Environmental carcinogens often exacerbate endogenous mutagenic processes to enhance tumor promotion

SUMMARY

Environmental exposures significantly influence cancer risk, but their mutational impact remains unclear. We perform whole-exome sequencing of hepatocellular carcinomas (HCCs) from B6C3F1/N mice that arise spontaneously with age (2 years old) or following chronic exposure to one of ten potential human carcinogens that operate through genotoxic or non-genotoxic mechanisms. HCCs from mice exposed to drinking water disinfection byproducts, such as bromochloroacetic acid (BCA) and bromodichloroacetic acid (BDCA), show dose-dependent increases in mutational burden, distinct mutational signatures (BCA-mSBS12 and BDCA-mSBS25), and enrichment of the aTn→aCn mutational motif. In contrast, HCCs from other exposures, as well as from spontaneous tumors, show comparable mutational burdens, mutational signatures, and enrichment of the nCg→nTg mutational motif. These findings suggest that many environmental carcinogens promote tumorigenesis by amplifying endogenous mutagenic processes rather than initiating distinct mutational events. Our results highlight the utility of rodent models for investigating environmental carcinogenesis and provide insights relevant to human cancer risk assessment.

Graphical Abstract

In brief

Xu et al. show that mouse hepatocellular carcinomas arising spontaneously or due to exposure to various carcinogens share mutational signatures with human cancers and that the majority of environmental carcinogens act as tumor promoters by exacerbating endogenous mutational processes. They also show that some genotoxic carcinogens exhibit a dose-dependent threshold effect.

INTRODUCTION

Hepatocellular carcinoma (HCC) is the fourth leading cause of cancer-related mortality globally, with the highest incidences in East Asia and West Africa.¹ In addition to well-established risk factors, including hepatitis viruses B and C, chronic alcohol abuse, metabolic syndrome, and aflatoxin B1 (AFB1) exposure,^2,3 increasing evidence implicates environmental chemical exposures in liver tumorigenesis.^4,5

The US National Toxicology Program (NTP) has conducted more than four decades of rodent carcinogenicity studies, evaluating over 600 chemicals relevant to occupational and environmental exposures.^6,7 The B6C3F1/N mouse, a standard model in these studies, exhibits a high incidence of both spontaneous and chemically induced HCC, making it a robust system to understand the interplay between endogenous and exogenous (exposure) factors in chemical carcinogenesis.

Stages in chemical carcinogenesis, such as initiation, promotion, and progression, have been documented in various rodent models.⁸ For regulatory assessment, chemical carcinogens are broadly classified as genotoxic (GTX) or non-genotoxic (NGTX) based on their mode of action (MoA). GTX hepatocarcinogens, such as AFB1,⁹ directly or indirectly via reactive metabolites, induce DNA adducts and lesions, which result in fixed mutations after inadequate DNA repair. Conversely, NGTX carcinogens—often described as tumor promoters or selectogens—induce HCCs via diverse MoAs, including hormonal perturbation, nuclear receptor activation, persistent cytotoxicity, mitogenesis, oxidative stress, inflammation, immunosuppression, and a variety of epigenetic modifications.¹⁰ In regulatory toxicology, GTX carcinogens are typically considered to have no safe exposure threshold, whereas NGTX carcinogens are assumed to exhibit a threshold below which carcinogenic risk is negligible.

Due to the multifactorial nature of the exposome in humans, the etiologies contributing to tumor initiation and promotion are poorly understood. However, rodent tumors resulting from well-defined exposures provide a powerful model to systematically investigate links between mutational signatures and carcinogenic mechanisms. Advances in high-throughput sequencing, catalogs of mutational signatures of cancers in humans, and knowledge-driven mutational motifs from model systems provide new insights into how environmental carcinogens contribute to cancer development.¹¹

To investigate the mutational landscape (mutational burden, mutational signature, and mutational motif) and cancer driver genes due to chronic exposure to GTX or NGTX carcinogens, we performed whole-exome sequencing (WES) on B6C3F1/N mouse HCCs arising spontaneously due to age (2-year-old mice, approximately equivalent to 65-year-old humans¹²) or due to chronic (2-year) repeated dose exposure to one of 10 potential human carcinogens.

This study includes B6C3F1/N mouse HCCs resulting from chronic exposure to 4 GTX chemicals and 6 NGTX chemicals based on the Ames assay (n = 100, 10/chemical) (Tables S1 and S2) at two dose levels obtained from the NTP 2-year rodent carcinogenicity bioassays. The GTX chemicals included bromochloroacetic acid (BCA), bromodichloroacetic acid (BDCA), Ginkgo biloba extract (GBE), and primaclone (PM), and the NGTX chemicals included anthraquinone (AQ), dibutyl phthalate (DBP), oxazepam (OZ), pentabromodiphenyl ether mixture (DE-71), p-chloro-α,α,α-trifluorotoluene (PCTFT), and tris(2-chloroisopropyl) phosphate (TCPP). In addition, 21 HCCs from control male B6C3F1/N mice that arose spontaneously due to age from the corresponding studies were also analyzed. Aged-matched normal tissues from vehicle control groups (representing germline genomes) (n = 20, 10/sex) were sequenced for generating a panel of normals (PON) to identify somatic variants in the tumors (see STAR Methods; Figure S1). Finally, 20 spontaneous HCCs from control female B6C3F1/N mice were also sequenced and compared to control male mice to understand the influence of sex in spontaneous cancer.

