Introduction

With the introduction of new genomic technologies such as tissue-based RNA microarrays, patterns of gene expression identified by microarray analyses have been discovered that stratify tumors and predict the clinical outcomes in different cancer types [1]. Some of them have been successfully used for identifying patients that can benefit of specific treatment and FDA (Food and Drug Administration) -approved tests based on breast cancer gene signatures are now commercially available. As a consequence, collecting in tissue banks surgical specimens that can be used for these analyses has become a mandatory issue for understanding the correlative results of the majority of current clinical trials. However, the variability in tissue handling and processing of surgical specimens may affect the reproducibility and interpretation of results. Several variables, including tissue manipulation, warm ex-vivo ischemia and storage times can potentially alter mRNA expression levels and adversely affect the validity of studies that used clinical specimens [2]. All these variables need to be carefully investigated in order to set up guidelines for tissue banking [3]. The organization of qualified and certified biobanks should be the basis to guarantee networking of activities and availability of quality-certified biological material.

The purpose of this pilot study was to explore the effects of tissue handling time in precisely documented tissue samples that followed the routine processing standards in our Institution (Fondazione IRCCS Istituto Nazionale dei Tumori, INT-MI), for developing a clinically applicable method of sampling tumors in tissue banks that can be safely used for microarray analyses. We used the colorectal cancer (CRC) model. CRC is the second most frequent cause of cancer death in Western countries and despite significant advances in its management, the overall survival for advanced and metastatic disease has changed little over the last 20 years, with five years at almost 90% for early and 15% for late tumors [4]. As a consequence, there is a pressing need for new biomarkers to improve the detection and the clinical treatment of CRC. Using high-density oligonucleotide microarrays we investigated sequential effects of tissue handling in specimens obtained from 15 CRCs and in their matched normal tissue collected at our Institution and left at room temperature at different time points after surgery. The primary study outcome was to evaluate the effect of the time on tumor samples and possibly select specific genes whose expression is time-related, that could be used as detectors of tissue degradation. Additionally, to identify genes influenced by time irrespectively of the sample type (normal or tumor), we also investigated the time effect in normal tissues and compared the differential expression between the two tissue types accounting for time.

Our results show that the impact of tissue handling time on the whole gene expression profile could be considered minor in colon and that a reasonable threshold for collecting specimens could be 60 minutes after surgery, when no gene alterations were observed. Such samples may be used to generate reproducible microarray profiles to aid treatment decision making utilizing clinicogenomic models.

Materials and Methods

2 Study Design and Sample Handling

The tumor and the normal counter-part samples used in the experiments were prospectively collected from 15 patients who underwent surgical resection at the INT-MI and whose tumors were representative of the different pathologic stages of this tumor type. At the histological routinely examination all tumor specimens were classified moderately differentiated colonic Adenocarcinomas NOS (grade G2 according to the American Joint Committee on Cancer 2010 http://www.cancerstaging.org/). The clinico-pathologic and histological details are listed in Table 1. Six fragments from each sample were obtained and were randomly left at room temperature at different time points as follows: three fragments at <20 min. (T0), one fragment at 60 min. (T1), one fragment at 180 min. (T2) and one fragment at 360 min. (T3). Time was measured starting from patient’s surgical excision and the first timepoint analysed (T0) was processed and frozen within 20 min. from surgery. Samples were transported from theater to pathology at room temperature without ice. The total number of analyzed fragments was 180 (90 normal samples and 90 tumor samples). Neoplastic samples were obtained from the central area of the neoplasia, avoiding selecting necrotic material or transition zones with healthy mucosa. Samples of colonic healthy mucosa were resected at least 20 centimeters far from the neoplasia and distant from the surgical resection margins. The control was costituited exclusively by normal mucosa, stripped from the luminal surface.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. Characteristics of the cases sampled for RNA isolation.

