Background: Arterio-venous malformations (AVMs) are rare congenital high-flow vascular anomalies characterized by abnormal direct artery-to-vein communications, lacking intervening capillaries. Complex extracranial AVMs, typically have a limited response to standard treatments such as sclerotherapy or surgery, leading to a progressive disabling pathology with lifelong consequences that include chronic pain, deformity, and disfunction. Somatic variants in the RAS-MEK pathway are implicated in AVMs, generating interest in the role of MEK inhibitors to treat this condition. However, open biopsy for molecular characterisation carries a potentially life-threatening bleeding risk.
We aimed to evaluate the effectiveness of liquid biopsy for somatic genotyping of AVMs in children and young adults. We hypotheses that liquid biopsy from efferent draining veins is a reproducible and feasible technique for obtaining DNA from AVM, and that detection of the underlying somatic mosaic variants from AVMs can assist in directing targeted pharmacotherapy.
Methods: 10 patients with complex AVMs for which endovascular treatment or diagnostic catheter angiography was being undertaken were enrolled and 15 blood samples collected (11 efferent vein, 3 synchronous peripheral vein, and 1 peripheral vein only). Ultra-deep panel next-generation sequencing was conducted on cell-free DNA (cfDNA), followed by error-correction sequencing data analysis and variant calling.
Results: Pathogenic variants were identified in 12 samples from 8 patients (levels 0.11%-20.06%). No procedural complications occurred from efferent sampling. Three patients with identified variants began genotype-directed pharmacotherapy, demonstrating marked clinical improvement.
Variant levels inversely correlated with the sampling distance from the AVM nidus (p=0.004). Similarly, variant levels in synchronous peripheral collections were reduced by 33-100% in 3 patients (p=0.035).
Conclusion: Somatic variants of the RAS-MEK pathway are common in extracranial AVMs and can be readily identified using efferent vein liquid biopsy, albeit with a moderate yield in peripheral samples. This approach reduces bleeding risk, enabling broader access to targeted pharmacotherapy for AVM patients.

Background: The development of new cancer therapies is constrained by the proteome's limited druggability, estimated at only 10-15%. Additionally, the increasing recognition of non-coding RNAs' roles in disease underscores the urgency for innovative therapeutics that move beyond conventional protein-targeting strategies. The emergence of RNA-targeting small molecules presents a promising avenue for novel therapeutic approaches, potentially expanding the range of actionable targets. Recent advancements include the use of splicing modulator compounds (SMCs) to suppress gene expression by introducing deleterious cryptic exons (cryptons) in a sequence specific manner.
Methods: In this study, we developed a computational method to systematically identify potential targets for SMCs, specifically cryptons within the human genome. This method combines deep learning with high-throughput RNA-sequencing data from human tissues to discover high confidence putative cryptons.
Results: We detected around 50k cryptons and predicted their therapeutic potential based on their ability to disrupt gene expression through PTCs, frameshift mutations, or the alterations of protein folding due to the introduction of novel peptide chains. Furthermore, our analysis revealed 41,664 previously unannotated cryptons, some of which are predicted to interfere with several oncogenes traditionally considered undruggable. We further validated the existence of several of cryptons in therapeutically relevant genes, by engineering one of the spliceosome components to match a specific cryptons 3’ splice sites.
Conclusion: Our findings provide a valuable resource for exploring potential therapeutic targets of SMCs and lay the groundwork for future drug screening efforts.

Background: Many paediatric cancer tumours are driven by rearrangements in neurotrophic tyrosine receptor kinase (NTRK) genes (NTRK1, NTRK2 and NTRK3). While several inhibitors targeting tropomyosin receptor kinases family genes have been either approved or are at an advanced stage of clinical development, many paediatric cancer patients develop resistance to these inhibitors. Tumour evolution studies in adult cancers suggest polyclonal tumours likely to develop therapy resistance. We hypothesized that characterizing the tumour evolution of pediatric cancers with NTRK fusions might reveal insights into resistance mechanisms.
Methods: In our extended ZERO-Childhood Cancer cohort, we identified 26 tumours (from 24 children) with NTRK-fusion events. This cohort includes central nervous system tumours (CNS; n=13), sarcomas (n=8), neuroblastomas (n=1) and other solid tumours (n=4). We performed ~90X whole genome sequencing on these tumours and corresponding germline samples (at ~30X) and analysed these data using computational genomic tools, including SAGE, PURPLE and LINX to identify somatic single nucleotide variants (SNV), copy number variants (CNV) and structural variants. We then applied SNV based tumour evolution framework to identify clonal and subclonal populations. The order of mutation clusters was inferred using evolutionary principles. We also identified four tumour samples that have undergone whole-genome doubling (WGD). The timing of WGD was inferred using GRITIC pipeline.
Results: In this cohort of NTRK-fusion tumours, SNV based clonality analysis revealed that 23 out of the 26 tumours were polyclonal in nature, while the monoclonal tumours exhibited low purity. Multi-time point analyses from two patients showed evidence of branching evolution. One CNS tumour presented with founding whole genome doubling event, while in other tumours WGD was relatively late event.
Conclusion: Polyclonality is a common feature of NTRK-fusion paediatric tumours, while timing of WGD may explain evolution of some tumours. Such inherent features of clonal evolution may help delineate therapy resistance in these tumours.

