Effects of ethanol and acetaldehyde on multi-organ genomic mutation landscapes Authors: Bérénice Chavanel, Eva Tibaldi, François Virard, Christine Carreira, Sergey Senkin, Behnoush Abedi-Ardekani, Vincent Cahais, Michael Korenjak, Zdenko Herceg, Silvia Balbo, Daniele Mandrioli, Jiri Zavadil Published: 2026 ABSTRACT Alcohol consumption is a major risk factor for cancer, particularly of the head and neck. Acetaldehyde, the main ethanol metabolite, is linked to oral carcinogenesis, but the mechanisms are not well understood. While some mutational signatures such as single-base substitution signature SBS16, doublet-base substitution signature DBS4, and indel signature ID11 have been linked to alcohol use, they are not limited to alcohol-related cancers. We examined the genome-wide in vivo mutation landscapes of rats chronically exposed to ethanol or acetaldehyde, focusing on head-and-neck tumors (involving the cheek and the Zymbal gland) and tumors of the forestomach and liver. Mutational signature analysis revealed early-onset, exposure-associated formation of signature SBS17 in ~ 38% of head-and-neck tumors, suggesting inflammation and oxidative damage may mediate carcinogenesis in these sites. We observed relative enrichment of driver-like mutations in Tp53, Mtor and Hras in exposed tumors, where all Tp53-mutated tumors were SBS17-positive, recapitulating a similar scenario from human gastro-esophageal adenocarcinomas. Next, the doublet-base, indel, copy number and structural variant changes were sporadic and not following exposure-specific patterns. No signatures previously associated with alcohol consumption in humans were observed in the exposed animals. Thus, in this animal cancer model, alcohol-related carcinogenesis across multiple organs is unlikely to be driven mainly by direct mutagenicity of ethanol or acetaldehyde. Rather, the results support a role for non-mutational processes in mediating their carcinogenic effects. Further investigation is warranted to clarify the relationship between mutational signature SBS17 and TP53 loss-of-function mutations as potential biomarkers of alcohol-related carcinogenesis in animals and humans.Graphical AbstractEthanol or acetaldehyde exposure in rats induces limited genome-wide mutations in head-and-neck, forestomach, and liver tissues.SBS17 appears early in 38% head-and-neck tumors following exposure, implicating inflammation and oxidative damage. Tp53, Mtor, and Hras mutations are enriched upon exposure, with all Tp53-mutant tumors SBS17-positive, paralleling human cancer observations.Exposure-linked carcinogenesis is unlikely driven by direct mutagenicity of ethanol/acetaldehyde. Supplementary InformationThe online version contains supplementary material available at https://doi.org/10.1007/s10565-026-10217-9. Introduction Alcohol consumption is a major risk factor for numerous diseases, including cancers at multiple anatomical sites. Exposures from ethanol (EtOH) and its primary metabolite, acetaldehyde (AcA), resulting from alcoholic beverage consumption, are both classified by IARC as carcinogenic to humans (Group 1) (IARC 2012). Alcohol intake increases the risk of at least seven cancer types, including those of the oral cavity, pharynx, larynx, esophagus, liver, breast and colorectum, which together accounted for approximately 4.1% of all new cancer cases worldwide in 2020 (Rumgay et al. 2021). The mechanisms underlying alcohol carcinogenicity are diverse and not fully defined, though AcA likely plays a central role. EtOH and AcA have been reported to exert various genotoxic and mutagenic effects, potentially contributing to carcinogenesis and cellular dysfunction (Brooks and Zakhari 2014; Vijayraghavan and Saini 2023; Thapa and Chan 2025). Mutagenic processes resulting from chemical exposures or endogenous pathways can be investigated by analyzing their induced mutational signatures (Nik-Zainal et al. 2012; Alexandrov et al. 2013; Koh et al. 2021, Degasperi et al. 2022). The presence of these signatures indicates that the associated mutagenic mechanisms play a role in carcinogenesis (Roberts and Gordenin 2014; Hollstein et al. 2017; Balmain 2020). Mutational signatures may arise from natural cellular processes such as DNA repair deficiencies or deregulated mutagenic enzymes, or result from DNA damage by environmental, lifestyle, or therapeutic exposures such as tobacco smoke, asbestos, UV radiation or alkylating drugs. Understanding the origins of mutational signatures reveals the biological mechanisms and environmental factors underlying tumor initiation and development. Although alcohol is an established human carcinogen (IARC Group 1), no mutational signature has been unambiguously associated with its consumption. COSMIC single-base substitution (SBS) mutational signature SBS16 has been tentatively linked to alcohol drinking and is observed most frequently in liver cancer cases (Alexandrov et al. 2020). Accumulation of SBS16-specific mutations was found in alcohol-drinking patients with liver tumors (Fujimoto et al. 2016; Letouze et al. 2017) and esophageal squamous cell carcinomas (Chang et al. 2017, Li et al. 2018, Lin et al. 2018, Moody et al. 2021, Ko et al. 2022). A correlation was reported between SBS16 formation and the reduced-activity variant of ALDH2, an enzyme that detoxifies AcA, suggesting that AcA accumulation from impaired ethanol metabolism may underlie the characteristic SBS16 mutation pattern (Chang et al. 2017, Moody et al. 2021, Wei et al. 2021). However, SBS16 has also been found in individuals with no alcohol exposure history (Lin et al. 2018; Brunner et al. 2019; Ng et al. 2021; Pinyol et al. 2021). Other studies found the signature in the context of exposure to tobacco smoke (Alexandrov et al. 2016), another source of acetaldehyde. SBS16 mutations were specifically identified in current and ex-smoker samples with a dose–response induction of the signature with increasing pack-years of tobacco use (Gillison et al. 2019; Yoshida et al. 2020). More recently, the double-base substitution (DBS) signature DBS4 and the indel (ID) signature ID11 were found, in addition to SBS16, in the tumor genomes of head and neck (Torrens et al. 2025). SBS16 was also reported as strongly associated with the combined exposure to alcohol and tobacco smoke (Moody et al. 2021, Plath et al. 2021; Lawson et al. 2025, Torrens et al. 2025, Deneuve et al. 2026). Alcohol may also exert non-genotoxic effects in cancer development, such as by altering liver stem cells prior to tumorigenesis by inducing a tumor-permissive microenvironment, e.g. through chronic inflammation or immunosuppression that can drive precancerous transitions and enhance liver cancer progression. (Ma et al. 2020, Nguyen et al. 2021). Previous in vitro experimental studies of mutagenic