| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|
|
|
|||||||
|
|||||||||
Technical Briefs |
1 Spanish National Cancer Research Centre (CNIO), Madrid, Spain; 2 Department of Oncology, Hospital Universitario Puerta de Hierro, Madrid, Spain; 3 Department of Pharmacogenetics and Pharmacogenomics, Hospital Universitario Gregorio Marañón, Madrid, Spain; 4 National Centre of Biotechnology (CNB-CSIC), Campus Universidad Autónoma, Madrid, Spain
aaddress correspondence to this author at: Spanish National Cancer Research Centre (CNIO), 3 Melchor Fernández Almagro St., 28029 Madrid, Spain; fax 34-91-732-8028, e-mail mserrano{at}cnio.es
Abstract
Background: The blood of cancer patients is known to contain fragments of RNA released from the tumor. The application of genomic profiling techniques to plasma RNA may allow the unbiased selection of cancer markers in the blood, but the informative value of genomic profiling of plasma RNA is currently unknown.
Methods: We used cDNA microarray hybridization to perform genomic profiling of plasma RNA from colorectal cancer (CRC) patients and from healthy donors. From a list of 40 genes differentially upregulated in cancer patients, we randomly selected 4 genes for further characterization. These 4 markers were analyzed by quantitative reverse-transcription PCR in a wide set of samples including paired samples from the same CRC patients before and after surgical resection of the tumor.
Results: Three of the selected markers—EPAS1, UBE2D3, and KIAA0101—were confirmed by PCR to be significantly increased in cancer compared to healthy donors. Importantly, 2 of the markers, EPAS1 and UBE2D3, showed a significant decrease after surgery, returning to the levels of healthy donors. Finally, supervised class prediction using these 3 markers correctly (77%) assigned presurgery samples to the CRC group and assigned postsurgery samples from the same patients to the healthy group.
Conclusions: Our findings demonstrate the usefulness of gene expression profiling of circulating plasma RNA to find cancer markers of potential clinical value.
The blood of cancer patients contains higher concentrations of DNA than does the blood of healthy individuals (1). The development of PCR amplification techniques has allowed the analysis of this circulating DNA, and a large body of evidence has demonstrated that the plasma DNA from cancer patients presents features of the cancer DNA, suggesting that it derives from tumor cells (2)(3)(4)(5). More recently, several groups have reported the extraction of RNA from the plasma of cancer patients and its subsequent analysis by reverse transcription (RT)-PCR (6)(7)(8)(9)(10). The potential use of plasma RNA for the analysis of cancer is highly attractive for several reasons: it requires a minimally invasive method (collection of a small amount of blood); it can be obtained at any time and in a repetitive fashion, allowing the analysis of disease progression and treatment response; and its simplicity is amenable for use in asymptomatic populations at risk. The analysis of plasma RNA has been restricted to a few markers assumed to be abundant and specifically associated with particular cancer types, for example, mammaglobin for breast cancer and tyrosinase for melanoma (7)(10). Further progress toward the clinical use of plasma RNA requires the unbiased identification of markers. Genome-wide profiling of plasma RNA is an obvious approach but has technical drawbacks that could prevent its application, such as the low abundance and lack of integrity of plasma RNA (11).
For this reason we evaluated the feasibility of a genomic approach to studying plasma RNA. We measured by cDNA microarray hybridization the relative abundance of the different RNA species in the plasma of colorectal cancer (CRC) patients (n = 12) and healthy donors (n = 8). All the patients and healthy donors along this study gave their informed consent following the rules of the Research Ethics Board of Hospital Universitario Puerta de Hierro. Each sample was competitively hybridized against a common reference formed by a pool of blood samples from 26 healthy donors different from those hybridized as individual healthy samples. Differential gene expression analysis between CRC and healthy donors identified a total of 87 genes, including 40 that were differentially upregulated in the cancer group (see Supplementary Fig. 1A and Supplementary Table 1 in the Data Supplement that accompanies the online version of this Technical Brief at http://www.clinchem.org/content/vol53/issue10). Comparison with previous gene expression analysis of CRC that used tissue from the tumor as the source of RNA revealed that 4 of our differentially upregulated genes, PSAM3, RANBP1, GCLC, and KIAA0101, had been previously identified in 2 different studies as upregulated in CRC (12)(13).