Background: Molecularly targeted therapies improve outcomes for childhood cancer patients and are made possible by personalised medicine approaches like the Zero Childhood Cancer Program (ZERO). Mutations in CBL, an E3 ubiquitin ligase, represent one such molecular target in acute myeloid leukaemia, where CBL mutation mediates activation of the Receptor Tyrosine Kinase (RTK) FLT3 and sensitivity to FLT3-targeted Tyrosine Kinase Inhibitors (TKIs). We have identified novel CBL mutations in paediatric CNS and solid tumours, raising the possibility that CBL mutations mediate RTK activation in other tumour types and these patients may benefit from TKI therapy.
Methods: We analysed the ZERO cohort for somatic mutations in CBL. Molecular (whole genome, RNA sequencing and methylation) and high-throughput drug screening data from CBL mutated-patients, performed as part of ZERO, were interrogated. In silico pathogenicity analysis (ESM-2) was performed on novel variants. CBL mutations were cloned from patients and modelled in vitro to analyse the functional and signalling impacts of CBL mutation.
Results: We identified 26 somatic CBL variants in 22 individual patients, including 14 patients with CNS tumours. Thirteen non-haematological tumour samples harboured CBL mutations that are established drivers (exon 8/9Δ) or suspected to be oncogenic given their location in critical domains and in silico pathogenicity analysis. Over half of these samples had molecular indications of RTK activation (7/13). Most CNS tumours did not fall into a pre-defined DNA methylation-based subtype classification, suggesting CBL mutation may represent a distinct molecular entity. In vitro modelling of exon 8/9Δ in a neuroblastoma cell line demonstrated that this variant enhances proliferation through RTK activation in other tumour types.
Conclusions: CBL mutation may be a marker of RTK activation and therapeutic target in CNS and solid tumours. This finding will extend the benefit of TKIs, which have proven efficacy in paediatric cancer, to more patients.

Background: CRISPR-Cas9 genome-wide screening has comprehensively mapped the dependency landscape in cancer, including paediatric subtypes. Whilst therapeutic discovery in certain cancers has benefitted from this, other high-risk subtypes, such as MLL-rearranged acute myeloid leukaemia (MLLr-AML), still lack therapeutic options. Existing work in the dependency-mapping space has focused predominantly on the coding genome, ignoring potential novel cancer dependencies in non-coding regions. Long non-coding RNAs (lncRNAs) in particular represent untranslated RNAs that are increasingly implicated in cancer development, including in MLLr-AML. In the current study, we perform both CRISPR-Cas9 screening and Cas13d secondary screening to identify novel lncRNA dependencies in MLLr-AML, with Cas13d as a validation tool that better emulates therapeutic conditions, where genes are often modulated instead of completely suppressed.
Methods: Stable Cas9-positive MLLr-AML cells were generated using lentiviral transduction, with enzymatic activity confirmed through single-target knockdown. In vitro genome-wide CRISPR-Cas9 lncRNA KO screening was performed in three Cas9-positive MLLr-AML cell lines transduced with single guide RNA (sgRNA) libraries at ~500x coverage. Following 21 days in culture, gDNA was isolated and sequenced to observe overall sgRNA representation. lncRNA dependencies were identified using both MAGeCK and CHRONOS analysis softwares. Stable Cas13d-positive MLLr-AML cells were similarly generated for pooled CRISPR-Cas13d secondary screening.
Results: MAGECK identified 5 lncRNA gene dependencies (fold-change < -1.5, FDR < 0.01) overlapping all three cell lines in the CRISPR-Cas9 screen. In CHRONOS, 18 genes showed strong dependency (gene-effect score < -1) across all three cell lines. A custom Cas13d library for secondary screening was designed containing sgRNAs targeting the identified dependencies. Three stable Cas13d-positive MLLr-AML cell lines were generated in preparation. Flow cytometry following transduction with CD45 sgRNAs showed successful knockdown in these lines.
Conclusion: Utilising a genome-wide screening approach we have identified lncRNA dependencies in MLLr-AML which will be further validated utilising a CRISPR-Cas13d secondary screening strategy.

