2.1.

Experimental Setup

The experiments were performed using two confocal Raman microscopes on porcine ear skin ex vivo in the biological fingerprint region (400 to $1800{\mathrm{cm}}^{-1}$ for excitation at 473 and 785 nm and 700 to $1800{\mathrm{cm}}^{-1}$ for excitation at 325 and 633 nm). For the excitation at 325, 473, and 633 nm, a LabRAM HR Evolution (HORIBA Jobin Yvon, Longjumeau Cedex, France) with a comprised BX41 microscope (Olympus, Tokyo, Japan) was used. For the excitation at 473 and 633 nm, a $600\text{-}\mathrm{g}/\mathrm{mm}$ grating, an MPlan N VIS2 $50\times$ objective (Olympus, Tokyo, Japan), and an $Ø100\text{-}\mu \mathrm{m}$ pinhole were used. A $2400\text{-}\mathrm{g}/\mathrm{mm}$ grating, an LMU-40X-NUV objective (Thorlabs, Newton, New Jersey), and an $Ø200\text{-}\mu \mathrm{m}$ pinhole were used for excitation at 325 nm. Both gratings were blazed at 500 nm. A 325 nm IK3301 R-G He-Cd laser (Kimmon Koha Co., Ltd., Tokyo, Japan), a 473 nm Cobolt Blues 25-mW diode laser (Cobolt AB, Solna, Sweden), and a 633-nm HeNe laser (Melles Griot, Carlsbad, California) were used as excitation sources, delivering 8.1, 6.1, and 8.4 mW of optical power, respectively, to the sample surface. The spectral resolution was $<4{\mathrm{cm}}^{-1}$. Data acquisition was performed using the associated LabSpec 6.4 software. This device was described in detail elsewhere.⁴⁷

For NIR excitation, a Model 3510 Skin Composition Analyzer (SCA) confocal Raman microscope (River Diagnostics B.V., Rotterdam, The Netherlands) suitable for in vivo and ex vivo measurements with a $50\times$ oil objective and a 785 nm laser (Innovative Photonic Solutions, Monmouth Junction, New Jersey) was used. The spot size was $5\mu \mathrm{m}$ and an optical power of 22 mW on the skin surface was used. An axial resolution of $<5\mu \mathrm{m}$, as well as a spectral resolution of $2{\mathrm{cm}}^{-1}$, was achieved. This device was described in detail elsewhere.^48,49

2.2.

Experiments

The experiments were conducted ex vivo on unpigmented, untreated, not preirradiated porcine ears, serving as a skin model,⁵⁰ at a maximum of 3 days after slaughter. The hair was carefully removed from the skin without influence on the SC, and the samples were stored at $4°\mathrm{C}$ before the measurements, which were performed at $22°\mathrm{C}$ room temperature. Measurements were performed in 21 consecutive acquisitions of 15 s, resulting in a total experiment time of 315 s with continuous irradiation for every sample. A total of $n=6$ measurements were taken at $\approx$0, 10, 15, 20, 50, 120, and $150\mu \mathrm{m}$ depths for every applied excitation wavelength. Depths 0 to $15\mu \mathrm{m}$ were assumed to be located within the SC,^{43,45,51–54} depths 20 to $50\mu \mathrm{m}$ in the viable epidermis,^44,45,55 and depths 120 to $150\mu \mathrm{m}$ within the papillary dermis. Deviations of the measured depth in the skin due to varying refractive index in the different skin layers⁴¹ were neglected.

The position of the skin surface was determined after the photobleaching experiment, by performing a $z$-scan of Raman measurements from $40\mu \mathrm{m}$ above to $40\mu \mathrm{m}$ below the visual focus plane in $2\mu \mathrm{m}$ increments, while maintaining the $x$- and $y$-positions of the sample. After baseline removal, the height of the $1654{\mathrm{cm}}^{-1}$ amide I Raman band was analyzed as a function of the $z$-position. For 325 nm excitation, this was done with the band around $1600{\mathrm{cm}}^{-1}$. The skin surface was then defined to be the position with 50% of the maximum peak height.⁵⁶ For the measurements with the Model 3510 SCA, a similar procedure could be automatically performed using the nonrestricted multiple least squares fit method of the keratin profile using the “Skin Tools 2.0” software (River Diagnostics B.V., Rotterdam, The Netherlands), which was explained in detail elsewhere.⁵⁷

2.3.

