Compromising human skin in vivo and ex vivo to study skin barrier repair

.


Introduction
The main skin barrier function is located in the uppermost epidermal layer, the stratum corneum (SC), consisting of terminally differentiated corneocytes embedded in a lipid matrix [1].A proper lipid organization and lipid composition in this matrix are important for a well-functioning skin barrier [2][3][4].The SC barrier lipids (e.g.ceramides (CERs), fatty acids, and cholesterol) are mainly assembled in a dense orthorhombic lateral packing while a smaller lipid fraction adopts a less dense hexagonal packing [5,6].The SC CER fraction consists of a specific set of CER subclasses defined by their sphingoid base and acyl chain.Both chains can vary in carbon chain length and polar head group, resulting in a wide array of differently structured CER species, see Fig. S2.
Tape-stripped healthy human skin in vivo is used as a model to study skin with a compromised barrier.This model has been employed to study several aspects of inflammatory skin diseases [7], to examine the penetration of compounds through the skin [8,9], to increase the bioavailability of topical products in the deeper epidermal layers [10], and to generate a compromised skin barrier to examine the biological processes of skin barrier repair [11][12][13].One of these processes is restoring the lipid composition and lipid ordering (lipid properties) in the SC.However, little is known about the effect of barrier disruption by tape-stripping on the lipid properties of the regenerated SC, and whether the regenerated SC lipid properties mimic healthy or diseased skin.
An alternative model to study skin barrier repair is by using stripped ex vivo skin which regenerates SC over time in the incubator (skin barrier repair (SkinBaR) model) [14].Several aspects of the lipid properties of this model have been studied and are known to mimic to some extent the lipid properties of inflammatory skin diseases [14,15].However, it is unknown to what extent the CER composition of the SkinBaR model reflects that of tape-stripped and regenerated human skin in vivo.In the present study we examine whether the SkinBaR model can potentially replace skin barrier repair studies in clinical settings.Therefore, we determined the lipid properties in regenerated SC of the ex vivo SkinBaR model and in regenerated SC of in vivo tapestripped skin in detail.
The SC of five skin conditions was examined.SC obtained from healthy in vivo control skin and regenerated tape-stripped skin is abbreviated as Ctrl In-vivo and Reg In-vivo , respectively.SC obtained from ex vivo control skin, cultured skin, and regenerated cyanoacrylate-stripped skin is hereafter called Ctrl SkinBaR , Cul SkinBaR , and Reg SkinBaR , see Fig. 1.
The results show that the changes in CER subclass composition and CER chain length of Reg SkinBaR SC reflect those observed in Reg In-vivo

T
Both models show that regeneration alters the CER composition and mimics important aspects of the CER composition encountered in inflammatory skin diseases, such as atopic dermatitis (AD).

Compromised in vivo skin barrier
15 healthy Caucasian volunteers (7 male, 18-29 years) participated in the study.The study was approved by the ethical committee of Leiden University Medical Center and performed according to the Declaration of Helsinki.Volunteers signed written informed consent.To exclude interference with topical products, the volunteers were asked not to use soaps or cosmetics on their ventral forearms during the whole study period.After a 1-week washout period, the SC barrier was disrupted by tape-stripping an area of 3.5 × 2.5 cm on the ventral forearm using D-Squame tape (CuDerm, Dallas, TX).Consecutive tape-strips were used until trans-epidermal water loss (TEWL) values over 60 g/ m 2 /h were reached and the skin had a shiny appearance [16].An AquaFlux AF200 (Biox, London, UK) was used to monitor the TEWL.To obtain SC after the 16-day recovery period, 21 tape-strips were harvested at the regenerated site and a control site using polyphenylene sulfide tape (Nichiban, Tokyo, Japan).The first tape-strip (tape 0) was discarded.An infrared spectrum was recorded after each second tapestrip in order to examine the lipid organization (see below).

Ex vivo human skin (SkinBaR) model
Ex vivo human skin was obtained from a local hospital, used within 12 h after surgery, and handled according to the Declaration of Helsinki principles.The skin was cleaned, processed, stripped, and cultured as described before [14].SC was isolated and SC sheets were stored over silica gel under argon atmosphere until use for CER extraction or infrared spectroscopy (see below).

