The LNCaP and DU145 cell lines were originally isolated from metastatic lesions of patients suffering from prostate cancer (Stone et al, 1978; Horoszewicz et al, 1983) and purchased from DSMZ (Braunschweig, Germany) as American Type Culture Collection (ATCC) cell lines. HeLa cells were derived from a cervical carcinoma (Scherer et al, 1953) obtained from Dr Aubee. The T47D-B8 and MCF-7 cell lines descended from human breast carcinomas (Soule et al, 1973; Soto et al, 1986; Ruiz de Almodovar et al, 1994). The MCF-7 subclones BB and Bus were described to differ from MCF-7 parental cells with respect to dsb induction and repair, p53 status, cell cycle and apoptosis (Nunez et al, 1995; Siles et al, 1995). These cells as well as the RT112 bladder carcinoma cell line (Masters et al, 1986) were gifted by Dr Ruiz de Almodevar. SCC4451 cells were established from a squamous cell carcinoma of the head and neck obtained from Dr Zölzer (Zölzer et al, 1995). All cells were either grown in RMPI 1640 or DMEM (Life Science Technology/BRL, Karlsruhe, Germany) supplemented with 10% FCS and penicillin/streptomycin in 5% CO₂.

Fibroblast cells, HF-1, -2, -7, -8, -46, -60 and HF, were established from punch biopsies as described elsewhere (Borgmann et al, 2002). CHO cells were kept in α-MEM medium. The experiments were performed with nearly confluent tumour cells and confluent fibroblasts.

Tumour cell survival was assessed by colony formation assay. Near-confluent cultures were X-irradiated at 220 kVp, at a dose-rate of 2 Gy min⁻¹. Irradiated cells were immediately plated and grown for 2–3 weeks with one medium change. Stained colonies of more than 50 cells were counted. Experiments were repeated thrice with three replicates each.

Subconfluent cells were suspended (3 × 10⁶ ml in 0.8% low melting point agarose (Bio-Rad, Munich, Germany) and pipetted into 180 μl plug moulds to solidify (Dahm-Daphi and Dikomey, 1995). Those agarose plugs were irradiated on ice to prevent DNA repair during irradiation and thereafter directly subjected to cell lysis (0.4 M EDTA, 2% N-lauryl sarcosine, and 1 mg ml⁻¹ proteinase K, all Sigma, Deisenhofen, Germany). Lysis was started on ice for 30 min and continued at 37°C overnight. The plugs were washed thrice with Tris-EDTA buffer and sliced into pieces containing about 10⁵ cells, which were inserted into a 14 × 20 cm 0.8% agarose gel (High-grade ultrapure, Bio-Rad). The gel was then covered with a thin overlayer of 0.8% agarose to avoid light fraction artefacts upon optical imaging. Electrophoresis was performed at 0.6 V cm⁻¹ for 30 h in 0.5 × TBE buffer (45 mM Tris base, 45 mM boric acid, 2 mM EDTA) in a conventional apparatus (Subcell, Bio-Rad). The gel was then stained overnight (0.5 μg ml⁻¹ ethidium bromide), destained overnight in ddH₂O and CCD camera (Sony XC-75CE) equipped with an image analysis system (Optimas, Silverspring, MD, USA) was used to quantify the FDR.

Cells were treated as before but electrophoresis was performed differently. Running conditions were now 0.6 V cm⁻¹ for 30 h followed by 1.5 V cm⁻¹ for 6 h. Compared to the regimen published previously (Dahm-Daphi and Dikomey, 1995) the present protocol was restricted to two different field strengths sufficient for the current purpose. Detailed analysis and mathematics were described elsewhere (Dahm-Daphi and Dikomey, 1995). In principle, analysis was based on an equation given by Blöcher (1990) describing the sigmoid curve of the total FDR obtained by PFGE. Here, the total FDR corresponds to the sum of the FDR of the two bands (GFGE at 0.6 and 1.5 V cm⁻¹, Figure 2). The curves of total FDR and FDR at 1.5 V cm⁻¹ were both sigmoid (Figure 4) and could, hence, be fitted by the equation described by Blöcher (1990). The third curve (band 1) corresponds to the difference between both fits. The χ² fits were calculated simultaneously and gave the number of dsb in unirradiated cells, of dsb induced per DNA-unit, the maximum DNA fragment size in each band and the factor of retention (f_ret).