RESULTS

Mutational burden increases with dose for some GTX carcinogens

The mouse HCC exome data identified 13,284 single-nucleotide variants (SNVs), 611 dinucleotide variants, and 357 insertion or deletion (indel) variants. Remarkably, compared to spontaneous tumors from control male mice, only HCCs from two chemical (BCA and BDCA) exposures showed significantly elevated mutational burdens (p < 0.05, Wilcoxon test) (Figure 1A). Interestingly, the mutational loads of 6 tumors from the high-dose BDCA group were much higher compared to the remaining 4 tumors from the low-dose group. In contrast, HCCs from chronic exposure to the remaining 8 chemicals did not differ in mutational load from spontaneous tumors. Spontaneous HCCs from control male and female mice also showed no significant difference, indicating no sex-based variation in overall mutational burden. Together, these findings suggest that most carcinogens studied, regardless of MoA, did not increase the mutational burden compared to tumors arising spontaneously due to age, with the exception of specific GTX carcinogens like BCA and BDCA.

Figure 1. The mutational landscape in mouse hepatocellular carcinomas arising spontaneously or due to chronic exposure to either a low or a high dose of various genotoxic or non-genotoxic carcinogens.

(A) Total counts of single-nucleotide variants.

(B) Minimum estimates of mutation load (MEML) associated with aTn mutational motifs.

(C) MEML associated with nCg mutational motifs.

(D) Mutational landscape in SBS96 format of mouse HCCs arising spontaneously (SpntF and SpntM) or due to exposure to genotoxic (BCA, BDCA, GBE, and PM) or non-genotoxic (AQ, DBP, OZ, PCTFT, DE-71, and TCPP) chemicals.

(E) Dose-dependent differences in SBS96 mutational landscape in mouse HCCs from BDCA exposure (H1-6, high dose; L1-4, low dose).

(F) Relative contributions of variant counts in transcribed strands vs. untranscribed strands from tumor samples exposed to BDCA.

(G) Log2-transformed relative contributions of 6 nucleotide changes obtained from transcribed strands vs. untranscribed strands from tumor samples from BDCA exposure. Significant strand bias is defined as a log2 ratio > 0.5 with p < 0.05, indicated by an asterisk.

Some chemical mutagens induce dose-dependent increase of mutational load associated with knowledge-defined mutational motifs

Given the dose-dependent increase in mutational burden observed with BCA and BDCA (Figure 1A), we next assessed whether known mutagenic processes were detectable in these tumors. For this purpose, we used the pattern of mutagenesis by APOBEC cytidine deminases (P-MACD) analytical pipeline designed to provide sample-specific statistical evaluation of knowledge-derived mutational motif(s) and calculate the minimum estimate of the motif mutational load caused by motif-specific mutagenesis. We identified two mutational motifs that were previously detected in human cancers.^11,13 The aTn→aCn (nAt→nGt) mutational motif is a feature of chemically induced mutations caused by lesions to adenine bases (a reverse complement motif is shown to underscore the origin from damaged adenines). It was discovered in the study of mutagenesis by a small epoxide glycidamide (GA) but was also implicated in mutagenesis with other S_N2-reacting electrophiles.¹¹ The other motif, nCg→nTg, is found ubiquitously in several cancers, can arise from the endogenous spontaneous deamination of methylated cytosines in meCpG dinucleotides, and constitutes the major component of the single base substitution (SBS1) mutational signature in the Catalogue of Somatic Mutations in Cancer (COSMIC).¹³ The P-MACD-calculated minimal estimates of mutation loads associated with aTn→aCn (nAt→nGt) and nCg→nTg showed the presence of both mechanisms in HCCs (Figures 1B and 1C; Table S3). HCCs from BCA and BDCA exposures showed dose-dependent elevation of aTn→aCn (nAt→nGt) mutagenesis, while only the low-dose DE-71 exposure showed this elevation. A previous study indicated that C→T changes in the aCn trinucleotide are predominant in human tumors¹¹; consistently, we see little evidence of aTn→aAn (nAt→nTt) and aTn→aGn (nAt→nCt) events in these samples (Figure S2). Unlike the dose-dependent load with the aTn (nAt→nGt) motif associated with only three chemicals, the mutation load of the ubiquitous endogenous mutagenesis nCg→nTg motif was prevalent in all samples of this study and showed no relation to the exposed chemicals or dosages (Figure S3). Importantly, the detection of this nCg→nTg motif within all groups indicates that the WES mutational catalogs in our study had sufficient numbers of mutational calls to detect ubiquitous mutational motifs. Consistent with previous research,¹¹ mutational loads associated with other base substitutions in the nCg trinucleotide, nCg→nAg and nCg→nGg, were not detected (Figure S4).

