Causal Network Analysis of Head and Neck Keloid Tissue Identifies Potential Master Regulators

Objectives/Hypothesis: To generate novel insights and hypotheses in keloid development from potential master regulators.
Study Design: Prospective cohort.Methods: Six fresh keloid and six normal skin samples from 12 anonymous donors were used in a prospective cohort study. Genome-wide profiling was done previously on the cohort using the Infinium HumanMethylation450 BeadChip (Illu- mina, San Diego, CA). The 190 statistically significant CpG islands between keloid and normal tissue mapped to 152 genes (P < .05). The top 10 statistically significant genes (VAMP5, ACTR3C, GALNT3, KCNAB2, LRRC61, SCML4, SYNGR1, TNS1, PLEKHG5, PPP1R13-a, false discovery rate <.015) were uploaded into the Ingenuity Pathway Analysis software’s Causal Net-work Analysis (QIAGEN, Redwood City, CA). To reflect expected gene expression direction in the context of methylation changes, the inverse of the methylation ratio from keloid versus normal tissue was used for the analysis. Causal Network Analysis identified disease-specific master regulator molecules based on downstream differentially expressed keloid-specific genes and expected directionality of expression (hypermethylated vs. hypomethylated).Results: Causal Network Analysis software identified four hierarchical networks that included four master regulators (pyroxamide, tributyrin, PRKG2, and PENK) and 19 intermediate regulators.Conclusions: Causal Network Analysis of differentiated methylated gene data of keloid versus normal skin demonstrated four causal networks with four master regulators. These hierarchical networks suggest potential driver roles for their down- stream keloid gene targets in the pathogenesis of the keloid phenotype, likely triggered due to perturbation/injury to normal tissue. INTRODUCTION Keloids were first described centuries ago dating back to the Smith Papyrus of Ancient Egypt surgical text. They are benign fibroproliferative tumors that extend beyond the original wound.1 Keloids predominate in darker skinned individuals, including blacks and His- panics.2 The incidence varies from 4.5% to 16%.3 Lower incidences have been suggested in the head and neck area after surgery.4 Initial studies explored observatio- nal linkages, but over time research has focused on finding genetic aberrations, which could be the etiological basis for keloids. Understanding the molecular genetics of keloid pathogenesis can lead to improved interven- tions and treatment.With the increased popularity of epigenetics, where changes in gene functioning/expression are not a result of changes in DNA sequence, many researchers are interested in its contribution to the pathogenesis of dis- ease.5 DNA methylation typically occurs in promoter regions and can interfere with normal transcription of genes.6,7 These factors can activate (hypomethylation) or repress (hypermethylation) gene expression, and they appear to be instrumental in cancer tumorigenesis. Stem cells and progenitor cells affected by epigenetically dys- regulated factors have led to tumor and cancer progres- sion.8 The reversible nature of the epigenetic factors led to identifying linkage networks between DNA and gene expression. Once networks are found, epigenetic drugs can be created to alter gene expression through either activation or repression.5,9 Within these networks, find- ing a master regulator could have large implications in clinical practice. The master regulator definition has been used loosely over the years. A consensus article described a master regulator as a gene that is expressed at the inception of a cell line or type, which then regulatesseveral downstream factors directly or indirectly to spec- ify differentiation.10 For example, in the Drosophila fly, BX-C and ANT-C genes are responsible for initial control of body segment differentiation.11 The importance of identifying master regulators in keloid studies has yet to be unveiled.Genome-wide profiling was previously done on the six keloid and six normal skin cohorts from 12 anony- mous donors using the Infinium HumanMethylation450 BeadChip assay (Illumina, San Diego, CA). A three-tier system for statistical rigor was used for the detection of biologically relevant methylation markers. The 190 stat- istically significant CpG islands between keloid and nor-mal tissue mapped to 152 genes (P < .05), of which 63%(96) are hypomethylated as compared to 37% (56) that are hypermethylated.7 The top 10 statistically signifi- cant genes (false discovery rate <0.015) were identifiedas VAMP5, ACTR3C, GALNT3, KCNAB2, LRRC61,SCML4, SYNGR1, TNS1, PLEKHG5, PPP1R13-a.QIAGEN’s