Data Analysis

All data analysis was performed in MATLAB^® R2016a (The MathWorks, Inc., Natick, Massachusetts). To analyze the fluorescence dynamics, the area under the curve (AUC) was calculated using trapezoidal numerical integration in the 400 to 700, 700 to 1100, 1100 to 1500, and 1500 to $1800{\mathrm{cm}}^{-1}$ ranges. This was done to examine a possible wavelength-dependent photobleaching effect. For analysis of the AUC, cosmic spikes were removed, using a seventh-order median filter. Due to reduced dispersion of the $600\text{-}\mathrm{g}/\mathrm{mm}$ grating at 633 nm and the $2400\text{-}\mathrm{g}/\mathrm{mm}$ grating at 325 nm excitation, the 400 to $700{\mathrm{cm}}^{-1}$ range was omitted in these cases. The autofluorescence intensity was normalized by dividing by the intensity of the initial acquisition at $t=15\mathrm{s}$.

The autofluorescence decay was fitted to a double-exponential decay function [Eq. (1)] using nonlinear regression for every measurement. Peak heights for the $1003{\mathrm{cm}}^{-1}$ phenylalanine and urea, the $1452{\mathrm{cm}}^{-1}$ $\delta ({\mathrm{CH}}_2)({\mathrm{CH}}_3)$, and the $1654{\mathrm{cm}}^{-1}$ amide I bands were analyzed by fitting a Gaussian function to the baseline removed spectra. For 325 nm excitation, this was done for the band around $1600{\mathrm{cm}}^{-1}$ due to the different spectral shape.

3.1.

Raw Spectra

Figure 1 shows exemplarily the mean of $n=6$ measurements for $10\mu \mathrm{m}$ depths in the skin for 325 (a), 473 (b), 633 (c), and 785 nm (d) at the consecutive acquisition times from 15 to 180 s. The total intensity of the fluorescence is highly dependent on the fluorophore concentration in the measured volume and varies strongly for repeating measurements, which becomes especially obvious for smaller measurement volumes, using confocal microscopy. However, the maximum fluorescence intensity of the mean of six measurements at 473 nm excitation is higher by a factor of $\approx 2$ compared to the excitation at 633 nm, despite a 1.3 times smaller excitation power (6.1 mW for 473 nm and 8.4 mW for 633 mW). This would be in agreement with the expected loss of fluorescence at increased excitation wavelengths. The fluorescence intensity at 325 nm excitation is considerably reduced, compared to 473 and 633 nm, which can be explained by the wavelength-dependent absorption- and scattering-based intensity loss, entailing smaller penetration depths at UV excitation. The results for 785 nm [Fig. 1(d)] were acquired using different instrumentation at 22 mW. For a detailed analysis of the total fluorescence, a wavelength-dependent light throughput of the detecting optics, as well as the detector sensitivity, would have to be taken into account.⁵⁸ Because of scattering, absorption, and anisotropy propagation, the total intensity of the spectra decreases with increasing depth in the skin (data not shown). The Raman/fluorescence ratio clearly increases during the photobleaching process, which is in accordance with previously published data.⁸

Fig. 1

Example for the photobleaching effect at (a) 325 nm (8.1 mW), (b) 473 nm (6.1 mW), (c) 633 nm (8.4 mW), and (d) 785 nm excitation (22.0 mW). The spectra show the mean of six measurements at $10\mu \mathrm{m}$ depths in the SC for 12 consecutive acquisitions of 15 s. The fluorescence intensity decreases with every acquisition.

In contrast to the fluorescence, the Raman intensities do not bleach over extended irradiation. To verify this expected behavior, the Raman peak heights for the 1003, 1452, and $1654{\mathrm{cm}}^{-1}$ bands were recorded during the 315 s irradiation. After baseline removal, no change in peak intensities could be observed. The spectral shape for 325 nm excitation [Fig. 1(a)] is clearly distinct from the visible and NIR excitation, which was previously also observed on cell measurements.⁵⁹ Only a band around $1600{\mathrm{cm}}^{-1}$ can be identified. The intensity of this band, however, does also not decrease during autofluorescence photobleaching.

3.2.

Autofluorescence Photobleaching

Figure 2 shows the mean of the normalized autofluorescence decay for the measured depths at 325, 473, 633, and 785 nm excitation. Although the fluorescence intensity shown in Fig. 1 is wavenumber dependent, as evident by the skew of the spectra, especially for 325 and 473 nm excitation, a difference in the bleaching rate depending on the wavenumber range as defined in the data analysis section could not be found. Therefore, Fig. 2 only shows the photobleaching of the complete 700 to $1800{\mathrm{cm}}^{-1}$ range. The standard deviation $1\sigma$ was $\le 0.1$ ($\le 0.2$ for 325 nm excitation at $0\mu \mathrm{m}$ depth in skin), but for clarity, error bars are omitted in the figures.