https://doi.org/10.1371/journal.pone.0053406.t001

Results

1 RIN Analysis

The RNA Integrity Number (RIN) of all samples over all time points had a median value of 5.8 with an inter-quantile range of 1.95. Tumor samples had on average a higher RIN value than the normal samples (median of 6.4 and 5.35 respectively), especially at the first time point T0. Similar distributions of RIN values could be observed across the different time points and considerable variation was observed for each sample within the first time point T0 (RIN variance median = 1.668, IQR = 2.084) which was more pronounced in the normal samples. The time course RIN analysis showed that RIN values had a tendency to decrease over time (overall p-value for Time = 0.0565 and for single contrasts compared to T0, T1 = 0.0214, T2 = 0.1705 and T3 = 0.0529). The distribution of RIN values differ significantly (p-value = 0.0008) in Tumor samples with respect to the Normal ones (Fig. 1).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Figure 1. RIN analysis.

RIN distributions at each time point for both tumor (A) and normal (B) samples. A higher RIN indicate a higher RNA quality.

https://doi.org/10.1371/journal.pone.0053406.g001

By considering all the four time points we identified a subset of 6 normal and 7 tumor cases the RIN values of which was ever higher than five and time course expression analysis was performed also for this subset of samples. It should be stressed that no association was observed between gene profiles on tumor samples at T0 and stage (p-value = 0.59) or location (p-value = 0.95) of the disease.

2 Time Course Expression Analysis

After the quality control of microarray data, the expression profiles of 4 samples (2 profiles from tumor and 2 from normal samples) were removed. These outliers belonged to different time points of two patients in the case of normal samples and to the same patient for tumor samples. Due to the nature of the study design it was crucial to have for each patients a full pictures across all the time points and thus all profiles associated with those patients were not take into consideration in the time course analysis. As a consequence, a total of 13 patients were used in the Normal dataset which contained 17895 probes and 14 in the Tumor dataset containing 18007 probes. Both datasets shared 16698 common probes.

The probes identified by the stringent Bonferroni correction (p-value <0.05) method in the Time factor or in at least one of the time contrasts (T0 baseline) in the Normal and Tumor datasets are listed in Table 2 (N = 32) and Table 3 (N = 31) respectively. In these tables, the rows have been ordered according to the raw p-values deriving from the Time factor and a star (*) under the columns ‘Time’, ‘T1’, ‘T2’, or ‘T3’ indicates from which contrast the probe was selected. Heatmaps in the panel A and B of Figure S1 graphically represent the direction of the expression alteration at each time point in the Tumor and Normal datasets, respectively.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 2. Top candidates identified in the Normal dataset (Bonferroni P<0.05).

https://doi.org/10.1371/journal.pone.0053406.t002

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 3. Top candidates identified in the Tumor dataset (Bonferroni P<0.05).

https://doi.org/10.1371/journal.pone.0053406.t003

Five probes (genes) exhibiting different means of expression across the four time points were identified both in the Normal and the Tumor datasets (HIST1H1D, HIST1H1E, HIST1H4E, HIST4H4 and 5960086). The majority of the variability observed derives from the last time point T3 at 360 minutes when comparing to the baseline T0. This is especially true in the Tumor dataset whereas some probes deriving from the T2 contrast emerge in the Normal dataset. All genes identified in the Tumor dataset are consistently up-regulated over time; the same applies to the Normal dataset with a few exceptions. On the whole however, most genes do not exhibit drastic changes in levels of expression. Functional annotation of the genes from the two lists (using both Gene Ontology and KEGG pathways databases) highlighted that 22 of the 32 genes modulated in normal tissue were histones, involved in nucleasome organization and chromatin assembly. Four of them were also time-dependent in the tumor samples, together with other 27 genes with various functions involved in cancer, including oncogenes, such as JUN, FOSB, ABL1 and EGR1. Another gene commonly modulated in Normal and Tumor tissues was RNU11, a small nuclear RNA.

An identical analysis was performed using the subset of samples with a RIN higher than five which identified only 6 probes in the Normal dataset and 6 in the Tumor dataset, dataset probably because of the reduced sample size. No common probes where found in the two lists. Five of 6 probes (HIST1H1E, HIST1H4B, HIST1H4E, HIST4H4, 5960086) in the Normal dataset were also present in the analysis using all samples and 2 (HSPA1A, IER5) were in common in Tumor dataset.

