Origin and maintenance of mice

This research complies with ethical regulations, with protocols approved by the Institutional Animal Care and Use Committee (Van Andel Institute (VAI); protocols 19-0026, 22-09-036, 18-10-028 and 21-08-023).

FVB/NJ.Trim28+/MommeD9 (Trim28+/D9) mice were originally generated in the Whitelaw laboratory26, and B6.129S4-Trp53tm3.1Tyj/J (Trp53+/R270H) mice were generated in the Jacks laboratory28 and purchased from Jackson Laboratories (stock 008182). Both lines were backcrossed for over ten generations (FVB/NJ and C57Bl/6J, respectively) and maintained in house by breeding with wild-type siblings and periodic background refreshment using WT from JAX. Approximately 349 F1 hybrids were generated by crossing an 8-week-old FVB.Trim28+/D9 male with two 8-week-old B6.Trp53R270H/+ females, which were separated after checking for plugs the next morning. Mating mice were randomly selected. All mice were fed breeder chow (LabDiet, 5021, 0006540) ad libitum and housed in individually ventilated cages (Tecniplast, Sealsafe Plus GM500 in DGM Racks) at a maximum density of five mice per cage. Each cage was enriched with Enviro-dri (the Andersons, Crink-l’Nest) and cardboard dome homes (Shepherd, Shepherd Shack Dome). Whenever possible, same-sex siblings and same-sex animals from different litters were combined (~20 days of age) to co-house isogenic animals. Animals were kept on a 12-h light–dark cycle at an average ambient temperature of 23 °C and 35% humidity.

A total of 270 mice were randomly selected for body composition data analysis, including 137 males (18 WT, 34 Trim28+/D9, 44 Trp53+/R270H and 41 Trp53R270H/+;Trim28+/D9) and 133 females (30 WT, 32 Trim28+/D9, 36 Trp53+/R270H and 35 Trp53R270H/+;Trim28+/D9). At 4, 8, 16, 32, 40, 50, 60 and 70 weeks of age (or at euthanasia), mice were weighed and scanned with the EchoMRI system for fat and lean mass composition in the morning (EchoMRI, EchoMRI-100H).

Tumor analysis was conducted on 114 mice: 60 males (6 WT, 15 Trim28+/D9, 17 Trp53+/R270H and 22 Trp53R270H/+;Trim28+/D9) and 54 females (8 WT, 8 Trim28+/D9, 21 Trp53+/R270H and 17 Trp53R270H/+;Trim28+/D9). We performed tumor analysis blinded for genotype and phenotype, temporally collecting mice according to the timing of health reports. We specify in the text every time we are only referring to one of the sexes.

The VAI Vivarium Core officially requested a reduction of mouse cages to decrease staff in the building during the COVID-19 pandemic. As a result, we reduced active experimental mouse cages due to the extensive mouse number of our experimental cohort. Those mice are appropriately statistically censored in our data.

Mice are checked daily by animal keepers and two to three times per week by expert VAI Vivarium Core Staff for health, well-being and mass or tumor presence. Mice were flagged in health check reports if they exhibited >20% weight loss, tumors ~15% of body weight as assessed by palpation (this maximal tumor size was never exceeded), tumor ulcerations, tumor discharge or hemorrhage, mobility issues, reduced appetite or hydration, limited defecation or urination, abnormal gait or posture, labored breathing, lack of movement or hypothermia. Mice with reported health concerns or those reaching the 70-week study endpoint were euthanized via CO2 asphyxiation and cervical dislocation.

Mice used for different analysis are reported in Supplementary Table 1. The complete data of the mouse cohort are reported in Supplementary Table 2.

