Differential Gene Expression in Erlotinib-treated Fibroblasts

Abstract

Background:

Therapies targeting the epidermal growth factor receptor (EGFR) result in a painful rash, the most common and debilitating toxicity among patients with non-small cell lung cancer (NSCLC) who take EGFR tyrosine kinase inhibitor (TKI) therapy; however, predicting the development and the severity of the rash is difficult.

Objective:

To examine how erlotinib—an EGFR TKI that NSCLC patients take to stop or slow tumor growth—altered the transcriptome of dermal fibroblasts.

Methods:

Dermal fibroblasts (ATCC® PCS-201-012™) were seeded in cell culture flasks, grown under standard conditions, and transferred to cell culture dishes. Cells were treated once daily for three days with erlotinib 100nM (n = 5), erlotinib 1μM (n = 5), vehicle 1μM (DMSO) (n = 5), or no treatment (n = 5). Total RNA was extracted using a standard TRIzol® method and hybridized using Affymetrix GeneChip® Human Genome U133 Plus 2.0 arrays. Raw intensities generated from the arrays were normalized using Robust Multi-array Average method and analyzed using ANOVA in Limma R software. Differentially expressed genes were analyzed using Ingenuity Pathway Analysis to identify canonical or noncanonical signaling pathways enriched in this dataset.

Results:

We selected genes for investigation based on their potential role in wound healing (AQP3), rash development (CCL2), fibroblast activation (PALLD), cancer and cancer progression (GDF-15, SLC7A11, MMP12, DIRAS3), and cell cycle control (CDC6). We were able to validate four of these genes by both Western blot analysis and qPCR (MMP12, CCL2, CDC6, and SLC7A11).

Discussion:

If found predictive of rash in future studies using patient samples, our findings may help to identify those at risk for severe rash so that (a) the dose of EGFR TKI therapy may be adjusted; (b) additional treatments for the rash can be developed; and/or (c) precise, patient-centered interventions can be developed so that patients with cancer can better self-manage their rash and adhere to EGFR TKI treatment.

Lung cancer is the leading cause of cancer deaths in the United States, with 154,050 deaths estimated for 2018 (Siegel, Miller, & Jemal, 2018). Lung cancer is usually diagnosed at an advanced stage and the five-year survival rate is only 14% (Siegel, Miller, & Jemal, 2017). Efforts to improve survival while maintaining quality of life (QoL) for patients with cancer have led to development of targeted therapies (Agostara, Carruba, & Usset, 2008; Mattia et al., 2018), such as tyrosine kinase inhibitors (TKIs). Targeted therapies act on or within specific pathways in cancer cells, while sparing normal cells, thereby increasing safety and therapeutic efficacy for patients with cancer. Use of targeted therapy represents one of the growing number of approaches to personalized treatment of cancer (Mattia et al., 2018) because patients receive therapy based on the unique genetic profile, or subtype, of his/her cancer.

Because epidermal growth factor receptor (EGFR) is overexpressed in 80%–85% of patients with non-small cell lung cancer (NSCLC) (Molina, Yang, Cassivi, Schild, & Adjei, 2008; Press & Lenz, 2007), targeted therapy development for this patient population has focused on inhibition of the EGFR, a family of four membrane-bound proteins (ErbB1 [EGFR]; Human Epidermal Growth Factor Receptor Type 1 [HER1], ErbB2 [HER2/neu], ErbB3 [HER3], and ErbB4 [HER4]) (Jones, 2003) that are structurally similar to tyrosine kinase proteins (Gaborit et al., 2011; Lynch et al., 2004). TKIs targeting the EGFR (e.g., erlotinib, afatinib, gefitinib) competitively inhibit binding of adenosine triphosphate to wild-type EGFR (Lacouture, 2007), preventing EGFR activation and inhibiting cell signaling events that regulate cell proliferation, differentiation, survival, angiogenesis, and metastasis (Chen et al., 1987).

EGFR TKIs are clinically effective for individuals with NSCLC (Shepherd et al., 2005; Yang et al., 2012) and are considered to be standard of care for first-line treatment of patients with NSCLC that have EGFR activating mutations (Rosell et al., 2012; Zhou et al., 2011). EGFR is also expressed in normal cells (e.g., skin, sebaceous glands) (Eames, Kroth, Flaig, Ruzicka, & Wollenberg, 2010; Mascia et al., 2010); thus, mechanism-based, or “on-target” toxicities (i.e., toxicities shared by all agents that inhibit the same target), are common. The most frequent and debilitating toxicity of EGFR TKI therapy is an acneiform rash that patients develop on the face, neck, scalp, and upper trunk (Lacouture et al., 2011; Solomon & Jatoi, 2011). The rash is physically and emotionally painful (Lacouture, 2011; Pérez-Soler et al., 2005; Wickersham et al., 2014; Wong et al., 2010), associated with itching and infections, and financially costly (Lacouture & Rodeck, 2013), resulting in decreased patient QoL (Joshi et al., 2010) and adherence to EGFR TKI therapy (Boone et al., 2007; Lacouture, 2007; Luu et al., 2011; Takeda, Okamoto, & Nakagawa, 2015). The incidence of rash has varied across erlotinib trials (33.0–79.0%) (Hirsh, 2011). One review of medical claims of over 22,000 patients receiving EGFR TKI therapy found that nearly 10% had a recorded rash (Chen et al., 2015).

EGFR TKI-related rash is associated with better clinical outcomes for patients with NSCLC (Pérez-Soler, 2006). Petrelli et al. (2012) found that skin rash was an independent predictor of overall survival (p < 0.00001) and progression-free survival (p < 0.00001). Moreover, individuals who developed a Grade 2–4 rash based on National Cancer Institute Common Terminology Criteria for Adverse Events were more likely to respond to TKI therapy (response rate 42% vs. 7%; p < 0.0001). As such, the rash is considered a biomarker of EGFR TKI therapy response, but the relationship between rash and response remains unexplained, and predicting who will develop the rash and how severely is difficult. Other known risks include age, gender, skin phototype, ultraviolet light exposure, smoking, and use of EGFR TKI therapy in combination with chemotherapy (Lacouture & Balagula, 2018). Age ≥ 70 years has also been associated with increased risk of rash with erlotinib treatment (Wheatley-Price, Ding, Seymour, Clark, & Shepherd, 2008).