effects of AcA, conducted either in human induced pluripotent stem cells (Kucab et al. 2019) or in yeast models (Thapa et al. 2022, Vijayraghavan et al. 2022) remained inconclusive, by failing to yield a concordant exposure-specific mutational signature. Thus, solid experimental evidence for specific, EtOH and AcA-directed genome-wide mutagenesis remains lacking. Studies of the effects of chemical compounds in rodents provide invaluable insights into how these agents initiate and promote cancer development, offering an approximation of the human physiological context. Rodent models are commonly used in chemical carcinogenicity bioassays to evaluate potential human carcinogens (Hoenerhoff et al. 2009; Gift et al. 2013), due to the similarity of relevant biological pathways between rodents and humans. An additional advantage of rodent bioassays is their ability to capture chemical metabolism in vivo, a feature that is often lacking in cell-based systems, which may not fully replicate the complex interactions among metabolic activation and detoxification pathways. Studies showed that rodent models reproduce mutational patterns induced in other experimental settings and/or found in human cancer data (Huang et al. 2017; Speer et al. 2023, Korenjak et al. 2024) and also allow assessing the effects of non-genotoxic promoters of carcinogenesis (Riva et al. 2020). Importantly, the IARC Monographs program considers the outcomes from animal bioassays as a key evidence stream in the carcinogen evaluation and classification program. Experimental studies in mice by the U.S. National Toxicology Program examined ethanol’s toxicological and carcinogenic effects, with inconclusive results on cancer risk from drinking-water exposure (NTP 2004). In contrast, Sprague Dawley rat exposure bioassays performed at the Cesare Maltoni Cancer Research Center of the Ramazzini Institute, Bologna, Italy, as part of the long-term experimental studies on the carcinogenicity of EtOH (Soffritti et al. 2002a) and AcA (Soffritti et al. 2002b) indicated organ-specific tumorigenic effects for either exposure. We thus hypothesized that direct DNA damage by EtOH/AcA can be effectively assessed in archived tissue samples derived from these longitudinal in vivo studies performed in well-controlled experimental settings. Here we report the genome-wide analysis of tumors formed in Sprague Dawley rats chronically exposed to EtOH or AcA through drinking water. By performing whole-genome sequencing (WGS) of the tumor and non-tumor control tissues, and through the analysis of multi-scale mutational signatures, copy number alterations and EtOH/AcA-associated mutations in cancer driver genes, we aimed to comprehensively delineate the in vivo underpinnings of alcohol-related mutagenesis and carcinogenesis. Materials and methods Rat tissues Alcohol-fixed paraffin-embedded (AFPE) tumor and non-tumor Sprague Dawley rat tissues from the Cesare Maltoni Cancer Research Center of the Ramazzini Institute repositories were analyzed. The rats had been exposed to EtOH (0% or 10%) (Soffritti et al. 2002a) or AcA (0, 50, 250, 500, 1500 or 2500 mg/L) (Soffritti et al. 2002b), with either chemical administered in their drinking water and supplied ad libitum until the time of spontaneous death. Based on a thorough histopathological assessment, our analysis focused on a subset of tumors from all organs that were directly exposed or involved in EtOH/AcA metabolism and showed elevated tumor incidence after EtOH or AcA treatment (Fig. S1), relative to the low sporadic rates observed in control (untreated) animals. The final set included tumor samples from the cheek, the Zymbal gland (an auditory gland considered a direct site of exposure), the larynx, the forestomach, and the liver. For each animal, unaffected, non-clonal kidney tissue samples were used as a source of germline DNA reference. A total of twenty-eight pairs of tumor and matching normal samples from both exposure studies were included in the WGS analysis (Table S1). Tissue processing and enrichment for tumor content Ten 10-μm sections of tumor tissue and ten sections of normal kidney were obtained for each animal. The tumor sections were flanked by two additional 3-μm sections stained with hematoxylin and eosin (H&E). The stained cuts were scanned using the Glissando Desktop Scanner (Objective Imaging) for histopathological evaluation and demarcation of the tumor area using Aperio eSlide Manager software (Leica Biosystems) (Fig. S2 and Table S2). For a subset of tumor samples with the actual 3D tumor size lacking (n = 16), the size was based on estimates reflecting the largest tumor dimension measured on the tissue section slides. Deparaffinization was performed by washing the slides with xylene, followed by a series of descending concentrations of ethanol as follows: two consecutive baths of xylene, followed by baths of 100% ethanol, 95% ethanol, and 70% ethanol and a final bath of distilled water. To enrich for tumor cell content, and to minimize the inclusion of non-malignant cells, such as immune cells, blood vessels, stromal cells, or necrotic tumor cells, careful macrodissection of the tumor areas was done by scraping the region of interest using H&E staining of neighboring sections as a visual guide. DNA extraction and NGS library preparation Genomic DNA was extracted from the fifty-six animal-matched tumor and non-tumor tissues, using the Macherey–Nagel Nucleospin Tissue XS kit. Ten sections of 10 μm were used for tumor tissues and three sections for non-tumor tissues. For large regions of interest (> 25 mm2), a maximum of three sections was applied per column and a maximum of five sections per column for small regions of interest (< 25 mm2). Libraries for WGS were constructed using the KAPA HyperPrep Library Preparation kit. Fragmentation of DNA was performed using a Covaris S220 ultrasonicator instrument. Post-extraction DNA and final DNA libraries were quantified using the Qubit fluorometer 3.0 from ThermoFisher. Agilent TapeStation 4200 was used to check the size distribution of genomic DNA and final libraries prior to sequencing (Fig. S3). Whole-genome sequencing, somatic variant calling and filtering The libraries were whole-genome sequenced on S4 flow cell of the Illumina NovaSeq 6000 platform (Genewiz, Azenta) in the paired-end 150 base-pair mode, loading six to seven samples per lane to achieve ~ 450 million paired-end reads per sample. The FASTQ files were assessed for quality using the FastQC tool, and processed by using the modules within the IARC bioinformatics platform (https://github.com/IARCbioinfo), as follows: raw sequencing reads were adapter- and quality-trimmed using Trim Galore (v0.6.10; Cutadapt v5.1), then aligned to the reference genome Rnor_6.0/Rn6 (Rattus norvegicus) with BWA-MEM (v0.7.17). Alignments (BAM files) were sorted and indexed using Sambamba (v0.7.1), and duplicates were marked with Samblaster (v0.1.26). Base quality score recalibration (BQSR) was performed using GATK4 (v4.1.7.0) following best-practice recommendations. The key metrics including the X coverage for the aligned reads (mean = 32.7x, median = 34.1x, range = 13.9–54.6x) and duplicate read rates (mean = 27%, median = 26%, range = 17–45%) are listed for all samples in Table S3. Somatic variants were subsequently called using the variant caller GATK4 Mutect2 (v4.2.0.0) (Benjamin et al. 2019) to detect single-base, doublet-base substitutions (SBS and DBS, respectively) and small insertions/deletions (indels, or IDs) in the tumor samples using the animal-matched normal kidney tissue as germline-like reference. Two tumor-normal pairs (LV_4/N_4 and LX_21/N_21) were excluded from further analysis after the relatedness analysis using the Somalier tool (v0.3.2) (Pedersen et al. 2020) (Table S3) indicated that the paired tissues originated from unrelated animals. Somatic variants for the remaining 26 tumor-normal pairs were annotated using ANNOVAR (2025Mar02 release) (Wang et al. 2010), Further filtering of the MuTect2 variant calling format (VCF) files was performed as follows: all public rat SNPs from the NCBI’s dbSNP Build 146 have been removed. No presence of a mutated read in the tumor sample was allowed in its normal tissue counterpart. A variant allele frequency (VAF) threshold of > = 0.4 was set for the tumor mutations to stringently remove all non-clonal events, such as sporadic DNA damage due to sample processing. Table S4 and Fig. S4 collectively summarize the resulting absolute and relative somatic mutation counts and their distribution across the genome segments. The final mutation spectra (SBS, DBS, ID) (Fig. S5, S6, S7, respectively) were obtained for all the tumor samples and processed in further analysis. Structural variants (SVs) were called from BAM files using GRIDSS (v2.13.2)(Cameron et al. 2017). Somatic SVs were subsequently processed through a custom IARC Nextflow pipeline (available at https://github.com/IARCbioinfo/mutacet_2026). Briefly, SV calls were filtered to retain variants with a minimum read coverage of 10x, strictly excluding any alternative variants present in the matched normal sample to ensure somatic origin. Filtered VCF files were converted into BEDPE format using a custom integration script designed for downstream compatibility, and the individual sample SV spectra were plotted prior to signature extraction analysis (Fig. S8). Extraction of de novo mutational signatures and their decomposition into COSMIC signature The filtered somatic mutation sets were processed by SigProfilerMatrixGenerator for de novo mutational signature analysis comprising extraction and decomposition into COSMIC content using SigProfilerExtractor (v1.2.6), integrated within the automated workflow. The initial number of SBS, DBS, ID and SV signatures to be extracted per data set was estimated based on the stability and mean sample cosine distance for different signature numbers. Final de novo signatures were extracted using the non-negative matrix factorization module of SigProfilerExtractor (Islam et al. 2022) (Table S5). SigProfilerExtractor was further used to decompose the de novo signatures into established known signatures from version 3.4 catalogue of the COSMIC Mutational Signatures (Tables S6, S7). When signature reconstruction cosine similarity fell below 0.8 (the case of ID signatures ID83B and ID83C, see Table S6), the decomposition result was not considered. Optimized attribution of activities of the SBS mutational signatures SigProfilerExtractor-identified COSMIC SBS signatures were attributed to each sample using the optimized signature attribution tool Mutational Signature Attribution Analysis (MSA) https://gitlab.com/s.senkin/MSA (Senkin 2021). For this step, only COSMIC reference signatures that were identified in the decomposition of de novo signatures were included in the panel used for attribution. MSA estimated confidence intervals by repeatedly simulating mutation profiles and reassigning signatures using a parametric bootstrap approach, to extract 95% confidence intervals for each attributed mutational signature activity (Fig. S9). Analysis of driver mutations in cancer genes Somatic variants from mutation annotation format (MAF) files were annotated using ANNOVAR (release 2025Mar02) and subsequently filtered to retain exonic and splicing (non-silent) variants with predicted functional consequences (e.g., nonsynonymous SNVs, stopgain/stoploss mutations, and frameshift indels). Gene symbols were parsed to account for multiple annotations per variant. Different variant allele frequency (VAF) cutoffs were tested (VAF ≥ 0.2, VAF ≥ 0.4), and somatic mutations with high variant allele frequency (VAF ≥ 0.4) were further analyzed, as they were potentially exposure-driven and clonally enriched early events. Furthermore, this threshold helped exclude potential effects arising from fixative exposure or other sporadic, processing-related damage, all of which would be expected to be non-clonal and occurring at low VAF. Two curated driver gene sets (IntOGen (release 2024–09–20) and the COSMIC Cancer Gene Census (CGC, release v101/v103, 2025) were used for annotation of the mutated cancer genes in the rat, upon conversion to the Rattus norvegicus orthologs using the babelgene R package (version 22.9) (Dolgalev 2022). The genes successfully mapped during ortholog conversion were retained, resulting in 469 genes from IntOGen and 554 from COSMIC CGC. The non-silently mutated candidate cancer driver genes from the rat tumors were matched and identified against the union of these two lists. Copy number alteration analysis Somatic copy number alterations (CNAs) were identified using CNVkit (version 0.9.11) (Talevich et al. 2016) on WGS data, following the workflow recommendations. A pooled reference was constructed by combining all available normal samples, with bias correction applied for GC content and repeat-masked regions using Rnor_6/Rn6 reference genome (Rattus norvegicus), downloaded from the UCSC Genome Browser and divided into bins of 1,000 base pairs. Gene annotation was performed using the rn6 RefFlat file, also retrieved from the UCSC Table Browser (hgTables). Results Exposure effects on tumorigenesis and tumor mutation burden An elevated overall number of malignant tumors was observed in animals from all treatment groups in both the ethanol (EtOH) and acetaldehyde (AcA) exposure studies, compared to the untreated control animals (Fig. S1). Neither treatment resulted in significant differences in body weight among the studied animals (Soffritti et al. 2002a, b). While the control group exhibited a