Comprehensive Genomic Analysis of Normal and Cancer Cells Elucidates the Elevated Mutation Burden in Cancer

Abstract

Self-renewing normal tissues generate several somatic mutations at each division. Previous studies have reported that cancer cells have more mutations than their normal counterparts. It is not obvious why dramatic differences in mutation burdens between normal tissues and cancers should exist. To fully understand human tumorigenesis, the increase of mutation burden in cancers will have to be understood. Here, we provided a systematic comparison of mutational burdens in normal and cancer cells from five different organs, revealing a four-fold increase of mutation burdens in cancerous vs. non-cancerous cells. Three proposed hypotheses that could account for the increased mutation burdens in cancer are: the classical hypothesis, where driver gene mutations explain the higher mutational burden; the catastrophic hypothesis, where extreme mutational events lead to large-scale genomic alterations; and the tail hypothesis, where differences in baseline mutation rates among individuals account for the differences. Testing through orthogonal observations showed that the observed medians and distributions of mutation burdens in cancers could be explained by the hypotheses to various degrees of significance, and only the tail hypothesis could easily explain the increase in median mutation burdens in the normal tissues of cancer patients compared to the normal tissues of non-cancer patients. Overall, this study characterizes an increased mutation burden across multiple types of cancer compared to normal tissue and provides insights into the contributing factors. A tenable hypothesis proposed in this study involving fundamental differences in baseline mutation rates among individuals could have implications for cancer prevention strategies.

Introduction

The advent of massively parallel sequencing has advanced our understanding of mutational burdens in cancers. Such sequencing has enabled the determination of the number of mutations in cancer cells (“mutation burdens”) because cancers are clonal, i.e., all cancer cells have the same mutations that were present in the cancer’s founder cell (1). Evidence based on those sequencing data indicated that also normal tissues accumulate a large number of somatic mutations (2). However, in contrast to cancer, healthy tissues are composed of numerous, highly divergent clones, requiring more methodologic and analytic approaches than bulk sequencing of cancer cells. Fortunately, recent technological developments have been able to quantify mutation burdens in normal tissues (3, 4). For example, sequencing of small pieces of normal tissues, such as single intestinal crypts, reveals the mutations present in the founder cells of those crypts – analogous to identifying the mutations in the founder cells of cancers.

Although it often has been observed that cancer cells have more mutations than their normal counterparts (5–8), research often stops at this observation without quantitatively addressing the magnitude and statistical distribution of these differences and their dynamics. In tissues that do not renew frequently (such as those of brain or muscle), it is obvious why cancers have higher mutation burdens than their normal counterparts. Every time a normal cell divides, it accumulates three to five new somatic mutations, so dividing cancer cells will clearly have a higher mutation burden than non-dividing cells that don’t self-renew (9–11). However, it is not intuitively obvious why cancer cells should have higher mutational burdens than self-renewing normal cells, given that many cancer cells don’t have microsatellite instability, and that several normal cell types divide as frequently as cancer cells. In the current study, we move beyond summary statistics to perform an in-depth statistical analysis of the probability distributions of mutational burdens in both normal and cancer cells in vivo in their natural circumstances rather than in in vitro or model systems. By leveraging publicly available whole-exome sequencing (WES) and whole-genome sequencing (WGS) data, we provide a distribution-based comparison that accounts for intra- and inter-individual variability. Our findings reveal that cancer cells exhibit a significantly higher mutational burden − a median of four times that of normal tissues − across several organs, with distinct statistical patterns that had not been readily apparent from previous studies. We then formulate hypotheses to explain these striking differences.

