Optimization of laser capture microdissection and RNA amplification for gene expression profiling of prostate cancer

Optimization of RNA quality in tissues processed for LCM

The quality of total RNA extracted from stained tissues was assessed using an Agilent bioanalyzer [18]. Significant RNA degradation occurred during the staining protocol, probably due to reactivation of endogenous nucleases in aqueous solutions. RNase inhibitor was, therefore, included in all staining solutions except xylene, in which it is insoluble. Electropherograms of total RNA extracted from serial sections of three representative prostate tissue specimens that were unstained, stained in the absence of RNase inhibitor, or stained in the presence of RNase inhibitor are presented in Fig. 1. Relative to unstained tissues, RNA degradation occurred in tissues stained in the absence of RNase inhibitor, as evidenced by a shortening and broadening of the 28S ribosomal RNA peak at about 48 seconds. In all cases, degradation was decreased when tissues were stained in the presence of RNase inhibitor, as evidenced by recovery of the 28S ribosomal RNA peak height.

Quantitative measures of RNA degradation, Degradation Factors and RNA Integrity Numbers [18,19], based on the electropherograms presented in Fig. 1 are provided [see Additional file 1]. These quantitative measures were in accordance with visual interpretations. We also developed a qPCR assay to characterize RNA integrity using primer pairs specific for the 5' and 3' ends of the PSA transcript. The qPCR results [see Additional file 2] correlated well with quality assessments based on electrophoretic traces. Interestingly, tissues that appeared similar in terms of RNA integrity when unstained were observed to differ significantly in RNA degradation after staining, especially in the absence of RNase inhibitor.

LCM and determination of RNA concentration and quality

Specific cell populations based on primary Gleason pattern [20,21] were collected using LCM. Epithelial cells were captured from five benign tissues and two cases with benign prostatic hyperplasia (BPH). Primary Gleason pattern 3, 4, and 5 cells were collected from seven, two, and eight cases, respectively. Tumor cells were also collected from three cases of prostate cancer lymph node metastases. LCM images taken before, during, and after capture of primary Gleason pattern 3 cells are presented [see Additional file 3].

We attempted to use a NanoDrop^® spectrophotometer to quantify the limited amounts of LCM RNA because the small sample requirement (1–2 μL) precludes the need for sample dilution. Measurements typically were not reproducible, presumably because the sample concentrations were at or below the detection limit of 1.5 ng/μL [22]. Determination of RNA quality using the Agilent bioanalyzer also lacked reproducibility, even though the total RNA Pico assay has a qualitative range of 200–5000 pg/μL [23]. We, therefore, developed a qPCR assay to assess RNA quantity and integrity.

A primer set specific for the 3' end of the β-actin transcript was designed to quantify relative RNA concentrations. β-actin was used for normalization rather than GAPDH because GAPDH expression levels have been shown to correlate with pathologic stage in human prostate tumors [24]. The 3' β-actin primer set was combined with another primer set specific for the middle (M) region of the β-actin transcript to determine 3'/M β-actin ratios for assessment of RNA quality. The primer sets are designated 3' and M because the amplicons they generate fall within the target sequences used to design the 3' and M beta-actin probes on the Affymetrix U133 Plus 2.0 array. The ratio of the RNA quantities determined using the two primer sets is indicative of RNA quality. This is because oligo(dT) primer is used to synthesize cDNA template for amplification in qPCR, allowing only transcripts with intact 3' ends to be detected. Transcripts shortened due to degradation are detected as having less amplification of their upstream ends and, therefore, higher 3'/M ratios.

Results of the analysis of RNA quality and concentration based on qPCR with β-actin primers are presented in Fig. 2 for samples obtained after LCM of 27 prostate tissues. Since the 3'/M β-actin ratios were all two or lower, this was established as a quality control parameter for proceeding with linear amplification. One case, 586-GP5, was observed to be an outlier with a 3'/M beta-actin ratio of 0.6, and this result was reproduced. Yields of RNA obtained for the different cases varied substantially. Yield was not observed to correlate with quality since even samples with low yield (e.g., 1593-GP5) had good quality RNA. We, therefore, sought to determine if there was a correlation between yield and number of cells collected for the different cases. We used the number of laser pulses performed during capture of cells as an approximation of the number of cells collected for each case. There was no discernable correlation between yield and number of laser pulses or tissue type [see Additional file 4]. Individual cases appear to differ significantly in RNA content, making it difficult to determine how many cells must be collected to obtain sufficient RNA for linear amplification.

