Mutational signatures reveal ternary relationships between homologous recombination repair, APOBEC, and mismatch repair in gynecological cancers

We accessed, cleaned, and reformatted Simple Somatic Mutation (SSM) files for WGS and WES of primary tumors from the data portal of the International Cancer Genome Consortium (ICGC, https://​dcc.​icgc.​org/​projects) and utilized them for subsequent analyses. WGS data included UCEC (n = 91), ovarian (n = 157) and cervical (n = 20) tumors. WES data included UCEC (n = 531), ovarian (n = 411) from ICGC and cervical (n = 289) tumors from the Cancer Genome Atlas (TCGA). In addition, 134 related gynecological cancer cell lines including UCEC related (n = 40), ovarian related (n = 74) and cervical related (n = 21) were accessed via Cancer Cell Line Encyclopedia (CCLE) [33].

Mutational signatures of WES and WGS of tumor samples were extracted using an NMF-based de novo mutational signatures analysis [34] decomposed by COSMIC, V3.1 which includes 96 Single-Base Substitutions (SBSs), 78 Double-Base Substitutions (DBSs) and 83 small Insertions/Deletions (IDs) signatures (http://​cancer.​sanger.​ac.​uk/​cosmic/​signatures).

The cosine similarity [4, 17], which is the cosine of the angle between two vectors in space, is the standard measure for comparing two spectra. For decomposing the identified NMF-based signatures to COSMIC signatures, cosine similarity was used.

Confirmatory mutational signature analysis was performed using the computational tool SigMA, using hg19 as the reference human genome. SigMA, which uses a likelihood-based approach, has been used to accurately detect the mutational signature associated with HR deficiency (SBS3) from WGS, WES and targeted gene panels, even from low mutation counts. SigMA was run in tuned MVA models [do_mva = T and do_assign = T] for UCEC and ovarian cancer, and in exploratory mode [do_mva = F and do_assign = F] for cervical cancer. For all tumors, run () function was set as check_msi = T and add_sig3 = T. A detailed description of the algorithm and its performance is provided in [35].

TMB was measured as number of all somatic mutations per Megabase (Mb) of each tumor genome, visualized by [34]. Following extraction of mutational signatures for each sample, the first and second dominant signatures were visualized based on their contributions over the exomes or the whole genomes of analyzed tumors. Hierarchical clustering was performed based on the relative contribution of signatures in each tumor. These data were displayed in the clustered heatmaps splitted-up and annotated by their dominant signatures. The contribution values of MMRd, HRd, and APOBEC signatures were used as inputs to draw the ternary plots. All visualizations were performed using the “car”, “ComplexHeatmap”, “stats”,“circlize”, “ggplot”, and “ternary” packages in R [36‐38].

Overall survival (OS) data were downloaded from ICGC Data Portal(https://​dcc.​icgc.​org) and used for survival analysis utilizing the “Survminer” and “survival” packages in R [39]. P values representing the significance were determined from log-rank.

P values were calculated by Hypergeometric test for assessing significance of co-occurrences and exclusivity of signatures; and by Fisher exact test for assessing significance of the dominant signature in signature groups. All analyses were conducted in the R statistical environment (R version 4.0.0 http://​www.​r-project.​org/​). All reported P values were two tailed; ≤ 0.05 was considered significant.

In total, we analyzed 455,621 SBSs, 2,940 IDs, and 1,561 DBSs from 91 UCEC whole genomes to detect NMF-based de novo mutational signatures. MMRd signatures, including SBS15 and SBS44, were present in 23% of tumors, with an average contribution of 0.43 (range: 0.2–0.81). MMRd was the first or second dominant signature when present. HRd (SBS3) was present in 14% of tumors with an average contribution of 0.49 (range: 0.36–0.73) as the first or second dominant signature.

The other detected signatures included SBS8 in 25%, APOBEC (SBS2,13) in 39%, and POLE (SBS10a, b) in 10% of tumors, with average contributions of 0.26 (range: 0.15–0.35), 0.23 (range: 0.04–0.81), and 0.28 (range: 0.11–0.88), respectively. Only in one tumor, we detected the MSI/POLE (SBS14) signature with the highest TMB. We also detected the ID1, ID2, ID3, ID7 and ID8 signatures in UCEC whole genomes. The etiology of the D7 and ID8 signatures are associated with MMRd and non-homologous end joining (NHEJ), respectively; they were the first/second dominant signatures in 23% and 53% of UCEC tumors respectively. Finally, we detected DBS2, DBS4, DBS6, DBS9, and DBS11 whose etiologies with the exception of DBS11’s association with APOBEC, remain unknown. DBS11 was the first or second dominant signatures in 21% of UCEC tumors (Fig. 1 and Additional file 1: Fig. S1).