Materials and Methods

Data sources

Somatic mutation burden data in morphologically normal tissues were obtained from previously published WGS and WES studies (3, 12–21), covering seven organs: bladder, colon, esophagus, liver, lungs (bronchus), prostate, and stomach. For comparative analyses with cancer tissues, we used only clinically normal tissues from cancer-free donors (Supplementary Tables S1-S2). For other analyses – including testing hypotheses about cancer predisposition and assessing inter-individual variability – we additionally included clinically normal tissues from both cancer-free individuals and patients with cancer, as well as matched tumors (15, 21, 22). Unless otherwise specified, references to normal tissues correspond to those from cancer-free donors. Somatic mutation burden data in cancer were extracted from TCGA masked MAF files (aliquot_ensemble_masked.maf). Only primary, untreated tumor samples were used. Unless otherwise specified, we excluded cases with mismatch repair deficiency or microsatellite instability (Supplementary Table S3). Complementary analysis of COSMIC mutational signatures across the seven organs and the three tissue groups (normal tissue from cancer-free individuals, normal tissue from cancer patients, and tumors) were based on published data (14, 15, 18–21, 23, 24).

Mutation burden estimation

Mutation burden was defined as the number of SNVs per Mb. For comparative analyses with cancer tissues, raw mutation counts in normal tissues were normalized by donor age and extrapolated to age 70 (Supplementary Fig. S1). Analyses of cancerous tissues included only autosomal SNVs with variant allele frequency >5% and were restricted to patients aged 65–75 years to match the age distribution of the normal tissue datasets.

Exposure adjustment

To estimate mutational burden distributions in normal tissues representative of the general U.S. population, we adjusted for exposures to known mutagenic factors, including tobacco use (CDC Current Cigarette Smoking Among Adults in the United States [cited 2026 June 26]. Available from: https://www.cdc.gov/tobacco/php/data-statistics/adult-data-cigarettes/index.html), alcohol (CDC Percentage of Adults Aged ≥ 65 Years Who Drank Four or More Alcoholic Drinks Per Week, by Sex and Age. [cited 2026 June 26]. Available from: http://dx.doi.org/10.15585/mmwr.mm7133a5), obesity (CDC National Health and Nutrition Examination Survey. [cited 2026 June 26]. Available from: https://dx.doi.org/10.15620/cdc:106273), and infectious agents (HHS Viral Hepatitis National Strategic Plan. [cited 2026 June 26]. Available from: https://www.hhs.gov/sites/default/files/Viral-Hepatitis-National-Strategic-Plan-2021-2025.pdf), (25). Where individual-level metadata were available, we resampled to match national exposure frequencies. For underrepresented exposures, we applied multiplicative adjustments based on literature-derived cancer risk attributable to each exposure (26) (Supplementary Table S4). Resulting distributions were modeled using Gamma probability density functions. Full methodological details are provided in Supplementary Methods, Section 2.

Hypothesis modeling

We evaluated three hypotheses to explain the elevated somatic mutation burdens observed in cancer cells compared to morphologically normal cells from the same tissues. In each case, we modeled the cancer mutation burden distribution as a mathematical transformation of the corresponding normal tissue burden distribution.

Hypothesis H1: Classical hypothesis.

Increased burden results from accelerated cell division or mutation rate due to driver mutations, which we modeled through a scaling transformation of the baseline distribution: X_cancer = α(s,m) X_normal. Here, α reflects the average compounded effect of three driver mutations on the proliferation rate (effect of each driver quantified by 1 + s) and, for MSI cancer patients, on the mutation probability (quantified by 1 + m; see Section 3.1 in the Supplementary Material). This multiplicative transformation stretches the baseline distribution, increasing both the median and the variance of mutation burden in cancer.

Hypothesis H2: Catastrophic hypothesis.

A high-impact event introduces a fixed number of mutations, which we modeled through an additive shift in the baseline distribution: X_cancer = X_normal + γ. Here γ represents the number of mutations introduced by the catastrophic event. This transformation increases the median without affecting the shape or variance of the distribution.

Hypothesis H3: Tail hypothesis.

Individuals with inherently higher mutation accumulation are more likely to develop cancer, which we modeled through a two-step transformation involving weighting and scaling of the normal cell distribution. We first applied a weighting function R(x) to the probability density of the mutation burden in normal cells, P_normal (x), to reflect increased cancer risk in the upper tail of the distribution: P_weighted (x) ∝ R(x) P_normal (x). The weighting emphasizes the high-burden end of the distribution, capturing the likely increase in baseline mutation rate for cancer patients. The resulting increase in median burden in normal tissues from cancer patients is reported in Supplementary Tables S5–S6. We then applied the scaling factor α = α(s,m), as defined in H1, to account for the effect of driver mutations. The final cancer mutation burden distribution is given by: P_cancer (x) ∝ R(α⁻¹x) P_normal (α⁻¹x). Detailed methods and definitions of R(x) and α(s,m) are provided in the Supplementary Material (Sections 3.3, 4.2 and 5).

