Evaluation of two highly-multiplexed custom panels for massively parallel semiconductor sequencing on paraffin DNA

Abstract

Aim: Massively parallel sequencing (MPS) holds promise for expanding cancer translational research and diagnostics. As yet, it has been applied on paraffin DNA (FFPE) with commercially available highly multiplexed gene panels (100s of DNA targets), while custom panels of low multiplexing are used for re-sequencing. Here, we evaluated the performance of two highly multiplexed custom panels on FFPE DNA.

Methods: Two custom multiplex amplification panels (B, 373 amplicons; T, 286 amplicons) were coupled with semiconductor sequencing on DNA samples from FFPE breast tumors and matched peripheral blood samples (n samples: 316; n libraries: 332). The two panels shared 37% DNA targets (common or shifted amplicons). Panel performance was evaluated in paired sample groups and quartets of libraries, where possible.

Results: Amplicon read ratios yielded similar patterns per gene with the same panel in FFPE and blood samples; however, performance of common amplicons differed between panels (p<0.001). FFPE genotypes were compared for 1267 coding and non-coding variant replicates, 999 out of which (78.8%) were concordant in different paired sample combinations. Variant frequency was highly reproducible (Spearman's rho 0.959). Repeatedly discordant variants were of high coverage / low frequency (p<0.001). Genotype concordance was (a) high, for intra-run duplicates with the same panel (mean±SD: 97.2±4.7, 95%CI: 94.8-99.7, p<0.001); (b) modest, when the same DNA was analyzed with different panels (mean±SD: 81.1±20.3, 95%CI: 66.1-95.1, p = 0.004); and (c) low, when different DNA samples from the same tumor were compared with the same panel (mean±SD: 59.9±24.0; 95%CI: 43.3-76.5; p = 0.282). Low coverage / low frequency variants were validated with Sanger sequencing even in samples with unfavourable DNA quality.

Conclusions: Custom MPS may yield novel information on genomic alterations, provided that data evaluation is adjusted to tumor tissue FFPE DNA. To this scope, eligibility of all amplicons along with variant coverage and frequency need to be assessed.

Conflict of interest statement

Competing Interests: Author Zoi Alexopoulou is employed by Health Data Specialists Ltd. There are no patents, products in development, or marketed products to declare. This does not alter the authors' adherence to all the PLoS ONE policies on sharing data and materials. The rest of the authors have declared that no competing interests exist.

Figures

Fig 1. Panel comparison and sample groups.

A. Comparison of the B and T panels with respect to amplicon targets. Common amplicons had the same ID in both panels. Common target—different amplicons had slightly shifted coordinates targeting the same genomic regions (±10 nts). Percentages among all amplicons in the two panels are shown. B. REMARK diagram of patients, samples, and sample groups in this study. In total, 316 DNA samples in 332 libraries were examined.

Fig 2. Performance of the B and T panels in DNA from matched blood and tumor FFPE samples.

A: B panel; B: T panel. Read ratios for all amplicons for matched samples sorted per gene are shown, corresponding to 70192 observations with the B and 26128 observations with the T panel. Lanes and dots therein represent mean read ratios per amplicon for all samples tested in the respective group. Amplicon order is the same in all graphs. Solid and dotted horizontal lines within graphs: mean values + 3xSD per panel per sample type, respectively. Amplicon reading efficiency was overall constant between blood—FFPE samples with the same panel. For some genes with frequent gains in breast cancer, e.g., CCND1, EGFR, ERBB2, PIK3CA (B panel) and AKT1, EGFR (T panel) outliers with maximal amplicon read ratios outside the Y-axis were observed (A & B, red stars in tumor graphs). By contrast, for genes frequently lost in breast cancer, e.g., TP53 with both panels, mean read ratios in tumor DNA were lower than in blood (A & B, turquoise stars). FFPE-specific over-representation of individual amplicons was occasionally observed (A & B, diagonal arrows). Importantly, patterns of read ratios occasionally differed for genes targeted with the same amplicons in panels B and T, e.g., for ARID1B, MAP3K1, TP53 (A & B, black stars with coloured outlines in blood graphs).

Fig 3. Performance of individual amplicons in blood and FFPE DNA.

A. Amplicon performance grading did not significantly differ between the two panels. Columns: combined evaluation of each panel in blood and FFPE. Numbers within boxes: actual amplicon number per category, as indicated. B. Performance of the 83 common amplicons was significantly different in the two panels (p<0.0001). The most unstable amplicons were of performance grade 1 and 2. C. A non-linear distribution of mean read ratios according to amplicon GC content was observed. Read ratios of amplicons with >75% or <25% GC content were almost uniformly below the 10th percentile cut off (dotted line in both graphs) (Kruskal-Wallis test p<0.0001 for each panel and sample group). D. Very high and very low GC content was significantly associated with failed amplicons (gr 0). This pattern was also present for the B-panel in blood samples, despite the absence of statistical significance. Except for these extreme cases, however, all other amplicon categories did not significantly differ with regards to GC%.

Fig 4. Variant concordance in FFPE samples.

For all graphs, solid color boxes represent concordant, whereas striped boxes represent discordant variants. Red color stands for MUT variants (amino acid changing, excluding TP53 p.P72R) and blue for nonMUT (non amino acid changing, coding and non-coding). Opposite directions of stripes stand for each paired sample. A: Five TN samples with the B panel. B: Five BR samples with the T panel. C: Four TN samples with the T panel (out of five on trial, one was ineligible). D: Panel comparison for the same sample. Only common and shifted amplicons in the two panels were evaluated. E: Different samples from the same tumor revealed distinct genotypes when tested with the same panel. F: Comparison between genotypes from subsequent runs performed 6 months apart for BR samples with the B panel. Old run only, new run only: patterns apply to all columns in this graph.

Fig 5. Effect of coverage and frequency on variant concordance with custom panels on FFPE samples.

A: Variant frequency (VF) was highly preserved in replicate measurements from the same samples. B and C: Concordant variants had statistically significant higher coverage and frequency as compared to discordant ones (red lines: mean values). However, both high and low coverage and frequency were observed for discordant variants and followed the same bimodal pattern in these categories. D. Out of the 181 different positions accounting for the 1267 variants under study, calls were constantly concordant in 83 (45.8%), partially concordant in 26.0%, and constantly discordant in 28.2%. E. Distribution of VF and amplicon GC% according to high and low coverage for the categories indicated on the bottom of the graphs. Red and lilac dots: high coverage in discordant and concordant variants, respectively; turquoise and green dots: low coverage, respectively. Among discordant only variants, those of high coverage were almost exclusively of low VF (boxed).

Fig 6. Sanger sequencing validation of MPS variants.

Five cases with matched tumor—blood samples are shown. Annotations in black letters are given for tumors (FFPE); in blue letters for germline (blood). TCC: approximate tumor cell content in the DNA sample; VC and PC: variant and position coverage with MPS; VAF: Variant frequency (VC/PC). The same DNA samples were used per case for both methods. A—C: same case, DNA quality unfavourable, three variants in tumor, two in germline. D—F: individual cases with unfavourable (D and E) and favourable (F and G) DNA quality. Sanger sequencing peaks usually but not always corresponded to MPS VAF. In A, D and F, perhaps in B as well, variants would have been missed with Sanger sequencing only. Note target specific differences in position coverage, which derives from amplicon reads. Red dots: low VAF. Black dot: the expected wild type allele at 17% frequency, based on MPS VAF, was not observed with Sanger sequencing.