Genotyping

Ear punch biopsies were collected at 10 days and digested in 20 µl genomic DNA lysis buffer (100 mM Tris-HCl, pH 8.5, 5 mM EDTA, 0.2% SDS, 100 mM NaCl) with 20 mg proteinase K (Thermo Scientific, EO0491). The thermal cycling protocol used was 55 °C for 16 h, 95 °C for 10 min and a hold at 4 °C (lid at 105 °C). Nuclease-free water (Invitrogen, AM9938) was added to each lysate for a final volume of 180 µl. PCR reactions for Trim28 and Trp53 alleles used 1 µl diluted biopsy lysate in a 19-µl master mix (1× DreamTaq Buffer, 0.2 mM dNTPs, 0.1 µM forward and reverse primer mix, 2 U DreamTaq DNA Polymerase in nuclease-free water; Thermo Scientific, EP0703). PCR primers and thermal cycling conditions are detailed in Supplementary Tables 3 and 4. Each PCR product (20 µl) was digested with either 0.5 µl XceI and NspI (for Trim28+/D9; Thermo Scientific, FD1474) or 0.5 µl MslI (for Trp53R270H/+; New England Biolabs, R0571L) in a final reaction volume of 30 µl. Restriction conditions are detailed in Supplementary Table 5. Digestion products (~700 bp, WT Trim28; ~250 bp + ~450 bp, Trim28+/D9; ~500 bp, WT Trp53; ~200 bp + ~300 bp, Trp53R270H/+) were visualized on a 3% agarose gel (Fisher Scientific, BP160-500) in 1× TAE, with GelRed as the intercalating dye (Biotium, 41003).

Statistical analyses of developmental heterogeneity

Levene’s test was used to assess homoscedasticity of body, fat or lean mass across genotypes, with P values adjusted by the Benjamini–Hochberg method. mclust (version 5.4.9)59 was used for iterative expectation–maximization maximum-likelihood estimation in parameterized Gaussian mixture models, with regularization to smooth BIC. Classification uncertainty was used as the graphical parameter for fat–lean mass plots and for weighing log-rank P values in mouse Kaplan–Meier plots. Because most mice were classified with high confidence, the effect of this correction is negligible. Rmixmod (version 2.1.8)60 validated mclust results via unsupervised classification and density estimation using BIC, ICL and NEC. Both methods clustered 16-week fat and lean mass data by genotype. For tests requiring it, normality and equal variances were formally tested. Data analysis was not blinded. The VAI Bioinformatics and Biostatistics Core applied generalized additive models (GAMs, version 1.22.2) to model fat and lean mass changes over time, using random-effect splines for individual slopes and intercepts. The ‘emmeans’ package (version 1.10.0)61 compared overall group fat–lean mass slope differences by group, while a separate GAM modeled fat–lean mass differences at each time point. We included a random-effect spline for each mouse but excluded the spine for a random slope by week. Cancer death proportions were analyzed with similar models, with P values adjusted by the Benjamini–Hochberg method.

Tissue collection

Tissues were dissected and fixed in 10% NBF solution (3.7–4% formaldehyde (37–40%), 0.03 M NaH2PO4, 0.05 M Na2HPO4 in distilled water with a final pH of 7.2 ± 0.5): epididymal white adipose tissue; uterus or preputial glands, seminal vesicles and testis; bladder; pancreas; spleen; intestine; stomach; mesenteric fat; liver; kidneys; heart; lungs; thymus; brain; breast (ninth); hindlimb muscles and bones. We also recovered spine, ribs, skull, skin and any other abnormal mass. The fixative volume was 15–20 times the tissue volume. Specimens >2.5 mm thick were cut to proper fixation. Most tissues were fixed for 40 h, while fat-rich tissues (epididymal white adipose tissue, mesenteric fat, uterus) were fixed for 72 h. Bones and spines were fixed for 1 week followed by 1 week of decalcification in 14% EDTA (14% free-acid EDTA at pH 7.2, adjusted with NH4OH). After incubation, all tissues were transferred to 70% ethanol. Data collection was performed blinded.

Tissue preparation for histology

All tissues were embedded in paraffin by the VAI Pathology and Biorepository Core. Dehydration and clearing were automated with a Tissue-Tek VIP 5 instrument (Sakura) using the following protocol: 60 min in 70% ethanol, 60 min in 80% ethanol, 2× 60 min in 95% ethanol, 3× 60 min in 100% ethanol, 2× 30 min in xylene and 75 min in paraffin. Embedding was performed with a Leica EG1150 system. Three 5-µm sections, spaced 150 µm apart, were cut from each tissue for hematoxylin and eosin (H&E) staining using a Leica rotary microtome. The remaining tissue was stored as a paraffin block. H&E staining was performed with a Tissue-Tek Prisma Plus Automated Slide Stainer (Sakura) and Prisma H&E Staining Kit 1.

Pathology evaluation

Standard 5-µm H&E-stained sections were assessed for tumors and dysplastic lesions by a board-certified pathologist at the VAI Pathology and Biorepository Core. Most samples were provided blindly. Tumors were classified as malignant or benign, with all malignant tumors being primary. Metastatic or secondary tumors were identified based on primary tumor characteristics and immunohistochemical validation but were not reported in this study. Tumors were categorized into carcinomas, germ cell tumors, leukemias, lymphomas and sarcomas, with detailed classification by tissue of origin.