Mixture models.

A proportion p_c of cancer cells follows the catastrophic hypothesis H2, while the remaining fraction follows either H1 or H3. The resulting cancer mutation burden distribution is given by the combination: P_cancer (x) = p_c P_H2 (x) + (1 - p_c) P_Hk (x), k = 1 or 3, where P_H1 (x), P_H2 (x) and P_H3 (x) are the probability densities of mutation burden in cancer cells under each hypothesis.

Model parameters (γ for H2, 1 + s for H1 and H3, along with 1 + m for MSI patients) were fitted to observed medians (Supplementary Table S5), or, where appropriate (i.e., for s and m), inferred from prior literature and data (16, 17), (HHS Viral Hepatitis National Strategic Plan. [cited 2026 June 26]. Available from: https://www.hhs.gov/sites/default/files/Viral-Hepatitis-National-Strategic-Plan-2021-2025.pdf), (25, 27–44) (Supplementary Table S6). For the mixture models, the fraction p_c was set to the estimated proportion of cancer patients with chromothripsis (45), the parameter s was taken from the literature as above, and γ was fitted to observed medians (Supplementary Table S7).

Statistical analyses

Inter-individual variation in somatic mutation burden across normal tissues was assessed using the Kruskal-Wallis test. Comparisons between mutation burdens in normal versus cancer tissues were performed with one-sided Wilcoxon rank-sum tests, testing the hypothesis of increased mutation burden in cancer samples. Distributional differences were further evaluated using one-sided Kolmogorov-Smirnov tests. To assess model fit between observed and hypothesis-driven mutation distributions, Kolmogorov-Smirnov statistics were computed. For analyses comparing normal tissues from cancer patients and cancer-free donors, one-sided Wilcoxon rank-sum tests were applied under the assumption of higher mutation burdens in cancer patients. Intra- and inter-individual variability in mutation burden was quantified using ANOVA and linear mixed-effects models, estimating variance components attributed to donor identity, with intraclass correlation coefficients calculated to measure within-donor consistency both with and without accounting for cancer status. All statistical analyses were performed using R statistical software (version 4.4.2, RRID:SCR_001905) with key packages including tidyverse for data manipulation, rstatix for non-parametric testing, and lme4 for mixed-effects modeling. Significance was set at p < 0.05.

Results

Variation in mutational burden across normal tissues

We analyzed the probability distributions of somatic single nucleotide variants (SNVs) in normal tissues from five human organs, normalized to age 70. We focused on SNVs rather than insertions or deletions (indels) because SNVs account for over 90% of somatic mutations and are more reliably sequenced on some sequencing instruments. Data were derived from WES or WGS of DNA samples from clinically normal donors, averaging 586 samples from 29 individuals per organ (3, 12–19). These distributions reflect both intra- and inter-individual variability. Our analysis revealed large heterogeneity in somatic mutational burden among normal individuals when age normalized (Kruskal-Wallis test, p < 0.001; examples in Fig. 1A, Supplementary Fig. S2, Supplementary Table S8). Across the five organs, the average ratio between the highest and lowest median mutational burdens across donors was 8-fold, illustrating substantial inter-individual differences among healthy individuals.

Fig. 1. Probability distributions of somatic mutational burden in normal and cancer cells, measured as the number of substitution mutations per cell per megabase pair (Mb) in a 70-year-old individual.

