Impact of biospecimens handling on biomarker research in breast cancer

Case material

Eleven histologically confirmed primary breast tumors were obtained from the Tissue Bank of the Fondazione IRCCS Istituto Nazionale Tumori between October 2006 and March 2007. For each sample a written informed consent signed by the patient was available authorizing the use of leftover material for research purposes. The study was approved by Independent Ethics Committee and Institutional Review Board. Samples used in the study were selected based on the size, carefully avoiding necrotic areas, fat and normal tissue. An adjacent section was stained and used for defining the percentage of tumor cells. Only specimens with more than 70% of tumor cells were included in the study. Each tissue sample was divided into 4 aliquots. One was immediately frozen while the remaining three were frozen after 2, 6 and 24 hours at room temperature.

RNA isolation and expression profiling

Tissue was pulverized using a Mikrodismembrator (Braun Biotech International, Germany). Total RNA was extracted with the Trizol reagent (Invitrogen, Carlsbad, CA) according to manufacture instructions and an additional DNase digestion was performed using the RNeasy kit (Qiagen, Valencia, CA). After each extraction a small fraction of RNA was used for quality and yield assessment. RNA total concentration and purity were determined by UV spectrometry. Total RNA electrophoretic profile was analysed by the Agilent RNA 6000 NanoLabChip kit on the Agilent 2100 Bioanalyzer (Agilent Technologies, Palo Alto, CA) using the software provided by the manufacturer for determination of RIN (RNA integrity number).

Probe labelling and hybridization were performed as previously described [11]. The samples and a reference RNA (Universal Human Reference RNA, Stratagene, La Jolla, CA) were labelled directly with Cy3-dCTP (reference RNA) or Cy5-dCTP (sample RNA) (Amersham Biosciences, Little Chalfont, UK) and indirectly with 3DNA Submicro Expression Array Detection kit (Genisphere, Montvala, NJ). Hybridization was carried out in a hybridization station (Genomic Solutions, Ann Arbour, MI), slides were scanned using the GenePix 4000B microarray scanner and quantified using GenePix Pro 5.0.1.24 (Axon Instruments/Molecular Devices). The RNAs were hybridized on two different cDNA microarrays containing a total of 17172 unique clones selected from the Human sequence-verified I.M.A.G.E. clone collection (Research Genetics/Invitrogen, Carlsbad, CA) and spotted in triplicate. Raw gene expression data have been deposited at the European Bioinformatics Institute (EBI) ArrayExpress and are accessible through accession no. E-MEXP-2035.

Quantitative real-time PCR

Total RNA isolated for the microarray analysis was used to verify the quantity of specific messengers by real-time PCR for 4 biologically relevant differentially expressed genes (i.e., FGFR4, ESR1, ERBB2, FBLN2). RNA was reverse transcribed using the High-Capacity cDNA Archive Kit (Applied Biosystems, Foster City, CA). Samples were amplified in multiplex PCR reactions using one of the assays of interest labelled with FAM. PPIA was used as housekeeping gene. Reactions were performed in a final reaction volume of 20 μl containing cDNA template, 10 μl 2× TaqMan Gene Expression Assay (Applied Biosystems, Foster City, CA) and thermal cycling was performed on an ABI PRISM 7700 Sequencer Detector (Applied Biosystems, Foster City, CA).

