The design of nbTMB and mlTNB, based on non-B and mutation co-localisation

Our investigation unveils two novel biomarkers: nbTMB (non-B-informed TMB) and mlTNB (mutation-localised tumour non-B DNA burden), aiming to quantify the multi-dimensions of genomic instability using both tumour mutations and non-B DNA at a sample level.

First, nbTMB quantifies mutations co-localised within non-B forming regions (Fig. 1a) as a non-B informed tumour mutation burden TMB. The mutation signatures are first extracted from each tumour profile and then mutation sites overlapped with non-B DNA motifs are further quantified at genomic-wide. We also calculate nbTMB percentage (nbTMBp) to describe the proportion of tumour mutations co-localised with non-B DNA motifs, relative to total TMB.

Fig. 1: nbTMBp predicts patient outcome and drug sensitivities.

a Schematic representation of non-B-informed mutation quantification, showing the calculation of nbTMBp and its role as an indicator of TMB composition, bTMB (traditional Tumour Mutation Burden), and nbTMB (non-B-informed Tumour Mutation Burden), emphasising the fraction of mutations associated with non-B DNA structures. b Kaplan-Meier survival analysis contrasting overall survival (OS) in patients for Pan-Can patients (N = 867) with high vs. low TMB undergoing immunotherapy, highlighting the prognostic significance of TMB levels. The high TMB is define defined as the top 20% within each cancer type. c Stratification of pan-cancer patient survival outcomes by TMB levels, with a secondary differentiation based on OS status. Among pan-can patients categorised by TMB levels (high or low), a further distinction into ‘alive’ and ‘deceased’ based on overall survival reveals that the deceased cohort consistently exhibits a higher nbTMBp percentage across both TMB categories. d Venn diagram depicting the gene overlap between the MSKCC-Panel-468, genes with non-B DNA mutations and genes linked to immune checkpoint inhibitor (ICI) response, suggesting a potential connection between non-B DNA mutations and immunotherapy effectiveness. e, f Comparison of survival rates within the high TMB patient subset, showing that higher nbTMBp is associated with decreased survival, thereby offering additional stratification within this group. g, h Analysis within the high TMB category, indicating that further dissecting patients on TMB exclusively does not significantly stratify patients’ survival outcomes. i, j Clustering of ovarian cell lines by nbTMBp to identify distinct groups with varying levels of nbTMBp, potentially reflecting differential drug sensitivity. k nbTMBp shows a linear trend of increasing drug resistance of Cisplatin. This is evident in both IC50 metrics (where a lower value indicates increased sensitivity) and dose-response AUC (where a higher value indicates increased sensitivity). It demonstrates that increased nbTMBp correlates with heightened resistance to Cisplatin in ovarian cancer cell lines, suggesting the utility of nbTMBp in predicting drug response. Such a correlation is absent in the case of another platinum-based compound Carboplatin. l Heatmap and ridge plot illustrating the lack of a clear relationship between TMB alone and drug sensitivity of Cisplatin and Carboplatin in ovarian cancer cell lines, indicating the potential advantage of incorporating nbTMBp for more refined predictions. m Distribution of ovarian cancer cell lines across TMB-based clusters, with the colour gradient representing the range of TMB values. n Box plots showing drug sensitivity metrics across TMB-defined clusters for Cisplatin and Carboplatin, indicating the absence of a consistent pattern between TMB and drug response, thus underscoring the potential value of nbTMBp in therapeutic decision-making.

Second, mlTNB refers to mutation-localised tumour non-B DNA burden as a quantification of non-B DNAs. Different from nbTMB, the marker, mlTNB, focuses on the counts of non-B motifs that contain mutation sites (Fig. 2a). Due the various non-B types, the mlTNB is further calculated by non-B motif types. For the comparison across non-B types and across samples, the burden value will also be normalised by both the number of mutations and the motif library size.

