This study was approved by Hospital Israelita Albert Einstein (HIAE) Institutional Review Board CAAE: 72996523.0.0000.0071; # 6.541.648. Informed consent was obtained from all patients who provided authorization for the use of their molecular data through an online consent form.

To validate the TSO500 system (Illumina, San Diego, CA, USA), we tested 80 samples for accuracy, two samples of DNA and RNA for reproducibility, and one sample in multiple runs to assess the limit of detection. Specifically, 46 DNA and 34 RNA samples were selected based on the criteria outlined in Figure 1S, Supplementary Material. For CGP and analysis of real-world data, 454 patients were included. Different tumor types containing at least 20% of the tumoral area in the histological slides were analyzed.

Nucleic acid extraction was performed using All Prep (QIAGEN, Hilden, Germany), QiaAmp FFPE (QIAGEN), or ReliaPrep (Promega, Madison, WI, USA) kits, and the samples were stored at -80°C until use. After nucleic acid quantification, quality was evaluated using Tapestation (Agilent Technologies, Santa Clara, CA, USA) equipment with the High Sensitivity D1000 Screen Tape and RNA Screen Tape reagents for DNA and RNA, respectively.

Libraries were prepared according to the TruSight Oncology 500 Reference Guide. First, the DNA samples were fragmented using the ME220 system (Covaris, Woburn, MA, USA). In summary, fragmented DNA and cDNA from RNA samples were indexed and underwent two cycles of hybridization and target region capture. After normalization, the samples were quantified and pooled. Sequencing was then performed using the Illumina NextSeq platform, and the resulting data were processed using the bioinformatics pipeline outlined below.

An in-house pipeline was developed based on the default configuration outcomes of the TruSight Oncology (TSO) pipeline v2.2.0 for downstream analysis. For single nucleotide polymorphism (SNP) and insertion-deletion (indel) analyses, stitched realigned BAM files were used as inputs for SNP calling with Freebayes v1.0.2 and Mutect2 from GATK v4.1.5.0, employing a minimum base quality score of 30 and a callable depth of 3. Variants identified by Pisces (the TSO standard variant caller), Freebayes, and Mutect2 were merged and annotated using ANNOVAR (version 2019Oct24). The annotated files were filtered based on a predefined list of target genes.

To assess the presence of MSI, the TSO pipeline evaluates 130 repeat loci;⁽⁶⁾ if a minimum of 40 microsatellites were evaluable, MSI analysis was performed using PercentageUnstableSites. Microsatellite instability was considered positive if at least 20% of the microsatellites were unstable. For TMB, if the value of adjusted TmbPerMb was >10 mutations/Mb, TMB was considered high (TMB-H).

Copy number variants (CNVs) were analyzed from the output of the TSO application. First, the VCF files were annotated using AnnotSV v2.3.2, and the variants were filtered based on a targeted gene list for further evaluation. To complement this analysis, particularly for deletion events in the CDKN2A/B genes, a pipeline based on the GATK4 somatic workflow v1.4.0 was implemented. A negative control tumor panel (NCTP) comprising 83 tumor samples with minimal CNVs was established to validate the detection of CDKN2A/B deletions.

For RNA analysis, fusions were evaluated using a combined approach that included the VCF output from the TSO pipeline and Arriba v.2.3.0. A list of hotspot fusions was used to filter and highlight the relevant clinical events. RNA splicing variant outputs in the VCF files from the TSO pipeline were also annotated using ANNOVAR and filtered based on the same list of targeted genes used for CNV analysis.

The TSO500 assay identifies 523 genes and 55 RNA transcripts. For clinical purposes, two reporting gene sets were used: an initial panel of 159 genes (v1) and an expanded panel implemented in June 2023 (v2), which retained all v1 genes and included other clinically relevant targets, totaling 351 genes (Table 1S, Supplementary Material), such as those with diagnostic, prognostic, and therapeutic potential, and genes associated with hereditary cancer predisposition syndromes.

For artifact exclusion, we developed a metric that evaluated the occurrence of each variant among all sequenced samples and the allele frequency distribution of the previously detected variant. Variants frequently detected in the historic cohort within the allele frequency range typical of artifacts were excluded. Cancer hotspots were not included in this filtering and were always flagged for detailed evaluation. Variants considered real with an allele frequency above 5% were included in the analysis.

Variant classification was performed based on the Variant Interpretation for Cancer Consortium (VICC) criteria.⁽⁷⁾ Variants were classified as benign, likely benign, variants of uncertain significance (VUS), likely oncogenic, or oncogenic. Specialized literature and public databases, including GNOMAD, ClinVar, Franklin, and cBioPortal were used for variant classification. Gene fusions were analyzed using the IGV program. If both genes had at least 20 supporting reads, including split reads and unmatched pairs, the fusion was considered true. Figure 2S, Supplementary Material shows a diagram of the TSO500 workflow from sample extraction to results reporting.

Statistical analyses were performed to evaluate the technical validation of the TSO500 and the molecular characterization of the Brazilian cohort of patients with solid tumors.

