Peter Tsai

Approximately 200 critically ill infants and children in New Zealand are in high-dependency care, many suspected of having genetic conditions, requiring scalable genomic testing. We adopted an acute care genomics protocol from an accredited laboratory and established a clinical pipeline using Oxford Nanopore Technologies PromethION 2 solo system and Fabric GEM™ software. Benchmarking of the pipeline was performed using Global Alliance for Genomics and Health benchmarking tools and Genome in a Bottle samples (HG002-HG007). Evaluation of single nucleotide variants resulted in a precision and recall of 0.997 and 0.992, respectively. Small indel identification approached a precision of 0.922 and recall of 0.838. Large genomic variations from Coriell Copy Number Variation Reference Panel 1 were reliably detected with ~2 M long reads. Finally, we present results obtained from fourteen trio samples, ten of which were processed in parallel with a clinically accredited short-read rapid genomic testing pipeline (Newborn Genomics Programme; NCT06081075; 2023-10-12).

Approximately 200 critically ill infants and children in New Zealand are in high-dependency neonatal/paediatric acute care at any given time, many with suspected genetic conditions, necessitating a scalable distributed solution for rapid genomic testing. We adopt the existing acute care genomics protocol of an accredited laboratory and established an expandable acute care clinical pipeline based around the Oxford Nanopore Technologies PromethION 2 solo system connected to a Bayesian AI-based clinical decision support tool (Fabric GEM™ software). In the establishment phase, we performed benchmarking using Global Alliance for Genomics and Health (GA4GH) benchmarking tools and Genome in a Bottle samples HG002-HG007. We evaluated single nucleotide variants (SNVs) and small insertions-deletions (indels) calls and achieved SNV precision and recall of 0.997 ± 0.0006 and 0.992 ± 0.001, respectively. Small indel identification approached a precision of 0.922 ± 0.019 and recall of 0.838 ± 0.043. Rarefaction analyses demonstrated that SNV identification plateaus at ∼20X coverage, while small indels plateaus at ∼40X coverage. Large genomic variations from Coriell Copy Number Variation Reference Panel 1 (CNVPANEL01) were reliably detected with ∼2M long reads. Finally, we present results obtained from ten trio samples that were processed through the pipeline validation phase, averaging a 5-day turnaround time, conducted in parallel with a clinically accredited short-read rapid genomic testing pipeline.

Supplementary Figure S5: Inference of putative sequences of genomic changes consistent with the data using the LICHeE and REVOLVER methods LICHeE and Revolver were used to further speculate the order of genomic changes in this patient's dataset. ClonEvol was also trialed, using PyClone putative clonal clusters as input, however the results did not converge. A, LICHeE was run according to the tool creator's instructions, using VAFs as input, to generate clonal relationships of tumor samples. Numbers inside colored circles represent the number of mutations defining a clone, and the squares represent the clonal structures of individual samples. Shaded regions indicate the proportion of cells belonging to that clone, where the white regions represent normal cells. Four dominant tumor groups were apparent. Where two or more colors are present, LICHeE has predicted the sample to contain a mixture of clones (most apparent in Pa1 and Lu1 but also present in In2, Pa6 and others). This feature is also visible in the VAF plot (Figure 5a) but not represented on the DNA phylogram (Figure 5b). Technical reasons prevented the inclusion of some samples, e.g. Lu8 and Pa5. B, Clonal relationships between tumor samples as predicted by REVOLVER, using PyClone putative clonal clusters as input (from VAF and chromosomal copy number), and default tool parameters. Clones are labeled according to one variant defining that clone (e.g., MASTL), and the circle size is proportional to the number of variants defining the clone. Number refers to clone number. REVOLVER highlighted the same pattern of progression seen in other analyses: from normal cells, common variants accumulate (labeled as MASTL in Figure 5b), and a further set of variants shared by most tumors was identified (labeled GPAM) before a split into two dominant tumor groups, each characterized by their own set of variants (labeled SLIT1 and PAQR6 respectively). REVOLVER highlights the progression of variant accumulation across all samples, rather than indicating individual samples. All coloring consistent with Figure 5. Overall, LICHeE and REVOLVER explore the clonal and subclonal structure of individual tumor samples and suggest the same sequence of genomic progression evident in the DNA phylogram (Figure 5b). None of these cancer-specific methods attempts to time the evolutionary divergence events.

