Unstained FFPE slides were reviewed and marked by a molecular pathologist to indicate the affected tumor area and cellularity for subsequent procedures. Total amount of tumor tissue used for DNA extraction was 0.25 cm² or less in some cases. Tumor tissue was scraped into a 0.2-ml PCR tube. DNA purification was performed using Ion AmpliSeq™ Direct FFPE DNA Kit (ThermoFisher Scientific) according to the manufacturer’s protocol. Deparaffinization of slides was not required and only done in cases where there was little to no tissue visible on slides. In these cases, deparaffinization was done using 100% xylene followed by 100% ethanol. DNA was quantitated using the Invitrogen Qubit 3.0 Fluorometer and Invitrogen Qubit dsDNA HS Assay Kit.

Libraries were prepared with 20 ng of genomic DNA and constructed by automated library preparation using the Ion Chef™ System and Ion Ampliseq™ Cancer Hotspot Panel v2 (Life Technologies). This panel included hotspot regions, including ~2,800 COSMIC mutations of 50 oncogenes and tumor suppressor genes with known implications in diagnosis, prognosis, and therapeutic decision-making. Regions covered and average coverage for this targeted panel are described in Table S1 . DNA libraries of eight samples were combined into one library, which was then diluted to a concentration of 30 pmol/L. The diluted pooled DNA library was used for template preparation and chip loading on the Ion Chef system using Ion 520 chips, followed by sequencing on either the Ion PGM™ System or Ion S5™ Sequencer (ThermoFisher Scientific). Parameters used for assessing run quality included key signal >30, ISP loading >30%, and usable reads >30%. Parameters used for assessing sample quality included mean depth of coverage >1,000× and uniformity >90%. In some cases, samples were assessed at >500× mean coverage with adjusted cutoffs for variant reporting (VAF 15% with 500× coverage).

Validation of analysis pipeline conforms to the recommendations from the Association for Molecular Pathology and the College of American Pathologists (20). We are using the Torrent Suite Software on the Torrent Server for automated sequencing data alignment and analysis. This process uses the Torrent Mapping Alignment Program (TMAP), which is specifically optimized for Ion Torrent data. Base calling, alignments, and run quality control were performed using the Ion Torrent Suite™ Software v5.8.0. Variant calling was performed by Torrent Suite™ Variant Caller plugin using standard settings. BAM and VCF files for variants were imported into Geneticist Assistant version 1.4.2 (SoftGenetics) for sample quality control assessment (minimum base coverage and mean amplicon coverage) and for databasing. As part of our analytical pipeline, we evaluate quality control by average read depth (>1,000×) and uniformity (>90%). Average coverage values for each region is listed in Supplementary Table 1 . Beyond the quality review of each region, variant quality is also evaluated with adapted guidelines proposed by Petrackova et al. (21). Genetic variants with 5% or greater variant frequency and minimum coverage of 250 were analyzed. Variant assessments were done by genome analysts (primary review) and a clinical molecular geneticist or a molecular pathologist (secondary review). Variants were classified into four tiers (Tier I to Tier IV) based on the consensus guidelines set by the Association for Molecular Pathology, the American Society of Clinical Oncology, and the College of American Pathologists (22). Variants classified as Tier I and II were of strong clinical significance, Tier III variants were with unknown clinical significance due to lack of significant evidence, and Tier IV variants were benign or likely benign. Sequence pileups for reportable variants were manually assessed to ensure no miscalls. Variants classified as Tier I/II (described as clinically relevant variant) were reported to the oncologist/physician for each sample.

NGS assay validation was performed using a retrospective set of tumor samples previously tested by other methods like allele-specific PCR with fluorescent hydrolysis probe detection (Enterogen) and sanger sequencing in our laboratory, and the limit of detection (LOD) was assessed by testing serially diluted DNA samples positive for known variants ( Supplementary File 1 ).