Comparison of the IVT and Enzo kits

To determine whether to use the IVT kit [25,26] or the Enzo kit in our linear amplification protocol, we performed a side-by-side comparison. Samples of total RNA from Gleason pattern 3 and Gleason pattern 5 microdissected cells were amplified and labeled using each kit. Different amounts of total RNA from Gleason pattern 3 and Gleason pattern 5 cells, 0.7 ng and 5.3 ng, respectively, were amplified to determine if 2- to 10-fold less than the minimum recommended amount of input RNA could yield sufficient labeled cRNA for hybridization to a microarray. This was important since it was not feasible to obtain 10–100 ng of RNA from microdissected cells for most of the 27 prostate tissues analyzed (Fig. 2). Furthermore, to avoid concentrating the RNA [see Additional file 5], the total amount of RNA needed for linear amplification should be contained in 4 μL or less (see Materials and Methods). This volume is limited by the minimum volume of buffer, 11 μL, needed to elute RNA from the purification column.

A sufficient yield of labeled cRNA (21–72 μg) was obtained for each sample to hybridize 15 μg to a U133 Plus 2.0 array. Correlation plots were generated using un-normalized expression data from GCOS 1.1. Signal correlations (R²) were 0.94 for inter-assay comparisons (Figures 3A and 3B). Saturation of signal intensities for transcripts expressed at higher levels was apparent for targets prepared using the Enzo kit since the plots assumed a banana shape at higher signals. In agreement with these results, previous studies indicated that use of the Enzo kit to prepare targets for hybridization to the U133 Plus 2.0 array resulted in some degree of saturation of signal intensities for transcripts expressed at very high levels compared with data obtained on the U133A array, and the IVT kit alleviates the high-end saturation effect observed with the Enzo kit [25]. We chose to use the IVT kit in our linear amplification protocol.

We investigated the reproducibility of the final steps of the procedure, namely fragmentation of additional aliquots of the Gleason pattern 5 cRNA samples labeled using the IVT kit or the Enzo kit, hybridization of the fragmented cRNA to U133 Plus 2.0 arrays from a different lot, washing, staining, and scanning (referred to as hybridization replicates). Enzo saturation was again evident in the correlation plot of the hybridization replicates, and the inter-assay signal correlation (R²) was 0.95 (Fig. 3C). For comparison, Fig. 3D shows a scatter plot correlation of expression data from replicate Gleason pattern 5 samples amplified and labeled on different days using the IVT kit. The replicates were treated the same except that different amounts of total RNA (approximately 2 ng and 5 ng) were used for amplification, and the microarrays were scanned using two different GeneChip^® Scanner 3000 instruments. The intra-assay correlation (R² = 0.96) was better than the inter-assay correlations (R² = 0.94–0.95), and no banana shape was evident.

The hybridization replicates allowed another interesting analysis to be performed. Signal correlation (R²) between replicate hybridizations was plotted against signal intensity for Gleason pattern 5 samples labeled using the IVT kit or the Enzo kit. Correlations between hybridization replicates decreased rapidly with decreasing signal intensity both for samples labeled using the IVT kit and the Enzo kit [see Additional file 6]. Average signal intensities obtained with the IVT kit were significantly lower than those obtained with the Enzo kit, shifting the entire plot to the left. Signal correlations between hybridization replicates represented a best-case scenario; therefore, replicates of the entire linear amplification procedure were also analyzed. For any given intensity value, the correlation was significantly higher between the hybridization replicates than the replicates of the entire linear amplification protocol.

Linear amplification of LCM RNA

Equivalent amounts (about 2 ng) of total RNA from normal epithelial cells, BPH cells, Gleason pattern 3 cells, Gleason pattern 5 cells, and lymph node metastases that were microdissected from 25 total cases were amplified and labeled using the IVT kit as described in Methods. SuperScript™ III, rather than SuperScript™ II, was used to synthesize cDNA since it was found to be more processive [see Additional file 7]. The average yield of labeled cRNA was 72 μg (ranging from 42–94 μg). The average size of the biotinylated cRNA was about 750 base pairs for each of the 25 cases. The average fold amplification during the first round was 60-fold, which was significantly greater than the 4–10-fold recommended by the Working Group [27]. The average fold amplification during the second round was about 800-fold, whereas it should have been about 400-fold according to the Working Group [27]. A so-called "normalizer sample" was amplified along with the other 25 samples. This sample consisted of RNA from microdissected Gleason pattern 4 cells. Aliquots of the normalizer sample were stored at -80°C to allow one aliquot to be amplified with each subsequent batch of test samples to serve as a replicate and an indicator of batch-to-batch variability.

