Whole genome sequencing of pre-treatment and post-treatment locally advanced rectal cancer using long and short read technologies Authors: Lauren McAuley, Lydia O’Sullivan, Liam Grogan, Adrian Murphy, Christina Skourou, Brendan Curran, Deborah McNamara, Joanna Fay, Brian O’Neill, Bryan T. Hennessy, Sinead Toomey, Simon J. Furney Published: 2026 ABSTRACT We performed a pilot study in which Illumina (ILMN) and nanopore (ONT) whole genome sequencing (WGS) was generated for samples obtained from a 37-year-old patient with locally advanced rectal cancer who did not respond to neoadjuvant chemoradiotherapy (nCRT). Pre-treatment and post-treatment tumour biopsy, adjacent normal tissue, and matched pre-treatment blood samples were sequenced to identify somatic alterations. We used DNA methylation from ONT WGS to detect changes in the tumour epigenomes compared to adjacent normal tissue. We employed established pipelines for the identification of somatic alterations to assess the concordance of SNV, short indel, copy number and structural variation from the short-read and long-read tumour genome data. Overall, 69.4% and 30.1% of SNVs were concordant between technologies in the pre-and post-treatment tumours, respectively. A multinomial logistic regression was performed to further understand discordant SNV calls, highlighting tumour purity, presence in a repetitive region, strand bias, and germline variant allele frequency as factors contributing to discordance. Indel concordance was poor (26.8% and 9.2% in the pre-and post-treatment tumour, respectively). Copy number alteration profiles were highly similar while minimal concordance was reported for structural variants (2.4% and 0.3% for pre-and post-treatment tumours, respectively). Finally, we used the ONT data to detect DNA modifications (5-methylcytosine and 5-hydroxymethylcytosine) between tumour and matched adjacent normal tissue. Based on genome-wide average CpG modification rates, global hypomethylation was observed in tumours compared to adjacent normal tissues. Biologically relevant epigenomic changes related to epithelial-mesenchymal transition were detected in the post-treatment tumour, as well as potential radiation-induced changes in the post-treatment adjacent normal tissue.Supplementary InformationThe online version contains supplementary material available at 10.1038/s41598-026-54265-x. Introduction Cancer is inherently a genetic disease. To capture the complexity of cancer, several large-scale genome sequencing initiatives such as the Pan-Cancer Analysis of Whole Genomes (PCAWG) Consortium, the Hartwig Medical Foundation, and the Cancer Programme of the 100,000 Genomes Project from Genomics England were established and generated large repositories of publicly available short read whole genome sequencing (WGS) data, including data from primary and metastatic tumours and matched normal tissues1–3. These highly valuable resources have been used extensively by researchers, making substantial contributions to the current understanding of cancer genomics and assisting in the development of new computational tools to analyse short-read genome sequencing data1–5. However, the findings of these datasets are restricted by inherent limitations of short reads, including the inability to accurately align to repetitive regions and limited sensitivity when detecting structural variants6. In recent years, long-read sequencing technologies like nanopore sequencing (Oxford Nanopore Technologies; ONT) and HiFi sequencing (Pacific Biosciences; PacBio) have gained traction; long-read technologies typically produce reads over 10 kb in length which enables the resolution of large repetitive regions and complex structural variants, effectively addressing the disadvantages of short-read sequencing. An additional benefit of long read sequencing is the ability to directly obtain native epigenetic modifications from both DNA and RNA sequencing data without additional library preparation steps. While several techniques are available to detect DNA methylation from short reads, these involve harsh bisulfite treatments and/or enzymatic conversion steps7–9. Moreover, long read RNA sequencing can sequence full length transcripts and detect RNA modifications such as m6A10. As the accuracy of long-read technologies approach that of Illumina (ILMN) near-perfect reads and sequencing costs fall, the application of long-read sequencing to cancer genomics is becoming more viable. The utility of long-reads to improve tumour genome, and epigenome, characterisation has been demonstrated across multiple cancer types, including osteosarcoma and paediatric medulloblastoma11–15. The Genomics England Cancer 2.0 and All of Us initiatives have launched pilot projects using ONT and PacBio sequencing, respectively, to assess how these technologies can improve the detection of small and large genomic variants and the feasibility of applying them in clinical and research settings16. Long read sequencing technologies have great potential; however, whether somatic variants detected by standard short-read sequencing and analyses protocols can be detected in patient samples using long-read sequencing data and currently available long-read sequencing tools must be further assessed. Here, we present a pilot study that focused on comparing the abilities of these sequencing technologies to detect somatic variants, including driver events, in patient tumour genomes using available bioinformatic tools. To this end, ILMN and ONT WGS was performed on samples obtained from a patient with locally advanced rectal cancer (LARC) who did not respond to therapy. Pre-treatment and post-treatment tumour biopsies, pre-treatment and post-treatment adjacent normal tissues, and matched pre-treatment blood underwent WGS with both ILMN and ONT. Since adjacent normal tissue was available, DNA methylation from ONT WGS was also examined to detect changes in the epigenetic profile of the tumour and adjacent normal tissue. Results Patient characteristics The patient was a 37-year-old male diagnosed with LARC. The patient underwent neoadjuvant chemoradiotherapy (nCRT), completing 6 cycles of 5-fluorouracil and receiving 50.4 Gy radiation over 28 fractions. The patient underwent surgical resection of the tumour 8 weeks after completing nCRT. Response to nCRT was poor with a Mandard tumour regression grade of 4. Immunohistochemistry of mismatch repair proteins was performed with no loss of expression of PMS-2, MLH-1, MSH-2, or MSH-6, indicating that the tumour was microsatellite stable (MSS). Time to distal recurrence after beginning nCRT was 9 months. Metastases were observed in the liver and omentum. Time to local recurrence was 18 months after beginning nCRT. Overall survival was 23 months. Whole genome sequencing of tumour and adjacent normal tissues WGS using short-read (ILMN) and long-read (ONT) sequencing technologies was performed on a total of 5 samples collected from one LARC patient, including pre-treatment and post-treatment primary tumour samples, pre-treatment and post-treatment adjacent normal tissues, and matched pre-treatment blood (Fig. 1A). Average coverage across all samples for ONT sequencing was 37X. For ILMN sequencing, average coverage across all samples was 45X (Supplementary Table 1). Whole genome sequencing data of tumour and adjacent normal tissues were compared to the matched blood sample to detect somatic single nucleotide variants (SNVs) and insertions and deletions (indels), somatic copy number alterations (sCNAs), and somatic structural variants (SVs). As adjacent normal tissues were available, changes in DNA methylation patterns between tumours and adjacent normal tissues were also investigated. Fig. 1(A) Schematic of the study design. (B) Concordance of SNVs (top) and indels (bottom) between Mutect and ClairS for pre- and post-treatment tumours. (C) SBS (top) and indel (bottom) signatures for each set in the pre- and