Methanol-based fixation is superior to buffered formalin for next-generation sequencing of DNA from clinical cancer samples

Abstract

Background: Next-generation sequencing (NGS) of tumour samples is a critical component of personalised cancer treatment, but it requires high-quality DNA samples. Routine neutral-buffered formalin (NBF) fixation has detrimental effects on nucleic acids, causing low yields, as well as fragmentation and DNA base changes, leading to significant artefacts.

Patients and methods: We have carried out a detailed comparison of DNA quality from matched samples isolated from high-grade serous ovarian cancers from 16 patients fixed in methanol and NBF. These experiments use tumour fragments and mock biopsies to simulate routine practice, ensuring that results are applicable to standard clinical biopsies.

Results: Using matched snap-frozen tissue as gold standard comparator, we show that methanol-based fixation has significant benefits over NBF, with greater DNA yield, longer fragment size and more accurate copy-number calling using shallow whole-genome sequencing (WGS). These data also provide a new approach to understand and quantify artefactual effects of fixation using non-negative matrix factorisation to analyse mutational spectra from targeted and WGS data.

Conclusion: We strongly recommend the adoption of methanol fixation for sample collection strategies in new clinical trials. This approach is immediately available, is logistically simple and can offer cheaper and more reliable mutation calling than traditional NBF fixation.

Keywords: HGSOC; NBF; SNVs; UMFIX; copy-number abnormalities; fixation; next-generation sequencing.

Figure 1.

Study design. (A) Operative specimens from women undergoing surgery for HGSOC were sampled with a scalpel to acquire three surgical tumour samples and a 16G needle was used to obtain three mock biopsies. Matched surgical and biopsy samples from each case, with matched control tissue, were processed in parallel with fixation in NBF or UMFIX, or SF before downstream analysis. (B) Sample workflow: numbers of patients (P) and samples (S) used for analysis. Bx, biopsy; Tu, surgical tumour fragment; Ctrl, control tissue.

Figure 2.

DNA yield and copy-number calling performance. Box plots show results of PCR assays for DNA size after extraction from SF, NBF and methanol (UMFIX) fixation from matched biopsy and surgical samples from 11 HGSOC patients. (A) Observed ΔCq values for DNA yield of 90 bp fragments (negative ΔCq values are shown for convenience). *P < 0.05, **P < 0.005, ***P < 0.0005. (B) Observed Q ratios for 129 bp/41 bp (top) and 305 bp/41 bp (bottom) fragments. Vertical brackets indicate Wilcoxon rank-sum test for difference in means: *P < 0.05, **P < 0.005, ***P < 0.0005. (C) Scatter plots show correlation between median normalised copy-number profiles from shallow WGS of SF compared with NBF or UMFIX biopsy and surgical samples from 12 patients. Spearman's rank-sum correlation rho is shown. Gray background (orange online) indicates plots with the highest correlation between UMFIX and NBF for each patient sample (biopsy or surgery). (D) Boxplots show an observed variance for each copy-number segment (n = 90 312) in 69 samples from 12 patients.

Figure 3.

Single-nucleotide noise profiles and variant calling performance. (A) Bar plots of the three somatic mutation signatures (S1–S3) identified by non-negative matrix factorisation using all non-reference bases observed in sWGS and TAm-SEQ sequencing data in 69 samples from 12 patients (n = 255 376). Bar plots are grouped by the observed base change with individual bars showing the proportion observed at different trinucleotide sequences. (B) Stacked bar plots show the proportion of the three mutation signatures observed only in SF and NBF or UMFIX fixed samples compared with signatures present in all samples from an individual patient (common). (C) Sensitivity (top) and specificity (bottom) for manually curated SNV calls (n = 546) from TAm-SEQ of biopsy and surgical samples from 11 patients. Bars indicate the 95% confidence interval around the indicted mean.

Figure 4.

H&E staining and IHC scoring. (A) H&E staining of tumour fragments (left) and biopsies (right) from matched tissues fixed in NBF and UMFIX. Bars represent 100 µm. (B) Representative images (left) show IHC staining for p53, PAX8, CK7, WT1 on matched NBF and UMFIX tissues. Quantitative histoscores (middle) of intensity of staining for each IHC marker on tumour fragments [dark gray points (pink online)], biopsies [light gray points (orange online)] and control tissue [black points (blue online)] samples. Spearman's rank-sum correlation rho is shown (P < 0.001 for all analyses). Paired data plots (right) show comparison of median histoscores from paired UMF- and NBF-fixed tissues for each IHC marker. Median scores were only significantly different for WT staining (P = 0.011).