1. Introduction
The intrinsic radiosensitivity of normal and tumour tissues is a major determinant of the outcome of any radiation-based treatment. A significant correlation between the intrinsic radiosensitivity of tumour cells, measured by in vitro clonogenic assays, and patient response to radiotherapy has previously been demonstrated [
1,
2]. The accurate prediction of radiation response and/or intrinsic radiosensitivity would be a major milestone towards effective personalised treatments.
Currently, radiotherapy, for a given cancer type, is typically given with a standard dose determined from clinical experience and population-level clinical trials. However, cancer is a highly heterogeneous disease, driven by numerous genetic alterations which significantly affect radiosensitivity. This genetic variability is reflected in estimates which suggest that cancers at the same site which receive the same treatment may vary in radiosensitivity by 25% or more [
3,
4]. As a result, a significant number of patients are certainly under- or over-dosed, reducing the clinical benefit of radiotherapy.
The clonogenic assay has been the gold standard method for measuring cellular radiosensitivity, providing a large amount of useful data. However, most of these studies were completed in different conditions and by different investigators, giving substantial variability in reported survival values, as elegantly demonstrated by Nuryadi et al. in the A549 cell model [
5]. This same survival variability is observed in published data from other cell models, such as HT116, PC3, H460, DU145, SW480, HT29, and HT1299 (
Figure 1) [
6]. This enormous variability makes it very difficult to produce accurate radiosensitivity biomarkers or predictive models.
Furthermore, the clonogenic assay has limitations, including the extended time required for colony growth (typically 7 to 10 days or longer), and the low efficiency of clone formation in certain cell lines, which can make this assay unsuitable for such cells. This highlights the need for the application of new rapid predictive assays and models of radiation response if intrinsic sensitivity is to be translated as a clinical tool.
The other obvious candidate as a radiosensitivity marker is DNA double-strand breaks (DSBs). Their relationship to clonogenic survival has been previously studied, and some investigators found correlation between the two assays [
7,
8,
9,
10,
11], but not all [
12]. The weakness of these studies has been that they were performed in a small number of cell lines. Additionally, micronucleus and apoptosis assays have also been studied as end points for cellular radiosensitivity; however, there is no clear agreement regarding the predictive power of these metrics [
11,
13,
14].
The aim of this study was to characterize radiosensitivity across a diverse panel of human cell lines. This combined clonogenic assays with measuring different biological features as surrogate markers of radiation response, including induction and repair of radiation-induced DSBs, cell cycle distribution, and ploidy. This was first performed in a controlled panel of cell lines with known DNA repair defects introduced using CRISPR-Cas9. This was then expanded into a more diverse panel of human cells with different genetic backgrounds. This enabled the evaluation of the predictive power of these baseline features, while at the same time beginning to build a robust dataset with reduced inter-experimental variabilities to support future modelling.
3. Discussion
It is widely accepted that there is a vast variability in intrinsic radiosensitivity between different cancer cells which is not currently accounted for. As a result, the identification of a biological endpoint predictive of tumour radiosensitivity would make an important contribution to enhancing the effectiveness of radiotherapy, by allowing treatments to be planned at an individual patient level.
This heterogeneity in radiosensitivity was seen in the present study, with SF2 values from 0.041 to 0.71 (MID from 0.70 to 3.36 Gy). This is in agreement with previous reports of radiosensitivity across a wide variety of human tumours by Amundson et al. (SF2 range from 0.038 to 0.95) [
16]. The lower range of SF2 reported in Amundson’s study was mostly seen in lymphoid tumours, which were not evaluated in our panel. By contrast, in our data, the only cell line that had extreme sensitivity was RPE-1 LIG4
−/−, and the second most sensitive cell line in our panel was RPE-1 ATM
−/−. ATM and LIG4 are major players in cellular DNA damage response and DNA repair, and their loss is well-established to be associated with radiation hypersensitivity [
15].
