Delineating the Effects of Passaging and Exposure in a Longitudinal Study of Arsenic-Induced Squamous Cell Carcinoma in a HaCaT Cell Line Model

Abstract

Cutaneous squamous cell carcinoma (cSCC) is a major deleterious health effect of chronic arsenic (iAs) exposure. The molecular mechanism of arsenic-induced cSCC remains poorly understood. We recently demonstrated that chronic iAs exposure leads to temporally regulated genome-wide changes in profiles of differentially expressed mRNAs and miRNAs at each stage of carcinogenesis (7, 19, and 28 weeks) employing a well-established passage-matched HaCaT cell line model of arsenic-induced cSCC. Here, we performed longitudinal differential expression analysis (miRNA and mRNA) between the different time points (7 vs 19 weeks and 19 vs 28 weeks) within unexposed and exposed groups, coupled to expression pairing and pathway analyses to differentiate the relative effects of long-term passaging and chronic iAs exposure. Data showed that 66–105 miRNA [p < .05; log₂(fold change) > I1I] and 2826–4079 mRNA [p < .001; log₂(fold change) > I1I] molecules were differentially expressed depending on the longitudinal comparison. Several mRNA molecules differentially expressed as a function of time, independent of iAs exposure were being targeted by miRNA molecules which were also differentially expressed in a time-dependent manner. Distinct pathways were predicted to be modulated as a function of time or iAs exposure. Some pathways were also modulated both by time and exposure. Thus, the HaCaT model can distinguish between the effects of passaging and chronic iAs exposure individually and corroborate our previously published data on effects of iAs exposure compared with unexposed passage matched HaCaT cells. In addition, this work provides a template for cell line-based longitudinal chronic exposure studies to follow for optimal efficacy.

Chronic arsenic (iAs) exposure primarily through contaminated drinking water puts >200 million individuals at risk globally (Naujokas et al., 2013; Podgorski and Berg, 2020) including >2 million in the United States (Ayotte et al., 2017; Baris et al., 2016; Frost et al., 2003). Although such chronic exposure results in multiorgan cancers, premalignant lesions, and noncancer adverse health outcomes (Argos et al., 2010; Banerjee, 2011; Flanagan et al., 2012; Ghosh et al., 2007a; Palma-Lara et al., 2020; States, 2015; Tseng, 2005, 2007), skin lesions are considered to be hallmarks of arsenicosis (Ghosh et al., 2007b; Hunt et al., 2014; Yu et al., 2006, 2018). Three different types of skin cancer lesions are caused by chronic arsenic exposure: Bowen’s disease, cutaneous basal cell carcinoma, and cutaneous squamous cell carcinoma (cSCC) (Banerjee, 2011; Martinez et al., 2011). Although it is known that the molecular etiology of arsenic-induced skin cancer is distinct from UV-induced skin cancers (Martinez et al., 2011), detailed understanding of the former is currently incomplete.

A major impediment in the exploration of molecular mechanisms involved in pathogenesis of arsenic-induced cSCC is the complete absence of an animal model. Irrespective of the iAs exposure level, rodents have never developed skin tumors unless treated with a cocarcinogen (Burns et al., 2004; Kitchin, 2001; Rossman et al., 2001; 2004; States et al., 2011; Tokar et al., 2010; Waalkes et al., 2004). Thus, most mechanistic studies are performed using cell lines. However, to emulate the prolonged time period and low level of exposure reflecting the blood level of chronically exposed human populations, a cell line was required that could be passaged for months without becoming senescent. To fulfill these requirements, HaCaT cell line model of arsenic exposure-induced cSCC was developed by Pi et al. (2008).

The HaCaT cell line was originally developed by Boukamp et al. by prolonged culturing of epidermal keratinocytes from adult human male (62 years old) back skin tissue obtained through surgical excision of the distant periphery of a melanoma (Boukamp et al., 1988). This cell line was maintained under low Ca²⁺ (0.2 mM) culture conditions at 38.5°C and immortalized spontaneously without the expression of SV40 T antigen, or the presence of HPV genomic sequences (Boukamp et al., 1988). HaCaT cells grew as a monolayer and expressed keratinocyte-specific markers as long as they were cultured under low Ca²⁺ conditions (Boukamp et al., 1988). These cells were aneuploid (hypotetrapolid) with most clones harboring 72–88 chromosomes (Boukamp et al., 1988). This range of chromosome numbers remained stable when the cell line was cultured for approximately 300 passages. Subcutaneous injection or transplantation into the muscle fascia of HaCaT cells at passage 100 in the interscapular region of 4–6 weeks old nude mice (BALB/c nu/nu backcrosses) did not give rise to tumors, demonstrating that this cell line is nontumorigenic (Boukamp et al., 1988). This nontumorigenic nature persisted even at 300 passages (Boukamp et al., 1997). HaCaT cell line has been demonstrated to be a suitable model for studying growth, differentiation, and carcinogenesis of human keratinocytes (Colombo et al., 2017).

