DeconstructSigs: delineating mutational processes in single tumors distinguishes DNA repair deficiencies and patterns of carcinoma evolution

Abstract

Background: Analysis of somatic mutations provides insight into the mutational processes that have shaped the cancer genome, but such analysis currently requires large cohorts. We develop deconstructSigs, which allows the identification of mutational signatures within a single tumor sample.

Results: Application of deconstructSigs identifies samples with DNA repair deficiencies and reveals distinct and dynamic mutational processes molding the cancer genome in esophageal adenocarcinoma compared to squamous cell carcinomas.

Conclusions: deconstructSigs confers the ability to define mutational processes driven by environmental exposures, DNA repair abnormalities, and mutagenic processes in individual tumors with implications for precision cancer medicine.

Fig. 1

deconstructSigs workflow and output. a Given an input tumor profile and reference input signatures, deconstructSigs iteratively infers the weighted contributions of each reference signature until an empirically chosen error threshold is reached. b Example of the plot generated by the command ‘plotSignatures’. The top panel is the tumor mutational profile displaying the fraction of mutations found in each trinucleotide context, the middle panel is the reconstructed mutational profile created by multiplying the calculated weights by the signatures, and the bottom panel is the error between the tumor mutational profile and reconstructed mutational profile, with SSE annotated

Fig. 2

Comparison of signature contributions identified with deconstructSigs and WTSI Mutational Signature Framework. Scatterplots represent the relationship between the weighted proportions calculated using the WTSI Mutational Signature Framework method on a set of TCGA tumors and those inferred with deconstructSigs from the same set of patients. Each point plotted represents the weights assigned by both methods to one signature detected in a patient

Fig. 3

Specific TCGA patient examples. Comparison of tumor mutational profiles and reconstructed profiles output from deconstructSigs and WTSI Mutational Signature Framework. The reconstructed tumor profiles generated by using the signature weights assigned by the deconstructSigs method and the WTSI Mutational Signature Framework method are given for three tumor samples. a A signature associated with POLE hypermutation, signature 10, was identified in TCGA patient TCGA-D5-6931 using the WTSI Mutational Signature Framework (signature weight = 0.204) but not with deconstructSigs. However, a POLE exonuclease domain mutation was not observed in this patient. b The mutational profile of patient TCGA-67-6215 showed activity of Signature 17 but as this signature was not considered a possible signature extracted in the first step of the WTSI Mutational Signature Framework output, it was only called with deconstructSigs (signature weight = 0.634). c A signature associated with DNA mismatch repair deficiency, signature 6, was identified by deconstructSigs (signature weight = 0.481) in patient TCGA-AN-A0AK but was not identified by the WTSI Mutational Signature Framework. An MSH6 frameshift mutation was identified in TCGA-AN-A0AK indicating the DNA mismatch repair deficiency signature identified is unlikely to be spurious

Fig. 4

Temporal dissection of mutational processes. Mutations called from a cohort of five LUAD (circles) and LUSC (triangles) patients with multi-region sequencing were temporally dissected into trunk and branch mutations, described previously in [9]. One patient (L002) had a tumor exhibiting an adenosquamous histological subtype, with separate regions being of different histology. For each patient, the fraction of contribution of signatures associated with smoking (a) and APOBEC activity (b) was determined in the trunk and branch mutations. The smoking signature was seen at higher fractions in the trunk mutations (blue) than the branch mutations (red), whereas the signature of APOBEC activity was seen to contribute more to the LUAD branch mutations than LUAD trunk mutations or LUSC trunk or branch mutations

Fig. 5

Signatures present in esophageal carcinoma. a The signatures identified in a cohort of ESCA tumors by deconstructSigs, using the input reference signatures from [4]. The prevalence, defined as the fraction of patients the signature was detected in, is plotted for each mutational signature identified, and the proportion of patients with a higher fraction of early (red) or late (blue) mutations corresponding to that signature is shown. b, c A specific analysis of two esophageal adenocarcinomas exhibiting signs of signature 17 activity. The mutational profiles are given for all the mutations identified in both tumors, as well as the mutations classified as early or late. Signature 17 was identified as the largest contributor to the early mutations of patient TCGA-2H-A9GR (b) whereas it was identified as the contributing to the generation of the majority of late mutations in patient TCGA-L5-A4OU (c)