Human skin equivalents cultured under hypoxia display enhanced epidermal morphogenesis and lipid barrier formation

Human skin equivalents (HSEs) are three-dimensional cell models mimicking characteristics of native human skin (NHS) in many aspects. However, a limitation of HSEs is the altered in vitro morphogenesis and barrier formation. Differences between in vitro and in vivo skin could have been induced by suboptimal cell culture conditions, of which the level of oxygen in vitro (20%) is much higher than in vivo (0.5–8%). Our aim is to study how external oxygen levels affect epidermal morphogenesis and barrier formation in HSEs. In the present study, fibroblast and keratinocyte monocultures, and HSEs were generated under 20% (normoxia) and 3% (hypoxia) oxygen level. In all cultures under hypoxia, expression of hypoxia-inducible factor target genes was increased. Characterization of HSEs generated under hypoxia using immunohistochemical analyses of morphogenesis biomarkers revealed a reduction in epidermal thickness, reduced proliferation, similar early differentiation, and an attenuated terminal differentiation program compared to normoxia, better mimicking NHS. The stratum corneum ceramide composition was studied with liquid chromatography coupled to mass spectrometry. Under hypoxia, HSEs exhibited a ceramide composition that more closely resembles that of NHS. Consequently, the lipid organization was improved. In conclusion, epidermal morphogenesis and barrier formation in HSEs reconstructed under hypoxia better mimics that of NHS.

Three-dimensional (3D) human skin equivalents (HSEs) are mainly used for toxicology screenings and for research purposes to increase the knowledge on skin biology, or particularly the skin barrier formation. Various types of 3D skin models are currently available, including epidermal skin models, full thickness skin models (FTMs) ( Supplementary Fig. 1a,b), explant models, and skin-on-a-chip models, listed by increased complexity and physiological relevance [1][2][3][4][5][6][7] . As reviewed recently by Niehues et al. 8 , while all of these models have advantages and drawbacks, a known limitation of the in vitro developed skin models is the altered barrier formation and resulting reduced functionality when compared to native human skin (NHS) [9][10][11] . This could lead to inadequate in vitro -in vivo correlations regarding pharmacokinetics and permeability testing of compounds or misinterpretation of adverse outcome pathways 12 .
The permeability barrier of the epidermis resides mainly in the uppermost layer, the stratum corneum (SC). In this layer of dead cells, corneocytes are keratinized and cross-linked to form impermeable structures 13,14 .
In between the corneocytes, the lipids form the only continuous pathway through the SC ( Supplementary  Fig. 1c,d). The lipids form a highly structured matrix organized in densely packed crystalline lamellar phases 15 ( Supplementary Fig. 1e,f). This lipid matrix contributes substantially to the barrier functions of the skin, which are protecting from external substance or allergen entry and restricting water loss 15,16 . The most abundant lipid classes within the SC are cholesterol, free fatty acids (FFAs) and ceramides 17 . Several ceramide subclasses exist, categorized based on their head group architecture [18][19][20][21] . When considering the total ceramides (CERs total ) subclasses analyzed in this study, these consist of ceramides (CERs) and acylceramides (CERs EO) ( Supplementary  Fig. 2). Alterations in lipid composition most probably lead to an altered organization and can thereby reduce

Results
Fibroblast and keratinocyte monocultures under hypoxia. To fully understand the effects of oxygen levels on skin cell cultures, we initially characterized monocultures of fibroblasts or keratinocytes developed at 20% O 2 for normoxia or 3% O 2 for hypoxia. The spindle shaped morphology of fibroblast monocultures was unaffected after 6 days in culture under hypoxia when compared to normoxia (Fig. 1a). Next, we performed gene expression analysis on pre-selected HIF target genes involved in metabolic reprogramming (glucose transporter 1 (GLUT1) and pyruvate dehydrogenase lipoamide kinase isozyme 1 (PDK1)) and the angiogenic molecule vascular endothelial growth factor A (VEGFA). Significant upregulation of these HIF target genes was detected in fibroblasts when cultured under hypoxia (Fig. 1b). To investigate the effect of oxygen levels on keratinocyte growth and differentiation, these were cultured in low (proliferating) and high (differentiating) calcium (Ca 2+ ) medium for several days in normoxia and hypoxia (Fig. 1c). Similar to