Tissue preparation for DNA and RNA extraction

Samples were randomly selected based on tumor type, genotype and phenotype. Curls from formalin-fixed, paraffin-embedded (FFPE) healthy tissues were prepared by the VAI Pathology and Biorepository Core, cutting three 50-µm curls using a microtome for storage at −80 °C. A 5-µm section was also cut for H&E staining to confirm the absence of tumors.

For tumor-containing FFPE tissues, microdissection or macrodissection was performed to separate the tumor from adjacent healthy tissue. All instrumentation and tools were treated with RNaseZAP (Invitrogen, AM9782) for decontamination, and RNase-free water was used in the tissue floating bath (Growcells.com, UPW-2000). Laser capture microdissection (LMD) was performed with the Leica LMD 6500 system. Glass slides (Leica, 11505189) were treated with UV light (Instrumedics UV curing lamp) for 30 min to prevent static and promote adherence. Mounted sections were cut at 10 µm and dried in an oven at 60 °C for 20 min. Slides were deparaffinized with three changes of xylene for 20 min followed by one change of 100% ethanol for 20 min. All slides were dissected within 24 h of sectioning. LMD-dissected areas were stored at −80 °C until further processing. Macrodissection of FFPE tissue sections was performed manually using a razor. The mounted sections were cut at 10 µm and dried in an oven at 60 °C for 20 min. All slides were dissected within 24 h of sectioning. Dissected areas were stored at −80 °C until further processing.

DNA extraction

Ear punch biopsies were collected and randomly selected based on tumor type, genotype and phenotype. DNA was purified using a DNeasy Blood & Tissue Kit (Qiagen, 69504) with minor modifications. After digestion, samples were brought to 220 µl with 1× PBS, and then steps 2–7 of the Quick-Start Protocol were followed. DNA was eluted with two washes of 100 µl buffer AE. For samples requiring both RNA and DNA, the Quick-DNA/RNA Microprep kit (Zymo, D7005) was used, following its specific protocol.

DNA from FFPE healthy and tumor tissues was extracted with the Quick-DNA/RNA FFPE kit (Zymo, R1009), with slight modifications based on tissue source. Curls and macrodissected tissues were deparaffinized in 800 µl Deparaffinization Solution at 55 °C for 5 min and then digested at 55 °C for 4 h. Microdissected tissues, previously deparaffinized, were digested at 55 °C for 1 h. Subsequent protocol steps were followed according to the kit manual, with an additional centrifugation step to remove residual ethanol before elution, which used 30 µl for macrodissected and microdissected tissues and 50 µl for curls, respectively. Purified DNA was quantified by Qubit fluorometry (Life Technologies).

Mouse DNA methylation array

DNA samples (6–500 ng) were bisulfite converted using the Zymo EZ DNA Methylation Kit (Zymo Research) following the manufacturer’s protocol with modifications for the Illumina Infinium methylation assays. After conversion, reactions were cleaned with Zymo-Spin columns and eluted in 12 µl Tris buffer. The bisulfite-converted DNA was processed using the Illumina mouse methylation array protocol. For the assay, 7 µl of converted DNA was denatured with 4 µl 0.1 M NaOH. The DNA was then amplified and hybridized to the Infinium BeadChip31 and underwent an extension reaction with fluorophore-labeled nucleotides according to the protocol. Arrays were scanned on the Illumina iScan platform, and probe-specific calls were made using Illumina GenomeStudio version 2011.1 to generate IDAT files. Data collection and analysis were conducted blind to experimental conditions.