Ingenuity Pathway Analysis (IPA) was used for uploading the 10 genes for analysis in the soft- ware’s Causal Network Analysis (CNA) (QIAGEN, Red- wood City, CA). CNA is a feature that helps in understanding causal connections between diseases, genes, and networks of upstream regulators. The data- base found in IPA, which makes CNA possible, is based on the Ingenuity Knowledge Base, which is a collection of 5 million observations made from biomedical literature.12CNA connects upstream regulators or master regulators to downstream differentially expressed genes, and the expected directionality of expression based database pre- dictions. These master regulators in causal networks can have a different number of “hops” based on the depth number, which translates into the number of “intermediate regulators” to reach target genes. For exam- ple, a depth of three consists of three hops such as master regulator!intermediate regulator #1!intermediate regu- lator #2!dataset target/molecule.12The gene direction was assessed by the inverse of the methylation ratio from keloid versus normal tissue, as gene expression is inverse of methylation direction. The top 10 statistically significant genes with the lowest false discovery rate were selected for this analysis. Fol- lowing CNA analysis, the top master regulator networks were ranked using an additional filter of an absolute z score value of 2.0. A positive 2 denotes significant pre- dictors of activation and a negative 2 as significant pre- dictors of inhibition. The Henry Ford Health System Institutional Review Board committee approved the study. RESULTS We identified four hierarchical networks via the CNA software, which included four master regulators (Fig. 1): pyroxamide (Fig. 2b), tributyrin (Fig. 3), PRKG2 (Fig. 4), and PENK (Fig. 5)13,14; and 19 intermediate regulators: HDAC1, HDAC3, HDAC4, PPAR-c, TP53, RUNX2, RB1, P38 MAPK, CDK4, Pkc(s), PPAR-c, MYOD1, PSEN1, TNS, FOXO1, ERK, Akt, GSK3 b, and VEGF. The figures also show the overlay function used to relate biological processes, in this case fibrosis/fibro- blast, to the causal network.The CNA demonstrated that pyroxamide and tribu- tyrin are predicted to be activating drugs, z score 2.0 and P value overlap (P 5 .025) and z score 2.0 and P value overlap (P 5.0025), respectively. In our hypothetical model, pyroxa- mide and tributyrin inhibit HDAC1, HDAC3, and HDAC4. The target downstream molecules GALNT3, PPP1R13L, and VAMP5 are upregulated, and TNS1 is downregulated in both of these networks (Figs. 1–3).The CNA demonstrated that PENK is predicted to be inactivated (z score 22.0, P 5 .037). In our hypotheti- cal model, an inhibited PENK leads to activation of P38 MAPK, CDK4, and Pkc(s) and inhibition of RB1, which are the first set of intermediate regulators. The result of this network is upregulation of VAMP5, GALNT3, and KCNAB2, and a downregulated TNS1 (Fig. 4).The CNA demonstrated that PRKG2 is predicted to be inactivated (z score 22.0, P 5 .031). In our hypotheti- cal model, an inhibited PRKG2 indicates activation of GSK3 b and inhibition of FOXO1, ERK, and Akt, which are the first set of intermediate regulators. The end result of this network is upregulation of VAMP5, PPP1R13L, SCML4, and KCNAB2 (Fig. 5). DISCUSSION Keloids have the propensity for continued growth due to increased proliferation and failure of apoptosis.15 Studies have shown the downregulation of apoptotic genes in keloid tissue including those that promote and inhibit apoptosis.16 Dysregulation of genes important in extracellular matrix formation and immunity have been implicated in the pathogenesis of keloids.1,17 Keloids are characterized by excessive deposition of collagen, result- ing from aberrant extracellular matrix production.18 Furthermore TGF-b is highly activated in fibrotic proc- esses, causing persistent activation of myofibroblasts, collagen gene expression, and excessive extracellular matrix synthesis. Inhibition of TGF-b by SIRT1 was shown to attenuate fibrotic responses, by inhibiting fibroblast migration/contraction, Smad-dependent TGF-b signaling, stimulation of collagen synthesis, and a-SMA expression.19 Also, bFGF has long been implicated in hypertrophic scars and keloids, because it regulates extracellular matrix synthesis and degradation and expression of TGF-b. bFGF can induce rapid apoptosis in myofibroblasts20,21 leading to inhibition of wound con- traction.20 Studies have demonstrated that bFGF can inhibit scarring by inhibiting TGF-b1/Smad-dependent pathway,20,21 a decrease in membrane metalloprotease (MMP)-1, and collagen type 1 and 