Stable Karyotypes in Epithelial Cancer Cell Lines Despite High Rates of Ongoing Structural and Numerical Chromosomal Instability

Abstract

Most human tumors and tumor cell lines exhibit numerical and structural chromosomal abnormalities. The goal of this study was to determine the ongoing rates of structural and numerical instability in selected cancer cell lines and to investigate the consequences of these rates to karyotypic progression. We studied two colorectal (HCT-116 and HT-29) and two ovarian (SKOV-3 and OVCAR-8) cancer cell lines and their single cell subclones. We found that the signature karyotypes of all four cell lines were distinct and each aberrant. Whereas high rates of ongoing structural and/or numerical chromosomal instability could be demonstrated in all cell lines, there was a relative stability of the consensus karyotype over many generations. No new clonal structural chromosomal reconfigurations emerged and the few numerical changes of karyotypes were restricted to abnormal chromosomes. This implies a kind of genomic optimization under the conditions of cell culture and suggests a link between genomic stabilization and cell propagation. We have been able to support this possibility by computer modeling. We did not observe a profound difference in the rates of numerical or structural instability in the cell lines with a replication error phenotype (RER+) versus the other cell lines.

Introduction

Cancer cells differ from normal cells in their genotypes. Genotypic differences are observed in three major forms: 1) aneuploidy, in which entire chromosomes are gained or lost; 2) structural chromosomal alterations, characterized by insertions, deletions, translocations, amplifications; 3) point or oligobase mutations of primary sequences. Genotypic changes in cancer cells are the consequence of genomic instability that appears to be fundamental characteristic of cancer, acquired at the early steps of tumor development [1–4]. The level of genomic instability cannot be determined by simply scoring the presence of genetic alterations in tumor cells because these alterations in different tumors may develop during different time intervals or with different rates of changes [5]. For instance, the presence of extensive numerical chromosomal changes in human epithelial cancers or cancer cell lines may indicate that these cells had chromosomal instability during some stages of carcinogenesis, but it does not necessarily follow that this instability persists in cancer cells during their continued growth. In a previous study [6], fluorescence in situ hybridization (FISH) was used to show that ongoing numerical chromosomal instability resulting in losses or gains of chromosomes persists in aneuploid colorectal cancer cell lines, suggesting a persistent defect in chromosomal segregation. In that study contrasting microsatellite unstable (MIN) and microsatellite stable but “chromosomally unstable” cell lines (CIN), the rate of numerical instability was found to be significantly higher in CIN cells than in MIN cells, and was estimated as approximately 10^-2 gains or losses per chromosome per generation.

Multiple rearrangements of chromosomal structure are also a very common manifestation of instability in epithelial tumors [7–10]. Cytoskeletal defects and breakage-fusion-bridges (BFB) cycles are thought to be a possible source of chromosomal instability and karyotypic heterogeneity in cancer cells [11,12]. Frequent generation of novel structural chromosomal rearrangements detected in subclones of human prostate cancer cell lines [13] led to suggestion that structural chromosomal instability, as well as numerical instability, could be an intrinsic characteristic of cancer cells.

The rate of ongoing structural chromosomal instability in human epithelial cancer cells remains to be elucidated, and in this study, we estimate the rate of structural instability in several human epithelial cancer cell lines, and compare the levels of ongoing structural and numerical chromosomal instability.

The existence of ongoing structural and numerical chromosomal instability, as well as the rate of numerical instability, was established on cancer cell lines grown in culture [6,12]. It is generally assumed that many cancer cell lines have relatively stable karyotypes over a long time of maintenance in steady-state culture conditions. If this assumption is correct and can be experimentally validated, such validation would lead us to an additional question: Is the rate of persisting chromosomal instability observed in cancer cells insufficient to produce karyotypic changes after many generations of cell line growth in culture, or are there selection forces working against the incorporation of new karyotypic rearrangements into the signature karyotypes?

We performed comprehensive characterization of single cell subclones of two colorectal (HCT-116 and HT-29) and two ovarian (SKOV-3 and OVCAR-8) cell lines, with subsequent comparison of karyotypes from early and late passages. We chose these particular cell lines not only to represent distinct tumor lineages, but also to represent RER+ (HCT-116, SKOV-3) and RER- (HT-29, OVCAR-8) phenotypes. Characterization included: 1) spectral karyotyping (SKY) to detect structural rearrangements; 2) FISH with a panel of centromeric probes to measure chromosome number changes; and 3) comparative genomic hybridization (CGH) to analyze total genome dosage alterations. In addition, we used computer simulations to estimate the consequences of the experimentally found rates of ongoing structural and numerical chromosomal instability on the karyotype of the cell population in the absence or presence of selection against new chromosomal aberrations.

Materials and Methods

Cell Lines and Clones

We used two colorectal and two ovarian human cancer cell lines: HCT-116, HT-29, SKOV-3, and OVCAR-8. These cell lines are in the NCI in vitro drug screening panel, and were obtained from the NCI cell repository (NCI-FCRDC, Frederick, MD). HCT-116 is a near-diploid cell line, SKOV-3 is near-tetraploid, and the others are grossly aneuploid. HCT-116 and SKOV-3 are defective in nucleotide mismatch recognition and repair, whereas HT-29 and OVCAR-8 are mismatch repair-competent [14,15].