It is believed that inflammation is responsible for many of the signs and symptoms associated with EGFR TKI rash, but the primary event leading to rash development appears to be altered EGFR signaling in basal keratinocytes (Lacouture, 2006). The commonly accepted model of EGFR inhibitor-induced skin toxicity by Lacouture (2006) proposes that EGFR inhibition in basal keratinocytes leads to cell growth arrest, chemokine expression, and abnormal cell maturation and differentiation. This in turn prompts inflammatory cell recruitment, which leads to a cutaneous injury and the development of patient signs and symptoms. Research by Mascia et al. (2013) and Lichtenberger et al. (2013) have provided immune-related mechanistic insight into EGFR inhibitor-related rash in keratinocytes of mice and of patients that supports this model. Further, preclinical data support the idea that epidermal keratinocytes affect epidermal immunity (Lacouture & Rodeck, 2013). As such, in vitro inhibition or activation of EGFR may result in regulatory effects on cytokine and chemokine production, such as increased production of chemokine (C-C motif) ligand2 (CCL2)/monocyte chemotactic protein-1 (MCP-1) in response to EGFR inhibition (Mascia, Mariani, Girolomoni, & Pastore 2003). On the other hand, activation of EGFR results in increased C-X-C motif chemokine ligand 8 (CXCL8) and granulocyte-macrophage colony-stimulating factor (GM-CSF) expression (Mascia et al., 2010).

Despite the research examining underlying mechanisms of EGFR TKI-related rash in keratinocytes, the standard of care for rash treatment has not changed significantly over the last 10 years. Current rash treatment includes a combination of topical anti-inflammatory agents (i.e., corticosteroids) and oral antibiotics (e.g., doxycycline, minocycline) that have antibacterial and anti-matrix metalloproteinase activity, both of which address the cutaneous inflammation and opportunistic bacterial infections associated with EGFR TKI-related rash (Lacouture & Rodeck, 2013). Unfortunately, up to 30% of patients will redevelop the rash despite standard rash treatment (Lacouture & Rodeck, 2013). Translation of preclinical studies into new effective treatments has been difficult, perhaps due to the necessity that topical or systemic agents used to reduce skin rash do not interfere with the therapeutic effects of the EGFR TKIs themselves.

One approach to better understand EGFR TKI-associated rash includes examination of EGFR TKI effects on other dermal cell types, such as fibroblasts. Dermal fibroblasts play a critical role in maintaining extracellular matrix integrity and wound healing (Lacouture, Reilly, Gerami, & Guitart, 2008; Ross, 1969) by migrating to the wound site, manufacturing extracellular matrix components (e.g., collagen), making granulation tissue (Grinnell, 1994), and promoting wound contraction (Gabbiani, Hirschel, Ryan, Statkov, & Majno, 1972). EGF activation of leads to activation of the EGFR, causing increased cell migration in skin fibroblasts (Blay & Brown, 1985; Chen et al., 1993). Though much research has focused on the inflammatory and immune-mediated effects of EGFR inhibition in keratinocytes, immunoreactivity can be found in the whole epidermis of normal skin (Nanney, Magid, Stoscheck, & King, 1984).

To develop precise, personalized treatments for patients experiencing EGFR TKI-rash, understanding the variability in development and severity of the rash is needed. Further, the mechanisms underlying the correlation of rash severity and drug activity has yet to be fully understood (Pérez-Soler, 2006). One approach to addressing these gaps is to examine whether erlotinib treatment of dermal fibroblasts produces gene expression changes that may inform biological pathways; however, little research has focused in this area. One study examined the correlation between single nucleotide polymorphisms in absorption, distribution, metabolism, and excretion (ADME) genes and skin rash by genotyping peripheral blood samples (Arbitrio et al., 2016), but the study was retrospective and exploratory in nature. Another demonstrated that variants of the ABCG2 (ATP Binding Cassette Subfamily G Member 2) −15622C/T and 1143C/T polymorphisms were associated with ABCG2 expression levels and with erlotinib serum concentrations, but a definitive association with EGFR TKI rash could not be made (Rudin et al., 2008). Therefore, our purpose was to examine how erlotinib treatment altered the transcriptome of dermal fibroblasts. Identification of rash-related genes or pathways could provide further mechanistic insight into EGFR TKI-related rash and a basis from which personalized treatments for EGFR TKI-rash could be developed.

Methods

We conducted wet-bench experiments using a commercially available immortalized fibroblast cell line to investigate how erlotinib treatment altered the transcriptome of dermal fibroblasts. Because no human subjects were enrolled, institutional review board approval was not required.

Cell Culture

Human dermal fibroblasts (ATCC® PCS-201-012™) were grown in cell culture flasks treated with poly-L-lysine hydrobromide (Sigma, St. Louis, MO). Cell culture medium was supplemented with 20% fetal bovine serum (FBS; Fisher Healthcare, Houston, TX) and 1% 10,000 U/ml penicillin G, 10,000μg/ml streptomycin and 25μg/ml amphotericin B (Invitrogen™ Life Technologies™, Grand Island, NY) according to a published protocol (Villegas & McPhaul, 2005). Fibroblasts were maintained in 75cm² tissue culture flasks (Corning, Tewksberry, MA) in an incubator at 37°C enriched with 5% carbon dioxide. Media was replaced every three to four days. Subculturing upon confluence was achieved with 0.25% trypsin/EDTA solution (Invitrogen™ Life Technologies™, Grand Island, NY), which was inactivated after two minutes with fibroblast culture media containing FBS. Only fibroblasts from passage five or less were used for the experiments to limit chromosomal rearrangement. Cells were transferred to 35mm cell culture dishes (~ 5,000 cells/dish) (Sigma, St. Louis, MO), grown to confluence, and then treated once daily for three days with erlotinib 100nM (n = 5), erlotinib 1μM (n = 5), vehicle 1μM (DMSO) (n = 5), or no treatment (n = 5) (Cao et al., 2006; Petty et al., 2004; Ware et al., 2013; Wertheimer et al., 2013).

