Effects of amniotic fluid on human keratinocyte gene expression: Implications for wound healing

Abstract Cutaneous wounds can lead to huge suffering for patients. Early fetal wounds have the capacity to regenerate without scar formation. Amniotic fluid (AF), containing hyaluronic acid (HA), may contribute to this regenerative environment. We aimed to analyse changes in gene expression when human keratinocytes are exposed to AF or HA. Human keratinocytes were cultured to subconfluence, starved for 12 h and then randomised to be maintained in (1) Dulbecco's modified Eagle's medium (DMEM), (2) DMEM with 50% AF, or (3) DMEM with 50% fetal calf serum (FCS). Transcriptional changes were analysed using microarray and enriched with WebGestalt and Enrichr. Additionally, eight diagnostic genes were analysed using semiquantitative real‐time PCR to investigate epidermal differentiation and cellular stress after HA exposure as an alternative for AF exposure. The AF and FCS treatments resulted in enrichment of genes relating to varied aspects of epidermal and keratinocyte biology. In particular, p63‐, AP1‐ and NFE2L2‐ (Nrf2) associated genes were found significantly regulated in both treatments. More genes regulated by FCS treatment were associated with inflammatory signalling, whilst AF treatment was dominantly associated with molecular establishment of epidermis and lipid metabolic activity. HA exposure mostly resulted in gene regulation that was congruent with the AF microarray group, with increased expression of ITGA6 and LOR. We conclude that AF exposure enhances keratinocyte differentiation in vitro, which suggests that AF constituents can be beneficial for wound‐healing applications.

human fetus is immersed in amniotic fluid (AF) 3,4 which is rich in epithelial growth factor (EGF), transforming growth factors alpha and beta 1 (TGFα, TGFβ1), insulin-like growth factor 1 (IGF1), erythropoietin (EPO) and tissue factor, as well as several protective factors such as lactoferrin, α-defensin, lysozyme, calprotectin and cathelicidin. 5 These factors play important roles in adult wound healing. 6 Hyaluronic acid (HA) is a major component in AF, and it also has higher concentration in younger than in older connective tissue. 7 The presence of HA in fetal wound healing is prolonged compared to that of the adult. 2 Exposure to AF shortens the time to re-epithelialization of human adult wounds, and this effect is mediated by HA. 8 Moreover, treatment of wounds with HA, in a minimally invasive in vivo human wound model, has been shown to accelerate re-epithelialisation and alter protein expression. 9 Several factors, including TGFβ1, EGF and HA, have seemed likely promising therapeutic candidates for wound regeneration. To date, their clinical impact has been moderate and it seems unlikely that single-agent therapies are the key to solving the complex processes of repair and regeneration in human cutaneous wound healing. 10 There is a need to further characterise the keratinocyte pheno-

| Collection of amniotic fluid
Residual AF was obtained from ultrasound-guided amniocenteses from women who were 14-18 weeks pregnant and gave informed consent. The collection was approved by the ethics committee at Linköping University, Sweden (register number 03-342). The AF from approximately 100 women were pooled and centrifuged at 2400× g for 5 min. The supernatant was filtered through a sterile 0.22 µm pore-size filter (EMD Millipore), and aliquots of AF were kept at −20°C.

| Cell culture treatments
At passage three (P3) subconfluence, six keratinocyte cultures in 75 cm 2 cell culture flasks were starved (DMEM only) for 12 h and then randomised into three groups. Two cultures were subsequently given only DMEM, two were kept in DMEM with 50% AF and two received DMEM with 50% FCS. All cultures were trypsinised and pelleted 24 h later. and 48 h and presented as mean with SD ( Figure S1).

