| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|
|
|
|||||||
|
|||||||||
Cancer Diagnostics |
Institut d’Investigació Biomèdica de Bellvitge (IDIBELL)-Institut Català d’Oncologia,1 Translational Research Laboratory, Unit of Biostatistics and Bioinformatics, Cancer Epidemiology Department, and 2 IDIBELL-Institut de Recerca Oncològica Molecular Oncology Center, L’Hospitalet de Llobregat, Barcelona, Spain.
aAddress correspondence to this author at: Institut Català d’Oncologia, Laboratori de Recerca Translacional, Av. Gran Via s/n, Km 2.7, 08907 L’Hospitalet de Llobregat, Barcelona, Spain. Fax 34-93-2607466; e-mail gcapella{at}ico.scs.es.
 |
Abstract |
|---|
 Top Abstract IntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
Methods: RAP-PCR products obtained from samples with limited availability of biological material, such as experimental metastases, were hybridized to conventional cDNA microarrays. We performed replicates of self-self hybridizations of RAP-PCR products and mathematical modeling to assess reproducibility and sources of variation.
Results: Gene/slide interaction (47.3%) and the PCR reaction (33.8%) accounted for the majority of the variability. From these observations, we designed a protocol using two pools of three independent RAP-PCR reactions coming from two independent reverse transcription reactions hybridized in duplicate and evaluated them in the analyses of paired xenograft-metastases samples. Using this approach, we found that HER2 and MMP7 may be down-regulated during distal dissemination of colorectal tumors.
Conclusion: RAP-PCR glass array hybridization can be used for transcriptome analysis of small samples.
|
Introduction |
|---|
|
TopAbstractIntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
Several strategies have been developed to allow the analysis of small samples in which the amount of available RNA (<1 µg of total RNA) is insufficient for massive gene expression studies (1)(2)(3)(4). Trenkle et al. (1) proposed the use of nonstoichiometric reduced-complexity probes for hybridization to cDNA arrays and noted the fitness of the RNA arbitrarily primed PCR (RAP-PCR)2 method. RAP-PCR is an unbiased fingerprinting PCR that samples a reproducible subset of message population based on the best matches with arbitrary primers (5). It allows the construction of a probe with reduced complexity, which increases the representation of rare messages, and uses small amounts of total RNA (10–100 ng) or mRNA (0.1–1 ng).
Although different RAP-PCR fingerprints give hybridization patterns with overlapping representations of expressed genes, the concordance is partial, and with several probes, a greater fraction of the message population can be screened than with total cDNA probes (1). Moreover, with reduced-complexity probes, the number of differentially expressed genes detected is 10 times higher than when total cDNA is used (3). This approach has been applied to the identification of genes differentially expressed in laser-microdissected colonic crypts (6)(7) and rheumatoid arthritis (8)(9). In all of these studies, including the original report (1), RAP-PCR products were hybridized to cDNA array membranes.
To gain insight into the molecular basis of the metastatic process, we used orthotopic implantation of human primary tumors in nude mice (10)(11)(12). Implanted xenografts resemble, in their early stages, primary tumors and reproduce, in part, their dissemination pattern, allowing completion of the metastatic process in a short time. Nevertheless, transcriptome analyses of experimental metastases may sometimes be hampered by their small size. In this study we evaluated the feasibility of hybridization of RAP-PCR product to slide arrays to allow use of small samples such as experimental metastases. By combining experimental designs and mathematical models, we investigated the distinct factors contributing to variability of hybridization of RAP-PCR probes on cDNA microarrays. From these observations, we designed a simpler procedure that minimizes the number of replicates and laboratory effort. This protocol was validated by the analysis of paired xenograft-metastases samples.
|
Materials and Methods |
|---|
|
TopAbstractIntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
). Mean passage time from mice to mice of xenografts was 4 months. Fresh samples were embedded in OCT and kept frozen at –80 °C until their analysis. Manual microdissection was performed to ensure that 80% of all nuclei analyzed were from tumoral cells in tumor samples and from healthy epithelial cells in nontumor samples. Reproducibility experiments were performed with a single healthy colonic mucosa that was collected under standard conditions in our center, likely to represent the usual setting in which this technique will be used.