post-treatment tumours. (D) Counts of SNVs by genomic annotations per SNV set. (E) Counts of SNVs by repetitive element annotation per SNV sets. (F) Proportions of SNVs that were validated in orthogonal sequencing technologies. (G) Multinomial regression analysis of factors contributing to SNV discordance. (H) Size distribution of indels identified by Mutect and ClairS. (A) Schematic of the study design. (B) Concordance of SNVs (top) and indels (bottom) between Mutect and ClairS for pre- and post-treatment tumours. (C) SBS (top) and indel (bottom) signatures for each set in the pre- and post-treatment tumours. (D) Counts of SNVs by genomic annotations per SNV set. (E) Counts of SNVs by repetitive element annotation per SNV sets. (F) Proportions of SNVs that were validated in orthogonal sequencing technologies. (G) Multinomial regression analysis of factors contributing to SNV discordance. (H) Size distribution of indels identified by Mutect and ClairS. Concordance of small somatic variants between callers SNVs were detected using Mutect2 in the ILMN data and ClairS in the ONT data (Table 1). The concordance of the SNVs detected between callers was examined for each sample (Fig. 1B), with 69.4% agreement in the pre-treatment tumour and 30.1% agreement in the post-treatment tumour. Few SNVs were concordant between callers in adjacent normal tissue samples (1.1% and 0.3% agreement in the pre-treatment and post-treatment adjacent normal tissues, respectively) (Supplementary Fig. 1A). The concordance of SNVs detected by a second short read small variant caller, Strelka2, was also assessed, with small differences in concordance observed with ClairS compared to Mutect2 (Supplementary Fig. 1B). Mutect2 was chosen for subsequent analyses due to its better performance at variant frequencies below 10%17. Sets of SNVs were defined as follows; all Mutect SNVs, all ClairS SNVs, SNVs detected by both Mutect and ClairS are termed ‘Mutect/ClairS intersection’ SNVs, while SNVs detected only in the ONT data or the ILMN data are referred to as ‘private to ClairS’ or ‘private to Mutect’, respectively. Table 1Counts of small somatic variants detected by Mutect2 and ClairS.VariantCallerTumourAdjacent normalPre-treatmentPost-treatmentPre-treatmentPost-treatmentSNVMutect801451651018450ClairS72984171600473IndelMutect672350124117ClairS714532378606 Table 1Counts of small somatic variants detected by Mutect2 and ClairS.VariantCallerTumourAdjacent normalPre-treatmentPost-treatmentPre-treatmentPost-treatmentSNVMutect801451651018450ClairS72984171600473IndelMutect672350124117ClairS714532378606 Counts of small somatic variants detected by Mutect2 and ClairS. Mutational signature analysis of single base substitutions (SBS) was performed on the SNV sets. In both pre- and post-treatment tumours, SBS5 (clock-like) and SBS18 (damage by reactive oxygen species) were consistently detected across all Mutect, all ClairS and Mutect/ClairS intersection sets (Fig. 1C). SBS signatures detected in the set private to ClairS were largely comparable to the SBS signatures detected for all ClairS SNVs. A similar pattern was observed for the private to Mutect SNVs, however SBS5 was not detected in this set in the pre-treatment tumour. High cosine similarity was observed between SNV sets private to a given caller and their Mutect/ClairS intersection counterparts, indicating only slight differences in the underlying trinucleotide contexts of SNVs private to callers (Supplementary Fig. 2A). The reasons behind SNV discordance between ILMN and ONT sequencing technologies in this patient’s tumour genomes were investigated using a total of 16,216 SNVs detected in tumour samples; 8,432 SNVs were in the Mutect/ClairS intersection set, 4,747 SNVs were private to Mutect and 3,037 SNVs were private to ClairS. To understand if genomic location affected SNV discordance, SNV sets were annotated with genomic regions as follows; 1-5 kb upstream of a gene, exonic, intronic, intergenic, and promoter regions. A higher proportion of SNVs private to ClairS were in intergenic regions compared to other sets (1,508/3,037 (49.7%) private to ClairS, 2,113/4,747 (44.5%) private to Mutect, and 3,535/8,432 (41.9%) in the Mutect/ClairS intersection) (Fig. 1D). As repetitive elements are known to pose difficulties for variant detection, SNV sets were annotated with various repetitive elements to assess differences in the abilities of the sequencing technologies to profile these regions in this patient’s tumour genomes (Fig. 1E). SNVs in satellite regions were more prevalent in the set private to ClairS (503/3,037 (16.6%) in private to ClairS set vs. 232/4,747 (4.9%) private to Mutect set vs. 58/8,432 (0.7%) in the Mutect/ClairS intersection set). These patterns were consistent in each tumour sample (Supplementary Fig. 3A, B). A multinomial logistic regression was performed to further understand the factors contributing to discordant SNV calls in this patient’s tumour samples. Evidence of SNVs that were private to a caller (i.e. private to a sequencing technology) was extracted from the orthogonal data type where a call was not made (termed validation calls). Where successfully force-called, total read depths, variant allele frequencies (VAFs), and strand bias information was obtained for a given SNV in both ONT and ILMN data. This data was available for 15,605/16,216 SNVs; 8,432 SNVs were in the Mutect/ClairS intersection, 4,574 SNVs private to Mutect were validated in ONT data, and 2,599 SNVs private to ClairS were validated in ILMN data (Fig. 1F, Supplementary Table 2). This variant information, along with sample and repetitive region annotations, was used as input for the regression model. Several factors contributed to SNV discordance between callers/sequencing technologies in this patient’s tumour genomes (Fig. 1G). Tumour purity estimates from short and long-read data differed (Supplementary Table 3), however the post-treatment tumour was consistently reported to have lower purity compared to the pre-treatment tumour. Thus, sample was used as a surrogate for tumour purity and was a contributor to discordance for both private SNV sets (SNVs private to Mutect relative risk (RR) = 1.57, p < 0.001; SNVs private to ClairS RR = 1.58, p < 0.001), which was in line with the disparity in concordance initially observed. Similar proportions of SNVs were in repetitive regions across the SNV sets (1,542/2,599 (59.3%) private to ClairS, 2,527/4,574 (55.2%) private to Mutect, and 4,348/8,432 (51.6%) in Mutect/ClairS intersection). Whether an SNV was in a repetitive region contributed to discordance, with SNVs private to ClairS 40% more likely to be discordant for this reason (RR = 1.40, p < 0.001). This was not a significant factor for discordance in the SNVs private to Mutect (RR = 1.18, p = 0.10). Additionally, strand bias in the ONT validation data of SNVs private to Mutect contributed to discordance of these SNVs (ONT validation of private Mutect SNVs; mean strand bias = 0.75, RR = 3.27, p < 0.001, private Mutect SNVs; mean strand bias = 0.57, RR = 1.40, p < 0.001). Germline support for alternate alleles also contributed to SNV discordance; ClairS calls in the ONT data were 61% more likely to be discordant based on VAF in the germline sample (RR = 1.61, p < 0.001), with validation of these calls in ILMN data showing an increased likelihood of discordance when considering VAF in the germline sample (RR = 2.99, p < 0.001). Similarly, validation of private to Mutect SNVs in the ONT data showed an increased likelihood of discordance due to VAF in the germline sample (RR = 2.43, p < 0.001). Indels Concordance of indels between Mutect and ClairS was 26.8% (293/1,093) in the pre-treatment tumour and 9.2% (74/808) in the post-treatment tumour of this patient (Fig. 1B). Concordance was minimal between callers in the patient’s adjacent normal tissues (Supplementary Fig. 4A). Interestingly, most concordant indels in tumours were of the