The NGS assay was clinically tested and validated by using a commercially available reference standard from Horizon Diagnostics (HDx Perkin Elmar), with known variant allele frequencies and a cohort of 39 retrospective tumor DNA samples selected from the laboratory archive, which were previously tested by other validated methods in our laboratory including allele-specific PCR with fluorescent hydrolysis probe detection (Enterogen) and sanger sequencing. These include 27 positive samples, each carrying at least one Tier I/II variant and 12 negative samples with no Tier I/II variants ( Table S1 ). All the positive control samples were concordant by the NGS assay and no Tier I/II variants were detected in the negative controls. Therefore, the sensitivity and specificity of this assay for the previously assessed variants were determined to be 100%. Repeat sequencing runs were performed to assess the reproducibility and were estimated to be 100%. By testing serial dilutions of tumor DNA samples with Tier I/II variants (BRAF V600E, KRAS G12C, and EGFR exon 19 deletion p.L747_E749del), the LOD for accurate and reproducible variant calling was determined to be 5% of variant allele frequency.

A total of 1,375 NSCLC samples were tested by the lung NGS subpanel, namely, 790 female samples (mean age: 69.6 years; range: 24.8 to 91.8) and 585 male samples (mean age: 69.7 years; range: 36 to 93.8 years) and 952 (69%) tested positive for reportable (Tier I or II) variants. Tier I/II variants identified are listed in Supplementary Table 2 . The detection rates in male and female samples were 65.1% and 72.2%, respectively. Variant detection rates in specific age groups are listed in Table 2 . KRAS was found to be the most commonly mutated gene (41.1%), followed by TP53 (24.2%) and EGFR (11.2%) ( Figure 1A ). A total of 731 samples had one variant, 205 had two variants, and 16 had three variants. Co-occurring variants are shown in Figure 1B . Out of 154 samples with an EGFR variant, 69 (42%) were exon 19 deletions, 42 (26%) were exon 21 L858R, 17 (10%) were exon 20 insertions, 11 (7%) were exon 18 G719S, 6 (4%) were exon 21 L861Q, 7 (4%) were T790M/L858R, 2 (1%) were T790M/Exon 19 del, and 9 (6%) were rare EGFR variants ( Figure 1C ). An interesting point noted is the high frequency of KRAS G12C mutations (15.2%; 209/1,375) in our study cohort, which was higher than the frequency of EGFR mutations (11.2%).

Out of 692 colorectal cancer samples (mean age:65.4 years; 404 male samples and 288 female samples) tested by the colon subpanel, 424 (61.2%) had a Tier I/II variant and 6 had variants in two genes ( Supplementary Table 3 ). A total of 322 (46.5%) had a KRAS variant, 82 (11.8%) had a BRAF variant, and 26 (3.7%) had a NRAS variant ( Table 3 ; Figure 2A ). The most common KRAS variants were in codon 12 followed by codons 13, 146, and 61. Rare clinically relevant variants were detected in six patients. The most frequent BRAF variant was Val600Glu (V600E), which occurred in 72 samples; 10.4% of all patients had colon cancer.

Out of 462 melanoma samples, 322 (69.7%) were positive for Tier I/II variant and 8 patients had variants in two genes (5 patients with BRAF and NRAS; 3 patients with BRAF and KIT). The highest frequency of variants was observed in the BRAF (42.2%) gene followed by NRAS (23.4%) and KIT (2.33%). The most common BRAF gene variants were in codon Val600 and NRAS codon 61 ( Table 4 ; Figure 2B ). All sample details and variants are listed in Supplementary File 4 .

Out of 135 brain tumor samples tested, 28 (20.7%) had one clinically relevant variant, 11 (8.1%) had a BRAF Val600Glu variant, 14 (10.3%) had an IDH1 Arg132 variant, and 3 (2.2%) had an IDH2 Arg172 variant ( Supplementary File 5 ).

A total of 500 tumor samples from patients presenting with various cancer types were tested by a comprehensive panel with 50 genes. Two hundred ninety samples (58%) were tested positive for at least one clinically relevant Tier I/II variant. Among these samples, 82 (16.4%) had two variants and 30 (6%) had three or more variants ( Supplementary File 6 ). Overall, TP53 (23%), KRAS (18.4%), and PIK3CA (9.4%) were among the most frequently mutated genes, followed by APC, BRAF, PEN, RET, NRAS, IDH1, SMAD4, AKT1, and CDKN2A with a detection rate ranging from 1.4% to 3.8% ( Figure 3 ; Supplementary File 6 ). Other less frequently mutated genes were FBXW7, PDGFRA, RB1, ATM, ERBB2, KIT, GNAS, HRAS, FGFR2, Flt3, MET, NOTCH1, STK11, VHL, and SMARCB1 with ≤1% detection rate ( Table 5 ). Oncoprint plots ( Figure 4 ) shows overview of alterations in each sample with in common tumor types analyzed in this study.