Mutational signatures in most chemical-induced mouse HCCs resemble spontaneous HCCs

We analyzed mutational spectra using the 96-trinucleotide substitution classification¹³ and found that most chemical-induced HCCs and spontaneous HCCs were dominated by cytosine-to-thymine (C>T) transitions (38%–45%) with notable cytosine-to-adenine (C>A) transversions (17%–23%) and thymine-to-cytosine (T>C) transitions (13%–17%), except for HCCs from BCA and BDCA exposures (Figure 1D). Interestingly, BCA-induced HCCs exhibited a distinct profile (C>T: 33%, T>C: 28%, and C>A: 12%), while BDCA-induced tumors showed predominant T>A (49%), T>C (23%), and C>T (14%) mutations. Six HCCs from the high-dose BDCA exposure group demonstrated distinct mutational spectra different from spontaneous HCCs (p <0 .05, chi-squared test) with predominant T>A mutations (46%–64%) and enriched T>C (14%–26%) and C>T (11%–13%) mutations. The rest of the 4 HCCs from the low-dose BDCA exposure group exhibited mutational spectra broadly resembling those observed in spontaneous HCCs (p > 0.05, chi-squared test) with C>T (22%–38%), T>A (17%–27%), and T>C (15%–27%) (Figure 1E), suggesting that GTX and NGTX MoAs are not mutually exclusive, i.e., lower doses of some GTX compounds might operate through an NGTX MoA.

To further explore the genomic effects of the chemicals analyzed in this study, we performed de novo mutational signature extraction using a hierarchical Dirichlet process (HDP; https://github.com/nicolaroberts/hdp) and revealed 5 trinucleotide mouse de novo mutational signatures (msig1–msig5) (Figure 2A). Based on a criterion of a cosine similarity > 0.65 when compared to known human COSMIC (v.3.2) signatures, the mouse de novo signatures decomposed into four mouse SBS (mSBS) signatures that were designated as mSBS3, mSBS5, mSBS12, and mSBS25 (which corresponded to respective COSMIC signatures with the same identification number). One of the mouse signatures with a cosine similarity < 0.65 was considered a novel signature and designated as mSBS_N1. Notably, mutational signatures mSBS12 and mSBS25 were exclusively observed in HCCs resulting from exposure to BCA and BDCA, respectively, and were thus considered exogenous signatures since they were absent in all HCCs arising spontaneously (Figure 2B).

Figure 2. Mutational landscape of mouse hepatocellular carcinomas arising spontaneously or due to chronic exposure to various genotoxic or non-genotoxic carcinogens.

(A) Cosine similarity of mouse de novo signatures with COSMIC signatures.

(B) Contribution of mSBS signatures across various chronic chemical exposures. The size of the dots matches the percentage of samples with a contribution of at least 10% from the signature. The color indicates the mean relative contribution for the samples where the signature contribution is ≥10%.

(C) Relative contribution of different SBS signatures in mouse HCCs across various chronic chemical exposures.

(D) Dose response of aTn mutational motif load in mSBS25 activity in BDCA.

(E) Dose response of aTn mutational motif load in mSBS12 activity in BDCA.

(F) Dose response of aTn mutational motif load in mSBS12 activity in BCA.

Since signatures mSBS3, mSBS5, and mSBS_N1 were found in the spontaneous tumors (control male and/or female), they are most likely of endogenous origin. All HCCs arising spontaneously (from control male and female mice) or due to chronic chemical exposure exhibited the mSBS5 signature. SBS5 has been defined as a “clock-like” signature and has previously been reported in most cancers and non-tumor cells.¹⁴ In addition, the mSBS3 signature was only identified in spontaneous HCCs from control female mice. mSBS_N1 was surprisingly absent in spontaneous HCCs from control male mice but was present in spontaneous HCCs from control female mice and all chemical exposures except PM, TCPP, and DBP. mSBS_N1 showed the highest similarity to SBS40 (a clock-like signature) but with a cosine of only 0.53. To further explore the mutational signatures across all mouse HCCs, we clustered individual tumors based on their signature profiles (Figure 2C). We noted that 6 HCCs from the high-dose BDCA exposure clustered distinctly from the other samples (data not shown). Interestingly, HCCs from the low-dose BDCA exposure, though with lower mutational burden, still retained mSBS25 at varying fractions. Overall, HCCs arising spontaneously resembled other HCCs arising due to various chemical exposures and primarily exhibited SBS5, suggestive of a clock-like endogenous signature.

Absence of distinct dinucleotide or indel mutational signatures in HCCs arising spontaneously or due to chronic chemical exposures

Regardless of whether HCCs arose spontaneously or due to chronic chemical exposures, dinucleotide mutations had a lower frequency compared to SNVs, with the mutation load of dinucleotides (Figure S5A) and the proportion of each dinucleotide mutation type (Figure S5B) very similar across all HCCs. The number of small indel mutations (usually between 1 and 50 bp) also showed no significant difference across all HCCs, either arising spontaneously or due to chemical exposures (Figure S5C). Because of the limited number of dinucleotide mutations and small indel mutations, no doublet base substitution signatures (mDBSs) or indel signatures (mIDs) were identified in this study.