DNA methylation analysis

Analysis of IDAT files was performed using the SeSAMe pipeline (version 1.22.2)62 and its wrapper SeSAMeStr (version 0.1.0)63. Fifty-eight independent biological replicates from ear biopsies of wild-type, Trim28+/D9, Trp53R270H/+ and Trp53R270H/+;Trim28+/D9 male mice at day 10 were analyzed. A second cohort of 22 samples from wild-type and Trim28+/D9 male mice was also analyzed independently. Additionally, 46 samples from intestines, lungs, stomachs and pancreata of Trim28+/D9 males were analyzed, comparing light and heavy mice. Data preprocessing and quality controls followed SeSAMe’s default parameters and the preprocessing code ‘TQCDPB’, with all samples showing a detection rate >90% and no dye bias. In differential DNA methylation analyses between genotypes, the effect size cutoff was set to 0.1 (10% differential DNA methylation) with a P-value cutoff of <0.05, unless specified otherwise. For analyses between isogenic Trim28+/D9-heavy and Trim28+/D9-light mice, the effect size cutoff was 0.05 (5% differential DNA methylation), with the same P-value threshold. Batch effect was included as a covariate, while other technical and biological effects (detection rate, initial DNA concentration, litter) were evaluated but not included due to co-linearity with the batch effect. Global DNA methylation correlation was assessed using the ‘chart.Correlation’ function from the ‘PerformanceAnalytics’ R package (version 2.0.4). Similarity between samples was calculated as the sum of squared residuals from linear regressions. Principal-component analysis of β values was performed on SeSAMeStr pipeline output using the R function ‘prcomp’ from the ‘stats’ package (version 3.6.2). Heatmap visualization of DML was created using the R package ‘ComplexHeatmap’ (version 2.20.0)64, with β values modeled and weighted using the mclust certainty score in limma (version 3.60.4)65. Probe enrichment analysis was performed using the SeSAMe knowYourCG module, with annotations based on the knowYourCG tool. Related information is available at http://zwdzwd.github.io/InfiniumAnnotation#mouse. The ChromHMM annotation is derived from a mouse consensus by the ENCODE project profiling 66 mouse epigenomes across 12 tissues at daily intervals from embryonic day 11.5 to birth66. Lists of probe-enriched genes are reported in Supplementary Table 6. Gene ontology analysis of probe-enriched genes was performed using the R package ‘clusterProfiler’ (version 4.12.1)67. Probe enrichment analysis on metabolism-related gene sets was performed by retrieving all gene sets with the term ‘metabolism’ or ‘metabolic’ from the Molecular Signatures Database (MSigDB)68. We analyzed 674 gene sets related to metabolism or metabolic processes from all MSigDB collections. Gene enrichment in the Jensen DISEASES database69 was performed using the R package ‘enrichR’ (version 3.2)70. Further data visualization of SeSAMe and SeSAMeStr output was performed using RStudio. For genomic snapshots of genes enriched for DML between Trim28+/D9-light and Trim28+/D9-heavy mice, we applied RUVSeq (version 1.38.0)71 to remove unwanted variation, defining empirical control probes based on differential analysis between Trim28+/D9-light and Trim28+/D9-heavy mice with limma65. The 100,000 least differentially methylated probes were used in the ‘RUVg’ command from RUVSeq, and RUVSeq-corrected, z-scored β values were plotted. Data analysis was not conducted blind to the experimental conditions.

Whole-exome sequencing

Samples were randomly selected based on tumor type and genotype–phenotype combinations relevant to the biological question. Libraries for whole-exome sequencing (WES) were prepared by the VAI Genomics Core from 100–200 ng of genomic DNA using the Twist Library Preparation EF Kit 2.0 version 5.0 (Twist Bioscience). DNA was enzymatically sheared to an average size of 250 bp, followed by end repair, A tailing and ligation to uniquely barcoded dual indexes (Twist Bioscience). PCR amplification (ten cycles) was performed, and the Twist Mouse Exome Panel was used to capture whole-exome regions before a final PCR round (six cycles). Quality and quantity of the libraries were assessed using the Agilent TapeStation 4200 (Agilent Technologies) and the QuantiFluor dsDNA System (Promega). Sequencing (2 × 100 bp) was performed on an Illumina NovaSeq 6000 sequencer (Illumina) to an average raw depth of 50 million paired reads per library. Base calling was performed with Illumina RTA3 (version 3.4.4), and the output was demultiplexed and converted to FastQ format with Illumina bcl2fastq2 (version 2.20). Data collection and analysis were conducted blind to the experimental conditions.

Whole-exome sequencing analysis

Adaptor sequences and low-quality reads were trimmed using Trim Galore (version 0.6.0; https://github.com/FelixKrueger/TrimGalore)72. Trimmed reads were aligned to the mm10 reference genome with BWA (version 0.7.17)73, and duplicates were marked using SAMBLASTER (version 0.1.26)74. Paired healthy and tumor BAM files were analyzed with GATK Mutect2 (version 4.1.8.1)75 to identity somatic single-nucleotide variants (SNVs), insertions and deletions (indels). BCFtools (version 1.17)76 was used to extract SNVs and indels based on the gene annotation file provided by Twist Bioscience. SnpEff (version 5.1)77 was used to predict the effect of each SNV and indel. Data visualization was performed using the ComplexHeatmap R package (version 2.20.0)64. Data collection and analysis were not conducted blind to the experimental conditions.

