Impact of Collection and Storage of Lung Tumor Tissue on Whole Genome Expression Profiling

Abstract

Gene expression profiling could assist in revealing biomarkers of lung cancer prognosis and progression. The handling of biological samples may strongly influence global gene expression, a fact that has not been addressed in many studies. We sought to investigate the changes in gene expression that may occur as a result of sample processing time and conditions. Using Illumina Human WG-6 arrays, we quantified gene expression in lung carcinoma samples from six patients obtained at chest opening before and immediately after lung resection with storage in RNAlater [T1a_(CO) and T1b_(LR)], after receipt of the sample for histopathology, placed in RNAlater [T2a_(HP)]; snap frozen [T2b_(HP.SF)]; or snap frozen and stored for 1 week [T2c_(HP.SFA)], as well as formalin-fixed, paraffin-embedded (FFPE) block samples. Sampling immediately after resection closely represented the tissue obtained in situ, with only 1% of genes differing more than twofold [T1a_(CO) versus T1b_(LR)]. Delaying tissue harvest for an average of 30 minutes from the operating theater had a significant impact on gene expression, with approximately 25% of genes differing between T1a_(CO) and T2a_(HP). Many genes previously identified as lung cancer biomarkers were altered during this period. Examination of FFPE specimens showed minimal correlation with fresh samples. This study shows that tissue collection immediately after lung resection with conservation in RNAlater is an optimal strategy for gene expression profiling.

Lung cancer (LC) is the leading cause of cancer death worldwide¹ with non-small cell LC accounting for approximately 87% of newly diagnosed cases.² LC is the most common cancer in the UK and it is predicted that it will remain so for at least the next 20 years.

Measurement of global gene expression is a powerful means of establishing the transcriptional activity of particular cells or tissues. Gene expression profiling can allow the identification of subgroups of cancer allowing early prediction of disease progression and survival.^3–6

Global gene expression studies in LC have had conflicting results and many signature-based outcome predictions have not been replicated independently (as reviewed^7,8). Multiple confounding factors have hampered these studies including the innate heterogeneity of cancerous tissue. Studies are often conducted with limited numbers of samples making it difficult to derive statistically stringent results from the measurement of thousands of transcripts.

The quality of the biological sample used and sample handling are key factors influencing global gene expression studies. In particular RNA in ex vivo tissues degrades rapidly with the potential to influence expression patterns and bias the interpretation of results.

The majority of published gene expression studies for LC have used tumor tissues that have been snap frozen or formalin-fixed paraffin embedded (FFPE), with many of the studies relying on archival or biobanked tissues.^3–5,9–12 It is uncertain how this material represents biological features in vivo.

To address the impact of timing of tissue sampling, processing, and storage, we have conducted a differential gene expression analysis in lung tumor tissue of patients with LC at different time points of specimen collection under various conservation conditions starting from in vivo state represented by viable tissue and ending at archival FFPE tissue.

Materials and Methods

Samples

Lung carcinoma tissue samples were obtained from six patients during tumor resection surgery at the Royal Brompton Hospital in London. Demographic and clinical characteristics of the patients are detailed in Table 1. All participants gave written informed consent for research on biobanked tissue and the biobank consent was approved by the Royal Brompton and Harefield Ethics Committee (REC reference number LREC 02-261).

Table 1.

Demographic Characteristics of Study Patients

Patient ID	Sex	Age at surgery (years)	Histological diagnosis	Tumor stage
03	Male	57	Squamous cell carcinoma	T2N0M0
05	Female	83	Adenocarcinoma	T1N0M0
06	Male	62	Adenocarcinoma	T1N0M0
07	Male	70	Broncho-alveolar cell carcinoma	T2N0M0
08	Female	80	Large cell undifferentiated	T2N0M0
09	Male	57	Pleomorphic carcinoma	T2N1M0