fibroblasts, keratinocyte morphology was not altered when cultured under hypoxia. However, keratinocytes showed a significant decrease in cell proliferation rate under hypoxia from 3 days in monoculture (Fig. 1d). Furthermore, gene expression analysis of subconfluent proliferating and confluent differentiating keratinocytes also showed upregulation of the pre-selected HIF target genes under hypoxia (Fig. 1e). This indicates that reduction from 20% to 3% oxygen level is sufficient to induce adaptation and metabolic reprogramming of fibroblasts and keratinocytes to hypoxic conditions. Human skin equivalents generated under hypoxia. Next, evaluation of the effect of oxygen level in 3D co-cultures of fibroblasts and keratinocytes was performed by generation of full thickness models (FTMs) under normoxia or hypoxia. In both conditions, FTMs have a similar appearance based on macroscopic observations and the detection of four distinguishable epidermal layers (i.e. the stratum basale, stratum spinosum, stratum granulosum and SC) in hematoxylin and eosin (HE) stained cross sections (Fig. 2a). The epidermis remained well-ordered under hypoxia and the thickness of the viable epidermis was significantly reduced, better resembling that of NHS (Fig. 2b). The number of corneocyte layers in the SC of FTMs in hypoxia was significantly reduced as compared to normoxia and was similar to NHS (Fig. 2c). Likewise as in monocultures, upregulation of HIF target genes VEGFA, PDK1 and GLUT1 was observed in the epidermis of FTMs generated under hypoxia (Fig. 2d).

Morphogenesis of FtMs generated under hypoxia.
To obtain further information on the influence of oxygen levels on FTMs, the expression of specific biomarkers for dermal and epidermal morphogenesis was examined in the various FTMs generated under normoxia and hypoxia and these were compared to NHS (Fig. 3). The expression of the cell proliferation biomarker Ki67 remained restricted to the basal cell layer in all conditions. The proliferation was considerably higher in FTMs generated under normoxia as compared to NHS. However, the proliferation index of FTMs generated under hypoxia was lower than in normoxia, thereby mimicking that of NHS better. The early differentiation program was executed correctly in all conditions as indicated by suprabasal keratin 10 (K10) expression. The late and terminal differentiation programs were attenuated, as the keratinocytes in the granular layer appeared more flattened based on the expression of involucrin (INV), loricrin (LOR) and filaggrin (FLG). Epidermal cell activation was determined by expression of biomarkers K16 and K17. In FTMs generated in hypoxia, increased expression of K16 was observed as compared to normoxia, whereas K17 remained unexpressed. Both biomarkers were absent in NHS. Next, we determined the deposition of the basement membrane (BM) proteins collagen type IV (COL IV) and laminin 332 (L332). These were more continuously present in NHS and had a patchier similar appearance in both FTMs. The distribution of fibroblasts was studied by the expression of the mesenchymal biomarker vimentin (VIM). In FTMs generated under normoxia or www.nature.com/scientificreports www.nature.com/scientificreports/ hypoxia, the distribution was continuous and comparable, whereas in NHS there was a clear distinction between papillary and reticular dermis. Finally, the fibroblasts differentiation was determined by assessment of expressed alpha smooth muscle actin (ASMA), specific for myofibroblasts. No differences for this fibroblast characteristic were found in the FTMs generated under normoxia or hypoxia. In NHS, ASMA expression was also observed in dermal regions surrounding the capillaries, which were absent in FTMs (data not shown). stratum corneum lipid composition of FtMs under hypoxia. Thereafter, the lipid barrier formation was studied in FTMs generated under normoxia or hypoxia and compared to NHS. To this end, lipids were extracted from the SC, which revealed an equal total lipid content in the SC of FTMs in both conditions and NHS (Fig. 4a). Next, we determined the CERs total composition within the lipid extract using a liquid chromatography coupled to mass spectrometry analysis. In FTMs generated under normoxia and hypoxia, all twelve major CER (d) Proliferation of keratinocytes in low calcium culture medium under hypoxia, indicated by mean ± 95% confidence interval. (e) Expression of HIF target genes in keratinocytes cultured in low or high calcium medium in hypoxia as compared to normoxia, indicated by mean ± s.e.m., n = 4. Statistical differences are noted as *, ** or ***, corresponding to P < 0.05, <0.01, <0.001. Scale bar indicates 50 µm.