Driver gene landscape in spontaneous and chemical-induced HCCs

In general, tumor occurrence is thought to require as many as 2–8 so-called driver gene mutations, as well as numerous accompanying passenger gene mutations.^15,16 Using defined criteria outlined in the STAR Methods, we have identified 21 putative driver genes (Figure 3), of which Hras, Ctnnb1, Npm1, Arid1b, and Muc4 were cataloged in the COSMIC cancer gene census. Based on the mutation frequency and predicted functional impact of somatic mutations in these genes, six driver genes (Sfr1, Ube2c, Hras, Muc4, Ctnnb1, and Flg2) were identified (Figure S6). Most HCCs analyzed harbored mutations in more than two of these candidate driver genes. Notably, Sfr1 was recurrently mutated across the majority of the HCCs, with a predominant T>G transversion localized to chromosome 19, position 47732754 (Figure S7).

Figure 3.

Putative cancer driver genes in mouse HCC arising spontaneously or due to chronic chemical exposures

We further examined the associations between driver genes and specific chemical exposures or sex differences (Table S4). Compared to HCCs arising spontaneously in control male mice, HCCs from control female mice exhibited significantly fewer mutations in candidate driver genes Flg2, Sfr1, and Muc4 (p < 0.05, Fisher’s exact test). Consistent with previous reports,¹⁷ Hras—a key activator of the mitogen-activated protein kinase/extracellular signal-regulated kinase (MAPK/ERK) pathway—was preferentially mutated in spontaneous HCCs. In contrast, HCCs resulting from chronic exposure to DE-71, GBE, OZ, and PCTFT had fewer Hras mutations (p < 0.005, Fisher’s exact test). Regardless of the MoA, the average numbers of driver genes identified in all male HCCs were similar (2.6–3.4 in GTX chemical HCCs; 2.6–3.8 in NGTX chemical HCCs vs. 3.38 in spontaneous HCCs).

Linkage of mouse HCC mutational signature with human cancers and other chemical exposures

To assess the relevance of mouse HCC mutational signatures to human disease, we applied the SignatureEstimation (v.1.0.0) algorithm to a comprehensive dataset comprising 4,645 whole-genome and 19,184 whole-exome sequences from human tumors (https://www.synapse.org/#!Synapse:syn11801889) (Figure 4A). This analysis revealed an association of mSBS5, mSBS12, and mSBS25 with liver tumors in humans (Figures 4A and 4B; Table S5). The etiologies of SBS12 and SBS25 are not well understood. SBS12 has been associated with hepatocellular tumors¹³ and correlated with signatures from styrene oxide exposure.¹⁸ SBS25 has been identified in Hodgkin’s lymphoma cell lines¹⁹ and has been reported to be potentially associated with chemotherapy.²⁰ Additionally, SBS25 also correlated with the GA signature,²¹ a cytochrome P450 2E1-catalyzed epoxide metabolite of acrylamide.

Figure 4. Comparison of mSBS signatures to human cancers.

(A) Identification of mutation signature from cancers in humans related to mSBS signature.

(B) mSBS12 and two mutational spectra of hepatocellular carcinomas in humans.

(C) mSBS25 and two mutational spectra of hepatocellular carcinomas in humans.

We have previously found that the correlation between mutational signatures and estimates of mutation load associated with mutational motifs can reveal biological mechanisms underlying mutagenesis.¹¹ Therefore, we correlated sample-specific activities of msig1 (mSBS3), msig2 (mSBS5), msig3 (mSBS25), msig4 (mSBS-N1), and msig5 (mSBS12) to the estimates of mutation load associated with aTn→aCn and nCg→nTg motifs generated by the P-MACD pipeline. There was no significant correlation between the endogenous nCg→nTg mutation load and any of the mSBSs (Figure S8). However, the mutation load associated with a ubiquitous motif of chemical mutagenesis, aTn→aCn, showed significant correlations with mSBS25 and mSBS12 from BDCA exposure and with mSBS12 activity from BCA exposure (Figures 2D–2F and S9; Table S3).

DISCUSSION

Although some agents, such as benzo(a)pyrene (BaP) and dietary contaminants (aflatoxins and ochratoxins),²² have been associated with an increased risk for human HCCs, establishing the translational relevance of numerous MoAs/chemicals resulting in mouse HCCs to human health is still challenging. Our analysis of mutational patterns and driver genes in B6C3F1/N mouse HCCs arising spontaneously due to aging or due to various chronic chemical exposures demonstrated that BDCA- and BCA-induced HCCs exhibit distinct mutational signatures and dose-dependent mutational burdens. HCCs from BCA and BDCA exposures exhibited mSBS12 and mSBS25 signatures, respectively. Incidentally, Riva and colleagues using whole-genome sequencing (WGS) on mouse HCCs from the same tumor archives also determined that BCA shows the mSBS12 signature.²³ The COSMIC database attempts to link each of these signatures from various cancers to their respective underlying mechanisms. For example, human liver cancers show signatures SBS1, SBS4, SBS5, SBS6, SBS12, SBS16, SBS17, SBS22, SBS23, and SBS24. SBS1, SBS5, and SBS6 are reflective of endogenous clock-like signatures or defective DNA mismatch repair, while SBS4, SBS22, and SBS24 are related to exogenous exposures to BaP, aristolochic acid (AA), and AFB1, respectively. However, the underlying etiologies related to SBS12, SBS16, SBS17 and SBS23 are unknown. mSBS12 has been reported in mouse HCCs, resulting from chronic exposures to BCA (this study and Riva et al.²³), cumene, and trichloropropane, suggesting a conserved mutagenesis mechanism related to these chemicals. Both BCA and trichloropropane are drinking water contaminants, and about 20% of human HCCs report SBS12, supporting a potential hypothesis for further research. Also, the presence of the mSBS25 signature in mouse HCCs from BDCA exposure provides an opportunity to better understand the underlying mechanism related to SBS25 that has previously been related to chemotherapy.