RNA extraction

Fresh ear punch biopsies were collected and in part randomly selected based on tumor type and genotype–phenotype combination, without using the genotyping buffer. Rather, RNA was extracted using the Quick-DNA/RNA Microprep Plus kit (Zymo, D7005), which includes in-column DNase I treatment. For FFPE tissues, the Quick-DNA/RNA FFPE kit (Zymo, R1009) was used with minor modifications based on the tissue type. Curls and macrodissected tissues were deparaffinized in 800 µl Deparaffinization Solution at 55 °C for 5 min, followed by digestion at 55 °C for 4 h. Already deparaffinized microdissected tissues were immediately digested for 1 h. The remaining steps were followed according to the kit manual, including except for an additional centrifugation step with the empty column before elution to remove residual ethanol. RNA was eluted with 50 µl of DNase- and RNase-free water for curls and 30 µl of DNase- and RNase-free water for macrodissected and microdissected tissues. Purified RNA was quantified by Qubit fluorometry (Life Technologies).

RNA sequencing

Libraries were prepared by the VAI Genomics Core from 142 ng of total RNA using the KAPA RNA HyperPrep Kit (Kapa Biosystems). Ribosomal RNA levels were reduced with the QIAseq FastSelect −rRNA HMR Kit (Qiagen). RNA was sheared to 300–400 bp, and cDNA fragments were ligated to IDT for Illumina TruSeq UD indexed adaptors (Illumina) before PCR amplification. Library quality and quantity were assessed using the Agilent TapeStation 4200 (Agilent Technologies) and the QuantiFluor dsDNA System (Promega). Individually indexed libraries were pooled, and 2 × 100-bp sequencing was performed on an Illumina NovaSeq 6000 sequencer to an average depth of 50 million raw paired reads per transcriptome. Base calling was performed with Illumina RTA3 (version 3.4.4), and outputs were demultiplexed and converted to FastQ format with Illumina bcl2fastq2 (version 2.20). Data collection and analysis were performed blind to the experimental conditions.

RNA-sequencing analysis

All reads from FastQ files underwent quality control, adaptor trimming, mapping and counting using the mRNA-seq pipeline of snakePipes (version 2.7.3)78. Mapping was performed with STAR (version 2.7.10b) on the GRCm38/mm10 mouse genome, and reads were counted with featureCounts (version 2.0.1). The resulting raw count matrices were input into DESeq2 (version 1.44.0)79, accounting for technical (RIN, library preparation and sequencing batches, initial RNA concentration) and biological (litter, age) covariates. Variance-stabilizing transformation was applied and was used for downstream analyses and data visualization. Dimensional reduction analysis was performed using the Rtsne (version 0.17, t-SNE, https://github.com/jkrijthe/Rtsne) and uwot (version 0.2.2, UMAP, https://CRAN.R-project.org/package=uwot) R packages. Heatmap visualization was performed using the ComplexHeatmap R package (version 2.20.0)64. Data analysis was not performed blind to the experimental conditions.

Immunofluorescence staining

Samples were randomly selected based on tumor type, genotype and phenotype. Paraffin sections (5 µm) were deparaffinized and subjected to antigen retrieval using DAKO EnVision FLEX High pH antigen retrieval buffer for 20 min at 97 °C by the VAI Pathology and Biorepository Core. Slides were blocked with 2% FBS for 1 h and then incubated overnight at 4 °C with anti-TRIM28 (Thermo Fisher, MA5-35303, clone ARC0047) antibody diluted 1:50 in DAKO EnVision FLEX antibody diluent. After washing 3× with 1× PBS for 5 min, slides were stained with Rb-647 secondary antibody (Invitrogen, A32733) at 1:500 for 2 h at room temperature, followed by three washes with 1× PBS for 5 min each. DAPI (Invitrogen, D21490) was applied for 10 min at room temperature, and slides were washed 3× in DI water for 5 min each before coverslipping with Prolong Gold mounting medium (Invitrogen, P36930).

