When we sequence a tumor genome, we typically look for "driver" mutations—the specific broken genes that cause uncontrolled cellular growth. However, every genome also contains thousands of "passenger" mutations. These mutations do not drive the cancer, but they are far from useless. They carry a historical record of the precise chemical and biological processes that damaged the cell's DNA over its lifetime.
By analyzing the specific patterns of these mutations, researchers can perform a kind of molecular archaeology, uncovering whether a tumor was caused by environmental carcinogens, inherited DNA repair deficiencies, or normal cellular aging. This framework is known as mutational signature analysis.
The Genome as a Chemical Canvas
To understand why mutational signatures work, we must look at mutations not as abstract letters changing in a text, but as the outcomes of specific chemical reactions. A mutagen—whether it is ultraviolet (UV) light, tobacco smoke, or an endogenous cellular enzyme—does not damage DNA entirely at random. It interacts with specific nucleotides situated within specific local sequence contexts.
In their seminal study, Signatures of mutational processes in human cancer, Ludmil B. Alexandrov et al. formalized a classification system based on this local context. They focused on single-nucleotide variants (SNVs) and looked at the immediate 5' and 3' flanking bases of each mutation.
There are six possible base substitutions if we account for DNA complementarity: $\text{C}\rightarrow\text{A}$, $\text{C}\rightarrow\text{G}$, $\text{C}\rightarrow\text{T}$, $\text{T}\rightarrow\text{A}$, $\text{T}\rightarrow\text{C}$, and $\text{T}\rightarrow\text{G}$. For each of these six types, there are 4 possible upstream bases and 4 possible downstream bases. Multiplying these together ($6 \times 4 \times 4$) yields 96 distinct mutational channels.
Flanking 5' Base Target Mutation Flanking 3' Base
[A, C, G, T] ---> C -> T ---> [A, C, G, T]
Every sequenced tissue sample can be represented as a discrete probability distribution across these 96 channels—a unique "mutation spectrum."
The Logic of Matrix Deconvolution
A single tumor is rarely the product of just one mutational process. Throughout its lineage, a cell might accumulate mutations from normal aging, then lose a critical DNA repair mechanism, and simultaneously be exposed to an external carcinogen. The observed 96-channel spectrum of a tumor is therefore a composite mixture of multiple overlapping processes.
To isolate these processes, researchers treat the problem as a form of blind source separation, akin to separating individual voices from a recorded room conversation. As explained by Susanne Gröbner et al. in The landscape of genomic alterations across childhood cancers, mathematical models can decompose these complex mutation patterns into their constituent parts.
Mathematically, if we have a mutation count matrix $V$ (where rows represent the 96 channels and columns represent different patient samples), we want to factorize it into two simpler matrices, $W$ and $H$:
$$V \approx W \times H$$
- Matrix $W$ (The Signatures): Represents the characteristic 96-channel shape of each mutational process. These shapes are highly conserved across different patients and tissue types.
- Matrix $H$ (The Exposures): Represents the "volume" or activity level of each signature within an individual patient's sample.
When analyzing large cohorts, unsupervised algorithms like Non-negative Matrix Factorization (NMF) are used to discover these signatures from scratch. However, when dealing with small cohorts or individual clinical cases where unsupervised discovery lacks statistical power, Gregor Warsow et al. note in Genomic features of renal cell carcinoma with venous tumor thrombus that researchers can pivot to supervised mutational signature analysis. This approach takes a predefined catalog of known signatures and uses optimization techniques—such as lasso-based regression—to find the linear combination of signatures that best explains the patient's observed spectrum, a method utilized by Andrew J. Armstrong et al. in Molecular medicine tumor board: whole-genome sequencing to inform on personalized medicine for a man with advanced prostate cancer.
Distinguishing Endogenous Clocks from Exogenous Scars
Deconvolving these profiles allows clinicians and researchers to clearly differentiate between endogenous (internal) and exogenous (external) sources of genomic instability.
