Background
5-FU is one of the standard drugs used in chemotherapeutic regimens for metastatic colorectal cancer worldwide, despite the fact that resistance to 5-FU is a major obstacle to successful therapy. Elucidation of the complex biological mechanisms involved in development of resistance to 5-FU is an essential step towards predicting or overcoming such resistance. Established colorectal cancer cell lines subjected to different 5-FU treatment protocols are often used as
in vitro model systems for investigations of downstream cellular responses to 5-FU and to generate 5-FU-resistant derivatives for the study of biological mechanisms involved in selection for drug resistance. We used the HCT116 colon cancer cell line for investigations of response to 5-FU and development of resistance because this cell line induces high and dose-dependent levels of apoptosis in response to increasing concentrations of 5-FU and can thus be considered to be sensitive to 5-FU. We generated treatment-derivative cell lines from parental HCT116 cells subjected to two different 5-FU treatment protocols. One protocol simulated the clinical bolus regimens using clinically-achievable 5-FU levels, the other utilized serial passage of HCT116 cells and continuous exposure to increasing 5-FU concentrations. The latter method is the more traditional one used to generate drug resistance, but one drawback is that such protocols often use higher, potentially clinically-irrelevant drug concentrations. The in vitro bolus 5-FU exposure protocol used was based on a previously-published protocol designed to simulate clinical bolus regimens in order to generate treatment-derivative cell lines [
1]. This was achieved by a stepwise dilution of medium after 5-FU addition that was timed to approximate clinical clearance kinetics. It was assumed that this protocol would generate 5-FU-resistant derivatives of HCT116 after only three cycles of treatment, since it was previously shown to do so for two other colorectal cell lines, HT29 and HCT8. We thus opted to use three cycles of bolus treatment and to subsequently characterize the resultant derivative(s) generated regardless of resistance status at treatment endpoint.
5-FU resistance phenotypes were assessed in the treatment derivatives generated by both exposure protocols using growth inhibition and apoptosis assays. We also compared gene expression in the HCT116 treatment-derivatives generated by both 5-FU exposure protocols in order to determine whether the signaling and regulatory pathways involved in mediating the downstream response to 5-FU exposure (and possibly involved in development of resistance) were similar or substantially different as a result of differential treatment. Gene expression in the treatment derivatives and parental cells was detected and quantified using oligonucleotide microarrays. Microarray technology facilitates a more complete and inclusive experimental approach whereby alterations at the transcript level can be simultaneously measured for entire genomes in response to a defined stimulus, e.g. drug treatment [
2]. This allows for a more rapid determination of whether specific signaling and/or regulatory pathways have in fact been affected in common for all cell lines examined, or if they are substantially different from one cell line to another. The results of our investigations demonstrate that there were strong similarities in gene expression among all the resultant treatment-derivatives despite the fact that they were generated by two different treatment protocols, and despite the fact that the bolus derivative, unlike the continuous exposure derivatives, had not become resistant to 5-FU after three cycles of treatment. Selection pressures on HCT116 cells as a result of 5-FU challenge are thus similar for both treatment protocols.
Discussion
We compared gene expression and specific growth- and apoptosis-associated phenotypes in HCT116 treatment-derivatives generated either by continuous exposure to 5-FU or by using an exposure protocol that more directly simulated the clinical bolus regimens. The aim was to determine whether the signaling and regulatory pathways involved in mediating the downstream response to 5-FU exposure were similar or substantially different as a result of these two different treatment protocols. A previous study [
1] attempted to systematically address this question by determining the activity of different enzymes involved in 5-FU metabolism in cell lines generated by these two methods, and demonstrated differences in enzyme activity levels between the respective resultant derivatives. However, the authors did not have DNA microarray technology available to them at that time in order to gain complete transcriptional profiles of the cell lines used in their study. Use of this technology in the present study yielded comparative information about expression levels simultaneously for the same 8500 genes for three derivative cell lines and one parental cell line (baseline) from which the other three were derived, allowing for a more rapid determination of whether specific signaling and/or regulatory pathways had in fact been affected in common for all derivative cell lines examined, or if they were substantially different from one cell line to another.
