Overview of the Genomics England TGCT cohortAll TGCT tumour-normal sample pairs were processed through 100kGP bioinformatic somatic-variant analysis pipelines (tumour coverage: 95–122.7×, mean: 108.7×; normal: 31.3–62.6×, mean: 39.3×). We restricted our analysis to high-quality data derived from fresh frozen material, involving 60 tumour samples from 57 individuals [age 17–77 years (y); median 35 y]; (55 untreated primary and five late-stage treated metastatic TGCT), including four primary tumour regions sampled from a single participant (Fig. 1a, Supplementary Data 1). The primary tumours comprised 39 pure seminomas and 16 NSGCT, including three EC cases and one undifferentiated teratoma. A bimodal age distribution was observed at diagnosis in participants, as expected, with most seminomas being diagnosed between age 20 y and 40 y (Supplementary Fig. 1).Fig. 1: Mutational landscape of adult testicular germ cell tumours (TGCT).a Genomic profiling of primary and metastatic TGCT samples with matched germline DNA from peripheral blood. Samples were collected from 60 participants recruited from seven NHS Genomic Medicine Centres (GMCs) across England as indicated on the map. The human silhouette drawing was modified from a template from V<underline>ecteezy.com</underline> (https://www.vecteezy.com/vector-art/299365-medical-infographic-of-human-body). b From top to bottom: number of coding mutations identified in each sample; number of insertions and deletions (indels) in each sample; total number of structural variants in each sample, separated into tandem duplications (TD), deletions (DEL), head-to-head (H2HINV) and tail-to-tail (T2TINV) inversions, transversions (TRANS); proportion of mutations assigned to single base substitution (SBS), insertion/deletion (ID), and doublet base substitution (DBS) mutational signatures; TGCT subtype; tumour type (primary or metastasis); clinical stage; age-group of participant; mutation status of KIT driver gene; mutation status of RAS (KRAS or NRAS) driver genes; presence or absence of 12p amplification according to GISTIC2. Exposures or processes linked with mutational signatures are listed. Two samples that were not sequenced via a PCR-free workflow are excluded from this figure. HRD homologous recombination deficiency, amp amplification, mut driver mutation, NHEJ non-homologous end joining, NSGCT non-seminomatous germ cell tumours, ROS reactive oxygen species, y years of age.Across the GEL cohort, we identified 80,760 individual single nucleotide variants (SNVs), 7412 small insertions and deletions (indels), and 1865 chromosomal rearrangements (Fig. 1b, Methods). As per previous reports, tumours were typified by a uniformly low rate of single nucleotide variants (SNVs; mean genome-wide substitution rate of 0.475/Mb; range 0.095–1.62), likely reflecting the embryological origins of TGCT4,5. No tumour displayed a hypermutated phenotype, i.e., excessively high SNV/indel mutation burden (maximum SNV/Mb = 1.62; maximum indel/Mb = 0.28).Identifying subtypespecific driver mutationsUsing the IntOGen pipeline, which aggregates seven complementary driver discovery algorithms, we searched for driver genes across the GEL TGCT cohort (Methods, Supplementary Data 2). Eight genes were significantly somatically mutated (KIT, KRAS, NRAS, RAC1, SPEN, EP300, KLF4, KMT2C). Consistent with The Cancer Genome Atlas (TCGA) study of TGCT5, KIT driver mutations defined a subset of seminomas (Fig. 2). Mutations in KIT were clustered primarily in exon 17, in a pattern similar to that previously reported in testicular seminomas and intracranial GCTs (Supplementary Fig. 2, Supplementary Data 2)14,15. Multiple mutations affecting the same oncogene (KIT) were observed in only one participant with a clinical stage II seminoma. Further analysis identified additional drivers defining distinct subgroups within the seminoma subtype. These encompassed gain-of-function mutations in the transcription factor KLF4 and the GTPase RAC1, as well as loss-of-function mutations in the histone acetyltransferase EP300 (Fig. 2). Additionally, we searched for non-coding drivers using three complementary algorithms, namely OncodriveFML16, OncodriveCLUSTL17, and ActiveDriverWGS18. However, we did not