Mutational motif-centered analysis of WES/WGS data in conjunction with agnostic signature extraction can provide testable hypotheses about the sources and mechanisms of mutagenesis in cancers and non-cancerous human tissue.^11,24,25 Here, we have shown the correlation of mutation load associated with aTn→aCn (nAt→nGt) motif mutation with activities (i.e., mutation load) associated with mSBS12 for BCA and mSBS12, as well as mSBS25 for BDCA, suggesting a chemical-specific mutagenic mechanism. Further evaluation of these signatures using defined exposures exploiting appropriate in vitro/in vivo models may validate these findings.

BDCA and BCA are haloacetic acid byproducts in the process of drinking water disinfection, which are listed as reasonably anticipated to be human carcinogens by the NTP’s Report on Carcinogens (RoC).^26,27 The potential MoAs of hepatocarcinogenesis of haloacetic acids include oxidative stress resulting in increases in 8-hydroxydeoxyguanosine (8-OHdG) adducts, DNA damage, PPAR-α activation, and cytotoxicity with regenerative proliferation.^28,29

In this study, BDCA and BCA exposures produced dose-dependent mutagenesis with an aTn→aCn (nAt→nGt) motif that is observed in all types of human cancers and is associated with copying of damaged adenines by error-prone translesion polymerase Rev1.¹¹ In addition, the mutation spectrum from BDCA exposure showed remarkable preponderance of T→A (A→T) mutations, which are similar to bulky adenine adducts from aristolactam nitrenium ions that are metabolites of AA.^30,31 We performed strand bias analysis to evaluate the balance of detected variants obtained from transcribed strands vs. untranscribed strands. T:A is significantly disproportional between two strands from the BDCA-induced tumors (Figure 1G). Strand biases (significant strand bias was defined as a log2 ratio > 0.5 with a statistical significance p < 0.05, indicated by an asterisk) were also observed in tumors exposed to other chemicals but not in tumors arising spontaneously (Figure S10). Thus, adenine lesions could be a prominent component of carcinogenic mutagenesis by BDCA and BCA. Notwithstanding specific mutagenic mechanisms, dose-related early incidences of BDCA-induced HCCs²⁸ suggest that there may be potential distinct initiating events. The dose-related differences in the mutational landscape in tumors from BCA and BDCA exposures in our study showed that different MoAs are likely operating across the dose spectrum. Also, delineation of dose-dependent mutational signatures may be possible, especially in the case of strong mutagens.^28,29

HCCs from the other two GTX agents, GBE and PM, did not show a higher mutation burden or distinct mutational signatures in our study. GBE was reported to induce DNA damage in cultured human hepatic cells partially via Topoisomerase II inhibition³² as well as by chronic activation of various nuclear receptors.^33,34 PM is partly metabolized to phenobarbital (an IARC group 2B carcinogen), a NGTX carcinogen that has been found to induce Cytochrome P450 enzymes (CYP2B1 and CYP2B2) and enhance cell proliferation and tumor promotion.³⁵ In spite of their mutagenicity in the Ames assay, these data suggest that the mechanisms of hepatocarcinogenesis of GBE and PM may be similar to those of other NGTX carcinogens.

The risk associated with chemical toxicity and carcinogenicity hazards is important in regulatory toxicology. The classification of chemicals by genotoxicity assays into GTX and NGTX based on MoAs provides a logical and actionable framework for cancer risk assessment, with the assumption that a threshold dose exists for NGTX carcinogens. Intriguingly, two Ames test-positive chemicals in our analysis (GBE and PM) did not exhibit a specific mutational pattern or a higher mutation burden compared to Ames test-negative chemicals. One possible explanation might be the extremely high concentration of some chemicals used in the in vitro bacterial mutation assays and relatively lower concentrations used in the in vivo mouse carcinogenesis studies,³⁶ which supports the concept of a carcinogenic threshold for genotoxicants.^37,38 Sometimes, carcinogens that are negative in bacterial mutation assays yield positive results in the in vivo transgenic rodent gene mutation assay, mainly due to the generation of intermediate GTX metabolites.³⁷ Also, the mutational load data suggest that a carcinogen like BDCA at high doses may be operating via a GTX mechanism but might be functioning via a NGTX mechanism at lower doses. Our study suggests that the current risk assessment paradigm of chemical carcinogens might be challenged, especially when a chemical is shown to be a genotoxicant in guideline in vitro genotoxicity assays but may actually cause tumors via a NGTX MoA, with the potential for a threshold effect at a lower dose.