2.22 MMP-2 (gelatin- ase-A) and MMP-9 (gelatinase-B) are involved in tissue remodeling and wound healing,23 as they bind to gelatin, collagens, and laminin.24 These findings could be exploited in the treatment of keloids.The hierarchical network master regulators, tribu- tyrin and pyroxamide (activated drugs) and PENK and PRKG2 (inactivated molecules), are discussed based on three hypotheses. Application of an inhibitory chemical or drug in a causal network experiment results in the opposite expression effect as seen in the literature.12,25 Hypothesis 1: Pyroxamide and Tributyrin Are Predicted to be Activated DrugsTributyrin and pyroxamide have the same effects on the signaling cascade with the same intermediate regulators and gene products (Figs. 1–3). There are four classes of histone deacetylases (HDAC): class I: 1, 2, 3, 8; class II: 4, 5, 6, 7, 9, 10; class III: SIRT 1, 2, 3, 4, 5, 6,7; and class IV 4: 11.26 To date, three histone deacetylase inhibitors (HDACIs) have been approved for lymphoma therapy by the US Food and Drug Administration: Vori- nostat (SAHA, Zolina), romidepsin (Istodax, FK228, FR901228, depsipeptide), and belinostat (Beleodaq, PXD-101).27 Histone acetyltransferases add an acetyl group to the N-terminus, causing activation of gene expression due to relaxation of chromatin into euchro- matin, making it more amenable to transcription.28 To silence genes, HDACs must remove the acetyl group that condenses chromatin into heterochromatin, making transcription more difficult.28 HDACIs have a multitude of functions, including a critical role in transcription reg- ulation and cell viability.29 This functions as an epige- netic therapy for cancer. Additionally, HDACIs have been found to decrease inflammatory mediators such as cytokines.30 Pyroxamide is an HDACI, thus increasing histone acetylation.31 Pyroxamide is a synthetic derivative of hydroxamic acid (vorinostat) that inhibits class I HDACs.26 Tributyrin is a colorless oily triglyceride found in butter. Tributyrin, prodrug of butyrin,33 is also an HDACI.34 As seen in Figures 2B and 3, these drugs pre- dict inhibition on HDAC 1, 3, and 4, which represent the first set of intermediate regulators and predicted activa- tion of the second set of intermediate regulators PPAR-c, TP53, and RUNX2.Consistent with the reported literature, trichostatin A, a structural derivative of vorinostat, inhibits class I and II HDACs and was found to decrease collagen, type 1,35 production by suppressing TGF-b in keloid fibroblasts.36 PPAR-c has predicted activation by HDAC3 and HDAC1 (Figs. 2B and 3). This relationship is supported by the obser- vation that PPAR-c agonists (troglitazone)37 and HDAC3 inhibitors cause activation of PPAR-c,38 resulting in inhibi- tion of the TGF-b1/Smad signaling cascade.35 In addition, the CNA relationship (Figs. 2B and 3) is supported by the development of hepatic fibrosis that is induced by HDACIs, leading to inhibition of PPAR-c.39P53 is a transcription factor that induces the intrinsic apoptotic pathway, and mutations result in deregulated cell growth and carcinogenesis.26 P53 under- expression from sequence mutations support the aber- rant unchecked growth of keloids due to inability to activate pro-apoptotic genes.15,40 Additionally, low P53 expression coupled with increased P63 (causes inhibition of transcriptional activation of P53) expression leads to dysregulated cell growth in keloids.41 Fortunately, HDA- CIs hyperacetylate the p53 tumor suppressor resulting in protein stabilization leading to increased expression of p53 target genes.26 RUNX2, another intermediate reg- ulator, codes for a transcription factor that is aberrantly active in cancer cells.42 HDACIs profoundly inhibit RUNX2 expression by disruption of its transcription.42 RUNX2 represses apoptosis via inhibition of the p53 tumor suppressor protein following DNA damage, thus leading to uncontrolled cell growth. Tributyrin has different biological functions, such as having a role in differentiation of fibroblast cell lines (Fig. 3). A study by Richards et al. investigated the role of tributyrin added to a fibrin sealant in decreasing pos- terior laminectomy adhesions in exposed dura in sheep. Notably, fibrosis scores on magnetic resonance imaging and histology were lower compared to controls, without adverse effects. In addition, there is an inverse relation- ship between tributyrin concentration and human neo- natal foreskin fibroblast proliferation.33 Tributyrin has been shown to inhibit colon cancer cells and other malig- nant cells through induction of apoptosis and upregula- tion of caspases and terminal differentiation