Single cell clones of each of the four cell lines were established by dilution. Cells were diluted to deliver a single cell to only one of every three wells in a 96-well microtiter plate (Costar, Corning, NY). Four plates were used for each cell line with two independent dilutions, and the number of clones developed after 2 weeks was counted. Two clones for each cell line (called A and B subclones) were expanded through a defined number of generations (∼25 generations) to have enough cells for analysis (4x10⁷ cells). These single cell clone outgrowths were designated as “p1.” The p1 subclones were grown in culture for an additional 25 passages (∼40 generations) and designated as “p25.” The parental cell lines and the p1 and p25 subclones were subjected to further analysis by SKY, FISH with chromosome-specific centromeric probes, and CGH.

Metaphase specimens were prepared according to standard cytogenetic procedure.

SKY

The SKY hybridization protocol has been described in detail [16]. Chromosome-specific painting probes were generated in our laboratory from chromosome-specific template DNA (kindly provided by Dr. Thomas Ried) using two consecutive rounds of degenerate oligonucleotide-primed polymerase chain reaction (DOP-PCR). Chromosome labeling was performed by incorporating Rhodamine 110-dUTP (Perkin Elmer, Foster City, CA), Spectrum Orange-dUTP, Texas Red-dUTP (Molecular Probes, Eugene, OR), biotin-16-dUTP, and digoxigenin-11-dUTP (Boehringer Mannheim, Indianapolis, IN) in a secondary PCR reaction. Combinatorial fluorescence was produced by combining differentially labeled chromosome painting probes. The biotinylated probe sequences were visualized using Avidin-Cy5 (Amersham, Piscataway, NJ), and the digoxigenin-labeled probe sequences by incubation with mouse antidigoxigenin antibody (Sigma, St. Louis, MO) following sheep antimouse-antibody conjugated to Cy5.5 (Amersham).

Image acquisition was performed using a SD200 Spectracube (Applied Spectral Imaging, Carlsbad, CA) mounted on a Leica DMRXA microscope (Leica, Wetzlar, Germany) through a custom-designed optical filter (SKY v.3; Chroma Technology, Brattleboro, VT). Applied Spectral Imaging software (Spectral Imaging and SkyView) was used for image acquisition and analysis.

Ten metaphases were analyzed for each subclone and parental cell line. Results were reported using an abbreviated format of the International System for Human Cytogenetic Nomenclature (ISCN). Based on all clonal aberrations found in 10 analyzed metaphases, we created a composite karyotype for each cell line. Chromosomal aberrations were considered as clonal if found in two or more metaphases of the same cell line (in three or more metaphases for chromosome loss), according to ISCN conventions. Aberrations found in one metaphase only were designated as unique or nonclonal.

FISH with Chromosome-Specific Centromeric Probes

FISH with chromosome-specific centromeric probes was performed in order to assess cell-to-cell variations in chromosome number. Slides with interphase nuclei were prepared from subclones of HCT-116, HT-29, SKOV-3, and OVCAR-8 cell lines using a standard cytogenetic technique. Slides were aged for 3 to 7 days, incubated with 2x SSC, dehydrated in 70%, 90%, and 100% ethanol and denatured in 70% formamide/0.65 xSSC solution at 80°C for 2 minutes.

For chromosomes 1, 3, and 15, we used directly labeled CEP DNA FISH probes (Vysis, Downers Grove, IL) according to the manufacturer's protocol. Chromosome-specific α-satellite probes for chromosomes 7, 11, 17, and 18 were produced according to a previously described protocol [17]. All α-satellite probes were tested first on normal lymphocyte slides to assure the absence of cross-hybridization with other chromosomes.

To count the number of signals, we used a Leica DMRXA fluorescent microscope (Leica) equipped with DAPI filter, rhodamine filter, and dual-band pass filter (Chroma Technology) to visualize DAPI and rhodamine simultaneously. Image acquisition was performed using a Sensys CCD camera (Photometrics, Tucson, AZ) and Q-FISH software (Leica Microsystems Imaging Solutions, Cambridge, UK).

At least 100 nuclei were counted in each experiment. The result was presented as a percentage of nuclei with a number of α-satellite signals different from the modal number of signals for that particular chromosome. Enlarged polyploid nuclei with polysomic number of signals, which represented 1% of cells in HCT-116 and HT-29 cell line clones, 4% in SKOV-3, and 8% in OVCAR-8, were not included in the analysis of aneusomy.

The time elapsed between the preparation of the slide and the hybridization, and the denaturation temperature were critical parameters for obtaining consistent and reproducible results in evaluation of cell-to-cell chromosome number variability in cancer cell lines. Side-by-side FISH experiments with only a difference in temperature of denaturation (experiments were done with interphase spreads of an HT-29 clone and chromosome-specific centromeric probes for chromosomes 3, 11, and 17) revealed the range of signals off the mode to be 25% to 37% after denaturation of slides at 72°C compared to 6% to 14% after denaturation at 80°C (5-to 7-day-old slides). The same experiment with older slides (20 to 30 days) showed up to 45% to 64% of signals off the mode after denaturation of slides at 72°C.