|
Total RNA was extracted by standard procedures (13). RNA quality and concentration were determined by spectrophotometry and gel electrophoresis. For selected experiments, mRNA was purified from total RNA by use of the PolyATract® mRNA Isolation System III reagent set (Promega) according to the manufacturer’s instructions.
rap-pcr
Reverse transcription was performed with 50 ng of total RNA, 200 U of M-MLV reverse transcriptase (Gibco BRL, Life Technologies), 15.3 units of RNAGuard RNase Inhibitor Porcine (Amersham Biosciences UK Limited), 0.5 mM deoxynucleotide triphosphates, 50 mM Tris-HCl (pH 8.3), 75 mM KCl, 3 mM MgCl2, 10 mM dithiothreitol, and 0.5 µM primer in a final volume of 20 µL. The reaction was performed for 1 h at 37 °C, followed by 5 min at 95 °C. Primers were chosen arbitrarily from those available in our stock (none of them was specifically designed to do RAP-PCR) and were evaluated for quality of the fingerprint generated when run in sequencing gel electrophoresis (complexity of the band pattern and reproducibility) as described previously (14). The sequences of the selected primers were as follows: pU6, 5'-GCTTCTGACTTATTCTTGCTCTTAG-3'; H12B, 5'-CGCCAGGCTCACCTCTATA-3'; TP53, 5'-AGGATACTATTCAGCCCGAGGTG-3'; DB2-122dw, 5'-GCCGACGTGCTCCTGATGTT-3'; Ml1 8up, 5'-GTTTCAGTCTCAGCCATGAG-3'; DB2, 5'-ACAGATCTGAAGGGTGAAATATTCTCC-3'; and p53, 5'-ATGGAGGAGCCGCAGTCA-3'.
PCRs were performed with 5 µL of the reverse transcription product in a final volume of 50 µL. The reagents added were 2.5 U of Taq DNA polymerase (Boehringer Mannheim), 20 mM Tris-HCl (pH 8), 50 mM KCl, 2.5 mM MgCl2, 0.1 mM deoxynucleotide triphosphates, and 2 µM primer, which was the same primer that was used in the reverse transcription. Before the PCR, tubes were heated 4 min at 94 °C. PCR amplification consisted of five low-stringency cycles (1 min at 94 °C; 45 s at 40–55 °C, depending on the primer; and 75 s at 72 °C) followed by 35 high-stringency cycles (1 min at 94 °C; 45 s at 55–60 °C, depending on the primer; 75 s at 72 °C. Finally, tubes were kept for 5 min at 72 °C. We diluted 3 µL of the RAP-PCR products in 9 µL of denaturing loading buffer, heated the mixture for 3 min at 95 °C, and electrophoresed it in denaturing sequencing gels (6% polyacrylamide and 8 mol/L urea for 3–4 h at 55 W). Resolved bands were visualized by silver nitrate staining.
probe labeling
RAP-PCR products were labeled with dCTP-Cy3 or dCTP-Cy5 (Amersham Biosciences UK Limited) by use of the Bioprime DNA Labeling System (Gibco BRL, Life Technologies). Probes were purified using ConcertTM Rapid Purification System (Gibco BRL, Life Technologies). DNA concentrations and rates of dye incorporation were quantified spectrophotometrically.
CDNA labeling methods for total amount of rna
Reverse transcription and labeling reactions were performed at the same time to get the cDNA for hybridizing to cDNA microarrays. We used 1 µg of total RNA, which was diluted in 2.4 µL. RNA was mixed with 6 µg of primer random hexamer in a final volume of 11 µL. This mixture was incubated at 70 °C for 10 min and then was mixed with 600 U of SuperScript II reverse transcriptase (Invitrogen BV/NOVEX); 30 units of RNase inhibitor; 50 mM Tris-HCl; 75 mM KCl; 3 mM MgCl2; 10 mM dithiothreitol; 0.5 mM each of dATP, dTTP, and dGTP; 0.2 mM dCTP; and 1 mM Cy dye in a final volume of 30 µL. The mixture was incubated at 42 °C for 2 h, and after the first hour, 1 µL of reverse transcriptase was added. Finally the reaction was stopped by addition of EDTA and NaOH to a final volume of 47.5 µL (final concentrations, 0.63 and 31 mM, respectively). The last incubation was at 70 °C for 10 min followed by addition of 24 mM HCl to a final volume of 62.5 µL. Probes were purified by use of the Concert Rapid Purification System. The cDNA concentration and rate of dye incorporation were quantified spectrophotometrically.
