Skip to main content

Advertisement

Pathway-specific differences between tumor cell lines and normal and tumor tissue cells

Molecular Cancervolume 5, Article number: 55 (2006) | Download Citation

Abstract

Background

Cell lines are used in experimental investigation of cancer but their capacity to represent tumor cells has yet to be quantified. The aim of the study was to identify significant alterations in pathway usage in cell lines in comparison with normal and tumor tissue.

Methods

This study utilized a pathway-specific enrichment analysis of publicly accessible microarray data and quantified the gene expression differences between cell lines, tumor, and normal tissue cells for six different tissue types. KEGG pathways that are significantly different between cell lines and tumors, cell lines and normal tissues and tumor and normal tissue were identified through enrichment tests on gene lists obtained using Significance Analysis of Microarrays (SAM).

Results

Cellular pathways that were significantly upregulated in cell lines compared to tumor cells and normal cells of the same tissue type included ATP synthesis, cell communication, cell cycle, oxidative phosphorylation, purine, pyrimidine and pyruvate metabolism, and proteasome. Results on metabolic pathways suggested an increase in the velocity nucleotide metabolism and RNA production. Pathways that were downregulated in cell lines compared to tumor and normal tissue included cell communication, cell adhesion molecules (CAMs), and ECM-receptor interaction. Only a fraction of the significantly altered genes in tumor-to-normal comparison had similar expressions in cancer cell lines and tumor cells. These genes were tissue-specific and were distributed sparsely among multiple pathways.

Conclusion

Significantly altered genes in tumors compared to normal tissue were largely tissue specific. Among these genes downregulation was a major trend. In contrast, cell lines contained large sets of significantly upregulated genes that were common to multiple tissue types. Pathway upregulation in cell lines was most pronounced over metabolic pathways including cell nucleotide metabolism and oxidative phosphorylation. Signaling pathways involved in adhesion and communication of cultured cancer cells were downregulated. The three way pathways comparison presented in this study brings light into the differences in the use of cellular pathways by tumor cells and cancer cell lines.

Background

Cell lines derived from tumors and tissues comprise the most frequently used living systems in research on cell biology. Limitations on the abundance of tissue samples necessitate the use of animal models and cell lines in the studies of tumor-related phenomena. Cancer cell lines have been extensively used in screening studies involving drug sensitivity and effectiveness of anti cancer drugs [1]. Other studies using cultured cells aimed at the determination of the phenotypic properties of cancer cells such as proliferation rates, migration capacity and ability to induce angiogenesis [2]. In other studies, human cultured cells were used to create tumors in the mice models [3].

Whether measurements on cell lines provide information about the metastatic behavior of cancer cells in vivo is currently under investigation. Unsupervised classification of gene expression profiles of cancer tissue and cancer cell lines result in separate clustering of cancer cell lines from tissue cells for both solid tumors and blood cancers [4]. Sets of genes responsible for differences between solid tumors and cell lines in their response to anti cancer drugs have been identified in the Serial Analysis of Gene Expression (SAGE) Database [5]. Most optimal cell lines to represent given tumor tissue types were determined with the use of a quantitative tissue similarity index [6]. Results were striking: only 34 of the 60 cell lines used in the analysis were most similar to the tumor types from which they were derived. The study provided valuable information about selection of most appropriate cell lines in pharmaceutical screening programs and other cancer research. In a more recent work Sandberg et al. [7] identified those gene function groups for which cell lines differed most significantly from tumors based on meta-analysis using Gene Ontology (GO). Genes involved in cell-cycle progression, protein processing and protein turnover as well as genes involved in metabolic pathways were found to be upregulated (an increase in expression reflected by mRNA transcript levels) in cell lines, whereas genes for cell adhesion molecules and membrane signaling proteins in cell lines were downregulated (a decrease in expression reflected by mRNA transcript levels) in comparison with tumors [7]. To build on this approach, functional enrichment analysis based on Kyoto Encyclopedia of Genes and Genomes (KEGG) pathways [8, 9] can be used to illustrate causal relationships between genes (gene products). While GO is organized into hierarchical annotations in the context of normal cellular function, the KEGG database organizes the genes (gene products) into pathway reaction maps and functional complexes, including some disease-specific pathways.