Breast: Out of 77 breast tumor samples, 45 (58.4%) had a Tier I/II variant with a high frequency of variants observed in PIK3CA (32.4%) and TP53 (22%) genes. Variants in more than one gene were detected in 13 patients and 7 of these patients had variants in both TP53 and PIK3CA gene ( Figure 4 ).

Colorectal: Out of 67 colorectal tumor samples, 52 (77.6%) tested positive and the most frequently mutated genes included KRAS (38.8%), TP53 (29.8%), and APC (25.35) followed by NRAS (11.9%) and PIK3CA (5.9%). Co-occurrence of variants in more than one gene was common in colorectal cancer samples as 32 patients tested positive for variant in more than one gene. The most common co-occurrence of variants was observed in KRAS, APC, and TP53 genes ( Figure 4 ).

Non-small cell lung cancer: Forty-eight NSCLC tumor samples tested by the comprehensive panel also showed high frequency of variants in KRAS (31.2%) and TP53 (18.7%) genes, which was in line with the targeted lung subpanel.

Pancreatic: Approximately 33 of 39 pancreatic tumor samples had a clinically relevant variant with KRAS (71.7%), TP53 (35.8%), and CDKN2A (10.2%) observed as the most common genes with a variant. A total of 13 pancreatic tumor samples had co-occurring variants in KRAS and TP53 genes ( Figure 4 ).

Endometrial: Out of 29 endometrial cancer samples, 21 (72.4%) tested positive and 34.4% had variants in TP53 gene, and 13.7% in PTEN and PIK3CA gene each.

Ovarian: Out of 28 ovarian cancer samples, 16 (57.1%) tested positive and 25% had variants in TP53 gene.

Thyroid tumor: Out of 23 thyroid tumor samples, 17 (73.9%) had genetic variants. BRAF gene was the most commonly mutated, in 39% of patients, followed by RET gene in 17.3% of patients.

Rare tumors: Some of the less frequent advanced tumors tested by the comprehensive panel in our study cohort include renal, esophagus, bladder, cervix, oral, gastric, sarcoma, melanoma, and prostate. The frequencies of variants in genes observed in these tumors are listed in Table 5 .

Target subpanels are designed to test specific genes and/or variants with known clinical utility for the tumor type. These genes are recommended by clinical guidelines and have a direct impact on patient management. Target panels have the advantage of streamlining data analysis and interpretation, expediting the results, and decreasing the rate of variants of unknown clinical significance. In our setting, NGS panel analysis resulted in high diagnostic yields across various clinical indications including lung (69%), colon (61.2%), melanoma (69.7%), and brain (20.7%), highlighting the utility of targeted subpanels in the clinical setting. In contrast, comprehensive panels have the ability to identify genes known to be targetable in other cancer and other genes with therapeutic or prognostic potential and/or diagnostic markers that may be incorporated in the clinical practice as more evidence become available. For example, comparison of data from colorectal tumor showed a significant diagnostic yield in genes such as TP53 and APC that are currently not part of the targeted panel. Overall, the diagnostic yield of the comprehensive panel was 58%, ranging from 77.6% in colorectal cancer to 2% in oral cancer ( Table 5 ).

Genetic mutations observed in NSCLC may vary among patients with different racial and demographic background and other clinical and pathological aspects (23, 24). However, we do not have the exact ethnicity information available for patients included in our study but overall frequency of variants in NSCLC observed in our study is in accordance with other reported studies from the Western population. For example, EGFR variants seen in 11% of our samples are reported in literature at nearly 15% among North Americans, in contrast to nearly 50% in Asians (23). The most common EGFR mutations are either exon 19 deletions or exon 21 L858R found at 42% and 26%, respectively, in our dataset, which is comparable but slightly lower than what has been reported in literature: exon 19 deletions at 42.8% and exon 21 L858R substitutions at 29.8% reported in a metanalysis of studies from many regions including North America (23). In our population, the frequency of less common EGFR mutations, exon 20 insertions (10%), G719X (7%), and L861Q (4%), are all higher than expected (25).

