiment samples. After each time period, the skin was removed from the diffusion cell, wiped and washed with 50 mL aqua dest. The specimen was mounted into a freezing medium (Leica, Germany), shock frozen by fluid nitrogen and cut into various 10 μm thick vertical slices in order to prepare penetration samples for microscopy.

2.6.

Spectral Imaging

Spectral imaging was performed using a TCS SP5 confocal laser-scanning microscope (Leica Microsystems, Mannheim, Germany) equipped with a detection unit that allows spectral discrimination to selectively detect narrow wavelength bands. The recorded fluorescence emission region (excitation at 476 nm) starting at 495 nm up to 650 nm (divided into 10 single detection bands that are 20 nm wide) was chosen to cover the spectra of porcine skin and Nile Red. A 0.7 NA water immersion objective ($20\times$) was used and the image resolution was set to $512\times 512\text{pixel}$ with a pixel size of $1.51{\mathrm{\mu m}}^2$, resulting in a field of view of $775\times 775{\mathrm{\mu m}}^2$. The generated image stacks were of the dimensions $xy\lambda$. The reference spectrum of Nile Red was recorded in the same way. Figures 2 and 3 give an overview of a 24 h penetration sample in comparison to natural (untreated) skin. The series of $\lambda$-stack pictures and an exemplary spectrum of the selected regions of interest (ROI) are shown to illustrate the different spectra. These spectral differences are used for the image processing scheme.

Fig. 2

Overview (a), images of the $\lambda$-stack (b) and spectra of ROI 1 and 2 (c) of a 24 h penetration skin sample. The given wavelengths of the $\lambda$-stack describe the center of the 20 nm wide wavebands.

Fig. 3

Overview (a), images of the $\lambda$-Stack, (b) spectra of ROI 1 and 2, (c) of an untreated skin sample. The given wavelengths of the $\lambda$-stack describe the center of the 20 nm wide wavebands.

2.7.

Image Processing Scheme

The image processing starts with the application of a threshold to exclude pixels with low intensities. The background or noisy image regions are excluded from the calculations. In the next step, the intensity data are normalized along the $\lambda$-axis to calculate intensity independent spectra for each image pixel. The normalized cross correlation ($\mathrm{C}\mathrm{C}$) metric is used to quantify the similarity between the normalized sample spectra and the reference spectra of Nile Red.

The resulting correlation coefficient map of the image is built on the spectral information in the sample and is therefore based on dye distribution. Hereby, the highest $\mathrm{CC}$ value of one represents complete accordance between sample and reference spectrum. Decreasing $\mathrm{CC}$ values indicate decreasing correlation of the spectra.

To enable a comparison of the different images, the uptake indicator ${\mathrm{U}\mathrm{I}}_x$ is introduced. It represents the quantity of coefficients higher than a threshold normalized to the total number of pixels representing tissue.

Because ${\mathrm{U}\mathrm{I}}_x$ depends on an arbitrary threshold, two evaluation experiments were carried out. In the first experiment untreated skin samples were used to calculate the fraction that is false positively classified as Nile Red pixels with increasing threshold values (see Table 1).

Table 1

Measurements of untreated skin samples (n=5) for the evaluation of the 0 parameter.

Table 2

Presentation of the skin penetration results in mean UI0.8 and standard deviation for the four time periods.

Table 1 indicates the fraction of pixels of an untreated skin sample that are false positively calculated to be similar to the dye reference spectrum.

In the second experiment the reference skin samples were used to evaluate the highest threshold that allows a clear discrimination between the used concentrations.

Figure 4 shows the results of the reference samples indicating that ${\mathrm{U}\mathrm{I}}_{0.9}$ is not suitable to allow discrimination of the two lowest concentrations, whereas it is possible when using ${\mathrm{U}\mathrm{I}}_{0.8}$. Considering the very low false positive classification of. 02%, ${\mathrm{U}\mathrm{I}}_{0.8}$ was used as the evaluation parameter for further experiments.

