2.1.

Clinical Study

This study was conducted in Toulouse, France, in accordance with the Declaration of Helsinki and was approved by the “Comité de Protection des Personnes Sud-Ouest et Outre Mer III” (ref.: 2014/44). Seven healthy volunteers aged between 20 and 30 years old and seven healthy volunteers aged 60 and older were recruited. The older subject group presented a photoaging score SCINEXA (score of intrinsic and extrinsic skin aging)¹⁷ superior or equal to 2, proving clear photoaging. All participants gave written informed consent before taking part in the study.

2.2.

Skin Samples

Four millimeters punch biopsies of sun-protected buttock and sun-exposed dorsal forearm skin were performed on each subject. Samples were immediately included in optimal cutting temperature compound and frozen in liquid nitrogen. Thin sections (50 μm) were fixed with acetone for 10 min and mounted between a microscope slide and a coverslip with a mounting medium (Dako, ref S3023).

2.3.

Multiphoton Microscopy Images

Image acquisition was performed on a Nikon A1-MP-Si microscope with a Spectra Physics MaiTai laser. The image collection is done with nondescanned GaAsP detectors or a descanned spectral detector. The lens used for the acquisitions is a $25\times W$ with a numerical aperture of 1.1. The excitation wavelength was fixed at 840 nm and all sample images were acquired with the same parameters (image size, detectors’ gain and offset, and filters). The SHG signal was collected between 400 and 492 nm and the AF between 500 and 550 nm. For each sample, three stacks of images were acquired on a depth of around $40\mu \mathrm{m}$. The pixel size was $0.21\mu \mathrm{m}$ and the $z$-step was $0.375\mu \mathrm{m}$.

Because the section samples were not well flat and to only consider images with robust information, we selected the image slices with a sufficient mean intensity for the calculations. More precisely, $I_{\text{mean}}$ being the mean intensity of a given slice and $I_{\text{max}}$ being the mean intensity of the most intense slice of the 3-D stack, we applied the criterion $I_{\text{mean}}>0.8\times I_{\max }$ to only consider the most intense slices. The parameter calculations were performed on the selected $35\pm 5$ slices, corresponding to a total depth of $13\pm 2\mu \mathrm{m}$.

As for controlling the separability between SHG and AF signals, spectral images were also acquired between 400 and 650 nm with a spectral detector, with a resolution of 10 nm.

All images were acquired below the dermis–epidermis junction and were centered in the upper reticular dermis.

2.4.

First-Order Analysis

All image analysis was performed with MATLAB (MathWorks, Natick, Massachusetts). Calculations were performed on the whole 3-D volume. A fixed threshold is applied, and the densities of collagen and elastin are calculated according to that threshold:

Several statistical moments¹⁸ and the entropy of the distribution¹⁹ were calculated on the histograms:

2.5.

Wavelet Analysis

In this part, maximum intensity projection (MIP) of 3-D images has been considered. The wavelet decomposition of the two-dimensional projections has been performed using Daubechies wavelet at the first scale.¹⁴ Horizontal, vertical, and diagonal coefficients are then put in a single histogram, which can be modeled by a GGD:

The parameters $\alpha$ and $\beta$ are then used to characterize the images.

2.6.

Statistics

For each parameter, a statistical analysis was made to determine the significance of the differences between zones and between groups. A Wilcoxon–Mann–Whitney test was done to compare values between groups and a paired Wilcoxon test to compare study areas inside each group. A $p$-value under 0.05 was considered significant.

3.1.

Spectral Analysis

To confirm that SHG and AF signals can be accurately analyzed separately, multispectral images were acquired on different skin samples. Mean gray values were spectrally analyzed according to the wavelength. They exhibited a clear bimodal response (cf. Fig. 3). A sharp peak can be noticed at 420 nm, which corresponds to the SHG signal (the laser excitation wavelength being 840 nm). A second broad peak is observed between 450 and 550 nm, which corresponds to the AF signal. Both signals appear well-defined and clearly different from each other.

Fig. 3

Spectral analysis of the signal: (a) acquisition of the 25 images corresponding to 25 different wavelength bands, (b) color reconstruction of the signal and delineation of the region of interest, and (c) mean intensity value of the signal according to the wavelength.

3.2.

Colocalization Analysis

The first channel has a broad spectral detection windows (400 to 492 nm), which includes part of the AF signal. A colocalization analysis was performed between the “blue” and “green” channels to evaluate the independence of the two measured signals.

From Fig. 4, we observe two different signal dynamics. A Pearson correlation coefficient was computed for each acquisition and we obtained a mean coefficient of 0.13. This analysis showed that the two signals were not highly correlated. We can thus assess that the two signals are mostly independent and then attribute the blue channel primarily to collagen and the green channel primarily to elastin.

Fig. 4

Example of a colocalization histogram for one single acquisition.