CGH

CGH was performed according to a previously described procedure [18] on normal blood lymphocyte metaphases (Vysis). Whole genomic DNA was extracted from parental cell lines HCT-116, HT-29, SKOV-3, OVCAR-8, and their single cell subclones at p1 and p25 according to standard procedure [19]. Nick translation was performed on the test DNA from cancer cell lines with bio-16-dUTP (Boehringer Mannheim) and on the reference normal DNA (Promega, Madison, WI) with digoxigenin-11-dUTP (Boehringer Mannheim). The biotin-labeled tumor DNA was visualized with an avidin-conjugated to FITC (Vector Laboratories, Burlingame, CA), and the digoxigenin-labeled reference DNA with a mouse antidigoxigenin antibody (Sigma), followed by detection with a goat antimouse antibody conjugated to TRITC (Sigma). For image acquisition, we used Leica Q-FISH software (Leica Microsystems Imaging Solutions) interfaced to a Sensys CCD camera (Photometrics) and a Leica DMRXA microscope (Leica) that was equipped with three aligned optical filters for DAPI, FITC, and TRITC (Chroma Technology). At least 10 metaphase cells from each hybridization experiment were acquired and analyzed. CGH ratio profiles were calculated using Leica CGH software (Leica Microsystem Imaging Solutions).

Models and Computer Simulations

To assess possible explanations for the frequently unchanged composite karyotype between p1 and p25, we developed a computer program to simulate the generations of a cell line. The program simulates the division of cells and acquisition of chromosomal aberrations in discrete units of “generations.” We estimated that the time point called p1 occurs after 25 generations and the time point p25 occurs after 65 generations.

In its simplest usage, the program starts with a single cell having 50 chromosomes at generation 1. Each cell doubles into two at each generation and each chromosome of the new cell may become aberrant with probability a. When the number of cells exceeds 10,000, the program uniformly samples 1/10 of the cells and keeps that sample as representative for future generations. Given a collection of cells at some generation, the karyotype is determined by sampling 10 cells uniformly at random and recording a chromosome as “unchanged” if it is unaltered in at least 9/10 cells, and “aberrant” if it has the same aberration in at least 2/10 cells. For numerical instability, we made the simplifying assumption that for any chromosome number c, there could be only one type of aberration (in reality there could be two types of aberrations, a gain or a loss, but this makes no qualitative change in the numbers). For structural instability, aberrations are generated with unique identifiers, so that if an aberration identifier is later seen in two distinct cells, those aberrations are necessarily clonal. Because it is unlikely for a given chromosome to acquire two distinct aberrations during the simulation, only the first aberration, if any, is kept for each chromosome. To diminish bias in the small karyotype sample, the program actually computes 100 instances of the consensus karyotype by sampling for 10 cells 100 separate times. Sampling over the 100 instances is done with replacement. To decide if the composite karyotype remains stable from p1 to p25, the program compares the ith karyotype (i=1,2,...,100) after 25 generations to the ith karyotype after 65 generations.

The program records whether the karyotypes are different or not. Because the program samples the karyotype 100 times, each run gives a number between 0 and 100 reflecting the probability P that the karyotypes acquired clonal changes. By convention, a low number means that the karyotype is very unlikely to change, and a high number means that the karyotype is likely to change.

For each model considered, we ran the program 1000 times with different random seeds, so as to be very confident in our estimates of P.

The program includes various user-controlled parameters that allow us to simulate different models of cell lines and aberrations. We allowed for the possibility that at each generation, the cell division occurs with probability d<1 (attrition), regardless of the presence or absence of aberrations. We allowed for the possibility that at each generation, the probability that a cell survives is further multiplied by either 1/c^t, where 1/c is a constant defined as a growth penalty value for a new aberration and t is the total number of aberrations acquired from p1 onward, or 1/cⁿ, where n is the number of aberrations newly acquired at this generation.

In the Results section, we restrict attention to three main set-ups of the program:

• Model I: a=1/1000, and all cells survive and propagate.

• Model IIA: a=1/1000, attrition d=1/3, and aberrant cells have a penalty 1/2^t that is multiplicative on their propagation.

• Model IIB: a=1/1000, d=1/3, and cells with newly acquired aberrations have a penalty 1/10 on their propagation.

As we shall see in the Results section, these set-ups help us define relevant parameters for yielding the experimentally obtained results.

TUNEL assay

For detection and quantification of apoptosis at the single cell level, we used In Situ Cell Death Detection Kit (Boehringer Mannheim).

Statistical analysis

The data are expressed as mean±SD. Student's t-test was performed to determine statistical differences between groups. Differences were considered significant at P<.05.

Results

Karyotypic Abnormalities of Parental Cell Lines HCT-116, HT-29, SKOV-3, and OVCAR-8

Spectral karyotypes and CGH profiles determined for two colorectal (HCT-116 and HT-29) and two ovarian (SKOV-3 and OVCAR-8) cancer cell lines showed that structural and numerical chromosomal rearrangements were present in all four cell lines. Based on spectral karyotypes, for each cell line we counted the total number of chromosomes per metaphase and the numbers of abnormal chromosomes with clonal and nonclonal aberrations. (Structural chromosomal rearrangements found in two or more metaphases of the same cell line were designated as clonal, and those found only once were designated as unique, nonclonal aberrations. ) Genome dosage changes were counted as numbers of gains and losses of whole chromosomes or any chromosome subregions revealed by CGH in each cell line. Ploidy status and clonal chromosomal aberrations in these cell lines are summarized in Table 1.

