Whole-genome sequencing uncovers the genomic determinants of therapeutic resistance to immune checkpoint blockade

doi:10.21203/rs.3.rs-3318337/v1

Download PDF

Article

Whole-genome sequencing uncovers the genomic determinants of therapeutic resistance to immune checkpoint blockade

https://doi.org/10.21203/rs.3.rs-3318337/v1

This work is licensed under a CC BY 4.0 License

Version 1

posted

You are reading this latest preprint version

Checkpoint inhibitors (CPI), ameliorate the anti-tumour response by blocking inhibitory immune checkpoint receptors, and have revolutionised the treatment of advanced cancers. However, the prediction of treatment response is suboptimal, and there remains a strong reliance on tumour mutation burden (TMB). Studies to date are limited to whole exome sequencing (WES), with no data yet reported on the utility of whole genome sequencing (WGS) in a pan-cancer cohort. Here we report a pan-cancer cohort of 318 tumour/normal genomes from the Genomics England 100,000 Genomes Project cohort treated with CPIs. Pan-cancer biomarkers previously reported from WES such as clonal TMB, total neoantigen burden and TMB had continued utility in predicting treatment response. Clonal TMB remained the strongest univariate predictor of positive treatment outcome, followed by infiltrating T cell fraction, and tobacco/UV mutational signatures. using whole genome assay, we additionally detected novel signatures associated with poor outcomes, including markers reflecting chemotherapy-induced mutations. Patients treated with chemotherapy prior to CPI displayed reduced survival irrespective of tumour type and had more subclonal mutations. Structural variants (SVs) were also predictive of poor therapeutic response and were enriched with non-coding intronic breakpoints, generating significantly fewer neoantigens than expected by chance. Global genomic features such as telomere length were associated with poor survival following CPI treatment, particularly in renal and bladder cancers. Together, these validated and novel biomarkers showed collective utility when combined to predict CPI outcomes. Our results highlight the value of WGS in detecting biomarkers of treatment resistance and highlight the promise of WGS for use in clinical practice.

Biological sciences/Cancer/Tumour immunology

Biological sciences/Computational biology and bioinformatics/Genome informatics

Biological sciences/Immunology/Immunogenetics

Biological sciences/Genetics/Immunogenetics

Immune checkpoint inhibitors (CPIs) have revolutionized advanced cancer care and are approved for use in over 15 solid tumour types and in metastatic solid tumours classified as microsatellite instability-high (MSI-h) or mismatch repair deficient (MMRd)¹. Clinical response to CPI is highly variable both between and within tissue types. In the absence of reliable tumour-specific markers of resistance or sensitivity, there remains a high clinical reliance on tumour mutation burden (TMB) which is associated with response to CTLA-4^2–4 and PD-1/PD-L1 blockade⁵ in melanoma, and PD-1/PD-L1 blockade in non-small-cell lung carcinoma^6–8, urothelial cancers⁹, microsatellite unstable cancers¹⁰ and pan-tumour cohorts^11,12. TMB remains the only marker being routinely clinically applied¹³. In a recent meta-analysis, TMB (and clonal TMB) ranked as the top pan-cancer predictor of CPI response across 13 studies, each deriving biomarkers from either whole exome sequencing (WES) or transcriptomic data¹⁴. The inclusion of genetic features such as CXCL9 expression, UV, tobacco and APOBEC mutational signatures to a multivariate model was found to improve prediction from TMB alone, hinting at the potential utility of biomarker panels¹⁴.

To date, most studies investigating additional biomarkers utilized WES or single nucleotide polymorphism (SNP) arrays to identify genetic biomarkers of CPI response. While useful, these technologies lose valuable genomic information and suffer from several technical drawbacks. Whole-genome sequencing (WGS) offers technical advantages over WES, including the lack of exome capture bias, reduced variability in coverage across exonic regions, and the ability to detect alterations within or overlapping non-coding regions such as copy-number alterations (CNAs) and structural variants (SVs), and to profile excluded global features such as telomeres. As most of the human genome is non-coding, WGS enables the discovery of more mutations, enhancing methods such as non-negative matrix factorization (NMF) for the identification of mutational signatures (owing to a reduced Poisson variance), or minimizing the sum-squared error (SSE) in deconvolution approaches^15,16. Similarly, as WGS offers greater resolution to detect CNAs and point mutations, this facilitates more accurate reconstruction of mutation phylogeny¹⁷.

Importantly, WGS empowers unique biological insights, including the identification of alterations creating non-coding epitopes^18–20, or those which affect coding sequences via regulatory elements such as promoters or enhancers^21,22. WGS facilitates the identification of large and complex SVs overlapping intronic regions which can be more accurately mapped, particularly in cases with breakpoints overlapping intronic regions²³. Moreover, without the enrichment for human exonic regions, WGS allows less biased meta-genomic analyses of non-human DNA and the microbiome²⁴.

Herein, we present the first pan-cancer analysis to assess the entire genomic landscape in a cohort of patients treated with checkpoint inhibitor (CPI) therapy and to discover novel genomic biomarkers predicting survival following treatment. WGS was performed in n = 318 CPI-treated patients spanning multiple tumour types, recruited as part of the Genomics England 100,000 Genomes Project. We confirmed the utility of previous WES-derived biomarkers in WGS and finding novel predictors of survival following treatment. Our results suggest WGS-derived biomarkers may add additional utility in predicting treatment response when combined with TMB, particularly in tumour-type specific biomarker panels.

Cohort demographic and genomic overview

Within the Genomics England 100,000 genomes project cohort (GEL cohort), we identified a total of n = 318 patients who had received either anti-PD-1, anti-PD-L1, anti-CTLA-4 or a combination treatment of CPIs as part of their cancer care and whose tumour was subject to WGS sequencing before CPI treatment (median 221 days). Our pan-cancer cohort comprised 38.99% (124/318) melanoma, 26.41% (84/318) lung, 18.87% (60/318) renal, 4.4% (14/318) bladder, 3.46% (11/318) ovarian, 2.83% (9/318) breast and 5% (16/318) other tumour types (Fig. 1a). The majority of the cohort, 76.73% (244/318), received PD-1/PD-L1 targeting agents as a first CPI therapy, while 22.6% (74/318) had CTLA-4 or CTLA-4/PD-1/PD-L1 combination therapy. As a measure of treatment response, we used the time from the first CPI treatment to the date of the last follow-up or death (days). The median survival time for the cohort was 744 days (95% CI: 544-1,098 days), with a total of 179 events (Fig. 1a). Median coding TMB (cTMB) across the cohort was 4.93 muts/Mb, the median whole-genome TMB (wgTMB) was 6.17 muts/Mb and the median ploidy was 2.10 (Fig. 1a). Demographically, 44.97% (143/318) of the cohort were female and 7.81% (25/318) were non-white participants (Fig. 1b).

Whole-exome derived biomarkers perform similarly in whole-genome data

We first compared TMB derived from coding regions (cTMB) compared to a whole-genome derived TMB (wgTMB). We observed a strong correlation between counts, both pan-cancer (Spearmans rho = 0.98, p < 2.2x10^− 16, Fig. 1c), and across individual tumour types (Spearmans rho ≥ 0.84, Extended data Fig. 1a), reflecting that noncoding mutations mirror distinct biological processes, affecting the entire genome²². WgTMB was marginally skewed towards slightly higher counts (Fig. 1d), therefore upon applying a 10 muts/Mb threshold to classify patients into TMB high or low, we observed an additional 4.1% of patients would be reclassified as potentially sensitive to treatment (TMB high) using wgTMB compared to just 0.5% of patients where cTMB predicted a higher estimate (Fig. 1e). Segregating patients into wgTMB high and low based on this threshold showed distinct survival profiles (Fig. 1f). Segregation of patients into discrete groups (high vs low) performed similarly in a tumour-type adjusted Cox model for cTMB (HR 0.608, p = 0.00256) and wgTMB (HR 0.568, p = 0.000392). Across the entire GEL cohort of n = 15,208 cancer patients, TMB showed a treatment-specific association with outcome (HR 0.85, p = 0.03), predicting improved time from diagnosis to date of last follow-up or death, when incorporated into a Cox model adjusted by propensity scores derived from core demographic variables (see methods).