Fig. 2

Mean normalized decay of AUC for six repeative measurements in the 700 to $1800{\mathrm{cm}}^{-1}$ wavenumber range using (a) 325 (8.1 mW), (b) 473 (6.1 mW), (c) 633 (8.4 mW), and (d) 785 nm excitation (22.0 mW) at different depths in the skin.

The results indicate that a nonlinear photobleaching occurs for all excitation wavelengths in all the measured skin depths. The largest decay occurs for 325 nm excitation, where the autofluorescence decreases from 18% to 29% of its initial intensity. For 473 nm excitation, the fluorescence intensity decreases from 39% to 54% of its initial value at all skin depths after 315 s irradiation. The decrease for 633 nm excitation is only slightly reduced from 36% to 60% of its initial intensity. Despite a $>2.5$ times higher excitation power at 785 nm, the autofluorescence decay is considerably smaller, as the fluorescence intensity decreases only from 50% to 80% of the initial value in this case. The initial fluorescence at 785 nm excitation is severely reduced compared to 325, 473, and 633 nm, and it is assumed that only few fluorophores in the skin are excited by NIR light.

Excitation at 325 nm has a considerably lower penetration depth, therefore, only depths in the SC are shown. The largest decay occurs at $10\mu \mathrm{m}$ depth, where the fluorescence intensity decreases to $<20\%$ after 315 s irradiation. Measurements below the SC cannot be reasonably performed at 325 nm excitation. Although a majority of endogenous fluorophores are excited in the UV wavelength range, the main effect of autofluorescence photobleaching in this case is likely related to keratin, which is the main skin component in the SC. For excitation at 473 nm and 633 nm, the largest decay occurs at $50\mu \mathrm{m}$ depth in the skin, which represents the stratum spinosum layer of the viable epidermis, with fluorescence intensity of 35% to 40% of the initial value after 315 s irradiation. The decay in the SC (0 and $10\mu \mathrm{m}$) was the smallest in this case (50% to 60% of the initial value). For the 785 nm excitation, the largest decay occurs in the SC between 0 and $15\mu \mathrm{m}$ in the skin (50% to 55% of the initial value after irradiation) and considerably smaller bleaching in the dermis (120 to $150\mu \mathrm{m}$), where the fluorescence intensity only dropped to 75% to 80% of the initial value after 315 s irradiation. The major fluorophore excited by 785 nm is assumed to be melanin, which is located mainly in the lower epidermis. These results indicate that the fluorescence photobleaching by 785 nm excitation mainly occurs for endogenous fluorophores located in the epidermis, which is likely melanin. These results disagree with the findings of Wang et al.,⁴⁶ who found only negligible fluorescence in the epidermis at 785 nm excitation. This needs to be further elucidated. The comparatively small photobleaching component in the dermis could possibly result from collagen and elastin,⁷ which are also the main dermal components. However, collagen and elastin mainly fluoresce under blue excitation, which is likely the origin of the moderate photobleaching at 473 nm excitation below $100\mu \mathrm{m}$ in the skin. The relatively small amount of autofluorescence photobleaching in the SC (0 to $10\mu \mathrm{m}$) at 473 nm excitation may result from keratin to a large extent, which is the major component of the SC.⁴¹ Our results do not indicate autofluorescence free layers in the skin at either of the applied excitation wavelengths.

Using the confocal Raman microscope, a possible long-term drift of the laser focus position due to the drying of the sample may occur; hence, measurements to examine a possible recovery of the autofluorescence could not be performed. Zeng¹⁷ noticed a fluorescence recovery in vivo after 6 days, while Stratonnikov et al.⁴ examined a recovery after 3 to 5 days after in vivo laser irradiation. In the study by Lihachev et al.,³ autofluorescence recovery to 80% of the initial value took $<5\text{days}$. The origin of the fluorescence recovery is not fully understood. A possible explanation is the diffusion of unbleached fluorophores to the irradiated area, as explained for FRAP measurements.^19,20 In contradiction, the results by Zeng¹⁷ indicated that this was not the explanation for the fluorescence recovery in their presented data; instead, a long-term photochemical reaction was suggested to be more probable than photo decomposition of the endogenous fluorophores. In the case of decomposition of the irradiated molecules, a decrease of the Raman signal would also be expected, which was not recognized in the presented data.

3.3.