www.nature.com/scientificreports www.nature.com/scientificreports/ subclasses were detected (Fig. 4b). As compared to the ion map of NHS, in that of FTMs there is an increased presence of CERs with a mass below 600, indicative for increased presence of CERs with a shorter carbon chain length. Quantification of the CERs total in absolute molar values revealed a similar cumulative amount of CERs total per mg of SC in FTMs and NHS (Fig. 4c). Subsequently, the CER and CER EO subclass profiles were analyzed (Fig. 4d). An altered CER subclass profile in hypoxic FTMs was detected with significant higher abundance of subclass NP. This also improved the CER subclass ratio defined by + + dS P H S ( ) , which mimics more closely that of NHS. The CER EO subclass profile of the FTMs did not differ, although the composition of CERs EO in FTMs deviated from that in NHS. Next, the mean chain length (MCL) of the CERs and CERs EO was determined (Fig. 4e,f). High similarity was observed between the FTMs for the MCL of the CERs and of the CERs EO. As compared to that of NHS, there is a significant reduction observed in the MCL of the CERs, but not in that of the CERs EO. In line with the MCL, no differences in the carbon chain length distribution of the CERs total in the FTMs generated at both conditions were detected ( Supplementary Fig. 3). When comparing the carbon chain length distribution of FTMs to that of NHS, the main difference is the increased presence of CERs with ≤42 carbon atoms and the reduced presence of CERs with ≥44 carbon atoms ( Supplementary Fig. 3a). No major differences were observed in the carbon chain length distribution of CERs EO of FTMs, although substantially altered as compared to that of NHS ( Supplementary Fig. 3b). Hereafter, the level of unsaturation in the CER subclasses AS and NS was determined. These subclasses were selected, as changes in the level of unsaturation in these subclasses properly indicate differences in the level of unsaturation for CERs total (Helder et al., in preparation). The unsaturation index indicates the percentage of monounsaturated CERs per subclasses. When comparing the unsaturation index between FTMs developed under normoxia to hypoxia, a similar unsaturation index was observed (Fig. 4g). As monounsaturation in ceramides is naturally low in NHS, this was not further analyzed. stratum corneum lipid organization of FtMs under hypoxia. After assessment of the ceramide composition, the organization of the lipids in the intercorneocyte space was determined in FTMs generated www.nature.com/scientificreports www.nature.com/scientificreports/ under normoxia or hypoxia and compared to NHS. The lamellar organization was examined by small-angle X-ray diffraction. The diffraction profile with a first, second, and third order of diffraction attributed to the long periodicity phase (LPP) were detected in FTMs, irrespective of oxygen level (Fig. 5a). From the positions of the diffraction peaks, the repeat distance of the LPP in both FTMs was calculated. This revealed an increase in the repeat distance of the LPP in FTMs generated under hypoxia (Fig. 5b). In NHS, by using recrystallization studies, the LPP repeat distance has been reported before to be around 13 nm 34 . Therefore, the LPP repeat distance in the FTMs generated in hypoxia better resembles the repeat distance in NHS. The lateral packing of FTMs generated under normoxia or hypoxia and NHS was evaluated by Fourier transformed infrared spectrometry (FTIR) measurements. Examination of the methylene rocking vibration in the 700-740 cm −1 region of the spectrum provides information about the lipid lateral packing; two peaks at 720 and 730 cm −1 indicate lipids assembled in the very dense orthorhombic packing, whereas a single peak at 720 cm −1 is visible when the lipids form a hexagonal packing 15 . In FTMs, a weak doublet shoulder was visible at low temperatures (Fig. 5c). This indicated that a small fraction of the lipids assembled in an orthorhombic lateral packing, but that most lipids form a hexagonal lateral packing. When increasing the temperature, the intensity of the 730 cm −1 shoulder gradually reduced. At 40 °C, the lipids assembled only in a hexagonal packing, which was similar for normoxia and hypoxia. In NHS, the 730 cm −1 peak intensity was higher at low temperatures. This indicated that a higher fraction of lipids was organized in the orthorhombic lateral packing in NHS. When increasing the temperature, the intensity of the 730 cm −1 peak reduced, but a doublet was still present at 40 °C, indicating a substantial level of lipids still adapted an orthorhombic packing. The