Abstract
The genome of every cancer cell carries a record of the mutational processes that have acted throughout its history. Mutational signature analysis, which infers the activity of mutagenic processes from their characteristic base-change patterns, has become an indispensable tool for interpreting somatic mutations. However, this framework captures only which types of mutations a process generates and not where in the genome they occur — a distribution influenced by replication timing, chromatin organization, transcription, DNA secondary structure, and other genomic features. Here, we present a generative probabilistic framework (MuTopia) that jointly infers mutational spectra and their genome-wide topography as nonlinear functions of genomic and epigenomic state. Applied to whole-genome sequencing data from 15 tumor types, MuTopia reveals that mutational processes fall into eight conserved topographic archetypes, or topotypes, shaped primarily by replication timing and chromatin state. Diverse mutational processes converge upon this limited repertoire, indicating that the genomic distribution of mutagenesis is constrained less by the source of damage than by how that damage is processed. Individual mutational processes exhibit state-dependent variation in their genomic distributions: the same signature can adopt distinct topotypes depending on repair proficiency and replication stress. For instance, SBS8 shifts from a canonical late-replicating profile in homologous recombination-proficient tumors to an early-replicating, stress-associated topotype in HR-deficient tumors, and replication stress similarly reshapes the genomic distribution of APOBEC editing. Topotypes, therefore, provide a classification of mutagenesis distinct from spectral signatures, capturing aspects of tumor biology that spectra alone cannot resolve.