background level of tumors consistent with spontaneous and sporadic tumor formation, a phenomenon commonly expected during long-term rodent carcinogenicity bioassays, the treated groups exhibited higher incidence of neoplastic lesions. This increase suggests that both EtOH and AcA exposures may promote or accelerate tumorigenesis beyond the baseline levels observed in untreated animals. We observed a statistically significant trend toward earlier spontaneous death in the treated animals, which on average died around 104 weeks of age, compared to 129 weeks in the untreated control group, considering all analyzed tumor sites. This difference was also observed upon separate analysis for head-and-neck tumors, indicating accelerated tumor development and/or lethality following exposure to EtOH or AcA (Fig. 1A–D). Furthermore, tumors were larger in treated animals compared to those forming spontaneously in untreated controls (Fig. 1E–H). The effects of treatment on both survival and tumor size reached statistical significance in head-and-neck tumors arising in the cheek, Zymbal gland, and larynx (Fig. 1C, D, G, H), and these differences were further accentuated upon grouping of tumors from EtOH- or AcA-treated animals (Fig. 1C, G).Fig. 1The age-at-death and tumor-size effects of EtOH or AcA exposures in Sprague Dawley rats. (A-D) The graphs show the age of the untreated control and exposed rats categorized by cancer site, at the time of spontaneous death. (E–H) Graphs show the size of the tumor samples, categorized by cancer site. All tumor sites (A, B, E, F, n = 28) or a subset of head-and-neck tumors (C, D, G, H, n = 18) are shown separately. Untreated and treated samples (combining EtOH- and AcA-treated samples) are shown in (A, C, E, G). Untreated and the individual EtOH and AcA exposure groups are shown in (B, D, F, H). Unt: Untreated, Treated: EtOH- or AcA-treated combined. The sample size for each experimental group is shown under the x-axis in parentheses. * p < 0.05, ** p < 0.01 using non-parametric Mann–Whitney test (combined EtOH and AcA treatment groups) or Kruskal–Wallis test (separate treatment groups) followed by uncorrected Dunn’s test for multi-group comparisons. ns: non-significant p value. Median is represented by black horizontal lines The age-at-death and tumor-size effects of EtOH or AcA exposures in Sprague Dawley rats. (A-D) The graphs show the age of the untreated control and exposed rats categorized by cancer site, at the time of spontaneous death. (E–H) Graphs show the size of the tumor samples, categorized by cancer site. All tumor sites (A, B, E, F, n = 28) or a subset of head-and-neck tumors (C, D, G, H, n = 18) are shown separately. Untreated and treated samples (combining EtOH- and AcA-treated samples) are shown in (A, C, E, G). Untreated and the individual EtOH and AcA exposure groups are shown in (B, D, F, H). Unt: Untreated, Treated: EtOH- or AcA-treated combined. The sample size for each experimental group is shown under the x-axis in parentheses. * p < 0.05, ** p < 0.01 using non-parametric Mann–Whitney test (combined EtOH and AcA treatment groups) or Kruskal–Wallis test (separate treatment groups) followed by uncorrected Dunn’s test for multi-group comparisons. ns: non-significant p value. Median is represented by black horizontal lines We next analyzed overall somatic mutation burdens in the whole-genome sequenced tumor samples and found that somatic SBS counts were relatively similar across tumors from the control, EtOH-treated, and AcA-treated animals (Fig. S4A). Three samples displayed relatively elevated mutation numbers, potentially indicating a hypermutator phenotype or technical artifacts related to sample quality, DNA preparation, or sequencing errors. The majority of mutations were SBS events evenly distributed across all tumor sites and exposure groups, with the expected genome-segment distribution (Fig. S4B, C). Exposure-related enrichment of single-base substitution mutational signature SBS17 We next investigated the mutational signatures in the rat tumors. From the SBS mutation spectra (Fig. S5) we identified four de novo SBS mutational signatures (SBS96 A, B, C, D) of high stability (Fig. S9A) and decomposed them further into COSMIC mutational signatures (Fig. 2A, Fig. S9B, S10, Tables S6, S7). These comprised clock-like signatures SBS1, SBS5 and SBS40a present across all or most tumors, regardless of the treatment conditions. We also observed sporadic, low-representation occurrence of signature SBS37, not associated with treatment. Possibly mis-characterized SBS2 and SBS10b (each in a single sample and with unexpectedly low counts) were subsequently pruned from the final list. Interestingly, signatures SBS17a and SBS17b were identified solely in the head-and-neck tumors of the exposed animals, including one cheek (sample CH_8) and 27% of the Zymbal gland tumors (ZG_9, ZG_23, ZG_24, ZG_25) derived from animals exposed to 10% EtOH or higher doses of AcA (1,500 and 2,500 mg/L) (Fig. 2B, E). The optimized mutational signature attribution by using MSA (Senkin 2021) confirmed the extraction/decomposition results described above (Fig. 2C) and namely the enrichment of SBS17b in the same set of samples (Fig. 2D, F). While the MSA identification of the T > G-based SBS17b appeared highly specific and in keeping with the SigProfilerExtractor findings, we observed limited, low-level attribution interference for the T > A, T > C-based SBS17a in a subset of samples (Fig. S11). In addition, we detected three COSMIC mutational signatures SBS45, SBS58 and SBS95, likely technical artifacts (Fig. 2B, C, Fig. S9B, S10), originating mainly from the de novo signature SBS96C exhibiting ~ 90% artifact-derived content. The artifactual signatures were observed somewhat more frequently in tissues from EtOH-treated rats, suggesting possible study-specific differences in sample processing or quality. Importantly, neither the extraction nor the attribution analysis of the tumors of exposed animals revealed any evidence of mutational signature SBS16, previously proposed to reflect the effects of alcohol’s metabolite AcA in humans.Fig. 2SBS mutational signatures identified in the rat tumors. (A) COSMIC SBS mutational signatures of biological (non-artefact) origin extracted from the sequenced tumor samples. (B, C) Graphs show relative distribution of the mutational signatures across tumor samples with corresponding total mutation counts on the top. * = signatures listed in COSMIC as possible sequencing artifacts. Mutational signatures were (B) extracted either by using SigProfilerExtractor or (C) attributed by using Mutational Signature Attribution (MSA). (D) Relative optimized attribution of SBS17b across tumor samples by using the MSA tool. (E) Summary of the presence of signature SBS17 (SBS17a