Skip to main content

Comparison of triple-negative breast cancer molecular subtyping using RNA from matched fresh-frozen versus formalin-fixed paraffin-embedded tissue.

Jovanović B, Sheng Q, Seitz RS, Lawrence KD, Morris SW, Thomas LR, Hout DR, Schweitzer BL, Guo Y, Pietenpol JA, Lehmann BD

Publication: BMC Cancer (2017) • Vol 17 • Page 241PMID: 28376728View on PubMed

Purpose

This paper investigated the effects of using formalin-fixed paraffin-embedded (FFPE) rather than frozen specimens and next-generation sequencing (NGS) system (MiSeq versus HiSeq) on NGS quality metrics and assignment of triple negative breast cancer (TNBC) subtype. The concordance of gene expression and TNBC subtype assignment between published microarray and RNA-Seq data in The Cancer Genome Atlas (TCGA) was also examined.

Conclusion

Overall, FFPE specimens performed comparably to frozen specimens with similar numbers of mapped reads and percentage unmapped reads, but FFPE specimens had a larger percentage of transcripts mapping to rRNA, snoRNA, snRNA, intronic regions, processed transcripts, miRNA, pseudogenes, antisense, and lincRNA and a lower percentage of on-target reads. Importantly, hierarchical clustering resulted in specimens clustering predominantly by preservation method with only two frozen specimens clustering with their FFPE counterparts but, when transcripts with differential expression were excluded, clustering was by patient.  The authors attribute the differences between FFPE and frozen specimens to the bias introduced by the poly-A selection process used for frozen but not FFPE specimens.  For specimens over 10 years old, there were significantly more unmapped reads in FFPE specimens than in the matched frozen specimen and there were lower correlations in expression between FFPE and frozen specimens than in specimens stored less than 4 years.

The percentage on-target, intronic, and intragenic reads were comparable between the HiSeq and MiSeq platforms but, because of the increased depth, the correlation in expression between matched FFPE and frozen specimens was significantly higher using HiSeq than MiSeq for all transcripts, including protein coding transcripts and the triple negative breast cancer centroid transcripts.

For specimens with more than 0.2 separation between the two highest subtypes, there was 100% concordance in subtype calls between FFPE and frozen specimens using either HiSeq or MiSeq and >95% concordance in assignment between microarray and RNAseq. A greater than 0.2 separation was achieved more often using HiSeq than MiSeq and using frozen specimens rather than FFPE specimens.   The assigned subtypes were the same when the whole centroid was used and when only the 1685 genes in the centroid that were not differentially expressed between FFPE and frozen specimens were considered, indicating that subtype assignment is not adversely affected by these genes. 

Studies (2)

Purpose

This study compared MiSeq and HiSeq performance in RNA from matched FFPE and frozen TNBC specimens subjected to different processing methods and investigated the effects on TNBC subtype assignment. Extra tissue from twenty-one triple negative breast tumors were obtained from the pathology lab and frozen or formalin-fixed and paraffin-embedded. No details of specimen acquisition or processing were included. RNA was extracted from frozen sections of unspecified thickness using the RNeasy mini kit and enriched for mRNA by using NEBNext Magnetic Oligo d(T)25 Beads. FFPE sections (unspecified thickness) were deparaffinized using the Qiagen deparaffinization solution and extracted using the RNeasy FFPE kit. Ribosomal RNA (rRNA) was removed from FFPE specimens that yielded greater than 0.2 µg RNA using Ribo-Zero Gold before library preparation using ScriptSeq. For FFPE specimens yielding less than 0.2 µg RNA, no rRNA depletion was performed and libraries were prepared using TotalScript. RNA was quantified by Qubit and integrity of RNA from FFPE specimens was evaluated using a Tapestation. Libraries were quantified using the Library Quantification kit and sequenced on an Illumina MiSeq (10 pairs) or HiSeq (11 pairs).