When HaCaT cells are continuously exposed to 100 nM NaAsO₂ (iAs) for 28 weeks, they become fully transformed and acquire aggressive cSCC phenotype (Pi et al., 2008; Sun et al., 2009). This exposure level corresponds to the blood arsenic level found in chronically exposed human populations in Inner Mongolia and Mexico (Gonsebatt et al., 1994, 1997; Pi et al., 2000) and thus faithfully replicates toxicologically and environmentally relevant exposure. Bilateral injection of such iAs-transformed HaCaT cells into the renal capsules of male nude mice produced highly aggressive metastatic SCC that invaded the renal parenchyma, as opposed to passage-matched unexposed HaCaT cells (Pi et al., 2008). Injection of HaCaT cells transformed by chronic exposure to 100 nM iAs either in the right hind leg of female BALB/c nude mice or intradermally in athymic nude mice also resulted in increased tumor incidence (Alexander et al., 2019; Barajas-Olmos et al., 2019). Employing this well validated model, we recently demonstrated that chronic iAs exposure leads to epithelial-mesenchymal transition (EMT) through a temporally regulated change in the landscape of differentially expressed miRNA and their target mRNA molecules at each stage in the development of cSCC (Banerjee et al., 2021). At each stage (7, 19, and 28 weeks of exposure, corresponding to early transformation related changes, transformation initiation and fully transformed, respectively), these molecular signatures are largely unique and dysregulate pathways and networks with well characterized roles in carcinogenesis. For example, major dysregulated pathways at 7-week time point are involved in cell cycle regulation, whereas those at 19 and 28-week time points include Rho-GTPase signaling and ubiquitin proteasomal system, respectively (Banerjee et al., 2021), demonstrating carcinogenic transition with time of exposure in HaCaT cell line.

Although the HaCaT cells maintained its non-tumorigenic status even at passage numbers approximately 300 (Boukamp et al., 1997; Fusenig and Boukamp, 1998), it is well known that cell lines at high passage number exhibit alterations in a wide variety of characteristics, including morphology, response to stimuli, expression of transporters, and growth rates (ATCC, n.d.; Briske-Anderson et al., 1997; Chang-Liu and Woloschak, 1997; Kwist et al., 2016; Raffetto et al., 1999; Wu et al., 2017). Moreover, long-term passaging often leads to novel mutations, resulting in modified stress response and a consequent shift in selective pressure, alterations in growth rate and survival (Hughes et al., 2007; Wenger et al., 2004). Passaging can also result in differential expression of numerous genes in a wide variety of cell lines, belonging to key biological pathways involved in stress response, EMT, cell cycle regulation, and cell adhesion (Lin et al., 2008; O'Driscoll et al., 2006; Wang et al., 2004).

Given the well-established effect of passaging on gene expression, we adopted passage matching to ensure robust control for our longitudinal gene expression analysis leading to iAs exposure-induced cSCC (Banerjee et al., 2021). The data unequivocally demonstrated that there is considerable temporal variation in the expression pattern of multiple molecules (ATF4, BCL2, β-Catenin, CHOP, IRE1, NRF2, and ZO-1) irrespective of the iAs exposure status (Banerjee et al., 2021). The presence of RNA-seq data individually for each of the conditions (with and without iAs exposure at 7, 19, and 28 weeks) provided us with an unique opportunity to examine if our model could effectively differentiate between the effects of passaging and iAs exposure. Analysis of this extensive dataset would also enable us to further empirically address whether passage matched controls that we employed previously was effective (Banerjee et al., 2021). These questions highlight an existing knowledge gap in the literature that needs to be addressed to determine the robustness of these long-term exposure model systems and the conclusions deduced therefrom. Even more importantly, answers to these questions would be of broader implication as they would be applicable to any cell culture-based longitudinal analysis and could provide clarity for better experimental design. Consequently, in the present study, we performed a longitudinal differential gene expression analysis (miRNA and mRNA) coupled to pathway analyses to compare the relative effects of long-term passaging and chronic iAs exposure in a well-established model of arsenic-induced cSCC.

MATERIALS AND METHODS

 

 

Chemicals

Sodium arsenite (NaAsO₂; CAS 7784-0698) was obtained from Thermo Fisher Scientific Inc. (Waltham, Massachusetts), prepared as 1000X stocks in UltraPure DNase/RNase-Free Distilled Water (Thermo Fisher Scientific Inc.) and stored as single-thaw aliquots. The aliquots were thawed immediately before use. Fresh stock was prepared on a monthly basis. Minimal essential media—alpha modification (MEM), trypsin, ethylene diamine tetraacetic acid, and penicillin/streptomycin were obtained from Thermo Fisher Scientific Inc. Fetal bovine serum (characterized) was obtained from Hyclone (Logan, Utah). All other chemicals were obtained from Thermo Fisher Scientific Inc., unless specifically mentioned.

Cell culture

The HaCaT model of Pi et al. (2008) was adopted for the present study as described. HaCaT cells were the kind gift of Dr TaiHao Quan, University of Michigan. Briefly, independent quadruplicate cell cultures were maintained for each condition (4 with 0 nM and 4 with 100 nM NaAsO₂) continuously for 28 weeks. Briefly, the cells were cultured in MEM supplemented with 10% fetal bovine serum, 100 units/ml penicillin/100 µg/ml streptomycin and 2 mM glutamine. Cultures were maintained at 37°C in a humidified 5% CO₂ atmosphere. Cells were passaged twice a week and 10⁶ cells were plated per 100 mm dish every time. Identity of cultures as HaCaT cells was confirmed by STR mapping (Genetica, Burlington, North Carolina). Detailed schema for study design is provided in Figure  1 and Supplementary Table 1.

Figure 1.

Study design depicting all the differential miRNA and differential mRNA expression comparisons performed. The top part (yellow circles) represents the unexposed HaCaT cells at 3 time points, whereas the bottom part (orange circles) represent the iAs exposed (100 nM) HaCaT cells at the same 3 time points. Vertical double-ended arrows represent the passage-matched differential miRNA and mRNA comparisons (DEG1-3), whereas the horizontal double-ended arrows represent the longitudinal comparisons within the unexposed (DEG4/DEG6) and exposed groups (DEG5/DEG7), respectively. For details regarding each comparison, please see Supplementary Table 1.