Similarly, KRAS mutation seen at 41.1% in our study is reported to range from one-fourth to one-third (∼15%–35%) of the NSCLC patients in the Western population to 4%–8% in Asians (26, 27). In our dataset, patients had KRAS G12C mutations in 15.2% of samples. This is higher than our rate of EGFR positivity, and nearly double the frequency of the common EGFR mutations (exon 19 deletions and L858R) combined.

BRAF and KRAS mutations have been reported to occur in ∼10% and 44% of patients with metastatic colorectal cancer, which is similar to frequency (KRAS: 46%; BRAF: 11.2%; NRAS: 3%) observed in our sample cohort (28). Frequency of variants identified in melanoma samples as listed in Table 4 is in line with reported studies (BRAF ∼50%, NRAS ∼25%; KIT ∼4.5%), as analyzed by Vanni et al. from different reported studies (29, 30). Glioma samples tested by the brain subpanel had BRAF variants at 8% and IDH1/IDH2 variants at 12.5%. As reported earlier, we noted a higher incidence of BRAF variants in younger patients (the median age of patients with a BRAF variant is 31.5 years vs. the median age of patients without a BRAF variants is 56.6 years) (31). However, it was noted that the overall frequency of variants across different tumor types identified by our targeted subpanels is comparable to reported literature but detailed analysis about the frequency of specific variant types based on precise tumor histology and clinical and demographic details is beyond the scope of this study as this information was not available in our laboratory-based database.

Out of 1,375 NSCLC samples tested by the lung subpanel, 69% (952) tested positive for a variant. Co-occurrence of mutation in more than one gene was observed in 260/1,375 (19%) patients. Co-occurring variants in oncogenic drivers and tumor suppressors in NSCLC tumors contribute towards complex molecular diversity of these tumors and may impact the efficacy of TKIs. The most common co-occurring genetic alteration observed in our study include TP53 (207 samples) with 121 KRAS, 40 EGFR, 25 PIK3CA, 17 BRAF, 3 NRAS, and 1 ERBB2 positive samples. Concurrent TP53 mutations with KRAS and EGFR are reported as negative prognostic markers for advanced NSCLC patients who are candidates for EGFR-TKI or ALK-TKI treatments (32). Dual mutations in the EGFR gene like T790M/Exon 19 del and T790M/L858R were observed and are known to alter the response to TKI therapy (33). We also observed tumor samples with co-occurring driver variants in EGFR, KRAS, and BRAF genes (seven patients with KRAS/EGFR variants; four patients with BRAF/KRAS; one patient with BRAF/EGFR/KRAS). Typically considered as mutually exclusive, concurrent driver mutations in EGFR/KRAS/BRAF genes are rare in literature (34, 35) and their clinical relevance is not well known but it has been suggested that patients with co-occurring actionable variants may require tailoring the combination or sequential treatment strategies (36). Among the samples tested by the colon subpanel, a high frequency of mutations was noted in KRAS (46.5%) and BRAF (11.8%) genes, and four patients had concomitant KRAS and BRAF (nonV600E) variants. BRAF and KRAS are two key oncogenes that determine response to anti-EGFR therapies in colorectal cancer patients. BRAF and KRAS variants are mutually exclusive and co-occurrence is rarely reported (37, 38). However, clinical outcomes for such patients is not well characterized but personalized combination therapeutic strategies have been suggested for such patients (28). Also, the co-occurrence of NRAS and BRAF, the two most frequently mutated genes in melanoma, is also rarely reported (2.9%), which is in agreement with our observations as we found concomitant NRAS/BRAF variants in 5 (1%) of our melanoma samples (39); however, their clinical impact is not known.

A growing number of genomic aberrations are impacting the treatment decisions for advanced metastatic tumors. We implemented a comprehensive NGS assay covering 50 oncogenes and tumor suppressor genes to identify variants in genes with a potential clinical impact in a variety of tumor types, including breast, colorectal, pancreas, endometrial, ovary, lung, and thyroid tumors. To evaluate the molecular profile, 500 tumor samples were assessed by this assay and a positive mutation rate of 58% was observed. Some of the frequently mutated genes identified in these tumors hold high clinical relevance.