Erlotinib Treatment

Erlotinib (Tarceva®, OSI-774; Roche Life Science, Indianapolis, IN) was purchased from LC Laboratories (Woburn, MA) and dissolved in 100% dimethyl sulfoxide (DMSO; Sigma, St. Louis, MO). We chose doses of erlotinib from the literature that were previously used in cell culture experiments (Cao et al., 2006; Petty et al., 2004; Ware et al., 2013; Wertheimer et al., 2013). Erlotinib was added to culture media to produce two final concentrations of 100nM and 1μM. We added the same volume of solvent (DMSO) to all samples, vehicle, and staurosporine (positive control for cell death; Enzo Life Sciences, Farmingdale, NY).

Indirect Immunofluorescence

To confirm that cultured cells in vitro were indeed fibroblasts, immunostaining was performed according to a published protocol (Goodpaster et al., 2008). Proliferating untreated cells were removed from 35mm tissue culture dishes with 0.25% trypsin/EDTA solution, which were inhibited with fibroblast culture medium containing 20% FBS. Fibroblasts were plated on 25mm glass cover slips at a density of approximately 5,000 cells and then incubated overnight at 37°C/5% CO₂. Fibroblasts were treated with vehicle (DMSO), erlotinib 1μM, or no treatment (three slips per condition) daily for three days at the same time each day. Coverslips were collected, washed twice with phosphate buffered solution (PBS; Quality Biological, Gaithersburg, MD), fixed with 3% paraffin (Sigma, St. Louis, MO) for 15 minutes at room temperature, permeabolized with 1% Triton X-100 (Sigma, St. Louis, MO) for 30 minutes at room temperature, and then blocked with 5% bovine serum albumin (BSA; Sigma, St. Louis, MO) for one hour at room temperature to prevent nonspecific binding of antibodies. Cells were incubated for one hour at room temperature with the primary antibody (TE-7; EMD Millipore, Billerica, MA) diluted (1:100) in buffer containing 1% BSA and 0.1% Tween20 (Sigma, St. Louis, MO) in PBS. Coverslips were washed five times with PBS and then incubated with the secondary antibody (FITC conjugated donkey anti-mouse IgG; Jackson ImmunoResearch Laboratories, Inc., West Grove, PA) diluted (1:100) in the same buffer for one hour at room temperature. Slips were mounted on slides with DAPI Prolong Gold (Invitrogen™ Life Technologies™, Grand Island, NY) to visualize the nucleus.

Survival Analysis: MTT assay

To examine whether erlotinib treatment affected cell viability, we performed an MTT (3-[4,5-dimethylthiazol-2-yl]-2,5-diphenyl tetrazolium bromide) assay using the Vybrant® MTT Cell Proliferation Kit (Invitrogen, Carlsbad, CA). Fibroblasts were plated in a 96-well plate according to published procedures (van Meerloo, Kaspers, & Cloos, 2011). Erlotinib was added to cell culture media to produce four final concentrations of 1μM, 2μM, 5μM and 10μM (Ali et al., 2008; X. Chen et al., 2013; Ling, Lin, & Pérez-Soler, 2008; Orzáez, Guevara, Sancho, & Pérez-Payá, 2012). These concentrations were chosen to reflect the range of toxicity of erlotinib in vitro and clinically achievable doses (1–2 μM). The same amount of solvent (DMSO) was added to all samples, vehicle (DMSO; positive control for erlotinib), and staurosporine (positive control for cell death; Enzo Life Sciences, Farmingdale, NY) to control for any potential toxicity associated with the solvent. Each well was treated with one dose daily for three days at the same time each day. The MTT assay was performed according to a standard protocol and read at 570nm/1(s). The experiment was performed in triplicate. The no treatment condition was designated as the reference group.

Total RNA Extraction

Total RNA was isolated using a standard TRIzol®-based extraction method (Invitrogen™ Life Technologies™, Grand Island, NY) according to established methods from the literature (Dorsey et al., 2009). Briefly, fibroblasts were homogenized in TRIzol® Reagent in cell culture dishes, homogenized into solution, and then transferred to 1.5mL Eppendorf tubes. Chloroform was added, and then the mixture was separated into phases via centrifugation. Total RNA was precipitated from the aqueous phase with isopropanol and then washed with 75% ethanol. Total RNA was re-suspended in 20μL diethylpyrocarbonate water (Invitrogen™ Life Technologies™, Grand Island, NY) and stored at −80°C until further analysis.

RNA Labeling, Human Genome Microarray Processing, and Analysis

We used the GeneChip® Human Genome U133 Plus 2.0 array from Affymetrix (Santa Clara, CA), which provides coverage of the whole human transcriptome (54,675 probe sets across transcriptome) and enables unbiased whole transcriptome analysis. All microarrays were processed by one person in the same laboratory following a standardized laboratory protocol to minimize nonbiological technical bias (Dorsey et al., 2009). Total RNA (1μg) was used to prepare double-stranded cDNA. The quality and quantity of RNA was assessed by spectrophotometer (Nanodrop 1000; Thermo Scientific) and by RNA Integrity Number using the Agilent Technologies Bioanalyzer. Samples with RNA integrity numbers (RIN) > 8 were used for cDNA synthesis. Double-stranded cDNA was then used as template in an in vitro transcription reaction to prepared biotinylated cRNA. The biotinylated target was fragmented and hybridized to the probes contained on the expression arrays. After 16-hour hybridization, the cocktail was removed and stored at −80°C. Arrays were washed and stained in the Affymetrix fluidics station and scanned in the 3000 7G scanner.

We used the R programming environment (v.3.2.2) (http://cran.r-project.org) to conduct microarray analysis using packages developed by the Bioconductor Project (www.bioconductor.org) and clustering, Principal Component Analysis, and the Bioconductor affyQCReport (v.1.48.0) package to check the quality of the raw intensities (.CEL files) generated for each array. Raw intensities were background-corrected for each array, normalized across arrays, and summarized using the Robust Multi-Array Average (RMA) method (Irizarry et al., 2003) from the Bioconductor affy (v. 1.48.0) package to generate normalized expression values at the probe level for each array. The genes that were differentially expressed between conditions: (a) erlotinib 1μM versus combined control (no treatment and vehicle); (b) erlotinib 100nM versus combined control (no treatment and vehicle); and (c) erlotinib 1μM versus 100nM, were identified by using one-way analysis of variance (ANOVA) embedded in Bioconductor Limma (v.1.48.0) package (Smyth, 2004). Error inflation due to multiple testing was corrected by using the false discovery rate (FDR) (Reiner, Yekutieli, & Benjamini, 2003) with an a priori alpha of 0.01. Thus, only genes that were significantly different at p < .01 after correction were considered for further analysis.