Total RNA isolation

Total RNA isolation from cells harvested at 7, 19, and 28-week time points followed by quality check was performed employing mirVana RNA isolation kit and Agilent RNA 6000 Pico kit, Eukaryote, version 2.6 respectively, following the manufacturer’s recommendations (Al-Eryani et al., 2018; Banerjee et al., 2021). All samples used had RIN (RNA integrity number) > 9.

RNA-seq, data mapping, and analysis

The work described in the current article is based on the same RNA-seq experiments that our group published recently (Banerjee et al., 2021). Library preparation, cluster generation, RNA-seq analysis, data mapping, and analyses (both for miRNA and mRNA) were as described. Data were deposited in the GEO database, accession numbers GSE153057 (miRNA) and GSE107054 (mRNA). Differentially expressed miRNA was defined as p < .05 and log₂(fold change) > I1I, whereas differentially expressed mRNA was defined as p < .001 and log₂(fold change) > I1I, respectively for this study.

Expression pairing and pathway analyses

Differentially expressed miRNA and mRNA molecules (DEG4-DEG7; Figure  1) were analyzed for expression pairing (miRNA-mRNA pairing) and prediction of dysregulated canonical pathways by ingenuity pathway analysis (Kramer et al., 2014). Pathways with −log(p value) > 1.3 and Z-score > 1 were defined as activated, whereas those with log(p value) > 1.3 and Z-score < −1 were defined as inhibited (Banerjee et al., 2021). Figures representing pathway interactions and node maps were generated employing ingenuity pathway analysis (Kramer et al., 2014).

Heat maps and Venn diagrams

Heat maps and bar graphs were generated using GraphPad Prism 9.0.1 (GraphPad Software, San Diego, California). Venn diagrams were generated employing Venn Diagram Plotter program (Pacific Northwest National Laboratory, https://omics.pnl.gov/software/venn-diagram-plotter).

RESULTS

 

Differential miRNA and mRNA Expression Profile With Passaging

Study design including details of all the different differential gene expression analyses performed is shown in Figure  1 and Supplementary Table 1. Details regarding the differentially expressed miRNA and mRNA molecules between passage-matched iAs exposed and unexposed HaCaT cells at each time point (7, 19, and 28 weeks; DEG1-DEG3) have been published (Banerjee et al., 2021). Figure  2 and Supplementary Tables 2 and 3 provide details on the profile of differentially expressed miRNA molecules for each comparison in the present study (DEG4-DEG7). Depending on the comparison, 66–105 miRNA molecules were found to be differentially expressed (Figure  2A). Between 7 and 19 weeks, more miRNA molecules were suppressed than induced, whereas the trend was reversed between 19 and 28 weeks (Figure  1A). Of the 163 and 121 differentially expressed miRNA molecules between 7 and 19 weeks (DEG4 + DEG5) and 19 and 28 weeks (DEG6 + DEG7), respectively (irrespective of the exposure status), 42 and 27 miRNA molecules were common in both unexposed and exposed groups in each those longitudinal comparisons, respectively (Figs.  2B–E and Supplementary Tables 2 and 3) and represented the effect of passaging irrespective of iAs exposure. The differentially expressed miRNA molecules that were uniquely present in either unexposed or exposed groups at any longitudinal comparison (Figure  1D and Supplementary Table 2) were considered to represent the effect of iAs exposure. Differentially expressed miRNAs exclusive to DEG5 and DEG7 are considered to be direct effects of iAs exposure, whereas those exclusive to DEG4 and DEG6 are considered to be effects of passaging that are prevented by iAs exposure.

Figure 2.

Differential expression of miRNA in longitudinal comparisons. A, Bar graph showing the number of miRNA molecules differentially expressed for each longitudinal comparison. miRNA molecules differentially expressed between 7 and 19 weeks are represented by DEG4 (unexposed) and DEG5 (exposed), and those between 19 and 28 weeks are represented by DEG6 (unexpoed) and DEG7 (exposed), respectively. Open bars represent induced miRNA molecules and closed bars represent suppressed miRNA molecules for each comparison. Differential miRNA expression is defined as p < .05; log₂(fold change) > I1I. B, Venn diagram depicting the distribution of differentially expressed miRNAs [p < .05; log₂(fold change) > I1I] between 7and 19 weeks in unexposed (DEG4) and exposed (DEG5) HaCaT cells, along with the number of overlaps. C, Venn diagram depicting the distribution of differentially expressed miRNAs [p < .05; log₂(fold change) > I1I] between 19 and 28 weeks in unexposed (DEG6) and exposed (DEG7) HaCaT cells, along with the number of overlaps. D, Heat map of differentially expressed miRNA molecules between 7and 19 weeks in unexposed (DEG4) and exposed (DEG5) HaCaT cells. E, Heat map of differentially expressed miRNA molecules between 19 and 28 weeks in unexposed (DEG6) and exposed (DEG7) HaCaT cells. The numerals on the y axis in (D) and (E) refer to the serial numbers assigned to the differentially expressed miRNA molecules in Supplementary Tables 2 and 3, respectively. The color code bar on top refers to the log₂(fold change) expression values. Absence of a bar (represented by white) signifies either that miRNA molecule was not detected or was not differentially expressed for that longitudinal comparison.