conformational ordering of the lipids was examined to determine the ordered-disordered phase behavior of the lipid matrix. This could be obtained by assessment of the methylene symmetric stretching vibration peak position in the FTIR spectrum. When lipid chains adapt a higher conformational ordering, these are fully extended. This is indicated by methylene stretching vibration peaks positioned below a wavenumber of 2850 cm −1 . Vibration peaks shifted to a higher wavenumber (2852 cm −1 ) when higher disordering of lipids is present. Peak positions of methylene symmetric stretching vibrations of FTMs and of NHS are plotted against temperature range of 0-40 °C (Fig. 5d). The conformational ordering of the lipids in FTMs generated at normoxia or hypoxia was comparable. In NHS, lowest vibrations were detected, indicative for highest conformational ordering of the lipid tails. In FTM and NHS, the lipids adopt a crystalline phase, as the observed frequency is below 2850 cm −1 at skin surface temperature of 32 °C. www.nature.com/scientificreports www.nature.com/scientificreports/

Discussion
The main objective of this study is to investigate the influence of oxygen level on dermal and epidermal morphogenesis and barrier formation in HSEs. Our results demonstrate that a reduced oxygen level (i) activates HIF signaling in vitro leading to metabolic reprogramming of primary human fibroblasts and keratinocytes in mono and 3D cultures, (ii) decreases epidermal thickness in FTMs, (iii) affects epidermal morphogenesis, and (iv) alters the SC lipid composition and organization. These alterations led to better resemblance of HSEs to NHS.
The results obtained in this study are in line with previous studies, which demonstrated an altered expression of several barrier proteins including FLG, LOR and INV under hypoxia, indicative for a restructured barrier formation 33,35 . Furthermore, silencing of HIF-α affected the terminal differentiation program in a human epidermal model that resulted in impaired stratification of the epidermis 32 . A similar observation was made in vivo, where it has been shown that the lack of HIF-α results in an impaired terminal differentiation program and barrier formation in a murine model 33 . This emphasizes the importance of oxygen levels and HIF signaling in the formation of the SC barrier. An abnormal differentiation under hypoxia has also been reported, as determined by a reduced K1/K10 and FLG/LOR expression in HSEs after 24 hours at 1% oxygen 36 . However, the level of ROS scavenger molecules (vitamin C and E) and the duration and level of oxygen reduction (1% vs 3%) differ between this and our study, in part explaining the divergent outcomes. Complementary to our results, hyperbaric oxygen treatment (hyperoxia) of HSEs accelerated the formation of the epidermis, concomitantly with an increased proliferation rate and enlargement of the granular cell layer, which again emphasizes the critical role of oxygen during skin tissue engineering 37,38 .
The lipid barrier formation is enhanced in FTMs generated under hypoxia, as it mimics that of NHS more closely than in FTMs generated under normoxia. Of interest is the significant improvement in CER subclass NP www.nature.com/scientificreports www.nature.com/scientificreports/ level and improved subclass ratio + + dS P H S ( ) under hypoxia. An increase in this subclass or in the subclass ratio is also one of the key alterations observed in vivo in diseased skin conditions 39,40 and is shown to be directly correlated with the skin barrier function 41 . Lipid elongation or the level of unsaturation in subclasses AS and NS is not affected by culturing under hypoxia, indicating that other regulatory pathways are involved in these processes. In this study, we thoroughly characterized the barrier formation in HSEs by full quantification of the ceramide composition. Due to the technical complexity of the data analysis, no unsaturated CERs EO with a linoleic chain (contributing ~17-19% to FTM and NHS, Helder et al., in preparation) and saturated CERs EO with an oleic chain (contributing ~23% to FTM, Helder et al., in preparation) were analyzed. However, the interpretation of the data is not affected by this. The ceramide composition affects the lipid organization as shown in this study and others 40,42 , but also for the barrier functionality, as demonstrated in in vivo (diseased) skin conditions 40,43 , in HSEs 9,42 and in pure lipid model membranes 44,45 . The influence of reduced oxygen level on the presence of other major lipid classes (free fatty acids, cholesterol) could be of interest in future studies.