We considered genes to be biologically, significantly differentially expressed if they were up- or downregulated greater than twofold (+2/−2) and had an FDR of less than 0.01. The pathways which include those upregulated or downregulated genes due to erlotinib treatment were determined by Ingenuity Pathways Analysis (QIAGEN, Redwood City, CA) and String software (© STRING CONSORTIUM 2018, version 10.5), a database of known and predicted protein-protein interactions (Szklarczyk et al., 2015). Raw and processed microarray data were deposited into the Gene-Expression Omnibus (GEO) database (GSE106151).

Real-Time Quantitative Polymerase Chain Reaction (qPCR) Verification of Identified Target Genes

Differentially expressed target genes identified from the microarray analysis were validated using qPCR in a new set of experiments. Human dermal fibroblasts (ATCC® PCS-201-012™) were grown in cell culture as outlined above, treated with erlotinib 1μM, vehicle (DMSO), or no treatment daily for three days. Total RNA was extracted on the fourth day as described above and was reverse transcribed using Superscript® III reverse transcriptase and oligo(dT) primers (Invitrogen™ Waltham, MA). Forty cycles of qPCR were performed using the LightCycler® 480 SYBR Green I Master Mix (Roche Life Science, Indianapolis, IN). The relative abundance of each sequence was computed using Roche LightCycler® 480 Relative Quantification software (Roche Life Science, Indianapolis, IN). The primer sequences (Integrated DNA Technologies, Inc., Coralville, IA) used to amplify each gene are listed in the supplemental tables. The β-actin gene was used as the reference gene.

Western Blot Analysis

We performed Western blot analysis to determine whether changes seen in each of the differentially expressed target genes resulted in corresponding changes to the protein. In a new experiment, human dermal fibroblasts (ATCC® PCS-201-012™) were grown in cell culture as outlined above and treated with erlotinib 1μM, vehicle (DMSO), or no treatment daily for three days. After 72 hours, cells were homogenized with boiling laemmli sample buffer, boiled at 100°C, fractionated on 4–12% NuPAGE bis-tris gels (Invitrogen), and transferred to a nitrocellulose membrane. Membranes were placed in nonfat dried milk to reduce nonspecific antibody binding and then incubated overnight at 4°C with a primary antibody to the protein of interest followed by incubation with a horseradish peroxidase-conjugated secondary antibody (supplemental tables). The membranes were visualized with chemiluminescence (ThermoFisher Scientific, Waltham, MA). To standardize samples for protein loading quantities, after processing, the blots were stripped and reprobed with a primary antibody to β-actin followed by incubation with a horseradish peroxidase-conjugated secondary antibody (supplemental tables). The membranes were then visualized with chemiluminescence (ThermoFisher Scientific). Blots were quantified by scanning JPEG files into ImageJ imaging software (National Institutes of Health, Bethesda, MD) to determine the optical density of each band.

Results

Cultured Cells (ATCC® PCS-201-012™) were Fibroblasts and Treatment with Erlotinib Did Not Affect Cell Viability

To confirm that the cells in vitro were predominantly fibroblasts, we performed immunostaining according to a published procedure (TE-7 1:100; EMD Millipore, Billerica, MA) (Goodpaster et al., 2008). We found no statistical difference among the treatment groups (F_2,6) = .088; p = 0.917) (Figure 1A, 1B). To confirm that treatment with erlotinib or vehicle did not affect the viability of fibroblasts, we performed MTT assays according to a published protocol (van Meerloo et al., 2011). There was a statistical difference (F_6,14 = 9.023; P<0.0004) between no treatment and the staurosporine (positive control for cell death; *p < 0.0001) group and no treatment and the erlotinib 10μM (**p = 0.005) group (Figure 1C). No other statistically significant differences were found among the groups.

Figure 1.

Confirmation that cultured cells (ATCC® PCS-201-012™) were fibroblasts and that treatment with the tyrosine kinase inhibitor erlotinib did not impinge upon cell viability. (A) Representative fluorescent micrographs show expression of indicated markers in fibroblasts. Cells were thawed, resuspended, and grown to 80% confluence per procedure, and then transferred to 25 mm coverslips (3 coverslips per condition). Cells were fixed and then immunolabeled with the primary antibody, anti-TE-7 (1:100). The secondary antibody was FITC conjugated donkey anti-mouse IgG (1:100). White arrows indicate cell bodies labeled with DAPI and anti-TE-7. Each experiment was performed in triplicate. (B) There was no statistical difference among the treatment groups (F_2,6 = 0.088; p = 0.917). Treatment with the tyrosine kinase inhibitor erlotinib did not impinge upon cell viability (C). Human fibroblasts (ATCC® PCS-201-012™, Manassas, VA) were cultured in flat-bottomed 96-well plates according to standard protocol. Fibroblasts were treated with vehicle, staurosporine, and erlotinib (1–10μM) every day for three days. After three days, the effects of treatment on cell viability were assessed using the Vybrant® MTT Cell Proliferation Kit (V-13154, Invitrogen, Carlsbad, CA) according to manufacturer’s instructions. Absorbance of the solution was measured at 540 nM(1sec) to quantify the amount of formazan produced by the reaction. Cell viability was computed as mean percentage of control with the no treatment group designated as the reference group. Staurosporine served as a positive control for cell death. The mean of three individual experiments is represented. There was a statistical difference (F_6,14 = 9.023; p < 0.0004) between no treatment and the staurosporine (*p < 0.0001) group and the no treatment and erlotinib 10μM (**p = 0.005) group.