The present study focuses on pathway specific differences in gene expression patterns between cancer cell lines and tumors as well as cancer cell lines and normal tissue and tumors and normal tissue. Extension of microarray data analysis to three-way comparison allows for the identification of gene expression patterns unique to cell lines. Such patterns might have arisen due to factors related to the cell culture environment. We used publicly accessible microarray data available for normal and cancer tissues and associated NCI60 cell lines in a pathway-specific quantitative analysis of gene expression profiles. A dominant theme that emerged from our analysis was that pathway-specific gene expression differences between cancer cell lines and cancer tissue were similar both in magnitude and direction to corresponding differences between cell lines and normal tissue cells. Cell cycle associated differences between normal and tumor tissue were amplified in cell lines. Results on metabolic pathways suggested an increase in the velocity of RNA and DNA production and increased flow of metabolites in the oxidative phosphorylation pathway. On the other hand, a small fraction of significantly altered genes in tumor-to-normal comparison had similar expressions in cancer cell lines and tumor cells. These genes were tissue-specific and were positioned sparsely along multiple pathways.

Materials and methods

Microarray datasets

Microarray datasets used in this study consisted of the publicly accessible gene expression profile dataset for NCI60 cell lines [10] and similar data for a panel of tumors and normal tissue samples [11]. This dataset contains measurements obtained using the Affymetrix Hu6800 arrays (Table 1). The tissue types considered in this study (breast, CNS, colon, prostate, ovary, and renal tissue) were restricted to those where the microarray results were available for normal and tumor tissue as well as corresponding cell lines. MDA-MB-435 and MDN cell line samples were excluded from these datasets because their tissue of origin, previously thought to be breast, is now suspect [6].

Table 1 Microarray data presented by Staunton et al. [10] and Ramaswamy et al. [11] used in the three way comparison of gene expression patterns in cell lines, tumors and normal tissue.
Full size table

Quality of probe set annotations

Quality of the Hu6800 GeneChip annotation was assessed because this platform is several versions away from current human microarrays. While the Hu6800 design is old and probe designs have since been greatly improved, the quality of probe annotation is maintained through regular updates by Affymetrix. The annotations used in this study are based on a July 12th 2006 update of Affymetrix annotations according to the March 2006 (NCBI Build 36.1) version of the human genome. A comparison was done between gene annotations for the Hu6800 GeneChip obtained from Webgestalt (web-based gene set analysis toolkit) [12] and from the Affymetrix website on August 7th, 2006. Out of the 7129 probesets on the chip, 6058 had the same annotations from both Webgestalt and Affymetrix. Of the remaining 1071 probesets, 692 were not annotated, 288 were annotated in the Affymetrix list but not in Webgestalt, 28 were annotated in Webgestalt but not Affymetrix, and 63 (~1%) probesets had conflicting annotations in Webgestalt and Affymetrix. Only 42 (~0.70% of all genes) genes belonging to any known KEGG pathway had discrepancies between Webgestalt and Affymetrix. While there were very few probes with discrepant annotations in any given pathway, this list of 42 probes was enriched for Antigen processing and presentation, Natural killer cell mediated cytotoxicity, Cell adhesion molecules (CAMs), Type I diabetes mellitus, and SNARE interactions in vesicular transport pathways. A review of this probe list revealed that discrepancies were merely due to updates and minor revisions to the official gene symbol that may reflect increased understanding of these genes functions. Genes associated with KEGG pathways represent a subset of well-studied and sequenced genes. Overall, the probe sets of genes belonging to KEGG pathways have well established and reliable annotations on the Hu6800 GeneChip. Annotations retrieved from Webgestalt were used for the remainder of the analysis.

Normalization

Gene expression data was normalized for each tissue type by computing the Robust Multichip Average (RMA) [13, 14] directly from the Affymetrix .CEL files for cell line, tumor, and normal samples. RMA consists of three steps: a background adjustment, quantile normalization and finally summarization. Quantile normalization method utilizes data from all arrays in an experiment in order to form the normalization relation [13, 14] RMA generated expression measure is on the log base 2 scale.

Normalized data was generated using the Bioconductor (package for R) [15] implementation of RMA. R 2.3.1 [16] was first installed on an Intel Xeon machine running a Windows Professional Operating System. The Biobase 1.10.1 (dated 20 June 2006) package which contains the base functions for Bioconductor was installed by accessing the getBioC.R script directly from the Bioconductor website [17]. The "readaffy" command was used to load all .CEL files for a single tissue type. The RMA expression measures for each tissue type were computed using the "rma" function with default settings, including the Perfect Match Adjustment Method setting as Perfect Match Only so that expression signal calculation was based upon the perfect match values from each probe set as described in [13]. The RMA computed expression values were written out to a comma separated text file.

