Diagnostic samples were obtained from the repository of the Institute of Pathology at the University Hospital Cologne, Germany. Formalin-fixed and paraffin-embedded (FFPE) tissue samples were obtained as part of routine clinical care under approved ethical protocols complied with the Ethics Committee of the Medical Faculty of the University of Cologne. The Ethics Committee waved the need for ethical approval for this study. Sixty-six primary non-small cell lung cancer (NSCLC) samples were included in this study, with cases initially testing negative by NGS intentionally selected to ensure thorough comparison with FISH results. Sample selection was based on availability and prior evidence suggesting particular gene amplifications. FISH analysis was performed for the amplification of the Mesenchymal Epithelial Transition (MET) gene on 26 samples, the Human Epidermal Growth Factor Receptor 2 (ERBB2) gene on 21 samples, the Phosphatidylinositol-4,5-Bisphosphate 3-Kinase Catalytic Subunit Alpha (PIK3CA) gene on 9 samples, and the Kirsten Rat Sarcoma Viral Oncogene Homolog (KRAS) gene on 9 samples. One tumor was tested for both MET and ERBB2 gene amplification. In the NGS analysis, copy number gains for each gene were determined by quantifying coverage increases as fold changes. For a given target gene, the fold change was calculated by dividing the highest observed coverage among the target regions of that gene by the mean coverage across all genes in the sequencing panel.

NGS provided data on targeted regions in 27 different genes ( Supplementary Table S1 ) regarding mean coverage (of all genes together) and range of coverage for each specific gene covered. The following sequences of interest were covered by our NGS panel: MET (exons 14 (+intron 13 and intron14), 16–19; reference sequence: NM_001127500), ERBB2 (exons 8, 19, 20; reference sequence: NM_004448), PIK3CA (exons 8, 10, 21; reference sequence: NM_006218), and KRAS (exons 11, 15; reference sequence: NM_033360). Parallel validation with FISH was performed on all samples to verify the NGS findings.

All samples underwent formalin fixation and paraffin embedding according to standard protocols (11). Sections of 10 µm thickness were sliced from the FFPE tissue blocks and subsequently deparaffinized. Tumor regions were macrodissected from unstained slides, using a marked hematoxylin-eosin (H&E) stained slide for reference. All slides were marked by an experienced pathologist. The samples were digested overnight with proteinase K, and DNA extraction was performed using the Maxwell RSC FFPE Plus DNA Kit (Promega, Mannheim, Germany) on the Maxwell RSC (Promega), following the manufacturer’s guidelines.

For sequencing, the DNA content was first measured using a Tecan Infinite 200 microplate reader (Tecan, Maennedorf, Switzerland) and the QuantiFluor ONE dsDNA Kit (Promega). Genomic DNA was fragmented by enzymatic fragmentation to a suitable size range (e.g. 150–200 bp) (Twist Bioscience, South San Francisco, California, United States). The fragmented DNA was then subjected to end-repair, adenylation, and ligation with sequencing adapters (Twist Library Preparation EF Kit).

A hybrid-capture enrichment protocol was employed to selectively capture the target regions of interest. Biotinylated probes complementary to the target sequences (Twist Bioscience NGS Target Enrichment Solutions with a panel of 27 genes relevant for lung cancer therapy) were hybridized to the fragmented DNA. The hybridized DNA-probe complexes were captured using streptavidin-coated magnetic beads (Dynabeads MyOne Streptavidin T1; Thermo Fisher Scientific, Waltham, Massachusetts, United States). Unbound DNA was washed away, and the captured DNA was eluted from the beads.

The enriched target DNA was then PCR-amplified using library-specific primers (KAPA HiFi HotStart ReadyMix; Roche, Basel, Switzerland) to prepare the sequencing library. The resulting library was quantified, diluted, and pooled in equal amounts. Finally, the constructed libraries were sequenced on a high-throughput sequencing platform (NextSeq 550 System; Illumina or NovaSeq; Illumina, San Diego, California, United States) using the appropriate sequencing reagent kit (NextSeq 500/550 High Output Kit v2.5; Illumina or NovaSeq 6000 SP (300c) Reagent Kit v1.5; Illumina) following the manufacturer’s recommendations. Subsequent data analysis was done using an in-house pipeline. Employing fg-bio tools (v. 2.2.2) (https://fulcrumgenomics.github.io/fgbio/), alignment steps with bwa mem (v. 0.7.17) (31) included read deduplication to exclude PCR related amplification artifacts. From GATK (v. 4.1.9.0), the CollectHsMetrics tool was used to determine coverage (12).