same length and position (Fig. 1H). Of all indels detected in the pre-treatment tumour, 428/672 (63.7%) and 451/714 (63.2%) detected by Mutect and ClairS, respectively, overlapped (minimum 1 bp) or were within 5 bp of a homopolymer. In the post-treatment tumour (195/350 (55.7%) of all Mutect indels and 368/532 (69.2%) of all ClairS indels overlapped or were within 5 bp of a homopolymer) (Supplementary Table 4). Of indels concordant between callers, 146/293 (49.8%) and 31/74 (41.9%) overlapped or were adjacent to homopolymers in the pre- and post-treatment tumours (Supplementary Fig. 4B). However, for each tumour and caller, 67–79% of discordant indels overlapped or were adjacent to homopolymers, indicating that homopolymers contributed to the high rate of discordance observed (Supplementary Fig. 4B). Indel concordance between callers in the pre- and post-treatment tumours increased to 40.8% and 15.6%, respectively, when indels overlapping/adjacent to homopolymers were excluded (Supplementary Fig. 4C). Similarly to SNV sets, indel sets were defined. Indel signatures were extracted using all indels detected as input regardless of homopolymer status (Fig. 1C). ID18 (colibactin exposure) and ID14 (aetiology unknown) were consistently detected by ClairS across tumours, while ID1 (slippage during DNA replication) and ID23 (aristolochic acid exposure) were consistently detected by Mutect. Mutect/ClairS intersection signatures more closely resembled signatures extracted from Mutect indels as ID1 and ID23 were detected in this indel set. Moreover, indel sets private to ClairS demonstrated low cosine similarity with Mutect/ClairS intersection and private to Mutect sets (Supplementary Fig. 2B). This indicates different nucleotide contexts underlying indels private to ClairS and is reflected in the different indel signatures detected (Fig. 1C). Concordance of small somatic variants between tumours Sequencing of this patient’s tumours with short and long read technologies enabled a comparison between technologies to detect tumour evolution. Concordance of SNVs between the patient’s tumours was similar between technologies with 43.5% and 41.1% of SNVs detected in both the pre- and post-treatment tumours by Mutect and ClairS, respectively (Fig. 2A). Of all 10,666 SNVs detected across the patient’s tumour samples by Mutect and/or ClairS, 1,932 (18.1%) were common to pre- and post-treatment tumours sequenced by ILMN and ONT (Fig. 2B). A total of 2,920 (27.4%) SNVs were detected by both sequencing technologies in a given tumour but detected by a single sequencing technology in the other tumour, i.e. detected in 3 of 4 samples (Fig. 2B). In some of these cases, low quality evidence from an orthogonal sequencing technology supported these SNVs in the remaining tumour; including these SNVs increased concordance between pre- and post-treatment tumours to 48.3% and 48.6% for Mutect and ClairS respectively, with an increase in overall concordance to 28.0% (2,988/10,666) (Supplementary Fig. 5). Fig. 2(A) Concordance of SNVs detected by Mutect and ClairS in pre-treatment and post-treatment tumours. (B) UpSet plot representing concordance of SNVs detected by Mutect and ClairS across pre- and post-treatment tumours. (C) Concordance of indels detected by Mutect and ClairS in pre-treatment and post-treatment tumours. (D) UpSet plot representing concordance of indels detected by Mutect and ClairS across pre- and post-treatment tumours, with indels overlapping or adjacent to homopolymers (HP) in red and indels not overlapping/adjacent to homopolymers (NonHP) in blue. (A) Concordance of SNVs detected by Mutect and ClairS in pre-treatment and post-treatment tumours. (B) UpSet plot representing concordance of SNVs detected by Mutect and ClairS across pre- and post-treatment tumours. (C) Concordance of indels detected by Mutect and ClairS in pre-treatment and post-treatment tumours. (D) UpSet plot representing concordance of indels detected by Mutect and ClairS across pre- and post-treatment tumours, with indels overlapping or adjacent to homopolymers (HP) in red and indels not overlapping/adjacent to homopolymers (NonHP) in blue. Indel concordance between the patient’s tumours was measured across all indels and when excluding indels overlapping or adjacent to homopolymers. Concordance of indels between tumours was higher using Mutect compared to ClairS, with 31.7% and 10.9% indel concordance between pre- and post-treatment tumours, respectively (Fig. 2C). Of all indels detected across the pre- and post-treatment tumours, 59/1,585 (3.7%) indels were detected in both the pre- and post-treatment tumours by both Mutect and ClairS (32/483 (6.6%) when excluding indels overlapping/adjacent to homopolymers) (Fig. 2D, Supplementary Fig. 4B). A further 144/1,585 (9.1%) of indels were detected in the pre- and post-treatment tumours but were not detected by one orthogonal technology in one tumour; 84/483 (17.4%) were concordant when excluding indels overlapping/adjacent to homopolymers (Fig. 2D, Supplementary Fig. 4B). A total of 105/1,585 indels (6.6%) were detected by both Mutect and ClairS in only the pre-treatment tumour (42/483 (8.7%) when excluding indels overlapping/adjacent to homopolymers), while no indels private to the post-treatment tumour were validated by an orthogonal sequencing technology (Fig. 2D, Supplementary Fig. 4B). Concordance of small driver mutations Small driver mutations were defined as any SNV or indel predicted or predicted and annotated as a driver by Cancer Genome Interpreter (CGI) with at least a one-star classification of oncogenicity on ClinVar. No small driver mutations were detected in adjacent normal tissues, and only SNV drivers were detected in tumours (Supplementary Table 5). Driver mutations in KRAS (p.G12D), PIK3CA (p.E542K) and SOCS1 (p.Q131*) were consistently detected by short and long reads in the pre-treatment tumour (Table 2). The KRAS variant was also successfully called in the short and long read post-treatment tumour. However, detection of PIK3CA and SOCS1 drivers in the post-treatment tumour was inconsistent. Though present at a VAF of 20%, the PIK3CA variant was not successfully called in the long read post-treatment tumour as it was also present at 2.9% VAF in the long-read germline sample. The variant was only supported by a single read in ONT WGS of the germline sample and was not detected in the short-read germline sample. Moreover, this germline VAF was only reported in the post-treatment tumour iteration of ClairS and not the pre-treatment tumour iteration – both of which used the same germline input. This highlights a discrepancy in the ClairS somatic calls at this position between pre- and post-treatment tumour iterations. The SOCS1 variant was not initially reported in the post-treatment tumour sequenced by short reads. However, force-calling with Mutect at that position detected evidence of the variant (VAF 4.2%) but FilterMutectCalls deemed this variant a likely sequencing error. Table 2Driver SNVs detected by short and long read sequencing.DRIVERTimepointVAF in germline (%)VAF in tumour (%)MutectClairSMutectClairS PIK3CA (p.E454K)Pre-treatment0.00.012.526.5Post-treatment0.02.921.4 20.0 KRAS (p.G12D)Pre-treatment0.00.046.927.3Post-treatment0.00.015.813.3 SOCS1 (p.Q131*)Pre-treatment0.00.019.528.9Post-treatment0.00.0 4.2 12.8Bold lettering indicates that the mutation did not pass quality thresholds in the algorithm indicated. Table 2Driver SNVs detected by short and long read sequencing.DRIVERTimepointVAF in germline (%)VAF in tumour (%)MutectClairSMutectClairS PIK3CA (p.E454K)Pre-treatment0.00.012.526.5Post-treatment0.02.921.4 20.0 KRAS (p.G12D)Pre-treatment0.00.046.927.3Post-treatment0.00.015.813.3 SOCS1 (p.Q131*)Pre-treatment0.00.019.528.9Post-treatment0.00.0 4.2 12.8Bold lettering indicates that the mutation did not pass quality thresholds in the algorithm indicated. Driver SNVs detected by short and long