Table 1.

Summary of Ploidy and Total Number of Clonal Chromosomal Aberrations in Parental Epithelial Cancer Cell Lines.

Cell Line	Origin	Ploidy	Clonal Chromosomal Aberrations
			Structural Rearrangements (SKY, Number of Abnormal Chromosomes)	Genome Dosage Changes (CGH, Number of Gains and Losses)
HCT-116	Colorectal carcinoma	near-diploid	5	5
HT-29	Colorectal carcinoma	near-triploid	16	19
SKOV-3	Ovarian carcinoma	near-tetraploid	28	12
OVCAR-8	Ovarian carcinoma	hyperdiploid	40	34

Cell lines had remarkable differences in their signature karyotype. The karyotype was relatively normal in the male-derived HCT-116 cell line (Table 1 and Figure 1A), and was basically the same as reported in other studies [20,21]. The only difference was the presence of a balanced translocation involving chromosomes 4 and 17 in 4 out 10 analyzed metaphases. Chromosome Y was missing in all cells analyzed, and clonal structural rearrangements involving chromosomes 8, 10, 16, 17, and 18 were present.

Figure 1.

SKY, CGH, and FISH with centromere-specific probes of the cell line HCT-116, subclones A and B. (A) Karyotype of the HCT-116 cell line, subclone B, in classification colors. The only karyotypic difference between subclones A and B was the presence in the B of a balanced translocation involving chromosome 4 and der(18)t(17;18). (B) The average CGH ratio profiles for the near-diploid cell line HCT-116, subclone B. Note that all aberrations detected by SKY were also seen by CGH analysis, except for a balanced translocation between chromosomes 4 and der(18)t(17;18), and a loss of chromosome Y (normal female metaphase spreads were used for CGH experiments). CGH profiles for parental cell line HCT-116 and subclones A and B were identical. (C) Interphase FISH with chromosome-specific centromeric probes for chromosome 1 and 11 (HCT-116, subclone A). Arrowheads show nuclei with number of signals different from the modal number of chromosomes for that specific probe. Note that the modal number for chromosomes 1 and 11 was two for all three methods of analysis (SKY, CGH, and FISH).

HT-29 had more structural and numerical abnormalities compared to HCT-116 (Table 1 and Figure 2A). This neartriploid cell line was grossly aneuploid with aneusomies of chromosomes 5, 7, 11, 13, 14, 15, 19, 20, and 21, and structurally rearranged chromosomes 2, 3, 4, 5, 6, 8, 13, 14, 17, 18, 19, 20, 22, and X. Most of chromosomal rearrangements found in HT-29 cell line were similar to those previously described [21,22]. Cell line SKOV-3 was near-tetraploid with three copies of chromosome 7, 13, and 14. Clonal structural aberrations involved chromosomes 1, 2, 3, 5, 9, 10, 11, 12, 14, 15, 17, 18, 19, 20, and X (Table 1 and Figure 3). Hyperdiploid cell line OVCAR-8 had the most heavily rearranged karyotype (Table 1 and Figure 4A), where 40 rearranged chromosomes were clonally present, of which 17 had complex rearrangements involving more than two chromosomes.

Figure 2.

SKY, CGH, and FISH with centromere-specific probes of cell line HT-29, subclone B. (A) Karyotype of the near-triploid HT-29 cell line, subclone B, in classification colors. The karyotype of this subclone was slightly different from the parental cell line. Three copies of chromosome 7 were present in the subclone, whereas the parental cell line had four copies. (B) The average CGH ratio profiles for the cell line HT-29, subclone B. Note that all aberrations detected by SKY were also seen by CGH analysis except a balanced translocation (6;14). (C) Interphase FISH with chromosome-specific centromeric probes for chromosomes 1, 11, and 17. Arrowheads show nuclei with number of signals different from the modal number of chromosomes for that specific probe. Note that the modal number for chromosomes 1 and 17 was three, and for chromosome 11 was four according to all three methods of analysis (SKY, CGH, and FISH). Also, the modal numbers for all other chromosomes studied in the HT-29, subclone B, were in complete agreement with FISH, SKY, and CGH.

Figure 3.

Spectral karyotype of the SKOV-3 parental cell line in display colors (A) and classification colors (B). Nonclonal chromosomal rearrangements (found in this cell only) are labeled with an asterisk (*).

Figure 4.

(A) Spectral karyotype of the OVCAR-8 cell line in display colors. Chromosomes frequently undergoing additional rearrangements are indicated with arrows. (B) Examples of nonclonal variations of a clonal aberration. Nonclonal variants were collected from SKY-analyzed cells of the OVCAR-8 parental cell line and subclones.

All parental cell lines exhibited karyotypic heterogeneity (cell-to-cell differences in karyotypes). The level of structural chromosomal heterogeneity was relatively low in HT-29 and HCT-116 cells. Only 0.1 nonclonal structural rearrangements per cell were found in each colorectal cell line, whereas in ovarian cancer cell lines higher average numbers of 1.6 (SKOV-3) and 1.8 (OVCAR-8) nonclonal structural aberrations per cell were seen.

