5. Methods

#### 5.1. Pathway analysis

We chose several public studies about hair loss in men with AGA using microarray technology from GEO database (http://www.ncbi.nlm.nih.gov/geo/). Data from GSE66664 dataset were log2 transformed, quantile normalized, and then imported into Pathway Studio® software. We used Pathway Studio® version 9 with ResNet® database version 12.

Differentially expressed genes were identified with two-class unpaired T-test implemented in Pathway Studio®. Multiple probes were averaged by best p-value or maximum magnitude. Selection of differentially expressed genes was made using samples BAB (GSM1623702-72) and BAN (GSM1627372-441) after 0 min and after 15 min, 30 min, 3 h, 24 h, and 48 h after DHT treatment.

Cell processes and pathways enriched in over (under) DE genes were set up with GSEA algorithm or Fisher exact test. From DE list, significantly expressed genes were chosen with 0.05 p-value and 2-fold change of the expression (log change interval [1; 1]) cutoffs. In GSEA, Mann-Whitney U-test with 0.05 p-value cutoff was applied, followed by expanding of the content of proteins groups. GO biological process groups and pathway collection from Pathway Studio were used as gene set. The pathway collection included Cell Process, Receptor Signaling, Expression targets, Canonical Signal Transduction Signaling, Immunology, Toxicity, and Disease sub-collections (2160 pathways in total). Moreover, 10 pathways were constructed specifically for analysis of hair follicle-specific genes expression. These are Dermal Papilla Cells Proliferation in Anagen; Dermal papilla Cell Regression in Catagen; Keratinocytes Apoptosis; Melanocytes Differentiation; Hair Stem Cells Maintenance; Androgen Synthesis in Sebocytes; Sebocytes Proliferation; Hair follicle in Telogen; Immune System Activation in AGA; Androgen Receptor Genomic, and Non-genomic Signaling (files with pathways available in Pathways Studio® or in XML format upon request). Total numbers of entities and relations in hair follicle-related pathways are 150 and 76, respectively.

To find statistically significant regulators of DE genes, SNEA algorithm was applied. SNEA is a sub-network analysis of DE genes or proteins based on GSEA method and 2 million biological relations in ResNet® database [71]. The choice of neighbors for sub-network generating was entities upstream of DE genes (proteins/genes, complexes, functional classes) connected by relations of "Expression" or "Promoter Binding" types. We used 0.05 p-value cutoff and limit of 100 generated sub-networks ranked by best p-value. We performed SNEA to find 100 significant miRNA regulators.

Pathways statistically enriched with discovered list of regulators were found with Fisher exact test. All images have been exported from Pathway Studio with special color specification (all overexpressed genes in experiment are red, all underexpressed genes are blue, entities not measured or filtered with p-value or log change cutoff are grey).

### 5.2. Signaling pathway reconstruction

and TGFB, BMP, and WNT signaling are believed to promote shortage on amount and functional activity of dermal papilla cells in bald scalp in AGA. WNT signaling is considered to be the main trigger of telogen-anagen transition that drives repopulation of dermal papilla progenitor cells from bulge niche. Its impairment could potentially lead to AGA phenotype. Detailed understanding of molecular causes of hair loss in AGA is still has not been reached. New standardized studies of hair follicle cell cycle in human scalp are required to find humanspecific information about dermal papilla and stem cells maintenance during their lifetime

Pathway analysis of cDNA microarray data from cultured immortalized human DP cells from balding (frontal) AGA scalps reveals activation of chromatin remodeling and metaphaseanaphase transition pathways. Observed slight up-regulation of cell cycle inhibitor protein CDKN2A confirms other studies and indicates up-regulation of DNA repairing pathways. Expression of AR-related genes NOS3, EGR1, SMAD6, BTG2, and LATS2 was significantly down-regulated exclusively after DHT treatment in frontal DP cells compared to occipital bald cells from AGA scalp. Differential expression of TGFB1I1, TGFB2, THBS1, PTHLH, and ANGPT2 was not changed dramatically, but they appeared among top 25 regulators of underexpressed genes in bald DP cells samples. TGFB1I1 is a reported AR cofactor. Ligands TGFB2, THBS1, PTHLH, and ANGPT2 launch receptors signaling related to catagen progression (and ligand CTGF to telogen). MIR106A was found among the top of significant miRNA

After 1 h of DHT treatment, EGR1 and EGR2 genes reduced their expression by more than 70% in balding versus nonbalding scalp. SPRY1 and CEPBD and NR4A2 genes had also negatively differential expression trend in bald samples. CYB1P1 and CTGF had positively increased in bald samples versus nonbald after 1 h of DHT treatment. DKK1 as WNT inhibitor was underexpressed in nonbald DP samples 1 h after DHT treatment. DKK4 was, vice versa, overexpressed in nonbald DP cells in the first 1 h. NR4A2, EGR1, HES1, and NR4A2 had remarkable difference in expression between bald and nonbald DP cell samples. Unfortunately, role of these proteins was acquired in cancer-related studies that make them hardly

The obtained results suggest that AGA dermal papilla in frontal scalp area differs from occipital ones unlikely due to downregulation of proliferation and increased expression of

We chose several public studies about hair loss in men with AGA using microarray technology from GEO database (http://www.ncbi.nlm.nih.gov/geo/). Data from GSE66664 dataset were log2 transformed, quantile normalized, and then imported into Pathway Studio® software. We

applicable to the hair follicle research. Further investigation is required.

catagen triggers, but rather due to reduced expression of anagen triggers.

used Pathway Studio® version 9 with ResNet® database version 12.

renewals.

162 Hair and Scalp Disorders

5. Methods

5.1. Pathway analysis

regulators of DE genes in bald DP cells.

Pathway Studio® software and ResNet® database from Elsevier were used for building hair follicle cells specific interactive model pathways. Model signaling pathways reconstruction was based on current scientific knowledge: settled facts and hypotheses (e.g., data acquired in animal models). Pathway creation relies on published papers and searching ResNet® database for neighbors, functions, and expression of chosen proteins and molecules.

Model of cell specific molecular interactions in this article consists in an interactive signaling pathway that is represented by the graphical scheme and the annotations. In the graphical representation, the members of interactions are shown linked together with functional relations. The annotations section includes the information about properties of pathway members and relations pulled out from ResNet® database. ResNet® database storages dictionary and ontology of biological-related entities (proteins, small molecules, diseases, cell processes, cells, treatments, etc.) linked together by relations (such as protein 1-protein 2 "binding") (Figure 7). Relations depict facts supporting by sentences (references) with scientific evidences which are extracted from more than 3 million articles and manually curated when adding to pathways. The relations indicate the effect of relation (negative or positive), the direction of the relation, mechanism of the relation (such as phosphorylation), etc. [71]. The method of the pathway reconstruction required looking for common relationships to entities in the database (common downstream cell process or disease, upstream regulators, downstream expression targets, etc.) and reviewing manually the results.

#### Supplemental Information:

Supplemental Table 1 and Supplemental Table 2 could be downloaded via link:

https://data.mendeley.com/datasets/yfnrkc7r9x/1