The resulting expression values for each sample were checked against the average expression across cell line, tumor, and normal populations by calculating their correlation coefficients. Two anomalous samples (one normal tissue sample from colon and one tumor sample from prostate) were identified having correlations well outside the remaining population (R < 0.9) and removed; RMA for those tissues was recomputed excluding the suspect samples. The RMA generated gene expression data for the Affymetrix chips was clustered using a hierarchical clustering algorithm with Pearson correlation coefficient as the distance metric using average linkage using TIGR MeV Version 3.1. For each of the six tissues under consideration, the cell line samples clustered together in a single branch distinct from the branches containing tumor and normal tissue samples. This result confirmed that all the cell line samples have characteristics that are significantly different from the tumor tissue.

Significance analysis for gene expression

The Significance Analysis of Microarray Data (SAM) implementation [18] in the TIGR MeV Version 3.1 software [19] was used to identify those genes that had statistically significant differences in expression between tumor samples, cell lines, and normal tissue. SAM analysis was performed using all default parameters and adjusting the delta-value to obtain a maximum number of genes while maintaining a conservative false discovery rate of zero. A list of significant genes was identified for cell line-tumor cell line-normal and normal-tumor combinations for each of the six tissue types. When the set of significant genes was deleted from the microarray data, clustering analysis based on the remaining genes interspersed microarray datasets for cell lines with corresponding datasets for tissue.

Identification of significantly altered pathways

Two different methods were used for identifying significantly altered pathways. First, Kyoto Encyclopedia of Genes and Genomes (KEGG) pathways [8, 9] were identified as significantly altered by performing a functional enrichment analysis on genes identified as significant by SAM analysis. The analysis was carried out using the Webgestalt system [12], comparing significant genes obtained by SAM against all genes in the Affymetrix HU6800 array, for each comparison under study. A p-value for pathway enrichment was obtained using the hypergeometiric test documented in [12]. Four different p-value cutoffs (0.001, 0.01, 0.05 and 0.1) were used in order to assess the dependence of the significant pathway identification on p value. This process was also applied to subsets of significant genes, for example, the intersection of significant genes from (CL - N) and (T - N).

A second method was applied to KEGG pathway genes in order to detect changes that were not apparent on a single-gene basis. For this method, KEGG pathways were deemed significantly altered if at least 80% of the genes for that pathway contained on the HU6800 array were shifted in the same direction for a given comparison. For each of the six tissues, three-way comparisons were performed between averaged cell line, tumor, and normal samples. Similar examples of how significant changes in functional pathways are revealed by a population of related genes that are not evident from observations of a single gene are found in [20, 21].

Results

Significant genes

This article presents a pathway-specific analysis of gene expression profile differences between cancer cell lines and normal and tumor tissue. The microarray data used in the three-way comparison of gene expression profiles covered breast, CNS, colon, ovary, prostate, and renal tissue (Table 1). Gene expression profiles of cancer cell lines derived from this data clustered together in a branch exclusive of tumor and normal tissue (3) within each tissue type and for all tissue types combined. Lists of significant genes (SAM genes) were determined using SAM analysis from the microarray data pairs of cell lines and tumors (CL - T), cell lines and normal tissue (CL - N) and tumor and normal tissue (T - N) for each of the six tissue types under consideration. Table 2 provides a summary of the numbers of significant genes for the three-way comparison. The table shows that the significant genes for (CL - T) and (CL - N) pairs ranged in numbers from low hundreds to thousands, depending on the tissue type. Significant genes for (T - N) pairs were lower in number than those for (CL - T) and (CL - N) pairs in all six tissues under consideration. Downregulation of significant genes was a trend in (T - N) comparisons while a majority of SAM genes were upregulated in cell lines compared to tumor and normal (CL - T; CL - N). Moreover, an overwhelming majority of the SAM genes in (T - N) comparison were not found as significantly altered in (CL - T) comparisons. The gene set (T - N) - (T - N ∩ CL - T) listed in Table 2 shows a vast majority of SAM genes in (T - N) comparison are not significantly altered in expression in (CL - T) comparison, suggesting that cancer cell lines may be good representation models for tumor cells in gene expression profile studies. On the other hand, the set (CL - T) contains many more genes than the (T - N) comparison, revealing that cancer cell lines have a large number of genes that are significantly altered in expression compared to tumor cells. The same trend holds true when cell lines are compared with normal tissue cells. These results indicate that global gene expression profiles of cultured cancer cell lines contain significantly different gene expression patterns compared to the corresponding profiles for normal and tumor tissue.