(A) Kernel density estimates of mutational burden in colonic cells from five normal donors from the same study (17), based on an average of 48 age-adjusted samples per donor. Significant inter-donor variability was observed (Kruskal-Wallis test, p < 0.001). (B–F) Distributions of mutational burden in normal (blue) and cancer cells (red) tissues. Distributions for normal tissue follow a Gamma model fitted to datasets (3, 12–19), adjusted for donor age and exposure (smoking, alcohol, obesity, infection) via scaling or duplication. Cancer distributions are kernel density estimates from TCGA WES data for MSS patients aged 65–75. Medians, marked on each curve, reflect age-normalized datasets as detailed in Supplementary Fig. S3 and Supplementary Table S9. Sample sizes used to derive each distribution are indicated in brackets.

Mutational burden in normal vs. cancer cells

We next compared the somatic mutational burden in normal and cancer cells across the same five organs, using age-normalized SNV distributions. Cancer distributions were based on WES data from untreated primary tumors in The Cancer Genome Atlas (TCGA) and limited to tumors without mismatch repair defects. Normal tissues consistently harbored lower mutational burdens than cancers, whether quantified by average or median numbers of mutations (Fig. 1B-F, Supplementary Fig. S3). We note that mutation burdens estimated from bulk tumor sequencing primarily reflect mutations present in the most recent common ancestor of the sampled cells, with limited contribution from mutations acquired thereafter. As a result, the total mutational burden estimated in cancer cells is likely underestimated in the data. We consider median burdens to be a more conservative metric than average burdens because it minimizes outliers, such as those caused by cancers containing mutations in repair systems other than mismatch repair (e.g., MutY or unrecognized repair systems). The difference in the number of mutations between cancers and normal tissues was highly significant, with an average 4.4-fold increase in median mutation burdens (Wilcoxon rank-sum test, one-sided p < 0.001). Additionally, a randomly selected normal sample had significantly fewer mutations than a randomly selected cancer sample (Kolmogorov-Smirnov test, one-sided p < 0.001). Detailed statistical comparisons are provided in Supplementary Table S9.

Rationale behind the higher mutational burden in cancer cells

Our current understanding of mutagenesis identifies three primary causes of mutations: (a) the natural processes of aging and cell replication, (b) environmental exposures and lifestyle factors, and (c) inherited factors (27, 28, 46–48). Common cancers generally harbor at least three driver gene mutations (49) and each of these mutations in a single cancer can be the result of one or more of these three causes. To account for the difference in mutation burdens among normal and cancer cells, we advance three different hypotheses.

Hypothesis H1: Classical hypothesis.

The most straightforward explanation attributes the increased mutational burden in cancers to the effects of driver gene mutations. These mutations, by definition, directly or indirectly confer a selective growth advantage, leading to either higher rates of cell birth or lower rates of cell death. These or other somatic mutations could concomitantly produce an increased mutation rate, depending on the affected signaling pathway (50). As illustrated in Fig. 2A, when such driver events (red dots) occur within a cancer cell lineage (red line), they could result in an increase in the rate of mutation accumulation, ultimately producing an abnormally high number of somatic mutations.

Fig. 2.

(A–C) Schematic representation of somatic mutation accumulation in a single normal cell lineage (blue) and in a cancer cell lineage (red) under each hypothesis.

Hypothesis H2: Catastrophic hypothesis.

An alternative explanation attributes the increase in mutational burden to an extreme event during the course of tumorigenesis (Fig. 2B). According to this hypothesis, the mutation rate surges during a single or limited number of divisions during the cancer’s evolution. H2 is distinguished from H1 in that the increase in mutation burden is not only due to the effects of founder mutations and does not assume that the mutation rate is constant over the lifetime of the neoplasm. For example, the catastrophe could be due to high levels of reactive oxygen species in a poorly vascularized section of a tumor, with one of the impacted cells coincidentally or subsequently becoming the dominant tumor clone. This hypothesis is strongly supported by the well-documented occurrence of catastrophic events in cancer evolution, such as chromothripsis and chromoplexy (51–53).

Hypothesis H3: Tail hypothesis.