Probesets detected as present were binned into low, medium, and high expression level categories [see Additional file 8] [see Additional file 9]. The coefficient of variance for the number of probesets in each bin was generally low (less than 6%). We observed no apparent correlation between the yield or RNA integrity based on the 3'/M β-actin assay and the number of probesets in different bins. Unsupervised clustering using less than 2000 genes grouped 5/5 normal and 2/2 BPH cases in one clade, 5/7 Gleason pattern 3 cases in another clade, and the remaining 2/7 pattern 3 cases in a third clade that also included 8/8 Gleason pattern 5 cases along with 3/3 lymph node metastases (Fig. 4). The normalizer sample (Gleason pattern 4) clustered with the high-grade samples.

Principal components analysis [see Additional file 10] showed segregation of samples along Eigengene vector 1 to be predominately by Gleason pattern. From left to right, a progression was observed from benign prostatic epithelial and BPH cells to Gleason pattern 3, Gleason pattern 5, and metastatic cells.

We used qPCR to confirm differential expression of alpha-methylacyl coenzyme A racemase (AMACR) and hepsin detected using microarrays. Aliquots of the same samples that were amplified and labeled to generate microarray results were also analyzed by qPCR before and after linear amplification. Relative expression levels of AMACR and hepsin measured by qPCR in unamplified samples correlated well with results obtained with amplified samples and with microarray results (Fig. 5). Both AMACR and hepsin were upregulated in prostate cancer compared to benign prostatic epithelial cells. The mean fold difference for AMACR expression between prostate cancer and benign cells was 5.3 (p = 0.02) in unamplified samples, 4.5 (p = 0.014) after linear amplification, and 4.2 (p = 0.004) as determined by microarray analysis. The mean fold difference for hepsin expression between prostate cancer and benign cells was 5.7 (p = 0.00004) for unamplified samples, 4.4 (p = 0.0003) after linear amplification, and 4.3 (p = 0.00003) as determined by microarray analysis. The fidelity of differential gene expression was, therefore, preserved during the linear amplification procedure. Correlations between microarray and qPCR data after linear amplification were 0.83 and 0.87 for AMACR and hepsin expression, respectively. Expression of hepsin was highest on average in Gleason pattern 3 cells (5.6 ± 2.9 hepsin/beta-actin ratio), and was observed to decrease in high grade, Gleason pattern 5 and lymph node metastasis, samples (2.6 ± 1.5) to a level intermediate between benign (0.7 ± 0.4) and low grade, Gleason pattern 3, cells (Figures 5D–5F). Interestingly, a similar pattern was observed for AMACR expression (Figures 5A–5C).

Frozen tissue sectioning

Surgical specimens with written patient consent were obtained from the Mayo Clinic Specialized Program of Research Excellence in Prostate Cancer (SPORE) tumor bank with Institutional Review Board approval (#1937-00). After radical prostatectomy, tissues from patients that had not received preoperative hormonal therapy, chemotherapy, or radiation therapy were flash frozen in liquid nitrogen and stored at -80°C. Fisherbrand Superfrost uncharged slides (Fisher Scientific, Hampton, NH) were baked at 220°C for at least two hours. Frozen tissue sections were cut at 5 μm in a -20°C cryostat and placed on slides pre-chilled at 4°C. After the tissue adhered to the slide, the slide was immediately placed on a flat piece of dry ice. The slides were placed in a chilled slide box on dry ice, transported under dry ice, and stored at -80°C. Sections from each tissue specimen were stained with hematoxylin and eosin and analyzed by a pathologist.

Staining of frozen tissue sections, LCM, and RNA isolation

Frozen tissue sections were stained using the HistoGene™ LCM Frozen Section Staining Kit (Arcturus, Mountain View, CA) according to the manufacturer's protocol [43], with the following modifications. ProtectRNA™ RNase inhibitor (1× final concentration, catalog number R7397, Sigma-Aldrich) was added to each solution, except xylene. Slides were transferred on dry ice from the -80°C freezer to the laboratory bench and immediately placed in 75% ethanol. Incubation in 100% ethanol was extended to five minutes. Following incubation in xylene for five minutes, slides were dried for 60 seconds under vacuum in a dessicator.

To analyze RNA integrity in stained tissue sections, RNA was prepared by pipetting 100 μL of extraction buffer directly onto the tissue section on the glass slide and using the pipet tip to gently scrape the tissue into the buffer, which was then transferred into an RNase-free microcentrifuge tube. RNA was extracted and purified using the PicoPure™ RNA Isolation Kit (Arcturus) according to the manufacturer's protocol, including on-column DNase treatment (Qiagen, Valencia, CA).