Table 2 Number of significant genes identified by SAM in comparisons of cell line-to-tumor (CL - T), cell line-to-normal (CL - N), and tumor-to-normal (T - N) comparisons.
Full size table

SAM genes common in (CL - T) comparisons for all six tissues were all upregulated. Table 3 shows the list of 51 significant genes in (CL - T) comparisons that are common to the six tissue types under consideration. In this list of 51 genes, the overrepresented KEGG pathways with a p-value cutoff of 0.01 are cell cycle, oxidative phosphorylation, proteasome, pyrimidine metabolism, and ubiquitin mediated proteolysis. The 18 genes shown in italics also appeared among 29 significant genes that were common to all (CL - N) comparisons. The 18 genes common to both lists again showed overrepresentation of cell cycle and ubiquitin mediated proteolysis pathways under a p-value cutoff of 0.01. Moreover these eighteen genes showed the same trend of upregulation in cell line-to-tumor (CL - T) and cell line-to-normal (CL - N) comparisons. No significant genes identified in the (T - N) comparisons were common to all six tissues.

Table 3 SAM genes that were upregulated in cell lines compared to tumors in all the 6 tissues considered in the study (CL - T).
Full size table

Significant pathways

KEGG pathways whose gene expression profiles differed significantly in (CL - T), (CL - N), and (T - N) pair comparisons were identified using a hypergeometric test as described in the Methods section. Figure 1 shows the most frequently observed KEGG pathways with altered gene expression profiles for (CL - T), (CL - N) and (T - N) pairs for breast, CNS, colon, ovary, prostate, and renal tissue. Cell cycle and a number of metabolic and transcription-related pathways emerged as significantly altered in almost all (CL - T) and (CL - N) comparison pairs. Cellular pathways that were significantly altered in cell lines compared to tumor cells and normal cells of the same tissue type in at least two tissue types included cell cycle, oxidative phosphorylation, purine and pyrimidine metabolism, proteasome, ribosome, and RNA polymerase. The most striking difference between cell lines and tumor tissue in Figure 1 is in the oxidative phosphorylation pathway. Oxidative phoshorylation is the final stage of cellular metabolism following glycolysis and the citric acid cycles. The loss of cancer cell dependence on oxidative metabolism may be an important factor in the development of tumors [22]. ECM-receptor interaction, which is thought to affect cell migration, appeared with more subtle differences between all comparisons (CL - T), (CL - N), and (T - N). This may reflect more tissue-specific composition of the migration machinery utilized in tumor cell invasion.

Figure 1
figure1

KEGG pathways identified to be significantly altered in cell lines and tumors (CL - T), cell lines and normal tissue (CL - N), and tumor and normal tissue (T - N) comparisons. The term frequency shown in the figure is defined as the ratio of tissue types for which a pathway identified as significantly altered to the total number of tissue types (6). KEGG pathways were identified as significantly altered by using a hypergeometric test with a p-value cutoff. The minimum number of SAM genes in each significantly altered pathway has been set to two. The error bars indicate the standard deviation of frequency for different p- value cutoffs (p = 0.001, 0.01, 0.05 and 0.1).

Full size image

Next we used pathway-specific analysis to identify up- and downregulation patterns in three-way comparisons. Figure 2 provides module maps showing the direction of regulation of KEGG pathways that were identified to be significantly different in at least 2 tissue types in (CL - T) comparisons. The pathways presented in Figure 2a were deemed significantly altered if the average gene expression between two conditions was altered in the same direction for at least 80% of the genes in the pathway. This criterion captured seven of the significant pathways from Figure 1 along with 23 additional pathways. The Figure 2a indicates a high degree of correlation in the direction of Aminoacyl-tRNA synthetases, Monoterpenoid biosynthesis, Proteosome, and RNA polymerase pathway shifts in cell line – tumor and cell line – normal comparisons. Many more pathways appear to be significantly altered in the module map if the criterion for percentage of genes altered in the same direction is reduced from 80% to 70% (Figure 2b). These two module maps illustrate how extensive the pathway alterations are in cell lines compared to tumor and normal tissue (CL - T; CL - N).

Figure 2
figure2

A module map showing the direction of regulation of cellular pathways that were identified as significantly altered in cell lines compared to tumor tissue (CL - T) in at least 2 of the 6 tissues considered in this study. In (a), a pathway is deemed significantly altered if at least 80% of the genes in the pathway are shifted in a common direction. In (b), a pathway is deemed significantly altered if at least 70% of the genes in the pathway are shifted in a common direction. The color red indicates an upregulated pathway, the color green indicates a downregulated pathway, and the color black indicates that the pathway was not significant in that comparison.