Fig. 2: mlTNB predicts prognosis in early-stage pancreatic cancer.

a Diagram delineating the categorisation of tumour non-B DNA burdens: total non-B burden (TNB), mutation-localised tumour non-B burden (mlTNB) and mutation-free tumour non-B burden (mfTNB). b Clustering analysis of early-stage pancreatic cancer patients with progression, resulting in seven distinct patient sample clusters. Each group is characterised by unique mlTNB loads, which are differentiated by the type of non-B DNA motifs present, such as direct repeats (DR), G-quadruplexes (G4), inverted repeats (IR), mirror repeats (MR), short tandem repeats (STR) and Z-DNA (ZDNA). c Kaplan-Meier plots illustrating progression-free survival for the seven patient clusters identified, providing insight into the prognostic value of mlTNB categorisation. d Summary of pathway enrichment analysis, which underscores the predominant gene mutation signatures associated with each of the mlTNB clusters, potentially linking molecular pathways with patient prognosis. e Comparative overview of the distribution of fractional genome alteration (FGA), TMB and microsatellite instability (as measured by MANTIS score) across the clusters, indicating a lack of significant variation among the groups, which all show low TMB (median TMB < 2), low FGA (median < 30%) and MSS (median MANTIS score < 0.3), underscoring mlTNB as a driving factor stratifying the patients. f Comparison of PFS between the cluster with the most extended median PFS (mlTNB-DR, cluster 3) and the cluster with the shortest median PFS (mlTNB-ZDNA, cluster 5), emphasising the differential impact of non-B DNA motif types on patient outcomes. g Visual representation of the chromosomal distribution of non-B DNA and mutation co-localisations contributing to the mlTNB burden across different chromosomes. The x-axis represents samples, and the y-axis represents chromosomes. Each column corresponds to a sample, and each coloured grid indicates a mutation/non-B DNA co-localisation within a specific chromosome for that sample.

The two markers are calculated for each tumour profile at sample level. The mutation signatures are extracted from each tumour profile. The genomic-wide non-B forming region are further overlapped with the mutated regions for each tumour profile. Using the overlapped region by counting separately the number of mutation and non-B motifs involved, we are able to derive the two metrics, nbTMB and mlTNB, as the new biomarker to reflect the interplay between mutation and non-B DNA. The metric was further refined by optional normalisations to predict patient prognosis and treatment responses.

In this study, nbTMBp was applied to immunotherapy treatment in ovarian cancer due to the high heterogeneity observed in patient responses within the high-TMB group. nbTMBp refines TMB by quantifying the percentage of mutations co-localised with non-B DNA, providing additional stratification. Conversely, mlTNB was developed for pancreatic cancer, an ‘immune-cold’ cancer with typically low TMB. mlTNB focuses on non-B DNA burden, accounting for the impact of non-B DNA structures on genomic instability.

nbTMB differentiates survival among TMB-High patient post-immunotherapy

We first describe a Pan-Can immunotherapy analyses in which nbTMBp appears linked with prognosis. High TMB has been reported to be associated with improved immunotherapy response [16,17,18]. However, within TMB-high patients, outcomes remain heterogenous. Herein, we explore the heterogeneity with TMB high/low groups using nbTMBp to investigate its role as a biomarker associated with post-immunotherapy survival.

We analysed the mutation data from patients who underwent immunotherapy based on the MSK-IMPACT study from 11 different cancer types [17]. Within each caner type, using the 80th percentile of TMB, we assigned patients into TMB -high and -low groups and compared their overall survival (OS) (Fig. 1b). We further stratified patients within each of TMB-High and -Low groups based on their status—alive/censored or deceased. We defined nbTMB for each patient sample by quantifying the numbers of mutations co-localised with non-B forming regions [19]. When comparing median nbTMBp across groups, the TMB-high group had a lower nbTMBp overall, relative to the TMB-low group (Fig. 1c). Within each TMB classified group, median nbTMBp was significantly higher in deceased patients, irrespective of their high/low status (Fig. 1c). A gene-level analysis of immune response signatures [20] revealed an 86% overlap between mutations co-localised with non-B motifs and immune checkpoint inhibitor-outcome-linked genes (n = 98) (Fig. 1d).

Next, we performed clustering on nbTMBp within the TMB-High patients, which revealed two patient subgroups (Fig. 1e), one with median nbTMPp of around 10% that was associated with significantly (p < 0.01) shorter OS as compared to TMB-High patients with less than 10% nbTMBp (Fig. 1f). For comparison, the same analysis was applied to TMB in which no significant OS difference was observed (Fig. 1g–h). Altogether, our findings lend support for the further study of nbTMBp as a potential marker of differential survival within TMB-high patients on immunotherapy. These results may reflect the potential contribution from non-B DNA genomic instability in some patients that have poor survival, despite having high TMB.