We next benchmarked the performance of canonical determinants of CPI response identified from WES studies against wgTMB and cTMB, including clonal and subclonal TMB¹⁴, estimated total predicted neoantigen burden, T cell infiltrate (predicted using T cell ExTRECT)²⁵, aneuploidy²⁶, Indel TMB²⁷ and HLA loss-of-heterozygosity (LOH, using LOHHLA)²⁸. Across the pan-cancer cohort, clonal TMB most significantly predicted improved overall survival (OS) following CPI treatment, with a hazard ratio (HR) of 0.684 (95% CI 0.549–0.852) p = 0.0007, followed by total neoantigen burden HR 0.722 (95% CI 0.566–0.920), p = 0.00841, cTMB HR 0.756 (95% CI 0.602–0.949), p = 0.0160 and wgTMB HR 0.759 (95% CI 0.603–0.954), p = 0.0182 (Fig. 1g). Tumour T cell infiltrate (TCRA fraction) showed a suggestive albeit nonsignificant association with survival following treatment, HR 0.851 (95% CI 0.723–1.0021), p = 0.0531, however, when also adjusted for wgTMB provided independent predictive value HR 0.847 (95% CI 0.721–0.994), p = 0.0429. In a subsample of the cohort, for patients with available digitised H&Es, we noted a moderate to strong correlation between tumour T cell infiltrate and manual TIL scoring performed by a pathologist (Spearman’s rho = 0.58, p = 4x10^− 4, Extended data Fig. 1b). Of five measures of lymphocyte spatial arrangement, we noted tumour T cell infiltrate estimated from the genomic TCRA fraction was correlated most strongly with overall stromal lymphocytes and central stromal TILs compared to intratumoural TILs or TILs at the invasive edge (Extended data Fig. 1c). We also observed within this subset, that patients with a high Clonal TMB had higher numbers of TILs (p = 0.0042, Extended data Fig. 1c). When corrected for baseline differences across tumour types a linear model showed clonal tmb and HLA-B LOH were both associated with reduced TIL infiltration on H&Es (adjusted linear model coefficients 8.785, p = 0.0119 and − 17.848, p = 0.0243 respectively).

As biomarkers were called relative to a matched normal DNA, we observed no significant racial inflation and underperformance for significant biomarkers (interaction with ancestry for clonal TMB p = 0.433, total neoantigen burden p = 0.373, cTMB p = 0.374 and wgTMB p = 0.297 respectively)²⁹. Most WES biomarkers also performed similarly across tumour types, however measures of TMB and total neoantigen burden associated more strongly in lung cancer (Extended data Fig. 1e).

Mutational signatures derived from WGS predict differential CPI response

Of 30 single-base WES-derived mutational signatures in COSMIC v2, those linked to tobacco, UV light and APOBEC activity have been previously associated with improved CPI response in a large meta-analysis conducted by our group¹⁴. COSMIC v3 now describes an updated compendium of signatures describing 77 single base substitutions (SBS), 11 double base substitutions (DBS) and 18 indel signatures^30,31. Within our cohort, signatures associated with improved CPI response were all associated with UV (SBS7a) or tobacco-induced mutational processes (DBS2. DBS4), after adjustment for tumour type and wgTMB (Fig. 2a). Many of these signatures were highly correlated to one another, reflecting the same underlying processes (Extended data Fig. 2a). UV and tobacco signatures were strongly specific for melanoma and lung tumours respectively (Fig. 2b). Interestingly, this analysis also identified four signatures associated with poor outcomes following CPI including SBS44 (defective DNA mismatch repair and microsatellite instability)³⁰, DBS3 (Polymerase epsilon exonuclease domain mutations))³⁰, DBS9 (associated with APOBEC activity)³² and SBS25 (platinum chemotherapy)³⁰. We did not observe any samples with SBS87, a thiopurine chemotherapy exposure-associated signature previously linked to improved CPI response in lung cancer³³. SBS38, previously described as CPI resistance-associated³⁴ in WGS-sequenced melanoma, was present in only 6.53% (8/123) of melanoma patients and was not associated with response (HR). Likewise, SBS5, also associated with resistance in melanoma³⁴, was found in all tumour groups and 35.22% (112/318) of the cohort, but did not predict survival after CPI in melanoma or pan-cancer (HR 1.185, p = 0.150 or HR 1.0714, p = 0.398).

As tumours with mutations attributed to chemotherapeutic agents were associated with poor overall survival after CPI agents, we compared the outcome of patients who had received chemotherapy either before (n = 156) or after CPI treatment (n = 71). We observed reduced survival in patients who had received chemotherapy before CPI (Fig. 2c), after correcting for tumour type (multivariate (MV) Cox HR 2.020, p = 0.0020) or tumour type and wgTMB (MV Cox HR 2.31, p = 0.00031). We recapitulated this finding in another independent cohort (p < 0.0001, Fig. 2d). We observed no significant differences in the wgTMB between samples taken before or after chemotherapy (12.571 vs 13.185 respectively, Wilcoxon test p = 0.943) nor a difference in total neoantigen burden (848.11 vs 745.58, p = 0.921). It has previously been noted that patients who have previously received chemotherapy may have higher numbers of subclonal neoantigens³⁵. We observed a higher number of low variant allele frequency (VAF < 5%) mutations in patients who received chemotherapy prior to CPI (one-way Wilcox test p = 0.00707, Fig. 2d), and a lower subclonal TMB (one-way Wilcox test, p = 0.033).

Immune germline SNPs do not strongly predict CPI response

Between 15–20% of the intratumoral variation of interferon signalling is heritable, therefore, modulating the immune response in cancer³⁷. Thus, we considered the effect of germline SNPs which can modulate the tumour-immune microenvironment (TIME)³⁷. Recent studies have identified germline variants in immune cells that can influence the tumour-immune microenvironment^37,38. These SNPs have been reported to influence follicular T cell infiltration³⁸, MHC class II expression³⁷, and associate with immune checkpoint blockade response, though the overall heritability of traits is less than 20%³⁷. To characterise the effect of TIME SNPs in our cohort, we considered any association of previously identified SNPs³⁷ with cTMB, wgTMB, and tumour T cell infiltration (quantified from DNA TCRA fraction²⁵. Only two SNPs, rs35356925 and rs1920021, which are negatively associated with interferon expression, were significantly positively associated with tumour TCRA fraction in renal cancer and malignant melanoma (Fig. 3a). These SNPs were significantly negatively associated with cTMB and wgTMB in bladder cancer (Fig. 3a). Next, we used Cox proportional hazards models to test for TIME SNP association with overall survival, in the context of TMB, sex, and disease type (see Methods). No SNPs were significantly associated with overall survival after false-discovery rate (FDR) correction (Fig. 3b), though four SNPs (6:32575658:C:G, OR: -0.19, p = 0.01, q = 0.07; 6:31783507:C:G, OR:-0.15, p = 0.049, q = 0.14, 6:32574305:A:G, OR: -0.16, p = 0.04, q = 0.1, 6:32575493:C:G, OR: -0.16, p = 0.04, q = 0.11), which are negatively associated with MHC-II expression, were nominally associated with overall survival following CPI therapy (Fig. 3b). Empirical power analysis confirmed that our current cohort of n = 318 is powered to detect SNPs with hazard ratios greater than 2, and with an event rate of 0.57 (Fig. 3c), therefore we are able to rule-out single SNPs which may double the risk of death following CPI therapy.

Structural variants facilitate genomic rearrangement with a minimal neoantigen footprint

Structural variants (SVs) represent a potential source of neoantigens³⁹, but also modulate the expression of key genes influencing tumour fates²³ and thus, may influence immunotherapy outcome. Measuring the total SV burden, we observed a median of 108 SVs per patient (range 16-1382). SV load significantly differed by tumour type and was highest in breast cancers, and lowest in renal cancers (Extended Fig. 3a). SV burden was only weakly correlated with wgTMB (Spearman’s rho = 0.270, p = 9.78x10^− 7).

We next examined how SV burden is associated with outcomes following CPI. Total SV burden was associated with worse outcomes, however, in particular, the total head-to-head (h2hINV), tail-to-tail inversions (t2tINV) and translocations (TRA) most strongly predicted poor survival after CPI treatment in univariable associations, compared to gene duplications (DUP) or deletions (DEL) (Fig. 4a). However, we noted strong collinearity between these SV measures (Extended data Fig. 3b). Tumours with a higher SV burden also had significantly fewer tumour-infiltrating T cells (TCRA fraction, Fig. 4b, two-way Wilcox test p = 6.08x10^− 6) when compared to patients with low SV burden. This held true when correcting for both tumour type and wgTMB (linear model coefficient for SV burden [continuous] -0.202, p = 0.000405). Thus, we compared the proportion of SVs which create predicted neoantigens (total SV neoantigens / total SV count), derived from either germline DNA; which are immune tolerated, or somatic DNA samples; which are subject to immune selection. The proportion of SVs forming predicted neoantigens was greatly reduced in somatic samples compared to the proportion in normal (p < 0.00001, Fig. 4c). While all tumours had at least one detected SV, only 27% of patients in the cohort (86/318) had one or more predicted SV neoantigens. In other mutation classes such as SNVs and indels, the total mutation count was strongly correlated with the number of predicted neoantigens (Spearman’s rho = 0.78, p < 2x10^− 16, Fig. 4d). In stark contrast, the total SV burden was weakly correlated with the number of predicted SV neoantigens (Spearman’s rho = 0.15, p = 0.014, Fig. 4e).