Figure 1 illustrates the pipeline for the tissue collection. When possible for each patient tissue samples were obtained at five collection points: viable tissue at the time of chest opening [T1a_(CO)]; immediately after resection [T1b_(LR)]; after transfer to histopathology [T2a_(HP)]; after transfer to histopathology and snap freezing [T2b_(HP.SF)]; and after transfer to histopathology and 1 week after being snap frozen and archived [T2c_(HP.SFA)]. FFPE samples of each patient's tumor was obtained from histopathology after tissue had been fixed in 10% formol saline for 24 hours before being embedded in paraffin wax according to standard procedure. The FFPE blocks have been kept at room temperature for 25 to 27 months before RNA extraction for the current study. The size of tissue samples taken for the experiment was approximately 6 × 6 × 3 mm. For the FFPE samples, five or six 10-micron sections were obtained.

Figure 1.

Sample collection pipeline.

After sampling the T1a_(CO), T1b_(LR), and T2a_(HP) tissues were immediately placed into RNAlater (Qiagen, Germantown, MD) for 24 hours at 4°C and then frozen at −80°C until RNA extraction. No RNAlater was used for the T2b_(HP.SF) and T2c_(HP.SFA) samples as these were snap frozen.

The average interval of samples collection after T1a_(CO) was 1.9 ± 0.6 hours for T1b_(LR), and 2.4 ± 0.5 hours for T2a_(HP), T2b_(HP.SF), and T2c_(HP.SFA). For patients 08 and 09, T2a_(HP), T2b_(HP.SF), and T2c_(HP.SFA) time point specimens were unavailable.

RNA Extraction

Total RNA was extracted using the RNeasy Fibrous Tissue Mini Kit (Qiagen) for the majority of time points, with the exception of the FFPE samples processed using the RecoverAll Total Nucleic Acid Isolation Kit for FFPE (Ambion, Foster City, CA). The frozen T1 and T2 specimens were placed into the lysis buffer at room temperature and immediately homogenized using the TissueRuptor (handheld rotor-stator homogenizer) with disposable probes (Qiagen) as per the manufacturer's recommendations. After homogenization the further steps for RNA extraction were performed following the manufacturer's protocol. Using sterile scalpel blades FFPE sections were cut up into smaller fragments before RNA extraction according to the manufacturer's protocol. Yield and purity of total RNA obtained was assessed using a Nanodrop ND-1000 spectrophotometer (NanoDrop, Thermo Scientific, Wilmington, DE) with RNA integrity determined by RNA Integrity Number (RIN) using a Bioanalyzer 2100 (Agilent Technologies, Santa Clara, CA).

Microarray Hybridization

Illumina human WG-6 v2 BeadChip microarrays (containing 48,804 probes corresponding to 43,186 genes derived from NCBI RefSeq [Build 36.2] and UniGene [Build 199] databases) (Illumina, Inc., San Diego, CA) were used to assess global gene expression for each sample.

Five hundred nanograms of total RNA was amplified, converted to cRNA, fragmented, and then biotin-labeled using the Illumina TotalPrep RNA-amplification kit (Ambion, USA). Then 1.5 μg of labeled cRNA was hybridized to each array according to the Illumina whole-genome gene expression direct hybridization assay protocol 11286331, Rev. A., after which arrays were scanned using the Illumina BeadArray Reader. The images were processed and converted into signal intensities using the Illumina GenomeStudio software Version 2009.2 (Illumina, Inc.). The same software was used to perform hybridization quality control (QC).

The expression data have been deposited in the EMBL-EBI Array Express database (http://www.ebi.ac.uk/arrayexpress) and are available through E-MTAB-581 accession number.