Differential Gene Expression Found in Fibroblasts Treated with Erlotinib

To gain mechanistic insight into the variability of development and severity of EGFR TKI-related rash, we conducted microarray analysis to examine the transcriptome in fibroblasts from a human cell line treated with erlotinib versus no treatment or versus vehicle. Initial one-way ANOVA identified 1060 differentially expressed genes (DEGs) at FDR < 0.01 and FC > |2|. When comparing the vehicle versus no treatment groups, we noted only 14 unique DEGs between the two groups; therefore, we combined the no treatment and vehicle groups into one combined control group and performed one-way ANOVA for the following comparisons: (a) erlotinib 1μM versus erlotinib 100nM; (b) erlotinib 1μM versus combined control; and (c) erlotinib 100nM versus combined control. Using parameters of FDR < 0.01 and FC > |2|, 361 DEGs were differentially expressed in the 1μM versus combined control group, versus 13 DEGs in the 100nM versus combined control group. Since these 13 DEGs were also found in the 1μM versus combined control group, we chose to focus our analysis on the erlotinib 1μM (also the more clinically relevant dose) versus combined control group.

Target genes were identified from the microarray analysis based on their possible role of rash, wound healing, fibroblast activation, or cancer progression. Although the following DEGS did not meet the FC and FDR cutoff values established a priori, we chose to further evaluate aquaporin 3 (Gill blood group) (AQP3; downregulated, FC = −1.3) for its potential role in wound healing, CCL2 (downregulated, FC = −1.9) for its potential role in rash development, and palladin, cytoskeletal associated protein (PALLD; upregulated, FC = 1.6) for its possible role in fibroblast activation. AQP3 was of particular interest because it is expressed in cultured human skin fibroblasts, and EGF has been shown to induce fibroblast migration, which in turn induces expression of AQP3 in a time- and dose-dependent manner (Cao et al., 2006) (supplemental figures).

Because AQP3, CCL2, and PALLD were not abundantly expressed in our sample, we also examined the top up- and downregulated DEGs in our dataset to further examine those that were involved in cancer and cancer progression pathways (Table 1). We identified several genes implicated in tumorigenesis, such as growth differentiation factor 15 (GDF-15), NSCLC (solute carrier family 7 [anionic amino acid transporter light chain] member 7, SLC7A11; matrix metallopeptidase 12, MMP12), and breast and ovarian cancers (DIRAS family, GTP-binding RAS-like 3, DIRAS3).

Table 1.

Top 20 genes differentially regulated in fibroblasts treated with 1μM erlotinib versus control.