FISH analysis was performed on 4 µm thick slides prepared from the same FFPE blocks used for NGS. Probes specific to the regions of interest were obtained from ZytoVision. The procedure followed the manufacturer’s instructions, with hybridization conditions optimized for each probe set. Fluorescence signals were detected using a Microscope KP-PLUS slides (Klinipath, Duiven, Netherlands). For each gene, tissue slides were hybridized overnight with the respective Zyto-Light SPEC Dual Color Probe (ZytoVision, Bremerhaven, Germany) — MET/CEN7, ERBB2/CEN17, PIK3CA/CEN3, and KRAS/CEN12. Twenty contiguous tumor cell nuclei were individually evaluated to calculate the gene/centromere (CEN) ratio and the average GCN per cell. The classification criteria for MET, ERBB2, PIK3CA, and KRAS amplifications were adapted from previously published studies (13–18). Signals were manually counted in cells with ≤15 copies and estimated using clusters for 15 and more gene copies (10, 14).

Descriptive statistics were used to summarize the data, including mean and range for NGS fold changes, FISH Gene/CEN ratios, and average GCN per cell. The relationship between NGS fold change and FISH Gene/CEN ratios as well as average GCN per cell was evaluated using Spearman Correlation to assess the strength and direction of the association. Linear regression analyses were conducted to model the relationship between NGS fold changes and both FISH Gene/CEN ratios and average GCN per cell, providing insights into the predictive value of NGS data for FISH-determined gene amplifications. Both Spearman Correlation and Linear regression analyses were performed using the online tool available at www.socscistatistics.com.

Tumor cell content (TCC), as determined through histopathological assessment of H&E-stained slides in our cohort ranged from 10% to 80%, providing suitable conditions for NGS analysis (20). Mean NGS coverage was highest in the MET group at 2996.3 reads and lowest in the KRAS group at 209.6 reads. Fold changes in coverage varied across groups, with MET ranging from 1.29 to 8.45, ERBB2 from 0.57 to 25.08, PIK3CA from 2.11 to 6.59, and KRAS from 2.32 to 6.04.

Amplification was identified in 46 samples (68.7%), including 17 cases of MET, 11 of ERBB2, and all cases of PIK3CA and KRAS. With the exception of four MET cases (two with high-level and two with low-level amplification), all FISH-positive cases in our cohort showed positivity in both Gene/CEN ratio and GCNs per cell. In the MET group, the Gene/CEN ratio varied from 1.02 to 9.46, while in the ERBB2 group, it ranged from 0.86 to 8.59. The PIK3CA group showed ratios between 2.9 and 11.29, and the KRAS group had the highest ratios, from 4.39 to 18.2. Correspondingly, GCNs per cell spanned from 2.65 to 30.95 in the MET group, 2.55 to 28.7 in the ERBB2 group, 8.55 to 28.35 in the PIK3CA group, and 6.8 to 23.95 in the KRAS group.

Among FISH-negative cases, NGS fold change ranged from 0.57 for ERBB2 to 1.95 for MET, while in FISH-positive cases, it ranged from 2.11 for PIK3CA to 25.08 for ERBB2. Two MET cases with low level amplification had corresponding NGS fold changes of 2.54 and 3.36. NGS fold changes showed a positive correlation with FISH Gene/CEN ratios (Spearman’s ρ = 0.720, p < 0.001), with the linear regression model: ŷ = 0.292X + 3.031 ( Figure 2a ). Similarly, a positive correlation was observed between NGS fold changes and gene copy number per cell (Spearman’s ρ = 0.847, p < 0.001), described by the model: ŷ = 1.182X + 8.081 ( Figure 2b ). The linear regression plots for each gene are presented in Figure 3 .

The strong positive correlations observed between NGS fold changes and both FISH Gene/CEN ratios and gene copy number per cell underscore the utility of NGS in evaluating gene amplification status. This suggests that NGS could reliably indicate amplification without requiring confirmatory FISH, especially when fold changes are markedly elevated. For setting a cutoff to distinguish FISH-positive from FISH-negative cases based on NGS fold change, our results suggest that values below 2.0 tend to correspond with FISH-negative findings, while values above 2.0 align with FISH-positive results.

