Skip to main content

Clinical whole-genome sequencing from routine formalin-fixed, paraffin-embedded specimens: pilot study for the 100,000 Genomes Project.

Robbe P, Popitsch N, Knight SJL, Antoniou P, Becq J, He M, Kanapin A, Samsonova A, Vavoulis DV, Ross MT, Kingsbury Z, Cabes M, Ramos SDC, Page S, Dreau H, Ridout K, Jones LJ, Tuff-Lacey A, Henderson S, Mason J, Buffa FM, Verrill C, Maldonado-Perez D, Roxanis I, Collantes E, Browning L, Dhar S, Damato S, Davies S, Caulfield M, Bentley DR, Taylor JC, Turnbull C, Schuh A

Publication: Genet Med (2018) • Vol 20 • Page 1196-1205PMID: 29388947View on PubMed

Purpose

This paper compared DNA yield and integrity; next-generation sequencing (NGS) metrics; and detection of single nucleotide variants (SNVs), insertions/deletions (InDels), and copy number variations in matched formalin-fixed paraffin-embedded (FFPE) and frozen tumor specimens. The effects of optimization of DNA extraction, reverse cross-linking, and the bioinformatics pipeline used for analysis on NGS metrics and SNV detection were also investigated.

Conclusion

As expected, DNA from FFPE specimens was more degraded than that from frozen specimens and NGS metrics; including read pass filter aligned ratios, mean sequencing depth, proportion of regions reaching a depth of 70X, median insert size, ratio of chimeric pairs, uniformity of coverage, and AT and GC dropout performance, were significantly worse in FFPE specimens. The number of SNVs and InDels detected depended on the variant detection method used. Overall, 6% of the variants were detected in both the FFPE and frozen specimen, 21% were only identified in the frozen specimen, and 27% in just the FFPE specimen with FFPE specimens having a greater proportion of C>A/G>T mutations and a lower proportion of C>T (24.2% versus 36%). Percentage overlap in mutations was strongly correlated with the mutant allele fraction such that as the minimal allele fraction was increased, the overlap between FFPE and frozen specimens increased. Although 90% of the genes and InDels from the Cosmic cancer gene census were more frequently mutated in FFPE than frozen specimens, less than 1% of functional mutations were more frequently mutated in FFPE specimens.

Use of any of the optimization protocols improved NGS metrics, with no clear preferred method. Further, the overlap in mutation detection between frozen and FFPE specimens was higher using the optimization protocols. While the variability prevented accurate copy number detection using the initial extraction protocols, the optimized protocols increased the reliability of copy number analysis.

Studies (2)

Purpose

This study compared DNA yield and integrity; NGS metrics; and detection of SNVs, InDels and copy number variations in matched FFPE and frozen tumor specimens. The effects of differences between bioinformatics pipelines used on SNV detection was also investigated. Matched tumor and peripheral blood specimens were collected from 52 cancer patients (10 breast, 12 colorectal, 7 endometrial, 4 prostatic, 14 renal, and 5 thoracic). A 5-mm punch of the tumor specimen was snap-frozen in liquid nitrogen and embedded in OCT. The remainder of the tumor was formalin-fixed in 10% formalin in sodium chloride or phosphate for an unspecified duration that was determined by specimen type and then paraffin-embedded. Frozen and FFPE sections were evaluated for cellularity, necrosis, and percentage viable cells. DNA was extracted from blood using QIAamp DNA Mini Kit from 30 mg frozen specimens using the All Prep DNA/RNA Mini Kit and from two 1-mm FFPE punches using truXTRAC FFPE DNA Kit with an overnight proteinase K digestion at 56˚C. DNA was quantified using Nanodrop and the Qubit dsDNA Broad Range Assay kit. Integrity was determined by electrophoresis and purity by A260/280 ratios. Quality was determined by the quantitative real-time PCR based FFPE QC PCR assay. Sequencing libraries were prepared using the TruSeq Nano DNA Library Prep and sequenced on a HiSeq2500 to 70X for tumor specimens and 30X for blood specimens.