hybridization
For hybridization, we used poly-L-lysine cDNA microarrays [Hu 4.6K; W.M. Keck Foundation Biotechnology Resource Laboratory (http://keck.med.yale.edu)] containing 4608 human cDNAs spotted in duplicate. A list of the genes covered can be obtained at http://keck.med.yale.edu/dna_arrays.htm. The hybridization area was 22 x 22 mm. We dried 1–1.5 µg of the labeled DNA probe (usually corresponding to 20–50 pmol of dye) and resuspended it in a solution containing 1.6 g/L yeast tRNA, 1.6 g/L poly(dA), 1.6 g/L Cot-1 DNA, 3.5x standard saline citrate (SSC), and 3 g/L sodium dodecyl sulfate in a final volume of 14.52 µL. The labeled DNA was denatured for 1 min at 95 °C and left for 30 min at room temperature to allow it to interact with the blocking agents. The slides were denatured for 2 min in boiling water and kept for 2 min at 76 °C. Finally, the probe was put on the slide, and hybridization was performed in a humidity chamber at 65 °C for 14–16 h. After hybridization, the slides were washed consecutively with 2x SSC–1 g/L sodium dodecyl sulfate, 1x SSC, and 0.2x SSC for 2 min each.
image analysis and data processing
Slides were scanned with a GSI Lumonics ScanArray 4000. Images were analyzed by use of Spot software (CSIRO, Mathematical and Informatics Sciences), and the resulting data were processed to filter out low-quality spots based on spot area and similarity of readings between the two replicates of each gene. Replicate intensities were averaged. Subsequently, the two channels were normalized by use of the lowess nonparametric smoother method of a set of invariant genes because correlation of Cy5 and Cy3 values varies with intensity (15).
experimental design and statistical analysis
To assess the sources of variability in measured log-intensities, we performed multiple analyses of the same sample. The main factors considered to introduce random variation were (a) slide/hybridization, which we considered as the same effect, because in one slide only one hybridization can be performed, and (b) the probe synthesis process, which includes the reverse transcription reaction, PCR reaction, and labeling. Starting from the same RNA sample, different products were obtained and hybridized in multiple combinations, including labeling of individual RAP-PCR reactions or pools of them and hybridization of the same PCR labeled with distinct fluorochromes, as illustrated in Fig. 1
.
|
We used the restricted maximum likelihood method to estimate the components of variance, allowing assessment of the relative magnitude of each source of variability. Initially, we developed a mathematical model for single-slide hybridization that included both the random variation (attributable to the sequence of a given gene and the amount of a given RNA in our target sample) and the variables introducing artifactual variability: the fluorochrome, the reverse transcription reaction, and the PCR reaction. Nevertheless, because of the nature of the experiments to be performed, comparison between intensities obtained with distinct slides is mandatory and may be particularly critical when comparing serial samples (i.e., normal-adenoma-carcinoma-metastasis). Thus, we further developed the model to allow for accurate sample-to-sample comparisons. Once the origin of variability was ascertained, replicates and/or pooling in the more critical steps were performed to produce a final protocol.
validation of changes observed in paired tumor-metastases samples
We compared changes in three specific genes, MMP7, HER2, and PCAF, that showed recurrent differences in the RAP-PCR/microarray system, with data obtained by quantitative reverse transcription-PCR (qRT-PCR) to validate the nature and magnitude of the change. The PCR reactions were performed in a LightCycler apparatus (Roche Diagnostics S.L. Applied Science), using the LC-FastStart DNA Master SYBR Green I reagent set (Roche Diagnostics S.L. Applied Science) according to the manufacturer’s instructions. Selection of adequate housekeeping control genes is critical. ß2-Microglobulin and cyclophilin were chosen based on our previous experience. The ratio between them was constant at 1, thus (16) one of them, ß2-microglobulin, was finally used to control for input RNA. A relative calibration curve was constructed for each gene with use of five serial dilutions starting with 100 ng of RNA. Relative quantification between samples was done by use of a mathematical model (17). Primer sequences and PCR conditions may be obtained from the authors on request.