Persistent Structural Chromosomal Instability in Colorectal and Ovarian Cancer Subclones

After single cell subcloning of HCT-116, HT-29, OVCAR-8, and SKOV-3 cell lines, two clones of each cell line were propagated in culture and analyzed when approximately 4x10⁷ cells had been generated (we defined this stage as “p1” clones).

As anticipated based on the karyotypic heterogeneity of the parental cell lines (see above), some of the single cell subclones showed variant karyotypes. HCT-116, subclone A, inherited the karyotype of the main clone of parental cell line. HCT-116, subclone B, inherited the same karyotype, but with additional balanced rearrangement involving chromosomes 4 and 17 (Figure 1A). Loss of an abnormal chromosome 7 was found in both p1 single cell subclones of HT-29 compared to the parental cell line (Figure 2A). OVCAR-8, subclone A, had the same karyotype as the parental cell line. In OVCAR-8, subclone B, we noted a gain of 5p14-q14 and a loss of 5q15-qter regions. SKOV-3, subclone A, showed a loss of chromosome 7 and gains of a dicentric der(5;10)t(5;10)(q35; p13)t(5;13)(p12;q21.1) and a third copy of del(12)(q22). In addition, der(19)t(11;19)(q11;q11) was missing, but a der(19;22)(p10;q10) was present. In subclone B, the loss of 14q12-qter was the only difference from the parental cell line.

As in our analysis of the parental cell lines, we found in their single cell subclones that the level of cell-to-cell karyotypic heterogeneity was higher in the ovarian compared to the colorectal cancer cell lines. SKY of p1 subclones revealed heterogeneity with the frequency of novel nonclonal aberrations varying from 0.1 per cell in the HT-29 subclone to 1.6 per cell in the OVCAR-8 subclone (Table 2). Because all characterized subclones were grown in culture in similar conditions for the same number of generations, we were able to compare directly the rates of occurrence of new structural rearrangements. For colorectal cancer cell lines, the rate was approximately 10^-4 per chromosome per generation. For the ovarian cancer cell lines, it was close to 10^-3 or approximately 10-fold higher than in the colorectal cancer cell lines. Assuming that every chromosome has an equal chance to be rearranged in any cell, this means that among each 20 ovarian cancer cells or among each 200 colorectal cancer cells in any given generation at least one new nonclonal rearrangement would occur.

Table 2.

Average Number of Novel Nonclonal Structural Aberrations Per Cell in Single Cell Subclones.

Subclone/Passage	HCT-116	HT-29	SKOV-3	OVCAR-8
A/p1	0.1	0.2	0.8	1.3
A/p25	0.4	0.2	1.2	1.6
B/p1	0.3	0.1	ND	1.6
B/p25	0.2	0.2	ND	1.5

ND: not determined.

However, not all chromosomes, in fact, participated equally in new rearrangements. In the metaphases that were analyzed, e.g., in the OVCAR-8 cell line, 11 chromosomes underwent additional rearrangements frequently (Figure 4A and B), 23 chromosomes were additionally rearranged only once, and 19 chromosomes did not participate in nonclonal rearrangements at all.

After 25 additional passages in culture, karyotypes of clones were analyzed again and compared to p1. We did not find an increase in the average number of novel nonclonal aberrations at p25 compared to p1 (Table 2).

Ongoing Numerical Chromosomal Instability

We then examined the rate of numerical instability in these cell lines. FISH of interphase cells with a panel of centromeric chromosome-specific probes performed on subclones of the four cell lines demonstrated the existence of cell-to-cell variability in all examined subclones. Variability was evaluated as a percent of nuclei with a number of α-signals different from the modal number. The modal number of signals determined for each chromosome in the interphase FISH experiment (Table 3, numbers in brackets) agrees perfectly with the modal number of the same chromosome found by SKY, and with chromosome copy number deviations from the ploidy value detected by CGH (Figures 1 and 2).

Table 3.

Numerical Chromosomal Instability in Subclones of Cancer Cell Lines.

Cell Line/Subclone/Passage	Chromosome 1	Chromosome 3	Chromosome 7	Chromosome 11	Chromosome 15	Chromosome 17	Chromosome 18	Average (%)
HT-29/B/p1	(3) 7%	(3) 6%	(3) 18%	(4) 14%	ND	(3) 12%	(3) 11%	11.3±4.1
HT-29/B/p25	ND	(3) 7%	(3) 7%	(4) 12%	(4) 14%	(3) 12%	ND	10.4±2.9
HCT-116/A/p1	(2) 6%	(2) 6%	(2) 8%	(2) 8%	(2) 8%	(2) 10%	(2) 9%	7.9±1.5
HCT-116/A/p25	(2) 11%	ND	(2) 6%	(2) 7%	(2) 8%	(2) 8%	ND	8.0±1.9*
OVCAR-8/B/p1	(5) ND	(3) 16%	(2) 6%	(3) 10%	(2) 18%	(2) 13%	ND	12.6±4.8
OVCAR-8/B/p25	(5) ND	(3) 11%	(2) 9%	(3) 7%	(2) 13%	(2) 5%	ND	9.0±3.2
SKOV-3/A/p1	(4) 9%	(4) 8%	(2) 6%	(4) 13%	ND	(4) 9%	(4) 18%	10.5±4.3
SKOV-3/A/p25	(4) 9%	(4) 14%	(2) 10%	(4) 19%	(4) 17%	(4) 19%	(4) 20%	15.4±4.5*
Normal lymphocytes	(2) 2%	(2) 3%	ND	(2) 2%	(2) 3%	ND	(2) 3%	2.6±0.5

Interphase FISH with a panel of chromosome-specific centromeric probes was performed on subclones of epithelial cancer cell lines. For each chromosome tested, the modal number of signals was determined (numbers in brackets). The fraction of cells with a number of signals different from the mode is shown. At least 100 nuclei were evaluated per clone with each chromosome probe.