Immediately after staining a tissue section, LCM was performed by a pathologist using a PixCell^® II instrument (Arcturus). Thirty minutes on average (± 15 minutes) were spent collecting cells from a tissue section. RNA was extracted from the captured cells and purified as described above. Extracts from homogeneous populations of cells captured on multiple caps from one to six serial sections of each case were combined onto a single RNA purification column to achieve a sufficient yield and concentration of RNA for linear amplification.

RNA quality was assessed using an Agilent 2100 bioanalyzer (Agilent Technologies, Palo Alto, CA) and/or a qPCR assay for 3'/5' PSA or 3'/M β-actin (described below). RNA quantity was assessed using a NanoDrop^® ND-1000 spectrophotometer (NanoDrop Technologies, Wilmington, DE) and/or a qPCR assay for the 3' end of the beta-actin transcript (described below).

Quantitative PCR

First-strand cDNA synthesis was performed using SuperScript™ II or III (Invitrogen, Carlsbad, CA) and oligo(dT)_12–18 (Invitrogen) according to the manufacturer's instructions. One μL of LCM RNA was used per 20 μL cDNA synthesis reaction. Typically, 0.25 μL of the resulting cDNA reaction was used per qPCR reaction, and samples were analyzed in triplicate. Primer sequences are provided [see Additional file 12]. Quantitative PCR was performed using a 7900HT instrument (Applied Biosystems, Foster City, CA) and SYBR^® Green PCR master mix (Applied Biosystems). Thermal cycle parameters were as follows: 50°C for 2 minutes, 95°C for 10 minutes, 40 cycles of denaturation at 95°C for 15 seconds and annealing and extension at 64°C for one minute. Dissociation curves were generated by denaturation at 95°C for 15 seconds and annealing at 60°C for 15 seconds followed by a gradual increase in temperature (2% ramp rate) to 95°C. Standard curves were generated for each primer pair with serial dilutions of cDNA prepared using a mixture of total RNA isolated from benign and malignant prostate tissues. The results from the tested samples were compared to the corresponding standard curves generated in the same experiment to determine the sample RNA yield. Melting curve analysis was used to assess PCR specificity.

Target synthesis and hybridization

Target amplification and labeling were performed according to the Affymetrix protocol [41], with the following modifications. Since the starting amount of total RNA was only two ng, the poly-A RNA control stock was diluted five-fold more than recommended for a starting amount of 10 ng of total RNA. The T7-oligo(dT) primer/poly-A controls mix was prepared two-fold more concentrated than recommended to allow one μL to be combined with four μL of total RNA sample for first cycle, first-strand cDNA synthesis. The total RNA sample/T7-oligo(dT) primer/poly-A controls mix was incubated at 65°C for three minutes. SuperScript™ III (Invitrogen) was used to synthesize cDNA for one hour at 50°C. In the second cycle, purified cRNA was incubated with random primers for five minutes at 65°C. SuperScript™ III (Invitrogen) was used to synthesize first-strand cDNA for one hour at 50°C. First-strand cDNA was incubated with T7-oligo(dT) primer for three minutes at 65°C in preparation for second-strand cDNA synthesis. After cleanup of double-stranded cDNA, 0.5 μL was saved for qPCR analysis, and the remaining cDNA was used to synthesize biotin-labeled cRNA. Labeling of cRNA transcripts with biotin was performed using the GeneChip^® IVT Labeling Kit (IVT kit; Affymetrix, Santa Clara, CA) or the Enzo BioArray™ HighYield™ RNA Transcript Labeling Kit (Enzo kit; Enzo Life Sciences, Farmingdale, NY). Labeling reactions were incubated for 16 hours. Yield of biotinylated cRNA was measured using a NanoDrop^® ND-1000 spectrophotometer (NanoDrop Technologies). Unfragmented and fragmented cRNA samples were analyzed on an Agilent 2100 bioanalyzer (Agilent Technologies). Fifteen μg of fragmented cRNA were hybridized to U133 Plus 2.0 arrays (Affymetrix), washed, stained, and scanned according to the Affymetrix protocol [41].

Data analysis

To generate signal correlation scatter plots for replicate samples prepared with the IVT kit or the Enzo kit, un-normalized expression data from GCOS 1.1 were plotted for all genes except the Affymetrix control genes.

Unsupervised clustering was performed using dChip version 1.3 software. All 26 arrays were first normalized to the array with median intensity (1041-GP3). The PM-only or PM/MM model was used with expression values that were or were not log₂ transformed. Approximately 1500–2000 genes with the highest standard deviation/mean (not log₂ transformed) or the highest standard deviation (log₂ transformed) were used to perform unsupervised hierarchical clustering.