Possibly due to differences in genetics, environmental exposure, or lifestyle, sex differences have been observed in the prevalence of human liver cancer, and CTNNB1 has been identified as a sex-specific index for the prediction and evaluation of treatment in patients with hepatocellular cancer.^39,40 Our study also uncovered sex differences in specific driver genes, such as Flg2, Sfr1, and Muc4, and in genome-wide phenomena, such as mSBS3, which was noted only in spontaneous female mice, albeit only in a few tumors. These sex biases suggest differences in the origins and progression of tumors between male and female mice, which may be influenced by different endogenous factors.

In summary, the GTX and NGTX chemicals analyzed in this study are environmental exposures with relevance to public health. Using mutational signature analysis and mutational motif enrichment analysis, this study demonstrates that rodent HCCs exhibit a similar mutational landscape to human tumors, and most environmental agents (especially at low doses) act as promoters of endogenous mutagenic processes upon chronic exposure, which are also likely to occur in humans. Additionally, this study also shows that the dose of exposure might influence the MoA of some GTX carcinogens, implying a carcinogenic threshold dose. Furthermore, examination of rodent tumors from defined chemical exposures may provide mechanistic insights into COSMIC signatures with unknown mechanisms, as shown for SBS12 in this study. Future rodent cancer genomic studies with additional chemical carcinogens with defined MoAs would provide mechanistic knowledge linking mutational signatures and environmental carcinogenesis.

Limitations of the study

The absence of matched genomic controls for each tumor may contribute to potential germline contamination. This was partially mitigated by generating a panel of normals (PON) from age-matched controls to filter out germline variants and setting stringent filtering criteria to minimize false positives in somatic variant calling in tumors. While this strategy helped mitigate germline interference, it also resulted in lower mutation counts, which limited the robustness of mutational signature extraction and necessitated the use of a relatively low cosine similarity threshold (>0.65) for COSMIC signature matching. Despite these limitations, our mutational signature results were consistent with findings from the whole-genome analysis of mouse tumors resulting from similar exposures,²³ supporting the biological validity of our approach. Finally, the availability of time-course samples would have strengthened the causal relationships between SBS12/SBS25 and the corresponding biology related to adduct formation, initiating mutations, and tumor promotion.

RESOURCE AVAILABILITY

Lead contact

Requests for further information and resources should be directed to and will be fulfilled by the lead contact, Arun R. Pandiri (arun.pandiri@nih.gov).

Materials availability

This study did not generate new unique reagents.

Data and code availability

• The sequencing raw data have been deposited at the SRA with a submission ID: PRJNA779957 (molecular analysis of mouse HCC tumors).

• All the codes used in this manuscript are indicated in the key resources table. No new code was generated in this manuscript.

• Any additional information required to reanalyze the data reported in this paper is available from the lead contact upon request.

KEY RESOURCES TABLE

REAGENT or RESOURCE	SOURCE	IDENTIFIER
Biological samples
Retrospective fresh frozen (FF) samples from the B6C3F1/N mice ~2 years of age from the archives. n = 131/male; n = 30/female	National Toxicology Program’s (NTP) tissue archives	N/A
Critical commercial assays
Gentra Puregene Tissue Kit	Qiagen	Cat#158066
SureSelectXT Mouse All Exon, 96 kit	Agilent	Cat #5190-4642
Deposited data
Raw and processed Exome seq data	Sequence read Archive (SRA)	https://www.ncbi.nlm.nih.gov/sra/?term=PRJNA779957
Software and algorithms
FastQC(V0.12.1)	https://www.bioinformatics.babraham.ac.uk/projects/fastqc/	N/A
Cutadapt(v4.5)	https://cutadapt.readthedocs.io/en/stable	N/A
BWA-MEM(V0.7.15-r1140)	https://bio-bwa.sourceforge.net/bwa.shtml	N/A
GATK(v4.2.4.0)	https://gatk.broadinstitute.org/hc/en-us	N/A
SnpEff(v5.1d)	https://pcingola.github.io/SnpEff/	N/A
MutationPatterns(v3.14.0)	https://bioconductor.org/packages/release/bioc/manuals/MutationalPatterns/man/MutationalPatterns.pdf	N/A
HDP(v0.1.5)	https://github.com/nicolaroberts/hdp	N/A
P-MACD	https://github.com/NIEHS/P-MACD	N/A
SignatureEstimation (v.1.0.0)	https://www.ncbi.nlm.nih.gov/CBBresearch/Przytycka/software/signatureestimation/Vignette.html	N/A
Dndscv(v0.0.1.0)	https://github.com/im3sanger/dndscv	N/A

STAR★METHODS

EXPERIMENTAL MODEL AND STUDY PARTICIPANT DETAILS

Fresh frozen (FF) samples preparation

All samples used in this study were obtained from the National Toxicology Program’s (NTP’s) tissue archives and no additional animal studies were conducted for this project. All animal studies that contributed tissues to the NTP archives were conducted in animal facilities fully accredited by the Association for the Assessment and Accreditation of Laboratory Animal Care International. These studies were approved by the respective contract research laboratories’ Animal Care and Use Committees and conducted in compliance of all relevant NTP animal care and use policies and applicable federal, state and local regulations and guidelines. In general, from each study, the B6C3F1/N mice were euthanized at ~105 weeks and extensive sample collection was done according to the NTP study protocols. Fresh frozen (FF) samples from chemically-induced HCCs, spontaneous HCCs and age-matched normal livers from vehicle control groups were used in this study. More details, including the doses of chemicals used and administration routes, are provided in Table S1. Liver tumors larger than 0.5 mm were flash frozen in liquid nitrogen and stored at −80°C until DNA isolation. The microscopic morphology of liver tumor samples was confirmed independently by two board certified veterinary pathologists (R.C.K. and A.R.P.). Details of each tumor sample are provided in Table S2.