Indeed, our findings show remarkable concordance with previously published results from a CRISPR screen studying radiation response in RPE-1 p53
−/− cells, despite employing different methods to assess the impact of DNA repair deficiencies on radiosensitivity. Specifically, while we investigated clonogenic survival, the screen study quantified raw numbers of cells with the loss of particular genes [
15]. This not only underscores the reproducibility of different approaches, but also increases confidence in the robustness of our data presented here. We observed a near-perfect correlation among various knockout models, with the exception of BRCA1 null cells (R
2 = 0.98,
p = 0.002). This discrepancy in BRCA1 data could potentially be attributed to the limitations of CRISPR dropout screens in genes that are essential for cell proliferation, as well as genes with multiple cellular functions, as has been previously reported in BRCA1 [
17]. Nonetheless, despite this discrepancy, our results align closely with previously reported data and strengthen the overall reliability of our findings.
The determination of radiosensitivity by the clonogenic assay is a lengthy process, so there is a need to understand cellular mechanisms and parameters that might enable the prediction of intrinsic radiosensitivity. The development of such rapid techniques would enable them to inform clinical decision-making and allow personalized treatment planning.
Here, we evaluated the relationship of cellular radiosensitivity with different cellular characteristics in a two-step approach; first in a controlled environment where the only difference between cell models is targeted mutations in specific DNA damage repair genes, then in diverse panel of human cell lines where cells have a more complex genetic profile. A wide variation was observed in cellular features, such as ploidy, plating efficiency, and DNA damage repair rates, among the cells used in this study.
Radiation-induced DSBs and their repair is attractive as a clinical and pre-clinical test, as results can be available in a few hours. Moreover, DSB repair is an obvious candidate marker for radiation sensitivity because DSBs are considered the most critical type of DNA damage caused by radiation. The relationship between residual levels of DSBs or repair rates to cell survival in cancer has been a topic of numerous investigations [
7,
8,
9,
10,
11,
12]. In CRISPR-modified cells with DNA repair defects, the number of residual DSBs is the metric which best predicts cell survival (
= 0.78,
p = 0.009). This excellent performance in the CRISPR-modified cell lines is likely due to the very controlled environment where the only difference between cell models is one mutation in a specific DNA damage repair gene.
On the contrary, in the whole panel of human cell lines, where cells have a more complex genetic profile, residual DSBs fail to predict survival (
= 0.02,
p = 0.542), as do other metrics associated with DSBs, such as baseline levels of damage and levels of induced damage. This is despite many of the cell lines we investigated having mutations in DSB repair pathways (
Supplementary Table S4). Interestingly, even among the subgroup of normal human cell lines that are assumed to have an intact genetic background, none of the analysed metrics demonstrate predictive power, despite significant variation in sensitivity. It is important to note that most of these normal cell lines were immortalized and the impact of immortalization on radiation response is complex and context-dependent, and further research is needed to fully understand the underlying mechanisms. Tissue type may also play a role in these differences, as the more varied backgrounds of these normal lines may impact on involved proliferative, metabolic, and cell death processes.
However, while CRISPR knockout generates severe homozygous loss of function mutations, many of these mutations have less clear significance or are heterozygous, potentially having a reduced impact on DSB repair. This is supported by more general mutational data, such as in the Cancer Cell Line Encyclopedia [
18], where 22% of cell lines have at least one damaging mutation in the NHEJ or HR pathways, but only 2% of cell lines have a homozygous damaging mutation of the kind assessed in most radiobiological DNA repair studies. This suggests that, in more diverse cell lines, DSBs alone may not be as strong a predictor of response.
Considering other possible parameters, in this study we did not find a correlation between survival and ploidy. This agrees with the study of West et al., where no statistically significant differences in SF2 values could be found between diploid and aneuploid cervical carcinomas [
19].
Cellular sensitivity is known to vary during different phases of the cell cycle. Typically, cells are most sensitive to irradiation during mitosis and in late G2, less sensitive in G1, and least sensitive during the latter part of S phase [
20]. Thus, cell cycle profile distribution is another possible predictor of radiosensitivity. Nonetheless, there is some controversy around the impact of cell cycle distribution in radiosensitivity with some studies showing correlations while others have not [
13,
21]. Here, we found suggestions of correlation between intrinsic sensitivity and cell cycle distribution, particularly with the percentage of cell in S phase in both groups of cells, CRISPR-modified (
= 0.46,
p = 0.044) and pooled cell lines (
= 0.37,
p = 0.005).