Increasing nbTMB is associated with decreased cisplatin sensitivity in ovarian cancer

We next explore the use of nbTMBp as a marker of altered cisplatin drug sensitivity in ovarian cancer. Cisplatin resistance is a major hurdle in effectively treating ovarian cancer [21]. Although cisplatin is commonly used for ovarian cancer treatment, drug resistance often arises due to a faulty apoptotic process, reducing treatment effectiveness [22,23,24,25,26]. Among the 57 ovarian cell lines with mutation profiles [27], ~40% have TMB greater than ten and a median fraction of genome altered (FGA) of ~50%, which supports the potential role of genomic instability in treatment resistance. Investigating how cells signal in response to chemotherapy from the non-B DNA perspective of genomic instability may shed light on treatment outcomes.

We defined an ovarian cell line specific mutation signature and corresponding nbTMBp for use in a cluster analysis that identified three cell line groups of varying (low to high) nbTMBp (Fig. 1i, j). Median nbTMBp significantly differed among the three clustered cell line groups. Tests of association between TMB, FGA and tumour grade with nbTMBp-derived cell line clusters lacked significance, as did a correlation between TMB and nbTMBp among the ovarian cell lines. We examined the effect of clusters on cisplatin drug sensitivity in which increasing nbTMBp was significantly associated with decreasing cisplatin sensitivity. This finding was consistent for dose-response AUC with cisplatin (Fig. 1k). For comparison, we performed the same analyses on carboplatin sensitivity which did not show the same result, suggesting a cisplatin specific nbTMBp effect. Additionally, the same cluster analysis on TMB failed to show a similar result (Fig. 1l, m). Altogether, our findings show support for the further exploration of nbTMBp as a potential marker of cisplatin sensitivity that may help to explain resistance when all other markers indicate otherwise.

mlTNB differentiates survival among early-stage pancreatic patient progressors

In contrast to nbTMBp, we next explore the use of mlTNB to quantify non-B burden and its association with survival in pancreatic adenocarcinoma (PAAD). PAAD is a highly aggressive cancer with poor outcome. Existing genomic instability measures have not proven informative in differentiating survival into clinically translatable patient groups for risk stratification [28]. As opposed to focusing on mutation numbers, mlTNB quantifies non-B DNA regions that contain mutation sites which can be classified according to non-B structure types to provide a more nuanced perspective [7] (Fig. 2a).

Using the mutation profiles of 76 TCGA early-stage pancreatic patients who progressed, we quantified mlTNB for each sample and used it in a cluster analysis resulting in seven patient groups with differentiated non-B structure types (Fig. 2b) that significantly differed in progression-free survival (PFS) (Fig. 2c). Patients with high mlTNB characterised mainly by direct repeat regions (high mlTNB-DR burden) was associated with the longest PFS (n = 7, median = 25 months), while patients with high mlTNB in Z-DNA regions had the shortest (n = 10, median = 5 months). PFS among other groups were similar: the patient group with high mlTNB from short tandem repeat regions (STR) (n = 13, median = 11 months); the sample group with mlTNB from mirror repeats (MR) without inverted repeats (IR) (n = 7, median = 12 months); and the group with MR with IR (n = 9, median = 15 months).

According to the pathway analysis on gene signatures of sample clusters (see Methods), the high mlTNB-DR burden patients had mutation signatures enriched in MAPK and Notch signalling pathways, as compared to the other clusters enriched with double-stranded break and mismatch repair (cluster 1-IR), hedgehog and WNT signalling (cluster 2-STR) and interleukin-4 signalling (cluster 6-MR) pathways (Fig. 2d). There was a lack of significant association between mlTNB clusters with age, race, sex, PAAD subtypes [29, 30], KRAS mutation status and tumour purity. Additionally, there was no significant association between the non-B DNA clusters with markers of genomic instability, TMB, FGA, and MSI-score (measured by MANIS score) [31], whose distributions were similar across clusters and on average, were low in value (Fig. 2e). Specifically, the median TMB was 1.43, the median FGA, 0.13 and the median MSI-score was 0.28 across all groups. In the shortest PFS (high mlTNB-ZDNA, cluster 5-ZDNA) (Fig. 2f), chromosome 7 had the highest prevalence of non-B mutation co-localisation, while the longest PFS (high mlTNB-DR, cluster 3-DR) patient group non-B mutation co-localisation resided mainly on chromosome 5 (Fig. 2g). Our results lend support to the further study of mlTNB as a differentiating marker of survival in PAAD patients that may further inform on their heterogeneous response to treatment.