METHOD DETAILS

DNA extraction and exome sequencing

Genomic DNA was extracted using Gentra Puregene Kit (Qiagen, Valencia, CA) using standard procedures.

Exome sequences of 141 tumor samples and 20 normal age-matched samples were enriched from the genomic libraries according to the manufacturer’s protocol with the SureSelect Mouse Exome kit (Agilent, Santa Clara, CA). Sequencing was performed on the Illumina HiSeq 2500 according to the manufacturer’s protocol generating 151 base pair paired-end reads.

QUANTIFICATION AND STATISTICAL ANALYSIS

Somatic variant calling

Paired-end whole exome sequencing data were assessed with FastQC (https://www.bioinformatics.babraham.ac.uk/projects/fastqc/) for quality control, and sequencing adapters were trimmed with Cutadapt (https://cutadapt.readthedocs.io/en/stable/) when necessary. The pre-processed short reads were aligned with BWA-MEM (https://bio-bwa.sourceforge.net/bwa.shtml) against reference mouse genome (mm10). Alignment bam files were further processed to retain high quality alignment results and passed through the base calling quality recalibration by GATK (https://gatk.broadinstitute.org/hc/en-us). Variant calling was accomplished by Mutect2 in the tumor-only mode, against an in-house gender specific panel of normals (PON) obtained from male and female normal control samples respectively. During the variant calling, we also included dbSNP (build 146) as the germline known variant source and guided the caller to limit the genomic region at vendor provided captured regions with 100 bp up/down stream flanking regions padded, and the minimum base quality score was kept at 30 and above. Variant calling results were filtered using “FilterMutectCalls” module within GATK with the default parameter to retain “passing” variants.

A separate set of variants were prepared from whole genome sequence data obtained from inhouse normal mice by samtools and haplotypercaller and was used as an internal germline variant panel. The resulting SNV, Doublet mutation and short INDEL were retained separately for further analyses. All variant annotation was carried out by SnpEff (version 4.3i).

Mutational spectrum generation and mutational signatures extraction

For each chemical induced tumor or spontaneous tumor sample, six types of base substitutions and the mutation context with flanking nucleotides were extracted with functions from the MutationPatterns R packages, the mutational spectrum of each sample was examined. De novo substitution signatures were extracted using the HDP package (https://github.com/nicolaroberts/hdp) with the parameters recommended by the authors. To compare to human cancer mutational signatures in the COSMIC database, mouse de novo signatures were normalized by the factors suggested by,²³ and compared to all available human cancer mutational signatures using cosine correlation. When the cosine correlation was greater than 0.6, human signature number was used with “m” as prefix; when the cosine correlation was less than 0.6, it was named as a “novel” mouse signature.

Revealing mutagenesis associated with knowledge-derived trinucleotide motifs

For each chemical induced tumor or spontaneous tumor sample, the SNV catalog was imported into P-MACD (https://github.com/NIEHS/P-MACD)^25,41 to estimate the enrichment and minimal estimate of mutational load of the mutation in aTn (to aCn (major), aAn, and aGn) and nCg (to nTg (major), nAg, and nTg). The enrichment of a mutational event (e.g., aTn→aCn is defined as the ratio between the specified mutational motif over random mutagenesis in the surrounding embedded sequence (set to ±20 bases from the mutated base). In addition to the conventional representation of trinucleotide mutational motifs in dsDNA, reverse complements were also considered, and each motif was represented by a pyrimidine nucleotide of a base pair. The IUPAC codes were used to represent ambiguous nucleotides. The mutated residue in the trinucleotide motif is in upper cases. The equation to calculate enrichment of a trinucleotide motif is provided below using the aTn→aCn event as example.

$$Enrichmen{t}_{aTn\to aCn}=\frac{(Mutation{s}_{aTn\to aCn})\times (Contex{t}_t)}{(Mutation{s}_{T\to C})\times (Contex{t}_{aTn})}$$

Mutations that happen at less than 10 bases apart from the mutated trinucleotide were excluded from the enrichment calculation as these “complex” mutations likely arise from the activity of translesion polymerases and may confound the analysis. A Fisher’s exact test was applied wherein the ratio of the number of mutations within the trinucleotide motif (Mutations_aTn→aCn) and those that do not conform to the trinucleotide motif (Mutations_T→C – Mutations_aTn→aCn) was compared to the number of bases in the context that either were in the trinucleotide motif (Context_aTn) versus those that were not in the motif (Context_t – Context_aTn). The false-discovery rate (FDR) of the p-values generated for each sample was controlled by the Benjamini-Hochberg procedure. For samples with the enrichment value larger than 1 and the corrected p-value lower than 0.05, the minimal estimate of mutation load (MEML) associated with motif-preferring mutagenic mechanism was calculated using the following equation with the aTn→aCn event as an example.

$$MEM{L}_{aTn\to aCn}=\frac{(Mutation{s}_{aTn\to aCn})\times (Enrichmen{t}_{aTn\to aCn}-1)}{(Enrichmen{t}_{aTn\to aCn})}$$

For samples without statistical significance in the FDR-adjusted Fisher’s exact test, the motif-specific mutation load was defined as 0. MEML can also be referred as motif mutational load in figures and throughout the manuscript.