The presence and contribution of detected signatures enabled us to distinguish five tumor groups with distinct patterns of mutational signatures. The dominant signatures were significantly associated with their corresponding signature group. (Fig. 1 and Additional file 1: Fig. S1). MMRd and HRd signatures occurred in a mutually exclusive manner with no overlap between the groups. Like HRd, the APOBEC signatures were mutually exclusive with MMRd. In contrast, ID7 and ID8 were mutually exclusive with HRd (SBS3) as well as DBS2, 4, and 6. However, ID7 co-occurred with the MMRd SBS signatures. DBS2, 4 and 6 co-occurred with SBS8, but were exclusive with the MMRd SBS and ID1, 2, 3, 7, and 8 signatures. Finally, DBS9 was mutually exclusive with the APOBEC SBS signatures. The group of tumors with a dominant POLE signature exhibited other mutational signatures the least; other than POLE, these tumors only had the ID1, 3, 8, and DBS9 signatures. These interactions revealed a ternary relationship between MMRd, HRd, and APOBEC, which respectively had the highest to the lowest prevalence in UCEC (Fig. 1 and Additional file 1: Fig. S1, Additional file 3: Table S2).

We also analyzed 740,535 SBSs, 8,265 IDs, and 653 DBSs detected in 531 UCEC exomes. The mutational signatures were consistent with those detected in UCEC whole genomes. The prevalence of mutational signatures and their contribution to UCEC tumors were comparable in both WGS and WES analyses, except for the MMRd signatures, SBS21, SBS26, SBS15, and SBS44, which were detected in 53% of the exomes vs. 23% of the whole genomes. Consistent with WGS analysis, MMRd and HRd signatures occurred in a mutually exclusive manner, and APOBEC co-occurred with HRd and was mutually exclusive with MMRd. A group of tumors (23%) showed SBS87 signature which was not detected by WGS data. The proposed etiology of SBS87 is thiopurine chemotherapy treatment (TCT) [40]; it was the dominant signature in 12% of cases with an average contribution value of 0.36 (range: 0.09–0.78). None of the samples in tumors with the TCT signature had HRd (Fig. 2 and Additional file 1: Fig. S2, Additional file 3: Table S3).

In the UCEC exomes, we detected ID1, ID10, and ID11 in addition to a de novo signature, ID83B, which had cosine similarity of 0.42 with COSMIC’s ID1. Among these signatures, only ID1 was detected by both WGS and WES. The only DBS signature detected in the exomes was DBS10 with an etiology associated with MMRd (Fig. 2 and Additional file 1: Fig. S2, Additional file 3: Table S3).

Finally, as a confirmatory approach we performed multivariate analysis for extracting SBS mutational signatures of the UCEC whole genomes and exomes. The data were consistent with those of NMF-based signature extraction. Multivariate analysis also confirmed the pattern of co-occurrence and exclusiveness among HRd, APOBEC and MMRd. Notably, SBS3 and SBS8 co-occurred in UCEC whole genomes (Additional file 1: Fig. S3, Additional file 3: Table S4).

In total, we analyzed 1,180,840 SBSs, 63,073 IDs, and 11,096 DBSs from 157 ovarian tumors’ whole genomes to detect NMF-based de novo mutational signatures. MMRd was present and dominant in only one sample. HRd (SBS3) was present in 52% of tumors in 98% of which it was the first dominant signature. HRd has an average contribution value of 0.55 (range: 0.31–0.86), with the third highest TMB and a median of ~ 1 somatic mutation per Megabase. APOBEC (SBS2, 13) was present in 35% of tumors with an average contribution value of 0.11 (range: 0.04–0.45). APOBEC was never the first dominant and was the second dominant signature in only 4% of tumors. SBS8 was the other major detected signature, present in 57% of tumors with an average contribution value of 0.24 (range: 0.11–0.52). SBS8 was the first or second dominant signature in 31% of tumors. (Fig. 3 and Additional file 1: Fig. S4).