| Gene expression analysis with microarray
The duplicate cell pellets were pooled and sent into plastic tubes on dry ice (−70°C) to the core facility for Bioinformatics and

| Quantitative real-time PCR
Exploring the relationships between the starvation treatment (DMEM), HA and FCS treatment in terms of keratinocyte differentiation and activity, we selected eight standard markers for semiquantitative real-time PCR ( Figure 4C). The starting primary human keratinocyte culture ("Ctrl") is used as reference (set to 1).
The DMEM serum starvation treatment ("Starve") is the same for the AF and FCS groups, but from 24 h and onwards, the treat-  (TaqMan Gene Expression Assays 20X, Applied Biosystems).
Relative quantification was performed using the ΔΔC(T)-method, resulting in fold changes relative to control. Technical replicates were averaged for each biological replicate, and duplicate cultures were used for each group. Graphs were created in Prism 7.0 (GraphPad Software).

| Microarray gene expression analysis
We performed basic quality control on the arrays, observing normalised signal distribution after RMA ( Figure 1A) that is highly correlated ( Figure 1B). Principal component analysis (PCA; Figure 1C) indicates that starvation had the greatest impact on variation (interpreted as PC1), whilst the difference between AF and FCS explained the rest of the variation in the data set (interpreted as PC2). When looking at the top 1000 expressed genes, the group clustering follows PCA results with FCS and AF groups slightly closer to each other than the DMEM group ( Figure 1D). The MvA plots for each comparison, AF versus DMEM ( Figure 1E), FCS versus DMEM ( Figure 1F) and AF versus FCS ( Figure 1G), show y-axis ("M") representing fold change (log2) and x-axis ("A") representing relative abundance. The overview of probeset-level expression changes is presented in Figure 1H ("increased") and Figure 1I  in Figure 1C), resulted in 576 probesets increased ( Figure 1H are also terms specifically related to epithelial cells ("regulation of epithelial cell differentiation" and "epithelial cell apoptotic process").
We compared the genes associated with these two terms to the 813 upregulated genes in AF treatment ( Figure 2D). Contrary to a view of set exclusivity, only three genes of 23 were unique to FCS treatment.
These were KLF7 ( Figure 2D Among the top five pathway terms in both treatments, we find "Vitamin D receptor pathway", "Nuclear receptor meta-pathway" and "TNF signalling pathway." The term "Tight junction" was the highest ranked KEGG pathway in AF treatment but did not make it into top five terms in FCS treatment. Instead, higher rankings were seen in FCS treatment for "IL-17 signaling pathway" and "Photodynamic therapy-induced NF-kB survival signaling". Analysis of the genes differentially regulated between the treatment groups (AvF; Figure 3C) leads to similar themes as the GO analysis, in Figure 2G. The top four KEGG terms clearly relate to lipid metabolism, whereas "AMPK signaling" further underscores increased cellular homeostatic activity. The IL-17, TNF and NF-kB activities are higher after FCS treatment than after AF treatment.
The relevance of the p63 (TP63) network for epidermal development and keratinocyte differentiation motivates an analysis of the representation of this network among the genes in this study.
Also important for keratinocytes is the AP1 transcriptional network.  Figure 3D-ii). Again, there is a numerical advantage to the AF treatment in AP1 representation in both upregulated and downregulated sets. In fact, about half of the AP1-associated upregulated genes in AF are shared with FCS treatment, whilst more than 75% of AP1associated genes upregulated by FCS are shared with AF. Unlike the p63-associated sets, the AP1-associated genes are more numerous among the upregulated rather than downregulated gene sets indicating an increase in AP1 activity overall. NFE2L2 ( Figure 3D-iii), also commonly known as Nrf2, 17 represents an important network related to stress, metabolism and wound-healing responses of keratinocytes. [18][19][20] NFE2L2 is represented in both treatment groups. The associated genes are more evenly distributed between treatments, but also more highly represented among the downregulated genes. The genes MAFF and MAF are regulated in both treatments, encoding for heterodimer binding partners to Nrf2. 21 Identified Nrf2 target genes, such as GCLM, SLC6A6 and PIR, 22

decrease in both treatments.
Overlaps between all the genes in Figure 3Di-iii are constructed in Figure 3D-iv to check confirm that this analysis does not rely on fully overlapping gene sets. In the upregulated Venn diagram, the  A summary of normalised and relative expression of these markers in the microarray data is shown in Figure 4D Figure 4D-ii), with KRT14 and ITGA6 remaining at starvation levels, unlike in the qPCR where ITGA6 markedly decreases. As in the qPCR data, KRT14 remains stable for the first 24 h of treatment, whilst TP63 shows an initial decrease in all treatments for the same period. The gene HSPB1 shows a slight decrease in the array data ( Figure 4D-iii), unlike in the qPCR data, and PTK2 is directionally more congruent with slight increase after AF treatment.