|
Results |
|---|
|
TopAbstractIntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
To analyze reproducibility and sources of variation, we performed several self-self hybridizations of RAP-PCR products, as described above, labeled with distinct fluorochromes and hybridized to conventional cDNA microarrays (Fig. 1, A and B
). We used the variance of log ratios as a measure of variability. Comparison of independent hybridizations of RAP-PCR probes obtained from the same sample showed a typical log ratio variance of 1.22 (range, 0.822–1.32) when no replicates or pooling was performed. To explore the sources of this variability, we focused on variation of signal intensity, not on variance of log ratios. Single-channel intensities of self-self hybridizations, previously normalized to allow for comparison between distinct fluorochromes, were analyzed for variance components. As expected, differences in the representativeness of the 4608 different genes of the sample used accounted for a major fraction of the variability. This effect was so large [73% of the total variability (95% confidence interval, 63–82%)] that it was taken into account and subsequently fixed in the model, allowing the analysis of artifactual variability. It has previously been suggested that fluorochromes may introduce artifactual variability (16). In our hands, hybridization of the same RAP-PCR product labeled with the two dyes did not vary, suggesting that dye/gene interaction was negligible. This allowed us to exclude a dye effect from the model and simplify it.
Modeling of the variability associated with this type of analysis indicated that the major source of variability (47.3%) was gene/slide interaction, i.e., the effect of performing hybridization in a different slide. The PCR reaction accounted for one third of the variability (33.8%), and the reverse transcription reaction for an additional 13.3% (Fig. 1
). Residual variance was very low (5.7%), as expected.
rap-pcr variability can be effectively reduced through pooling
On the basis of these observations, we attempted to generate an experimental design that could offer an acceptable artifactual variability. To determine whether several hybridizations of replicates performed better than hybridization of pooled replicates, we performed two distinct reverse transcription reactions with our sample template and three independent RAP-PCR reactions for each reverse transcription, yielding six PCR probes (Fig. 2
). Variance calculated from the mean values of three hybridizations of PCRs coming from independent reverse transcriptions (0.812; range, 0.552-0.932; Fig. 2A
) was significantly lower than the basal log ratio variance of 1.22 from a single hybridization. Because of the high cost of repeated hybridizations, we did not consider hybridization in duplicate of the same PCR product. Rather, we evaluated the effect on reproducibility of pooled RAP-PCR products of the same RNA (Fig. 2B
). Interestingly, variances of the hybridization of two pools of independent RAP-PCR reactions coming from two independent reverse transcription reactions were comparable (0.912; range, 0.632–1.1052). On the basis of these results, we decided to establish the protocol described in Fig. 2C
, which consists of hybridizing, in duplicate, two pools of three independent RAP-PCR reactions coming from two independent reverse transcription reactions. We validated this protocol by analyzing three additional mucosae. For sample 1, the variance for a single hybridization was 0.712, and the variance for the pooling hybridization was 0.592 (range, 0.532–0.642). The corresponding results for sample 2 were 0.922 and 0.802 (0.782–0.812), and for sample 3 were 1.102 and 0.832 (0.782–0.882).
|
coverage of rap-pcr vs CDNA
We next wanted to determine whether the transcriptome coverage of the RAP-PCR products compared with that of whole cDNA. Each RAP-PCR product as well as cDNA obtained from the same healthy colonic sample was labeled exclusively with Cy5 to avoid variations in intensity from competition. All products were independently hybridized in distinct slides. With this design, no normalization of the data was performed because distribution of intensities for each primer was a priori unknown, and thus distributions could not be assumed to be equal.
