Transcriptional changes associated with resistance to inhibitors of epidermal growth factor receptor revealed using metaanalysis

Inhibitors have been studied principally to find out their mechanism of action in different cancer types. We here aim to study the differential gene expression in EGFR inhibitor-sensitive vs. resistant cancer cell lines. We searched GEO Datasets using key term “EGFR Resistance” and, limiting our choices to human cell lines, selected studies that directly compare transcriptional profiles of matched EGFR inhibitor-sensitive vs. resistant datasets (Table 1). We found 8 appropriate studies comprising 15 datasets and 274 microarrays. In seven datasets EGFR inhibitor-sensitive vs. resistant cell lines were compared without any other treatment. In one study comprising four datasets (GSE38310), two different types of Erlotinib-resistant cell lines were originated from Erlotinib-sensitive cell lines and then treated either with DMSO or Erlotinib. We compared DMSO-treated sensitive cell lines with DMSO-treated resistant cell lines and Erlotinib-treated sensitive cell lines with Erlotinib-treated resistant cell lines. In a Gefitinib study comprising three datasets [10], sensitive and resistant cell lines were treated with Gefitinib and EGF separately or in combination (GSE34228). In one study Erlotinib-sensitive and resistant cell lines were treated with miR-7 (GSE40130). In another study, GSE38404, resistant cell lines were produced by exposing the sensitive cell lines to inhibitor for different time periods. We have combined data from all the resistant cell lines for metaanalysis.

Illumina microarrays were used in majority of datasets, comprising 3 studies and 7 datasets. In four studies, comparing four datasets Affymetrix microarrays were used. Agilent microarrays were used in one study comparing four datasets. Different studies used different types of cell lines for example Non-Small Cell Lung Cancer, Head and Neck Cancer, A431 and Skin cancer cell lines; therefore, we have also individually compared datasets from individual studies (data not shown).

Information about the type of sensitive and resistant cells, pretreatment and type of EGFR inhibitor, labeling accession number, platform, number of datasets, number of sensitive and resistant microarray chips in each dataset is summarized in Table 1.

We have compared these datasets using different approaches; 1) datasets comparing all EGFR inhibitor-sensitive vs. resistant cell lines; 2) Gefitinib- and 3) Erlotinib-sensitive vs. resistant cell lines; 4) irreversible inhibitor-sensitive vs. resistant cell lines; 5) Cetuximab-sensitive vs. resistant cell lines [11]. We used RankProd software to select differentially expressed genes in these individual groups i.e. sensitive vs. resistant, with results graphically presented in (Additional file 2: Figure S2).

Ontological categories overrepresented in lists of differentially expressed genes were identified using DAVID [19]; complete lists are given in Additional files 3, 4, 5, 6 and 7. To condense redundancies, we have also identified clusters for differentially expressed ontological categories in the sensitive vs. Erlotinib-, Irreversible- and Cetuximab-resistant groups (see below).

We have compared overexpressed ontological categories in all sensitive vs. resistant cell lines. Table 2a contains the 10 categories with the best p-values; in Table 2b, we selected the characteristically different categories between sensitive and resistant cells with p-values better than 10⁻⁴.

In Table 2a, we find that in the sensitive cell lines the genes related to systems involved in transport of materials across the membranes are overexpressed, whereas in the resistant cell lines the genes related to the metabolism and macromolecule biosynthesis, especially of proteins, are overexpressed. Although we have found genes associated with regulation of the protein translation in both groups, this category is more significant in the resistant cell lines. Both EGFR inhibitor-sensitive and resistant cell lines overexpress cell proliferation genes, however the sensitive cell lines tend to overexpress apoptosis and cancer-related genes as well. Interestingly, the resistant cell lines seem significantly more responsive to various endogenous and exogenous stimuli than the sensitive ones.

Oncogenesis-related genes and pathways are up-regulated in both sensitive and resistant cell lines. As reported in previous studies, in sensitive cell lines tumor growth seems dependent on the oncogene activation through ErbB receptor kinases [15]. In contrast, in resistant cell lines often the Ras pathway is activated independently of receptors (Table 2b). We note that while Ras is considered a downstream target of EGFR signaling in noncancerous cells [20], the EGFR- and Ras-associated genes comprise widely different groups (Table 2c).

Interestingly, we have observed that in the sensitive cell lines lipids are preferentially metabolized as the source of energy. But as cells become resistant, both carbohydrates and lipids are metabolized to provide energy. In addition, resistant cell lines are inducing the expression of energy generating genes, including the oxidoreductases. This observation suggests that energy production is an important matter for development of resistant cell lines.

Genes involved in responses to various extracellular and intracellular stimuli, for example steroid hormone, hydrogen peroxide and stress, are over expressed significantly in the resistant cell lines compared to sensitive ones. These responses may be related either to cells’ protection from toxins or overall cell survival, as stress and hydrogen peroxide responses prepare the cell for a toxic environment and steroids stimulate proliferation in certain cancers [21,22]. Those cancer cells that increase expression of genes responsive to these stimuli tend to adapt and survive [23].