| DISCUSS ION
The results of our gene set enrichment analyses show two important aspects of AF treatment of primary human keratinocytes in the short term. The first is that the poststarvation rescue using AF is comparable with FCS in terms of re-establishing an expanding culture with stratification potential. We previously found similar effects, at 50% concentrations, on the re-epithelialisation of in vitro human skin wounds, 8 showing some coherence between the results on epidermal regeneration and these keratinocyte culture data. The second is that there are differences in some of the core networkassociated transcriptional activities within the phenotypic space of keratinocytes attributable to AF. Through our use of multiple  chr5q32   chr1q21   chr4q12   ADAMTSL4  ANXA9  CGN  ECM1  FLG  GJA5  HIST2H2BE  IVL  LCE3D  MUC1  NOTCH2NL  PRR9  RPS27  S100A12  S100A7  S100A9  SPRR1A  SPRR1B  SPRR2C  SPRR2G  SPRR3  SPRR4  TCHH   ADAMTSL4  CGN  HIST2H2BE  IVL  LCE3D  MLLT11  MUC1  PRR9  RPS27  S100A12  S100A7  S100A9  SH2D2A  SPRR1A  SPRR1B  SPRR2G  SPRR3  TCHH   ANXA9  CGN  CRABP2  ECM1  FLG  IVL  LCE3D  PRR9  S100A12  S100A7  SPRR1A  SPRR1B  SPRR2C  SPRR2G  SPRR3  SPRR4   chr5q32   chr1q21   chr21q   chr1q21   chr18q12   chr13q22   chr10p12   chr12q Top 100 genes Our analyses show that there is substantial similarity between the AF and FCS treatments, some of which is attributable to the initial starvation condition. Serum starvation is a well-used technique in cell culture experiments, although it is variably implemented and defined. 24 Typically, starvation involves DMEM with no or low amounts of serum. KSFM contains more amino acids and hormones (more akin to Ham's F12) than the very basal DMEM formulation and was designed with enough supplements to typically allow serumfree in vitro cultures. In this case, we chose to implement DMEMbased starvation instead of KSFM without EGF and BPE to better reproduce previous uses of starvation paradigms.
Starvation is not unequivocally related to metabolic inactivity. 24 Growing cells, such as in cultures that have not yet reached confluency, require higher levels of glucose owing to cellular anabolic activity and rRNA activity due to increased translation. The increased turnover of rRNA is helped by the UBTF-dependent transcriptional promotion of rRNA genes, 25 which we see enriched for in both AF and FCS poststarvation treatments. The enrichment of terms related to the NFE2L2 network also indicates an increase in cellular stress activity. NRFs can activate target genes termed antioxidant response elements, are crucial for mediating cellular stress responses and may be important for chemical and toxic protection in keratinocytes. 26 The NFE2L2/Nrf2 is also a KGF target that regulates gene expression and inflammation in wound healing. 18 Calcium concentration is a major determinant for keratinocyte differentiation. 27 The basic expansion medium of KSFM contains low amounts of calcium to minimise differentiation and maintain a proliferative phenotype. We did not measure or control calcium levels in this study, but DMEM, AF and FCS are all expected to contain higher levels of calcium, which likely drives an initial stimulus towards differentiation. According to manufacturer, DMEM contains 1.8-mmol calcium chloride and FCS may contain around 3.5-4 mmol, 28 whilst AF contains around 2 mmol (2-2.5 mEq 29 ). These levels are above recommended concentration for maintaining keratinocytes undifferentiated, which may accentuate the transcriptional activity around loci related to keratinocyte differentiation and epidermal development, such as chr1q21.3, and the p63 and AP1 networks.
We identified the p63-extended transcriptional network as a central regulatory theme. This can be expected, as we are dealing with keratinocyte cultures, and p63 is considered a key transcription factor for the epithelial lineages, including keratinocytes. 30,31 Simultaneously, we saw that the two treatments regulate different players within the p63 network, despite their significant gene set overlap and shared net directional change. The keratinocyte phenotype is heavily reliant on several clusters of genes, all regulated downstream of p63 and co-factors. The epithelial differentiation complex (EDC) located on chr1q21.3 contains genes for S100A1-A13, involucrin, loricrin, late cornified envelope (LCE) genes and other important keratinocyte genes. 