Correlation analysis between SBS activities and the mutational load of motifs

The activities of de novo mouse signatures can be estimated by multiplying the relative contribution (fraction) of each signature to the total SNVs. The signature activities were used for the correlation analysis against the minimum estimates of mutation load of the two major mutational events, aTn→aCn and nCg→nTg. The correlation analysis was done on HCCs arising spontaneously or due to chemical exposures using Spearman’s rank correlation adjusted for the FDR by the Benjamini-Yekutieli procedure.

Driver gene detection

To identify cancer driver genes in mouse hepatocellular carcinomas in our study, we used dNdScv,⁴² a maximum-likelihood based method to quantify the dN/dS ratios of mutations obtained from variant calling. This method allowed identification of non-synonymous substitutions, including missense, nonsense and essential splice site substitutions. Genes with high recurrent synonymous mutations and indels were identified using a negative binomial background model.^42,43 The mutation annotations were carried out within the mm10 refseq framework. Only variants that underwent positive selection (dNdScv p < 0.05) and located in the coding regions were considered for further refinement. Additional criteria used for candidate driver gene selection include: first, all genes should be recurrently mutated genes (occurrence greater than or equal to 2); second, all genes harboring the mutations are predicted to affect protein function encoded by the gene (moderate or high putative impact predicted by SNPEff); third, all genes should be known cancer driver genes based on a mouse database (the Sleeping Beauty Cancer Driver Database (SBCDDB, http://sbcddb.moffitt.org) and the Candidate Cancer Gene Database (CCGB, http://ccgd-starrlab.oit.umn.edu/)); fourth, all genes harbored at least one mutation with higher cancer cell fraction (CCF) reflecting the early presence of the mutation in tumor evolution (>30% CCF).

Relationship of mouse signatures in the human mutational catalogue

In addition to using the cosine similarity for the de novo mouse mutational signature to human COSMIC signature database, we also evaluated them against the variants reported by the Pan Cancer analysis working group^13,44 from a vareity of cancer types that included renal cell carcinoma, hepatocellular carcinoma and alveolar/bronchiolar adenocarcinoma, etc. We obtained a collection of 4,645 variants from whole-genome sequences and 19,184 variants from whole-exome sequences (https://www.synapse.org/#!Synapse:syn11801889) and used SignatureEstimation (v.1.0.0) to estimate the significant exposures of the five de novo signatures in relation to human cancer type reported by.²³ We selected the tumor samples having a minimal contribution level of 5% of at least 1 of the 5 signatures from our study and reported the significant association.

Chemical association or gender association with driver mutation

To evaluate whether driver gene mutation are associated with exposure, we tested for each driver the balance between chemical exposed samples harboring the mutation vs. those that did not harbor such a mutation against the proportion of those in the spontaneous male samples. To evaluate whether driver gene mutations are associated with gender, we tested for each driver the balance between spontaneous female samples harboring the mutation vs. those that did not harbor such a mutation against the proportion of those in spontaneous male samples. We used the Fisher’s exact test to determine if the proportions differ significantly from what is expected. A nominal p-value was used to indicate the statistical significance.

Dose effect of chemical exposure

To evaluate the dose effect of the chemical exposure, we selected three chemicals and tested dose level association compared to the spontaneous tumors. We collected the six nucleotides mutation count without respect to their flanking nucleotides, for each of the three chemical exposed groups, and we retained high-level exposed samples separated from low-level exposed samples. We performed a Chi-square tests for each group/per chemical compared to the spontaneous tumor samples. A nominal p-value was used to indicate the statistical significance.

Strand bias analysis

To evaluate the balance of detected mutants obtained from transcribed strand vs. untranscribed strand, we used the method in MutationalPattern R package to tally 96 trinucleotides counts from both transcription and replication strand. The ratio between variants from the transcribed and untranscribed strands was tested in HCCs arising spontaneously or due to chemical exposures. Significant strand bias is defined as a log2 ratio >0.5 with statistical significance p-value <0.05, indicated by an asterisk.

Other statistical analysis

Most statistical analyses were conducted with the open-source program R or Microsoft Excel. Principal component analysis (PCA) was conducted with Partek (Partek Inc., Chesterfield, MO).

Supplementary Material

Supplemental information can be found online at https://doi.org/10.1016/j.celrep.2025.115978.

Highlights.

• Many environmental carcinogens amplify endogenous mutational processes

• Exacerbation of endogenous mutagenic processes may contribute to tumor promotion

• Genotoxicity of some mutagenic agents appears to be dose dependent

• Carcinogen-specific mutational signatures in mice are also noted in human cancers