Retrospective evaluation of gene signatures using archived FFPE samples can be an important capability for toxicology. Testing the accuracy and specificity of gene signatures formulated by predictive models for carcinogenicity or screening for new or hypothesis-driven transcripts would be facilitated by the use of archival samples. In this study, we demonstrated that molecular analyses could be performed on RNA extracted and amplified from FFPE rat tissues preserved for four years. Critical features of the process included deparaffinization, protein digestion, silica-gel isolation, and DNase I digestion to isolate degraded but analytically useful RNA. Use of random hexamers for reverse transcriptase reactions and primers for relatively short amplicons were important steps in using degraded RNA as a template for amplification in qPCR.7, 48
qPCR has been used to obtain gene expression data from FFPE samples from different tissue types and organs, often with tumor specimens to improve prognostic accuracy in clinical care.7, 12, 15, 49
For example, RNA from FFPE samples of 153 breast cancer patients were used to develop an 8-gene qPCR gene expression score of prognostic value that could be reasonably implemented in clinical settings without the need for fresh frozen RNA or more costly commercial signature arrays.7
One recent study identified a 14 gene expression signature using qPCR with FFPE RNA from a 361 patient cohort to predict survival in nonsquamous, non-small-cell lung cancer.50
Other investigations have also defined classifiers of disease progression or prognostic markers for invasive bladder cancer (eight genes)51
or squamous cell carcinoma of lung (12 genes)52
by qPCR analysis after gene selection from extensive microarray data sets, albeit from fresh-frozen materials. Importantly, the principle of qPCR-based screening of archival samples for focused sets of gene expression targets can similarly be exploited in toxicology studies.
Here, we tested a focused gene signature comprised of 14 transcripts and found agreement in direction and proportional change in transcript expression for Adam8, Mybl2, Akr7a3, Chd13, C8orf46 homolog, Ddit4l, Fhit, Abcc3, Abcb1b, Akr7a2 and Gst5a in FFPE samples compared to fresh frozen and microarray formats. Further studies will be needed to determine how the transcripts in this expression pattern might be of prognostic value for chemically-induced rat liver tumors. Ideally, a gene signature can be formulated whose combined expression pattern can accurately predict the occurrence of hepatocellular carcinoma in premalignant tissue in rodents. Such attempts are already underway by preclinical safety testing consortiums; one such study used data generated from more than 900 study-diverse liver samples in studies involving 66 various compound exposures to predict non-genotoxic heptocarcinogens from a 22-gene signature qPCR array platform.53
Signature specificity and sensitivity for predicting nongenotoxic carcinogens were 67% and 59%, respectively, in which compounds were best classified with expression data from short-term repeat dose exposures. Among their findings was that modes of action for nongenotoxic and genotoxic compound could be discriminated based on the expression of specific genes.
The selection of transcripts for qPCR testing in our FFPE samples was based on informative genes from modeling of liver microarray data and genes responding to AFB1 exposure. These included probes representing nine upregulated (>3 fold), three moderately upregulated (2–3 fold) and two downregulated (<3 fold) transcripts. While eleven genes were concordant across platforms, three transcripts showed some variance, including Grin2c (upregulated), Wwox (moderately upregulated), and Cxlc1 (down regulated). Cxcl1 (GRO/Cinc-1) is a rat chemokine for neutrophils (similar to IL-8) that was reduced about 2 to 3 fold by microarray and qPCR, respectively, using RNA from fresh frozen tissue but it increased slightly above control in FFPE liver (1.2 fold). Since no significant differences in Cxcl1 were found among the three expression platforms, we would attribute these discrepancies to a minimal immune response in liver to AFB1 exposure. For Wwox, this transcript was generally lowered by hepatocarcinogen models17
but showed a marginal increase of 1.3 fold in AFB1-treated rats by microarray analysis compared to a slightly decreased expression by qPCR of RNA from fresh frozen liver (1.6 fold decrease) and no change in FFPE liver (1.03 fold decrease) from the same animals. The reason for this discordance is not known but it may either reflect regional differences within the liver or that such small expression differences cannot be discriminated by qPCR. Finally, the transcript for the glutamate receptor subunit Grin2c was significantly increased by AFB1 in microarray analysis (18.1 fold increase) and by qPCR of RNA from FFPE sampling (4.5 fold increase) but could not be similarly reproduced using RNA from fresh frozen tissue (4.5 fold reduction) despite testing multiple qPCR primer sets across the length of the transcript including the 3′ portion. The length of the Grinc2c transcript is 4388 nt and the microarray probe was complementary to the 3′-end of the transcript. Our qPCR data indicated Grin2C was a relatively low expression transcript (Ct values of 27–32 for various primer sets). We used the same RNA extracted from fresh frozen tissue for qPCR that was used for microarray analysis, that was of high quality with intact 28S and 18S peaks and that showed concordance of all transcripts except Wwox and Grin2c for these two platforms (). We surmise that varying results for Grin2C might relate to differences in amplification efficiencies between microarray and qPCR procedures for this particularly long transcript of low copy number. Based on our limited gene set, retrospective testing of FFPE samples for generating gene signatures might include (where possible) moderately well expressed genes that are not inordinately large in length and at least a 2–3 fold expression difference from reference groups. We also recognize that amplification of low copy number transcripts can be particularly challenging with suboptimal RNA which can be addressed by using larger input amounts of RNA, mRNA isolation, use of gene-specific primers (instead of random hexamers) or nested primer sets. Use of gene signatures comprised of multiple transcripts chosen from the numerous critical biological pathways altered by chemical exposure can minimize reliance on single transcript biomarkers.54
Furthermore, qPCR screening of archival training sets of multiple transcripts to remove technically or biologically variable transcripts is common practice in designing a robust gene signature prior to application in test samples.50
In our study, nucleic acid preservation took place under near optimal conditions in which 24hr formalin fixation was immediately followed by dehydration and paraffin embedding. The stability of the nucleic acids is dependent on fixation time in NBF and other variables such as pH, temperature, thickness of the tissues, type of tissues, tissue quality and isolation method, penetration of the fixation buffer into the tissue and efficiency of protein cross-linking.1, 15, 48, 55
Among these factors, sample duration in formalin is likely of highest importance since increased tissue crosslinking occurs during prolonged fixation and reduces RNA yield and quality.2
One study found that the yield of extractable RNA and DNA of human FFPE liver samples fixed in NBF from 2 weeks to 8 months was usable for qPCR but signal diminished with long fixation times as did RNA fragment size.56
Once tissues are embedded in paraffin, nucleic acids and protein are relatively well maintained over time. However, RNA from archival FFPE breast cancer tissue isolated after six years was less degraded and of larger molecular weight than RNA isolated after seventeen years, suggesting that some nucleic acid fragmentation might still occur in paraffin blocks. Until such time that newer methods of tissue fixation without formaldehyde57, 58
come into routine use for improved perservation of nucleic acids and cell structure for histology, the available data indicates that minimizing formalin-based fixation time before paraffin embedding is extremely useful for histopathology and downstream molecular analyses. Furthermore, the precise amount of time that tissues spend in fixative prior to parffin embedding is valuable metadata worth capturing for future archival studies.