While the use of ctDNA-based minimal residual disease (MRD) monitoring is gaining popularity in adult cancers, this approach is not yet widely adopted in paediatric cancers (PCs). However, this approach is not widely implemented in pediatric cancers. Enabled by the ZERO program, Australian’s national Childhood Cancer Precision Medicine program, we developed a robust and sensitive detection method to monitor ctDNA in PCs. This method overcomes PC-specific challenges including low tumor mutational burden, lacking of hotspot mutations and limited blood from children. Specifically, our assay includes personalized panel design which targets primary tumor mutations from each patient, a fixed panel targeting hotspot mutations, error suppression from duplex deep sequencing and customized software for UMI de-duplexing (REDUX) and tumor fraction estimation (WISP).
We retrospectively analyzed 661 blood or cerebrospinal fluid (CSF) samples from 168 PC patients. We achieved an average detection limit of 0.005% ctDNA fraction and a minimal limit of 0.0001%. We detected ctDNA burdens ranged widely from 0.001% to 100% with brain (CNS) tumors consistently showing a low ctDNA burden (average 0.03%). When patients were identified with high disease burden, our approach identified ctDNA in over 93% of extra-cranial solid tumours, 90% of leukaemia, 89% of sarcoma, and 90% of Neuroblastoma. Our approach showed that ctDNA burden dynamics correlated with disease progression or remission. Importantly, we identified circulating disease in ~20% of clinically complete response assessments, predicting relapse 1–6 months before clinical manifestation. Finally, with the fixed panel, our assay identified emergence of novel variants, suggesting potential new treatment options during tumour evolution. Overall, these results demonstrated that our ctDNA-based disease monitoring approach for pediatric cancer is an important addition to standard of care, quantifying disease burden and tracking disease evolution for precision-guided treatment.

Background: The Zero Childhood Cancer Program (ZERO) is a precision medicine program with currently ~1600 patients enrolled to date, with the program forecasted to have 3500 patients by end of 2025. This poses two data challenges: how to store and analyse billions of germline variants from this cohort, and how to integrate different data sources and types from within and outside the program to interpret analysis outcomes. For these large and complex datasets, we selected a Data Lakehouse architecture for its combination of performance, data-type flexibility and cost-effectiveness.
Method: To address these challenges, we implemented a cloud Data Lakehouse underpinned by the Delta file format and distributed computation engine Spark, on the Databricks platform. Additionally, we created a scalable data pipeline to investigate pharmacogenic interactions with adverse drug reaction (ADR) in ZERO patients to demonstrate the ability to efficiently integrate heterogenous data.
Results: We successfully populated a distributed germline variant database with >1600 whole genomes for query and analysis; this achieved genome range cohort queries for over 8.3 billion variants in under 5 seconds and a 60:1 compression ratio from raw VCF (10TB to 165GB).
We then integrated PharmKGB’s drug-pharmacogenic interactions and ZERO’s ADR dataset with ZERO’s germline database, enabling detection of patients with ADR patients that had pharmacogene alleles related to their ADR-causing drugs. We identified 3 ADR patients out of 34 patient-drug combinations with possible pharmacogene links, one having a well-documented pharmacogene (UGT1A1*93) related to their ADR therapy.
Conclusion: This project demonstrates how the Data Lakehouse architecture can be successfully implemented to meet the scale of the ZERO program’s analytical needs.

Germline alterations in cancer predisposition genes (CPGs) are associated with increased risk of developing certain cancers. Identifying underlying germline genetic contribution to cancer development not only assists in developing a personalised management plan for the patient, but also provides cancer risk clarification and guidance for family. Studies suggest 10-15% of paediatric cancer patients carry germline pathogenic variants in CPGs. However, a proportion of patients with features suggestive of underlying cancer predisposition, have no pathogenic alterations identified in known CPGs. The PREDICT study was conducted to identify germline variants in patients <21 years, newly diagnosed with cancer across NSW, together with biological parents where possible (trio analysis), using whole genome sequencing (WGS). Our team aimed to comprehensively interrogate the genome sequencing data to identify and functionally validate novel germline variants that may contribute to cancer predisposition.
We have established bioinformatics pipelines to analyse PREDICT germline WGS data against a panel of ~1250 cancer-related genes, to investigate genomic alterations within and beyond protein-coding regions, including exonic single nucleotide variants and small insertions/deletions, splicing variants, non-coding region variants (e.g. promoter regions), and structural alterations. Trio-based WGS data was run on The University of Sydney's High Performance Computing system, with analysis incorporating RStudio and online curation tools.
Clinically actionable pathogenic/likely pathogenic variants were detected in 51/230 (22.2%) enrolled patients (e.g. PMS2, NF1), and 62 novel candidate variants were identified in 53/230 (23.0%). Functional validation of prioritised variants-of-interest recurrently identified in B cell leukaemia/lymphoma, is currently being conducted in transduced murine pro-B cell line Ba/F3, which stably express wildtype and each mutant, to assess their functional impacts on the relevant pathway.
Our comprehensive WGS analysis has identified known and novel germline alterations. Further characterisation of novel findings may broaden our understanding regarding underlying mechanisms of paediatric cancer predisposition, and lead to newly defined genotype-phenotype associations.