Statistical Analysis

The signal intensities corresponding to gene expression levels of individual arrays were background corrected and imported into text files using the Illumina GenomeStudio 2009.2 software. All subsequent analyses were performed in R language environment (R Foundation for Statistical Computing, Vienna, Austria) using the suite of programs within Bioconductor v.2.5.¹³

Text files containing gene expression values were imported into R using the lumi package.¹⁴ Variance stabilizing normalization was applied to reduce between-arrays variation. The substantial differences in gene expression between FFPE and other time point samples required that normalization for FFPE was done separately. Post normalization, genes with low detection rates (P > 0.01) were removed. Two datasets were generated in anticipation of absence of gene expression signal in FFPE samples due to the high level of RNA degradation recognized to occur in this type of sample. The first comprised samples from all time points except FFPE and included 18,597 genes, each found to be expressed significantly above background in at least one of these samples. The second dataset comprised samples from all time points and included 4555 genes, each found to be expressed significantly above background in at least one FFPE sample only. Paired comparisons were performed to assess differentially expressed genes between all data points. Using the limma package robust regression was applied and individual t-statistics were calculated for each gene and each comparison followed by empirical Bayesian method application to moderate the standard deviations between genes.¹⁵ Raw P values were adjusted for multiple comparisons using the false-discovery rate approach of Benjamini and Hochberg.¹⁶

Hierarchical cluster analysis was applied to datasets to evaluate the “proximity” between the time points. Using the publicly available database and research tools DAVID^17,18 and Ingenuity Pathway Analysis (IPA; Ingenuity Systems, Redwood City, CA), gene ontology, and pathway analyses were performed to consider biological meaning of differential expression of genes between the data points. In network analysis, maximum 25 networks and maximum 35 molecules per network were stipulated.

Results

RNA Quality Analysis

High quality of RNA (RIN ≥7) is the ideal to enable robust expression microarray results to be generated. RINs <7 indicate RNA degradation.

RIN scores revealed that the highest quality of RNA was obtained for T1a_(CO) and T1b_(LR) samples: mean 7.4 (range 6.7 to 8.5) and 7.8 (range 6.7 to 8.9), respectively;. For T2a_(HP) samples mean RIN was 7.7 (range 5.4 to 8.3), whereas for snap-frozen samples [T2b_(HP.SF) and T2c_(HP.SFA)] the median values were 4.4 (range 3.3 to 5.6) and 5.2 (range 4.8 to 5.6), respectively. These figures highlight the beneficial use of RNAlater stabilization in preventing RNA degradation. As anticipated, the lowest RINs, range 2.2 to 2.5, were obtained for FFPE samples.

It is unlikely that poor RINs in T2b_(HP.SF) and T2c_(HP.SFA) samples resulted from snap-freezing or long-term low-temperature storage. It is more likely that RNA degradation has occurred during the thawing of the specimens for RNA extraction.¹⁹

Despite differences in RNA integrity, the total yield of RNA was independent of the time points of collection and ranged between 43 and 367 μg (average 198.3 μg). For FFPE samples the yields ranged between ∼5 and 17 μg (average 8.7 μg).

Gene Expression Profiling

The Illumina Human WG-6 v2 BeadChip microarray was used to analyze whole-genome gene expression in samples from the six patients.

The average number of genes significantly (P < 0.01) expressed above background was similar at all data points for unfixed samples and ranged between 12,000 and 13,000 genes. For FFPE samples the expression of approximately 3000 genes was detected above background. The highest ratio of average signal to background was found for T1a_(CO) and T1b_(LR) samples (average 3.2 ± 0.8), in other samples it was lower (average 2.7 ± 0.9), and in FFPE it was < 1 (average 0.2 ± 0.03). Two samples were identified as outliers based on the number of significantly expressed genes and ratio of signal to background and excluded from subsequent analyses.

Following variance stabilizing normalization, 18,597 genes were significantly present at least one time point in non-fixed samples and were included in subsequent analyses. Exclusion of genes that were not significantly expressed in at least one of the FFPE samples left 4555 transcripts.

As time from initial chest opening at surgery progressed there was a notable and significant increase in the number of genes that were significantly differentially expressed (Table 2). Less than 5% of genes were differentially expressed between T1a_(CO) and T1b_(LR) time points, and only 1% of genes differed more than twofold, indicating that the T1b_(LR) point is a good representation of the in vivo state. The number of differences between T1a_(CO) or T1b_(LR) and subsequent points was much higher (Table 2).

Table 2.