Incorporation of the optimized oxygen micro-environment in future projects could open new possibilities to study different aspects of skin biology. One of these is a 3D co-culture with vascular endothelial cells. These are often supplemented with high concentrations of synthetic VEGF [46][47][48] , whereas under hypoxia the production of endogenous VEGF is upregulated, potentially leading to a more supportive micro-environment. This facilitates studies on dermal angiogenesis, which is an important aspect in wound healing and tumorigenesis 49,50 . Furthermore, a low oxygen micro-environment could also be applied in human skin organ culture (hSOC) 51 . This enables the determination of the effect of hypoxia on other skin resident cell populations, which are currently missing in FTMs. Additionally, the interplay between oxygen levels in the skin micro-environment with the ability of the epidermis to induce anti-microbial responses during bacterial (S. aureus) infections was reported recently 52,53 . Essentially, hypoxia lowers the threshold for induction of these protective responses, suggesting a role of oxygen level during the dynamic anti-microbial barrier 54 , besides its role on the structural barrier formation.
In summary, we have shown in monocultures and in FTMs that external oxygen level directly affects cellular proliferation rate, epidermal morphogenesis and barrier formation. Importantly, we have shown that sub-optimized external conditions are in part responsible for altered in vitro skin tissue reconstruction. www.nature.com/scientificreports www.nature.com/scientificreports/

Materials and Methods
Isolation and culturing of primary fibroblasts and keratinocytes. Declaration of Helsinki principles are followed during obtainment of primary cells from human skin tissue originating from surplus breast tissue of healthy donors. Experiments were conducted in accordance with article 7:467 of the Dutch Law on Medical Treatment Agreement and the Code for proper Use of Human Tissue of the Dutch Federation of Biomedical Scientific Societies (https://www.federa.org/codes-conduct). As of this national legislation, coded surplus tissue can be used for scientific research purposes when no written objection in made by the informed donor, i.e. an informed opt-out system. Therefore, additional approval of an ethics committee regarding scientific use of surplus tissue was neither requested nor required, as stated earlier 55 . Primary cells were isolated from surplus human mamma skin after separation of the epidermis and dermis by incubation in 2.4 U/mL dispase II (Roche, Almere, The Netherlands). Isolation and culturing of primary cell suspensions was performed as described before 42,56,57 .
All isolated primary cells were tested and found negative for mycoplasma contamination. Amount of viable proliferating primary cells was determined by trypan blue exclusion assay using a Bio-Rad TC20 ™ automatic cell counter (Bio-Rad, Hercules, CA, USA) according to manufacturer's instructions.
Generation of full thickness models. Full thickness models (FTMs) were generated in two steps and cultured on inert membranes using a transwell system (Corning Transwell cell culture inserts, membrane diameter 24 mm, pore size 3 μm; Corning Life Sciences, The Netherlands) as described before 9,42 . In the first step, the dermal equivalents were generated using primary fibroblasts (1.2-1.5 × 10 5 ) in passage number 2-5, hydrated rat-tail tendon collagen (4 mg/mL), 1 M NaOH, Hank's Balanced Salt Solution, and 5% fetal bovine serum (FBS; GE Healthcare, Chicago, IL, USA). After polymerization at 37 °C, the dermal equivalents were cultured 7 days as described elsewhere 9,42 . In the second step, the epidermal equivalents were generated by seeding primary keratinocytes (2.5 × 10 5 /model) in passage number 1 directly onto each dermal equivalent as reported before 9,57 . The FTMs were kept submerged for total of four days. Hereafter, the FTMs were lifted to the air-liquid interface and cultured for 14-15 days under normoxia (20%) in the Memmert INC153med CO 2 incubator (Memmert, Schwabach, Germany) or under hypoxia (3%) in the Heracell ™ 240 CO 2 incubator (ThermoFisher, Waltham, MA, USA). FTM batches were generated with 5 different primary cell donors.