identify any significant non-coding elements under positive selection (Supplementary Fig. 3).Fig. 2: Heatmap of molecular mutations in testicular germ cell tumours (TGCT).In total, 57 individual adult participant samples were analysed. Point mutation and indel drivers independently identified in The Cancer Genome Atlas (TCGA) and Genomics England (GEL) TGCT cohorts are shown alongside annotated GISTIC2 focal segments. Driver presence/absence in the Memorial Sloan Kettering—Metastatic Events and Tropisms (MSK-MET) TGCT cohort is also shown. Seventeen recurrently mutated genes were found in the cohort, respectively, with KIT being the most frequently altered gene (Supplementary Data 2). The colour code indicates mutation type(s) or TGCT subtype (see legends). Samples with more than one type of mutation (missense, nonsense or in-frame insertion) in the same gene correspond to ‘Multi Hit’ events. Samples with driver mutations and amplifications/deletions in the same gene are indicated with an overlay of two colours. Amp amplification, Del deletion, CNA copy number alteration, NSGCT non-seminomatous germ cell tumours.To supplement our analysis of GEL tumours, we reanalysed data from TGCT cohorts within the TCGA5 (128 samples) and Memorial Sloan Kettering – Metastatic Events and Tropisms (MSK-MET)19 (128 samples) studies, allowing us to identify further subtype-specific coding drivers. Within the TCGA dataset, somatic mutations in 10 genes reached significance, including NOTCH1, PIK3CA, BIRC6, ARID1B, and LRP1B. Two putative driver genes, PTMA and FAT4, were identified in NSGCT subtypes and primarily subject to loss-of-function mutations (Supplementary Data 2). GEL cohort data also provided support for PTMA (prothymosin alpha) as a putative driver gene, previously implicated in TGCT, though not currently included in the COSMIC Cancer Gene Census20,21.Finally, we assessed the clinical actionability of identified driver gene mutations by referencing the OncoKB Knowledge Base (http://oncokb.org/)22, and found that 17% (19/110) of alterations annotated by OncoKB were targetable (OncoKB Level 1-4). Most targetable mutations (18/19) were Level 3B, indicating predictive biomarkers that are considered standard-of-care for a different tumour type.Cancer driver genes in focal genomic alterationsThe Battenberg algorithm was used to estimate clonal and subclonal copy number variation across the cohort23. Applying GISTIC224 to these profiles, we identified 29 genomic regions recurrently affected by focal amplifications and deletions (Methods, Supplementary Fig. 4, Supplementary Data 3). In addition to established recurrent copy number alterations (CNAs), including chromosome arm-level gains spanning KRAS (12p), amplifications involving KIT (4q12; 19% cases) and MDS2 (1p36.32; 17% cases), and deletions spanning DMRT1 (9p24.3; 37% cases), which is associated with testicular germ cell tumour susceptibility25, we identified 26 additional novel events. Although KIT mutations appeared to be restricted to a subset of seminomas, amplifications spanning KIT were also observed in NSGCT (Fig. 2). Segments 1q21.3 (14% cases), 7q11.23 (46% cases), and 22q11.1 (25% cases) spanning oncogenes SETDB1, CDK6, and DGCR8 respectively, were found to be recurrently amplified. Focal deletions spanning cyclin A1 (CCNA1) and the transcription factor FOXO1 (13q13.3), critical for successful spermatogenesis, were found to occur exclusively in seminomas. Notably, focal gains spanning AFP (4q13.2) were also restricted to a subset of seminomas (8/57). Although alpha-fetoprotein is a serum tumour marker typically associated with non-seminomatous germ cell tumours, previous reports have noted elevated serum AFP levels in some cases of histologically pure seminomas26,27. Several recurrent deletions spanned WNT signalling-related genes including the cadherins CDH1 and CDH11, CREBBP (16q24.2) and SMAD4 (18q22.2). Mutual exclusivity analysis revealed that the most prominent driver events were largely not co-occurring, although the most significant driver interactions identified were cooperating events including PIK3CA-MCL1 amplifications, and RB1-FLI1, RB1-MEN1, and MAF-SMAD4 deletions (Supplementary Fig. 5). Mutually exclusive events were