The performance metrics for the analytical validation were calculated using the FoundationOne test as a reference. Variant concordance between the TSO500 panel and F1 test was assessed using the percentage of shared variants.

The frequencies of genomic alterations were calculated according to variant type (SNVs, indels, gene fusions, and CNVs) and the affected genes. The proportion of variants classified as oncogenic, likely oncogenic, or VUS was determined. For the TMB and MSI analyses, measures of central tendency (mean, median), dispersion (standard deviation), and proportion of high-level cases (e.g., TMB ≥10 mut/Mb) were calculated. Data distributions were evaluated across tumor types to explore potential associations between genomic burden and histological subtypes.

The TSO500 test (Illumina) was validated in our laboratory following the CAP guidelines. The validation process included accuracy, intra- and inter-assay reproducibility, and definition of limit of detection. The evaluated parameters included DNA variant detection (SNV and Indels), TMB, MSI, CNVs, and gene fusion detection (RNA-based). The panel was validated using paraffin-embedded tissue samples obtained from different sites (Table 2S, Supplementary Material).

Quality control was performed by evaluating the minimum number of target reads, average fragment size, median absolute deviation for CNVs, uniformity, usable sites for MSI, contamination score, and average read coverage according to the manufacturer’s recommendations. Additionally, bioinformatics analyses were performed to validate the pipeline developed by the laboratory using the VarStation platform.

To assess the accuracy of somatic variant identification, TMB, and MSI, we used 46 DNA samples and compared their results with those of the FoundationOne test, which is considered the gold standard.

With the TSO500 assay, we identified a total of 305 DNA variants. Among them, 212 variants were detected by FoundationOne, resulting in an overall concordance of 88.70%. Regarding discrepancies, 81 variants detected by TSO500 were not reported by FoundationOne: 4 were in genes not covered by the FoundationOne panel (ERCC2 and PHOX2B), whereas 77 were confirmed upon IGV inspection. In contrast, 15 variants reported by FoundationOne were not detected by TSO500, primarily owing to their variant allele frequency (VAF) being below 5%, high population frequency, or absence in IGV inspection. Additionally, 11 variants detected by both assays were classified as benign by our team, and 1 variant was considered an artifact.

Performance analysis of the TSO500 assay revealed a sensitivity of 93.39%, indicating a good ability to detect variants identified by the FoundationOne assay. Additionally, the positive predictive value was calculated to be 73.36%, reflecting the proportion of true positive variants among all those reported as positive by the TSO500.

To classify the 212 variants detected by both assays, we compared the designation of the variants as either VUS or oncogenic/probably oncogenic. Among these, we observed three discrepancies: one variant classified as oncogenic by FoundationOne was classified as a VUS by our team (NM_015125.4:c.310G>A, p.Glu104Lys), whereas two variants classified as VUS by FoundationOne were categorized as oncogenic/probably oncogenic by our team (NM_000455.4:c.526G>A, p.Asp176Asn and NM_020975.6:c.2410G>T, p.Val804Leu). This resulted in 98.60% correlation between the two assays for variant classification.

Tumor mutational burden accuracy was assessed in two ways: as a continuous variable compared with correlation among all cases, resulting in a correlation coefficient (R) of 0.91, and when using the FDA-recognized cutoff of 10 mutations per megabase (mut/Mb). The accuracy of classifying cases above or below the cutoff was 86% compared to that of FoundationOne (R=0.86) (Figure 1).

Figure 1

For the fusion analysis, 34 RNA samples were used and compared with previous results obtained from FoundationOne CDx (Foundation Medicine) or Oncomine (ThermoFisher). Additionally, a commercial control, SeraSeq RNA Fusion (SeraCare), was used to assess the accuracy of fusion detection. The correlation achieved for fusion detection was 96%.

To evaluate gene amplification (CNVs), 46 samples with previous results (20 positive and 26 negative) from the FoundationOne CDx Test (Foundation Medicine) were used. Amplifications in 18 samples were confirmed using TSO500, resulting in a sensitivity of 92.30%. A specificity of 99.80% was achieved when we used a copy number cutoff ≥6. If the values of the fold change were ≥2.1 and <2.3 and the copy numbers were ≥5.1 and <6, amplification was considered suspicious.

To validate deletion detection with TSO500, we focused on the genes CDKN2A and CDKN2B as they are located in close proximity, and deletions in this region frequently span several hundred nucleotides. Two groups of samples were used: an NCTP consisting of 83 tumor samples (TC > 60%) with no CNVs detected in prior TSO500 runs, and 48 samples used for accuracy assessment, which already had a previous result using FoundationOne, of which 39 were negative and 9 were positive for homozygous deletions in CDKN2A/B. The overall accuracy of TSO500 was 93.70%, with 3 out of the 48 samples showing discordant results with FoundationOne (1 ≤ copy number ≤ 1.35 or 0.7 ≤ copy ratio ≤ 0.85).