Number and Percentage of Significant Differentially Expressed Genes Between Different Time Points

	Chest opening	Lung resection	Histo-path RNA later	Histo-path snap frozen	Histo-path snap frozen and stored
	T1a(CO)	T1b(LR)	T2a(HP)	T2b(HP.SF)	T2c(HP.SFA)	FFPE
T1a(CO)	–	914 (4.9)	4508 (24.2)	5363 (28.8)	6164 (33.1)	4172 (91.6)
T1b(LR)	334/580	–	4378 (23.5)	5160 (27.7)	6324 (34.0)	4110 (90.2)
T2a(HP)	1560/2948	1881/2497	–	79 (0.4)	684 (3.7)	4083 (89.6)
T2b(HP.SF)	1411/3952	1623/3537	13/66	–	32 (0.2)	4030 (88.5)
T2c(HP.SFA)	1363/4801	1797/4527	108/576	18/14	–	4071 (89.4)
FFPE	527/3645	475/3635	448/3635	400/3630	405/3666	–

The total number of differentially expressed genes (FDR-adjusted, P < 0.05) is given above the central diagonal, with the percentage of total genes measured given in parentheses. Percentage was calculated using a total of 18,579 genes for all time points except for formalin-fixed and paraffin embedded tissue (FFPE). For FFPE, the denominator was the 4555 genes with measurable transcripts. Beneath the diagonal the number of genes showing increased expression is shown before “/”and the number decreased in expression is shown after “/”. Subsequent time points are compared with the one immediately prior.

The T2a_(HP), T2b_(HP.SF), and T2c_(HP.SFA) points were similar in gene expression (Table 2). The low numbers of differentially expressed genes between T2a_(HP) and T2b_(HP.SF), and between T2b_(HP.SF) and T2c_(HP.SFA), suggested that snap freezing and storage at low temperature had a low impact on gene expression. At the same time, the number of genes differentially expressed between T1a_(CO) or T1b_(LR) and T2a_(HP) was lower than it was between T1a_(CO) or T1b_(LR) and T2b_(HP.SF), confirming the efficiency of RNAlater conservation before freezing.

As expected, FFPE showed most genes to be differentially expressed when compared with the unfixed samples (Table 2), reflecting a high level of RNA degradation. Some correlation was observed between gene expression in T1a_(CO) and FFPE samples (R² 0.12 to 0.45, P < 0.0001, for the 4555 transcripts detectable in FFPE).

We next considered the effects of specimen handling and processing on the expression of genes previously established by others as potential markers of LC development and prognosis of outcome. We chose 145 genes that from the literature were reported to be highly relevant for clinical application as biomarkers^4–6 (see Supplemental Table S1 at http://jmd.amjpathol.org). Of these genes, 119 (82%) were significantly expressed in the tissues of our study and 68 of them were differentially expressed between T1a_(CO)/T1b_(LR) and later time points (Table 3).^4–6

Table 3.

Genes Previously Considered as Potential Biomarkers of Lung Cancer Prognosis and Survival^4–6 Found to be Significantly Differentially Expressed in Transition Between in Situ After Chest Opening or Lung Resection (T1a_(CO)/T1b_(LR)) and Later Time Points