All derivative cell lines demonstrated similar transcriptional profiles, despite the facts that they were generated by two different 5-FU exposure protocols and that they differed substantially in their levels of resistance to 5-FU. This suggests that similar signaling and regulatory pathways were involved in mediating the downstream response to 5-FU in each of the derivative lines, i.e., that the selection pressures on the cells in terms of dealing with 5-FU challenge are similar for both exposure protocols. Despite the fact that use of the in vitro bolus model for resistance development did not result in 5-FU-resistant derivatives after 3 cycles of treatment, we opted to proceed with characterizations of the resultant bolus derivatives generated. We reasoned that these derivatives could provide valuable information about the early downstream responses to 5-FU. The Bolus3 derivative had a transcriptional profile that was very similar to that seen in the strongly-resistant ContinB derivative. The actual genes selected for and expressed in response to 5-FU exposure, and the levels of gene expression were very similar between the two cell lines; additionally, many of the same genes were scored as up- or down-regulated in both derivatives. The Bolus3 and Bolus1 derivatives were also strongly correlated for gene expression levels and gene expression level changes, not surprisingly since the Bolus3 derivative is a simply a sequential continuation of the Bolus1 cell line, differing only in the number of bolus cycles it has undergone. Gene expression analysis of Bolus1 cells was included in the study in order to gain valuable information about the transcriptional profile associated with one cycle of 5-FU treatment and recovery. The results indicate that already after one round of 5-FU treatment and recovery that these cells had a transcriptional profile that was very similar to that seen in the Bolus3 and ContinB cells. The ContinD derivative demonstrated strong correlations with the ContinB, Bolus3 and Bolus1 derivatives for gene expression levels, but only moderate correlations with the ContinB, Bolus3 and Bolus1 derivatives for gene expression changes. A strong correlation was demonstrated with a replicate ContinD sample for gene expression levels, indicating a high degree of reproducibility of the data. All HCT116 treatment-derivative lines also demonstrated gene expression that was very similar to the parental HCT116 cell line, reflecting their origin from this cell line. However, the Bolus 3 and ContinB cell lines were the treatment-derivatives that differed most from the parental line, yet were most like each other. The ContinD derivative demonstrated the strongest correlation with the parental HCT116 cell line with regard to gene expression levels of all the derivatives, followed by the Bolus1 derivative, thus the ContinD derivative is most like the parental line. This was also demonstrated by cluster analysis using gene expression level data for all cell lines. The ContinD line clustered together with the Bolus1 line, whereas the ContinB and Bolus3 lines clustered together; the parental HCT116 cell line clustered together with each of these individual clusters, but was much closer distance-wise to the ContinD/Bolus1 cluster. Further analysis of the gene expression data for all of the HCT116 treatment-derivatives, with specific focus on the differences in gene expression between these cell lines, will lead to an explanation of what defines resistance to 5-FU in these cells. We conclude that the ContinB and ContinD derivatives can be useful in vitro models for the study of the molecular mechanisms underlying 5-FU resistance, whereas the Bolus3 derivative, while not resistant to 5-FU, can be useful for the study of early cellular responses to 5-FU exposure.
Recent articles about resistance to anticancer agents have discussed the importance of drug target alterations, DNA repair, and evasion of apoptosis as relevant mechanisms of drug resistance [
4,
5]. In the present study, genes on signaling and regulatory pathways associated with 5-FU metabolism, apoptosis, and DNA repair were among the many genes shown to be significantly altered during selection for 5-FU resistance. The main mechanism of action of 5-FU is to inhibit thymidylate synthase, a key enzyme in the de novo synthesis of dTMP from dUMP [
2,
6]. Cells continually exposed to 5-FU must compensate for the inhibition of thymidylate synthase and deal with the DNA damage / eventual apoptosis that may result due to FdUTP misincorporation into DNA. Overexpression of the thymidylate synthase (
TYMS) gene is a characteristic of 5-FU-resistant cell lines [
7,
8] and is predictive of poor response to 5-FU treatment in colorectal cancer [
9,
10]. The
TYMS gene was overexpressed nearly 2-fold in all of the HCT116 treatment-derivatives in the present study compared to the parental cells. This suggests that up-regulation of
TYMS is one of the first cellular responses to repeated 5-FU exposure, but that it may not necessarily define a 'resistance' phenotype, since the Bolus3 derivative, while not 'resistant' to 5-FU as assessed by growth inhibition assays, expressed levels of this gene similar to those seen in the strongly-resistant ContinB and ContinD derivatives. Several other genes involved in 5-FU metabolism (e.g.