A threshold had to be defined to classify the genes as represented (18). The distribution of intensities of RAP-PCR and cDNA hybridizations precluded the definition of an objective threshold. Initially, spots with signal intensities less than twofold higher than the background (<1% of the 4608 genes in the slide) were assumed to have no signal. The representativeness (number of spots with specific hybridization) of the different RAP-PCR products and the cDNA was affected equally when we used distinct arbitrary thresholds based on exclusion of spots depicting the lowest intensity (Fig. 3
). When we fixed as a threshold twice the mean intensity of the 15% lowest genes, cDNA hybridization showed relevant signals in 68% of the genes spotted on the slide, in the same range as the sum of the three RAP-PCR products mean of all 3-to-3 combinations; Fig. 3B
). The addition of the three best primers accounted for most of this representativeness (Fig. 3
) and clearly outperformed cDNA. Finally, the cumulated information provided by all seven RAP-PCR products analyzed reached 91% for the genes, 23% more than was obtained when we used cDNA (Fig. 3
). The overrepresentation of RAP-PCR products compared with cDNA is likely to reflect the advantage of RAP-PCR to amplify low-abundance mRNAs that are poorly detected in whole cDNA hybridizations.
|
An additional theoretical advantage of using RAP-PCR products is the possibility of having multiple controls if the tag gene is represented in independent RAP-PCRs. When four or more different RAP-PCR products are hybridized, up to 70% of the genes are scored in two or more hybridizations, which indicates the powerful autovalidating feature of this approach even if no replicates are performed (Fig. 4
).
|
confirmation of differential expression between different samples
With this protocol, we studied three sets of related samples including a primary tumor, corresponding orthotopic xenograft, and one to three experimental metastases. Three arbitrarily selected genes previously associated with cancer development (MMP7, HER2, and PCAF) were used to ascertain the usefulness of our approach. According to microarray hybridizations, MMP7 expression changed in a nonconsistent manner in xenografts compared with the corresponding primary tumors. Unexpectedly, in five of seven metastases analyzed, MMP7 expression was further diminished. Similar results were obtained for HER2 and PCAF.
All changes with log ratios higher than 1.5, as assessed by RAP-PCR hybridization, were confirmed by qRT-PCR (Fig. 5
). The magnitude of ratios from microarray experiments tended to be lower than those from qRT-PCR, suggesting that kinetics of hybridization in the slide setting tend to smooth actual changes. On the other hand, discordant results were obtained for a portion of HER2 (n = 3) and PCAF (n = 5) analyses, where log ratios of 7–9 were not detected by slide hybridizations, a fact that may be also attributed to the kinetics of hybridization. For the analyzed genes, no false-positive changes were detected.
|
|
Discussion |
|---|
|
TopAbstractIntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
The hybridization of RAP-PCR to filter arrays has been shown to solve, in part, these limitations (1). The reduction of the complexity of the transcriptome together with its expansion by RAP-PCR was critical (a) to increase the sensitivity of the technique, which is seriously limited by the overwhelming complexity of the sample’s transcriptome; and (b) to overcome the small sample sizes routinely available in the clinical setting. Because very low amounts of biological material (<50 ng of total RNA/experiment or 10 pg of mRNA/experiment) are required, the use of microdissected samples is allowed. Additionally, the use of lower-complexity representations is likely to allow the unbiased representation of rare and common mRNAs. This technique has already been applied (6)(7)(8) to a filter array platform that requires the use of radioactivity and is difficult to automate. We therefore decided to develop and evaluate a glass microarray system that requires no radioactive handling, facilitates automation, and offers the possibility of cohybridization of two or more samples (if distinct fluorochromes are to be used).
The use of a PCR step in hybridizing a glass array is a major source of variation in the procedure that might well preclude its routine use. In spite of the importance of quantification of artifactual variability, few studies have addressed this issue in a systematic manner in array hybridizations. In the present study, we performed several self-self hybridizations, using distinct conditions, and created a mathematical model to assess the sources of artifactual variability in RAP-PCR hybridization to glass arrays. In agreement with previous studies showing that the array itself is the most common source of error in array-based analyses (16), gene/slide interaction was shown to contribute to 50% of the artifactual variability observed. This can be minimized only by the use of replicates, which will, on the other hand, substantially increase costs. As expected, the RAP-PCR procedure and, to a lesser extent, the reverse transcription reaction also contribute to artifactual variability. Fortunately, pooling of reactions can minimize this variability in an affordable manner, as shown for other DNA amplification procedures such as whole genome amplification (19).