Gene ID	Gene name	Fold change⁎
AASS	Aminoadipate-semialdehyde synthase	1.95
ABCC4	ATP-binding cassette, sub-family C (CFTR/MRP), member 4	1.54
ADM	Adrenomedullin	1.10
AKAP12	A kinase (PRKA) anchor protein 12	1.13
ALDOA	Aldolase A, fructose-bisphosphate	1.34
ALG8	Asparagine-linked glycosylation 8, alpha-1,3-glucosyltransferase homolog (S. cerevisiae)	1.19
C6ORF15	Chromosome 6 open reading frame 15	0.89
CASK	Calcium/calmodulin-dependent serine protein kinase (MAGUK family)	1.49
CASP4	Caspase 4, apoptosis-related cysteine peptidase	1.52
CDS1	CDP-diacylglycerol synthase (phosphatidate cytidylyltransferase) 1	1.48
COL3A1	Collagen, type III, alpha 1	1.40
CPA3	Carboxypeptidase A3 (mast cell)	3.42
CRK	V-crk sarcoma virus CT10 oncogene homolog (avian)	2.19
CSTB	Cystatin B (stefin B)	0.72
CTSL	Cathepsin L1	1.63
DBP	D site of albumin promoter (albumin D-box)-binding protein	0.81
DPAGT1	Dolichyl-phosphate (UDP-N-acetylglucosamine) N-acetylglucosaminephosphotransferase 1 (GlcNAc-1-P transferase)	1.43
EVI1	MDS1 and EVI1 complex locus	3.20
FADD	Fas (TNFRSF6)-associated via death domain	1.63
FEZ2	Fasciculation and elongation protein zeta 2 (zygin II)	0.81
FGFR2	Fibroblast growth factor receptor 2	2.01
FLJ20397	HEAT repeat-containing 2	1.45
FUCA1	Fucosidase, alpha-L- 1, tissue	0.83
GAPDH	Glyceraldehyde-3-phosphate dehydrogenase	1.33
GGA3	Golgi-associated, gamma adaptin ear containing, ARF-binding protein 3	1.34
GRB7	Growth factor receptor-bound protein 7	1.46
H2AFZ	H2A histone family, member Z	1.03
HLA-G	Major histocompatibility complex, class I, G	0.56
HLF	Hepatic leukemia factor	0.68
HMBS	Hydroxymethylbilane synthase	1.50
HRB	ArfGAP with FG repeats 1	1.62
KIAA0746	Sel-1 suppressor of lin-12-like 3 (C. elegans)	1.65
KLF10	Kruppel-like factor 10	2.10
KLF6	Kruppel-like factor 6	0.78
KRTDAP	Keratinocyte differentiation-associated protein	1.06
LRIG1	Leucine-rich repeats and immunoglobulin-like domains 1	1.52
MAP4	Microtubule-associated protein 4	1.34
MAPK14	Mitogen-activated protein kinase 14	2.19
MSH3	MutS homolog 3 (E. coli)	1.32
MT2A	Metallothionein 2A	0.77
NME2	Non-metastatic cells 2, protein (NM23B) expressed in	1.52
NP	Purine nucleoside phosphorylase	1.45
NTRK2	Neurotrophic tyrosine kinase, receptor, type 2	0.43
NTS	Neurotensin	12.57
PDE7A	Phosphodiesterase 7A	1.13
PELI2	Pellino homolog 2 (Drosophila)	1.29
PPIF	Peptidylprolyl isomerase F	1.37
RAB11A	RAB11A, member RAS oncogene family	1.11
RND3	Rho family GTPase 3	1.31
RPLP0	Ribosomal protein, large, P0	1.45
RPS26	Ribosomal protein S26	1.49
RPS3	Ribosomal protein S3	1.56
SC4MOL	Sterol-C4-methyl oxidase-like	0.78
SFTPC	Surfactant protein C	0.50
SLC2A1	Solute carrier family 2 (facilitated glucose transporter), member 1	2.83
SPRR2E	Small proline-rich protein 2E	1.28
STARD3	StAR-related lipid transfer (START) domain-containing 3	0.92
STC1	Stanniocalcin 1	0.64
TIA1	TIA1 cytotoxic granule-associated RNA-binding protein	1.78
TKTL1	Transketolase-like 1	0.29
TMEM126B	Transmembrane protein 126B	1.37
TMF1	TATA element modulatory factor 1	2.04
TPBG	Trophoblast glycoprotein	1.39
TTR	Transthyretin	0.76
TUBA4A	Tubulin, alpha 4a	0.64
UGP2	UDP-glucose pyrophosphorylase 2	1.39
WNT10B	Wingless-type MMTV integration site family, member 10B	0.51
ZNF552	Zinc finger protein 552	1.11

^⁎Calculated as median value for all pair-wise comparisons, FDR-adjusted P < 0.05.