DTYMK, UP, DHFR) were also up-regulated in both the Bolus3 derivative and in one or both of the continuous derivatives, suggesting that their overexpression contributes to but may not necessarily define a 'resistance' phenotype.
Since we have a documented reduced apoptosis phenotype for the ContinB and ContinD derivatives (but not for the Bolus3 derivative), we conclude that loss of sensitivity to apoptosis induction accompanies selection for resistance to 5-FU in these cell lines. This phenotype cannot be explained by loss of wild-type p53 function, since both the ContinB and ContinD cell lines retained a wild-type
TP53 genotype [
3], and both cell lines induce p53 in response to transient 5-FU challenge in similar fashion to the parental cell line. It was thus relevant to examine expression levels of other genes involved in the regulation of apoptosis. Genes involved in apoptosis promotion, e.g.,
ANXA4, BNIP3L, and
PDCD4, were strongly down-regulated in the ContinB and Bolus3 derivatives (but not in the ContinD derivative) in the present study. Genes involved in apoptosis inhibition, e.g.
BIRC5 and
MALT1, were strongly up-regulated in the ContinB derivative but not in the Bolus3 or ContinD derivatives, whereas
IRAK1 was strongly up-regulated in both the ContinB and Bolus3 derivatives, but not in the ContinD derivative. Overexpression of
BIRC5,
MALT1, and
IRAK1 may thus contribute to the reduced apoptosis phenotype in the ContinB derivative cell line, but other (still unknown) apoptotic-regulatory genes might account for the same in the ContinD derivative. Overall, our results are consistent with recent publications that suggest that overexpression of apoptosis-inhibitory genes or underexpression of apoptosis-promoting genes contribute to a drug resistance phenotype [
4,
5], at least for the ContinB derivative.
Likewise, DNA repair genes such as
FEN1 and
FANCG were overexpressed in both continuous-treatment derivatives, suggesting that the latter can repair DNA more effectively than the parental line, consistent with recent studies that suggest that DNA repair contributes to general drug resistance [
4,
5]. It is unknown at present whether overexpression of these genes results in more efficient DNA repair since we have not measured DNA repair activity. The ContinB derivative also overexpressed
RAD23B. The Bolus3 cell line also overexpressed all three of these DNA repair genes at levels similar to those measured for the continuous derivatives, despite its lack of resistance to 5-FU. An investigation of what underlies the strong 5-FU resistance demonstrated in the ContinB and ContinD cells must therefore focus on other highly-up-regulated or down-regulated genes and their levels of expression, as well as on the alterations that have occurred at the gene level (e.g. increase in copy number). These will involve both functional studies using gene silencing and the use of genomic arrays, respectively, and this work is currently underway in our laboratory.
The bolus exposure regimen employed in the present study was previously shown to generate resistant cell populations after only 3 cycles of treatment in two different colorectal cancer cell lines, HCT8 and HT29 [
1], but this was not the result for the HCT116 cells used in our study. One probable explanation for this is that HCT116 cells may be more difficult to make resistant to 5-FU than HCT8 and HT29 cells, and that it would take exposure to higher concentrations of the drug or to increased numbers of treatment cycles in order to achieve resistance. The fact that HCT116 cells are DNA-mismatch repair-deficient due to lack of hMHL1 expression may explain their tolerance to 5-FU, since loss of DNA mismatch repair has been shown to confer resistance to 5-FU [
11,
12]. It has been shown that lack of hMHL1 expression is the result of hypermethylation of the hMHL1 gene, and a recent publication demonstrated that it was possible to restore sensitivity to 5-FU (in effect, overcome
in vitro resistance) by re-expression of hMHL1 protein through demethylation in hypermethylated cell lines [
13]. We have not assessed the methylation status of this gene in any of the cell lines used in this study.
Methods
Cell lines, culture conditions and chemicals
Early passage HCT116 cells (ATCC CCL 247) derived from a poorly-differentiated colonic adenocarcinoma were used. These cells have a wild-type
TP53 genotype, a somatic frameshift mutation in exon 3 of the
BAX gene [
14,
15], DNA-mismatch repair-deficiency due to lack of hMHL1 expression [
16], two different thymidylate synthase enzymes [
17,
18] and a near-diploid DNA complement. Cells were maintained as monolayers in RPMI 1640 supplemented with 10% heat-inactivated fetal calf serum, 2 mM glutamine and 0.6% Pen-Strep at 37°C in a 5% CO
2 atmosphere (all cell culture reagents and plasticware purchased from Invitrogen Gibco). 5-FU was obtained from Amersham Life Sciences, England.