identified involving KIT-MAF and PTMA-MCL1. The sole intra-chromosomal pair identified consisted of co-occurring MEN1-FLI1 deletions.A primary somatic feature in TGCT development is copy number gain of chromosome 12p, typically structured as an isochromosome (i12p)7,28. We observed allelic copy number profiles consistent with the presence of at least one i12p in 75% (43/57) of tumours. A subset of these (5/43; 12%) were categorised as canonical chromosomes (Supplementary Methods) but characterised by complex rearrangements of the 12p arm. Complex i12p cases were all seminomas with recurrent focal loss at 11q24.3 encompassing the ETS transcription factor FLI1 (Supplementary Data 3). Most tumours lacking the i12 p event were seminomas (13/14; 93%) and instead had at least four copies of 12p. Only two samples exhibited 12q loss of heterozygosity (LOH), suggesting that most tumours had undergone duplication of chromosome 12 or a second WGD before i12p formation, as previously described29.Hotspots of structural variation in TGCTUsing methods described by Glodzik et al.30, we identified a single structural variant hotspot involving large (>100 kilobases, kb) tandem duplications (TD) and eight deletion hotspots (Supplementary Data 4). We observed one TD hotspot in the region of chr19:55–58 Mb spanning the histone methyltransferase, GLP. Interestingly, a gain-of-function mutation in the Caenorhabditis elegans Notch receptor glp-1 has been described, leading to germline tumour formation31. However, this hotspot did not overlap with any GISTIC-defined focal amplifications. Deletion hotspots associated with copy number loss were centred on chr3:60 Mb spanning the fragile histidine triad (FHIT) gene, chr9:7-12 Mb covering the tyrosine phosphatase PTPRD, and chr16: 78–84 Mb targeting cadherin 13 (CDH13). We observed chromothripsis in one tumour (GEL-TGCT-0056), a rare case of metastatic teratoma with somatic-type malignancy, in which a cluster of 23 structural variants arose in a single catastrophic event affecting chromosomes 7 and 17, including amplification of PPM1D (Supplementary Data 4). Canonical translocations and fusions associated with Ewing’s sarcoma and related primitive neuroectodermal tumours were not detected in this participant.KRAS amplification on extrachromosomal DNAWe next leveraged the GEL dataset to explore the landscape of extrachromosomal DNA (ecDNA) formation in testicular cancer. EcDNA is often associated with oncogene amplification and poor clinical outcomes in many cancers32. Amplicon structures were detected and classified in TGCT using the Amplicon Architect tool33. Amplicons were identified in 85% (46/54) of the TGCT samples (Methods, Supplementary Data 5). The size of single-interval amplicons detected ranged from 116 kb to 76 Mb (median 4 Mb), and over 85% (113/130) were >1 Mb. Complex rearrangements identified in at least two samples spanned the established TGCT oncogenes KRAS, MYC, EGFR, and members of signalling pathways commonly dysregulated in cancers including WNT (SOX2), RTK (PDGFRA), and the p53 pathway inhibitors, CDK4/6 and MDM2 (Supplementary Data 5). The only oncogene identified within cyclic amplicon structures, including ecDNA in one instance, was KRAS, and only in seminomas. Amplicons showing a signature34 of having been created by a breakage-fusion-bridge mechanism were also exclusively identified in seminomas. Seminomas also carry a significantly higher number of amplicon structures relative to NSGCT (p = 0.018; Supplementary Fig. 6).Complete repertoire of mutational signaturesTo gain insight into the aetiological basis of mutation, we extracted mutational signatures (Supplementary Figs. 7–8, Supplementary Data 6). In most tumours, the majority of single base substitutions (SBS) could be assigned to signatures SBS5/SBS40 and SBS1 (using nomenclature established in ref.35), thought to result from endogenous clock-like mutagenic processes (Supplementary Fig. 8); however only SBS5 and the number of C > T mutations at NpCpG trinucleotides correlated with age (p = 4.3 × 10−8 and p = 0.02, respectively; Supplementary Fig. 9). Seminomas with mutant KIT had significantly lower SBS1 than either wild-type seminomas (p = 0.0028) or NSGCT (p = 5.5 × 10−6).Some TGCT