We have also observed that certain processes related to immunity are overexpressed in the sensitive as well as in resistant cell lines, but we found that ontological categories for these processes have significantly better p-values in the resistant ones. Further, detailed study of the genes involved in immunity revealed that the innate immunity involving the complement component system is active in sensitive cells. In contrast, the resistant cells utilize both innate and adaptive immune systems, especially involving the cytokines and chemokines (Table 2d).

In EGFR inhibitor-sensitive cell lines, cell death and proliferation are relatively suppressed. But in resistant cell lines proliferation is relatively increased, while differentiation is decreased, thus favoring cancer cells growth and persistence. Another major distinction is that sensitive cell lines strongly express adhesion-related genes, whereas in resistant cells the genes related to cell movement are overexpressed. This suggests that the resistant cells have increased tendency to metastasize, perhaps via EMT [13,24].

Surprisingly, protein biosynthesis was the only ontological category in the top ten differentially expressed ones in Erlotinib-sensitive cell lines (Table 3a). In the resistant cells, we have also found mitochondria, immunity and cytoskeleton genes in the top ten ontological categories. In Erlotinib-sensitive cell lines, carbohydrate and protein metabolism genes are overexpressed with very good p-values, 10⁻²¹ (Table 3b). These lines seem to utilize the maximum of their energy reservoirs from glucose and protein molecules. Conversely, in the Erlotinib-resistant cell lines, glycolysis and gluconeogenesis are suppressed and, importantly, genes related to mitochondria and mitochondrial processes are remarkably boosted. These results suggest that increased production of energy to support cellular metabolic processes and survival are very important in the development of Erlotinib resistance. From these observations we suggest that energy level could be the limiting factor for the tumor cells survival in the presence of EGFR inhibitors.

In the Table 4a, we show that the cancer-related genes and pathways are preferentially activated in the sensitive cell lines. But in the cells resistant to Gefitinib, the ontological categories related to macromolecule biosynthesis, specifically of proteins, are most significantly overexpressed. Apparently, the sensitive cell lines have higher tendency for attachment, as the adhesion related genes are upregulated in these, whereas the resistant cell lines are significantly more responsive to endogenous and exogenous stimuli. In the sensitive cell lines ontological categories related to cell cycle, metabolism of proteins and signaling are also up regulated, but the p-values are not as significant as in the resistant cell lines. Tumor related pathways and cell cycle-related genes are also relatively upregulated in resistant cell lines.

Interestingly, we have not found any ontological category related to immune system development overexpressed in the sensitive cell lines. In contrast, in the resistant cell lines immunity systems, specifically biosynthesis of cytokines genes, are boosted (Table 4b). This suggests that Gefitinib-resistant cell lines use cytokines of the immune systems to develop resistance against Gefitinib and thus maintain tumor cells growth and progression [25].

In resistant cells the ontological categories related to responses to various stimuli and angiogenesis were robustly and consistently overexpressed. We have also found increased expression of genes responsive to reactive oxygen species, which ultimately tends to activate apoptosis and immunity pathways in these cells. Expression of genes related to hypoxia has also been seen in the resistant cell lines, which may increase their drug resistance and ultimately their growth rate [26].

Both in Gefitinib-sensitive and in resistant cell lines proliferation and apoptosis are regulated. However, in the resistant cell lines, the cell death processes are negatively regulated and the cell division is enhanced, suggesting that resistant cell lines have higher propensity to survive. Meanwhile, responses to various endogenous and exogenous stimuli are enhanced, maybe to cope with stresses from environment. In addition, immunity genes and cytokine biosynthesis are upregulated to help cancer cells to survive and grow.

Interestingly, we have observed that lipid metabolism genes are prominent in the sensitive cell lines, whereas oxidoreductases, as well as both glucose and lipid metabolism, i.e. genes related to production of energy, are more prominent in the resistant cell lines. Energy requirements for the tumor cells are met preferentially by formation of new blood vessels but also by glucose and lipid metabolism.

We have combined the microarray data from the cell lines selected as resistant to irreversible EGFR inhibitors WZ4002 and PF299804. Unexpectedly, in the top ten categories from the sensitive cell lines, genes related to membrane systems are found (Table 5a). In contrast, the extracellular region-related ontological categories are upregulated in the resistant cell lines. Cell cycle and lipid metabolism categories are relatively upregulated in the sensitive cell lines, similar to what was observed in reversible EGFR inhibitors-sensitive cell lines (Table 4b). Ontological categories related to cell division were found in both groups, however, in the resistant cell lines, cell division is positively regulated with better p-value than in sensitive cell lines. Importantly, the immunity genes are relatively overexpressed in resistant cell lines, reinforcing our observation that cytokine overexpression is one of the survival strategies gained by resistant cell lines (Table 5b).