To assess the reproducibility of the TSO500 test, two DNA samples and two RNA samples were tested in five inter- and intra-assay replicates. We obtained 100% reproducibility for all replicates.

To establish the analytical sensitivity of the assay, we assessed its limit of detection (LoD) using a sample containing an MEN1 variant (NM_000244: exon10:c.1534G>A, p.G512S) with a VAF of 5%. This variant was reliably detected across all three replicates, with VAFs ranging from 0.058 to 0.09, confirming an LoD of 5%. This consistency in detection underscores the sensitivity of the test at lower allele frequencies, which is critical for the accurate identification of variants present at low levels in heterogeneous samples.

Comprehensive Genomic Profiling was performed on 454 samples using TSO500. Although the TSO500 assay assessed a total of 578 genes, we focused our gene analysis on 250 genes considered clinically relevant. Oncogenic/likely oncogenic mutations were identified in 131 genes and VUS in 148. The 30 genes with the highest number of oncogenic variants were TP53, KRAS, APC, and EGFR (Figure 2).

Figure 2

Among the 454 patients with CGP results, 37 did not present with any oncogenic or likely oncogenic variants (Table 3S, Supplementary Material), However, in the 417 positive cases, we identified 1,188 oncogenic alteration events and 1,630 VUS across all variant types, including SNVs, indels, fusions, and amplifications. Overall, 57.85% of the detected variants were VUS, and 42.15% were oncogenic (Table 4S, Supplementary Material), with an average of 2.61 oncogenic variants and 3.10 VUS per case. Analysis of the distribution of variant types among cases showed that the most common variants were SNVs, followed by indels, CNV, and gene fusions (Figure 3S, Supplementary Material).

We profiled 57 different tumor types, the most common being lung (66 cases), colorectal (52 cases), and pancreatic adenocarcinoma (45 cases). Among the total of 57 tumor types, 40 less frequent types were represented in figure 3 as “Others,” as they each occurred in only 1 to 7 cases.

Figure 3

The average TMB was 7.47 mut/Mb, ranging from 0.78 to 85.6 mut/Mb. TMB was reported in 424 cases, with 72 cases presenting levels above 10 mut/Mb, indicating that 16.98% of the analyzed cases were TMB-H.

Figure 4 illustrates the 13 tumor types that had a significant representation of TMB cases, with melanoma showing the greatest variation. The median TMB varied significantly among tumor types, a common phenomenon that plays a crucial role in influencing the potential response to immune checkpoint inhibitors.⁽⁸⁾

Figure 4

The presence of numerous outliers in tumors such as sarcoma, lung cancer, and colorectal adenocarcinoma suggests that there are individual cases within these tumor types with exceptionally high TMB, which could be particularly relevant for therapeutic decisions.

Regarding microsatellite analysis, only seven cases exhibited MSI, defined as having 20% or more unstable sites; all these MSI cases also exhibited TMB-H. The median TMB was 61.3 and 4.7 mut/Mb in the MSI and microsatellite stability (MSS) cases, respectively. The correlation between TMB and MSI status in 424 patients with TMB>0 is shown in figure 5. Interestingly, all cases with MSS had <10% unstable microsatellites, with one exception: a case of a soft tissue sarcoma with 12.63% unstable microsatellites and 13.4 mut/Mb, which was classified as MSS. Nevertheless, this patient had immunohistochemistry results showing a loss of nuclear expression of the DNA repair system enzymes MLH-1 and PMS2, and had a clinical response to immunotherapy.⁽⁹⁾

Figure 5

A total of 66 gene fusions were identified in 59 of the 454 cases, representing 13% of the samples analyzed. Among them, 55 were unique. Of these, 25 were classified as VUS and 30 were deemed oncogenic (Figure 4S, Supplementary Material). Nineteen of the oncogenic fusions were canonical, as they are well documented and characterized in the scientific literature, as represented in orange in Figure 4S, Supplementary Material and detailed in table 1.

Table 1

Canonical oncogenic fusions
SLC45A3::ERG	Prostate adenocarcinoma
TMPRSS2::ERG	Prostate adenocarcinoma (8)
SND1::BRAF	Prostate adenocarcinoma
TMPRSS2::ETV4	Prostate adenocarcinoma
KIF5B:: RET	Lung adenocarcinoma
CD74::ROS1	Lung adenocarcinoma
CD74:: NRG1	Lung adenocarcinoma
ALK::EML4	Lung adenocarcinoma
SDC4::ROS1	Lung adenocarcinoma
MET (exon 14 skipping)	Lung adenocarcinoma (2)
KIAA1549::BRAF	Glioma (5)
FGFR2::SHTN1	Glioma
SEC 61G::EGFR	Glioma
EWSR1::WT1	Sarcoma
SCD5::ALK	Sarcoma
LMNA::NTRK1	Sarcoma
PAX3::FOXO1	Sarcoma
PAX3::NCOA2	Sarcoma
GOPC::ROS1	Colorectal adenocarcinoma