Interestingly, although S-phase is typically suggested to be a more resistant phase of the cell cycle, in the CRISPR-modified cell lines this correlation is negative (
Figure 3E). This suggests that this may be an indirect effect, resulting from, e.g., slowed progression through S phase when DNA repair defects are present. A positive correlation is seen as expected between S-phase population and MID in the pooled cell lines. Somewhat surprising is the substantial correlation of percentage of cell in G2 phase of the cycle in the subgroup of lung cancer cell lines (
= 0.72,
p = 0.032), as this metric performed poorly for all the other subgroups (
< 0.35,
p > 0.12).
Similar behaviour was observed with plating efficiency, where it had some predictive power for survival (
= 0.31,
p = 0.013) across the pooled cell lines. This metric performed well in the subgroup of lung cancer cell lines (
= 0.79,
p = 0.018), but poorly for the prostate cancer cells (
= 0.03
p = 0.77). The lack of correlation between plating efficiency and survival has also been previously reported [
22]. Plating efficiency had a similar predictive power on the CRISPR cell lines subgroup (
= 0.45,
p = 0.048), although this is dominated by the LIG4 knockout line. Taken together, we can conclude that plating efficiency may give some indications of sensitivity but is not a generally reliable predictive feature alone.
Moving beyond individual correlations, these data may be useful, in combination with other resources, to help develop new mechanistic models of radiation response which can incorporate these different parameters together in predictive tools [
23]. To support this, results from this paper are made available in full in a
supplementary data file.
For future model development, the variability between and within each of the subgroups, combined with the lack of correlation with DNA damage metrics, leads us to hypothesize that genetic alterations in proliferation and metabolic pathways may have a significant impact in cellular radiation response, as cellular metabolism, proliferation, and antioxidant pathways play crucial roles in the response of cells to radiation.
As an example, high expression of the mTOR protein, an important protein in the regulation of cell proliferation and metabolism, was found to be associated with poor prognosis in cervical cancer treated with radiotherapy [
24]. Therefore, pathways such as PI3K/Akt/mTOR and MAPK/ERK have an important role in intrinsic radiosensitivity and clinical radiation response and should be investigated.
Furthermore, metabolic pathways also control the production of energy and essential molecules needed for cell survival and repair. Understanding how radiation affects cellular metabolism can provide insights into the mechanisms of radiation response and may lead to strategies for enhancing radiation efficacy. The disruption of cellular metabolism with mannose has been shown to result in impaired cancer cell growth and survival [
25]. This effect is thought to be mediated through multiple mechanisms, including the disruption of glycolysis, leading to a decreased energy production, modulation of signalling pathways involved in cancer progression, such as the PI3K/Akt pathway and enhancement of anti-cancer immune response [
26].
Similarly, radiation exposure increases the levels of reactive oxygen species (ROS) and other damaging molecules, overwhelming the cellular antioxidant defence mechanisms. Subsequently, this can cause oxidative stress, further DNA damage and cell death. Antioxidant pathways, such as the glutathione system, superoxide dismutase, and catalase, help to neutralize ROS and prevent oxidative damage [
27]. Modulating these antioxidant pathways may enhance the cellular antioxidant capacity and reduce radiation-induced damage [
28].
In summary a better understanding of the interplay between cellular metabolism, proliferation, antioxidant pathways, and radiation response is essential for elucidating the mechanisms of radiation response and may lead to the development of strategies to increase radiation efficacy in cancer treatments.
4. Materials and Methods
4.1. Cell Lines
In this study a total of 27 cell lines were used, divided in four groups: (1) 6 Human lung cancer cell lines acquired from ATCC (H460, A549, H1792, SW1573, H23, and H441); (2) 5 Human prostate cancer cell lines acquired from ATCC (PC-3, DU145, 22RV1, C42B, and LNCAP); (3) 8 Human normal cell lines acquired from ATCC or ECACC (PNT1A, MRC5, RPE-1, WPMY1, AGO1522, RWPE1, MCF10A, and PNT2); and (4) 8 CRISPR-modified cell lines, derived from the RPE-1 cell line with stable knock-out of p53, ATM, DCLRE1C (Artemis), BRCA1, BAX, LIG4, FANCD2, and PRKDC (DNA-PKcs).
