MYTH: An algorithm to score intratumour heterogeneity based on alterations of DNA methylation profiles

To the Editor: Intratumour heterogeneity (ITH) has significant associations with tumour development and therapeutic responses. The evaluation of ITH at the methylation level may gain an advantage over that at the genetic and transcriptional levels, although such algorithms remain lacking. We proposed a novel algorithm to score methylationyielding tumour heterogeneity (MYTH) of a tumour sample TS, given a DNA methylation profiling dataset containingm genes and t tumour samples, as follows:


MYTH: An algorithm to score intratumour heterogeneity based on alterations of DNA methylation profiles
To the Editor: Intratumour heterogeneity (ITH) has significant associations with tumour development and therapeutic responses. The evaluation of ITH at the methylation level may gain an advantage over that at the genetic and transcriptional levels, although such algorithms remain lacking. We proposed a novel algorithm to score methylationyielding tumour heterogeneity (MYTH) of a tumour sample TS, given a DNA methylation profiling dataset containing m genes and t tumour samples, as follows: To prove the effectiveness of MYTH in measuring ITH, we associated MYTH scores with clinical and phenotypic features, genomic features, antitumour immunity and drug response in 32 cancer types from The Cancer Genome Atlas (TCGA) (https://portal.gdc.cancer.gov/) (Table S1). We found that higher MYTH scores correlated with worse survival in pan-cancer and many individual cancer types ( Figure 1A). MYTH scores were significantly higher in metastatic than in primary tumours in six can-  Figure 1B). MYTH scores correlated positively with tumour stemness scores in pan-cancer and 21 cancer types ( Figure 1C). Moreover, MYTH scores correlated positively with proliferation signature scores in pan-cancer and 24 cancer types ( Figure 1D). MYTH scores were significantly lower in EGFR-mutated than EGFR-wildtype LUAD and higher in BRAF-mutated than BRAF-wildtype COAD ( Figure 1E). In BRCA, MYTH scores were higher in basallike than HER2-enriched and luminal A&B (ER+) subtypes ( Figure 1E), consistent with the higher genomic ITH in basal-like versus other subtypes of breast cancer. 1 Altogether, these results indicate that the MYTH ITH level is an adverse prognostic factor in diverse cancers.
Tumour mutation burden (TMB) correlated positively with MYTH scores in pan-cancer and 13 cancer types (Figure 2A). In 25 cancer types, copy number alteration scores correlated positively with MYTH scores. MYTH scores were higher in TP53-mutated than TP53-wildtype tumours in nine cancer types ( Figure 2B). In six cancer types prevalent with MSI tumours, MYTH scores were higher in MSI-high than MSS/MSI-low tumours ( Figure 2C). These results suggest an association between MYTH ITH and genomic instability in cancer. The enrichment scores of immune signatures (CD8+ T cells, NK cells and immune cytolytic activity) correlated inversely with MYTH scores in pan-cancer and 24, 20 and 25 cancer types, respectively ( Figure 2D), supporting that ITH inhibits antitumour  Figure 2E), suggesting that MYTH ITH increases with the increase of tumour purity. Moreover, MYTH scores were lower in normal samples than tumour samples in pan-cancer and eight cancer types with normal samples' methylation data available ( Figure 2F). These results reflect that MYTH does represent ITH among tumour cells.
The global methylation levels correlated inversely with MYTH scores in 20 of the 22 cancer types with related data available 2 ( Figure 3A). WNT pathway is associated with hypermethylation in cancer, 3 whose enrichment scores correlated inversely with MYTH scores in pancancer and 22 cancer types ( Figure 3B). These results suggest a negative association between MYTH scores and global methylation levels in cancer. The pathways highly enriched in high-MYTH-score tumours included cell cycle, p53 signalling, DNA replication and homologous recombination ( Figure 3C), suggesting that increased activities of these pathways may promote ITH. The pathways highly enriched in low-MYTH-score tumours were mainly involved in immune and stromal signatures ( Figure 3D), accordant with the negative association between MYTH scores and immune signatures.
We explored associations between MYTH ITH scores and ITH scores by other seven algorithms, including MATH, 4 EXPANDS, 5 PhyloWGS, 6 ABSOLUTE, 7 DEPTH, 8 tITH 9 and sITH. 10 Among them, MATH, EXPANDS, Phy-loWGS and ABSOLUTE evaluate ITH at the DNA level, and DEPTH, tITH and sITH at the mRNA level. In pancancer, MYTH scores showed the strongest correlations with DEPTH and tITH scores and the weakest correlations with MATH and PhyloWGS scores ( Figure 4A). Overall, MYTH scores had stronger correlations with ITH scores by the mRNA-based algorithms than with those by the DNAbased algorithms. A potential reason could be that the ITH at the DNA methylation level directly impacts mRNA expression.
We compared MYTH with the seven algorithms in the 32 cancer types. TMB correlated positively with ITH scores by MATH, EXPANDS, PhyloWGS, ABSOLUTE, DEPTH, tITH and sITH in 11, 9, 21, 6, 16, 8 and 5 cancer types, respectively, compared to MYTH in 13 cancer types ( Figure 4B). In the six cancer types prevalent with MSI tumours, ITH scores by EXPANDS, PhyloWGS, DEPTH and tITH were significantly higher in MSI-high than MSS/MSI-low tumours in one, one, three and four cancer types, respectively, compared to MYTH in six cancer types ( Figure 4B). In contrast, ITH scores by MYTH, ABSOLUTE and sITH were significantly lower in MSI-high than in MSS/MSI-low tumours in four, three and one cancer types, respectively. The immune cytolytic activity scores correlated negatively with ITH scores by MATH, EXPANDS, PhyloWGS, ABSO-LUTE, DEPTH, tITH and sITH in eight, one, three, nine, eleven, eight and four cancer types, respectively, compared to MYTH in 24 cancer types ( Figure 4C). Tumour purity correlated positively with ITH scores by MATH, EXPANDS, PhyloWGS, ABSOLUTE, DEPTH, tITH and sITH in nine, four, eight, six, twenty-two, twenty and eight cancer types, respectively, compared to MYTH in 30 cancer types ( Figure 4D). It indicates that MYTH is more likely to capture the ITH among tumour cells than the other algorithms. MATH scores correlated negatively with survival in more cancer types, compared to the other algorithms ( Figure 4E). ITH scores by MATH, EXPANDS, PhyloWGS, ABSOLUTE, DEPTH, tITH and sITH were significantly higher in metastatic than primary tumours in three, two, three, two, two, two and three cancer types, respectively, compared to MYTH in six cancer types ( Figure 4F).
In cancer cell lines, MYTH scores correlated positively with the expression of MKI67 (a marker for cell proliferation) and DNA repair genes; MYTH scores were higher in MSI-high than MSS/MSI-low cell lines; MYTH scores correlated inversely with IC50 values of the compounds targeting chromatin ( Figure S1). These results are consistent with the findings in the TCGA bulk tumours and suggest p-values are shown. *p < .05, **p < .01, ***p < .001; ns, not significant. TMB, tumour mutation burden; MSI, microsatellite instability; MSS, microsatellite stability that higher MYTH ITH tumours are more sensitive to epigenetic therapies.
In conclusion, the MYTH algorithm is superior or comparable to established algorithms in characterising ITH.