We do not know whether the magnitude of artifactual variability resulting from use of RAP-PCR products as a probe compares with that resulting from use of cDNA probes. Nevertheless, we point out that estimates of variability—based on experimental data—are needed to establish the need for either replicate hybridizations or pooling of reactions. Estimation of variability may be also of help in choosing the strategy of statistical analyses to be performed.
Only the combined use of sample pooling—because it may reduce random amplification events—and replicate hybridizations can reduce the artifactual variability observed. Thus, based on our observations, we finally propose performing two hybridizations of distinct pools of RAP-PCR reactions, which is readily amenable for analysis of small samples, either experimental or clinical, increasing the feasibility of using RAP-PCR in the routine clinical setting. Although this strategy offers a good balance between cost and feasibility, there is still room for improvement. We must take into account that, in our approach, true replicates (hybridization of the same probe on distinct slides) are not used. Some artifactual variability attributable to true replicates may be expected (data not shown). However, although our protocol does not correct for this type of variability, the use of two hybridizations is likely to make these results sufficiently robust.
Another potential advantage of hybridization of RAP-PCR probes is the generation of data that may not be easily obtained by conventional cDNA or oligonucleotide-based microarrays because of nonstoichiometric selection of the represented transcripts. We have shown that, in the stringent hybridization conditions (65 °C) used in our setting, coverage of the best RAP-PCR product is comparable to that of whole cDNA, suggesting that a less-complex probe offers a better relative fraction. Moreover, coverage may be improved by the addition of independent hybridizations of different RAP-PCRs, which would also add confidence to the results because of the overlapping of coverages observed between distinct RAP-PCR products. The concomitant hybridization of distinct RAP-PCR products could theoretically be a good option to obtain, in a single hybridization, more coverage. However, experimental data obtained in nylon filters suggest that simultaneous hybridization of distinct RAP-PCR probes may lead to reduced target detection, limiting its potential benefits (20).
In the experimental samples used to validate our protocol, all changes in expression detected by RAP-PCR array hybridization with log ratios >1.5 were confirmed by qRT-PCR. In agreement with previous reports, qRT-PCR confirmed the trend of poor concordance regarding log-ratio values (17). Most of the studies with cDNA hybridizations have used arbitrary cutoff values such as two- or threefold changes to define a true change, but few of them have provided the experimental data to support the criteria used. Thus, in our experimental setting and with use as a reference a threshold value based on the variance of self-self experiments (18), threefold changes should be considered as true changes. However, our results also show the limitations of the sensitivity of our protocol. In some paired samples, qRT-PCR changes showing log ratios of 7–9 were not detected by RAP-PCR slide hybridization. It is likely that the compressed range for detecting intensity changes in slide hybridizations compared with qRT-PCR leads to loss of sensitivity. Finally, the systematic bias of the technique (21), which is not always improved by the use of replicates, may account in part for a loss of accuracy. This is not apparently the case because this bias, if present, tended to improve the performance of our methodology when we used our protocol.
In summary, we have systematically evaluated the sources of artifactual variability for RAP-PCR products hybridized to glass arrays by combining replicates of self-self hybridizations of RAP-PCR products and performing mathematical modeling. Gene/slide interaction and the PCR reaction are the major sources of artifactual variability, which can be reduced by combining hybridization replicates and pooling of samples. We have proposed a protocol that has been essentially validated by qRT-PCR analyses. Using this approach, we found that HER2, MMP7, and PCAF may be down-regulated during distal dissemination of colorectal tumors. We therefore consider that we have developed a consistent hybridization protocol that allows the use of the RAP-PCR/microarray approach for large-scale study of transcriptomes in small samples from experimental protocols or obtained in the routine clinical setting.
|
Acknowledgments |
|---|
|
Footnotes |
|---|
2 Nonstandard abbreviations: RAP-PCR, RNA arbitrarily primed PCR; SSC, standard saline citrate; and qRT-PCR, quantitative reverse transcription-PCR.
|
References |
|---|
|
TopAbstractIntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