Data depicting the profile of differentially expressed mRNA molecules for each comparison in the present study (DEG4-DEG7) are presented in Figure  3 and Supplementary Tables 4 and 5. The number of differentially expressed mRNA molecules for different comparisons ranged between 2826 and 4079 (Figure  3A). Similar to differentially expressed miRNA, more mRNA molecules were suppressed than induced between 7 and 19 weeks, whereas the trend was reversed between 19 and 28 weeks (Figure  3A). Between 7 and 19 weeks, 1846 mRNA molecules were found to be differentially expressed irrespective of iAs exposure (DEG4 and DEG5), whereas the same was true for 1910 mRNA molecules between 19 and 28 weeks (DEG6 and DEG7), as shown in Figure  3B–E and Supplementary Tables 4 and 5. Such differentially expressed mRNA molecules common in exposed and unexposed groups at one longitudinal comparison (1846 at 7 vs 19 weeks; 1910 at 19 vs 28 weeks) were defined to represent the effect of passaging irrespective of iAs exposure, whereas the differentially expressed mRNA molecules that were uniquely present in either unexposed or exposed group at any longitudinal comparison were considered to represent the effect of iAs exposure (direct effect or prevention of passaging related changes) as explained in the previous paragraph.

Figure 3.

Differential expression of mRNA in longitudinal comparisons. A, Bar graph showing the number of mRNA molecules differentially expressed for each longitudinal comparison. mRNA molecules differentially expressed between 7and 19 weeks are represented by DEG4 (unexposed) and DEG5 (exposed), and those between 19 and 28 weeks are represented by DEG6 (unexposed) and DEG7 (exposed), respectively. Open bars represent induced mRNA molecules and closed bars represent suppressed mRNA molecules for each comparison. Differential mRNA expression is defined as p < .001; log₂(fold change) > I1I. B, Venn diagram depicting the distribution of differentially expressed mRNAs [p < .001; log₂(fold change) > I1I] between 7and 19 weeks in unexposed (DEG4) and exposed (DEG5) HaCaT cells, along with the number of overlaps. C, Venn diagram depicting the distribution of differentially expressed mRNAs [p < 0.001; log₂(fold change) > I1I] between 19 and 28 weeks in unexposed (DEG6) and exposed (DEG7) HaCaT cells, along with the number of overlaps. D, Heat map of differentially expressed mRNA molecules between 7 and 19 weeks in unexposed (DEG4) and exposed (DEG5) HaCaT cells. E, Heat map of differentially expressed mRNA molecules between 19 and 28 weeks in unexposed (DEG6) and exposed (DEG7) HaCaT cells. The numerals on the y axis in (D) and (E) refer to the serial numbers assigned to the differentially expressed miRNA molecules in Supplementary Tables 4 and 5, respectively. The color code bar on top refers to the log₂(fold change) expression values. Absence of a bar (represented by white) signifies either that miRNA molecule was not detected or was not differentially expressed for that longitudinal comparison.

Interaction Between miRNA-mRNA Differentially Expressed by Passaging

Next, we performed expression pairing to determine if the miRNA molecules differentially expressed due to passaging were targeting the mRNA molecules that also were differentially expressed due to passaging. The longitudinal comparison of these miRNA and mRNA molecules along with their induction/suppression status is presented in Figure  4. Of the 42 and 27 miRNA molecules that were differentially expressed in both unexposed and exposed HaCaT cells in 19 versus 7 weeks and 28 versus 19 weeks comparisons, respectively, 7 were found to be common (Figure  4A). Notably, all these miRNA molecules were differentially expressed in the same direction (induced or suppressed) in both unexposed and exposed cells at a given time point (42 and 27, respectively) (Figure  4B). For the mRNA molecules that were differentially expressed in both unexposed and exposed HaCaT cells in 7 versus 19 weeks and 19 versus 28 weeks comparisons respectively, 731 were found to be common (Figure  4C). Notably, for each comparison, differential expression of all but one mRNA molecule was in the same direction in passage-matched unexposed and exposed cells (Figure  4D).

Figure 4.

Interaction between miRNA-mRNA differentially expressed by passaging. A, Venn diagram depicting distribution of differentially expressed miRNAs [p < .05; log₂(fold change) > I1I] both in unexposed and exposed HaCaT cells between 7 and 19 weeks (both DEG4 and DEG5) and between 19 and 28 weeks (both DEG6 and DEG7) along with the number of overlaps. B, Bar graph showing the number of miRNA molecules induced and suppressed for each longitudinal comparison in both unexposed and exposed HaCaT cells. Open bars represent induced miRNA molecules, closed bars represent suppressed miRNA molecules, and gray bars represent miRNA molecules that are differentially expressed in opposite directions in unexposed and exposed cells for each comparison. C, Venn diagram depicting distribution of differentially expressed mRNAs [p < .001; log₂(fold change) > I1I] both in unexposed and exposed HaCaT cells between 7 and 19 weeks (both DEG4 and DEG5) and between 19 and 28 weeks (both DEG6 and DEG7) along with the number of overlaps. D, Bar graph showing the number of mRNA molecules induced and suppressed for each longitudinal comparison in both unexposed and exposed HaCaT cells. Open bars represent induced mRNA molecules, closed bars represent suppressed mRNA molecules, and gray bars represent mRNA molecules that are differentially expressed in opposite directions in unexposed and exposed cells for each comparison. E, Expression pairing between differentially expressed miRNA and differentially expressed mRNA between 7 and 19 weeks and between 19 and 28 weeks, respectively. For each longitudinal comparison, the total number of differentially expressed miRNA molecules that are targeting one or more differentially expressed mRNA molecules are shown as light gray bars; the number of differentially mRNA molecules that are targeted by one or more or differentially expressed miRNA molecules are shown as closed bars, total number of miRNA-mRNA pairings are shown as open bars, and the number of differentially expressed mRNA molecules that are not predicted to be targeted by any differentially expressed miRNA molecule are shown as hatched bars. F, Concordance-discordance relationship between differentially expressed miRNA and differentially expressed mRNA for the longitudinal comparisons. The data are presented as % of total interactions predicted at that time point.