Sequential Number	Affymetrix Probe set ID	Entrez ID	Gene symbol	Gene name	F Statistic	Fold Change	FDR	Summary
1	221577_x_at	9518	GDF-15	Growth differentiation factor 15	106.512338779809	4.65	2.96E-08	Location: 19p13.1. A.K.A. NAG-1, MIC-1 Protein-coding gene. Member of TGF-beta family. GDF-15 is activated by p53 and induced by some antitumorigenic and apoptosis-inducing compounds. Over expression of GDF-15 has been linked to apoptosis, and reduced expression of GDF-15 may increase tumorigenesis (Kadara, Schroeder, Lotan, Pisano, & Lotan, 2006; Yu et al., 2012).
2	217678_at	23657	SLC7A11	Solute carrier family 7 (anionic amino acid transporter light chain, xc- system), member 11	53.573309199192	4.45	1.33E-06	Location: 4q28–32. A.K.A. xCT, CCBR1. Protein-coding gene. Related super pathways include amino acid transport across the plasma membrane and transmembrane transport of small molecules. The Xc− system plays a critical role in growth and progression of cancer and glutathione-based drug resistance. (Baek et al., 2012) xCT (SLC7A11) is a subunit of the Xc− system. In an exploration of nine lung cancer cell lines, (Huang, Dai, Barbacioru, & Sadée, 2005) five showed high expression of SLC7A11, including A549, HOP-62, NCI-H226, NCI-H322M, and NCI-H460. Of these, HOP-62 showed the highest SLC7A11 expression. Xie, An, Jiang, & Wang (2012) investigated the use of the Xc− system as a biomarker for resistance to radiation in NSCLC and found that irradiation-resistant tumors expressed genes related to the human solute carrier family. Another study combined treatment with inhibitors of xCT-dependent cystine transport and EGFR and found a synergistic reduction of EGFR-expressing HNSCC tumor growth (Yoshikawa et al., 2013).
3	209921_at	23657	SLC7A11	Solute carrier family 7 (anionic amino acid transporter light chain, xc- system), member 11	49.0796818347496	4.13	2.24E-06	Per #2.
4	212314_at	23231	SEL1L3	Sel-1 suppressor of lin-12-like 3	99.8413798369012	3.92	4.13E-08	Location: 4p15.2 Protein-coding gene. A.K.A. KIAA0746 SEL1L3 has been explored in the context of genotyping of persons with low-dependence versus high-dependence in smoking cessation (Rose, Behm, Drgon, Johnson, & Uhl, 2010).
5	217966_at	116946	FAM129A	Family with sequence similarity 129, member A	117.468294453297	3.88	2.15E-08	Location: 1q25.3 A.K.A. NIBAN, cell growth-inhibiting gene 39 protein Protein-coding gene. Regulates phosphorylation of many proteins involved in translation regulation. Marker of renal cancer in rats and humans and is upregulated in different types of thyroid cancer (Matsumoto et al., 2006). FAM129B (A.K.A. MINERVA) is upregulated in many types of cancers such as breast, kidney, and lung. In one study, EGFR phosphorylated FAM129B at Y593, which promoted the interaction between FAM129B and Ras (Ji et al., 2016).
6	218145_at	57761	TRIB3	Tribbles pseudokinase 3	80.6798758411509	3.67	1.36E-07	Location: 20p13-p12.2. A.K.A. TRB3, TRB-3, NIPK Protein-coding gene. The protein encoded by the TRIB3 gene is induced by NFκB. TRIB3 negatively regulates NFκB and AKT1 and interacts with AKT1, AKT2, ATF4, MAP2K1 and MAP2K7. TRIB3 has been studied in diabetes (i.e., it disrupts insulin signaling by binding directly to AKT kinases and blocking their activation (Ding et al., 2014). In one study (Lorente et al., 2009) amphiregulin (an EGFR ligand) expression was associated with increased extracellular signal-regulated kinase activation, which mediated resistance to THC through blunting expression of p8 and TRB3 (two genes that are involved in cannabinoid-induced apoptosis of glioma cells).
7	239370_at	100505633	LINC01133	Long intergenic non-protein coding RNA 1133	32.65857495	3.61	2.65E-05	Location: 1q23.2 RNA gene. No data available.
8	228575_at	53833	IL20RB	Interleukin 20 receptor beta	58.8633421204388	3.58	7.56E-07	Location: 3q22.3. Protein-coding gene. Super-related pathways include the STAT3 Pathway. Chada et al. (2005) examined the potential of melanoma differentiation-associated gene-7 (MDA-7) as a target for pancreatic cancer. MDA-7 (IL24) has two receptors, 1IL20R and 2IL20R. They found that exposure to MDA-7 induces apoptosis in pancreatic cancer cells. One study (Gupta et al., 2008) found that combining either GST-MDA-7 or GST-M4 (0.1uM) and erlotinib (10uM) (suboptimal apoptosis-inducing concentrations) synergistically enhanced growth inhibition and apoptosis induction over that observed with either agent alone. Combination treatment also augmented inhibition of EGFR signaling, analyzed by phosphorylation of EGFR and its downstream effectors AKT and ERK1/2, over that with single-agent therapy.
9	223195_s_at	83667	SESN2	Sestrin 2	81.4607362289605	3.48	1.29E-07	Location 1p35.3. A.K.A Hi95, hypoxia-induced gene 95, SES2 Protein-coding gene. SESN2 encodes a member of the sestrin family of PA26-related proteins. The encoded protein may function in the regulation of cell growth and survival and may be involved in cellular response to different stress conditions. Related super pathways include cell cycle/checkpoint control, DNA damage response, and p53. Yang et al. (2013) performed bioinformatic analyses for radiation-induced transcriptome alteration. They selected seven significant radiation-altered genes (SESN2, FN1, TRAF4, CDKN1A, COX-2, DDB2 and FDXR), and then compared radiation effects in two types of NSCLC cells with different radiosensitivity. They determined that COX-2 is a putative biomarker for radioresistance in NSCLC cells. Another group (Lee et al., 2012) investigated SESN2 in p53 wild type A549 and H460 cells. SESN2 was overexpressed in most lung cancer patients and was a negative prognostic factor for patient survival.
10	219270_at	79094	CHAC1	ChaC glutathione-specific gamma-glutamylcy-clotransferase 1glutamylcyclc	110.109570226206	3.37	2.50E-08	Location: 15q15.1 A.K.A. Cation transport regulator-like protein; Gamma-GCT Acting on Glutathione Homolog, Botch, Gamma-GCG Protein-coding gene. Related pathways include glutathione metabolism, unfolded protein response, notch signaling pathway, glutathione synthesis, neurogenesis biological processes. CHAC1 catalyzes the cleavage of glutathione. Depletion of glutathione is a critical part of apoptosis. Higher CHAC1 mRNA expression levels have been shown to positively correlate with poor tumor differentiation in breast and ovarian cancers (Goebel et al., 2012).
11	203967_at	990	CDC6	Cell division cycle 6 homolog	54.9159400040564	−3.75	1.16E-06	Location: 17q21.3 Protein encoding gene. The CDC6 protein functions as a regulator at the early steps of DNA replication. CDC6 localizes in the cell nucleus during cell cycle G1, but translocates to the cytoplasm at the start of S phase. The subcellular translocation of CDC6 during the cell cycle is regulated through phosphorylation by cyclin dependent kinases. CDC6 may be implicated in the squamous subtype of NSCLC (Qian, Luo, Zhao, & Huang, 2014). One research group (Allera-Moreau et al., 2012) investigated gene expression and overall survival of patients with early or midstage NSCLC and found that expression of POLQ, CDC6, PLK1, RAD51, and CLASPIN was associated with overall, relapse-free, and disease-free survival, independent of treatment and stage of disease.
12	223381_at	83540	NUF2	NUF2, NDC80 kinetochore complex component	45.6508322108842	−3.77	3.47E-06	Location: 1q23.3 A.K.A. NUF2, NDC80 Kinetochore Complex Component; Cell Division Cycle-Associated Protein 1; CDCA1 Protein coding gene; associated with centromeres of mitotic HeLa cells, which suggests that this protein is a functional homolog of yeast Nuf2. Elevated NUF2 gene expression has been found in a range of tumor types and cell lines and is associated with poor outcome in cancer patients. Silencing of NUF2 inhibits tumor growth and leads to apoptosis in cancer cell lines (Kamburov et al., 2015).
13	220651_s_at	55388	MCM10	Minichromo-some maintenance 10 replication initiation factor	84.24540582	−3.81	1.07E-07	Location: 10p13 Protein encoding gene. The hexameric protein complex formed by MCM proteins is a key component of the pre-replication complex and it may be involved in the formation of replication forks and in the recruitment of other DNA replication-related proteins. MCM10 is regulated by proteolysis and phosphorylation in a cell cycle-dependent manner.
14	209773_s_at	6241	RRM2	ribonucleotide reductase regulatory subunit M2	72.2037544	−3.82	2.60E-07	Location: 2p25.1 Protein-coding gene. Hydroxyurea; gemcitabine Mammalian ribonucleotide reductase is composed of two different dimeric protein components often called Rl and R2, which are necessary for reduction of ribonucleoside diphosphates to their corresponding deoxyribonucleotides, a rate-limiting process in DNA synthesis (Fan et al., 1998).
15	203968_at	990	CDC6	Cell division cycle 6 homolog	89.82962444	−3.90	7.63E-08	Per #11.
16	201890_at	6241	RRM2	ribonucleotide reductase regulatory subunit M2	107.5843363	−3.96	2.92E-08	Per #14.
17	217979_at	27075	TSPAN13	tetraspanin 13	151.0494677	−4.71	4.68E-09	Location: 7p21.1 A.K.A. Transmembrane 4 Superfamily Member 13; Tspan-13 Cell surface protein with four hydrophobic domains; plays a role in regulation of cell development, activation, growth, and motility. TSPAN13 is over-expressed in a number of cancers, including prostate cancer (Yan et al., 2015). TSPAN13 is also being pursued as a potential biomarker for prostate cancer progression.
18	205034_at	9134	CCNE2	cyclin E2	100.9603775	−5.28	3.88E-08	Location: A.K.A. CYCE2 Cyclins are regulators of CDK kinases; CCNE2 plays an important role in G1/S transition and is overexpressed in a number of cancers including breast, gastric, prostate cancers.
19	204580_at	4321	MMP12	Matrix metallopeptidase 12	69.7882905	−5.47	3.00E-07	Location: 11q22.3 A.K.A. HME MMP12 is a protein-coding gene involved in the breakdown of extracellular matrix in normal physiological processes, such as embryonic development, reproduction, and tissue remodeling, as well as in disease processes, such as arthritis and metastasis. Most MMPs are secreted as inactive pro-proteins, which are activated when cleaved by extracellular proteinases. It is thought that the protein encoded by this gene is cleaved at both ends to yield the active enzyme, but this processing has not been fully described. The enzyme degrades soluble and insoluble elastin. It may play a role in aneurysm formation and studies in mice suggest a role in the development of emphysema. MMP-7 (matrilysin) levels were tested in patients with NSCLC (n = 114) and healthy controls (n = 100) (Sun et al., 2012). Protein levels were assessed by ELISA; plasma protein levels of MMP-7 in lung cancer patients were significantly higher than those found in healthy control subjects (p < 0.001). In another study, (Qian et al., 2014) CypA increased NSCLC cell invasion by regulating the activity of secreted matrix metallopeptidase 9 (MMP9). Suppression of CypA with 239836 CypA inhibitor decreased cell proliferation and MMP9 activity. The variant genotype −1562 T/T in the MMP9 gene has been associated with decreased risk of lung cancer (González-Arriaga et al., 2012). The MMP2 −735 T/T genotype has been associated with a lower survival rate in lung cancer than those with C/T or C/C (González-Arriaga et al., 2012).
20	215506_s_at	9077	DIRAS3	DIRAS family, GTP-binding RAS-like 3	43.97232471	−12.62	4.39E-06	Location: 1p31. Protein-coding gene. DIRAS3 is a member of the ras superfamily and is expressed in normal ovarian and breast epithelial cells but not in ovarian and breast cancers. DIRAS3 is an imprinted gene, with monoallelic expression of the paternal allele, which is associated with growth suppression.