Serial passage of HCT116 cells in increasing 5-FU concentrations (continuous exposure)
Derivative cell lines were generated by serial passage of parental HCT116 cells in the presence of increasing concentrations of 5-FU, and recently characterized for various phenotypes and genotypes [
3]. Briefly, intermittent 5-FU treatments of exponentially-growing HCT116 cells for 48 hours, followed by recovery periods in drug-free medium until the cells regained exponential growth (an interval corresponding to circa 18 doubling times), were used to generate independent treatment derivatives. The concentrations of 5-FU used ranged from 5–770 μM. Two independent derivatives, designated ContinB and ContinD (original designation ResB and ResD; designation changed for the present study in order to emphasize difference between derivatives generated by continuous 5-FU exposure versus those generated by bolus treatment), were selected and cultured further. Drug treatments continued until both derivative cell lines demonstrated stable resistance to 5-FU challenge, as assessed by growth inhibition assays.
Bolus 5-FU exposure of HCT116 cells (clinical clearance kinetics model)
Pizzorno and Handschumacher [
1] used 5-FU concentrations that were within the achievable clinical range to treat two different colon cancer cell lines, using an exposure protocol that more directly simulated the clinical regimens currently in use. The bolus protocol provided a drug concentration profile similar to that obtained in patients after a 600 mg/m
2 dose of 5-FU [
1]. This was achieved by a stepwise dilution of medium after 5-FU addition that was timed to approximate clinical clearance kinetics. The initial 5-FU concentration of 500 μM was diluted out to 250 μM, 100 μM, 20 μM, 2 μM and finally to 0.5 μM over the course of approximately 5 hours. Cells were then allowed to incubate in 0.5 μM 5-FU for 24 hours. The following day, the same regimen was repeated. This protocol was repeated daily for 5 consecutive days, at which point the cells were removed to drug-free media via a medium shift and allowed to recover. The entire regimen was repeated 3 times at intervals corresponding to about 18 doubling times [
1]. We performed 3 cycles of treatment since these authors demonstrated that their exposure regimen resulted in the selection of (transiently)-resistant cell populations after only three cycles of treatment in two other colorectal cancer cell lines, HCT8 and HT29. Cell cultures that had recovered following each treatment cycle were frozen down for subsequent gene expression and growth inhibition analyses; these were labeled as HCT116 Bolus1, Bolus2, and Bolus3. Assessments of growth inhibition were done only for HCT116 Bolus3, since this derivative should in principle have developed the most resistance to 5-FU after three cycles of treatment. Gene expression levels were quantified for both HCT116 Bolus3 and Bolus1 cells; gene expression in the latter was used as a basis for comparison with the same in HCT116 Bolus3 cells. Untreated exponentially-growing parental HCT116 cells were used as controls.
Subsequent characterizations of all HCT116 derivative cell lines were done on exponentially-growing cells that had been grown without 5-FU in the media for at least one month, taking into account recommendations for
in vitro resistance work presented in a recent review [
2].
Assessment of resistance levels in HCT116 parental and treatment-derivative cell lines using growth inhibition assays
HCT116 parental and treatment-derivative cells that were maintained in regular medium without 5-FU were seeded in duplicate in 12-well plates and allowed to grow until they reached early exponential growth. Media was shifted 24 hours before the addition of 5-FU. Exponentially growing cells were incubated in the absence or presence of 8–10 different 5-FU concentrations for 24 hours, at which time the medium was aspirated off, the cells trypsinized and counted using a trypan blue viability assay. The drug concentration that resulted in a 50% growth inhibition (GI50) was determined graphically from sigmoidal dose-response curves of log-transformed dose values versus cell counts (percent of control) at 24 hours of continuous drug exposure. The resistance level was calculated as the ratio between the GI50 concentrations of the drug in the growth medium of the resistant and the sensitive cells, respectively. Two replicate experiments were performed.
