Deficiency of replication-independent DNA mismatch repair drives a 5-methylcytosine deamination mutational signature in cancer

Abstract

Multiple mutational signatures have been associated with DNA mismatch repair (MMR)–deficient cancers, but the mechanisms underlying their origin remain unclear. Here, using mutation data from cancer genomes, we identify a previously unknown function of MMR that is able to protect genomes from 5-methylcytosine (5mC) deamination–induced somatic mutations in a replication-independent manner. Cancers with deficiency of MMR proteins MSH2/MSH6 (MutSα) exhibit mutational signature contributions distinct from those that are deficient in MLH1/PMS2 (MutLα). This disparity arises from unrepaired 5mC deamination–induced mismatches rather than replicative DNA polymerase errors. In cancers with biallelic loss of MBD4 DNA glycosylase, repair of 5mC deamination damage is strongly associated with H3K36me3 chromatin, implicating MutSα as the essential factor in its repair. We thus uncover a noncanonical role of MMR in the protection against 5mC deamination–induced mutation in human cancers.

Fig. 1.. Landscape of MSI-H samples.

(A) Profile of mutation burden and six types of mutation frequency as well as the aberrant status of MMR genes including DNA mutation, RNA expression, and methylation. Cancer types and mutants classification are also indicated. (B) Principal components (PC) analysis of MSI-H cancer samples based on the frequency of 96 types of mutational contexts. The fractions of the six types of mutations are represented by the area of the sectors, and MutSα mutants are highlighted in red. SNV, single nucleotide variant; nDel, insertion and deletion.

Fig. 2.. Mutation signature contribution in MutSα and MutLα mutants.

(A) Fraction of MMRd-associated signatures and age-related signature SBS1 contribution in MutSα and MutLα mutants. (B) The spectrum of de novo signatures extracted from TCGA MSI-H cancer samples. (C to F) The boxplot of SigA contribution for MutSα and MutLα mutants in TCGA-MMRd, MSK-CRC, MSK-UCEC, and DepMap MMRd cohort. P values were calculated by two-tailed Student’s t test.

Fig. 3.. Replication asymmetry for MMRd samples.

The landscape of replication asymmetry for all observed mutations (A) and expected mutations (B) in MutSα and MutLα mutants. The expected mutations were obtained from simulation data that consider the abundance of trinucleotide mutational contexts. (C and D) Boxplot of replication strand bias for A > G/T > C and C > T/G > A mutations in MutSα and MutLα mutants. (E) Boxplot of replication strand bias for CpG C > T and non-CpG C > T mutations in MutSα and MutLα mutants. WXS, whole exome sequencing. ***P <0.001; n.s., >0.05, two-tailed Student’s t test.

Fig. 4.. Replication asymmetry for MBD4 and POLE mutants.

The landscape of replication asymmetry for all observed mutations (A) and expected mutations (B) in MBD4 and POLE mutants. The expected mutations are obtained from simulation data that consider the abundance of trinucleotide mutational contexts. (C and D) Boxplot of replication strand bias for CpG C > T and non-CpG C > T mutations in MBD4 and POLE mutants. (E and F) Boxplot of replication strand bias for CpG C > T mutations in highly methylated and lowly methylated regions for MMRd samples and POLE mutants. Sites with a β > 0.3 are defined as highly methylated, while <0.3 as lowly methylated. The range of mutation counts in lowly and highly methylated sites for calculating strand bias in MMRd samples was (52 to 156) and (2347 to 6145), respectively, and for POLE mutants, (72 to 703) and (4032 to 39,580), respectively.

Fig. 5.. Association of mutation frequency with local determinants for different samples.

(A to D) Observed and expected mutation densities in exons and introns across MSS, POLE mutants, MMRd samples, and MBD4 mutants. The expected mutations are obtained from simulation data that consider the abundance of trinucleotide mutational contexts. The decrease of observed (obs) and expected (exp) mutation density in exonic regions is indicated and calculated as (obs-exp)/exp. (E) Correlation of CpG C > T mutation ratio (obs/exp) with gene expression for MMRd samples, MSS, and MBD4 mutants. The P values of the correlation are 7.7 × 10⁻⁴, <2.2 × 10⁻¹⁶, and 0.167 for MBD4 mutants, MMRd samples, and MSS, respectively, and they were obtained from the linear regression model by fitting observed mutation density with unbinned gene expression. (F to H) Boxplot of transcription strand bias for CpG C > T and non-CpG C > T mutations in MMRd samples, MSS samples, and MBD4 mutants. (I) The hazard ratio (HR) of different epigenetics marks for CpG C > T mutation formation from multivariable logistic regression model. The 95% confidence level is indicated. P value is calculated by Wald’s test. (J) Correlation between mutations in MBD4 mutants and H3K36me3 signal from mobilized CD34⁺ primary cells.

Fig. 6.. Association of MBD4 binding sites, histone mark H3K36me3, and mutations.

(A) Venn diagram indicating the number of regions classified as top and bottom MBD4 and H3K36me3 signal based on the HepG2 cell line. (B) The ratio of observed and expected CpG C > T mutations in different regions for MBD4 mutants, MSS, and MMRd cancers. The HR of different epigenetics marks for CpG C > T mutation formation from multivariable logistic regression model for MMRd (C) and MSS (D) cancers. The 95% confidence level is indicated. P value was calculated by Wald’s test. (E) Schematic of the proposed mechanism of mismatch formation, repair, and mutations.