Chromothripsis, detected using ShatterSeek was present in 31.76% (101/318) of patients in the cohort. The presence of chromothripsis was associated with significantly reduced OS following CPI treatment (Log-Rank p = 0.003, Fig. 4f) and held true when adjusting for tumour type (MV Cox regression HR 1.441, p = 0.0195). The presence of chromothripsis was associated with a significantly higher SV burden (positive mean 146.009 vs negative mean 30.732; two-way Wilcox test p = 2.2x10^− 16) and reduced T cell infiltrate (positive mean 0.0683 vs negative mean 0.1207; two-way Wilcox test p = 1.083x10^− 5).

TP53 loss-of-function (LOF) mutations are suggested biomarkers for CPI response, often resulting in an increased mutational burden and increased neoantigen load^40–43. However, TP53-induced genomic instability is similarly associated with the development of chromothripsis^44–47. A significantly higher TMB and neoantigen burden was evident in TP53 LOF mutants (mean 14.636 vs 20.782; p = 0.00898 and 849.182 vs 1219.242; p = 0.00721; Extended data Fig. 3d&e), however, tumours with TP53 mutations were also more likely to have chromothripsis, occurring in 67.56% of tumours with TP53 LOF mutations versus just 37.62% in those without (Chi-squared test of independence p = 1.67x10^− 6, Fig. 4g). Patients with TP53 LOF mutations subsequently had lower tumour T cell fractions than those without (Extended data Fig. 3f).

Alterations in the MHC class I pathway associates with CPI resistance

Loss of antigen presentation is a known mechanism of CPI resistance^48,49. We aimed to assess the landscape of non-coding short variants affecting regulatory elements (REs) in the MHC class I antigen presentation pathway (see methods, Table 1). We considered mutations in promoter or enhancer regions regulating genes in the class I MHC presentation pathway. We utilized experimentally defined REs inferred from consensus across a range of RE databases, compiled and scored by Genehancer⁵⁰. REs listed in Genehancer were defined from five databases; dbSUPER, ENCODE, Ensembl, FANTOM5 and VISTA. We then applied stringent filtering to this database to improve confidence in RE:target interactions (see methods). From this filtered list, we assessed mutations in promoters, enhancers and enhancer/promoters targeting genes within the MHC class I presentation and survival following CPI therapy. For mutations occurring in a single enhancer, only one site had a significant association after correction for multiple testing. GH06J033061 (chr6:33061202–33063000 [GRCh38/hg38]), a ~ 1.8kb super-enhancer located on chromosome 6 which targets members of the TAP complex (TAP1, TAP2 and TAPBP), responsible for peptide transport into the ER and acts to stabilize the peptide-MHC complex during loading⁵¹. Patients with alterations in this enhancer had significantly shorter survival following CPI treatment (Extended data Fig. 4a). Finally, we also assessed LOH occurring within genes in this pathway and found only loss-of-heterozygosity in ER-resident aminopeptidase ERAP1 was associated with reduced survival following CPI treatment after controlling for multiple testing (Extended data Fig. 4b).

Telomere lengthening reduces CPI efficacy in renal and bladder cancers

Telomere elongation is a common adaptation in tumour cells and increased telomerase reverse transcriptase (TERT) activity has been linked with a more immunosuppressive TME⁵² and CPI response^53–55. Telomere length was estimated in both germline (normal, whole-blood) and tumour WGS samples using Telomerecat⁵⁶. Consistent with previous observations, telomere attrition was associated with age in normal (whole-blood) derived DNA (Spearman’s rho: -0.3, p = 7x10^− 7) but not in tumour samples (Spearman’s rho: -0.041, p = 0.500, Fig. 5a). Telomere length derived from tumour DNA differed marginally across tumour types, with significantly lower length in renal cancer with respect to lung and melanoma (pairwise Dunn’s test p = 0.0238 and p = 0.0293 respectively, Fig. 5b). Overall, compared to normal tissue, tumour tissue had a significantly reduced telomere length (two-sided Mann-Whitney U p = 5.3x10^− 5), however, when stratified by tumour type, this divergence was driven primarily by differences in renal and bladder cancers (one-sided Mann-Whitney U with Benjamini-Hochberg correction, p = 1.5x10^− 5 and p = 3.55x10^− 2 respectively, Fig. 5c&d).

Overall, across the pan-cancer cohort, increased telomere length in tumour tissue did not predict reduced survival following CPI therapy (Cox HR:1.00, p = 0.96). However, it was significantly associated with reduced survival in renal cancers (Cox HR 1.629, p = 0.0046) and bladder cancers (Cox HR 1.963, p = 0.039). Comparing those tumours within the top quartile of telomere lengths to those below this threshold highlights marked differences in overall survival following CPI therapy (Fig. 5e&f). Increased telomere lengths were not a consequence of canonical TERT activating mutations (C250T or C228T) in these cancer types as we observed no significant differences in tumours with or without these mutations (two-way Wilcox test, p = 0.505 and p = 0.1419 for renal and bladder respectively).

Increased intratumoral microbial presence is associated with reduced CPI response

CPI response is influenced by the gut microbiome, with increased gut microbiome diversity associated with improved treatment response. However, there is a paucity of evidence investigating CPI response in relation to the intratumoral microbiome. We therefore explored the association of the intratumoral microbiome with CPI response in a pan-cancer setting. The proportion of reads aligning to the microbiome was similar between tumour types (Kruskal-Wallis p = 0.72, Extended data Fig. 5a). Increased intratumoral microbiome proportion was found to associate with reduced survival following CPI treatment (Extended data Fig. 5b), which remained true when adjusted for tumour type (HR 1.176, p = 0.0236) and Tumour type with wgTMB (HR 1.171, p = 0.03049). Species diversity, measured by Shannon index, was not significantly associated with outcome (HR 1.083, p = 0.2714).

A multivariable model improves the prediction of treatment response compared to TMB alone

Finally, to determine whether WGS-derived biomarkers can improve the prediction of survival following CPI treatment compared to the clinical standard of TMB alone, we created a series of multivariable Cox models. Variables showing an association with survival after CPI in all prior analyses were taken forward for inclusion. To remove strong collinearity, we clustered all variables based on Spearman’s correlation and in highly correlated clusters, selected variables with the strongest univariate association with survival across the pan-cancer cohort. Finally, we applied elastic net penalisation with 5-fold cross validation over 1000 random seeds to remove variables with a coefficient of zero in all permutations. Models were constructed on the pan-cancer dataset and in the most represented tumour types, melanoma, lung, renal and all others. The multivariable pan-cancer model significantly outperformed that of TMB alone (partial likelihood test, p = 0.015), however, it had the lowest overall accuracy (c-index = 0.626, Fig. 6a). Constructing a separate model for the main three tumour types (Melanoma, Lung, Renal) and all others showed promise in improving survival prediction following CPI therapy, with all models outperforming TMB alone (Fig. 6a). This was most apparent in the lung (partial likelihood test p = 0.036, c-index 0.762 vs 0.627) and all others (partial likelihood test p = 0.046, c-index 0.683 vs 0.512). In the multi-variable models, clonal TMB most strongly predicted a positive therapeutic outcome across tumour types, except for the mixed all others group (Fig. 5b). Collectively, these genomic and clinical variables reflect five biological features, firstly at the genomic level; mutation quantity, neoantigen fitness, the broader tumour microenvironment, processes facilitating immune evasion and finally, host immunity (Fig. 5c).

This study represents the first in-depth pan-cancer investigation of the utility of WGS to uncover biomarkers which improve the prediction of CPI treatment outcomes and highlights the potential promise of WGS to develop improved predictive models and uncover novel biology.

Whole-exome-derived biomarkers such as total neoantigen burden and TMB remained strong as pan-cancer biomarkers in this cohort. We observed little difference in TMB derived from consensus coding regions⁵⁷ which comprise traditional panel-based arrays or the whole genome. Our results suggested that an additional ~ 4% of patients may meet the mutational burden threshold for CPI therapy if sequenced using WGS. Clonal TMB remained the strongest pan-cancer biomarker as previously observed in large meta-analyses¹⁴. In an updated assessment of mutational signatures, UV and tobacco-induced mutations were most strongly associated with positive CPI response, in Melanoma and Lung cancer, respectively. These observations are in line with previous observations based on WES data in earlier COSMIC versions^14,58–60 but highlight more specific substitution patterns associated with treatment response. For example, it is notable that only SBS7a is strongly associated with outcome following treatment, whereas no such association was seen for other UV signatures SBS7b or SBS7c. These results differed from a recent WGS analysis of CPI response in melanoma (n = 77) where UV signatures did not correlate with CPI response³⁴. This study also found SBS38 and SBS5 to be associated with CPI resistance, which we did not observe³⁴. We noted a strong association between SBS4 (tobacco) and outcome after CPI, which was also recently associated with durable clinical benefit in a WGS cohort of lung cancer patients³³.