Cluster Analysis of Gene Expression

We performed a non-supervised hierarchical cluster analysis to identify genes characterizing each of the times studied. Considering the 661 genes that differed significantly across T1a_(CO) to T2c_(HP.SFA) (FDR-adjusted P < 0.0001), we observed that the specimens collected in the operating theater [T1a_(CO) and T1b_(LR)] clustered together and were distinct from a second cluster based around retrieval during routine histopathology procedures [T2a_(HP), T2b_(HP.SF), and T2c_(HP.SFA)] (Figure 2).

Figure 2.

Clusters of the most statistically significant (FDR adjusted P < 0.0001) differentially expressed genes. A: A heat map and hierarchical cluster analysis of 661 genes differentially expressed (P < 0.0001) at two surgical and three histopathology lab sample times. B: A heat map and hierarchical cluster of 3700 genes detectable in the FFPE samples and differentially expressed (P < 0.0001) compared with earlier time points.

When we considered transitions across all time points between T1a_(CO) to FFPE (including 3700 genes that were statistically significant at FDR-adjusted P < 0.0001) it was obvious that FFPE samples differed markedly from all other time points (Figure 2).

Gene Ontology and Pathway Analyses

To consider the possible biological meaning of changes in gene expression between time points we performed gene ontology (GO) and pathway analyses. We restricted these analyses to genes differentially expressed between T1a_(CO) and later time points with FDR-adjusted P < 0.01 (Table 4).

Table 4.

Top Gene Ontology (GO) Terms for Genes Differentially Expressed in Transition Between in Situ Biopsy After Chest Opening (T1a_(CO)), and Other Time Points

Time point	GO ID	Term	P value
Lung resection	GO:0006954	Inflammatory response	0.014
	GO:0009611	Response to wounding	0.049
Routine Histopathology (RNAlater)⁎	GO:0034960GO:0006139	Cellular biopolymer metabolic processNucleobase, nucleoside, nucleotide and nucleic acid metabolic process	4.29 × 10−91.00 × 10−5
FFPE	GO:0006412GO:0044237GO:0003735	TranslationCellular metabolic processStructural constituent of ribosome	1.97 × 10−74.64 × 10−67.95 × 10−6

^⁎For routine histopathology, only the RNAlater point is shown, as the results are very similar to snap-frozen samples.

Two marginally statistically significant GO terms were found for genes differentially expressed between the earliest time points [T1a_(CO) and T1b_(LR)] (Table 4). These related to the inflammatory response and the response to wounding.

For genes differentially expressed in the transition between T1a_(CO) and downstream time points [T2a_(HP), T2b_(HP.SF), and T2c_(HP.SFA)] the top GO terms obtained were similar for all comparisons and mainly characterized biopolymer and macromolecule metabolism. This was most likely a reflection of the activation and depression of cell reactions to lack of external supplements.

Finally, statistically significant GO terms for transition between T1a_(CO) and FFPE time points were found to encompass metabolic and biosynthesis processes, translation, and ribosome structure. These all likely reflect the tissue activity in the period after resection and before fixation in formalin.

We then performed a database-derived pathway analysis using genes that were at least threefold differentially expressed within the T1a_(CO), T1b_(LR), T2a_(HP), T2b_(HP.SF), and T2c_(HP.SFA) time points with an FDR-adjusted P < 0.001. There were 295 out of the 386 genes corresponding to these criteria that were successfully mapped in IPA and used for the pathway analysis. The top bio-functions that characterized interaction networks contained several terms and molecules related to cancer (Table 5), indicating that tissue handling significantly influences gene expression related to the phenotype of interest. Examination of canonical pathways revealed significant numbers of differentially expressed genes to be related to innate immunity and inflammation. These effects were mediated mainly by tumor necrosis factor (TNF) and IL-6 cytokines or associated with the FOS oncogene.

Table 5.