After oral AFB1 exposure, gene expression changes were most evident in liver, the primary organ for AFB1 activation and conjugation. Even though kidney is a secondary target for AFB1 carcinogenesis18
, only Abcb1b was significantly increased in this organ. This observation suggests the xenosensor activity of Abcb1b was likely responding to AFB1 or its metabolites, but we speculate there was insufficient toxicity at this exposure (90 days, 1 ppm) to induce detoxication genes such as Akr7a3, Gst5a or other transcripts in the signature. In addition, it is possible that a microarray analysis of kidney might show expression changes in other transcripts not measured here, given availability of fresh frozen RNA.
In lung, we found relatively low and inconsistent gene expression from extracted FFPE lung RNA even though the RIN values were comparable among all three tissues. We considered whether genes in the signature were not expressed in rat lung and therefore not detectable. However, a query of a curated gene expression database (http://www.nextbio.com
) suggested that each transcript could be detected under various treatment conditions (Supplemental Table 1
), although we cannot completely discount that relatively lower copy numbers in lung compared to liver and kidney could make detection of some transcripts more challenging. A more attractive explanation for the inconsistent qPCR data observed from lung was the comparatively poor RNA quality. The assessment of RNA quality by the RIN value takes into account the fractional area of 18S and 28S regions compared to total, as well as the heights of 28S (most degradation-sensitive) and 18S peaks and other factors in deriving a score that ranges from 1 to 10 (1 is the most degraded profile and 10 is the most intact).59
RIN values of 2 for FFPE samples as found in our study indicate a loss of both 18S and 28S peaks and accumulation of low molecular weight RNA species.59
However, we found the RIN value was not as discriminating for downstream analysis of FFPE RNA as the size range of RNA since liver and kidney contained RNA sizes of 2000–4000 nt which gave more consistent qPCR data than the smaller range of lung (≤500 nt). Freidin et al60
described poor gene expression in FFPE lung samples compared to robust profiling from RNA of fresh or frozen lung samples. Others have found that prolonged NBF fixation (>24 hr) of lung tissue reduces RT-PCR success.58
These reports and the less robust qPCR data from lung in our study suggest special care may be needed for archival preservation of this organ for later molecular analysis. Our observations suggest that expression of housekeeping genes like β-actin, S18 or others provide a functional readout of FFPE RNA integrity and, in combination with molecular sizing, are informative criteria to judge the usefulness of specific archival specimens.
One finding of interest was that amplification products up to 500 nt in length could be formed from cDNA of Akr7a3 and Ddit4l transcripts in FFPE RNA that could be cloned and sequenced. This capability could be of value in querying archival samples to detect specific mutations, synonomous substitutions as we report for Ddit4l, polymorphisms, splice variants, or other alterations in transcript expression. The target amplicon size for successful amplification in FFPE RNA samples is partly a function of FFPE RNA quality, median length of RNA fragments and relative cellular expression level of the specific transcript of interest. For example, the Akr7a3 is a well-expressed gene which could be cloned up to 499 nt (~500 nt) in our experiments although this band was not as intense as bands from amplicons 200 to 300 nt in length. The upper limit for amplification and cloning might exceed 500 nt for even more highly expressed transcripts and conversely might be somewhat less for lower expressed genes. Although we did not try additional measures to improve amplification of FFPE RNA, others have found that heating RNA at 70°C for 1 hr in Tris-EDTA buffer prior to RT-PCR removed almost all methylol groups which restored template integrity and PCR amplification to produce longer amplicons than without heating.61
As previously mentioned, the general integrity, quality, extractability and fate of nucleic acids will decline with increasing time spent in formaldehyde solution. Inclusion of neutral buffer in 4% formalin solutions will slow but not stop the process of nucleic acid degradation. While the tissue processing steps prior to paraffin embedding are not without hazard to cellular structures and biomolecules, nucleic acids are better preserved in FFPE blocked tissues over time than in formalin solutions. However, there are many reasons aside from commitment of resources that not all preserved clinical and preclinical tissues are reduced to paraffin blocks. A less obvious but important reason is to maintain the structural integrity of the specimen for the flexibility of choosing the cut surface for histologic examination at some later time. Use of advanced, non-formaldehyde fixatives will provide even further flexibility to molecular toxicology in the future.