Gene expression. RNA isolation. Isolation of RNA from 2D and 3D cell cultures was performed according to instructions provided by the manufacturer using a FavorPrep Tissue Total RNA Mini Kit (Favorgen, Ping-Tung, Taiwan). A single exception was the incorporation of a DNA digestion step, which occured during 15 minutes after the initial washing steps using an RNAse-free DNAse solution (Qiagen, Hilden, Germany). Subsequently, the isolated RNA was eluted and the concentration of the nucleic acid content was determined using a NanoDrop ™ UV-Vis Spectrophotometer (ThermoFisher).
Quantitative real-time polymerase chain reaction. A total of 500 ng isolated RNA was used to synthesize complementary DNA with an iScript ™ cDNA Synthesis Kit (Bio-Rad). Then, quantitative real-time polymerase chain reactions were performed with the SYBR Green Supermix (Bio-Rad) using the CFX384 Touch ™ real-time PCR detection system (Bio-Rad). Used settings were described previously by Mieremet et al. 58 . Details on primer sequences used in this study are described in Supplementary Table 1 Protein expression. General morphological assessments were performed on 5 μm paraffin embedded tissue sections after staining with hematoxylin and eosin (HE; Klinipath, Duiven, The Netherlands). Immunohistochemical and immunofluorescence analyses of protein expression were performed on 5 μm sections. FFPE material was deparaffinized and rehydration, whereas frozen material was dried overnight and fixed in acetone for 10 minutes. For FFPE material, heat-mediated antigen retrieval in citrate buffer was performed, followed by blocking with normal human serum (Sanquin, Leiden, The Netherlands). Stainings were performed using the streptavidin-biotin-peroxidase system (GE Healthcare) according to the manufacturer's instructions or using indirect immunofluorescence, as described elsewhere 42,59 . Exceptions were the ASMA and collagen type IV staining, for which no or protease-mediated antigen retrieval was performed respectively 42 . Visualization of the sections occurred using a light microscope (Zeiss Axioplan 2, Carl Zeiss BV, Breda, The Netherlands) or a fluorescence microscope (Leica CTR5000, Leica, Wetzlar, Germany). Application of solely secondary antibodies revealed no interfering background or unspecific staining (Supplementary Fig. 4a). Specifications of the materials are provided in Supplementary Table 2.
Determination of the number of corneocyte layers. Snap-frozen samples were cut 5 μm thickness using a Leica CM3050S cryostat (Leica Biosystems, Nussloch, Germany). Sections were placed on a Superfrost ® Plus Micro Slide (VWR, Radnor, USA), air-dried overnight, and fixated in ice-cold acetone for a minimum of 10 minutes. Samples were stained with a 1% (w/v) safranin O (Sigma-Aldrich, St. Louis, MO, USA) solution for 1 minute. Subsequently, samples were immersed with a 2% (w/v) KOH solution for 10-15 minutes, which induced swelling of the corneocytes. Samples were covered and imaged directly after the procedure to determine the number of corneocyte layers (Supplementary Fig. 4b).
www.nature.com/scientificreports www.nature.com/scientificreports/ Estimation of epidermal thickness and proliferation index. Procedures to estimate the epidermal thickness and the proliferation index were described before 60 . In short, epidermal thickness was determined by digital quantification of the viable epidermis in the HE stained samples. Proliferation index was determined by counting the number of Ki67 positive nuclei in a region of the basal epidermal layer covering at least 100 keratinocytes. A minimum of three different epidermal regions were included. stratum corneum lipid composition. Stratum corneum isolation. Isolation of the stratum corneum occurred after overnight incubation in a 0.1% (w/v) trypsin/PBS solution at 4 °C followed by incubation at 37 °C for 1 hour. The SC was peeled off the epidermis and was washed with 0.1% trypsin inhibitor (Sigma) and Millipore water. The isolates were air-dried and stored at dry conditions.