According to this hypothesis, some individuals naturally accumulate mutations at a faster rate throughout their entire lives or during specific periods, making them more susceptible to cancer (Fig. 2C). This higher rate of accumulation could be due to constitutionally higher replication error rates, such as inherited mutations in repair genes. Inter-individual differences in replication error rates have been documented (13, 14, 20, 23, 54, 55) but the importance of these to cancer incidence has been debated (except for patients with syndromic diseases such as xeroderma pigmentosum or Lynch syndrome). While germline variants and environmental exposures have important effects on somatic mutational processes, as of today they still account for only part of the observed variability. Large-scale analyses of mutational processes have shown that many mutational signatures remain of unknown origin (56), and only a limited number of inherited influences on mutation rates have been identified to date (57). Thus, a substantial fraction of inter-individual variation in somatic mutation accumulation remains unexplained by currently characterized factors. In this framework, H3 captures such variation in a statistical sense, without attributing it to a specific underlying mechanism. This hypothesis includes any additional source of mutations, including those due to environmental factors, and allows for the contribution of driver mutations, as in H1, whose effects may further amplify mutation accumulation once present. As an example, some individuals who smoke might have more mutations in their normal lung tissue than others who have smoked to equivalent extents. According to H3, the individuals who developed lung cancer would largely be those who had a higher mutation burden in their normal lung.

These hypotheses are not mutually exclusive, and the increased mutation burden observed in cancer cells is likely to result from a combination of processes acting across patients. To better reflect this biological complexity, we additionally considered mixture models combining distinct hypotheses.

Mixture model (H1 + H2 or H3 + H2).

In this framework, a fraction of cancer cells follows the catastrophic hypothesis H2, while the remainder evolves according to either the classical hypothesis H1 or the tail hypothesis H3. Because H3 can be viewed as encompassing H1 together with an additional “tail” component reflecting elevated baseline mutation accumulation, it is not necessary to consider mixtures that include both H1 and H3.

Evaluating the three hypotheses

In the following, we describe experimental observations that provide evidence for or against each hypothesis, or at least define boundary conditions that impact their likelihood.

1). Median mutation burden in cancer cells.

We first examine the mutational events required for the hypothesis-based mutation distributions to align with the actual mutation distributions. For example, a median of 3.8 mutations per megabase pair (Mb) is actually observed in colon cancer patients aged 70 years. According to the classical hypothesis H1, the selective growth advantage conferred by a driver mutation affecting cell division rate, generally denoted as 1 + s, would have to be 5.1 to account for the observed median mutation burden (Supplementary Table S5). This model assumes that s is constant over time, which is likely an over-simplification, but has the advantage that it minimizes the number of variables used for modeling. For H2 to align with the observed mutation burden at age 70, a sudden addition of 9,000 mutations would have to occur in the cancer cell through a single catastrophic event. This is roughly 3,000-fold higher than the mutations expected per division in normal cells (~ 10⁻⁹ mutations per base pair). Mutation rates in cultured cancer cells are similarly estimated to average ~ 1.6·10⁻⁹ mutations per base pair per division (58). For comparison, a sudden exposure to 7.7 Gy of X-irradiation would result in approximately 200 mutations, an order of magnitude lower than required by H2. The tail hypothesis H3 assumes that cancer patients begin with a higher baseline mutation burden in their normal tissues, with a median burden increased by a factor of 2.7 compared to non-cancer individuals, as a result of genetic or environmental factors. Under this assumption, the observed mutation burden in cancer cells from 70-year-old colon cancer patients could be explained by a selective growth advantage of 1 + s = 1.9 for driver mutations. This value is broadly consistent with estimates based on cell division rates in normal and tumor tissues, which suggest 1 + s ≈ 1.2 (Supplementary Tables S5-S6).

2). Distribution of mutation burdens among cancer cells.