ND: not determined.

^*Statistically significant difference (P<.001) was found among HCT-116 subclone, passage 25, and SKOV-3 subclone, passage 25.

The level of variability of α-satellite signals for different chromosomes in HCT-116 was in the range of 6% to 11%, with an average of 7.9%. It was higher, from 6% to 20% for HT-29, SKOV-3, and OVCAR-8, with an average of 11.3%, 10.5%, and 12.6%, respectively (Table 3). The differences in the variability of α-satellite signals between cell lines at p1 were not statistically significant.

The average variability of α-satellite signals for normal lymphocytes was 2.6% (Table 3). This was significantly (P<.001) lower when compared with the average variability seen in each of the cancer cell lines.

Taking into account the number of chromosomes and cell generations, the rate of chromosome gains and losses was estimated from interphase FISH as ∼10^-3 per chromosome per generation for all four cell lines (3.5x10^-3 for HT-29, 2x10^-3 for HCT-116, 4x10^-3 for OVCAR-8, 3x10^-3 for SKOV-3). That means that for a cell line with the modal number of 50 chromosomes, one would expect to find one gain or loss of a chromosome for each 20-cell sample in any given generation.

The variability of centromere numbers was also obtained by counting the number of copies of each centromere (in normal and rearranged copies of chromosome) in each of metaphases of the same clones analyzed by SKY. Percentage of metaphases that have deviations from the modal centromere number was averaged over all centromeres (as we did for interphase FISH with centromere-specific probes). The rate of chromosome gains and losses estimated by this method, which can produce less accurate results due to mechanical losses of chromosomes, was basically at the same level of ∼10^-3 per chromosome per generation for all four cell lines.

Interphase FISH analysis with centromeric probes was again performed on the same cell lines at p25. There was no significant difference in the numerical instability between p1 and p25 for any of the cell lines. An increase in numerical instability (average 15.4±4.5%) was seen in the SKOV-3 cell line, which was significantly different only from the variability seen in p25 of the HCT-116 cell line (Table 3).

The level of numerical instability in HT-29 clones was lower than previously reported [6]. In that report, the average percentages of centromere signals different from the modal number were 46% and 51% for two HT-29 clones analyzed. To rule out the possibility that the HT-29 cell line analyzed here was different than in that previous study, we performed SKY of two clones of HT-29 described in the previous publication (kindly provided by C. Lengauer). Karyotypes of these two clones of HT-29 were found to be identical to HT-29 clones A and B in our study, except for a del(6)(q11), which was only found in our clones. In a FISH experiment with a panel of centromeric chromosome-specific probes, the average number of abnormal signals was 12.4±5.3% for one clone, and 13.3±3.5% for another, essentially the same as the values found for our clones of HT-29 in this experiment (Table 3). Possible explanations for the discrepancy with the previous publication could be different temperatures of denaturation and/or the age of the slides used in the analysis. After changing the temperature of denaturation from 80°C to 72°C and using 3-week-old slides, we found an increase in the variability of signals up to 45% to 64% in our HT-29 clones, imitating the values of the previous report.

Stability of the Consensus Karyotype1 of Each Subclone Over Many Generations

While we found ample evidence of persistent structural and numerical instability, the composite SKY karyotypes of subclones were relatively stable and consistent, especially in colorectal cell lines, which was confirmed by CGH analysis. Only two of eight subclones displayed clonal differences in their own karyotypes after propagation of each subclone for 25 passages (approximately 40 generations) in culture. Subclone A, SKOV-3, lost dicentric derivative of chromosomes 5 and 10. Subclone B, OVCAR-8, lost der(5) and der(18)t(10;18), but gained a second copy of der(3)t(3;8). All clonal changes in karyotypes detected by SKY at the single metaphase cell level were confirmed by CGH analysis of genome dosage changes at the population level.

Models and Computer Simulations

To help define the parameters relating ongoing chromosomal instability and its effect on the karyotypes, models were developed and tested using computer simulation techniques (Figure 5).

Figure 5.

The scheme for computer simulations.

Model I is based on assumptions that cells acquire chromosomal changes with the rate of 10^-3 per chromosome per generation (this was the highest rate we observed, and was the rate determined for both OVCAR-8 and SKOV-3), and all cells have an equal chance of surviving regardless of the number of new aberrations. After running simulations of the first model, it was found that typically about 65 of 100 trials yielded karyotypes with new clonal configurations of chromosomes. The probability of seeing unchanged karyotypes in four independent experiments (subclones) would be (1-0.65) raised to the fourth power, which equals 0.015 (less then a 1.5% chance). This result is not consistent with our findings.