and SBS17b combined) in head-and-neck tumors (cheek, Zymbal gland, larynx) of animals exposed to either treatment (left panels, EtOH and AcA combined), EtOH only (middle panels) or AcA only (right panels) using SigProfilerExtractor. (F) Validation of the presence of signature SBS17 (considering only SBS17b) in head-and-neck tumors (cheek, Zymbal gland, larynx) of animals exposed to either treatment (left, EtOH and AcA combined), EtOH only (middle) or AcA only (right) using MSA. Statistical analysis was done using nonparametric Mann–Whitney test considering all values (including zeros). n.a.: p-value non applicable. Abbreviations: AcA—acetaldehyde; EtOH—ethanol; NNLS—non-negative least squares; SBS—single base substitution; cancer sites referred to by sample ID’s: CH—cheek, FS—forestomach, LV—liver, ZG—Zymbal gland SBS mutational signatures identified in the rat tumors. (A) COSMIC SBS mutational signatures of biological (non-artefact) origin extracted from the sequenced tumor samples. (B, C) Graphs show relative distribution of the mutational signatures across tumor samples with corresponding total mutation counts on the top. * = signatures listed in COSMIC as possible sequencing artifacts. Mutational signatures were (B) extracted either by using SigProfilerExtractor or (C) attributed by using Mutational Signature Attribution (MSA). (D) Relative optimized attribution of SBS17b across tumor samples by using the MSA tool. (E) Summary of the presence of signature SBS17 (SBS17a and SBS17b combined) in head-and-neck tumors (cheek, Zymbal gland, larynx) of animals exposed to either treatment (left panels, EtOH and AcA combined), EtOH only (middle panels) or AcA only (right panels) using SigProfilerExtractor. (F) Validation of the presence of signature SBS17 (considering only SBS17b) in head-and-neck tumors (cheek, Zymbal gland, larynx) of animals exposed to either treatment (left, EtOH and AcA combined), EtOH only (middle) or AcA only (right) using MSA. Statistical analysis was done using nonparametric Mann–Whitney test considering all values (including zeros). n.a.: p-value non applicable. Abbreviations: AcA—acetaldehyde; EtOH—ethanol; NNLS—non-negative least squares; SBS—single base substitution; cancer sites referred to by sample ID’s: CH—cheek, FS—forestomach, LV—liver, ZG—Zymbal gland The identified DBS and indel signatures do not correlate with exposure To further investigate the mutational patterns in the tumors of the animals, we analyzed the DBS and ID signatures. One de novo DBS signature and four de novo ID signatures were extracted in the tumor samples and decomposed into reference COSMIC signatures (Fig. S12, S13, Tables S5, S6, S7). We did not observe clear exposure-related differences for DBS and ID signature distribution, except for a possible trend for signature ID2 present in 7 of 20 (35%) exposed samples compared to 1 of 6 (17%) unexposed. However, the trend did not achieve statistical significance (Mann-Whitney test, p = 0.35). Similarly to SBS16, signatures DBS4 and ID11 previously associated with the combination of alcohol drinking and tobacco smoking (Torrens, Moody et al. 2025) were not detected in the EtOH- or AcA-treated rat tumor genomes. Genes with recurrent/multiple non-silent mutations and exposure-related profiles of cancer driver genes To assess the functional impact of the somatic mutations in the analyzed tumors, we extracted all genes that were mutated by non-silent (protein sequence changing) events, focusing on either genes mutated in at least 2 samples or genes mutated more than once in the same sample (multi-hit scenario) (Table S8). The resulting set of 20 recurrently mutated genes is shown in Fig. 3A. However, these genes include a mix of cancer-associated and non-cancer genes. We thus explored the potential cancer driver mutations and their distribution across the treatment conditions. Among the candidate cancer driver genes analyzed, 38 well-established genes with pivotal roles in oncogenesis were detected as non-silently mutated in the tumors (Fig. 3B, Table S9). Of these, 25 (66%) carried loss-of-function mutations, 7 (18%) gain-of-function mutations and 6 (16%) loss-of-function/dominant negative mutations (see Fig. 3C and Table S9 for detailed annotations). This analysis highlighted a noteworthy divergence between untreated and EtOH/AcA-induced tumors, with treated samples harboring recurrent mutations in 3 well-established driver genes, overlapping with the analysis shown in Fig. 3A. These comprised the tumor suppressor gene Tumor protein p53 (Tp53), harboring established or likely loss-of-function amino acid substitutions (p.G106A, p.T123M, p.Y124C, p.A157V, p.V141A, and p.S239F), which are likely to disrupt Tp53’s tumor-suppressive function. Tp53 mutations mostly occured in AcA-exposed animals (in 5 out of 11 (45%) tumors of the AcA-exposed animals) and in 2 out of 10 (20%) tumors of the EtOH-exposed animals. Notably, 5 of the 6 (83%) Tp53-mutated samples overlapped with the majority (5 of 7, or 71%) of samples positive for signature SBS17 (Fig. 4A), as detected by either mutational signature extraction (by SPE) or mutational signature attribution (by MSA) (Fig. 2B and C and Fig. S11). Additionally, the gene mechanistic target of rapamycin (Mtor) carried recurrent mutations (p.R731H and p.A1344D), though the functional significance of these variants remains unknown. These alterations were found in tumors of the liver (LV_14), and the Zymbal gland (ZG_27), corresponding to respective treatments by EtOH 10% or AcA 2500 mg/L. Lastly, we observed recurrent driver mutations in the Harvey rat sarcoma viral oncogene homolog (Hras), with gain-of-function substitutions (p.G12E and p.Q61L) in exposed samples, and another p.G12E mutant in the unexposed group. Both p.G12E and p.Q61L are well-established oncogenic driver mutations associated with aggressive, proliferative tumors across many human cancers.Fig. 3Non-silently and recurrently mutated genes and cancer drivers. (A) Oncoplot-like summary of genes with non-silent, recurrent mutations identified across the analyzed samples. Colored rectangles denote the type of mutation detected: red – non-synonymous single-nucleotide variants (SNVs); green—stop-gain mutations; dark gray—frameshift deletions; blue—frameshift insertions; and diagonally divided squares indicate samples with multiple mutations in the same gene. The bar plots on the right summarize the percentage of samples with recurrently mutated genes (left, in blue) and the percentage of non-silent mutations across the sample set (right, in red), respectively. Gene names are shown on the y-axis and sample IDs on the x-axis. See Table S8 for further detailed annotations. (B) Distribution of the mutated cancer driver genes (orthologous to human cancer genes from the Cancer Gene Census or the IntOgen databases) across treated and untreated sample groups. The relative frequencies of the number of samples affected by at least one mutation occurring in the listed cancer driver genes are represented by the colored dots. The corresponding changes in protein (AA—amino acid) sequence are detailed on the right of the graph. Refer to Materials and methods for additional details. (C) Functional classification of mutated driver genes identified in the analyzed samples. The pie chart shows the proportion of cancer driver mutations known or predicted to result in loss-of-function, gain-of-function, or loss-of-function/dominant-negative effects, based on the reported amino acid changes and known functional annotations of the affected driver genes. See Table S9 for further detailed annotationsFig. 