We then performed quantitative RT-PCR (Q-RT-PCR) on the same samples to analyze expression for a subset of 4 randomly selected genes, KIAA0101, UBE2D3, EPAS1, and DDX46, from the list of 40 differentially upregulated genes. Two of them were validated by PCR as significantly upregulated in the CRC samples (KIAA0101 and UBE2D3; both with Kruskal–Wallis (KW) P value <0.05), one showed clearly increased expression although with lower statistical significance (EPAS1; KW P value = 0.098), and the last one could not be validated (DDX46; P value = 0.96) (Table 1
and Supplementary Fig. 1B in the online Data Supplement).
|
One of these 4 genes, KIAA0101, also known as p15PAF, has been previously identified as a commonly overexpressed gene in a variety of solid tumors by 2 independent groups who used large-scale metaanalysis of cancer DNA microarray data (14)(15). EPAS1 encodes hypoxia-inducible factor 2-alpha (HIF2
), an important angiogenic factor whose high expression in CRC has been shown to play an important role in tumor progression and to possess prognostic value (16). UBE2D3, another of our selected markers, encodes a ubiquitin-conjugating enzyme, also known as UBCH5C, involved in the regulated degradation of important cellular factors such as the tumor suppressor p53 and the NF
B regulator, IB (17)(18). Finally, DDX46 encodes a member of the DEAD box protein family that has putative helicase activity and is involved in pre-mRNA splicing as part of the 17S U2 small nuclear ribonucleoprotein complex.
To test the consistency of the detection of these genes in the blood of CRC patients, we analyzed their expression by Q-RT-PCR on a set of 29 new CRC plasma samples and 36 healthy donor samples, different from the ones that were part of the microarray study. With this external set we verified the increased expression of 2 of the markers, although only 1 of them (EPAS1, KW P value <0.05) was significantly higher in CRC patients than in healthy donors. The other marker (UBE2D3, KW P = 0.09), although its mean value was increased, had lower statistical significance (Table 1
and Supplementary Fig. 2A in the online Data Supplement). Finally, the overall analysis of the above markers by Q-RT-PCR across all the samples used in this study, i.e., the internal and the external sets together, showed a clear and statistically significant (KW P <0.05) increase of KIAA0101, UBE2D3, and EPAS1 in the plasma of CRC patients compared to healthy individuals (Fig. 1A
).
|
To test the power of the identified markers in enabling differentiation of the tumor from the healthy condition, we measured the levels of these markers in the plasma of CRC patients (n = 11), from whom it was possible to obtain blood samples after surgical removal of the tumor. Importantly, EPAS1 and UBE2D3 were significantly decreased in the postsurgery samples compared to the presurgery samples from the same patients (Table 1
and Supplementary Fig. 2B in the online Data Supplement).
To further explore the discriminating power of the markers identified in this study, we applied a supervised learning algorithm to the Q-RT-PCR dataset, excluding the pre- and postsurgery data and using the resulting dataset as the training set. Support vector machine (SVM) analysis with leave-one-out cross-validation of this training set using the 3 validated genes UBE2D3, EPAS1, and KIAA0101, showed that they enabled classification of up to 71% of the training samples correctly (52 of 73 samples) (see Fig. 1B
). Use of only 2 markers, UBE2D3 and EPAS1, did not improve scores obtained with the 3 genes together. Using the model generated by SVM with the 3 markers, we performed class prediction on a test set composed of the Q-RT-PCR data derived from the pre- and postsurgery group. In this way, we classified 77% of the samples correctly (17 of 22 samples), i.e., presurgery samples were classified as CRC and postsurgery samples were classified as normal (Fig. 1B
). The misidentified samples were 3 presurgery samples that were wrongly classified as normal and 2 postsurgery samples assigned to the CRC group.
Simultaneous monitoring of the expression of numerous genes by DNA microarrays provides a powerful tool in medical research, but the widespread clinical application of DNA microarrays is hindered by the need for sample collection directly from the tumors. Analysis of circulating RNA in the plasma circumvents this limitation, making sample collection easy and reproducible, and allowing for reiterative extractions during treatment response. This proof-of-concept study demonstrates the feasibility of such an approach. Our results provide an example of the power of plasma RNA analysis to differentiate tumor from the healthy condition in a clinical setting. We observed that some of our markers, present at high concentrations in CRC, returned to normal after surgical removal of the tumor. Furthermore, class prediction using SVM classified the presurgery samples as members of the CRC group and the postsurgery samples as part of the normal group.
On the basis of our results, large-scale gene expression profiling of a large number of samples should yield candidate markers of potential diagnostic and prognostic value.
Acknowledgments
Grant/funding support: This work was mainly funded by Grant FUGEDAD from the Spanish Ministry of Education and Science (MEC) (to M.S., F.B., and A.Z.). Additional support was obtained for the laboratory of M.S. from the CNIO, the MEC, and the European Union (INTACT and PROTEOMAGE). The laboratory of F.B. was also funded by MEC and Fundacion Mutua Madrileña.
Financial disclosures: None declared.
Acknowledgments: We thank Joaquin Dopazo and Ignacio Medina from Centro de Investigación Príncipe Felipe, Valencia, for their excellent support with class prediction.
References
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
: training set). The models generated were used to classify the pre- and postsurgery samples (
: test set). The graph shows the percentage of correctly classified samples in each case. Correctly classified and total number of samples are shown on top of each column.