Protein Profile Analysis

Following tissue pulverization, tumor samples were processed as described in [12]. Briefly, samples were lysed in an ice-cold buffer containing 50 mM HEPES (pH 7.6), 150 mM NaCl, 10% glycerol, 1% Triton X-100, 1.5 mM MgCl₂, 1 mM EGTA, 10 mM Na₄P₂O₇, 100 mM NaF, protease inhibitors and in presence or absence of 1 mM sodium orthovanadate. After 30 min incubation with gentle rocking at 4°C, lysates were cleared by centrifugation (10 min at 13,000 rpm). Supernatants were collected and, after protein concentration determination, tested for the presence of specific proteins by Western blotting. For Western blotting, cell lysates were resolved by 4-12% SDS-PAGE (precast gel NuPAGE, Invitrogen). Proteins were transferred to nitrocellulose membranes (Hybond-C Super, Amersham Bioscience, Little Chalfont, UK), checked for equal sample loading by Red Ponceau S staining, and probed with the appropriate antibodies. Immunoreactive bands were visualized using horseradish peroxidase-linked secondary antiserum (Amersham Biosciences, Little Chalfont, UK) and detected using an enhanced ECL system (Amersham Biosciences, Little Chalfont, UK). For Western blotting, the following antibodies were used: anti-p-Tyr 4G10 (Upstate Biotechnology, Millipore Corporation, Billerica, MA); p-Neu (Tyr 1248)-R (Santa Cruz, Biothecnology, Santa Cruz, CA) for ERBB2 and anti-FAK (Transduction Laboratories, BD, Franklin Lakes, NJ).

Data analysis and Statistics

Raw microarray images and quantifications were stored and processed in Bio Array Software Environment (BASE, Lund Sweden) [13]. Poor signal quality of background-corrected Cy3 and Cy5 intensities were flagged and a lowess normalization [14] was applied to each slide. Replicated spots were averaged and their log (base2) expression ratios (tumor/reference) were downloaded from BASE and imported into BRB-ArrayTools version 3.4.0. (Bethesda, MD) developed by Richard Simon and Amy Peng Lam [15]. ANOVA analysis was performed following their instructions. All I.M.A.G.E clones annotations were updated with the latest release of NCBI Unigene (Build No. 194) using SOURCE[16]. In order to find the overlap between our list of genes and genes included in public available signatures, we employed the Merge function of the web-application MatchMiner [17]. Hierarchical clustering was performed using centered correlation and average linkage method.

For gene function analyses, a score obtained calculating the -Log(p-value) of the Fisher Exact Test indicates the probability that a function is obtained by chance; scores equal or greater than 3 give a 99.9% confidence level of not being generated by chance.

The genes identified by microarray analyses were challenged for gene function. After having imported their accession numbers into the Ingenuity Pathway Analysis (IPA) software (Ingenuity Systems, Mountain View, CA), they were categorized according to the gene functions present in the database.

Data analysis for real-time PCR was done using the Sequence Detector version 1.9 software. Relative log-expression of the genes (-ΔC_t) was obtained subtracting the number of cycle threshold observed for the 18S gene from that observed for the gene of interest. After being exported further data analysis was performed in a BRB_ArrayToolv3.4.0.

The association between gene expression evaluated by cDNA microarry and real-time PCR was evaluated by linear regression analysis.

Gene expression analysis

The evaluation of RNA showed a time dependent decrease of 28S and 18S ribosomal bands and a trend towards lower RIN numbers for samples stored for longer times at room temperature. Two examples are shown in Fig. 1, one with limited and one with more extended RNA degradation. Overall, for 8 breast cancers the serial samples showed valuable results, i.e., RNA not degraded (RIN>5) after prolonged exposure of samples to room temperature, and these cases were submitted to gene expression analysis [18, 19]. Unsupervised hierarchical clustering analysis showed that samples belonging to the same patient tended preferentially to cluster together (Fig. 2), highlighting the patient-specific gene expression features despite different tissue freezing times. In fact only in 3 of the 8 samples (Pb6, Pb3 and Pb5) aliquots frozen at different times clustered separately.

Notwithstanding the relatively limited RNA degradation, a time-dependent modulation of genes was observed in the different serial aliquots, likely expected as a cellular reaction to stress conditions following the surgical resection of blood vessels and the suboptimal pre-analytical conditions. The gene expression profile of these serial specimens was performed using cDNA arrays containing 17,172 unique clones and analyzed by the ANOVA model. Using such approaches, genes whose modulation varied over the time independently of the intrinsic patient-specific differences were disclosed. A separate ANOVA model at significant threshold of p < 0.01 was fitted for each gene to relate the log-ratios expression to the factors under investigation (i.e. the time points and the different patients, the latter considered as blocking factor). Our results pointed out a significant variation as a function of time. Considering the number of genes modulated at each single time point compared to the sample immediately frozen as baseline, we noticed an increase in the number of modulated genes especially after 24 hours: 121 genes (0.76% of genes present in our chip) were modulated after 2 hours (66/121 were up-regulated); 157 genes (0.98%) were modulated after 6 hours (110/157 were up-regulated), 657 genes (4.1%) were modulated after 24 hours (287/657 were up-regulated) following tissue resection. Moreover, taking into account the overall time course of gene expression, we found 461 genes regulated at the significant threshold of p < 0.01, representing 2.88% of genes present in our chip. Around 1.03% of 461 genes were expected to be false positives according to the multiple testing correction method of Benjamini-Hochberg [20].