of different cell lines.44–47 A different study demonstrated that butyrate inhibited the growth of human gingival fibro- blasts, likely due to induction of G0/G1 and arrest of the G2/M cycle.48 Additionally, other HDACIs have been found to cause cell cycle arrest in G1/G2, leading to cell growth inhibition/apoptosis26 without affecting normal cells.28,49 In a different study, butyrate was shown to induce senescence of rat embryo fibroblasts showing strong tumor suppressor activity, protein degradation, and cytoskeleton reorganization.50The PENK gene codes for a preprotein, which results in multiple protein products, that bind to the opioid receptors l and d (Figs. 1 and 4). These facilitate the perception of pain.13 P38 MAPK is highly involved in cancer cells. P38 MAPK inhibition leads to dysregu- lated cell growth and acts as the cellular break (tumor suppressor activity) by activating P53 when exposed to oncogenic stress.51 Similarly, the MEK-ERK pathway can activate P38 MAPK, leading to premature senes- cence in mice51 and human fibroblasts impacting growth negatively by conferring tumor suppressor activity and by stabilizing P53 mRNA.52 It has been shown that induction of PENK via oxidative stress can elicit the translocation and phosphorylation (activation)53 of P38 MAPK.54 P38 MAPK has also been implicated as a regulator of inflammatory molecules and inhibitors55 and MMPs.56 More specifically, MMP-2 and MMP-9 expres- sion has been found to be regulated by the P38 MAPK pathway.56 Overall, activation of P38 MAPK can depend on the affected cell type; it can play a role in the prolif- eration of cancer cells but also has been implicated in cell senescence, cell cycle arrest, and apoptosis.53 The vast range of biologic function is due to the variety of hundreds of substrates per MAPK. The PRKG2 gene codes for a protein that inhibits renin secretion and helps chloride/water secretion in the small intestine, amongst many other functions (Figs. 1 and 5).58 PRKG2 is abundant in intestinal mucosa, brain nuclei, chondrocytes, and lung tissue.59 It is an ubiqui- tous protein involved in multiple biological processes.60 Decreased levels of PRKG2 cause P53 downregulation, which leads to impaired terminal differentiation and enhanced proliferation.60A study by Cao et al. demonstrated that PRKG2 over- activity attenuates bFGF-induced proliferation and migra- tion by inhibiting the MAPK/ERK signaling pathway.61 Another mechanism in which there is an increased fibrotic response is activation of the ERK pathway, another inter- mediate regulator in our analysis, which then helps bFGF inhibit TGF-b-induced aSmad in fibroblasts.21One of the limitations of the study is the lack of confirmation of the network analysis results at the molecular level for mechanistic verification. Neverthe- less, this information may help to identify novel hypoth- eses to investigate molecular mechanisms underlying keloid pathogenesis. Furthermore, some genes may appear within a network whose function is unknown or irrelevant to the question. Another limitation preventing replication of study results in an in vivo model would be lack of availability of the chemical or molecule in the causal network. There are multitudes of software for pathway analysis such as IPA, DAVID (Database for Annotation, Visualization and Integrated Discovery), PANTHER (Protein Analysis Through Evolutionary Relationships), and others that access different knowl- edge databases, and as such pose inherent limitations to capturing similar relationships. In addition, databases may not be completely up to date, thus increasing the chances of missing key relationships. The statistical test- ing within the databases also varies. CONCLUSION CNA of differentiated methylated gene data of keloid versus normal skin identified four master regulators. As noted earlier, the causal networks had intermediate regu- lators that were involved in cell proliferation, senescence, apoptosis, and tumor suppression indicating a causative relation, indirect or direct, with the development to keloids. CNA provides a conceptual snapshot to hypothe- size relationships based on published data with our experi- mental data. This study is an illustration of IPA’s CNA module to decipher novel master regulators for causal rela- tionships with downstream epigenetically deregulated keloid target genes for further consideration as potential therapeutic targets. Future directions for translation of this novel data include application to a biological system, such as keloid fibroblasts, and potentially evaluating gene expression for mechanistic Pyroxamide corroboration.