Full size image

The pathway-specific results on cell line-tumor microarray data comparisons presented in this study are in agreement with the results recently published by Sandberg et al. [7] on the gene expressions patterns associated with gene ontology categories in cell lines and tumors. These authors have used the same microarray databases used in our study and reached highly similar conclusions on the directions of difference between cell lines and tumors along equivalent pathways and gene ontology categories. Table 4 provides a comparison of the KEGG pathways (from Figure 1) against the most related gene ontology categories from Sandberg et al. [7]. KEGG pathways for complement and coagulation cascade and phenylalanine metabolism passed the significance criteria based on the (T - N) comparison in our study but we could not located the corresponding GO categories in the Sandberg et al. study on cell lines vs. tumor tissue.

Table 4 Comparison of results obtained from this study with those based on Gene Ontology Processes by Sandberg et al. [7]
Full size table

Gene expression changes in metabolic pathways

Metabolic pathways such as oxidative phosphorylation, pyrimidine and purine metabolism account for some of the most significant alterations among the three-way comparisons. The alterations in the oxidative phosphorylation pathway were discussed briefly in the previous section. Purine and pyrimidine metabolic pathways synthesize the nucleotides that make RNA and DNA. All of the nitrogens in the purine and pyrimidine bases (as well as some of the carbons) are derived from amino acids glutamine, aspartic acid, and glycine, whereas the ribose and deoxyribose sugars are derived from glucose. Figure 3 shows the KEGG diagram of pyrimidine metabolism with the expression values (averaged over six tissues) overlaid for (CL - T) (3a), (CL - N) (3b), and (T - N) (3c) comparisons. This KEGG pathway is altered with upregulated expression for a majority of genes in cell lines and tumors when compared to normal tissue. The increased levels of pyrimidine metabolism gene expression are most pronounced in cell lines (Fig 3a). A predicted increase in the velocity of RNA and DNA base production in cell lines is consistent with trends of increasing rates of cell division observed in cell cultures [23]. The observation that nucleotide metabolism accelerates in cancer has been discussed in the literature. Development of pyrimidine and purine analogs as potential antineoplastic agents evolved from an early presumption that cancer is a disease of uncontrolled growth and nucleic acids are involved in growth control [24].

Figure 3
figure3

KEGG pyrimidine metabolism diagram. Gene expression shifts are projected from comparisons of cell line-to-tumor (CL - T), cell line-to-normal (CL - N), and tumor-to-normal (T - N) comparisons averaged over all six tissues. The color red indicates upregulated genes, green indicates downregulated genes and grey indicates the genes that are not on the microarray. Uncolored genes are not in the organism-specific pathway for Homo sapiens. A gene is identified as upregulated (downregulated) if its gene expression value averaged over 6 tissue types were greater (or lesser) in cell lines compared to tumor or normal tissue. Colored genes with white lettering were also identified with SAM in at least two tissues.

Full size image

Gene expression pattern changes in cell cycle

In contrast to the pyrimidine metabolism pathway discussed above, the gene expression alterations along the cell cycle pathway appear to be more complex and tissue-specific. Figure 4a shows the KEGG diagram of cell division cycle with genes specific to Homo sapiens shaded light green. Figure 4b shows the extent of alteration of these genes in the three-way comparisons for each tissue type with a graded color map representing maximum upregulation in red and maximum downregulation in green.

Figure 4
figure4

KEGG cell cycle diagram. Genes are shown (a) in a pathway map with genes specific to homo-sapiens shaded light green and (b) tabulated with a color map showing average gene expression shifts for samples within the six tissues. Red indicates a positive change and green indicates a negative change in average RMA value for the respective cell line-tumor (CL - T), cell line-normal (CL - N), and tumor-normal (T - N) comparisons, with color scale limits set to -2 and +2.

Full size image

Perhaps the most obvious feature of this color map is how subtle the changes in (T - N) comparisons are relative to (CL - T) and (CL - N) comparisons in all six tissues under consideration. Genes such as CCNA2, CCNB1, CDC20, CDK4, and MDM2 through MDM7 are consistently upregulated in cell lines compared to tumors and normal tissue. On the other hand, genes such as CCND1, CCND3, CDC16, and CDK2 do not exhibit quickly a recognizable pattern. A multitude of gene expression profiles in cell cycle may point towards the same disease process.