We did observe signatures linked to poor CPI outcomes in the pan-cancer cohort such as SBS44 and DBS3. SBS44 is purported to be reflective of DNA mismatch repair deficiency³¹ and DSB3 is associated with polymerase mutations³⁰. If such processes do underly these signatures, we may have expected a positive association, as more favourable CPI outcomes have been observed in both MSI^61,62 and POLD1/POLE mutants tumours^63,64. However, given that our analyses were adjusted for wgTMB, it conceivable that such signatures do not provide any further independent benefit beyond increasing neoantigen burden. Additionally, while rare, we noted patients with a high proportion of mutations attributed to chemotherapy exposure (SBS25), correlated with a worse treatment outcome. Moreover, patients treated with chemotherapy before CPI had reduced overall survival, independent of tumour type. This association contradicts that of thiopurine chemotherapy exposure (SBS87) which was recently shown to be enriched in CPI responders in NSCLC³³ but was absent in our cohort. Platinum chemotherapies such as cisplatin are DNA-damaging agents which could theoretically increase total mutational burden and neoantigen load, however, we did not detect such changes in these tumours. Instead, we observed a greater burden of subclonal mutations in tumours treated with prior chemotherapy. The ratio of clonal to subclonal mutations or intratumoral heterogeneity (ITH) is a known driver of resistance to CPI therapy, and indeed, greater ITH is associated with a more suppressive tumour microenvironment, and reduced capacity for tumour rejection⁶⁵. Our observations highlight the ongoing imperative to study the best order for therapeutic interventions.

This study highlighted SV burden as a strong pan-cancer predictor of poor survival following CPI which was linked with reduced infiltrating T cells. This was despite SVs representing a potential source of tumour neoantigens^39,66. Somatic SVs underwent strong negative selection compared to germline SVs and formed fewer neoantigens compared to other mutation types. SVs could therefore, feasibly represent an immune-surveillance-efficient method of rearranging the genome in a pro-tumorigenic manner compared to short indels or SNVs. Many of the tumours with the highest SV burden also had chromothripsis which presented similarly, with poor outcome and low tumour T cell infiltrate. These results complement those of Newell et al., who identified SV burden and chromothripsis as CPI-resistance associated in melanoma patients³⁴. We were able to extend these observations, confirming these genomic features as pan-cancer markers of treatment resistance. An important related observation was the strong co-occurrence between TP53 LOF mutations and chromothripsis. TP53 mutations have been suggested as biomarkers of improved CPI-response^41–43, as disruption of DNA repair yields an increased mutational burden and subsequently higher neoantigen load. While this was also true in our cohort, we observed a higher SV burden and ~ 65% rate of chromothripsis in those with LOF TP53 mutations. This may partly explain the variability of association with CPI-response seen between studies for this biomarker¹⁴. TP53 loss is already established to be intrinsically linked to chromothripsis^44–47, therefore, our results highlight a clear dichotomy between the potential for TP53 inactivation to improve the immune response, promoting increased mutational and neoantigen load, or conversely, facilitate genomic rearrangement with minimal neoantigen production.

We noted some alterations in class I antigen presentation pathway which also showed promise in predicting poor therapeutic response to CPI, such as mutations in enhancer regions, structural variants and loss-of-heterozygosity. It stands to reason that evolutionary pressures would favour the inactivation of such key members through multiple mechanistic routes. Our results, however, should be cautiously interpreted until they are further validated.

We observed a strong association with longer tumour telomere length and reduced survival following CPI treatment, particularly in renal and bladder tumours where telomeres were most elongated compared to DNA derived from normal tissue. Increased TERT activity has been linked with an immunosuppressive microenvironment, inducing CXCL10 expression, which stimulates the infiltration of suppressive T-cell populations and increases the percentage of CD4 + and FOXP3 + cells⁵². To our knowledge, our study is the first to link somatic telomere elongation and CPI outcome, however, recurrent TERT promoter mutations which increase transcriptional activation, namely, C250T and C228T⁶⁷ have been studied in the context of CPI response. Our results support evidence in renal cell carcinoma which identified an enrichment of TERT promoter mutations in patients with reduced CPI benefit⁵³. In our cohort, the presence of these alterations did not explain the differences in telomere length, therefore the exact mechanism of elongation remains uncertain. Other studies of TERT promoter mutations, both in urothelial carcinoma and pan-cancer have noted a potential therapeutic benefit from CPI^54,55.

In our study, tumours with a higher intratumoral microbial proportion had a worse therapeutic response to CPI. There are several possible mechanisms by which intratumoral microbes could reduce anti-tumour immunity, including the upregulation of reactive oxygen species, promoting an anti-inflammatory environment, T cell inactivation, and immunosuppression⁶⁸. A recent study of early-stage lung cancer found that bacterial burden was significantly higher in tumour cells compared to normal cells, suggesting this may be a feature of the more aggressive disease⁶⁹. These results run contrary to data linking intratumoral microbiota to improved increased immune infiltrate⁷⁰ or novel antigen presentation⁷¹. Our results build on early evidence of the significance of the microbiome in the TME and emphasize the importance of intratumoral microbiome composition in influencing CPI response.

Finally, this work highlights the potential for WGS to detect genomic alterations which can complement TMB in predicting treatment response. We note that these models require further validation as new WGS data becomes available. However, our findings are promising particularly in cancer types where TMB appears relatively uninformative. Whole-genome sequencing of cancers is increasingly recognized as an important component of oncological care. WGS offers a more uniform distribution of coverage depth, genotype quality, and minor read ratio for the detection of SNVs and indels, compared to WES⁷². Moreover, studies have indicated more than a four-fold reduction in the proportion of false-positive SNVs than for WES, in addition to improved detection of copy-number alterations and copy-neutral SVs⁷², The clinical implementation of WGS is highly feasible, with NHS England already rolling out WGS for all children with cancer ⁷³. In this context, WGS has proven to have clinical utility, beyond standard-of-care assays⁷³.

Patient selection

Patients were selected from the 100,000 Genomes Project, a UK initiative led by Genomics England (GEL), aiming to study health and disease through the whole-genome sequencing of 100,000 genomes in the UK⁷³. Patients were selected using the systemic anti-cancer therapy (SACT) treatment data (version 13) identifying any patients treated with immune checkpoint blockade inhibition therapy and then filtered for those who received this therapy as part of their cancer care. Finally, only patients who had their tumours sampled for WGS before treatment were used in this cohort. This formed the pan-cancer GEL CPI cohort. The cohort consists of patients with different types of cancer, including Melanoma (n = 124), Lung (n = 84), Renal (n = 60), Other (n = 16), Bladder (n = 14), Ovarian (n = 11), and Breast cancer (n = 9). Other tumour types included adult glioma, prostate, colorectal, endometrial, oral oropharyngeal, sarcoma, hepatobiliary, haemonc and upper gastrointestinal cancers. Overall survival was calculated from the start of CPI treatment until either the date of the last follow-up or death. At the time of analysis, RECIST response data were not available for this cohort.

Processing of WGS data

The genomes of the cohort were aligned to the hg38 reference genome (GRCh38 Decoy assembly) using the illumina Isaac pipeline v03.16.02.1973, which has been shown to introduce bias by preferentially soft-clipping reads supporting non-reference alleles (semi-aligned reads), potentially affecting downstream analyses. Therefore, the FixVAF tool was used to correct this bias⁷⁴. General file manipulations were conducted with BEDtools v2.27.1, BCFtools v1.10.2, and tabix v0.2.6. VEP v89.4 was used to annotate variants. Read clipping was carried out using SAMtools v1.9 and GATK v4.1.4.1, while HLA HD was employed for HLA allele calling⁷⁵. Indel TMB was calculated by counting the total number of indels per Mb in coding regions defined by the Friends TMB harmonisation project⁵⁷. Estimation of copy number alterations (CNAs) was conducted using Battenberg¹⁷. Reconstruction of clonal and subclonal mutations was performed using DPclust¹⁷.

Estimation of coding and whole genome mutational metrics

Coding TMB was estimated by counting the total number of filtered variants (passing all filters) overlapping all regions defined by coding regions defined by the Friends TMB harmonization project, per patient. The total variants were then divided by the effective genome size. Indel TMB was calculated as for coding TMB however, first filtering for insertion or deletion mutations. Whole-genome TMB was calculated similarly however taking into account the total number of filtered mutations in any region of the genome.

Image acquisition process summary

Haematoxylin and eosin-stained sections of tumours from participants of Genomics England’s 100,000 Genome Project were scanned at the National Pathology Imaging Cooperative (NPIC) Centre in Leeds. Slides were pseudoanomysed prior to scanning. All the slides were scanned on a single Leica GT450 DX scanner at a resolution of 0.26mm/pixel (equivalent to approximately 40x magnification) producing an SVS format image. These images were JPEG compressed with a quality value of 91 with the Segue_V1 ICC colour calibration profile applied. Manual quality control checks were performed on all images and if required images were re-scanned.