subtypes exhibited distinct SBS patterns. SBS18, a signature linked with damage by reactive oxygen species (ROS), was detected in two tumours, both NSGCT with minor YST components. Notably, in GEL-TGCT-0038, the majority of variants were attributable to SBS18. This signature has previously been described in multiple paediatric cancers, placental tissue, and most recently in patients with pre- and peripubertal YSTs11,36. A signature attributable to platinum chemotherapy exposure, SBS35, was detected in two post-chemotherapy metastases, as expected. SBS31, another signature related to platinum drug treatment, was also found in a clinical stage I primary seminoma treated with radical orchiectomy and carboplatin after sampling. SBS32, a signature not reported in prior TGCT studies, and associated with azathioprine treatment37, was detected in 11% (6/57) of participants, despite no documented medical history indicating that any of these participants had received such treatment. Of note, a similar finding was recently reported in acute myeloid leukaemia patients, implying mutational mechanisms other than exposure to azathioprine may contribute to SBS3238. Changes in mutational signature activity between clonal and subclonal mutations were observed (Supplementary Fig. 10), with a general trend towards a lower proportion of subclonal mutations attributed to SBS5 (p = 2.2 × 10−16; test for trend in proportions). Analysis of the indel (ID) mutational spectra revealed a predominance of ID1 and ID2, both due to slippage during DNA replication35. Deletion patterns characterised by ID6 and ID8 and arising from distinct mechanisms of DNA double-strand break repair35 were mutually exclusive (Supplementary Data 6). The majority of doublet base signatures (DBS) identified in TGCT were of unknown aetiology, except for those associated with tobacco smoking (DBS2) and platinum chemotherapy (DBS5).We next examined mutational processes generating genomic rearrangements in TGCT. To detect these, we first applied a recently developed framework39 for classifying chromosomal instability in cancer from 21 pan-cancer copy number signatures (CN1-CN21) (Supplementary Figs. 11–12). The tetraploidy-associated signature CN2 was found in most samples, across both seminomas and NSGCT. We also identified an attribution of both CN1 and CN2 signatures together across a number of tumours, indicating a hyperdiploid or sub-tetraploid profile39. We identified contributions from CN13-CN15, a family of signatures characteristic of specific numerical chromosomal instability, encompassing whole-arm or whole-chromosome-scale loss of heterozygosity events. CN13, which is dominated by LOH segments of total copy number 1, was restricted to NSGCT. Co-occurrence of signatures CN1, CN13, and CN15 was observed in a small number of participants (3/57; 5%) with copy number profiles showing significant amounts of copy-neutral LOH and only in metastatic samples or, notably, primary cases that reported subsequent metastases, suggesting potential clinical relevance for this signature in TGCT.Next, we classified structural rearrangements in subclasses considering their type and size (Methods), applying the same statistical framework used for other classes of mutational signatures40. This approach revealed two structural variant signatures (S1, S2) (Supplementary Figs. 13–14), present in both seminomas and NSGCT. Signatures S1 and S2 were similar to recently described rearrangement reference signatures characterised by unclustered translocations (RefSig R2) and unclustered deletions up to 100 kb (RefSig R5), respectively (Supplementary Fig. 15)40. Previously described associations include RefSig R5 with BRCA2 mutations and RefSig R2 with driver mutations in TP5340. Although BRCA2 mutations were not detected in the GEL cohort, tumours exhibiting recurrent deletions spanning BRCA2 displayed a significantly higher prevalence of signature S2 rearrangements (p = 0.001528, Wilcoxon rank sum test). Other signatures associated with inefficient homologous recombination repair are either not detected in the GEL TGCT cohort (SBS3) or are present in a small number of cases (ID6/ID8). Thus, it is not clear that loss of BRCA2 contributes to the overall signature