Photobleaching Decay Parameters

The two-exponential decay function was used for a nonlinear regression approximation of the autofluorescence decay. The offset parameter $A$ in the decay function [Eq. (1)] is a measure of the baseline fluorescence intensity of the spectra that is not bleached. This parameter varies depending on the measured skin position and, hence, the availability of fluorophores in the measured volume,¹⁶ which is in agreement with other published data.¹² ${\tau }_1$ and $a$ describe a fast decrease component of the decay function while ${\tau }_2$ and $b$ describe a slow decrease component. The parameters ${\tau }_1$ and ${\tau }_2$ indicate the rate while parameters $a$ and $b$ indicate the fraction of the photobleaching decay for each of the components.¹⁶ Figure 3 shows an example of the experimental decay data in the 1100 to $1500{\mathrm{cm}}^{-1}$ region for 785 nm excitation and the two-exponential decay function fitted by nonlinear regression ($R\text{-}\text{squared}=0.997$).

Fig. 3

Example for autofluorescence decay normalized AUC for 15 to 315 s 785 nm excitation. Measured in the SC ($10\mu \mathrm{m}$ skin depth) in the 1100 to $1500{\mathrm{cm}}^{-1}$ wavenumber range. The data are fitted to a double-exponential decay function [Eq. (1)] by nonlinear regression ($R\text{-}\text{squared}=0.997$). ${\tau }_1=12.262$, ${\tau }_2=103.081$, $a=0.800$, $b=0.131$, and $A=0.651$.

The photobleaching could be described by the two-exponential decay function in all skin depths at all applied excitation wavelengths. Former suggestions that the two components of the decay function represent the photobleaching of the SC and the dermis, respectively,^16,17 could not be verified by our results because every measurement is limited to a single skin layer due to the use of confocal microscopy.

The differences in the amount of photobleaching, as discussed in Fig. 2, correlate to the offset parameter $A$, as shown in Fig. 4. A significant difference of the photobleaching depending on the detected wavenumber ranges (400 to 700, 700 to 1100, 1100 to 1500, and 1500 to $1800{\mathrm{cm}}^{-1}$) could not be determined using one-way analysis of variance (ANOVA, $p>0.25$) for either of the applied excitation wavelengths (data not shown). A reduced residual fluorescence that is not bleached during excitation can clearly be noticed for UV excitation. The largest photobleaching component at a depth of $50\mu \mathrm{m}$, as explained in Fig. 2, is represented by the lowest parameter $A$ for 473 and 633 nm excitation. For 785 nm excitation, the residual fluorescence is lowest in the SC and increases in the viable epidermis, especially in the papillary dermis.

Fig. 4

Autofluorescence photobleaching decay parameter $A$ of double-exponential decay function [Eq. (1)] using 325, 473, 633, and 785 nm excitation for different depths in the skin in the 700 to $1800{\mathrm{cm}}^{-1}$ wavenumber range. The mean of six measurements with error bars representing $\pm \text{standard deviation}$ $1\sigma$ is shown.

These findings are not reflected in the bleaching rate parameters ${\tau }_1$ and ${\tau }_2$, which means that a large bleaching does not necessarily occur at a fast rate.

Figure 5 shows the decay parameters ${\tau }_1$ and ${\tau }_2$ for the 700 to $1800{\mathrm{cm}}^{-1}$ wavenumber range. No significant differences of the photobleaching rates depending on the detected wavenumber ranges were found (ANOVA, $p>0.25$). In the case of influence of multiple fluorophores emitting with differing emission maxima, this could be an indication that the photobleaching rates for these fluorophores are equal. In the in vivo study by Zeng et al.,¹⁶ using 442 nm excitation, only minor wavelength dependence of the fitting parameters was found.

Fig. 5

Autofluorescence photobleaching decay parameters (a) ${\tau }_1$ and (b) ${\tau }_2$ of double-exponential decay function [Eq. (1)] using 325, 473, 633, and 785 nm excitation for different depths in the skin in the 700 to $1800{\mathrm{cm}}^{-1}$ wavenumber range. The mean of six measurements with error bars representing $\pm \text{standard deviation}$ $1\sigma$ is shown.

For excitation at 473 and 633 nm, the parameter ${\tau }_1$ is largest in the dermis (120 to $150\mu \mathrm{m}$), representing a slower bleaching of the fast decay component (Fig. 5), while a moderate amount of photobleaching in the dermis can be seen in Fig. 2. For 785 nm excitation, the parameters ${\tau }_1$ and ${\tau }_2$ are reduced at the skin surface and in the SC, which represents a faster bleaching process. Large deviations can be seen for excitation at 325 nm. At $0\mu \mathrm{m}$ depth, ${\tau }_1$ is 2.6 to 4.3 times larger compared to the deeper measurements, which reach to the bottom of the SC. However, the standard deviation is 137% of ${\tau }_1$ in this case. In the case of UV excitation, a considerable dark spot on the skin sample and the formation of an air gap between the sample and the cover slide in the microscopic video image were noticed. It is likely that the skin surface position was not stable during the measurements at UV excitation, which could explain the large deviations. ${\tau }_2$, which represents the slow bleaching component, generally increases with increasing depth in the skin. A possible explanation could be the depth-dependent intensity loss of the excitation.