All cell lines were maintained in supplier recommended complete medium (
Supplementary Table S5). All cultures were incubated at humidified 37 °C in 5% CO
2.
The RPE-1 cells were selected for CRISPR-Cas9 studies due to their stable epithelial cell characteristics, with minimal genetic modifications, which closely mimic normal tissue biology. Additionally, its widespread usage in previous CRISPR studies provides a robust foundation of data for comparison, confirming its suitability as a reliable and well-established cellular model.
The process of selecting tumoral and normal human cell lines in this study was carried out based on factors such as cell availability, previously described levels of radiosensitivity and their widespread use in radiation studies. Lung and prostate cell lines were selected in this initial comparison as radiotherapy is frequently used as a primary treatment in these cancers.
4.2. Establishment of DNA Repair Deficient CRISPR/Cas9 Stable Cell Lines
CRISPR-Cas9 modified cell lines were generated using a transient transfection of CRISPR-Cas9 in RPE-1 cells, by delivering a ribonucleoprotein (RNP) complex to recombinant Cas9 coupled to a specific guide RNA following manufacturer’s instructions. In brief, an RNP complex was assembled by mixing the tracrRNA (100 µM)(#1072532 IDT, Coralville, AI, USA) and crRNA (100 µM) (predesigned crRNA were obtained from IDT, Coralville, AI, USA) in equimolar concentrations and Cas9 Nuclease V3 (61 µM) (#1081058 IDT, Coralville, AI, USA). A list of crRNA used in this study can be found in
Supplementary Table S6.
Delivery of the RNP complex to cells was performed via Lonza nucleofection. For that, 3.5 × 105 cells suspended in 20 µL of Nucleofector Solution (P3 Primary Cell 4D-NucleofectorTM X Kit S, #V4XP-3032, Lonza, Basel, Switzerland) supplemented with 5 µL of previously assembled RNP complex and 1 µL of IDT electroporation enhancer (100 µM) (#1075916 IDT, Coralville, AI, USA) were used for each nucleofection. After nucleofection cells were expanded for clonal selection.
All cell lines were generated as single knockouts, apart from BRCA1 which was combined with P53 knockout, as cells with only BRCA1 knockout were not viable.
The RPE-1 cells were selected for CRISPR-Cas9 studies due to their stable epithelial cell characteristics, with minimal genetic modifications, which closely mimic normal tissue biology. Additionally, its widespread usage in previous CRISPR studies provides a robust foundation of data for comparison, confirming its suitability as a reliable and well-established cellular model.
4.3. Knock-Out Validation by Western Blot
Stable knock-out was validated by Western Blot of the protein of interest in populations derived from single clones (
Supplementary Figure S3). Cells were lysed in RIPA buffer (150 mM Tris-HCl pH 8; 50 mM NaCl; 1%
v/
v NP-40, 0.5%
v/
v sodium deoxycholate (10%) and 0.1%
v/
v SDS (10%)) supplemented with protease inhibitor (cOmplete Mini, Roche, Basel, Switzerland).
A 40 µg total protein sample was loaded onto an SDS-PAGE gel, and, after electrophoresis, the proteins were blotted on a nitrocellulose membrane (Life Technologies, Carlsbad, CA, USA). The membranes were blocked with 5% non-fat milk in PBS-Tween (0.1%) and incubated overnight at 4 °C with primary antibody diluted in milk (
Supplementary Table S7). After washing with PBS-Tween, membranes were incubated in their secondary anti-rabbit and anti-mouse horseradish peroxidase-conjugated antibodies diluted at 1:2000 at room temperature for 1 h. The membranes were then washed and developed with Luminata Crescendo Western Blot HRP substrate (Milipore, Burlington, MA, USA) using the GBox Imager by Synagene (Synagene UK, Cambridge, UK).