Expression pairing demonstrated that in the 7 versus 19 weeks comparison (DEG4-DEG5), 24 differentially expressed miRNA molecules were predicted to target 613 differentially expressed mRNAs resulting in 972 predicted pairings (Figure  4E and Supplementary Table 6). In the case of the 19 versus 28 weeks comparison (DEG6-DEG7), 17 differentially expressed miRNA molecules were predicted to target 546 differentially expressed molecules resulting in 799 predicted pairings (Figure  4E and Supplementary Table 7). Overall, approximately 35% and 50% of all the predicted pairings in the 7 versus 19 weeks and 19 versus 28 weeks comparisons, respectively were found to be concordant (Figure  4F and Supplementary Tables 6 and 7).

We further investigated the top concordant miRNA-mRNA relationships for each longitudinal comparison (7 versus 19 weeks and 19 versus 28 weeks). The top 25 induced mRNA molecules for both DEG 4 and DEG5 that were found to be targeted by 9 different miRNAs that were suppressed in both DEG 4 and DEG5, whereas top 25 suppressed mRNA molecules for the same comparisons were targeted by 10 induced miRNAs. Similarly, the top 25 mRNA molecules induced between 19 and 29 weeks (DEG6, DEG7) were targeted by 4 suppressed miRNAs, whereas the top 25 suppressed mRNA molecules were targeted by 10 induced miRNAs. Predictably, each of these miRNA molecules differentially expressed as an outcome of passaging are targeting multiple top induced/suppressed mRNA molecules that are differentially expressed due to passaging. These top concordant miRNA-mRNA interactions are depicted in Supplementary Figure 1.

Effect of Passaging on Pathway Dysregulation

We performed pathway analysis on the differentially expressed datasets individually (DEG4-DEG7) to identify the pathways dysregulated as a function of time or iAs exposure. Results demonstrate that 66–104 pathways were predicted to be dysregulated depending on the comparison (Figure  5A). Interestingly, the majority of dysregulated pathways was predicted to be inhibited between 7 and 19 weeks (DEG4-DEG5). In contrast, most of the pathways dysregulated between 19 and 28 weeks (DEG6-DEG7) were predicted to be activated (Figure  5A). A total of 43 pathways were dysregulated in both DEG4 and DEG5 (Supplementary Table 8). The major pathways commonly dysregulated between DEG4 and DEG5 included those involved in cholesterol/sterol synthesis (superpathway of cholesterol biosynthesis; cholesterol biosynthesis I; cholesterol biosynthesis II [via 24,25-dihydrolanosterol]; cholesterol biosynthesis III [via desmosterol]; mevalonate pathway I) and immune response (natural killer cell signaling; IL-6 signaling; role of IL-17F in allergic inflammatory airway diseases; ILK Signaling). Similarly, 43 pathways were dysregulated in both DEG6 and DEG7 (Figure  5A and Supplementary Table 8). The major pathways commonly dysregulated between DEG6 and DEG7 included those involved in kinase signaling (PKCθ signaling in T lymphocytes; protein kinase A signaling; Tec kinase signaling) and endocrine regulation (estrogen biosynthesis; noradrenaline and adrenaline degradation; MIF-mediated glucocorticoid regulation). Of these, 7 pathways (inhibition of matrix metalloproteases; PD-1, PD-L1 cancer immunotherapy pathway; aldosterone signaling in epithelial cells; calcium signaling; sperm motility; endothelin-1 signaling; dendritic cell maturation) were found to be common across all 4 comparisons (DEG4-7), as depicted in Figure  5B and Supplementary Table 8. These 2 sets of 43 pathways between DEG4-5 and DEG6-7, respectively (including 7 overlaps) were considered to be modulated as a function of passaging, independent of iAs exposure. The activation/inhibition status of these temporally modulated pathways are depicted in Figure  5B and Supplementary Table 8. All the common pathways at any one of the longitudinal comparisons (DEG4-5 or DEG6-7) were found to be activated/inhibited in the same direction (Figure  5B and Supplementary Table 8). Intriguingly, the 7 pathways common to all 4 comparisons were found to be dysregulated in opposite directions between DEG4-5 (7 and 19 weeks) and DEG6-7 (19 and 28 weeks) (Figure  5B and Supplementary Table 8).

Figure 5.

Effect of passaging on pathway dysregulation. A, Bar graph showing the number of predicted dysregulated pathways for each longitudinal comparison along with the number of pathways that are predicted to be dysregulated in both unexposed (DEG4) and exposed (DEG5) HaCaT cells between 7 and 19 weeks and in both unexposed (DEG6) and exposed (DEG7) HaCaT cells between 19 and 28 weeks. Activated pathways as represented by open bars, whereas inhibited pathways are represented by closed bars. Activation is defined as −log(p value) > 1.3; Z-score > 1, whereas inhibition is defined as −log(p value) > 1.3; Z-score < −1. B, Heat map of predicted activated/inhibited pathways that are common between unexposed and exposed HaCaT cells between 7and 19 weeks (DEG4, DEG5) and 19 and 28 weeks (DEG6, DEG7). The color code bar on top refers to the Z-score values. Absence of a bar (represented by white) signifies that the pathway was not predicted to be activated or inhibited for that comparison. The pathways depicted in this panel are presented in the same order as in Supplementary Table 8.