In the whole genomes of ovarian tumors, we detected ID1, ID2, and ID8, in addition to a new signature, ID83C, which had cosine similarity of 0.77 with COSMIC’s ID15, ID2 and ID1. ID1 and ID2’s etiology is associated with replication slippage, and ID8’s is NHEJ [17]. ID8 (NHEJ) was observed in 89% and was first or second dominant ID signature in 83% of cases. These tumors also showed combinations of DBS2, DBS4, DBS6, DBS9, and DBS5. DBS5, which is associated with prior platinum therapy (PT), had the highest TMB. (Fig. 3 and Additional file 1: Fig. S4).

In ovarian tumors, the presence and contribution of detected signatures enabled us to distinguish three main tumor groups with distinct patterns of mutational signatures two of which were dominated by HRd (SBS3) and SBS8 (Fig. 3 and Additional file 1: Fig. S4, Additional file 3: Table S5). APOBEC was present across the groups, and along with ID8(NHEJ), co-occurred significantly with both HRd (SBS3) and SBS8. Notably, DBS2, DBS4 and DBS6 also had co-occurrence with APOBEC, HRd, and SBS8. In these tumors, the ternary relationship between MMRd, HRd, and APOBEC was dominated by the latter two with almost no contribution from MMRd.

We also analyzed 47,902 SBSs, 1,796 IDs, and 547 DBSs from 411 ovarian tumor exomes. The mutational signatures were consistent with those detected in the whole genomes. In contrast, to detecting MMRd in only one whole genome, we detected the MMRd-associated SBS15 signature in 16% of the exomes; it was the first or second dominant signature in 5% of tumors with as average contribution of 0.19 (range: 0.07–0.41). HRd (SBS3) was present and was the first or second dominant signature in 74% of tumors with an average contribution value of 0.69 (range: 0.28–1); it had the third highest TMB with a median of ~ 1 somatic mutation per Megabase. APOBEC (SBS2, 13) was present among 18% of tumors, with an average contribution of 0.10 (range: 0.05–0.20). APOBEC was never the first dominant and was the second dominant signature in only 2% of cases (Fig. 4 and Additional file 1: Fig. S5).

We detected ID14, ID16 with unknown etiologies in ovarian tumor exomes, along with ID83B, which had cosine similarity of 0.18 with COSMIC’s ID7. These signatures were all different from the ID signature profile detected in ovarian tumor whole genomes. The exomes also showed DBS2, DBS4, DBS6, DBS9, and DBS11(APOBEC) among which DBS2 and DBS4 co-occurred with HRd (SBS3) (Fig. 4 and Additional file 1: Fig. S5, Additional file 3: Table S6).

Finally, as a confirmatory approach we performed multivariate analysis for extracting SBS mutational signatures of the ovarian tumors’ whole genomes and exomes. The data were consistent with those of NMF-based signature extraction. Notably, SBS3 and SBS8 had co-occurred in ovarian tumors’ whole genomes (Additional file 1: Fig. S6, Additional file 3: Table S7).

Only 20 cervical tumors had available WGS data. In these tumors, we detected three groups based on the first or second dominant detected signature (Additional file 1: Fig. S7 and Additional file 1: Fig. S8). APOBEC was the most prevalent, detected in 65% of cases followed by POLE present in 20%. The remaining samples had SBS5/Aging as the first or second dominant signature.

Using WES data from a larger cohort of 289 cervical tumors, we analyzed 81,126 SBSs, 1,929 IDs, and 164 DBSs to detect NMF-based de novo mutational signatures. The MMRd signature, included only SBS15 and was present in 19% of tumors with an average contribution of 0.26 (range: 0.07–0.64). MMRd was the first or second dominant signature among 9% of cases. HRd (SBS3) was present and the first dominant signature in 4% of tumors with an average contribution value of 0.50 (range: 0.39–0.68). APOBEC (SBS2, 13) was present among 85% of tumors with an average contribution of 0.49 (range: 0.08–0.91). APOBEC was first or second dominant signature among 68% of cases. Other major detected signatures, POLE (SBS10a, b) and TCT (SBS87), were present among 48% and 10% of tumors with average contributions of 0.17 (range: 0.08–0.64) and 0.32 (range: 0.13–0.75), and the first or second dominance in 19% and 7% of cases, respectively. Similar to UCEC, the TCT signature was only detected in the exomes and not in the whole genomes of cervical tumors (Fig. 5 and Additional file 1: Fig. S9).