Using the parameter for each hypothesis that best aligns to the median observed mutation burden, we could determine how well the distribution of predicted mutation burdens among cancers matched the observed ones. These distributions are plotted in Fig. 3A for colon cancer (actual distribution compared to the predicted distributions using fitted parameters). Though no individual hypothesis accurately recapitulated the observed data distribution, the tail hypothesis provides the best fit based on the Kolmogorov-Smirnov distance (D_KS = 0.07 for H3 vs. 0.17 for H1 and 0.25 for H2). Similar models and predictions were constructed for the other four cancer types evaluated in this study (Fig. 3B–E and Supplementary Fig. S4). In each case, the fitted parameter used to determine the predictions of each model is included in Supplementary Table S5, along with their goodness-of-fit to the observed distributions. When applying literature-derived parameters rather than fitting to the observed medians (Supplementary Fig. S5A), H3 aligned more closely with empirical data than H1 (D_KS = 0.30 vs. 0.78), though both significantly deviated from observations (p < 0.001). The literature-based parameters for H1 and H3 (not defined for H2), along with the predicted medians and the distances between predicted and observed distributions across cancer types, are detailed in Supplementary Table S6, with the associated distributions shown in Supplementary Fig. S5B–E and S6. While combining hypotheses can improve fit due to the increased number of parameters, we restricted analyses to realistic scenarios by setting the fraction of cancer cells following H2 to the known proportion of patients with chromothripsis (45), and using the literature-derived parameters for H1 and H3. In this framework, the only free parameter is the number of mutations added in a cancer cell lineage by a catastrophic event in H2, which was chosen to align the predicted median with the observed data. Using this approach, the combination H3 + H2 provides a slightly better fit across the five cancer types than H1 + H2 (mean D_KS = 0.23 vs. 0.30), while requiring a smaller, and therefore more realistic, number of mutations introduced during the catastrophic event (2.9 vs. 3.4/Mb on average). Parameter values and Kolmogorov-Smirnov distances are reported in Supplementary Table S7, with the corresponding distributions shown in Supplementary Fig. S7.

Fig. 3. (A–E) Modeling somatic mutational burden distributions in cancer cells using parameters optimized to match the observed median number of mutations.

Solid lines represent the observed probability distributions of mutational burden in normal and cancer cells from 70-year-old individuals. Dashed and dotted lines show model predictions for cancer cells under different hypotheses. Fitted parameters for each hypothesis and cancer type are provided in Supplementary Table S5.

3). Mutational burden in normal tissue from cancerous patients.

The mutation burdens in normal tissues provides another way to test the three hypotheses. The tail hypothesis predicts a higher mutational burden in the normal tissues of cancer patients compared to cancer-free individuals with similar demographics and exposures, an average 3.0-fold increase across the five organs studied. Under both the classical and catastrophic hypotheses, however, individuals of similar age, exposure, and lifestyle should exhibit comparable mutational burdens in normal tissues, regardless of whether they develop cancer. Published studies show that physiologically normal esophageal tissue from individuals with esophageal squamous cell carcinoma harbors higher mutational burdens than those in patients without esophageal cancer (15). There is in fact a 2.4-fold increase in median mutation burdens in the normal esophageal tissues of cancer patients compared to non-cancer patients (Wilcoxon rank-sum test, one-sided p < 0.001, Fig. 4). A similar pattern is observed in morphologically normal prostatic tissue from men with or without prostate cancer (21), where the median SNV accumulation rate is 2.9 times higher in cancer patients (Wilcoxon rank-sum test, one-sided p = 0.013). As depicted in Fig. 4, increases in mutation burdens are also seen in morphologically normal bladder and stomach tissues from cancer patients, but not in colon (14, 19, 20). For reference, Supplementary Fig. S8 presents matched normal and tumor tissues from the same cancer patients across colon, esophagus, and prostate (15, 21, 22). Complementary analysis of COSMIC mutational signatures across seven organs and the three tissue groups (normal tissue from cancer-free individuals, normal tissue from cancer patients, and tumors) (14, 15, 18–21, 23, 24) highlights how mutational processes vary across organs and tissue categories (Supplementary Fig. S9–S10). Cosine similarity analysis further indicates that the overall pattern of mutational signatures is largely preserved within organs, with cancer cells retaining a profile broadly similar to their tissue of origin, both from cancer patients and cancer-free individuals. Across the seven organs, the mean similarity between normal tissue in healthy individuals and cancer cells is 0.81; in the subset of organs with all three tissue types available (esophagus, prostate, and stomach), similarities are 0.86, 0.87, and 0.91 for healthy normal tissue vs. cancer cells, normal vs. cancer cells in cancer patients, and normal tissue from cancer-free vs. cancer patients, respectively (Supplementary Fig. S11–S12).