This result is in keeping with published data [3, 4, 6–9] in which a certain degree of gene expression variation as a function of collection procedures is reported. However, the modulation we noticed in breast cancer specimens was not as remarkable as the one reported by in Spruessel et al [4], where 20% of all detectable genes are modulated after 30 minutes of colon resection. In our study the majority of the genes (about two-thirds) were down-regulated, and up-regulated genes represented less than 1% of all the genes in our chip. Similarly Dash et al reported less than 0.6% of over-expressed genes in their study on radical prostatectomy specimens [8].

To better assess the impact of sample handling on the expression of critical genes, some of the modulated genes with an established biological role in breast cancer were subjected to a technical validation by qPCR. The statistically significant correlations between expression evaluated by cDNA microarray and Real Time PCR for FGFR4 (r = 0.57, p = 0.0008), FBLN2 (r = 0.595, p = 0.0004), ESR1 (r = 0.87, p = 0.0000008), ERBB2 (r = 0.754, p = 0.000002) clearly supports the fact that such a modulation is not imputable to technical artifacts, but represents a real biological phenomenon.

Effect of sample collection procedure on gene pathways

The 461 genes modulated as a function of delayed freezing of samples were analyzed using the IPA tool to understand if they were representative of specific biological categories. We focused our attention on the twelve most significant gene functions (Fig 3) that contain 69.6% of all genes which were present in the IPA database. Inflammatory, immunological disease, and cell cycle were the most significantly enriched gene classes. Globally various genes crucial for breast cancer were included. Among them, PIK3CD (p110δ) which is a catalytic subunit member of the class IA PI3K heterodimer, a signal transduction enzyme that regulates a broad spectrum of cellular functions as cell growth, proliferation and ultimately survival [21]. Inhibition of PIK3CD activity reduces VEGF activity leading to lack of vascular permeability [22]. As survival pathways play a crucial role both in prognostic outcome of breast cancer as well as in affecting treatment sensitivity, an artifact induction of such genes could be dramatically misleading. Similarly, among the genes down-regulated, STAT1 transcription was found repressed because of hypoxia mediated by HIF-1 and STRA13 [23]. CD44 was also among the down-regulated genes. This gene encodes a cell surface receptor for hyaluronate and is a direct target of miR373/miR520c, two recently identified metastasis-promoting microRNAs in breast cancer [24]. CD44 expression is associated with increased survival in node negative breast cancer [25] and decreased metastatic invasion [26].

Also genes reported to be associated with response to antiestrogen were found to be down-regulated by delayed tissue processing. Among them FBLN2, a gene reported to be up-regulated in elderly breast cancer patients who are not responsive to pre-operative treatment with toremifene [11] and FGFR4 whose up-regulation predicts poor response to tamoxifen [27]. Results are reported in Fig 4 panel A.

Comparison with published breast cancer microarray studies

The use of gene signatures for predicting prognosis and response to treatment is quickly moving from a research area to the clinical practice. The reliability of such tools as well as they performance are strictly related to accurate gene expression determination which could be affected by sample handling and storage. To better understand to which extent commonly used signatures in breast cancer could be affected by artifacts linked to sample handling we verified how many of our 461 genes were included in such gene lists [2, 28–33].

We then assumed that the extent of the impact of tissue handling on the prognostic/predictive reliability of a given gene signature, could be indirectly evaluated looking at its capability to maintain the clustering of individual samples collected at different times.

