High-Resolution Copy Number Patterns From Clinically Relevant FFPE Material

Abstract

Systematic tumour profiling is essential for biomarker research and clinically for assessing response to therapy. Solving the challenge of delivering informative copy number (CN) profiles from formalin-fixed paraffin embedded (FFPE) material, the only likely readily available biospecimen for most cancers, involves successful processing of small quantities of degraded DNA. To investigate the potential for analysis of such lesions, whole-genome CNVseq was applied to 300 FFPE primary tumour samples, obtained from a large-scale epidemiological study of melanoma. The quality and the discriminatory power of CNVseq was assessed. Libraries were successfully generated for 93% of blocks, with input DNA quantity being the only predictor of success (success rate dropped to 65% if <20 ng available); 3% of libraries were dropped because of low sequence alignment rates. Technical replicates showed high reproducibility. Comparison with targeted CN assessment showed consistency with the Next Generation Sequencing (NGS) analysis. We were able to detect and distinguish CN changes with a resolution of ≤10 kb. To demonstrate performance, we report the spectrum of genomic CN alterations (CNAs) detected at 9p21, the major site of CN change in melanoma. This successful analysis of CN in FFPE material using NGS provides proof of principle for intensive examination of population-based samples.

Figure 1

Workflow of samples from tumours sampled from the Leeds Melanoma Cohort. The path below shows how sample numbers are reduced through the process. The Figure also shows the bioinformatics stages through the process. This figure was constructed using SmartDraw 2007 v8.16 Healthcare Edition (www.smartdraw.com).

Figure 2

Nucleotide summaries for each library. Overall, for the 335 libraries sequenced at 5 per lane, sequencing yielded an average of 82.1 × 10⁶ reads (21.3 × 10⁶–191.4 × 10⁶). For the 20 libraries sequenced at 1 per lane, there was an average of 368.2 × 10⁶ reads (215.3 × 10⁶–425.6 × 10⁶). The total nucleotide count (across both pairs) for each library is represented by the bar height. This is divided into nucleotides removed during preprocessing (“failed cutadapt” in orange) and nucleotides removed during alignment and read post-processing (“failed alignment” in red). The remaining nucleotides (“passed” in green) are present in the downstream analyses. Library annotations are shown below the plot. The libraries run at a higher depth at one library per lane are marked (“single lane”). The 10 libraries identified as low quality and subsequently rejected are marked (“rejected”), as are the seven libraries from normal samples (“normals”). Following pre-processing (adapter trimming, dropping of reads less than 20 bp, and quality score less than 20), a median of 79.1 × 10⁶ (5.5 × 10⁶–189.3 × 10⁶) reads (multiplexed) and 366.1 × 10⁶ (219.6 × 10⁶–423.8 × 10⁶) reads (single) were successfully aligned. After alignment, deduplication, and read quality filtering, a median of 65.1 × 10⁶ (multiplexed) and 320.0 × 10⁶ (single) reads were retained. 10 libraries which were clear outliers, showing exceptionally low alignment rates (3.1 × 10⁶–34.3 × 10⁶ aligned reads retained), were rejected from subsequent analyses. Overall, the non-rejected libraries gave a median alignment rate of 0.82 (0.47–0.90) yielding a median coverage of 1.8x (multiplexed) and 9.1x (single).

Figure 3

Summary of 303 distinct tumour samples examined for chromosome 9p21 in the CDKN2A region (CDKN2A (chr9:21,967,753–21,995,301); MTAP (chr9:21,802,636–21,865,971); and CDKN2B (chr9:22,002,903–22,009,363) showing that 204 of these samples (67%) showed no evidence of a CN change. Among the 99 samples showing a CN change, various patterns are observed, the majority of which impacted in CDKN2A, the known target of common alteration for melanoma. Three lines of evidence support the findings from the bioinformatic analysis: (A) (bottom left) gene expression analysis of regions of p16 (a protein product of CDKN2A) are associated with the extent of deletion in keeping with tumour heterogeneity and subsets of cells having a deletion (see also Supplementary Fig. 11); (B) (bottom centre) replication of results in a subset of tumours via MLPA. MLPA mean ratios (red dots) are superimposed (see Supplementary Table 2); and (C) evidence that a germline CN variant, esv36200012 (chr9: 23362412–23378071), can be detected reliably. The bottom right panel shows that the measured extent of CN loss matches the expected genotype.

Figure 4

A summary of the regions deleted on 9p21 in the vicinity of CDKN2A. The limits of the coding regions of the genes in the region is shown vertically along the chromosome while horizontally the boundaries of the deleted regions produced by the segmentation assay are shown. As expected, the target of the deletions is CDKN2A. Gene boundaries: MTAP = green; CDKN2A = light blue; CDKN2B = blue.