Fig. 4. Mutation accumulation rate in histologically normal tissue from cancer patients and cancer-free donors.

Rates are shown as the number of somatic mutations per cell, normalized by donor age, and measured across the exome (bladder, esophagus) or genome (colon, prostate, stomach). Boxplots display the median, interquartile range, and outliers; means are indicated by dashed lines. Sample sizes are as follows: Bladder, 483 samples from 15 donors and 72 from 4 cancer patients (20); Colon, 330 samples from 29 donors and 116 from 15 cancer patients (14); Esophagus, 106 samples from 40 donors and 47 from 19 cancer patients (15); Prostate, 6 samples from 6 donors and 30 from 24 cancer patients (21); Stomach, 79 samples from 12 donors and 138 from 17 cancer patients (19). One-sided Wilcoxon tests indicate significantly higher mutation burdens in normal tissues from cancer patients compared to cancer-free donors in the bladder, esophagus, prostate, and stomach; p-values shown reflect these tests. Sample details are provided in Supplementary Material.

4). Intra- versus inter-individual variation in somatic mutation burden in normal tissues.

In many of the published studies on morphologically normal tissues (14, 15, 19, 20), multiple samples of normal tissues from the same patient were evaluated. According to the tail hypothesis (but not the other two hypotheses) the difference in mutation burdens between any two samples of normal tissues from the same patient should be less than the difference between two samples of normal tissues randomly chosen from two different patients, whether or not these patients have cancer. This comparison is akin to measuring intra-individual variation vs. inter-individual variation and was performed for four morphologically normal tissues. These results are depicted in Supplementary Fig. S13, and show that across all tissues analyzed, mutation accumulation rates were significantly more similar between samples from the same donor than between donors (one-sided Wilcoxon rank-sum test, p < 0.001). Intra-individual consistency was particularly strong in colon and esophagus, where donor identity explained 72% and 63% of the variance, respectively, even after adjusting for cancer status (Supplementary Table S10).

Discussion

Our study had two goals. First, to provide a comprehensive analysis of the mutation burdens in normal tissues and cancers in a fashion that would allow direct comparison. This required careful attention to the technologies used to measure mutation burden among multiple different laboratories and the design of normalization procedures. Once this first goal was achieved, and it was clear that there was a striking difference in mutation burdens between normal and cancerous tissues, we attempted to disentangle the factors contributing to the increased mutation burden in cancer cells. We evaluated three hypotheses: the classical hypothesis, where driver gene mutations alone explain the higher mutational burden; the catastrophic hypothesis, where extreme mutational events lead to large-scale genomic alterations; and the tail hypothesis, where differences in baseline mutation rates among individuals, in addition to driver gene mutations, account for the differences. These hypotheses are not mutually exclusive and rather represent complementary processes which together contribute to mutation accumulation in cancer.

To test these three hypotheses, we determined the parameters necessary for each of them to account for the observed increases in the median mutation burdens of all cancer types evaluated. Using these same parameters, we then evaluated the ability of the models to explain the distributions (rather than medians) of mutation burdens among healthy individuals’ cells and among patients’ cancer cells. Finally, we evaluated mutational burdens in the normal tissues from patients with and without cancer and measured the variation in normal tissues from the same patient to provide evidence for or against each hypothesis.

We found that the observed medians and distributions of mutation burdens in cancers could be explained by the models we considered, though some performed better than others and some required parameter values that were not consistent with prior estimates. For example, for the classical hypothesis to explain the results, a higher selective growth advantage conferred by driver gene mutations would be needed than estimated by some previous studies (59–64).

An additional orthogonal test of the hypotheses showed that only one of them, i.e., the tail hypothesis, could easily explain the increase in median mutation burdens in the normal tissues of cancer patients compared to the normal tissues of non-cancer patients. The tail hypothesis is based on the idea that a small subset of individuals has a higher baseline mutation rate that predisposes them to cancer. The corresponding normal tissues from these patients would be expected to have higher baseline mutation rates than the general population. This would then lead to higher mutation burdens in the normal tissues from cancer patients compared to patients without cancer. It is difficult to reconcile this particular observation with the other two hypotheses, which posit that the mutation burdens arise through cancer-specific changes in driver genes or catastrophic events, neither of which should occur in healthy tissues.