Supplementary Figure S3: Mutational signatures across all samples collected at autopsy A, Overall mutational burden does not reveal significant mutational signatures (MuSiCa). The signature contribution is indicated by the degree of shading (key to right). The signature with the highest contribution was Signature 3, however it was not statistically significant in most samples according to tool Signal. B, The scarHRD method for quantifying homologous recombination deficiency did not identify any tumors to be HR-deficient; with all having an HRD-sum score < 20 (known HR-deficient tumors usually have HRD-sums of > 40)28.

Supplementary Figure S3: Mutational signatures across all samples collected at autopsy A, Overall mutational burden does not reveal significant mutational signatures (MuSiCa). The signature contribution is indicated by the degree of shading (key to right). The signature with the highest contribution was Signature 3, however it was not statistically significant in most samples according to tool Signal. B, The scarHRD method for quantifying homologous recombination deficiency did not identify any tumors to be HR-deficient; with all having an HRD-sum score < 20 (known HR-deficient tumors usually have HRD-sums of > 40)28.

Supplementary Figure S4: Absence of ARID1A variant in targeted panel sequencing of Biopsy 1 despite adequate depth Top panel shows Biopsy 1 targeted panel sequencing. Bottom panel shows representative tumor sampled at autopsy (Sc6) with heterozygous deletion in ARID1A, clearly absent from Biopsy 1 sequencing despite adequate depth (7240 unique molecules). The probability of not sampling the ARID1A variant in Biopsy 1 due to chance alone was calculated assuming tumor cellularity of 80% and the heterozygous ARID1A variant being present in 50% of tumor reads. 7240 unique molecules covered this genomic position. The binomial distribution in R was used to calculate the probability of not sampling this variant: dbinom(7240, size=7240, prob=0.6).

Supplementary Figure S4: Absence of ARID1A variant in targeted panel sequencing of Biopsy 1 despite adequate depth Top panel shows Biopsy 1 targeted panel sequencing. Bottom panel shows representative tumor sampled at autopsy (Sc6) with heterozygous deletion in ARID1A, clearly absent from Biopsy 1 sequencing despite adequate depth (7240 unique molecules). The probability of not sampling the ARID1A variant in Biopsy 1 due to chance alone was calculated assuming tumor cellularity of 80% and the heterozygous ARID1A variant being present in 50% of tumor reads. 7240 unique molecules covered this genomic position. The binomial distribution in R was used to calculate the probability of not sampling this variant: dbinom(7240, size=7240, prob=0.6).

Supplementary Table S1 includes a description of the tumour samples analysed.Supplementary Table S2 includes DNA WES summary statistics.Supplementary Table S3 includes RNA-Seq summary statistics.Supplementary Table S4 includes WGS linked-read summary statistics.Supplementary Table S5 includes regions covered in the custom ThermoFisher Scientific Ampliseq HD panel. Supplementary Table S6 includes targeted DNA sequencing custom panel summary statistics. Supplementary Table S7 includes low-coverage DNA WES summary statistics.Supplementary Table S8 includes gene set enrichment of genes lost from Chr 11.Supplementary Table S9 includes variants identified in two or more tumour sites.Supplementary Table S10 includes the consequence of somatic variants in Biopsy 1.

Supplementary Figure S6: ctDNA detection of private variants While only four private variants were reported by the sequencing platform’s IonReporter software (indicated in orange, key top right), there were sequencing reads to support other private variants. Private variants unique to samples La1 (3 variants) and Pa6 (1 variant) were supported by 2 to 3 variant molecules, and there was a single variant molecule detected to support private variant detection in the following samples: Br2, Cr1, In2, Ki2, La1, Lh1, Lh2, Lu6, Lu7, Lu9, Pa2, Pa3, Pa6, Sc7, Th1, Th2, Ut1, Ve2, Ve3.

Supplementary Figure S2: Breakpoints in 10x Loupe software linked-reads view.Each bar represents a sequencing read, and those joined with a horizontal line share the same barcode.Reads are grouped by haplotype (green and purple), with unphased reads in grey. Vertical orange lines indicate putative breakpoints defined by the software. Given that the maximum region of the chromosome phased was significantly shorter than the length of Chr 11 (as indicated in the ‘phase block view’ plot), some breakpoints are assigned to different haplotypes, therefore haplotypes should not be compared between regions (the breakpoint appears in purple in some regions and in green in others).The copy number patterns are consistent with all breakpoints occurring on the same copy of the chromosome.