Fig. 4

Determination of the appropriate value for ${\mathrm{U}\mathrm{I}}_x$. The graph shows the ${\mathrm{U}\mathrm{I}}_{0.7}$, ${\mathrm{U}\mathrm{I}}_{0.8}$ and ${\mathrm{U}\mathrm{I}}_{0.9}$ values for $0.2\mathrm{\mu g}/\mathrm{mL}$, $1\mathrm{\mu g}/\mathrm{mL}$ and $2\mathrm{\mu g}/\mathrm{mL}$ Nile Red reference skin samples. It is not possible to discriminate the $0.2\mathrm{\mu g}/\mathrm{mL}$ and $1\mathrm{\mu g}/\mathrm{mL}$ sample when using ${\mathrm{U}\mathrm{I}}_{0.9}$.

2.8.

Validation Measurements

To confirm that the scheme is not depending on absolute intensity values, the samples were imaged with different intensification by varying photomultiplier (PMT) voltages in a range from 500 V to 1200 V in eight steps. ${\mathrm{U}\mathrm{I}}_{0.8}$ was calculated for every image and was expected to be constant for different image intensity levels. For the second experiment the focal depth was changed during imaging of two skin samples and ${\mathrm{U}\mathrm{I}}_{0.8}$ was calculated to assess the influence of focusing.

3.1.

Scheme Validation

The results of the validation experiment for varying signal intensities are shown in Fig. 5(a). Regarding the variation of the PMT voltage to modify the image intensity, ${\mathrm{U}\mathrm{I}}_{0.8}$ is nearly constant over a 200 to 300 V region for the 24 h and 8 h samples. The ${\mathrm{U}\mathrm{I}}_{0.8}$ increase at 500 to 600 V can be explained with the very low overall intensity of the spectral images at 500 V. Thus, the spectra cannot be properly measured. The ${\mathrm{U}\mathrm{I}}_{0.8}$ decrease starting at 900 to 1000 V can be observed due to a strong signal saturation effect in single spectral bands for the 8 h and 24 h samples. Therefore, the spectral information is lost because of detector saturation. The 6 h sample has a ${\mathrm{U}\mathrm{I}}_{0.8}$ value that is constant over the range of 600 to 1200 V.

Fig. 5

The graphs show the results of the varying PMT (a) and focus (b) settings. The image sequences illustrate the changes in the data when varying the PMT voltage and the focus depth, respectively.

These measurements indicate that the ${\mathrm{U}\mathrm{I}}_{0.8}$ values do not change significantly over a broad intensity interval when using reasonable imaging parameters.

Focus position, in contrast, is a very sensitive parameter. The graph in Fig. 5(b) shows a Gaussian-like distribution of ${\mathrm{U}\mathrm{I}}_{0.8}$ with varying focus depths for the 10 µm thick sample. The measurements indicate that it is important to focus carefully for getting reliable results. Considering the moderate ${\mathrm{U}\mathrm{I}}_{0.8}$ change around the $-1\mathrm{\mu m}$ to 1 µm relative focus position, a roughly 3-μm tolerance margin for focusing can be supposed.

3.2.

Skin Penetration Study

The results of the penetration study show a strong time dependence of dye uptake in skin (see Fig. 6 and Table 2). The assessment of this uptake using the ${\mathrm{U}\mathrm{I}}_{0.8}$ parameter allows the quantitative comparison of the four time periods. It shows that the 24 h samples have a more than four times higher mean ${\mathrm{U}\mathrm{I}}_{0.8}$ value compared to the 8 h samples.

Fig. 6

Skin penetration study results ($n=20$ per penetration period). Data are presented in mean $\pm \mathrm{SD}$. *$P<0.05$ and ***$P<0.001$ evaluated by paired Student’s $t$-test. The 24 h samples show a significantly higher uptake of Nile Red.