3 RT-PCR Validation

The three genes that were significant for ‘Time’, ‘T2’ and the ‘T3’ contrasts in the Tumor samples (JUN, FOSB and ABL1) were selected for technical validation with Real Time PCR (RT-PCR) analysis in the same Tumor tissues belonging to the 14 different patients analyzed by microarrays. Two of them (JUN and FOSB) confirmed the results from the arrays. Figure 2 reports the expression dynamic over time of the genes, normalized to 18S. Normalization using GAPDH and ACTB housekeeping genes confirmed these results (data not shown). ABL1 was not validated as ischemia associated gene likely due to the weak level of association between microarray and RT-PCR data. This can be partially explained by the limited expression variability of this gene in our samples compared with JUN and FOSB (Fig. S2). Four genes commonly modulated in normal and tumor tissue, were also analyzed (HIST1H1D, HIST1H1E, HIST1H4E and HIST4H4); only one of them, HIST1H4E, confirmed the microarray data.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Figure 2. Time course for selected genes.

RT-PCR values of FOSB, JUN and ABL1 (−ΔCt) normalized to the house-keeping gene 18S.

https://doi.org/10.1371/journal.pone.0053406.g002

Discussion

We studied the effect of tissue handling time on global gene expression using CRC specimens sub-sampled and snap-frozen at different time points post-surgery, following routine tissue handling protocols of a Pathology Unit. RNA quality assessed by the RNA Integrity Number (RIN) showed a slight trend of decrease associated to time with tumor samples exhibiting higher and less variable RIN values at T0 compared to the normal samples. The presence of degradation at T0 could indicate the variability of the procedure or a specific sensitivity of RNA from colon tissue. Results obtained for tumor samples are in agreement with what found by Bray et al. [15]. All the samples were used for microarray analysis irrespective of their RIN number to understand if the microarray platform could highlight better proxies of quality for tissue handling. The distribution of feature intensities per array was highly similar across all the 180 arrays analyzed, suggesting that all RNAs performed at the same level and that hybridizations were done similarly; only 4 samples did not pass the quality controls.

To assess the variability of gene expression measurements as a function of time, each feature (probe) was modeled respect to the Time categories and adjusted for the Patient factor. This was done separately in both Tumor and Normal datasets. Using a Bonferroni correction of P<0.05, thirty-one probes were identified as time-dependent in the tumor specimens and thirty-two in the normal samples. Many of the genes identified in the tumor dataset were DNA-binding proteins involved in different processes: some corresponded to genes involved in cancer, such as ABL1, JUN, FOSB, NFKB1A and CCL5; four probes belonged to histones (HIST1H1E, HIST1H4E, HIST1H1D and HIST4H4), others were transcription factors (DDIT3) or genes involved in transcription such as EGR1 and PPP1R15A. Six genes were involved in apoptosis, one of the processes induced by tissue resection. Most of the genes identified (N = 22) were significant in the Time comparison involving all time points, 16 were identified in T3 versus T0 contrast and only 3 genes (ABL1, JUN and FOS) at both time points T2 and T3. Nine genes changed only at T3. On the other hand, many of the genes identified in the normal samples (N = 21) corresponded to genes coding for Histone proteins, and three were small nuclear RNAs (RNU2-1, RNU11 and RNU12). Thirty-two genes changed over all the four time points analyzed; 17 of them were significantly different at time point T3 and 8 at both time points T2 and T3. Two genes changed only at T3. By performing the same analysis only using the samples with a RIN above five very few genes were identified (6 probes in both datasets) of which mainly histones were in common with the previously identified genes. The results from these two approaches indicate that the changes observed over time are fairly negligible. Indeed, when we compared the tissue types between both Normal and Tumor datasets a very large number of differences where observed, as expected, irrespective of the quality of the RIN. Further, functional analysis of these probes using the Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway database showed that these differences matched with eleven significant pathways involved in cancer progression including Cell cycle, DNA replication and the p53 signaling pathway (data not shown).