Sequencing studies analyzing somatic driver mutations in genetically complex tumors like breast cancer have identified PIK3CA as a frequently mutated gene in breast tumors followed by TP53 with co-occurrence reported in 10%–15% patients (40, 41). We detected variants in PIK3CA at 33%, TP53 at 22%, and co-occurring variants in 9% patients (7/77). Variants in other clinically relevant genes were less frequent with EGFR and AKT1 variants found in 5% patients. The PI3-kinase inhibitor in combination with the estrogen receptor (ER) antagonist is FDA-approved for the treatment of patients with PIK3CA mutant ER+/HER2− breast cancer and the presence of PIK3CA/TP53 mutations has evidence of prognostic relevance (42, 43).

Colorectal tumors had the highest frequency of variants (77.6%) as compared to other tumors analyzed in our study. In line with previous studies, colorectal cancer samples had high frequency of variants in genes including KRAS, TP53, APC, NRAS, PTEN, and PIK3CA. Other than RAS genes that predict anti-EGFR sensitivity, PIK3CA mutations have prognostic implication in colorectal tumors. PIK3ACA mutations are associated with a worse response to first-line chemotherapy (11) and a phase I clinical trial has shown evidence for the sensitivity of PIK3CA mutated colorectal cancer to the PIK3a-selective inhibitor (44). Early trial data suggest that PI3K inhibitors may be of benefit to solid tumors harboring PTEN mutation (45). There is evidence that co-occurring variants in APC, PIK3CA, TP53, and KRAS may impact the overall disease progression and response but it is not yet well studied and larger trials are needed to fully understand the impact on disease outcome (46).

Following TP53 (34.4%), PTEN and PIK3CA were the most commonly mutated genes in endometrial tumors, and these variants have predictive value in endometrial tumors. PTEN has been associated with a favorable outcome in endometrial cancer, and pre-clinical data have shown that inactivating mutations in the PTEN gene may confer sensitivity to PI3K-AKT inhibitors as well as PI3K/mTOR inhibitors (14, 45). Since PIK3CA is another frequently mutated gene in endometrial tumors, PIK3CA-directed inhibitors may show benefits but its utility is still under investigation (47). Recurrent TP53 mutations in endometrial tumors have been associated with higher rates of recurrence in grade 1–2, stage I and II endometrioid adenocarcinoma (48, 49). CTNNB1 gene variants were less frequent at 7% and are associated with increased risk of recurrence in grade 1–2 early-stage endometrial endometrioid adenocarcinoma (50).

KRAS gene alterations are the most frequent, observed at a frequency of 72% in pancreatic tumor samples, with TP53 at 36% as the second most common followed by CDKN2A. The high prevalence of KRAS variants in pancreatic tumors suggests the clinical utility of therapies targeting the RAS signaling pathway, but studies done in this direction so far have shown little clinical benefit (51). However, KRAS mutations have been reported to have a negative impact on prognosis and improve the performance of classic cytopathology to diagnose pancreatic tumors (52).

Forty-eight NSCLC tumor samples tested by the comprehensive panel had KRAS (31.2%) and TP53 (18.7%) as the most frequently mutated, followed by ERBB2, RB1, and EGFR. EGFR mutations predict response to anti-EGFR therapies in NSCLC tumors, and TP53 mutations have been reported to have a negative prognostic effect (6). Oncogenic variants in ERBB2 gene may confer sensitivity to anti-HER2-directed therapies and trastuzumab deruxtecan is now FDA-approved for these patients (53–55).

Thyroid cancer is another tumor type with a high mutation positivity rate of 74% with BRAF at 40% followed by RET at 17.3%. BRAF positive thyroid tumors have shown sensitivity to RAF and/or MEK inhibitors (56) and RET variants correlate with aggressive phenotype and worse outcome (57). Co-occurring variants also predict survival in thyroid tumors (58, 59).

A molecular testing model implemented in Ontario as per CCO guidelines represents the integration of both comprehensive and targeted subpanels for tumor samples in the clinical setting. The data collected here are not sufficient to make any conclusions about the utility of one over the other, but this model definitely highlights the feasibility of administering both in a clinical environment with evidence of high variant detection rates. We have targeted ongoing studies to establish the utility of subpanel vs. full panel testing and also to investigate disease-specific clinical scenarios suited best for each type of testing.