3.3.

First-Order Analysis

The SAAID was calculated on MIP images (results are shown in Fig. 5). This parameter exhibited a clear difference between the older and younger skin ($p=0.009$ and 0.015 on the buttock and the forearm, respectively). Moreover, chronically UV-exposed older skin appeared more affected than the UV-protected skin (group 2, $p=0.016$).

Fig. 5

SAAID values.

To precisely analyze the differences between intrinsic and extrinsic aging of the dermal fibers, SHG, and AF signals were also studied separately.

First-order parameters were calculated on the SHG images (cf. Fig. 6 and Appendix, Table 2 for statistical analysis). Mean intensity and skewness of the images’ distributions were significantly different between the buttock and the forearm for both groups (group 1, $p=0.016$ and 0.047 for mean intensity and skewness, respectively; group 2, $p=0.016$ and 0.016 for mean intensity and skewness, respectively). We observed that the median intensity histograms (Fig. 7) exhibited a more compact distribution on the forearm, corresponding to a weaker SHG signal. We also observed that the density of collagen was significantly weaker on the forearm than on the buttock ($p=0.016$ for group 1 and group 2). The collagen density and the SHG intensity tend to increase on the buttock whereas we see the opposite effects on the forearm. This could be explained by different mechanisms occurring in intrinsic and extrinsic aging.

Fig. 6

First-order analysis of the SHG signal.

Fig. 7

Mean SHG intensity histograms for the two groups and the two zones.

The same parameters have been calculated on the AF signal that gives information on the presence of elastin in the dermis. Results are shown in Fig. 8 and complete statistical analysis in the Appendix (Table 3). We first notice that in the younger group, there are few differences between the AF signals on the buttock and on the forearm. Their histograms appear very similar (cf. Fig. 9) and only the mean intensity parameter exhibits a significant difference ($p=0.0313$). The differences between groups are more significant: elastin occupies more space (higher density, $p=0.019$ on the buttock), its intensity appears greater ($p=0.024$ and 0.012 on the buttock and the forearm, respectively), and the signal has a higher entropy in the dermis of older skin ($p=0.024$ on the buttock and the forearm).

Fig. 8

First-order analysis of the AF signal.

Fig. 9

Mean AF intensity histograms for the two groups and the two zones.

These observations are in accordance with what we see in the images: the elastin network seems to be less organized in older skin and its fibers spread throughout the whole dermis. Therefore, the images appear more homogeneous, in accordance with a lower entropy level. However, aging acts differentially on the buttock than on the forearm. Looking at the median histograms for the buttock (Fig. 9), the intensity distribution spreads with age. This leads to a higher proportion of high intensity pixels, corresponding to the well-defined elastin fibers. On the contrary, on the forearm, the intensity distribution exhibits a very thin tail, thus giving a large majority of low intensity pixels on the image. This is coherent with the fact that we observe very homogeneous texture on the older group forearm skin. Therefore, we observe different changes in elastin structure with intrinsic aging than with photoaging.

3.4.

Wavelet Analysis

The analysis of the wavelet coefficients gives specific information about the images’ texture. This analysis was performed on SHG and AF signals; the results are shown in Figs. 10 and 11 for the SHG signal and Figs. 12 and 13 for the AF signal. Statistical analysis is presented in the Appendix (Tables 4 and 5).

Fig. 10

Wavelet analysis on SHG images.

Fig. 11

GGD distributions generated with the mean $\alpha$ and $\beta$ parameters calculated for SHG signal.

Fig. 12

Wavelet analysis on AF images.

Fig. 13

GGD distributions generated with the mean $\alpha$ and $\beta$ parameters calculated for AF signal.

For the SHG signal, we observe larger differences between the two skin zones than between the two age groups, as was seen in the first-order analysis. Lower $\alpha$ and $\beta$ parameters were measured on the buttock than on the forearm. The buttock skin area also presented a more compact wavelet coefficient distribution. This might reflect a lower diversity of textures as well as a lower prevalence of the most common collagen texture. No clear variations relative to aging were observed for these parameters.

The AF signal exhibits a more complex response in regards to aging and photoaging. The $\alpha$ parameter increases with aging on both skin zones (comparison between groups: $p=0.024$ on the buttock and $p=0.031$ on the forearm). This scale parameter increase demonstrates the fact that the GGD distribution is more widely spread, and it also reflects the larger diversity of image textures.

The $\beta$ value increases with aging only on the forearm skin area ($p=0.039$). It thus appears to be modified by photoaging. An increase of this shape parameter means that the distribution is flatter for high wavelet coefficients, leading to lower variance and entropy of their distribution. Thus, the texture of the elastin network in the photoaged dermis is different from that observed on the other skin areas, as it is more homogeneous. The combined increase of $\alpha$ and $\beta$ implies a reduction in the texture variety.