4Cross-species cancer relevance of the proposed SBS17 formation and concurrent TP53 loss. (A) Venn diagram showing the overlap between SBS17-positive tumors and tumors harboring driver mutations in key cancer genes (Tp53, Hras, and Mtor). Numbers indicate the number of tumors within each intersecting category. (B) Distribution of SBS17-positive tumors across alcohol-associated cancer types. Bubble size indicates the percentage of tumors positive for mutational signature SBS17 among all analyzed tumors of the same cancer type (data source: COSMIC). The number of SBS17-positive tumors and total analyzed tumors are indicated on the right, together with the estimated proportion of tumors harboring both SBS17 and TP53 mutations. (C) Proposed mechanism linking ethanol exposure and acidic reflux to SBS17-associated mutagenesis. Ethanol metabolism to AcA and chronic inflammation induced by gastric or pancreatic acidic reflux may promote reactive oxygen species (ROS) production. ROS can oxidize the free nucleotide pool, converting dGTP into 8-oxo-dGTP, which may subsequently mispair with adenine during DNA replication and generate T > G transversions characteristic of the SBS17b mutational pattern, whereas TP53 loss may confer a selective advantage for cancer outgrowth on an SBS17 mutagenic background. LoF—loss of function, TLS pol— translesion synthesis polymerase Non-silently and recurrently mutated genes and cancer drivers. (A) Oncoplot-like summary of genes with non-silent, recurrent mutations identified across the analyzed samples. Colored rectangles denote the type of mutation detected: red – non-synonymous single-nucleotide variants (SNVs); green—stop-gain mutations; dark gray—frameshift deletions; blue—frameshift insertions; and diagonally divided squares indicate samples with multiple mutations in the same gene. The bar plots on the right summarize the percentage of samples with recurrently mutated genes (left, in blue) and the percentage of non-silent mutations across the sample set (right, in red), respectively. Gene names are shown on the y-axis and sample IDs on the x-axis. See Table S8 for further detailed annotations. (B) Distribution of the mutated cancer driver genes (orthologous to human cancer genes from the Cancer Gene Census or the IntOgen databases) across treated and untreated sample groups. The relative frequencies of the number of samples affected by at least one mutation occurring in the listed cancer driver genes are represented by the colored dots. The corresponding changes in protein (AA—amino acid) sequence are detailed on the right of the graph. Refer to Materials and methods for additional details. (C) Functional classification of mutated driver genes identified in the analyzed samples. The pie chart shows the proportion of cancer driver mutations known or predicted to result in loss-of-function, gain-of-function, or loss-of-function/dominant-negative effects, based on the reported amino acid changes and known functional annotations of the affected driver genes. See Table S9 for further detailed annotations Cross-species cancer relevance of the proposed SBS17 formation and concurrent TP53 loss. (A) Venn diagram showing the overlap between SBS17-positive tumors and tumors harboring driver mutations in key cancer genes (Tp53, Hras, and Mtor). Numbers indicate the number of tumors within each intersecting category. (B) Distribution of SBS17-positive tumors across alcohol-associated cancer types. Bubble size indicates the percentage of tumors positive for mutational signature SBS17 among all analyzed tumors of the same cancer type (data source: COSMIC). The number of SBS17-positive tumors and total analyzed tumors are indicated on the right, together with the estimated proportion of tumors harboring both SBS17 and TP53 mutations. (C) Proposed mechanism linking ethanol exposure and acidic reflux to SBS17-associated mutagenesis. Ethanol metabolism to AcA and chronic inflammation induced by gastric or pancreatic acidic reflux may promote reactive oxygen species (ROS) production. ROS can oxidize the free nucleotide pool, converting dGTP into 8-oxo-dGTP, which may subsequently mispair with adenine during DNA replication and generate T > G transversions characteristic of the SBS17b mutational pattern, whereas TP53 loss may confer a selective advantage for cancer outgrowth on an SBS17 mutagenic background. LoF—loss of function, TLS pol— translesion synthesis polymerase Importantly, the mutations detected in the cancer driver genes in Fig. 3B and Table S8 exhibited a minimum variant allele frequency (VAF) of 0.42 (average VAF of 0.51, median VAF ~ 0.58), indicating these mutations pervade the tumors in a clonal manner, upon arising early and becoming positively selected for during the tumor development. This indicates their possible functional importance and roles as drivers. Nevertheless, the overall mutation rates in established cancer driver genes were low, consistent with the relatively low mutation burdens and a lack of more pronounced treatment-associated mutation patterns. Analysis of structural variants and copy number alterations The analysis of structural variants (SV) revealed highly variable SV counts per sample (average = 42/sample, median = 12/sample, range: 0 to 345; see Table S4). Three COSMIC SV signatures were varyingly distributed in the tumor genomes (Fig. S14): signature SV2 which encompasses translocations and larger inversions (from 100 kb to > 10 Mb), larger tandem duplications (100 kb to > 10 Mb) and larger deletions (from 1 Mb to > 10 Mb), was present in almost all samples regardless the exposure groups, Next, we observed signature SV5 (rich in deletions of 1–10 kb, followed by tandem duplications (mainly from 1 to 100 kb, reaching 1 Mb and above). SV5 was somewhat enriched in exposed samples (11 of 20, 55%) compared to untreated samples (1 of 6, 17%) but the difference did not reach statistical significance (Mann-Whitney test, p-value of 0.12). Lastly, SV8 