The image data were generated as part of the collaboration between Genomics England’s Cancer 2.0 Multimodal Program and NPIC to generate research pathology data for the 100,000 Genome Project cancer cohort⁷⁶. Original diagnostic tissue specimens for the 100,000 Genome Project underwent routine processing at the participating NHS pathology laboratories where the diagnostic glass slides were generated and stored. Individual trusts were funded by Genomics England to send the relevant slides to the NPIC Centre for scanning.

Data access will be available by application to the Genomics England Research Environment in due course. This project was a pilot study for access and analysis of image data.

Histological assessment of tumour infiltrating lymphocytes (TILs)

Slides were included from n = 74 participants in this study. Up to five scanned haematoxylin and eosin (H&E)-stained slides containing invasive tumour were identified from each participant. If present, at least one slide was selected from the periphery of the tumour. If suitable slides were not identified, the participant was excluded from the study.

For each selected slide, TILs were classified as lymphocytes and plasma cells. These were subdivided into intratumoural (iTILs) and stromal TIL (sTILs) and were scored according to the guidance published by The International Immuno-Oncology Biomarker Working Group^77,78. If the tumour periphery, defined as tumour lying within 1mm of the invasive edge of the tumour margin, was present in the tissue sections, the invasive edge sTILs and the central sTILs were scored independently. If scanned slides were only represented by metastatic disease to a lymph node, only iTILs were scored. The presence and absence of a lymphoid aggregate and or an aggregate with a germinal centre were also recorded.

All scoring was undertaken by one pathologist (LR) following discussion with KL and AMF. LR was blinded to the results of the genomic analysis.

Fitting of known mutational signatures

SBS96, DBS78 and ID83 mutational signatures were estimated using DeconstructSigs¹⁶ and COSMIC v3.1. For testing associations with OS following CPI, signatures were included in a multivariable Cox regression model along with tumour type (to account for tumour-specific survival differences) and wgTMB (to account for overall mutational burden). P-values were corrected for multiple comparisons using the Benjamini-Hochberg correction. For plotting purposes, only signatures with FDR < 0.99 were shown.

Investigating non-coding mutations in the class I antigen presentation pathway

Distal regulatory elements listed in Genehancer were defined from five databases; dbSUPER, ENCODE, Ensembl, FANTOM5 and VISTA⁵⁰. We considered regulatory elements listed in at least four or more independent databases. Promoters were defined as listed on Ensembl. Elements were filtered further based on Genehancer (GH) score (an overall measure of confidence) and gene association (GA) score (indicates the confidence that a given regulatory element effects the transcription of a target gene) considering only the top 30% of elements. As promoters, enhancers and enhancer promoters were defined as separate regulatory element types, percentile cut-offs for the top 30% were calculated relative to each group separately. For promoters the top 70% of GH and GA scores were 0.71 and 0.65, for enhancers 0.7 and 3.17, and for enhancer/promoters was 1.9 and 2.1 respectively. From this filtered list we assessed variants in enhancers targeting genes within the MHC class I presentation pathway.

In our analyses, we considered core components of the MHC class I antigen presentation pathway defined by the Molecular signatures database, specifically genes listed under the gene ontology (GO) biological process: “antigen processing and presentation of peptide antigen via MHC class I” which included 29 members (Table 1).

Table 1

MHC Class I genes considered in analyses.
NCBI Gene Id	Gene symbol	Description
10437	IFI30	IFI30 lysosomal thiol reductase
1636	ACE	angiotensin I converting enzyme
2207	FCER1G	Fc fragment of IgE receptor Ig
23457	ABCB9	ATP binding cassette subfamily B member 9
2923	PDIA3	protein disulfide isomerase family A member 3
3077	HFE	homeostatic iron regulator
3105	HLA-A	major histocompatibility complex, class I, A
3106	HLA-B	major histocompatibility complex, class I, B
3107	HLA-C	major histocompatibility complex, class I, C
3133	HLA-E	major histocompatibility complex, class I, E
3134	HLA-F	major histocompatibility complex, class I, F
3135	HLA-G	major histocompatibility complex, class I, G
3136	HLA-H	major histocompatibility complex, class I, H
3140	MR1	major histocompatibility complex, class I-rel
3416	IDE	insulin degrading enzyme
3551	IKBKB	inhibitor of nuclear factor kappa B kinase subunit beta
4012	LNPEP	leucyl and cystinyl aminopeptidase
50856	CLEC4A	C-type lectin domain family 4 member A
51128	SAR1B	secretion associated Ras related GTPase 1B
51752	ERAP1	endoplasmic reticulum aminopeptidase 1
54842	MFSD6	major facilitator superfamily domain containing 6
55080	TAPBPL	TAP binding protein like
563	AZGP1	alpha-2-glycoprotein 1, zinc-binding
567	B2M	beta-2-microglobulin
64167	ERAP2	endoplasmic reticulum aminopeptidase 2
6890	TAP1	transporter 1, ATP binding cassette subfamily...
6891	TAP2	transporter 2, ATP binding cassette subfamily...
6892	TAPBP	TAP binding protein
811	CALR	calreticulin

Testing for association between germline TIME SNPs and CPI response

Genome-wide significant tumour-immune microenvironment (TIME) associated SNPS were obtained from Sayaman et al³⁷. 170 SNPs identified in gVCFs that passed filtering were retained and used for analysis. The Pearson R statistical test was used to test individual SNP association with whole genome TMB, coding TMB, and TCRA fraction, respectively, within individual disease types. SNP GT values were converted to 0/1/2 numeric values depending on major or minor alleles associated with the published phenotype in Sayaman et al. For overall survival analysis, the Cox PH model with default parameters was used from the lifelines python package (v0.27.2). SNPs were evaluated for association with overall survival in the context of sex, scaled whole genome TMB, and correcting for tumour type. Benjamini-Hochberg multiple testing correction was applied for all analyses.

Testing for association between TP53 loss of function variants and chromothripsis

Loss-of-function variants were defined as those in known hotspots affecting TP53 function structurally, altering DNA-binding or resulting in carboxy-terminal truncation as defined by Hainaut et al.,⁷⁹. These mutations are outlined in table 2. Additionally, we considered frameshift indel variants which would disrupt this gene.

Table 2

TP53 loss-of-function mutations.
Position (codon)	cDNA/base change	Protein/ amino acid change	Mutation effect	Chrom	start	end
175	c.524G > A	p.R175H	Structural	17	7675088	7675088
196	c.586C > T	p.R196*	Structural	17	7674945	7674945
213	c.637C > T	p.R213*	Null	17	7674894	7674894
220	c.659A > G	p.Y220C	Structural	17	7674872	7674872
245	c.733G > A	p.G245S	Structural	17	7674230	7674230
248	c.742C > T	p.R248W	DNA-binding	17	7674221	7674221
248	c.743G > A	p.R248Q	DNA-binding	17	7674220	7674220
273	c.818G > A	p.R273H	DNA-binding	17	7673802	7673802
273	c.817C > T	p.R273C	DNA-binding	17	7673803	7673803
282	c.844C > T	p.R282W	DNA-binding	17	7673776	7673776
306	c.916C > T	p.R306*	Carboxy-terminal truncation	17	7673704	7673704
342	c.1024C > T	p.R342*	Carboxy-terminal truncation	17	7670685	7670685

Telomere length estimation and telomere altering mutations

Telomerecat v3.30 was used to estimate telomere length from tumour and normal (whole blood) WGS bam files using default parameters. Telomerecat is a ploidy-agnostic method for estimating telomere length which accounts for both aneuploidy and interstitial telomeric reads, resulting in estimations more appropriate in the analysis of cancer genomes⁵⁶.Telomere content was estimated from tumour and germline bam files by the methods Telomere Hunter v1.1.0⁸⁰ and Telomerecat v3.3.0⁵⁶ using default parameters. Telomere content was normalized by log2(Tumour/Germline).

TERT promoter mutations

Mutations in positions C250T (chr5:1,295,113 [G > A]) and C228T (chr5:1,295,135 [G > A]) for hg38 were considered as canonical TERT activating promoter mutations as previously defined⁶⁷. Mean tumour telomere length was compared between patients with or without either promoter mutation.

Metagenomic and microbiome estimation

Kraken2 was used to estimate the total microbial proportion in samples⁸¹. The Kraken method utilises a database which contains records consisting of a k-mer and the last common ancestor (LCA) of all organisms whose genomes contain that k-mer. Here, reads which are unaligned to the human genome are then supplied for classification. Reads are searched against the database for each segment of k letters in a sequence, and using the common ancestor of all the resulting taxa to assign a label to the sequence⁸².