repertoire.Prevalence of whole genome duplicationWhole genome duplications (WGD) are near universal in TGCT, with recent work showing these events occur early in embryogenesis5,11. In all but one case (56/57), tumours from the GEL cohort were shown to have undergone WGD (Supplementary Fig. 16). Using MutationTimeR41, we timed somatic mutations relative to copy number gains and calculated the relative timing of these gains. We then timed the occurrence of WGD, using the ratio of clock-like mutations occurring before and after WGD (Methods). We observed a median of  ~ 9 substitutions (range 0-375) occurring prior to WGD, and in seven cases we did not observe any pre-WGD substitutions, supporting early occurrence of genome duplication, likely in utero (Fig. 3a). This observation is in stark contrast to most solid cancers, where WGD events are broadly distributed throughout clonal evolution and likely stochastic (Fig. 3b). However, in three cases, genome doubling events were estimated to occur much later relative to the rest of the cohort. One of these samples, an extensively metastatic GCT with a predominant EC component, carried an estimated 375 pre-duplication substitutions. A further two cases, both clinical stage I seminomas, also exhibited relatively late WGD. Both were metachronous bilateral testicular tumours; one participant had their first TGCT diagnosis almost 30 years before 100kGP sample collection, and the other was diagnosed for a second time five years after sampling. In the metastatic case, there was a past history of a bilateral retractile testis but with no previous report of bilateral TGCT. Recent single cell analyses suggest that neonates possess a small pool of gonadal cells with characteristics of primordial germ cells (PGCs) in their testes42. It is therefore conceivable that PGC-like cells lingering into infancy could undergo the same WGD process.Fig. 3: Timing of whole genome duplication (WGD) events across Pan-Cancer Analysis of Whole Genomes (PCAWG) and Genomics England (GEL) testicular germ cell tumour (TGCT) cohorts.a Bar plot showing estimated pre-duplication mutation burden (yellow) and total clonal mutation burden (dark grey) per TGCT. The dashed line indicates the median pre-duplication burden across all samples. Samples where WGD occurred relatively late are shaded in light grey. b Number of samples with a WGD event in each cancer type is shown alongside the corresponding violinplot. Data points from PCAWG appear in grey. Data points from GEL appear in red. PCAWG cancer types with less than four WGD samples not shown. Tumour abbreviations reported as per PCAWG study (ref. 41). c Distribution of synchronous (sync) and asynchronous (async) gain patterns across GEL TGCT genomes, split by ploidy status (WGD whole genome duplication, ND near diploid). Uninformative samples had too few mutations or gained segments to allow accurate timing.We then estimated the time point during PGC development that WGD occurred by dividing pre-duplication substitution burden estimates by the reported mutation rate per cell division within PGCs43, as described in Oliver et al.11 (Supplementary Methods). Excluding the late WGD cases, median WGD was estimated to occur at ~11 cell divisions in TGCT (range 0–71.5, lower and upper bounds of post-PGC cell divisions), setting the genetic hallmark of TGCT initiation in the developmental period. Most tumours with WGD (42/56; 75%) had synchronous chromosomal gains (Supplementary Methods, Fig. 3c), broadly in line with the distribution of gain patterns reported by the Pan-Cancer Analysis of Whole Genomes (PCAWG) in tumours with WGD41. A subset of tumours (12/56; 21%) that had undergone genome duplication evidenced asynchronous gains; asynchronous gains were only observed in pure seminomas or NSGCT with a predominant EC or seminoma component, suggesting divergent patterns of chromosomal evolution underlying histogenesis (Fig. 3c). Moreover, the proportion of CNAs attributed to signature CN2 was significantly higher in samples with synchronous gain patterns (P = 0.029, Wilcoxon rank sum test), while the proportion of CNAs attributed to CN14 was higher (P = 0.006, Wilcoxon rank sum test) in asynchronous genomes.Relative timing of genome doubling