4.4. Clonogenic Survival Assay
The colony formation assay was carried out according to published methods [
29]. Cells were seeded into six-well plates with an optimal cell density according to the absorbed dose. On the following day, cells were irradiated with doses of 0 to 8 Gy of X-rays, at a dose rate of 0.59 Gy/min using a 225 kVp, 13.3 mA X-RAD 225 radiation source (Precision X-ray Inc., Madison, CT, USA).
After irradiation, cells were incubated for 7 to 11 days depending on the cell line. The colonies were stained with 4% crystal violet solution in ethanol and were manually counted, with a colony defined as consisting of at least 50 cells. From these counts, plating efficiency (PE) and survival fraction (SF) were calculated. Survival fraction was defined as the number of colonies formed after irradiation divided by the number of cells seeded, corrected for the PE of unirradiated cells. Data were fit to the linear quadratic equation () using non-linear regression weighted for standard deviation. The Mean Inactivation Dose (MID) is defined as the area under the dose–response curve.
4.5. DNA Damage Immunofluorescence Assay
Following 2 Gy irradiation, cells were fixed in a 50:50 methanol–acetone solution and permeabilized (0.5% Triton X-100 in PBS) at predetermined time points before being blocked in blocking buffer (5% FBS an 0.1% Triton X-100 in PBS) and stained with 53BP1 primary antibody (1:5000) (#NB100-304, Novus Biologicals, Centennial, CO, USA) and
H2AX primary antibody (1:10000) (#05-636-I, Merks Chemicals, Frankfurt, Germany) for 1 h before being washed three times and stained with Alexa Flour 568 goat anti-rabbit IgG secondary antibody (#A21429, Life Technologies, Carlsbad, CA, USA) and Alexa Flour 488 goat anti-mouse IgG secondary antibody (#A21131, Life Technologies, Carlsbad, CA, USA) (1:2000) in the dark for 1 h. Following staining, cells were washed three times and mounted onto microscope slides using Prolong Gold anti-fade reagent with DAPI (#P36930, Invitrogen, Carlsbad, CA, USA). Foci were manually counted from the whole nucleus of 50 randomly selected cells on each sample with a Nikon Eclipse Ti microscope (Nikon Corporation, Tokyo, Japan), using a 60× objective, representative images of 53BP1 foci can be found in
Supplementary Figure S4. Data are presented as the mean values of foci per cell and the respective standard deviation of three independent experiments. Data presented here are corrected for control samples by subtracting off the number of foci in unirradiated cells. For repair kinetic analysis, foci data were then fit in GraphPad Prism 9 using non-linear regression to an exponential decay,
, where
represents the initial number of foci,
represents the residual damage and
is the rate of DSB repair.
4.6. Cell Cycle Profile and Ploidy Analysis
Exponentially growing cells were harvested before being fixed in 100% ice-cold ethanol and left at 4 °C overnight. Samples were then centrifuged, the excess ethanol was removed, the cell pellets were then resuspended in PBS and centrifuged again before being resuspended in 500
L of PI/RNAse A. The samples were incubated at 37 °C for 30 min before being analysed in a BD Accuri C6 Plus Flow Cytometer (BD Biosciences, Franklin Lakes, NJ, USA). In total, 10,000 flow cytometer events were collected and analysed per sample. Quantification was carried out using the BD Accuri C6 Plus Analysis software. To perform ploidy analysis, the value of each sample’s G1 median fluorescence intensity (MFI) peak was then compared to the value of the MFI peak of Chicken Erythrocyte Nuclei (CEN) Singlets (#1013, BioSure, Grass Valley, CA, USA). The ratio of the G1 MFI peak value to the CEN singlets MFI peak value was then compared with reported ploidy value of cells by Sanger (
https://cellmodelpassports.sanger.ac.uk (accessed on 8 November 2021)) where available, showing an
of 0.93 (
Supplementary Figure S1).
4.7. Statistical Analysis
All experiments were performed in triplicate, data are presented as mean values and respective standard deviation. Unpaired Student’s t-test and one-way ANOVA were used for statistical evaluation as appropriate. All statistics and graph plotting used GraphPad Prism 9.0 (GraphPad, Boston, MA, USA).