Effect of Chronic iAs Exposure on Pathway Dysregulation

Pathway analysis demonstrated that several pathways were predicted to be dysregulated only in the unexposed (DEG4 and DEG6) or exposed (DEG5 and DEG7) longitudinal comparisons (Supplementary Figure 2 and Table 9). Examination of longitudinal comparisons in unexposed cells revealed 34 and 23 dysregulated pathways between 7 and 19 weeks (DEG4) and between 19 and 28 weeks (DEG6), respectively (Supplementary Figure 2 and Table 9). These pathways were considered to highlight passage-regulated alterations that were prevented by chronic iAs exposure. The major dysregulated pathways in these comparisons include those involved in cell cycle regulation and DNA repair (DEG4: cell cycle control of chromosomal replication; cyclins and cell cycle regulation; mitotic roles of Polo-Like kinase; cell cycle: G2/M DNA damage checkpoint regulation; nucleotide excision repair [NER] pathway; role of BRCA1 in DNA damage response) and in signal transduction (DEG5: PPAR signaling; NF-κB signaling; Toll-like receptor signaling; iNOS signaling; Oncostatin M signaling; ERK5 signaling). Similarly, analysis of longitudinal comparisons in exposed cells revealed 61 and 32 dysregulated pathways between 7 and 19 weeks (DEG5) and between 19 and 28 weeks (DEG7), respectively (Supplementary Figure 2 and Table 9). These pathways represent the changes triggered by chronic iAs exposure independent of passaging. The major dysregulated pathways in these comparisons include signal transduction pathways (DEG6: Netrin signaling; Thrombin signaling; FGF signaling; GNRH signaling; SAPK/JNK signaling; Ephrin B signaling; PI3K/AKT signaling; Ephrin receptor signaling) and receptor-mediated signaling pathways (DEG7: VEGF family ligand-receptor interactions; Fc Epsilon RI signaling; Glutamate receptor signaling; Neuregulin signaling).

DISCUSSION

As the number of arsenic-exposed individuals exceeds 225 million globally (Naujokas et al., 2013; Podgorski and Berg, 2020), it is imperative to develop a thorough understanding of the molecular mechanisms involved in the etiology of its hallmark adverse health outcome, cSCC (Hunt et al., 2014). As mentioned in the Introduction section, in the absence of appropriate animal model of arsenic-induced skin cancer, most of our knowledge in this area is derived from cell culture-based studies (States et al., 2011). Although most studies have not employed a chronic exposure regime, HaCaT cell lines exposed continuously to 100 nM iAs for 28 weeks have emerged as a validated model system to study chronic iAs exposure-induced cSCC (Pi et al., 2008; Sun et al., 2009). Recently, our group has employed this model to demonstrate that each phase of arsenic-induced skin carcinogenesis is characterized by unique molecular signatures stemming from the differential expression of mRNAs and miRNAs regulating each transition (Banerjee et al., 2021). In that study, we used unexposed passage-matched HaCaT cells as the frame of reference to evaluate the effect of chronic arsenic exposure while nullifying the possible effects of passaging cells for long periods. In this study, we empirically examine whether our model system could effectively delineate between the effects of passaging from the effects of chronic iAs exposure. The current work thus demonstrates the robustness of our previous work while further establishing the importance of employing proper passage-matched controls in any longitudinal study employing cell lines.

In this work, our approach was first to determine the highly significant differential miRNA and mRNA expression events with respect to time as well as iAs exposure (DEG4-DEG7). This allowed us to stratify these differential expression events further into those that are common for a longitudinal comparison irrespective of the exposure status (common between DEG4-DEG5 and DEG6-DEG7), providing us with a clear picture of molecules whose expression changes as a function of time (representing passaging) but is not affected by iAs exposure. This novel approach enabled us to investigate the nature of miRNA-mRNA interactions and the pathway dysregulation that are modulated solely by passaging.

The data presented show that a specific set of miRNA and mRNA molecules are being modulated as a function of time (Figs.  2 and 3). Similar transcriptome-wide differential mRNA expression profile has been reported previously by others in different cell lines including mouse insulinoma, rheumatoid arthritis synovial fibroblasts, human retinal pigment epithelium, and porcine human temporomandibular joint disc cells (Allen and Athanasiou, 2007; Lidgerwood et al., 2020; Mouriaux et al., 2016; Neumann et al., 2010; O'Driscoll et al., 2006). Furthermore, it was recently shown that global miRNA expression profiles can also be altered due to passage time (Ikari et al., 2015). Our results are consistent with previous studies examining the effects of passaging time on miRNA and mRNA expression profiles in multiple cell line models. Interestingly, we observed a distinct pattern of expression modulation in miRNAs that are differentially expressed due to passaging. Majority of the common differentially expressed miRNAs between 7 and 19 weeks (DEG4-5) were suppressed, whereas most of the differentially expressed miRNAs between 19 and 28 weeks (DEG6-7) were induced. Although this is a novel finding of the study, we are not sure whether this is a cell line-specific observation or is a common passaging-related outcome irrespective of the cell line. Future longitudinal studies employing different cell lines will be required to answer this question, as well as their possible functional implication.