Ceramide analysis by LC/MS
Lipid extraction, liquid chromatography combined with mass spectrometry (LC/MS) measurements, and quantification of tape-strips and SC sheets was performed as described elsewhere [17].The obtained tape-strips were punched to Ø 16 mm, and extracted using a modified 4 step Bligh and Dyer at 40 °C.Extracts of tapes 5-8 were combined in one sample.After extraction and evaporation of the extraction solvents, samples were dissolved in heptane:chloroform:methanol (95:2.5:2.5)(v:v:v).CERs were analyzed using an Acquity UPLC H-class (Waters, Milford, MA) connected to an XEVO TQ-S mass spectrometer (Waters, Milford, MA).Processing and post-processing were performed as described before [17].The total molar amount of all CERs was determined quantitatively and used to calculate the molar percentage of each individual CER.The analyzed CER subclasses are depicted in Fig. S2, synthetic CERs that were used for calibration are listed in Table S1.

SC lipid lateral organization and conformational ordering
Fourier-transform infrared spectroscopy (FTIR) was used to examine the SC lateral lipid organization and the conformational ordering of the SC lipids of in vivo and ex vivo skin samples [18].All FTIR spectra were recorded using a Varian 670-IR spectrometer (Agilent Technologies, Santa Clara, CA).For more details, see the Supplementary Methods.

Statistical analyses
SPSS (v23, IBM, New York, NY) was used for group-wise comparisons using linear mixed models (LMMs) with nested terms.LMMs were used because of the advantages of i) analysis of multiple variables and their interactions in one model, ii) data pairing (e.g.multiple conditions within one subject) can be examined, iii) the ability to handle missing data, and iv) the use of nested variables.LMMs were used to examine differences between the Ctrl In-vivo and Ctrl SkinBaR samples, and the change in the parameter (effect size) of culturing and regeneration of SC.For more details, see Supplementary Methods.

Results
To examine if SC regeneration induced changes in CER profile, and whether SC regenerated ex vivo was affected differently than SC regenerated in vivo, the SC CER composition of the tape-strips obtained in the in vivo study and the isolated SC of the SkinBaR model were quantified using a CER lipodomics method, LC/MS.In the present study we quantified the level in each ceramide subclass, (Fig. S2 explains abbreviations used for CERs according to [19]) and calculated the mean carbon chain length (MCL, total number of carbon atoms of the sphingoid base and acyl chain) and the percentage of CERs with a total chain length of 34 carbon atoms (C34 CERs).We focus on these parameters as they correlated with skin barrier function in previous studies [3,13,20,21].

SC regeneration results in a change in subclass profile and MCL of the CERs
The changes in CER subclass profiles of the 5 different skin conditions were examined.Fig. 2 shows an altered CER profile due to regeneration with similar trends in both Reg In-vivo and Reg SkinBaR compared to their controls (Reg In-vivo vs. Ctrl In-vivo and Reg SkinBaR vs. Cul SkinBaR ).Both regenerated models showed increased percentages of S subclass CERs, whereas all three P subclasses and NH subclass CERs were decreased.The percentages of AH subclass CERs were only affected in the Reg SkinBaR samples and not in the Reg In-vivo samples.No variation in the dS CER subclasses was observed.All CER O subclasses increased in both Reg In-vivo and Reg SkinBaR samples.Table S2 shows the results of the LMM analysis which indicated that the increased S subclasses and decreased P and NH subclasses in the regenerated SC (both in vivo and SkinBaR) compared to their respective controls were statistically significant.
Changes in the percentage of acyl-CERs (EO subclasses) did not show a similar trend, but showed inverse effects in the SkinBaR model The first two conditions were obtained from human in vivo skin.The first condition is healthy in vivo control skin (hereafter called Ctrl In-vivo ).The second condition is tape-stripped and regenereted in vivo skin (named Reg In-vivo ).The other three conditions were obtained from ex vivo skin.These were a control condition for the ex vivo skin (called Ctrl SkinBaR ), non-stripped cultured ex vivo skin (Cul SkinBaR ), and ex vivo skin that was stripped and cultured to regenerate the SC (Reg SkinBaR ).
(decrease in percentage of EO subclasses in regenerated SC) compared to in vivo skin (increase in percentage EO subclasses in regenerated SC).When the percentages of EO and O CERs within each sample type were combined, there was only a minor difference in total (EO and O) percentage CERs induced by regeneration compared to control (see Fig. 2B and Table S3).
Because the average CER chain length is important for the skin barrier function, the MCL of all CERs was calculated (Fig. 2C and Table S3).In Ctrl In-vivo skin the CERs had a MCL of 46.9 carbon atoms, and in Ctrl SkinBaR and Cul SkinBaR skin the MCL did not show a significant difference and were 47.1 and 46.8, respectively.Significant reductions in MCL of the CERs, with approximately 0.4 and 1 carbon atom, were observed in the Reg In-vivo and Reg SkinBaR , respectively (Table S4).