Note. NSCLC = non-small cell lung cancer; EGFR = epidermal growth factor receptor; FDR = false discovery rate

All gene locations, also known as (A.K.A.) abbreviations, and summaries are from Gene cards: http://www.genecards.org/. Other references are delineated in the references section.

Next, we conducted pathway analysis of the erlotinib 1μM versus combined control group using Ingenuity Pathway Analysis to understand the biological context of how erlotinib may alter the transcriptional activity of human adult dermal fibroblasts. Canonical pathways that were significantly affected in the erlotinib 1μM versus combined control analysis are displayed in Figure 2. While none of the most highly enriched pathways were related to rash development, the cell division cycle 6 homolog (CDC6) gene, a regulator at the early steps of DNA replication (FC = −3.75), was downregulated in the cell cycle control of chromosomal replication canonical pathway. We chose to further investigate CDC6 because it has been associated with overall, relapse-free, and disease-free survival, independent of treatment and stage of disease. Finally, we used String software to query a curated database for potential protein-protein interactions and found possible predicted co-expression between CCL2 and TIMP1 (Tissue Inhibitor of Metalloproteinase-1, acts on MMP12) and CCL2 and GDF-15 (Supplemental tables).

Figure 2.

Unbiased pathway analysis demonstrating the top 10 differentially enriched canonical signaling pathways is presented. Green indicates downregulation and red indicates upregulation. The orange line shows the −log10 p for each pathway, e.g., the ratio of the number of genes in each pathway that met the cut off value/the total number of genes in the pathway. The numbers to the right represent the total number of genes in each pathway.

Gene Expression Validation with qPCR Validated Selected Genes

Of the eight DEGs we selected for validation (supplemental tables), we were able to examine five by qPCR: SLC7A11, MMP12, DIRAS3, CDC6, and CCL2 (Figure 3). Three genes had expression changes in the same direction as the microarray results, but at different magnitudes of fold change; however, two (MMP12, CCL2) showed gene expression changes in the opposite direction from the microarray.

Figure 3.

Validation of differentially expressed genes from microarray analysis using qPCR. Fresh fibroblasts were cultured and underwent experimental procedures as described. Results are reported as linear fold changes in erlotinib 1uM as compared to collapsed control; n = 3 per group for each microarray and qPCR. qPCR = quantitative real-time polymerase chain reaction. Solute carrier family 7 [anionic amino acid transporter light chain] member 7, = SLC7A11; Matrix metallopeptidase 12 = MMP12; chemokine (C-C motif) ligand 2 = CCL2; cell division cycle 6 homolog = CDC6; DIRAS3 = DIRAS family, GTP-binding RAS-like 3

Western Blot Analysis to Quantify Protein Expression Further Validated DEGs

To confirm that levels of mRNA expression for each gene corresponded to similar protein levels, we performed Western blot analyses of protein lysate from fresh fibroblasts (Figure 4). Protein levels for MMP12 and SCL7A11 were upregulated 240% and 262%, respectively, in the erlotinib 1μM condition compared to control (Figure 4). The protein levels of CCL2 and CDC6 were downregulated 41% and 27%, respectively, in the erlotinib 1μM condition compared to control. β-actin was the reference protein.

Figure 4.

Validation of gene expression analyses using Western blot analysis. Analysis of protein lysate from fibroblasts in the combined control group (no treatment, vehicle) and the erlotinib 1μM identified target protein level for MMP12, CCL2, CDC6, and SLC7A11. The protein level of MMP12 and SCL7A11 was upregulated 240%, and 262%, respectively, in the erlotinib 1μM condition compared to control. The protein level of CCL2 and CDC6 was downregulated 41% and 27%, respectively, in the in the erlotinib 1μM condition compared to control. Beta actin was used as the reference product. Solute carrier family 7 [anionic amino acid transporter light chain] member 7, = SLC7A11; Matrix metallopeptidase 12 = MMP12; chemokine (C-C motif) ligand 2 = CCL2; cell division cycle 6 homolog = CDC6.