read sequencing. PIK3CA (p.E454K) KRAS (p.G12D) SOCS1 (p.Q131*) Bold lettering indicates that the mutation did not pass quality thresholds in the algorithm indicated. Somatic copy number alterations Somatic copy number alterations (sCNAs) were detected using FACETS on the ILMN data and Wakhan on the ONT data. These tools also estimated ploidy and purity for each sample (Supplementary Table 3). The genomes of pre-treatment adjacent normal tissues were observed to be diploid by short and long reads. The post-treatment adjacent normal genome was reported as diploid by short reads, with a ploidy of 3.84 reported by Wakhan potentially due to the lower coverage of this sample and poor phasing (Supplementary Fig. 6). Ploidy estimates between orthogonal sequencing data were highly similar for the pre-treatment tumour with estimates of 2.56 and 2.53 for ILMN and ONT, respectively, with similar ploidy estimates also observed in the post-treatment tumour (2.45 and 2.49 for ILMN and ONT, respectively). Additionally, purity estimates based on ONT data were notably higher compared to the ILMN data (0.63 and 0.54 for ONT vs. 0.4 and 0.21 for ILMN for the pre- and post-treatment tumours, respectively). Fig. 3(A) FACETS and Wakhan copy number profiles for the pre- and post-treatment tumours. (B) Copy number signatures detected in the pre- and post-treatment tumours using segmentation profiles from FACETS and Wakhan. (C) Mutational profile of copy number signatures detected in tumours by FACETS and Wakhan. (D) Copy number status of driver genes across tumours. (A) FACETS and Wakhan copy number profiles for the pre- and post-treatment tumours. (B) Copy number signatures detected in the pre- and post-treatment tumours using segmentation profiles from FACETS and Wakhan. (C) Mutational profile of copy number signatures detected in tumours by FACETS and Wakhan. (D) Copy number status of driver genes across tumours. In this patient’s tumours, similar sCNA patterns were evident between ILMN and ONT WGS data (Fig. 3A). Hyperploidy of multiple chromosomes (1q, 2, 6, 7, 8, 12, 13, 14, 16, 19, 20) was consistently reported across technologies across this patient’s tumours. A genome is classified as having undergone whole genome duplication (WGD) if over 50% of the autosomal genome has a major copy number of at least 218. Using this metric, it was determined that the pre-treatment tumour may have undergone a WGD event. Copy number signature analysis also supported a past WGD event; CN1 (diploid) and CN2 (tetraploid) signatures were detected in the FACETS and Wakhan segmentation data of pre- and post-treatment tumours (Fig. 3B). Contributions of CN2 were lower in Wakhan and CN23 (an oversegmentation artefact) was detected (Fig. 3B). CN23 is a result of segments < 100kb and is typically observed in CN data generated by SNP6 arrays. Segment profiles derived from Wakhan displayed an increased number of segments < 100kb compared to the FACETS segments profiles, though whether this is a sequencing artifact or genuine CN difference remains unclear (Fig. 3C). Consistent with histopathological findings of microsatellite stability in this patient’s tumour, the CN signature linked to deficient mismatch repair (CN25) was not detected in short or long read WGS data. Copy number changes were defined based on the occurrence of WGD as described previously19. The timing of sCNAs can be deduced from the total and minor copy numbers. FACETS and Wakhan detected loss of heterozygosity (LOH) of the entirety of chromosome 17 and of a 41.56Mb region of chromosome 3 (chr3p26.3-p22.1) in the pre-treatment tumour, which occurred prior to WGD. Additionally, a loss occurred in chromosome 1q along with a gain in chromosome 13 occurred in the pre-treatment tumour after WGD. Subsequent losses of single copies of chromosomes 2, 6–8, 12, 14, 16, 19, 20 and double losses in chromosomes 1p, 3–5, 9–11, 15, 17, 18, 21 and 22 were observed. Copy number states in the post-treatment tumour remained predominately stable compared to the pre-treatment tumour, however the majority of chromosome 12 experienced a loss with the exception of a 15.64Mb region containing KRAS. Additionally, single and double losses of segments of chromosome 13 are observed in the post-treatment tumour sequenced by ILMN; however, this pattern is not evident in the ONT data. Copy number alterations were observed in several driver genes previously reported to experience copy number changes in primary MSS stable colorectal cancer (CRC) and in driver genes in the Cancer Gene Census (Fig. 3D)19,20. Loss of heterozygosity in CDK12, CTNNB1, ERBB2, TP53, and USP36 was consistently detected by short and long read technologies, while gains were less consistently identified. Homozygous deletions of BCL9, PDE4DIP, and TPTE were observed in long reads but not in short reads. Structural variants Somatic SVs were detected using Manta and Severus in ILMN and ONT data, respectively. An increased number of smaller SVs (< 10kb) was detected by Severus compared to Manta across all samples (Fig. 4A, Supplementary Fig. 7A). Additionally, Severus detected a markedly greater number of insertions, and this observation was particularly pronounced in tumour samples (Fig. 4A, Supplementary Fig. 7A). In this patient, we note that the proportion of classes of SVs detected in ILMN and ONT data differed; while just a single insertion was detected across both ILMN tumour samples, this SV class was the most prominent class detected in ONT samples (Table 3). The majority of SVs detected by Manta across ILMN tumour samples were breakends. Fig. 4(A) Distribution of SV sizes detected across all samples by Manta and Severus. (B) Concordance of SVs between Manta and Severus for the pre- and post-treatment tumours. (C) Concordance of SVs across all tumours and SV callers. (D) Circos plots of SVs detected in the pre- and post-treatment tumours by Manta and Severus. (E) Relative contributions of SV signatures differed between tumour samples and SV callers. (F) Mutational profile of SV signatures. (A) Distribution of SV sizes detected across all samples by Manta and Severus. (B) Concordance of SVs between Manta and Severus for the pre- and post-treatment tumours. (C) Concordance of SVs across all tumours and SV callers. (D) Circos plots of SVs detected in the pre- and post-treatment tumours by Manta and Severus. (E) Relative contributions of SV signatures differed between tumour samples and SV callers. (F) Mutational profile of SV signatures. Table 3Counts of structural variants detected by Manta and Severus.CallerSV classTumourAdjacent normalPre-treatmentPost-treatmentPre-treatmentPost-treatmentMantaTotal438 (100)95 (100)324 (100)129 (100)INS1 (0.2)0 (0)0 (0)0 (0)DEL13 (3.0)2 (2.1)1 (0.3)4 (3.1)BND412 (94.1)86 (90.5)316 (97.5)112 (86.8)INV10 (2.3)3 (3.2)4 (1.2)7 (5.4)DUP2 (0.5)4 (4.2)3 (0.9)6 (4.7)SeverusTotal368 (100)247 (100)35 (100)9 (100)INS313 (85.1)199 (80.6)14 (40.0)4 (44.4)DEL27 (7.3)27 (10.9)9 (25.7)2 (22.2)BND16 (4.3)7 (2.8)5 (14.3)0 (0)INV10 (2.7)2 (0.8)1 (2.9)1 (11.1)DUP2 (0.5)3 (1.2)0 (0.0)1 (11.1)INS, insertion; DEL, deletion; BND, breakend; INV, inversion; DUP, duplication. Table 3Counts of structural variants detected by Manta and Severus.CallerSV classTumourAdjacent normalPre-treatmentPost-treatmentPre-treatmentPost-treatmentMantaTotal438 (100)95 (100)324 (100)129 (100)INS1 (0.2)0 (0)0 (0)0 (0)DEL13 (3.0)2 (2.1)1 (0.3)4 (3.1)BND412 (94.1)86 (90.5)316 (97.5)112 (86.8)INV10 (2.3)3 (3.2)4 (1.2)7 (5.4)DUP2 (0.5)4 (4.2)3 (0.9)6 (4.7)SeverusTotal368 (100)247 (100)35 (100)9 (100)INS313 (85.1)199 (80.6)14 (40.0)4 (44.4)DEL27 (7.3)27 (10.9)9 (25.7)2 (22.2)BND16 (4.3)7 (2.8)5 (14.3)0 (0)INV10 (2.7)2 (0.8)1 (2.9)1 (11.1)DUP2 (0.5)3 (1.2)0 (0.0)1 (11.1)INS, insertion; DEL, deletion; BND, breakend; INV, inversion; DUP, duplication. Counts of structural variants detected by Manta and Severus. INS, insertion; DEL, deletion; BND, breakend; INV, inversion; DUP, duplication. An