Lipid extraction and liquid chromatography -mass spectrometry. Extraction of barrier lipids from the SC was carried out using a modified Bligh and Dyer procedure 61 as described by Boiten et al. 62 . To determine the amount of material extracted from the SC, the SC was dried and weighed before as well as after the extraction procedure. The obtained lipid extract was dissolved in a suitable volume of heptane:chloroform:methanol (95:2.5:2.5 v:v:v). The lipids were analyzed and quantified using normal phase liquid chromatography -mass spectrometry (LC-MS) analysis, as described by Boiten et al. 62 . In brief, SC lipid extracts were separated on a PVA-Sil column (5 μm particles, 100 × 2.1 mm ID; YMC, Kyoto, Japan) with an Acquity UPLC H-class (Waters, Milford, MA, USA). CERs total were detected by an XEVO TQ-S mass spectrometer (Waters). LC-MS measurements occurred in full scan mode with settings for time between 1.25-8.00 minutes at m/z 350-1350 and for time between 8.0-12.5 minutes at m/z 500-1350.
Data quantification. The area under the curve (AUC) of a chromatogram from the mono-isotopic mass of the main fragment was integrated. For a selected number of CERs, the relative abundance of the corresponding unsaturated CER containing two 13 C was more than 10%. When this occurred, the AUC of the two 13 C containing CER was calculated using its natural isotope distribution. This AUC was subtracted from the one of with which it overlapped. Quantification of the molar amount of ceramides occurred using the AUC of the chromatogram of any specific ceramide, the internal standard (INST) of CER N(24deuterated)S(18) and a three dimensional response model based on compound properties and a calibration curve from a limited number of synthetic ceramides according to the method described by Boiten et al. 62 (Supplementary Fig. 5). Nomenclature of the CER subclasses is followed according to Motta et al. 63 and is shown in Supplementary Fig. 2. stratum corneum lipid organization. Small-angle X-ray diffraction. Small-angle x-ray diffraction analyses occurred at station BM26B of the European Synchrotron Radiation Facility using methods specified before by Mojumdar et al. 64 . The detector was calibrated with silver behenate and cholesterol. Sample measurements occurred twice for 90 seconds at two detector positions. Distance of the sample to the pilatus1M detector was approximately 2.1 meter. Profiles were obtained by converting the detector image from Cartesian (x, y) to polar (r, θ) coordinates by integrating over scattering angle θ. The scattering intensity was determined as a function of the scattering vector q, which was defined as = , where n indicates the order of the diffraction peak. Determination of the repeat distance of each lamellar phase was performed using combined data from the three main diffraction peaks.
Fourier transform infrared spectrometry. Fourier transform infrared spectroscopy (FTIR) spectra were obtained using a Varian 670-IR spectrometer (Agilent Technologies, Santa Clara, CA, USA). The system was equipped with a broad-band mercury cadmium telluride detector, cooled with liquid nitrogen, and connected to an external controlled heating device. Before measurements, the SC was sandwiched between two AgBr cells in parallel to their surface. A constant stream of dry air was applied for 30 minutes to prevent signal interference by water condensations. The FTIR settings are described by Mieremet et al. 58 . Spectra were analyzed and deconvoluted using Varian Resolutions Pro 5.2.0 software (Agilent Technologies).
statistics. Statistical examinations were performed with GraphPad Prism v7.00 (GraphPad Software, La Jolla, CA, USA). In general, statistical testing was performed with 1-way or 2-way ANOVA with Tukey's multiple comparisons post-test, otherwise indicated. Significance is shown for comparison between normoxia and hypoxia (with lines) and for comparison between FTMs and NHS (above NHS), otherwise indicated. Significant differences were indicated by *, ** or ***, which correspond to P values < 0.05, <0.01, <0.001, respectively.

Data Availability
The raw data obtained during this study and processed data used for analysis is available upon reasonable request from the senior authors.