Regenerated SC increased percentage of C34 ceramides and unsaturated ceramides
Another parameter of interest is the fraction of C34 CERs.Ctrl In-vivo  and Ctrl SkinBaR had similar percentages of C34 CERs.Compared to these controls, the mol% C34 CERs was significantly increased due to culturing and regeneration.The effect of regeneration was significantly larger in Reg SkinBaR than in Reg In-vivo (Fig. S3, Table S5).To determine if the increase in the fraction of C34 CERs was independent of changes in subclass, the percentages of C34s within each subclass was determined.Again, the mol% C34 CERs within subclasses did not significantly differ for Ctrl In-vivo and SkinBaR .Within the subclasses, the percentages of C34 CERs were increased in Cul SkinBaR samples compared to the control.S4).Dotted lines: 95% confidence interval.Bars: mean ± SD, *p < 0.05 compared to its respective control, ‡ Ctrl SkinBaR is different from Ctrl In-vivo (p < 0.05).

Furthermore, in both Reg
In-vivo and Reg SkinBaR , a further increase in the percentage of C34 within the subclasses NS, AS, NH, and AH was observed (Fig. 3, Table S5).For the CER subclasses NP, NdS, AP, and AdS a similar increase was observed (Fig. S3).Due to the limited amount of SC acquired by tape-stripping (in vivo), C34 percentages of these CER subclasses fall below the detection limit.
As C34 CERs together with EO CERs have been shown to influence the MCL in previous studies, we investigated whether an LMM with the parameters EO CERs and C34 CERs could be used to predict the MCL (Table S6).A high correlation between the predicted and measured MCL was observed (Figure 2D), indicating 88% of the variation in MCL could be explained by the change in percentage EO CERs and C34 CERs.
Another important CER composition parameter is the percentage of mono-unsaturated CERs (MuCERs).Although MuCERs were detected in in vivo tape-stripping samples, these could not be accurately determined for all samples.The MuCER percentages of the three SkinBaR sample groups were compared using an LMM (Table S7).In Ctrl SkinBaR SC 2.8 mol% MuCERs were present.This fraction significantly increased to 4.4 mol% in Cul SkinBaR samples (p = 0.04) and to 8.0 mol% in Reg SkinBaR SC (p < 0.01), see Fig. 3B.

SC regeneration does not affect lipid ordering
The lipid ordering was analyzed using FTIR.For the SkinBaR samples, CH 2 stretching vibrations in the FTIR spectra could be obtained over a temperature range from 0 to 90 °C.In order to compare the results obtained from the in vivo study and the SkinBaR model, the lipid ordering was examined using the center of gravity of the CH 2 symmetric stretching vibration peak in the FTIR spectra of SC at skin temperature (32 °C), see Fig. 4A.A peak position below 2850 cm −1 indicates a conformational ordering of the lipids.If the peak position is increased to wavenumbers above 2852 cm −1 , the lipids in the SC are mainly in a conformational disordered state, which is a liquid state.In spectra from both Ctrl In-vivo and Reg In-vivo samples, the CH 2 peak was positioned at 2848.7 cm −1 .For the Ctrl SkinBaR and Cul SkinBaR samples the peak position was located at wavenumbers of 2849.2 cm −1 and 2849.3 cm −1 , respectively, which was significantly higher than in the spectra of the in vivo samples.Although the peak position in the spectra of Reg SkinBaR is substantially higher than that of Ctrl SkinBaR , this difference was not statistically significant in the LMM (Table S8).