Summary of Findings

FFPE specimens had significantly lower yields of double-stranded DNA than matched frozen specimens (P=1.41 x 10-6) and electrophoresis revealed the DNA was fragmented (<0.5-40 kb versus >40 kb). Further confirming the effect was due to degradation, the spectrophotometric yield and purity were comparable between frozen and FFPE specimens. FFPE specimens had lower median insert size (P=1.798 x 10-9), read pass filter aligned ratios (P=3.618 x 10-9), mean sequencing depth (77x versus 93x), and proportion of regions reaching a depth of 70X (0.351 versus 0.782), and a higher ratio of chimeric pairs (P=1.330 x 10-8) than matched frozen specimens. Further, FFPE specimens had a higher AT dropout and lower GC dropout than matched frozen specimens (P=1.804 x 10-9, both) and 1.1 to 5.8-fold higher standard deviation in the uniformity of coverage (P<2.2 x 10-16).

The number of SNVs and InDels detected depended on the variant detection method used. Mutect and Strelka had the highest SNV sensitivity (0.77 for Strelka and 0.76 for Mutect versus 0.49 for Shimmer and 0.51 for a Combination), but the positive predictive value (PPV) was highest for the combination or Mutect (0.89 for combination and 0.72 for Mutect versus 0.35 for Shimmer and 0.21 for Strelka). The number of C>T mutations in FFPE specimens was correlated with the increase in intragenic variants and was enriched in FFPE specimens, especially when analyzed by Strelka. The sensitivity for InDels was highest using Strelka (0.5 for Strelka, 0.28 for Shimmer, and 0.3 for the combination), but the PPV was highest using the combination or Shimmer (0.65 for combination, 0.62 for Shimmer, and 0.11 for Strelka).

Overall, 6% of the variants were detected in both the FFPE and frozen specimens, 21% were only identified in frozen specimens, and 27% in just FFPE specimens with FFPE specimens having a greater proportion of C>A/G>T mutations and a lower proportion of C>T (24.2% versus 36%). The authors state the lower proportion of C>T mutations indicates that the mutations in FFPE specimens are not deamination artifacts. The overlap in SNV detection between FFPE and frozen specimens increased with increasing minimum allele frequency such that it was 71% in reliable areas of the genome and 63% overall when the minimal allele fraction was set to be 0.067 in one specimen. Further, percentage overlap in mutations was strongly correlated with the mutant allele fraction (r=0.731, P=1.09e-9); modestly correlated with the number of C>A/G>T mutations (r=0.437, P=0.0013), proportion of reads aligned (r=0.492, P=0.000245), proportion of chimeric pairs (r=0.45, P=0.000934), median of reads covered at 70X (r=0.437, P=0.00134), and number of SNVs in the FFPE sample (r=0.458, P=0.000718); and weakly correlated with median insert size (r=0.368, P=0.00786). The authors consequently believe they are indicative of intra-tumoral variability. There was no increase in HighRisk InDels in FFPE specimens. Although 90% of the genes and InDels from the Cosmic cancer gene census were more frequently mutated in FFPE than frozen specimens, this was true for less than 1% of the functional mutations. Further, the sensitivities of frozen and FFPE specimens for mutations in Ion AmpliSeq Cancer Hotspot Panel v2 were 0.86 and 0.82, respectively. FFPE specimens had much more noise in the copy number intensity plots resulting in decreased reliability of copy number calls.

The tumor content determined by the pathologists and the specimen purity determined by sequencing data were weakly correlated (r=0.347, P=0.00376). Among the tumor types, thoracic tumors had the most SNVs and the highest agreement in SNVs between FFPE and frozen specimens.

  • Biospecimens

    TissueKidney
    TissueBreast
    TissueColorectal
    TissueUterus
    TissueProstate
    TissueOvary
    TissueAdrenal Gland
    FluidBlood
  • Preservation Types

    FormalinFrozen
  • Diagnoses

    NeoplasticCarcinoma
  • Platforms

    DNAElectrophoresis
    DNAAutomated electrophoresis/Bioanalyzer
    DNANext generation sequencing
    DNASpectrophotometry
    DNAFluorometry
  • Pre-analytical Factors

    ClassificationPre-Analytical FactorValue(s)
    Biospecimen PreservationType of fixation/preservation
    Formalin (buffered)
    Frozen
    Next generation sequencing SpecificData handling
    Combination
    Mutect
    Shimmer
    Strelka
    Spectrophotometry SpecificTechnology platform
    Qubit