SAM genes common to cancer cell lines and tumor cells

It is of interest to cell biologists to identify similarities between cancer cell lines and tumors. Towards that goal, one can determine the list of SAM genes belonging to both (T - N) and (CL - N) comparisons but do not appear to be significant in (CL - T) comparison. This list is shown in Table 5 for all six tissues under consideration. Table 5 gives an indication of the size of the SAM gene subsets that are preserved and commonly regulated in cell lines and tumors but not in normal tissues. The list of genes in Table 5 comprises mostly downregulated genes for breast, colon, ovary, prostate, and renal tissue, with CNS as the only exception. When these lists were projected onto KEGG pathways, the probability of enrichment score could not be used as an indication that the pathways are similar because KEGG pathways that include genes from these lists also included SAM genes from (CL - T) comparisons. In conclusion, it was not possible to assert pathway similarity with statistical confidence using this analysis.

Table 5 Genes that were identified by SAM in both (T - N) and (CL - N) comparisons but not in (CL - T) comparisons; (T - N ∩ CL - N) – (T - N ∩ CL - N ∩ CL - T).
Full size table

Conclusion

Our study shows that a large portion of genes implicated in the emergence and progression of cancer have similar gene expression values in tumors and cancer cell lines indicating the value of cultured cell lines in cancer research. However, the pair-wise comparisons of gene expression profiles of CL, T, and N across all tissues illustrate that there are pronounced changes in gene expression specific to cell lines (CL - T; CL - N) that may not represent a disease process. This study also identified the signaling and metabolic pathways in cell lines that have distinctly different gene expression patterns than those associated with normal and tumor tissue. Pathway-specific gene expression changes in (CL - T) and (CL - N) comparisons were more consistent than (T - N) comparisons in the set of six tissues under consideration. Just as the gene expression changes in tumor – normal tissue comparison were largely tissue-specific, the significantly altered pathways among tumor – normal comparisons were limited to a small number of tissues. Functional enrichment analysis allows us to explore significant changes in pathways despite having heterogeneous changes in gene expression across different tissues. Cellular pathways that were significantly upregulated in cell lines compared to tumor cells and normal cells of the same tissue type included ATP synthesis, cell cycle, oxidative phosphorylation, purine, pyrimidine and pyruvate metabolism, and proteasome. Results on metabolic pathways suggested an increase in the velocity nucleotide metabolism and RNA production.

The dominant trend in the gene expression profiles along significantly altered pathways in cell lines appeared to be upregulation of genes when compared either to tumor or normal tissue. Exceptions included genes in the cell adhesion molecules, cell communication, and ECM-receptor interaction, focal adhesion, and complement/coagulation cascade pathways. The apparent downregulation of the complement/coagulation cascade in cell lines may be due to the heterogeneous mixture of cells in tumor samples including immune cells as well as tissue-specific cells.

The composition of the cell culture medium may be the reason why gene expression patterns that differentiate cancer cell lines from tumor tissue are similar to those patterns that differentiate between cell lines and normal tissue. Typical cell culture medium is replete with metabolites, growth factors, and cytokines, among others, for which cells normally must compete in vivo [24]. Multicellular interfaces with which tumor cells interact in vivo are not replicated for cells grown in cell culture plates [26–29]. The differences in environmental selection pressures may help explain the differential gene expression patterns between the tumor tissue and the cell lines. Our finding about the upregulation of oxidative phosphorylation in cell lines is supported by previous metabolic studies [30, 31]. The documentation of gene expression differences along signaling and metabolic pathways is important in compound screening during the drug discovery process. Compounds may affect significantly altered pathways between cell lines and tumor tissue differently. Recent studies are taking advantage of the technological advances in microfluidics and tissue engineering to develop three-dimensional cell culture systems that aim simulating in vivo culture conditions. Whether cell lines can be made to mimic tumor cell gene expression patterns by altering the culture medium conditions is a question yet to be fully explored.

References

  1. 1.

    Yamori T: Panel of human cancer cell lines provides valuable database for drug discovery and bioinformatics. Cancer Chemother Pharmacol. 2003, 52 (Suppl 1): S74-9. 10.1007/s00280-003-0649-1

  2. 2.

    Kim JB, Stein R, O'Hare MJ: Three-dimensional in vitro tissue culture models of breast cancer – a review. Breast Cancer Res Treat. 2004, 85 (3): 281-91. 10.1023/B:BREA.0000025418.88785.2b

  3. 3.

    Price JE, Zhang RD: Studies of human breast cancer metastasis using nude mice. Cancer Metastasis Rev. 1990, 8: 285-297. 10.1007/BF00052605

  4. 4.

    Ross DT, Perou CM: A comparison of gene expression signatures from breast tumors and breast tissue derived cell lines. Dis Markers. 2001, 17 (2): 99-109.

  5. 5.