Ceramide composition can predict the lipid ordering
Due to the relation between the lipid composition and conformational ordering, the difference between the conformational ordering of the lipids in the SC of the in vivo and SkinBaR samples might be explained by changes in lipid composition.To obtain insight in the contribution of parameters related to ceramide composition, these were used to make a predictive model for the lipid ordering (i.e.CH 2 stretching peak position).We included ceramide related parameters in an LMM (Table S9).These parameters were i) C34 and EO percentage (together being an MCL predictor), and ii) the subclass molar ratio + + dS P H S (excluding EO ceramides) which correlated with barrier function in a previous study [17].For more details about the subclass ratio, see Supplementary Methods.Fig. 4B depicts the actual CH 2 -peak position compared to the predicted position.The r 2 indicates that 64% of  the variation in CH 2 -peak position was explained, with normally distributed residuals.This model showed that the combination of these CER-related parameters could predict the lipid conformational ordering in SkinBaR and In vivo samples.

Discussion
An important property of the validity of a skin model is that the model translates to physiological conditions.Here, we examined the translation (to physiological conditions) of the previously developed SkinBaR model [14] to the in vivo situation with a focus on the SC lipids.

CER composition in regenerated SC of SkinBaR model is similar to that in regenerated SC in vivo
In general, the CER percentages and lipid properties in the Reg SkinBaR samples were changed compared to control samples, but resembled those in Reg In-vivo samples.However, the changes observed in Reg SkinBaR were more pronounced, as discussed below.i) Both regenerated SC models showed decreased mol% of CER P subclasses and increased percentage of CER S subclasses compared to control SC samples.This indicates an imbalance in activity of the enzymes acid sphingomyelinase and glucocerebrosidase, both involved in post-synthetic modification of CERs [22].Although the expression of these enzymes was not affected in the SkinBaR model [14], their activity could have been altered [23].Besides the changes in S and P subclasses, both the Reg SkinBaR and Reg In-vivo models also showed similarities in a decreased CER MCL, and an increase in percentages of C34 CERs compared to their controls.ii) The percentage of unsaturated CERs was predominantly increased in Reg SkinBaR SC.In agreement with the elevated percentage of unsaturated CERs, the expression of Stearoyl-CoA desaturase, a lipid processing enzyme responsible for hydrocarbon chain unsaturation, was increased in the SkinBaR model [14].iii) The only substantial difference between the in vivo model and the SkinBaR model was the change in percentages of EO CERs, which was increased in the Reg In-vivo samples and decreased in the Reg SkinBaR samples.iv) The lipid ordering was not significantly affected by regeneration in both in vivo and ex vivo conditions.This might be due to less deep stripping in the present study compared to previous studies.Previously, lipids in the regenerated SC of the SkinBaR model were less ordered when almost all SC was removed [15].Although the difference in CH 2 stretching peak position between Ctrl In-vivo and Ctrl SkinBaR was significant, a difference in spectrum collection methods did not play an important role: the CER composition could be used to accurately predict the lipid ordering.Therefore, the difference in peak position mainly originated from a difference in CER composition.

SC regeneration process in vivo and ex vivo: a comparison
Previously, it has been shown that both in clinical settings and in the SkinBaR model the proliferation rate immediately after barrier disruption is higher than after a recovery period of several days [7,14].In the SkinBaR model, the proliferation rate was normalized after 8 days of culturing [14], whereas in in vivo skin, the proliferation rate remained elevated for at least 10 days [7].This indicates that there is a difference in velocity at which skin barrier repairs.This may be caused by an inflammatory response in healthy in vivo skin during skin barrier repair that slows down the repair process [24].The fact that the systemic response is lacking in the SkinBaR model substantiates the faster barrier repair process as described above.This also indicates that, during skin barrier repair, the in vivo skin is closer to skin homeostasis, which may be an underlying factor for the smaller deviation in changes in CER composition in vivo compared to the ex vivo SkinBaR model.Since barrier function recovery of stripped healthy human skin takes at least 14 days [25], this period and the period of 8 days in the SkinBaR model are sufficient long to study the effect of formulations on skin barrier repair.