Multivariable model construction

Variables showing an association with survival after CPI in all prior analyses were taken forward for inclusion. Additionally, we considered clinical and demographic variables such as age of diagnosis, prior chemotherapy, sex and ethnicity. To remove strong collinearity, we clustered all genomic variables based on Spearman’s correlation and in highly correlated clusters, selected variables with the strongest univariate association with survival across the pan-cancer cohort. Finally, we applied elastic net penalisation with 5-fold cross validation over 1000 random seeds to remove variables with a coefficient of zero in all permutations. Standard Cox regression models were then fit with the remaining variables.

ACKNOWLEDGEMENTS

This research was made possible through access to the data and findings generated by the 100,000 Genomes Project. The 100,000 Genomes Project is managed by Genomics England Limited (a wholly owned company of the Department of Health and Social Care). The 100,000 Genomes Project is funded by the National Institute for Health Research and NHS England. The Wellcome Trust, Cancer Research UK and the Medical Research Council also funded research infrastructure. The 100,000 Genomes Project uses data provided by patients and collected by the National Health Service as part of their care and support.

GENOMICS ENGLAND RESEARCH CONSORTIUM

Ambrose, J. C.¹; Arumugam, P.¹; Bevers, R.¹; Bleda, M.¹; Boardman-Pretty, F. ^1,2; Boustred, C. R.¹; Brittain, H.¹ ; Brown, M.A.¹; Caulfield, M. J.^1,2; Chan, G. C.¹; Giess A.¹; Griffin, J. N.¹; Hamblin, A.¹; Henderson, S.^1,2; Hubbard, T. J. P. ¹; Jackson, R.¹ ; Jones, L. J.^1,2; Kasperaviciute, D.^1,2; Kayikci, M.¹; Kousathanas, A.¹; Lahnstein, L.¹; Lakey, A.¹; Leigh, S. E. A.¹; Leong, I. U. S.¹; Lopez, F. J.¹; Maleady-Crowe, F.¹; McEntagart, M.¹; Minneci F.¹; Mitchell, J.¹; Moutsianas, L.^1,2; Mueller, M.^1,2; Murugaesu, N.¹; Need, A. C.^1,2; O‘Donovan P.¹; Odhams, C. A.¹; Patch, C.^1,2; Perez-Gil, D.¹; Pereira, M. B.¹; Pullinger, J.¹; Rahim, T.¹; Rendon, A.¹; Rogers, T.¹; Savage, K.¹; Sawant, K.¹; Scott, R. H.¹; Siddiq, A.¹; Sieghart, A.¹; Smith, S. C.¹; Sosinsky, A.^1,2; Stuckey, A.¹; Tanguy M.¹; Taylor Tavares, A. L.¹; Thomas, E. R. A.^1,2; Thompson, S. R.¹; Tucci, A.^1,2; Welland, M. J.¹; Williams, E.¹; Witkowska, K.^1,2; Wood, S. M.^1,2; Zarowiecki, M.¹.

1. Genomics England, London, UK.

2. William Harvey Research Institute, Queen Mary University of London, London, UK.

DATA AVAILABILITY

The National Genomic Research Library (NGRL) functions as a "reading library," which means that direct data extraction is not possible. However, all Whole Genome Sequencing (WGS) data, BAM files, and processed files mentioned can be accessed in situ through the Lung Cancer Genomics England Clinical Interpretation Partnership (GECIP) once an individual's data access request has been approved. To become a GECIP member for access, please visit the link provided at https://www.genomicsengland.co.uk/research/academic/join-gecip. This process involves completing an online application, undergoing verification by the applicant's institution, successfully finishing a concise information governance training course (approximately 30 minutes), and receiving approval from the Lung Cancer domain lead (Charles Swanton). Further information can be found at https://www.genomicsengland.co.uk/research/academic.

All open source tools used in this manuscript were cited in the Methods. No custom code was used for any aspect of data processing or analysis.

Yes, I can confirm we applied and received approval to access the data. Here is the link to our approved project in the research registry:

https://research.genomicsengland.co.uk/research-registry/browse/Browsetheresearchregistrypublic/?id=98a5c154-78d5-ea11-a813-000d3a7ed588