and driver mutations in TGCTUsing a permutation approach, we identified CNAs with evidence for significant enrichment or depletion across the GEL TGCT cohort and in the seminoma and NSGCT subgroups (Methods, Supplementary Data 7). A probabilistic timing model was used to reconstruct the order of acquisition of recurrent genomic aberrations, including WGD, enriched CNAs, and putative driver mutations across all TGCT genomes and within each of seminomas and NSGCT (Methods, Fig. 4). Enriched gains spanned known cancer and TGCT drivers including MYC (8q11-q24), EGFR (7p11.2), and BRAF (7q34). Similarly, enriched LOH events covered tumour suppressor genes such as APC (5q22.2), ATM (11q22.3), and CDX2 (13q12.2). No evidence was found for enriched homozygous deletion events. We further identified CNAs with evidence for significant negative enrichment in TGCT, implying that these events are less important for, or perhaps incompatible with, driving tumourigenesis in TGCT or in the context of widespread WGD (Supplementary Data 7).Fig. 4: Probabilistic ordering reveals most likely timing of copy number and driver events in TGCT.a Genome-wide landscape of clonal and subclonal loss of heterozygosity (LOH), gain, and homozygous deletion (HD) events in Genomics England (GEL) testicular germ cell tumour (TGCT) cohort. The y-axis corresponds to the fraction of tumours with a particular event. Events identified as enriched by our model and genes of interest within these regions are labelled. b Probabilistic ordering (left panel) of significantly enriched copy number events, whole genome duplication (WGD), and IntOGen-identified mutational drivers. A Plackett-Luce model was used to order events by sampling from all possible tumour phylogenies across the entire dataset (1000 iterations). Events are ordered along a timing scale (x-axis) from early to late by the mean value of the relative timing estimates. Horizontal lines show the range of time scale values inferred across the cohort for each event. The vertical lines and points for each event represent the mean and standard deviation for each distribution. The grey dashed vertical line represents the mean timing estimate of WGD across all samples. The proportion of these events present in different subtypes and clinical stages is shown in two central panels. Horizontal lines (right panel) indicate the minimum and maximum age of diagnosis in individuals harbouring a mutational event. The shaded grey circle indicates the median age of diagnosis corresponding to each mutational event. The dashed red line shows the median age of the cohort. c. Plackett-Luce-based probabilistic ordering of enriched events and mutational drivers only in seminomas. Seminomas were split into two groups (young-onset, late-onset; Supplementary Methods). The grey dashed vertical line represents the mean timing estimate of WGD across all samples. LOH loss of heterozygosity, NSGCT non-seminomatous germ cell tumours excluding tumours with seminomatous components, NSGCT (Sem) NSGCT with seminomatous components.In line with our analysis of WGD developmental timing, tetraploidisation was consistently the earliest event seen, followed by 12p gains spanning the KRAS locus, which may imply an initiating tumourigenic role in adult TGCT. To more accurately estimate the timing of high-level copy number gains specific to chromosome 12, we used AmplificationTimeR44, a method for timing individual amplification events (Supplementary Methods, Supplementary Data 7). Within most samples analysed, findings were in keeping with early timing of whole genome doubling. However, there is evidence to suggest that in some participants, chromosome 12 gains instead represent the earliest occurring events in the evolutionary history of the tumour, occurring pre-WGD (Supplementary Fig. 17).Most of the early events following WGD were gains and showed balanced representation across TGCTs (Fig. 3b), although later enriched events were specific to TGCT subtypes, such as the 12q11 gain spanning KIF21A and restricted to NSGCT. The only CNA event uniquely enriched in seminomas was a recurrent LOH spanning BRCA2 (chr13:18–114 Mb). Whilst not statistically significant, most individuals (12/14; 86%) harbouring this event