The IALLNP is a commercially available amplicon-based DNA-sequencing assay that targets 60 genes (the complete coding regions of 17 genes and exonic hotspots in 43 genes) recurrently mutated in lymphoid malignancies. The panel includes two primer pools generating 1,337 amplicons. It detects SNVs and insertions/deletions (indels) but not copy number variation, loss of heterozygosity, structural rearrangements, and aneuploidies.

For assay validation, we used 54 samples, including FFPE lymph-node tissues (N=22), bone-marrow (BM) aspirates (N=22), and peripheral blood (PB; N=10), previously tested using a hybridization capture-based NGS assay [22] at an external accredited laboratory. Additionally, one commercial reference standard, SeraSeq Lymphoma DNA Mutation Mix (0710-2203; Seracare, Milford, MA, USA), was used. The testing indications included a wide spectrum of lymphoid malignancies: CLL, mantle cell lymphoma, DLBCL, marginal zone lymphoma, primary mediastinal B-cell lymphoma, Waldenstrom’s macroglobulinemia, and T-cell lymphomas. The publication of this study data was approved by the Human Research Ethics Committee of St. Vincent’s Hospital (Sydney, Australia).

DNA from BM aspirates and PB was extracted using the EZ1&2 DNA Blood 350 μL Kit on an EZ2 Connect MDX instrument (Qiagen, Hilden, Germany). The neoplastic content in FFPE samples was assessed using hematoxylin and eosin staining. Macro-dissection was performed when necessary to enrich tumor tissue or to exclude poor-quality tissue. FFPE DNA was extracted using the MagMAX FFPE DNA/RNA Ultra Kit on a Kingfisher Duo Prime Magnetic Particle Processor (Thermo Fisher Scientific). Extracted DNA was quantified using the Qubit HS dsDNA Assay Kit (Thermo Fisher Scientific) and diluted to a working concentration.

The Ion Torrent Genexus Integrated Sequencer performs automated barcoded library preparation, templating, and sequencing. Each run can include a maximum of 15 samples and a no-template control (NTC) when utilizing all four lanes on the Ion Torrent GX5 semiconductor chip. The minimum input amount is 27.75 ng gDNA per sample.

Genexus software v6.8.1.1 (and the latest update, 6.8.2.0; Thermo Fisher Scientific) was used to align reads against the GRCh37/hg19 human reference genome using the Torrent Mapping Alignment Program, and Torrent Variant Caller v5.20 was used to evaluate and annotate variants. Variants are described according to Human Genome Variation Society (HGVS) nomenclature v20.05 (manual correction of indels and duplications is often required). The provided hotspot file was expanded via a comprehensive literature search for clinically significant variants in the genes covered by the panel. Optimized variant-finding parameter settings included a minimum cutoff VAF of 2% for indels and 2.5% for hotspots and SNVs; minimum assigned quality score of 25 for indels and 10 for hotspots and SNVs; and maximum strand bias tolerance of 0.9 for SNVs, 0.85 for indels, and 0.96 for hotspots. A customized filter chain was created using the Genexus software to include variants that 1) were included in the Oncomine Variant Annotation Database; 2) were listed as pathogenic or likely pathogenic in ClinVar; or 3) remained present after the exclusion of University of California, Santa Cruz common single-nucleotide polymorphisms, non-exonic (allowing a 5-bp splicing region) and synonymous variants (variant effect settings were restricted to: missense, nonframeshiftInsertion, nonframeshiftDeletion, nonframeshiftBlockSubstitution, nonsense, stoploss, frameshiftInsertion, frameshiftDeletion, and frameshiftBlockSubstitution), and variants present in population databases at a frequency >10⁻⁴.

Candidate variants were visually inspected using Integrative Genomics Viewer (IGV) v2.16 (Broad Institute, Cambridge, MA, USA) and manually annotated using the Catalogue of Somatic Mutations in Cancer, Genome Aggregation Database v4.0, The TP53 Database (when relevant), UniProt, and literature review [23–26]. Clinically significant variants were defined as those having an impact on patient diagnosis, prognosis, and/or treatment (predicting therapeutic response or resistance) according to a modified version of the College of American Pathologists/Association for Molecular Pathology joint consensus guidelines [27]. Variants of uncertain clinical significance are reported, whereas variants deemed benign or likely benign are not. Concordance analysis was limited to genomic regions covered by both the IALLNP and the externally accredited laboratory panel. Data were analyzed using Microsoft Office Excel software v2411 (Microsoft Corporation, Redmond, WA, USA).