Next, we wanted to understand if the miRNA molecules differentially expressed as an effect of passaging could be modulating the mRNA molecules that are differentially expressed in a similar manner. The data presented demonstrate that a considerable proportion of the mRNA molecules differentially expressed as an effect of passaging is being targeted by the miRNA molecules whose expression is also modulated by time (Figure  4). This targeting is a novel finding of our study and has not been previously examined as far as we are aware. The data also show that 35%–50% of all the predicted miRNA-mRNA pairings are concordant (Figure  4F). We further demonstrate that several of the top induced/suppressed mRNA molecules are being targeted by multiple members of a small subset of differentially expressed miRNA molecules (Supplementary Figure 1). This raises the possibility that there might be redundance built into the way HaCaT cells respond to passaging. However, further studies will be needed to test if this is true.

In addition, we also observed predicted miRNA-mRNA interactions that are discordant in nature (Figure  4F). However, it is entirely possible that the “discordant” interactions represent homeostatic feedback mechanisms to balance between the mRNA and protein levels of a gene. For example, we have previously shown that protein level of BCL2 is higher at the 19-week time point compared with 7-week time point irrespective of the exposure status (Banerjee et al., 2021). This induction is also reflected in our mRNA seq data that show that BCL2 mRNA is induced in both exposed and unexposed groups between 7- and 19-week time point (Supplementary Table 3). Our expression pairing data shows that BCL2 is targeted by 2 differentially expressed miRNAs (hsa-miR-181c and hsa-miR-34c), both of which are induced and could be preventing further induction of BCL2 post-transcriptionally as is evident from the much lower protein expression level of BCL2 at 28 weeks (Banerjee et al., 2021). Furthermore, miRNAs often could be exerting their regulatory effects on translation repression without impacting the mRNA levels (Bhattacharyya et al., 2006; Wilczynska and Bushell, 2015). In addition, discrepancy in databases regarding targeting information between miRNA and mRNA could also be responsible for potentially underrepresented pairings (Banerjee et al., 2021). Taken together, the data suggest that differentially expressed miRNAs as a function of passaging time can at least in part explain regulating which mRNA molecules are differentially expressed in a time-dependent manner. However, there are likely multiple other molecular mechanisms involved such as epigenetic drift, a known outcome of prolonged serial passaging (Franzen et al., 2021; Garitaonandia et al., 2015; Rubin, 1993; Weissbein et al., 2017).

We went one step further and performed pathway analysis to garner a deeper understanding of how pathways are being modulated through a complex interplay of passaging and chronic iAs exposure. Inspection of the data presented in Figure  5B reveals modulation of several pathways that are known to be affected by passage number. For example, estrogen receptor signaling is predicted to be inhibited between 7 and 19 weeks irrespective of the exposure status (Figure  5B). Suppression of the estrogen receptor mRNA levels with increasing passage number has been demonstrated (Campbell et al., 2002). Interestingly, senescence pathway was predicted to be inhibited between 7 and 19 weeks in both exposed and unexposed cells (Figure  5B). This could possibly be an outcome of activation of multiple cholesterol biosynthesis pathways (Figure  5B) as induction of cholesterol synthesis is known to suppress senescence marker expression (Zhang et al., 2016). The interaction of multiple cholesterol biosynthetic pathways by sharing differentially expressed key rate limiting molecules are depicted in Supplementary Figure 3. Intriguingly, aryl hydrocarbon receptor (AhR) pathway is predicted to be inhibited between 19 and 28 weeks in an iAs independent manner (Figure  5B). AhR pathway also interacts with multiple xenobiotic metabolism pathways by sharing important differentially expressed molecules as depicted in Supplementary Figure 4. Although acute iAs exposure is known to activate AhR pathway in HaCaT cells (Mohammadi-Bardbori et al., 2015), data regarding iAs effect at environmentally and toxicologically relevant levels in the nanomolar range are lacking. Recent data suggest that besides its well-established role in toxic response, AhR signaling also mediates a host of normal physiological processes including cell cycle, apoptosis, and cell migration (Lindsey and Papoutsakis, 2012). Thus, further investigation is required to understand the role of AhR signaling inhibition under conditions of prolonged cell culture.

In addition, several pathways show a complex pattern of modulation dependent upon both time and iAs exposure. In our previous publication, we showed that estrogen-mediated S phase entry was activated in iAs-exposed cells at 7-week time point compared with passage-matched unexposed cells, but not at 19 weeks (Banerjee et al., 2021). This change can be explained by the fact that between 7 and 19 weeks, this pathway is activated in both unexposed and exposed cells, with higher activation in the former (Figure  5B and Supplementary Table 8). Similarly, unfolded protein response pathway is inhibited in exposed cells at 7 weeks, but not at 19 weeks (Banerjee et al., 2021), as the pathway is inhibited both in unexposed and exposed cells between 7 and 19 weeks (Figure  5B and Supplementary Table 8). Such changes are found to be occurring between 19 and 28 weeks as well. Phospholipase C signaling is inhibited in exposed cells at 19 weeks, but not at 28 weeks (Banerjee et al., 2021) as this pathway is getting activated in both unexposed and exposed cells between 19 and 28 weeks (Figure  5B and Supplementary Table 8). Pathways that are predicted to be activated/inhibited unilaterally in unexposed or exposed cells at any longitudinal comparison also agree with our previously published passage-matched data. NER pathway was predicted to be activated in iAs-exposed cells at 7-week time point compared with passage-matched unexposed cells (Banerjee et al., 2021). However, this activation is lost at 19 weeks, as the pathway is activated as a function of time in unexposed cells only, between 7 and 19 weeks (Supplementary Figure 2 and Table 9). Similarly, ERK/MAPK pathway is predicted to be inhibited in 19-week passage-matched exposed cells (Banerjee et al., 2021), because this pathway gets inhibited unilaterally in response to iAs exposure between 7 and 19-week time points (Supplementary Figure 2 and Table 9). Such pattern of iAs-dependent changes are also observed in a plethora of pathways involved in cell cycle regulation and DNA damage response, well-known players in carcinogenesis as well as arsenic toxicity (Curtin, 2012; Hunt et al., 2014; Muenyi et al., 2015).

