"Special care was taken to repeat the study by Tillmann et al. [4] as accurately as possible. Male C3H/HeNCrl (n 1⁄4 43) and female C57Bl/6N (n 1⁄4 290) mice were purchased in a staggered design from Charles River Germany, Sulzfeld, Germany, at an age of 8e9 weeks. After acclimatization, at the age of 12 weeks (females), the males and 128 females were mated for one week (ratio 3 fe- males: 1 male) in two rounds, thus a total of 256 potentially pregnant females were obtained. They were distributed to the 128 cages of the exposure devices, two animals per cage. Exposure or sham-exposure of the pregnant females thus started at day 6 p.c. (post conception). All females were weighted at day 13 p.c., and the ones with the highest weight gains remained in the exposure de- vices while the others were sacrificed (CO2 overdose). The remaining 34 females, age 12 weeks, were mated with the males, and the female offspring served as the untreated, unexposed cage control (n 1⁄4 96, three animals per cage). At day 14 p.c., the females in the exposure devices were injected (i.p.) with ethylnitrosourea (ENU; SigmaeAldrich, Taufkirchen, Germany) at a dose of 40 mg/kg in saline. Six days after birth, after sexing three female F1 animals were left with their mothers, while the surplus females and the males were removed and sacrificed. Litters with too few female pups were filled up with surplus females from other litters of the same exposure group. In total, four groups of female F1 mice were obtained, 96 animals per group. At day 21, pups were weaned, and the dams were sacrificed.

2.2. Exposure to electromagnetic fields

The exposure devices consisted of eight radial waveguides with 16 cages each, arranged in stacks of two and connected to power amplifiers and RF-generators. Details have been published earlier [5]. Extensive numerical calculations of the field distributions and the corresponding SAR values revealed unavoidable substantial variations for animals in different positions and within animals (local maximum SAR values) which could be as much as 3e5 times higher than the whole-body SAR. Two waveguides per exposure group with 16 cages each (32 cages in total, 96 animals) were one out of four groups with the following nominal whole-body SAR levels: sham-exposed (0 W/kg), 0.04 W/kg (low), 0.4 W/kg (mod- erate) and 2 W/kg (high) for a reference configuration of three mice (body weight 20 g each) per cage, with a standard deviation for this configuration of around 36%. The exposure was comparably ho- mogeneous with standard deviations of the whole body SAR within the cages between 30% (adult animals) and 91% (pups) due to spatial electric field variations and movement of the groups of animals. In order to equalize the exposure between the 16 cages of each waveguide (max. 12% variation of the cages' mean electric field strengths), the cages were permuted every second day by one exposure section. Exposure was 24/7 for the entire period (72 weeks) with the exception of half an hour each night at 3 a.m. in order to shut down the generators and to reboot the system. It should be noted that exposure was interrupted for 4 h daily in the previous study [4] for animal caretaking. Another difference was due to different geometries of the radial waveguides used in the previous study and in the present one (diameter of the three waveguides with 20 cages each were 1.9 m in Ref. [4], and 2.2 m in the present study). Thus, the ratios of whole body SAR values to the respective incident field strengths were different. This difference, however, is not biologically important but only relevant for the SAR calculations.

Electrical fields inside the waveguides as well as temperatures were measured automatically. One power amplifier in the highest exposure group malfunctioned for 20 days (weeks 33e35), pro- ducing only one fourth of the SAR during this period, affecting half of this exposure group. All other generators and amplifiers worked without failure so that the total time without exposure was mini- mal (0.5%). Exposure conditions were not known to the persons handling the animals or otherwise being involved in the experi- ment. Only after all data have been analyzed, they were sent to the cooperating partners (University of Wuppertal) in exchange with the exposure codes.

2.3. Procedures during and at the end of the experiment

The mice were routinely inspected visually, and their body weights were recorded weekly during the first 14 weeks, and thereafter every two weeks. After one year (week 52), pooled feces

samples of all cages were collected and checked for the presence of Helicobacter ssp. (PCR analysis, Charles River). None of the samples was positive. When signs of disease were noted, or when the body weight of an animal showed a sudden drop, these animals were removed from their cages, sacrificed by CO2, and immediately dissected. All surviving animals, including the cage controls, were sacrificed when survival rates of the ENU-treated animals dropped below 25% (OECD guideline No. 451 ‘Carcinogenicity Studies’). Due to a technical mishap, sham-exposed and exposed (2 W/kg) ani- mals were sacrificed 1e2 weeks too early or too late, respectively. Gross morphological abnormalities were noted, and the following organs were immersion-fixed: brain, kidneys, spleen, liver, and lymph nodes. Lungs were immersion-fixed after intratracheal instillation of formalin (37%). Tissues (except lymph nodes) were embedded in paraffin, and sections of 4 mm were stained with hematoxylin/eosin. Histopathological examination was done ac- cording to international standards [6]. To ensure that our diagnoses were correct, a set of 107 sections with different tumors were cross- checked by a professional pathologist. No deviations from our di- agnoses were found.

2.4. Statistical analysis

Comparisons of body weights was done by parametric analysis of variance, followed by repeated measures post-hoc test. Survival times were compared by log-rank test (SPSS v. 22, IBM). Incidences of tumors were compared by Fisher's exact test (one-tailed) using the program GraphPad Prism (GraphPad Software, San Diego, CA, U.S.A.). For the Bayesian analysis, a script was programmed in R [7]."

C57Bl/6N mice, female

HAWC