composed of 1–10 Kb inversions, was observed across the samples regardless of the exposure groups. Comprehensive analysis of copy number alterations (CNAs) across the whole-genome sequencing data did not reveal remarkable recurrent or well-established amplifications or deletions in the analyzed tumor samples. The observed copy ratio profiles remained largely within the expected diploid range, with only isolated and low-amplitude fluctuations not reflecting the treatment conditions (Fig. S15). These minor variations appeared to be sporadic. Overall, the CNA landscape suggests stable genomic architecture in response to the treatment by EtOH or AcA, with no evidence of widespread genomic gains or losses indicative of selective pressure or clonal evolution under either exposure regimens. Discussion Our study documents that chronic exposure to either EtOH or AcA increases tumor formation in rats and affects survival and tumor size in head-and-neck cancer contexts. The enrichment of signature SBS17 in a subset of head and neck tumors from ethanol- or high-dose AcA–treated animals suggests a link between alcohol metabolism and SBS17 formation. Next, the concurrent presence of SBS17 mutational activity and TP53 loss-of-function mutations is notable as it mirrors a reported pattern in human gastro-esophageal adenocarcinomas (Secrier et al. 2016; Abbas et al. 2023). In up to 70% of these tumors (Fig. 4B), SBS17a/b is strongly associated with TP53 mutations, chromosomal instability, and progressive disease states, particularly in the Barrett’s esophagus-esophageal adenocarcinoma sequence. These reports (Secrier et al. 2016; Abbas et al. 2023) suggest that TP53 dysfunction may permit tolerance of the DNA damage or replication stress underlying signature SBS17. Thus, the observation of the same co-occurrence in EtOH or AcA-exposed rat tumors recapitulates a key feature of this human disease context, supporting the interpretation that similar mutagenic processes and selective pressures may be operating. In particular, this raises the possibility that the exposure induces SBS17-like damage, while TP53 mutations provides a permissive background for clonal expansion under this mutational burden, paralleling the evolutionary dynamics observed in human gastro-esophageal cancers. More generally, SBS17-positive cancer sites in humans (esophagus, breast, stomach, pancreas, colorectum) have been epidemiologically associated with alcohol intake (Alexandrov et al. 2020) (Fig. 4B). The mechanisms underlying SBS17 are not fully characterized, although hypotheses include oxidative damage of the free nucleotide pool, leading to SBS17-specific mutation patterns (Dulak et al. 2013; Tomkova et al. 2018). Barrett’s esophagus and its progression to esophageal adenocarcinoma, where SBS17 occurs frequently, are associated with chronic gastro-esophageal reflux causing persistent inflammation and ROS-driven oxidative stress (Dvorak et al. 2007; Yoshida 2007; Kauppi et al. 2016). Pancreatic reflux containing bile and enzymes can also affect the stomach and esophagus. Acid- and bile-induced oxidative stress can damage DNA bases and the nucleotide pool, forming 8-oxo-dGTP. Its incorporation by translesion synthesis (TLS) polymerases interferes with replication and repair and increases T > G transversions, the hallmark of SBS17 (Satou et al. 2007; Suzuki and Kamiya 2017; Tomkova et al. 2018) (Fig. 4C). Consistent with inflammation driving SBS17, reduced SBS17-specific mutations have been observed in Barrett’s esophagus patients using nonsteroidal anti-inflammatory drugs (NSAIDs) (Galipeau et al. 2018). In human cancers, SBS17 is common in gastrointestinal tumors—esophageal adenocarcinoma (Weaver et al. 2014, Secrier et al. 2016), Barrett’s esophagus (Abbas et al. 2023), and gastric cancers (Han et al. 2024)—where acid and bile reflux and chronic inflammation cause oxidative stress. Chronic alcohol consumption can increase gastro-esophageal reflux (Pehl et al. 2002, Seidl et al. 2011, Pan et al. 2019) and, as an inducer of ROS through ethanol/AcA metabolism, may exacerbate oxidative damage and trigger mutational processes resulting in SBS17. However, the formation of SBS17 is not exclusively linked to alcohol exposure. Multiple studies have reported its presence in rodent tissues in the absence of exposure to any known exogenous mutagen. In cultured mouse embryonic fibroblasts (MEF), the signature forms naturally during senescence-bypass and clonal immortalization (Olivier et al. 2014; Nik-Zainal et al. 2015; Zhivagui et al. 2019; Riva et al. 2020). SBS17 also arose in subclones expanded from mouse small-bowel single-organoid stem cells under prolonged culture (Behjati et al. 2014) and in murine single cells and clones derived from primary fibroblasts (Milholland et al. 2017). In vivo models likewise show SBS17 formation: in aflatoxin B1–treated mice developing HCC-like liver cancer, SBS17 appears as a late-onset pattern independent of exposure (Huang et al. 2017), and mouse models of diffuse human gastric cancer exhibit a mutational pattern reproducible with human gastric cancer, including SBS17 (Zhang et al. 2022). Genome-wide analysis of colorectal tissues across 16 species shows SBS17 only in rodents (mouse, rat, naked mole-rat) (Cagan et al. 2022). The pattern is also detected in both tumor and normal mouse tissues (Riva et al. 2020). Together, these observations suggest that SBS17 may partly reflect endogenous DNA-damage processes exacerbated under stress conditions such as those induced by alcohol exposure and/or oxidative stress. Genotoxicity and mutagenicity were proposed as modes of action of the EtOH/AcA pathway, and aldehydes, more generally, were reported as predominant forces inducing DNA damage and inhibiting DNA repair such as in the context of tobacco smoke carcinogenesis (Weng et al. 2018). However, growing evidence indicates that EtOH/AcA mutagenicity may not be the predominant effect in the context of alcohol drinking. AcA forms covalent DNA adducts such as N2-ethylidene-2'-deoxyguanosine (N2-ethylidene-dG), N6-ethylidene-deoxyadenosine (N6-ethylidene-dA), which have been detected after stabilization into N2-ethyl-dG and N6-ethyl-dA, and N1-oxydiethylidene-dA, and it is known to induce DNA–protein and interstrand crosslinks (Brooks and Theruvathu 2005; Matsuda et al. 2006; Brooks and Zakhari 2014, Betsuyaku et al. 2025), yet it remains unclear if these lesions result in permanent mutations (Betsuyaku et al. 2025). Indeed, adduct formation alone might not be a reliable predictor of mutagenesis, likely due to context-dependent DNA repair, lesion tolerance, and cell-type-specific detoxification processes. Many AcA-induced modifications are bulky, transient, or repaired before replication, escaping detection by NGS, which identifies stable