SV benchmarking tool, Minda, was used to compare SVs between Manta and Severus. When comparing two SVs, Minda will declare them as the same variant if both the start and end positions of each variant are within a given window between samples21. SV concordance rates between technologies were low; only 19 (2.4%) and 1 (0.3%) SVs detected by Manta and Severus in the pre-treatment and post-treatment tumours, respectively, were concordant (Fig. 4B). A single SV was concordant between Manta and Severus in both the pre- and post-treatment tumours; a 50,484bp deletion in CDH1 (Fig. 4C). Between pre- and post-treatment tumours, the concordance of SVs detected by Manta was low (27/506, 5.3%), with these condordant SVs predominately consisting of breakends (24/27), along with 2 inversions and 1 deletion. Higher concordance was observed within Severus, with 168/447 (37.6%) of SVs called in both the pre- and post-treatment tumours, consisting of 161 insertions and 7 deletions. It was hypothesised that SV concordance was likely higher upon visual examination of circos plots (Fig. 4D). As an example, Manta detected a high density of translocations between chromosome 13 and chromosome 7 in the pre-treatment tumour. In the same region of chromosome 13, Severus detected a similar hotspot of insertions in the pre- and post-treatment tumours. Comparison of genomic coordinates found that for 7 of 24 insertions detected by Severus in chromosome 13 of the pre-treatment tumour, Manta detected breakends within 500bp of the insertion start position (4/7 had breakends within 10bp of the insertion start position) (Supplementary Table 6). The mate pairs of 6/7 of these Manta SVs mapped to a 541bp region in an intron of PKD1L1 on chromosome 7. Insertion sequences reported by Severus in chromosome 13 had high sequence similarity to PKD1L1 (Supplementary Table 7). Thus, the insertions in chromosome 13 reported by Severus are likely concordant with the translocations between chromosome 7 and chromosome 13 reported by Manta. Minda did not deem interchromosomal translocations detected by Manta as concordant with insertions reported by Severus because the second breakend detected by Manta was in a different chromosome to the insertion, i.e. the end positions were not in agreement. Determining the concordance of additional Severus insertions with Manta translocations would require a similar manual comparison of genomic coordinates and sequence similarities for all insertion sequences and breakends as outlined above, which is laborious and time consuming. Using ShatterSeek, chromothripsis was not detected between chromosomes 7 and 13 in either short or long read data. Notably, the single deletion that was consistently detected between short and long read technologies in both pre- and post-treatment tumours was a 50,484bp deletion in CDH1 (Table 4). This deletion encompasses approximately 4.2kb upstream of exon 1, the entirety of exons 1 and 2, and extends into intron 2, disrupting the structure of this tumour suppressor gene. The only insertion detected in short reads was a 136bp insertion in intron 2 of CTNBB1 in the pre-treatment tumour which was predicted to result in the loss of the acceptor site by SpliceAI (probability 0.89). This insertion was also detected by long reads in the pre- and post-treatment tumour (Table 4). Additionally, LOH of CTNNB1 was detected by long and short reads (Fig. 3D). Table 4Driver SVs detected by short and/or long read sequencing.GeneStart positionSV typeSV lengthTimepointTechnologyILMNONT CTNNB1 chr3:41224520INS136bpPre-treatmentYESYESPost-treatmentNOYES CDH1 chr16:68733044DEL50,484bpPre-treatmentYESYESPost-treatmentYESYES Table 4Driver SVs detected by short and/or long read sequencing.GeneStart positionSV typeSV lengthTimepointTechnologyILMNONT CTNNB1 chr3:41224520INS136bpPre-treatmentYESYESPost-treatmentNOYES CDH1 chr16:68733044DEL50,484bpPre-treatmentYESYESPost-treatmentYESYES Driver SVs detected by short and/or long read sequencing. SV signature analysis was performed. However, insertions are not considered by SigProfiler, along with any deletions, inversions, or duplications smaller than 1kb, resulting in the exclusion of over 80% of SVs detected in tumour samples by Severus. Despite this, some similarity in signature profiles was observed between Manta and Severus with SV2, which is characterised by a high proportion of non-clustered translocations, detected for both callers (Fig. 4E, F). For Manta, SV4 was detected in the pre-treatment tumour which is attributed to a high proportion of clustered complex translocations. SV signatures detected for Severus were more varied, potentially due to the low amounts of variants considered; SV5 and SV8 were detected in the pre-treatment tumour which are attributed to short (< 10kb) deletions and inversions, respectively. DNA modifications DNA modifications (5-methylcytosine; 5mC and 5-hydroxymethylcytosine; 5hmC) were detected using ONT sequencing. As only single replicates were available, a statistical approach accounting for this was utilised22. Average rates of DNA modifications were examined across various genomic regions (entire genes, exons, introns, promoters, enhancers, CpG islands, and repetitive regions). Based on genome-wide average CpG modification rates, global hypomethylation was observed in tumours compared to adjacent normal tissues with average global 5mC rates of 70.4% and 72.0% in the pre- and post-treatment tumours and 74.7% and 74.2% in the pre- and post-treatment adjacent normal tissues, respectively (Fig. 5A). On average, all genomic features were hypomethylated in tumours compared to matched adjacent normal tissues except in CpG islands which were, on average, hypermethylated in tumours (22.0% and 18.5% in pre- and post-treatment tumours, respectively, vs. 19.0% and 18.0% in pre- and post-treatment adjacent normal tissues, respectively). Additionally, average global hydroxymethylation rates were similar across samples (2.2%, 2.8%, 3.0% in pre-treatment tumour, pre-treatment adjacent normal and post-treatment adjacent normal, respectively), with the exception of an increased average rate of hydroxymethylation in the post-treatment tumour (4.5%). Higher average rates of hydroxymethylation in the post-treatment tumour were consistent across all genomic regions (Fig. 5A). In post-treatment adjacent normal tissue, the genome-wide methylation rate was lower compared to pre-treatment adjacent normal tissue. This pattern was largely consistent across various genomic features, along with increased average 5hmC rates compared to pre-treatment adjacent normal tissue. Fig. 5(A) Average methylation and hydroxymethylation rates genome-wide and across genomic features. (B) Counts of DMRs and DhMRs detected between samples. (C) Counts of differentially modified regions affecting genomic features across sample comparisons. (D) Functional genomic enrichment analysis of D(h)MRs detected across sample comparisons. (A) Average methylation and hydroxymethylation rates genome-wide and across genomic features. (B) Counts of DMRs and DhMRs detected between samples. (C) Counts of differentially modified regions affecting genomic features across sample comparisons. (D) Functional genomic enrichment analysis of D(h)MRs detected across sample comparisons. Differentially methylated regions (DMRs) and differentially hydroxymethylated regions (DhMRs) were detected between pre- and post-treatment adjacent normal tissues, between tumours and adjacent normal tissues, and between pre- and post-treatment tumours. The pre-treatment adjacent normal sample was considered as baseline because the post-treatment adjacent normal had been exposed to chemoradiotherapy. Fewer D(h)MRs were detected between adjacent normal tissues relative to the amount of DMRs detected between adjacent normal tissues and matched tumour samples and between pre- and post-treatment tumours, suggesting more extensive