Protein expression corroboration of RNA-seq predictions is imperative for drawing strong conclusions about the modulation of biological pathways and networks. Our recently published work on passage-matched analysis of this RNAseq dataset (Banerjee et al., 2021) provides considerable support for the conclusions drawn in the current work. We showed a clear pattern of temporal variation in protein expression corresponding to the variations in mRNA expression regardless of iAs exposure. Examples include the epithelial markers ZO-1 and β-Catenin that are both induced at 19 weeks in iAs exposed and unexposed control cells compared with their expression at 7-week time points (Banerjee et al., 2021). Similar temporal variations were also observed in the protein expression of molecules involved in the endoplasmic reticulum stress pathways (ATF4, CHOP, IRE1, NRF2, BCL2), in accordance with the longitudinal RNA-seq predictions (Banerjee et al., 2021). Taken together, the data from this study unequivocally demonstrate that the HaCaT cell line model can distinguish between molecular changes that are brought about by chronic iAs exposure and those that are regulated by time in culture. The results also reinforce our findings from the previous passage matching analyses (Banerjee et al., 2021).

Our study design was optimized with multiple quality control measures to ensure robustness of the data generated, data analysis, and interpretations. We incorporated STR mapping at regular intervals to ensure that the genetic identity of the cells remained uniform throughout the duration of the experiment and was identical to the parental cell line (Banerjee et al., 2021). Long-term, low-dose exposure scenarios are often difficult to simulate in a laboratory setting but are critical for understanding the effects of such exposure. It is absolutely essential for a toxicant like iAs, which is well known to have hormetic effects at multiple biological levels (Calabrese and Baldwin, 2003; Hashmi et al., 2014) as well as very distinct mechanisms and outcomes for acute and chronic exposures (Nurchi et al., 2020; Ratnaike, 2003). However, innumerable studies investigating a mechanistic basis of chronic iAs toxicity and diseases employ µM-mM ranges of iAs exposure coupled to short time periods (hours to days) which is closer to acute iAs exposure than reflecting chronic exposure scenario. This exposure difference highlights the importance of the use of dose and exposure time used in this study for a better simulation of chronic iAs exposure-induced cSCC to generate findings that can be extended to chronically exposed populations. However, the long exposure period (28 weeks) makes it imperative that time of exposure be considered as an important parameter that affects the biological processes. As mentioned in the Introduction section, long-term cellular passaging affects multiple biological processes, including differential gene expression, cell cycle regulation, and stress response (ATCC, n.d.; Briske-Anderson et al., 1997; O'Driscoll et al., 2006). Many of these processes are also involved in regulation of carcinogenesis (Liang and Pardee, 2003; Matthews et al., forthcoming; Moreno-Smith et al., 2010; O'Malley et al., 2020; Wang et al., 2020). Clearly, failure to account for the effect of passaging will lead to erroneous conclusions. The data presented in this work unequivocally demonstrates that some of the molecules and pathways are affected by passaging alone (Figure  5B), some are affected by iAs exposure alone (Supplementary Figure 2), whereas others are affected by both (Supplementary Figure 2). Thus, adopting a study design with inbuilt controls for both exposure and passaging time allowed us to empirically infer how each of these 2 parameters modulate the behavior of molecules, pathways, and networks during the process of chronic iAs exposure-induced cSCC development. This inference would have been impossible if we had adopted a standard time course experimental setup that does not allow for a negative control to rule out the effects of passaging time.

The impacts of this study are far reaching. Although we have used chronic iAs exposure in this case, the conclusions about the effects of passaging and exposure, as well as their interaction, will hold for any chronic exposure scenario, irrespective of the exact nature of the toxicant/treatment under study. Herein, we provide a blueprint for any longitudinal study involving a chronic exposure/treatment to follow by highlighting the necessity and importance of passage matching as a negative control. Studies with high passage number are often necessary in chronic carcinogen exposure-induced malignant transformation but must be designed with utmost caution and consist of inbuilt quantifiable quality control measures at multiple steps to ensure the quality of the data generated. Time course experiments without passage matched controls only control for exposure but not for passaging. Given that passaging time affects many of the miRNA, mRNA, biological pathways and networks, the absence of a proper passaging control leaves the data analysis and interpretation susceptible to include false positive and false negatives as outcomes of exposure erroneously. Thus, the conclusions drawn from studies with such inadequate design are also likely to be flawed. Longitudinal exposure analysis with passage-matched unexposed control for each time point includes negative controls for both the exposure as well as the passaging components. This study design allows for the analysis of data with respect to the effects of exposure and time individually, as well as their interactions empirically with respect to each molecule, pathway, and network. Failure to do so often results in spurious data and cost in terms of wasted research investment of time and finance (Drexler et al., 1999; Hughes et al., 2007).

In conclusion, we pay heed to the dictum that “all models are wrong, some are useful” (Box and Draper, 1987, 2007) and provide empirical evidence to demonstrate that despite its potential drawbacks, our model system and the data derived from it are robust and stand up to scrutiny. In the process, we also provide a blueprint for other long-term longitudinal studies to emulate.

SUPPLEMENTARY DATA

Supplementary data are available at Toxicological Sciences online.