Ramos-Casals M, Brahmer JR, Callahan MK, et al. Immune-related adverse events of checkpoint inhibitors. Nat Rev Dis Primers. 2020;6(1):38.
Snyder A, Makarov V, Merghoub T, et al. Genetic basis for clinical response to CTLA-4 blockade in melanoma. N Engl J Med. 2014;371(23):2189-2199.
Van Allen EM, Miao D, Schilling B, et al. Genomic correlates of response to CTLA-4 blockade in metastatic melanoma. Science. 2015;350(6257):207-211.
Hugo W, Zaretsky JM, Sun L, et al. Genomic and Transcriptomic Features of Response to Anti-PD-1 Therapy in Metastatic Melanoma. Cell. 2017;168(3):542.
Riaz N, Havel JJ, Makarov V, et al. Tumor and Microenvironment Evolution during Immunotherapy with Nivolumab. Cell. 2017;171(4):934-949 e916.
Rizvi NA, Hellmann MD, Snyder A, et al. Cancer immunology. Mutational landscape determines sensitivity to PD-1 blockade in non-small cell lung cancer. Science. 2015;348(6230):124-128.
Carbone DP, Reck M, Paz-Ares L, et al. First-Line Nivolumab in Stage IV or Recurrent Non-Small-Cell Lung Cancer. N Engl J Med. 2017;376(25):2415-2426.
Rizvi H, Sanchez-Vega F, La K, et al. Molecular Determinants of Response to Anti-Programmed Cell Death (PD)-1 and Anti-Programmed Death-Ligand 1 (PD-L1) Blockade in Patients With Non-Small-Cell Lung Cancer Profiled With Targeted Next-Generation Sequencing. J Clin Oncol. 2018;36(7):633-641.
Rosenberg JE, Hoffman-Censits J, Powles T, et al. Atezolizumab in patients with locally advanced and metastatic urothelial carcinoma who have progressed following treatment with platinum-based chemotherapy: a single-arm, multicentre, phase 2 trial. Lancet. 2016;387(10031):1909-1920.
Le DT, Uram JN, Wang H, et al. PD-1 Blockade in Tumors with Mismatch-Repair Deficiency. N Engl J Med. 2015;372(26):2509-2520.
Cristescu R, Mogg R, Ayers M, et al. Pan-tumor genomic biomarkers for PD-1 checkpoint blockade-based immunotherapy. Science. 2018;362(6411).
Samstein RM, Lee CH, Shoushtari AN, et al. Tumor mutational load predicts survival after immunotherapy across multiple cancer types. Nat Genet. 2019;51(2):202-206.
Marcus L, Fashoyin-Aje LA, Donoghue M, et al. FDA Approval Summary: Pembrolizumab for the Treatment of Tumor Mutational Burden-High Solid Tumors. Clin Cancer Res. 2021;27(17):4685-4689.
Litchfield K, Reading JL, Puttick C, et al. Meta-analysis of tumor- and T cell-intrinsic mechanisms of sensitization to checkpoint inhibition. Cell. 2021;184(3):596-614 e514.
Alexandrov LB, Nik-Zainal S, Wedge DC, Campbell PJ, Stratton MR. Deciphering signatures of mutational processes operative in human cancer. Cell Rep. 2013;3(1):246-259.
Rosenthal R, McGranahan N, Herrero J, Taylor BS, Swanton C. DeconstructSigs: delineating mutational processes in single tumors distinguishes DNA repair deficiencies and patterns of carcinoma evolution. Genome Biol. 2016;17:31.
Nik-Zainal S, Van Loo P, Wedge DC, et al. The life history of 21 breast cancers. Cell. 2012;149(5):994-1007.
Laumont CM, Vincent K, Hesnard L, et al. Noncoding regions are the main source of targetable tumor-specific antigens. Sci Transl Med. 2018;10(470).
Panda A, de Cubas AA, Stein M, et al. Endogenous retrovirus expression is associated with response to immune checkpoint blockade in clear cell renal cell carcinoma. JCI Insight. 2018;3(16).
Smith CC, Beckermann KE, Bortone DS, et al. Endogenous retroviral signatures predict immunotherapy response in clear cell renal cell carcinoma. J Clin Invest. 2018;128(11):4804-4820.
Consortium ITP-CAoWG. Pan-cancer analysis of whole genomes. Nature. 2020;578(7793):82-93.
Rheinbay E, Nielsen MM, Abascal F, et al. Analyses of non-coding somatic drivers in 2,658 cancer whole genomes. Nature. 2020;578(7793):102-111.
Li Y, Roberts ND, Wala JA, et al. Patterns of somatic structural variation in human cancer genomes. Nature. 2020;578(7793):112-121.
Poore GD, Kopylova E, Zhu Q, et al. Microbiome analyses of blood and tissues suggest cancer diagnostic approach. Nature. 2020;579(7800):567-574.
Bentham R, Litchfield K, Watkins TBK, et al. Using DNA sequencing data to quantify T cell fraction and therapy response. Nature. 2021;597(7877):555-560.
Davoli T, Uno H, Wooten EC, Elledge SJ. Tumor aneuploidy correlates with markers of immune evasion and with reduced response to immunotherapy. Science. 2017;355(6322).
Turajlic S, Litchfield K, Xu H, et al. Insertion-and-deletion-derived tumour-specific neoantigens and the immunogenic phenotype: a pan-cancer analysis. Lancet Oncol. 2017;18(8):1009-1021.
McGranahan N, Rosenthal R, Hiley CT, et al. Allele-Specific HLA Loss and Immune Escape in Lung Cancer Evolution. Cell. 2017;171(6):1259-1271 e1211.
Nassar AH, Adib E, Abou Alaiwi S, et al. Ancestry-driven recalibration of tumor mutational burden and disparate clinical outcomes in response to immune checkpoint inhibitors. Cancer Cell. 2022;40(10):1161-1172 e1165.
Alexandrov LB, Kim J, Haradhvala NJ, et al. The repertoire of mutational signatures in human cancer. Nature. 2020;578(7793):94-101.
Drost J, van Boxtel R, Blokzijl F, et al. Use of CRISPR-modified human stem cell organoids to study the origin of mutational signatures in cancer. Science. 2017;358(6360):234-238.
Farmanbar A, Firouzi S, Kneller R, Khiabanian H. Mutational signatures reveal ternary relationships between homologous recombination repair, APOBEC, and mismatch repair in gynecological cancers. J Transl Med. 2022;20(1):65.
Donker HC, van Es B, Tamminga M, et al. Using genomic scars to select immunotherapy beneficiaries in advanced non-small cell lung cancer. Sci Rep. 2023;13(1):6581.
Newell F, Pires da Silva I, Johansson PA, et al. Multiomic profiling of checkpoint inhibitor-treated melanoma: Identifying predictors of response and resistance, and markers of biological discordance. Cancer Cell. 2022;40(1):88-102 e107.
McGranahan N, Furness AJ, Rosenthal R, et al. Clonal neoantigens elicit T cell immunoreactivity and sensitivity to immune checkpoint blockade. Science. 2016;351(6280):1463-1469.
Chowell D, Yoo SK, Valero C, et al. Improved prediction of immune checkpoint blockade efficacy across multiple cancer types. Nat Biotechnol. 2022;40(4):499-506.
Sayaman RW, Saad M, Thorsson V, et al. Germline genetic contribution to the immune landscape of cancer. Immunity. 2021;54(2):367-386 e368.
Shahamatdar S, He MX, Reyna MA, et al. Germline Features Associated with Immune Infiltration in Solid Tumors. Cell Rep. 2020;30(9):2900-2908 e2904.
Hundal J, Miller CA, Griffith M, et al. Cancer Immunogenomics: Computational Neoantigen Identification and Vaccine Design. Cold Spring Harb Symp Quant Biol. 2016;81:105-111.
Blagih J, Zani F, Chakravarty P, et al. Cancer-Specific Loss of p53 Leads to a Modulation of Myeloid and T Cell Responses. Cell Rep. 2020;30(2):481-496 e486.
Aredo JV, Padda SK, Kunder CA, et al. Impact of KRAS mutation subtype and concurrent pathogenic mutations on non-small cell lung cancer outcomes. Lung Cancer. 2019;133:144-150.
Assoun S, Theou-Anton N, Nguenang M, et al. Association of TP53 mutations with response and longer survival under immune checkpoint inhibitors in advanced non-small-cell lung cancer. Lung Cancer. 2019;132:65-71.
Lin X, Wang L, Xie X, et al. Prognostic Biomarker TP53 Mutations for Immune Checkpoint Blockade Therapy and Its Association With Tumor Microenvironment of Lung Adenocarcinoma. Front Mol Biosci. 2020;7:602328.
Cortes-Ciriano I, Lee JJ, Xi R, et al. Comprehensive analysis of chromothripsis in 2,658 human cancers using whole-genome sequencing. Nat Genet. 2020;52(3):331-341.
Grobner SN, Worst BC, Weischenfeldt J, et al. The landscape of genomic alterations across childhood cancers. Nature. 2018;555(7696):321-327.
Tolomeo D, L'Abbate A, Lonoce A, et al. Concurrent chromothripsis events in a case of TP53 depleted acute myeloid leukemia with myelodysplasia-related changes. Cancer Genet. 2019;237:63-68.
Rucker FG, Dolnik A, Blatte TJ, et al. Chromothripsis is linked to TP53 alteration, cell cycle impairment, and dismal outcome in acute myeloid leukemia with complex karyotype. Haematologica. 2018;103(1):e17-e20.
Sade-Feldman M, Jiao YJ, Chen JH, et al. Resistance to checkpoint blockade therapy through inactivation of antigen presentation. Nat Commun. 2017;8(1):1136.
Kalbasi A, Ribas A. Tumour-intrinsic resistance to immune checkpoint blockade. Nat Rev Immunol. 2020;20(1):25-39.
Fishilevich S, Nudel R, Rappaport N, et al. GeneHancer: genome-wide integration of enhancers and target genes in GeneCards. Database (Oxford). 2017;2017.
Mantel I, Sadiq BA, Blander JM. Spotlight on TAP and its vital role in antigen presentation and cross-presentation. Mol Immunol. 2022;142:105-119.
Mao J, Zhang Q, Wang Y, et al. TERT activates endogenous retroviruses to promote an immunosuppressive tumour microenvironment. EMBO Rep. 2022;23(4):e52984.
Dizman N, Lyou Y, Salgia N, et al. Correlates of clinical benefit from immunotherapy and targeted therapy in metastatic renal cell carcinoma: comprehensive genomic and transcriptomic analysis. J Immunother Cancer. 2020;8(2).
Li H, Li J, Zhang C, Zhang C, Wang H. TERT mutations correlate with higher TMB value and unique tumor microenvironment and may be a potential biomarker for anti-CTLA4 treatment. Cancer Med. 2020;9(19):7151-7160.
de Kouchkovsky I, Zhang L, Philip EJ, et al. TERT promoter mutations and other prognostic factors in patients with advanced urothelial carcinoma treated with an immune checkpoint inhibitor. J Immunother Cancer. 2021;9(5).
Farmery JHR, Smith ML, Diseases NB-R, Lynch AG. Telomerecat: A ploidy-agnostic method for estimating telomere length from whole genome sequencing data. Sci Rep. 2018;8(1):1300.
Merino DM, McShane LM, Fabrizio D, et al. Establishing guidelines to harmonize tumor mutational burden (TMB): in silico assessment of variation in TMB quantification across diagnostic platforms: phase I of the Friends of Cancer Research TMB Harmonization Project. J Immunother Cancer. 2020;8(1).
Anagnostou V, Niknafs N, Marrone K, et al. Multimodal genomic features predict outcome of immune checkpoint blockade in non-small-cell lung cancer. Nat Cancer. 2020;1(1):99-111.
Miao D, Margolis CA, Vokes NI, et al. Genomic correlates of response to immune checkpoint blockade in microsatellite-stable solid tumors. Nat Genet. 2018;50(9):1271-1281.
Trucco LD, Mundra PA, Hogan K, et al. Ultraviolet radiation-induced DNA damage is prognostic for outcome in melanoma. Nat Med. 2019;25(2):221-224.
Marcus L, Lemery SJ, Keegan P, Pazdur R. FDA Approval Summary: Pembrolizumab for the Treatment of Microsatellite Instability-High Solid Tumors. Clin Cancer Res. 2019;25(13):3753-3758.
Lemery S, Keegan P, Pazdur R. First FDA Approval Agnostic of Cancer Site - When a Biomarker Defines the Indication. N Engl J Med. 2017;377(15):1409-1412.
Wang F, Zhao Q, Wang YN, et al. Evaluation of POLE and POLD1 Mutations as Biomarkers for Immunotherapy Outcomes Across Multiple Cancer Types. JAMA Oncol. 2019;5(10):1504-1506.
Garmezy B, Gheeya J, Lin HY, et al. Clinical and Molecular Characterization of POLE Mutations as Predictive Biomarkers of Response to Immune Checkpoint Inhibitors in Advanced Cancers. JCO Precis Oncol. 2022;6:e2100267.
Wolf Y, Bartok O, Patkar S, et al. UVB-Induced Tumor Heterogeneity Diminishes Immune Response in Melanoma. Cell. 2019;179(1):219-235 e221.
Mansfield AS, Peikert T, Vasmatzis G. Chromosomal rearrangements and their neoantigenic potential in mesothelioma. Transl Lung Cancer Res. 2020;9(Suppl 1):S92-S99.
Hasanau T, Pisarev E, Kisil O, Nonoguchi N, Le Calvez-Kelm F, Zvereva M. Detection of TERT Promoter Mutations as a Prognostic Biomarker in Gliomas: Methodology, Prospects, and Advances. Biomedicines. 2022;10(3).
Yang L, Li A, Wang Y, Zhang Y. Intratumoral microbiota: roles in cancer initiation, development and therapeutic efficacy. Signal Transduct Target Ther. 2023;8(1):35.
Wong-Rolle A, Dong Q, Zhu Y, et al. Spatial meta-transcriptomics reveal associations of intratumor bacteria burden with lung cancer cells showing a distinct oncogenic signature. J Immunother Cancer. 2022;10(7).
Zhu G, Su H, Johnson CH, Khan SA, Kluger H, Lu L. Intratumour microbiome associated with the infiltration of cytotoxic CD8+ T cells and patient survival in cutaneous melanoma. Eur J Cancer. 2021;151:25-34.
Kalaora S, Nagler A, Nejman D, et al. Identification of bacteria-derived HLA-bound peptides in melanoma. Nature. 2021;592(7852):138-143.
Belkadi A, Bolze A, Itan Y, et al. Whole-genome sequencing is more powerful than whole-exome sequencing for detecting exome variants. Proc Natl Acad Sci U S A. 2015;112(17):5473-5478.
Trotman J, Armstrong R, Firth H, et al. The NHS England 100,000 Genomes Project: feasibility and utility of centralised genome sequencing for children with cancer. Br J Cancer. 2022;127(1):137-144.
Cornish AJ, Chubb D, Frangou A, et al. Reference bias in the Illumina Isaac aligner. Bioinformatics. 2020;36(17):4671-4672.
Kawaguchi S, Higasa K, Shimizu M, Yamada R, Matsuda F. HLA-HD: An accurate HLA typing algorithm for next-generation sequencing data. Hum Mutat. 2017;38(7):788-797.
Jennings CN, Humphries MP, Wood S, et al. Bridging the gap with the UK Genomics Pathology Imaging Collection. Nat Med. 2022;28(6):1107-1108.
Hendry S, Salgado R, Gevaert T, et al. Assessing Tumor-Infiltrating Lymphocytes in Solid Tumors: A Practical Review for Pathologists and Proposal for a Standardized Method from the International Immuno-Oncology Biomarkers Working Group: Part 2: TILs in Melanoma, Gastrointestinal Tract Carcinomas, Non-Small Cell Lung Carcinoma and Mesothelioma, Endometrial and Ovarian Carcinomas, Squamous Cell Carcinoma of the Head and Neck, Genitourinary Carcinomas, and Primary Brain Tumors. Adv Anat Pathol. 2017;24(6):311-335.
Hendry S, Salgado R, Gevaert T, et al. Assessing Tumor-infiltrating Lymphocytes in Solid Tumors: A Practical Review for Pathologists and Proposal for a Standardized Method From the International Immunooncology Biomarkers Working Group: Part 1: Assessing the Host Immune Response, TILs in Invasive Breast Carcinoma and Ductal Carcinoma In Situ, Metastatic Tumor Deposits and Areas for Further Research. Adv Anat Pathol. 2017;24(5):235-251.
Hainaut P, Pfeifer GP. Somatic TP53 Mutations in the Era of Genome Sequencing. Cold Spring Harb Perspect Med. 2016;6(11).
Feuerbach L, Sieverling L, Deeg KI, et al. TelomereHunter - in silico estimation of telomere content and composition from cancer genomes. BMC Bioinformatics. 2019;20(1):272.
Wood DE, Lu J, Langmead B. Improved metagenomic analysis with Kraken 2. Genome Biol. 2019;20(1):257.
Wood DE, Salzberg SL. Kraken: ultrafast metagenomic sequence classification using exact alignments. Genome Biol. 2014;15(3):R46.