Modulation in lipid barrier composition in regenerated skin is very similar to that in inflammatory skin diseases
It was shown that tape-stripping healthy skin induced parakeratosis [7], and induced an inflammatory response by secretion of various cytokines [26,27].Some of these cytokines also play a role in AD and/ or psoriasis [28,29] and have been shown to induce changes in the lipid biosynthesis in the viable epidermis, such as a reduction in lipid chain length, also observed in the present study [30,31].
When comparing the lipid properties of regenerated SC to that in SC of inflammatory diseased skin, many similarities were observed.These were i) similar changes in CER subclasses [20], ii) a decreased MCL [32], iii) an increase in percentages of C34 CERs [20,32], and iv) an increased percentage of unsaturated lipids [33].Most of these changes correlated with an impaired skin barrier function [3,13,20].Although the degree of changes vary, many of these alterations in CER composition have also been observed in other inflammatory skin diseases like psoriasis, and Netherton syndrome [31,34].Additionally, a less dense lipid packing observed in Reg SkinBaR SC [15] also corresponds to findings in SC of these inflammatory skin diseases: a higher fraction of lipids adopting a less dense hexagonal lateral packing has been reported [35].
In the present study, we focused on the relation between lipid composition and lipid organization in SC and how changes in the lipid composition in regenerated SC of the SkinBaR model are related to the changes in lipid composition and organization in regenerated SC in a clinical setting.For this study we focused on the composition and organization of the free extractable lipids, yet we acknowledge that the bound ceramides are an important part of the stratum corneum.Analysis of the bound ceramides might be the subject of future studies.

Conclusion
This study shows that the changes in lipid properties in both the Reg SkinBaR and Reg In-vivo models are very similar and mimic the lipid properties in inflammatory skin diseases.This concerns changes in CER subclass profiles, MCL, percentage of CER unsaturation, and conformational ordering.Therefore, the SkinBaR model can be used to predict the in vivo response with regards to the lipid composition after application of topical barrier repair treatments aiming to restore the normal CER composition.By doing so, the need for clinical studies is reduced.

Fig. 1 .
Fig. 1.Explanation of the five different SC samples.The first two conditions were obtained from human in vivo skin.The first condition is healthy in vivo control skin (hereafter called CtrlIn-vivo ).The second condition is tape-stripped and regenereted in vivo skin (named RegIn-vivo ).The other three conditions were obtained from ex vivo skin.These were a control condition for the ex vivo skin (called Ctrl SkinBaR ), non-stripped cultured ex vivo skin (Cul SkinBaR ), and ex vivo skin that was stripped and cultured to regenerate the SC (Reg SkinBaR ).

Fig. 2 .
Fig. 2. In vivo and SkinBaR model ceramide profiles and mean ceramide chain length show a very similar alteration compared to their controls.A-B) The total molar amount of all CERs was set at 100%.Bars show the CER subclass distribution in molar percentage of the total amount of CERs.N = 15 for in vivo samples, N = 3, 6, and 9 for Ctrl SkinBaR , Cul SkinBaR , and Reg SkinBaR , respectively.C) The MCL of all CERs.D) Correlation of the MCL with the predicted MCL (for residuals: see Fig. S4).Dotted lines: 95% confidence interval.Bars: mean ± SD, *p < 0.05 compared to its respective control, ‡ Ctrl SkinBaR is different from Ctrl In-vivo (p < 0.05).

Fig. 3 .
Fig. 3.The percentage of C34 ceramides and unsaturated CERs in the SC increased in the regenerated SC.A) The percentage of C34 per CER subclass.B) The percentage of the monounsaturated CERs in mol%.Data of 3 donors was analyzed, indicated by different shades of gray.Bars: mean ± SD, *p < 0.05 compared to its respective control.

Fig. 4 .
Fig. 4. CH 2 symmetric stretching vibrations in spectra of in vivo skin and SkinBaR models and correlation of predicted values for the CH 2 stretching peak with observed values.A) The CH 2 stretching peak positions in the FTIR spectra at skin temperature of 32 °C.Bars: mean ± SD, ‡ Ctrl SkinBaR is different from Ctrl In-vivo (LMM: p < 0.05).B) The linear mixed model used to predict the CH 2 stretching peak includes percentages of C34 CERs and EO CERs, and two CER subclass ratios (for residuals: see Fig. S5).Dotted lines: 95% confidence interval.