Quantification of apoptotic response to 5-FU in HCT116 parental and treatment-derivative cell lines
Apoptotic response to transient 5-FU challenge was quantified in the parental and treatment-derivative HCT116 cell lines at 24 hours following 5-FU addition to the medium, using the TUNEL method for apoptosis detection. Different concentrations of 5-FU were added to the medium of exponentially-growing cells that had otherwise been maintained in 5-FU-free medium for one month. Apoptotic cells in control and treated samples were identified by end-labeling apoptotic DNA fragments with a fluorescent marker (streptavidin-FITC labeling of incorporated biotin-conjugated dUTP), mediated by terminal deoxynucleotidyl transferase (TdT) [
19]. Cells were counterstained with 5 μg/ml propidium iodide to stain cellular DNA. The resulting labeled apoptotic fractions were quantified in the control and treated cell samples using a FACSCalibur laser flow cytometer (BDIS, San Jose, CA), after appropriate gating using pulse-width analysis of the DNA content signal to exclude doublets and aggregates. Two replicate experiments were performed.
Oligonucleotide array analyses and quantification of gene expression levels
Harvesting of the cells for RNA extraction for comparative investigations of gene expression in the treatment-derivative and parental HCT116 cell lines was done on exponentially growing cells that were maintained in regular medium without 5-FU at the time of the actual microarray investigations. Analyses were done using Gene Chip Human Genome 8.5 K Focus Arrays (Affymetrix, Santa Clara, CA). Information about these arrays can be found at
http://www.affymetrix.com. RNA isolation, target labeling and hybridization to the arrays, array washing, staining, scanning and data analysis were performed using Affymetrix standard protocols, instrumentation, and Microarray Suite version 5.0 software. One replicate hybridization was performed for the ContinD derivative (same RNA sample, labeled anew and hybridized) in order to confirm the reproducibility of gene expression alterations between two same sample hybridizations.
The Microarray Suite software contains Detection, Signal, Change, and Signal log ratio algorithms that are used to evaluate individual gene expression for all gene probe sets (including controls) on the microarrays. These algorithms are described in detail in the Statistical Algorithms Description Document published by Affymetrix and available on their website. Briefly, the Detection algorithm assesses probe pair saturation, calculates a detection p-value, and assigns a present (P), marginal (M), or absent (A) call. The Signal algorithm calculates a quantitative metric for each probe set using the stray and real signal hybridization intensities, corresponding to Mismatch (MM) and Perfect Match (PM) minus MM, respectively. For comparison analysis of two samples (in the present study, a treatment derivative versus the parental cell line), a Change algorithm is used to generate a change p-value and an associated 'change', which indicates whether gene expression is increased (I), marginally-increased (MI), decreased (D), marginally-decreased (MD) or no change (NC) in a resistant derivative compared to the parental cell line. The Signal log ratio algorithm produces a quantitative estimate of these gene expression changes in the form of signal log2 ratios.
Correlation and cluster analyses for HCT116 derivative and parental cell lines were done for the following parameters: a) raw signal intensity values: these values are actual gene expression levels, and were compared for all genes in all cell lines (derivatives and parental) using Spearman correlation analysis; b) signal log2 ratios: these values are quantitative estimates of gene expression changes, and were compared using Spearman correlation analysis. Hierarchical cluster analyses for the same parameters were done using the Joining:Tree clustering method, Complete linkage, distance metric was Euclidean distances.
Real-time RT-PCR
Alterations of gene expression detected using Gene Chip arrays were confirmed by real-time RT-PCR for 4 genes,
TYMS, F3, FEN1, and
TOP2A, in ContinB and ContinD cells. These genes were chosen for RT-PCR analysis because they were shown to be strongly up- or down-regulated by microarray analysis in ContinB and/or ContinD cells [
3]. The
GAPDH gene was used as an internal control. Information about the cDNA-specific primers and procedure used for these investigations are available in the same study.
Statistical analyses
Correlation and hierarchical cluster analyses were performed using Statistica version 5.5 software (StatSoft Inc., Tulsa, OK). p-values ≤ 0.05 were considered to denote statistical significance.
Authors' contributions
PMD conceived the study, carried out one-half of the resistance development work, evaluated the microarray data, and did the statistical analyses. KLK carried out all maintenance cell culture work and one-half of the resistance development work. Both PMD and KLK performed the growth inhibition assays and did the RNA isolations necessary for microarray work. SHT performed the oligonucleotide microarray hybridizations under supervision from TH, and WR performed real-time RT-PCR. All authors read and approved the final version.