belonged to the ‘young-onset’ group (<40 years; Supplementary Methods). Additionally, a subset of enriched CNAs were specific to young-onset seminomas, such as chr8:45–129 Mb and chr7:60–159 Mb (Fig. 4c).A Dirichlet Process clustering algorithm was used to cluster SNVs and indels according to their cancer cell fraction45. There was no significant variation in the proportion of SNVs, indels, or CNAs identified as subclonal across participants according to tumour stage or subtype (Supplementary Fig. 18a, b; Supplementary Data 7). Multi-site clonality analysis of four whole-genome sequenced regions from one participant (GEL-TGCT-0058; pure seminoma) point towards limited intra-patient tumour heterogeneity (Supplementary Fig. 18c). Across all subtypes, mutations in driver genes including KIT, KRAS, and NRAS were relatively late events, occurring post-WGD and after corresponding copy number gains or other CNAs (Fig. 4b, c). Participants with KRAS and KMT2C drivers typically had a higher age at diagnosis.HLA loss enriched in seminomasNone of the 60 tumours harboured nonsynonymous mutations in human leucocyte antigen (HLA) genes (Methods). However, LOH at the HLA locus, where either the maternal or paternal allele is lost, was identified in six tumours using the LOHHLA algorithm46 (Supplementary Data 8). HLA LOH affected a single type-I gene in two seminoma cases, and the HLA-A and -C genes in another three. HLA LOH potentially affected all three HLA genes in a single case, which was the only NSGCT case affected. In GEL-TGCT-0053, a post-chemotherapy lymph node metastasis, LOH was detected in both HLA-A and HLA-B, and although LOH could not be established in this case where two highly similar HLA-C haplotypes were observed (C07:02 and C07:01), the ordering of HLA genes suggested it was likely that there was also loss of HLA-C. We found no significant associations between HLA homozygosity (Methods) and either age of diagnosis, clinical stage, or pathological stage.Allelic imbalance without LOH, i.e., HLA imbalance as a result of unequal copy gain at the HLA locus or LOH not reaching statistical significance, was observed in a further 17 cases and in both seminomas and NSGCT. The majority of cases where HLA imbalance was observed were in seminomas (9/17; 53%). In most other tumours with HLA imbalance, the major histological component was EC (6/17; 35%), suggesting subtype-specific mechanisms of immune disruption in TGCT. HLA or B2M mutations, which can disrupt neoantigen-MHC binding, were not observed (Methods). Scanning for somatic mutations in genes involved in antigen presentation and processing (Methods), we found one seminoma exhibiting HLA LOH had also acquired a mutation in the proteasome regulator PSME4, which plays a key role in immunoproteasome activity and generating immunopeptidome diversity. Collectively, these findings suggest that while HLA mutations are unlikely to be a major mechanism of immune evasion in TGCT overall, HLA LOH could represent a mechanism of immune disruption and/or escape, primarily in a subset of seminomas, though further study is required.TGCT samples had a median of 10 neoantigenic mutations, mostly arising from SNVs (Supplementary Data 8). We found no significant difference in neoantigen burden between samples with HLA LOH, HLA imbalance, or an intact HLA locus. We next explored tumours with HLA LOH and evaluated whether the LOH event affected their neoantigen landscape (Supplementary Fig. 19). To do this, we computed the number of antigenic peptides predicted to bind the allele lost in the LOH event and compared it with the number of peptides binding to the retained allele. We found no significant difference overall, but observed a trend in 3/6 samples (GEL-TGCT-0007, GEL-TGCT-0018, GEL-TGCT-0050) for a higher number of binders associated with the lost, rather than the retained, allele. In 2/6 samples we observed the opposite trend, though the difference between lost/kept-associated binders was less striking. These observations could suggest that in some seminomas, HLA LOH provides functional escape from immune selection pressure, whereas in the other samples, immune selection is negligible, or HLA LOH is secondary to other non-genetic escape mechanisms.