Endogenous Clocks and Enzymatic Failures
Some signatures act as steady, internal cellular clocks. For example, Srinivasa R. Rao et al. in Whole-genome sequencing of bladder cancers reveals somatic CDKN1A mutations and clinicopathological associations with mutation burden demonstrated that many tumors are dominated by the spontaneous deamination of methylated cytosines. This is a natural consequence of biological aging that leaves a highly specific trail of $\text{C}\rightarrow\text{T}$ transitions at CpG dinucleotides.
Other internal signatures reveal cellular machinery gone rogue. The APOBEC family of cytidine deaminases normally aids in antiviral defense by mutating viral genomes, but when misregulated, it attacks the host genome instead. In Genomic characterization of tumor mutational burden-high breast carcinomas, Theodore Vougiouklakis et al. used signature deconvolution to show that tumors with exceptionally high mutation burdens are frequently driven by hyperactive APOBEC mutagenesis rather than random background damage.
Similarly, Youngwook KIM et al. in Integrative genomic analysis of salivary duct carcinoma leveraged signature decomposition to identify homologous recombination deficiencies (known as "BRCAness") in tumors that lacked inherited driver mutations, opening up unexpected avenues for targeted therapies like PARP inhibitors.
Exogenous Scars
Environmental carcinogens leave entirely different chemical footprints. In Dissecting the Mutational Landscape of Cutaneous Melanoma: An Omic Analysis Based on Patients from Greece, Georgia Kontogianni et al. mapped the exomes of melanoma patients and found a stark enrichment of $\text{C}\rightarrow\text{T}$ transitions occurring specifically at dipyrimidine (CC or TC) sites. This pattern is the unmistakable molecular scar of ultraviolet (UV) radiation damage, confirming the exogenous etiology of the disease at the single-nucleotide level.
Mapping the History of Development and Disease
Because mutational signatures accumulate continuously, they provide a temporal record of a tissue's life history. Mutations that occur early in embryonic development will be shared across many somatic cells, while mutations occurring late in life will be restricted to small sub-clones.
This principle allows researchers to study genetic mosaicism in healthy individuals. In Postzygotic single-nucleotide mosaicisms in whole-genome sequences of clinically unremarkable individuals, August Yue Huang et al. isolated low-frequency postzygotic mutations in healthy subjects. They discovered that the mutational signatures operating during normal development closely mirror those found in cancerous tumors, proving that the fundamental physics of DNA damage and repair remains uniform across both normal development and disease progression.
[Fertilized Egg] ---> [Early Development] ---> [Adult Tissue]
| |
Signature: Early Clock Signature: Environmental Exposure
(High Variant Frequency) (Low Variant Frequency)
In clinical oncology, tracking these signatures across space and time helps chart how a tumor evolves. For instance, Ren‐Chin Wu et al. in Elucidating the pathogenesis of synchronous and metachronous tumors in a woman with endometrioid carcinomas using a whole-exome sequencing approach used signature analysis to untangle a complex diagnostic riddle, proving that a tumor appearing eight years after an initial diagnosis was not a new primary cancer, but a late-stage recurrence sharing an ancestral signature of mismatch repair failure.
Furthermore, Naveed Ishaque et al. in Whole genome sequencing puts forward hypotheses on metastasis evolution and therapy in colorectal cancer quantified signature exposures to demonstrate that while primary tumors and their metastases share a common ancestral footprint, they undergo divergent evolution, experiencing different active mutational forces as they migrate to new microenvironments.
By shifting the analytical focus from which genes are mutated to the mathematical decomposition of the mutation spectrum, mutational signature analysis transforms passenger mutations from genomic noise into a precise, diagnostic chronicle of cellular history.
Key Foundational Work
Signatures of mutational processes in human cancer (2013) — contributed 1 concept elements to this direction.
Postzygotic single-nucleotide mosaicisms in whole-genome sequences of clinically unremarkable individuals (2014) — contributed 1 concept elements to this direction.
Whole-genome sequencing of bladder cancers reveals somatic CDKN1A mutations and clinicopathological associations with mutation burden (2014) — contributed 1 concept elements to this direction.