epigenetic remodelling in tumour tissues (Fig. 5B). To understand the genomic landscape of DNA modifications, D(h)MRs were annotated with genomic regions. The vast majority of D(h)MRs included multiple genomic features, with a greater number of genomic features of all classes affected by D(h)MRs in tumours. This was in line with the observation of more widespread epigenetic changes in tumours compared to adjacent normal tissue (Supplementary Fig. 8, Supplementary Table 8). D(h)MRs predominately affected introns, enhancers, and repetitive genomic elements, followed by exons (Fig. 5C). DhMRs were predominately detected in the post-treatment tumour (Fig. 5B), with the largest proportion of DhMRs affecting Alu elements (Supplementary Fig. 8B). Epigenetic modification of 61 driver genes was assessed. The list of drivers was chosen based on the COSMIC cancer gene census and genes with potential driver mutations identified in the tumours in this study were also added. Of these driver genes, 46 were affected by D(h)MRs in at least one comparison, with D(h)MRs predominantly encompassing enhancers linked to these genes and fewer promoters and gene bodies affected (Supplementary Fig. 8C). While gene expression data was not available in this study, a functional enrichment of genomic regions was performed using the Genomic Regions Enrichment of Annotations Tool (GREAT) to associate DMRs and DhMRs with biological functions23. GREAT analysis of DMRs revealed an enrichment of several cancer-related pathways in pre- and post-treatment tumours, and in post-treatment adjacent normal tissue (Fig. 5D). DMRs in the post-treatment adjacent normal tissue were most significantly enriched near genomic regions associated with the P53 pathway. Enrichment of this pathway was also observed in the post-treatment tumour compared to adjacent normal tissues, but not in the pre-treatment tumour. Compared to timepoint-matched adjacent normal tissues, DMRs in the pre-treatment and post-treatment tumours were most significantly enriched near genomic regions involved in the early estrogen response. D(h)MRs near genomic regions linked to TGFβ signalling were significantly enriched in all comparisons, though it was the most significantly enriched pathway identified for DMRs and DhMRs detected between the pre-treatment adjacent normal tissue and post-treatment tumour. Between pre- and post-treatment tumours, the DMRs and DhMRs detected were most significantly enriched near regions associated with genomic regions linked to the Wnt β-catenin and EMT pathways, respectively. Discussion In this pilot study, we compared somatic variants detected in short- and long-read WGS data derived from pre- and post-treatment tumour samples of a LARC patient. Standard tools were utlised for short-read analyses, with recently available somatic variant calling algorithms employed for long-read analyses. A limitation of this study is the assessment of one variant calling tool per mutation type. Moreover, tumour purity is an important factor when detecting all classes of somatic variants as these somatic variant calling tools consistently demonstrate poorer performance when tumour purity is low, particularly for low frequency variants17,24. In this study, most SNVs were detected in both short and long-read sequencing data but low tumour purity was a contributing factor to SNV discordance. The presence of reads supporting alternative alleles in germline samples also contributed to SNV discordance and resulted in discrepancies in driver SNV detection in this study; a PIK3CA E545K mutation was not initially detected in the long read post-treatment tumour despite its presence at a VAF of 20.0%. Moreover, SNVs in repetitive regions also contributed to SNV discordance. Similarly, we observed limited indel concordance between short and long read technologies which was driven by a lack of concordance among indels within simple homopolymers. This is consistent with previous work highlighting homopolymers as difficult regions for indel calling25. Generally, copy number calls across the genome were consistent between short and long reads, with both technologies indicating a past-WGD event and sub-tetraploid tumour genome26. However, copy number calls for specific driver genes were less congruous; homozygous deletions of several driver genes were reported in long read but not short read WGS data. While short reads can reliably detect most large (> 10kb) somatic SVs within mappable regions, long reads are better able to detect smaller SVs, insertions, and complex SVs21,27. In this case, we observed that Severus detected a greater number of insertions and smaller deletions compared to Manta, as well as a high concordance rate between insertions detected by Severus. However, concordance of SVs was low between technologies.This was exacerbated by differences in how SV calling tools denote variants; several insertions reported by Severus were reported as translocations by Manta which resulted in false negative concordance calls. Tumour evolution was assessed in this study as pre- and post-treatment tumour samples were available. Observed in both short- and long-read data, SNV and indel concordance rates were lower between the pre-treatment and post-treatment tumours. In terms of sCNAs, copy number profiles were generally stable between pre- and post-treatment tumours in both short and long reads. For short reads, most SVs were detected in the pre- or post-treatment tumour only. While this was also true for long reads, many SVs were still concordant between the pre- and post-treatment tumour. Low concordance of somatic variants between the pre- and post-treatment tumour may reflect the lower purity of the post-treatment tumour as well as the loss of subclones contributing to tumour heterogeneity. Mutational signature analysis provides insight into the biological processes acting in a tumour. SNVs, indels, sCNAs, and SVs have been assigned to specific mutational profiles and associated with exogenous and/or endogenous mutagens26,28,29. Mutations included in these profiles have thus far been derived from short read sequencing data. We sought to assess the relevance of these established short-read-based mutational signatures to long read mutational profiles. Similar single base substitution signatures were detected between short and long read data while indel signatures were inconsistent. While similar trinucleotide contexts were seen between SNVs private to short or long reads, platform-specific differences in the nucleotide contexts of indels were observed. However, inconsistencies in indel signatures were indicative of low indel concordance rather than unsuitability of indel signatures for long-read data. Copy number signatures were largely consistent between short and long read data though a signature associated with sequencing artefacts was detected in the long read data. Finally, SV signatures were not suitable for use with long read data as insertions and other SVs shorter than 1kb are not accepted as input for SV signature analysis using SigProfiler. It will be necessary to include data from long reads in the development of future copy number and SV signatures in order to make these signatures more applicable to long reads. While the addition of technical replicates would improve the robustness of the observations made, long-read WGS provided an additional epigenomic information that may be biologically relevant to this patient’s tumour. Overall, differential methylation analysis showed changes in methylation patterns that are commonly observed in tumours, including global hypomethylation and local hypermethylation of enhancers, promoters and CpG islands30. Finally, regions linked to TGFβ, Wnt, and EMT signalling pathways underwent epigenetic changes in the post-treatment tumour. These pathways have been linked to chemotherapy resistance31–33. Simultaneous generation of DNA sequencing and DNA modification data