A challenge arises when FISH shows a low Gene/CEN ratio but a high average GCN. These cases could indicate true amplifications or high polysomy, yet may not exhibit a corresponding increase in NGS coverage (24). Visually identifying such cases by coverage alone can be challenging, as NGS might not show the expected increase in coverage. This is because NGS coverage reflects the total number of sequence reads mapped to a gene’s target regions, and if these regions are limited, subtle increases in gene copies might not significantly impact overall coverage (10). Moreover, technical factors such as TCC, sequencing depth, and the specific regions of the gene covered by the NGS panel can also influence the observed fold change. For instance, if a gene has multiple copies in a small percentage of tumor cells, the overall increase in coverage should be diluted by non-tumor cells, leading to an underestimation of amplification. In our study, we encountered 20 cases with low tumor cellularity (≤20%), which could potentially affect the accuracy of NGS-based copy number assessments. Although our analysis did not reveal significant discrepancies between NGS and FISH results in these cases, we acknowledge that low tumor cellularity is a known factor that can lead to underestimation of copy number amplification. This underscores the importance of considering tumor cellularity when interpreting NGS results and highlights the need for integrating multiple diagnostic modalities. Therefore, relying solely on NGS coverage as a predictive tool requires further validation on larger cohorts, along with more refined analytical methods and expanded gene coverage in NGS panels to enhance detection accuracy (25, 26). Additionally, assessing more genes with FISH is essential to determine if a 2.0 cutoff is universally optimal across various genes.

Our NGS panel focused on selected exons of particular interest, which could influence both coverage and correlation results. By targeting specific exons, the panel might not fully represent the gene’s overall amplification status, potentially leading to underestimation or overestimation of gene amplification (25, 26). This limitation is particularly relevant when the regions selected do not encompass the most critical areas of amplification across the entire gene. This selective coverage may also contribute to the variability observed in the regression slopes across different genes, as illustrated in Figure 3 . For example, genes with exons more susceptible to technical factors—such as variations in probe efficiency or sequencing depth—might exhibit inconsistent fold changes compared to FISH results. Furthermore, the use of a single reference sequence for each gene in NGS adds another layer of variability to fold change calculations. Discrepancies between the reference sequence and the actual genomic sequence in the tumor samples could introduce inaccuracies in fold change measurements, affecting the reliability of NGS in predicting gene amplifications.

Integrating NGS into clinical diagnostics offers significant potential for advancing the detection and characterization of genomic alterations in cancer. Unlike FISH, which is restricted to specific probes and lower throughput, NGS can simultaneously detect small variants, GCN changes across numerous genes, and other key biomarkers such as tumor mutational burden (TMB), microsatellite instability (MSI), and homologous recombination deficiency (HRD) (27–30). This comprehensive capability makes NGS an invaluable tool for personalized cancer treatment, offering a more complete genomic profile with less observer dependence and reduced risk of tissue sectioning artifacts.

However, to fully harness the potential of NGS, further research is required to refine predictive models and establish standardized thresholds for NGS fold changes. While our study demonstrates the utility of NGS in detecting MET and ERBB2 amplifications in NSCLC, the absence of EGFR amplifications in our cohort may potentially limit the generalizability of our findings. This absence reflects our sample selection strategy, which prioritized cases with other alterations rather than a cross-section of NSCLC cases. Future studies should include a broader range of gene amplifications, including EGFR, to comprehensively evaluate the performance of NGS across the diverse molecular landscape of NSCLC. Similarly, this study lacks non-amplified PIK3CA and KRAS cases that probably results in higher regression lines ( Figure 3 ). A robust validation cohort should include not only cases with clear amplification and FISH-negative samples, but also low-coverage cases, and those with high GCNs but low Gene/CEN ratios. This would allow for the determination of a reliable cut-off for negative results and improve the specificity of NGS. Additionally, examining cases with amplification in the context of high polysomy could provide insights into how NGS distinguishes these from true gene amplifications. Notably, the amplification status in our study was determined based on established FISH criteria, which account for both Gene/CEN ratios and average GCNs. While this approach effectively identifies FISH-positive cases, it does not exclude the possibility of high polysomy being classified as amplification. Future analyses should explore this distinction further to enhance the accuracy of NGS-based classifications.

To further enhance the accuracy of NGS, expanding the panel to cover more regions in genes of interest could improve detection sensitivity. Additionally, in future studies, reporting the range of Gene/CEN ratios and GCNs for individual cells could provide valuable insights into intratumor heterogeneity and help further explore the correlation between NGS and FISH results. Refining this aspect could make NGS reliable alternative to FISH in routine diagnostics, particularly for cases where multiple biomarkers need to be assessed simultaneously.