This paper compared data obtained from different gene expression and single nucleotide variant (SNV) discovery technologies (Nanostring, whole transcriptome sequencing) among and between case-matched breast cancer specimens preserved by snap-freezing and formalin-fixation and paraffin-embedding.
Gene expression, single nucleotide variant and fusion transcript discovery in archival material from breast tumors.
Norton N, Sun Z, Asmann YW, Serie DJ, Necela BM, Bhagwate A, Jen J, Eckloff BW, Kalari KR, Thompson KJ, Carr JM, Kachergus JM, Geiger XJ, Perez EA, Thompson EA
Purpose
Conclusion
Gene expression levels in case-matched formalin-fixed paraffin-embedded (FFPE) and frozen specimens were strongly to very strongly correlated when analyzed by NanoString and modestly to strongly correlated by whole transcriptome sequencing of ScriptSeq2 libraries. Although frozen and FFPE specimens had a comparable percentage of overall mapped reads and ratio of known transition to transversion mutations, FFPE specimens displayed lower average RNA integrity numbers (RIN), percentages of reads mapped to gene or junctions, and reduced detection of fusion transcripts and SNVs, but a higher ratio of novel transition to transversion mutations. Interestingly, correlation coefficients for expression data generated with NanoString and NGS of ScriptSeq2 libraries were higher for FFPE than frozen specimens.
Studies (2)
Purpose
This study compared data obtained from different gene expression and SNV discovery technologies (Nanostring, whole transcriptome sequencing) among and between case-matched snap-frozen and FFPE breast cancer specimens. Case-matched frozen and FFPE specimens were generated from nine breast tumor resections. Frozen specimens were snap-frozen in liquid nitrogen and stored at -80˚C. Additional details on FFPE processing were not provided. RNA was extracted from using the Qiagen miRNeasy kit and from six 10 µm FFPE sections using the Qiagen AllPrep FFPE kit. Gene expression in FFPE and frozen specimens was quantified using the NanoString Cancer Gene expression panel and by whole transcriptome sequencing of rRNA-depleted ScriptSeq V2 libraries.
Summary of Findings
Although mean RNA yield from FFPE and frozen specimens were comparable (7.99 µg versus 7.75 µg, respectively), RIN scores from FFPE specimens were lower than those from matched frozen specimens (1.8-2.3 versus 6.8-8.9, respectively). Expression levels in matched FFPE and frozen specimens were strongly to very strongly correlated by NanoString (r=0.728-0.973, P<2x10-16) but only modestly to strongly correlated by whole transcriptome sequencing of ScriptSeq2 libraries (r=0.598-0.830, P< 2x 10-16). When expression profiles generated by NanoString and NGS of ScriptSeq2 libraries were compared by Pearson's correlation, a strong correlation was observed for the two methods among FFPE specimens (0.838, range 0.468-0.924), while a modest correlation was observed among snap-frozen specimens (0.664, range 0.372-0.827). When the same expression data were compared by Spearman's correlation, correlation coefficients were more consistent among both FFPE (0.734, range 0.707-0.771) and frozen (0.731, range 0.707-0.762) specimens.
NGS of ScriptSeq2 libraries revealed the percentage of reads mapping to the genome were comparable for FFPE and frozen specimens (79.2% versus 81.3%), but frozen specimens demonstrated a higher percentage of reads mapping to genes or junctions than FFPE specimens (50.6% versus 24.8% and 10.29% versus 3.98%, respectively). Expression of long Intergenic non-coding RNAs (lncRNAs) was strongly correlated between matched FFPE and frozen specimens (r=0.969-0.997), but the percentage of reads mapping to lncRNAs was higher for FFPE than frozen specimens (mean 5.47% versus 3.58%).oOf 32 fusion transcripts, 22 were found only in frozen specimens and two were found only in FFPE specimens, indicating FFPE-specific exon loss; an effect that was not able to be mitigated by increasing the read count by 3-fold.