Model II assumes the same rate of chromosomal changes as the first one (10^-3 per chromosome per generation). We stipulated an attrition factor modulating cell growth. We did that because at the time of single cell subcloning, we had diluted the cells such as to deliver a single cell to only one of every three wells in a 96-well microtiter plate. We had therefore anticipated generating about 32 subclones per plate. For OVCAR-8 and SKOV-3 cell lines, we repeatedly generated only about 33% of the anticipated subclone number (about 10 clones per plate) for the given time of culture growth before subclone harvest (about 2 weeks). Therefore, not all the cells in the parental culture were equally capable of propagation as single cells. The attrition rate stipulated in model II was 1/3 based on these results. A parallel experiment performed within our laboratory used these same cell lines as substrates for clonogenic assays in which multiple clones were grown together on the same plate in the same media. Here again, the SKOV-3 and OVCAR-8 cell lines were less than 50% clonogenic than anticipated strictly on the basis of number of cells plated (O. Glebov, unpublished observations). This observed problem in cell growth might have been due to cellular death or, more generally, to differences in propagation ability of cells, which nevertheless are viable. A TUNEL assay on these exponentially growing cell lines (see Materials and Methods section) revealed about 2% of apoptotic cells, suggesting that the pathway of programmed cell death was not the basis of the decreased clonogenicity that we observed.

We also stipulated that new aberrations be deleterious to cell division because we observed that the single cell subclones of the ovarian cell lines that emerged showed on average 0.75 new clonal aberrations per cell despite the fact that our analyses of metaphase cells from the parental cell lines revealed on average 1.7 new aberrant chromosomes per cell. If everyone of these cells had been equally capable of developing into a single cell subclone, we would have expected to see more clonal variations among the subclones that developed and that we called “p1.” Therefore, there appears to be a growth “penalty” for the acquisition of new chromosomal aberrations. We considered two types of penalty, one in which the penalty compounded with each new aberration that occurred, the other where, if a cell “survived” the acquisition of an aberration, only the next new aberration would be penalized. Penalty values were set at 1/2^t in model IIA (the “compounded” example; see Materials and Methods section). In subsequent model II variations, we changed both the magnitude of the penalty (1/2 or 1/10) and whether the penalty was “compounded” or only restricted to the next “new” aberration n (see Materials and Methods section).

Simulations using the criteria of model IIA supported the notion that penalizing the acquisition of new aberrations increased the probability of observing unchanged consensus karyotypes after propagation of cells for 40 generations. In a subsequent simulation (model IIB) using the same mutation (10^-3) and attrition (1/3) rates but making the penalty 1/10 for the next new aberration, the probability of seeing no numerical chromosomal changes among the eight “p25” subclones analyzed was (0.53)⁸=0.006. Therefore, we should anticipate seeing occasional numerical changes in composite karyotypes and, in fact, we did. We saw numerical changes in two of eight subclones. In contrast, we saw no structural changes in any of the four p25 ovarian subclones analyzed. Under the conditions of the simulation, the probability of that is 0.2 (20% chance). Therefore, the penalty imposed in this particular simulation, while only one of many possibilities, is consistent with the observed experimental data.

Discussion

Our results support the view that chromosomal instability can be a persistent characteristic of cancer cells [5,6,13]. We found that not only numerical chromosomal instability persists in cancer cells, but also ongoing structural instability was present in colorectal and ovarian cancer cell lines.

We have grouped genomic translocations, deletions, duplications, amplifications, inversions, and insertions, all as structural chromosomal rearrangements; however, these processes are distinct and their underlying mechanisms could be different and very complex.

The level of numerical instability was approximately the same (roughly 10^-3 per chromosome per generation) for each of the colorectal or ovarian cancer cell lines studied. In contrast, the level of structural chromosomal instability in the ovarian cell lines (10^-3 per chromosome per generation) was nearly 10 times higher than in the colorectal cell lines (10^-4 per chromosome per generation). Therefore, rates of structural and numerical chromosomal instability do not necessarily correlate in cancer cells. The existence of cancers with predominantly structural or numerical aberrations supports this conclusion [7,8,10]. However, the majority of karyotypically rearranged epithelial cancers or cancer cell lines have both extensive structural and numerical chromosomal alterations.

Despite persistence of structural and numerical chromosomal instability in human epithelial cancer cell lines, the consequences of ongoing instability at the population level were notable: only two of four ovarian subclones and none of four colorectal subclones displayed differences in consensus karyotypes after propagation of each subclone for 25 passages (approximately 40 generations) in culture and the clonal changes observed in the ovarian cancer cell lines were mainly reductions in the genome and not new structural configurations. To interpret these data, we developed models and tested them using computer simulation. With the assumptions (obtained from analysis of chromosomal instability in ovarian carcinoma cell lines OVCAR-8 and SKOV-3) that the rate of structural instability was 10^-3 per chromosome per generation, and the ability of cells to divide was decreased after acquisition of new aberrant chromosomes, simulations predicted relative stability of the structural futures of the karyotype over time.