Top IPA bio-functions of gene networks characterizing the transition between samples harvested at operation (T1a_(CO)) or the routine histopathology laboratory (T2a_(HP))

Bio-functions	P value	Number of molecules
Diseases and Disorders
Inflammatory response	1.18 × 10−9 – 2.42 × 10−3	51
Dermatological diseases and conditions	7.46 × 10−8 – 2.29 × 10−3	30
Cancer	2.12 × 10−6 – 2.38 × 10−3	79
Hematological disease	8.25 × 10−6 – 2.28 × 10−3	16
Organismal injury and abnormalities	8.25 × 10−6 – 2.28 × 10−3	25
Molecular and Cellular Functions
Cell-to-cell signaling and interaction	1.52 × 10−9 – 2.49 × 10−3	55
Cellular movement	3.12 × 10−9 – 2.42 × 10−3	48
Cellular growth and proliferation	1.80 × 10−6 – 2.28 × 10−3	70
Cell death	2.87 × 10−6 – 2.42 × 10−3	69
Cell cycle	3.25 × 10−6 – 2.16 × 10−3	26
Physiological System Development and Function
Hematological system Development and function	1.52 × 10−9 – 2.42 × 10−3	54
Immune cell trafficking	1.52 × 10−9 – 2.42 × 10−3	35
Tissue development	1.52 × 10−9 – 2.28 × 10−3	42
Tissue morphology	2.08 × 10−6 – 2.29 × 10−3	38
Organ development	3.79 × 10−6 – 2.16 × 10−3	11

We used a further IPA analysis to test how well FFPE samples might represent in vivo gene expression, concentrating on the 383 genes that were not differentially expressed between T1a_(CO) and FFPE (FDR-adjusted P > 0.05). A total of 163 genes were successfully mapped in IPA. Only two weakly significant pathways were revealed for these genes; FGF signaling (P = 0.03; FGF17, PIK3C2B, and FGF18 proteins) and TR/RXR activation (P = 0.035; PIK3C2B, NXPH2, RCAN2). This suggested that the FFPE samples provided very limited information compared to earlier time points.

Discussion

Given the increasing importance of microarray profiling for clinical cancer management, it is perhaps surprising that there is little published information examining the effect of time of collection, tissue handling and storage on global gene expression analysis in LC and other solid tumors.

Our investigations suggest some simple guidelines for optimal management of lung tumor specimens and gene expression studies.

Tissue obtained in situ in the chest [T1a_(CO)] may be considered as a gold standard, but in situ harvest may not always be practical or feasible. We found sampling immediately after resection [T1b_(LR)] closely represented the in vivo state, as only 193 (1%) of genes differed more than twofold between T1a_(CO) and T1b_(LR) and genes related to cancer did not appear in network and pathways analyses.

The second major stage in handling of tumor specimens typically occurs in histopathology departments. We found that an average transition time of 30 minutes from the operating theater had a significant impact on gene expression profiles, as we observed differences in expression for approximately 25% of genes (or more than 5% if at least twofold changes were counted) when either T1a_(CO) or T1b_(LR) were compared with T2a_(HP).

Snap freezing without conservation in RNAlater had a significant impact on RNA quality and integrity, resulting in lower expression signals (but no additional bias in gene expression patterns). Storage after freezing did not significantly affect gene expression scores because few genes were differentially expressed between T2b_(HP.SF) and T2c_(HP.SFA) time points.