Libraries generated from snap-frozen specimen yielded significantly higher mean detection of total SNV (180,000 versus 115,000) and a higher percentage of non-exonic SNV (76.2% versus 53.4%) than libraries generated from FFPE specimens. Although the ratio of transition to transversion mutations were comparable among FFPE and snap-frozen specimens for known SNVs (snap-frozen: mean=2.15, range=2.06-2.27; FFPE: mean=2.21, range=2.17-2.25), the same ratio for novel SNVs was higher among FFPE than frozen specimens (FFPE: mean=2.23,range=2.08-2.39; snap-frozen: mean=1.81, range=1.52-2.39), indicating that transversion mutations may be introduced during FFPE fixation and processing. Only 39-61% of SNVs identified in case-matched frozen specimens were also observed in FFPE specimens. Importantly, the percentage of SNVs identified in FFPE specimens was strongly and positively correlated with library insert size (R=0.73).
Biospecimens
TissueBreastPreservation Types
FormalinFrozenDiagnoses
NeoplasticCarcinomaPlatforms
RNAAutomated electrophoresis/BioanalyzerRNANext generation sequencingRNADNA microarrayPre-analytical Factors
Classification Pre-Analytical Factor Value(s) Biospecimen Preservation Type of fixation/preservation Formalin (buffered)Snap frozenNext generation sequencing Specific Technology platform NanoString
Purpose
The purpose of this study was to compare two different library construction techniques, rRNA RiboZeroGold/ScriptSeq2 versus TruSeq, based upon gene expression and SNV detection data generated using NGS. Expression data obtained by NGS was compared to NanoString generated data by Pearson's correlation. Nine breast tumor resections were snap-frozen in liquid nitrogen and then stored at -80˚C. RNA was extracted using the Qiagen miRNeasy kit. Gene expression and SNV discovery were conducted by whole transcriptome sequencing of RiboZero Gold rRNA-depleted ScriptSeq V2 libraries and TruSeq V2 libraries.
Summary of Findings
When Expression profiles generated by NanoString were compared to NGS results obtained with TruSeq libraries correlations between the two technologies were strong (r=0.790, range= 0.558-0.922), but NanoString and NGS were only modestly correlated when NGS results were obtained with ScriptSeq2 libraries (r=0.664, range= 0.372-0.827) . Further, the correlation between the two NGS platforms was modest (r=0.680, range =0.579-0.8212). While the percentage of reads mapped to the genome were comparable for NGS of RiboZeroGold/ScriptSeq2 and TruSeq libraries (81.3% versus 80.8%), TruSeq libraries demonstrated a higher percentage of reads mapped to genes or junctions than RiboZeroGold/ScriptSeq libraries (77.4% versus 50.6% and 15.8% versus 10.29%, respectively). A comparison of gene expression profiles between library construction methods revealed 116 genes that were differentially expressed, 107 (92.2%) of which were more highly expressed in the RiboZeroGold/ScriptSeq libraries. Of these, 23.4% were histone H1 cluster genes and 26.2% were small nucleolar RNA and small Cajal body-specific RNAs. Expression of lncRNAs were only weakly to modestly correlated (r=0.355-0.486) among library construction methods, with a larger percentage of RiboZeroGold/ScriptSeq than TruSeq reads mapping to LincRNA (2.63-4.89% versus 1.10-1.85%). Of the lncRNAs that were differentially expressed in the two libraries, 179 lncRNA were more highly expressed in the RiboZeroGold/ScriptSeq library and 72 more highly expressed in TruSeq libraries; additionally, longer transcripts were more likely to have a higher expression in RiboZeroGold/ScriptSeq libraries. Twenty-five of the 43 fusion transcripts identified in frozen specimens were observed in both libraries, 13 were only observed in TruSeq libraries, and 5 were only observed in RiboZeroGold/ScriptSeq libraries. RiboZeroGold/ScriptSeq libraries had a significantly higher number of SNV calls and a slightly lower ratio of transition to transversion mutations than TruSeq libraries, both for known mutations (RiboZeroGold/ScriptSeq: mean=2.15, range=2.06-2.27; TruSeq: mean=2.44, range=2.4-2.51) and novel mutations (RiboZeroGold/ScriptSeq: mean=1.81, range=1.52-2.39; TruSeq: mean=1.91, range=1.51-2.39).
Biospecimens
TissueBreastPreservation Types
FrozenDiagnoses
NeoplasticCarcinomaPlatforms
RNANext generation sequencingRNADNA microarrayPre-analytical Factors
Classification Pre-Analytical Factor Value(s) Next generation sequencing Specific Technology platform NanoStringRiboZeroGold/ScriptSeqTruSeq