The changes observed in the genes for both normal and tumor tissues mainly occurred at T3, although some genes were already altered at T2, suggesting that a critical time point lays at 60 minutes at room temperature. Interestingly, the three genes that showed changes already at T2 in tumor samples were all oncogenes, which suggests to carefully consider alterations of these genes while handling tumor samples and to use a more conservative time threshold (i.e. 60 minutes) for sample collection. The expression levels over time of most of the genes were increased. The present finding is in agreement with what found by Dumur et al. [16] on ovary cancer cases and by Bray et al. [15], in CRC, where the number of probes with increased expression augmented with time. As discussed by the authors of the two studies, this is opposite to what expected since handling time should favor degradation of RNA transcripts and may in part at least reflect an active modulation of gene expression. Comparison between GO classification of differentially expressed genes in our (FDR adjusted p-values<0.05) and Bray’s analyses showed few common enriched terms (‘protein dimerization activity’ and ‘transcription factor activity’). Possible explanations for these results could be the different time points evaluated (up to 360 minutes in our study and up to 120 minutes in Bray’s study), different procedures followed for sample collection (surgical specimens that followed the routine processing standard vs tumor biopsies) and different microarray platforms used for expression analyses (Illumina vs Affymetrix).

Two of the three genes selected for the validation in tumor tissues (JUN and FOSB) confirmed the array results. These genes were also among the mRNAs most significantly affected by time to freezing in a breast cancer study [17]. As reported by these authors, JUN and FOSB are stress induced immediate early transcription factors which are components of AP-1 dimers, and these dimers have been found altered by ischemia in different tissues including prostate and colon cancer. In ischemic and riperfused cells the two genes induce proliferation and apoptosis and could be closely related to attemps for degradation or regeneration of injuried tissues [18]. Five probes were shared among normal and tumor specimens. They identified genes from the histone family (HIST1H1D, HIST1H1E, HIST1H4E and HIST4H4) involved in DNA organization, and a small nuclear RNA (RNU11). Only one of these genes (HIST1H4E) validated the microarray data. The high sequence similarity of the histones could explain the high number of modulated histones identified, that could be due to partially aspecific hybridization and could be the reason for the lack of validation with an independent technique.

In this paper we have described for the first time the effect over time on handling up to 6 hours of normal colorectal tissue, which is frequently used as control in genome wide analyses. Interestingly, normal tissue showed less degradation than its corresponding tumor specimen, both in terms of RNA quality (RIN value) and of modulated genes that were mainly histones. Our results favor the storage of normal tissue, in addition to the tumoral.

The overall changes in gene expression seen in the specimens analyzed in the current study, where more than 16000 probes were investigated, do not seem to correlate with a global transcriptome event that one might expect, at least during the time frame analyzed in this study (<20 minutes, 60 minutes, 180 minutes and 360 minutes). Considering the very low number of affected genes found, the impact of tissue handling time on the overall gene expression profiling, at least in CRC, could be considered minor and would not be expected to play a major role in gene expression-based tumor stratification.

Our results agree with that of Dumor et al. [16] and of Micke et al. [19] that showed no relevant changes in ovary cancer and with what Hatzis et al. found in breast cancer [20]. Other studies have shown significant alterations in RNA after 30 minutes of tissue extirpation[15]; [21]; [22]. Since our design did not include early times, we cannot exclude that relevant changes have already occurred before our first time point (20 minutes). However, the aforementioned studies typically have very low sample sizes and would need further validation.

Significant alteration of gene expression profile were observed in our previous study on breast cancer specimens where tissue handling time altered the expression of genes included in the commonest breast cancer predictive gene signatures [23]. These data were in agreement with Borgan et al. [17], that found miRNA (microRNA) and mRNA expression alterated in breast cancer with ischemia time up to six hours. Indeed, Hatzis and colleagues [20] in breast cancer showed no relevant changes in expression levels of single genes and multigene signatures, probably because they considered 40 minutes as longest time point at room temperature or up to 180 minutes but using the RNA-later to preserve RNA integrity.

Despite partial disagreement on its extent, time of tissue handling can have an effect on gene expression, and probably might vary with tissue and with tumor type, including those that could exhibit greater sensitivity to hypoxia effects, such as brain tumors [24].

Supporting Information