Fifteen validation sequencing runs were performed. Analysis of the initial runs revealed that some amplicons, often in repetitive and GC-rich areas, consistently underperformed (mean depth consistently <100×). These amplicons were omitted from the target region browser extensible display file as the consistent lack of coverage compromised variant calling reliability in these regions. Moreover, the omitted amplicons were confirmed to be devoid of any recurrent clinically significant variants (except for the minor hotspot, BCL2 G47). The final panel coverage is provided in Table 1 (excluded regions are listed in Supplemental Data Table S1). With these improvements, the mean overall run metrics were: Ion Sphere Particle loading, 91.70% (range, 85.18%–94.81%); enrichment, 98.94% (97.47%–99.76%); final reads, 45.18% (38.85%–50.00%); polyclonality, 43.17% (39.08%–47.96%); and raw read accuracy, 98.06% (97.68%–98.38%). The mean overall sample metrics were: mean read length, 122.7 bp; mapped reads, 3,178,028; mean depth, 2,494×; and uniformity, 94%. FFPE samples had a lower mean read length (6.7 bp on average), similar mean depth (2,482× vs. 2,503×), and 1.1% lower uniformity than PB/BM samples (Supplemental Data Table S2). The average NTC metrics were: mean read length, 45.5 bp and mean depth, 9.4×. For the retained 1,319 amplicons, an average of 96.57% targeted bases were covered at 350× depth, and 99.27% at 100×. The most common amplicons not reaching 350× are detailed in Supplemental Data Table S3. The QC metric values are provided in Table 2.

The external laboratory identified 163 variants across the 54 validation samples, 15 of which were in regions not covered by the IALLNP, resulting in 148 candidate variants. The IALLNP detected 182 variants, representing 133 of 148 of the candidate variants, and 49 that were in regions not covered by the external reference laboratory (with no orthogonal option available for validation). These 182 variants included 136 missense, 14 splicing, and 32 indel variants. The commercial reference standard included 20 assessable variants, all of which were detected by the panel. Overall, 153 of 168 (91%) assessable variants were detected (Supplemental Data Table S2). Six false-negative calls were variants with VAFs of 3%–4.9%, and nine with VAFs of <3%. Notably, the IALLNP does not incorporate unique molecular identifiers. The detection rate was 135/135 (100%) for variants with VAF ≥5%; 16/22 (73%) for variants with VAF 3%–4.9%, and 2/11 (18%) for variants with VAF <3%. The Pearson correlation coefficient between IALLNP and reference laboratory VAFs was R²=0.9347 (Fig. 1).

The limit of detection was determined using the SeraSeq Lymphoma DNA Mutation Mix diluted in a background of wild-type DNA. At expected VAFs of 3%–5% (1:1 dilution), the assay correctly detected 20 out of 20 variants, including 18 SNVs and two indels, twice (Table 3). At 1:2 dilution (expected VAFs, 2.5%–3.5%), all 20 variants were present; however, 5 of 20 were low-confidence calls held back by the filter settings. Combined with the above accuracy data, the lower limit of detection was established to be 5% VAF (100% sensitivity).

The original IALLNP was provided without a sequence variant baseline (SVB) file, resulting in considerable numbers of false-positive calls. A baseline Oncomine Tumor Specific Panel-derived SVB file has since been provided. Unresolved recurrent false positives were identified and confirmed via manual interrogation using IGV. This required assessing each variant call against the sequencing results of other samples in the same run to detect sequencing artifacts (Supplemental Data Fig. S1). These artifacts were subsequently optimized bioinformatically (Supplemental Data Table S4). Following these improvements, the mean per-sample number of false-positive variant calls was 0.4. Five samples known to carry no variants had zero false-positive calls. The persisting false-positive variant calls were all easily identifiable as artifacts upon review in IGV and excluded.

To determine reproducibility, nine clinical samples were analyzed across two runs, and three samples were analyzed in duplicate over two runs. In total, 77 of 83 variants were reliably detected (92.8%). The six nonreproducible calls all had VAFs <5% (Supplemental Data Table S5). The SeraSeq standard was also analyzed using four separate runs, including twice in duplicate. Given the 20 assessable variants and six occasions, 119/120 variants (99.2%) were called reproducibly. BCL2 NM_000633.3: c.302G>C was missed on one occasion in a GC-rich area known to be challenging.

We used NGS workflows existing in our laboratory for IALLNP validation. Microtomy, hematoxylin and eosin staining, and tumor slide assessment took approximately 2 hrs per batch of FFPE samples (5–10 samples). DNA extraction required 3–4 hrs for FFPE samples and 1 hr for fresh PB and BM samples. Fluorometric quantification, sample dilution, plate preparation, and Genexus instrument loading took approximately 2 hrs of hands-on time. The runtime on the instrument is 19 hrs for an eight-sample run using two lanes on the GX5 chip.