In cervical tumor exomes, we detected ID1, ID2, and ID7 (MMRd), in addition to a de novo signature, ID83B, which had cosine similarity of 0.78 with COSMIC’s ID8 and ID3, and was present in 81% of cases. These tumors also showed DBS4, DBS10 (MMRd) and DBS11(APOBEC) (Fig. 5 and Additional file 1: Fig. S9).

We identified six cervical tumor groups with distinct patterns of mutational signatures. APOBEC (SBS2, 13) was the most prevalent signature; it co-occurred significantly with POLE (SBS10a, b) and APOBEC (DBS11) and was mutually exclusive with the MMRd SBS signatures. MMRd and HRd signature distinguished tumor groups in which APOBEC was also present in a dispersed pattern. The MMRd SBS signatures co-occurred with TCT (SBS87) (Fig. 5 and Additional file 3: Table S8).

Finally, similar to the analyses of the UCEC and ovarian tumors, a multivariate analysis for extracting SBS mutational signatures, confirmed the results from NMF-based signature extraction in the exomes and whole genomes of cervical tumors (Additional file 1: Fig. S10).

We compared 5-year OS of patients stratified by mutational signatures suffering from UCEC, ovarian and cervical cancers. In UCEC tumors, the group of patients with HRd and APPBEC had significantly worse survival than the group with MMRd signature (P = 0.0037) (Additional file 1: Fig. S11). In ovarian tumors, patients with HRd, APOBEC and MMRd signatures together showed better survival compared to the rest of samples (P = 0.0015) (Additional file 1: Fig. S12). In cervical tumors, no significance difference was observed, probably due to limited number of samples in each group restricting the statistical power (Additional file 1: Fig. S13). Finally, we compared survival of patients stratified based on their mutational signatures, independent from their original diagnosis. The results showed that patients with HRd had the poorer outcome followed by APOBEC compared to those with MMRd signature (P = 0.016, 0.0001 and 0.0077) (Additional file 1: Fig. S14).

We analyzed 106,917 SBSs, 10,925 IDs, and 135 DBSs detected by WES in 40 uterine cell lines. The MMRd signatures, including SBS6, SBS20, and SBS26, were present in 24 cell lines with an average contribution value of 0.71. When present, MMRd was the first or second dominant signature. POLE (SBS10a, b) was present in four cell lines in three of which it was the first dominant signature; it had an average contribution of was 0.55. Moreover, DBS11 (APOBEC) was present and dominant in 19 cell lines with an average contribution of 0.98.

For the MMRd and POLE SBS signatures, the results from multivariate analysis were consistent with the signature extracted using the NMF-based approach. However, HRd and APOBEC were only detected by multivariate analysis of cell lines. HRd (SBS3) was present in 12 cell lines with an average contribution of 0.07. HRd was the second dominant signature in two cell lines. Finally, APOBEC was present in 21 cell lines with an average contribution of 0.05; it was the second dominant signature in two cell lines (Additional file 1: Fig. S15).

Next, we analyzed 43,757 SBSs, 4,717 IDs, and 134 DBSs, detected in the exomes of 74 ovarian cell lines. The MMRd signatures, including SBS15 and SBS44, were present in 38 cell lines with an average contribution of 0.24. MMRd was the first or second dominant signature in 15 cell lines. HRd (SBS3) was present among 18 cell lines with an average contribution value of 0.42; it was the first or second dominant signature when present. POLE (SBS10a, b) was present in 8 cell lines with an average contribution value of was 0.14; it was the first dominant signature in only one tumor. DBS11 (APOBEC) was present and dominant in 22 cell lines with the average contribution value of 0.95. Multivariate analysis was consistent with NMF-based signature extraction (Additional file 1: Fig. S16, Additional file 3: Table S8).

Finally, we analyzed 10,842 SBSs, 725 IDs, and 47 DBSs detected in 21 cervical cell line exomes. The MMRd signatures, including SBS6, and SBS26 were present in four cell lines with an average contribution of 0.34. MMRd was the first or second dominant signature in three cell lines. APOBEC was present in 15 cell lines with an average contribution value of 0.31; it was the first or second dominant signature in 12 cell lines. Multivariate analysis did not detect MMRd; however, it found HRd(SBS3) present in 11 cell lines with an average contribution value of 0.1. HRd was second dominant signature in one cell line. Multivariate analysis also detected APOBEC in 20 cell lines with an average contribution value of 0.27; APOBEC was first/second dominant signature in 14 cell lines (Additional file 1: Fig. S17, Additional file 3: Table S8).