32 We find the EDC highly enriched in our gene sets in all comparisons ( Figure 4A). There are three more documented clusters of relevance to keratinocytes: the keratin type I loci at chr17q12-q21 containing the acidic keratins (KRT9-20), 33,34 the type II loci at chr12q11-q14 containing the basic (type II) keratins (including KRT1-8) 34 and a 40 kbp locus on chr19. 35 These were represented among the transcriptional changes in all groups, but not highly enriched. To understand processes in both physiological and pathological states, we need to systematically explore and understand the involvement of chromatin remodelers at discrete stages of epidermal differentiation. 36 In our data, circumspect regulation is found for the involvement of genes such as PBX1, a known epigenetic regulator of the LCE subcluster of the EDC, 37 in contributing to differential regulation between AF and FCS treatment.
We looked for evidence on the balance between proliferation and differentiation and whether this was differentially regulated.
Firstly, the GO terms, pathways and transcription factor results did not fit into such an interpretation between AF and FCS treatments, whilst there seemed to be clear differences in cell growth, metabolism and differentiation compared to DMEM (starvation).
Differences in p63 and AP1 networks are not easily interpreted in this light, either. Whilst there is a slight advantage to the AF treatment in the number of genes associated with the subset of postulated progenitor-related p63-associated genes 38  Both themes can be related to increased AP1 network activity in AF-treated cells. Unlike the p63-associated sets, the AP1-associated genes are more numerous among the upregulated rather than downregulated genes indicating an increase in AP1 activity overall. This may be congruent with a view of AF being more supportive of in vitro epidermal development and keratinocyte differentiation, as suggested by microarchitectural attributes such as cell adhesion, polarity, hippo signalling and lipid metabolism. The dominance of "Tight junction" and "extracellular structure organisation" exemplifies that cell-cell interactions and the development of the epidermal microarchitecture are stronger in the AF treatment group. This may also be congruent with the appearance in the enrichment analysis of YAP activity and the hippo pathway. The hippo pathway, which is part of the developmental processes controlling organ size, 39  for stem cell maintenance, and its role to promote differentiation, with extracellular chemical and mechanical cues. 43 The underlying mechanisms behind the previously shown positive effects of AF on adult cutaneous wound healing 9 might be better understood by a closer look at the genes regulated in cell culture.
As an example, we found enrichment of the SALL4 transcriptional network, which is implicated in scar-free healing through a role in ECM remodelling. 44 The AF may be triggering keratinocyte differentiation that includes a higher degree of regeneration. Further studies are required to determine which factors are the crucial drivers, but a central role for HA has already been determined. 8,9 In our initial comparisons of HA and AF stimulus, we observe similar regulation and distribution of keratinocyte differentiation.
We conclude that the vast networks that govern epidermal homeostasis and the keratinocyte phenotypes show a metastability in regulation that allows both AF and FCS treatments to regulate similar functions through diverse gene set regulation. AF seems to drive maturation and regeneration in favour of inflammation in this experimental model, and the phenotype distribution in culture seems to be maintained also with HA. The analysis of gene expression changes may clarify future experiments in more complex models.

ACK N OWLED G EM ENT
The authors wish to acknowledge the invaluable technical support of Kristina Briheim and Annika Starkenberg, as well as the core facility at IBK, Linköping University.

CO N FLI C T O F I NTE R E S T S
The authors declare that there is no conflict of interest regarding the publication of this paper.

AUTH O R CO NTR I B UTI O N S
EN and GK planned the study; EN collected the AF, performed cell