Discussion

Therapies targeting EGFR have been shown to be clinically effective for patients with tumors that overexpress EGFR (e.g., NSCLC), but EGFR TKI therapy often results in a painful papulopustular rash that may limit chronic treatment. We performed this microarray study to further understand how the transcriptome changes in a fibroblast cell line after introduction of erlotinib treatment and to identify potential genes that were mechanistically linked to EGFR TKI-related rash for further study. Of the eight genes we selected for validation, we were able to confirm four by both Western blot analysis and qPCR (MMP12, CCL2, CDC6, and SLC7A11).

The advantage of using microarray technology in this experiment was the opportunity to visualize simultaneous changes in expression of almost all genes in the human genome in dermal fibroblasts treated with erlotinib versus combined control (vehicle or no treatment) (Ness, 2006). However, microarray lacks specificity in that only the genes on the array may be examined. As such, there may be different isoforms of our genes of interest that were not detected on the array. In addition, background levels of hybridization (genes that hybridize to a probe regardless of expression level) also limit accuracy of microarray analysis, especially for genes in low abundance (Marioni, Mason, Mane, Stephens, & Gilad, 2008). Given this, it is not surprising we were unable to validate AQP3 and PALLD; still, our results provide a basis for further study of these genes in patients with cancer who experience an EGFR TKI-induced rash.

We selected several genes for further study because of their potential roles in wound healing, rash development, or fibroblast activation, but these genes were not highly expressed in our dataset. CCL2 has been shown to be robustly upregulated in human skin and across different mouse models (Lichtenberger et al., 2013; Mascia et al., 2013) suggesting an anti-inflammatory effect of EGFR TKIs in skin. In our microarray study, treatment with erlotinib 1μM on a fibroblast cell line resulted in downregulation of CCL2 but upregulated when confirmed with qPCR in a new experiment, which is consistent with the described studies. Given immunoreactivity is found in all skin cell types, but more prominently in keratinocytes (Nanney et al., 1984), it is plausible that CCL2 may be less abundant in the dermal fibroblasts. When examining the potential associations of the proteins, String analysis showed some association between CCL2 and CDC6, which functions as a regulator of the early steps of DNA replication. CDC6 has been implicated in the squamous subtype of NSCLC (Qian et al., 2014) and associated with overall, relapse-free, and disease-free survival, independent of treatment and stage of disease for patients with NSCLC (Allera-Moreau et al., 2012). The relationship of the genes and their interaction regarding EGFR TKI-induced rash remains an area of exploration.

We chose to validate two genes, MMP12 and SLC7A11 because of their roles in cancer and cancer progression. MMP12 is a protein-coding gene involved in the breakdown of extracellular matrix in normal physiological processes, such as tissue remodeling, and has been shown to have an important role in tumor invasion and metastasis. Although we found no published studies investigating MMP12 and EGFR-TKI rash specifically, MMP12 has been examined in patients with dermatitis hepetiformis (DH) and was expressed by subepithelial macrophages of spontaneous and induced DH rash (Salmela, Pender, Reunala, MacDonald, & Saarialho-Kere, 2001), and may be a target for future exploration for EGFR TKI-induced rash.

Our results should be interpreted considering several limitations. First, we used a single human fibroblast cell line treated with one EGFR TKI, and then assessed for changes in gene expression; therefore, the cells in our experiments should have behaved similarly and at the same rate of change. We are not able to generalize our results to other cell lines (e.g., mutant EGFR cell lines, lung cancer cell lines such as HOP-62) or to primary cells derived from human skin biopsies that contain a mixture of dermal cell types with cells at different rates of change. Still, our findings provide a basis for further evaluation of EGFR TKI-induced rash in patients with and without cancer, and with and without a rash, to understand how inhibition of EGFR changes the transcriptome in fibroblasts cultured from skin biopsies, considering confounding variables such as prior cancer treatment and EGFR mutational status.

Given that upregulation of EGFR has been linked to cancer-related processes, such as inhibition of apoptosis and uncontrolled cell proliferation (Keese et al., 2005; Mendelsohn, 2004), assuming a direct pathway, inhibition of EGFR should result in downregulation of these processes. While our results are consistent with this hypothesis, treatment with erlotinib or vehicle in our study did not affect cell viability as measured by MTT. Because the amount of MTT is (a) directly proportional to the number of living, rather than apoptotic, cells during MTT exposure (Sylvester, 2011); (b) may vary among different cell types; and (c) may vary with duration of cell culture, fibroblasts in our experiments may have been in the process of apoptosis and therefore not represented by MTT.

Finally, first- (e.g., erlotinib) and second- (e.g., afatinib) generation EGFR TKIs target wild-type EGFR, but third-generation EGFR TKIs, such as osimertinib and rocelitinib, spare wild-type EGFR, resulting in less rash and better activity against cancer (Singh & Jadhav, 2018). However, other problematic side effects (e.g., hyperglycemia, QTc prolongation) have been reported with newer EGFR TKIs, and patients develop resistance to third generation EGFR TKIs due to a C797S mutation (Patel, Pawara, Ansari, & Surana, 2017). Future studies should compare changes in gene expression among different generations and types of EGFR TKIs.

Conclusion

We performed a microarray study to understand how the transcriptome changes in fibroblasts treated with erlotinib. We validated changes in the expression of four genes, CCL2, CDC6, SLC7A11, and MMP12, and they are candidates for further investigation because of their potential role in development and severity of EGFR TKI-related rash in dermal fibroblasts. If found predictive of rash in future studies using patient fibroblast samples, our findings may help to identify those at risk for severe rash so that (a) the dose of EGFR TKI therapy may be adjusted; (b) additional treatments for the rash can be developed; and/or (c) precise, patient-centered interventions can be developed so that patients with cancer can better self-manage their rash and adhere to EGFR TKI treatment.