Comparison of clinical targeted next-generation sequence data from formalin-fixed and fresh-frozen tissue specimens

Abstract

Next-generation sequencing (NGS) has emerged as a powerful technique for the detection of genetic variants in the clinical laboratory. NGS can be performed using DNA from FFPE tissue, but it is unknown whether such specimens are truly equivalent to unfixed tissue for NGS applications. To address this question, we performed hybridization-capture enrichment and multiplexed Illumina NGS for 27 cancer-related genes using DNA from 16 paired fresh-frozen and routine FFPE lung adenocarcinoma specimens and conducted extensive comparisons between the sequence data from each sample type. This analysis revealed small but detectable differences between FFPE and frozen samples. Compared with frozen samples, NGS data from FFPE samples had smaller library insert sizes, greater coverage variability, and an increase in C to T transitions that was most pronounced at CpG dinucleotides, suggesting interplay between DNA methylation and formalin-induced changes; however, the error rate, library complexity, enrichment performance, and coverage statistics were not significantly different. Comparison of base calls between paired samples demonstrated concordances of >99.99%, with 96.8% agreement in the single-nucleotide variants detected and >98% accuracy of NGS data when compared with genotypes from an orthogonal single-nucleotide polymorphism array platform. This study demonstrates that routine processing of FFPE samples has a detectable but negligible effect on NGS data and that these samples can be a reliable substrate for clinical NGS testing.

Figure 1

Experimental overview. A: Tissue from 16 collected lung adenocarcinomas was selected from the Siteman Cancer Center tumor bank. For each carcinoma, paired fresh frozen and routinely processed (overnight fixation with 10% w/w buffered formalin) FFPE samples were available. Slides from both fresh frozen and FFPE tissue were reviewed for adequacy. The mean age of patients at the time of sequencing was 8.1 years. DNA was extracted from both fresh frozen and FFPE tissue using the same method. FFPE-derived DNA was subjected to a PCR size control ladder assay to determine the extent of DNA fragmentation. A subset (eight cases) of fresh tissue–derived DNA was further analyzed by the Affymetrix V6 SNP array. Extracted DNA was then sheared, indexed, adapter ligated, and captured using WU-CaMP27 gene enrichment probes that target 27 commonly mutated genes in cancer. Enriched DNA libraries were then subjected to low-cycle PCR amplification (eight cycles) and sequenced on an Illumina HiSeq 2000 instrument using V3 chemistry and 2 × 101-bp paired end reads. All NGS data were aligned to the National Center for Biotechnology Information build 37 (hg19) reference using Novoalign in paired-end mode. Variants were called using the GATK version 2.1. B: To evaluate the effect of preanalytic factors on NGS, we subjected fresh tissue to both prolonged formalin fixation and prolonged ischemic time. DNA from these samples was extracted, processed, analyzed as above, and finally compared with a fresh tissue time point.

Figure 2

The distribution of library insert sizes for reads from FFPE and frozen samples. The distribution of insert sizes from both sample types, determined from the distance between properly mapped forward and reverse read pairs, demonstrating smaller insert sizes in FFPE samples due to formalin-induced DNA fragmentation.

Figure 3

Coverage statistics for FFPE and frozen NGS data. A: Distribution of mean coverage for the total target region for 16 frozen (red) and 16 FFPE (purple) samples in unique reads with mapping quality >20. The box plot shows the median (solid line), interquartile range (IQR) (colored bar), and range of the mean coverage by sample type. No significant difference was observed between the sample types (P = 0.26). B: Coverage uniformity, determined by calculating the fraction of targeted positions at various coverage thresholds. Bar height indicates the mean proportion of positions at the indicated coverage level for frozen and FFPE samples; error bars indicate the SD. No significant differences occurred at any coverage level. C: Distribution of mean coverage in high-quality bases by gene and sample type. Box plots show median, IQR, and range. No significant differences between FFPE and frozen samples. D: Mean coverage and IQR in the mutational hotspot region spanning KRAS codons 12 and 13 (depicted) for FFPE (red) and frozen (purple) samples. Solid lines depict the median for each set of samples.

Figure 4

Comparison of SNV calls. The overall concordance between FFPE and frozen tissue was >99.99% for all 300,000 sequenced positions. A comparison of variant calls (depicted above) demonstrated that 4281 variant calls (97%) agreed between FFPE and frozen samples; 56 calls (1.3%) were unique to frozen samples, and 70 calls (1.8%) were unique to FFPE. A comparison to Affymetrix SNP6 arrays (eight cases) demonstrated that 533 of the 550 variant calls (98%) were concordant among all three methods.

Figure 5

Observed spectrum of high-quality base changes in FFPE and frozen NGS data. A: Distributions of the mean transition and transversion frequency for frozen (red) versus FFPE (purple) samples among high-quality discrepancies (see Materials and Methods). The ratio of transitions to transversions was significantly larger in FFPE compared with fresh tissue (P = 6.4 × 10⁻⁸). B: Distributions of the mean frequencies for each possible base change for each sample type. Only C to T and G to A transitions were significantly different between fresh and frozen samples (P = 3.9 × 10⁻¹⁰ and 1.2 × 10⁻⁹, respectively). In both panels, box plots display the median and interquartile range (IQR) of the per-sample mean frequency for each base change by sample type, with whiskers extending to the last data point within 1.5 times the IQR and outliers indicated by circles.

Figure 6

Effect of antecedent and subsequent base on C to T transitions. Distributions of the mean C to T transition frequency with different antecedent bases (A) and different subsequent bases (B) by sample type. All C to T dinucleotide conversions occur at significantly higher rates in FFPE compared with frozen tissue. The conversion frequency is independent of the antecedent base (A) but is much larger for the case of CG to TG conversions (B) (mean odds ratio = 0.19, 95% CI, 0.18 to 0.21) (see Supplemental Table S5). In both panels, box plots show the median and interquartile range (IQR) of the per sample frequencies, with whiskers extending to the last data point within 1.5 times the IQR.

Figure 7

Effect of preanalytic factors on NGS. A: Percentage of mapped, on-target, and unique reads for a single surgical resection specimen that was snap frozen (red) or subjected to 24-hour (light purple), 48-hour (medium purple), or 72-hour (dark purple) formalin fixation. Percentage of mapped, on-target, and unique reads for a single surgical resection specimen that was snap frozen (red) or subjected to 24 (light purple) or 48 (medium purple) hours of ischemia followed by routine processing. B: The fraction of positions at unique on-target coverage levels between ×50 and ×1000 for the frozen specimen (red) or after formalin fixation for 24 (light purple), 48 (medium purple), or 72 (dark purple) hours. The fraction of positions at unique on-target coverage levels between ×50 and ×1000 for the frozen specimen (red) or after a 24-hour (light purple) or 48-hour (medium purple) ischemic time. C: The coverage coefficient of variation (bar plot, left scale) for snap frozen (red) compared with tissue after 24, 48, or 72 hours of formalin fixation. The superimposed plot (right scale) shows the percentage of SNVs detected in each sample. D: The coverage coefficient of variation (bar plot, left scale) for snap frozen (red) compared with tissue after a 24- or 48-hour ischemic time. The superimposed plot (right scale) shows the percentage of SNVs detected in each sample.