Cell cycle distribution of PI-stained cells was determined by flow cytometry in an FACScan (Becton Dickenson, Heidelberg, Germany).

Each experiment was repeated three times and the data were given as a mean with its standard error (±s.e.m.). Statistical analysis, data fitting and graphics were performed by means of the Prism 3.1 computer program (GraphPad Software, San Diego, USA).

For the nine tumuor cell lines tested, the number of dsb induced was found to vary by a factor of 2 from 5.75 to 11.0 × 10⁻¹² dsb/Gy/Da (Table 2), which in principle agreed with most previous studies (Kelland et al, 1988; McMillan et al, 1990; Schwartz et al, 1990, 1991; Ruiz de Almodovar et al, 1994; Zaffaroni et al, 1994; Whitaker et al, 1995; Woudstra et al, 1998; Eastham et al, 2001) using either PFGE or neutral filter elution. Some authors did not mention differences in particular; however, variations were in the same order of magnitude as the above reports (Giaccia et al, 1992; Olive et al, 1994; McKay and Kefford, 1995). There was only one report (Olive et al, 1994) that found differences among six tumour cell lines using neutral filter elution but factually not by PFGE. Our result is important for three reasons. Firstly, the technique used is superior to others (see below) and, second, the variation in damage induction appears to be a property of tumour, but not of normal cells (see below). Most importantly, for eight out of the nine tumour cell lines the variation in the frequency of induced dsb showed a significant correlation with the respective variation in radiosensitivity (Figure 7A). Cell lines with a high number of dsb induced found to be much more sensitive than cell lines with a low number of induced dsb. This was similarly found by Ruiz de Almodovar et al (McMillan et al, 1990; Ruiz de Almodovar et al, 1994; Whitaker et al, 1995), other studies showed an insignificant trend (Schwartz et al, 1988; Giaccia et al, 1992; Zaffaroni et al, 1994; McKay and Kefford, 1995; Woudstra et al, 1998). In our study, one (LNCaP, Figure 7A) out of the nine cell lines fell off the general relationship between initial damage and cell survival indicating that the cellular radiosensitivity is eventually not only determined by the number of induced dsb but also by other still unknown factors.

The variations in the dsb induction frequency found for tumour cells are most likely due to different chromatin structures. High condensation of chromatin structure and tight DNA–protein association should efficiently protect from oxygen radical attack to DNA, and vice versa. Such structural variations among tumour cells have in fact been shown by means of the Halo-assay (Schwartz and Vaughan, 1989; Lynch et al, 1991; Woudstra et al, 1998). However, direct evidence for chromatin structure being responsible for differences in the dsb induction among tumour cells is lacking yet. It should be noted that the large differences in the dsb induction observed are mainly due to the three extreme values. This may indicate that not all tumour cell lines have an altered chromatin structure.

Contrary to tumour cells, the seven normal fibroblasts tested showed only scarce variation in the number of dsb induced, which confirmed previous results of five and twelve fibroblast lines (Wurm et al, 1994; Dikomey et al, 2000). Taken all together, it can be assumed that the number of dsb induced varies substantially in tumour cells, but not in normal human cells. The latter might reflect the general interindividual stability of the human genome.