with ONT WGS enabled further biological insights into this patient’s tumour genome, without the need for additional sample material or sequencing costs required for bisulfite sequencing. The clinically relevant genomic characteristics of the LARC genome studied here were consistent between short and long reads. This patient’s tumour was histologically confirmed as MSS. Consistent with this, no observations of MSI were made in short or long read data. KRAS G12D and PIK3CA E545K mutations were detected in the tumour by short and long read WGS data. These driver mutations occur in 11.8% and 5.4% of MSS stable CRCs, respectively19. Approximately 50% of non-hypermutated early onset CRCs display early TP53 mutations with whole genome doubling34. In the 37-year-old patient involved in this study, we observed a non-hypermutated tumour genome and LOH of TP53 with a subsequent WGD event in both short and long read data. SVs affecting the CTNNB1 and CDH1 genes were consistently supported by short and long read data. Both CDH1 and CTNNB1 have roles in EMT and are known to drive CRC tumourigenesis35,36. The deletion in CDH1 affected exons 1 and 2 which encode the signal peptide of E-cadherin. This peptide is necessary for the translation and translocation of E-cadherein, with mutations affecting the signal peptide linked to hereditary diffuse gastric cancer37–39. Moreover, the CTNNB1 insertion detected was predicted to disrupt splicing, potentially resulting in a more stable form of β-catenin that could promote EMT and therefore, promote chemoresistance in this patient31,33,40–42. The additional DNA modification data from long-read WGS provided further context as to how genomic regions linked to EMT-related pathways underwent epigenetic remodelling in this patient’s tumour. This is a potentially relevant finding as this patient developed metastatic disease affecting the liver and omentum 9 months after beginning nCRT. Further validation of general SV and methylation observations would be needed to confirm biological relevance in future studies. Computational tools for somatic variant calling with long read WGS data are in early stages of development relative to short-read tools and thus care should be taken when applying these tools. Compared to short-read technologies, long-read technologies were capable of detecting key characteristics of this patient’s tumour genome and DNA modification data revealed additional potentially relevant biological insights. However, comparisons of these technologies in large patient cohorts are required to reach a conclusive result regarding the application of long-read technologies to clinical samples. Methods Clinical samples and ethical approval The study was performed in accordance with the Declaration of Helsinki. Informed consent was provided by the patient to perform whole genome sequencing (WGS) and ethical approval was obtained for the study from the Beaumont Hospital Research Ethics Committee (reference: 13/82 and 15/09). Clinical trial number: not applicable. In total, 5 samples were collected from the patient; a pre-treatment primary tumour sample and adjacent normal tissue were obtained, as well as matched blood. During surgical resection, post-treatment primary tumour and adjacent normal tissues were collected. As previously described, adjacent normal samples were obtained from a distal region of the rectal tract that was macroscopically unaffected and disease-free43. All specimens were snap frozen. Fresh-frozen tissues were split into three parts. Sections from the top and bottom of each tissue sample were stained with haematoxylin and eosin. A trained pathologist examined the stained tissues to ensure a minimum of 30% tumour in tumour tissues and no tumour present in the adjacent normal samples43. DNA extraction High molecular weight DNA was extracted from all samples following the Oxford Nanopore Rabbit Liver DNA extraction protocol, with one modification due to small sample size; the Blood and Cell Culture DNA Mini Kit (QIAGEN) was used instead of the Blood and Cell Culture DNA Midi Kit (QIAGEN). DNA was quantified using the Qubit dsDNA Broad Range Assay Kit (Invitrogen). DNA quality was determined using the Bioanalyser DNA 1000 kit (Agilent Technologies). Oxford nanopore whole genome sequencing DNA libraries were prepared using the Ligation Sequencing Kit (SQK-LSK114, ONT), as per the manufacturer’s protocol. Sequencing was performed with R10.4.1 PromethION flow cells (FLO-PRO144M) by a PromethION 2 Solo machine with a runtime of 72 h per flowell. Each flowcell was washed and reloaded as was necessary based on pore performance. Live basecalling was performed using the MinKNOW software (v23.11.4) with Dorado (v7.2.13) on high accuracy mode with 5mC and 5hmC calling enabled. Reads with a quality score less than 7 were discarded. FASTQ files were aligned to GRCh38 using minimap2 (v2.26) with the following settings; -ax map-ont44. BAM files were sorted and indexed using samtools (v1.9)45. Genome coverage was determined using mosdepth (v0.3.3)46. Somalier (v0.2.19) was used to ensure samples were related47. Illumina whole genome sequencing Extracted DNA was diluted using elution buffer. At Novogene, DNA libraries were prepared and paired-end sequencing carried out on an Illumina NovoSeq 6000. Adapter trimming was performed on FASTQ files using Trimmomatic (v0.39) with a modified TruSeq2-PE.fa file to remove partial adapter contamination (Supplementary Materials File 1)48. The parameters were ILLUMINACLIP:< adapters.fa>:4:4:5:2:True LEADING:3 TRAILING:3 MINLEN:36. Trimmed FASTQ files were aligned to GRCh38 using BWA-mem (v0.7.17)49. Using samtools (v1.12)45, BAMs were sorted and where samples were sequenced across multiple lanes, BAM files were merged prior to marking duplicates. Samtools was used to index BAM files. Using GATK (v4.5.0.0), duplicate reads were marked (MarkDuplicates) and base quality score recalibrated (BaseRecalibrator, ApplyBQSR)50. Genome coverage was assessed using mosdepth (v0.3.3)46. Somalier (v0.2.19) was used to confirm sample relatedness47. Small somatic variant calling Small somatic variants were detected in Illumina WGS by Mutect2 (GATK v.4.3.0.0) using by comparing tumour/adjacent normal tissues to matched blood. A panel of normals from the 1000 Genomes Project (1000g_pon.hg38.vcf.gz) and a germline resource from gnomAD (af-only-gnomad.hg38.vcf.gz) were supplied to Mutect2 to reduce false positive somatic calls. The ExAC common germline variants (small_exac_common_3.hg38.vcf.gz) were used with GATK (v4.3.0.0) GetPileupSummaries and CalculateContamination. Outputs from these tools were used in FilterMutectCalls to filter technical artefacts, non-somatic mutations, and sequencing errors. After these filtering steps, the VCF was split into SNVs and indels using bcftools view. Using ClairS (v0.4.0), single nucleotide variants and indels were detected in ONT WGS data by comparing tumour/adjacent normal tissues to matched blood24. The following parameters were set; --platform ont_r10_dorado_hac_5khz --enable-indel-calling --enable_clair3_germline_output –q8 –longphase_for_phasing. Enabling germline output was necessary to obtain germline variant VCFs required as input for other tools (i.e. Wakhan and Severus). Somatic small variants were required to meet the following criteria; within chromosomes 1–22, X and Y, marked as PASS, and with read depths of at least 4 in both the tissue of interest and the germline sample. Multiallelic variants were split using bcftools norm and indels normalised using GATK (v4.3.0.0) LeftAlignAndTrimVariants. Indels over 50 bp long were removed. Unique variant IDs in the format of ‘< chromosome>:< position>:< reference_allele>:< alternate_allele>’ were assigned to each variant using bcftools annotate. Variant effect predictor (VEP; v.107) was used to annotate small somatic variants51. To remove common variants, filter_vep was used to remove