Many plausible settings of the parameters (mutation rate, attrition, and penalty values) of our simulation program (data not shown) gave the same qualitative result of a low probability of change in the composite karyotype between generations 25 and 65. Thus, our results are mathematically plausible, but we are not claiming to define the specific parameters of cell division and acquisition of new aberrations that may precisely obtain in all cell lines. While we tested many settings of parameters, we focused on models that incorporated our experimentally derived mutation and attrition values; thus, the major variable became the penalty applied to a cell that acquires additional aberrations. For simplicity of analysis, we stipulated that the penalty for a newly acquired aberration would be charged either cumulatively for every aberration acquired by the cell or for only the next new aberration acquired by the cell. However, there is no reason why some combination of these two charges could not occur. Furthermore, the penalty for a newly acquired numerical aberration need not be the same as for a newly acquired structural aberration, nor must all numerical and structural aberrations have equivalent penalties. In addition, it is quite plausible that some (rare) chromosomal changes can provide a selective advantage to growth in cell culture. Despite these qualifications, our observations are most consistent with some sort of growth penalty being charged for the majority of cells that acquired new aberrations.

Thus, the data and simulations suggest that in steady-state conditions of cell culture, the emergence of new clonal chromosomal rearrangements occurs very slowly as long as the environment remains constant. It can be documented that the karyotypes of some well-studied epithelial cancer cell lines have remained relatively unchanged after years of passaging in culture in different laboratories [23,24]. This, again, implies a selection against the emergence of new structural rearrangements under steady-state conditions. If selection factors are applied, the chromosomal instability of the individual cancer cells, as manifested by the observed karyotypic heterogeneity, may provide a mechanism for the generation of clonal karyotypic variants that have a selective advantage. An example of karyotype changes under selection in culture is described in a study of the MCF-7 cell line — karyotypically stable for many generations until a selection for mitoxantrone resistance was applied [25]. All three independent MCF-7 mitoxantrone-resistant derivative sublines demonstrated new structural clonal rearrangements: translocations involving chromosome 4 and amplifications of 4q21-q22.

Our data also suggest a selection against the emergence of clonal numerical changes in the karyotype. We saw clonal chromosome losses and gains only in two ovarian subclones among four ovarian and four colorectal subclones studied, despite the presence of numerical chromosomal instability in all of them.

We found that numerical and structural chromosomal instability were present in all four cell lines despite the fact that two cell lines (HCT-116 and SKOV-3) were mismatch repair-defective and exhibit RER+ phenotype. Genome destabilization, therefore, exists in RER+ cells not only at the single nucleotide level, but involves chromosome structure and number as well. Karyotypic characterization of RER+ cell lines [21,26] and tumors [27] showed that RER+ cancer cells always have karyotypic abnormalities. Therefore, three kinds of instability — structural chromosomal instability, numerical chromosomal instability, and microsatellite instability — can coexist in RER+ cancer cells.

What then could be the explanation for the signature karyotypic differences that generally distinguish these two groups of tumors (in general, RER+ cells are near-diploid and RER- are grossly aneuploid), if chromosomal instability is present in both of them? For example, the HT-29 cell line is grossly aneuploid with 16 abnormal clonal chromosomal reconfigurations, whereas HCT-116 is near-diploid with only five clonal structural aberrations. The rate of structural chromosomal instability in these cell lines was approximately the same, and the rate of numerical chromosomal instability was less than two-fold higher in HT-29 compared to HCT-116. In our computer simulations, a two-fold difference in the rate of structural or numerical instability did not alter the outcome of predicted karyotypic progression (data not shown). Thus, the rate of instability by itself does not seem to account for karyotypic difference, and therefore, one might consider whether it is the length of the time over which the rate has been present or the selective pressures to which the cells were subjected during tumor development that might, more likely, explain the difference in karyotypic pattern. It is also possible that the types of numerical and structural instability that we have measured here are not the major mechanisms that cause the karyotypic signature difference. If some tumors passed through a time of “cataclysmic” karyotypic destabilization, which somehow other tumors were able to bypass, this could explain the general karyotypic dichotomy that is observed. For example, catastrophic mitoses brought about by telomere dysfunction during growth crisis [28,29] could be such a mechanism in grossly aneuploid tumor development. Restoration of telomere function stabilizes telomere ends and thus abrogates this cause of karyotypic lability, which may allow crisis resolution and continued tumor growth. Mismatch repair-defective cells may be capable of bypassing this crisis by allowing telomeres to be replenished even in the absence of telomerase [30,31]. Distinguishing between near-diploid versus grossly aneuploid tumors is possible at a relatively early time in tumorigenesis. It has been reported that adenomatous polyps in the colon (or Barrett's esophagus lesions as well) can already demonstrate gross aneuploidy [32–34].

In summary, we found evidence of dramatic structural and numerical chromosomal instability having occurred in the past history of some of these cell lines. We found clear evidence of ongoing structural and numerical instability as well, although that instability was not particularly distinguishable based on mismatch repair status. Inspite all of these evidence of previous and persistent instability, there was relative stability of the karyotype at the population level of each single cell subclone over many generations of continuous culture.

Abbreviations

RER

replication error

SKY

spectral karyotyping

CGH

comparative genomic hybridization

del

deleted chromosome

der

derivative chromosome

TUNEL

TdT-mediated dUTP-X nick end labeling Composite karyotype — a karyotype with all clonal aberrations listed Consensus karyotype — a composite karyotype, which includes clonal aberrations detected by SKY, and where the gains and the losses of genetic material are also confirmed by CGH