FFPE samples have been considered to be a promising source of biological information because of their availability.²⁰ A number of studies have been published in which the capability of FFPE samples to provide RNA of adequate quality for expression assays in cancer with use of real-time polymerase chain reaction and different microarray platforms in comparison with snap frozen material has been analyzed.^{9,11,12,21,22} Overall there was a poor correlation between FFPE and snap frozen tissues in our study, although a limited set of genes demonstrated correlation that might be useful in particular circumstances. It has also been shown that the time of FFPE samples storage is critical, with samples older than 1 year have remarkably decreased RNA quality.¹² Our current study of approximately 2-year-old FFPE samples has shown that the number of genes significantly expressed above background was four times lower than in non-FFPE samples. This suggests that the potential of FFPE samples to represent an in vivo state is circumscribed. These results however may have been influenced by the use of oligo(dT) primers in the cDNA synthesis step in the Illumina optimized TotalPrep RNA-amplification kit protocol. A proportion of the RNA from FFPE samples is likely to be fragmented with species present that lack polyA tails. Consequently, the TotalPrep RNA-amplification kit that is required to be used in combination with the Illumina Human WG-6 v2 BeadChip microarrays may for FFPE samples result in a lower yield of target cRNA. Recently it has been proposed that using Affymetrix Exon 1.0ST arrays (Affymetrix, Santa Clara, CA), originally designed to allow study of alternative splicing, may be important when dealing with archival FFPE samples. The probe sets on the exon arrays instead of targeting the 3′ end of each transcript span the entire length of each gene. This may allow the potential to detect and measure more robustly in archival FFPE RNA more genes within the genome.²³

Inflammatory response and related pathways mediated mainly by TNF and IL-6 molecules were highly affected by sample handling in transition between T1a_(CO)/T1b_(LR) and later time points. TNF and IL-6 are pluripotent cytokines involved in regulation of immune responses and inflammation. They have been found to be essential in the development and clinical heterogeneity of cancer as well as other lung diseases such as asthma and chronic obstructive pulmonary disease (COPD).^24–28 In particular, IL6 is considered as one of the key genes involved in the development and progression of LC through up-regulation of the EGFR/IL-6/STAT3 signaling cascade²⁹ and the potential utility of anti-IL-6 therapy in cancer has been discussed.³⁰

We also found that pathways associated with the FOS oncogene to be up-regulated in the transition from T1 to T2 time points. Human FOS acts as a trans-activating regulator of gene expression and has an important role in signal transduction, cell proliferation, and differentiation in early response of cells to growth factors.^31–33

Global gene expression assays are considered to be a promising tool to improving our understanding of etiology and pathogenesis of LC and may be of use in screening, disease classification, and outcome prediction.⁷ Our observation that many genes previously considered to be biomarkers for LC development and prognosis^4–6 were affected by the time and handling of samples indicate that the accuracy and standardization of gene expression measurements is essential for clinical applications of the technology. Our observations also highlight the need for caution, with regard to the reproducibility and utility for meta-analyses of data from some microarray-based gene expression studies, especially those where snap-frozen samples have been used and others where fresh samples have not been immediately processed. A significant number of highly cited articles on gene expression profiling in lung or breast cancer involved studies where fresh-frozen tissues have been examined. Some of these studies have reported predictive expression signatures, but there has been little if any overlap in particular genes between the studies.^34,35 Even though the studies differ in many methodological aspects, as shown by our current study, tissue handling has a significant influence on gene expression profiles obtained. It is worth noting that the quality of samples for microarray experiments is not a lung cancer specific issue and is increasingly recognized to be a crucial consideration in studies of other cancer types and noncancerous tissues.^36–38

There are a few additional points to highlight about our study. The sample size of our study is small (n = 6) and therefore the study statistically may be underpowered and potentially lead to bias in results due to outliers. We have at least partially addressed this by using robust regression to calculate differential expression. This approach is claimed to be more powerful and less sensitive to outliers than a number of other statistical procedures.³⁹ Another issue arises from using samples of different histological types. This produces additional heterogeneity in gene expression profiles and can reduce the number of genes differentially expressed between the conditions. Even though this cannot be excluded, we presume that the initial heterogeneity pattern is kept throughout the experiment and overall results are unlikely significantly affected. Also, according to our unpublished data, different cancer subtypes share a significant number of common genes underlying common cancer-related processes, such as inflammation or cell cycle. This means that we can expect a lot of similarity in global gene expression profiles for different lung cancer subtypes despite the presence of subtype specific expression signatures.

In summary, our study demonstrates that sample handling and processing are critical factors when conducting and interpreting results for LC gene expression profiling. Tissue collection for gene expression analysis at lung resection with conservation in RNAlater is the optimal strategy for gene expression profiling.