    Stein WD, Litman T, Fojo T, Bates SE: A Serial Analysis of Gene Expression (SAGE) database analysis of chemosensitivity: comparing solid tumors with cell lines and comparing solid tumors from different tissue origins. Cancer Res. 2004, 64 (8): 2805-16. 10.1158/0008-5472.CAN-03-3383

  6. 6.

    Sandberg R, Ernberg I: Assessment of tumor characteristic gene expression in cell lines using a tissue similarity index (TSI). Proc Natl Acad Sci USA. 2005, 102 (6): 2052-2057. 10.1073/pnas.0408105102

  7. 7.

    Sandberg R, Ernberg I: The molecular portrait of in vitro growth by meta-analysis of gene-expression profiles. Genome Biol. 2005, 6 (8): R65- 10.1186/gb-2005-6-8-r65

  8. 8.

    Kanehisa M: A database for post-genome analysis. Trends Genet. 1997, 13: 375-376. 10.1016/S0168-9525(97)01223-7

  9. 9.

    Kanehisa M, Goto S: KEGG: Kyoto Encyclopedia of Genes and Genomes. Nucleic Acids Res. 2000, 28: 27-30. 10.1093/nar/28.1.27

  10. 10.

    Staunton JE, Slonim DK, Coller HA, Tamayo P, Angelo MJ, Park J, Scherf U, Lee JK, Reinhold WO, Weinstein JN, Mesirov JP, Lander ES, Golub TR: Chemosensitivity prediction by transcriptional profiling. Proc Natl Acad Sci USA. 2001, 98 (19): 10787-92. 10.1073/pnas.191368598

  11. 11.

    Ramaswamy S, Tamayo P, Rifkin R, Mukherjee S, Yeang CH, Angelo M, Ladd C, Reich M, Latulippe E, Mesirov JP, Poggio T, Gerald W, Loda M, Lander ES, Golub TR: Multiclass cancer diagnosis using tumor gene expression signatures. Proc Natl Acad Sci USA. 2001, 98 (26): 15149-54. 10.1073/pnas.211566398

  12. 12.

    Zhang B, Kirov S, Snoddy J: WebGestalt: an integrated system for exploring gene sets in various biological contexts. Nucleic Acids Res. 2005, 33: W741-8. 10.1093/nar/gki475

  13. 13.

    Irizarry RA, Hobbs B, Collin F, Beazer-Barclay YD, Antonellis KJ, Scherf U, Speed TP: Exploration, normalization, and summaries of high density oligonucleotide array probe level data. Biostatistics. 2003, 4 (2): 249-64. 10.1093/biostatistics/4.2.249

  14. 14.

    Bolstad BM, Irizarry RA, Astrand M, Speed TP: A Comparison of Normalization Methods for High Density Oligonucleotide Array Data Based on Bias and Variance. Bioinformatics. 2003, 19 (2): 185-193. Supplemental information, 10.1093/bioinformatics/19.2.185

  15. 15.

    Gentleman RC, Carey VJ, Bates DM, Bolstad B, Dettling M, Dudoit S, Ellis B, Gautier L, Ge Y, Gentry J, Hornik K, Hothorn T, Huber W, Iacus S, Irizarry R, Leisch F, Li C, Maechler M, Rossini AJ, Sawitzki G, Smith C, Smyth G, Tierney L, Yang JY, Zhang J: Bioconductor: open software development for computational biology and bioinformatics. Genome Biol. 2004, 5 (10): R80- 10.1186/gb-2004-5-10-r80

  16. 16.

    Ihaka R, Gentleman R: R: A Language for Data Analysis and Graphics. Journal of Computational and Graphical Statistics. 1996, 3 (5): 299-314. 10.2307/1390807. 10.2307/1390807

  17. 17.

    Bioconductor for R Installation Script. [http://www.bioconductor.org/getBioC.R]

  18. 18.

    Tusher VG, Tibshirani R, Chu G: Significance analysis of microarrays applied to the ionizing radiation response. Proc Natl Acad Sci USA. 2001, 98: 5116-5121. 10.1073/pnas.091062498

  19. 19.

    Saeed AI, Sharov V, White J, Li J, Liang W, Bhagabati N, Braisted J, Klapa M, Currier T, Thiagarajan M: TM4: a free, open-source system for microarray data management and analysis. Biotechniques. 2003, 34: 374-378.

  20. 20.

    Segal E, Friedman N, Koller D, Regev A: A module map showing conditional activity of expression modules in cancer. Nat Genet. 2004, 36 (10): 1090-8.

  21. 21.