variants that were not listed in the Catalogue of Somatic Mutations in Cancer (COSMIC) database (v95) or were not novel variants (--filter “(Existing_variation match COS or not Existing_variant)”)52. Variant allele frequency was calculated as read depth of alternate allele/total read depth reported by a given caller. Small driver mutations were defined as any SNV or indel predicted or predicted and annotated as a driver by Cancer Genome Interpreter (CGI) with at least a one-star classification of oncogenicity on ClinVar53. CGI was used with hg38 and rectal adenocarcinoma (READ) tissue specified. Small variant concordance SNVs were considered concordant between sequencing technologies on the basis of SNV IDs; the chromosome, position, reference and alternate allele must agree. To evaluate indel concordance, several bedtools (v2.30.0) approaches were assessed; (1) complete and reciprocal overlap of indels (--f 1, -r), (2) minimum overlap of 1 bp (default settings), and (3) 1 bp overlap within a 5 bp window (slop –b 5)54. Similar bedtools approaches were used to assess overlap of indels with homopolymers. Simple repeat homopolymer files (GRCh38_SimpleRepeat_homopolymer_*_slop5.bed.gz) were obtained from Genome in a Bottle genome stratifications (https://github.com/usnistgov/giab-stratifications)55. Factors affecting SNV concordance were examined in further detail. Variants detected only in short-read WGS data by Mutect2 were evaluated in ONT data through another iteration of ClairS in which an additional options were used as follows: --genotyping_mode_vcf_fn was used to supply a VCF with SNVs private to Mutect and –phase_tumor to re-activate phasing of ONT data when supplying a VCF. Variants detected only in ONT WGS data were evaluated in ILMN data by using Mutect2 to force call at the SNV positions by supplying a VCF with SNVs private to ClairS to the –alleles option. Using the annotatr R package (v1.24.0), SNVs were annotated with genomic region (“hg38_genes_1to5kb”, “hg38_genes_exons”, “hg38_genes_promoters”, “hg38_genes_introns”, “hg38_genes_intergenic”). SNVs were annotated with repetitive regions using RepeatMasker information for hg38 available from the AnnotationHub R package (v3.6.0). Structural variant calling and comparison Structural variants in Illumina WGS data were detected using Manta (v1.6.0) with default parameters56. The supplementary convertInversion.py script was applied to the output VCF to reformat inversion entries to single inverted sequence junctions. Structural variants in ONT WGS data were detected using Severus (v1.3)21. BAMs were phased using WhatsHap (v1.7) haplotag57. Phased germline SNVs output by ClairS were provided to the haplotag command to ensure consistent phasing between samples. Phased tumour and germline BAMs and the phased germline VCF were used as input for Severus. Tandem repeat regions (human_GRCh38_no_alt_analysis_set.trf.bed.gz) were supplied to Severus using –vntr-bed. For Manta and Severus, structural variants marked as PASS and within chromosomes 1:22, X, and Y were selected for further analysis. Comparison of structural variants as performed using Minda (https://github.com/KolmogorovLab/minda, commit 2066ef5) with default parameters. Genomic coordinates of SV breakends were further compared using the GenomicRanges R package. Sequence similarity of insertion sequences was determined using BLAST58. Circos plots were generating using the circlize R package (v0.4.15). The effect of structural variants on splicing was investigated using the online version of SpliceAI (https://spliceailookup.broadinstitute.org/, accessed 31/07/2025)59. ShatterSeek (v1.1) was used to detect chromothripsis60. Somatic copy number alterations facetsSuite (https://github.com/mskcc/facets-suite, v2.0.9) was used to run FACETS (v0.6.2) on the Illumina WGS data to detect sCNAs61. FACETS requires a base-level coverage file as input which must be generated using snp-pileup (v0.6.2). The parameters -q15 -Q20 -P100 -r25,0 were applied as well as a file outlining common SNPs (obtained from dbSNP, build 151). The resulting pileup was used as input to the facetsSuite script run_facets_wrapper.R (--purity-cval = 1000 and –cval 500). Wakhan (https://github.com/KolmogorovLab/Wakhan, v0.1.0) was used to detected somatic copy number alterations in ONT WGS data. BAMs haplotagged with using phased normal VCFs output from ClairS were used as input. The normal phased VCF from ClairS and somatic SV breakpoints from Severus were provided to Wakhan using the –normal-phased-vcf and –breakpoints options, respectively. Subclonal analysis (--copy-numbers-subclonal-enable) and detection of loss of heterozygosity (--loh-enable) were enabled. A purity range was provided (0-0.60). For comparative purposes, a list of all genes assessed by FACETS was provided to Wakhan using --user-input-genes. Copy number profiles were visualised using R. Mutational signature analysis Single base substitutions, indels, copy number, and structural variant signatures were assigned using SigProfilerAssignment (v0.1.8) with COSMIC v3.462. Cosine similarities of SNV and indel sets were calculated and plotted using the MutationalPatterns R package (v3.8.1). To perform copy number signature analysis on ONT data, haplotype-specific bed files output by Wakhan were combined using awk and modified to replicate the output format of the FACETS segmentation file. The version of Wakhan used in this analysis (v0.1.0) does not perform copy number calling in centromeric regions and as a result segments surrounding centromeres reported 0 coverage. Segments flanking centromeres were extended to fill in centromeric gaps and avoid introduction of false homozygous deletions to the copy number signature analysis. For structural variant signature analysis, output VCFs of Manta and Severus were filtered for PASS variants and converted to BEDPE format. All mutational signature contributions and mutational matrices were visualised in R. Methylation analysis Unmapped BAMs containing methylation information were output by the MinKNOW software. Unmapped BAMs were sorted by read name and merged. Merged uBAMs were converted to FASTQ format using –T * to ensure all tags were carried over and subsequently aligned to GRCh38 using minimap2 (v2.26) with the –ax map-ont –y options. Aligned BAMs were sorted by coordinate and reads with MAPQ < 20 were removed. Longphase (v1.7.3) haplotag was used to phase methylBAMs using germline SNVs detected by ClairS and SVs detected by Sniffles2 (v2.6.0)63,64. Modkit (https://github.com/nanoporetech/modkit, v0.5.0) pileup was used to generate bedmethyl files. Bedmethyl files were filtered for all sites within autosomes and sex chromosomes. The DSS-single method available in the DSS R package (v 2.54.0) was used to detect differentially modified regions between conditions using single replicates by enabling smoothing with the DMLtest() function22. CallDMR() was used with minlen = 100, minCG = 15, dis.merge = 100. The minimum change in modification (delta) was set to 15% for DMRs and 5% for DhMRs. Differentially modified regions were linked to genomic features using bedtools intersect. The genomic coordinates of canonical transcripts (knownGene, exons, introns) for GRCh38 were obtained from UCSC table browser. Promoters were defined as − 2Kb and + 200bp from the transcription start site. RepeatMasker and CpG island coordinates for GRCh38 were also obtained from UCSC table browser. Enhancers were obtained from GeneHancer v5.2465. Only elements with elite status classed as enhancers or enhancer/promoters were included in this analysis. Functional enrichment of genomic regions was performed using rGREAT (v2.9.2) with windows extending from genes and the UCSC knownGene transcription start sites23. Hallmark MSigDB gene sets were used. Supplementary Information Below is the link to the electronic supplementary material. Supplementary Material 1 Supplementary Material 1