Zeng¹⁷ found in vivo that the fast and slow decay parameters ${\tau }_1$ and ${\tau }_2$ differed by an order of magnitude, which was confirmed by our results. Also in the in vivo study by Wang et al.⁷ using 785 nm excitation, ${\tau }_1$ and ${\tau }_2$ differed by an order of magnitude. The ratio ${\tau }_1/{\tau }_2$ was between 0.08 and 0.12 in our results, except for excitation at 325 nm, where ${\tau }_1/{\tau }_2$ was 0.3 at $0\mu \mathrm{m}$ and 0.05 at 10 and $20\mu \mathrm{m}$ depth in the skin.

Figure 6 shows the two-exponential decay parameters $a$ and $b$. Analog to ${\tau }_1$ and ${\tau }_2$ a one-way ANOVA showed no significant differences depending on the wavenumber ranges ($p>0.25$); therefore, only the 700 to $1800{\mathrm{cm}}^{-1}$ range is shown. For clarity, the parameter $a$ is not shown for 325 nm excitation; the value was increased by a factor of $\approx 4$ for $15\mu \mathrm{m}$ and was 3 orders of magnitude larger for 0, 10, and $20\mu \mathrm{m}$ depth in the skin, with standard deviations $1\sigma$ of up to 180% of $a$. The large deviations in the case of UV excitation could be explained by a possible movement of the skin surface during irradiation. Large deviations also occur for the $b$ parameter for 325 nm excitation at $0\mu \mathrm{m}$ depth in the skin, which has a standard deviation $1\sigma$ of 50%. For 473 and 633 nm excitation, an increase of $a$ and $b$ with increasing depth in the skin can be observed, with the exemption of a decrease of $a$ in the papillary dermis. In reverse, $a$ and $b$ decrease with increasing depths in the skin for 785 nm excitation.

Fig. 6

Autofluorescence photobleaching decay parameters (a) $a$ and (b) $b$ of double-exponential decay function [Eq. (1)] using 325, 473, 633, and 785 nm excitation for different depths in the skin in the 700 to $1800{\mathrm{cm}}^{-1}$ wavenumber range. For clarity, parameter $a$ is not shown for 325 nm. Compared to the other excitation wavelengths, the value was increased by a factor of $\approx 4$ for $15\mu \mathrm{m}$ and was 3 orders of magnitude larger for 0, 10, and $20\mu \mathrm{m}$. The mean of six measurements with error bars representing $\pm \text{standard deviation}$ $1\sigma$ is shown.

The $b/a$ ratio for 473, 633, and 785 nm is between 0.3 and 0.4 for all depths in the skin, with the only exception of 0.2 for 785 nm at $0\mu \mathrm{m}$ depth, which can be explained by the increased $a$ in this case. For 325 nm excitation, the $b/a$ ratio is between 0.001 and 0.06, indicating a larger contribution of the fast decay component (parameter $a$ is not shown for 325 nm for clarity).

A direct comparison of the photobleaching rates is challenging because they are dependent on the irradiation intensities. Linear and nonlinear relations between excitation intensity and fluorescence have been found, depending on the excitation wavelength.¹³ The excitation intensity is in return dependent on the depth within the skin due to absorption- and scattering-based attenuation. Furthermore, the absorption and scattering in the skin are wavelength dependent, entailing a higher penetration depth at 785 nm than at the 473 and 633 nm visible excitation wavelengths and severely reduced penetration depth only at 325 nm. Further investigations are necessary to compensate the depth- and wavelength-dependent irradiation and autofluorescence intensities. One approach could be the fluorescence normalization based on the keratin contribution of the recorded Raman spectra,⁵⁸ as the keratin distribution within the SC is considered to be homogenous.^41,56 However, for deeper skin layers, the normalization is a very complicated task due to the inhomogeneous distribution of the fluorophors. Franzen et al.⁶⁰ developed an algorithm for compensated depth-dependent intensity loss that was based on measurements on a skin phantom incubated with known homogenous drug concentrations and obtained homogenous drug-based Raman intensities in the corrected depth profiles. Challenges for such approaches might arise for physiologic variations, such as varying thickness of skin layers depending on the body site.