    Segal E, Friedman N, Kaminski N, Regev A, Koller D: From signatures to models: understanding cancer using microarrays. Nat Genet. 2005, S38-45. 37 Suppl.

  22. 22.

    Chevrollier A, Loiseau D, Gautier F, Malthiery Y, Stepien G: ANT2 expression under hypoxic conditions produces opposite cell-cycle behavior in 143B and HepG2 cancer cells. Mol Carcinog. 2005, 42 (1): 1-8. 10.1002/mc.20059

  23. 23.

    A predicted increase in the velocity of RNA and DNA base production in cell lines is in line with trend of increasing rates of cell division observed in cell cultures.

  24. 24.

    Kufe Donald W, Pollock Raphael E, Weichselbaum Ralph R, Bast Robert C, Gansler Ted S, Holland James F, Frei Emil : Pyrimidine and Purine Antimetabolites. Cancer Medicine Inc. Edited by: Decker BC. 2003, 50-6.

  25. 25.

    Vogel TW, Zhuang Z, Li J, Okamoto H, Furuta M, Lee YS, Zeng W, Oldfield EH, Vortmeyer AO, Weil RJ: Proteins and protein pattern differences between glioma cell lines and glioblastoma multiforme. Clin Cancer Res. 2005, 11 (10): 3624-32. 10.1158/1078-0432.CCR-04-2115

  26. 26.

    Dipasquale B, Colombatti M, Tridente G: Morphological heterogeneity and phenotype modifications during long term in vitro cultures of six new glioblastoma cell lines. Tumori. 1990, 76 (2): 172-8.

  27. 27.

    Chen MH, Yang WK, Whang-Peng J, Lee LS, Huang TS: Differential inducibilities of GFAP expression, cytostasis and apoptosis in primary cultures of human astrocytic tumours. Apoptosis. 1998, 3: 171-82. 10.1023/A:1009698822305

  28. 28.

    Pandita A, Aldape KD, Zadeh G, Guha A, James CD: Contrasting in vivo and in vitro fates of glioblastoma cell subpopulations with amplified EGFR. Genes Chromosomes Cancer. 2004, 39: 29-33. 10.1002/gcc.10300

  29. 29.

    Pieper RO: Defined human cellular systems in the study of glioma development. Front Biosci. 2003, 8: s19-27.

  30. 30.

    Faure Vigny H, Heddi A, Giraud S, Chautard D, Stepien G: Expression of oxidative phosphorylation genes in renal tumors and tumoral cell lines. Mol Carcinog. 1996, 16 (3): 165-72. 10.1002/(SICI)1098-2744(199607)16:3<165::AID-MC7>3.0.CO;2-G

  31. 31.

    Franks SE, Kuesel AC, Lutz NW, Hull WE: 31P MRS of human tumor cells: effects of culture media and conditions on phospholipid metabolite concentrations. Anticancer Res. 1996, 16 (3B): 1365-74.

Download references

Acknowledgements

This study was supported by the National institute of Health (NIH) grant #232240 and by the National science Foundation (NSF) grant # 235327.

Author information

Affiliations

  1. Center for Integrated Bioinformatics, School of Biomedical Engineering, Science and Health Systems, Bossone 714, Drexel University, 3143 Chestnut Street, Philadelphia, PA, 19104, USA

    • Adam Ertel
    • , Arun Verghese
    •  & Aydin Tozeren

  2. Lombardi Comprehensive Cancer Center at Georgetown University, 3800 Reservoir Road, Washington, DC, 20057, USA

    • Stephen W Byers

  3. Division of Bioinformatics, Fox Chase Cancer Center, 333 Cottman Avenue, Philadelphia, PA, 19111-2497, USA

    • Michael Ochs

Authors

  1. Search for Adam Ertel in:

  2. Search for Arun Verghese in:

  3. Search for Stephen W Byers in:

  4. Search for Michael Ochs in:

  5. Search for Aydin Tozeren in:

Corresponding author

Correspondence to Aydin Tozeren.

Authors’ original submitted files for images

Below are the links to the authors’ original submitted files for images.

Authors’ original file for figure 1

Authors’ original file for figure 2

Authors’ original file for figure 3

Authors’ original file for figure 4

Authors’ original file for figure 6

Authors’ original file for figure 7

Authors’ original file for figure 8

Rights and permissions

Reprints and Permissions

About this article

Keywords

  • Functional Enrichment Analysis
  • Gene Ontology Category
  • Pyrimidine Metabolism
  • Robust Multichip Average
  • NCI60 Cell Line

Molecular Cancer

ISSN: 1476-4598