Most measurements of dsb were performed with neutral filter elution, PFGE or CFGE, which all rely on quantification of overall fragments released from the bulk DNA. The newly applied GFGE has now the advantage to allow the direct determination of the number of dsb. It is shown here, for the first time, that this number was in fact correlated with the initial slope (up to 20 Gy) of the FDR curve of CFGE. However, this does not mean that the initial slope can reliably taken as an indicator for the amount of initial damage. In many cases, the variation in the initial slopes of FDR curves may not be large enough to reveal significant differences between the cell lines. The initial slope further depend on S-phase cells and on the retention factor f_ret. High fractions of replicating cells and high retention values decrease the slope and apparently the number of induced dsb. This might well be the reason why only three out of 13 studies found a relationship between induced damage and tumour cell radiosensitivity (Kelland et al, 1988; Schwartz et al, 1988, 1990, 1991; McMillan et al, 1990; Giaccia et al, 1992; Olive et al, 1994; Ruiz de Almodovar et al, 1994; Zaffaroni et al, 1994; McKay and Kefford, 1995; Whitaker et al, 1995; Woudstra et al, 1998; Eastham et al, 2001). In conclusion, GFGE is the preferable method over a standard FDR assay to measure dsb induction in tumour cells.

The mean dsb induction frequency of all 16 cell lines was 8.1 × 10⁻¹² dsb/Gy/Da. (Table 2). This value confirmed our previous results (Dahm-Daphi and Dikomey, 1995; El-Awady et al, 2001) and also agreed well with other data (range 8–15 × 10⁻¹² dsb/Gy/Da) based on PFGE, from which a number of dsb can be calculated when it is combined with either ¹²⁵I-decay, analysis of fragment size distribution after high doses, or restriction digest (Blöcher and Pohlit, 1982; Ager and Dewey, 1990; Iliakis et al, 1991a, 1991b; Lawrence et al, 1993; Cedervall et al, 1994, 1995; Löbrich et al, 1994a, 1994b; Rothkamm and Löbrich, 1999, 2003). Only, Ruiz de Almodovar et al (1994) found a much higher induction frequency of about 66 dsb × 10⁻¹²/Gy/Da.

The amount of nonrepaired damage showed a broad variation among the nine tumour cell lines. However, it was shown that apoptosis (Chukhlovin et al, 1995) and also cell cycle progression could have an impact on dsb measurements, which means that the residual damage recorded does not exclusively reflect repair capacity. In line with that we did not find a correlation between residual damage and cell survival, which also agreed with other reports (Kelland et al, 1988; Olive et al, 1994; McKay and Kefford, 1995; Nunez et al, 1995; Whitaker et al, 1995; Woudstra et al, 1996). Surprisingly, four studies found such a correlation but only when the residual damage was measured 1–2 h after irradiation (Schwartz et al, 1988, 1990; Giaccia et al, 1992; Zaffaroni et al, 1994). It may well be that the impact of cell cycle and apoptosis was minimum after such short repair intervals and the data, thus, reflect mainly the amount of dsb induced and initial repair efficiency. Of note, for our data, the association between induced damage and radiosensitivity appears to be close, which means that the impact of the repair capacity should be only marginal.

The recently introduced technique to visualise sites of histone γ-H2AX phosphorylation (Rogakou et al, 1998) needs much lower doses than gel electropheresis to monitor dsb repair (Rothkamm and Löbrich, 2003) and may thus reduce apoptosis and cell cycle pertubation, although this technique is generally also sensitive to DNA degradation (Rogakou et al, 2000) and stalled replication (Ward and Chen, 2001). However, it needs to be shown whether recording of γ-H2AX will be an advantage over conventional gel electrophoresis for monitoring residual damage in tumour cells.

In contrast to tumour cells, for normal human and rodent fibroblasts, the variation in cellular sensitivity did not result from differences in the induction, but the repair of dsb (Dahm-Daphi et al, 1994; Kiltie et al, 1997; Zhou et al, 1997a, 1997b; Dikomey et al, 1998, 2000). These results suggest that the mechanisms affecting radiosensitivity are different for tumour and normal cell lines.