Yes there is potential Competing Interest. J.L. has received research funding from Bristol-Myers Squibb, Merck, Novartis, Pfizer, Achilles Therapeutics, Roche, Nektar Therapeutics, Covance, Immunocore, Pharmacyclics, and Aveo, and served as a consultant to Achilles, AstraZeneca, Boston Biomedical, Bristol-Myers Squibb, Eisai, EUSA Pharma, GlaxoSmithKline, Ipsen, Imugene, Incyte, iOnctura, Kymab, Merck Serono, Nektar, Novartis, Pierre Fabre, Pfizer, Roche Genentech, Secarna, and Vitaccess. ST is funded by CRUK [grant number C50947/A18176), the NIHR]Biomedical Research Centre at the Royal Marsden Hospital and the Institute of Cancer Research [grant number A109], the Kidney and Melanoma Cancer Fund of the Royal Marsden Cancer Charity, the Rosetrees Trust [grant number A2204], Ventana Medical Systems [grant numbers 10467, 10530), the National Institutes of Health (US) and the Melanoma Research Alliance. ST has received speaker fees from Roche, Astra Zeneca, Novartis and Ipsen. ST has patents filed Indel mutations as a therapeutic target and predictive biomarker PCTGB2018/051892 and PCTGB2018/051893; Clear Cell Renal Cell Carcinoma Biomarkers P113326GB. C.S. acknowledges grants from AstraZeneca, Boehringer-Ingelheim, Bristol Myers Squibb, Pfizer, Roche-Ventana, Invitae (previously Archer Dx Inc—collaboration in minimal residual disease sequencing technologies), Ono Pharmaceutical, and Personalis. He is Chief Investigator for the AZ MeRmaiD 1 and 2 clinical trials and is the Steering Committee Chair. He is also Co-Chief Investigator of the NHS Galleri trial funded by GRAIL and a paid member of GRAIL’s Scientific Advisory Board. He receives consultant fees from Achilles Therapeutics (also SAB member), Bicycle Therapeutics (also a SAB member), Genentech, Medicxi, China Innovation Centre of Roche (CICoR) formerly Roche Innovation Centre—Shanghai, Metabomed (until July 2022), and the Sarah Cannon Research Institute C.S has received honoraria from Amgen, AstraZeneca, Bristol Myers Squibb, GlaxoSmithKline, Illumina, MSD, Novartis, Pfizer, and Roche-Ventana. C.S. has previously held stock options in Apogen Biotechnologies and GRAIL, and currently has stock options in Epic Bioscience, Bicycle Therapeutics, and has stock options and is co-founder of Achilles Therapeutics. C.S. also declares a patent application (PCT/US2017/028013) for methods to lung cancer; targeting neoantigens (PCT/EP2016/059401); identifying patent response to immune checkpoint blockade (PCT/EP2016/071471), determining HLA LOH (PCT/GB2018/052004); predicting survival rates of patients with cancer (PCT/GB2020/050221), identifying patients who respond to cancer treatment (PCT/GB2018/051912); methods for lung cancer detection (US20190106751A1). C.S. is an inventor on a European patent application (PCT/GB2017/053289) relating to assay technology to detect tumor recurrence. This patent has been licensed to a commercial entity and under their terms of employment C.S is due a revenue share of any revenue generated from such license(s). M.L. reports grants or contracts from BMS, Shionogi, and AstraZeneca; consulting fees from BioNTech, Bicycle Therapeutics, Janssen, Merck Sorano, Pfizer, and ADC Therapeutics; honoraria from AstraZeneca and Pfizer; and support for attending meetings or travel from MSD, Janssen, and Bayer. K.L. has a patent on indel burden and CPI response pending and speaker fees from Roche tissue diagnostics, research funding from CRUK TDL/Ono/LifeArc alliance, and a consulting role with Monopteros Therapeutics.

Extendeddatafigures.docx

Download PDF

Version 1

posted

You are reading this latest preprint version

Whole-genome sequencing uncovers the genomic determinants of therapeutic resistance to immune checkpoint blockade

Status:

Version 1

Abstract

Figures

Main

Results

Cohort demographic and genomic overview

Whole-exome derived biomarkers perform similarly in whole-genome data

Mutational signatures derived from WGS predict differential CPI response

Immune germline SNPs do not strongly predict CPI response

Structural variants facilitate genomic rearrangement with a minimal neoantigen footprint

Alterations in the MHC class I pathway associates with CPI resistance

Telomere lengthening reduces CPI efficacy in renal and bladder cancers

Increased intratumoral microbial presence is associated with reduced CPI response

A multivariable model improves the prediction of treatment response compared to TMB alone

Discussion

Methods

Patient selection

Processing of WGS data

Estimation of coding and whole genome mutational metrics

Image acquisition process summary

Histological assessment of tumour infiltrating lymphocytes (TILs)

Fitting of known mutational signatures

Investigating non-coding mutations in the class I antigen presentation pathway

Testing for association between germline TIME SNPs and CPI response

Testing for association between TP53 loss of function variants and chromothripsis

Telomere length estimation and telomere altering mutations

TERT promoter mutations

Metagenomic and microbiome estimation

Multivariable model construction

Declarations

References

Additional Declarations

Supplementary Files

Status:

Version 1