Surprisingly, we found increased expression of EGFR ligands in cells selected using irreversible inhibitors (Table 5b). Perhaps these cells are still dependent on EGFR-mediated signaling cascades. This differs from observations in cell lines selected for resistance to reversible inhibitors, which use alternative pathways for oncogenes activation (Tables 2b and 3b).

For the analysis of cells selected as resistant to antibody inhibitor Cetuximab, we could find only a single study including six microarrays [27], which means that the statistical robustness of the results is reduced. Interestingly, we found that the differential gene expression in Cetuximab-sensitive and resistant cell lines is quite unlike the differential gene expression seen in the cell lines obtained using tyrosine kinase inhibitors. This agrees well with our previous study documenting that the transcriptional responses to EGFR antibody inhibitors are different from those to kinase inhibitors [9]. In sensitive cell lines, ontological categories related to immunity, cell proliferation, cell migration and response to external stimulus are overexpressed (Table 6a). In contrast, in the resistant cell lines epithelium development- and differentiation-associated gene ontologies are frequently and significantly overexpressed. Intriguing, the ontological category for wound response was overexpressed in resistant cells as well, although it had relatively less significant p-value.

In Table 6b, the overregulated processes in Cetuximab-sensitive cell line comprise immunity responses, signal transduction involving JAK-STAT pathway and angiogenesis. In contrast, in the resistant cell line genes for cell cycle, cell structure and reduction of transcription are increased.

The charts comparing overexpressed ontological categories, although very informative, contain many redundant and overlapping ontological categories (Tables 2,3,4,5 and 6). To get around this problem we have clustered the ontological categories using DAVID software. Cluster outputs for Erlotinib-, irreversible inhibitor and Cetuximab-sensitive versus resistant cell lines are presented in Additional file 8.

In Erlotinib-sensitive cell lines, transcription, translation and protein transport processes, marked as ‘T’ in Additional file 8a, are statistically more significant. In contrast, we found mitochondria-related genes, designated as ‘E’ for ‘energy’, the most frequent ontological category in resistant cell lines. Ontologies related to apoptosis, ‘A’, are the second most upregulated category in sensitive cells. In resistant cell lines genes related to cell cycle, ‘CC’, and cytoskeleton, ‘CY’, are the most expressed categories after energy production. In sensitive cell lines apparently energy is spent for synthesizing proteins and regulating cell death processes, whereas in resistant cells energy is mainly produced and utilized in cell growth maintenance. Thus, if this hypothesis is confirmed by the laboratory approaches, energy starvation strategies could be used to treat cancer cells.

Unexpectedly, categories ‘M’, describing motion of cells, and ‘V’, denoting the vascularization and angiogenesis, were found in sensitive cell lines, whereas adhesion-related genes, ‘AD’, were found relatively overexpressed in resistant cell lines. We have also found that certain categories related to energy metabolism; marked as ‘M-c’ for metabolism of carbohydrates and ‘E’ for energy-related processes are also up regulated in sensitive cells (Additional file 8a).

Although extracellular region ‘EC’ is the top regulated category in irreversible inhibitor-resistant cell lines, important EGFR-activated cellular processes, including immunity, ‘IM’, angiogenesis, ‘V’, and cytoskeleton, ‘CY’, were upregulated (Additional file 8b). In sensitive cell lines membrane systems, ‘MM’, response to hormones, ‘R’, and cell cycle, ‘CC’, were upregulated.

In Cetuximab antibody-resistant cell lines, many ontological categories are suppressed. For example, ontological category ‘IM’, immunity, is statistically the most prominent, biosynthesis of inflammatory cytokine and responses to these cytokines are specifically downregulated. We have observed that gene ontologies related to migration of the cells, marked ‘M’, are prominent in Additional file 8c. Cell adhesion categories, designated as ‘AD’, are also present there but the enrichment score and p-values are comparatively less. Blood vessels development processes, ‘V’, are overexpressed in Cetuximab-sensitive cell lines.

We have also found increased expression of signaling, ‘S’, apoptosis inhibition, ‘A’, and homeostasis-related genes, ‘H’, in sensitive cell lines, but not in resistant cell lines. Some of the overregulated clusters were also related to cell cycle, ‘CC’, transcription, ‘T’, and responses to external stimuli, ‘R’. Ectoderm development category was overexpressed with higher p-value in antibody-resistant cell lines, reflecting epidermal cell origin. Other ontological categories were extracellular matrix ‘EC’ and adhesion (Additional file 8c).

As observed above, (Tables 3,4,5 and 6), we have seen that responses of antibody-resistant cell line are reverse of those observed in tyrosine kinase inhibitors-resistant cell lines (comparing Additional files 8a and b with Table 6c). All ontological categories observed in Cetuximab-sensitive cell lines seem responsible for tumor cell survival and progression and were previously seen overexpressed in tyrosine kinase inhibitor-resistant cell lines.