The overlay intensity-based images and the spectrally resolved dye distribution are shown in Fig. 7. The intensity-based images were recorded with optimized parameters to achieve high image quality. The dye distribution was visualized using the correlation coefficient map and the ${\mathrm{U}\mathrm{I}}_{0.8}$ parameter as a threshold. The epidermal skin layer can be distinguished from the lower dermal layer due to its bright gray appearance. The darker dermal layer, however, shows brighter structures throughout the whole area. Dye distribution, shown in red, indicates that for the 3 h, 6 h and 8 h samples, the uptake does not cover the whole epidermis in the way that the ${\mathrm{U}\mathrm{I}}_{0.8}$ criteria is fulfilled. The 24 h sample shows complete uptake in the epidermis and additional uptake in the dermis layer.

Fig. 7

Overlay of the intensity based confocal images of the skin and the spectrally resolved dye distribution (shown in red, please refer to the online version of this manuscript) for the four time periods.

4.1.

Pharmaceutical In Vitro Experiments

High dermal transport is important to address skin diseases occurring in the upper layers. Tape-stripping methods are commonly used to stepwise ablate small parts of the outer skin for quantitatively analyze depth profiles.²⁹ However, it is limited to the stratum corneum and does not give any information of the homogeneity and the distribution of drugs in the skin layer itself. In contrast, microscopic evaluation of vertical slices has the capability to show profiles directly throughout the whole skin, but lacks the possibility to quantify the amount of penetrated substances. Schmidts³⁰ presented a skin penetration study that correlated an ELISA based quantification with an intensity-based microscopy evaluation method. The results were of good qualitative accordance indicating the suitability of the microscopy approach for evaluating dermal delivery systems. The developed uptake indicator of this paper was a consequent advancement of Schmidts’ microscopy analysis. It allows indirect quantitative measurements and is a step towards the quantitative evaluation of dermal delivery systems and processes. The results of the presented penetration study support this statement as they show a clear discrimination of the four time periods.

Our proposed method may also be beneficial to studies involving applications of spectral and hyperspectral imaging of penetration studies. Roberts; et al.; performed delivery studies using a FLIM multiphoton system⁴ that is capable of spectrally detecting fluorescence signal to compare liposome systems. They concluded that the deformability of the liposomes increases penetration based on visual evaluation. In this case, the uptake indicator could be used to obtain comparable results and further increase insight into underlying penetration processes.

Hernandez-Palacios¹⁴ introduced a hyperspectral camera system with 20 µm spatial resolution to evaluate penetration studies. The hyperspectral camera detected the emission peak of Alexa 488 (518 to 524 nm) and therefore avoided the use of fluorescence filters. However, the evaluation is based on the fluorescence intensity that is composed of Alexa dye and skin autofluorescence background. Skin structures like melanin and keratin, with an emission peak at 520 nm,^2,4 contribute to this intensity. Subtraction of a constant offset value compensates the overlaying signal. Figure 7 illustrates that, when using microscopic resolution, structures of the skin are brightly visible that are not covered with dye. In this case, the overlapping and superimposing signal of the skin and dye vary depending on microscopic structures that are not equally distributed. Hence, a constant autofluorescent background offset cannot be defined and mapping of the dye based on intensity in the microscopic scale is not possible.

4.2.

Advantages and Drawbacks for Delivery Studies

The spectral imaging scheme has several benefits. Additional to the advantage of accurate dye mapping, spectral analysis is superior to intensity-based methods in terms of experiment procedures because it does not require standardized imaging. In general, the emission intensity in confocal microscopy depends on various parameters. Therefore, when using intensity-based methods, it is a comprehensive but very important task to use constant settings in order to assure the comparability of the results. This standardization process requires the assessment and control of imaging setup and sample conditions. While the setup may be kept constant for a whole experiment with certain effort, the sample conditions cannot be controlled easily due to their biological nature.