Results are reported in Fig. 4 for various gene signatures. After clustering the sample aliquots with the genes derived from the signature of Chang et al (343 genes in common with our data set) aliquots from 5 of 8 samples clustered together and similar data were obtained using the signature of Sorlie et al (271 genes in common), where in 6 out of 8 samples aliquots collected at different times clustered together. As expected, clustering of samples using signatures with a low number of genes as that of Paik et al [35], or with a small number of genes after searching for common genes within our data set, as that of Ivshina et al [31], were instead affected by inappropriate tissue handling as tissue aliquots obtained from the same patient did not cluster together. This underlines the fact that tissue mishandling has a limited effect on signatures with high gene numbers as already suggested in the Background section.

Interestingly, using the signatures with an intermediate number of genes (respectively 27 and 19 common genes) as in the case of the signature by Wang et al [34] and by van't Veer [32], we still observed that in most samples (5 out of 8 and 4 out of 8, respectively) aliquots from the same patient obtained at different times, did in fact cluster together. In these dendrograms the relatively low number of genes was probably partially compensated by the lack of overlapping (only 2 overlapping genes with the genes by Wang et al and none with genes by van't Veer) between such genes and the 461 genes that we identified as genes modulated by temperature. This may be an indirect proof of the fact that our 461 genes may be considered as reliable indicators of tissue mishandling

We also considered the expression of single genes implicated in breast cancer and currently used as predictive markers in the clinical management (ESR1, ERBB2, FBLN2, FGFR4, PGR). As reported in Fig 5 independently of ER base levels, ESR1 was consistently down-regulated through the four considered time points with the maximal down-regulation after 24 hours at room temperature. Fig 5 panel B reports the ESR1 microarray intensity expression for three samples where the effect is more evident (Pb3, Pb5 and Pb7). A similar down-regulation was observed for FBLN2 and FGFR4. On the contrary, no significant modulation was found for ERBB2.

Impact of sample collection procedures on protein profile

In order to set up conditions to study signaling pathways in post resection specimens, we also explored the effects of pre-analytical procedures on protein status, examining tyrosine phosphorylation or expression in Western blotting experiments and time course conditions. We firstly explored the tyrosine phosphoprotein profiles testing the reactivity to antiphosphotyrosine antibody on total extracts of 2 serial samples from a breast cancer specimen (Pb2, Pb3) frozen at different times from surgery and obtained in absence or in presence of orthovanadate, a phosphatase inhibitor. Results are shown for Pb3 taken as a representative sample in Fig 6A. With the addition of orthovanadate in lysis buffer, the loss of phosphotyrosine proteins due to delayed freezing was clearly prevented up to 6 hours. However, after 24 hours at room temperature no major differences were observed between the pattern of phosphoproteins in the presence and in the absence of orthovanadate. These results provide the information that phosphatase inhibition is not sufficient to stabilize protein phosphorylation at room temperature. We then performed time course experiments on other two serial samples from breast cancers (Pb1 and Pb2) lysed in presence of orthovanadate (Fig 6B). Antiphosphotyrosine hybridization results highlighted the progressive loss of phosphoproteins during time course with individual variability probably depending on which specific phosphoprotein pathways become activated or suppressed ex vivo.

Detection of phosphoepitopes is often very important for understanding whether a specific pathway is active, thus inadequate pre-analytical procedure may dramatically modify biological interpretation of data obtained on clinical samples. Therefore sample Pb1 and Pb2, for which ERBB2 immuhistochemistry positivity was detected (Table 1) and gene expression analysis showed no significant change as a function of the time delay in freezing, were tested in Western blotting for the level of Tyr 1248 phosphorylation of ERBB2 using a phospho-specific antibody (Fig 6C). The result was the same for both tumor specimens: signal decreased until disappearing within 24 hours post resection. When an anti-FAK antibody was used to representatively testing protein integrity during time course, it was found that the signal intensity disappeared at 24 hours of incubation (Fig 6C). Still considering Western blotting a non-quantitative assay it is clear that at 24 hours time point protein degradation was maximal.