sequence changes (Brooks and Theruvathu 2005). For instance, interstrand crosslinks may stall polymerases and activate Fanconi anemia repair without leaving detectable mutations (Langevin et al. 2011; Garaycoechea et al. 2018). DNA adducts like N2-ethylidene-dG are unstable and may be removed physiologically, further questioning their mutagenic impact in vivo (Brooks and Zakhari 2014). EtOH and AcA mutagenicity may also be context-dependent and secondary to other stress responses (Seitz and Stickel 2007, Hoes et al. 2021). Importantly, a recent in vitro experimental study (Lozsa et al. 2026) investigated human cell lines exposed long-term to physiologically relevant, analytically validated AcA treatments. The study found no increase in single base-substitution or indel mutagenesis and no induction of the alcohol-associated SBS16 signature, similar to our in vivo results presented here. Instead, AcA caused limited structural genomic alterations, similar in type and sizes to our findings in the rat, and paralleling the 32 kb to 1 Mb deletions and duplications observed in a Japanese gastric cancer cohort, indicating that DNA breaks requiring homologous recombination may drive AcA’s carcinogenic effects. AcA may also promote carcinogenesis through non-mutagenic mechanisms, including epigenetic alterations, as it can disrupt histone modifications and DNA methylation, leading to transcriptional dysregulation without sequence changes (Varela-Rey et al. 2013; Zakhari 2013). Chronic alcohol intake can foster an inflammatory or otherwise altered microenvironment that supports selection and expansion of pre-cancerous cells (Lieber 2004; Pinyol et al. 2021). Importantly, EtOH and AcA metabolism is closely linked to oxidative stress, a key mechanism in alcohol-related carcinogenesis (Comporti et al. 2010; Cederbaum 2012, Hoes et al. 2021). As noted above, SBS17 may result from ROS accumulation due to chronic, alcohol-induced oxidative stress (Fig. 4). These findings suggest that the role of EtOH and AcA in cancer development involves a complex interplay of genotoxic and non-genotoxic mechanisms, and that mutagenicity may not be the sole or primary driver of pathogenicity. In our study, the relatively low number of mutational events observed in tumors from exposed animals, together with the lack of strong association of mutations with treatment groups, suggests that direct ethanol- or acetaldehyde-induced mutagenicity is unlikely to be the principal mechanism underlying alcohol-related carcinogenesis across multiple organs in this model, based on genome-scale analyses of SBSs, indels, CNAs, and SVs. We speculate that the consequences of EtOH/AcA exposure from alcohol drinking are multiple, combining deleterious cellular effects with broader microenvironmental changes that may not be detectable through genomic analysis using whole-genome sequencing. Such mechanisms warrant investigations in future studies. Importantly, and in keeping with other experimental results (Lozsa et al. 2026), our analysis of tumors from EtOH- and AcA-treated rats did not detect signature SBS16, previously associated with alcohol, or the alcohol- and smoking-related signatures DBS4 and ID11. This result, consistent across experimental studies, suggests that mutational processes previously correlated with alcohol consumption in human cancers, such as SBS16, are not reproduced in the rat model under the applied exposure conditions. Other mutation types, including DBS, indels, and CNAs, showed no consistent exposure-related changes, and driver gene analysis revealed only a small number of recurrent non-silent mutations. Although Tp53 and Mtor were recurrently mutated in some exposed animals, overall low rates of mutations affecting genes suggest that direct mutagenicity is not the primary mechanism of alcohol-driven carcinogenesis in this model. The relative lack of pronounced exposure-related mutation spectra and signatures points to possible species-specific differences in alcohol-related mutagenesis or a requirement for additional co-factors, underscoring the complexity of modeling alcohol-induced mutational processes. Our study also presents a technical innovation: the analyzed samples were alcohol-fixed paraffin-embedded (AFPE), unlike the formalin-fixed tissues (FFPE) increasingly more often analyzed by WGS. Alcohol fixatives better preserve tissue morphology, protein, and nucleic acids (Panzacchi et al. 2019), likely ensuring higher-quality sequencing with fewer fixation-related artifacts. In this study, we improved the detection of mutational signatures in comparison with recent FFPE-based studies, including our own (Panzacchi et al. 2019; Guo et al. 2022; Basyuni et al. 2024; Chavanel et al. 2024; Litchfield et al. 2025). This study has several limitations, including the relatively small number of tumor samples available for genomic analysis, which somewhat limited statistical power and dose–response resolution. In addition, tumors from different organs were pooled to compensate for the small sample size, although the effects of EtOH and AcA may vary by tissue. Nonetheless, this is the first investigation of alcohol exposure–specific mutational signatures or higher-level genomic alterations in a mammalian in vivo cancer model, providing novel insights into mechanisms of alcohol-related carcinogenesis. Conclusion Our results suggest a relationship between alcohol consumption and SBS17 formation, implicating damage from ethanol metabolism. The partial presence of SBS17 in exposed animals, along with the absence of presumed alcohol-linked signatures such as SBS16, indicates that in this model neither ethanol nor AcA robustly induces consistent, sequence-altering or structural mutations detectable by WGS. SBS17 formation, likely from oxidation of the free nucleotide pool, may be exacerbated in rodents by exposure-related stress and weaker repair mechanisms but due to its incomplete penetrance it is not likely to be directly attributable to AcA mutagenicity. The lack of robust exposure-related mutation spectra and mutational signatures does not imply absence of genotoxic stress but highlights the limits of mutation-centric methods. Therefore, complementary integrative approaches combining innovative macromolecule adductomics, crosslink assays, and epigenomic or microenvironment analyses are warranted in future studies. Overall, our findings propose a scenario in which directly mutagenic mechanisms play a limited role in alcohol-associated carcinogenesis, notwithstanding the debated mutagenic potential of AcA-mediated DNA damage. Supplementary Information Below is the link to the electronic supplementary material. Supplementary file1 (PDF 17.5 MB)Supplementary file2 (XLSX 88 KB) Supplementary file1 (PDF 17.5 MB) Supplementary file2 (XLSX 88 KB)