Immunofluorescence acquisition

Whole-tissue images were collected in a single plane on a Zeiss Axioscan 7 slide scanner using ZEN blue (version 3.7) by the VAI Imaging Core. DAPI and AF647-stained samples were excited by a Colibri 7 LED light source at 385 nm and 630 nm. Emission was detected with a Hamamatsu ORCA-Flash 4.0 camera and a Zeiss Plan-Apochromat ×20, 0.8-NA air objective. Resulting 14-bit images, scaled to 0.1725 × 0.1725 µm per pixel, were compressed by JpgXr at 85%. Data collection and analysis were conducted blind to the experimental conditions.

Immunofluorescence analysis

Full-resolution pyramidal CZI images of TRIM28-stained tissue sections were analyzed using QuPath80 (version 0.5.1) by the VAI Imaging Core. Annotations were made to contour target tissue while excluding bubbles, large folds, nontarget tissues and bright autofluorescence caused by debris. Nuclei were detected using the QuPath StarDist (version 0.4.4) plugin via dsb2018_heavy_augment.pb (https://qupath.readthedocs.io/en/0.4/docs/deep/stardist.html). QuPath’s Train Object Classifier was trained on ROIs selected from two samples of each tissue type and their batch-paired positive controls to classify nuclei as positive or negative for TRIM28. The classifier was trained using the random tree model and applied across images. Key measurements (for example, number of detections, percent positive and fluorescence) were exported for statistical analysis, with single-nuclear signals averaged per biological replicate. At least three independent biological replicates per group (totaling 22,964,279 nuclei) were analyzed. TRIM28 nuclear fluorescence was corrected for known technical and biological confounders (antibody lot, litter, cause of death, age at death and tumor type) using correctBatchEffects from the R package ‘limma’ (ref. 65). Data collection and analysis were performed blind to the experimental conditions.

cBioPortal and TCGA Pan-Cancer Atlas data analyses

The TCGA Pan-Cancer Atlas39, comprising 32 studies and 10,967 samples, was used for preliminary analyses in cBioPortal81. Outputs were revisualized in RStudio. Kaplan–Meier curves were generated for samples harboring mutations in either the TRIM28+/D9-heavy or TRIM28+/D9-light gene signatures, compared to those without mutations, unless otherwise noted. Statistical significance was tested using log-rank tests (P < 0.05). Co-occurrence or exclusivity of pairwise mutations within these gene signatures was tested by one-sided Fisher’s exact test, with Benjamini–Hochberg Padj < 0.05. Overlap with the COSMIC Cancer Gene Census38 was similarly tested. The effects of mutations in TRIM28+/D9-light signature genes were assessed across samples from TCGA and non-overlapping samples from cBioPortal (n = 69,223 samples), with sex-specific analyses performed on 30,203 samples from 28,966 female patients and 34,276 samples from 32,192 male patients. Mutations were categorized as ‘harmful’ or ‘protective’ based on median survival ratios (months) between altered and unaltered samples (>1, ‘harmful’; <1, ‘protective’). To address sample bias, tissues were divided into low (<3,000) and high (>3,000) sample groups for heatmap visualization. The analysis was performed both on overall survival and disease-free survival. Gene expression and DNA methylation analyses of the TRIM28+/D9-light signature were performed on the Pan-Cancer Atlas from TCGA samples for which both RNA-seq and DNA methylation (HM27 and HM450 arrays) data were available, comprising 10,013 samples from 10,005 patients (of a total of 10,967 samples from 10,953 patients present in the Atlas) and 33 cancer types. DNA methylation analysis focused on probes within ±50 bp of TSS regions. Linear correlations between mRNA expression, DNA methylation and survival were visualized as heatmaps, with P < 0.05 for RNA-seq and 0.01 for DNA methylation. Data collection and analysis were not conducted blind to the experimental conditions.

Statistics and reproducibility

Power analysis was performed by the VAI Bioinformatics and Biostatistics Core using the pwr R package for power analysis (R version 3.5.2)82 to determine sample size. The goal was to assess cancer incidence differences (carcinoma, sarcoma, lymphoma and leukemia) between Trim28+/D9-light and Trim28+/D9-heavy mice. A two-sample test of proportions was performed using Firth logistic regression, with power set to 80%, α = 0.05 and equal sample sizes assumed for each group.

Owing to COVID-related reductions, 118 mice were randomly excluded for tumor analysis. Additionally, 17 mice were excluded because they were found dead and too stiff to harvest. The final cohort included all animals from litters of 4–12 pups.

Experiments were randomized, and investigators were blinded to group allocation and outcome assessment whenever possible.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.