Note that the increased baseline mutation rate invoked by the tail hypothesis could occur through two different mechanisms – through exposure of certain tissues to mutagens or through a change in replication fidelity through constitutional defects in DNA polymerization or repair. The first mechanism is not novel. For example, the concept of field cancerization from environmental factors posits that individuals predisposed to cancer often exhibit elevated mutational burdens in histologically normal tissues (65). This could be due to the direct effects of exogenous or endogenous mutagens, or increased cell division rates triggered by environmental factors such as chronic inflammation or stress. Elevated proliferation rates have also been observed in individuals with precancerous conditions that later progress to cancer (66), further supporting this connection. However, the second potential mechanism underlying the tail hypothesis is more intriguing. It posits that constitutional or inherited defects in replication fidelity accounts for many, if not the majority, of many cancer types.

Our study has several limitations. First, mutation burdens in cancers were estimated from bulk sequencing data, which primarily capture mutations present in the most recent common ancestor (MRCA) of the tumor cells, which may have existed many years ago. As a result, mutations acquired afterwards are underrepresented, and the total mutational burden of cancer cells is likely underestimated. In other words, the “red curve” in our figures is less shifted to the right than it would be if all post-MRCA mutations were included. This would further strengthen the relative support for the tail hypothesis, as the classical and catastrophic hypotheses would be even less able to account for the observed data on their own. Second, our age normalization assumes linearity of mutation accumulation with years of life, which may not be true. Although we attempt to explain our observations through three different hypotheses acting individually, the increased mutation burden in cancers is probably due to some combination of these hypotheses, with the relative importance of each hypothetical factor varying among patients. We show that simple mixture models combining these mechanisms consistent with estimates reported in the literature can reproduce the observed mutation burden distributions. Moreover, each of the models examined is purposefully simplistic as we wished to use as few parameters as possible. One critical aspect of this simplification is that we do not consider changes over time in the parameters. Similarly, we assume that there is only one major catastrophic event, but there could be many that occur during the development of a cancer.

Future research could address some of these limitations by investigating the causes of mutational burden variability, particularly through studies of younger individuals and twin cohorts to disentangle the influences of germline mutations, environmental exposures, and somatic mutation accumulation. Familial risk studies, such as those focusing on pre-malignant lesions (67), could further clarify the interplay of genetic predisposition, shared environment, and endogenous factors in shaping cancer susceptibility. Similarly, analysis of mutational signatures can also inform the etiology of these somatic mutations. In our analyses the pattern of mutational signatures seems largely preserved within organs, with cancer cells retaining a profile broadly similar to their tissue of origin, providing again support for the tail hypothesis, but further studies on larger datasets are required.

The only one of the three hypothesis that has practical implications for cancer prevention is the tail hypothesis. Reducing exposures to agents that increase mutation burdens, through direct mutagenesis or increased cell division rate, is already recognized as a key way to reduce cancer deaths (68–70). However, if most cancers of some types occur only in patients with increased baseline mutation rates as a result of inherited genetic variations, then identification of those individuals could improve cancer risk prediction and secondary prevention measures (68). The degree of variation in mutation rates among healthy individuals estimated in this study, and its later refinements, may have important implications, as previous studies suggest a nonlinear, power-law relationship between somatic mutation burden and cancer risk with, e.g., smokers have an observed 3 to 4-fold higher mutational burden associated with a more than 18-fold increase in lung cancer incidence (28, 29) (The Health Consequences of Smoking: A Report of the Surgeon General. [cited 2026 June 26]. Available from: https://www.ncbi.nlm.nih.gov/books/NBK44695).

Statement of Significance.

Comprehensive analysis of mutation burden distributions across normal and cancerous tissues reveals consistently higher burden in cancers, with a proposed explanation that may improve risk prediction and cancer prevention.

Data Availability

The data analyzed in this study were obtained from The Cancer Genome Atlas and from previously published research articles (3, 12–22, 24), as cited in the manuscript.