Summary of Findings

The number of mapped reads and the percentage of unmapped reads were comparable in frozen and FFPE specimens but, when only specimens stored for >10 years were examined, the percentage of unmapped reads was higher in FFPE than matched frozen specimens (13% versus 4%, P=0.0066). The percentage on-target, intronic, and intragenic reads were comparable between the HiSeq and MiSeq platforms but frozen specimens had a much higher percentage on-target reads than FFPE specimens (82% versus 28%, P<0.0001 on MiSeq and 86% versus 34%, P<0.0001 on HiSeq) and lower percentage of reads that aligned to introns (10.7% versus 58.4%, P<0.0001 on MiSeq and 14% versus 60%, P<0.0001 on HiSeq) or intragenic regions (3% versus 9%, P=0.0187 on MiSeq and 3% versus 8%, P=0.001 on HiSeq). The authors state these differences stem from the poly-A capture used to enrich frozen specimens. The correlation in expression between matched FFPE and frozen specimens was significantly higher using HiSeq than for MiSeq for all transcripts, including protein coding transcripts and the triple negative breast cancer centroid transcripts (P<0.001, all). Correlations in expression between FFPE and frozen specimens were lower for specimens >10 years old than specimens that were less than 4 years old (0.78 versus 0.88, P=0.0014) and the difference in correlations between platforms disappeared for specimens that were less than 4 years old. In hierarchical clustering, specimens clustered predominantly by preservation method with only two frozen specimens clustering with their FFPE counterparts. Removal of the transcripts that were differentially expressed between FFPE and frozen resulted in hierarchical clustering predominantly by specimen rather than by preservation method. The authors note that the majority of differentially expressed transcripts can be attributed to differences in the methodology, as the poly-A selection method used for frozen specimens enriches for coding transcripts and selects against noncoding RNA species. This is demonstrated by only small increases in the number of protein coding transcripts expressed predominantly in frozen versus FFPE tissue (52.5% versus 47.5%) and by the much larger percentage of transcripts mapping to rRNA, snoRNA, snRNA, intronic regions, processed transcripts, miRNA, pseudogenes, antisense, and lincRNA in FFPE than frozen specimens (93.5%-100% versus 0-6.5%). Pathway analysis showed some enrichment for plasma membrane protein transcripts in FFPE specimens. The authors found that the enriched proteins tended to be long and those that were differentially-expressed, were, on average, longer in FFPE specimens than in frozen specimens (6309 versus 3770 bp, P<2.2e-16). Using two long genes, the authors found a 3’ selection bias in frozen specimens that is believed to be introduced by the poly-A selection.

There was only 67% concordance in TNBC subtype between FFPE and frozen specimens using HiSeq and only 58% concordance using MiSeq. However, concordant subtypes for matched FFPE and frozen specimens were always assigned when a more than 0.2 separation existed between the two highest subtypes on HiSeq (6 of 11 specimens) and MiSeq (3 of 10 specimens). The correlations to predicted subtypes were greatest for frozen specimens sequenced using HiSeq and lowest for FFPE specimens sequenced with MiSeq. The assigned subtypes were the same when the whole centroid was used and when only the 1685 genes in the centroid that were not differentially expressed between FFPE and frozen specimens were considered, indicating that subtype assignment is not adversely affected by these genes.

  • Biospecimens

    TissueBreast
  • Preservation Types

    FrozenFormalin
  • Diagnoses

    NeoplasticCarcinoma
  • Platforms

    RNANext generation sequencing
    RNAFluorometry
  • Pre-analytical Factors

    ClassificationPre-Analytical FactorValue(s)
    Biospecimen PreservationType of fixation/preservation
    Formalin (buffered)
    Frozen
    StorageStorage duration
    <4 years
    >10 years
    Next generation sequencing SpecificTechnology platform
    HiSeq
    MiSeq
    Next generation sequencing SpecificTemplate modification
    FFPE specimens rRNA depleted
    Frozen specimens Poly-A selected