It is known that the pH-value changes in different skin layers³¹ and may influence the intensity of dyes that have a strong pH-dependency. Apart from the pH value, fluorescence emission can also be affected by polarity. In the case of Nile Red, the fluorescence is unaffected by pH-values between 4.5 and 8.5 but the emission peak shifts depending on the polarity of the environment.²⁵ A possible shift must be considered for choosing the appropriate reference spectrum for image processing calculations. Additionally, photobleaching is a serious problem that requires attention. Borgia⁶ reports a 4 to 16% bleaching rate after a one second illumination of Nile Red. It is concluded that the images have to be taken within 10 ms to avoid a strong signal variation.

The spectral data do not depend on absolute intensity values but on relative intensity changes between single wavebands. Consequently they are less sensitive to condition variations and photobleaching resulting in a high robustness of the calculated results. Figure 5 shows that the calculated assessment parameter is constant for reasonable changes of the imaging parameters. This allows measuring every sample with individual settings without the need to use standardized imaging parameters for comparing the results. The advantage is a broad, analyzable concentration range of fluorescent dye in skin specimens. This is especially useful for comparing completely independent experiment setups.

Another advantage of the spectral method is that it is still applicable whenever dyes with emission spectra that are clearly separated from the skin spectrum are not available. Although the spectra overlap, an explicit correlation to either dye or tissue can still be performed. As a result, a reliable assessment of the spatial distribution of dye in tissue is possible.

The ${\mathrm{U}\mathrm{I}}_{0.8}$ parameter that is chosen for the evaluation of the penetration study is an arbitrary measure. It is important to understand that it cannot be used to quantitatively assess the total amount of dye in skin. Due to the overlapping dye and skin spectra, it is difficult to determine the exact correlation coefficient that describes the transition between the two possible signal origins. The ${\mathrm{U}\mathrm{I}}_{0.x}$ parameters are used to arbitrarily classify the calculation results into dye and tissue pixels according to a threshold value. For this reason the proposed spectral method is not applicable for tasks that need exact quantitative results. In these cases intensity-based methods can be used that need to be calibrated or intensity corrected. Monte Carlo simulations that determine and compensate for effects of signal attenuation due to scattering and absorption³² are often applied for these applications. Combining these techniques with the spectral evaluation for penetration studies would add valuable information to the localization and penetration depth of dye and agents.

4.3.

Pharmaceutical In Vivo Experiments

It is arguable that in vitro results can be used to assess delivery systems designed for in vivo use. No clear data is available in the literature as there are many studies with good and poor correlation between in vitro and in vivo results.³³ Therefore, in vitro results for drug delivery cannot be transferred directly to in vivo considerations without further investigation. For this reason, studies are performed with a parallel in vivo/in vitro setup³⁴35.^–36 using living mice for the application and penetration of delivery systems and substances. The mice either are sacrificed or biopsies are taken for analysis afterwards. An approach for in vivo imaging without sacrificing the animals is the use of animal window models.^37,38 Here, the living animal can be microscopically examined directly through windows that are surgically inserted into the region of interest. Whenever living animals are observed motion artifacts have to be considered and either fixation or synchronizing the scanning system to the heartbeat needs to be performed. Amornphimoltham³⁹ reviewed further in vivo applications for drug delivery research and showed the technical efforts that have to be made for data acquisition.

4.4.

Feasibility for Clinical In Vivo Applications

Gareau⁴⁰ presented an approach of in vivo confocal reflectance microscopy for the detection of early-stage melanoma of murine skin without further surgical interventions. Lange-Asschenfeld⁴¹ and Caspers⁴² showed confocal laser-scanning microscopes for in vivo imaging of human skin morphology and wound healing. Both studies mention the potential use for pharmaceutical research on drug delivery. A combination of the proposed microscope setup and our spectrally resolved detection unit could be used for optical sectioning and evaluation of fluorescent dyes in living tissue. Additionally, the in vivo microscopy scheme presented by Gareau combined with our proposed spectral uptake indicator could be used in clinical applications like melanoma detection presented by Kuzmina¹⁵ for monitoring size and shape variations.

Extending established microscopic in vivo imaging schemes with our proposed spectral evaluation method has the chance to provide additional information for pharmaceutical and clinical applications. However, further investigation is necessary to validate these potential approaches.